| 1 | /* regcomp.c |
| 2 | */ |
| 3 | |
| 4 | /* |
| 5 | * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee |
| 6 | * |
| 7 | * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"] |
| 8 | */ |
| 9 | |
| 10 | /* This file contains functions for compiling a regular expression. See |
| 11 | * also regexec.c which funnily enough, contains functions for executing |
| 12 | * a regular expression. |
| 13 | * |
| 14 | * This file is also copied at build time to ext/re/re_comp.c, where |
| 15 | * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT. |
| 16 | * This causes the main functions to be compiled under new names and with |
| 17 | * debugging support added, which makes "use re 'debug'" work. |
| 18 | */ |
| 19 | |
| 20 | /* NOTE: this is derived from Henry Spencer's regexp code, and should not |
| 21 | * confused with the original package (see point 3 below). Thanks, Henry! |
| 22 | */ |
| 23 | |
| 24 | /* Additional note: this code is very heavily munged from Henry's version |
| 25 | * in places. In some spots I've traded clarity for efficiency, so don't |
| 26 | * blame Henry for some of the lack of readability. |
| 27 | */ |
| 28 | |
| 29 | /* The names of the functions have been changed from regcomp and |
| 30 | * regexec to pregcomp and pregexec in order to avoid conflicts |
| 31 | * with the POSIX routines of the same names. |
| 32 | */ |
| 33 | |
| 34 | #ifdef PERL_EXT_RE_BUILD |
| 35 | #include "re_top.h" |
| 36 | #endif |
| 37 | |
| 38 | /* |
| 39 | * pregcomp and pregexec -- regsub and regerror are not used in perl |
| 40 | * |
| 41 | * Copyright (c) 1986 by University of Toronto. |
| 42 | * Written by Henry Spencer. Not derived from licensed software. |
| 43 | * |
| 44 | * Permission is granted to anyone to use this software for any |
| 45 | * purpose on any computer system, and to redistribute it freely, |
| 46 | * subject to the following restrictions: |
| 47 | * |
| 48 | * 1. The author is not responsible for the consequences of use of |
| 49 | * this software, no matter how awful, even if they arise |
| 50 | * from defects in it. |
| 51 | * |
| 52 | * 2. The origin of this software must not be misrepresented, either |
| 53 | * by explicit claim or by omission. |
| 54 | * |
| 55 | * 3. Altered versions must be plainly marked as such, and must not |
| 56 | * be misrepresented as being the original software. |
| 57 | * |
| 58 | * |
| 59 | **** Alterations to Henry's code are... |
| 60 | **** |
| 61 | **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
| 62 | **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
| 63 | **** by Larry Wall and others |
| 64 | **** |
| 65 | **** You may distribute under the terms of either the GNU General Public |
| 66 | **** License or the Artistic License, as specified in the README file. |
| 67 | |
| 68 | * |
| 69 | * Beware that some of this code is subtly aware of the way operator |
| 70 | * precedence is structured in regular expressions. Serious changes in |
| 71 | * regular-expression syntax might require a total rethink. |
| 72 | */ |
| 73 | #include "EXTERN.h" |
| 74 | #define PERL_IN_REGCOMP_C |
| 75 | #include "perl.h" |
| 76 | |
| 77 | #define REG_COMP_C |
| 78 | #ifdef PERL_IN_XSUB_RE |
| 79 | # include "re_comp.h" |
| 80 | EXTERN_C const struct regexp_engine my_reg_engine; |
| 81 | #else |
| 82 | # include "regcomp.h" |
| 83 | #endif |
| 84 | |
| 85 | #include "invlist_inline.h" |
| 86 | #include "unicode_constants.h" |
| 87 | |
| 88 | #define HAS_NONLATIN1_FOLD_CLOSURE(i) \ |
| 89 | _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i) |
| 90 | #define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \ |
| 91 | _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i) |
| 92 | #define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c) |
| 93 | #define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c) |
| 94 | |
| 95 | #ifndef STATIC |
| 96 | #define STATIC static |
| 97 | #endif |
| 98 | |
| 99 | /* this is a chain of data about sub patterns we are processing that |
| 100 | need to be handled separately/specially in study_chunk. Its so |
| 101 | we can simulate recursion without losing state. */ |
| 102 | struct scan_frame; |
| 103 | typedef struct scan_frame { |
| 104 | regnode *last_regnode; /* last node to process in this frame */ |
| 105 | regnode *next_regnode; /* next node to process when last is reached */ |
| 106 | U32 prev_recursed_depth; |
| 107 | I32 stopparen; /* what stopparen do we use */ |
| 108 | |
| 109 | struct scan_frame *this_prev_frame; /* this previous frame */ |
| 110 | struct scan_frame *prev_frame; /* previous frame */ |
| 111 | struct scan_frame *next_frame; /* next frame */ |
| 112 | } scan_frame; |
| 113 | |
| 114 | /* Certain characters are output as a sequence with the first being a |
| 115 | * backslash. */ |
| 116 | #define isBACKSLASHED_PUNCT(c) memCHRs("-[]\\^", c) |
| 117 | |
| 118 | |
| 119 | struct RExC_state_t { |
| 120 | U32 flags; /* RXf_* are we folding, multilining? */ |
| 121 | U32 pm_flags; /* PMf_* stuff from the calling PMOP */ |
| 122 | char *precomp; /* uncompiled string. */ |
| 123 | char *precomp_end; /* pointer to end of uncompiled string. */ |
| 124 | REGEXP *rx_sv; /* The SV that is the regexp. */ |
| 125 | regexp *rx; /* perl core regexp structure */ |
| 126 | regexp_internal *rxi; /* internal data for regexp object |
| 127 | pprivate field */ |
| 128 | char *start; /* Start of input for compile */ |
| 129 | char *end; /* End of input for compile */ |
| 130 | char *parse; /* Input-scan pointer. */ |
| 131 | char *copy_start; /* start of copy of input within |
| 132 | constructed parse string */ |
| 133 | char *save_copy_start; /* Provides one level of saving |
| 134 | and restoring 'copy_start' */ |
| 135 | char *copy_start_in_input; /* Position in input string |
| 136 | corresponding to copy_start */ |
| 137 | SSize_t whilem_seen; /* number of WHILEM in this expr */ |
| 138 | regnode *emit_start; /* Start of emitted-code area */ |
| 139 | regnode_offset emit; /* Code-emit pointer */ |
| 140 | I32 naughty; /* How bad is this pattern? */ |
| 141 | I32 sawback; /* Did we see \1, ...? */ |
| 142 | SSize_t size; /* Number of regnode equivalents in |
| 143 | pattern */ |
| 144 | Size_t sets_depth; /* Counts recursion depth of already- |
| 145 | compiled regex set patterns */ |
| 146 | U32 seen; |
| 147 | |
| 148 | I32 parens_buf_size; /* #slots malloced open/close_parens */ |
| 149 | regnode_offset *open_parens; /* offsets to open parens */ |
| 150 | regnode_offset *close_parens; /* offsets to close parens */ |
| 151 | HV *paren_names; /* Paren names */ |
| 152 | |
| 153 | /* position beyond 'precomp' of the warning message furthest away from |
| 154 | * 'precomp'. During the parse, no warnings are raised for any problems |
| 155 | * earlier in the parse than this position. This works if warnings are |
| 156 | * raised the first time a given spot is parsed, and if only one |
| 157 | * independent warning is raised for any given spot */ |
| 158 | Size_t latest_warn_offset; |
| 159 | |
| 160 | I32 npar; /* Capture buffer count so far in the |
| 161 | parse, (OPEN) plus one. ("par" 0 is |
| 162 | the whole pattern)*/ |
| 163 | I32 total_par; /* During initial parse, is either 0, |
| 164 | or -1; the latter indicating a |
| 165 | reparse is needed. After that pass, |
| 166 | it is what 'npar' became after the |
| 167 | pass. Hence, it being > 0 indicates |
| 168 | we are in a reparse situation */ |
| 169 | I32 nestroot; /* root parens we are in - used by |
| 170 | accept */ |
| 171 | I32 seen_zerolen; |
| 172 | regnode *end_op; /* END node in program */ |
| 173 | I32 utf8; /* whether the pattern is utf8 or not */ |
| 174 | I32 orig_utf8; /* whether the pattern was originally in utf8 */ |
| 175 | /* XXX use this for future optimisation of case |
| 176 | * where pattern must be upgraded to utf8. */ |
| 177 | I32 uni_semantics; /* If a d charset modifier should use unicode |
| 178 | rules, even if the pattern is not in |
| 179 | utf8 */ |
| 180 | |
| 181 | I32 recurse_count; /* Number of recurse regops we have generated */ |
| 182 | regnode **recurse; /* Recurse regops */ |
| 183 | U8 *study_chunk_recursed; /* bitmap of which subs we have moved |
| 184 | through */ |
| 185 | U32 study_chunk_recursed_bytes; /* bytes in bitmap */ |
| 186 | I32 in_lookbehind; |
| 187 | I32 in_lookahead; |
| 188 | I32 contains_locale; |
| 189 | I32 override_recoding; |
| 190 | I32 recode_x_to_native; |
| 191 | I32 in_multi_char_class; |
| 192 | int code_index; /* next code_blocks[] slot */ |
| 193 | struct reg_code_blocks *code_blocks;/* positions of literal (?{}) |
| 194 | within pattern */ |
| 195 | SSize_t maxlen; /* mininum possible number of chars in string to match */ |
| 196 | scan_frame *frame_head; |
| 197 | scan_frame *frame_last; |
| 198 | U32 frame_count; |
| 199 | AV *warn_text; |
| 200 | HV *unlexed_names; |
| 201 | SV *runtime_code_qr; /* qr with the runtime code blocks */ |
| 202 | #ifdef DEBUGGING |
| 203 | const char *lastparse; |
| 204 | I32 lastnum; |
| 205 | U32 study_chunk_recursed_count; |
| 206 | AV *paren_name_list; /* idx -> name */ |
| 207 | SV *mysv1; |
| 208 | SV *mysv2; |
| 209 | |
| 210 | #define RExC_lastparse (pRExC_state->lastparse) |
| 211 | #define RExC_lastnum (pRExC_state->lastnum) |
| 212 | #define RExC_paren_name_list (pRExC_state->paren_name_list) |
| 213 | #define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count) |
| 214 | #define RExC_mysv (pRExC_state->mysv1) |
| 215 | #define RExC_mysv1 (pRExC_state->mysv1) |
| 216 | #define RExC_mysv2 (pRExC_state->mysv2) |
| 217 | |
| 218 | #endif |
| 219 | bool seen_d_op; |
| 220 | bool strict; |
| 221 | bool study_started; |
| 222 | bool in_script_run; |
| 223 | bool use_BRANCHJ; |
| 224 | bool sWARN_EXPERIMENTAL__VLB; |
| 225 | bool sWARN_EXPERIMENTAL__REGEX_SETS; |
| 226 | }; |
| 227 | |
| 228 | #define RExC_flags (pRExC_state->flags) |
| 229 | #define RExC_pm_flags (pRExC_state->pm_flags) |
| 230 | #define RExC_precomp (pRExC_state->precomp) |
| 231 | #define RExC_copy_start_in_input (pRExC_state->copy_start_in_input) |
| 232 | #define RExC_copy_start_in_constructed (pRExC_state->copy_start) |
| 233 | #define RExC_save_copy_start_in_constructed (pRExC_state->save_copy_start) |
| 234 | #define RExC_precomp_end (pRExC_state->precomp_end) |
| 235 | #define RExC_rx_sv (pRExC_state->rx_sv) |
| 236 | #define RExC_rx (pRExC_state->rx) |
| 237 | #define RExC_rxi (pRExC_state->rxi) |
| 238 | #define RExC_start (pRExC_state->start) |
| 239 | #define RExC_end (pRExC_state->end) |
| 240 | #define RExC_parse (pRExC_state->parse) |
| 241 | #define RExC_latest_warn_offset (pRExC_state->latest_warn_offset ) |
| 242 | #define RExC_whilem_seen (pRExC_state->whilem_seen) |
| 243 | #define RExC_seen_d_op (pRExC_state->seen_d_op) /* Seen something that differs |
| 244 | under /d from /u ? */ |
| 245 | |
| 246 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 247 | # define RExC_offsets (RExC_rxi->u.offsets) /* I am not like the |
| 248 | others */ |
| 249 | #endif |
| 250 | #define RExC_emit (pRExC_state->emit) |
| 251 | #define RExC_emit_start (pRExC_state->emit_start) |
| 252 | #define RExC_sawback (pRExC_state->sawback) |
| 253 | #define RExC_seen (pRExC_state->seen) |
| 254 | #define RExC_size (pRExC_state->size) |
| 255 | #define RExC_maxlen (pRExC_state->maxlen) |
| 256 | #define RExC_npar (pRExC_state->npar) |
| 257 | #define RExC_total_parens (pRExC_state->total_par) |
| 258 | #define RExC_parens_buf_size (pRExC_state->parens_buf_size) |
| 259 | #define RExC_nestroot (pRExC_state->nestroot) |
| 260 | #define RExC_seen_zerolen (pRExC_state->seen_zerolen) |
| 261 | #define RExC_utf8 (pRExC_state->utf8) |
| 262 | #define RExC_uni_semantics (pRExC_state->uni_semantics) |
| 263 | #define RExC_orig_utf8 (pRExC_state->orig_utf8) |
| 264 | #define RExC_open_parens (pRExC_state->open_parens) |
| 265 | #define RExC_close_parens (pRExC_state->close_parens) |
| 266 | #define RExC_end_op (pRExC_state->end_op) |
| 267 | #define RExC_paren_names (pRExC_state->paren_names) |
| 268 | #define RExC_recurse (pRExC_state->recurse) |
| 269 | #define RExC_recurse_count (pRExC_state->recurse_count) |
| 270 | #define RExC_sets_depth (pRExC_state->sets_depth) |
| 271 | #define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed) |
| 272 | #define RExC_study_chunk_recursed_bytes \ |
| 273 | (pRExC_state->study_chunk_recursed_bytes) |
| 274 | #define RExC_in_lookbehind (pRExC_state->in_lookbehind) |
| 275 | #define RExC_in_lookahead (pRExC_state->in_lookahead) |
| 276 | #define RExC_contains_locale (pRExC_state->contains_locale) |
| 277 | #define RExC_recode_x_to_native (pRExC_state->recode_x_to_native) |
| 278 | |
| 279 | #ifdef EBCDIC |
| 280 | # define SET_recode_x_to_native(x) \ |
| 281 | STMT_START { RExC_recode_x_to_native = (x); } STMT_END |
| 282 | #else |
| 283 | # define SET_recode_x_to_native(x) NOOP |
| 284 | #endif |
| 285 | |
| 286 | #define RExC_in_multi_char_class (pRExC_state->in_multi_char_class) |
| 287 | #define RExC_frame_head (pRExC_state->frame_head) |
| 288 | #define RExC_frame_last (pRExC_state->frame_last) |
| 289 | #define RExC_frame_count (pRExC_state->frame_count) |
| 290 | #define RExC_strict (pRExC_state->strict) |
| 291 | #define RExC_study_started (pRExC_state->study_started) |
| 292 | #define RExC_warn_text (pRExC_state->warn_text) |
| 293 | #define RExC_in_script_run (pRExC_state->in_script_run) |
| 294 | #define RExC_use_BRANCHJ (pRExC_state->use_BRANCHJ) |
| 295 | #define RExC_warned_WARN_EXPERIMENTAL__VLB (pRExC_state->sWARN_EXPERIMENTAL__VLB) |
| 296 | #define RExC_warned_WARN_EXPERIMENTAL__REGEX_SETS (pRExC_state->sWARN_EXPERIMENTAL__REGEX_SETS) |
| 297 | #define RExC_unlexed_names (pRExC_state->unlexed_names) |
| 298 | |
| 299 | /* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set |
| 300 | * a flag to disable back-off on the fixed/floating substrings - if it's |
| 301 | * a high complexity pattern we assume the benefit of avoiding a full match |
| 302 | * is worth the cost of checking for the substrings even if they rarely help. |
| 303 | */ |
| 304 | #define RExC_naughty (pRExC_state->naughty) |
| 305 | #define TOO_NAUGHTY (10) |
| 306 | #define MARK_NAUGHTY(add) \ |
| 307 | if (RExC_naughty < TOO_NAUGHTY) \ |
| 308 | RExC_naughty += (add) |
| 309 | #define MARK_NAUGHTY_EXP(exp, add) \ |
| 310 | if (RExC_naughty < TOO_NAUGHTY) \ |
| 311 | RExC_naughty += RExC_naughty / (exp) + (add) |
| 312 | |
| 313 | #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?') |
| 314 | #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \ |
| 315 | ((*s) == '{' && regcurly(s))) |
| 316 | |
| 317 | /* |
| 318 | * Flags to be passed up and down. |
| 319 | */ |
| 320 | #define WORST 0 /* Worst case. */ |
| 321 | #define HASWIDTH 0x01 /* Known to not match null strings, could match |
| 322 | non-null ones. */ |
| 323 | |
| 324 | /* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single |
| 325 | * character. (There needs to be a case: in the switch statement in regexec.c |
| 326 | * for any node marked SIMPLE.) Note that this is not the same thing as |
| 327 | * REGNODE_SIMPLE */ |
| 328 | #define SIMPLE 0x02 |
| 329 | #define SPSTART 0x04 /* Starts with * or + */ |
| 330 | #define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */ |
| 331 | #define TRYAGAIN 0x10 /* Weeded out a declaration. */ |
| 332 | #define RESTART_PARSE 0x20 /* Need to redo the parse */ |
| 333 | #define NEED_UTF8 0x40 /* In conjunction with RESTART_PARSE, need to |
| 334 | calcuate sizes as UTF-8 */ |
| 335 | |
| 336 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1) |
| 337 | |
| 338 | /* whether trie related optimizations are enabled */ |
| 339 | #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION |
| 340 | #define TRIE_STUDY_OPT |
| 341 | #define FULL_TRIE_STUDY |
| 342 | #define TRIE_STCLASS |
| 343 | #endif |
| 344 | |
| 345 | |
| 346 | |
| 347 | #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3] |
| 348 | #define PBITVAL(paren) (1 << ((paren) & 7)) |
| 349 | #define PAREN_OFFSET(depth) \ |
| 350 | (RExC_study_chunk_recursed + (depth) * RExC_study_chunk_recursed_bytes) |
| 351 | #define PAREN_TEST(depth, paren) \ |
| 352 | (PBYTE(PAREN_OFFSET(depth), paren) & PBITVAL(paren)) |
| 353 | #define PAREN_SET(depth, paren) \ |
| 354 | (PBYTE(PAREN_OFFSET(depth), paren) |= PBITVAL(paren)) |
| 355 | #define PAREN_UNSET(depth, paren) \ |
| 356 | (PBYTE(PAREN_OFFSET(depth), paren) &= ~PBITVAL(paren)) |
| 357 | |
| 358 | #define REQUIRE_UTF8(flagp) STMT_START { \ |
| 359 | if (!UTF) { \ |
| 360 | *flagp = RESTART_PARSE|NEED_UTF8; \ |
| 361 | return 0; \ |
| 362 | } \ |
| 363 | } STMT_END |
| 364 | |
| 365 | /* Change from /d into /u rules, and restart the parse. RExC_uni_semantics is |
| 366 | * a flag that indicates we need to override /d with /u as a result of |
| 367 | * something in the pattern. It should only be used in regards to calling |
| 368 | * set_regex_charset() or get_regex_charset() */ |
| 369 | #define REQUIRE_UNI_RULES(flagp, restart_retval) \ |
| 370 | STMT_START { \ |
| 371 | if (DEPENDS_SEMANTICS) { \ |
| 372 | set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \ |
| 373 | RExC_uni_semantics = 1; \ |
| 374 | if (RExC_seen_d_op && LIKELY(! IN_PARENS_PASS)) { \ |
| 375 | /* No need to restart the parse if we haven't seen \ |
| 376 | * anything that differs between /u and /d, and no need \ |
| 377 | * to restart immediately if we're going to reparse \ |
| 378 | * anyway to count parens */ \ |
| 379 | *flagp |= RESTART_PARSE; \ |
| 380 | return restart_retval; \ |
| 381 | } \ |
| 382 | } \ |
| 383 | } STMT_END |
| 384 | |
| 385 | #define REQUIRE_BRANCHJ(flagp, restart_retval) \ |
| 386 | STMT_START { \ |
| 387 | RExC_use_BRANCHJ = 1; \ |
| 388 | *flagp |= RESTART_PARSE; \ |
| 389 | return restart_retval; \ |
| 390 | } STMT_END |
| 391 | |
| 392 | /* Until we have completed the parse, we leave RExC_total_parens at 0 or |
| 393 | * less. After that, it must always be positive, because the whole re is |
| 394 | * considered to be surrounded by virtual parens. Setting it to negative |
| 395 | * indicates there is some construct that needs to know the actual number of |
| 396 | * parens to be properly handled. And that means an extra pass will be |
| 397 | * required after we've counted them all */ |
| 398 | #define ALL_PARENS_COUNTED (RExC_total_parens > 0) |
| 399 | #define REQUIRE_PARENS_PASS \ |
| 400 | STMT_START { /* No-op if have completed a pass */ \ |
| 401 | if (! ALL_PARENS_COUNTED) RExC_total_parens = -1; \ |
| 402 | } STMT_END |
| 403 | #define IN_PARENS_PASS (RExC_total_parens < 0) |
| 404 | |
| 405 | |
| 406 | /* This is used to return failure (zero) early from the calling function if |
| 407 | * various flags in 'flags' are set. Two flags always cause a return: |
| 408 | * 'RESTART_PARSE' and 'NEED_UTF8'. 'extra' can be used to specify any |
| 409 | * additional flags that should cause a return; 0 if none. If the return will |
| 410 | * be done, '*flagp' is first set to be all of the flags that caused the |
| 411 | * return. */ |
| 412 | #define RETURN_FAIL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \ |
| 413 | STMT_START { \ |
| 414 | if ((flags) & (RESTART_PARSE|NEED_UTF8|(extra))) { \ |
| 415 | *(flagp) = (flags) & (RESTART_PARSE|NEED_UTF8|(extra)); \ |
| 416 | return 0; \ |
| 417 | } \ |
| 418 | } STMT_END |
| 419 | |
| 420 | #define MUST_RESTART(flags) ((flags) & (RESTART_PARSE)) |
| 421 | |
| 422 | #define RETURN_FAIL_ON_RESTART(flags,flagp) \ |
| 423 | RETURN_FAIL_ON_RESTART_OR_FLAGS( flags, flagp, 0) |
| 424 | #define RETURN_FAIL_ON_RESTART_FLAGP(flagp) \ |
| 425 | if (MUST_RESTART(*(flagp))) return 0 |
| 426 | |
| 427 | /* This converts the named class defined in regcomp.h to its equivalent class |
| 428 | * number defined in handy.h. */ |
| 429 | #define namedclass_to_classnum(class) ((int) ((class) / 2)) |
| 430 | #define classnum_to_namedclass(classnum) ((classnum) * 2) |
| 431 | |
| 432 | #define _invlist_union_complement_2nd(a, b, output) \ |
| 433 | _invlist_union_maybe_complement_2nd(a, b, TRUE, output) |
| 434 | #define _invlist_intersection_complement_2nd(a, b, output) \ |
| 435 | _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output) |
| 436 | |
| 437 | /* We add a marker if we are deferring expansion of a property that is both |
| 438 | * 1) potentiallly user-defined; and |
| 439 | * 2) could also be an official Unicode property. |
| 440 | * |
| 441 | * Without this marker, any deferred expansion can only be for a user-defined |
| 442 | * one. This marker shouldn't conflict with any that could be in a legal name, |
| 443 | * and is appended to its name to indicate this. There is a string and |
| 444 | * character form */ |
| 445 | #define DEFERRED_COULD_BE_OFFICIAL_MARKERs "~" |
| 446 | #define DEFERRED_COULD_BE_OFFICIAL_MARKERc '~' |
| 447 | |
| 448 | /* What is infinity for optimization purposes */ |
| 449 | #define OPTIMIZE_INFTY SSize_t_MAX |
| 450 | |
| 451 | /* About scan_data_t. |
| 452 | |
| 453 | During optimisation we recurse through the regexp program performing |
| 454 | various inplace (keyhole style) optimisations. In addition study_chunk |
| 455 | and scan_commit populate this data structure with information about |
| 456 | what strings MUST appear in the pattern. We look for the longest |
| 457 | string that must appear at a fixed location, and we look for the |
| 458 | longest string that may appear at a floating location. So for instance |
| 459 | in the pattern: |
| 460 | |
| 461 | /FOO[xX]A.*B[xX]BAR/ |
| 462 | |
| 463 | Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating |
| 464 | strings (because they follow a .* construct). study_chunk will identify |
| 465 | both FOO and BAR as being the longest fixed and floating strings respectively. |
| 466 | |
| 467 | The strings can be composites, for instance |
| 468 | |
| 469 | /(f)(o)(o)/ |
| 470 | |
| 471 | will result in a composite fixed substring 'foo'. |
| 472 | |
| 473 | For each string some basic information is maintained: |
| 474 | |
| 475 | - min_offset |
| 476 | This is the position the string must appear at, or not before. |
| 477 | It also implicitly (when combined with minlenp) tells us how many |
| 478 | characters must match before the string we are searching for. |
| 479 | Likewise when combined with minlenp and the length of the string it |
| 480 | tells us how many characters must appear after the string we have |
| 481 | found. |
| 482 | |
| 483 | - max_offset |
| 484 | Only used for floating strings. This is the rightmost point that |
| 485 | the string can appear at. If set to OPTIMIZE_INFTY it indicates that the |
| 486 | string can occur infinitely far to the right. |
| 487 | For fixed strings, it is equal to min_offset. |
| 488 | |
| 489 | - minlenp |
| 490 | A pointer to the minimum number of characters of the pattern that the |
| 491 | string was found inside. This is important as in the case of positive |
| 492 | lookahead or positive lookbehind we can have multiple patterns |
| 493 | involved. Consider |
| 494 | |
| 495 | /(?=FOO).*F/ |
| 496 | |
| 497 | The minimum length of the pattern overall is 3, the minimum length |
| 498 | of the lookahead part is 3, but the minimum length of the part that |
| 499 | will actually match is 1. So 'FOO's minimum length is 3, but the |
| 500 | minimum length for the F is 1. This is important as the minimum length |
| 501 | is used to determine offsets in front of and behind the string being |
| 502 | looked for. Since strings can be composites this is the length of the |
| 503 | pattern at the time it was committed with a scan_commit. Note that |
| 504 | the length is calculated by study_chunk, so that the minimum lengths |
| 505 | are not known until the full pattern has been compiled, thus the |
| 506 | pointer to the value. |
| 507 | |
| 508 | - lookbehind |
| 509 | |
| 510 | In the case of lookbehind the string being searched for can be |
| 511 | offset past the start point of the final matching string. |
| 512 | If this value was just blithely removed from the min_offset it would |
| 513 | invalidate some of the calculations for how many chars must match |
| 514 | before or after (as they are derived from min_offset and minlen and |
| 515 | the length of the string being searched for). |
| 516 | When the final pattern is compiled and the data is moved from the |
| 517 | scan_data_t structure into the regexp structure the information |
| 518 | about lookbehind is factored in, with the information that would |
| 519 | have been lost precalculated in the end_shift field for the |
| 520 | associated string. |
| 521 | |
| 522 | The fields pos_min and pos_delta are used to store the minimum offset |
| 523 | and the delta to the maximum offset at the current point in the pattern. |
| 524 | |
| 525 | */ |
| 526 | |
| 527 | struct scan_data_substrs { |
| 528 | SV *str; /* longest substring found in pattern */ |
| 529 | SSize_t min_offset; /* earliest point in string it can appear */ |
| 530 | SSize_t max_offset; /* latest point in string it can appear */ |
| 531 | SSize_t *minlenp; /* pointer to the minlen relevant to the string */ |
| 532 | SSize_t lookbehind; /* is the pos of the string modified by LB */ |
| 533 | I32 flags; /* per substring SF_* and SCF_* flags */ |
| 534 | }; |
| 535 | |
| 536 | typedef struct scan_data_t { |
| 537 | /*I32 len_min; unused */ |
| 538 | /*I32 len_delta; unused */ |
| 539 | SSize_t pos_min; |
| 540 | SSize_t pos_delta; |
| 541 | SV *last_found; |
| 542 | SSize_t last_end; /* min value, <0 unless valid. */ |
| 543 | SSize_t last_start_min; |
| 544 | SSize_t last_start_max; |
| 545 | U8 cur_is_floating; /* whether the last_* values should be set as |
| 546 | * the next fixed (0) or floating (1) |
| 547 | * substring */ |
| 548 | |
| 549 | /* [0] is longest fixed substring so far, [1] is longest float so far */ |
| 550 | struct scan_data_substrs substrs[2]; |
| 551 | |
| 552 | I32 flags; /* common SF_* and SCF_* flags */ |
| 553 | I32 whilem_c; |
| 554 | SSize_t *last_closep; |
| 555 | regnode_ssc *start_class; |
| 556 | } scan_data_t; |
| 557 | |
| 558 | /* |
| 559 | * Forward declarations for pregcomp()'s friends. |
| 560 | */ |
| 561 | |
| 562 | static const scan_data_t zero_scan_data = { |
| 563 | 0, 0, NULL, 0, 0, 0, 0, |
| 564 | { |
| 565 | { NULL, 0, 0, 0, 0, 0 }, |
| 566 | { NULL, 0, 0, 0, 0, 0 }, |
| 567 | }, |
| 568 | 0, 0, NULL, NULL |
| 569 | }; |
| 570 | |
| 571 | /* study flags */ |
| 572 | |
| 573 | #define SF_BEFORE_SEOL 0x0001 |
| 574 | #define SF_BEFORE_MEOL 0x0002 |
| 575 | #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL) |
| 576 | |
| 577 | #define SF_IS_INF 0x0040 |
| 578 | #define SF_HAS_PAR 0x0080 |
| 579 | #define SF_IN_PAR 0x0100 |
| 580 | #define SF_HAS_EVAL 0x0200 |
| 581 | |
| 582 | |
| 583 | /* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the |
| 584 | * longest substring in the pattern. When it is not set the optimiser keeps |
| 585 | * track of position, but does not keep track of the actual strings seen, |
| 586 | * |
| 587 | * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but |
| 588 | * /foo/i will not. |
| 589 | * |
| 590 | * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble" |
| 591 | * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be |
| 592 | * turned off because of the alternation (BRANCH). */ |
| 593 | #define SCF_DO_SUBSTR 0x0400 |
| 594 | |
| 595 | #define SCF_DO_STCLASS_AND 0x0800 |
| 596 | #define SCF_DO_STCLASS_OR 0x1000 |
| 597 | #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR) |
| 598 | #define SCF_WHILEM_VISITED_POS 0x2000 |
| 599 | |
| 600 | #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */ |
| 601 | #define SCF_SEEN_ACCEPT 0x8000 |
| 602 | #define SCF_TRIE_DOING_RESTUDY 0x10000 |
| 603 | #define SCF_IN_DEFINE 0x20000 |
| 604 | |
| 605 | |
| 606 | |
| 607 | |
| 608 | #define UTF cBOOL(RExC_utf8) |
| 609 | |
| 610 | /* The enums for all these are ordered so things work out correctly */ |
| 611 | #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET) |
| 612 | #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \ |
| 613 | == REGEX_DEPENDS_CHARSET) |
| 614 | #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET) |
| 615 | #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \ |
| 616 | >= REGEX_UNICODE_CHARSET) |
| 617 | #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \ |
| 618 | == REGEX_ASCII_RESTRICTED_CHARSET) |
| 619 | #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \ |
| 620 | >= REGEX_ASCII_RESTRICTED_CHARSET) |
| 621 | #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \ |
| 622 | == REGEX_ASCII_MORE_RESTRICTED_CHARSET) |
| 623 | |
| 624 | #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD) |
| 625 | |
| 626 | /* For programs that want to be strictly Unicode compatible by dying if any |
| 627 | * attempt is made to match a non-Unicode code point against a Unicode |
| 628 | * property. */ |
| 629 | #define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE)) |
| 630 | |
| 631 | #define OOB_NAMEDCLASS -1 |
| 632 | |
| 633 | /* There is no code point that is out-of-bounds, so this is problematic. But |
| 634 | * its only current use is to initialize a variable that is always set before |
| 635 | * looked at. */ |
| 636 | #define OOB_UNICODE 0xDEADBEEF |
| 637 | |
| 638 | #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv)) |
| 639 | |
| 640 | |
| 641 | /* length of regex to show in messages that don't mark a position within */ |
| 642 | #define RegexLengthToShowInErrorMessages 127 |
| 643 | |
| 644 | /* |
| 645 | * If MARKER[12] are adjusted, be sure to adjust the constants at the top |
| 646 | * of t/op/regmesg.t, the tests in t/op/re_tests, and those in |
| 647 | * op/pragma/warn/regcomp. |
| 648 | */ |
| 649 | #define MARKER1 "<-- HERE" /* marker as it appears in the description */ |
| 650 | #define MARKER2 " <-- HERE " /* marker as it appears within the regex */ |
| 651 | |
| 652 | #define REPORT_LOCATION " in regex; marked by " MARKER1 \ |
| 653 | " in m/%" UTF8f MARKER2 "%" UTF8f "/" |
| 654 | |
| 655 | /* The code in this file in places uses one level of recursion with parsing |
| 656 | * rebased to an alternate string constructed by us in memory. This can take |
| 657 | * the form of something that is completely different from the input, or |
| 658 | * something that uses the input as part of the alternate. In the first case, |
| 659 | * there should be no possibility of an error, as we are in complete control of |
| 660 | * the alternate string. But in the second case we don't completely control |
| 661 | * the input portion, so there may be errors in that. Here's an example: |
| 662 | * /[abc\x{DF}def]/ui |
| 663 | * is handled specially because \x{df} folds to a sequence of more than one |
| 664 | * character: 'ss'. What is done is to create and parse an alternate string, |
| 665 | * which looks like this: |
| 666 | * /(?:\x{DF}|[abc\x{DF}def])/ui |
| 667 | * where it uses the input unchanged in the middle of something it constructs, |
| 668 | * which is a branch for the DF outside the character class, and clustering |
| 669 | * parens around the whole thing. (It knows enough to skip the DF inside the |
| 670 | * class while in this substitute parse.) 'abc' and 'def' may have errors that |
| 671 | * need to be reported. The general situation looks like this: |
| 672 | * |
| 673 | * |<------- identical ------>| |
| 674 | * sI tI xI eI |
| 675 | * Input: --------------------------------------------------------------- |
| 676 | * Constructed: --------------------------------------------------- |
| 677 | * sC tC xC eC EC |
| 678 | * |<------- identical ------>| |
| 679 | * |
| 680 | * sI..eI is the portion of the input pattern we are concerned with here. |
| 681 | * sC..EC is the constructed substitute parse string. |
| 682 | * sC..tC is constructed by us |
| 683 | * tC..eC is an exact duplicate of the portion of the input pattern tI..eI. |
| 684 | * In the diagram, these are vertically aligned. |
| 685 | * eC..EC is also constructed by us. |
| 686 | * xC is the position in the substitute parse string where we found a |
| 687 | * problem. |
| 688 | * xI is the position in the original pattern corresponding to xC. |
| 689 | * |
| 690 | * We want to display a message showing the real input string. Thus we need to |
| 691 | * translate from xC to xI. We know that xC >= tC, since the portion of the |
| 692 | * string sC..tC has been constructed by us, and so shouldn't have errors. We |
| 693 | * get: |
| 694 | * xI = tI + (xC - tC) |
| 695 | * |
| 696 | * When the substitute parse is constructed, the code needs to set: |
| 697 | * RExC_start (sC) |
| 698 | * RExC_end (eC) |
| 699 | * RExC_copy_start_in_input (tI) |
| 700 | * RExC_copy_start_in_constructed (tC) |
| 701 | * and restore them when done. |
| 702 | * |
| 703 | * During normal processing of the input pattern, both |
| 704 | * 'RExC_copy_start_in_input' and 'RExC_copy_start_in_constructed' are set to |
| 705 | * sI, so that xC equals xI. |
| 706 | */ |
| 707 | |
| 708 | #define sI RExC_precomp |
| 709 | #define eI RExC_precomp_end |
| 710 | #define sC RExC_start |
| 711 | #define eC RExC_end |
| 712 | #define tI RExC_copy_start_in_input |
| 713 | #define tC RExC_copy_start_in_constructed |
| 714 | #define xI(xC) (tI + (xC - tC)) |
| 715 | #define xI_offset(xC) (xI(xC) - sI) |
| 716 | |
| 717 | #define REPORT_LOCATION_ARGS(xC) \ |
| 718 | UTF8fARG(UTF, \ |
| 719 | (xI(xC) > eI) /* Don't run off end */ \ |
| 720 | ? eI - sI /* Length before the <--HERE */ \ |
| 721 | : ((xI_offset(xC) >= 0) \ |
| 722 | ? xI_offset(xC) \ |
| 723 | : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \ |
| 724 | IVdf " trying to output message for " \ |
| 725 | " pattern %.*s", \ |
| 726 | __FILE__, __LINE__, (IV) xI_offset(xC), \ |
| 727 | ((int) (eC - sC)), sC), 0)), \ |
| 728 | sI), /* The input pattern printed up to the <--HERE */ \ |
| 729 | UTF8fARG(UTF, \ |
| 730 | (xI(xC) > eI) ? 0 : eI - xI(xC), /* Length after <--HERE */ \ |
| 731 | (xI(xC) > eI) ? eI : xI(xC)) /* pattern after <--HERE */ |
| 732 | |
| 733 | /* Used to point after bad bytes for an error message, but avoid skipping |
| 734 | * past a nul byte. */ |
| 735 | #define SKIP_IF_CHAR(s, e) (!*(s) ? 0 : UTF ? UTF8_SAFE_SKIP(s, e) : 1) |
| 736 | |
| 737 | /* Set up to clean up after our imminent demise */ |
| 738 | #define PREPARE_TO_DIE \ |
| 739 | STMT_START { \ |
| 740 | if (RExC_rx_sv) \ |
| 741 | SAVEFREESV(RExC_rx_sv); \ |
| 742 | if (RExC_open_parens) \ |
| 743 | SAVEFREEPV(RExC_open_parens); \ |
| 744 | if (RExC_close_parens) \ |
| 745 | SAVEFREEPV(RExC_close_parens); \ |
| 746 | } STMT_END |
| 747 | |
| 748 | /* |
| 749 | * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given |
| 750 | * arg. Show regex, up to a maximum length. If it's too long, chop and add |
| 751 | * "...". |
| 752 | */ |
| 753 | #define _FAIL(code) STMT_START { \ |
| 754 | const char *ellipses = ""; \ |
| 755 | IV len = RExC_precomp_end - RExC_precomp; \ |
| 756 | \ |
| 757 | PREPARE_TO_DIE; \ |
| 758 | if (len > RegexLengthToShowInErrorMessages) { \ |
| 759 | /* chop 10 shorter than the max, to ensure meaning of "..." */ \ |
| 760 | len = RegexLengthToShowInErrorMessages - 10; \ |
| 761 | ellipses = "..."; \ |
| 762 | } \ |
| 763 | code; \ |
| 764 | } STMT_END |
| 765 | |
| 766 | #define FAIL(msg) _FAIL( \ |
| 767 | Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \ |
| 768 | msg, UTF8fARG(UTF, len, RExC_precomp), ellipses)) |
| 769 | |
| 770 | #define FAIL2(msg,arg) _FAIL( \ |
| 771 | Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \ |
| 772 | arg, UTF8fARG(UTF, len, RExC_precomp), ellipses)) |
| 773 | |
| 774 | #define FAIL3(msg,arg1,arg2) _FAIL( \ |
| 775 | Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \ |
| 776 | arg1, arg2, UTF8fARG(UTF, len, RExC_precomp), ellipses)) |
| 777 | |
| 778 | /* |
| 779 | * Simple_vFAIL -- like FAIL, but marks the current location in the scan |
| 780 | */ |
| 781 | #define Simple_vFAIL(m) STMT_START { \ |
| 782 | Perl_croak(aTHX_ "%s" REPORT_LOCATION, \ |
| 783 | m, REPORT_LOCATION_ARGS(RExC_parse)); \ |
| 784 | } STMT_END |
| 785 | |
| 786 | /* |
| 787 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL() |
| 788 | */ |
| 789 | #define vFAIL(m) STMT_START { \ |
| 790 | PREPARE_TO_DIE; \ |
| 791 | Simple_vFAIL(m); \ |
| 792 | } STMT_END |
| 793 | |
| 794 | /* |
| 795 | * Like Simple_vFAIL(), but accepts two arguments. |
| 796 | */ |
| 797 | #define Simple_vFAIL2(m,a1) STMT_START { \ |
| 798 | S_re_croak(aTHX_ UTF, m REPORT_LOCATION, a1, \ |
| 799 | REPORT_LOCATION_ARGS(RExC_parse)); \ |
| 800 | } STMT_END |
| 801 | |
| 802 | /* |
| 803 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2(). |
| 804 | */ |
| 805 | #define vFAIL2(m,a1) STMT_START { \ |
| 806 | PREPARE_TO_DIE; \ |
| 807 | Simple_vFAIL2(m, a1); \ |
| 808 | } STMT_END |
| 809 | |
| 810 | |
| 811 | /* |
| 812 | * Like Simple_vFAIL(), but accepts three arguments. |
| 813 | */ |
| 814 | #define Simple_vFAIL3(m, a1, a2) STMT_START { \ |
| 815 | S_re_croak(aTHX_ UTF, m REPORT_LOCATION, a1, a2, \ |
| 816 | REPORT_LOCATION_ARGS(RExC_parse)); \ |
| 817 | } STMT_END |
| 818 | |
| 819 | /* |
| 820 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3(). |
| 821 | */ |
| 822 | #define vFAIL3(m,a1,a2) STMT_START { \ |
| 823 | PREPARE_TO_DIE; \ |
| 824 | Simple_vFAIL3(m, a1, a2); \ |
| 825 | } STMT_END |
| 826 | |
| 827 | /* |
| 828 | * Like Simple_vFAIL(), but accepts four arguments. |
| 829 | */ |
| 830 | #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \ |
| 831 | S_re_croak(aTHX_ UTF, m REPORT_LOCATION, a1, a2, a3, \ |
| 832 | REPORT_LOCATION_ARGS(RExC_parse)); \ |
| 833 | } STMT_END |
| 834 | |
| 835 | #define vFAIL4(m,a1,a2,a3) STMT_START { \ |
| 836 | PREPARE_TO_DIE; \ |
| 837 | Simple_vFAIL4(m, a1, a2, a3); \ |
| 838 | } STMT_END |
| 839 | |
| 840 | /* A specialized version of vFAIL2 that works with UTF8f */ |
| 841 | #define vFAIL2utf8f(m, a1) STMT_START { \ |
| 842 | PREPARE_TO_DIE; \ |
| 843 | S_re_croak(aTHX_ UTF, m REPORT_LOCATION, a1, \ |
| 844 | REPORT_LOCATION_ARGS(RExC_parse)); \ |
| 845 | } STMT_END |
| 846 | |
| 847 | #define vFAIL3utf8f(m, a1, a2) STMT_START { \ |
| 848 | PREPARE_TO_DIE; \ |
| 849 | S_re_croak(aTHX_ UTF, m REPORT_LOCATION, a1, a2, \ |
| 850 | REPORT_LOCATION_ARGS(RExC_parse)); \ |
| 851 | } STMT_END |
| 852 | |
| 853 | /* Setting this to NULL is a signal to not output warnings */ |
| 854 | #define TURN_OFF_WARNINGS_IN_SUBSTITUTE_PARSE \ |
| 855 | STMT_START { \ |
| 856 | RExC_save_copy_start_in_constructed = RExC_copy_start_in_constructed;\ |
| 857 | RExC_copy_start_in_constructed = NULL; \ |
| 858 | } STMT_END |
| 859 | #define RESTORE_WARNINGS \ |
| 860 | RExC_copy_start_in_constructed = RExC_save_copy_start_in_constructed |
| 861 | |
| 862 | /* Since a warning can be generated multiple times as the input is reparsed, we |
| 863 | * output it the first time we come to that point in the parse, but suppress it |
| 864 | * otherwise. 'RExC_copy_start_in_constructed' being NULL is a flag to not |
| 865 | * generate any warnings */ |
| 866 | #define TO_OUTPUT_WARNINGS(loc) \ |
| 867 | ( RExC_copy_start_in_constructed \ |
| 868 | && ((xI(loc)) - RExC_precomp) > (Ptrdiff_t) RExC_latest_warn_offset) |
| 869 | |
| 870 | /* After we've emitted a warning, we save the position in the input so we don't |
| 871 | * output it again */ |
| 872 | #define UPDATE_WARNINGS_LOC(loc) \ |
| 873 | STMT_START { \ |
| 874 | if (TO_OUTPUT_WARNINGS(loc)) { \ |
| 875 | RExC_latest_warn_offset = MAX(sI, MIN(eI, xI(loc))) \ |
| 876 | - RExC_precomp; \ |
| 877 | } \ |
| 878 | } STMT_END |
| 879 | |
| 880 | /* 'warns' is the output of the packWARNx macro used in 'code' */ |
| 881 | #define _WARN_HELPER(loc, warns, code) \ |
| 882 | STMT_START { \ |
| 883 | if (! RExC_copy_start_in_constructed) { \ |
| 884 | Perl_croak( aTHX_ "panic! %s: %d: Tried to warn when none" \ |
| 885 | " expected at '%s'", \ |
| 886 | __FILE__, __LINE__, loc); \ |
| 887 | } \ |
| 888 | if (TO_OUTPUT_WARNINGS(loc)) { \ |
| 889 | if (ckDEAD(warns)) \ |
| 890 | PREPARE_TO_DIE; \ |
| 891 | code; \ |
| 892 | UPDATE_WARNINGS_LOC(loc); \ |
| 893 | } \ |
| 894 | } STMT_END |
| 895 | |
| 896 | /* m is not necessarily a "literal string", in this macro */ |
| 897 | #define warn_non_literal_string(loc, packed_warn, m) \ |
| 898 | _WARN_HELPER(loc, packed_warn, \ |
| 899 | Perl_warner(aTHX_ packed_warn, \ |
| 900 | "%s" REPORT_LOCATION, \ |
| 901 | m, REPORT_LOCATION_ARGS(loc))) |
| 902 | #define reg_warn_non_literal_string(loc, m) \ |
| 903 | warn_non_literal_string(loc, packWARN(WARN_REGEXP), m) |
| 904 | |
| 905 | #define ckWARN2_non_literal_string(loc, packwarn, m, a1) \ |
| 906 | STMT_START { \ |
| 907 | char * format; \ |
| 908 | Size_t format_size = strlen(m) + strlen(REPORT_LOCATION)+ 1;\ |
| 909 | Newx(format, format_size, char); \ |
| 910 | my_strlcpy(format, m, format_size); \ |
| 911 | my_strlcat(format, REPORT_LOCATION, format_size); \ |
| 912 | SAVEFREEPV(format); \ |
| 913 | _WARN_HELPER(loc, packwarn, \ |
| 914 | Perl_ck_warner(aTHX_ packwarn, \ |
| 915 | format, \ |
| 916 | a1, REPORT_LOCATION_ARGS(loc))); \ |
| 917 | } STMT_END |
| 918 | |
| 919 | #define ckWARNreg(loc,m) \ |
| 920 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 921 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 922 | m REPORT_LOCATION, \ |
| 923 | REPORT_LOCATION_ARGS(loc))) |
| 924 | |
| 925 | #define vWARN(loc, m) \ |
| 926 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 927 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 928 | m REPORT_LOCATION, \ |
| 929 | REPORT_LOCATION_ARGS(loc))) \ |
| 930 | |
| 931 | #define vWARN_dep(loc, m) \ |
| 932 | _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \ |
| 933 | Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \ |
| 934 | m REPORT_LOCATION, \ |
| 935 | REPORT_LOCATION_ARGS(loc))) |
| 936 | |
| 937 | #define ckWARNdep(loc,m) \ |
| 938 | _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \ |
| 939 | Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \ |
| 940 | m REPORT_LOCATION, \ |
| 941 | REPORT_LOCATION_ARGS(loc))) |
| 942 | |
| 943 | #define ckWARNregdep(loc,m) \ |
| 944 | _WARN_HELPER(loc, packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ |
| 945 | Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \ |
| 946 | WARN_REGEXP), \ |
| 947 | m REPORT_LOCATION, \ |
| 948 | REPORT_LOCATION_ARGS(loc))) |
| 949 | |
| 950 | #define ckWARN2reg_d(loc,m, a1) \ |
| 951 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 952 | Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \ |
| 953 | m REPORT_LOCATION, \ |
| 954 | a1, REPORT_LOCATION_ARGS(loc))) |
| 955 | |
| 956 | #define ckWARN2reg(loc, m, a1) \ |
| 957 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 958 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 959 | m REPORT_LOCATION, \ |
| 960 | a1, REPORT_LOCATION_ARGS(loc))) |
| 961 | |
| 962 | #define vWARN3(loc, m, a1, a2) \ |
| 963 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 964 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 965 | m REPORT_LOCATION, \ |
| 966 | a1, a2, REPORT_LOCATION_ARGS(loc))) |
| 967 | |
| 968 | #define ckWARN3reg(loc, m, a1, a2) \ |
| 969 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 970 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 971 | m REPORT_LOCATION, \ |
| 972 | a1, a2, \ |
| 973 | REPORT_LOCATION_ARGS(loc))) |
| 974 | |
| 975 | #define vWARN4(loc, m, a1, a2, a3) \ |
| 976 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 977 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 978 | m REPORT_LOCATION, \ |
| 979 | a1, a2, a3, \ |
| 980 | REPORT_LOCATION_ARGS(loc))) |
| 981 | |
| 982 | #define ckWARN4reg(loc, m, a1, a2, a3) \ |
| 983 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 984 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 985 | m REPORT_LOCATION, \ |
| 986 | a1, a2, a3, \ |
| 987 | REPORT_LOCATION_ARGS(loc))) |
| 988 | |
| 989 | #define vWARN5(loc, m, a1, a2, a3, a4) \ |
| 990 | _WARN_HELPER(loc, packWARN(WARN_REGEXP), \ |
| 991 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), \ |
| 992 | m REPORT_LOCATION, \ |
| 993 | a1, a2, a3, a4, \ |
| 994 | REPORT_LOCATION_ARGS(loc))) |
| 995 | |
| 996 | #define ckWARNexperimental(loc, class, m) \ |
| 997 | STMT_START { \ |
| 998 | if (! RExC_warned_ ## class) { /* warn once per compilation */ \ |
| 999 | RExC_warned_ ## class = 1; \ |
| 1000 | _WARN_HELPER(loc, packWARN(class), \ |
| 1001 | Perl_ck_warner_d(aTHX_ packWARN(class), \ |
| 1002 | m REPORT_LOCATION, \ |
| 1003 | REPORT_LOCATION_ARGS(loc)));\ |
| 1004 | } \ |
| 1005 | } STMT_END |
| 1006 | |
| 1007 | /* Convert between a pointer to a node and its offset from the beginning of the |
| 1008 | * program */ |
| 1009 | #define REGNODE_p(offset) (RExC_emit_start + (offset)) |
| 1010 | #define REGNODE_OFFSET(node) ((node) - RExC_emit_start) |
| 1011 | |
| 1012 | /* Macros for recording node offsets. 20001227 mjd@plover.com |
| 1013 | * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in |
| 1014 | * element 2*n-1 of the array. Element #2n holds the byte length node #n. |
| 1015 | * Element 0 holds the number n. |
| 1016 | * Position is 1 indexed. |
| 1017 | */ |
| 1018 | #ifndef RE_TRACK_PATTERN_OFFSETS |
| 1019 | #define Set_Node_Offset_To_R(offset,byte) |
| 1020 | #define Set_Node_Offset(node,byte) |
| 1021 | #define Set_Cur_Node_Offset |
| 1022 | #define Set_Node_Length_To_R(node,len) |
| 1023 | #define Set_Node_Length(node,len) |
| 1024 | #define Set_Node_Cur_Length(node,start) |
| 1025 | #define Node_Offset(n) |
| 1026 | #define Node_Length(n) |
| 1027 | #define Set_Node_Offset_Length(node,offset,len) |
| 1028 | #define ProgLen(ri) ri->u.proglen |
| 1029 | #define SetProgLen(ri,x) ri->u.proglen = x |
| 1030 | #define Track_Code(code) |
| 1031 | #else |
| 1032 | #define ProgLen(ri) ri->u.offsets[0] |
| 1033 | #define SetProgLen(ri,x) ri->u.offsets[0] = x |
| 1034 | #define Set_Node_Offset_To_R(offset,byte) STMT_START { \ |
| 1035 | MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \ |
| 1036 | __LINE__, (int)(offset), (int)(byte))); \ |
| 1037 | if((offset) < 0) { \ |
| 1038 | Perl_croak(aTHX_ "value of node is %d in Offset macro", \ |
| 1039 | (int)(offset)); \ |
| 1040 | } else { \ |
| 1041 | RExC_offsets[2*(offset)-1] = (byte); \ |
| 1042 | } \ |
| 1043 | } STMT_END |
| 1044 | |
| 1045 | #define Set_Node_Offset(node,byte) \ |
| 1046 | Set_Node_Offset_To_R(REGNODE_OFFSET(node), (byte)-RExC_start) |
| 1047 | #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse) |
| 1048 | |
| 1049 | #define Set_Node_Length_To_R(node,len) STMT_START { \ |
| 1050 | MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \ |
| 1051 | __LINE__, (int)(node), (int)(len))); \ |
| 1052 | if((node) < 0) { \ |
| 1053 | Perl_croak(aTHX_ "value of node is %d in Length macro", \ |
| 1054 | (int)(node)); \ |
| 1055 | } else { \ |
| 1056 | RExC_offsets[2*(node)] = (len); \ |
| 1057 | } \ |
| 1058 | } STMT_END |
| 1059 | |
| 1060 | #define Set_Node_Length(node,len) \ |
| 1061 | Set_Node_Length_To_R(REGNODE_OFFSET(node), len) |
| 1062 | #define Set_Node_Cur_Length(node, start) \ |
| 1063 | Set_Node_Length(node, RExC_parse - start) |
| 1064 | |
| 1065 | /* Get offsets and lengths */ |
| 1066 | #define Node_Offset(n) (RExC_offsets[2*(REGNODE_OFFSET(n))-1]) |
| 1067 | #define Node_Length(n) (RExC_offsets[2*(REGNODE_OFFSET(n))]) |
| 1068 | |
| 1069 | #define Set_Node_Offset_Length(node,offset,len) STMT_START { \ |
| 1070 | Set_Node_Offset_To_R(REGNODE_OFFSET(node), (offset)); \ |
| 1071 | Set_Node_Length_To_R(REGNODE_OFFSET(node), (len)); \ |
| 1072 | } STMT_END |
| 1073 | |
| 1074 | #define Track_Code(code) STMT_START { code } STMT_END |
| 1075 | #endif |
| 1076 | |
| 1077 | #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS |
| 1078 | #define EXPERIMENTAL_INPLACESCAN |
| 1079 | #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/ |
| 1080 | |
| 1081 | #ifdef DEBUGGING |
| 1082 | int |
| 1083 | Perl_re_printf(pTHX_ const char *fmt, ...) |
| 1084 | { |
| 1085 | va_list ap; |
| 1086 | int result; |
| 1087 | PerlIO *f= Perl_debug_log; |
| 1088 | PERL_ARGS_ASSERT_RE_PRINTF; |
| 1089 | va_start(ap, fmt); |
| 1090 | result = PerlIO_vprintf(f, fmt, ap); |
| 1091 | va_end(ap); |
| 1092 | return result; |
| 1093 | } |
| 1094 | |
| 1095 | int |
| 1096 | Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...) |
| 1097 | { |
| 1098 | va_list ap; |
| 1099 | int result; |
| 1100 | PerlIO *f= Perl_debug_log; |
| 1101 | PERL_ARGS_ASSERT_RE_INDENTF; |
| 1102 | va_start(ap, depth); |
| 1103 | PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, ""); |
| 1104 | result = PerlIO_vprintf(f, fmt, ap); |
| 1105 | va_end(ap); |
| 1106 | return result; |
| 1107 | } |
| 1108 | #endif /* DEBUGGING */ |
| 1109 | |
| 1110 | #define DEBUG_RExC_seen() \ |
| 1111 | DEBUG_OPTIMISE_MORE_r({ \ |
| 1112 | Perl_re_printf( aTHX_ "RExC_seen: "); \ |
| 1113 | \ |
| 1114 | if (RExC_seen & REG_ZERO_LEN_SEEN) \ |
| 1115 | Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \ |
| 1116 | \ |
| 1117 | if (RExC_seen & REG_LOOKBEHIND_SEEN) \ |
| 1118 | Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \ |
| 1119 | \ |
| 1120 | if (RExC_seen & REG_GPOS_SEEN) \ |
| 1121 | Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \ |
| 1122 | \ |
| 1123 | if (RExC_seen & REG_RECURSE_SEEN) \ |
| 1124 | Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \ |
| 1125 | \ |
| 1126 | if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \ |
| 1127 | Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \ |
| 1128 | \ |
| 1129 | if (RExC_seen & REG_VERBARG_SEEN) \ |
| 1130 | Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \ |
| 1131 | \ |
| 1132 | if (RExC_seen & REG_CUTGROUP_SEEN) \ |
| 1133 | Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \ |
| 1134 | \ |
| 1135 | if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \ |
| 1136 | Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \ |
| 1137 | \ |
| 1138 | if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \ |
| 1139 | Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \ |
| 1140 | \ |
| 1141 | if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \ |
| 1142 | Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \ |
| 1143 | \ |
| 1144 | Perl_re_printf( aTHX_ "\n"); \ |
| 1145 | }); |
| 1146 | |
| 1147 | #define DEBUG_SHOW_STUDY_FLAG(flags,flag) \ |
| 1148 | if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag) |
| 1149 | |
| 1150 | |
| 1151 | #ifdef DEBUGGING |
| 1152 | static void |
| 1153 | S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str, |
| 1154 | const char *close_str) |
| 1155 | { |
| 1156 | if (!flags) |
| 1157 | return; |
| 1158 | |
| 1159 | Perl_re_printf( aTHX_ "%s", open_str); |
| 1160 | DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL); |
| 1161 | DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL); |
| 1162 | DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF); |
| 1163 | DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR); |
| 1164 | DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR); |
| 1165 | DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL); |
| 1166 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR); |
| 1167 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND); |
| 1168 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR); |
| 1169 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS); |
| 1170 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS); |
| 1171 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY); |
| 1172 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT); |
| 1173 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY); |
| 1174 | DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE); |
| 1175 | Perl_re_printf( aTHX_ "%s", close_str); |
| 1176 | } |
| 1177 | |
| 1178 | |
| 1179 | static void |
| 1180 | S_debug_studydata(pTHX_ const char *where, scan_data_t *data, |
| 1181 | U32 depth, int is_inf) |
| 1182 | { |
| 1183 | GET_RE_DEBUG_FLAGS_DECL; |
| 1184 | |
| 1185 | DEBUG_OPTIMISE_MORE_r({ |
| 1186 | if (!data) |
| 1187 | return; |
| 1188 | Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf, |
| 1189 | depth, |
| 1190 | where, |
| 1191 | (IV)data->pos_min, |
| 1192 | (IV)data->pos_delta, |
| 1193 | (UV)data->flags |
| 1194 | ); |
| 1195 | |
| 1196 | S_debug_show_study_flags(aTHX_ data->flags," [","]"); |
| 1197 | |
| 1198 | Perl_re_printf( aTHX_ |
| 1199 | " Whilem_c: %" IVdf " Lcp: %" IVdf " %s", |
| 1200 | (IV)data->whilem_c, |
| 1201 | (IV)(data->last_closep ? *((data)->last_closep) : -1), |
| 1202 | is_inf ? "INF " : "" |
| 1203 | ); |
| 1204 | |
| 1205 | if (data->last_found) { |
| 1206 | int i; |
| 1207 | Perl_re_printf(aTHX_ |
| 1208 | "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf, |
| 1209 | SvPVX_const(data->last_found), |
| 1210 | (IV)data->last_end, |
| 1211 | (IV)data->last_start_min, |
| 1212 | (IV)data->last_start_max |
| 1213 | ); |
| 1214 | |
| 1215 | for (i = 0; i < 2; i++) { |
| 1216 | Perl_re_printf(aTHX_ |
| 1217 | " %s%s: '%s' @ %" IVdf "/%" IVdf, |
| 1218 | data->cur_is_floating == i ? "*" : "", |
| 1219 | i ? "Float" : "Fixed", |
| 1220 | SvPVX_const(data->substrs[i].str), |
| 1221 | (IV)data->substrs[i].min_offset, |
| 1222 | (IV)data->substrs[i].max_offset |
| 1223 | ); |
| 1224 | S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]"); |
| 1225 | } |
| 1226 | } |
| 1227 | |
| 1228 | Perl_re_printf( aTHX_ "\n"); |
| 1229 | }); |
| 1230 | } |
| 1231 | |
| 1232 | |
| 1233 | static void |
| 1234 | S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state, |
| 1235 | regnode *scan, U32 depth, U32 flags) |
| 1236 | { |
| 1237 | GET_RE_DEBUG_FLAGS_DECL; |
| 1238 | |
| 1239 | DEBUG_OPTIMISE_r({ |
| 1240 | regnode *Next; |
| 1241 | |
| 1242 | if (!scan) |
| 1243 | return; |
| 1244 | Next = regnext(scan); |
| 1245 | regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state); |
| 1246 | Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)", |
| 1247 | depth, |
| 1248 | str, |
| 1249 | REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv), |
| 1250 | Next ? (REG_NODE_NUM(Next)) : 0 ); |
| 1251 | S_debug_show_study_flags(aTHX_ flags," [ ","]"); |
| 1252 | Perl_re_printf( aTHX_ "\n"); |
| 1253 | }); |
| 1254 | } |
| 1255 | |
| 1256 | |
| 1257 | # define DEBUG_STUDYDATA(where, data, depth, is_inf) \ |
| 1258 | S_debug_studydata(aTHX_ where, data, depth, is_inf) |
| 1259 | |
| 1260 | # define DEBUG_PEEP(str, scan, depth, flags) \ |
| 1261 | S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags) |
| 1262 | |
| 1263 | #else |
| 1264 | # define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP |
| 1265 | # define DEBUG_PEEP(str, scan, depth, flags) NOOP |
| 1266 | #endif |
| 1267 | |
| 1268 | |
| 1269 | /* ========================================================= |
| 1270 | * BEGIN edit_distance stuff. |
| 1271 | * |
| 1272 | * This calculates how many single character changes of any type are needed to |
| 1273 | * transform a string into another one. It is taken from version 3.1 of |
| 1274 | * |
| 1275 | * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS |
| 1276 | */ |
| 1277 | |
| 1278 | /* Our unsorted dictionary linked list. */ |
| 1279 | /* Note we use UVs, not chars. */ |
| 1280 | |
| 1281 | struct dictionary{ |
| 1282 | UV key; |
| 1283 | UV value; |
| 1284 | struct dictionary* next; |
| 1285 | }; |
| 1286 | typedef struct dictionary item; |
| 1287 | |
| 1288 | |
| 1289 | PERL_STATIC_INLINE item* |
| 1290 | push(UV key, item* curr) |
| 1291 | { |
| 1292 | item* head; |
| 1293 | Newx(head, 1, item); |
| 1294 | head->key = key; |
| 1295 | head->value = 0; |
| 1296 | head->next = curr; |
| 1297 | return head; |
| 1298 | } |
| 1299 | |
| 1300 | |
| 1301 | PERL_STATIC_INLINE item* |
| 1302 | find(item* head, UV key) |
| 1303 | { |
| 1304 | item* iterator = head; |
| 1305 | while (iterator){ |
| 1306 | if (iterator->key == key){ |
| 1307 | return iterator; |
| 1308 | } |
| 1309 | iterator = iterator->next; |
| 1310 | } |
| 1311 | |
| 1312 | return NULL; |
| 1313 | } |
| 1314 | |
| 1315 | PERL_STATIC_INLINE item* |
| 1316 | uniquePush(item* head, UV key) |
| 1317 | { |
| 1318 | item* iterator = head; |
| 1319 | |
| 1320 | while (iterator){ |
| 1321 | if (iterator->key == key) { |
| 1322 | return head; |
| 1323 | } |
| 1324 | iterator = iterator->next; |
| 1325 | } |
| 1326 | |
| 1327 | return push(key, head); |
| 1328 | } |
| 1329 | |
| 1330 | PERL_STATIC_INLINE void |
| 1331 | dict_free(item* head) |
| 1332 | { |
| 1333 | item* iterator = head; |
| 1334 | |
| 1335 | while (iterator) { |
| 1336 | item* temp = iterator; |
| 1337 | iterator = iterator->next; |
| 1338 | Safefree(temp); |
| 1339 | } |
| 1340 | |
| 1341 | head = NULL; |
| 1342 | } |
| 1343 | |
| 1344 | /* End of Dictionary Stuff */ |
| 1345 | |
| 1346 | /* All calculations/work are done here */ |
| 1347 | STATIC int |
| 1348 | S_edit_distance(const UV* src, |
| 1349 | const UV* tgt, |
| 1350 | const STRLEN x, /* length of src[] */ |
| 1351 | const STRLEN y, /* length of tgt[] */ |
| 1352 | const SSize_t maxDistance |
| 1353 | ) |
| 1354 | { |
| 1355 | item *head = NULL; |
| 1356 | UV swapCount, swapScore, targetCharCount, i, j; |
| 1357 | UV *scores; |
| 1358 | UV score_ceil = x + y; |
| 1359 | |
| 1360 | PERL_ARGS_ASSERT_EDIT_DISTANCE; |
| 1361 | |
| 1362 | /* intialize matrix start values */ |
| 1363 | Newx(scores, ( (x + 2) * (y + 2)), UV); |
| 1364 | scores[0] = score_ceil; |
| 1365 | scores[1 * (y + 2) + 0] = score_ceil; |
| 1366 | scores[0 * (y + 2) + 1] = score_ceil; |
| 1367 | scores[1 * (y + 2) + 1] = 0; |
| 1368 | head = uniquePush(uniquePush(head, src[0]), tgt[0]); |
| 1369 | |
| 1370 | /* work loops */ |
| 1371 | /* i = src index */ |
| 1372 | /* j = tgt index */ |
| 1373 | for (i=1;i<=x;i++) { |
| 1374 | if (i < x) |
| 1375 | head = uniquePush(head, src[i]); |
| 1376 | scores[(i+1) * (y + 2) + 1] = i; |
| 1377 | scores[(i+1) * (y + 2) + 0] = score_ceil; |
| 1378 | swapCount = 0; |
| 1379 | |
| 1380 | for (j=1;j<=y;j++) { |
| 1381 | if (i == 1) { |
| 1382 | if(j < y) |
| 1383 | head = uniquePush(head, tgt[j]); |
| 1384 | scores[1 * (y + 2) + (j + 1)] = j; |
| 1385 | scores[0 * (y + 2) + (j + 1)] = score_ceil; |
| 1386 | } |
| 1387 | |
| 1388 | targetCharCount = find(head, tgt[j-1])->value; |
| 1389 | swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount; |
| 1390 | |
| 1391 | if (src[i-1] != tgt[j-1]){ |
| 1392 | scores[(i+1) * (y + 2) + (j + 1)] = MIN(swapScore,(MIN(scores[i * (y + 2) + j], MIN(scores[(i+1) * (y + 2) + j], scores[i * (y + 2) + (j + 1)])) + 1)); |
| 1393 | } |
| 1394 | else { |
| 1395 | swapCount = j; |
| 1396 | scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore); |
| 1397 | } |
| 1398 | } |
| 1399 | |
| 1400 | find(head, src[i-1])->value = i; |
| 1401 | } |
| 1402 | |
| 1403 | { |
| 1404 | IV score = scores[(x+1) * (y + 2) + (y + 1)]; |
| 1405 | dict_free(head); |
| 1406 | Safefree(scores); |
| 1407 | return (maxDistance != 0 && maxDistance < score)?(-1):score; |
| 1408 | } |
| 1409 | } |
| 1410 | |
| 1411 | /* END of edit_distance() stuff |
| 1412 | * ========================================================= */ |
| 1413 | |
| 1414 | /* Mark that we cannot extend a found fixed substring at this point. |
| 1415 | Update the longest found anchored substring or the longest found |
| 1416 | floating substrings if needed. */ |
| 1417 | |
| 1418 | STATIC void |
| 1419 | S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, |
| 1420 | SSize_t *minlenp, int is_inf) |
| 1421 | { |
| 1422 | const STRLEN l = CHR_SVLEN(data->last_found); |
| 1423 | SV * const longest_sv = data->substrs[data->cur_is_floating].str; |
| 1424 | const STRLEN old_l = CHR_SVLEN(longest_sv); |
| 1425 | GET_RE_DEBUG_FLAGS_DECL; |
| 1426 | |
| 1427 | PERL_ARGS_ASSERT_SCAN_COMMIT; |
| 1428 | |
| 1429 | if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) { |
| 1430 | const U8 i = data->cur_is_floating; |
| 1431 | SvSetMagicSV(longest_sv, data->last_found); |
| 1432 | data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min; |
| 1433 | |
| 1434 | if (!i) /* fixed */ |
| 1435 | data->substrs[0].max_offset = data->substrs[0].min_offset; |
| 1436 | else { /* float */ |
| 1437 | data->substrs[1].max_offset = (l |
| 1438 | ? data->last_start_max |
| 1439 | : (data->pos_delta > OPTIMIZE_INFTY - data->pos_min |
| 1440 | ? OPTIMIZE_INFTY |
| 1441 | : data->pos_min + data->pos_delta)); |
| 1442 | if (is_inf |
| 1443 | || (STRLEN)data->substrs[1].max_offset > (STRLEN)OPTIMIZE_INFTY) |
| 1444 | data->substrs[1].max_offset = OPTIMIZE_INFTY; |
| 1445 | } |
| 1446 | |
| 1447 | if (data->flags & SF_BEFORE_EOL) |
| 1448 | data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL); |
| 1449 | else |
| 1450 | data->substrs[i].flags &= ~SF_BEFORE_EOL; |
| 1451 | data->substrs[i].minlenp = minlenp; |
| 1452 | data->substrs[i].lookbehind = 0; |
| 1453 | } |
| 1454 | |
| 1455 | SvCUR_set(data->last_found, 0); |
| 1456 | { |
| 1457 | SV * const sv = data->last_found; |
| 1458 | if (SvUTF8(sv) && SvMAGICAL(sv)) { |
| 1459 | MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8); |
| 1460 | if (mg) |
| 1461 | mg->mg_len = 0; |
| 1462 | } |
| 1463 | } |
| 1464 | data->last_end = -1; |
| 1465 | data->flags &= ~SF_BEFORE_EOL; |
| 1466 | DEBUG_STUDYDATA("commit", data, 0, is_inf); |
| 1467 | } |
| 1468 | |
| 1469 | /* An SSC is just a regnode_charclass_posix with an extra field: the inversion |
| 1470 | * list that describes which code points it matches */ |
| 1471 | |
| 1472 | STATIC void |
| 1473 | S_ssc_anything(pTHX_ regnode_ssc *ssc) |
| 1474 | { |
| 1475 | /* Set the SSC 'ssc' to match an empty string or any code point */ |
| 1476 | |
| 1477 | PERL_ARGS_ASSERT_SSC_ANYTHING; |
| 1478 | |
| 1479 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 1480 | |
| 1481 | /* mortalize so won't leak */ |
| 1482 | ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX)); |
| 1483 | ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */ |
| 1484 | } |
| 1485 | |
| 1486 | STATIC int |
| 1487 | S_ssc_is_anything(const regnode_ssc *ssc) |
| 1488 | { |
| 1489 | /* Returns TRUE if the SSC 'ssc' can match the empty string and any code |
| 1490 | * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys |
| 1491 | * us anything: if the function returns TRUE, 'ssc' hasn't been restricted |
| 1492 | * in any way, so there's no point in using it */ |
| 1493 | |
| 1494 | UV start, end; |
| 1495 | bool ret; |
| 1496 | |
| 1497 | PERL_ARGS_ASSERT_SSC_IS_ANYTHING; |
| 1498 | |
| 1499 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 1500 | |
| 1501 | if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) { |
| 1502 | return FALSE; |
| 1503 | } |
| 1504 | |
| 1505 | /* See if the list consists solely of the range 0 - Infinity */ |
| 1506 | invlist_iterinit(ssc->invlist); |
| 1507 | ret = invlist_iternext(ssc->invlist, &start, &end) |
| 1508 | && start == 0 |
| 1509 | && end == UV_MAX; |
| 1510 | |
| 1511 | invlist_iterfinish(ssc->invlist); |
| 1512 | |
| 1513 | if (ret) { |
| 1514 | return TRUE; |
| 1515 | } |
| 1516 | |
| 1517 | /* If e.g., both \w and \W are set, matches everything */ |
| 1518 | if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) { |
| 1519 | int i; |
| 1520 | for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) { |
| 1521 | if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) { |
| 1522 | return TRUE; |
| 1523 | } |
| 1524 | } |
| 1525 | } |
| 1526 | |
| 1527 | return FALSE; |
| 1528 | } |
| 1529 | |
| 1530 | STATIC void |
| 1531 | S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc) |
| 1532 | { |
| 1533 | /* Initializes the SSC 'ssc'. This includes setting it to match an empty |
| 1534 | * string, any code point, or any posix class under locale */ |
| 1535 | |
| 1536 | PERL_ARGS_ASSERT_SSC_INIT; |
| 1537 | |
| 1538 | Zero(ssc, 1, regnode_ssc); |
| 1539 | set_ANYOF_SYNTHETIC(ssc); |
| 1540 | ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP); |
| 1541 | ssc_anything(ssc); |
| 1542 | |
| 1543 | /* If any portion of the regex is to operate under locale rules that aren't |
| 1544 | * fully known at compile time, initialization includes it. The reason |
| 1545 | * this isn't done for all regexes is that the optimizer was written under |
| 1546 | * the assumption that locale was all-or-nothing. Given the complexity and |
| 1547 | * lack of documentation in the optimizer, and that there are inadequate |
| 1548 | * test cases for locale, many parts of it may not work properly, it is |
| 1549 | * safest to avoid locale unless necessary. */ |
| 1550 | if (RExC_contains_locale) { |
| 1551 | ANYOF_POSIXL_SETALL(ssc); |
| 1552 | } |
| 1553 | else { |
| 1554 | ANYOF_POSIXL_ZERO(ssc); |
| 1555 | } |
| 1556 | } |
| 1557 | |
| 1558 | STATIC int |
| 1559 | S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state, |
| 1560 | const regnode_ssc *ssc) |
| 1561 | { |
| 1562 | /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only |
| 1563 | * to the list of code points matched, and locale posix classes; hence does |
| 1564 | * not check its flags) */ |
| 1565 | |
| 1566 | UV start, end; |
| 1567 | bool ret; |
| 1568 | |
| 1569 | PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT; |
| 1570 | |
| 1571 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 1572 | |
| 1573 | invlist_iterinit(ssc->invlist); |
| 1574 | ret = invlist_iternext(ssc->invlist, &start, &end) |
| 1575 | && start == 0 |
| 1576 | && end == UV_MAX; |
| 1577 | |
| 1578 | invlist_iterfinish(ssc->invlist); |
| 1579 | |
| 1580 | if (! ret) { |
| 1581 | return FALSE; |
| 1582 | } |
| 1583 | |
| 1584 | if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) { |
| 1585 | return FALSE; |
| 1586 | } |
| 1587 | |
| 1588 | return TRUE; |
| 1589 | } |
| 1590 | |
| 1591 | #define INVLIST_INDEX 0 |
| 1592 | #define ONLY_LOCALE_MATCHES_INDEX 1 |
| 1593 | #define DEFERRED_USER_DEFINED_INDEX 2 |
| 1594 | |
| 1595 | STATIC SV* |
| 1596 | S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state, |
| 1597 | const regnode_charclass* const node) |
| 1598 | { |
| 1599 | /* Returns a mortal inversion list defining which code points are matched |
| 1600 | * by 'node', which is of type ANYOF. Handles complementing the result if |
| 1601 | * appropriate. If some code points aren't knowable at this time, the |
| 1602 | * returned list must, and will, contain every code point that is a |
| 1603 | * possibility. */ |
| 1604 | |
| 1605 | dVAR; |
| 1606 | SV* invlist = NULL; |
| 1607 | SV* only_utf8_locale_invlist = NULL; |
| 1608 | unsigned int i; |
| 1609 | const U32 n = ARG(node); |
| 1610 | bool new_node_has_latin1 = FALSE; |
| 1611 | const U8 flags = (inRANGE(OP(node), ANYOFH, ANYOFRb)) |
| 1612 | ? 0 |
| 1613 | : ANYOF_FLAGS(node); |
| 1614 | |
| 1615 | PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC; |
| 1616 | |
| 1617 | /* Look at the data structure created by S_set_ANYOF_arg() */ |
| 1618 | if (n != ANYOF_ONLY_HAS_BITMAP) { |
| 1619 | SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]); |
| 1620 | AV * const av = MUTABLE_AV(SvRV(rv)); |
| 1621 | SV **const ary = AvARRAY(av); |
| 1622 | assert(RExC_rxi->data->what[n] == 's'); |
| 1623 | |
| 1624 | if (av_tindex_skip_len_mg(av) >= DEFERRED_USER_DEFINED_INDEX) { |
| 1625 | |
| 1626 | /* Here there are things that won't be known until runtime -- we |
| 1627 | * have to assume it could be anything */ |
| 1628 | invlist = sv_2mortal(_new_invlist(1)); |
| 1629 | return _add_range_to_invlist(invlist, 0, UV_MAX); |
| 1630 | } |
| 1631 | else if (ary[INVLIST_INDEX]) { |
| 1632 | |
| 1633 | /* Use the node's inversion list */ |
| 1634 | invlist = sv_2mortal(invlist_clone(ary[INVLIST_INDEX], NULL)); |
| 1635 | } |
| 1636 | |
| 1637 | /* Get the code points valid only under UTF-8 locales */ |
| 1638 | if ( (flags & ANYOFL_FOLD) |
| 1639 | && av_tindex_skip_len_mg(av) >= ONLY_LOCALE_MATCHES_INDEX) |
| 1640 | { |
| 1641 | only_utf8_locale_invlist = ary[ONLY_LOCALE_MATCHES_INDEX]; |
| 1642 | } |
| 1643 | } |
| 1644 | |
| 1645 | if (! invlist) { |
| 1646 | invlist = sv_2mortal(_new_invlist(0)); |
| 1647 | } |
| 1648 | |
| 1649 | /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS |
| 1650 | * code points, and an inversion list for the others, but if there are code |
| 1651 | * points that should match only conditionally on the target string being |
| 1652 | * UTF-8, those are placed in the inversion list, and not the bitmap. |
| 1653 | * Since there are circumstances under which they could match, they are |
| 1654 | * included in the SSC. But if the ANYOF node is to be inverted, we have |
| 1655 | * to exclude them here, so that when we invert below, the end result |
| 1656 | * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We |
| 1657 | * have to do this here before we add the unconditionally matched code |
| 1658 | * points */ |
| 1659 | if (flags & ANYOF_INVERT) { |
| 1660 | _invlist_intersection_complement_2nd(invlist, |
| 1661 | PL_UpperLatin1, |
| 1662 | &invlist); |
| 1663 | } |
| 1664 | |
| 1665 | /* Add in the points from the bit map */ |
| 1666 | if (! inRANGE(OP(node), ANYOFH, ANYOFRb)) { |
| 1667 | for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) { |
| 1668 | if (ANYOF_BITMAP_TEST(node, i)) { |
| 1669 | unsigned int start = i++; |
| 1670 | |
| 1671 | for (; i < NUM_ANYOF_CODE_POINTS |
| 1672 | && ANYOF_BITMAP_TEST(node, i); ++i) |
| 1673 | { |
| 1674 | /* empty */ |
| 1675 | } |
| 1676 | invlist = _add_range_to_invlist(invlist, start, i-1); |
| 1677 | new_node_has_latin1 = TRUE; |
| 1678 | } |
| 1679 | } |
| 1680 | } |
| 1681 | |
| 1682 | /* If this can match all upper Latin1 code points, have to add them |
| 1683 | * as well. But don't add them if inverting, as when that gets done below, |
| 1684 | * it would exclude all these characters, including the ones it shouldn't |
| 1685 | * that were added just above */ |
| 1686 | if (! (flags & ANYOF_INVERT) && OP(node) == ANYOFD |
| 1687 | && (flags & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)) |
| 1688 | { |
| 1689 | _invlist_union(invlist, PL_UpperLatin1, &invlist); |
| 1690 | } |
| 1691 | |
| 1692 | /* Similarly for these */ |
| 1693 | if (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP) { |
| 1694 | _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist); |
| 1695 | } |
| 1696 | |
| 1697 | if (flags & ANYOF_INVERT) { |
| 1698 | _invlist_invert(invlist); |
| 1699 | } |
| 1700 | else if (flags & ANYOFL_FOLD) { |
| 1701 | if (new_node_has_latin1) { |
| 1702 | |
| 1703 | /* Under /li, any 0-255 could fold to any other 0-255, depending on |
| 1704 | * the locale. We can skip this if there are no 0-255 at all. */ |
| 1705 | _invlist_union(invlist, PL_Latin1, &invlist); |
| 1706 | |
| 1707 | invlist = add_cp_to_invlist(invlist, LATIN_SMALL_LETTER_DOTLESS_I); |
| 1708 | invlist = add_cp_to_invlist(invlist, LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE); |
| 1709 | } |
| 1710 | else { |
| 1711 | if (_invlist_contains_cp(invlist, LATIN_SMALL_LETTER_DOTLESS_I)) { |
| 1712 | invlist = add_cp_to_invlist(invlist, 'I'); |
| 1713 | } |
| 1714 | if (_invlist_contains_cp(invlist, |
| 1715 | LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE)) |
| 1716 | { |
| 1717 | invlist = add_cp_to_invlist(invlist, 'i'); |
| 1718 | } |
| 1719 | } |
| 1720 | } |
| 1721 | |
| 1722 | /* Similarly add the UTF-8 locale possible matches. These have to be |
| 1723 | * deferred until after the non-UTF-8 locale ones are taken care of just |
| 1724 | * above, or it leads to wrong results under ANYOF_INVERT */ |
| 1725 | if (only_utf8_locale_invlist) { |
| 1726 | _invlist_union_maybe_complement_2nd(invlist, |
| 1727 | only_utf8_locale_invlist, |
| 1728 | flags & ANYOF_INVERT, |
| 1729 | &invlist); |
| 1730 | } |
| 1731 | |
| 1732 | return invlist; |
| 1733 | } |
| 1734 | |
| 1735 | /* These two functions currently do the exact same thing */ |
| 1736 | #define ssc_init_zero ssc_init |
| 1737 | |
| 1738 | #define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp)) |
| 1739 | #define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX) |
| 1740 | |
| 1741 | /* 'AND' a given class with another one. Can create false positives. 'ssc' |
| 1742 | * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be |
| 1743 | * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */ |
| 1744 | |
| 1745 | STATIC void |
| 1746 | S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc, |
| 1747 | const regnode_charclass *and_with) |
| 1748 | { |
| 1749 | /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either |
| 1750 | * another SSC or a regular ANYOF class. Can create false positives. */ |
| 1751 | |
| 1752 | SV* anded_cp_list; |
| 1753 | U8 and_with_flags = inRANGE(OP(and_with), ANYOFH, ANYOFRb) |
| 1754 | ? 0 |
| 1755 | : ANYOF_FLAGS(and_with); |
| 1756 | U8 anded_flags; |
| 1757 | |
| 1758 | PERL_ARGS_ASSERT_SSC_AND; |
| 1759 | |
| 1760 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 1761 | |
| 1762 | /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract |
| 1763 | * the code point inversion list and just the relevant flags */ |
| 1764 | if (is_ANYOF_SYNTHETIC(and_with)) { |
| 1765 | anded_cp_list = ((regnode_ssc *)and_with)->invlist; |
| 1766 | anded_flags = and_with_flags; |
| 1767 | |
| 1768 | /* XXX This is a kludge around what appears to be deficiencies in the |
| 1769 | * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag, |
| 1770 | * there are paths through the optimizer where it doesn't get weeded |
| 1771 | * out when it should. And if we don't make some extra provision for |
| 1772 | * it like the code just below, it doesn't get added when it should. |
| 1773 | * This solution is to add it only when AND'ing, which is here, and |
| 1774 | * only when what is being AND'ed is the pristine, original node |
| 1775 | * matching anything. Thus it is like adding it to ssc_anything() but |
| 1776 | * only when the result is to be AND'ed. Probably the same solution |
| 1777 | * could be adopted for the same problem we have with /l matching, |
| 1778 | * which is solved differently in S_ssc_init(), and that would lead to |
| 1779 | * fewer false positives than that solution has. But if this solution |
| 1780 | * creates bugs, the consequences are only that a warning isn't raised |
| 1781 | * that should be; while the consequences for having /l bugs is |
| 1782 | * incorrect matches */ |
| 1783 | if (ssc_is_anything((regnode_ssc *)and_with)) { |
| 1784 | anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER; |
| 1785 | } |
| 1786 | } |
| 1787 | else { |
| 1788 | anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with); |
| 1789 | if (OP(and_with) == ANYOFD) { |
| 1790 | anded_flags = and_with_flags & ANYOF_COMMON_FLAGS; |
| 1791 | } |
| 1792 | else { |
| 1793 | anded_flags = and_with_flags |
| 1794 | &( ANYOF_COMMON_FLAGS |
| 1795 | |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER |
| 1796 | |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP); |
| 1797 | if (ANYOFL_UTF8_LOCALE_REQD(and_with_flags)) { |
| 1798 | anded_flags &= |
| 1799 | ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD; |
| 1800 | } |
| 1801 | } |
| 1802 | } |
| 1803 | |
| 1804 | ANYOF_FLAGS(ssc) &= anded_flags; |
| 1805 | |
| 1806 | /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes. |
| 1807 | * C2 is the list of code points in 'and-with'; P2, its posix classes. |
| 1808 | * 'and_with' may be inverted. When not inverted, we have the situation of |
| 1809 | * computing: |
| 1810 | * (C1 | P1) & (C2 | P2) |
| 1811 | * = (C1 & (C2 | P2)) | (P1 & (C2 | P2)) |
| 1812 | * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2)) |
| 1813 | * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2)) |
| 1814 | * <= ((C1 & C2) | P1 | P2) |
| 1815 | * Alternatively, the last few steps could be: |
| 1816 | * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2)) |
| 1817 | * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2)) |
| 1818 | * <= (C1 | C2 | (P1 & P2)) |
| 1819 | * We favor the second approach if either P1 or P2 is non-empty. This is |
| 1820 | * because these components are a barrier to doing optimizations, as what |
| 1821 | * they match cannot be known until the moment of matching as they are |
| 1822 | * dependent on the current locale, 'AND"ing them likely will reduce or |
| 1823 | * eliminate them. |
| 1824 | * But we can do better if we know that C1,P1 are in their initial state (a |
| 1825 | * frequent occurrence), each matching everything: |
| 1826 | * (<everything>) & (C2 | P2) = C2 | P2 |
| 1827 | * Similarly, if C2,P2 are in their initial state (again a frequent |
| 1828 | * occurrence), the result is a no-op |
| 1829 | * (C1 | P1) & (<everything>) = C1 | P1 |
| 1830 | * |
| 1831 | * Inverted, we have |
| 1832 | * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2) |
| 1833 | * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2)) |
| 1834 | * <= (C1 & ~C2) | (P1 & ~P2) |
| 1835 | * */ |
| 1836 | |
| 1837 | if ((and_with_flags & ANYOF_INVERT) |
| 1838 | && ! is_ANYOF_SYNTHETIC(and_with)) |
| 1839 | { |
| 1840 | unsigned int i; |
| 1841 | |
| 1842 | ssc_intersection(ssc, |
| 1843 | anded_cp_list, |
| 1844 | FALSE /* Has already been inverted */ |
| 1845 | ); |
| 1846 | |
| 1847 | /* If either P1 or P2 is empty, the intersection will be also; can skip |
| 1848 | * the loop */ |
| 1849 | if (! (and_with_flags & ANYOF_MATCHES_POSIXL)) { |
| 1850 | ANYOF_POSIXL_ZERO(ssc); |
| 1851 | } |
| 1852 | else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) { |
| 1853 | |
| 1854 | /* Note that the Posix class component P from 'and_with' actually |
| 1855 | * looks like: |
| 1856 | * P = Pa | Pb | ... | Pn |
| 1857 | * where each component is one posix class, such as in [\w\s]. |
| 1858 | * Thus |
| 1859 | * ~P = ~(Pa | Pb | ... | Pn) |
| 1860 | * = ~Pa & ~Pb & ... & ~Pn |
| 1861 | * <= ~Pa | ~Pb | ... | ~Pn |
| 1862 | * The last is something we can easily calculate, but unfortunately |
| 1863 | * is likely to have many false positives. We could do better |
| 1864 | * in some (but certainly not all) instances if two classes in |
| 1865 | * P have known relationships. For example |
| 1866 | * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print: |
| 1867 | * So |
| 1868 | * :lower: & :print: = :lower: |
| 1869 | * And similarly for classes that must be disjoint. For example, |
| 1870 | * since \s and \w can have no elements in common based on rules in |
| 1871 | * the POSIX standard, |
| 1872 | * \w & ^\S = nothing |
| 1873 | * Unfortunately, some vendor locales do not meet the Posix |
| 1874 | * standard, in particular almost everything by Microsoft. |
| 1875 | * The loop below just changes e.g., \w into \W and vice versa */ |
| 1876 | |
| 1877 | regnode_charclass_posixl temp; |
| 1878 | int add = 1; /* To calculate the index of the complement */ |
| 1879 | |
| 1880 | Zero(&temp, 1, regnode_charclass_posixl); |
| 1881 | ANYOF_POSIXL_ZERO(&temp); |
| 1882 | for (i = 0; i < ANYOF_MAX; i++) { |
| 1883 | assert(i % 2 != 0 |
| 1884 | || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i) |
| 1885 | || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1)); |
| 1886 | |
| 1887 | if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) { |
| 1888 | ANYOF_POSIXL_SET(&temp, i + add); |
| 1889 | } |
| 1890 | add = 0 - add; /* 1 goes to -1; -1 goes to 1 */ |
| 1891 | } |
| 1892 | ANYOF_POSIXL_AND(&temp, ssc); |
| 1893 | |
| 1894 | } /* else ssc already has no posixes */ |
| 1895 | } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC |
| 1896 | in its initial state */ |
| 1897 | else if (! is_ANYOF_SYNTHETIC(and_with) |
| 1898 | || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with)) |
| 1899 | { |
| 1900 | /* But if 'ssc' is in its initial state, the result is just 'and_with'; |
| 1901 | * copy it over 'ssc' */ |
| 1902 | if (ssc_is_cp_posixl_init(pRExC_state, ssc)) { |
| 1903 | if (is_ANYOF_SYNTHETIC(and_with)) { |
| 1904 | StructCopy(and_with, ssc, regnode_ssc); |
| 1905 | } |
| 1906 | else { |
| 1907 | ssc->invlist = anded_cp_list; |
| 1908 | ANYOF_POSIXL_ZERO(ssc); |
| 1909 | if (and_with_flags & ANYOF_MATCHES_POSIXL) { |
| 1910 | ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc); |
| 1911 | } |
| 1912 | } |
| 1913 | } |
| 1914 | else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc) |
| 1915 | || (and_with_flags & ANYOF_MATCHES_POSIXL)) |
| 1916 | { |
| 1917 | /* One or the other of P1, P2 is non-empty. */ |
| 1918 | if (and_with_flags & ANYOF_MATCHES_POSIXL) { |
| 1919 | ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc); |
| 1920 | } |
| 1921 | ssc_union(ssc, anded_cp_list, FALSE); |
| 1922 | } |
| 1923 | else { /* P1 = P2 = empty */ |
| 1924 | ssc_intersection(ssc, anded_cp_list, FALSE); |
| 1925 | } |
| 1926 | } |
| 1927 | } |
| 1928 | |
| 1929 | STATIC void |
| 1930 | S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc, |
| 1931 | const regnode_charclass *or_with) |
| 1932 | { |
| 1933 | /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either |
| 1934 | * another SSC or a regular ANYOF class. Can create false positives if |
| 1935 | * 'or_with' is to be inverted. */ |
| 1936 | |
| 1937 | SV* ored_cp_list; |
| 1938 | U8 ored_flags; |
| 1939 | U8 or_with_flags = inRANGE(OP(or_with), ANYOFH, ANYOFRb) |
| 1940 | ? 0 |
| 1941 | : ANYOF_FLAGS(or_with); |
| 1942 | |
| 1943 | PERL_ARGS_ASSERT_SSC_OR; |
| 1944 | |
| 1945 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 1946 | |
| 1947 | /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract |
| 1948 | * the code point inversion list and just the relevant flags */ |
| 1949 | if (is_ANYOF_SYNTHETIC(or_with)) { |
| 1950 | ored_cp_list = ((regnode_ssc*) or_with)->invlist; |
| 1951 | ored_flags = or_with_flags; |
| 1952 | } |
| 1953 | else { |
| 1954 | ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with); |
| 1955 | ored_flags = or_with_flags & ANYOF_COMMON_FLAGS; |
| 1956 | if (OP(or_with) != ANYOFD) { |
| 1957 | ored_flags |
| 1958 | |= or_with_flags |
| 1959 | & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER |
| 1960 | |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP); |
| 1961 | if (ANYOFL_UTF8_LOCALE_REQD(or_with_flags)) { |
| 1962 | ored_flags |= |
| 1963 | ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD; |
| 1964 | } |
| 1965 | } |
| 1966 | } |
| 1967 | |
| 1968 | ANYOF_FLAGS(ssc) |= ored_flags; |
| 1969 | |
| 1970 | /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes. |
| 1971 | * C2 is the list of code points in 'or-with'; P2, its posix classes. |
| 1972 | * 'or_with' may be inverted. When not inverted, we have the simple |
| 1973 | * situation of computing: |
| 1974 | * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2) |
| 1975 | * If P1|P2 yields a situation with both a class and its complement are |
| 1976 | * set, like having both \w and \W, this matches all code points, and we |
| 1977 | * can delete these from the P component of the ssc going forward. XXX We |
| 1978 | * might be able to delete all the P components, but I (khw) am not certain |
| 1979 | * about this, and it is better to be safe. |
| 1980 | * |
| 1981 | * Inverted, we have |
| 1982 | * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2) |
| 1983 | * <= (C1 | P1) | ~C2 |
| 1984 | * <= (C1 | ~C2) | P1 |
| 1985 | * (which results in actually simpler code than the non-inverted case) |
| 1986 | * */ |
| 1987 | |
| 1988 | if ((or_with_flags & ANYOF_INVERT) |
| 1989 | && ! is_ANYOF_SYNTHETIC(or_with)) |
| 1990 | { |
| 1991 | /* We ignore P2, leaving P1 going forward */ |
| 1992 | } /* else Not inverted */ |
| 1993 | else if (or_with_flags & ANYOF_MATCHES_POSIXL) { |
| 1994 | ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc); |
| 1995 | if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) { |
| 1996 | unsigned int i; |
| 1997 | for (i = 0; i < ANYOF_MAX; i += 2) { |
| 1998 | if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1)) |
| 1999 | { |
| 2000 | ssc_match_all_cp(ssc); |
| 2001 | ANYOF_POSIXL_CLEAR(ssc, i); |
| 2002 | ANYOF_POSIXL_CLEAR(ssc, i+1); |
| 2003 | } |
| 2004 | } |
| 2005 | } |
| 2006 | } |
| 2007 | |
| 2008 | ssc_union(ssc, |
| 2009 | ored_cp_list, |
| 2010 | FALSE /* Already has been inverted */ |
| 2011 | ); |
| 2012 | } |
| 2013 | |
| 2014 | PERL_STATIC_INLINE void |
| 2015 | S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd) |
| 2016 | { |
| 2017 | PERL_ARGS_ASSERT_SSC_UNION; |
| 2018 | |
| 2019 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 2020 | |
| 2021 | _invlist_union_maybe_complement_2nd(ssc->invlist, |
| 2022 | invlist, |
| 2023 | invert2nd, |
| 2024 | &ssc->invlist); |
| 2025 | } |
| 2026 | |
| 2027 | PERL_STATIC_INLINE void |
| 2028 | S_ssc_intersection(pTHX_ regnode_ssc *ssc, |
| 2029 | SV* const invlist, |
| 2030 | const bool invert2nd) |
| 2031 | { |
| 2032 | PERL_ARGS_ASSERT_SSC_INTERSECTION; |
| 2033 | |
| 2034 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 2035 | |
| 2036 | _invlist_intersection_maybe_complement_2nd(ssc->invlist, |
| 2037 | invlist, |
| 2038 | invert2nd, |
| 2039 | &ssc->invlist); |
| 2040 | } |
| 2041 | |
| 2042 | PERL_STATIC_INLINE void |
| 2043 | S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end) |
| 2044 | { |
| 2045 | PERL_ARGS_ASSERT_SSC_ADD_RANGE; |
| 2046 | |
| 2047 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 2048 | |
| 2049 | ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end); |
| 2050 | } |
| 2051 | |
| 2052 | PERL_STATIC_INLINE void |
| 2053 | S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp) |
| 2054 | { |
| 2055 | /* AND just the single code point 'cp' into the SSC 'ssc' */ |
| 2056 | |
| 2057 | SV* cp_list = _new_invlist(2); |
| 2058 | |
| 2059 | PERL_ARGS_ASSERT_SSC_CP_AND; |
| 2060 | |
| 2061 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 2062 | |
| 2063 | cp_list = add_cp_to_invlist(cp_list, cp); |
| 2064 | ssc_intersection(ssc, cp_list, |
| 2065 | FALSE /* Not inverted */ |
| 2066 | ); |
| 2067 | SvREFCNT_dec_NN(cp_list); |
| 2068 | } |
| 2069 | |
| 2070 | PERL_STATIC_INLINE void |
| 2071 | S_ssc_clear_locale(regnode_ssc *ssc) |
| 2072 | { |
| 2073 | /* Set the SSC 'ssc' to not match any locale things */ |
| 2074 | PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE; |
| 2075 | |
| 2076 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 2077 | |
| 2078 | ANYOF_POSIXL_ZERO(ssc); |
| 2079 | ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS; |
| 2080 | } |
| 2081 | |
| 2082 | #define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C |
| 2083 | |
| 2084 | STATIC bool |
| 2085 | S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc) |
| 2086 | { |
| 2087 | /* The synthetic start class is used to hopefully quickly winnow down |
| 2088 | * places where a pattern could start a match in the target string. If it |
| 2089 | * doesn't really narrow things down that much, there isn't much point to |
| 2090 | * having the overhead of using it. This function uses some very crude |
| 2091 | * heuristics to decide if to use the ssc or not. |
| 2092 | * |
| 2093 | * It returns TRUE if 'ssc' rules out more than half what it considers to |
| 2094 | * be the "likely" possible matches, but of course it doesn't know what the |
| 2095 | * actual things being matched are going to be; these are only guesses |
| 2096 | * |
| 2097 | * For /l matches, it assumes that the only likely matches are going to be |
| 2098 | * in the 0-255 range, uniformly distributed, so half of that is 127 |
| 2099 | * For /a and /d matches, it assumes that the likely matches will be just |
| 2100 | * the ASCII range, so half of that is 63 |
| 2101 | * For /u and there isn't anything matching above the Latin1 range, it |
| 2102 | * assumes that that is the only range likely to be matched, and uses |
| 2103 | * half that as the cut-off: 127. If anything matches above Latin1, |
| 2104 | * it assumes that all of Unicode could match (uniformly), except for |
| 2105 | * non-Unicode code points and things in the General Category "Other" |
| 2106 | * (unassigned, private use, surrogates, controls and formats). This |
| 2107 | * is a much large number. */ |
| 2108 | |
| 2109 | U32 count = 0; /* Running total of number of code points matched by |
| 2110 | 'ssc' */ |
| 2111 | UV start, end; /* Start and end points of current range in inversion |
| 2112 | XXX outdated. UTF-8 locales are common, what about invert? list */ |
| 2113 | const U32 max_code_points = (LOC) |
| 2114 | ? 256 |
| 2115 | : (( ! UNI_SEMANTICS |
| 2116 | || invlist_highest(ssc->invlist) < 256) |
| 2117 | ? 128 |
| 2118 | : NON_OTHER_COUNT); |
| 2119 | const U32 max_match = max_code_points / 2; |
| 2120 | |
| 2121 | PERL_ARGS_ASSERT_IS_SSC_WORTH_IT; |
| 2122 | |
| 2123 | invlist_iterinit(ssc->invlist); |
| 2124 | while (invlist_iternext(ssc->invlist, &start, &end)) { |
| 2125 | if (start >= max_code_points) { |
| 2126 | break; |
| 2127 | } |
| 2128 | end = MIN(end, max_code_points - 1); |
| 2129 | count += end - start + 1; |
| 2130 | if (count >= max_match) { |
| 2131 | invlist_iterfinish(ssc->invlist); |
| 2132 | return FALSE; |
| 2133 | } |
| 2134 | } |
| 2135 | |
| 2136 | return TRUE; |
| 2137 | } |
| 2138 | |
| 2139 | |
| 2140 | STATIC void |
| 2141 | S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc) |
| 2142 | { |
| 2143 | /* The inversion list in the SSC is marked mortal; now we need a more |
| 2144 | * permanent copy, which is stored the same way that is done in a regular |
| 2145 | * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit |
| 2146 | * map */ |
| 2147 | |
| 2148 | SV* invlist = invlist_clone(ssc->invlist, NULL); |
| 2149 | |
| 2150 | PERL_ARGS_ASSERT_SSC_FINALIZE; |
| 2151 | |
| 2152 | assert(is_ANYOF_SYNTHETIC(ssc)); |
| 2153 | |
| 2154 | /* The code in this file assumes that all but these flags aren't relevant |
| 2155 | * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared |
| 2156 | * by the time we reach here */ |
| 2157 | assert(! (ANYOF_FLAGS(ssc) |
| 2158 | & ~( ANYOF_COMMON_FLAGS |
| 2159 | |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER |
| 2160 | |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP))); |
| 2161 | |
| 2162 | populate_ANYOF_from_invlist( (regnode *) ssc, &invlist); |
| 2163 | |
| 2164 | set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist, NULL, NULL); |
| 2165 | SvREFCNT_dec(invlist); |
| 2166 | |
| 2167 | /* Make sure is clone-safe */ |
| 2168 | ssc->invlist = NULL; |
| 2169 | |
| 2170 | if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) { |
| 2171 | ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL; |
| 2172 | OP(ssc) = ANYOFPOSIXL; |
| 2173 | } |
| 2174 | else if (RExC_contains_locale) { |
| 2175 | OP(ssc) = ANYOFL; |
| 2176 | } |
| 2177 | |
| 2178 | assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale); |
| 2179 | } |
| 2180 | |
| 2181 | #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ] |
| 2182 | #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid ) |
| 2183 | #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate ) |
| 2184 | #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \ |
| 2185 | ? (TRIE_LIST_CUR( idx ) - 1) \ |
| 2186 | : 0 ) |
| 2187 | |
| 2188 | |
| 2189 | #ifdef DEBUGGING |
| 2190 | /* |
| 2191 | dump_trie(trie,widecharmap,revcharmap) |
| 2192 | dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc) |
| 2193 | dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc) |
| 2194 | |
| 2195 | These routines dump out a trie in a somewhat readable format. |
| 2196 | The _interim_ variants are used for debugging the interim |
| 2197 | tables that are used to generate the final compressed |
| 2198 | representation which is what dump_trie expects. |
| 2199 | |
| 2200 | Part of the reason for their existence is to provide a form |
| 2201 | of documentation as to how the different representations function. |
| 2202 | |
| 2203 | */ |
| 2204 | |
| 2205 | /* |
| 2206 | Dumps the final compressed table form of the trie to Perl_debug_log. |
| 2207 | Used for debugging make_trie(). |
| 2208 | */ |
| 2209 | |
| 2210 | STATIC void |
| 2211 | S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap, |
| 2212 | AV *revcharmap, U32 depth) |
| 2213 | { |
| 2214 | U32 state; |
| 2215 | SV *sv=sv_newmortal(); |
| 2216 | int colwidth= widecharmap ? 6 : 4; |
| 2217 | U16 word; |
| 2218 | GET_RE_DEBUG_FLAGS_DECL; |
| 2219 | |
| 2220 | PERL_ARGS_ASSERT_DUMP_TRIE; |
| 2221 | |
| 2222 | Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ", |
| 2223 | depth+1, "Match","Base","Ofs" ); |
| 2224 | |
| 2225 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) { |
| 2226 | SV ** const tmp = av_fetch( revcharmap, state, 0); |
| 2227 | if ( tmp ) { |
| 2228 | Perl_re_printf( aTHX_ "%*s", |
| 2229 | colwidth, |
| 2230 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
| 2231 | PL_colors[0], PL_colors[1], |
| 2232 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | |
| 2233 | PERL_PV_ESCAPE_FIRSTCHAR |
| 2234 | ) |
| 2235 | ); |
| 2236 | } |
| 2237 | } |
| 2238 | Perl_re_printf( aTHX_ "\n"); |
| 2239 | Perl_re_indentf( aTHX_ "State|-----------------------", depth+1); |
| 2240 | |
| 2241 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) |
| 2242 | Perl_re_printf( aTHX_ "%.*s", colwidth, "--------"); |
| 2243 | Perl_re_printf( aTHX_ "\n"); |
| 2244 | |
| 2245 | for( state = 1 ; state < trie->statecount ; state++ ) { |
| 2246 | const U32 base = trie->states[ state ].trans.base; |
| 2247 | |
| 2248 | Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state); |
| 2249 | |
| 2250 | if ( trie->states[ state ].wordnum ) { |
| 2251 | Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum ); |
| 2252 | } else { |
| 2253 | Perl_re_printf( aTHX_ "%6s", "" ); |
| 2254 | } |
| 2255 | |
| 2256 | Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base ); |
| 2257 | |
| 2258 | if ( base ) { |
| 2259 | U32 ofs = 0; |
| 2260 | |
| 2261 | while( ( base + ofs < trie->uniquecharcount ) || |
| 2262 | ( base + ofs - trie->uniquecharcount < trie->lasttrans |
| 2263 | && trie->trans[ base + ofs - trie->uniquecharcount ].check |
| 2264 | != state)) |
| 2265 | ofs++; |
| 2266 | |
| 2267 | Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs); |
| 2268 | |
| 2269 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { |
| 2270 | if ( ( base + ofs >= trie->uniquecharcount ) |
| 2271 | && ( base + ofs - trie->uniquecharcount |
| 2272 | < trie->lasttrans ) |
| 2273 | && trie->trans[ base + ofs |
| 2274 | - trie->uniquecharcount ].check == state ) |
| 2275 | { |
| 2276 | Perl_re_printf( aTHX_ "%*" UVXf, colwidth, |
| 2277 | (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next |
| 2278 | ); |
| 2279 | } else { |
| 2280 | Perl_re_printf( aTHX_ "%*s", colwidth," ." ); |
| 2281 | } |
| 2282 | } |
| 2283 | |
| 2284 | Perl_re_printf( aTHX_ "]"); |
| 2285 | |
| 2286 | } |
| 2287 | Perl_re_printf( aTHX_ "\n" ); |
| 2288 | } |
| 2289 | Perl_re_indentf( aTHX_ "word_info N:(prev,len)=", |
| 2290 | depth); |
| 2291 | for (word=1; word <= trie->wordcount; word++) { |
| 2292 | Perl_re_printf( aTHX_ " %d:(%d,%d)", |
| 2293 | (int)word, (int)(trie->wordinfo[word].prev), |
| 2294 | (int)(trie->wordinfo[word].len)); |
| 2295 | } |
| 2296 | Perl_re_printf( aTHX_ "\n" ); |
| 2297 | } |
| 2298 | /* |
| 2299 | Dumps a fully constructed but uncompressed trie in list form. |
| 2300 | List tries normally only are used for construction when the number of |
| 2301 | possible chars (trie->uniquecharcount) is very high. |
| 2302 | Used for debugging make_trie(). |
| 2303 | */ |
| 2304 | STATIC void |
| 2305 | S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie, |
| 2306 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
| 2307 | U32 depth) |
| 2308 | { |
| 2309 | U32 state; |
| 2310 | SV *sv=sv_newmortal(); |
| 2311 | int colwidth= widecharmap ? 6 : 4; |
| 2312 | GET_RE_DEBUG_FLAGS_DECL; |
| 2313 | |
| 2314 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST; |
| 2315 | |
| 2316 | /* print out the table precompression. */ |
| 2317 | Perl_re_indentf( aTHX_ "State :Word | Transition Data\n", |
| 2318 | depth+1 ); |
| 2319 | Perl_re_indentf( aTHX_ "%s", |
| 2320 | depth+1, "------:-----+-----------------\n" ); |
| 2321 | |
| 2322 | for( state=1 ; state < next_alloc ; state ++ ) { |
| 2323 | U16 charid; |
| 2324 | |
| 2325 | Perl_re_indentf( aTHX_ " %4" UVXf " :", |
| 2326 | depth+1, (UV)state ); |
| 2327 | if ( ! trie->states[ state ].wordnum ) { |
| 2328 | Perl_re_printf( aTHX_ "%5s| ",""); |
| 2329 | } else { |
| 2330 | Perl_re_printf( aTHX_ "W%4x| ", |
| 2331 | trie->states[ state ].wordnum |
| 2332 | ); |
| 2333 | } |
| 2334 | for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) { |
| 2335 | SV ** const tmp = av_fetch( revcharmap, |
| 2336 | TRIE_LIST_ITEM(state, charid).forid, 0); |
| 2337 | if ( tmp ) { |
| 2338 | Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ", |
| 2339 | colwidth, |
| 2340 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), |
| 2341 | colwidth, |
| 2342 | PL_colors[0], PL_colors[1], |
| 2343 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
| 2344 | | PERL_PV_ESCAPE_FIRSTCHAR |
| 2345 | ) , |
| 2346 | TRIE_LIST_ITEM(state, charid).forid, |
| 2347 | (UV)TRIE_LIST_ITEM(state, charid).newstate |
| 2348 | ); |
| 2349 | if (!(charid % 10)) |
| 2350 | Perl_re_printf( aTHX_ "\n%*s| ", |
| 2351 | (int)((depth * 2) + 14), ""); |
| 2352 | } |
| 2353 | } |
| 2354 | Perl_re_printf( aTHX_ "\n"); |
| 2355 | } |
| 2356 | } |
| 2357 | |
| 2358 | /* |
| 2359 | Dumps a fully constructed but uncompressed trie in table form. |
| 2360 | This is the normal DFA style state transition table, with a few |
| 2361 | twists to facilitate compression later. |
| 2362 | Used for debugging make_trie(). |
| 2363 | */ |
| 2364 | STATIC void |
| 2365 | S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie, |
| 2366 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
| 2367 | U32 depth) |
| 2368 | { |
| 2369 | U32 state; |
| 2370 | U16 charid; |
| 2371 | SV *sv=sv_newmortal(); |
| 2372 | int colwidth= widecharmap ? 6 : 4; |
| 2373 | GET_RE_DEBUG_FLAGS_DECL; |
| 2374 | |
| 2375 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE; |
| 2376 | |
| 2377 | /* |
| 2378 | print out the table precompression so that we can do a visual check |
| 2379 | that they are identical. |
| 2380 | */ |
| 2381 | |
| 2382 | Perl_re_indentf( aTHX_ "Char : ", depth+1 ); |
| 2383 | |
| 2384 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { |
| 2385 | SV ** const tmp = av_fetch( revcharmap, charid, 0); |
| 2386 | if ( tmp ) { |
| 2387 | Perl_re_printf( aTHX_ "%*s", |
| 2388 | colwidth, |
| 2389 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
| 2390 | PL_colors[0], PL_colors[1], |
| 2391 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | |
| 2392 | PERL_PV_ESCAPE_FIRSTCHAR |
| 2393 | ) |
| 2394 | ); |
| 2395 | } |
| 2396 | } |
| 2397 | |
| 2398 | Perl_re_printf( aTHX_ "\n"); |
| 2399 | Perl_re_indentf( aTHX_ "State+-", depth+1 ); |
| 2400 | |
| 2401 | for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) { |
| 2402 | Perl_re_printf( aTHX_ "%.*s", colwidth,"--------"); |
| 2403 | } |
| 2404 | |
| 2405 | Perl_re_printf( aTHX_ "\n" ); |
| 2406 | |
| 2407 | for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) { |
| 2408 | |
| 2409 | Perl_re_indentf( aTHX_ "%4" UVXf " : ", |
| 2410 | depth+1, |
| 2411 | (UV)TRIE_NODENUM( state ) ); |
| 2412 | |
| 2413 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { |
| 2414 | UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next ); |
| 2415 | if (v) |
| 2416 | Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v ); |
| 2417 | else |
| 2418 | Perl_re_printf( aTHX_ "%*s", colwidth, "." ); |
| 2419 | } |
| 2420 | if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) { |
| 2421 | Perl_re_printf( aTHX_ " (%4" UVXf ")\n", |
| 2422 | (UV)trie->trans[ state ].check ); |
| 2423 | } else { |
| 2424 | Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n", |
| 2425 | (UV)trie->trans[ state ].check, |
| 2426 | trie->states[ TRIE_NODENUM( state ) ].wordnum ); |
| 2427 | } |
| 2428 | } |
| 2429 | } |
| 2430 | |
| 2431 | #endif |
| 2432 | |
| 2433 | |
| 2434 | /* make_trie(startbranch,first,last,tail,word_count,flags,depth) |
| 2435 | startbranch: the first branch in the whole branch sequence |
| 2436 | first : start branch of sequence of branch-exact nodes. |
| 2437 | May be the same as startbranch |
| 2438 | last : Thing following the last branch. |
| 2439 | May be the same as tail. |
| 2440 | tail : item following the branch sequence |
| 2441 | count : words in the sequence |
| 2442 | flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/ |
| 2443 | depth : indent depth |
| 2444 | |
| 2445 | Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node. |
| 2446 | |
| 2447 | A trie is an N'ary tree where the branches are determined by digital |
| 2448 | decomposition of the key. IE, at the root node you look up the 1st character and |
| 2449 | follow that branch repeat until you find the end of the branches. Nodes can be |
| 2450 | marked as "accepting" meaning they represent a complete word. Eg: |
| 2451 | |
| 2452 | /he|she|his|hers/ |
| 2453 | |
| 2454 | would convert into the following structure. Numbers represent states, letters |
| 2455 | following numbers represent valid transitions on the letter from that state, if |
| 2456 | the number is in square brackets it represents an accepting state, otherwise it |
| 2457 | will be in parenthesis. |
| 2458 | |
| 2459 | +-h->+-e->[3]-+-r->(8)-+-s->[9] |
| 2460 | | | |
| 2461 | | (2) |
| 2462 | | | |
| 2463 | (1) +-i->(6)-+-s->[7] |
| 2464 | | |
| 2465 | +-s->(3)-+-h->(4)-+-e->[5] |
| 2466 | |
| 2467 | Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers) |
| 2468 | |
| 2469 | This shows that when matching against the string 'hers' we will begin at state 1 |
| 2470 | read 'h' and move to state 2, read 'e' and move to state 3 which is accepting, |
| 2471 | then read 'r' and go to state 8 followed by 's' which takes us to state 9 which |
| 2472 | is also accepting. Thus we know that we can match both 'he' and 'hers' with a |
| 2473 | single traverse. We store a mapping from accepting to state to which word was |
| 2474 | matched, and then when we have multiple possibilities we try to complete the |
| 2475 | rest of the regex in the order in which they occurred in the alternation. |
| 2476 | |
| 2477 | The only prior NFA like behaviour that would be changed by the TRIE support is |
| 2478 | the silent ignoring of duplicate alternations which are of the form: |
| 2479 | |
| 2480 | / (DUPE|DUPE) X? (?{ ... }) Y /x |
| 2481 | |
| 2482 | Thus EVAL blocks following a trie may be called a different number of times with |
| 2483 | and without the optimisation. With the optimisations dupes will be silently |
| 2484 | ignored. This inconsistent behaviour of EVAL type nodes is well established as |
| 2485 | the following demonstrates: |
| 2486 | |
| 2487 | 'words'=~/(word|word|word)(?{ print $1 })[xyz]/ |
| 2488 | |
| 2489 | which prints out 'word' three times, but |
| 2490 | |
| 2491 | 'words'=~/(word|word|word)(?{ print $1 })S/ |
| 2492 | |
| 2493 | which doesnt print it out at all. This is due to other optimisations kicking in. |
| 2494 | |
| 2495 | Example of what happens on a structural level: |
| 2496 | |
| 2497 | The regexp /(ac|ad|ab)+/ will produce the following debug output: |
| 2498 | |
| 2499 | 1: CURLYM[1] {1,32767}(18) |
| 2500 | 5: BRANCH(8) |
| 2501 | 6: EXACT <ac>(16) |
| 2502 | 8: BRANCH(11) |
| 2503 | 9: EXACT <ad>(16) |
| 2504 | 11: BRANCH(14) |
| 2505 | 12: EXACT <ab>(16) |
| 2506 | 16: SUCCEED(0) |
| 2507 | 17: NOTHING(18) |
| 2508 | 18: END(0) |
| 2509 | |
| 2510 | This would be optimizable with startbranch=5, first=5, last=16, tail=16 |
| 2511 | and should turn into: |
| 2512 | |
| 2513 | 1: CURLYM[1] {1,32767}(18) |
| 2514 | 5: TRIE(16) |
| 2515 | [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1] |
| 2516 | <ac> |
| 2517 | <ad> |
| 2518 | <ab> |
| 2519 | 16: SUCCEED(0) |
| 2520 | 17: NOTHING(18) |
| 2521 | 18: END(0) |
| 2522 | |
| 2523 | Cases where tail != last would be like /(?foo|bar)baz/: |
| 2524 | |
| 2525 | 1: BRANCH(4) |
| 2526 | 2: EXACT <foo>(8) |
| 2527 | 4: BRANCH(7) |
| 2528 | 5: EXACT <bar>(8) |
| 2529 | 7: TAIL(8) |
| 2530 | 8: EXACT <baz>(10) |
| 2531 | 10: END(0) |
| 2532 | |
| 2533 | which would be optimizable with startbranch=1, first=1, last=7, tail=8 |
| 2534 | and would end up looking like: |
| 2535 | |
| 2536 | 1: TRIE(8) |
| 2537 | [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1] |
| 2538 | <foo> |
| 2539 | <bar> |
| 2540 | 7: TAIL(8) |
| 2541 | 8: EXACT <baz>(10) |
| 2542 | 10: END(0) |
| 2543 | |
| 2544 | d = uvchr_to_utf8_flags(d, uv, 0); |
| 2545 | |
| 2546 | is the recommended Unicode-aware way of saying |
| 2547 | |
| 2548 | *(d++) = uv; |
| 2549 | */ |
| 2550 | |
| 2551 | #define TRIE_STORE_REVCHAR(val) \ |
| 2552 | STMT_START { \ |
| 2553 | if (UTF) { \ |
| 2554 | SV *zlopp = newSV(UTF8_MAXBYTES); \ |
| 2555 | unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \ |
| 2556 | unsigned char *const kapow = uvchr_to_utf8(flrbbbbb, val); \ |
| 2557 | *kapow = '\0'; \ |
| 2558 | SvCUR_set(zlopp, kapow - flrbbbbb); \ |
| 2559 | SvPOK_on(zlopp); \ |
| 2560 | SvUTF8_on(zlopp); \ |
| 2561 | av_push(revcharmap, zlopp); \ |
| 2562 | } else { \ |
| 2563 | char ooooff = (char)val; \ |
| 2564 | av_push(revcharmap, newSVpvn(&ooooff, 1)); \ |
| 2565 | } \ |
| 2566 | } STMT_END |
| 2567 | |
| 2568 | /* This gets the next character from the input, folding it if not already |
| 2569 | * folded. */ |
| 2570 | #define TRIE_READ_CHAR STMT_START { \ |
| 2571 | wordlen++; \ |
| 2572 | if ( UTF ) { \ |
| 2573 | /* if it is UTF then it is either already folded, or does not need \ |
| 2574 | * folding */ \ |
| 2575 | uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \ |
| 2576 | } \ |
| 2577 | else if (folder == PL_fold_latin1) { \ |
| 2578 | /* This folder implies Unicode rules, which in the range expressible \ |
| 2579 | * by not UTF is the lower case, with the two exceptions, one of \ |
| 2580 | * which should have been taken care of before calling this */ \ |
| 2581 | assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \ |
| 2582 | uvc = toLOWER_L1(*uc); \ |
| 2583 | if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \ |
| 2584 | len = 1; \ |
| 2585 | } else { \ |
| 2586 | /* raw data, will be folded later if needed */ \ |
| 2587 | uvc = (U32)*uc; \ |
| 2588 | len = 1; \ |
| 2589 | } \ |
| 2590 | } STMT_END |
| 2591 | |
| 2592 | |
| 2593 | |
| 2594 | #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \ |
| 2595 | if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \ |
| 2596 | U32 ging = TRIE_LIST_LEN( state ) * 2; \ |
| 2597 | Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \ |
| 2598 | TRIE_LIST_LEN( state ) = ging; \ |
| 2599 | } \ |
| 2600 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \ |
| 2601 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \ |
| 2602 | TRIE_LIST_CUR( state )++; \ |
| 2603 | } STMT_END |
| 2604 | |
| 2605 | #define TRIE_LIST_NEW(state) STMT_START { \ |
| 2606 | Newx( trie->states[ state ].trans.list, \ |
| 2607 | 4, reg_trie_trans_le ); \ |
| 2608 | TRIE_LIST_CUR( state ) = 1; \ |
| 2609 | TRIE_LIST_LEN( state ) = 4; \ |
| 2610 | } STMT_END |
| 2611 | |
| 2612 | #define TRIE_HANDLE_WORD(state) STMT_START { \ |
| 2613 | U16 dupe= trie->states[ state ].wordnum; \ |
| 2614 | regnode * const noper_next = regnext( noper ); \ |
| 2615 | \ |
| 2616 | DEBUG_r({ \ |
| 2617 | /* store the word for dumping */ \ |
| 2618 | SV* tmp; \ |
| 2619 | if (OP(noper) != NOTHING) \ |
| 2620 | tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \ |
| 2621 | else \ |
| 2622 | tmp = newSVpvn_utf8( "", 0, UTF ); \ |
| 2623 | av_push( trie_words, tmp ); \ |
| 2624 | }); \ |
| 2625 | \ |
| 2626 | curword++; \ |
| 2627 | trie->wordinfo[curword].prev = 0; \ |
| 2628 | trie->wordinfo[curword].len = wordlen; \ |
| 2629 | trie->wordinfo[curword].accept = state; \ |
| 2630 | \ |
| 2631 | if ( noper_next < tail ) { \ |
| 2632 | if (!trie->jump) \ |
| 2633 | trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \ |
| 2634 | sizeof(U16) ); \ |
| 2635 | trie->jump[curword] = (U16)(noper_next - convert); \ |
| 2636 | if (!jumper) \ |
| 2637 | jumper = noper_next; \ |
| 2638 | if (!nextbranch) \ |
| 2639 | nextbranch= regnext(cur); \ |
| 2640 | } \ |
| 2641 | \ |
| 2642 | if ( dupe ) { \ |
| 2643 | /* It's a dupe. Pre-insert into the wordinfo[].prev */\ |
| 2644 | /* chain, so that when the bits of chain are later */\ |
| 2645 | /* linked together, the dups appear in the chain */\ |
| 2646 | trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \ |
| 2647 | trie->wordinfo[dupe].prev = curword; \ |
| 2648 | } else { \ |
| 2649 | /* we haven't inserted this word yet. */ \ |
| 2650 | trie->states[ state ].wordnum = curword; \ |
| 2651 | } \ |
| 2652 | } STMT_END |
| 2653 | |
| 2654 | |
| 2655 | #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \ |
| 2656 | ( ( base + charid >= ucharcount \ |
| 2657 | && base + charid < ubound \ |
| 2658 | && state == trie->trans[ base - ucharcount + charid ].check \ |
| 2659 | && trie->trans[ base - ucharcount + charid ].next ) \ |
| 2660 | ? trie->trans[ base - ucharcount + charid ].next \ |
| 2661 | : ( state==1 ? special : 0 ) \ |
| 2662 | ) |
| 2663 | |
| 2664 | #define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \ |
| 2665 | STMT_START { \ |
| 2666 | TRIE_BITMAP_SET(trie, uvc); \ |
| 2667 | /* store the folded codepoint */ \ |
| 2668 | if ( folder ) \ |
| 2669 | TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \ |
| 2670 | \ |
| 2671 | if ( !UTF ) { \ |
| 2672 | /* store first byte of utf8 representation of */ \ |
| 2673 | /* variant codepoints */ \ |
| 2674 | if (! UVCHR_IS_INVARIANT(uvc)) { \ |
| 2675 | TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \ |
| 2676 | } \ |
| 2677 | } \ |
| 2678 | } STMT_END |
| 2679 | #define MADE_TRIE 1 |
| 2680 | #define MADE_JUMP_TRIE 2 |
| 2681 | #define MADE_EXACT_TRIE 4 |
| 2682 | |
| 2683 | STATIC I32 |
| 2684 | S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, |
| 2685 | regnode *first, regnode *last, regnode *tail, |
| 2686 | U32 word_count, U32 flags, U32 depth) |
| 2687 | { |
| 2688 | /* first pass, loop through and scan words */ |
| 2689 | reg_trie_data *trie; |
| 2690 | HV *widecharmap = NULL; |
| 2691 | AV *revcharmap = newAV(); |
| 2692 | regnode *cur; |
| 2693 | STRLEN len = 0; |
| 2694 | UV uvc = 0; |
| 2695 | U16 curword = 0; |
| 2696 | U32 next_alloc = 0; |
| 2697 | regnode *jumper = NULL; |
| 2698 | regnode *nextbranch = NULL; |
| 2699 | regnode *convert = NULL; |
| 2700 | U32 *prev_states; /* temp array mapping each state to previous one */ |
| 2701 | /* we just use folder as a flag in utf8 */ |
| 2702 | const U8 * folder = NULL; |
| 2703 | |
| 2704 | /* in the below add_data call we are storing either 'tu' or 'tuaa' |
| 2705 | * which stands for one trie structure, one hash, optionally followed |
| 2706 | * by two arrays */ |
| 2707 | #ifdef DEBUGGING |
| 2708 | const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa")); |
| 2709 | AV *trie_words = NULL; |
| 2710 | /* along with revcharmap, this only used during construction but both are |
| 2711 | * useful during debugging so we store them in the struct when debugging. |
| 2712 | */ |
| 2713 | #else |
| 2714 | const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu")); |
| 2715 | STRLEN trie_charcount=0; |
| 2716 | #endif |
| 2717 | SV *re_trie_maxbuff; |
| 2718 | GET_RE_DEBUG_FLAGS_DECL; |
| 2719 | |
| 2720 | PERL_ARGS_ASSERT_MAKE_TRIE; |
| 2721 | #ifndef DEBUGGING |
| 2722 | PERL_UNUSED_ARG(depth); |
| 2723 | #endif |
| 2724 | |
| 2725 | switch (flags) { |
| 2726 | case EXACT: case EXACT_REQ8: case EXACTL: break; |
| 2727 | case EXACTFAA: |
| 2728 | case EXACTFUP: |
| 2729 | case EXACTFU: |
| 2730 | case EXACTFLU8: folder = PL_fold_latin1; break; |
| 2731 | case EXACTF: folder = PL_fold; break; |
| 2732 | default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] ); |
| 2733 | } |
| 2734 | |
| 2735 | trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) ); |
| 2736 | trie->refcount = 1; |
| 2737 | trie->startstate = 1; |
| 2738 | trie->wordcount = word_count; |
| 2739 | RExC_rxi->data->data[ data_slot ] = (void*)trie; |
| 2740 | trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) ); |
| 2741 | if (flags == EXACT || flags == EXACT_REQ8 || flags == EXACTL) |
| 2742 | trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 ); |
| 2743 | trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc( |
| 2744 | trie->wordcount+1, sizeof(reg_trie_wordinfo)); |
| 2745 | |
| 2746 | DEBUG_r({ |
| 2747 | trie_words = newAV(); |
| 2748 | }); |
| 2749 | |
| 2750 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, GV_ADD); |
| 2751 | assert(re_trie_maxbuff); |
| 2752 | if (!SvIOK(re_trie_maxbuff)) { |
| 2753 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); |
| 2754 | } |
| 2755 | DEBUG_TRIE_COMPILE_r({ |
| 2756 | Perl_re_indentf( aTHX_ |
| 2757 | "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n", |
| 2758 | depth+1, |
| 2759 | REG_NODE_NUM(startbranch), REG_NODE_NUM(first), |
| 2760 | REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth); |
| 2761 | }); |
| 2762 | |
| 2763 | /* Find the node we are going to overwrite */ |
| 2764 | if ( first == startbranch && OP( last ) != BRANCH ) { |
| 2765 | /* whole branch chain */ |
| 2766 | convert = first; |
| 2767 | } else { |
| 2768 | /* branch sub-chain */ |
| 2769 | convert = NEXTOPER( first ); |
| 2770 | } |
| 2771 | |
| 2772 | /* -- First loop and Setup -- |
| 2773 | |
| 2774 | We first traverse the branches and scan each word to determine if it |
| 2775 | contains widechars, and how many unique chars there are, this is |
| 2776 | important as we have to build a table with at least as many columns as we |
| 2777 | have unique chars. |
| 2778 | |
| 2779 | We use an array of integers to represent the character codes 0..255 |
| 2780 | (trie->charmap) and we use a an HV* to store Unicode characters. We use |
| 2781 | the native representation of the character value as the key and IV's for |
| 2782 | the coded index. |
| 2783 | |
| 2784 | *TODO* If we keep track of how many times each character is used we can |
| 2785 | remap the columns so that the table compression later on is more |
| 2786 | efficient in terms of memory by ensuring the most common value is in the |
| 2787 | middle and the least common are on the outside. IMO this would be better |
| 2788 | than a most to least common mapping as theres a decent chance the most |
| 2789 | common letter will share a node with the least common, meaning the node |
| 2790 | will not be compressible. With a middle is most common approach the worst |
| 2791 | case is when we have the least common nodes twice. |
| 2792 | |
| 2793 | */ |
| 2794 | |
| 2795 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
| 2796 | regnode *noper = NEXTOPER( cur ); |
| 2797 | const U8 *uc; |
| 2798 | const U8 *e; |
| 2799 | int foldlen = 0; |
| 2800 | U32 wordlen = 0; /* required init */ |
| 2801 | STRLEN minchars = 0; |
| 2802 | STRLEN maxchars = 0; |
| 2803 | bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the |
| 2804 | bitmap?*/ |
| 2805 | |
| 2806 | if (OP(noper) == NOTHING) { |
| 2807 | /* skip past a NOTHING at the start of an alternation |
| 2808 | * eg, /(?:)a|(?:b)/ should be the same as /a|b/ |
| 2809 | * |
| 2810 | * If the next node is not something we are supposed to process |
| 2811 | * we will just ignore it due to the condition guarding the |
| 2812 | * next block. |
| 2813 | */ |
| 2814 | |
| 2815 | regnode *noper_next= regnext(noper); |
| 2816 | if (noper_next < tail) |
| 2817 | noper= noper_next; |
| 2818 | } |
| 2819 | |
| 2820 | if ( noper < tail |
| 2821 | && ( OP(noper) == flags |
| 2822 | || (flags == EXACT && OP(noper) == EXACT_REQ8) |
| 2823 | || (flags == EXACTFU && ( OP(noper) == EXACTFU_REQ8 |
| 2824 | || OP(noper) == EXACTFUP)))) |
| 2825 | { |
| 2826 | uc= (U8*)STRING(noper); |
| 2827 | e= uc + STR_LEN(noper); |
| 2828 | } else { |
| 2829 | trie->minlen= 0; |
| 2830 | continue; |
| 2831 | } |
| 2832 | |
| 2833 | |
| 2834 | if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */ |
| 2835 | TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte |
| 2836 | regardless of encoding */ |
| 2837 | if (OP( noper ) == EXACTFUP) { |
| 2838 | /* false positives are ok, so just set this */ |
| 2839 | TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S); |
| 2840 | } |
| 2841 | } |
| 2842 | |
| 2843 | for ( ; uc < e ; uc += len ) { /* Look at each char in the current |
| 2844 | branch */ |
| 2845 | TRIE_CHARCOUNT(trie)++; |
| 2846 | TRIE_READ_CHAR; |
| 2847 | |
| 2848 | /* TRIE_READ_CHAR returns the current character, or its fold if /i |
| 2849 | * is in effect. Under /i, this character can match itself, or |
| 2850 | * anything that folds to it. If not under /i, it can match just |
| 2851 | * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN |
| 2852 | * all fold to k, and all are single characters. But some folds |
| 2853 | * expand to more than one character, so for example LATIN SMALL |
| 2854 | * LIGATURE FFI folds to the three character sequence 'ffi'. If |
| 2855 | * the string beginning at 'uc' is 'ffi', it could be matched by |
| 2856 | * three characters, or just by the one ligature character. (It |
| 2857 | * could also be matched by two characters: LATIN SMALL LIGATURE FF |
| 2858 | * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI). |
| 2859 | * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also |
| 2860 | * match.) The trie needs to know the minimum and maximum number |
| 2861 | * of characters that could match so that it can use size alone to |
| 2862 | * quickly reject many match attempts. The max is simple: it is |
| 2863 | * the number of folded characters in this branch (since a fold is |
| 2864 | * never shorter than what folds to it. */ |
| 2865 | |
| 2866 | maxchars++; |
| 2867 | |
| 2868 | /* And the min is equal to the max if not under /i (indicated by |
| 2869 | * 'folder' being NULL), or there are no multi-character folds. If |
| 2870 | * there is a multi-character fold, the min is incremented just |
| 2871 | * once, for the character that folds to the sequence. Each |
| 2872 | * character in the sequence needs to be added to the list below of |
| 2873 | * characters in the trie, but we count only the first towards the |
| 2874 | * min number of characters needed. This is done through the |
| 2875 | * variable 'foldlen', which is returned by the macros that look |
| 2876 | * for these sequences as the number of bytes the sequence |
| 2877 | * occupies. Each time through the loop, we decrement 'foldlen' by |
| 2878 | * how many bytes the current char occupies. Only when it reaches |
| 2879 | * 0 do we increment 'minchars' or look for another multi-character |
| 2880 | * sequence. */ |
| 2881 | if (folder == NULL) { |
| 2882 | minchars++; |
| 2883 | } |
| 2884 | else if (foldlen > 0) { |
| 2885 | foldlen -= (UTF) ? UTF8SKIP(uc) : 1; |
| 2886 | } |
| 2887 | else { |
| 2888 | minchars++; |
| 2889 | |
| 2890 | /* See if *uc is the beginning of a multi-character fold. If |
| 2891 | * so, we decrement the length remaining to look at, to account |
| 2892 | * for the current character this iteration. (We can use 'uc' |
| 2893 | * instead of the fold returned by TRIE_READ_CHAR because for |
| 2894 | * non-UTF, the latin1_safe macro is smart enough to account |
| 2895 | * for all the unfolded characters, and because for UTF, the |
| 2896 | * string will already have been folded earlier in the |
| 2897 | * compilation process */ |
| 2898 | if (UTF) { |
| 2899 | if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) { |
| 2900 | foldlen -= UTF8SKIP(uc); |
| 2901 | } |
| 2902 | } |
| 2903 | else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) { |
| 2904 | foldlen--; |
| 2905 | } |
| 2906 | } |
| 2907 | |
| 2908 | /* The current character (and any potential folds) should be added |
| 2909 | * to the possible matching characters for this position in this |
| 2910 | * branch */ |
| 2911 | if ( uvc < 256 ) { |
| 2912 | if ( folder ) { |
| 2913 | U8 folded= folder[ (U8) uvc ]; |
| 2914 | if ( !trie->charmap[ folded ] ) { |
| 2915 | trie->charmap[ folded ]=( ++trie->uniquecharcount ); |
| 2916 | TRIE_STORE_REVCHAR( folded ); |
| 2917 | } |
| 2918 | } |
| 2919 | if ( !trie->charmap[ uvc ] ) { |
| 2920 | trie->charmap[ uvc ]=( ++trie->uniquecharcount ); |
| 2921 | TRIE_STORE_REVCHAR( uvc ); |
| 2922 | } |
| 2923 | if ( set_bit ) { |
| 2924 | /* store the codepoint in the bitmap, and its folded |
| 2925 | * equivalent. */ |
| 2926 | TRIE_BITMAP_SET_FOLDED(trie, uvc, folder); |
| 2927 | set_bit = 0; /* We've done our bit :-) */ |
| 2928 | } |
| 2929 | } else { |
| 2930 | |
| 2931 | /* XXX We could come up with the list of code points that fold |
| 2932 | * to this using PL_utf8_foldclosures, except not for |
| 2933 | * multi-char folds, as there may be multiple combinations |
| 2934 | * there that could work, which needs to wait until runtime to |
| 2935 | * resolve (The comment about LIGATURE FFI above is such an |
| 2936 | * example */ |
| 2937 | |
| 2938 | SV** svpp; |
| 2939 | if ( !widecharmap ) |
| 2940 | widecharmap = newHV(); |
| 2941 | |
| 2942 | svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 ); |
| 2943 | |
| 2944 | if ( !svpp ) |
| 2945 | Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc ); |
| 2946 | |
| 2947 | if ( !SvTRUE( *svpp ) ) { |
| 2948 | sv_setiv( *svpp, ++trie->uniquecharcount ); |
| 2949 | TRIE_STORE_REVCHAR(uvc); |
| 2950 | } |
| 2951 | } |
| 2952 | } /* end loop through characters in this branch of the trie */ |
| 2953 | |
| 2954 | /* We take the min and max for this branch and combine to find the min |
| 2955 | * and max for all branches processed so far */ |
| 2956 | if( cur == first ) { |
| 2957 | trie->minlen = minchars; |
| 2958 | trie->maxlen = maxchars; |
| 2959 | } else if (minchars < trie->minlen) { |
| 2960 | trie->minlen = minchars; |
| 2961 | } else if (maxchars > trie->maxlen) { |
| 2962 | trie->maxlen = maxchars; |
| 2963 | } |
| 2964 | } /* end first pass */ |
| 2965 | DEBUG_TRIE_COMPILE_r( |
| 2966 | Perl_re_indentf( aTHX_ |
| 2967 | "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n", |
| 2968 | depth+1, |
| 2969 | ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count, |
| 2970 | (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount, |
| 2971 | (int)trie->minlen, (int)trie->maxlen ) |
| 2972 | ); |
| 2973 | |
| 2974 | /* |
| 2975 | We now know what we are dealing with in terms of unique chars and |
| 2976 | string sizes so we can calculate how much memory a naive |
| 2977 | representation using a flat table will take. If it's over a reasonable |
| 2978 | limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory |
| 2979 | conservative but potentially much slower representation using an array |
| 2980 | of lists. |
| 2981 | |
| 2982 | At the end we convert both representations into the same compressed |
| 2983 | form that will be used in regexec.c for matching with. The latter |
| 2984 | is a form that cannot be used to construct with but has memory |
| 2985 | properties similar to the list form and access properties similar |
| 2986 | to the table form making it both suitable for fast searches and |
| 2987 | small enough that its feasable to store for the duration of a program. |
| 2988 | |
| 2989 | See the comment in the code where the compressed table is produced |
| 2990 | inplace from the flat tabe representation for an explanation of how |
| 2991 | the compression works. |
| 2992 | |
| 2993 | */ |
| 2994 | |
| 2995 | |
| 2996 | Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32); |
| 2997 | prev_states[1] = 0; |
| 2998 | |
| 2999 | if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) |
| 3000 | > SvIV(re_trie_maxbuff) ) |
| 3001 | { |
| 3002 | /* |
| 3003 | Second Pass -- Array Of Lists Representation |
| 3004 | |
| 3005 | Each state will be represented by a list of charid:state records |
| 3006 | (reg_trie_trans_le) the first such element holds the CUR and LEN |
| 3007 | points of the allocated array. (See defines above). |
| 3008 | |
| 3009 | We build the initial structure using the lists, and then convert |
| 3010 | it into the compressed table form which allows faster lookups |
| 3011 | (but cant be modified once converted). |
| 3012 | */ |
| 3013 | |
| 3014 | STRLEN transcount = 1; |
| 3015 | |
| 3016 | DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n", |
| 3017 | depth+1)); |
| 3018 | |
| 3019 | trie->states = (reg_trie_state *) |
| 3020 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, |
| 3021 | sizeof(reg_trie_state) ); |
| 3022 | TRIE_LIST_NEW(1); |
| 3023 | next_alloc = 2; |
| 3024 | |
| 3025 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
| 3026 | |
| 3027 | regnode *noper = NEXTOPER( cur ); |
| 3028 | U32 state = 1; /* required init */ |
| 3029 | U16 charid = 0; /* sanity init */ |
| 3030 | U32 wordlen = 0; /* required init */ |
| 3031 | |
| 3032 | if (OP(noper) == NOTHING) { |
| 3033 | regnode *noper_next= regnext(noper); |
| 3034 | if (noper_next < tail) |
| 3035 | noper= noper_next; |
| 3036 | /* we will undo this assignment if noper does not |
| 3037 | * point at a trieable type in the else clause of |
| 3038 | * the following statement. */ |
| 3039 | } |
| 3040 | |
| 3041 | if ( noper < tail |
| 3042 | && ( OP(noper) == flags |
| 3043 | || (flags == EXACT && OP(noper) == EXACT_REQ8) |
| 3044 | || (flags == EXACTFU && ( OP(noper) == EXACTFU_REQ8 |
| 3045 | || OP(noper) == EXACTFUP)))) |
| 3046 | { |
| 3047 | const U8 *uc= (U8*)STRING(noper); |
| 3048 | const U8 *e= uc + STR_LEN(noper); |
| 3049 | |
| 3050 | for ( ; uc < e ; uc += len ) { |
| 3051 | |
| 3052 | TRIE_READ_CHAR; |
| 3053 | |
| 3054 | if ( uvc < 256 ) { |
| 3055 | charid = trie->charmap[ uvc ]; |
| 3056 | } else { |
| 3057 | SV** const svpp = hv_fetch( widecharmap, |
| 3058 | (char*)&uvc, |
| 3059 | sizeof( UV ), |
| 3060 | 0); |
| 3061 | if ( !svpp ) { |
| 3062 | charid = 0; |
| 3063 | } else { |
| 3064 | charid=(U16)SvIV( *svpp ); |
| 3065 | } |
| 3066 | } |
| 3067 | /* charid is now 0 if we dont know the char read, or |
| 3068 | * nonzero if we do */ |
| 3069 | if ( charid ) { |
| 3070 | |
| 3071 | U16 check; |
| 3072 | U32 newstate = 0; |
| 3073 | |
| 3074 | charid--; |
| 3075 | if ( !trie->states[ state ].trans.list ) { |
| 3076 | TRIE_LIST_NEW( state ); |
| 3077 | } |
| 3078 | for ( check = 1; |
| 3079 | check <= TRIE_LIST_USED( state ); |
| 3080 | check++ ) |
| 3081 | { |
| 3082 | if ( TRIE_LIST_ITEM( state, check ).forid |
| 3083 | == charid ) |
| 3084 | { |
| 3085 | newstate = TRIE_LIST_ITEM( state, check ).newstate; |
| 3086 | break; |
| 3087 | } |
| 3088 | } |
| 3089 | if ( ! newstate ) { |
| 3090 | newstate = next_alloc++; |
| 3091 | prev_states[newstate] = state; |
| 3092 | TRIE_LIST_PUSH( state, charid, newstate ); |
| 3093 | transcount++; |
| 3094 | } |
| 3095 | state = newstate; |
| 3096 | } else { |
| 3097 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc ); |
| 3098 | } |
| 3099 | } |
| 3100 | } else { |
| 3101 | /* If we end up here it is because we skipped past a NOTHING, but did not end up |
| 3102 | * on a trieable type. So we need to reset noper back to point at the first regop |
| 3103 | * in the branch before we call TRIE_HANDLE_WORD() |
| 3104 | */ |
| 3105 | noper= NEXTOPER(cur); |
| 3106 | } |
| 3107 | TRIE_HANDLE_WORD(state); |
| 3108 | |
| 3109 | } /* end second pass */ |
| 3110 | |
| 3111 | /* next alloc is the NEXT state to be allocated */ |
| 3112 | trie->statecount = next_alloc; |
| 3113 | trie->states = (reg_trie_state *) |
| 3114 | PerlMemShared_realloc( trie->states, |
| 3115 | next_alloc |
| 3116 | * sizeof(reg_trie_state) ); |
| 3117 | |
| 3118 | /* and now dump it out before we compress it */ |
| 3119 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap, |
| 3120 | revcharmap, next_alloc, |
| 3121 | depth+1) |
| 3122 | ); |
| 3123 | |
| 3124 | trie->trans = (reg_trie_trans *) |
| 3125 | PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) ); |
| 3126 | { |
| 3127 | U32 state; |
| 3128 | U32 tp = 0; |
| 3129 | U32 zp = 0; |
| 3130 | |
| 3131 | |
| 3132 | for( state=1 ; state < next_alloc ; state ++ ) { |
| 3133 | U32 base=0; |
| 3134 | |
| 3135 | /* |
| 3136 | DEBUG_TRIE_COMPILE_MORE_r( |
| 3137 | Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp) |
| 3138 | ); |
| 3139 | */ |
| 3140 | |
| 3141 | if (trie->states[state].trans.list) { |
| 3142 | U16 minid=TRIE_LIST_ITEM( state, 1).forid; |
| 3143 | U16 maxid=minid; |
| 3144 | U16 idx; |
| 3145 | |
| 3146 | for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) { |
| 3147 | const U16 forid = TRIE_LIST_ITEM( state, idx).forid; |
| 3148 | if ( forid < minid ) { |
| 3149 | minid=forid; |
| 3150 | } else if ( forid > maxid ) { |
| 3151 | maxid=forid; |
| 3152 | } |
| 3153 | } |
| 3154 | if ( transcount < tp + maxid - minid + 1) { |
| 3155 | transcount *= 2; |
| 3156 | trie->trans = (reg_trie_trans *) |
| 3157 | PerlMemShared_realloc( trie->trans, |
| 3158 | transcount |
| 3159 | * sizeof(reg_trie_trans) ); |
| 3160 | Zero( trie->trans + (transcount / 2), |
| 3161 | transcount / 2, |
| 3162 | reg_trie_trans ); |
| 3163 | } |
| 3164 | base = trie->uniquecharcount + tp - minid; |
| 3165 | if ( maxid == minid ) { |
| 3166 | U32 set = 0; |
| 3167 | for ( ; zp < tp ; zp++ ) { |
| 3168 | if ( ! trie->trans[ zp ].next ) { |
| 3169 | base = trie->uniquecharcount + zp - minid; |
| 3170 | trie->trans[ zp ].next = TRIE_LIST_ITEM( state, |
| 3171 | 1).newstate; |
| 3172 | trie->trans[ zp ].check = state; |
| 3173 | set = 1; |
| 3174 | break; |
| 3175 | } |
| 3176 | } |
| 3177 | if ( !set ) { |
| 3178 | trie->trans[ tp ].next = TRIE_LIST_ITEM( state, |
| 3179 | 1).newstate; |
| 3180 | trie->trans[ tp ].check = state; |
| 3181 | tp++; |
| 3182 | zp = tp; |
| 3183 | } |
| 3184 | } else { |
| 3185 | for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) { |
| 3186 | const U32 tid = base |
| 3187 | - trie->uniquecharcount |
| 3188 | + TRIE_LIST_ITEM( state, idx ).forid; |
| 3189 | trie->trans[ tid ].next = TRIE_LIST_ITEM( state, |
| 3190 | idx ).newstate; |
| 3191 | trie->trans[ tid ].check = state; |
| 3192 | } |
| 3193 | tp += ( maxid - minid + 1 ); |
| 3194 | } |
| 3195 | Safefree(trie->states[ state ].trans.list); |
| 3196 | } |
| 3197 | /* |
| 3198 | DEBUG_TRIE_COMPILE_MORE_r( |
| 3199 | Perl_re_printf( aTHX_ " base: %d\n",base); |
| 3200 | ); |
| 3201 | */ |
| 3202 | trie->states[ state ].trans.base=base; |
| 3203 | } |
| 3204 | trie->lasttrans = tp + 1; |
| 3205 | } |
| 3206 | } else { |
| 3207 | /* |
| 3208 | Second Pass -- Flat Table Representation. |
| 3209 | |
| 3210 | we dont use the 0 slot of either trans[] or states[] so we add 1 to |
| 3211 | each. We know that we will need Charcount+1 trans at most to store |
| 3212 | the data (one row per char at worst case) So we preallocate both |
| 3213 | structures assuming worst case. |
| 3214 | |
| 3215 | We then construct the trie using only the .next slots of the entry |
| 3216 | structs. |
| 3217 | |
| 3218 | We use the .check field of the first entry of the node temporarily |
| 3219 | to make compression both faster and easier by keeping track of how |
| 3220 | many non zero fields are in the node. |
| 3221 | |
| 3222 | Since trans are numbered from 1 any 0 pointer in the table is a FAIL |
| 3223 | transition. |
| 3224 | |
| 3225 | There are two terms at use here: state as a TRIE_NODEIDX() which is |
| 3226 | a number representing the first entry of the node, and state as a |
| 3227 | TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) |
| 3228 | and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) |
| 3229 | if there are 2 entrys per node. eg: |
| 3230 | |
| 3231 | A B A B |
| 3232 | 1. 2 4 1. 3 7 |
| 3233 | 2. 0 3 3. 0 5 |
| 3234 | 3. 0 0 5. 0 0 |
| 3235 | 4. 0 0 7. 0 0 |
| 3236 | |
| 3237 | The table is internally in the right hand, idx form. However as we |
| 3238 | also have to deal with the states array which is indexed by nodenum |
| 3239 | we have to use TRIE_NODENUM() to convert. |
| 3240 | |
| 3241 | */ |
| 3242 | DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n", |
| 3243 | depth+1)); |
| 3244 | |
| 3245 | trie->trans = (reg_trie_trans *) |
| 3246 | PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 ) |
| 3247 | * trie->uniquecharcount + 1, |
| 3248 | sizeof(reg_trie_trans) ); |
| 3249 | trie->states = (reg_trie_state *) |
| 3250 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, |
| 3251 | sizeof(reg_trie_state) ); |
| 3252 | next_alloc = trie->uniquecharcount + 1; |
| 3253 | |
| 3254 | |
| 3255 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
| 3256 | |
| 3257 | regnode *noper = NEXTOPER( cur ); |
| 3258 | |
| 3259 | U32 state = 1; /* required init */ |
| 3260 | |
| 3261 | U16 charid = 0; /* sanity init */ |
| 3262 | U32 accept_state = 0; /* sanity init */ |
| 3263 | |
| 3264 | U32 wordlen = 0; /* required init */ |
| 3265 | |
| 3266 | if (OP(noper) == NOTHING) { |
| 3267 | regnode *noper_next= regnext(noper); |
| 3268 | if (noper_next < tail) |
| 3269 | noper= noper_next; |
| 3270 | /* we will undo this assignment if noper does not |
| 3271 | * point at a trieable type in the else clause of |
| 3272 | * the following statement. */ |
| 3273 | } |
| 3274 | |
| 3275 | if ( noper < tail |
| 3276 | && ( OP(noper) == flags |
| 3277 | || (flags == EXACT && OP(noper) == EXACT_REQ8) |
| 3278 | || (flags == EXACTFU && ( OP(noper) == EXACTFU_REQ8 |
| 3279 | || OP(noper) == EXACTFUP)))) |
| 3280 | { |
| 3281 | const U8 *uc= (U8*)STRING(noper); |
| 3282 | const U8 *e= uc + STR_LEN(noper); |
| 3283 | |
| 3284 | for ( ; uc < e ; uc += len ) { |
| 3285 | |
| 3286 | TRIE_READ_CHAR; |
| 3287 | |
| 3288 | if ( uvc < 256 ) { |
| 3289 | charid = trie->charmap[ uvc ]; |
| 3290 | } else { |
| 3291 | SV* const * const svpp = hv_fetch( widecharmap, |
| 3292 | (char*)&uvc, |
| 3293 | sizeof( UV ), |
| 3294 | 0); |
| 3295 | charid = svpp ? (U16)SvIV(*svpp) : 0; |
| 3296 | } |
| 3297 | if ( charid ) { |
| 3298 | charid--; |
| 3299 | if ( !trie->trans[ state + charid ].next ) { |
| 3300 | trie->trans[ state + charid ].next = next_alloc; |
| 3301 | trie->trans[ state ].check++; |
| 3302 | prev_states[TRIE_NODENUM(next_alloc)] |
| 3303 | = TRIE_NODENUM(state); |
| 3304 | next_alloc += trie->uniquecharcount; |
| 3305 | } |
| 3306 | state = trie->trans[ state + charid ].next; |
| 3307 | } else { |
| 3308 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc ); |
| 3309 | } |
| 3310 | /* charid is now 0 if we dont know the char read, or |
| 3311 | * nonzero if we do */ |
| 3312 | } |
| 3313 | } else { |
| 3314 | /* If we end up here it is because we skipped past a NOTHING, but did not end up |
| 3315 | * on a trieable type. So we need to reset noper back to point at the first regop |
| 3316 | * in the branch before we call TRIE_HANDLE_WORD(). |
| 3317 | */ |
| 3318 | noper= NEXTOPER(cur); |
| 3319 | } |
| 3320 | accept_state = TRIE_NODENUM( state ); |
| 3321 | TRIE_HANDLE_WORD(accept_state); |
| 3322 | |
| 3323 | } /* end second pass */ |
| 3324 | |
| 3325 | /* and now dump it out before we compress it */ |
| 3326 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap, |
| 3327 | revcharmap, |
| 3328 | next_alloc, depth+1)); |
| 3329 | |
| 3330 | { |
| 3331 | /* |
| 3332 | * Inplace compress the table.* |
| 3333 | |
| 3334 | For sparse data sets the table constructed by the trie algorithm will |
| 3335 | be mostly 0/FAIL transitions or to put it another way mostly empty. |
| 3336 | (Note that leaf nodes will not contain any transitions.) |
| 3337 | |
| 3338 | This algorithm compresses the tables by eliminating most such |
| 3339 | transitions, at the cost of a modest bit of extra work during lookup: |
| 3340 | |
| 3341 | - Each states[] entry contains a .base field which indicates the |
| 3342 | index in the state[] array wheres its transition data is stored. |
| 3343 | |
| 3344 | - If .base is 0 there are no valid transitions from that node. |
| 3345 | |
| 3346 | - If .base is nonzero then charid is added to it to find an entry in |
| 3347 | the trans array. |
| 3348 | |
| 3349 | -If trans[states[state].base+charid].check!=state then the |
| 3350 | transition is taken to be a 0/Fail transition. Thus if there are fail |
| 3351 | transitions at the front of the node then the .base offset will point |
| 3352 | somewhere inside the previous nodes data (or maybe even into a node |
| 3353 | even earlier), but the .check field determines if the transition is |
| 3354 | valid. |
| 3355 | |
| 3356 | XXX - wrong maybe? |
| 3357 | The following process inplace converts the table to the compressed |
| 3358 | table: We first do not compress the root node 1,and mark all its |
| 3359 | .check pointers as 1 and set its .base pointer as 1 as well. This |
| 3360 | allows us to do a DFA construction from the compressed table later, |
| 3361 | and ensures that any .base pointers we calculate later are greater |
| 3362 | than 0. |
| 3363 | |
| 3364 | - We set 'pos' to indicate the first entry of the second node. |
| 3365 | |
| 3366 | - We then iterate over the columns of the node, finding the first and |
| 3367 | last used entry at l and m. We then copy l..m into pos..(pos+m-l), |
| 3368 | and set the .check pointers accordingly, and advance pos |
| 3369 | appropriately and repreat for the next node. Note that when we copy |
| 3370 | the next pointers we have to convert them from the original |
| 3371 | NODEIDX form to NODENUM form as the former is not valid post |
| 3372 | compression. |
| 3373 | |
| 3374 | - If a node has no transitions used we mark its base as 0 and do not |
| 3375 | advance the pos pointer. |
| 3376 | |
| 3377 | - If a node only has one transition we use a second pointer into the |
| 3378 | structure to fill in allocated fail transitions from other states. |
| 3379 | This pointer is independent of the main pointer and scans forward |
| 3380 | looking for null transitions that are allocated to a state. When it |
| 3381 | finds one it writes the single transition into the "hole". If the |
| 3382 | pointer doesnt find one the single transition is appended as normal. |
| 3383 | |
| 3384 | - Once compressed we can Renew/realloc the structures to release the |
| 3385 | excess space. |
| 3386 | |
| 3387 | See "Table-Compression Methods" in sec 3.9 of the Red Dragon, |
| 3388 | specifically Fig 3.47 and the associated pseudocode. |
| 3389 | |
| 3390 | demq |
| 3391 | */ |
| 3392 | const U32 laststate = TRIE_NODENUM( next_alloc ); |
| 3393 | U32 state, charid; |
| 3394 | U32 pos = 0, zp=0; |
| 3395 | trie->statecount = laststate; |
| 3396 | |
| 3397 | for ( state = 1 ; state < laststate ; state++ ) { |
| 3398 | U8 flag = 0; |
| 3399 | const U32 stateidx = TRIE_NODEIDX( state ); |
| 3400 | const U32 o_used = trie->trans[ stateidx ].check; |
| 3401 | U32 used = trie->trans[ stateidx ].check; |
| 3402 | trie->trans[ stateidx ].check = 0; |
| 3403 | |
| 3404 | for ( charid = 0; |
| 3405 | used && charid < trie->uniquecharcount; |
| 3406 | charid++ ) |
| 3407 | { |
| 3408 | if ( flag || trie->trans[ stateidx + charid ].next ) { |
| 3409 | if ( trie->trans[ stateidx + charid ].next ) { |
| 3410 | if (o_used == 1) { |
| 3411 | for ( ; zp < pos ; zp++ ) { |
| 3412 | if ( ! trie->trans[ zp ].next ) { |
| 3413 | break; |
| 3414 | } |
| 3415 | } |
| 3416 | trie->states[ state ].trans.base |
| 3417 | = zp |
| 3418 | + trie->uniquecharcount |
| 3419 | - charid ; |
| 3420 | trie->trans[ zp ].next |
| 3421 | = SAFE_TRIE_NODENUM( trie->trans[ stateidx |
| 3422 | + charid ].next ); |
| 3423 | trie->trans[ zp ].check = state; |
| 3424 | if ( ++zp > pos ) pos = zp; |
| 3425 | break; |
| 3426 | } |
| 3427 | used--; |
| 3428 | } |
| 3429 | if ( !flag ) { |
| 3430 | flag = 1; |
| 3431 | trie->states[ state ].trans.base |
| 3432 | = pos + trie->uniquecharcount - charid ; |
| 3433 | } |
| 3434 | trie->trans[ pos ].next |
| 3435 | = SAFE_TRIE_NODENUM( |
| 3436 | trie->trans[ stateidx + charid ].next ); |
| 3437 | trie->trans[ pos ].check = state; |
| 3438 | pos++; |
| 3439 | } |
| 3440 | } |
| 3441 | } |
| 3442 | trie->lasttrans = pos + 1; |
| 3443 | trie->states = (reg_trie_state *) |
| 3444 | PerlMemShared_realloc( trie->states, laststate |
| 3445 | * sizeof(reg_trie_state) ); |
| 3446 | DEBUG_TRIE_COMPILE_MORE_r( |
| 3447 | Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n", |
| 3448 | depth+1, |
| 3449 | (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount |
| 3450 | + 1 ), |
| 3451 | (IV)next_alloc, |
| 3452 | (IV)pos, |
| 3453 | ( ( next_alloc - pos ) * 100 ) / (double)next_alloc ); |
| 3454 | ); |
| 3455 | |
| 3456 | } /* end table compress */ |
| 3457 | } |
| 3458 | DEBUG_TRIE_COMPILE_MORE_r( |
| 3459 | Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n", |
| 3460 | depth+1, |
| 3461 | (UV)trie->statecount, |
| 3462 | (UV)trie->lasttrans) |
| 3463 | ); |
| 3464 | /* resize the trans array to remove unused space */ |
| 3465 | trie->trans = (reg_trie_trans *) |
| 3466 | PerlMemShared_realloc( trie->trans, trie->lasttrans |
| 3467 | * sizeof(reg_trie_trans) ); |
| 3468 | |
| 3469 | { /* Modify the program and insert the new TRIE node */ |
| 3470 | U8 nodetype =(U8)(flags & 0xFF); |
| 3471 | char *str=NULL; |
| 3472 | |
| 3473 | #ifdef DEBUGGING |
| 3474 | regnode *optimize = NULL; |
| 3475 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 3476 | |
| 3477 | U32 mjd_offset = 0; |
| 3478 | U32 mjd_nodelen = 0; |
| 3479 | #endif /* RE_TRACK_PATTERN_OFFSETS */ |
| 3480 | #endif /* DEBUGGING */ |
| 3481 | /* |
| 3482 | This means we convert either the first branch or the first Exact, |
| 3483 | depending on whether the thing following (in 'last') is a branch |
| 3484 | or not and whther first is the startbranch (ie is it a sub part of |
| 3485 | the alternation or is it the whole thing.) |
| 3486 | Assuming its a sub part we convert the EXACT otherwise we convert |
| 3487 | the whole branch sequence, including the first. |
| 3488 | */ |
| 3489 | /* Find the node we are going to overwrite */ |
| 3490 | if ( first != startbranch || OP( last ) == BRANCH ) { |
| 3491 | /* branch sub-chain */ |
| 3492 | NEXT_OFF( first ) = (U16)(last - first); |
| 3493 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 3494 | DEBUG_r({ |
| 3495 | mjd_offset= Node_Offset((convert)); |
| 3496 | mjd_nodelen= Node_Length((convert)); |
| 3497 | }); |
| 3498 | #endif |
| 3499 | /* whole branch chain */ |
| 3500 | } |
| 3501 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 3502 | else { |
| 3503 | DEBUG_r({ |
| 3504 | const regnode *nop = NEXTOPER( convert ); |
| 3505 | mjd_offset= Node_Offset((nop)); |
| 3506 | mjd_nodelen= Node_Length((nop)); |
| 3507 | }); |
| 3508 | } |
| 3509 | DEBUG_OPTIMISE_r( |
| 3510 | Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n", |
| 3511 | depth+1, |
| 3512 | (UV)mjd_offset, (UV)mjd_nodelen) |
| 3513 | ); |
| 3514 | #endif |
| 3515 | /* But first we check to see if there is a common prefix we can |
| 3516 | split out as an EXACT and put in front of the TRIE node. */ |
| 3517 | trie->startstate= 1; |
| 3518 | if ( trie->bitmap && !widecharmap && !trie->jump ) { |
| 3519 | /* we want to find the first state that has more than |
| 3520 | * one transition, if that state is not the first state |
| 3521 | * then we have a common prefix which we can remove. |
| 3522 | */ |
| 3523 | U32 state; |
| 3524 | for ( state = 1 ; state < trie->statecount-1 ; state++ ) { |
| 3525 | U32 ofs = 0; |
| 3526 | I32 first_ofs = -1; /* keeps track of the ofs of the first |
| 3527 | transition, -1 means none */ |
| 3528 | U32 count = 0; |
| 3529 | const U32 base = trie->states[ state ].trans.base; |
| 3530 | |
| 3531 | /* does this state terminate an alternation? */ |
| 3532 | if ( trie->states[state].wordnum ) |
| 3533 | count = 1; |
| 3534 | |
| 3535 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { |
| 3536 | if ( ( base + ofs >= trie->uniquecharcount ) && |
| 3537 | ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && |
| 3538 | trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) |
| 3539 | { |
| 3540 | if ( ++count > 1 ) { |
| 3541 | /* we have more than one transition */ |
| 3542 | SV **tmp; |
| 3543 | U8 *ch; |
| 3544 | /* if this is the first state there is no common prefix |
| 3545 | * to extract, so we can exit */ |
| 3546 | if ( state == 1 ) break; |
| 3547 | tmp = av_fetch( revcharmap, ofs, 0); |
| 3548 | ch = (U8*)SvPV_nolen_const( *tmp ); |
| 3549 | |
| 3550 | /* if we are on count 2 then we need to initialize the |
| 3551 | * bitmap, and store the previous char if there was one |
| 3552 | * in it*/ |
| 3553 | if ( count == 2 ) { |
| 3554 | /* clear the bitmap */ |
| 3555 | Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char); |
| 3556 | DEBUG_OPTIMISE_r( |
| 3557 | Perl_re_indentf( aTHX_ "New Start State=%" UVuf " Class: [", |
| 3558 | depth+1, |
| 3559 | (UV)state)); |
| 3560 | if (first_ofs >= 0) { |
| 3561 | SV ** const tmp = av_fetch( revcharmap, first_ofs, 0); |
| 3562 | const U8 * const ch = (U8*)SvPV_nolen_const( *tmp ); |
| 3563 | |
| 3564 | TRIE_BITMAP_SET_FOLDED(trie,*ch, folder); |
| 3565 | DEBUG_OPTIMISE_r( |
| 3566 | Perl_re_printf( aTHX_ "%s", (char*)ch) |
| 3567 | ); |
| 3568 | } |
| 3569 | } |
| 3570 | /* store the current firstchar in the bitmap */ |
| 3571 | TRIE_BITMAP_SET_FOLDED(trie,*ch, folder); |
| 3572 | DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "%s", ch)); |
| 3573 | } |
| 3574 | first_ofs = ofs; |
| 3575 | } |
| 3576 | } |
| 3577 | if ( count == 1 ) { |
| 3578 | /* This state has only one transition, its transition is part |
| 3579 | * of a common prefix - we need to concatenate the char it |
| 3580 | * represents to what we have so far. */ |
| 3581 | SV **tmp = av_fetch( revcharmap, first_ofs, 0); |
| 3582 | STRLEN len; |
| 3583 | char *ch = SvPV( *tmp, len ); |
| 3584 | DEBUG_OPTIMISE_r({ |
| 3585 | SV *sv=sv_newmortal(); |
| 3586 | Perl_re_indentf( aTHX_ "Prefix State: %" UVuf " Ofs:%" UVuf " Char='%s'\n", |
| 3587 | depth+1, |
| 3588 | (UV)state, (UV)first_ofs, |
| 3589 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6, |
| 3590 | PL_colors[0], PL_colors[1], |
| 3591 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | |
| 3592 | PERL_PV_ESCAPE_FIRSTCHAR |
| 3593 | ) |
| 3594 | ); |
| 3595 | }); |
| 3596 | if ( state==1 ) { |
| 3597 | OP( convert ) = nodetype; |
| 3598 | str=STRING(convert); |
| 3599 | setSTR_LEN(convert, 0); |
| 3600 | } |
| 3601 | assert( ( STR_LEN(convert) + len ) < 256 ); |
| 3602 | setSTR_LEN(convert, (U8)(STR_LEN(convert) + len)); |
| 3603 | while (len--) |
| 3604 | *str++ = *ch++; |
| 3605 | } else { |
| 3606 | #ifdef DEBUGGING |
| 3607 | if (state>1) |
| 3608 | DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "]\n")); |
| 3609 | #endif |
| 3610 | break; |
| 3611 | } |
| 3612 | } |
| 3613 | trie->prefixlen = (state-1); |
| 3614 | if (str) { |
| 3615 | regnode *n = convert+NODE_SZ_STR(convert); |
| 3616 | assert( NODE_SZ_STR(convert) <= U16_MAX ); |
| 3617 | NEXT_OFF(convert) = (U16)(NODE_SZ_STR(convert)); |
| 3618 | trie->startstate = state; |
| 3619 | trie->minlen -= (state - 1); |
| 3620 | trie->maxlen -= (state - 1); |
| 3621 | #ifdef DEBUGGING |
| 3622 | /* At least the UNICOS C compiler choked on this |
| 3623 | * being argument to DEBUG_r(), so let's just have |
| 3624 | * it right here. */ |
| 3625 | if ( |
| 3626 | #ifdef PERL_EXT_RE_BUILD |
| 3627 | 1 |
| 3628 | #else |
| 3629 | DEBUG_r_TEST |
| 3630 | #endif |
| 3631 | ) { |
| 3632 | regnode *fix = convert; |
| 3633 | U32 word = trie->wordcount; |
| 3634 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 3635 | mjd_nodelen++; |
| 3636 | #endif |
| 3637 | Set_Node_Offset_Length(convert, mjd_offset, state - 1); |
| 3638 | while( ++fix < n ) { |
| 3639 | Set_Node_Offset_Length(fix, 0, 0); |
| 3640 | } |
| 3641 | while (word--) { |
| 3642 | SV ** const tmp = av_fetch( trie_words, word, 0 ); |
| 3643 | if (tmp) { |
| 3644 | if ( STR_LEN(convert) <= SvCUR(*tmp) ) |
| 3645 | sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert)); |
| 3646 | else |
| 3647 | sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp)); |
| 3648 | } |
| 3649 | } |
| 3650 | } |
| 3651 | #endif |
| 3652 | if (trie->maxlen) { |
| 3653 | convert = n; |
| 3654 | } else { |
| 3655 | NEXT_OFF(convert) = (U16)(tail - convert); |
| 3656 | DEBUG_r(optimize= n); |
| 3657 | } |
| 3658 | } |
| 3659 | } |
| 3660 | if (!jumper) |
| 3661 | jumper = last; |
| 3662 | if ( trie->maxlen ) { |
| 3663 | NEXT_OFF( convert ) = (U16)(tail - convert); |
| 3664 | ARG_SET( convert, data_slot ); |
| 3665 | /* Store the offset to the first unabsorbed branch in |
| 3666 | jump[0], which is otherwise unused by the jump logic. |
| 3667 | We use this when dumping a trie and during optimisation. */ |
| 3668 | if (trie->jump) |
| 3669 | trie->jump[0] = (U16)(nextbranch - convert); |
| 3670 | |
| 3671 | /* If the start state is not accepting (meaning there is no empty string/NOTHING) |
| 3672 | * and there is a bitmap |
| 3673 | * and the first "jump target" node we found leaves enough room |
| 3674 | * then convert the TRIE node into a TRIEC node, with the bitmap |
| 3675 | * embedded inline in the opcode - this is hypothetically faster. |
| 3676 | */ |
| 3677 | if ( !trie->states[trie->startstate].wordnum |
| 3678 | && trie->bitmap |
| 3679 | && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) ) |
| 3680 | { |
| 3681 | OP( convert ) = TRIEC; |
| 3682 | Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char); |
| 3683 | PerlMemShared_free(trie->bitmap); |
| 3684 | trie->bitmap= NULL; |
| 3685 | } else |
| 3686 | OP( convert ) = TRIE; |
| 3687 | |
| 3688 | /* store the type in the flags */ |
| 3689 | convert->flags = nodetype; |
| 3690 | DEBUG_r({ |
| 3691 | optimize = convert |
| 3692 | + NODE_STEP_REGNODE |
| 3693 | + regarglen[ OP( convert ) ]; |
| 3694 | }); |
| 3695 | /* XXX We really should free up the resource in trie now, |
| 3696 | as we won't use them - (which resources?) dmq */ |
| 3697 | } |
| 3698 | /* needed for dumping*/ |
| 3699 | DEBUG_r(if (optimize) { |
| 3700 | regnode *opt = convert; |
| 3701 | |
| 3702 | while ( ++opt < optimize) { |
| 3703 | Set_Node_Offset_Length(opt, 0, 0); |
| 3704 | } |
| 3705 | /* |
| 3706 | Try to clean up some of the debris left after the |
| 3707 | optimisation. |
| 3708 | */ |
| 3709 | while( optimize < jumper ) { |
| 3710 | Track_Code( mjd_nodelen += Node_Length((optimize)); ); |
| 3711 | OP( optimize ) = OPTIMIZED; |
| 3712 | Set_Node_Offset_Length(optimize, 0, 0); |
| 3713 | optimize++; |
| 3714 | } |
| 3715 | Set_Node_Offset_Length(convert, mjd_offset, mjd_nodelen); |
| 3716 | }); |
| 3717 | } /* end node insert */ |
| 3718 | |
| 3719 | /* Finish populating the prev field of the wordinfo array. Walk back |
| 3720 | * from each accept state until we find another accept state, and if |
| 3721 | * so, point the first word's .prev field at the second word. If the |
| 3722 | * second already has a .prev field set, stop now. This will be the |
| 3723 | * case either if we've already processed that word's accept state, |
| 3724 | * or that state had multiple words, and the overspill words were |
| 3725 | * already linked up earlier. |
| 3726 | */ |
| 3727 | { |
| 3728 | U16 word; |
| 3729 | U32 state; |
| 3730 | U16 prev; |
| 3731 | |
| 3732 | for (word=1; word <= trie->wordcount; word++) { |
| 3733 | prev = 0; |
| 3734 | if (trie->wordinfo[word].prev) |
| 3735 | continue; |
| 3736 | state = trie->wordinfo[word].accept; |
| 3737 | while (state) { |
| 3738 | state = prev_states[state]; |
| 3739 | if (!state) |
| 3740 | break; |
| 3741 | prev = trie->states[state].wordnum; |
| 3742 | if (prev) |
| 3743 | break; |
| 3744 | } |
| 3745 | trie->wordinfo[word].prev = prev; |
| 3746 | } |
| 3747 | Safefree(prev_states); |
| 3748 | } |
| 3749 | |
| 3750 | |
| 3751 | /* and now dump out the compressed format */ |
| 3752 | DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1)); |
| 3753 | |
| 3754 | RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap; |
| 3755 | #ifdef DEBUGGING |
| 3756 | RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words; |
| 3757 | RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap; |
| 3758 | #else |
| 3759 | SvREFCNT_dec_NN(revcharmap); |
| 3760 | #endif |
| 3761 | return trie->jump |
| 3762 | ? MADE_JUMP_TRIE |
| 3763 | : trie->startstate>1 |
| 3764 | ? MADE_EXACT_TRIE |
| 3765 | : MADE_TRIE; |
| 3766 | } |
| 3767 | |
| 3768 | STATIC regnode * |
| 3769 | S_construct_ahocorasick_from_trie(pTHX_ RExC_state_t *pRExC_state, regnode *source, U32 depth) |
| 3770 | { |
| 3771 | /* The Trie is constructed and compressed now so we can build a fail array if |
| 3772 | * it's needed |
| 3773 | |
| 3774 | This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and |
| 3775 | 3.32 in the |
| 3776 | "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, |
| 3777 | Ullman 1985/88 |
| 3778 | ISBN 0-201-10088-6 |
| 3779 | |
| 3780 | We find the fail state for each state in the trie, this state is the longest |
| 3781 | proper suffix of the current state's 'word' that is also a proper prefix of |
| 3782 | another word in our trie. State 1 represents the word '' and is thus the |
| 3783 | default fail state. This allows the DFA not to have to restart after its |
| 3784 | tried and failed a word at a given point, it simply continues as though it |
| 3785 | had been matching the other word in the first place. |
| 3786 | Consider |
| 3787 | 'abcdgu'=~/abcdefg|cdgu/ |
| 3788 | When we get to 'd' we are still matching the first word, we would encounter |
| 3789 | 'g' which would fail, which would bring us to the state representing 'd' in |
| 3790 | the second word where we would try 'g' and succeed, proceeding to match |
| 3791 | 'cdgu'. |
| 3792 | */ |
| 3793 | /* add a fail transition */ |
| 3794 | const U32 trie_offset = ARG(source); |
| 3795 | reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset]; |
| 3796 | U32 *q; |
| 3797 | const U32 ucharcount = trie->uniquecharcount; |
| 3798 | const U32 numstates = trie->statecount; |
| 3799 | const U32 ubound = trie->lasttrans + ucharcount; |
| 3800 | U32 q_read = 0; |
| 3801 | U32 q_write = 0; |
| 3802 | U32 charid; |
| 3803 | U32 base = trie->states[ 1 ].trans.base; |
| 3804 | U32 *fail; |
| 3805 | reg_ac_data *aho; |
| 3806 | const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T")); |
| 3807 | regnode *stclass; |
| 3808 | GET_RE_DEBUG_FLAGS_DECL; |
| 3809 | |
| 3810 | PERL_ARGS_ASSERT_CONSTRUCT_AHOCORASICK_FROM_TRIE; |
| 3811 | PERL_UNUSED_CONTEXT; |
| 3812 | #ifndef DEBUGGING |
| 3813 | PERL_UNUSED_ARG(depth); |
| 3814 | #endif |
| 3815 | |
| 3816 | if ( OP(source) == TRIE ) { |
| 3817 | struct regnode_1 *op = (struct regnode_1 *) |
| 3818 | PerlMemShared_calloc(1, sizeof(struct regnode_1)); |
| 3819 | StructCopy(source, op, struct regnode_1); |
| 3820 | stclass = (regnode *)op; |
| 3821 | } else { |
| 3822 | struct regnode_charclass *op = (struct regnode_charclass *) |
| 3823 | PerlMemShared_calloc(1, sizeof(struct regnode_charclass)); |
| 3824 | StructCopy(source, op, struct regnode_charclass); |
| 3825 | stclass = (regnode *)op; |
| 3826 | } |
| 3827 | OP(stclass)+=2; /* convert the TRIE type to its AHO-CORASICK equivalent */ |
| 3828 | |
| 3829 | ARG_SET( stclass, data_slot ); |
| 3830 | aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) ); |
| 3831 | RExC_rxi->data->data[ data_slot ] = (void*)aho; |
| 3832 | aho->trie=trie_offset; |
| 3833 | aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) ); |
| 3834 | Copy( trie->states, aho->states, numstates, reg_trie_state ); |
| 3835 | Newx( q, numstates, U32); |
| 3836 | aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) ); |
| 3837 | aho->refcount = 1; |
| 3838 | fail = aho->fail; |
| 3839 | /* initialize fail[0..1] to be 1 so that we always have |
| 3840 | a valid final fail state */ |
| 3841 | fail[ 0 ] = fail[ 1 ] = 1; |
| 3842 | |
| 3843 | for ( charid = 0; charid < ucharcount ; charid++ ) { |
| 3844 | const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 ); |
| 3845 | if ( newstate ) { |
| 3846 | q[ q_write ] = newstate; |
| 3847 | /* set to point at the root */ |
| 3848 | fail[ q[ q_write++ ] ]=1; |
| 3849 | } |
| 3850 | } |
| 3851 | while ( q_read < q_write) { |
| 3852 | const U32 cur = q[ q_read++ % numstates ]; |
| 3853 | base = trie->states[ cur ].trans.base; |
| 3854 | |
| 3855 | for ( charid = 0 ; charid < ucharcount ; charid++ ) { |
| 3856 | const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 ); |
| 3857 | if (ch_state) { |
| 3858 | U32 fail_state = cur; |
| 3859 | U32 fail_base; |
| 3860 | do { |
| 3861 | fail_state = fail[ fail_state ]; |
| 3862 | fail_base = aho->states[ fail_state ].trans.base; |
| 3863 | } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) ); |
| 3864 | |
| 3865 | fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ); |
| 3866 | fail[ ch_state ] = fail_state; |
| 3867 | if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum ) |
| 3868 | { |
| 3869 | aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum; |
| 3870 | } |
| 3871 | q[ q_write++ % numstates] = ch_state; |
| 3872 | } |
| 3873 | } |
| 3874 | } |
| 3875 | /* restore fail[0..1] to 0 so that we "fall out" of the AC loop |
| 3876 | when we fail in state 1, this allows us to use the |
| 3877 | charclass scan to find a valid start char. This is based on the principle |
| 3878 | that theres a good chance the string being searched contains lots of stuff |
| 3879 | that cant be a start char. |
| 3880 | */ |
| 3881 | fail[ 0 ] = fail[ 1 ] = 0; |
| 3882 | DEBUG_TRIE_COMPILE_r({ |
| 3883 | Perl_re_indentf( aTHX_ "Stclass Failtable (%" UVuf " states): 0", |
| 3884 | depth, (UV)numstates |
| 3885 | ); |
| 3886 | for( q_read=1; q_read<numstates; q_read++ ) { |
| 3887 | Perl_re_printf( aTHX_ ", %" UVuf, (UV)fail[q_read]); |
| 3888 | } |
| 3889 | Perl_re_printf( aTHX_ "\n"); |
| 3890 | }); |
| 3891 | Safefree(q); |
| 3892 | /*RExC_seen |= REG_TRIEDFA_SEEN;*/ |
| 3893 | return stclass; |
| 3894 | } |
| 3895 | |
| 3896 | |
| 3897 | /* The below joins as many adjacent EXACTish nodes as possible into a single |
| 3898 | * one. The regop may be changed if the node(s) contain certain sequences that |
| 3899 | * require special handling. The joining is only done if: |
| 3900 | * 1) there is room in the current conglomerated node to entirely contain the |
| 3901 | * next one. |
| 3902 | * 2) they are compatible node types |
| 3903 | * |
| 3904 | * The adjacent nodes actually may be separated by NOTHING-kind nodes, and |
| 3905 | * these get optimized out |
| 3906 | * |
| 3907 | * XXX khw thinks this should be enhanced to fill EXACT (at least) nodes as full |
| 3908 | * as possible, even if that means splitting an existing node so that its first |
| 3909 | * part is moved to the preceeding node. This would maximise the efficiency of |
| 3910 | * memEQ during matching. |
| 3911 | * |
| 3912 | * If a node is to match under /i (folded), the number of characters it matches |
| 3913 | * can be different than its character length if it contains a multi-character |
| 3914 | * fold. *min_subtract is set to the total delta number of characters of the |
| 3915 | * input nodes. |
| 3916 | * |
| 3917 | * And *unfolded_multi_char is set to indicate whether or not the node contains |
| 3918 | * an unfolded multi-char fold. This happens when it won't be known until |
| 3919 | * runtime whether the fold is valid or not; namely |
| 3920 | * 1) for EXACTF nodes that contain LATIN SMALL LETTER SHARP S, as only if the |
| 3921 | * target string being matched against turns out to be UTF-8 is that fold |
| 3922 | * valid; or |
| 3923 | * 2) for EXACTFL nodes whose folding rules depend on the locale in force at |
| 3924 | * runtime. |
| 3925 | * (Multi-char folds whose components are all above the Latin1 range are not |
| 3926 | * run-time locale dependent, and have already been folded by the time this |
| 3927 | * function is called.) |
| 3928 | * |
| 3929 | * This is as good a place as any to discuss the design of handling these |
| 3930 | * multi-character fold sequences. It's been wrong in Perl for a very long |
| 3931 | * time. There are three code points in Unicode whose multi-character folds |
| 3932 | * were long ago discovered to mess things up. The previous designs for |
| 3933 | * dealing with these involved assigning a special node for them. This |
| 3934 | * approach doesn't always work, as evidenced by this example: |
| 3935 | * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches |
| 3936 | * Both sides fold to "sss", but if the pattern is parsed to create a node that |
| 3937 | * would match just the \xDF, it won't be able to handle the case where a |
| 3938 | * successful match would have to cross the node's boundary. The new approach |
| 3939 | * that hopefully generally solves the problem generates an EXACTFUP node |
| 3940 | * that is "sss" in this case. |
| 3941 | * |
| 3942 | * It turns out that there are problems with all multi-character folds, and not |
| 3943 | * just these three. Now the code is general, for all such cases. The |
| 3944 | * approach taken is: |
| 3945 | * 1) This routine examines each EXACTFish node that could contain multi- |
| 3946 | * character folded sequences. Since a single character can fold into |
| 3947 | * such a sequence, the minimum match length for this node is less than |
| 3948 | * the number of characters in the node. This routine returns in |
| 3949 | * *min_subtract how many characters to subtract from the the actual |
| 3950 | * length of the string to get a real minimum match length; it is 0 if |
| 3951 | * there are no multi-char foldeds. This delta is used by the caller to |
| 3952 | * adjust the min length of the match, and the delta between min and max, |
| 3953 | * so that the optimizer doesn't reject these possibilities based on size |
| 3954 | * constraints. |
| 3955 | * |
| 3956 | * 2) For the sequence involving the LATIN SMALL LETTER SHARP S (U+00DF) |
| 3957 | * under /u, we fold it to 'ss' in regatom(), and in this routine, after |
| 3958 | * joining, we scan for occurrences of the sequence 'ss' in non-UTF-8 |
| 3959 | * EXACTFU nodes. The node type of such nodes is then changed to |
| 3960 | * EXACTFUP, indicating it is problematic, and needs careful handling. |
| 3961 | * (The procedures in step 1) above are sufficient to handle this case in |
| 3962 | * UTF-8 encoded nodes.) The reason this is problematic is that this is |
| 3963 | * the only case where there is a possible fold length change in non-UTF-8 |
| 3964 | * patterns. By reserving a special node type for problematic cases, the |
| 3965 | * far more common regular EXACTFU nodes can be processed faster. |
| 3966 | * regexec.c takes advantage of this. |
| 3967 | * |
| 3968 | * EXACTFUP has been created as a grab-bag for (hopefully uncommon) |
| 3969 | * problematic cases. These all only occur when the pattern is not |
| 3970 | * UTF-8. In addition to the 'ss' sequence where there is a possible fold |
| 3971 | * length change, it handles the situation where the string cannot be |
| 3972 | * entirely folded. The strings in an EXACTFish node are folded as much |
| 3973 | * as possible during compilation in regcomp.c. This saves effort in |
| 3974 | * regex matching. By using an EXACTFUP node when it is not possible to |
| 3975 | * fully fold at compile time, regexec.c can know that everything in an |
| 3976 | * EXACTFU node is folded, so folding can be skipped at runtime. The only |
| 3977 | * case where folding in EXACTFU nodes can't be done at compile time is |
| 3978 | * the presumably uncommon MICRO SIGN, when the pattern isn't UTF-8. This |
| 3979 | * is because its fold requires UTF-8 to represent. Thus EXACTFUP nodes |
| 3980 | * handle two very different cases. Alternatively, there could have been |
| 3981 | * a node type where there are length changes, one for unfolded, and one |
| 3982 | * for both. If yet another special case needed to be created, the number |
| 3983 | * of required node types would have to go to 7. khw figures that even |
| 3984 | * though there are plenty of node types to spare, that the maintenance |
| 3985 | * cost wasn't worth the small speedup of doing it that way, especially |
| 3986 | * since he thinks the MICRO SIGN is rarely encountered in practice. |
| 3987 | * |
| 3988 | * There are other cases where folding isn't done at compile time, but |
| 3989 | * none of them are under /u, and hence not for EXACTFU nodes. The folds |
| 3990 | * in EXACTFL nodes aren't known until runtime, and vary as the locale |
| 3991 | * changes. Some folds in EXACTF depend on if the runtime target string |
| 3992 | * is UTF-8 or not. (regatom() will create an EXACTFU node even under /di |
| 3993 | * when no fold in it depends on the UTF-8ness of the target string.) |
| 3994 | * |
| 3995 | * 3) A problem remains for unfolded multi-char folds. (These occur when the |
| 3996 | * validity of the fold won't be known until runtime, and so must remain |
| 3997 | * unfolded for now. This happens for the sharp s in EXACTF and EXACTFAA |
| 3998 | * nodes when the pattern isn't in UTF-8. (Note, BTW, that there cannot |
| 3999 | * be an EXACTF node with a UTF-8 pattern.) They also occur for various |
| 4000 | * folds in EXACTFL nodes, regardless of the UTF-ness of the pattern.) |
| 4001 | * The reason this is a problem is that the optimizer part of regexec.c |
| 4002 | * (probably unwittingly, in Perl_regexec_flags()) makes an assumption |
| 4003 | * that a character in the pattern corresponds to at most a single |
| 4004 | * character in the target string. (And I do mean character, and not byte |
| 4005 | * here, unlike other parts of the documentation that have never been |
| 4006 | * updated to account for multibyte Unicode.) Sharp s in EXACTF and |
| 4007 | * EXACTFL nodes can match the two character string 'ss'; in EXACTFAA |
| 4008 | * nodes it can match "\x{17F}\x{17F}". These, along with other ones in |
| 4009 | * EXACTFL nodes, violate the assumption, and they are the only instances |
| 4010 | * where it is violated. I'm reluctant to try to change the assumption, |
| 4011 | * as the code involved is impenetrable to me (khw), so instead the code |
| 4012 | * here punts. This routine examines EXACTFL nodes, and (when the pattern |
| 4013 | * isn't UTF-8) EXACTF and EXACTFAA for such unfolded folds, and returns a |
| 4014 | * boolean indicating whether or not the node contains such a fold. When |
| 4015 | * it is true, the caller sets a flag that later causes the optimizer in |
| 4016 | * this file to not set values for the floating and fixed string lengths, |
| 4017 | * and thus avoids the optimizer code in regexec.c that makes the invalid |
| 4018 | * assumption. Thus, there is no optimization based on string lengths for |
| 4019 | * EXACTFL nodes that contain these few folds, nor for non-UTF8-pattern |
| 4020 | * EXACTF and EXACTFAA nodes that contain the sharp s. (The reason the |
| 4021 | * assumption is wrong only in these cases is that all other non-UTF-8 |
| 4022 | * folds are 1-1; and, for UTF-8 patterns, we pre-fold all other folds to |
| 4023 | * their expanded versions. (Again, we can't prefold sharp s to 'ss' in |
| 4024 | * EXACTF nodes because we don't know at compile time if it actually |
| 4025 | * matches 'ss' or not. For EXACTF nodes it will match iff the target |
| 4026 | * string is in UTF-8. This is in contrast to EXACTFU nodes, where it |
| 4027 | * always matches; and EXACTFAA where it never does. In an EXACTFAA node |
| 4028 | * in a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the |
| 4029 | * problem; but in a non-UTF8 pattern, folding it to that above-Latin1 |
| 4030 | * string would require the pattern to be forced into UTF-8, the overhead |
| 4031 | * of which we want to avoid. Similarly the unfolded multi-char folds in |
| 4032 | * EXACTFL nodes will match iff the locale at the time of match is a UTF-8 |
| 4033 | * locale.) |
| 4034 | * |
| 4035 | * Similarly, the code that generates tries doesn't currently handle |
| 4036 | * not-already-folded multi-char folds, and it looks like a pain to change |
| 4037 | * that. Therefore, trie generation of EXACTFAA nodes with the sharp s |
| 4038 | * doesn't work. Instead, such an EXACTFAA is turned into a new regnode, |
| 4039 | * EXACTFAA_NO_TRIE, which the trie code knows not to handle. Most people |
| 4040 | * using /iaa matching will be doing so almost entirely with ASCII |
| 4041 | * strings, so this should rarely be encountered in practice */ |
| 4042 | |
| 4043 | STATIC U32 |
| 4044 | S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, |
| 4045 | UV *min_subtract, bool *unfolded_multi_char, |
| 4046 | U32 flags, regnode *val, U32 depth) |
| 4047 | { |
| 4048 | /* Merge several consecutive EXACTish nodes into one. */ |
| 4049 | |
| 4050 | regnode *n = regnext(scan); |
| 4051 | U32 stringok = 1; |
| 4052 | regnode *next = scan + NODE_SZ_STR(scan); |
| 4053 | U32 merged = 0; |
| 4054 | U32 stopnow = 0; |
| 4055 | #ifdef DEBUGGING |
| 4056 | regnode *stop = scan; |
| 4057 | GET_RE_DEBUG_FLAGS_DECL; |
| 4058 | #else |
| 4059 | PERL_UNUSED_ARG(depth); |
| 4060 | #endif |
| 4061 | |
| 4062 | PERL_ARGS_ASSERT_JOIN_EXACT; |
| 4063 | #ifndef EXPERIMENTAL_INPLACESCAN |
| 4064 | PERL_UNUSED_ARG(flags); |
| 4065 | PERL_UNUSED_ARG(val); |
| 4066 | #endif |
| 4067 | DEBUG_PEEP("join", scan, depth, 0); |
| 4068 | |
| 4069 | assert(PL_regkind[OP(scan)] == EXACT); |
| 4070 | |
| 4071 | /* Look through the subsequent nodes in the chain. Skip NOTHING, merge |
| 4072 | * EXACT ones that are mergeable to the current one. */ |
| 4073 | while ( n |
| 4074 | && ( PL_regkind[OP(n)] == NOTHING |
| 4075 | || (stringok && PL_regkind[OP(n)] == EXACT)) |
| 4076 | && NEXT_OFF(n) |
| 4077 | && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) |
| 4078 | { |
| 4079 | |
| 4080 | if (OP(n) == TAIL || n > next) |
| 4081 | stringok = 0; |
| 4082 | if (PL_regkind[OP(n)] == NOTHING) { |
| 4083 | DEBUG_PEEP("skip:", n, depth, 0); |
| 4084 | NEXT_OFF(scan) += NEXT_OFF(n); |
| 4085 | next = n + NODE_STEP_REGNODE; |
| 4086 | #ifdef DEBUGGING |
| 4087 | if (stringok) |
| 4088 | stop = n; |
| 4089 | #endif |
| 4090 | n = regnext(n); |
| 4091 | } |
| 4092 | else if (stringok) { |
| 4093 | const unsigned int oldl = STR_LEN(scan); |
| 4094 | regnode * const nnext = regnext(n); |
| 4095 | |
| 4096 | /* XXX I (khw) kind of doubt that this works on platforms (should |
| 4097 | * Perl ever run on one) where U8_MAX is above 255 because of lots |
| 4098 | * of other assumptions */ |
| 4099 | /* Don't join if the sum can't fit into a single node */ |
| 4100 | if (oldl + STR_LEN(n) > U8_MAX) |
| 4101 | break; |
| 4102 | |
| 4103 | /* Joining something that requires UTF-8 with something that |
| 4104 | * doesn't, means the result requires UTF-8. */ |
| 4105 | if (OP(scan) == EXACT && (OP(n) == EXACT_REQ8)) { |
| 4106 | OP(scan) = EXACT_REQ8; |
| 4107 | } |
| 4108 | else if (OP(scan) == EXACT_REQ8 && (OP(n) == EXACT)) { |
| 4109 | ; /* join is compatible, no need to change OP */ |
| 4110 | } |
| 4111 | else if ((OP(scan) == EXACTFU) && (OP(n) == EXACTFU_REQ8)) { |
| 4112 | OP(scan) = EXACTFU_REQ8; |
| 4113 | } |
| 4114 | else if ((OP(scan) == EXACTFU_REQ8) && (OP(n) == EXACTFU)) { |
| 4115 | ; /* join is compatible, no need to change OP */ |
| 4116 | } |
| 4117 | else if (OP(scan) == EXACTFU && OP(n) == EXACTFU) { |
| 4118 | ; /* join is compatible, no need to change OP */ |
| 4119 | } |
| 4120 | else if (OP(scan) == EXACTFU && OP(n) == EXACTFU_S_EDGE) { |
| 4121 | |
| 4122 | /* Under /di, temporary EXACTFU_S_EDGE nodes are generated, |
| 4123 | * which can join with EXACTFU ones. We check for this case |
| 4124 | * here. These need to be resolved to either EXACTFU or |
| 4125 | * EXACTF at joining time. They have nothing in them that |
| 4126 | * would forbid them from being the more desirable EXACTFU |
| 4127 | * nodes except that they begin and/or end with a single [Ss]. |
| 4128 | * The reason this is problematic is because they could be |
| 4129 | * joined in this loop with an adjacent node that ends and/or |
| 4130 | * begins with [Ss] which would then form the sequence 'ss', |
| 4131 | * which matches differently under /di than /ui, in which case |
| 4132 | * EXACTFU can't be used. If the 'ss' sequence doesn't get |
| 4133 | * formed, the nodes get absorbed into any adjacent EXACTFU |
| 4134 | * node. And if the only adjacent node is EXACTF, they get |
| 4135 | * absorbed into that, under the theory that a longer node is |
| 4136 | * better than two shorter ones, even if one is EXACTFU. Note |
| 4137 | * that EXACTFU_REQ8 is generated only for UTF-8 patterns, |
| 4138 | * and the EXACTFU_S_EDGE ones only for non-UTF-8. */ |
| 4139 | |
| 4140 | if (STRING(n)[STR_LEN(n)-1] == 's') { |
| 4141 | |
| 4142 | /* Here the joined node would end with 's'. If the node |
| 4143 | * following the combination is an EXACTF one, it's better to |
| 4144 | * join this trailing edge 's' node with that one, leaving the |
| 4145 | * current one in 'scan' be the more desirable EXACTFU */ |
| 4146 | if (OP(nnext) == EXACTF) { |
| 4147 | break; |
| 4148 | } |
| 4149 | |
| 4150 | OP(scan) = EXACTFU_S_EDGE; |
| 4151 | |
| 4152 | } /* Otherwise, the beginning 's' of the 2nd node just |
| 4153 | becomes an interior 's' in 'scan' */ |
| 4154 | } |
| 4155 | else if (OP(scan) == EXACTF && OP(n) == EXACTF) { |
| 4156 | ; /* join is compatible, no need to change OP */ |
| 4157 | } |
| 4158 | else if (OP(scan) == EXACTF && OP(n) == EXACTFU_S_EDGE) { |
| 4159 | |
| 4160 | /* EXACTF nodes are compatible for joining with EXACTFU_S_EDGE |
| 4161 | * nodes. But the latter nodes can be also joined with EXACTFU |
| 4162 | * ones, and that is a better outcome, so if the node following |
| 4163 | * 'n' is EXACTFU, quit now so that those two can be joined |
| 4164 | * later */ |
| 4165 | if (OP(nnext) == EXACTFU) { |
| 4166 | break; |
| 4167 | } |
| 4168 | |
| 4169 | /* The join is compatible, and the combined node will be |
| 4170 | * EXACTF. (These don't care if they begin or end with 's' */ |
| 4171 | } |
| 4172 | else if (OP(scan) == EXACTFU_S_EDGE && OP(n) == EXACTFU_S_EDGE) { |
| 4173 | if ( STRING(scan)[STR_LEN(scan)-1] == 's' |
| 4174 | && STRING(n)[0] == 's') |
| 4175 | { |
| 4176 | /* When combined, we have the sequence 'ss', which means we |
| 4177 | * have to remain /di */ |
| 4178 | OP(scan) = EXACTF; |
| 4179 | } |
| 4180 | } |
| 4181 | else if (OP(scan) == EXACTFU_S_EDGE && OP(n) == EXACTFU) { |
| 4182 | if (STRING(n)[0] == 's') { |
| 4183 | ; /* Here the join is compatible and the combined node |
| 4184 | starts with 's', no need to change OP */ |
| 4185 | } |
| 4186 | else { /* Now the trailing 's' is in the interior */ |
| 4187 | OP(scan) = EXACTFU; |
| 4188 | } |
| 4189 | } |
| 4190 | else if (OP(scan) == EXACTFU_S_EDGE && OP(n) == EXACTF) { |
| 4191 | |
| 4192 | /* The join is compatible, and the combined node will be |
| 4193 | * EXACTF. (These don't care if they begin or end with 's' */ |
| 4194 | OP(scan) = EXACTF; |
| 4195 | } |
| 4196 | else if (OP(scan) != OP(n)) { |
| 4197 | |
| 4198 | /* The only other compatible joinings are the same node type */ |
| 4199 | break; |
| 4200 | } |
| 4201 | |
| 4202 | DEBUG_PEEP("merg", n, depth, 0); |
| 4203 | merged++; |
| 4204 | |
| 4205 | NEXT_OFF(scan) += NEXT_OFF(n); |
| 4206 | assert( ( STR_LEN(scan) + STR_LEN(n) ) < 256 ); |
| 4207 | setSTR_LEN(scan, (U8)(STR_LEN(scan) + STR_LEN(n))); |
| 4208 | next = n + NODE_SZ_STR(n); |
| 4209 | /* Now we can overwrite *n : */ |
| 4210 | Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char); |
| 4211 | #ifdef DEBUGGING |
| 4212 | stop = next - 1; |
| 4213 | #endif |
| 4214 | n = nnext; |
| 4215 | if (stopnow) break; |
| 4216 | } |
| 4217 | |
| 4218 | #ifdef EXPERIMENTAL_INPLACESCAN |
| 4219 | if (flags && !NEXT_OFF(n)) { |
| 4220 | DEBUG_PEEP("atch", val, depth, 0); |
| 4221 | if (reg_off_by_arg[OP(n)]) { |
| 4222 | ARG_SET(n, val - n); |
| 4223 | } |
| 4224 | else { |
| 4225 | NEXT_OFF(n) = val - n; |
| 4226 | } |
| 4227 | stopnow = 1; |
| 4228 | } |
| 4229 | #endif |
| 4230 | } |
| 4231 | |
| 4232 | /* This temporary node can now be turned into EXACTFU, and must, as |
| 4233 | * regexec.c doesn't handle it */ |
| 4234 | if (OP(scan) == EXACTFU_S_EDGE) { |
| 4235 | OP(scan) = EXACTFU; |
| 4236 | } |
| 4237 | |
| 4238 | *min_subtract = 0; |
| 4239 | *unfolded_multi_char = FALSE; |
| 4240 | |
| 4241 | /* Here, all the adjacent mergeable EXACTish nodes have been merged. We |
| 4242 | * can now analyze for sequences of problematic code points. (Prior to |
| 4243 | * this final joining, sequences could have been split over boundaries, and |
| 4244 | * hence missed). The sequences only happen in folding, hence for any |
| 4245 | * non-EXACT EXACTish node */ |
| 4246 | if (OP(scan) != EXACT && OP(scan) != EXACT_REQ8 && OP(scan) != EXACTL) { |
| 4247 | U8* s0 = (U8*) STRING(scan); |
| 4248 | U8* s = s0; |
| 4249 | U8* s_end = s0 + STR_LEN(scan); |
| 4250 | |
| 4251 | int total_count_delta = 0; /* Total delta number of characters that |
| 4252 | multi-char folds expand to */ |
| 4253 | |
| 4254 | /* One pass is made over the node's string looking for all the |
| 4255 | * possibilities. To avoid some tests in the loop, there are two main |
| 4256 | * cases, for UTF-8 patterns (which can't have EXACTF nodes) and |
| 4257 | * non-UTF-8 */ |
| 4258 | if (UTF) { |
| 4259 | U8* folded = NULL; |
| 4260 | |
| 4261 | if (OP(scan) == EXACTFL) { |
| 4262 | U8 *d; |
| 4263 | |
| 4264 | /* An EXACTFL node would already have been changed to another |
| 4265 | * node type unless there is at least one character in it that |
| 4266 | * is problematic; likely a character whose fold definition |
| 4267 | * won't be known until runtime, and so has yet to be folded. |
| 4268 | * For all but the UTF-8 locale, folds are 1-1 in length, but |
| 4269 | * to handle the UTF-8 case, we need to create a temporary |
| 4270 | * folded copy using UTF-8 locale rules in order to analyze it. |
| 4271 | * This is because our macros that look to see if a sequence is |
| 4272 | * a multi-char fold assume everything is folded (otherwise the |
| 4273 | * tests in those macros would be too complicated and slow). |
| 4274 | * Note that here, the non-problematic folds will have already |
| 4275 | * been done, so we can just copy such characters. We actually |
| 4276 | * don't completely fold the EXACTFL string. We skip the |
| 4277 | * unfolded multi-char folds, as that would just create work |
| 4278 | * below to figure out the size they already are */ |
| 4279 | |
| 4280 | Newx(folded, UTF8_MAX_FOLD_CHAR_EXPAND * STR_LEN(scan) + 1, U8); |
| 4281 | d = folded; |
| 4282 | while (s < s_end) { |
| 4283 | STRLEN s_len = UTF8SKIP(s); |
| 4284 | if (! is_PROBLEMATIC_LOCALE_FOLD_utf8(s)) { |
| 4285 | Copy(s, d, s_len, U8); |
| 4286 | d += s_len; |
| 4287 | } |
| 4288 | else if (is_FOLDS_TO_MULTI_utf8(s)) { |
| 4289 | *unfolded_multi_char = TRUE; |
| 4290 | Copy(s, d, s_len, U8); |
| 4291 | d += s_len; |
| 4292 | } |
| 4293 | else if (isASCII(*s)) { |
| 4294 | *(d++) = toFOLD(*s); |
| 4295 | } |
| 4296 | else { |
| 4297 | STRLEN len; |
| 4298 | _toFOLD_utf8_flags(s, s_end, d, &len, FOLD_FLAGS_FULL); |
| 4299 | d += len; |
| 4300 | } |
| 4301 | s += s_len; |
| 4302 | } |
| 4303 | |
| 4304 | /* Point the remainder of the routine to look at our temporary |
| 4305 | * folded copy */ |
| 4306 | s = folded; |
| 4307 | s_end = d; |
| 4308 | } /* End of creating folded copy of EXACTFL string */ |
| 4309 | |
| 4310 | /* Examine the string for a multi-character fold sequence. UTF-8 |
| 4311 | * patterns have all characters pre-folded by the time this code is |
| 4312 | * executed */ |
| 4313 | while (s < s_end - 1) /* Can stop 1 before the end, as minimum |
| 4314 | length sequence we are looking for is 2 */ |
| 4315 | { |
| 4316 | int count = 0; /* How many characters in a multi-char fold */ |
| 4317 | int len = is_MULTI_CHAR_FOLD_utf8_safe(s, s_end); |
| 4318 | if (! len) { /* Not a multi-char fold: get next char */ |
| 4319 | s += UTF8SKIP(s); |
| 4320 | continue; |
| 4321 | } |
| 4322 | |
| 4323 | { /* Here is a generic multi-char fold. */ |
| 4324 | U8* multi_end = s + len; |
| 4325 | |
| 4326 | /* Count how many characters are in it. In the case of |
| 4327 | * /aa, no folds which contain ASCII code points are |
| 4328 | * allowed, so check for those, and skip if found. */ |
| 4329 | if (OP(scan) != EXACTFAA && OP(scan) != EXACTFAA_NO_TRIE) { |
| 4330 | count = utf8_length(s, multi_end); |
| 4331 | s = multi_end; |
| 4332 | } |
| 4333 | else { |
| 4334 | while (s < multi_end) { |
| 4335 | if (isASCII(*s)) { |
| 4336 | s++; |
| 4337 | goto next_iteration; |
| 4338 | } |
| 4339 | else { |
| 4340 | s += UTF8SKIP(s); |
| 4341 | } |
| 4342 | count++; |
| 4343 | } |
| 4344 | } |
| 4345 | } |
| 4346 | |
| 4347 | /* The delta is how long the sequence is minus 1 (1 is how long |
| 4348 | * the character that folds to the sequence is) */ |
| 4349 | total_count_delta += count - 1; |
| 4350 | next_iteration: ; |
| 4351 | } |
| 4352 | |
| 4353 | /* We created a temporary folded copy of the string in EXACTFL |
| 4354 | * nodes. Therefore we need to be sure it doesn't go below zero, |
| 4355 | * as the real string could be shorter */ |
| 4356 | if (OP(scan) == EXACTFL) { |
| 4357 | int total_chars = utf8_length((U8*) STRING(scan), |
| 4358 | (U8*) STRING(scan) + STR_LEN(scan)); |
| 4359 | if (total_count_delta > total_chars) { |
| 4360 | total_count_delta = total_chars; |
| 4361 | } |
| 4362 | } |
| 4363 | |
| 4364 | *min_subtract += total_count_delta; |
| 4365 | Safefree(folded); |
| 4366 | } |
| 4367 | else if (OP(scan) == EXACTFAA) { |
| 4368 | |
| 4369 | /* Non-UTF-8 pattern, EXACTFAA node. There can't be a multi-char |
| 4370 | * fold to the ASCII range (and there are no existing ones in the |
| 4371 | * upper latin1 range). But, as outlined in the comments preceding |
| 4372 | * this function, we need to flag any occurrences of the sharp s. |
| 4373 | * This character forbids trie formation (because of added |
| 4374 | * complexity) */ |
| 4375 | #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ |
| 4376 | || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ |
| 4377 | || UNICODE_DOT_DOT_VERSION > 0) |
| 4378 | while (s < s_end) { |
| 4379 | if (*s == LATIN_SMALL_LETTER_SHARP_S) { |
| 4380 | OP(scan) = EXACTFAA_NO_TRIE; |
| 4381 | *unfolded_multi_char = TRUE; |
| 4382 | break; |
| 4383 | } |
| 4384 | s++; |
| 4385 | } |
| 4386 | } |
| 4387 | else { |
| 4388 | |
| 4389 | /* Non-UTF-8 pattern, not EXACTFAA node. Look for the multi-char |
| 4390 | * folds that are all Latin1. As explained in the comments |
| 4391 | * preceding this function, we look also for the sharp s in EXACTF |
| 4392 | * and EXACTFL nodes; it can be in the final position. Otherwise |
| 4393 | * we can stop looking 1 byte earlier because have to find at least |
| 4394 | * two characters for a multi-fold */ |
| 4395 | const U8* upper = (OP(scan) == EXACTF || OP(scan) == EXACTFL) |
| 4396 | ? s_end |
| 4397 | : s_end -1; |
| 4398 | |
| 4399 | while (s < upper) { |
| 4400 | int len = is_MULTI_CHAR_FOLD_latin1_safe(s, s_end); |
| 4401 | if (! len) { /* Not a multi-char fold. */ |
| 4402 | if (*s == LATIN_SMALL_LETTER_SHARP_S |
| 4403 | && (OP(scan) == EXACTF || OP(scan) == EXACTFL)) |
| 4404 | { |
| 4405 | *unfolded_multi_char = TRUE; |
| 4406 | } |
| 4407 | s++; |
| 4408 | continue; |
| 4409 | } |
| 4410 | |
| 4411 | if (len == 2 |
| 4412 | && isALPHA_FOLD_EQ(*s, 's') |
| 4413 | && isALPHA_FOLD_EQ(*(s+1), 's')) |
| 4414 | { |
| 4415 | |
| 4416 | /* EXACTF nodes need to know that the minimum length |
| 4417 | * changed so that a sharp s in the string can match this |
| 4418 | * ss in the pattern, but they remain EXACTF nodes, as they |
| 4419 | * won't match this unless the target string is is UTF-8, |
| 4420 | * which we don't know until runtime. EXACTFL nodes can't |
| 4421 | * transform into EXACTFU nodes */ |
| 4422 | if (OP(scan) != EXACTF && OP(scan) != EXACTFL) { |
| 4423 | OP(scan) = EXACTFUP; |
| 4424 | } |
| 4425 | } |
| 4426 | |
| 4427 | *min_subtract += len - 1; |
| 4428 | s += len; |
| 4429 | } |
| 4430 | #endif |
| 4431 | } |
| 4432 | } |
| 4433 | |
| 4434 | #ifdef DEBUGGING |
| 4435 | /* Allow dumping but overwriting the collection of skipped |
| 4436 | * ops and/or strings with fake optimized ops */ |
| 4437 | n = scan + NODE_SZ_STR(scan); |
| 4438 | while (n <= stop) { |
| 4439 | OP(n) = OPTIMIZED; |
| 4440 | FLAGS(n) = 0; |
| 4441 | NEXT_OFF(n) = 0; |
| 4442 | n++; |
| 4443 | } |
| 4444 | #endif |
| 4445 | DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl", scan, depth, 0);}); |
| 4446 | return stopnow; |
| 4447 | } |
| 4448 | |
| 4449 | /* REx optimizer. Converts nodes into quicker variants "in place". |
| 4450 | Finds fixed substrings. */ |
| 4451 | |
| 4452 | /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set |
| 4453 | to the position after last scanned or to NULL. */ |
| 4454 | |
| 4455 | #define INIT_AND_WITHP \ |
| 4456 | assert(!and_withp); \ |
| 4457 | Newx(and_withp, 1, regnode_ssc); \ |
| 4458 | SAVEFREEPV(and_withp) |
| 4459 | |
| 4460 | |
| 4461 | static void |
| 4462 | S_unwind_scan_frames(pTHX_ const void *p) |
| 4463 | { |
| 4464 | scan_frame *f= (scan_frame *)p; |
| 4465 | do { |
| 4466 | scan_frame *n= f->next_frame; |
| 4467 | Safefree(f); |
| 4468 | f= n; |
| 4469 | } while (f); |
| 4470 | } |
| 4471 | |
| 4472 | /* the return from this sub is the minimum length that could possibly match */ |
| 4473 | STATIC SSize_t |
| 4474 | S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, |
| 4475 | SSize_t *minlenp, SSize_t *deltap, |
| 4476 | regnode *last, |
| 4477 | scan_data_t *data, |
| 4478 | I32 stopparen, |
| 4479 | U32 recursed_depth, |
| 4480 | regnode_ssc *and_withp, |
| 4481 | U32 flags, U32 depth) |
| 4482 | /* scanp: Start here (read-write). */ |
| 4483 | /* deltap: Write maxlen-minlen here. */ |
| 4484 | /* last: Stop before this one. */ |
| 4485 | /* data: string data about the pattern */ |
| 4486 | /* stopparen: treat close N as END */ |
| 4487 | /* recursed: which subroutines have we recursed into */ |
| 4488 | /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */ |
| 4489 | { |
| 4490 | dVAR; |
| 4491 | SSize_t final_minlen; |
| 4492 | /* There must be at least this number of characters to match */ |
| 4493 | SSize_t min = 0; |
| 4494 | I32 pars = 0, code; |
| 4495 | regnode *scan = *scanp, *next; |
| 4496 | SSize_t delta = 0; |
| 4497 | int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF); |
| 4498 | int is_inf_internal = 0; /* The studied chunk is infinite */ |
| 4499 | I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0; |
| 4500 | scan_data_t data_fake; |
| 4501 | SV *re_trie_maxbuff = NULL; |
| 4502 | regnode *first_non_open = scan; |
| 4503 | SSize_t stopmin = OPTIMIZE_INFTY; |
| 4504 | scan_frame *frame = NULL; |
| 4505 | GET_RE_DEBUG_FLAGS_DECL; |
| 4506 | |
| 4507 | PERL_ARGS_ASSERT_STUDY_CHUNK; |
| 4508 | RExC_study_started= 1; |
| 4509 | |
| 4510 | Zero(&data_fake, 1, scan_data_t); |
| 4511 | |
| 4512 | if ( depth == 0 ) { |
| 4513 | while (first_non_open && OP(first_non_open) == OPEN) |
| 4514 | first_non_open=regnext(first_non_open); |
| 4515 | } |
| 4516 | |
| 4517 | |
| 4518 | fake_study_recurse: |
| 4519 | DEBUG_r( |
| 4520 | RExC_study_chunk_recursed_count++; |
| 4521 | ); |
| 4522 | DEBUG_OPTIMISE_MORE_r( |
| 4523 | { |
| 4524 | Perl_re_indentf( aTHX_ "study_chunk stopparen=%ld recursed_count=%lu depth=%lu recursed_depth=%lu scan=%p last=%p", |
| 4525 | depth, (long)stopparen, |
| 4526 | (unsigned long)RExC_study_chunk_recursed_count, |
| 4527 | (unsigned long)depth, (unsigned long)recursed_depth, |
| 4528 | scan, |
| 4529 | last); |
| 4530 | if (recursed_depth) { |
| 4531 | U32 i; |
| 4532 | U32 j; |
| 4533 | for ( j = 0 ; j < recursed_depth ; j++ ) { |
| 4534 | for ( i = 0 ; i < (U32)RExC_total_parens ; i++ ) { |
| 4535 | if (PAREN_TEST(j, i) && (!j || !PAREN_TEST(j - 1, i))) { |
| 4536 | Perl_re_printf( aTHX_ " %d",(int)i); |
| 4537 | break; |
| 4538 | } |
| 4539 | } |
| 4540 | if ( j + 1 < recursed_depth ) { |
| 4541 | Perl_re_printf( aTHX_ ","); |
| 4542 | } |
| 4543 | } |
| 4544 | } |
| 4545 | Perl_re_printf( aTHX_ "\n"); |
| 4546 | } |
| 4547 | ); |
| 4548 | while ( scan && OP(scan) != END && scan < last ){ |
| 4549 | UV min_subtract = 0; /* How mmany chars to subtract from the minimum |
| 4550 | node length to get a real minimum (because |
| 4551 | the folded version may be shorter) */ |
| 4552 | bool unfolded_multi_char = FALSE; |
| 4553 | /* Peephole optimizer: */ |
| 4554 | DEBUG_STUDYDATA("Peep", data, depth, is_inf); |
| 4555 | DEBUG_PEEP("Peep", scan, depth, flags); |
| 4556 | |
| 4557 | |
| 4558 | /* The reason we do this here is that we need to deal with things like |
| 4559 | * /(?:f)(?:o)(?:o)/ which cant be dealt with by the normal EXACT |
| 4560 | * parsing code, as each (?:..) is handled by a different invocation of |
| 4561 | * reg() -- Yves |
| 4562 | */ |
| 4563 | if (PL_regkind[OP(scan)] == EXACT && OP(scan) != LEXACT |
| 4564 | && OP(scan) != LEXACT_REQ8) |
| 4565 | join_exact(pRExC_state, scan, &min_subtract, &unfolded_multi_char, |
| 4566 | 0, NULL, depth + 1); |
| 4567 | |
| 4568 | /* Follow the next-chain of the current node and optimize |
| 4569 | away all the NOTHINGs from it. */ |
| 4570 | if (OP(scan) != CURLYX) { |
| 4571 | const int max = (reg_off_by_arg[OP(scan)] |
| 4572 | ? I32_MAX |
| 4573 | /* I32 may be smaller than U16 on CRAYs! */ |
| 4574 | : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX)); |
| 4575 | int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan)); |
| 4576 | int noff; |
| 4577 | regnode *n = scan; |
| 4578 | |
| 4579 | /* Skip NOTHING and LONGJMP. */ |
| 4580 | while ( (n = regnext(n)) |
| 4581 | && ( (PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n))) |
| 4582 | || ((OP(n) == LONGJMP) && (noff = ARG(n)))) |
| 4583 | && off + noff < max) |
| 4584 | off += noff; |
| 4585 | if (reg_off_by_arg[OP(scan)]) |
| 4586 | ARG(scan) = off; |
| 4587 | else |
| 4588 | NEXT_OFF(scan) = off; |
| 4589 | } |
| 4590 | |
| 4591 | /* The principal pseudo-switch. Cannot be a switch, since we look into |
| 4592 | * several different things. */ |
| 4593 | if ( OP(scan) == DEFINEP ) { |
| 4594 | SSize_t minlen = 0; |
| 4595 | SSize_t deltanext = 0; |
| 4596 | SSize_t fake_last_close = 0; |
| 4597 | I32 f = SCF_IN_DEFINE; |
| 4598 | |
| 4599 | StructCopy(&zero_scan_data, &data_fake, scan_data_t); |
| 4600 | scan = regnext(scan); |
| 4601 | assert( OP(scan) == IFTHEN ); |
| 4602 | DEBUG_PEEP("expect IFTHEN", scan, depth, flags); |
| 4603 | |
| 4604 | data_fake.last_closep= &fake_last_close; |
| 4605 | minlen = *minlenp; |
| 4606 | next = regnext(scan); |
| 4607 | scan = NEXTOPER(NEXTOPER(scan)); |
| 4608 | DEBUG_PEEP("scan", scan, depth, flags); |
| 4609 | DEBUG_PEEP("next", next, depth, flags); |
| 4610 | |
| 4611 | /* we suppose the run is continuous, last=next... |
| 4612 | * NOTE we dont use the return here! */ |
| 4613 | /* DEFINEP study_chunk() recursion */ |
| 4614 | (void)study_chunk(pRExC_state, &scan, &minlen, |
| 4615 | &deltanext, next, &data_fake, stopparen, |
| 4616 | recursed_depth, NULL, f, depth+1); |
| 4617 | |
| 4618 | scan = next; |
| 4619 | } else |
| 4620 | if ( |
| 4621 | OP(scan) == BRANCH || |
| 4622 | OP(scan) == BRANCHJ || |
| 4623 | OP(scan) == IFTHEN |
| 4624 | ) { |
| 4625 | next = regnext(scan); |
| 4626 | code = OP(scan); |
| 4627 | |
| 4628 | /* The op(next)==code check below is to see if we |
| 4629 | * have "BRANCH-BRANCH", "BRANCHJ-BRANCHJ", "IFTHEN-IFTHEN" |
| 4630 | * IFTHEN is special as it might not appear in pairs. |
| 4631 | * Not sure whether BRANCH-BRANCHJ is possible, regardless |
| 4632 | * we dont handle it cleanly. */ |
| 4633 | if (OP(next) == code || code == IFTHEN) { |
| 4634 | /* NOTE - There is similar code to this block below for |
| 4635 | * handling TRIE nodes on a re-study. If you change stuff here |
| 4636 | * check there too. */ |
| 4637 | SSize_t max1 = 0, min1 = OPTIMIZE_INFTY, num = 0; |
| 4638 | regnode_ssc accum; |
| 4639 | regnode * const startbranch=scan; |
| 4640 | |
| 4641 | if (flags & SCF_DO_SUBSTR) { |
| 4642 | /* Cannot merge strings after this. */ |
| 4643 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 4644 | } |
| 4645 | |
| 4646 | if (flags & SCF_DO_STCLASS) |
| 4647 | ssc_init_zero(pRExC_state, &accum); |
| 4648 | |
| 4649 | while (OP(scan) == code) { |
| 4650 | SSize_t deltanext, minnext, fake; |
| 4651 | I32 f = 0; |
| 4652 | regnode_ssc this_class; |
| 4653 | |
| 4654 | DEBUG_PEEP("Branch", scan, depth, flags); |
| 4655 | |
| 4656 | num++; |
| 4657 | StructCopy(&zero_scan_data, &data_fake, scan_data_t); |
| 4658 | if (data) { |
| 4659 | data_fake.whilem_c = data->whilem_c; |
| 4660 | data_fake.last_closep = data->last_closep; |
| 4661 | } |
| 4662 | else |
| 4663 | data_fake.last_closep = &fake; |
| 4664 | |
| 4665 | data_fake.pos_delta = delta; |
| 4666 | next = regnext(scan); |
| 4667 | |
| 4668 | scan = NEXTOPER(scan); /* everything */ |
| 4669 | if (code != BRANCH) /* everything but BRANCH */ |
| 4670 | scan = NEXTOPER(scan); |
| 4671 | |
| 4672 | if (flags & SCF_DO_STCLASS) { |
| 4673 | ssc_init(pRExC_state, &this_class); |
| 4674 | data_fake.start_class = &this_class; |
| 4675 | f = SCF_DO_STCLASS_AND; |
| 4676 | } |
| 4677 | if (flags & SCF_WHILEM_VISITED_POS) |
| 4678 | f |= SCF_WHILEM_VISITED_POS; |
| 4679 | |
| 4680 | /* we suppose the run is continuous, last=next...*/ |
| 4681 | /* recurse study_chunk() for each BRANCH in an alternation */ |
| 4682 | minnext = study_chunk(pRExC_state, &scan, minlenp, |
| 4683 | &deltanext, next, &data_fake, stopparen, |
| 4684 | recursed_depth, NULL, f, depth+1); |
| 4685 | |
| 4686 | if (min1 > minnext) |
| 4687 | min1 = minnext; |
| 4688 | if (deltanext == OPTIMIZE_INFTY) { |
| 4689 | is_inf = is_inf_internal = 1; |
| 4690 | max1 = OPTIMIZE_INFTY; |
| 4691 | } else if (max1 < minnext + deltanext) |
| 4692 | max1 = minnext + deltanext; |
| 4693 | scan = next; |
| 4694 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 4695 | pars++; |
| 4696 | if (data_fake.flags & SCF_SEEN_ACCEPT) { |
| 4697 | if ( stopmin > minnext) |
| 4698 | stopmin = min + min1; |
| 4699 | flags &= ~SCF_DO_SUBSTR; |
| 4700 | if (data) |
| 4701 | data->flags |= SCF_SEEN_ACCEPT; |
| 4702 | } |
| 4703 | if (data) { |
| 4704 | if (data_fake.flags & SF_HAS_EVAL) |
| 4705 | data->flags |= SF_HAS_EVAL; |
| 4706 | data->whilem_c = data_fake.whilem_c; |
| 4707 | } |
| 4708 | if (flags & SCF_DO_STCLASS) |
| 4709 | ssc_or(pRExC_state, &accum, (regnode_charclass*)&this_class); |
| 4710 | } |
| 4711 | if (code == IFTHEN && num < 2) /* Empty ELSE branch */ |
| 4712 | min1 = 0; |
| 4713 | if (flags & SCF_DO_SUBSTR) { |
| 4714 | data->pos_min += min1; |
| 4715 | if (data->pos_delta >= OPTIMIZE_INFTY - (max1 - min1)) |
| 4716 | data->pos_delta = OPTIMIZE_INFTY; |
| 4717 | else |
| 4718 | data->pos_delta += max1 - min1; |
| 4719 | if (max1 != min1 || is_inf) |
| 4720 | data->cur_is_floating = 1; |
| 4721 | } |
| 4722 | min += min1; |
| 4723 | if (delta == OPTIMIZE_INFTY |
| 4724 | || OPTIMIZE_INFTY - delta - (max1 - min1) < 0) |
| 4725 | delta = OPTIMIZE_INFTY; |
| 4726 | else |
| 4727 | delta += max1 - min1; |
| 4728 | if (flags & SCF_DO_STCLASS_OR) { |
| 4729 | ssc_or(pRExC_state, data->start_class, (regnode_charclass*) &accum); |
| 4730 | if (min1) { |
| 4731 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 4732 | flags &= ~SCF_DO_STCLASS; |
| 4733 | } |
| 4734 | } |
| 4735 | else if (flags & SCF_DO_STCLASS_AND) { |
| 4736 | if (min1) { |
| 4737 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &accum); |
| 4738 | flags &= ~SCF_DO_STCLASS; |
| 4739 | } |
| 4740 | else { |
| 4741 | /* Switch to OR mode: cache the old value of |
| 4742 | * data->start_class */ |
| 4743 | INIT_AND_WITHP; |
| 4744 | StructCopy(data->start_class, and_withp, regnode_ssc); |
| 4745 | flags &= ~SCF_DO_STCLASS_AND; |
| 4746 | StructCopy(&accum, data->start_class, regnode_ssc); |
| 4747 | flags |= SCF_DO_STCLASS_OR; |
| 4748 | } |
| 4749 | } |
| 4750 | |
| 4751 | if (PERL_ENABLE_TRIE_OPTIMISATION && |
| 4752 | OP( startbranch ) == BRANCH ) |
| 4753 | { |
| 4754 | /* demq. |
| 4755 | |
| 4756 | Assuming this was/is a branch we are dealing with: 'scan' |
| 4757 | now points at the item that follows the branch sequence, |
| 4758 | whatever it is. We now start at the beginning of the |
| 4759 | sequence and look for subsequences of |
| 4760 | |
| 4761 | BRANCH->EXACT=>x1 |
| 4762 | BRANCH->EXACT=>x2 |
| 4763 | tail |
| 4764 | |
| 4765 | which would be constructed from a pattern like |
| 4766 | /A|LIST|OF|WORDS/ |
| 4767 | |
| 4768 | If we can find such a subsequence we need to turn the first |
| 4769 | element into a trie and then add the subsequent branch exact |
| 4770 | strings to the trie. |
| 4771 | |
| 4772 | We have two cases |
| 4773 | |
| 4774 | 1. patterns where the whole set of branches can be |
| 4775 | converted. |
| 4776 | |
| 4777 | 2. patterns where only a subset can be converted. |
| 4778 | |
| 4779 | In case 1 we can replace the whole set with a single regop |
| 4780 | for the trie. In case 2 we need to keep the start and end |
| 4781 | branches so |
| 4782 | |
| 4783 | 'BRANCH EXACT; BRANCH EXACT; BRANCH X' |
| 4784 | becomes BRANCH TRIE; BRANCH X; |
| 4785 | |
| 4786 | There is an additional case, that being where there is a |
| 4787 | common prefix, which gets split out into an EXACT like node |
| 4788 | preceding the TRIE node. |
| 4789 | |
| 4790 | If x(1..n)==tail then we can do a simple trie, if not we make |
| 4791 | a "jump" trie, such that when we match the appropriate word |
| 4792 | we "jump" to the appropriate tail node. Essentially we turn |
| 4793 | a nested if into a case structure of sorts. |
| 4794 | |
| 4795 | */ |
| 4796 | |
| 4797 | int made=0; |
| 4798 | if (!re_trie_maxbuff) { |
| 4799 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); |
| 4800 | if (!SvIOK(re_trie_maxbuff)) |
| 4801 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); |
| 4802 | } |
| 4803 | if ( SvIV(re_trie_maxbuff)>=0 ) { |
| 4804 | regnode *cur; |
| 4805 | regnode *first = (regnode *)NULL; |
| 4806 | regnode *prev = (regnode *)NULL; |
| 4807 | regnode *tail = scan; |
| 4808 | U8 trietype = 0; |
| 4809 | U32 count=0; |
| 4810 | |
| 4811 | /* var tail is used because there may be a TAIL |
| 4812 | regop in the way. Ie, the exacts will point to the |
| 4813 | thing following the TAIL, but the last branch will |
| 4814 | point at the TAIL. So we advance tail. If we |
| 4815 | have nested (?:) we may have to move through several |
| 4816 | tails. |
| 4817 | */ |
| 4818 | |
| 4819 | while ( OP( tail ) == TAIL ) { |
| 4820 | /* this is the TAIL generated by (?:) */ |
| 4821 | tail = regnext( tail ); |
| 4822 | } |
| 4823 | |
| 4824 | |
| 4825 | DEBUG_TRIE_COMPILE_r({ |
| 4826 | regprop(RExC_rx, RExC_mysv, tail, NULL, pRExC_state); |
| 4827 | Perl_re_indentf( aTHX_ "%s %" UVuf ":%s\n", |
| 4828 | depth+1, |
| 4829 | "Looking for TRIE'able sequences. Tail node is ", |
| 4830 | (UV) REGNODE_OFFSET(tail), |
| 4831 | SvPV_nolen_const( RExC_mysv ) |
| 4832 | ); |
| 4833 | }); |
| 4834 | |
| 4835 | /* |
| 4836 | |
| 4837 | Step through the branches |
| 4838 | cur represents each branch, |
| 4839 | noper is the first thing to be matched as part |
| 4840 | of that branch |
| 4841 | noper_next is the regnext() of that node. |
| 4842 | |
| 4843 | We normally handle a case like this |
| 4844 | /FOO[xyz]|BAR[pqr]/ via a "jump trie" but we also |
| 4845 | support building with NOJUMPTRIE, which restricts |
| 4846 | the trie logic to structures like /FOO|BAR/. |
| 4847 | |
| 4848 | If noper is a trieable nodetype then the branch is |
| 4849 | a possible optimization target. If we are building |
| 4850 | under NOJUMPTRIE then we require that noper_next is |
| 4851 | the same as scan (our current position in the regex |
| 4852 | program). |
| 4853 | |
| 4854 | Once we have two or more consecutive such branches |
| 4855 | we can create a trie of the EXACT's contents and |
| 4856 | stitch it in place into the program. |
| 4857 | |
| 4858 | If the sequence represents all of the branches in |
| 4859 | the alternation we replace the entire thing with a |
| 4860 | single TRIE node. |
| 4861 | |
| 4862 | Otherwise when it is a subsequence we need to |
| 4863 | stitch it in place and replace only the relevant |
| 4864 | branches. This means the first branch has to remain |
| 4865 | as it is used by the alternation logic, and its |
| 4866 | next pointer, and needs to be repointed at the item |
| 4867 | on the branch chain following the last branch we |
| 4868 | have optimized away. |
| 4869 | |
| 4870 | This could be either a BRANCH, in which case the |
| 4871 | subsequence is internal, or it could be the item |
| 4872 | following the branch sequence in which case the |
| 4873 | subsequence is at the end (which does not |
| 4874 | necessarily mean the first node is the start of the |
| 4875 | alternation). |
| 4876 | |
| 4877 | TRIE_TYPE(X) is a define which maps the optype to a |
| 4878 | trietype. |
| 4879 | |
| 4880 | optype | trietype |
| 4881 | ----------------+----------- |
| 4882 | NOTHING | NOTHING |
| 4883 | EXACT | EXACT |
| 4884 | EXACT_REQ8 | EXACT |
| 4885 | EXACTFU | EXACTFU |
| 4886 | EXACTFU_REQ8 | EXACTFU |
| 4887 | EXACTFUP | EXACTFU |
| 4888 | EXACTFAA | EXACTFAA |
| 4889 | EXACTL | EXACTL |
| 4890 | EXACTFLU8 | EXACTFLU8 |
| 4891 | |
| 4892 | |
| 4893 | */ |
| 4894 | #define TRIE_TYPE(X) ( ( NOTHING == (X) ) \ |
| 4895 | ? NOTHING \ |
| 4896 | : ( EXACT == (X) || EXACT_REQ8 == (X) ) \ |
| 4897 | ? EXACT \ |
| 4898 | : ( EXACTFU == (X) \ |
| 4899 | || EXACTFU_REQ8 == (X) \ |
| 4900 | || EXACTFUP == (X) ) \ |
| 4901 | ? EXACTFU \ |
| 4902 | : ( EXACTFAA == (X) ) \ |
| 4903 | ? EXACTFAA \ |
| 4904 | : ( EXACTL == (X) ) \ |
| 4905 | ? EXACTL \ |
| 4906 | : ( EXACTFLU8 == (X) ) \ |
| 4907 | ? EXACTFLU8 \ |
| 4908 | : 0 ) |
| 4909 | |
| 4910 | /* dont use tail as the end marker for this traverse */ |
| 4911 | for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) { |
| 4912 | regnode * const noper = NEXTOPER( cur ); |
| 4913 | U8 noper_type = OP( noper ); |
| 4914 | U8 noper_trietype = TRIE_TYPE( noper_type ); |
| 4915 | #if defined(DEBUGGING) || defined(NOJUMPTRIE) |
| 4916 | regnode * const noper_next = regnext( noper ); |
| 4917 | U8 noper_next_type = (noper_next && noper_next < tail) ? OP(noper_next) : 0; |
| 4918 | U8 noper_next_trietype = (noper_next && noper_next < tail) ? TRIE_TYPE( noper_next_type ) :0; |
| 4919 | #endif |
| 4920 | |
| 4921 | DEBUG_TRIE_COMPILE_r({ |
| 4922 | regprop(RExC_rx, RExC_mysv, cur, NULL, pRExC_state); |
| 4923 | Perl_re_indentf( aTHX_ "- %d:%s (%d)", |
| 4924 | depth+1, |
| 4925 | REG_NODE_NUM(cur), SvPV_nolen_const( RExC_mysv ), REG_NODE_NUM(cur) ); |
| 4926 | |
| 4927 | regprop(RExC_rx, RExC_mysv, noper, NULL, pRExC_state); |
| 4928 | Perl_re_printf( aTHX_ " -> %d:%s", |
| 4929 | REG_NODE_NUM(noper), SvPV_nolen_const(RExC_mysv)); |
| 4930 | |
| 4931 | if ( noper_next ) { |
| 4932 | regprop(RExC_rx, RExC_mysv, noper_next, NULL, pRExC_state); |
| 4933 | Perl_re_printf( aTHX_ "\t=> %d:%s\t", |
| 4934 | REG_NODE_NUM(noper_next), SvPV_nolen_const(RExC_mysv)); |
| 4935 | } |
| 4936 | Perl_re_printf( aTHX_ "(First==%d,Last==%d,Cur==%d,tt==%s,ntt==%s,nntt==%s)\n", |
| 4937 | REG_NODE_NUM(first), REG_NODE_NUM(prev), REG_NODE_NUM(cur), |
| 4938 | PL_reg_name[trietype], PL_reg_name[noper_trietype], PL_reg_name[noper_next_trietype] |
| 4939 | ); |
| 4940 | }); |
| 4941 | |
| 4942 | /* Is noper a trieable nodetype that can be merged |
| 4943 | * with the current trie (if there is one)? */ |
| 4944 | if ( noper_trietype |
| 4945 | && |
| 4946 | ( |
| 4947 | ( noper_trietype == NOTHING ) |
| 4948 | || ( trietype == NOTHING ) |
| 4949 | || ( trietype == noper_trietype ) |
| 4950 | ) |
| 4951 | #ifdef NOJUMPTRIE |
| 4952 | && noper_next >= tail |
| 4953 | #endif |
| 4954 | && count < U16_MAX) |
| 4955 | { |
| 4956 | /* Handle mergable triable node Either we are |
| 4957 | * the first node in a new trieable sequence, |
| 4958 | * in which case we do some bookkeeping, |
| 4959 | * otherwise we update the end pointer. */ |
| 4960 | if ( !first ) { |
| 4961 | first = cur; |
| 4962 | if ( noper_trietype == NOTHING ) { |
| 4963 | #if !defined(DEBUGGING) && !defined(NOJUMPTRIE) |
| 4964 | regnode * const noper_next = regnext( noper ); |
| 4965 | U8 noper_next_type = (noper_next && noper_next < tail) ? OP(noper_next) : 0; |
| 4966 | U8 noper_next_trietype = noper_next_type ? TRIE_TYPE( noper_next_type ) :0; |
| 4967 | #endif |
| 4968 | |
| 4969 | if ( noper_next_trietype ) { |
| 4970 | trietype = noper_next_trietype; |
| 4971 | } else if (noper_next_type) { |
| 4972 | /* a NOTHING regop is 1 regop wide. |
| 4973 | * We need at least two for a trie |
| 4974 | * so we can't merge this in */ |
| 4975 | first = NULL; |
| 4976 | } |
| 4977 | } else { |
| 4978 | trietype = noper_trietype; |
| 4979 | } |
| 4980 | } else { |
| 4981 | if ( trietype == NOTHING ) |
| 4982 | trietype = noper_trietype; |
| 4983 | prev = cur; |
| 4984 | } |
| 4985 | if (first) |
| 4986 | count++; |
| 4987 | } /* end handle mergable triable node */ |
| 4988 | else { |
| 4989 | /* handle unmergable node - |
| 4990 | * noper may either be a triable node which can |
| 4991 | * not be tried together with the current trie, |
| 4992 | * or a non triable node */ |
| 4993 | if ( prev ) { |
| 4994 | /* If last is set and trietype is not |
| 4995 | * NOTHING then we have found at least two |
| 4996 | * triable branch sequences in a row of a |
| 4997 | * similar trietype so we can turn them |
| 4998 | * into a trie. If/when we allow NOTHING to |
| 4999 | * start a trie sequence this condition |
| 5000 | * will be required, and it isn't expensive |
| 5001 | * so we leave it in for now. */ |
| 5002 | if ( trietype && trietype != NOTHING ) |
| 5003 | make_trie( pRExC_state, |
| 5004 | startbranch, first, cur, tail, |
| 5005 | count, trietype, depth+1 ); |
| 5006 | prev = NULL; /* note: we clear/update |
| 5007 | first, trietype etc below, |
| 5008 | so we dont do it here */ |
| 5009 | } |
| 5010 | if ( noper_trietype |
| 5011 | #ifdef NOJUMPTRIE |
| 5012 | && noper_next >= tail |
| 5013 | #endif |
| 5014 | ){ |
| 5015 | /* noper is triable, so we can start a new |
| 5016 | * trie sequence */ |
| 5017 | count = 1; |
| 5018 | first = cur; |
| 5019 | trietype = noper_trietype; |
| 5020 | } else if (first) { |
| 5021 | /* if we already saw a first but the |
| 5022 | * current node is not triable then we have |
| 5023 | * to reset the first information. */ |
| 5024 | count = 0; |
| 5025 | first = NULL; |
| 5026 | trietype = 0; |
| 5027 | } |
| 5028 | } /* end handle unmergable node */ |
| 5029 | } /* loop over branches */ |
| 5030 | DEBUG_TRIE_COMPILE_r({ |
| 5031 | regprop(RExC_rx, RExC_mysv, cur, NULL, pRExC_state); |
| 5032 | Perl_re_indentf( aTHX_ "- %s (%d) <SCAN FINISHED> ", |
| 5033 | depth+1, SvPV_nolen_const( RExC_mysv ), REG_NODE_NUM(cur)); |
| 5034 | Perl_re_printf( aTHX_ "(First==%d, Last==%d, Cur==%d, tt==%s)\n", |
| 5035 | REG_NODE_NUM(first), REG_NODE_NUM(prev), REG_NODE_NUM(cur), |
| 5036 | PL_reg_name[trietype] |
| 5037 | ); |
| 5038 | |
| 5039 | }); |
| 5040 | if ( prev && trietype ) { |
| 5041 | if ( trietype != NOTHING ) { |
| 5042 | /* the last branch of the sequence was part of |
| 5043 | * a trie, so we have to construct it here |
| 5044 | * outside of the loop */ |
| 5045 | made= make_trie( pRExC_state, startbranch, |
| 5046 | first, scan, tail, count, |
| 5047 | trietype, depth+1 ); |
| 5048 | #ifdef TRIE_STUDY_OPT |
| 5049 | if ( ((made == MADE_EXACT_TRIE && |
| 5050 | startbranch == first) |
| 5051 | || ( first_non_open == first )) && |
| 5052 | depth==0 ) { |
| 5053 | flags |= SCF_TRIE_RESTUDY; |
| 5054 | if ( startbranch == first |
| 5055 | && scan >= tail ) |
| 5056 | { |
| 5057 | RExC_seen &=~REG_TOP_LEVEL_BRANCHES_SEEN; |
| 5058 | } |
| 5059 | } |
| 5060 | #endif |
| 5061 | } else { |
| 5062 | /* at this point we know whatever we have is a |
| 5063 | * NOTHING sequence/branch AND if 'startbranch' |
| 5064 | * is 'first' then we can turn the whole thing |
| 5065 | * into a NOTHING |
| 5066 | */ |
| 5067 | if ( startbranch == first ) { |
| 5068 | regnode *opt; |
| 5069 | /* the entire thing is a NOTHING sequence, |
| 5070 | * something like this: (?:|) So we can |
| 5071 | * turn it into a plain NOTHING op. */ |
| 5072 | DEBUG_TRIE_COMPILE_r({ |
| 5073 | regprop(RExC_rx, RExC_mysv, cur, NULL, pRExC_state); |
| 5074 | Perl_re_indentf( aTHX_ "- %s (%d) <NOTHING BRANCH SEQUENCE>\n", |
| 5075 | depth+1, |
| 5076 | SvPV_nolen_const( RExC_mysv ), REG_NODE_NUM(cur)); |
| 5077 | |
| 5078 | }); |
| 5079 | OP(startbranch)= NOTHING; |
| 5080 | NEXT_OFF(startbranch)= tail - startbranch; |
| 5081 | for ( opt= startbranch + 1; opt < tail ; opt++ ) |
| 5082 | OP(opt)= OPTIMIZED; |
| 5083 | } |
| 5084 | } |
| 5085 | } /* end if ( prev) */ |
| 5086 | } /* TRIE_MAXBUF is non zero */ |
| 5087 | } /* do trie */ |
| 5088 | |
| 5089 | } |
| 5090 | else if ( code == BRANCHJ ) { /* single branch is optimized. */ |
| 5091 | scan = NEXTOPER(NEXTOPER(scan)); |
| 5092 | } else /* single branch is optimized. */ |
| 5093 | scan = NEXTOPER(scan); |
| 5094 | continue; |
| 5095 | } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB) { |
| 5096 | I32 paren = 0; |
| 5097 | regnode *start = NULL; |
| 5098 | regnode *end = NULL; |
| 5099 | U32 my_recursed_depth= recursed_depth; |
| 5100 | |
| 5101 | if (OP(scan) != SUSPEND) { /* GOSUB */ |
| 5102 | /* Do setup, note this code has side effects beyond |
| 5103 | * the rest of this block. Specifically setting |
| 5104 | * RExC_recurse[] must happen at least once during |
| 5105 | * study_chunk(). */ |
| 5106 | paren = ARG(scan); |
| 5107 | RExC_recurse[ARG2L(scan)] = scan; |
| 5108 | start = REGNODE_p(RExC_open_parens[paren]); |
| 5109 | end = REGNODE_p(RExC_close_parens[paren]); |
| 5110 | |
| 5111 | /* NOTE we MUST always execute the above code, even |
| 5112 | * if we do nothing with a GOSUB */ |
| 5113 | if ( |
| 5114 | ( flags & SCF_IN_DEFINE ) |
| 5115 | || |
| 5116 | ( |
| 5117 | (is_inf_internal || is_inf || (data && data->flags & SF_IS_INF)) |
| 5118 | && |
| 5119 | ( (flags & (SCF_DO_STCLASS | SCF_DO_SUBSTR)) == 0 ) |
| 5120 | ) |
| 5121 | ) { |
| 5122 | /* no need to do anything here if we are in a define. */ |
| 5123 | /* or we are after some kind of infinite construct |
| 5124 | * so we can skip recursing into this item. |
| 5125 | * Since it is infinite we will not change the maxlen |
| 5126 | * or delta, and if we miss something that might raise |
| 5127 | * the minlen it will merely pessimise a little. |
| 5128 | * |
| 5129 | * Iow /(?(DEFINE)(?<foo>foo|food))a+(?&foo)/ |
| 5130 | * might result in a minlen of 1 and not of 4, |
| 5131 | * but this doesn't make us mismatch, just try a bit |
| 5132 | * harder than we should. |
| 5133 | * */ |
| 5134 | scan= regnext(scan); |
| 5135 | continue; |
| 5136 | } |
| 5137 | |
| 5138 | if ( |
| 5139 | !recursed_depth |
| 5140 | || !PAREN_TEST(recursed_depth - 1, paren) |
| 5141 | ) { |
| 5142 | /* it is quite possible that there are more efficient ways |
| 5143 | * to do this. We maintain a bitmap per level of recursion |
| 5144 | * of which patterns we have entered so we can detect if a |
| 5145 | * pattern creates a possible infinite loop. When we |
| 5146 | * recurse down a level we copy the previous levels bitmap |
| 5147 | * down. When we are at recursion level 0 we zero the top |
| 5148 | * level bitmap. It would be nice to implement a different |
| 5149 | * more efficient way of doing this. In particular the top |
| 5150 | * level bitmap may be unnecessary. |
| 5151 | */ |
| 5152 | if (!recursed_depth) { |
| 5153 | Zero(RExC_study_chunk_recursed, RExC_study_chunk_recursed_bytes, U8); |
| 5154 | } else { |
| 5155 | Copy(PAREN_OFFSET(recursed_depth - 1), |
| 5156 | PAREN_OFFSET(recursed_depth), |
| 5157 | RExC_study_chunk_recursed_bytes, U8); |
| 5158 | } |
| 5159 | /* we havent recursed into this paren yet, so recurse into it */ |
| 5160 | DEBUG_STUDYDATA("gosub-set", data, depth, is_inf); |
| 5161 | PAREN_SET(recursed_depth, paren); |
| 5162 | my_recursed_depth= recursed_depth + 1; |
| 5163 | } else { |
| 5164 | DEBUG_STUDYDATA("gosub-inf", data, depth, is_inf); |
| 5165 | /* some form of infinite recursion, assume infinite length |
| 5166 | * */ |
| 5167 | if (flags & SCF_DO_SUBSTR) { |
| 5168 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5169 | data->cur_is_floating = 1; |
| 5170 | } |
| 5171 | is_inf = is_inf_internal = 1; |
| 5172 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
| 5173 | ssc_anything(data->start_class); |
| 5174 | flags &= ~SCF_DO_STCLASS; |
| 5175 | |
| 5176 | start= NULL; /* reset start so we dont recurse later on. */ |
| 5177 | } |
| 5178 | } else { |
| 5179 | paren = stopparen; |
| 5180 | start = scan + 2; |
| 5181 | end = regnext(scan); |
| 5182 | } |
| 5183 | if (start) { |
| 5184 | scan_frame *newframe; |
| 5185 | assert(end); |
| 5186 | if (!RExC_frame_last) { |
| 5187 | Newxz(newframe, 1, scan_frame); |
| 5188 | SAVEDESTRUCTOR_X(S_unwind_scan_frames, newframe); |
| 5189 | RExC_frame_head= newframe; |
| 5190 | RExC_frame_count++; |
| 5191 | } else if (!RExC_frame_last->next_frame) { |
| 5192 | Newxz(newframe, 1, scan_frame); |
| 5193 | RExC_frame_last->next_frame= newframe; |
| 5194 | newframe->prev_frame= RExC_frame_last; |
| 5195 | RExC_frame_count++; |
| 5196 | } else { |
| 5197 | newframe= RExC_frame_last->next_frame; |
| 5198 | } |
| 5199 | RExC_frame_last= newframe; |
| 5200 | |
| 5201 | newframe->next_regnode = regnext(scan); |
| 5202 | newframe->last_regnode = last; |
| 5203 | newframe->stopparen = stopparen; |
| 5204 | newframe->prev_recursed_depth = recursed_depth; |
| 5205 | newframe->this_prev_frame= frame; |
| 5206 | |
| 5207 | DEBUG_STUDYDATA("frame-new", data, depth, is_inf); |
| 5208 | DEBUG_PEEP("fnew", scan, depth, flags); |
| 5209 | |
| 5210 | frame = newframe; |
| 5211 | scan = start; |
| 5212 | stopparen = paren; |
| 5213 | last = end; |
| 5214 | depth = depth + 1; |
| 5215 | recursed_depth= my_recursed_depth; |
| 5216 | |
| 5217 | continue; |
| 5218 | } |
| 5219 | } |
| 5220 | else if ( OP(scan) == EXACT |
| 5221 | || OP(scan) == LEXACT |
| 5222 | || OP(scan) == EXACT_REQ8 |
| 5223 | || OP(scan) == LEXACT_REQ8 |
| 5224 | || OP(scan) == EXACTL) |
| 5225 | { |
| 5226 | SSize_t bytelen = STR_LEN(scan), charlen; |
| 5227 | UV uc; |
| 5228 | assert(bytelen); |
| 5229 | if (UTF) { |
| 5230 | const U8 * const s = (U8*)STRING(scan); |
| 5231 | uc = utf8_to_uvchr_buf(s, s + bytelen, NULL); |
| 5232 | charlen = utf8_length(s, s + bytelen); |
| 5233 | } else { |
| 5234 | uc = *((U8*)STRING(scan)); |
| 5235 | charlen = bytelen; |
| 5236 | } |
| 5237 | min += charlen; |
| 5238 | if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */ |
| 5239 | /* The code below prefers earlier match for fixed |
| 5240 | offset, later match for variable offset. */ |
| 5241 | if (data->last_end == -1) { /* Update the start info. */ |
| 5242 | data->last_start_min = data->pos_min; |
| 5243 | data->last_start_max = is_inf |
| 5244 | ? OPTIMIZE_INFTY : data->pos_min + data->pos_delta; |
| 5245 | } |
| 5246 | sv_catpvn(data->last_found, STRING(scan), bytelen); |
| 5247 | if (UTF) |
| 5248 | SvUTF8_on(data->last_found); |
| 5249 | { |
| 5250 | SV * const sv = data->last_found; |
| 5251 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? |
| 5252 | mg_find(sv, PERL_MAGIC_utf8) : NULL; |
| 5253 | if (mg && mg->mg_len >= 0) |
| 5254 | mg->mg_len += charlen; |
| 5255 | } |
| 5256 | data->last_end = data->pos_min + charlen; |
| 5257 | data->pos_min += charlen; /* As in the first entry. */ |
| 5258 | data->flags &= ~SF_BEFORE_EOL; |
| 5259 | } |
| 5260 | |
| 5261 | /* ANDing the code point leaves at most it, and not in locale, and |
| 5262 | * can't match null string */ |
| 5263 | if (flags & SCF_DO_STCLASS_AND) { |
| 5264 | ssc_cp_and(data->start_class, uc); |
| 5265 | ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING; |
| 5266 | ssc_clear_locale(data->start_class); |
| 5267 | } |
| 5268 | else if (flags & SCF_DO_STCLASS_OR) { |
| 5269 | ssc_add_cp(data->start_class, uc); |
| 5270 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 5271 | |
| 5272 | /* See commit msg 749e076fceedeb708a624933726e7989f2302f6a */ |
| 5273 | ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING; |
| 5274 | } |
| 5275 | flags &= ~SCF_DO_STCLASS; |
| 5276 | } |
| 5277 | else if (PL_regkind[OP(scan)] == EXACT) { |
| 5278 | /* But OP != EXACT!, so is EXACTFish */ |
| 5279 | SSize_t bytelen = STR_LEN(scan), charlen; |
| 5280 | const U8 * s = (U8*)STRING(scan); |
| 5281 | |
| 5282 | /* Replace a length 1 ASCII fold pair node with an ANYOFM node, |
| 5283 | * with the mask set to the complement of the bit that differs |
| 5284 | * between upper and lower case, and the lowest code point of the |
| 5285 | * pair (which the '&' forces) */ |
| 5286 | if ( bytelen == 1 |
| 5287 | && isALPHA_A(*s) |
| 5288 | && ( OP(scan) == EXACTFAA |
| 5289 | || ( OP(scan) == EXACTFU |
| 5290 | && ! HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(*s)))) |
| 5291 | { |
| 5292 | U8 mask = ~ ('A' ^ 'a'); /* These differ in just one bit */ |
| 5293 | |
| 5294 | OP(scan) = ANYOFM; |
| 5295 | ARG_SET(scan, *s & mask); |
| 5296 | FLAGS(scan) = mask; |
| 5297 | /* we're not EXACTFish any more, so restudy */ |
| 5298 | continue; |
| 5299 | } |
| 5300 | |
| 5301 | /* Search for fixed substrings supports EXACT only. */ |
| 5302 | if (flags & SCF_DO_SUBSTR) { |
| 5303 | assert(data); |
| 5304 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5305 | } |
| 5306 | charlen = UTF ? (SSize_t) utf8_length(s, s + bytelen) : bytelen; |
| 5307 | if (unfolded_multi_char) { |
| 5308 | RExC_seen |= REG_UNFOLDED_MULTI_SEEN; |
| 5309 | } |
| 5310 | min += charlen - min_subtract; |
| 5311 | assert (min >= 0); |
| 5312 | delta += min_subtract; |
| 5313 | if (flags & SCF_DO_SUBSTR) { |
| 5314 | data->pos_min += charlen - min_subtract; |
| 5315 | if (data->pos_min < 0) { |
| 5316 | data->pos_min = 0; |
| 5317 | } |
| 5318 | data->pos_delta += min_subtract; |
| 5319 | if (min_subtract) { |
| 5320 | data->cur_is_floating = 1; /* float */ |
| 5321 | } |
| 5322 | } |
| 5323 | |
| 5324 | if (flags & SCF_DO_STCLASS) { |
| 5325 | SV* EXACTF_invlist = make_exactf_invlist(pRExC_state, scan); |
| 5326 | |
| 5327 | assert(EXACTF_invlist); |
| 5328 | if (flags & SCF_DO_STCLASS_AND) { |
| 5329 | if (OP(scan) != EXACTFL) |
| 5330 | ssc_clear_locale(data->start_class); |
| 5331 | ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING; |
| 5332 | ANYOF_POSIXL_ZERO(data->start_class); |
| 5333 | ssc_intersection(data->start_class, EXACTF_invlist, FALSE); |
| 5334 | } |
| 5335 | else { /* SCF_DO_STCLASS_OR */ |
| 5336 | ssc_union(data->start_class, EXACTF_invlist, FALSE); |
| 5337 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 5338 | |
| 5339 | /* See commit msg 749e076fceedeb708a624933726e7989f2302f6a */ |
| 5340 | ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING; |
| 5341 | } |
| 5342 | flags &= ~SCF_DO_STCLASS; |
| 5343 | SvREFCNT_dec(EXACTF_invlist); |
| 5344 | } |
| 5345 | } |
| 5346 | else if (REGNODE_VARIES(OP(scan))) { |
| 5347 | SSize_t mincount, maxcount, minnext, deltanext, pos_before = 0; |
| 5348 | I32 fl = 0, f = flags; |
| 5349 | regnode * const oscan = scan; |
| 5350 | regnode_ssc this_class; |
| 5351 | regnode_ssc *oclass = NULL; |
| 5352 | I32 next_is_eval = 0; |
| 5353 | |
| 5354 | switch (PL_regkind[OP(scan)]) { |
| 5355 | case WHILEM: /* End of (?:...)* . */ |
| 5356 | scan = NEXTOPER(scan); |
| 5357 | goto finish; |
| 5358 | case PLUS: |
| 5359 | if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) { |
| 5360 | next = NEXTOPER(scan); |
| 5361 | if ( OP(next) == EXACT |
| 5362 | || OP(next) == LEXACT |
| 5363 | || OP(next) == EXACT_REQ8 |
| 5364 | || OP(next) == LEXACT_REQ8 |
| 5365 | || OP(next) == EXACTL |
| 5366 | || (flags & SCF_DO_STCLASS)) |
| 5367 | { |
| 5368 | mincount = 1; |
| 5369 | maxcount = REG_INFTY; |
| 5370 | next = regnext(scan); |
| 5371 | scan = NEXTOPER(scan); |
| 5372 | goto do_curly; |
| 5373 | } |
| 5374 | } |
| 5375 | if (flags & SCF_DO_SUBSTR) |
| 5376 | data->pos_min++; |
| 5377 | min++; |
| 5378 | /* FALLTHROUGH */ |
| 5379 | case STAR: |
| 5380 | next = NEXTOPER(scan); |
| 5381 | |
| 5382 | /* This temporary node can now be turned into EXACTFU, and |
| 5383 | * must, as regexec.c doesn't handle it */ |
| 5384 | if (OP(next) == EXACTFU_S_EDGE) { |
| 5385 | OP(next) = EXACTFU; |
| 5386 | } |
| 5387 | |
| 5388 | if ( STR_LEN(next) == 1 |
| 5389 | && isALPHA_A(* STRING(next)) |
| 5390 | && ( OP(next) == EXACTFAA |
| 5391 | || ( OP(next) == EXACTFU |
| 5392 | && ! HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(* STRING(next))))) |
| 5393 | { |
| 5394 | /* These differ in just one bit */ |
| 5395 | U8 mask = ~ ('A' ^ 'a'); |
| 5396 | |
| 5397 | assert(isALPHA_A(* STRING(next))); |
| 5398 | |
| 5399 | /* Then replace it by an ANYOFM node, with |
| 5400 | * the mask set to the complement of the |
| 5401 | * bit that differs between upper and lower |
| 5402 | * case, and the lowest code point of the |
| 5403 | * pair (which the '&' forces) */ |
| 5404 | OP(next) = ANYOFM; |
| 5405 | ARG_SET(next, *STRING(next) & mask); |
| 5406 | FLAGS(next) = mask; |
| 5407 | } |
| 5408 | |
| 5409 | if (flags & SCF_DO_STCLASS) { |
| 5410 | mincount = 0; |
| 5411 | maxcount = REG_INFTY; |
| 5412 | next = regnext(scan); |
| 5413 | scan = NEXTOPER(scan); |
| 5414 | goto do_curly; |
| 5415 | } |
| 5416 | if (flags & SCF_DO_SUBSTR) { |
| 5417 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5418 | /* Cannot extend fixed substrings */ |
| 5419 | data->cur_is_floating = 1; /* float */ |
| 5420 | } |
| 5421 | is_inf = is_inf_internal = 1; |
| 5422 | scan = regnext(scan); |
| 5423 | goto optimize_curly_tail; |
| 5424 | case CURLY: |
| 5425 | if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM) |
| 5426 | && (scan->flags == stopparen)) |
| 5427 | { |
| 5428 | mincount = 1; |
| 5429 | maxcount = 1; |
| 5430 | } else { |
| 5431 | mincount = ARG1(scan); |
| 5432 | maxcount = ARG2(scan); |
| 5433 | } |
| 5434 | next = regnext(scan); |
| 5435 | if (OP(scan) == CURLYX) { |
| 5436 | I32 lp = (data ? *(data->last_closep) : 0); |
| 5437 | scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX); |
| 5438 | } |
| 5439 | scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; |
| 5440 | next_is_eval = (OP(scan) == EVAL); |
| 5441 | do_curly: |
| 5442 | if (flags & SCF_DO_SUBSTR) { |
| 5443 | if (mincount == 0) |
| 5444 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5445 | /* Cannot extend fixed substrings */ |
| 5446 | pos_before = data->pos_min; |
| 5447 | } |
| 5448 | if (data) { |
| 5449 | fl = data->flags; |
| 5450 | data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL); |
| 5451 | if (is_inf) |
| 5452 | data->flags |= SF_IS_INF; |
| 5453 | } |
| 5454 | if (flags & SCF_DO_STCLASS) { |
| 5455 | ssc_init(pRExC_state, &this_class); |
| 5456 | oclass = data->start_class; |
| 5457 | data->start_class = &this_class; |
| 5458 | f |= SCF_DO_STCLASS_AND; |
| 5459 | f &= ~SCF_DO_STCLASS_OR; |
| 5460 | } |
| 5461 | /* Exclude from super-linear cache processing any {n,m} |
| 5462 | regops for which the combination of input pos and regex |
| 5463 | pos is not enough information to determine if a match |
| 5464 | will be possible. |
| 5465 | |
| 5466 | For example, in the regex /foo(bar\s*){4,8}baz/ with the |
| 5467 | regex pos at the \s*, the prospects for a match depend not |
| 5468 | only on the input position but also on how many (bar\s*) |
| 5469 | repeats into the {4,8} we are. */ |
| 5470 | if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY)) |
| 5471 | f &= ~SCF_WHILEM_VISITED_POS; |
| 5472 | |
| 5473 | /* This will finish on WHILEM, setting scan, or on NULL: */ |
| 5474 | /* recurse study_chunk() on loop bodies */ |
| 5475 | minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, |
| 5476 | last, data, stopparen, recursed_depth, NULL, |
| 5477 | (mincount == 0 |
| 5478 | ? (f & ~SCF_DO_SUBSTR) |
| 5479 | : f) |
| 5480 | ,depth+1); |
| 5481 | |
| 5482 | if (flags & SCF_DO_STCLASS) |
| 5483 | data->start_class = oclass; |
| 5484 | if (mincount == 0 || minnext == 0) { |
| 5485 | if (flags & SCF_DO_STCLASS_OR) { |
| 5486 | ssc_or(pRExC_state, data->start_class, (regnode_charclass *) &this_class); |
| 5487 | } |
| 5488 | else if (flags & SCF_DO_STCLASS_AND) { |
| 5489 | /* Switch to OR mode: cache the old value of |
| 5490 | * data->start_class */ |
| 5491 | INIT_AND_WITHP; |
| 5492 | StructCopy(data->start_class, and_withp, regnode_ssc); |
| 5493 | flags &= ~SCF_DO_STCLASS_AND; |
| 5494 | StructCopy(&this_class, data->start_class, regnode_ssc); |
| 5495 | flags |= SCF_DO_STCLASS_OR; |
| 5496 | ANYOF_FLAGS(data->start_class) |
| 5497 | |= SSC_MATCHES_EMPTY_STRING; |
| 5498 | } |
| 5499 | } else { /* Non-zero len */ |
| 5500 | if (flags & SCF_DO_STCLASS_OR) { |
| 5501 | ssc_or(pRExC_state, data->start_class, (regnode_charclass *) &this_class); |
| 5502 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 5503 | } |
| 5504 | else if (flags & SCF_DO_STCLASS_AND) |
| 5505 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &this_class); |
| 5506 | flags &= ~SCF_DO_STCLASS; |
| 5507 | } |
| 5508 | if (!scan) /* It was not CURLYX, but CURLY. */ |
| 5509 | scan = next; |
| 5510 | if (((flags & (SCF_TRIE_DOING_RESTUDY|SCF_DO_SUBSTR))==SCF_DO_SUBSTR) |
| 5511 | /* ? quantifier ok, except for (?{ ... }) */ |
| 5512 | && (next_is_eval || !(mincount == 0 && maxcount == 1)) |
| 5513 | && (minnext == 0) && (deltanext == 0) |
| 5514 | && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 5515 | && maxcount <= REG_INFTY/3) /* Complement check for big |
| 5516 | count */ |
| 5517 | { |
| 5518 | _WARN_HELPER(RExC_precomp_end, packWARN(WARN_REGEXP), |
| 5519 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), |
| 5520 | "Quantifier unexpected on zero-length expression " |
| 5521 | "in regex m/%" UTF8f "/", |
| 5522 | UTF8fARG(UTF, RExC_precomp_end - RExC_precomp, |
| 5523 | RExC_precomp))); |
| 5524 | } |
| 5525 | |
| 5526 | min += minnext * mincount; |
| 5527 | is_inf_internal |= deltanext == OPTIMIZE_INFTY |
| 5528 | || (maxcount == REG_INFTY && minnext + deltanext > 0); |
| 5529 | is_inf |= is_inf_internal; |
| 5530 | if (is_inf) { |
| 5531 | delta = OPTIMIZE_INFTY; |
| 5532 | } else { |
| 5533 | delta += (minnext + deltanext) * maxcount |
| 5534 | - minnext * mincount; |
| 5535 | } |
| 5536 | /* Try powerful optimization CURLYX => CURLYN. */ |
| 5537 | if ( OP(oscan) == CURLYX && data |
| 5538 | && data->flags & SF_IN_PAR |
| 5539 | && !(data->flags & SF_HAS_EVAL) |
| 5540 | && !deltanext && minnext == 1 ) { |
| 5541 | /* Try to optimize to CURLYN. */ |
| 5542 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; |
| 5543 | regnode * const nxt1 = nxt; |
| 5544 | #ifdef DEBUGGING |
| 5545 | regnode *nxt2; |
| 5546 | #endif |
| 5547 | |
| 5548 | /* Skip open. */ |
| 5549 | nxt = regnext(nxt); |
| 5550 | if (!REGNODE_SIMPLE(OP(nxt)) |
| 5551 | && !(PL_regkind[OP(nxt)] == EXACT |
| 5552 | && STR_LEN(nxt) == 1)) |
| 5553 | goto nogo; |
| 5554 | #ifdef DEBUGGING |
| 5555 | nxt2 = nxt; |
| 5556 | #endif |
| 5557 | nxt = regnext(nxt); |
| 5558 | if (OP(nxt) != CLOSE) |
| 5559 | goto nogo; |
| 5560 | if (RExC_open_parens) { |
| 5561 | |
| 5562 | /*open->CURLYM*/ |
| 5563 | RExC_open_parens[ARG(nxt1)] = REGNODE_OFFSET(oscan); |
| 5564 | |
| 5565 | /*close->while*/ |
| 5566 | RExC_close_parens[ARG(nxt1)] = REGNODE_OFFSET(nxt) + 2; |
| 5567 | } |
| 5568 | /* Now we know that nxt2 is the only contents: */ |
| 5569 | oscan->flags = (U8)ARG(nxt); |
| 5570 | OP(oscan) = CURLYN; |
| 5571 | OP(nxt1) = NOTHING; /* was OPEN. */ |
| 5572 | |
| 5573 | #ifdef DEBUGGING |
| 5574 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
| 5575 | NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */ |
| 5576 | NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */ |
| 5577 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
| 5578 | OP(nxt + 1) = OPTIMIZED; /* was count. */ |
| 5579 | NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */ |
| 5580 | #endif |
| 5581 | } |
| 5582 | nogo: |
| 5583 | |
| 5584 | /* Try optimization CURLYX => CURLYM. */ |
| 5585 | if ( OP(oscan) == CURLYX && data |
| 5586 | && !(data->flags & SF_HAS_PAR) |
| 5587 | && !(data->flags & SF_HAS_EVAL) |
| 5588 | && !deltanext /* atom is fixed width */ |
| 5589 | && minnext != 0 /* CURLYM can't handle zero width */ |
| 5590 | |
| 5591 | /* Nor characters whose fold at run-time may be |
| 5592 | * multi-character */ |
| 5593 | && ! (RExC_seen & REG_UNFOLDED_MULTI_SEEN) |
| 5594 | ) { |
| 5595 | /* XXXX How to optimize if data == 0? */ |
| 5596 | /* Optimize to a simpler form. */ |
| 5597 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */ |
| 5598 | regnode *nxt2; |
| 5599 | |
| 5600 | OP(oscan) = CURLYM; |
| 5601 | while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/ |
| 5602 | && (OP(nxt2) != WHILEM)) |
| 5603 | nxt = nxt2; |
| 5604 | OP(nxt2) = SUCCEED; /* Whas WHILEM */ |
| 5605 | /* Need to optimize away parenths. */ |
| 5606 | if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) { |
| 5607 | /* Set the parenth number. */ |
| 5608 | regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/ |
| 5609 | |
| 5610 | oscan->flags = (U8)ARG(nxt); |
| 5611 | if (RExC_open_parens) { |
| 5612 | /*open->CURLYM*/ |
| 5613 | RExC_open_parens[ARG(nxt1)] = REGNODE_OFFSET(oscan); |
| 5614 | |
| 5615 | /*close->NOTHING*/ |
| 5616 | RExC_close_parens[ARG(nxt1)] = REGNODE_OFFSET(nxt2) |
| 5617 | + 1; |
| 5618 | } |
| 5619 | OP(nxt1) = OPTIMIZED; /* was OPEN. */ |
| 5620 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
| 5621 | |
| 5622 | #ifdef DEBUGGING |
| 5623 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
| 5624 | OP(nxt + 1) = OPTIMIZED; /* was count. */ |
| 5625 | NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */ |
| 5626 | NEXT_OFF(nxt + 1) = 0; /* just for consistency. */ |
| 5627 | #endif |
| 5628 | #if 0 |
| 5629 | while ( nxt1 && (OP(nxt1) != WHILEM)) { |
| 5630 | regnode *nnxt = regnext(nxt1); |
| 5631 | if (nnxt == nxt) { |
| 5632 | if (reg_off_by_arg[OP(nxt1)]) |
| 5633 | ARG_SET(nxt1, nxt2 - nxt1); |
| 5634 | else if (nxt2 - nxt1 < U16_MAX) |
| 5635 | NEXT_OFF(nxt1) = nxt2 - nxt1; |
| 5636 | else |
| 5637 | OP(nxt) = NOTHING; /* Cannot beautify */ |
| 5638 | } |
| 5639 | nxt1 = nnxt; |
| 5640 | } |
| 5641 | #endif |
| 5642 | /* Optimize again: */ |
| 5643 | /* recurse study_chunk() on optimised CURLYX => CURLYM */ |
| 5644 | study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt, |
| 5645 | NULL, stopparen, recursed_depth, NULL, 0, |
| 5646 | depth+1); |
| 5647 | } |
| 5648 | else |
| 5649 | oscan->flags = 0; |
| 5650 | } |
| 5651 | else if ((OP(oscan) == CURLYX) |
| 5652 | && (flags & SCF_WHILEM_VISITED_POS) |
| 5653 | /* See the comment on a similar expression above. |
| 5654 | However, this time it's not a subexpression |
| 5655 | we care about, but the expression itself. */ |
| 5656 | && (maxcount == REG_INFTY) |
| 5657 | && data) { |
| 5658 | /* This stays as CURLYX, we can put the count/of pair. */ |
| 5659 | /* Find WHILEM (as in regexec.c) */ |
| 5660 | regnode *nxt = oscan + NEXT_OFF(oscan); |
| 5661 | |
| 5662 | if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */ |
| 5663 | nxt += ARG(nxt); |
| 5664 | nxt = PREVOPER(nxt); |
| 5665 | if (nxt->flags & 0xf) { |
| 5666 | /* we've already set whilem count on this node */ |
| 5667 | } else if (++data->whilem_c < 16) { |
| 5668 | assert(data->whilem_c <= RExC_whilem_seen); |
| 5669 | nxt->flags = (U8)(data->whilem_c |
| 5670 | | (RExC_whilem_seen << 4)); /* On WHILEM */ |
| 5671 | } |
| 5672 | } |
| 5673 | if (data && fl & (SF_HAS_PAR|SF_IN_PAR)) |
| 5674 | pars++; |
| 5675 | if (flags & SCF_DO_SUBSTR) { |
| 5676 | SV *last_str = NULL; |
| 5677 | STRLEN last_chrs = 0; |
| 5678 | int counted = mincount != 0; |
| 5679 | |
| 5680 | if (data->last_end > 0 && mincount != 0) { /* Ends with a |
| 5681 | string. */ |
| 5682 | SSize_t b = pos_before >= data->last_start_min |
| 5683 | ? pos_before : data->last_start_min; |
| 5684 | STRLEN l; |
| 5685 | const char * const s = SvPV_const(data->last_found, l); |
| 5686 | SSize_t old = b - data->last_start_min; |
| 5687 | assert(old >= 0); |
| 5688 | |
| 5689 | if (UTF) |
| 5690 | old = utf8_hop_forward((U8*)s, old, |
| 5691 | (U8 *) SvEND(data->last_found)) |
| 5692 | - (U8*)s; |
| 5693 | l -= old; |
| 5694 | /* Get the added string: */ |
| 5695 | last_str = newSVpvn_utf8(s + old, l, UTF); |
| 5696 | last_chrs = UTF ? utf8_length((U8*)(s + old), |
| 5697 | (U8*)(s + old + l)) : l; |
| 5698 | if (deltanext == 0 && pos_before == b) { |
| 5699 | /* What was added is a constant string */ |
| 5700 | if (mincount > 1) { |
| 5701 | |
| 5702 | SvGROW(last_str, (mincount * l) + 1); |
| 5703 | repeatcpy(SvPVX(last_str) + l, |
| 5704 | SvPVX_const(last_str), l, |
| 5705 | mincount - 1); |
| 5706 | SvCUR_set(last_str, SvCUR(last_str) * mincount); |
| 5707 | /* Add additional parts. */ |
| 5708 | SvCUR_set(data->last_found, |
| 5709 | SvCUR(data->last_found) - l); |
| 5710 | sv_catsv(data->last_found, last_str); |
| 5711 | { |
| 5712 | SV * sv = data->last_found; |
| 5713 | MAGIC *mg = |
| 5714 | SvUTF8(sv) && SvMAGICAL(sv) ? |
| 5715 | mg_find(sv, PERL_MAGIC_utf8) : NULL; |
| 5716 | if (mg && mg->mg_len >= 0) |
| 5717 | mg->mg_len += last_chrs * (mincount-1); |
| 5718 | } |
| 5719 | last_chrs *= mincount; |
| 5720 | data->last_end += l * (mincount - 1); |
| 5721 | } |
| 5722 | } else { |
| 5723 | /* start offset must point into the last copy */ |
| 5724 | data->last_start_min += minnext * (mincount - 1); |
| 5725 | data->last_start_max = |
| 5726 | is_inf |
| 5727 | ? OPTIMIZE_INFTY |
| 5728 | : data->last_start_max + |
| 5729 | (maxcount - 1) * (minnext + data->pos_delta); |
| 5730 | } |
| 5731 | } |
| 5732 | /* It is counted once already... */ |
| 5733 | data->pos_min += minnext * (mincount - counted); |
| 5734 | #if 0 |
| 5735 | Perl_re_printf( aTHX_ "counted=%" UVuf " deltanext=%" UVuf |
| 5736 | " OPTIMIZE_INFTY=%" UVuf " minnext=%" UVuf |
| 5737 | " maxcount=%" UVuf " mincount=%" UVuf "\n", |
| 5738 | (UV)counted, (UV)deltanext, (UV)OPTIMIZE_INFTY, (UV)minnext, (UV)maxcount, |
| 5739 | (UV)mincount); |
| 5740 | if (deltanext != OPTIMIZE_INFTY) |
| 5741 | Perl_re_printf( aTHX_ "LHS=%" UVuf " RHS=%" UVuf "\n", |
| 5742 | (UV)(-counted * deltanext + (minnext + deltanext) * maxcount |
| 5743 | - minnext * mincount), (UV)(OPTIMIZE_INFTY - data->pos_delta)); |
| 5744 | #endif |
| 5745 | if (deltanext == OPTIMIZE_INFTY |
| 5746 | || -counted * deltanext + (minnext + deltanext) * maxcount - minnext * mincount >= OPTIMIZE_INFTY - data->pos_delta) |
| 5747 | data->pos_delta = OPTIMIZE_INFTY; |
| 5748 | else |
| 5749 | data->pos_delta += - counted * deltanext + |
| 5750 | (minnext + deltanext) * maxcount - minnext * mincount; |
| 5751 | if (mincount != maxcount) { |
| 5752 | /* Cannot extend fixed substrings found inside |
| 5753 | the group. */ |
| 5754 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5755 | if (mincount && last_str) { |
| 5756 | SV * const sv = data->last_found; |
| 5757 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? |
| 5758 | mg_find(sv, PERL_MAGIC_utf8) : NULL; |
| 5759 | |
| 5760 | if (mg) |
| 5761 | mg->mg_len = -1; |
| 5762 | sv_setsv(sv, last_str); |
| 5763 | data->last_end = data->pos_min; |
| 5764 | data->last_start_min = data->pos_min - last_chrs; |
| 5765 | data->last_start_max = is_inf |
| 5766 | ? OPTIMIZE_INFTY |
| 5767 | : data->pos_min + data->pos_delta - last_chrs; |
| 5768 | } |
| 5769 | data->cur_is_floating = 1; /* float */ |
| 5770 | } |
| 5771 | SvREFCNT_dec(last_str); |
| 5772 | } |
| 5773 | if (data && (fl & SF_HAS_EVAL)) |
| 5774 | data->flags |= SF_HAS_EVAL; |
| 5775 | optimize_curly_tail: |
| 5776 | if (OP(oscan) != CURLYX) { |
| 5777 | while (PL_regkind[OP(next = regnext(oscan))] == NOTHING |
| 5778 | && NEXT_OFF(next)) |
| 5779 | NEXT_OFF(oscan) += NEXT_OFF(next); |
| 5780 | } |
| 5781 | continue; |
| 5782 | |
| 5783 | default: |
| 5784 | Perl_croak(aTHX_ "panic: unexpected varying REx opcode %d", |
| 5785 | OP(scan)); |
| 5786 | case REF: |
| 5787 | case CLUMP: |
| 5788 | if (flags & SCF_DO_SUBSTR) { |
| 5789 | /* Cannot expect anything... */ |
| 5790 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5791 | data->cur_is_floating = 1; /* float */ |
| 5792 | } |
| 5793 | is_inf = is_inf_internal = 1; |
| 5794 | if (flags & SCF_DO_STCLASS_OR) { |
| 5795 | if (OP(scan) == CLUMP) { |
| 5796 | /* Actually is any start char, but very few code points |
| 5797 | * aren't start characters */ |
| 5798 | ssc_match_all_cp(data->start_class); |
| 5799 | } |
| 5800 | else { |
| 5801 | ssc_anything(data->start_class); |
| 5802 | } |
| 5803 | } |
| 5804 | flags &= ~SCF_DO_STCLASS; |
| 5805 | break; |
| 5806 | } |
| 5807 | } |
| 5808 | else if (OP(scan) == LNBREAK) { |
| 5809 | if (flags & SCF_DO_STCLASS) { |
| 5810 | if (flags & SCF_DO_STCLASS_AND) { |
| 5811 | ssc_intersection(data->start_class, |
| 5812 | PL_XPosix_ptrs[_CC_VERTSPACE], FALSE); |
| 5813 | ssc_clear_locale(data->start_class); |
| 5814 | ANYOF_FLAGS(data->start_class) |
| 5815 | &= ~SSC_MATCHES_EMPTY_STRING; |
| 5816 | } |
| 5817 | else if (flags & SCF_DO_STCLASS_OR) { |
| 5818 | ssc_union(data->start_class, |
| 5819 | PL_XPosix_ptrs[_CC_VERTSPACE], |
| 5820 | FALSE); |
| 5821 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 5822 | |
| 5823 | /* See commit msg for |
| 5824 | * 749e076fceedeb708a624933726e7989f2302f6a */ |
| 5825 | ANYOF_FLAGS(data->start_class) |
| 5826 | &= ~SSC_MATCHES_EMPTY_STRING; |
| 5827 | } |
| 5828 | flags &= ~SCF_DO_STCLASS; |
| 5829 | } |
| 5830 | min++; |
| 5831 | if (delta != OPTIMIZE_INFTY) |
| 5832 | delta++; /* Because of the 2 char string cr-lf */ |
| 5833 | if (flags & SCF_DO_SUBSTR) { |
| 5834 | /* Cannot expect anything... */ |
| 5835 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5836 | data->pos_min += 1; |
| 5837 | if (data->pos_delta != OPTIMIZE_INFTY) { |
| 5838 | data->pos_delta += 1; |
| 5839 | } |
| 5840 | data->cur_is_floating = 1; /* float */ |
| 5841 | } |
| 5842 | } |
| 5843 | else if (REGNODE_SIMPLE(OP(scan))) { |
| 5844 | |
| 5845 | if (flags & SCF_DO_SUBSTR) { |
| 5846 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 5847 | data->pos_min++; |
| 5848 | } |
| 5849 | min++; |
| 5850 | if (flags & SCF_DO_STCLASS) { |
| 5851 | bool invert = 0; |
| 5852 | SV* my_invlist = NULL; |
| 5853 | U8 namedclass; |
| 5854 | |
| 5855 | /* See commit msg 749e076fceedeb708a624933726e7989f2302f6a */ |
| 5856 | ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING; |
| 5857 | |
| 5858 | /* Some of the logic below assumes that switching |
| 5859 | locale on will only add false positives. */ |
| 5860 | switch (OP(scan)) { |
| 5861 | |
| 5862 | default: |
| 5863 | #ifdef DEBUGGING |
| 5864 | Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", |
| 5865 | OP(scan)); |
| 5866 | #endif |
| 5867 | case SANY: |
| 5868 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
| 5869 | ssc_match_all_cp(data->start_class); |
| 5870 | break; |
| 5871 | |
| 5872 | case REG_ANY: |
| 5873 | { |
| 5874 | SV* REG_ANY_invlist = _new_invlist(2); |
| 5875 | REG_ANY_invlist = add_cp_to_invlist(REG_ANY_invlist, |
| 5876 | '\n'); |
| 5877 | if (flags & SCF_DO_STCLASS_OR) { |
| 5878 | ssc_union(data->start_class, |
| 5879 | REG_ANY_invlist, |
| 5880 | TRUE /* TRUE => invert, hence all but \n |
| 5881 | */ |
| 5882 | ); |
| 5883 | } |
| 5884 | else if (flags & SCF_DO_STCLASS_AND) { |
| 5885 | ssc_intersection(data->start_class, |
| 5886 | REG_ANY_invlist, |
| 5887 | TRUE /* TRUE => invert */ |
| 5888 | ); |
| 5889 | ssc_clear_locale(data->start_class); |
| 5890 | } |
| 5891 | SvREFCNT_dec_NN(REG_ANY_invlist); |
| 5892 | } |
| 5893 | break; |
| 5894 | |
| 5895 | case ANYOFD: |
| 5896 | case ANYOFL: |
| 5897 | case ANYOFPOSIXL: |
| 5898 | case ANYOFH: |
| 5899 | case ANYOFHb: |
| 5900 | case ANYOFHr: |
| 5901 | case ANYOFHs: |
| 5902 | case ANYOF: |
| 5903 | if (flags & SCF_DO_STCLASS_AND) |
| 5904 | ssc_and(pRExC_state, data->start_class, |
| 5905 | (regnode_charclass *) scan); |
| 5906 | else |
| 5907 | ssc_or(pRExC_state, data->start_class, |
| 5908 | (regnode_charclass *) scan); |
| 5909 | break; |
| 5910 | |
| 5911 | case NANYOFM: |
| 5912 | case ANYOFM: |
| 5913 | { |
| 5914 | SV* cp_list = get_ANYOFM_contents(scan); |
| 5915 | |
| 5916 | if (flags & SCF_DO_STCLASS_OR) { |
| 5917 | ssc_union(data->start_class, cp_list, invert); |
| 5918 | } |
| 5919 | else if (flags & SCF_DO_STCLASS_AND) { |
| 5920 | ssc_intersection(data->start_class, cp_list, invert); |
| 5921 | } |
| 5922 | |
| 5923 | SvREFCNT_dec_NN(cp_list); |
| 5924 | break; |
| 5925 | } |
| 5926 | |
| 5927 | case ANYOFR: |
| 5928 | case ANYOFRb: |
| 5929 | { |
| 5930 | SV* cp_list = NULL; |
| 5931 | |
| 5932 | cp_list = _add_range_to_invlist(cp_list, |
| 5933 | ANYOFRbase(scan), |
| 5934 | ANYOFRbase(scan) + ANYOFRdelta(scan)); |
| 5935 | |
| 5936 | if (flags & SCF_DO_STCLASS_OR) { |
| 5937 | ssc_union(data->start_class, cp_list, invert); |
| 5938 | } |
| 5939 | else if (flags & SCF_DO_STCLASS_AND) { |
| 5940 | ssc_intersection(data->start_class, cp_list, invert); |
| 5941 | } |
| 5942 | |
| 5943 | SvREFCNT_dec_NN(cp_list); |
| 5944 | break; |
| 5945 | } |
| 5946 | |
| 5947 | case NPOSIXL: |
| 5948 | invert = 1; |
| 5949 | /* FALLTHROUGH */ |
| 5950 | |
| 5951 | case POSIXL: |
| 5952 | namedclass = classnum_to_namedclass(FLAGS(scan)) + invert; |
| 5953 | if (flags & SCF_DO_STCLASS_AND) { |
| 5954 | bool was_there = cBOOL( |
| 5955 | ANYOF_POSIXL_TEST(data->start_class, |
| 5956 | namedclass)); |
| 5957 | ANYOF_POSIXL_ZERO(data->start_class); |
| 5958 | if (was_there) { /* Do an AND */ |
| 5959 | ANYOF_POSIXL_SET(data->start_class, namedclass); |
| 5960 | } |
| 5961 | /* No individual code points can now match */ |
| 5962 | data->start_class->invlist |
| 5963 | = sv_2mortal(_new_invlist(0)); |
| 5964 | } |
| 5965 | else { |
| 5966 | int complement = namedclass + ((invert) ? -1 : 1); |
| 5967 | |
| 5968 | assert(flags & SCF_DO_STCLASS_OR); |
| 5969 | |
| 5970 | /* If the complement of this class was already there, |
| 5971 | * the result is that they match all code points, |
| 5972 | * (\d + \D == everything). Remove the classes from |
| 5973 | * future consideration. Locale is not relevant in |
| 5974 | * this case */ |
| 5975 | if (ANYOF_POSIXL_TEST(data->start_class, complement)) { |
| 5976 | ssc_match_all_cp(data->start_class); |
| 5977 | ANYOF_POSIXL_CLEAR(data->start_class, namedclass); |
| 5978 | ANYOF_POSIXL_CLEAR(data->start_class, complement); |
| 5979 | } |
| 5980 | else { /* The usual case; just add this class to the |
| 5981 | existing set */ |
| 5982 | ANYOF_POSIXL_SET(data->start_class, namedclass); |
| 5983 | } |
| 5984 | } |
| 5985 | break; |
| 5986 | |
| 5987 | case NPOSIXA: /* For these, we always know the exact set of |
| 5988 | what's matched */ |
| 5989 | invert = 1; |
| 5990 | /* FALLTHROUGH */ |
| 5991 | case POSIXA: |
| 5992 | my_invlist = invlist_clone(PL_Posix_ptrs[FLAGS(scan)], NULL); |
| 5993 | goto join_posix_and_ascii; |
| 5994 | |
| 5995 | case NPOSIXD: |
| 5996 | case NPOSIXU: |
| 5997 | invert = 1; |
| 5998 | /* FALLTHROUGH */ |
| 5999 | case POSIXD: |
| 6000 | case POSIXU: |
| 6001 | my_invlist = invlist_clone(PL_XPosix_ptrs[FLAGS(scan)], NULL); |
| 6002 | |
| 6003 | /* NPOSIXD matches all upper Latin1 code points unless the |
| 6004 | * target string being matched is UTF-8, which is |
| 6005 | * unknowable until match time. Since we are going to |
| 6006 | * invert, we want to get rid of all of them so that the |
| 6007 | * inversion will match all */ |
| 6008 | if (OP(scan) == NPOSIXD) { |
| 6009 | _invlist_subtract(my_invlist, PL_UpperLatin1, |
| 6010 | &my_invlist); |
| 6011 | } |
| 6012 | |
| 6013 | join_posix_and_ascii: |
| 6014 | |
| 6015 | if (flags & SCF_DO_STCLASS_AND) { |
| 6016 | ssc_intersection(data->start_class, my_invlist, invert); |
| 6017 | ssc_clear_locale(data->start_class); |
| 6018 | } |
| 6019 | else { |
| 6020 | assert(flags & SCF_DO_STCLASS_OR); |
| 6021 | ssc_union(data->start_class, my_invlist, invert); |
| 6022 | } |
| 6023 | SvREFCNT_dec(my_invlist); |
| 6024 | } |
| 6025 | if (flags & SCF_DO_STCLASS_OR) |
| 6026 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 6027 | flags &= ~SCF_DO_STCLASS; |
| 6028 | } |
| 6029 | } |
| 6030 | else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) { |
| 6031 | data->flags |= (OP(scan) == MEOL |
| 6032 | ? SF_BEFORE_MEOL |
| 6033 | : SF_BEFORE_SEOL); |
| 6034 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 6035 | |
| 6036 | } |
| 6037 | else if ( PL_regkind[OP(scan)] == BRANCHJ |
| 6038 | /* Lookbehind, or need to calculate parens/evals/stclass: */ |
| 6039 | && (scan->flags || data || (flags & SCF_DO_STCLASS)) |
| 6040 | && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) |
| 6041 | { |
| 6042 | if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY |
| 6043 | || OP(scan) == UNLESSM ) |
| 6044 | { |
| 6045 | /* Negative Lookahead/lookbehind |
| 6046 | In this case we can't do fixed string optimisation. |
| 6047 | */ |
| 6048 | |
| 6049 | SSize_t deltanext, minnext, fake = 0; |
| 6050 | regnode *nscan; |
| 6051 | regnode_ssc intrnl; |
| 6052 | int f = 0; |
| 6053 | |
| 6054 | StructCopy(&zero_scan_data, &data_fake, scan_data_t); |
| 6055 | if (data) { |
| 6056 | data_fake.whilem_c = data->whilem_c; |
| 6057 | data_fake.last_closep = data->last_closep; |
| 6058 | } |
| 6059 | else |
| 6060 | data_fake.last_closep = &fake; |
| 6061 | data_fake.pos_delta = delta; |
| 6062 | if ( flags & SCF_DO_STCLASS && !scan->flags |
| 6063 | && OP(scan) == IFMATCH ) { /* Lookahead */ |
| 6064 | ssc_init(pRExC_state, &intrnl); |
| 6065 | data_fake.start_class = &intrnl; |
| 6066 | f |= SCF_DO_STCLASS_AND; |
| 6067 | } |
| 6068 | if (flags & SCF_WHILEM_VISITED_POS) |
| 6069 | f |= SCF_WHILEM_VISITED_POS; |
| 6070 | next = regnext(scan); |
| 6071 | nscan = NEXTOPER(NEXTOPER(scan)); |
| 6072 | |
| 6073 | /* recurse study_chunk() for lookahead body */ |
| 6074 | minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, |
| 6075 | last, &data_fake, stopparen, |
| 6076 | recursed_depth, NULL, f, depth+1); |
| 6077 | if (scan->flags) { |
| 6078 | if ( deltanext < 0 |
| 6079 | || deltanext > (I32) U8_MAX |
| 6080 | || minnext > (I32)U8_MAX |
| 6081 | || minnext + deltanext > (I32)U8_MAX) |
| 6082 | { |
| 6083 | FAIL2("Lookbehind longer than %" UVuf " not implemented", |
| 6084 | (UV)U8_MAX); |
| 6085 | } |
| 6086 | |
| 6087 | /* The 'next_off' field has been repurposed to count the |
| 6088 | * additional starting positions to try beyond the initial |
| 6089 | * one. (This leaves it at 0 for non-variable length |
| 6090 | * matches to avoid breakage for those not using this |
| 6091 | * extension) */ |
| 6092 | if (deltanext) { |
| 6093 | scan->next_off = deltanext; |
| 6094 | ckWARNexperimental(RExC_parse, |
| 6095 | WARN_EXPERIMENTAL__VLB, |
| 6096 | "Variable length lookbehind is experimental"); |
| 6097 | } |
| 6098 | scan->flags = (U8)minnext + deltanext; |
| 6099 | } |
| 6100 | if (data) { |
| 6101 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 6102 | pars++; |
| 6103 | if (data_fake.flags & SF_HAS_EVAL) |
| 6104 | data->flags |= SF_HAS_EVAL; |
| 6105 | data->whilem_c = data_fake.whilem_c; |
| 6106 | } |
| 6107 | if (f & SCF_DO_STCLASS_AND) { |
| 6108 | if (flags & SCF_DO_STCLASS_OR) { |
| 6109 | /* OR before, AND after: ideally we would recurse with |
| 6110 | * data_fake to get the AND applied by study of the |
| 6111 | * remainder of the pattern, and then derecurse; |
| 6112 | * *** HACK *** for now just treat as "no information". |
| 6113 | * See [perl #56690]. |
| 6114 | */ |
| 6115 | ssc_init(pRExC_state, data->start_class); |
| 6116 | } else { |
| 6117 | /* AND before and after: combine and continue. These |
| 6118 | * assertions are zero-length, so can match an EMPTY |
| 6119 | * string */ |
| 6120 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &intrnl); |
| 6121 | ANYOF_FLAGS(data->start_class) |
| 6122 | |= SSC_MATCHES_EMPTY_STRING; |
| 6123 | } |
| 6124 | } |
| 6125 | } |
| 6126 | #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY |
| 6127 | else { |
| 6128 | /* Positive Lookahead/lookbehind |
| 6129 | In this case we can do fixed string optimisation, |
| 6130 | but we must be careful about it. Note in the case of |
| 6131 | lookbehind the positions will be offset by the minimum |
| 6132 | length of the pattern, something we won't know about |
| 6133 | until after the recurse. |
| 6134 | */ |
| 6135 | SSize_t deltanext, fake = 0; |
| 6136 | regnode *nscan; |
| 6137 | regnode_ssc intrnl; |
| 6138 | int f = 0; |
| 6139 | /* We use SAVEFREEPV so that when the full compile |
| 6140 | is finished perl will clean up the allocated |
| 6141 | minlens when it's all done. This way we don't |
| 6142 | have to worry about freeing them when we know |
| 6143 | they wont be used, which would be a pain. |
| 6144 | */ |
| 6145 | SSize_t *minnextp; |
| 6146 | Newx( minnextp, 1, SSize_t ); |
| 6147 | SAVEFREEPV(minnextp); |
| 6148 | |
| 6149 | if (data) { |
| 6150 | StructCopy(data, &data_fake, scan_data_t); |
| 6151 | if ((flags & SCF_DO_SUBSTR) && data->last_found) { |
| 6152 | f |= SCF_DO_SUBSTR; |
| 6153 | if (scan->flags) |
| 6154 | scan_commit(pRExC_state, &data_fake, minlenp, is_inf); |
| 6155 | data_fake.last_found=newSVsv(data->last_found); |
| 6156 | } |
| 6157 | } |
| 6158 | else |
| 6159 | data_fake.last_closep = &fake; |
| 6160 | data_fake.flags = 0; |
| 6161 | data_fake.substrs[0].flags = 0; |
| 6162 | data_fake.substrs[1].flags = 0; |
| 6163 | data_fake.pos_delta = delta; |
| 6164 | if (is_inf) |
| 6165 | data_fake.flags |= SF_IS_INF; |
| 6166 | if ( flags & SCF_DO_STCLASS && !scan->flags |
| 6167 | && OP(scan) == IFMATCH ) { /* Lookahead */ |
| 6168 | ssc_init(pRExC_state, &intrnl); |
| 6169 | data_fake.start_class = &intrnl; |
| 6170 | f |= SCF_DO_STCLASS_AND; |
| 6171 | } |
| 6172 | if (flags & SCF_WHILEM_VISITED_POS) |
| 6173 | f |= SCF_WHILEM_VISITED_POS; |
| 6174 | next = regnext(scan); |
| 6175 | nscan = NEXTOPER(NEXTOPER(scan)); |
| 6176 | |
| 6177 | /* positive lookahead study_chunk() recursion */ |
| 6178 | *minnextp = study_chunk(pRExC_state, &nscan, minnextp, |
| 6179 | &deltanext, last, &data_fake, |
| 6180 | stopparen, recursed_depth, NULL, |
| 6181 | f, depth+1); |
| 6182 | if (scan->flags) { |
| 6183 | assert(0); /* This code has never been tested since this |
| 6184 | is normally not compiled */ |
| 6185 | if ( deltanext < 0 |
| 6186 | || deltanext > (I32) U8_MAX |
| 6187 | || *minnextp > (I32)U8_MAX |
| 6188 | || *minnextp + deltanext > (I32)U8_MAX) |
| 6189 | { |
| 6190 | FAIL2("Lookbehind longer than %" UVuf " not implemented", |
| 6191 | (UV)U8_MAX); |
| 6192 | } |
| 6193 | |
| 6194 | if (deltanext) { |
| 6195 | scan->next_off = deltanext; |
| 6196 | } |
| 6197 | scan->flags = (U8)*minnextp + deltanext; |
| 6198 | } |
| 6199 | |
| 6200 | *minnextp += min; |
| 6201 | |
| 6202 | if (f & SCF_DO_STCLASS_AND) { |
| 6203 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &intrnl); |
| 6204 | ANYOF_FLAGS(data->start_class) |= SSC_MATCHES_EMPTY_STRING; |
| 6205 | } |
| 6206 | if (data) { |
| 6207 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 6208 | pars++; |
| 6209 | if (data_fake.flags & SF_HAS_EVAL) |
| 6210 | data->flags |= SF_HAS_EVAL; |
| 6211 | data->whilem_c = data_fake.whilem_c; |
| 6212 | if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) { |
| 6213 | int i; |
| 6214 | if (RExC_rx->minlen<*minnextp) |
| 6215 | RExC_rx->minlen=*minnextp; |
| 6216 | scan_commit(pRExC_state, &data_fake, minnextp, is_inf); |
| 6217 | SvREFCNT_dec_NN(data_fake.last_found); |
| 6218 | |
| 6219 | for (i = 0; i < 2; i++) { |
| 6220 | if (data_fake.substrs[i].minlenp != minlenp) { |
| 6221 | data->substrs[i].min_offset = |
| 6222 | data_fake.substrs[i].min_offset; |
| 6223 | data->substrs[i].max_offset = |
| 6224 | data_fake.substrs[i].max_offset; |
| 6225 | data->substrs[i].minlenp = |
| 6226 | data_fake.substrs[i].minlenp; |
| 6227 | data->substrs[i].lookbehind += scan->flags; |
| 6228 | } |
| 6229 | } |
| 6230 | } |
| 6231 | } |
| 6232 | } |
| 6233 | #endif |
| 6234 | } |
| 6235 | else if (OP(scan) == OPEN) { |
| 6236 | if (stopparen != (I32)ARG(scan)) |
| 6237 | pars++; |
| 6238 | } |
| 6239 | else if (OP(scan) == CLOSE) { |
| 6240 | if (stopparen == (I32)ARG(scan)) { |
| 6241 | break; |
| 6242 | } |
| 6243 | if ((I32)ARG(scan) == is_par) { |
| 6244 | next = regnext(scan); |
| 6245 | |
| 6246 | if ( next && (OP(next) != WHILEM) && next < last) |
| 6247 | is_par = 0; /* Disable optimization */ |
| 6248 | } |
| 6249 | if (data) |
| 6250 | *(data->last_closep) = ARG(scan); |
| 6251 | } |
| 6252 | else if (OP(scan) == EVAL) { |
| 6253 | if (data) |
| 6254 | data->flags |= SF_HAS_EVAL; |
| 6255 | } |
| 6256 | else if ( PL_regkind[OP(scan)] == ENDLIKE ) { |
| 6257 | if (flags & SCF_DO_SUBSTR) { |
| 6258 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 6259 | flags &= ~SCF_DO_SUBSTR; |
| 6260 | } |
| 6261 | if (data && OP(scan)==ACCEPT) { |
| 6262 | data->flags |= SCF_SEEN_ACCEPT; |
| 6263 | if (stopmin > min) |
| 6264 | stopmin = min; |
| 6265 | } |
| 6266 | } |
| 6267 | else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */ |
| 6268 | { |
| 6269 | if (flags & SCF_DO_SUBSTR) { |
| 6270 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 6271 | data->cur_is_floating = 1; /* float */ |
| 6272 | } |
| 6273 | is_inf = is_inf_internal = 1; |
| 6274 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
| 6275 | ssc_anything(data->start_class); |
| 6276 | flags &= ~SCF_DO_STCLASS; |
| 6277 | } |
| 6278 | else if (OP(scan) == GPOS) { |
| 6279 | if (!(RExC_rx->intflags & PREGf_GPOS_FLOAT) && |
| 6280 | !(delta || is_inf || (data && data->pos_delta))) |
| 6281 | { |
| 6282 | if (!(RExC_rx->intflags & PREGf_ANCH) && (flags & SCF_DO_SUBSTR)) |
| 6283 | RExC_rx->intflags |= PREGf_ANCH_GPOS; |
| 6284 | if (RExC_rx->gofs < (STRLEN)min) |
| 6285 | RExC_rx->gofs = min; |
| 6286 | } else { |
| 6287 | RExC_rx->intflags |= PREGf_GPOS_FLOAT; |
| 6288 | RExC_rx->gofs = 0; |
| 6289 | } |
| 6290 | } |
| 6291 | #ifdef TRIE_STUDY_OPT |
| 6292 | #ifdef FULL_TRIE_STUDY |
| 6293 | else if (PL_regkind[OP(scan)] == TRIE) { |
| 6294 | /* NOTE - There is similar code to this block above for handling |
| 6295 | BRANCH nodes on the initial study. If you change stuff here |
| 6296 | check there too. */ |
| 6297 | regnode *trie_node= scan; |
| 6298 | regnode *tail= regnext(scan); |
| 6299 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
| 6300 | SSize_t max1 = 0, min1 = OPTIMIZE_INFTY; |
| 6301 | regnode_ssc accum; |
| 6302 | |
| 6303 | if (flags & SCF_DO_SUBSTR) { /* XXXX Add !SUSPEND? */ |
| 6304 | /* Cannot merge strings after this. */ |
| 6305 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 6306 | } |
| 6307 | if (flags & SCF_DO_STCLASS) |
| 6308 | ssc_init_zero(pRExC_state, &accum); |
| 6309 | |
| 6310 | if (!trie->jump) { |
| 6311 | min1= trie->minlen; |
| 6312 | max1= trie->maxlen; |
| 6313 | } else { |
| 6314 | const regnode *nextbranch= NULL; |
| 6315 | U32 word; |
| 6316 | |
| 6317 | for ( word=1 ; word <= trie->wordcount ; word++) |
| 6318 | { |
| 6319 | SSize_t deltanext=0, minnext=0, f = 0, fake; |
| 6320 | regnode_ssc this_class; |
| 6321 | |
| 6322 | StructCopy(&zero_scan_data, &data_fake, scan_data_t); |
| 6323 | if (data) { |
| 6324 | data_fake.whilem_c = data->whilem_c; |
| 6325 | data_fake.last_closep = data->last_closep; |
| 6326 | } |
| 6327 | else |
| 6328 | data_fake.last_closep = &fake; |
| 6329 | data_fake.pos_delta = delta; |
| 6330 | if (flags & SCF_DO_STCLASS) { |
| 6331 | ssc_init(pRExC_state, &this_class); |
| 6332 | data_fake.start_class = &this_class; |
| 6333 | f = SCF_DO_STCLASS_AND; |
| 6334 | } |
| 6335 | if (flags & SCF_WHILEM_VISITED_POS) |
| 6336 | f |= SCF_WHILEM_VISITED_POS; |
| 6337 | |
| 6338 | if (trie->jump[word]) { |
| 6339 | if (!nextbranch) |
| 6340 | nextbranch = trie_node + trie->jump[0]; |
| 6341 | scan= trie_node + trie->jump[word]; |
| 6342 | /* We go from the jump point to the branch that follows |
| 6343 | it. Note this means we need the vestigal unused |
| 6344 | branches even though they arent otherwise used. */ |
| 6345 | /* optimise study_chunk() for TRIE */ |
| 6346 | minnext = study_chunk(pRExC_state, &scan, minlenp, |
| 6347 | &deltanext, (regnode *)nextbranch, &data_fake, |
| 6348 | stopparen, recursed_depth, NULL, f, depth+1); |
| 6349 | } |
| 6350 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) |
| 6351 | nextbranch= regnext((regnode*)nextbranch); |
| 6352 | |
| 6353 | if (min1 > (SSize_t)(minnext + trie->minlen)) |
| 6354 | min1 = minnext + trie->minlen; |
| 6355 | if (deltanext == OPTIMIZE_INFTY) { |
| 6356 | is_inf = is_inf_internal = 1; |
| 6357 | max1 = OPTIMIZE_INFTY; |
| 6358 | } else if (max1 < (SSize_t)(minnext + deltanext + trie->maxlen)) |
| 6359 | max1 = minnext + deltanext + trie->maxlen; |
| 6360 | |
| 6361 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 6362 | pars++; |
| 6363 | if (data_fake.flags & SCF_SEEN_ACCEPT) { |
| 6364 | if ( stopmin > min + min1) |
| 6365 | stopmin = min + min1; |
| 6366 | flags &= ~SCF_DO_SUBSTR; |
| 6367 | if (data) |
| 6368 | data->flags |= SCF_SEEN_ACCEPT; |
| 6369 | } |
| 6370 | if (data) { |
| 6371 | if (data_fake.flags & SF_HAS_EVAL) |
| 6372 | data->flags |= SF_HAS_EVAL; |
| 6373 | data->whilem_c = data_fake.whilem_c; |
| 6374 | } |
| 6375 | if (flags & SCF_DO_STCLASS) |
| 6376 | ssc_or(pRExC_state, &accum, (regnode_charclass *) &this_class); |
| 6377 | } |
| 6378 | } |
| 6379 | if (flags & SCF_DO_SUBSTR) { |
| 6380 | data->pos_min += min1; |
| 6381 | data->pos_delta += max1 - min1; |
| 6382 | if (max1 != min1 || is_inf) |
| 6383 | data->cur_is_floating = 1; /* float */ |
| 6384 | } |
| 6385 | min += min1; |
| 6386 | if (delta != OPTIMIZE_INFTY) { |
| 6387 | if (OPTIMIZE_INFTY - (max1 - min1) >= delta) |
| 6388 | delta += max1 - min1; |
| 6389 | else |
| 6390 | delta = OPTIMIZE_INFTY; |
| 6391 | } |
| 6392 | if (flags & SCF_DO_STCLASS_OR) { |
| 6393 | ssc_or(pRExC_state, data->start_class, (regnode_charclass *) &accum); |
| 6394 | if (min1) { |
| 6395 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 6396 | flags &= ~SCF_DO_STCLASS; |
| 6397 | } |
| 6398 | } |
| 6399 | else if (flags & SCF_DO_STCLASS_AND) { |
| 6400 | if (min1) { |
| 6401 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &accum); |
| 6402 | flags &= ~SCF_DO_STCLASS; |
| 6403 | } |
| 6404 | else { |
| 6405 | /* Switch to OR mode: cache the old value of |
| 6406 | * data->start_class */ |
| 6407 | INIT_AND_WITHP; |
| 6408 | StructCopy(data->start_class, and_withp, regnode_ssc); |
| 6409 | flags &= ~SCF_DO_STCLASS_AND; |
| 6410 | StructCopy(&accum, data->start_class, regnode_ssc); |
| 6411 | flags |= SCF_DO_STCLASS_OR; |
| 6412 | } |
| 6413 | } |
| 6414 | scan= tail; |
| 6415 | continue; |
| 6416 | } |
| 6417 | #else |
| 6418 | else if (PL_regkind[OP(scan)] == TRIE) { |
| 6419 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
| 6420 | U8*bang=NULL; |
| 6421 | |
| 6422 | min += trie->minlen; |
| 6423 | delta += (trie->maxlen - trie->minlen); |
| 6424 | flags &= ~SCF_DO_STCLASS; /* xxx */ |
| 6425 | if (flags & SCF_DO_SUBSTR) { |
| 6426 | /* Cannot expect anything... */ |
| 6427 | scan_commit(pRExC_state, data, minlenp, is_inf); |
| 6428 | data->pos_min += trie->minlen; |
| 6429 | data->pos_delta += (trie->maxlen - trie->minlen); |
| 6430 | if (trie->maxlen != trie->minlen) |
| 6431 | data->cur_is_floating = 1; /* float */ |
| 6432 | } |
| 6433 | if (trie->jump) /* no more substrings -- for now /grr*/ |
| 6434 | flags &= ~SCF_DO_SUBSTR; |
| 6435 | } |
| 6436 | else if (OP(scan) == REGEX_SET) { |
| 6437 | Perl_croak(aTHX_ "panic: %s regnode should be resolved" |
| 6438 | " before optimization", reg_name[REGEX_SET]); |
| 6439 | } |
| 6440 | |
| 6441 | #endif /* old or new */ |
| 6442 | #endif /* TRIE_STUDY_OPT */ |
| 6443 | |
| 6444 | /* Else: zero-length, ignore. */ |
| 6445 | scan = regnext(scan); |
| 6446 | } |
| 6447 | |
| 6448 | finish: |
| 6449 | if (frame) { |
| 6450 | /* we need to unwind recursion. */ |
| 6451 | depth = depth - 1; |
| 6452 | |
| 6453 | DEBUG_STUDYDATA("frame-end", data, depth, is_inf); |
| 6454 | DEBUG_PEEP("fend", scan, depth, flags); |
| 6455 | |
| 6456 | /* restore previous context */ |
| 6457 | last = frame->last_regnode; |
| 6458 | scan = frame->next_regnode; |
| 6459 | stopparen = frame->stopparen; |
| 6460 | recursed_depth = frame->prev_recursed_depth; |
| 6461 | |
| 6462 | RExC_frame_last = frame->prev_frame; |
| 6463 | frame = frame->this_prev_frame; |
| 6464 | goto fake_study_recurse; |
| 6465 | } |
| 6466 | |
| 6467 | assert(!frame); |
| 6468 | DEBUG_STUDYDATA("pre-fin", data, depth, is_inf); |
| 6469 | |
| 6470 | *scanp = scan; |
| 6471 | *deltap = is_inf_internal ? OPTIMIZE_INFTY : delta; |
| 6472 | |
| 6473 | if (flags & SCF_DO_SUBSTR && is_inf) |
| 6474 | data->pos_delta = OPTIMIZE_INFTY - data->pos_min; |
| 6475 | if (is_par > (I32)U8_MAX) |
| 6476 | is_par = 0; |
| 6477 | if (is_par && pars==1 && data) { |
| 6478 | data->flags |= SF_IN_PAR; |
| 6479 | data->flags &= ~SF_HAS_PAR; |
| 6480 | } |
| 6481 | else if (pars && data) { |
| 6482 | data->flags |= SF_HAS_PAR; |
| 6483 | data->flags &= ~SF_IN_PAR; |
| 6484 | } |
| 6485 | if (flags & SCF_DO_STCLASS_OR) |
| 6486 | ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp); |
| 6487 | if (flags & SCF_TRIE_RESTUDY) |
| 6488 | data->flags |= SCF_TRIE_RESTUDY; |
| 6489 | |
| 6490 | DEBUG_STUDYDATA("post-fin", data, depth, is_inf); |
| 6491 | |
| 6492 | final_minlen = min < stopmin |
| 6493 | ? min : stopmin; |
| 6494 | |
| 6495 | if (!(RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN)) { |
| 6496 | if (final_minlen > OPTIMIZE_INFTY - delta) |
| 6497 | RExC_maxlen = OPTIMIZE_INFTY; |
| 6498 | else if (RExC_maxlen < final_minlen + delta) |
| 6499 | RExC_maxlen = final_minlen + delta; |
| 6500 | } |
| 6501 | return final_minlen; |
| 6502 | } |
| 6503 | |
| 6504 | STATIC U32 |
| 6505 | S_add_data(RExC_state_t* const pRExC_state, const char* const s, const U32 n) |
| 6506 | { |
| 6507 | U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0; |
| 6508 | |
| 6509 | PERL_ARGS_ASSERT_ADD_DATA; |
| 6510 | |
| 6511 | Renewc(RExC_rxi->data, |
| 6512 | sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1), |
| 6513 | char, struct reg_data); |
| 6514 | if(count) |
| 6515 | Renew(RExC_rxi->data->what, count + n, U8); |
| 6516 | else |
| 6517 | Newx(RExC_rxi->data->what, n, U8); |
| 6518 | RExC_rxi->data->count = count + n; |
| 6519 | Copy(s, RExC_rxi->data->what + count, n, U8); |
| 6520 | return count; |
| 6521 | } |
| 6522 | |
| 6523 | /*XXX: todo make this not included in a non debugging perl, but appears to be |
| 6524 | * used anyway there, in 'use re' */ |
| 6525 | #ifndef PERL_IN_XSUB_RE |
| 6526 | void |
| 6527 | Perl_reginitcolors(pTHX) |
| 6528 | { |
| 6529 | const char * const s = PerlEnv_getenv("PERL_RE_COLORS"); |
| 6530 | if (s) { |
| 6531 | char *t = savepv(s); |
| 6532 | int i = 0; |
| 6533 | PL_colors[0] = t; |
| 6534 | while (++i < 6) { |
| 6535 | t = strchr(t, '\t'); |
| 6536 | if (t) { |
| 6537 | *t = '\0'; |
| 6538 | PL_colors[i] = ++t; |
| 6539 | } |
| 6540 | else |
| 6541 | PL_colors[i] = t = (char *)""; |
| 6542 | } |
| 6543 | } else { |
| 6544 | int i = 0; |
| 6545 | while (i < 6) |
| 6546 | PL_colors[i++] = (char *)""; |
| 6547 | } |
| 6548 | PL_colorset = 1; |
| 6549 | } |
| 6550 | #endif |
| 6551 | |
| 6552 | |
| 6553 | #ifdef TRIE_STUDY_OPT |
| 6554 | #define CHECK_RESTUDY_GOTO_butfirst(dOsomething) \ |
| 6555 | STMT_START { \ |
| 6556 | if ( \ |
| 6557 | (data.flags & SCF_TRIE_RESTUDY) \ |
| 6558 | && ! restudied++ \ |
| 6559 | ) { \ |
| 6560 | dOsomething; \ |
| 6561 | goto reStudy; \ |
| 6562 | } \ |
| 6563 | } STMT_END |
| 6564 | #else |
| 6565 | #define CHECK_RESTUDY_GOTO_butfirst |
| 6566 | #endif |
| 6567 | |
| 6568 | /* |
| 6569 | * pregcomp - compile a regular expression into internal code |
| 6570 | * |
| 6571 | * Decides which engine's compiler to call based on the hint currently in |
| 6572 | * scope |
| 6573 | */ |
| 6574 | |
| 6575 | #ifndef PERL_IN_XSUB_RE |
| 6576 | |
| 6577 | /* return the currently in-scope regex engine (or the default if none) */ |
| 6578 | |
| 6579 | regexp_engine const * |
| 6580 | Perl_current_re_engine(pTHX) |
| 6581 | { |
| 6582 | if (IN_PERL_COMPILETIME) { |
| 6583 | HV * const table = GvHV(PL_hintgv); |
| 6584 | SV **ptr; |
| 6585 | |
| 6586 | if (!table || !(PL_hints & HINT_LOCALIZE_HH)) |
| 6587 | return &PL_core_reg_engine; |
| 6588 | ptr = hv_fetchs(table, "regcomp", FALSE); |
| 6589 | if ( !(ptr && SvIOK(*ptr) && SvIV(*ptr))) |
| 6590 | return &PL_core_reg_engine; |
| 6591 | return INT2PTR(regexp_engine*, SvIV(*ptr)); |
| 6592 | } |
| 6593 | else { |
| 6594 | SV *ptr; |
| 6595 | if (!PL_curcop->cop_hints_hash) |
| 6596 | return &PL_core_reg_engine; |
| 6597 | ptr = cop_hints_fetch_pvs(PL_curcop, "regcomp", 0); |
| 6598 | if ( !(ptr && SvIOK(ptr) && SvIV(ptr))) |
| 6599 | return &PL_core_reg_engine; |
| 6600 | return INT2PTR(regexp_engine*, SvIV(ptr)); |
| 6601 | } |
| 6602 | } |
| 6603 | |
| 6604 | |
| 6605 | REGEXP * |
| 6606 | Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags) |
| 6607 | { |
| 6608 | regexp_engine const *eng = current_re_engine(); |
| 6609 | GET_RE_DEBUG_FLAGS_DECL; |
| 6610 | |
| 6611 | PERL_ARGS_ASSERT_PREGCOMP; |
| 6612 | |
| 6613 | /* Dispatch a request to compile a regexp to correct regexp engine. */ |
| 6614 | DEBUG_COMPILE_r({ |
| 6615 | Perl_re_printf( aTHX_ "Using engine %" UVxf "\n", |
| 6616 | PTR2UV(eng)); |
| 6617 | }); |
| 6618 | return CALLREGCOMP_ENG(eng, pattern, flags); |
| 6619 | } |
| 6620 | #endif |
| 6621 | |
| 6622 | /* public(ish) entry point for the perl core's own regex compiling code. |
| 6623 | * It's actually a wrapper for Perl_re_op_compile that only takes an SV |
| 6624 | * pattern rather than a list of OPs, and uses the internal engine rather |
| 6625 | * than the current one */ |
| 6626 | |
| 6627 | REGEXP * |
| 6628 | Perl_re_compile(pTHX_ SV * const pattern, U32 rx_flags) |
| 6629 | { |
| 6630 | PERL_ARGS_ASSERT_RE_COMPILE; |
| 6631 | return re_op_compile_wrapper(pattern, rx_flags, 0); |
| 6632 | } |
| 6633 | |
| 6634 | REGEXP * |
| 6635 | S_re_op_compile_wrapper(pTHX_ SV * const pattern, U32 rx_flags, const U32 pm_flags) |
| 6636 | { |
| 6637 | SV *pat = pattern; /* defeat constness! */ |
| 6638 | |
| 6639 | PERL_ARGS_ASSERT_RE_OP_COMPILE_WRAPPER; |
| 6640 | |
| 6641 | return Perl_re_op_compile(aTHX_ &pat, 1, NULL, |
| 6642 | #ifdef PERL_IN_XSUB_RE |
| 6643 | &my_reg_engine, |
| 6644 | #else |
| 6645 | &PL_core_reg_engine, |
| 6646 | #endif |
| 6647 | NULL, NULL, rx_flags, pm_flags); |
| 6648 | } |
| 6649 | |
| 6650 | |
| 6651 | static void |
| 6652 | S_free_codeblocks(pTHX_ struct reg_code_blocks *cbs) |
| 6653 | { |
| 6654 | int n; |
| 6655 | |
| 6656 | if (--cbs->refcnt > 0) |
| 6657 | return; |
| 6658 | for (n = 0; n < cbs->count; n++) { |
| 6659 | REGEXP *rx = cbs->cb[n].src_regex; |
| 6660 | if (rx) { |
| 6661 | cbs->cb[n].src_regex = NULL; |
| 6662 | SvREFCNT_dec_NN(rx); |
| 6663 | } |
| 6664 | } |
| 6665 | Safefree(cbs->cb); |
| 6666 | Safefree(cbs); |
| 6667 | } |
| 6668 | |
| 6669 | |
| 6670 | static struct reg_code_blocks * |
| 6671 | S_alloc_code_blocks(pTHX_ int ncode) |
| 6672 | { |
| 6673 | struct reg_code_blocks *cbs; |
| 6674 | Newx(cbs, 1, struct reg_code_blocks); |
| 6675 | cbs->count = ncode; |
| 6676 | cbs->refcnt = 1; |
| 6677 | SAVEDESTRUCTOR_X(S_free_codeblocks, cbs); |
| 6678 | if (ncode) |
| 6679 | Newx(cbs->cb, ncode, struct reg_code_block); |
| 6680 | else |
| 6681 | cbs->cb = NULL; |
| 6682 | return cbs; |
| 6683 | } |
| 6684 | |
| 6685 | |
| 6686 | /* upgrade pattern pat_p of length plen_p to UTF8, and if there are code |
| 6687 | * blocks, recalculate the indices. Update pat_p and plen_p in-place to |
| 6688 | * point to the realloced string and length. |
| 6689 | * |
| 6690 | * This is essentially a copy of Perl_bytes_to_utf8() with the code index |
| 6691 | * stuff added */ |
| 6692 | |
| 6693 | static void |
| 6694 | S_pat_upgrade_to_utf8(pTHX_ RExC_state_t * const pRExC_state, |
| 6695 | char **pat_p, STRLEN *plen_p, int num_code_blocks) |
| 6696 | { |
| 6697 | U8 *const src = (U8*)*pat_p; |
| 6698 | U8 *dst, *d; |
| 6699 | int n=0; |
| 6700 | STRLEN s = 0; |
| 6701 | bool do_end = 0; |
| 6702 | GET_RE_DEBUG_FLAGS_DECL; |
| 6703 | |
| 6704 | DEBUG_PARSE_r(Perl_re_printf( aTHX_ |
| 6705 | "UTF8 mismatch! Converting to utf8 for resizing and compile\n")); |
| 6706 | |
| 6707 | /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */ |
| 6708 | Newx(dst, *plen_p + variant_under_utf8_count(src, src + *plen_p) + 1, U8); |
| 6709 | d = dst; |
| 6710 | |
| 6711 | while (s < *plen_p) { |
| 6712 | append_utf8_from_native_byte(src[s], &d); |
| 6713 | |
| 6714 | if (n < num_code_blocks) { |
| 6715 | assert(pRExC_state->code_blocks); |
| 6716 | if (!do_end && pRExC_state->code_blocks->cb[n].start == s) { |
| 6717 | pRExC_state->code_blocks->cb[n].start = d - dst - 1; |
| 6718 | assert(*(d - 1) == '('); |
| 6719 | do_end = 1; |
| 6720 | } |
| 6721 | else if (do_end && pRExC_state->code_blocks->cb[n].end == s) { |
| 6722 | pRExC_state->code_blocks->cb[n].end = d - dst - 1; |
| 6723 | assert(*(d - 1) == ')'); |
| 6724 | do_end = 0; |
| 6725 | n++; |
| 6726 | } |
| 6727 | } |
| 6728 | s++; |
| 6729 | } |
| 6730 | *d = '\0'; |
| 6731 | *plen_p = d - dst; |
| 6732 | *pat_p = (char*) dst; |
| 6733 | SAVEFREEPV(*pat_p); |
| 6734 | RExC_orig_utf8 = RExC_utf8 = 1; |
| 6735 | } |
| 6736 | |
| 6737 | |
| 6738 | |
| 6739 | /* S_concat_pat(): concatenate a list of args to the pattern string pat, |
| 6740 | * while recording any code block indices, and handling overloading, |
| 6741 | * nested qr// objects etc. If pat is null, it will allocate a new |
| 6742 | * string, or just return the first arg, if there's only one. |
| 6743 | * |
| 6744 | * Returns the malloced/updated pat. |
| 6745 | * patternp and pat_count is the array of SVs to be concatted; |
| 6746 | * oplist is the optional list of ops that generated the SVs; |
| 6747 | * recompile_p is a pointer to a boolean that will be set if |
| 6748 | * the regex will need to be recompiled. |
| 6749 | * delim, if non-null is an SV that will be inserted between each element |
| 6750 | */ |
| 6751 | |
| 6752 | static SV* |
| 6753 | S_concat_pat(pTHX_ RExC_state_t * const pRExC_state, |
| 6754 | SV *pat, SV ** const patternp, int pat_count, |
| 6755 | OP *oplist, bool *recompile_p, SV *delim) |
| 6756 | { |
| 6757 | SV **svp; |
| 6758 | int n = 0; |
| 6759 | bool use_delim = FALSE; |
| 6760 | bool alloced = FALSE; |
| 6761 | |
| 6762 | /* if we know we have at least two args, create an empty string, |
| 6763 | * then concatenate args to that. For no args, return an empty string */ |
| 6764 | if (!pat && pat_count != 1) { |
| 6765 | pat = newSVpvs(""); |
| 6766 | SAVEFREESV(pat); |
| 6767 | alloced = TRUE; |
| 6768 | } |
| 6769 | |
| 6770 | for (svp = patternp; svp < patternp + pat_count; svp++) { |
| 6771 | SV *sv; |
| 6772 | SV *rx = NULL; |
| 6773 | STRLEN orig_patlen = 0; |
| 6774 | bool code = 0; |
| 6775 | SV *msv = use_delim ? delim : *svp; |
| 6776 | if (!msv) msv = &PL_sv_undef; |
| 6777 | |
| 6778 | /* if we've got a delimiter, we go round the loop twice for each |
| 6779 | * svp slot (except the last), using the delimiter the second |
| 6780 | * time round */ |
| 6781 | if (use_delim) { |
| 6782 | svp--; |
| 6783 | use_delim = FALSE; |
| 6784 | } |
| 6785 | else if (delim) |
| 6786 | use_delim = TRUE; |
| 6787 | |
| 6788 | if (SvTYPE(msv) == SVt_PVAV) { |
| 6789 | /* we've encountered an interpolated array within |
| 6790 | * the pattern, e.g. /...@a..../. Expand the list of elements, |
| 6791 | * then recursively append elements. |
| 6792 | * The code in this block is based on S_pushav() */ |
| 6793 | |
| 6794 | AV *const av = (AV*)msv; |
| 6795 | const SSize_t maxarg = AvFILL(av) + 1; |
| 6796 | SV **array; |
| 6797 | |
| 6798 | if (oplist) { |
| 6799 | assert(oplist->op_type == OP_PADAV |
| 6800 | || oplist->op_type == OP_RV2AV); |
| 6801 | oplist = OpSIBLING(oplist); |
| 6802 | } |
| 6803 | |
| 6804 | if (SvRMAGICAL(av)) { |
| 6805 | SSize_t i; |
| 6806 | |
| 6807 | Newx(array, maxarg, SV*); |
| 6808 | SAVEFREEPV(array); |
| 6809 | for (i=0; i < maxarg; i++) { |
| 6810 | SV ** const svp = av_fetch(av, i, FALSE); |
| 6811 | array[i] = svp ? *svp : &PL_sv_undef; |
| 6812 | } |
| 6813 | } |
| 6814 | else |
| 6815 | array = AvARRAY(av); |
| 6816 | |
| 6817 | pat = S_concat_pat(aTHX_ pRExC_state, pat, |
| 6818 | array, maxarg, NULL, recompile_p, |
| 6819 | /* $" */ |
| 6820 | GvSV((gv_fetchpvs("\"", GV_ADDMULTI, SVt_PV)))); |
| 6821 | |
| 6822 | continue; |
| 6823 | } |
| 6824 | |
| 6825 | |
| 6826 | /* we make the assumption here that each op in the list of |
| 6827 | * op_siblings maps to one SV pushed onto the stack, |
| 6828 | * except for code blocks, with have both an OP_NULL and |
| 6829 | * and OP_CONST. |
| 6830 | * This allows us to match up the list of SVs against the |
| 6831 | * list of OPs to find the next code block. |
| 6832 | * |
| 6833 | * Note that PUSHMARK PADSV PADSV .. |
| 6834 | * is optimised to |
| 6835 | * PADRANGE PADSV PADSV .. |
| 6836 | * so the alignment still works. */ |
| 6837 | |
| 6838 | if (oplist) { |
| 6839 | if (oplist->op_type == OP_NULL |
| 6840 | && (oplist->op_flags & OPf_SPECIAL)) |
| 6841 | { |
| 6842 | assert(n < pRExC_state->code_blocks->count); |
| 6843 | pRExC_state->code_blocks->cb[n].start = pat ? SvCUR(pat) : 0; |
| 6844 | pRExC_state->code_blocks->cb[n].block = oplist; |
| 6845 | pRExC_state->code_blocks->cb[n].src_regex = NULL; |
| 6846 | n++; |
| 6847 | code = 1; |
| 6848 | oplist = OpSIBLING(oplist); /* skip CONST */ |
| 6849 | assert(oplist); |
| 6850 | } |
| 6851 | oplist = OpSIBLING(oplist);; |
| 6852 | } |
| 6853 | |
| 6854 | /* apply magic and QR overloading to arg */ |
| 6855 | |
| 6856 | SvGETMAGIC(msv); |
| 6857 | if (SvROK(msv) && SvAMAGIC(msv)) { |
| 6858 | SV *sv = AMG_CALLunary(msv, regexp_amg); |
| 6859 | if (sv) { |
| 6860 | if (SvROK(sv)) |
| 6861 | sv = SvRV(sv); |
| 6862 | if (SvTYPE(sv) != SVt_REGEXP) |
| 6863 | Perl_croak(aTHX_ "Overloaded qr did not return a REGEXP"); |
| 6864 | msv = sv; |
| 6865 | } |
| 6866 | } |
| 6867 | |
| 6868 | /* try concatenation overload ... */ |
| 6869 | if (pat && (SvAMAGIC(pat) || SvAMAGIC(msv)) && |
| 6870 | (sv = amagic_call(pat, msv, concat_amg, AMGf_assign))) |
| 6871 | { |
| 6872 | sv_setsv(pat, sv); |
| 6873 | /* overloading involved: all bets are off over literal |
| 6874 | * code. Pretend we haven't seen it */ |
| 6875 | if (n) |
| 6876 | pRExC_state->code_blocks->count -= n; |
| 6877 | n = 0; |
| 6878 | } |
| 6879 | else { |
| 6880 | /* ... or failing that, try "" overload */ |
| 6881 | while (SvAMAGIC(msv) |
| 6882 | && (sv = AMG_CALLunary(msv, string_amg)) |
| 6883 | && sv != msv |
| 6884 | && !( SvROK(msv) |
| 6885 | && SvROK(sv) |
| 6886 | && SvRV(msv) == SvRV(sv)) |
| 6887 | ) { |
| 6888 | msv = sv; |
| 6889 | SvGETMAGIC(msv); |
| 6890 | } |
| 6891 | if (SvROK(msv) && SvTYPE(SvRV(msv)) == SVt_REGEXP) |
| 6892 | msv = SvRV(msv); |
| 6893 | |
| 6894 | if (pat) { |
| 6895 | /* this is a partially unrolled |
| 6896 | * sv_catsv_nomg(pat, msv); |
| 6897 | * that allows us to adjust code block indices if |
| 6898 | * needed */ |
| 6899 | STRLEN dlen; |
| 6900 | char *dst = SvPV_force_nomg(pat, dlen); |
| 6901 | orig_patlen = dlen; |
| 6902 | if (SvUTF8(msv) && !SvUTF8(pat)) { |
| 6903 | S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &dst, &dlen, n); |
| 6904 | sv_setpvn(pat, dst, dlen); |
| 6905 | SvUTF8_on(pat); |
| 6906 | } |
| 6907 | sv_catsv_nomg(pat, msv); |
| 6908 | rx = msv; |
| 6909 | } |
| 6910 | else { |
| 6911 | /* We have only one SV to process, but we need to verify |
| 6912 | * it is properly null terminated or we will fail asserts |
| 6913 | * later. In theory we probably shouldn't get such SV's, |
| 6914 | * but if we do we should handle it gracefully. */ |
| 6915 | if ( SvTYPE(msv) != SVt_PV || (SvLEN(msv) > SvCUR(msv) && *(SvEND(msv)) == 0) || SvIsCOW_shared_hash(msv) ) { |
| 6916 | /* not a string, or a string with a trailing null */ |
| 6917 | pat = msv; |
| 6918 | } else { |
| 6919 | /* a string with no trailing null, we need to copy it |
| 6920 | * so it has a trailing null */ |
| 6921 | pat = sv_2mortal(newSVsv(msv)); |
| 6922 | } |
| 6923 | } |
| 6924 | |
| 6925 | if (code) |
| 6926 | pRExC_state->code_blocks->cb[n-1].end = SvCUR(pat)-1; |
| 6927 | } |
| 6928 | |
| 6929 | /* extract any code blocks within any embedded qr//'s */ |
| 6930 | if (rx && SvTYPE(rx) == SVt_REGEXP |
| 6931 | && RX_ENGINE((REGEXP*)rx)->op_comp) |
| 6932 | { |
| 6933 | |
| 6934 | RXi_GET_DECL(ReANY((REGEXP *)rx), ri); |
| 6935 | if (ri->code_blocks && ri->code_blocks->count) { |
| 6936 | int i; |
| 6937 | /* the presence of an embedded qr// with code means |
| 6938 | * we should always recompile: the text of the |
| 6939 | * qr// may not have changed, but it may be a |
| 6940 | * different closure than last time */ |
| 6941 | *recompile_p = 1; |
| 6942 | if (pRExC_state->code_blocks) { |
| 6943 | int new_count = pRExC_state->code_blocks->count |
| 6944 | + ri->code_blocks->count; |
| 6945 | Renew(pRExC_state->code_blocks->cb, |
| 6946 | new_count, struct reg_code_block); |
| 6947 | pRExC_state->code_blocks->count = new_count; |
| 6948 | } |
| 6949 | else |
| 6950 | pRExC_state->code_blocks = S_alloc_code_blocks(aTHX_ |
| 6951 | ri->code_blocks->count); |
| 6952 | |
| 6953 | for (i=0; i < ri->code_blocks->count; i++) { |
| 6954 | struct reg_code_block *src, *dst; |
| 6955 | STRLEN offset = orig_patlen |
| 6956 | + ReANY((REGEXP *)rx)->pre_prefix; |
| 6957 | assert(n < pRExC_state->code_blocks->count); |
| 6958 | src = &ri->code_blocks->cb[i]; |
| 6959 | dst = &pRExC_state->code_blocks->cb[n]; |
| 6960 | dst->start = src->start + offset; |
| 6961 | dst->end = src->end + offset; |
| 6962 | dst->block = src->block; |
| 6963 | dst->src_regex = (REGEXP*) SvREFCNT_inc( (SV*) |
| 6964 | src->src_regex |
| 6965 | ? src->src_regex |
| 6966 | : (REGEXP*)rx); |
| 6967 | n++; |
| 6968 | } |
| 6969 | } |
| 6970 | } |
| 6971 | } |
| 6972 | /* avoid calling magic multiple times on a single element e.g. =~ $qr */ |
| 6973 | if (alloced) |
| 6974 | SvSETMAGIC(pat); |
| 6975 | |
| 6976 | return pat; |
| 6977 | } |
| 6978 | |
| 6979 | |
| 6980 | |
| 6981 | /* see if there are any run-time code blocks in the pattern. |
| 6982 | * False positives are allowed */ |
| 6983 | |
| 6984 | static bool |
| 6985 | S_has_runtime_code(pTHX_ RExC_state_t * const pRExC_state, |
| 6986 | char *pat, STRLEN plen) |
| 6987 | { |
| 6988 | int n = 0; |
| 6989 | STRLEN s; |
| 6990 | |
| 6991 | PERL_UNUSED_CONTEXT; |
| 6992 | |
| 6993 | for (s = 0; s < plen; s++) { |
| 6994 | if ( pRExC_state->code_blocks |
| 6995 | && n < pRExC_state->code_blocks->count |
| 6996 | && s == pRExC_state->code_blocks->cb[n].start) |
| 6997 | { |
| 6998 | s = pRExC_state->code_blocks->cb[n].end; |
| 6999 | n++; |
| 7000 | continue; |
| 7001 | } |
| 7002 | /* TODO ideally should handle [..], (#..), /#.../x to reduce false |
| 7003 | * positives here */ |
| 7004 | if (pat[s] == '(' && s+2 <= plen && pat[s+1] == '?' && |
| 7005 | (pat[s+2] == '{' |
| 7006 | || (s + 2 <= plen && pat[s+2] == '?' && pat[s+3] == '{')) |
| 7007 | ) |
| 7008 | return 1; |
| 7009 | } |
| 7010 | return 0; |
| 7011 | } |
| 7012 | |
| 7013 | /* Handle run-time code blocks. We will already have compiled any direct |
| 7014 | * or indirect literal code blocks. Now, take the pattern 'pat' and make a |
| 7015 | * copy of it, but with any literal code blocks blanked out and |
| 7016 | * appropriate chars escaped; then feed it into |
| 7017 | * |
| 7018 | * eval "qr'modified_pattern'" |
| 7019 | * |
| 7020 | * For example, |
| 7021 | * |
| 7022 | * a\bc(?{"this was literal"})def'ghi\\jkl(?{"this is runtime"})mno |
| 7023 | * |
| 7024 | * becomes |
| 7025 | * |
| 7026 | * qr'a\\bc_______________________def\'ghi\\\\jkl(?{"this is runtime"})mno' |
| 7027 | * |
| 7028 | * After eval_sv()-ing that, grab any new code blocks from the returned qr |
| 7029 | * and merge them with any code blocks of the original regexp. |
| 7030 | * |
| 7031 | * If the pat is non-UTF8, while the evalled qr is UTF8, don't merge; |
| 7032 | * instead, just save the qr and return FALSE; this tells our caller that |
| 7033 | * the original pattern needs upgrading to utf8. |
| 7034 | */ |
| 7035 | |
| 7036 | static bool |
| 7037 | S_compile_runtime_code(pTHX_ RExC_state_t * const pRExC_state, |
| 7038 | char *pat, STRLEN plen) |
| 7039 | { |
| 7040 | SV *qr; |
| 7041 | |
| 7042 | GET_RE_DEBUG_FLAGS_DECL; |
| 7043 | |
| 7044 | if (pRExC_state->runtime_code_qr) { |
| 7045 | /* this is the second time we've been called; this should |
| 7046 | * only happen if the main pattern got upgraded to utf8 |
| 7047 | * during compilation; re-use the qr we compiled first time |
| 7048 | * round (which should be utf8 too) |
| 7049 | */ |
| 7050 | qr = pRExC_state->runtime_code_qr; |
| 7051 | pRExC_state->runtime_code_qr = NULL; |
| 7052 | assert(RExC_utf8 && SvUTF8(qr)); |
| 7053 | } |
| 7054 | else { |
| 7055 | int n = 0; |
| 7056 | STRLEN s; |
| 7057 | char *p, *newpat; |
| 7058 | int newlen = plen + 7; /* allow for "qr''xx\0" extra chars */ |
| 7059 | SV *sv, *qr_ref; |
| 7060 | dSP; |
| 7061 | |
| 7062 | /* determine how many extra chars we need for ' and \ escaping */ |
| 7063 | for (s = 0; s < plen; s++) { |
| 7064 | if (pat[s] == '\'' || pat[s] == '\\') |
| 7065 | newlen++; |
| 7066 | } |
| 7067 | |
| 7068 | Newx(newpat, newlen, char); |
| 7069 | p = newpat; |
| 7070 | *p++ = 'q'; *p++ = 'r'; *p++ = '\''; |
| 7071 | |
| 7072 | for (s = 0; s < plen; s++) { |
| 7073 | if ( pRExC_state->code_blocks |
| 7074 | && n < pRExC_state->code_blocks->count |
| 7075 | && s == pRExC_state->code_blocks->cb[n].start) |
| 7076 | { |
| 7077 | /* blank out literal code block so that they aren't |
| 7078 | * recompiled: eg change from/to: |
| 7079 | * /(?{xyz})/ |
| 7080 | * /(?=====)/ |
| 7081 | * and |
| 7082 | * /(??{xyz})/ |
| 7083 | * /(?======)/ |
| 7084 | * and |
| 7085 | * /(?(?{xyz}))/ |
| 7086 | * /(?(?=====))/ |
| 7087 | */ |
| 7088 | assert(pat[s] == '('); |
| 7089 | assert(pat[s+1] == '?'); |
| 7090 | *p++ = '('; |
| 7091 | *p++ = '?'; |
| 7092 | s += 2; |
| 7093 | while (s < pRExC_state->code_blocks->cb[n].end) { |
| 7094 | *p++ = '='; |
| 7095 | s++; |
| 7096 | } |
| 7097 | *p++ = ')'; |
| 7098 | n++; |
| 7099 | continue; |
| 7100 | } |
| 7101 | if (pat[s] == '\'' || pat[s] == '\\') |
| 7102 | *p++ = '\\'; |
| 7103 | *p++ = pat[s]; |
| 7104 | } |
| 7105 | *p++ = '\''; |
| 7106 | if (pRExC_state->pm_flags & RXf_PMf_EXTENDED) { |
| 7107 | *p++ = 'x'; |
| 7108 | if (pRExC_state->pm_flags & RXf_PMf_EXTENDED_MORE) { |
| 7109 | *p++ = 'x'; |
| 7110 | } |
| 7111 | } |
| 7112 | *p++ = '\0'; |
| 7113 | DEBUG_COMPILE_r({ |
| 7114 | Perl_re_printf( aTHX_ |
| 7115 | "%sre-parsing pattern for runtime code:%s %s\n", |
| 7116 | PL_colors[4], PL_colors[5], newpat); |
| 7117 | }); |
| 7118 | |
| 7119 | sv = newSVpvn_flags(newpat, p-newpat-1, RExC_utf8 ? SVf_UTF8 : 0); |
| 7120 | Safefree(newpat); |
| 7121 | |
| 7122 | ENTER; |
| 7123 | SAVETMPS; |
| 7124 | save_re_context(); |
| 7125 | PUSHSTACKi(PERLSI_REQUIRE); |
| 7126 | /* G_RE_REPARSING causes the toker to collapse \\ into \ when |
| 7127 | * parsing qr''; normally only q'' does this. It also alters |
| 7128 | * hints handling */ |
| 7129 | eval_sv(sv, G_SCALAR|G_RE_REPARSING); |
| 7130 | SvREFCNT_dec_NN(sv); |
| 7131 | SPAGAIN; |
| 7132 | qr_ref = POPs; |
| 7133 | PUTBACK; |
| 7134 | { |
| 7135 | SV * const errsv = ERRSV; |
| 7136 | if (SvTRUE_NN(errsv)) |
| 7137 | /* use croak_sv ? */ |
| 7138 | Perl_croak_nocontext("%" SVf, SVfARG(errsv)); |
| 7139 | } |
| 7140 | assert(SvROK(qr_ref)); |
| 7141 | qr = SvRV(qr_ref); |
| 7142 | assert(SvTYPE(qr) == SVt_REGEXP && RX_ENGINE((REGEXP*)qr)->op_comp); |
| 7143 | /* the leaving below frees the tmp qr_ref. |
| 7144 | * Give qr a life of its own */ |
| 7145 | SvREFCNT_inc(qr); |
| 7146 | POPSTACK; |
| 7147 | FREETMPS; |
| 7148 | LEAVE; |
| 7149 | |
| 7150 | } |
| 7151 | |
| 7152 | if (!RExC_utf8 && SvUTF8(qr)) { |
| 7153 | /* first time through; the pattern got upgraded; save the |
| 7154 | * qr for the next time through */ |
| 7155 | assert(!pRExC_state->runtime_code_qr); |
| 7156 | pRExC_state->runtime_code_qr = qr; |
| 7157 | return 0; |
| 7158 | } |
| 7159 | |
| 7160 | |
| 7161 | /* extract any code blocks within the returned qr// */ |
| 7162 | |
| 7163 | |
| 7164 | /* merge the main (r1) and run-time (r2) code blocks into one */ |
| 7165 | { |
| 7166 | RXi_GET_DECL(ReANY((REGEXP *)qr), r2); |
| 7167 | struct reg_code_block *new_block, *dst; |
| 7168 | RExC_state_t * const r1 = pRExC_state; /* convenient alias */ |
| 7169 | int i1 = 0, i2 = 0; |
| 7170 | int r1c, r2c; |
| 7171 | |
| 7172 | if (!r2->code_blocks || !r2->code_blocks->count) /* we guessed wrong */ |
| 7173 | { |
| 7174 | SvREFCNT_dec_NN(qr); |
| 7175 | return 1; |
| 7176 | } |
| 7177 | |
| 7178 | if (!r1->code_blocks) |
| 7179 | r1->code_blocks = S_alloc_code_blocks(aTHX_ 0); |
| 7180 | |
| 7181 | r1c = r1->code_blocks->count; |
| 7182 | r2c = r2->code_blocks->count; |
| 7183 | |
| 7184 | Newx(new_block, r1c + r2c, struct reg_code_block); |
| 7185 | |
| 7186 | dst = new_block; |
| 7187 | |
| 7188 | while (i1 < r1c || i2 < r2c) { |
| 7189 | struct reg_code_block *src; |
| 7190 | bool is_qr = 0; |
| 7191 | |
| 7192 | if (i1 == r1c) { |
| 7193 | src = &r2->code_blocks->cb[i2++]; |
| 7194 | is_qr = 1; |
| 7195 | } |
| 7196 | else if (i2 == r2c) |
| 7197 | src = &r1->code_blocks->cb[i1++]; |
| 7198 | else if ( r1->code_blocks->cb[i1].start |
| 7199 | < r2->code_blocks->cb[i2].start) |
| 7200 | { |
| 7201 | src = &r1->code_blocks->cb[i1++]; |
| 7202 | assert(src->end < r2->code_blocks->cb[i2].start); |
| 7203 | } |
| 7204 | else { |
| 7205 | assert( r1->code_blocks->cb[i1].start |
| 7206 | > r2->code_blocks->cb[i2].start); |
| 7207 | src = &r2->code_blocks->cb[i2++]; |
| 7208 | is_qr = 1; |
| 7209 | assert(src->end < r1->code_blocks->cb[i1].start); |
| 7210 | } |
| 7211 | |
| 7212 | assert(pat[src->start] == '('); |
| 7213 | assert(pat[src->end] == ')'); |
| 7214 | dst->start = src->start; |
| 7215 | dst->end = src->end; |
| 7216 | dst->block = src->block; |
| 7217 | dst->src_regex = is_qr ? (REGEXP*) SvREFCNT_inc( (SV*) qr) |
| 7218 | : src->src_regex; |
| 7219 | dst++; |
| 7220 | } |
| 7221 | r1->code_blocks->count += r2c; |
| 7222 | Safefree(r1->code_blocks->cb); |
| 7223 | r1->code_blocks->cb = new_block; |
| 7224 | } |
| 7225 | |
| 7226 | SvREFCNT_dec_NN(qr); |
| 7227 | return 1; |
| 7228 | } |
| 7229 | |
| 7230 | |
| 7231 | STATIC bool |
| 7232 | S_setup_longest(pTHX_ RExC_state_t *pRExC_state, |
| 7233 | struct reg_substr_datum *rsd, |
| 7234 | struct scan_data_substrs *sub, |
| 7235 | STRLEN longest_length) |
| 7236 | { |
| 7237 | /* This is the common code for setting up the floating and fixed length |
| 7238 | * string data extracted from Perl_re_op_compile() below. Returns a boolean |
| 7239 | * as to whether succeeded or not */ |
| 7240 | |
| 7241 | I32 t; |
| 7242 | SSize_t ml; |
| 7243 | bool eol = cBOOL(sub->flags & SF_BEFORE_EOL); |
| 7244 | bool meol = cBOOL(sub->flags & SF_BEFORE_MEOL); |
| 7245 | |
| 7246 | if (! (longest_length |
| 7247 | || (eol /* Can't have SEOL and MULTI */ |
| 7248 | && (! meol || (RExC_flags & RXf_PMf_MULTILINE))) |
| 7249 | ) |
| 7250 | /* See comments for join_exact for why REG_UNFOLDED_MULTI_SEEN */ |
| 7251 | || (RExC_seen & REG_UNFOLDED_MULTI_SEEN)) |
| 7252 | { |
| 7253 | return FALSE; |
| 7254 | } |
| 7255 | |
| 7256 | /* copy the information about the longest from the reg_scan_data |
| 7257 | over to the program. */ |
| 7258 | if (SvUTF8(sub->str)) { |
| 7259 | rsd->substr = NULL; |
| 7260 | rsd->utf8_substr = sub->str; |
| 7261 | } else { |
| 7262 | rsd->substr = sub->str; |
| 7263 | rsd->utf8_substr = NULL; |
| 7264 | } |
| 7265 | /* end_shift is how many chars that must be matched that |
| 7266 | follow this item. We calculate it ahead of time as once the |
| 7267 | lookbehind offset is added in we lose the ability to correctly |
| 7268 | calculate it.*/ |
| 7269 | ml = sub->minlenp ? *(sub->minlenp) : (SSize_t)longest_length; |
| 7270 | rsd->end_shift = ml - sub->min_offset |
| 7271 | - longest_length |
| 7272 | /* XXX SvTAIL is always false here - did you mean FBMcf_TAIL |
| 7273 | * intead? - DAPM |
| 7274 | + (SvTAIL(sub->str) != 0) |
| 7275 | */ |
| 7276 | + sub->lookbehind; |
| 7277 | |
| 7278 | t = (eol/* Can't have SEOL and MULTI */ |
| 7279 | && (! meol || (RExC_flags & RXf_PMf_MULTILINE))); |
| 7280 | fbm_compile(sub->str, t ? FBMcf_TAIL : 0); |
| 7281 | |
| 7282 | return TRUE; |
| 7283 | } |
| 7284 | |
| 7285 | STATIC void |
| 7286 | S_set_regex_pv(pTHX_ RExC_state_t *pRExC_state, REGEXP *Rx) |
| 7287 | { |
| 7288 | /* Calculates and sets in the compiled pattern 'Rx' the string to compile, |
| 7289 | * properly wrapped with the right modifiers */ |
| 7290 | |
| 7291 | bool has_p = ((RExC_rx->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY); |
| 7292 | bool has_charset = RExC_utf8 || (get_regex_charset(RExC_rx->extflags) |
| 7293 | != REGEX_DEPENDS_CHARSET); |
| 7294 | |
| 7295 | /* The caret is output if there are any defaults: if not all the STD |
| 7296 | * flags are set, or if no character set specifier is needed */ |
| 7297 | bool has_default = |
| 7298 | (((RExC_rx->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD) |
| 7299 | || ! has_charset); |
| 7300 | bool has_runon = ((RExC_seen & REG_RUN_ON_COMMENT_SEEN) |
| 7301 | == REG_RUN_ON_COMMENT_SEEN); |
| 7302 | U8 reganch = (U8)((RExC_rx->extflags & RXf_PMf_STD_PMMOD) |
| 7303 | >> RXf_PMf_STD_PMMOD_SHIFT); |
| 7304 | const char *fptr = STD_PAT_MODS; /*"msixxn"*/ |
| 7305 | char *p; |
| 7306 | STRLEN pat_len = RExC_precomp_end - RExC_precomp; |
| 7307 | |
| 7308 | /* We output all the necessary flags; we never output a minus, as all |
| 7309 | * those are defaults, so are |
| 7310 | * covered by the caret */ |
| 7311 | const STRLEN wraplen = pat_len + has_p + has_runon |
| 7312 | + has_default /* If needs a caret */ |
| 7313 | + PL_bitcount[reganch] /* 1 char for each set standard flag */ |
| 7314 | |
| 7315 | /* If needs a character set specifier */ |
| 7316 | + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0) |
| 7317 | + (sizeof("(?:)") - 1); |
| 7318 | |
| 7319 | PERL_ARGS_ASSERT_SET_REGEX_PV; |
| 7320 | |
| 7321 | /* make sure PL_bitcount bounds not exceeded */ |
| 7322 | assert(sizeof(STD_PAT_MODS) <= 8); |
| 7323 | |
| 7324 | p = sv_grow(MUTABLE_SV(Rx), wraplen + 1); /* +1 for the ending NUL */ |
| 7325 | SvPOK_on(Rx); |
| 7326 | if (RExC_utf8) |
| 7327 | SvFLAGS(Rx) |= SVf_UTF8; |
| 7328 | *p++='('; *p++='?'; |
| 7329 | |
| 7330 | /* If a default, cover it using the caret */ |
| 7331 | if (has_default) { |
| 7332 | *p++= DEFAULT_PAT_MOD; |
| 7333 | } |
| 7334 | if (has_charset) { |
| 7335 | STRLEN len; |
| 7336 | const char* name; |
| 7337 | |
| 7338 | name = get_regex_charset_name(RExC_rx->extflags, &len); |
| 7339 | if (strEQ(name, DEPENDS_PAT_MODS)) { /* /d under UTF-8 => /u */ |
| 7340 | assert(RExC_utf8); |
| 7341 | name = UNICODE_PAT_MODS; |
| 7342 | len = sizeof(UNICODE_PAT_MODS) - 1; |
| 7343 | } |
| 7344 | Copy(name, p, len, char); |
| 7345 | p += len; |
| 7346 | } |
| 7347 | if (has_p) |
| 7348 | *p++ = KEEPCOPY_PAT_MOD; /*'p'*/ |
| 7349 | { |
| 7350 | char ch; |
| 7351 | while((ch = *fptr++)) { |
| 7352 | if(reganch & 1) |
| 7353 | *p++ = ch; |
| 7354 | reganch >>= 1; |
| 7355 | } |
| 7356 | } |
| 7357 | |
| 7358 | *p++ = ':'; |
| 7359 | Copy(RExC_precomp, p, pat_len, char); |
| 7360 | assert ((RX_WRAPPED(Rx) - p) < 16); |
| 7361 | RExC_rx->pre_prefix = p - RX_WRAPPED(Rx); |
| 7362 | p += pat_len; |
| 7363 | |
| 7364 | /* Adding a trailing \n causes this to compile properly: |
| 7365 | my $R = qr / A B C # D E/x; /($R)/ |
| 7366 | Otherwise the parens are considered part of the comment */ |
| 7367 | if (has_runon) |
| 7368 | *p++ = '\n'; |
| 7369 | *p++ = ')'; |
| 7370 | *p = 0; |
| 7371 | SvCUR_set(Rx, p - RX_WRAPPED(Rx)); |
| 7372 | } |
| 7373 | |
| 7374 | /* |
| 7375 | * Perl_re_op_compile - the perl internal RE engine's function to compile a |
| 7376 | * regular expression into internal code. |
| 7377 | * The pattern may be passed either as: |
| 7378 | * a list of SVs (patternp plus pat_count) |
| 7379 | * a list of OPs (expr) |
| 7380 | * If both are passed, the SV list is used, but the OP list indicates |
| 7381 | * which SVs are actually pre-compiled code blocks |
| 7382 | * |
| 7383 | * The SVs in the list have magic and qr overloading applied to them (and |
| 7384 | * the list may be modified in-place with replacement SVs in the latter |
| 7385 | * case). |
| 7386 | * |
| 7387 | * If the pattern hasn't changed from old_re, then old_re will be |
| 7388 | * returned. |
| 7389 | * |
| 7390 | * eng is the current engine. If that engine has an op_comp method, then |
| 7391 | * handle directly (i.e. we assume that op_comp was us); otherwise, just |
| 7392 | * do the initial concatenation of arguments and pass on to the external |
| 7393 | * engine. |
| 7394 | * |
| 7395 | * If is_bare_re is not null, set it to a boolean indicating whether the |
| 7396 | * arg list reduced (after overloading) to a single bare regex which has |
| 7397 | * been returned (i.e. /$qr/). |
| 7398 | * |
| 7399 | * orig_rx_flags contains RXf_* flags. See perlreapi.pod for more details. |
| 7400 | * |
| 7401 | * pm_flags contains the PMf_* flags, typically based on those from the |
| 7402 | * pm_flags field of the related PMOP. Currently we're only interested in |
| 7403 | * PMf_HAS_CV, PMf_IS_QR, PMf_USE_RE_EVAL, PMf_WILDCARD. |
| 7404 | * |
| 7405 | * For many years this code had an initial sizing pass that calculated |
| 7406 | * (sometimes incorrectly, leading to security holes) the size needed for the |
| 7407 | * compiled pattern. That was changed by commit |
| 7408 | * 7c932d07cab18751bfc7515b4320436273a459e2 in 5.29, which reallocs the size, a |
| 7409 | * node at a time, as parsing goes along. Patches welcome to fix any obsolete |
| 7410 | * references to this sizing pass. |
| 7411 | * |
| 7412 | * Now, an initial crude guess as to the size needed is made, based on the |
| 7413 | * length of the pattern. Patches welcome to improve that guess. That amount |
| 7414 | * of space is malloc'd and then immediately freed, and then clawed back node |
| 7415 | * by node. This design is to minimze, to the extent possible, memory churn |
| 7416 | * when doing the the reallocs. |
| 7417 | * |
| 7418 | * A separate parentheses counting pass may be needed in some cases. |
| 7419 | * (Previously the sizing pass did this.) Patches welcome to reduce the number |
| 7420 | * of these cases. |
| 7421 | * |
| 7422 | * The existence of a sizing pass necessitated design decisions that are no |
| 7423 | * longer needed. There are potential areas of simplification. |
| 7424 | * |
| 7425 | * Beware that the optimization-preparation code in here knows about some |
| 7426 | * of the structure of the compiled regexp. [I'll say.] |
| 7427 | */ |
| 7428 | |
| 7429 | REGEXP * |
| 7430 | Perl_re_op_compile(pTHX_ SV ** const patternp, int pat_count, |
| 7431 | OP *expr, const regexp_engine* eng, REGEXP *old_re, |
| 7432 | bool *is_bare_re, const U32 orig_rx_flags, const U32 pm_flags) |
| 7433 | { |
| 7434 | dVAR; |
| 7435 | REGEXP *Rx; /* Capital 'R' means points to a REGEXP */ |
| 7436 | STRLEN plen; |
| 7437 | char *exp; |
| 7438 | regnode *scan; |
| 7439 | I32 flags; |
| 7440 | SSize_t minlen = 0; |
| 7441 | U32 rx_flags; |
| 7442 | SV *pat; |
| 7443 | SV** new_patternp = patternp; |
| 7444 | |
| 7445 | /* these are all flags - maybe they should be turned |
| 7446 | * into a single int with different bit masks */ |
| 7447 | I32 sawlookahead = 0; |
| 7448 | I32 sawplus = 0; |
| 7449 | I32 sawopen = 0; |
| 7450 | I32 sawminmod = 0; |
| 7451 | |
| 7452 | regex_charset initial_charset = get_regex_charset(orig_rx_flags); |
| 7453 | bool recompile = 0; |
| 7454 | bool runtime_code = 0; |
| 7455 | scan_data_t data; |
| 7456 | RExC_state_t RExC_state; |
| 7457 | RExC_state_t * const pRExC_state = &RExC_state; |
| 7458 | #ifdef TRIE_STUDY_OPT |
| 7459 | int restudied = 0; |
| 7460 | RExC_state_t copyRExC_state; |
| 7461 | #endif |
| 7462 | GET_RE_DEBUG_FLAGS_DECL; |
| 7463 | |
| 7464 | PERL_ARGS_ASSERT_RE_OP_COMPILE; |
| 7465 | |
| 7466 | DEBUG_r(if (!PL_colorset) reginitcolors()); |
| 7467 | |
| 7468 | |
| 7469 | pRExC_state->warn_text = NULL; |
| 7470 | pRExC_state->unlexed_names = NULL; |
| 7471 | pRExC_state->code_blocks = NULL; |
| 7472 | |
| 7473 | if (is_bare_re) |
| 7474 | *is_bare_re = FALSE; |
| 7475 | |
| 7476 | if (expr && (expr->op_type == OP_LIST || |
| 7477 | (expr->op_type == OP_NULL && expr->op_targ == OP_LIST))) { |
| 7478 | /* allocate code_blocks if needed */ |
| 7479 | OP *o; |
| 7480 | int ncode = 0; |
| 7481 | |
| 7482 | for (o = cLISTOPx(expr)->op_first; o; o = OpSIBLING(o)) |
| 7483 | if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL)) |
| 7484 | ncode++; /* count of DO blocks */ |
| 7485 | |
| 7486 | if (ncode) |
| 7487 | pRExC_state->code_blocks = S_alloc_code_blocks(aTHX_ ncode); |
| 7488 | } |
| 7489 | |
| 7490 | if (!pat_count) { |
| 7491 | /* compile-time pattern with just OP_CONSTs and DO blocks */ |
| 7492 | |
| 7493 | int n; |
| 7494 | OP *o; |
| 7495 | |
| 7496 | /* find how many CONSTs there are */ |
| 7497 | assert(expr); |
| 7498 | n = 0; |
| 7499 | if (expr->op_type == OP_CONST) |
| 7500 | n = 1; |
| 7501 | else |
| 7502 | for (o = cLISTOPx(expr)->op_first; o; o = OpSIBLING(o)) { |
| 7503 | if (o->op_type == OP_CONST) |
| 7504 | n++; |
| 7505 | } |
| 7506 | |
| 7507 | /* fake up an SV array */ |
| 7508 | |
| 7509 | assert(!new_patternp); |
| 7510 | Newx(new_patternp, n, SV*); |
| 7511 | SAVEFREEPV(new_patternp); |
| 7512 | pat_count = n; |
| 7513 | |
| 7514 | n = 0; |
| 7515 | if (expr->op_type == OP_CONST) |
| 7516 | new_patternp[n] = cSVOPx_sv(expr); |
| 7517 | else |
| 7518 | for (o = cLISTOPx(expr)->op_first; o; o = OpSIBLING(o)) { |
| 7519 | if (o->op_type == OP_CONST) |
| 7520 | new_patternp[n++] = cSVOPo_sv; |
| 7521 | } |
| 7522 | |
| 7523 | } |
| 7524 | |
| 7525 | DEBUG_PARSE_r(Perl_re_printf( aTHX_ |
| 7526 | "Assembling pattern from %d elements%s\n", pat_count, |
| 7527 | orig_rx_flags & RXf_SPLIT ? " for split" : "")); |
| 7528 | |
| 7529 | /* set expr to the first arg op */ |
| 7530 | |
| 7531 | if (pRExC_state->code_blocks && pRExC_state->code_blocks->count |
| 7532 | && expr->op_type != OP_CONST) |
| 7533 | { |
| 7534 | expr = cLISTOPx(expr)->op_first; |
| 7535 | assert( expr->op_type == OP_PUSHMARK |
| 7536 | || (expr->op_type == OP_NULL && expr->op_targ == OP_PUSHMARK) |
| 7537 | || expr->op_type == OP_PADRANGE); |
| 7538 | expr = OpSIBLING(expr); |
| 7539 | } |
| 7540 | |
| 7541 | pat = S_concat_pat(aTHX_ pRExC_state, NULL, new_patternp, pat_count, |
| 7542 | expr, &recompile, NULL); |
| 7543 | |
| 7544 | /* handle bare (possibly after overloading) regex: foo =~ $re */ |
| 7545 | { |
| 7546 | SV *re = pat; |
| 7547 | if (SvROK(re)) |
| 7548 | re = SvRV(re); |
| 7549 | if (SvTYPE(re) == SVt_REGEXP) { |
| 7550 | if (is_bare_re) |
| 7551 | *is_bare_re = TRUE; |
| 7552 | SvREFCNT_inc(re); |
| 7553 | DEBUG_PARSE_r(Perl_re_printf( aTHX_ |
| 7554 | "Precompiled pattern%s\n", |
| 7555 | orig_rx_flags & RXf_SPLIT ? " for split" : "")); |
| 7556 | |
| 7557 | return (REGEXP*)re; |
| 7558 | } |
| 7559 | } |
| 7560 | |
| 7561 | exp = SvPV_nomg(pat, plen); |
| 7562 | |
| 7563 | if (!eng->op_comp) { |
| 7564 | if ((SvUTF8(pat) && IN_BYTES) |
| 7565 | || SvGMAGICAL(pat) || SvAMAGIC(pat)) |
| 7566 | { |
| 7567 | /* make a temporary copy; either to convert to bytes, |
| 7568 | * or to avoid repeating get-magic / overloaded stringify */ |
| 7569 | pat = newSVpvn_flags(exp, plen, SVs_TEMP | |
| 7570 | (IN_BYTES ? 0 : SvUTF8(pat))); |
| 7571 | } |
| 7572 | return CALLREGCOMP_ENG(eng, pat, orig_rx_flags); |
| 7573 | } |
| 7574 | |
| 7575 | /* ignore the utf8ness if the pattern is 0 length */ |
| 7576 | RExC_utf8 = RExC_orig_utf8 = (plen == 0 || IN_BYTES) ? 0 : SvUTF8(pat); |
| 7577 | RExC_uni_semantics = 0; |
| 7578 | RExC_contains_locale = 0; |
| 7579 | RExC_strict = cBOOL(pm_flags & RXf_PMf_STRICT); |
| 7580 | RExC_in_script_run = 0; |
| 7581 | RExC_study_started = 0; |
| 7582 | pRExC_state->runtime_code_qr = NULL; |
| 7583 | RExC_frame_head= NULL; |
| 7584 | RExC_frame_last= NULL; |
| 7585 | RExC_frame_count= 0; |
| 7586 | RExC_latest_warn_offset = 0; |
| 7587 | RExC_use_BRANCHJ = 0; |
| 7588 | RExC_warned_WARN_EXPERIMENTAL__VLB = 0; |
| 7589 | RExC_warned_WARN_EXPERIMENTAL__REGEX_SETS = 0; |
| 7590 | RExC_total_parens = 0; |
| 7591 | RExC_open_parens = NULL; |
| 7592 | RExC_close_parens = NULL; |
| 7593 | RExC_paren_names = NULL; |
| 7594 | RExC_size = 0; |
| 7595 | RExC_seen_d_op = FALSE; |
| 7596 | #ifdef DEBUGGING |
| 7597 | RExC_paren_name_list = NULL; |
| 7598 | #endif |
| 7599 | |
| 7600 | DEBUG_r({ |
| 7601 | RExC_mysv1= sv_newmortal(); |
| 7602 | RExC_mysv2= sv_newmortal(); |
| 7603 | }); |
| 7604 | |
| 7605 | DEBUG_COMPILE_r({ |
| 7606 | SV *dsv= sv_newmortal(); |
| 7607 | RE_PV_QUOTED_DECL(s, RExC_utf8, dsv, exp, plen, PL_dump_re_max_len); |
| 7608 | Perl_re_printf( aTHX_ "%sCompiling REx%s %s\n", |
| 7609 | PL_colors[4], PL_colors[5], s); |
| 7610 | }); |
| 7611 | |
| 7612 | /* we jump here if we have to recompile, e.g., from upgrading the pattern |
| 7613 | * to utf8 */ |
| 7614 | |
| 7615 | if ((pm_flags & PMf_USE_RE_EVAL) |
| 7616 | /* this second condition covers the non-regex literal case, |
| 7617 | * i.e. $foo =~ '(?{})'. */ |
| 7618 | || (IN_PERL_COMPILETIME && (PL_hints & HINT_RE_EVAL)) |
| 7619 | ) |
| 7620 | runtime_code = S_has_runtime_code(aTHX_ pRExC_state, exp, plen); |
| 7621 | |
| 7622 | redo_parse: |
| 7623 | /* return old regex if pattern hasn't changed */ |
| 7624 | /* XXX: note in the below we have to check the flags as well as the |
| 7625 | * pattern. |
| 7626 | * |
| 7627 | * Things get a touch tricky as we have to compare the utf8 flag |
| 7628 | * independently from the compile flags. */ |
| 7629 | |
| 7630 | if ( old_re |
| 7631 | && !recompile |
| 7632 | && !!RX_UTF8(old_re) == !!RExC_utf8 |
| 7633 | && ( RX_COMPFLAGS(old_re) == ( orig_rx_flags & RXf_PMf_FLAGCOPYMASK ) ) |
| 7634 | && RX_PRECOMP(old_re) |
| 7635 | && RX_PRELEN(old_re) == plen |
| 7636 | && memEQ(RX_PRECOMP(old_re), exp, plen) |
| 7637 | && !runtime_code /* with runtime code, always recompile */ ) |
| 7638 | { |
| 7639 | DEBUG_COMPILE_r({ |
| 7640 | SV *dsv= sv_newmortal(); |
| 7641 | RE_PV_QUOTED_DECL(s, RExC_utf8, dsv, exp, plen, PL_dump_re_max_len); |
| 7642 | Perl_re_printf( aTHX_ "%sSkipping recompilation of unchanged REx%s %s\n", |
| 7643 | PL_colors[4], PL_colors[5], s); |
| 7644 | }); |
| 7645 | return old_re; |
| 7646 | } |
| 7647 | |
| 7648 | /* Allocate the pattern's SV */ |
| 7649 | RExC_rx_sv = Rx = (REGEXP*) newSV_type(SVt_REGEXP); |
| 7650 | RExC_rx = ReANY(Rx); |
| 7651 | if ( RExC_rx == NULL ) |
| 7652 | FAIL("Regexp out of space"); |
| 7653 | |
| 7654 | rx_flags = orig_rx_flags; |
| 7655 | |
| 7656 | if ( (UTF || RExC_uni_semantics) |
| 7657 | && initial_charset == REGEX_DEPENDS_CHARSET) |
| 7658 | { |
| 7659 | |
| 7660 | /* Set to use unicode semantics if the pattern is in utf8 and has the |
| 7661 | * 'depends' charset specified, as it means unicode when utf8 */ |
| 7662 | set_regex_charset(&rx_flags, REGEX_UNICODE_CHARSET); |
| 7663 | RExC_uni_semantics = 1; |
| 7664 | } |
| 7665 | |
| 7666 | RExC_pm_flags = pm_flags; |
| 7667 | |
| 7668 | if (runtime_code) { |
| 7669 | assert(TAINTING_get || !TAINT_get); |
| 7670 | if (TAINT_get) |
| 7671 | Perl_croak(aTHX_ "Eval-group in insecure regular expression"); |
| 7672 | |
| 7673 | if (!S_compile_runtime_code(aTHX_ pRExC_state, exp, plen)) { |
| 7674 | /* whoops, we have a non-utf8 pattern, whilst run-time code |
| 7675 | * got compiled as utf8. Try again with a utf8 pattern */ |
| 7676 | S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen, |
| 7677 | pRExC_state->code_blocks ? pRExC_state->code_blocks->count : 0); |
| 7678 | goto redo_parse; |
| 7679 | } |
| 7680 | } |
| 7681 | assert(!pRExC_state->runtime_code_qr); |
| 7682 | |
| 7683 | RExC_sawback = 0; |
| 7684 | |
| 7685 | RExC_seen = 0; |
| 7686 | RExC_maxlen = 0; |
| 7687 | RExC_in_lookbehind = 0; |
| 7688 | RExC_in_lookahead = 0; |
| 7689 | RExC_seen_zerolen = *exp == '^' ? -1 : 0; |
| 7690 | RExC_recode_x_to_native = 0; |
| 7691 | RExC_in_multi_char_class = 0; |
| 7692 | |
| 7693 | RExC_start = RExC_copy_start_in_constructed = RExC_copy_start_in_input = RExC_precomp = exp; |
| 7694 | RExC_precomp_end = RExC_end = exp + plen; |
| 7695 | RExC_nestroot = 0; |
| 7696 | RExC_whilem_seen = 0; |
| 7697 | RExC_end_op = NULL; |
| 7698 | RExC_recurse = NULL; |
| 7699 | RExC_study_chunk_recursed = NULL; |
| 7700 | RExC_study_chunk_recursed_bytes= 0; |
| 7701 | RExC_recurse_count = 0; |
| 7702 | RExC_sets_depth = 0; |
| 7703 | pRExC_state->code_index = 0; |
| 7704 | |
| 7705 | /* Initialize the string in the compiled pattern. This is so that there is |
| 7706 | * something to output if necessary */ |
| 7707 | set_regex_pv(pRExC_state, Rx); |
| 7708 | |
| 7709 | DEBUG_PARSE_r({ |
| 7710 | Perl_re_printf( aTHX_ |
| 7711 | "Starting parse and generation\n"); |
| 7712 | RExC_lastnum=0; |
| 7713 | RExC_lastparse=NULL; |
| 7714 | }); |
| 7715 | |
| 7716 | /* Allocate space and zero-initialize. Note, the two step process |
| 7717 | of zeroing when in debug mode, thus anything assigned has to |
| 7718 | happen after that */ |
| 7719 | if (! RExC_size) { |
| 7720 | |
| 7721 | /* On the first pass of the parse, we guess how big this will be. Then |
| 7722 | * we grow in one operation to that amount and then give it back. As |
| 7723 | * we go along, we re-allocate what we need. |
| 7724 | * |
| 7725 | * XXX Currently the guess is essentially that the pattern will be an |
| 7726 | * EXACT node with one byte input, one byte output. This is crude, and |
| 7727 | * better heuristics are welcome. |
| 7728 | * |
| 7729 | * On any subsequent passes, we guess what we actually computed in the |
| 7730 | * latest earlier pass. Such a pass probably didn't complete so is |
| 7731 | * missing stuff. We could improve those guesses by knowing where the |
| 7732 | * parse stopped, and use the length so far plus apply the above |
| 7733 | * assumption to what's left. */ |
| 7734 | RExC_size = STR_SZ(RExC_end - RExC_start); |
| 7735 | } |
| 7736 | |
| 7737 | Newxc(RExC_rxi, sizeof(regexp_internal) + RExC_size, char, regexp_internal); |
| 7738 | if ( RExC_rxi == NULL ) |
| 7739 | FAIL("Regexp out of space"); |
| 7740 | |
| 7741 | Zero(RExC_rxi, sizeof(regexp_internal) + RExC_size, char); |
| 7742 | RXi_SET( RExC_rx, RExC_rxi ); |
| 7743 | |
| 7744 | /* We start from 0 (over from 0 in the case this is a reparse. The first |
| 7745 | * node parsed will give back any excess memory we have allocated so far). |
| 7746 | * */ |
| 7747 | RExC_size = 0; |
| 7748 | |
| 7749 | /* non-zero initialization begins here */ |
| 7750 | RExC_rx->engine= eng; |
| 7751 | RExC_rx->extflags = rx_flags; |
| 7752 | RXp_COMPFLAGS(RExC_rx) = orig_rx_flags & RXf_PMf_FLAGCOPYMASK; |
| 7753 | |
| 7754 | if (pm_flags & PMf_IS_QR) { |
| 7755 | RExC_rxi->code_blocks = pRExC_state->code_blocks; |
| 7756 | if (RExC_rxi->code_blocks) { |
| 7757 | RExC_rxi->code_blocks->refcnt++; |
| 7758 | } |
| 7759 | } |
| 7760 | |
| 7761 | RExC_rx->intflags = 0; |
| 7762 | |
| 7763 | RExC_flags = rx_flags; /* don't let top level (?i) bleed */ |
| 7764 | RExC_parse = exp; |
| 7765 | |
| 7766 | /* This NUL is guaranteed because the pattern comes from an SV*, and the sv |
| 7767 | * code makes sure the final byte is an uncounted NUL. But should this |
| 7768 | * ever not be the case, lots of things could read beyond the end of the |
| 7769 | * buffer: loops like |
| 7770 | * while(isFOO(*RExC_parse)) RExC_parse++; |
| 7771 | * strchr(RExC_parse, "foo"); |
| 7772 | * etc. So it is worth noting. */ |
| 7773 | assert(*RExC_end == '\0'); |
| 7774 | |
| 7775 | RExC_naughty = 0; |
| 7776 | RExC_npar = 1; |
| 7777 | RExC_parens_buf_size = 0; |
| 7778 | RExC_emit_start = RExC_rxi->program; |
| 7779 | pRExC_state->code_index = 0; |
| 7780 | |
| 7781 | *((char*) RExC_emit_start) = (char) REG_MAGIC; |
| 7782 | RExC_emit = 1; |
| 7783 | |
| 7784 | /* Do the parse */ |
| 7785 | if (reg(pRExC_state, 0, &flags, 1)) { |
| 7786 | |
| 7787 | /* Success!, But we may need to redo the parse knowing how many parens |
| 7788 | * there actually are */ |
| 7789 | if (IN_PARENS_PASS) { |
| 7790 | flags |= RESTART_PARSE; |
| 7791 | } |
| 7792 | |
| 7793 | /* We have that number in RExC_npar */ |
| 7794 | RExC_total_parens = RExC_npar; |
| 7795 | } |
| 7796 | else if (! MUST_RESTART(flags)) { |
| 7797 | ReREFCNT_dec(Rx); |
| 7798 | Perl_croak(aTHX_ "panic: reg returned failure to re_op_compile, flags=%#" UVxf, (UV) flags); |
| 7799 | } |
| 7800 | |
| 7801 | /* Here, we either have success, or we have to redo the parse for some reason */ |
| 7802 | if (MUST_RESTART(flags)) { |
| 7803 | |
| 7804 | /* It's possible to write a regexp in ascii that represents Unicode |
| 7805 | codepoints outside of the byte range, such as via \x{100}. If we |
| 7806 | detect such a sequence we have to convert the entire pattern to utf8 |
| 7807 | and then recompile, as our sizing calculation will have been based |
| 7808 | on 1 byte == 1 character, but we will need to use utf8 to encode |
| 7809 | at least some part of the pattern, and therefore must convert the whole |
| 7810 | thing. |
| 7811 | -- dmq */ |
| 7812 | if (flags & NEED_UTF8) { |
| 7813 | |
| 7814 | /* We have stored the offset of the final warning output so far. |
| 7815 | * That must be adjusted. Any variant characters between the start |
| 7816 | * of the pattern and this warning count for 2 bytes in the final, |
| 7817 | * so just add them again */ |
| 7818 | if (UNLIKELY(RExC_latest_warn_offset > 0)) { |
| 7819 | RExC_latest_warn_offset += |
| 7820 | variant_under_utf8_count((U8 *) exp, (U8 *) exp |
| 7821 | + RExC_latest_warn_offset); |
| 7822 | } |
| 7823 | S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen, |
| 7824 | pRExC_state->code_blocks ? pRExC_state->code_blocks->count : 0); |
| 7825 | DEBUG_PARSE_r(Perl_re_printf( aTHX_ "Need to redo parse after upgrade\n")); |
| 7826 | } |
| 7827 | else { |
| 7828 | DEBUG_PARSE_r(Perl_re_printf( aTHX_ "Need to redo parse\n")); |
| 7829 | } |
| 7830 | |
| 7831 | if (ALL_PARENS_COUNTED) { |
| 7832 | /* Make enough room for all the known parens, and zero it */ |
| 7833 | Renew(RExC_open_parens, RExC_total_parens, regnode_offset); |
| 7834 | Zero(RExC_open_parens, RExC_total_parens, regnode_offset); |
| 7835 | RExC_open_parens[0] = 1; /* +1 for REG_MAGIC */ |
| 7836 | |
| 7837 | Renew(RExC_close_parens, RExC_total_parens, regnode_offset); |
| 7838 | Zero(RExC_close_parens, RExC_total_parens, regnode_offset); |
| 7839 | } |
| 7840 | else { /* Parse did not complete. Reinitialize the parentheses |
| 7841 | structures */ |
| 7842 | RExC_total_parens = 0; |
| 7843 | if (RExC_open_parens) { |
| 7844 | Safefree(RExC_open_parens); |
| 7845 | RExC_open_parens = NULL; |
| 7846 | } |
| 7847 | if (RExC_close_parens) { |
| 7848 | Safefree(RExC_close_parens); |
| 7849 | RExC_close_parens = NULL; |
| 7850 | } |
| 7851 | } |
| 7852 | |
| 7853 | /* Clean up what we did in this parse */ |
| 7854 | SvREFCNT_dec_NN(RExC_rx_sv); |
| 7855 | |
| 7856 | goto redo_parse; |
| 7857 | } |
| 7858 | |
| 7859 | /* Here, we have successfully parsed and generated the pattern's program |
| 7860 | * for the regex engine. We are ready to finish things up and look for |
| 7861 | * optimizations. */ |
| 7862 | |
| 7863 | /* Update the string to compile, with correct modifiers, etc */ |
| 7864 | set_regex_pv(pRExC_state, Rx); |
| 7865 | |
| 7866 | RExC_rx->nparens = RExC_total_parens - 1; |
| 7867 | |
| 7868 | /* Uses the upper 4 bits of the FLAGS field, so keep within that size */ |
| 7869 | if (RExC_whilem_seen > 15) |
| 7870 | RExC_whilem_seen = 15; |
| 7871 | |
| 7872 | DEBUG_PARSE_r({ |
| 7873 | Perl_re_printf( aTHX_ |
| 7874 | "Required size %" IVdf " nodes\n", (IV)RExC_size); |
| 7875 | RExC_lastnum=0; |
| 7876 | RExC_lastparse=NULL; |
| 7877 | }); |
| 7878 | |
| 7879 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 7880 | DEBUG_OFFSETS_r(Perl_re_printf( aTHX_ |
| 7881 | "%s %" UVuf " bytes for offset annotations.\n", |
| 7882 | RExC_offsets ? "Got" : "Couldn't get", |
| 7883 | (UV)((RExC_offsets[0] * 2 + 1)))); |
| 7884 | DEBUG_OFFSETS_r(if (RExC_offsets) { |
| 7885 | const STRLEN len = RExC_offsets[0]; |
| 7886 | STRLEN i; |
| 7887 | GET_RE_DEBUG_FLAGS_DECL; |
| 7888 | Perl_re_printf( aTHX_ |
| 7889 | "Offsets: [%" UVuf "]\n\t", (UV)RExC_offsets[0]); |
| 7890 | for (i = 1; i <= len; i++) { |
| 7891 | if (RExC_offsets[i*2-1] || RExC_offsets[i*2]) |
| 7892 | Perl_re_printf( aTHX_ "%" UVuf ":%" UVuf "[%" UVuf "] ", |
| 7893 | (UV)i, (UV)RExC_offsets[i*2-1], (UV)RExC_offsets[i*2]); |
| 7894 | } |
| 7895 | Perl_re_printf( aTHX_ "\n"); |
| 7896 | }); |
| 7897 | |
| 7898 | #else |
| 7899 | SetProgLen(RExC_rxi,RExC_size); |
| 7900 | #endif |
| 7901 | |
| 7902 | DEBUG_DUMP_PRE_OPTIMIZE_r({ |
| 7903 | SV * const sv = sv_newmortal(); |
| 7904 | RXi_GET_DECL(RExC_rx, ri); |
| 7905 | DEBUG_RExC_seen(); |
| 7906 | Perl_re_printf( aTHX_ "Program before optimization:\n"); |
| 7907 | |
| 7908 | (void)dumpuntil(RExC_rx, ri->program, ri->program + 1, NULL, NULL, |
| 7909 | sv, 0, 0); |
| 7910 | }); |
| 7911 | |
| 7912 | DEBUG_OPTIMISE_r( |
| 7913 | Perl_re_printf( aTHX_ "Starting post parse optimization\n"); |
| 7914 | ); |
| 7915 | |
| 7916 | /* XXXX To minimize changes to RE engine we always allocate |
| 7917 | 3-units-long substrs field. */ |
| 7918 | Newx(RExC_rx->substrs, 1, struct reg_substr_data); |
| 7919 | if (RExC_recurse_count) { |
| 7920 | Newx(RExC_recurse, RExC_recurse_count, regnode *); |
| 7921 | SAVEFREEPV(RExC_recurse); |
| 7922 | } |
| 7923 | |
| 7924 | if (RExC_seen & REG_RECURSE_SEEN) { |
| 7925 | /* Note, RExC_total_parens is 1 + the number of parens in a pattern. |
| 7926 | * So its 1 if there are no parens. */ |
| 7927 | RExC_study_chunk_recursed_bytes= (RExC_total_parens >> 3) + |
| 7928 | ((RExC_total_parens & 0x07) != 0); |
| 7929 | Newx(RExC_study_chunk_recursed, |
| 7930 | RExC_study_chunk_recursed_bytes * RExC_total_parens, U8); |
| 7931 | SAVEFREEPV(RExC_study_chunk_recursed); |
| 7932 | } |
| 7933 | |
| 7934 | reStudy: |
| 7935 | RExC_rx->minlen = minlen = sawlookahead = sawplus = sawopen = sawminmod = 0; |
| 7936 | DEBUG_r( |
| 7937 | RExC_study_chunk_recursed_count= 0; |
| 7938 | ); |
| 7939 | Zero(RExC_rx->substrs, 1, struct reg_substr_data); |
| 7940 | if (RExC_study_chunk_recursed) { |
| 7941 | Zero(RExC_study_chunk_recursed, |
| 7942 | RExC_study_chunk_recursed_bytes * RExC_total_parens, U8); |
| 7943 | } |
| 7944 | |
| 7945 | |
| 7946 | #ifdef TRIE_STUDY_OPT |
| 7947 | if (!restudied) { |
| 7948 | StructCopy(&zero_scan_data, &data, scan_data_t); |
| 7949 | copyRExC_state = RExC_state; |
| 7950 | } else { |
| 7951 | U32 seen=RExC_seen; |
| 7952 | DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "Restudying\n")); |
| 7953 | |
| 7954 | RExC_state = copyRExC_state; |
| 7955 | if (seen & REG_TOP_LEVEL_BRANCHES_SEEN) |
| 7956 | RExC_seen |= REG_TOP_LEVEL_BRANCHES_SEEN; |
| 7957 | else |
| 7958 | RExC_seen &= ~REG_TOP_LEVEL_BRANCHES_SEEN; |
| 7959 | StructCopy(&zero_scan_data, &data, scan_data_t); |
| 7960 | } |
| 7961 | #else |
| 7962 | StructCopy(&zero_scan_data, &data, scan_data_t); |
| 7963 | #endif |
| 7964 | |
| 7965 | /* Dig out information for optimizations. */ |
| 7966 | RExC_rx->extflags = RExC_flags; /* was pm_op */ |
| 7967 | /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */ |
| 7968 | |
| 7969 | if (UTF) |
| 7970 | SvUTF8_on(Rx); /* Unicode in it? */ |
| 7971 | RExC_rxi->regstclass = NULL; |
| 7972 | if (RExC_naughty >= TOO_NAUGHTY) /* Probably an expensive pattern. */ |
| 7973 | RExC_rx->intflags |= PREGf_NAUGHTY; |
| 7974 | scan = RExC_rxi->program + 1; /* First BRANCH. */ |
| 7975 | |
| 7976 | /* testing for BRANCH here tells us whether there is "must appear" |
| 7977 | data in the pattern. If there is then we can use it for optimisations */ |
| 7978 | if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN)) { /* Only one top-level choice. |
| 7979 | */ |
| 7980 | SSize_t fake; |
| 7981 | STRLEN longest_length[2]; |
| 7982 | regnode_ssc ch_class; /* pointed to by data */ |
| 7983 | int stclass_flag; |
| 7984 | SSize_t last_close = 0; /* pointed to by data */ |
| 7985 | regnode *first= scan; |
| 7986 | regnode *first_next= regnext(first); |
| 7987 | int i; |
| 7988 | |
| 7989 | /* |
| 7990 | * Skip introductions and multiplicators >= 1 |
| 7991 | * so that we can extract the 'meat' of the pattern that must |
| 7992 | * match in the large if() sequence following. |
| 7993 | * NOTE that EXACT is NOT covered here, as it is normally |
| 7994 | * picked up by the optimiser separately. |
| 7995 | * |
| 7996 | * This is unfortunate as the optimiser isnt handling lookahead |
| 7997 | * properly currently. |
| 7998 | * |
| 7999 | */ |
| 8000 | while ((OP(first) == OPEN && (sawopen = 1)) || |
| 8001 | /* An OR of *one* alternative - should not happen now. */ |
| 8002 | (OP(first) == BRANCH && OP(first_next) != BRANCH) || |
| 8003 | /* for now we can't handle lookbehind IFMATCH*/ |
| 8004 | (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) || |
| 8005 | (OP(first) == PLUS) || |
| 8006 | (OP(first) == MINMOD) || |
| 8007 | /* An {n,m} with n>0 */ |
| 8008 | (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) || |
| 8009 | (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END )) |
| 8010 | { |
| 8011 | /* |
| 8012 | * the only op that could be a regnode is PLUS, all the rest |
| 8013 | * will be regnode_1 or regnode_2. |
| 8014 | * |
| 8015 | * (yves doesn't think this is true) |
| 8016 | */ |
| 8017 | if (OP(first) == PLUS) |
| 8018 | sawplus = 1; |
| 8019 | else { |
| 8020 | if (OP(first) == MINMOD) |
| 8021 | sawminmod = 1; |
| 8022 | first += regarglen[OP(first)]; |
| 8023 | } |
| 8024 | first = NEXTOPER(first); |
| 8025 | first_next= regnext(first); |
| 8026 | } |
| 8027 | |
| 8028 | /* Starting-point info. */ |
| 8029 | again: |
| 8030 | DEBUG_PEEP("first:", first, 0, 0); |
| 8031 | /* Ignore EXACT as we deal with it later. */ |
| 8032 | if (PL_regkind[OP(first)] == EXACT) { |
| 8033 | if ( OP(first) == EXACT |
| 8034 | || OP(first) == LEXACT |
| 8035 | || OP(first) == EXACT_REQ8 |
| 8036 | || OP(first) == LEXACT_REQ8 |
| 8037 | || OP(first) == EXACTL) |
| 8038 | { |
| 8039 | NOOP; /* Empty, get anchored substr later. */ |
| 8040 | } |
| 8041 | else |
| 8042 | RExC_rxi->regstclass = first; |
| 8043 | } |
| 8044 | #ifdef TRIE_STCLASS |
| 8045 | else if (PL_regkind[OP(first)] == TRIE && |
| 8046 | ((reg_trie_data *)RExC_rxi->data->data[ ARG(first) ])->minlen>0) |
| 8047 | { |
| 8048 | /* this can happen only on restudy */ |
| 8049 | RExC_rxi->regstclass = construct_ahocorasick_from_trie(pRExC_state, (regnode *)first, 0); |
| 8050 | } |
| 8051 | #endif |
| 8052 | else if (REGNODE_SIMPLE(OP(first))) |
| 8053 | RExC_rxi->regstclass = first; |
| 8054 | else if (PL_regkind[OP(first)] == BOUND || |
| 8055 | PL_regkind[OP(first)] == NBOUND) |
| 8056 | RExC_rxi->regstclass = first; |
| 8057 | else if (PL_regkind[OP(first)] == BOL) { |
| 8058 | RExC_rx->intflags |= (OP(first) == MBOL |
| 8059 | ? PREGf_ANCH_MBOL |
| 8060 | : PREGf_ANCH_SBOL); |
| 8061 | first = NEXTOPER(first); |
| 8062 | goto again; |
| 8063 | } |
| 8064 | else if (OP(first) == GPOS) { |
| 8065 | RExC_rx->intflags |= PREGf_ANCH_GPOS; |
| 8066 | first = NEXTOPER(first); |
| 8067 | goto again; |
| 8068 | } |
| 8069 | else if ((!sawopen || !RExC_sawback) && |
| 8070 | !sawlookahead && |
| 8071 | (OP(first) == STAR && |
| 8072 | PL_regkind[OP(NEXTOPER(first))] == REG_ANY) && |
| 8073 | !(RExC_rx->intflags & PREGf_ANCH) && !pRExC_state->code_blocks) |
| 8074 | { |
| 8075 | /* turn .* into ^.* with an implied $*=1 */ |
| 8076 | const int type = |
| 8077 | (OP(NEXTOPER(first)) == REG_ANY) |
| 8078 | ? PREGf_ANCH_MBOL |
| 8079 | : PREGf_ANCH_SBOL; |
| 8080 | RExC_rx->intflags |= (type | PREGf_IMPLICIT); |
| 8081 | first = NEXTOPER(first); |
| 8082 | goto again; |
| 8083 | } |
| 8084 | if (sawplus && !sawminmod && !sawlookahead |
| 8085 | && (!sawopen || !RExC_sawback) |
| 8086 | && !pRExC_state->code_blocks) /* May examine pos and $& */ |
| 8087 | /* x+ must match at the 1st pos of run of x's */ |
| 8088 | RExC_rx->intflags |= PREGf_SKIP; |
| 8089 | |
| 8090 | /* Scan is after the zeroth branch, first is atomic matcher. */ |
| 8091 | #ifdef TRIE_STUDY_OPT |
| 8092 | DEBUG_PARSE_r( |
| 8093 | if (!restudied) |
| 8094 | Perl_re_printf( aTHX_ "first at %" IVdf "\n", |
| 8095 | (IV)(first - scan + 1)) |
| 8096 | ); |
| 8097 | #else |
| 8098 | DEBUG_PARSE_r( |
| 8099 | Perl_re_printf( aTHX_ "first at %" IVdf "\n", |
| 8100 | (IV)(first - scan + 1)) |
| 8101 | ); |
| 8102 | #endif |
| 8103 | |
| 8104 | |
| 8105 | /* |
| 8106 | * If there's something expensive in the r.e., find the |
| 8107 | * longest literal string that must appear and make it the |
| 8108 | * regmust. Resolve ties in favor of later strings, since |
| 8109 | * the regstart check works with the beginning of the r.e. |
| 8110 | * and avoiding duplication strengthens checking. Not a |
| 8111 | * strong reason, but sufficient in the absence of others. |
| 8112 | * [Now we resolve ties in favor of the earlier string if |
| 8113 | * it happens that c_offset_min has been invalidated, since the |
| 8114 | * earlier string may buy us something the later one won't.] |
| 8115 | */ |
| 8116 | |
| 8117 | data.substrs[0].str = newSVpvs(""); |
| 8118 | data.substrs[1].str = newSVpvs(""); |
| 8119 | data.last_found = newSVpvs(""); |
| 8120 | data.cur_is_floating = 0; /* initially any found substring is fixed */ |
| 8121 | ENTER_with_name("study_chunk"); |
| 8122 | SAVEFREESV(data.substrs[0].str); |
| 8123 | SAVEFREESV(data.substrs[1].str); |
| 8124 | SAVEFREESV(data.last_found); |
| 8125 | first = scan; |
| 8126 | if (!RExC_rxi->regstclass) { |
| 8127 | ssc_init(pRExC_state, &ch_class); |
| 8128 | data.start_class = &ch_class; |
| 8129 | stclass_flag = SCF_DO_STCLASS_AND; |
| 8130 | } else /* XXXX Check for BOUND? */ |
| 8131 | stclass_flag = 0; |
| 8132 | data.last_closep = &last_close; |
| 8133 | |
| 8134 | DEBUG_RExC_seen(); |
| 8135 | /* |
| 8136 | * MAIN ENTRY FOR study_chunk() FOR m/PATTERN/ |
| 8137 | * (NO top level branches) |
| 8138 | */ |
| 8139 | minlen = study_chunk(pRExC_state, &first, &minlen, &fake, |
| 8140 | scan + RExC_size, /* Up to end */ |
| 8141 | &data, -1, 0, NULL, |
| 8142 | SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag |
| 8143 | | (restudied ? SCF_TRIE_DOING_RESTUDY : 0), |
| 8144 | 0); |
| 8145 | |
| 8146 | |
| 8147 | CHECK_RESTUDY_GOTO_butfirst(LEAVE_with_name("study_chunk")); |
| 8148 | |
| 8149 | |
| 8150 | if ( RExC_total_parens == 1 && !data.cur_is_floating |
| 8151 | && data.last_start_min == 0 && data.last_end > 0 |
| 8152 | && !RExC_seen_zerolen |
| 8153 | && !(RExC_seen & REG_VERBARG_SEEN) |
| 8154 | && !(RExC_seen & REG_GPOS_SEEN) |
| 8155 | ){ |
| 8156 | RExC_rx->extflags |= RXf_CHECK_ALL; |
| 8157 | } |
| 8158 | scan_commit(pRExC_state, &data,&minlen, 0); |
| 8159 | |
| 8160 | |
| 8161 | /* XXX this is done in reverse order because that's the way the |
| 8162 | * code was before it was parameterised. Don't know whether it |
| 8163 | * actually needs doing in reverse order. DAPM */ |
| 8164 | for (i = 1; i >= 0; i--) { |
| 8165 | longest_length[i] = CHR_SVLEN(data.substrs[i].str); |
| 8166 | |
| 8167 | if ( !( i |
| 8168 | && SvCUR(data.substrs[0].str) /* ok to leave SvCUR */ |
| 8169 | && data.substrs[0].min_offset |
| 8170 | == data.substrs[1].min_offset |
| 8171 | && SvCUR(data.substrs[0].str) |
| 8172 | == SvCUR(data.substrs[1].str) |
| 8173 | ) |
| 8174 | && S_setup_longest (aTHX_ pRExC_state, |
| 8175 | &(RExC_rx->substrs->data[i]), |
| 8176 | &(data.substrs[i]), |
| 8177 | longest_length[i])) |
| 8178 | { |
| 8179 | RExC_rx->substrs->data[i].min_offset = |
| 8180 | data.substrs[i].min_offset - data.substrs[i].lookbehind; |
| 8181 | |
| 8182 | RExC_rx->substrs->data[i].max_offset = data.substrs[i].max_offset; |
| 8183 | /* Don't offset infinity */ |
| 8184 | if (data.substrs[i].max_offset < OPTIMIZE_INFTY) |
| 8185 | RExC_rx->substrs->data[i].max_offset -= data.substrs[i].lookbehind; |
| 8186 | SvREFCNT_inc_simple_void_NN(data.substrs[i].str); |
| 8187 | } |
| 8188 | else { |
| 8189 | RExC_rx->substrs->data[i].substr = NULL; |
| 8190 | RExC_rx->substrs->data[i].utf8_substr = NULL; |
| 8191 | longest_length[i] = 0; |
| 8192 | } |
| 8193 | } |
| 8194 | |
| 8195 | LEAVE_with_name("study_chunk"); |
| 8196 | |
| 8197 | if (RExC_rxi->regstclass |
| 8198 | && (OP(RExC_rxi->regstclass) == REG_ANY || OP(RExC_rxi->regstclass) == SANY)) |
| 8199 | RExC_rxi->regstclass = NULL; |
| 8200 | |
| 8201 | if ((!(RExC_rx->substrs->data[0].substr || RExC_rx->substrs->data[0].utf8_substr) |
| 8202 | || RExC_rx->substrs->data[0].min_offset) |
| 8203 | && stclass_flag |
| 8204 | && ! (ANYOF_FLAGS(data.start_class) & SSC_MATCHES_EMPTY_STRING) |
| 8205 | && is_ssc_worth_it(pRExC_state, data.start_class)) |
| 8206 | { |
| 8207 | const U32 n = add_data(pRExC_state, STR_WITH_LEN("f")); |
| 8208 | |
| 8209 | ssc_finalize(pRExC_state, data.start_class); |
| 8210 | |
| 8211 | Newx(RExC_rxi->data->data[n], 1, regnode_ssc); |
| 8212 | StructCopy(data.start_class, |
| 8213 | (regnode_ssc*)RExC_rxi->data->data[n], |
| 8214 | regnode_ssc); |
| 8215 | RExC_rxi->regstclass = (regnode*)RExC_rxi->data->data[n]; |
| 8216 | RExC_rx->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
| 8217 | DEBUG_COMPILE_r({ SV *sv = sv_newmortal(); |
| 8218 | regprop(RExC_rx, sv, (regnode*)data.start_class, NULL, pRExC_state); |
| 8219 | Perl_re_printf( aTHX_ |
| 8220 | "synthetic stclass \"%s\".\n", |
| 8221 | SvPVX_const(sv));}); |
| 8222 | data.start_class = NULL; |
| 8223 | } |
| 8224 | |
| 8225 | /* A temporary algorithm prefers floated substr to fixed one of |
| 8226 | * same length to dig more info. */ |
| 8227 | i = (longest_length[0] <= longest_length[1]); |
| 8228 | RExC_rx->substrs->check_ix = i; |
| 8229 | RExC_rx->check_end_shift = RExC_rx->substrs->data[i].end_shift; |
| 8230 | RExC_rx->check_substr = RExC_rx->substrs->data[i].substr; |
| 8231 | RExC_rx->check_utf8 = RExC_rx->substrs->data[i].utf8_substr; |
| 8232 | RExC_rx->check_offset_min = RExC_rx->substrs->data[i].min_offset; |
| 8233 | RExC_rx->check_offset_max = RExC_rx->substrs->data[i].max_offset; |
| 8234 | if (!i && (RExC_rx->intflags & (PREGf_ANCH_SBOL|PREGf_ANCH_GPOS))) |
| 8235 | RExC_rx->intflags |= PREGf_NOSCAN; |
| 8236 | |
| 8237 | if ((RExC_rx->check_substr || RExC_rx->check_utf8) ) { |
| 8238 | RExC_rx->extflags |= RXf_USE_INTUIT; |
| 8239 | if (SvTAIL(RExC_rx->check_substr ? RExC_rx->check_substr : RExC_rx->check_utf8)) |
| 8240 | RExC_rx->extflags |= RXf_INTUIT_TAIL; |
| 8241 | } |
| 8242 | |
| 8243 | /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere) |
| 8244 | if ( (STRLEN)minlen < longest_length[1] ) |
| 8245 | minlen= longest_length[1]; |
| 8246 | if ( (STRLEN)minlen < longest_length[0] ) |
| 8247 | minlen= longest_length[0]; |
| 8248 | */ |
| 8249 | } |
| 8250 | else { |
| 8251 | /* Several toplevels. Best we can is to set minlen. */ |
| 8252 | SSize_t fake; |
| 8253 | regnode_ssc ch_class; |
| 8254 | SSize_t last_close = 0; |
| 8255 | |
| 8256 | DEBUG_PARSE_r(Perl_re_printf( aTHX_ "\nMulti Top Level\n")); |
| 8257 | |
| 8258 | scan = RExC_rxi->program + 1; |
| 8259 | ssc_init(pRExC_state, &ch_class); |
| 8260 | data.start_class = &ch_class; |
| 8261 | data.last_closep = &last_close; |
| 8262 | |
| 8263 | DEBUG_RExC_seen(); |
| 8264 | /* |
| 8265 | * MAIN ENTRY FOR study_chunk() FOR m/P1|P2|.../ |
| 8266 | * (patterns WITH top level branches) |
| 8267 | */ |
| 8268 | minlen = study_chunk(pRExC_state, |
| 8269 | &scan, &minlen, &fake, scan + RExC_size, &data, -1, 0, NULL, |
| 8270 | SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS|(restudied |
| 8271 | ? SCF_TRIE_DOING_RESTUDY |
| 8272 | : 0), |
| 8273 | 0); |
| 8274 | |
| 8275 | CHECK_RESTUDY_GOTO_butfirst(NOOP); |
| 8276 | |
| 8277 | RExC_rx->check_substr = NULL; |
| 8278 | RExC_rx->check_utf8 = NULL; |
| 8279 | RExC_rx->substrs->data[0].substr = NULL; |
| 8280 | RExC_rx->substrs->data[0].utf8_substr = NULL; |
| 8281 | RExC_rx->substrs->data[1].substr = NULL; |
| 8282 | RExC_rx->substrs->data[1].utf8_substr = NULL; |
| 8283 | |
| 8284 | if (! (ANYOF_FLAGS(data.start_class) & SSC_MATCHES_EMPTY_STRING) |
| 8285 | && is_ssc_worth_it(pRExC_state, data.start_class)) |
| 8286 | { |
| 8287 | const U32 n = add_data(pRExC_state, STR_WITH_LEN("f")); |
| 8288 | |
| 8289 | ssc_finalize(pRExC_state, data.start_class); |
| 8290 | |
| 8291 | Newx(RExC_rxi->data->data[n], 1, regnode_ssc); |
| 8292 | StructCopy(data.start_class, |
| 8293 | (regnode_ssc*)RExC_rxi->data->data[n], |
| 8294 | regnode_ssc); |
| 8295 | RExC_rxi->regstclass = (regnode*)RExC_rxi->data->data[n]; |
| 8296 | RExC_rx->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
| 8297 | DEBUG_COMPILE_r({ SV* sv = sv_newmortal(); |
| 8298 | regprop(RExC_rx, sv, (regnode*)data.start_class, NULL, pRExC_state); |
| 8299 | Perl_re_printf( aTHX_ |
| 8300 | "synthetic stclass \"%s\".\n", |
| 8301 | SvPVX_const(sv));}); |
| 8302 | data.start_class = NULL; |
| 8303 | } |
| 8304 | } |
| 8305 | |
| 8306 | if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) { |
| 8307 | RExC_rx->extflags |= RXf_UNBOUNDED_QUANTIFIER_SEEN; |
| 8308 | RExC_rx->maxlen = REG_INFTY; |
| 8309 | } |
| 8310 | else { |
| 8311 | RExC_rx->maxlen = RExC_maxlen; |
| 8312 | } |
| 8313 | |
| 8314 | /* Guard against an embedded (?=) or (?<=) with a longer minlen than |
| 8315 | the "real" pattern. */ |
| 8316 | DEBUG_OPTIMISE_r({ |
| 8317 | Perl_re_printf( aTHX_ "minlen: %" IVdf " RExC_rx->minlen:%" IVdf " maxlen:%" IVdf "\n", |
| 8318 | (IV)minlen, (IV)RExC_rx->minlen, (IV)RExC_maxlen); |
| 8319 | }); |
| 8320 | RExC_rx->minlenret = minlen; |
| 8321 | if (RExC_rx->minlen < minlen) |
| 8322 | RExC_rx->minlen = minlen; |
| 8323 | |
| 8324 | if (RExC_seen & REG_RECURSE_SEEN ) { |
| 8325 | RExC_rx->intflags |= PREGf_RECURSE_SEEN; |
| 8326 | Newx(RExC_rx->recurse_locinput, RExC_rx->nparens + 1, char *); |
| 8327 | } |
| 8328 | if (RExC_seen & REG_GPOS_SEEN) |
| 8329 | RExC_rx->intflags |= PREGf_GPOS_SEEN; |
| 8330 | if (RExC_seen & REG_LOOKBEHIND_SEEN) |
| 8331 | RExC_rx->extflags |= RXf_NO_INPLACE_SUBST; /* inplace might break the |
| 8332 | lookbehind */ |
| 8333 | if (pRExC_state->code_blocks) |
| 8334 | RExC_rx->extflags |= RXf_EVAL_SEEN; |
| 8335 | if (RExC_seen & REG_VERBARG_SEEN) |
| 8336 | { |
| 8337 | RExC_rx->intflags |= PREGf_VERBARG_SEEN; |
| 8338 | RExC_rx->extflags |= RXf_NO_INPLACE_SUBST; /* don't understand this! Yves */ |
| 8339 | } |
| 8340 | if (RExC_seen & REG_CUTGROUP_SEEN) |
| 8341 | RExC_rx->intflags |= PREGf_CUTGROUP_SEEN; |
| 8342 | if (pm_flags & PMf_USE_RE_EVAL) |
| 8343 | RExC_rx->intflags |= PREGf_USE_RE_EVAL; |
| 8344 | if (RExC_paren_names) |
| 8345 | RXp_PAREN_NAMES(RExC_rx) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names)); |
| 8346 | else |
| 8347 | RXp_PAREN_NAMES(RExC_rx) = NULL; |
| 8348 | |
| 8349 | /* If we have seen an anchor in our pattern then we set the extflag RXf_IS_ANCHORED |
| 8350 | * so it can be used in pp.c */ |
| 8351 | if (RExC_rx->intflags & PREGf_ANCH) |
| 8352 | RExC_rx->extflags |= RXf_IS_ANCHORED; |
| 8353 | |
| 8354 | |
| 8355 | { |
| 8356 | /* this is used to identify "special" patterns that might result |
| 8357 | * in Perl NOT calling the regex engine and instead doing the match "itself", |
| 8358 | * particularly special cases in split//. By having the regex compiler |
| 8359 | * do this pattern matching at a regop level (instead of by inspecting the pattern) |
| 8360 | * we avoid weird issues with equivalent patterns resulting in different behavior, |
| 8361 | * AND we allow non Perl engines to get the same optimizations by the setting the |
| 8362 | * flags appropriately - Yves */ |
| 8363 | regnode *first = RExC_rxi->program + 1; |
| 8364 | U8 fop = OP(first); |
| 8365 | regnode *next = regnext(first); |
| 8366 | U8 nop = OP(next); |
| 8367 | |
| 8368 | if (PL_regkind[fop] == NOTHING && nop == END) |
| 8369 | RExC_rx->extflags |= RXf_NULL; |
| 8370 | else if ((fop == MBOL || (fop == SBOL && !first->flags)) && nop == END) |
| 8371 | /* when fop is SBOL first->flags will be true only when it was |
| 8372 | * produced by parsing /\A/, and not when parsing /^/. This is |
| 8373 | * very important for the split code as there we want to |
| 8374 | * treat /^/ as /^/m, but we do not want to treat /\A/ as /^/m. |
| 8375 | * See rt #122761 for more details. -- Yves */ |
| 8376 | RExC_rx->extflags |= RXf_START_ONLY; |
| 8377 | else if (fop == PLUS |
| 8378 | && PL_regkind[nop] == POSIXD && FLAGS(next) == _CC_SPACE |
| 8379 | && nop == END) |
| 8380 | RExC_rx->extflags |= RXf_WHITE; |
| 8381 | else if ( RExC_rx->extflags & RXf_SPLIT |
| 8382 | && ( fop == EXACT || fop == LEXACT |
| 8383 | || fop == EXACT_REQ8 || fop == LEXACT_REQ8 |
| 8384 | || fop == EXACTL) |
| 8385 | && STR_LEN(first) == 1 |
| 8386 | && *(STRING(first)) == ' ' |
| 8387 | && nop == END ) |
| 8388 | RExC_rx->extflags |= (RXf_SKIPWHITE|RXf_WHITE); |
| 8389 | |
| 8390 | } |
| 8391 | |
| 8392 | if (RExC_contains_locale) { |
| 8393 | RXp_EXTFLAGS(RExC_rx) |= RXf_TAINTED; |
| 8394 | } |
| 8395 | |
| 8396 | #ifdef DEBUGGING |
| 8397 | if (RExC_paren_names) { |
| 8398 | RExC_rxi->name_list_idx = add_data( pRExC_state, STR_WITH_LEN("a")); |
| 8399 | RExC_rxi->data->data[RExC_rxi->name_list_idx] |
| 8400 | = (void*)SvREFCNT_inc(RExC_paren_name_list); |
| 8401 | } else |
| 8402 | #endif |
| 8403 | RExC_rxi->name_list_idx = 0; |
| 8404 | |
| 8405 | while ( RExC_recurse_count > 0 ) { |
| 8406 | const regnode *scan = RExC_recurse[ --RExC_recurse_count ]; |
| 8407 | /* |
| 8408 | * This data structure is set up in study_chunk() and is used |
| 8409 | * to calculate the distance between a GOSUB regopcode and |
| 8410 | * the OPEN/CURLYM (CURLYM's are special and can act like OPEN's) |
| 8411 | * it refers to. |
| 8412 | * |
| 8413 | * If for some reason someone writes code that optimises |
| 8414 | * away a GOSUB opcode then the assert should be changed to |
| 8415 | * an if(scan) to guard the ARG2L_SET() - Yves |
| 8416 | * |
| 8417 | */ |
| 8418 | assert(scan && OP(scan) == GOSUB); |
| 8419 | ARG2L_SET( scan, RExC_open_parens[ARG(scan)] - REGNODE_OFFSET(scan)); |
| 8420 | } |
| 8421 | |
| 8422 | Newxz(RExC_rx->offs, RExC_total_parens, regexp_paren_pair); |
| 8423 | /* assume we don't need to swap parens around before we match */ |
| 8424 | DEBUG_TEST_r({ |
| 8425 | Perl_re_printf( aTHX_ "study_chunk_recursed_count: %lu\n", |
| 8426 | (unsigned long)RExC_study_chunk_recursed_count); |
| 8427 | }); |
| 8428 | DEBUG_DUMP_r({ |
| 8429 | DEBUG_RExC_seen(); |
| 8430 | Perl_re_printf( aTHX_ "Final program:\n"); |
| 8431 | regdump(RExC_rx); |
| 8432 | }); |
| 8433 | |
| 8434 | if (RExC_open_parens) { |
| 8435 | Safefree(RExC_open_parens); |
| 8436 | RExC_open_parens = NULL; |
| 8437 | } |
| 8438 | if (RExC_close_parens) { |
| 8439 | Safefree(RExC_close_parens); |
| 8440 | RExC_close_parens = NULL; |
| 8441 | } |
| 8442 | |
| 8443 | #ifdef USE_ITHREADS |
| 8444 | /* under ithreads the ?pat? PMf_USED flag on the pmop is simulated |
| 8445 | * by setting the regexp SV to readonly-only instead. If the |
| 8446 | * pattern's been recompiled, the USEDness should remain. */ |
| 8447 | if (old_re && SvREADONLY(old_re)) |
| 8448 | SvREADONLY_on(Rx); |
| 8449 | #endif |
| 8450 | return Rx; |
| 8451 | } |
| 8452 | |
| 8453 | |
| 8454 | SV* |
| 8455 | Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value, |
| 8456 | const U32 flags) |
| 8457 | { |
| 8458 | PERL_ARGS_ASSERT_REG_NAMED_BUFF; |
| 8459 | |
| 8460 | PERL_UNUSED_ARG(value); |
| 8461 | |
| 8462 | if (flags & RXapif_FETCH) { |
| 8463 | return reg_named_buff_fetch(rx, key, flags); |
| 8464 | } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) { |
| 8465 | Perl_croak_no_modify(); |
| 8466 | return NULL; |
| 8467 | } else if (flags & RXapif_EXISTS) { |
| 8468 | return reg_named_buff_exists(rx, key, flags) |
| 8469 | ? &PL_sv_yes |
| 8470 | : &PL_sv_no; |
| 8471 | } else if (flags & RXapif_REGNAMES) { |
| 8472 | return reg_named_buff_all(rx, flags); |
| 8473 | } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) { |
| 8474 | return reg_named_buff_scalar(rx, flags); |
| 8475 | } else { |
| 8476 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags); |
| 8477 | return NULL; |
| 8478 | } |
| 8479 | } |
| 8480 | |
| 8481 | SV* |
| 8482 | Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey, |
| 8483 | const U32 flags) |
| 8484 | { |
| 8485 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER; |
| 8486 | PERL_UNUSED_ARG(lastkey); |
| 8487 | |
| 8488 | if (flags & RXapif_FIRSTKEY) |
| 8489 | return reg_named_buff_firstkey(rx, flags); |
| 8490 | else if (flags & RXapif_NEXTKEY) |
| 8491 | return reg_named_buff_nextkey(rx, flags); |
| 8492 | else { |
| 8493 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", |
| 8494 | (int)flags); |
| 8495 | return NULL; |
| 8496 | } |
| 8497 | } |
| 8498 | |
| 8499 | SV* |
| 8500 | Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv, |
| 8501 | const U32 flags) |
| 8502 | { |
| 8503 | SV *ret; |
| 8504 | struct regexp *const rx = ReANY(r); |
| 8505 | |
| 8506 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH; |
| 8507 | |
| 8508 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 8509 | HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 ); |
| 8510 | if (he_str) { |
| 8511 | IV i; |
| 8512 | SV* sv_dat=HeVAL(he_str); |
| 8513 | I32 *nums=(I32*)SvPVX(sv_dat); |
| 8514 | AV * const retarray = (flags & RXapif_ALL) ? newAV() : NULL; |
| 8515 | for ( i=0; i<SvIVX(sv_dat); i++ ) { |
| 8516 | if ((I32)(rx->nparens) >= nums[i] |
| 8517 | && rx->offs[nums[i]].start != -1 |
| 8518 | && rx->offs[nums[i]].end != -1) |
| 8519 | { |
| 8520 | ret = newSVpvs(""); |
| 8521 | CALLREG_NUMBUF_FETCH(r, nums[i], ret); |
| 8522 | if (!retarray) |
| 8523 | return ret; |
| 8524 | } else { |
| 8525 | if (retarray) |
| 8526 | ret = newSVsv(&PL_sv_undef); |
| 8527 | } |
| 8528 | if (retarray) |
| 8529 | av_push(retarray, ret); |
| 8530 | } |
| 8531 | if (retarray) |
| 8532 | return newRV_noinc(MUTABLE_SV(retarray)); |
| 8533 | } |
| 8534 | } |
| 8535 | return NULL; |
| 8536 | } |
| 8537 | |
| 8538 | bool |
| 8539 | Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key, |
| 8540 | const U32 flags) |
| 8541 | { |
| 8542 | struct regexp *const rx = ReANY(r); |
| 8543 | |
| 8544 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS; |
| 8545 | |
| 8546 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 8547 | if (flags & RXapif_ALL) { |
| 8548 | return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0); |
| 8549 | } else { |
| 8550 | SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags); |
| 8551 | if (sv) { |
| 8552 | SvREFCNT_dec_NN(sv); |
| 8553 | return TRUE; |
| 8554 | } else { |
| 8555 | return FALSE; |
| 8556 | } |
| 8557 | } |
| 8558 | } else { |
| 8559 | return FALSE; |
| 8560 | } |
| 8561 | } |
| 8562 | |
| 8563 | SV* |
| 8564 | Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags) |
| 8565 | { |
| 8566 | struct regexp *const rx = ReANY(r); |
| 8567 | |
| 8568 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY; |
| 8569 | |
| 8570 | if ( rx && RXp_PAREN_NAMES(rx) ) { |
| 8571 | (void)hv_iterinit(RXp_PAREN_NAMES(rx)); |
| 8572 | |
| 8573 | return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY); |
| 8574 | } else { |
| 8575 | return FALSE; |
| 8576 | } |
| 8577 | } |
| 8578 | |
| 8579 | SV* |
| 8580 | Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags) |
| 8581 | { |
| 8582 | struct regexp *const rx = ReANY(r); |
| 8583 | GET_RE_DEBUG_FLAGS_DECL; |
| 8584 | |
| 8585 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY; |
| 8586 | |
| 8587 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 8588 | HV *hv = RXp_PAREN_NAMES(rx); |
| 8589 | HE *temphe; |
| 8590 | while ( (temphe = hv_iternext_flags(hv, 0)) ) { |
| 8591 | IV i; |
| 8592 | IV parno = 0; |
| 8593 | SV* sv_dat = HeVAL(temphe); |
| 8594 | I32 *nums = (I32*)SvPVX(sv_dat); |
| 8595 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { |
| 8596 | if ((I32)(rx->lastparen) >= nums[i] && |
| 8597 | rx->offs[nums[i]].start != -1 && |
| 8598 | rx->offs[nums[i]].end != -1) |
| 8599 | { |
| 8600 | parno = nums[i]; |
| 8601 | break; |
| 8602 | } |
| 8603 | } |
| 8604 | if (parno || flags & RXapif_ALL) { |
| 8605 | return newSVhek(HeKEY_hek(temphe)); |
| 8606 | } |
| 8607 | } |
| 8608 | } |
| 8609 | return NULL; |
| 8610 | } |
| 8611 | |
| 8612 | SV* |
| 8613 | Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags) |
| 8614 | { |
| 8615 | SV *ret; |
| 8616 | AV *av; |
| 8617 | SSize_t length; |
| 8618 | struct regexp *const rx = ReANY(r); |
| 8619 | |
| 8620 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR; |
| 8621 | |
| 8622 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 8623 | if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) { |
| 8624 | return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx))); |
| 8625 | } else if (flags & RXapif_ONE) { |
| 8626 | ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES)); |
| 8627 | av = MUTABLE_AV(SvRV(ret)); |
| 8628 | length = av_tindex(av); |
| 8629 | SvREFCNT_dec_NN(ret); |
| 8630 | return newSViv(length + 1); |
| 8631 | } else { |
| 8632 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", |
| 8633 | (int)flags); |
| 8634 | return NULL; |
| 8635 | } |
| 8636 | } |
| 8637 | return &PL_sv_undef; |
| 8638 | } |
| 8639 | |
| 8640 | SV* |
| 8641 | Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags) |
| 8642 | { |
| 8643 | struct regexp *const rx = ReANY(r); |
| 8644 | AV *av = newAV(); |
| 8645 | |
| 8646 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL; |
| 8647 | |
| 8648 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 8649 | HV *hv= RXp_PAREN_NAMES(rx); |
| 8650 | HE *temphe; |
| 8651 | (void)hv_iterinit(hv); |
| 8652 | while ( (temphe = hv_iternext_flags(hv, 0)) ) { |
| 8653 | IV i; |
| 8654 | IV parno = 0; |
| 8655 | SV* sv_dat = HeVAL(temphe); |
| 8656 | I32 *nums = (I32*)SvPVX(sv_dat); |
| 8657 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { |
| 8658 | if ((I32)(rx->lastparen) >= nums[i] && |
| 8659 | rx->offs[nums[i]].start != -1 && |
| 8660 | rx->offs[nums[i]].end != -1) |
| 8661 | { |
| 8662 | parno = nums[i]; |
| 8663 | break; |
| 8664 | } |
| 8665 | } |
| 8666 | if (parno || flags & RXapif_ALL) { |
| 8667 | av_push(av, newSVhek(HeKEY_hek(temphe))); |
| 8668 | } |
| 8669 | } |
| 8670 | } |
| 8671 | |
| 8672 | return newRV_noinc(MUTABLE_SV(av)); |
| 8673 | } |
| 8674 | |
| 8675 | void |
| 8676 | Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren, |
| 8677 | SV * const sv) |
| 8678 | { |
| 8679 | struct regexp *const rx = ReANY(r); |
| 8680 | char *s = NULL; |
| 8681 | SSize_t i = 0; |
| 8682 | SSize_t s1, t1; |
| 8683 | I32 n = paren; |
| 8684 | |
| 8685 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH; |
| 8686 | |
| 8687 | if ( n == RX_BUFF_IDX_CARET_PREMATCH |
| 8688 | || n == RX_BUFF_IDX_CARET_FULLMATCH |
| 8689 | || n == RX_BUFF_IDX_CARET_POSTMATCH |
| 8690 | ) |
| 8691 | { |
| 8692 | bool keepcopy = cBOOL(rx->extflags & RXf_PMf_KEEPCOPY); |
| 8693 | if (!keepcopy) { |
| 8694 | /* on something like |
| 8695 | * $r = qr/.../; |
| 8696 | * /$qr/p; |
| 8697 | * the KEEPCOPY is set on the PMOP rather than the regex */ |
| 8698 | if (PL_curpm && r == PM_GETRE(PL_curpm)) |
| 8699 | keepcopy = cBOOL(PL_curpm->op_pmflags & PMf_KEEPCOPY); |
| 8700 | } |
| 8701 | if (!keepcopy) |
| 8702 | goto ret_undef; |
| 8703 | } |
| 8704 | |
| 8705 | if (!rx->subbeg) |
| 8706 | goto ret_undef; |
| 8707 | |
| 8708 | if (n == RX_BUFF_IDX_CARET_FULLMATCH) |
| 8709 | /* no need to distinguish between them any more */ |
| 8710 | n = RX_BUFF_IDX_FULLMATCH; |
| 8711 | |
| 8712 | if ((n == RX_BUFF_IDX_PREMATCH || n == RX_BUFF_IDX_CARET_PREMATCH) |
| 8713 | && rx->offs[0].start != -1) |
| 8714 | { |
| 8715 | /* $`, ${^PREMATCH} */ |
| 8716 | i = rx->offs[0].start; |
| 8717 | s = rx->subbeg; |
| 8718 | } |
| 8719 | else |
| 8720 | if ((n == RX_BUFF_IDX_POSTMATCH || n == RX_BUFF_IDX_CARET_POSTMATCH) |
| 8721 | && rx->offs[0].end != -1) |
| 8722 | { |
| 8723 | /* $', ${^POSTMATCH} */ |
| 8724 | s = rx->subbeg - rx->suboffset + rx->offs[0].end; |
| 8725 | i = rx->sublen + rx->suboffset - rx->offs[0].end; |
| 8726 | } |
| 8727 | else |
| 8728 | if (inRANGE(n, 0, (I32)rx->nparens) && |
| 8729 | (s1 = rx->offs[n].start) != -1 && |
| 8730 | (t1 = rx->offs[n].end) != -1) |
| 8731 | { |
| 8732 | /* $&, ${^MATCH}, $1 ... */ |
| 8733 | i = t1 - s1; |
| 8734 | s = rx->subbeg + s1 - rx->suboffset; |
| 8735 | } else { |
| 8736 | goto ret_undef; |
| 8737 | } |
| 8738 | |
| 8739 | assert(s >= rx->subbeg); |
| 8740 | assert((STRLEN)rx->sublen >= (STRLEN)((s - rx->subbeg) + i) ); |
| 8741 | if (i >= 0) { |
| 8742 | #ifdef NO_TAINT_SUPPORT |
| 8743 | sv_setpvn(sv, s, i); |
| 8744 | #else |
| 8745 | const int oldtainted = TAINT_get; |
| 8746 | TAINT_NOT; |
| 8747 | sv_setpvn(sv, s, i); |
| 8748 | TAINT_set(oldtainted); |
| 8749 | #endif |
| 8750 | if (RXp_MATCH_UTF8(rx)) |
| 8751 | SvUTF8_on(sv); |
| 8752 | else |
| 8753 | SvUTF8_off(sv); |
| 8754 | if (TAINTING_get) { |
| 8755 | if (RXp_MATCH_TAINTED(rx)) { |
| 8756 | if (SvTYPE(sv) >= SVt_PVMG) { |
| 8757 | MAGIC* const mg = SvMAGIC(sv); |
| 8758 | MAGIC* mgt; |
| 8759 | TAINT; |
| 8760 | SvMAGIC_set(sv, mg->mg_moremagic); |
| 8761 | SvTAINT(sv); |
| 8762 | if ((mgt = SvMAGIC(sv))) { |
| 8763 | mg->mg_moremagic = mgt; |
| 8764 | SvMAGIC_set(sv, mg); |
| 8765 | } |
| 8766 | } else { |
| 8767 | TAINT; |
| 8768 | SvTAINT(sv); |
| 8769 | } |
| 8770 | } else |
| 8771 | SvTAINTED_off(sv); |
| 8772 | } |
| 8773 | } else { |
| 8774 | ret_undef: |
| 8775 | sv_set_undef(sv); |
| 8776 | return; |
| 8777 | } |
| 8778 | } |
| 8779 | |
| 8780 | void |
| 8781 | Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren, |
| 8782 | SV const * const value) |
| 8783 | { |
| 8784 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE; |
| 8785 | |
| 8786 | PERL_UNUSED_ARG(rx); |
| 8787 | PERL_UNUSED_ARG(paren); |
| 8788 | PERL_UNUSED_ARG(value); |
| 8789 | |
| 8790 | if (!PL_localizing) |
| 8791 | Perl_croak_no_modify(); |
| 8792 | } |
| 8793 | |
| 8794 | I32 |
| 8795 | Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv, |
| 8796 | const I32 paren) |
| 8797 | { |
| 8798 | struct regexp *const rx = ReANY(r); |
| 8799 | I32 i; |
| 8800 | I32 s1, t1; |
| 8801 | |
| 8802 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH; |
| 8803 | |
| 8804 | if ( paren == RX_BUFF_IDX_CARET_PREMATCH |
| 8805 | || paren == RX_BUFF_IDX_CARET_FULLMATCH |
| 8806 | || paren == RX_BUFF_IDX_CARET_POSTMATCH |
| 8807 | ) |
| 8808 | { |
| 8809 | bool keepcopy = cBOOL(rx->extflags & RXf_PMf_KEEPCOPY); |
| 8810 | if (!keepcopy) { |
| 8811 | /* on something like |
| 8812 | * $r = qr/.../; |
| 8813 | * /$qr/p; |
| 8814 | * the KEEPCOPY is set on the PMOP rather than the regex */ |
| 8815 | if (PL_curpm && r == PM_GETRE(PL_curpm)) |
| 8816 | keepcopy = cBOOL(PL_curpm->op_pmflags & PMf_KEEPCOPY); |
| 8817 | } |
| 8818 | if (!keepcopy) |
| 8819 | goto warn_undef; |
| 8820 | } |
| 8821 | |
| 8822 | /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */ |
| 8823 | switch (paren) { |
| 8824 | case RX_BUFF_IDX_CARET_PREMATCH: /* ${^PREMATCH} */ |
| 8825 | case RX_BUFF_IDX_PREMATCH: /* $` */ |
| 8826 | if (rx->offs[0].start != -1) { |
| 8827 | i = rx->offs[0].start; |
| 8828 | if (i > 0) { |
| 8829 | s1 = 0; |
| 8830 | t1 = i; |
| 8831 | goto getlen; |
| 8832 | } |
| 8833 | } |
| 8834 | return 0; |
| 8835 | |
| 8836 | case RX_BUFF_IDX_CARET_POSTMATCH: /* ${^POSTMATCH} */ |
| 8837 | case RX_BUFF_IDX_POSTMATCH: /* $' */ |
| 8838 | if (rx->offs[0].end != -1) { |
| 8839 | i = rx->sublen - rx->offs[0].end; |
| 8840 | if (i > 0) { |
| 8841 | s1 = rx->offs[0].end; |
| 8842 | t1 = rx->sublen; |
| 8843 | goto getlen; |
| 8844 | } |
| 8845 | } |
| 8846 | return 0; |
| 8847 | |
| 8848 | default: /* $& / ${^MATCH}, $1, $2, ... */ |
| 8849 | if (paren <= (I32)rx->nparens && |
| 8850 | (s1 = rx->offs[paren].start) != -1 && |
| 8851 | (t1 = rx->offs[paren].end) != -1) |
| 8852 | { |
| 8853 | i = t1 - s1; |
| 8854 | goto getlen; |
| 8855 | } else { |
| 8856 | warn_undef: |
| 8857 | if (ckWARN(WARN_UNINITIALIZED)) |
| 8858 | report_uninit((const SV *)sv); |
| 8859 | return 0; |
| 8860 | } |
| 8861 | } |
| 8862 | getlen: |
| 8863 | if (i > 0 && RXp_MATCH_UTF8(rx)) { |
| 8864 | const char * const s = rx->subbeg - rx->suboffset + s1; |
| 8865 | const U8 *ep; |
| 8866 | STRLEN el; |
| 8867 | |
| 8868 | i = t1 - s1; |
| 8869 | if (is_utf8_string_loclen((U8*)s, i, &ep, &el)) |
| 8870 | i = el; |
| 8871 | } |
| 8872 | return i; |
| 8873 | } |
| 8874 | |
| 8875 | SV* |
| 8876 | Perl_reg_qr_package(pTHX_ REGEXP * const rx) |
| 8877 | { |
| 8878 | PERL_ARGS_ASSERT_REG_QR_PACKAGE; |
| 8879 | PERL_UNUSED_ARG(rx); |
| 8880 | if (0) |
| 8881 | return NULL; |
| 8882 | else |
| 8883 | return newSVpvs("Regexp"); |
| 8884 | } |
| 8885 | |
| 8886 | /* Scans the name of a named buffer from the pattern. |
| 8887 | * If flags is REG_RSN_RETURN_NULL returns null. |
| 8888 | * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name |
| 8889 | * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding |
| 8890 | * to the parsed name as looked up in the RExC_paren_names hash. |
| 8891 | * If there is an error throws a vFAIL().. type exception. |
| 8892 | */ |
| 8893 | |
| 8894 | #define REG_RSN_RETURN_NULL 0 |
| 8895 | #define REG_RSN_RETURN_NAME 1 |
| 8896 | #define REG_RSN_RETURN_DATA 2 |
| 8897 | |
| 8898 | STATIC SV* |
| 8899 | S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) |
| 8900 | { |
| 8901 | char *name_start = RExC_parse; |
| 8902 | SV* sv_name; |
| 8903 | |
| 8904 | PERL_ARGS_ASSERT_REG_SCAN_NAME; |
| 8905 | |
| 8906 | assert (RExC_parse <= RExC_end); |
| 8907 | if (RExC_parse == RExC_end) NOOP; |
| 8908 | else if (isIDFIRST_lazy_if_safe(RExC_parse, RExC_end, UTF)) { |
| 8909 | /* Note that the code here assumes well-formed UTF-8. Skip IDFIRST by |
| 8910 | * using do...while */ |
| 8911 | if (UTF) |
| 8912 | do { |
| 8913 | RExC_parse += UTF8SKIP(RExC_parse); |
| 8914 | } while ( RExC_parse < RExC_end |
| 8915 | && isWORDCHAR_utf8_safe((U8*)RExC_parse, (U8*) RExC_end)); |
| 8916 | else |
| 8917 | do { |
| 8918 | RExC_parse++; |
| 8919 | } while (RExC_parse < RExC_end && isWORDCHAR(*RExC_parse)); |
| 8920 | } else { |
| 8921 | RExC_parse++; /* so the <- from the vFAIL is after the offending |
| 8922 | character */ |
| 8923 | vFAIL("Group name must start with a non-digit word character"); |
| 8924 | } |
| 8925 | sv_name = newSVpvn_flags(name_start, (int)(RExC_parse - name_start), |
| 8926 | SVs_TEMP | (UTF ? SVf_UTF8 : 0)); |
| 8927 | if ( flags == REG_RSN_RETURN_NAME) |
| 8928 | return sv_name; |
| 8929 | else if (flags==REG_RSN_RETURN_DATA) { |
| 8930 | HE *he_str = NULL; |
| 8931 | SV *sv_dat = NULL; |
| 8932 | if ( ! sv_name ) /* should not happen*/ |
| 8933 | Perl_croak(aTHX_ "panic: no svname in reg_scan_name"); |
| 8934 | if (RExC_paren_names) |
| 8935 | he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 ); |
| 8936 | if ( he_str ) |
| 8937 | sv_dat = HeVAL(he_str); |
| 8938 | if ( ! sv_dat ) { /* Didn't find group */ |
| 8939 | |
| 8940 | /* It might be a forward reference; we can't fail until we |
| 8941 | * know, by completing the parse to get all the groups, and |
| 8942 | * then reparsing */ |
| 8943 | if (ALL_PARENS_COUNTED) { |
| 8944 | vFAIL("Reference to nonexistent named group"); |
| 8945 | } |
| 8946 | else { |
| 8947 | REQUIRE_PARENS_PASS; |
| 8948 | } |
| 8949 | } |
| 8950 | return sv_dat; |
| 8951 | } |
| 8952 | |
| 8953 | Perl_croak(aTHX_ "panic: bad flag %lx in reg_scan_name", |
| 8954 | (unsigned long) flags); |
| 8955 | } |
| 8956 | |
| 8957 | #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \ |
| 8958 | if (RExC_lastparse!=RExC_parse) { \ |
| 8959 | Perl_re_printf( aTHX_ "%s", \ |
| 8960 | Perl_pv_pretty(aTHX_ RExC_mysv1, RExC_parse, \ |
| 8961 | RExC_end - RExC_parse, 16, \ |
| 8962 | "", "", \ |
| 8963 | PERL_PV_ESCAPE_UNI_DETECT | \ |
| 8964 | PERL_PV_PRETTY_ELLIPSES | \ |
| 8965 | PERL_PV_PRETTY_LTGT | \ |
| 8966 | PERL_PV_ESCAPE_RE | \ |
| 8967 | PERL_PV_PRETTY_EXACTSIZE \ |
| 8968 | ) \ |
| 8969 | ); \ |
| 8970 | } else \ |
| 8971 | Perl_re_printf( aTHX_ "%16s",""); \ |
| 8972 | \ |
| 8973 | if (RExC_lastnum!=RExC_emit) \ |
| 8974 | Perl_re_printf( aTHX_ "|%4zu", RExC_emit); \ |
| 8975 | else \ |
| 8976 | Perl_re_printf( aTHX_ "|%4s",""); \ |
| 8977 | Perl_re_printf( aTHX_ "|%*s%-4s", \ |
| 8978 | (int)((depth*2)), "", \ |
| 8979 | (funcname) \ |
| 8980 | ); \ |
| 8981 | RExC_lastnum=RExC_emit; \ |
| 8982 | RExC_lastparse=RExC_parse; \ |
| 8983 | }) |
| 8984 | |
| 8985 | |
| 8986 | |
| 8987 | #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \ |
| 8988 | DEBUG_PARSE_MSG((funcname)); \ |
| 8989 | Perl_re_printf( aTHX_ "%4s","\n"); \ |
| 8990 | }) |
| 8991 | #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({\ |
| 8992 | DEBUG_PARSE_MSG((funcname)); \ |
| 8993 | Perl_re_printf( aTHX_ fmt "\n",args); \ |
| 8994 | }) |
| 8995 | |
| 8996 | /* This section of code defines the inversion list object and its methods. The |
| 8997 | * interfaces are highly subject to change, so as much as possible is static to |
| 8998 | * this file. An inversion list is here implemented as a malloc'd C UV array |
| 8999 | * as an SVt_INVLIST scalar. |
| 9000 | * |
| 9001 | * An inversion list for Unicode is an array of code points, sorted by ordinal |
| 9002 | * number. Each element gives the code point that begins a range that extends |
| 9003 | * up-to but not including the code point given by the next element. The final |
| 9004 | * element gives the first code point of a range that extends to the platform's |
| 9005 | * infinity. The even-numbered elements (invlist[0], invlist[2], invlist[4], |
| 9006 | * ...) give ranges whose code points are all in the inversion list. We say |
| 9007 | * that those ranges are in the set. The odd-numbered elements give ranges |
| 9008 | * whose code points are not in the inversion list, and hence not in the set. |
| 9009 | * Thus, element [0] is the first code point in the list. Element [1] |
| 9010 | * is the first code point beyond that not in the list; and element [2] is the |
| 9011 | * first code point beyond that that is in the list. In other words, the first |
| 9012 | * range is invlist[0]..(invlist[1]-1), and all code points in that range are |
| 9013 | * in the inversion list. The second range is invlist[1]..(invlist[2]-1), and |
| 9014 | * all code points in that range are not in the inversion list. The third |
| 9015 | * range invlist[2]..(invlist[3]-1) gives code points that are in the inversion |
| 9016 | * list, and so forth. Thus every element whose index is divisible by two |
| 9017 | * gives the beginning of a range that is in the list, and every element whose |
| 9018 | * index is not divisible by two gives the beginning of a range not in the |
| 9019 | * list. If the final element's index is divisible by two, the inversion list |
| 9020 | * extends to the platform's infinity; otherwise the highest code point in the |
| 9021 | * inversion list is the contents of that element minus 1. |
| 9022 | * |
| 9023 | * A range that contains just a single code point N will look like |
| 9024 | * invlist[i] == N |
| 9025 | * invlist[i+1] == N+1 |
| 9026 | * |
| 9027 | * If N is UV_MAX (the highest representable code point on the machine), N+1 is |
| 9028 | * impossible to represent, so element [i+1] is omitted. The single element |
| 9029 | * inversion list |
| 9030 | * invlist[0] == UV_MAX |
| 9031 | * contains just UV_MAX, but is interpreted as matching to infinity. |
| 9032 | * |
| 9033 | * Taking the complement (inverting) an inversion list is quite simple, if the |
| 9034 | * first element is 0, remove it; otherwise add a 0 element at the beginning. |
| 9035 | * This implementation reserves an element at the beginning of each inversion |
| 9036 | * list to always contain 0; there is an additional flag in the header which |
| 9037 | * indicates if the list begins at the 0, or is offset to begin at the next |
| 9038 | * element. This means that the inversion list can be inverted without any |
| 9039 | * copying; just flip the flag. |
| 9040 | * |
| 9041 | * More about inversion lists can be found in "Unicode Demystified" |
| 9042 | * Chapter 13 by Richard Gillam, published by Addison-Wesley. |
| 9043 | * |
| 9044 | * The inversion list data structure is currently implemented as an SV pointing |
| 9045 | * to an array of UVs that the SV thinks are bytes. This allows us to have an |
| 9046 | * array of UV whose memory management is automatically handled by the existing |
| 9047 | * facilities for SV's. |
| 9048 | * |
| 9049 | * Some of the methods should always be private to the implementation, and some |
| 9050 | * should eventually be made public */ |
| 9051 | |
| 9052 | /* The header definitions are in F<invlist_inline.h> */ |
| 9053 | |
| 9054 | #ifndef PERL_IN_XSUB_RE |
| 9055 | |
| 9056 | PERL_STATIC_INLINE UV* |
| 9057 | S__invlist_array_init(SV* const invlist, const bool will_have_0) |
| 9058 | { |
| 9059 | /* Returns a pointer to the first element in the inversion list's array. |
| 9060 | * This is called upon initialization of an inversion list. Where the |
| 9061 | * array begins depends on whether the list has the code point U+0000 in it |
| 9062 | * or not. The other parameter tells it whether the code that follows this |
| 9063 | * call is about to put a 0 in the inversion list or not. The first |
| 9064 | * element is either the element reserved for 0, if TRUE, or the element |
| 9065 | * after it, if FALSE */ |
| 9066 | |
| 9067 | bool* offset = get_invlist_offset_addr(invlist); |
| 9068 | UV* zero_addr = (UV *) SvPVX(invlist); |
| 9069 | |
| 9070 | PERL_ARGS_ASSERT__INVLIST_ARRAY_INIT; |
| 9071 | |
| 9072 | /* Must be empty */ |
| 9073 | assert(! _invlist_len(invlist)); |
| 9074 | |
| 9075 | *zero_addr = 0; |
| 9076 | |
| 9077 | /* 1^1 = 0; 1^0 = 1 */ |
| 9078 | *offset = 1 ^ will_have_0; |
| 9079 | return zero_addr + *offset; |
| 9080 | } |
| 9081 | |
| 9082 | STATIC void |
| 9083 | S_invlist_replace_list_destroys_src(pTHX_ SV * dest, SV * src) |
| 9084 | { |
| 9085 | /* Replaces the inversion list in 'dest' with the one from 'src'. It |
| 9086 | * steals the list from 'src', so 'src' is made to have a NULL list. This |
| 9087 | * is similar to what SvSetMagicSV() would do, if it were implemented on |
| 9088 | * inversion lists, though this routine avoids a copy */ |
| 9089 | |
| 9090 | const UV src_len = _invlist_len(src); |
| 9091 | const bool src_offset = *get_invlist_offset_addr(src); |
| 9092 | const STRLEN src_byte_len = SvLEN(src); |
| 9093 | char * array = SvPVX(src); |
| 9094 | |
| 9095 | const int oldtainted = TAINT_get; |
| 9096 | |
| 9097 | PERL_ARGS_ASSERT_INVLIST_REPLACE_LIST_DESTROYS_SRC; |
| 9098 | |
| 9099 | assert(is_invlist(src)); |
| 9100 | assert(is_invlist(dest)); |
| 9101 | assert(! invlist_is_iterating(src)); |
| 9102 | assert(SvCUR(src) == 0 || SvCUR(src) < SvLEN(src)); |
| 9103 | |
| 9104 | /* Make sure it ends in the right place with a NUL, as our inversion list |
| 9105 | * manipulations aren't careful to keep this true, but sv_usepvn_flags() |
| 9106 | * asserts it */ |
| 9107 | array[src_byte_len - 1] = '\0'; |
| 9108 | |
| 9109 | TAINT_NOT; /* Otherwise it breaks */ |
| 9110 | sv_usepvn_flags(dest, |
| 9111 | (char *) array, |
| 9112 | src_byte_len - 1, |
| 9113 | |
| 9114 | /* This flag is documented to cause a copy to be avoided */ |
| 9115 | SV_HAS_TRAILING_NUL); |
| 9116 | TAINT_set(oldtainted); |
| 9117 | SvPV_set(src, 0); |
| 9118 | SvLEN_set(src, 0); |
| 9119 | SvCUR_set(src, 0); |
| 9120 | |
| 9121 | /* Finish up copying over the other fields in an inversion list */ |
| 9122 | *get_invlist_offset_addr(dest) = src_offset; |
| 9123 | invlist_set_len(dest, src_len, src_offset); |
| 9124 | *get_invlist_previous_index_addr(dest) = 0; |
| 9125 | invlist_iterfinish(dest); |
| 9126 | } |
| 9127 | |
| 9128 | PERL_STATIC_INLINE IV* |
| 9129 | S_get_invlist_previous_index_addr(SV* invlist) |
| 9130 | { |
| 9131 | /* Return the address of the IV that is reserved to hold the cached index |
| 9132 | * */ |
| 9133 | PERL_ARGS_ASSERT_GET_INVLIST_PREVIOUS_INDEX_ADDR; |
| 9134 | |
| 9135 | assert(is_invlist(invlist)); |
| 9136 | |
| 9137 | return &(((XINVLIST*) SvANY(invlist))->prev_index); |
| 9138 | } |
| 9139 | |
| 9140 | PERL_STATIC_INLINE IV |
| 9141 | S_invlist_previous_index(SV* const invlist) |
| 9142 | { |
| 9143 | /* Returns cached index of previous search */ |
| 9144 | |
| 9145 | PERL_ARGS_ASSERT_INVLIST_PREVIOUS_INDEX; |
| 9146 | |
| 9147 | return *get_invlist_previous_index_addr(invlist); |
| 9148 | } |
| 9149 | |
| 9150 | PERL_STATIC_INLINE void |
| 9151 | S_invlist_set_previous_index(SV* const invlist, const IV index) |
| 9152 | { |
| 9153 | /* Caches <index> for later retrieval */ |
| 9154 | |
| 9155 | PERL_ARGS_ASSERT_INVLIST_SET_PREVIOUS_INDEX; |
| 9156 | |
| 9157 | assert(index == 0 || index < (int) _invlist_len(invlist)); |
| 9158 | |
| 9159 | *get_invlist_previous_index_addr(invlist) = index; |
| 9160 | } |
| 9161 | |
| 9162 | PERL_STATIC_INLINE void |
| 9163 | S_invlist_trim(SV* invlist) |
| 9164 | { |
| 9165 | /* Free the not currently-being-used space in an inversion list */ |
| 9166 | |
| 9167 | /* But don't free up the space needed for the 0 UV that is always at the |
| 9168 | * beginning of the list, nor the trailing NUL */ |
| 9169 | const UV min_size = TO_INTERNAL_SIZE(1) + 1; |
| 9170 | |
| 9171 | PERL_ARGS_ASSERT_INVLIST_TRIM; |
| 9172 | |
| 9173 | assert(is_invlist(invlist)); |
| 9174 | |
| 9175 | SvPV_renew(invlist, MAX(min_size, SvCUR(invlist) + 1)); |
| 9176 | } |
| 9177 | |
| 9178 | PERL_STATIC_INLINE void |
| 9179 | S_invlist_clear(pTHX_ SV* invlist) /* Empty the inversion list */ |
| 9180 | { |
| 9181 | PERL_ARGS_ASSERT_INVLIST_CLEAR; |
| 9182 | |
| 9183 | assert(is_invlist(invlist)); |
| 9184 | |
| 9185 | invlist_set_len(invlist, 0, 0); |
| 9186 | invlist_trim(invlist); |
| 9187 | } |
| 9188 | |
| 9189 | #endif /* ifndef PERL_IN_XSUB_RE */ |
| 9190 | |
| 9191 | PERL_STATIC_INLINE bool |
| 9192 | S_invlist_is_iterating(SV* const invlist) |
| 9193 | { |
| 9194 | PERL_ARGS_ASSERT_INVLIST_IS_ITERATING; |
| 9195 | |
| 9196 | return *(get_invlist_iter_addr(invlist)) < (STRLEN) UV_MAX; |
| 9197 | } |
| 9198 | |
| 9199 | #ifndef PERL_IN_XSUB_RE |
| 9200 | |
| 9201 | PERL_STATIC_INLINE UV |
| 9202 | S_invlist_max(SV* const invlist) |
| 9203 | { |
| 9204 | /* Returns the maximum number of elements storable in the inversion list's |
| 9205 | * array, without having to realloc() */ |
| 9206 | |
| 9207 | PERL_ARGS_ASSERT_INVLIST_MAX; |
| 9208 | |
| 9209 | assert(is_invlist(invlist)); |
| 9210 | |
| 9211 | /* Assumes worst case, in which the 0 element is not counted in the |
| 9212 | * inversion list, so subtracts 1 for that */ |
| 9213 | return SvLEN(invlist) == 0 /* This happens under _new_invlist_C_array */ |
| 9214 | ? FROM_INTERNAL_SIZE(SvCUR(invlist)) - 1 |
| 9215 | : FROM_INTERNAL_SIZE(SvLEN(invlist)) - 1; |
| 9216 | } |
| 9217 | |
| 9218 | STATIC void |
| 9219 | S_initialize_invlist_guts(pTHX_ SV* invlist, const Size_t initial_size) |
| 9220 | { |
| 9221 | PERL_ARGS_ASSERT_INITIALIZE_INVLIST_GUTS; |
| 9222 | |
| 9223 | /* First 1 is in case the zero element isn't in the list; second 1 is for |
| 9224 | * trailing NUL */ |
| 9225 | SvGROW(invlist, TO_INTERNAL_SIZE(initial_size + 1) + 1); |
| 9226 | invlist_set_len(invlist, 0, 0); |
| 9227 | |
| 9228 | /* Force iterinit() to be used to get iteration to work */ |
| 9229 | invlist_iterfinish(invlist); |
| 9230 | |
| 9231 | *get_invlist_previous_index_addr(invlist) = 0; |
| 9232 | SvPOK_on(invlist); /* This allows B to extract the PV */ |
| 9233 | } |
| 9234 | |
| 9235 | SV* |
| 9236 | Perl__new_invlist(pTHX_ IV initial_size) |
| 9237 | { |
| 9238 | |
| 9239 | /* Return a pointer to a newly constructed inversion list, with enough |
| 9240 | * space to store 'initial_size' elements. If that number is negative, a |
| 9241 | * system default is used instead */ |
| 9242 | |
| 9243 | SV* new_list; |
| 9244 | |
| 9245 | if (initial_size < 0) { |
| 9246 | initial_size = 10; |
| 9247 | } |
| 9248 | |
| 9249 | new_list = newSV_type(SVt_INVLIST); |
| 9250 | initialize_invlist_guts(new_list, initial_size); |
| 9251 | |
| 9252 | return new_list; |
| 9253 | } |
| 9254 | |
| 9255 | SV* |
| 9256 | Perl__new_invlist_C_array(pTHX_ const UV* const list) |
| 9257 | { |
| 9258 | /* Return a pointer to a newly constructed inversion list, initialized to |
| 9259 | * point to <list>, which has to be in the exact correct inversion list |
| 9260 | * form, including internal fields. Thus this is a dangerous routine that |
| 9261 | * should not be used in the wrong hands. The passed in 'list' contains |
| 9262 | * several header fields at the beginning that are not part of the |
| 9263 | * inversion list body proper */ |
| 9264 | |
| 9265 | const STRLEN length = (STRLEN) list[0]; |
| 9266 | const UV version_id = list[1]; |
| 9267 | const bool offset = cBOOL(list[2]); |
| 9268 | #define HEADER_LENGTH 3 |
| 9269 | /* If any of the above changes in any way, you must change HEADER_LENGTH |
| 9270 | * (if appropriate) and regenerate INVLIST_VERSION_ID by running |
| 9271 | * perl -E 'say int(rand 2**31-1)' |
| 9272 | */ |
| 9273 | #define INVLIST_VERSION_ID 148565664 /* This is a combination of a version and |
| 9274 | data structure type, so that one being |
| 9275 | passed in can be validated to be an |
| 9276 | inversion list of the correct vintage. |
| 9277 | */ |
| 9278 | |
| 9279 | SV* invlist = newSV_type(SVt_INVLIST); |
| 9280 | |
| 9281 | PERL_ARGS_ASSERT__NEW_INVLIST_C_ARRAY; |
| 9282 | |
| 9283 | if (version_id != INVLIST_VERSION_ID) { |
| 9284 | Perl_croak(aTHX_ "panic: Incorrect version for previously generated inversion list"); |
| 9285 | } |
| 9286 | |
| 9287 | /* The generated array passed in includes header elements that aren't part |
| 9288 | * of the list proper, so start it just after them */ |
| 9289 | SvPV_set(invlist, (char *) (list + HEADER_LENGTH)); |
| 9290 | |
| 9291 | SvLEN_set(invlist, 0); /* Means we own the contents, and the system |
| 9292 | shouldn't touch it */ |
| 9293 | |
| 9294 | *(get_invlist_offset_addr(invlist)) = offset; |
| 9295 | |
| 9296 | /* The 'length' passed to us is the physical number of elements in the |
| 9297 | * inversion list. But if there is an offset the logical number is one |
| 9298 | * less than that */ |
| 9299 | invlist_set_len(invlist, length - offset, offset); |
| 9300 | |
| 9301 | invlist_set_previous_index(invlist, 0); |
| 9302 | |
| 9303 | /* Initialize the iteration pointer. */ |
| 9304 | invlist_iterfinish(invlist); |
| 9305 | |
| 9306 | SvREADONLY_on(invlist); |
| 9307 | SvPOK_on(invlist); |
| 9308 | |
| 9309 | return invlist; |
| 9310 | } |
| 9311 | |
| 9312 | STATIC void |
| 9313 | S__append_range_to_invlist(pTHX_ SV* const invlist, |
| 9314 | const UV start, const UV end) |
| 9315 | { |
| 9316 | /* Subject to change or removal. Append the range from 'start' to 'end' at |
| 9317 | * the end of the inversion list. The range must be above any existing |
| 9318 | * ones. */ |
| 9319 | |
| 9320 | UV* array; |
| 9321 | UV max = invlist_max(invlist); |
| 9322 | UV len = _invlist_len(invlist); |
| 9323 | bool offset; |
| 9324 | |
| 9325 | PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST; |
| 9326 | |
| 9327 | if (len == 0) { /* Empty lists must be initialized */ |
| 9328 | offset = start != 0; |
| 9329 | array = _invlist_array_init(invlist, ! offset); |
| 9330 | } |
| 9331 | else { |
| 9332 | /* Here, the existing list is non-empty. The current max entry in the |
| 9333 | * list is generally the first value not in the set, except when the |
| 9334 | * set extends to the end of permissible values, in which case it is |
| 9335 | * the first entry in that final set, and so this call is an attempt to |
| 9336 | * append out-of-order */ |
| 9337 | |
| 9338 | UV final_element = len - 1; |
| 9339 | array = invlist_array(invlist); |
| 9340 | if ( array[final_element] > start |
| 9341 | || ELEMENT_RANGE_MATCHES_INVLIST(final_element)) |
| 9342 | { |
| 9343 | Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list, final=%" UVuf ", start=%" UVuf ", match=%c", |
| 9344 | array[final_element], start, |
| 9345 | ELEMENT_RANGE_MATCHES_INVLIST(final_element) ? 't' : 'f'); |
| 9346 | } |
| 9347 | |
| 9348 | /* Here, it is a legal append. If the new range begins 1 above the end |
| 9349 | * of the range below it, it is extending the range below it, so the |
| 9350 | * new first value not in the set is one greater than the newly |
| 9351 | * extended range. */ |
| 9352 | offset = *get_invlist_offset_addr(invlist); |
| 9353 | if (array[final_element] == start) { |
| 9354 | if (end != UV_MAX) { |
| 9355 | array[final_element] = end + 1; |
| 9356 | } |
| 9357 | else { |
| 9358 | /* But if the end is the maximum representable on the machine, |
| 9359 | * assume that infinity was actually what was meant. Just let |
| 9360 | * the range that this would extend to have no end */ |
| 9361 | invlist_set_len(invlist, len - 1, offset); |
| 9362 | } |
| 9363 | return; |
| 9364 | } |
| 9365 | } |
| 9366 | |
| 9367 | /* Here the new range doesn't extend any existing set. Add it */ |
| 9368 | |
| 9369 | len += 2; /* Includes an element each for the start and end of range */ |
| 9370 | |
| 9371 | /* If wll overflow the existing space, extend, which may cause the array to |
| 9372 | * be moved */ |
| 9373 | if (max < len) { |
| 9374 | invlist_extend(invlist, len); |
| 9375 | |
| 9376 | /* Have to set len here to avoid assert failure in invlist_array() */ |
| 9377 | invlist_set_len(invlist, len, offset); |
| 9378 | |
| 9379 | array = invlist_array(invlist); |
| 9380 | } |
| 9381 | else { |
| 9382 | invlist_set_len(invlist, len, offset); |
| 9383 | } |
| 9384 | |
| 9385 | /* The next item on the list starts the range, the one after that is |
| 9386 | * one past the new range. */ |
| 9387 | array[len - 2] = start; |
| 9388 | if (end != UV_MAX) { |
| 9389 | array[len - 1] = end + 1; |
| 9390 | } |
| 9391 | else { |
| 9392 | /* But if the end is the maximum representable on the machine, just let |
| 9393 | * the range have no end */ |
| 9394 | invlist_set_len(invlist, len - 1, offset); |
| 9395 | } |
| 9396 | } |
| 9397 | |
| 9398 | SSize_t |
| 9399 | Perl__invlist_search(SV* const invlist, const UV cp) |
| 9400 | { |
| 9401 | /* Searches the inversion list for the entry that contains the input code |
| 9402 | * point <cp>. If <cp> is not in the list, -1 is returned. Otherwise, the |
| 9403 | * return value is the index into the list's array of the range that |
| 9404 | * contains <cp>, that is, 'i' such that |
| 9405 | * array[i] <= cp < array[i+1] |
| 9406 | */ |
| 9407 | |
| 9408 | IV low = 0; |
| 9409 | IV mid; |
| 9410 | IV high = _invlist_len(invlist); |
| 9411 | const IV highest_element = high - 1; |
| 9412 | const UV* array; |
| 9413 | |
| 9414 | PERL_ARGS_ASSERT__INVLIST_SEARCH; |
| 9415 | |
| 9416 | /* If list is empty, return failure. */ |
| 9417 | if (high == 0) { |
| 9418 | return -1; |
| 9419 | } |
| 9420 | |
| 9421 | /* (We can't get the array unless we know the list is non-empty) */ |
| 9422 | array = invlist_array(invlist); |
| 9423 | |
| 9424 | mid = invlist_previous_index(invlist); |
| 9425 | assert(mid >=0); |
| 9426 | if (mid > highest_element) { |
| 9427 | mid = highest_element; |
| 9428 | } |
| 9429 | |
| 9430 | /* <mid> contains the cache of the result of the previous call to this |
| 9431 | * function (0 the first time). See if this call is for the same result, |
| 9432 | * or if it is for mid-1. This is under the theory that calls to this |
| 9433 | * function will often be for related code points that are near each other. |
| 9434 | * And benchmarks show that caching gives better results. We also test |
| 9435 | * here if the code point is within the bounds of the list. These tests |
| 9436 | * replace others that would have had to be made anyway to make sure that |
| 9437 | * the array bounds were not exceeded, and these give us extra information |
| 9438 | * at the same time */ |
| 9439 | if (cp >= array[mid]) { |
| 9440 | if (cp >= array[highest_element]) { |
| 9441 | return highest_element; |
| 9442 | } |
| 9443 | |
| 9444 | /* Here, array[mid] <= cp < array[highest_element]. This means that |
| 9445 | * the final element is not the answer, so can exclude it; it also |
| 9446 | * means that <mid> is not the final element, so can refer to 'mid + 1' |
| 9447 | * safely */ |
| 9448 | if (cp < array[mid + 1]) { |
| 9449 | return mid; |
| 9450 | } |
| 9451 | high--; |
| 9452 | low = mid + 1; |
| 9453 | } |
| 9454 | else { /* cp < aray[mid] */ |
| 9455 | if (cp < array[0]) { /* Fail if outside the array */ |
| 9456 | return -1; |
| 9457 | } |
| 9458 | high = mid; |
| 9459 | if (cp >= array[mid - 1]) { |
| 9460 | goto found_entry; |
| 9461 | } |
| 9462 | } |
| 9463 | |
| 9464 | /* Binary search. What we are looking for is <i> such that |
| 9465 | * array[i] <= cp < array[i+1] |
| 9466 | * The loop below converges on the i+1. Note that there may not be an |
| 9467 | * (i+1)th element in the array, and things work nonetheless */ |
| 9468 | while (low < high) { |
| 9469 | mid = (low + high) / 2; |
| 9470 | assert(mid <= highest_element); |
| 9471 | if (array[mid] <= cp) { /* cp >= array[mid] */ |
| 9472 | low = mid + 1; |
| 9473 | |
| 9474 | /* We could do this extra test to exit the loop early. |
| 9475 | if (cp < array[low]) { |
| 9476 | return mid; |
| 9477 | } |
| 9478 | */ |
| 9479 | } |
| 9480 | else { /* cp < array[mid] */ |
| 9481 | high = mid; |
| 9482 | } |
| 9483 | } |
| 9484 | |
| 9485 | found_entry: |
| 9486 | high--; |
| 9487 | invlist_set_previous_index(invlist, high); |
| 9488 | return high; |
| 9489 | } |
| 9490 | |
| 9491 | void |
| 9492 | Perl__invlist_union_maybe_complement_2nd(pTHX_ SV* const a, SV* const b, |
| 9493 | const bool complement_b, SV** output) |
| 9494 | { |
| 9495 | /* Take the union of two inversion lists and point '*output' to it. On |
| 9496 | * input, '*output' MUST POINT TO NULL OR TO AN SV* INVERSION LIST (possibly |
| 9497 | * even 'a' or 'b'). If to an inversion list, the contents of the original |
| 9498 | * list will be replaced by the union. The first list, 'a', may be |
| 9499 | * NULL, in which case a copy of the second list is placed in '*output'. |
| 9500 | * If 'complement_b' is TRUE, the union is taken of the complement |
| 9501 | * (inversion) of 'b' instead of b itself. |
| 9502 | * |
| 9503 | * The basis for this comes from "Unicode Demystified" Chapter 13 by |
| 9504 | * Richard Gillam, published by Addison-Wesley, and explained at some |
| 9505 | * length there. The preface says to incorporate its examples into your |
| 9506 | * code at your own risk. |
| 9507 | * |
| 9508 | * The algorithm is like a merge sort. */ |
| 9509 | |
| 9510 | const UV* array_a; /* a's array */ |
| 9511 | const UV* array_b; |
| 9512 | UV len_a; /* length of a's array */ |
| 9513 | UV len_b; |
| 9514 | |
| 9515 | SV* u; /* the resulting union */ |
| 9516 | UV* array_u; |
| 9517 | UV len_u = 0; |
| 9518 | |
| 9519 | UV i_a = 0; /* current index into a's array */ |
| 9520 | UV i_b = 0; |
| 9521 | UV i_u = 0; |
| 9522 | |
| 9523 | /* running count, as explained in the algorithm source book; items are |
| 9524 | * stopped accumulating and are output when the count changes to/from 0. |
| 9525 | * The count is incremented when we start a range that's in an input's set, |
| 9526 | * and decremented when we start a range that's not in a set. So this |
| 9527 | * variable can be 0, 1, or 2. When it is 0 neither input is in their set, |
| 9528 | * and hence nothing goes into the union; 1, just one of the inputs is in |
| 9529 | * its set (and its current range gets added to the union); and 2 when both |
| 9530 | * inputs are in their sets. */ |
| 9531 | UV count = 0; |
| 9532 | |
| 9533 | PERL_ARGS_ASSERT__INVLIST_UNION_MAYBE_COMPLEMENT_2ND; |
| 9534 | assert(a != b); |
| 9535 | assert(*output == NULL || is_invlist(*output)); |
| 9536 | |
| 9537 | len_b = _invlist_len(b); |
| 9538 | if (len_b == 0) { |
| 9539 | |
| 9540 | /* Here, 'b' is empty, hence it's complement is all possible code |
| 9541 | * points. So if the union includes the complement of 'b', it includes |
| 9542 | * everything, and we need not even look at 'a'. It's easiest to |
| 9543 | * create a new inversion list that matches everything. */ |
| 9544 | if (complement_b) { |
| 9545 | SV* everything = _add_range_to_invlist(NULL, 0, UV_MAX); |
| 9546 | |
| 9547 | if (*output == NULL) { /* If the output didn't exist, just point it |
| 9548 | at the new list */ |
| 9549 | *output = everything; |
| 9550 | } |
| 9551 | else { /* Otherwise, replace its contents with the new list */ |
| 9552 | invlist_replace_list_destroys_src(*output, everything); |
| 9553 | SvREFCNT_dec_NN(everything); |
| 9554 | } |
| 9555 | |
| 9556 | return; |
| 9557 | } |
| 9558 | |
| 9559 | /* Here, we don't want the complement of 'b', and since 'b' is empty, |
| 9560 | * the union will come entirely from 'a'. If 'a' is NULL or empty, the |
| 9561 | * output will be empty */ |
| 9562 | |
| 9563 | if (a == NULL || _invlist_len(a) == 0) { |
| 9564 | if (*output == NULL) { |
| 9565 | *output = _new_invlist(0); |
| 9566 | } |
| 9567 | else { |
| 9568 | invlist_clear(*output); |
| 9569 | } |
| 9570 | return; |
| 9571 | } |
| 9572 | |
| 9573 | /* Here, 'a' is not empty, but 'b' is, so 'a' entirely determines the |
| 9574 | * union. We can just return a copy of 'a' if '*output' doesn't point |
| 9575 | * to an existing list */ |
| 9576 | if (*output == NULL) { |
| 9577 | *output = invlist_clone(a, NULL); |
| 9578 | return; |
| 9579 | } |
| 9580 | |
| 9581 | /* If the output is to overwrite 'a', we have a no-op, as it's |
| 9582 | * already in 'a' */ |
| 9583 | if (*output == a) { |
| 9584 | return; |
| 9585 | } |
| 9586 | |
| 9587 | /* Here, '*output' is to be overwritten by 'a' */ |
| 9588 | u = invlist_clone(a, NULL); |
| 9589 | invlist_replace_list_destroys_src(*output, u); |
| 9590 | SvREFCNT_dec_NN(u); |
| 9591 | |
| 9592 | return; |
| 9593 | } |
| 9594 | |
| 9595 | /* Here 'b' is not empty. See about 'a' */ |
| 9596 | |
| 9597 | if (a == NULL || ((len_a = _invlist_len(a)) == 0)) { |
| 9598 | |
| 9599 | /* Here, 'a' is empty (and b is not). That means the union will come |
| 9600 | * entirely from 'b'. If '*output' is NULL, we can directly return a |
| 9601 | * clone of 'b'. Otherwise, we replace the contents of '*output' with |
| 9602 | * the clone */ |
| 9603 | |
| 9604 | SV ** dest = (*output == NULL) ? output : &u; |
| 9605 | *dest = invlist_clone(b, NULL); |
| 9606 | if (complement_b) { |
| 9607 | _invlist_invert(*dest); |
| 9608 | } |
| 9609 | |
| 9610 | if (dest == &u) { |
| 9611 | invlist_replace_list_destroys_src(*output, u); |
| 9612 | SvREFCNT_dec_NN(u); |
| 9613 | } |
| 9614 | |
| 9615 | return; |
| 9616 | } |
| 9617 | |
| 9618 | /* Here both lists exist and are non-empty */ |
| 9619 | array_a = invlist_array(a); |
| 9620 | array_b = invlist_array(b); |
| 9621 | |
| 9622 | /* If are to take the union of 'a' with the complement of b, set it |
| 9623 | * up so are looking at b's complement. */ |
| 9624 | if (complement_b) { |
| 9625 | |
| 9626 | /* To complement, we invert: if the first element is 0, remove it. To |
| 9627 | * do this, we just pretend the array starts one later */ |
| 9628 | if (array_b[0] == 0) { |
| 9629 | array_b++; |
| 9630 | len_b--; |
| 9631 | } |
| 9632 | else { |
| 9633 | |
| 9634 | /* But if the first element is not zero, we pretend the list starts |
| 9635 | * at the 0 that is always stored immediately before the array. */ |
| 9636 | array_b--; |
| 9637 | len_b++; |
| 9638 | } |
| 9639 | } |
| 9640 | |
| 9641 | /* Size the union for the worst case: that the sets are completely |
| 9642 | * disjoint */ |
| 9643 | u = _new_invlist(len_a + len_b); |
| 9644 | |
| 9645 | /* Will contain U+0000 if either component does */ |
| 9646 | array_u = _invlist_array_init(u, ( len_a > 0 && array_a[0] == 0) |
| 9647 | || (len_b > 0 && array_b[0] == 0)); |
| 9648 | |
| 9649 | /* Go through each input list item by item, stopping when have exhausted |
| 9650 | * one of them */ |
| 9651 | while (i_a < len_a && i_b < len_b) { |
| 9652 | UV cp; /* The element to potentially add to the union's array */ |
| 9653 | bool cp_in_set; /* is it in the the input list's set or not */ |
| 9654 | |
| 9655 | /* We need to take one or the other of the two inputs for the union. |
| 9656 | * Since we are merging two sorted lists, we take the smaller of the |
| 9657 | * next items. In case of a tie, we take first the one that is in its |
| 9658 | * set. If we first took the one not in its set, it would decrement |
| 9659 | * the count, possibly to 0 which would cause it to be output as ending |
| 9660 | * the range, and the next time through we would take the same number, |
| 9661 | * and output it again as beginning the next range. By doing it the |
| 9662 | * opposite way, there is no possibility that the count will be |
| 9663 | * momentarily decremented to 0, and thus the two adjoining ranges will |
| 9664 | * be seamlessly merged. (In a tie and both are in the set or both not |
| 9665 | * in the set, it doesn't matter which we take first.) */ |
| 9666 | if ( array_a[i_a] < array_b[i_b] |
| 9667 | || ( array_a[i_a] == array_b[i_b] |
| 9668 | && ELEMENT_RANGE_MATCHES_INVLIST(i_a))) |
| 9669 | { |
| 9670 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a); |
| 9671 | cp = array_a[i_a++]; |
| 9672 | } |
| 9673 | else { |
| 9674 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b); |
| 9675 | cp = array_b[i_b++]; |
| 9676 | } |
| 9677 | |
| 9678 | /* Here, have chosen which of the two inputs to look at. Only output |
| 9679 | * if the running count changes to/from 0, which marks the |
| 9680 | * beginning/end of a range that's in the set */ |
| 9681 | if (cp_in_set) { |
| 9682 | if (count == 0) { |
| 9683 | array_u[i_u++] = cp; |
| 9684 | } |
| 9685 | count++; |
| 9686 | } |
| 9687 | else { |
| 9688 | count--; |
| 9689 | if (count == 0) { |
| 9690 | array_u[i_u++] = cp; |
| 9691 | } |
| 9692 | } |
| 9693 | } |
| 9694 | |
| 9695 | |
| 9696 | /* The loop above increments the index into exactly one of the input lists |
| 9697 | * each iteration, and ends when either index gets to its list end. That |
| 9698 | * means the other index is lower than its end, and so something is |
| 9699 | * remaining in that one. We decrement 'count', as explained below, if |
| 9700 | * that list is in its set. (i_a and i_b each currently index the element |
| 9701 | * beyond the one we care about.) */ |
| 9702 | if ( (i_a != len_a && PREV_RANGE_MATCHES_INVLIST(i_a)) |
| 9703 | || (i_b != len_b && PREV_RANGE_MATCHES_INVLIST(i_b))) |
| 9704 | { |
| 9705 | count--; |
| 9706 | } |
| 9707 | |
| 9708 | /* Above we decremented 'count' if the list that had unexamined elements in |
| 9709 | * it was in its set. This has made it so that 'count' being non-zero |
| 9710 | * means there isn't anything left to output; and 'count' equal to 0 means |
| 9711 | * that what is left to output is precisely that which is left in the |
| 9712 | * non-exhausted input list. |
| 9713 | * |
| 9714 | * To see why, note first that the exhausted input obviously has nothing |
| 9715 | * left to add to the union. If it was in its set at its end, that means |
| 9716 | * the set extends from here to the platform's infinity, and hence so does |
| 9717 | * the union and the non-exhausted set is irrelevant. The exhausted set |
| 9718 | * also contributed 1 to 'count'. If 'count' was 2, it got decremented to |
| 9719 | * 1, but if it was 1, the non-exhausted set wasn't in its set, and so |
| 9720 | * 'count' remains at 1. This is consistent with the decremented 'count' |
| 9721 | * != 0 meaning there's nothing left to add to the union. |
| 9722 | * |
| 9723 | * But if the exhausted input wasn't in its set, it contributed 0 to |
| 9724 | * 'count', and the rest of the union will be whatever the other input is. |
| 9725 | * If 'count' was 0, neither list was in its set, and 'count' remains 0; |
| 9726 | * otherwise it gets decremented to 0. This is consistent with 'count' |
| 9727 | * == 0 meaning the remainder of the union is whatever is left in the |
| 9728 | * non-exhausted list. */ |
| 9729 | if (count != 0) { |
| 9730 | len_u = i_u; |
| 9731 | } |
| 9732 | else { |
| 9733 | IV copy_count = len_a - i_a; |
| 9734 | if (copy_count > 0) { /* The non-exhausted input is 'a' */ |
| 9735 | Copy(array_a + i_a, array_u + i_u, copy_count, UV); |
| 9736 | } |
| 9737 | else { /* The non-exhausted input is b */ |
| 9738 | copy_count = len_b - i_b; |
| 9739 | Copy(array_b + i_b, array_u + i_u, copy_count, UV); |
| 9740 | } |
| 9741 | len_u = i_u + copy_count; |
| 9742 | } |
| 9743 | |
| 9744 | /* Set the result to the final length, which can change the pointer to |
| 9745 | * array_u, so re-find it. (Note that it is unlikely that this will |
| 9746 | * change, as we are shrinking the space, not enlarging it) */ |
| 9747 | if (len_u != _invlist_len(u)) { |
| 9748 | invlist_set_len(u, len_u, *get_invlist_offset_addr(u)); |
| 9749 | invlist_trim(u); |
| 9750 | array_u = invlist_array(u); |
| 9751 | } |
| 9752 | |
| 9753 | if (*output == NULL) { /* Simply return the new inversion list */ |
| 9754 | *output = u; |
| 9755 | } |
| 9756 | else { |
| 9757 | /* Otherwise, overwrite the inversion list that was in '*output'. We |
| 9758 | * could instead free '*output', and then set it to 'u', but experience |
| 9759 | * has shown [perl #127392] that if the input is a mortal, we can get a |
| 9760 | * huge build-up of these during regex compilation before they get |
| 9761 | * freed. */ |
| 9762 | invlist_replace_list_destroys_src(*output, u); |
| 9763 | SvREFCNT_dec_NN(u); |
| 9764 | } |
| 9765 | |
| 9766 | return; |
| 9767 | } |
| 9768 | |
| 9769 | void |
| 9770 | Perl__invlist_intersection_maybe_complement_2nd(pTHX_ SV* const a, SV* const b, |
| 9771 | const bool complement_b, SV** i) |
| 9772 | { |
| 9773 | /* Take the intersection of two inversion lists and point '*i' to it. On |
| 9774 | * input, '*i' MUST POINT TO NULL OR TO AN SV* INVERSION LIST (possibly |
| 9775 | * even 'a' or 'b'). If to an inversion list, the contents of the original |
| 9776 | * list will be replaced by the intersection. The first list, 'a', may be |
| 9777 | * NULL, in which case '*i' will be an empty list. If 'complement_b' is |
| 9778 | * TRUE, the result will be the intersection of 'a' and the complement (or |
| 9779 | * inversion) of 'b' instead of 'b' directly. |
| 9780 | * |
| 9781 | * The basis for this comes from "Unicode Demystified" Chapter 13 by |
| 9782 | * Richard Gillam, published by Addison-Wesley, and explained at some |
| 9783 | * length there. The preface says to incorporate its examples into your |
| 9784 | * code at your own risk. In fact, it had bugs |
| 9785 | * |
| 9786 | * The algorithm is like a merge sort, and is essentially the same as the |
| 9787 | * union above |
| 9788 | */ |
| 9789 | |
| 9790 | const UV* array_a; /* a's array */ |
| 9791 | const UV* array_b; |
| 9792 | UV len_a; /* length of a's array */ |
| 9793 | UV len_b; |
| 9794 | |
| 9795 | SV* r; /* the resulting intersection */ |
| 9796 | UV* array_r; |
| 9797 | UV len_r = 0; |
| 9798 | |
| 9799 | UV i_a = 0; /* current index into a's array */ |
| 9800 | UV i_b = 0; |
| 9801 | UV i_r = 0; |
| 9802 | |
| 9803 | /* running count of how many of the two inputs are postitioned at ranges |
| 9804 | * that are in their sets. As explained in the algorithm source book, |
| 9805 | * items are stopped accumulating and are output when the count changes |
| 9806 | * to/from 2. The count is incremented when we start a range that's in an |
| 9807 | * input's set, and decremented when we start a range that's not in a set. |
| 9808 | * Only when it is 2 are we in the intersection. */ |
| 9809 | UV count = 0; |
| 9810 | |
| 9811 | PERL_ARGS_ASSERT__INVLIST_INTERSECTION_MAYBE_COMPLEMENT_2ND; |
| 9812 | assert(a != b); |
| 9813 | assert(*i == NULL || is_invlist(*i)); |
| 9814 | |
| 9815 | /* Special case if either one is empty */ |
| 9816 | len_a = (a == NULL) ? 0 : _invlist_len(a); |
| 9817 | if ((len_a == 0) || ((len_b = _invlist_len(b)) == 0)) { |
| 9818 | if (len_a != 0 && complement_b) { |
| 9819 | |
| 9820 | /* Here, 'a' is not empty, therefore from the enclosing 'if', 'b' |
| 9821 | * must be empty. Here, also we are using 'b's complement, which |
| 9822 | * hence must be every possible code point. Thus the intersection |
| 9823 | * is simply 'a'. */ |
| 9824 | |
| 9825 | if (*i == a) { /* No-op */ |
| 9826 | return; |
| 9827 | } |
| 9828 | |
| 9829 | if (*i == NULL) { |
| 9830 | *i = invlist_clone(a, NULL); |
| 9831 | return; |
| 9832 | } |
| 9833 | |
| 9834 | r = invlist_clone(a, NULL); |
| 9835 | invlist_replace_list_destroys_src(*i, r); |
| 9836 | SvREFCNT_dec_NN(r); |
| 9837 | return; |
| 9838 | } |
| 9839 | |
| 9840 | /* Here, 'a' or 'b' is empty and not using the complement of 'b'. The |
| 9841 | * intersection must be empty */ |
| 9842 | if (*i == NULL) { |
| 9843 | *i = _new_invlist(0); |
| 9844 | return; |
| 9845 | } |
| 9846 | |
| 9847 | invlist_clear(*i); |
| 9848 | return; |
| 9849 | } |
| 9850 | |
| 9851 | /* Here both lists exist and are non-empty */ |
| 9852 | array_a = invlist_array(a); |
| 9853 | array_b = invlist_array(b); |
| 9854 | |
| 9855 | /* If are to take the intersection of 'a' with the complement of b, set it |
| 9856 | * up so are looking at b's complement. */ |
| 9857 | if (complement_b) { |
| 9858 | |
| 9859 | /* To complement, we invert: if the first element is 0, remove it. To |
| 9860 | * do this, we just pretend the array starts one later */ |
| 9861 | if (array_b[0] == 0) { |
| 9862 | array_b++; |
| 9863 | len_b--; |
| 9864 | } |
| 9865 | else { |
| 9866 | |
| 9867 | /* But if the first element is not zero, we pretend the list starts |
| 9868 | * at the 0 that is always stored immediately before the array. */ |
| 9869 | array_b--; |
| 9870 | len_b++; |
| 9871 | } |
| 9872 | } |
| 9873 | |
| 9874 | /* Size the intersection for the worst case: that the intersection ends up |
| 9875 | * fragmenting everything to be completely disjoint */ |
| 9876 | r= _new_invlist(len_a + len_b); |
| 9877 | |
| 9878 | /* Will contain U+0000 iff both components do */ |
| 9879 | array_r = _invlist_array_init(r, len_a > 0 && array_a[0] == 0 |
| 9880 | && len_b > 0 && array_b[0] == 0); |
| 9881 | |
| 9882 | /* Go through each list item by item, stopping when have exhausted one of |
| 9883 | * them */ |
| 9884 | while (i_a < len_a && i_b < len_b) { |
| 9885 | UV cp; /* The element to potentially add to the intersection's |
| 9886 | array */ |
| 9887 | bool cp_in_set; /* Is it in the input list's set or not */ |
| 9888 | |
| 9889 | /* We need to take one or the other of the two inputs for the |
| 9890 | * intersection. Since we are merging two sorted lists, we take the |
| 9891 | * smaller of the next items. In case of a tie, we take first the one |
| 9892 | * that is not in its set (a difference from the union algorithm). If |
| 9893 | * we first took the one in its set, it would increment the count, |
| 9894 | * possibly to 2 which would cause it to be output as starting a range |
| 9895 | * in the intersection, and the next time through we would take that |
| 9896 | * same number, and output it again as ending the set. By doing the |
| 9897 | * opposite of this, there is no possibility that the count will be |
| 9898 | * momentarily incremented to 2. (In a tie and both are in the set or |
| 9899 | * both not in the set, it doesn't matter which we take first.) */ |
| 9900 | if ( array_a[i_a] < array_b[i_b] |
| 9901 | || ( array_a[i_a] == array_b[i_b] |
| 9902 | && ! ELEMENT_RANGE_MATCHES_INVLIST(i_a))) |
| 9903 | { |
| 9904 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a); |
| 9905 | cp = array_a[i_a++]; |
| 9906 | } |
| 9907 | else { |
| 9908 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b); |
| 9909 | cp= array_b[i_b++]; |
| 9910 | } |
| 9911 | |
| 9912 | /* Here, have chosen which of the two inputs to look at. Only output |
| 9913 | * if the running count changes to/from 2, which marks the |
| 9914 | * beginning/end of a range that's in the intersection */ |
| 9915 | if (cp_in_set) { |
| 9916 | count++; |
| 9917 | if (count == 2) { |
| 9918 | array_r[i_r++] = cp; |
| 9919 | } |
| 9920 | } |
| 9921 | else { |
| 9922 | if (count == 2) { |
| 9923 | array_r[i_r++] = cp; |
| 9924 | } |
| 9925 | count--; |
| 9926 | } |
| 9927 | |
| 9928 | } |
| 9929 | |
| 9930 | /* The loop above increments the index into exactly one of the input lists |
| 9931 | * each iteration, and ends when either index gets to its list end. That |
| 9932 | * means the other index is lower than its end, and so something is |
| 9933 | * remaining in that one. We increment 'count', as explained below, if the |
| 9934 | * exhausted list was in its set. (i_a and i_b each currently index the |
| 9935 | * element beyond the one we care about.) */ |
| 9936 | if ( (i_a == len_a && PREV_RANGE_MATCHES_INVLIST(i_a)) |
| 9937 | || (i_b == len_b && PREV_RANGE_MATCHES_INVLIST(i_b))) |
| 9938 | { |
| 9939 | count++; |
| 9940 | } |
| 9941 | |
| 9942 | /* Above we incremented 'count' if the exhausted list was in its set. This |
| 9943 | * has made it so that 'count' being below 2 means there is nothing left to |
| 9944 | * output; otheriwse what's left to add to the intersection is precisely |
| 9945 | * that which is left in the non-exhausted input list. |
| 9946 | * |
| 9947 | * To see why, note first that the exhausted input obviously has nothing |
| 9948 | * left to affect the intersection. If it was in its set at its end, that |
| 9949 | * means the set extends from here to the platform's infinity, and hence |
| 9950 | * anything in the non-exhausted's list will be in the intersection, and |
| 9951 | * anything not in it won't be. Hence, the rest of the intersection is |
| 9952 | * precisely what's in the non-exhausted list The exhausted set also |
| 9953 | * contributed 1 to 'count', meaning 'count' was at least 1. Incrementing |
| 9954 | * it means 'count' is now at least 2. This is consistent with the |
| 9955 | * incremented 'count' being >= 2 means to add the non-exhausted list to |
| 9956 | * the intersection. |
| 9957 | * |
| 9958 | * But if the exhausted input wasn't in its set, it contributed 0 to |
| 9959 | * 'count', and the intersection can't include anything further; the |
| 9960 | * non-exhausted set is irrelevant. 'count' was at most 1, and doesn't get |
| 9961 | * incremented. This is consistent with 'count' being < 2 meaning nothing |
| 9962 | * further to add to the intersection. */ |
| 9963 | if (count < 2) { /* Nothing left to put in the intersection. */ |
| 9964 | len_r = i_r; |
| 9965 | } |
| 9966 | else { /* copy the non-exhausted list, unchanged. */ |
| 9967 | IV copy_count = len_a - i_a; |
| 9968 | if (copy_count > 0) { /* a is the one with stuff left */ |
| 9969 | Copy(array_a + i_a, array_r + i_r, copy_count, UV); |
| 9970 | } |
| 9971 | else { /* b is the one with stuff left */ |
| 9972 | copy_count = len_b - i_b; |
| 9973 | Copy(array_b + i_b, array_r + i_r, copy_count, UV); |
| 9974 | } |
| 9975 | len_r = i_r + copy_count; |
| 9976 | } |
| 9977 | |
| 9978 | /* Set the result to the final length, which can change the pointer to |
| 9979 | * array_r, so re-find it. (Note that it is unlikely that this will |
| 9980 | * change, as we are shrinking the space, not enlarging it) */ |
| 9981 | if (len_r != _invlist_len(r)) { |
| 9982 | invlist_set_len(r, len_r, *get_invlist_offset_addr(r)); |
| 9983 | invlist_trim(r); |
| 9984 | array_r = invlist_array(r); |
| 9985 | } |
| 9986 | |
| 9987 | if (*i == NULL) { /* Simply return the calculated intersection */ |
| 9988 | *i = r; |
| 9989 | } |
| 9990 | else { /* Otherwise, replace the existing inversion list in '*i'. We could |
| 9991 | instead free '*i', and then set it to 'r', but experience has |
| 9992 | shown [perl #127392] that if the input is a mortal, we can get a |
| 9993 | huge build-up of these during regex compilation before they get |
| 9994 | freed. */ |
| 9995 | if (len_r) { |
| 9996 | invlist_replace_list_destroys_src(*i, r); |
| 9997 | } |
| 9998 | else { |
| 9999 | invlist_clear(*i); |
| 10000 | } |
| 10001 | SvREFCNT_dec_NN(r); |
| 10002 | } |
| 10003 | |
| 10004 | return; |
| 10005 | } |
| 10006 | |
| 10007 | SV* |
| 10008 | Perl__add_range_to_invlist(pTHX_ SV* invlist, UV start, UV end) |
| 10009 | { |
| 10010 | /* Add the range from 'start' to 'end' inclusive to the inversion list's |
| 10011 | * set. A pointer to the inversion list is returned. This may actually be |
| 10012 | * a new list, in which case the passed in one has been destroyed. The |
| 10013 | * passed-in inversion list can be NULL, in which case a new one is created |
| 10014 | * with just the one range in it. The new list is not necessarily |
| 10015 | * NUL-terminated. Space is not freed if the inversion list shrinks as a |
| 10016 | * result of this function. The gain would not be large, and in many |
| 10017 | * cases, this is called multiple times on a single inversion list, so |
| 10018 | * anything freed may almost immediately be needed again. |
| 10019 | * |
| 10020 | * This used to mostly call the 'union' routine, but that is much more |
| 10021 | * heavyweight than really needed for a single range addition */ |
| 10022 | |
| 10023 | UV* array; /* The array implementing the inversion list */ |
| 10024 | UV len; /* How many elements in 'array' */ |
| 10025 | SSize_t i_s; /* index into the invlist array where 'start' |
| 10026 | should go */ |
| 10027 | SSize_t i_e = 0; /* And the index where 'end' should go */ |
| 10028 | UV cur_highest; /* The highest code point in the inversion list |
| 10029 | upon entry to this function */ |
| 10030 | |
| 10031 | /* This range becomes the whole inversion list if none already existed */ |
| 10032 | if (invlist == NULL) { |
| 10033 | invlist = _new_invlist(2); |
| 10034 | _append_range_to_invlist(invlist, start, end); |
| 10035 | return invlist; |
| 10036 | } |
| 10037 | |
| 10038 | /* Likewise, if the inversion list is currently empty */ |
| 10039 | len = _invlist_len(invlist); |
| 10040 | if (len == 0) { |
| 10041 | _append_range_to_invlist(invlist, start, end); |
| 10042 | return invlist; |
| 10043 | } |
| 10044 | |
| 10045 | /* Starting here, we have to know the internals of the list */ |
| 10046 | array = invlist_array(invlist); |
| 10047 | |
| 10048 | /* If the new range ends higher than the current highest ... */ |
| 10049 | cur_highest = invlist_highest(invlist); |
| 10050 | if (end > cur_highest) { |
| 10051 | |
| 10052 | /* If the whole range is higher, we can just append it */ |
| 10053 | if (start > cur_highest) { |
| 10054 | _append_range_to_invlist(invlist, start, end); |
| 10055 | return invlist; |
| 10056 | } |
| 10057 | |
| 10058 | /* Otherwise, add the portion that is higher ... */ |
| 10059 | _append_range_to_invlist(invlist, cur_highest + 1, end); |
| 10060 | |
| 10061 | /* ... and continue on below to handle the rest. As a result of the |
| 10062 | * above append, we know that the index of the end of the range is the |
| 10063 | * final even numbered one of the array. Recall that the final element |
| 10064 | * always starts a range that extends to infinity. If that range is in |
| 10065 | * the set (meaning the set goes from here to infinity), it will be an |
| 10066 | * even index, but if it isn't in the set, it's odd, and the final |
| 10067 | * range in the set is one less, which is even. */ |
| 10068 | if (end == UV_MAX) { |
| 10069 | i_e = len; |
| 10070 | } |
| 10071 | else { |
| 10072 | i_e = len - 2; |
| 10073 | } |
| 10074 | } |
| 10075 | |
| 10076 | /* We have dealt with appending, now see about prepending. If the new |
| 10077 | * range starts lower than the current lowest ... */ |
| 10078 | if (start < array[0]) { |
| 10079 | |
| 10080 | /* Adding something which has 0 in it is somewhat tricky, and uncommon. |
| 10081 | * Let the union code handle it, rather than having to know the |
| 10082 | * trickiness in two code places. */ |
| 10083 | if (UNLIKELY(start == 0)) { |
| 10084 | SV* range_invlist; |
| 10085 | |
| 10086 | range_invlist = _new_invlist(2); |
| 10087 | _append_range_to_invlist(range_invlist, start, end); |
| 10088 | |
| 10089 | _invlist_union(invlist, range_invlist, &invlist); |
| 10090 | |
| 10091 | SvREFCNT_dec_NN(range_invlist); |
| 10092 | |
| 10093 | return invlist; |
| 10094 | } |
| 10095 | |
| 10096 | /* If the whole new range comes before the first entry, and doesn't |
| 10097 | * extend it, we have to insert it as an additional range */ |
| 10098 | if (end < array[0] - 1) { |
| 10099 | i_s = i_e = -1; |
| 10100 | goto splice_in_new_range; |
| 10101 | } |
| 10102 | |
| 10103 | /* Here the new range adjoins the existing first range, extending it |
| 10104 | * downwards. */ |
| 10105 | array[0] = start; |
| 10106 | |
| 10107 | /* And continue on below to handle the rest. We know that the index of |
| 10108 | * the beginning of the range is the first one of the array */ |
| 10109 | i_s = 0; |
| 10110 | } |
| 10111 | else { /* Not prepending any part of the new range to the existing list. |
| 10112 | * Find where in the list it should go. This finds i_s, such that: |
| 10113 | * invlist[i_s] <= start < array[i_s+1] |
| 10114 | */ |
| 10115 | i_s = _invlist_search(invlist, start); |
| 10116 | } |
| 10117 | |
| 10118 | /* At this point, any extending before the beginning of the inversion list |
| 10119 | * and/or after the end has been done. This has made it so that, in the |
| 10120 | * code below, each endpoint of the new range is either in a range that is |
| 10121 | * in the set, or is in a gap between two ranges that are. This means we |
| 10122 | * don't have to worry about exceeding the array bounds. |
| 10123 | * |
| 10124 | * Find where in the list the new range ends (but we can skip this if we |
| 10125 | * have already determined what it is, or if it will be the same as i_s, |
| 10126 | * which we already have computed) */ |
| 10127 | if (i_e == 0) { |
| 10128 | i_e = (start == end) |
| 10129 | ? i_s |
| 10130 | : _invlist_search(invlist, end); |
| 10131 | } |
| 10132 | |
| 10133 | /* Here generally invlist[i_e] <= end < array[i_e+1]. But if invlist[i_e] |
| 10134 | * is a range that goes to infinity there is no element at invlist[i_e+1], |
| 10135 | * so only the first relation holds. */ |
| 10136 | |
| 10137 | if ( ! ELEMENT_RANGE_MATCHES_INVLIST(i_s)) { |
| 10138 | |
| 10139 | /* Here, the ranges on either side of the beginning of the new range |
| 10140 | * are in the set, and this range starts in the gap between them. |
| 10141 | * |
| 10142 | * The new range extends the range above it downwards if the new range |
| 10143 | * ends at or above that range's start */ |
| 10144 | const bool extends_the_range_above = ( end == UV_MAX |
| 10145 | || end + 1 >= array[i_s+1]); |
| 10146 | |
| 10147 | /* The new range extends the range below it upwards if it begins just |
| 10148 | * after where that range ends */ |
| 10149 | if (start == array[i_s]) { |
| 10150 | |
| 10151 | /* If the new range fills the entire gap between the other ranges, |
| 10152 | * they will get merged together. Other ranges may also get |
| 10153 | * merged, depending on how many of them the new range spans. In |
| 10154 | * the general case, we do the merge later, just once, after we |
| 10155 | * figure out how many to merge. But in the case where the new |
| 10156 | * range exactly spans just this one gap (possibly extending into |
| 10157 | * the one above), we do the merge here, and an early exit. This |
| 10158 | * is done here to avoid having to special case later. */ |
| 10159 | if (i_e - i_s <= 1) { |
| 10160 | |
| 10161 | /* If i_e - i_s == 1, it means that the new range terminates |
| 10162 | * within the range above, and hence 'extends_the_range_above' |
| 10163 | * must be true. (If the range above it extends to infinity, |
| 10164 | * 'i_s+2' will be above the array's limit, but 'len-i_s-2' |
| 10165 | * will be 0, so no harm done.) */ |
| 10166 | if (extends_the_range_above) { |
| 10167 | Move(array + i_s + 2, array + i_s, len - i_s - 2, UV); |
| 10168 | invlist_set_len(invlist, |
| 10169 | len - 2, |
| 10170 | *(get_invlist_offset_addr(invlist))); |
| 10171 | return invlist; |
| 10172 | } |
| 10173 | |
| 10174 | /* Here, i_e must == i_s. We keep them in sync, as they apply |
| 10175 | * to the same range, and below we are about to decrement i_s |
| 10176 | * */ |
| 10177 | i_e--; |
| 10178 | } |
| 10179 | |
| 10180 | /* Here, the new range is adjacent to the one below. (It may also |
| 10181 | * span beyond the range above, but that will get resolved later.) |
| 10182 | * Extend the range below to include this one. */ |
| 10183 | array[i_s] = (end == UV_MAX) ? UV_MAX : end + 1; |
| 10184 | i_s--; |
| 10185 | start = array[i_s]; |
| 10186 | } |
| 10187 | else if (extends_the_range_above) { |
| 10188 | |
| 10189 | /* Here the new range only extends the range above it, but not the |
| 10190 | * one below. It merges with the one above. Again, we keep i_e |
| 10191 | * and i_s in sync if they point to the same range */ |
| 10192 | if (i_e == i_s) { |
| 10193 | i_e++; |
| 10194 | } |
| 10195 | i_s++; |
| 10196 | array[i_s] = start; |
| 10197 | } |
| 10198 | } |
| 10199 | |
| 10200 | /* Here, we've dealt with the new range start extending any adjoining |
| 10201 | * existing ranges. |
| 10202 | * |
| 10203 | * If the new range extends to infinity, it is now the final one, |
| 10204 | * regardless of what was there before */ |
| 10205 | if (UNLIKELY(end == UV_MAX)) { |
| 10206 | invlist_set_len(invlist, i_s + 1, *(get_invlist_offset_addr(invlist))); |
| 10207 | return invlist; |
| 10208 | } |
| 10209 | |
| 10210 | /* If i_e started as == i_s, it has also been dealt with, |
| 10211 | * and been updated to the new i_s, which will fail the following if */ |
| 10212 | if (! ELEMENT_RANGE_MATCHES_INVLIST(i_e)) { |
| 10213 | |
| 10214 | /* Here, the ranges on either side of the end of the new range are in |
| 10215 | * the set, and this range ends in the gap between them. |
| 10216 | * |
| 10217 | * If this range is adjacent to (hence extends) the range above it, it |
| 10218 | * becomes part of that range; likewise if it extends the range below, |
| 10219 | * it becomes part of that range */ |
| 10220 | if (end + 1 == array[i_e+1]) { |
| 10221 | i_e++; |
| 10222 | array[i_e] = start; |
| 10223 | } |
| 10224 | else if (start <= array[i_e]) { |
| 10225 | array[i_e] = end + 1; |
| 10226 | i_e--; |
| 10227 | } |
| 10228 | } |
| 10229 | |
| 10230 | if (i_s == i_e) { |
| 10231 | |
| 10232 | /* If the range fits entirely in an existing range (as possibly already |
| 10233 | * extended above), it doesn't add anything new */ |
| 10234 | if (ELEMENT_RANGE_MATCHES_INVLIST(i_s)) { |
| 10235 | return invlist; |
| 10236 | } |
| 10237 | |
| 10238 | /* Here, no part of the range is in the list. Must add it. It will |
| 10239 | * occupy 2 more slots */ |
| 10240 | splice_in_new_range: |
| 10241 | |
| 10242 | invlist_extend(invlist, len + 2); |
| 10243 | array = invlist_array(invlist); |
| 10244 | /* Move the rest of the array down two slots. Don't include any |
| 10245 | * trailing NUL */ |
| 10246 | Move(array + i_e + 1, array + i_e + 3, len - i_e - 1, UV); |
| 10247 | |
| 10248 | /* Do the actual splice */ |
| 10249 | array[i_e+1] = start; |
| 10250 | array[i_e+2] = end + 1; |
| 10251 | invlist_set_len(invlist, len + 2, *(get_invlist_offset_addr(invlist))); |
| 10252 | return invlist; |
| 10253 | } |
| 10254 | |
| 10255 | /* Here the new range crossed the boundaries of a pre-existing range. The |
| 10256 | * code above has adjusted things so that both ends are in ranges that are |
| 10257 | * in the set. This means everything in between must also be in the set. |
| 10258 | * Just squash things together */ |
| 10259 | Move(array + i_e + 1, array + i_s + 1, len - i_e - 1, UV); |
| 10260 | invlist_set_len(invlist, |
| 10261 | len - i_e + i_s, |
| 10262 | *(get_invlist_offset_addr(invlist))); |
| 10263 | |
| 10264 | return invlist; |
| 10265 | } |
| 10266 | |
| 10267 | SV* |
| 10268 | Perl__setup_canned_invlist(pTHX_ const STRLEN size, const UV element0, |
| 10269 | UV** other_elements_ptr) |
| 10270 | { |
| 10271 | /* Create and return an inversion list whose contents are to be populated |
| 10272 | * by the caller. The caller gives the number of elements (in 'size') and |
| 10273 | * the very first element ('element0'). This function will set |
| 10274 | * '*other_elements_ptr' to an array of UVs, where the remaining elements |
| 10275 | * are to be placed. |
| 10276 | * |
| 10277 | * Obviously there is some trust involved that the caller will properly |
| 10278 | * fill in the other elements of the array. |
| 10279 | * |
| 10280 | * (The first element needs to be passed in, as the underlying code does |
| 10281 | * things differently depending on whether it is zero or non-zero) */ |
| 10282 | |
| 10283 | SV* invlist = _new_invlist(size); |
| 10284 | bool offset; |
| 10285 | |
| 10286 | PERL_ARGS_ASSERT__SETUP_CANNED_INVLIST; |
| 10287 | |
| 10288 | invlist = add_cp_to_invlist(invlist, element0); |
| 10289 | offset = *get_invlist_offset_addr(invlist); |
| 10290 | |
| 10291 | invlist_set_len(invlist, size, offset); |
| 10292 | *other_elements_ptr = invlist_array(invlist) + 1; |
| 10293 | return invlist; |
| 10294 | } |
| 10295 | |
| 10296 | #endif |
| 10297 | |
| 10298 | #ifndef PERL_IN_XSUB_RE |
| 10299 | void |
| 10300 | Perl__invlist_invert(pTHX_ SV* const invlist) |
| 10301 | { |
| 10302 | /* Complement the input inversion list. This adds a 0 if the list didn't |
| 10303 | * have a zero; removes it otherwise. As described above, the data |
| 10304 | * structure is set up so that this is very efficient */ |
| 10305 | |
| 10306 | PERL_ARGS_ASSERT__INVLIST_INVERT; |
| 10307 | |
| 10308 | assert(! invlist_is_iterating(invlist)); |
| 10309 | |
| 10310 | /* The inverse of matching nothing is matching everything */ |
| 10311 | if (_invlist_len(invlist) == 0) { |
| 10312 | _append_range_to_invlist(invlist, 0, UV_MAX); |
| 10313 | return; |
| 10314 | } |
| 10315 | |
| 10316 | *get_invlist_offset_addr(invlist) = ! *get_invlist_offset_addr(invlist); |
| 10317 | } |
| 10318 | |
| 10319 | SV* |
| 10320 | Perl_invlist_clone(pTHX_ SV* const invlist, SV* new_invlist) |
| 10321 | { |
| 10322 | /* Return a new inversion list that is a copy of the input one, which is |
| 10323 | * unchanged. The new list will not be mortal even if the old one was. */ |
| 10324 | |
| 10325 | const STRLEN nominal_length = _invlist_len(invlist); |
| 10326 | const STRLEN physical_length = SvCUR(invlist); |
| 10327 | const bool offset = *(get_invlist_offset_addr(invlist)); |
| 10328 | |
| 10329 | PERL_ARGS_ASSERT_INVLIST_CLONE; |
| 10330 | |
| 10331 | if (new_invlist == NULL) { |
| 10332 | new_invlist = _new_invlist(nominal_length); |
| 10333 | } |
| 10334 | else { |
| 10335 | sv_upgrade(new_invlist, SVt_INVLIST); |
| 10336 | initialize_invlist_guts(new_invlist, nominal_length); |
| 10337 | } |
| 10338 | |
| 10339 | *(get_invlist_offset_addr(new_invlist)) = offset; |
| 10340 | invlist_set_len(new_invlist, nominal_length, offset); |
| 10341 | Copy(SvPVX(invlist), SvPVX(new_invlist), physical_length, char); |
| 10342 | |
| 10343 | return new_invlist; |
| 10344 | } |
| 10345 | |
| 10346 | #endif |
| 10347 | |
| 10348 | PERL_STATIC_INLINE UV |
| 10349 | S_invlist_lowest(SV* const invlist) |
| 10350 | { |
| 10351 | /* Returns the lowest code point that matches an inversion list. This API |
| 10352 | * has an ambiguity, as it returns 0 under either the lowest is actually |
| 10353 | * 0, or if the list is empty. If this distinction matters to you, check |
| 10354 | * for emptiness before calling this function */ |
| 10355 | |
| 10356 | UV len = _invlist_len(invlist); |
| 10357 | UV *array; |
| 10358 | |
| 10359 | PERL_ARGS_ASSERT_INVLIST_LOWEST; |
| 10360 | |
| 10361 | if (len == 0) { |
| 10362 | return 0; |
| 10363 | } |
| 10364 | |
| 10365 | array = invlist_array(invlist); |
| 10366 | |
| 10367 | return array[0]; |
| 10368 | } |
| 10369 | |
| 10370 | STATIC SV * |
| 10371 | S_invlist_contents(pTHX_ SV* const invlist, const bool traditional_style) |
| 10372 | { |
| 10373 | /* Get the contents of an inversion list into a string SV so that they can |
| 10374 | * be printed out. If 'traditional_style' is TRUE, it uses the format |
| 10375 | * traditionally done for debug tracing; otherwise it uses a format |
| 10376 | * suitable for just copying to the output, with blanks between ranges and |
| 10377 | * a dash between range components */ |
| 10378 | |
| 10379 | UV start, end; |
| 10380 | SV* output; |
| 10381 | const char intra_range_delimiter = (traditional_style ? '\t' : '-'); |
| 10382 | const char inter_range_delimiter = (traditional_style ? '\n' : ' '); |
| 10383 | |
| 10384 | if (traditional_style) { |
| 10385 | output = newSVpvs("\n"); |
| 10386 | } |
| 10387 | else { |
| 10388 | output = newSVpvs(""); |
| 10389 | } |
| 10390 | |
| 10391 | PERL_ARGS_ASSERT_INVLIST_CONTENTS; |
| 10392 | |
| 10393 | assert(! invlist_is_iterating(invlist)); |
| 10394 | |
| 10395 | invlist_iterinit(invlist); |
| 10396 | while (invlist_iternext(invlist, &start, &end)) { |
| 10397 | if (end == UV_MAX) { |
| 10398 | Perl_sv_catpvf(aTHX_ output, "%04" UVXf "%cINFTY%c", |
| 10399 | start, intra_range_delimiter, |
| 10400 | inter_range_delimiter); |
| 10401 | } |
| 10402 | else if (end != start) { |
| 10403 | Perl_sv_catpvf(aTHX_ output, "%04" UVXf "%c%04" UVXf "%c", |
| 10404 | start, |
| 10405 | intra_range_delimiter, |
| 10406 | end, inter_range_delimiter); |
| 10407 | } |
| 10408 | else { |
| 10409 | Perl_sv_catpvf(aTHX_ output, "%04" UVXf "%c", |
| 10410 | start, inter_range_delimiter); |
| 10411 | } |
| 10412 | } |
| 10413 | |
| 10414 | if (SvCUR(output) && ! traditional_style) {/* Get rid of trailing blank */ |
| 10415 | SvCUR_set(output, SvCUR(output) - 1); |
| 10416 | } |
| 10417 | |
| 10418 | return output; |
| 10419 | } |
| 10420 | |
| 10421 | #ifndef PERL_IN_XSUB_RE |
| 10422 | void |
| 10423 | Perl__invlist_dump(pTHX_ PerlIO *file, I32 level, |
| 10424 | const char * const indent, SV* const invlist) |
| 10425 | { |
| 10426 | /* Designed to be called only by do_sv_dump(). Dumps out the ranges of the |
| 10427 | * inversion list 'invlist' to 'file' at 'level' Each line is prefixed by |
| 10428 | * the string 'indent'. The output looks like this: |
| 10429 | [0] 0x000A .. 0x000D |
| 10430 | [2] 0x0085 |
| 10431 | [4] 0x2028 .. 0x2029 |
| 10432 | [6] 0x3104 .. INFTY |
| 10433 | * This means that the first range of code points matched by the list are |
| 10434 | * 0xA through 0xD; the second range contains only the single code point |
| 10435 | * 0x85, etc. An inversion list is an array of UVs. Two array elements |
| 10436 | * are used to define each range (except if the final range extends to |
| 10437 | * infinity, only a single element is needed). The array index of the |
| 10438 | * first element for the corresponding range is given in brackets. */ |
| 10439 | |
| 10440 | UV start, end; |
| 10441 | STRLEN count = 0; |
| 10442 | |
| 10443 | PERL_ARGS_ASSERT__INVLIST_DUMP; |
| 10444 | |
| 10445 | if (invlist_is_iterating(invlist)) { |
| 10446 | Perl_dump_indent(aTHX_ level, file, |
| 10447 | "%sCan't dump inversion list because is in middle of iterating\n", |
| 10448 | indent); |
| 10449 | return; |
| 10450 | } |
| 10451 | |
| 10452 | invlist_iterinit(invlist); |
| 10453 | while (invlist_iternext(invlist, &start, &end)) { |
| 10454 | if (end == UV_MAX) { |
| 10455 | Perl_dump_indent(aTHX_ level, file, |
| 10456 | "%s[%" UVuf "] 0x%04" UVXf " .. INFTY\n", |
| 10457 | indent, (UV)count, start); |
| 10458 | } |
| 10459 | else if (end != start) { |
| 10460 | Perl_dump_indent(aTHX_ level, file, |
| 10461 | "%s[%" UVuf "] 0x%04" UVXf " .. 0x%04" UVXf "\n", |
| 10462 | indent, (UV)count, start, end); |
| 10463 | } |
| 10464 | else { |
| 10465 | Perl_dump_indent(aTHX_ level, file, "%s[%" UVuf "] 0x%04" UVXf "\n", |
| 10466 | indent, (UV)count, start); |
| 10467 | } |
| 10468 | count += 2; |
| 10469 | } |
| 10470 | } |
| 10471 | |
| 10472 | #endif |
| 10473 | |
| 10474 | #if defined(PERL_ARGS_ASSERT__INVLISTEQ) && !defined(PERL_IN_XSUB_RE) |
| 10475 | bool |
| 10476 | Perl__invlistEQ(pTHX_ SV* const a, SV* const b, const bool complement_b) |
| 10477 | { |
| 10478 | /* Return a boolean as to if the two passed in inversion lists are |
| 10479 | * identical. The final argument, if TRUE, says to take the complement of |
| 10480 | * the second inversion list before doing the comparison */ |
| 10481 | |
| 10482 | const UV len_a = _invlist_len(a); |
| 10483 | UV len_b = _invlist_len(b); |
| 10484 | |
| 10485 | const UV* array_a = NULL; |
| 10486 | const UV* array_b = NULL; |
| 10487 | |
| 10488 | PERL_ARGS_ASSERT__INVLISTEQ; |
| 10489 | |
| 10490 | /* This code avoids accessing the arrays unless it knows the length is |
| 10491 | * non-zero */ |
| 10492 | |
| 10493 | if (len_a == 0) { |
| 10494 | if (len_b == 0) { |
| 10495 | return ! complement_b; |
| 10496 | } |
| 10497 | } |
| 10498 | else { |
| 10499 | array_a = invlist_array(a); |
| 10500 | } |
| 10501 | |
| 10502 | if (len_b != 0) { |
| 10503 | array_b = invlist_array(b); |
| 10504 | } |
| 10505 | |
| 10506 | /* If are to compare 'a' with the complement of b, set it |
| 10507 | * up so are looking at b's complement. */ |
| 10508 | if (complement_b) { |
| 10509 | |
| 10510 | /* The complement of nothing is everything, so <a> would have to have |
| 10511 | * just one element, starting at zero (ending at infinity) */ |
| 10512 | if (len_b == 0) { |
| 10513 | return (len_a == 1 && array_a[0] == 0); |
| 10514 | } |
| 10515 | if (array_b[0] == 0) { |
| 10516 | |
| 10517 | /* Otherwise, to complement, we invert. Here, the first element is |
| 10518 | * 0, just remove it. To do this, we just pretend the array starts |
| 10519 | * one later */ |
| 10520 | |
| 10521 | array_b++; |
| 10522 | len_b--; |
| 10523 | } |
| 10524 | else { |
| 10525 | |
| 10526 | /* But if the first element is not zero, we pretend the list starts |
| 10527 | * at the 0 that is always stored immediately before the array. */ |
| 10528 | array_b--; |
| 10529 | len_b++; |
| 10530 | } |
| 10531 | } |
| 10532 | |
| 10533 | return len_a == len_b |
| 10534 | && memEQ(array_a, array_b, len_a * sizeof(array_a[0])); |
| 10535 | |
| 10536 | } |
| 10537 | #endif |
| 10538 | |
| 10539 | /* |
| 10540 | * As best we can, determine the characters that can match the start of |
| 10541 | * the given EXACTF-ish node. This is for use in creating ssc nodes, so there |
| 10542 | * can be false positive matches |
| 10543 | * |
| 10544 | * Returns the invlist as a new SV*; it is the caller's responsibility to |
| 10545 | * call SvREFCNT_dec() when done with it. |
| 10546 | */ |
| 10547 | STATIC SV* |
| 10548 | S_make_exactf_invlist(pTHX_ RExC_state_t *pRExC_state, regnode *node) |
| 10549 | { |
| 10550 | dVAR; |
| 10551 | const U8 * s = (U8*)STRING(node); |
| 10552 | SSize_t bytelen = STR_LEN(node); |
| 10553 | UV uc; |
| 10554 | /* Start out big enough for 2 separate code points */ |
| 10555 | SV* invlist = _new_invlist(4); |
| 10556 | |
| 10557 | PERL_ARGS_ASSERT_MAKE_EXACTF_INVLIST; |
| 10558 | |
| 10559 | if (! UTF) { |
| 10560 | uc = *s; |
| 10561 | |
| 10562 | /* We punt and assume can match anything if the node begins |
| 10563 | * with a multi-character fold. Things are complicated. For |
| 10564 | * example, /ffi/i could match any of: |
| 10565 | * "\N{LATIN SMALL LIGATURE FFI}" |
| 10566 | * "\N{LATIN SMALL LIGATURE FF}I" |
| 10567 | * "F\N{LATIN SMALL LIGATURE FI}" |
| 10568 | * plus several other things; and making sure we have all the |
| 10569 | * possibilities is hard. */ |
| 10570 | if (is_MULTI_CHAR_FOLD_latin1_safe(s, s + bytelen)) { |
| 10571 | invlist = _add_range_to_invlist(invlist, 0, UV_MAX); |
| 10572 | } |
| 10573 | else { |
| 10574 | /* Any Latin1 range character can potentially match any |
| 10575 | * other depending on the locale, and in Turkic locales, U+130 and |
| 10576 | * U+131 */ |
| 10577 | if (OP(node) == EXACTFL) { |
| 10578 | _invlist_union(invlist, PL_Latin1, &invlist); |
| 10579 | invlist = add_cp_to_invlist(invlist, |
| 10580 | LATIN_SMALL_LETTER_DOTLESS_I); |
| 10581 | invlist = add_cp_to_invlist(invlist, |
| 10582 | LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE); |
| 10583 | } |
| 10584 | else { |
| 10585 | /* But otherwise, it matches at least itself. We can |
| 10586 | * quickly tell if it has a distinct fold, and if so, |
| 10587 | * it matches that as well */ |
| 10588 | invlist = add_cp_to_invlist(invlist, uc); |
| 10589 | if (IS_IN_SOME_FOLD_L1(uc)) |
| 10590 | invlist = add_cp_to_invlist(invlist, PL_fold_latin1[uc]); |
| 10591 | } |
| 10592 | |
| 10593 | /* Some characters match above-Latin1 ones under /i. This |
| 10594 | * is true of EXACTFL ones when the locale is UTF-8 */ |
| 10595 | if (HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(uc) |
| 10596 | && (! isASCII(uc) || (OP(node) != EXACTFAA |
| 10597 | && OP(node) != EXACTFAA_NO_TRIE))) |
| 10598 | { |
| 10599 | add_above_Latin1_folds(pRExC_state, (U8) uc, &invlist); |
| 10600 | } |
| 10601 | } |
| 10602 | } |
| 10603 | else { /* Pattern is UTF-8 */ |
| 10604 | U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' }; |
| 10605 | const U8* e = s + bytelen; |
| 10606 | IV fc; |
| 10607 | |
| 10608 | fc = uc = utf8_to_uvchr_buf(s, s + bytelen, NULL); |
| 10609 | |
| 10610 | /* The only code points that aren't folded in a UTF EXACTFish |
| 10611 | * node are are the problematic ones in EXACTFL nodes */ |
| 10612 | if (OP(node) == EXACTFL && is_PROBLEMATIC_LOCALE_FOLDEDS_START_cp(uc)) { |
| 10613 | /* We need to check for the possibility that this EXACTFL |
| 10614 | * node begins with a multi-char fold. Therefore we fold |
| 10615 | * the first few characters of it so that we can make that |
| 10616 | * check */ |
| 10617 | U8 *d = folded; |
| 10618 | int i; |
| 10619 | |
| 10620 | fc = -1; |
| 10621 | for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < e; i++) { |
| 10622 | if (isASCII(*s)) { |
| 10623 | *(d++) = (U8) toFOLD(*s); |
| 10624 | if (fc < 0) { /* Save the first fold */ |
| 10625 | fc = *(d-1); |
| 10626 | } |
| 10627 | s++; |
| 10628 | } |
| 10629 | else { |
| 10630 | STRLEN len; |
| 10631 | UV fold = toFOLD_utf8_safe(s, e, d, &len); |
| 10632 | if (fc < 0) { /* Save the first fold */ |
| 10633 | fc = fold; |
| 10634 | } |
| 10635 | d += len; |
| 10636 | s += UTF8SKIP(s); |
| 10637 | } |
| 10638 | } |
| 10639 | |
| 10640 | /* And set up so the code below that looks in this folded |
| 10641 | * buffer instead of the node's string */ |
| 10642 | e = d; |
| 10643 | s = folded; |
| 10644 | } |
| 10645 | |
| 10646 | /* When we reach here 's' points to the fold of the first |
| 10647 | * character(s) of the node; and 'e' points to far enough along |
| 10648 | * the folded string to be just past any possible multi-char |
| 10649 | * fold. |
| 10650 | * |
| 10651 | * Unlike the non-UTF-8 case, the macro for determining if a |
| 10652 | * string is a multi-char fold requires all the characters to |
| 10653 | * already be folded. This is because of all the complications |
| 10654 | * if not. Note that they are folded anyway, except in EXACTFL |
| 10655 | * nodes. Like the non-UTF case above, we punt if the node |
| 10656 | * begins with a multi-char fold */ |
| 10657 | |
| 10658 | if (is_MULTI_CHAR_FOLD_utf8_safe(s, e)) { |
| 10659 | invlist = _add_range_to_invlist(invlist, 0, UV_MAX); |
| 10660 | } |
| 10661 | else { /* Single char fold */ |
| 10662 | unsigned int k; |
| 10663 | U32 first_fold; |
| 10664 | const U32 * remaining_folds; |
| 10665 | Size_t folds_count; |
| 10666 | |
| 10667 | /* It matches itself */ |
| 10668 | invlist = add_cp_to_invlist(invlist, fc); |
| 10669 | |
| 10670 | /* ... plus all the things that fold to it, which are found in |
| 10671 | * PL_utf8_foldclosures */ |
| 10672 | folds_count = _inverse_folds(fc, &first_fold, |
| 10673 | &remaining_folds); |
| 10674 | for (k = 0; k < folds_count; k++) { |
| 10675 | UV c = (k == 0) ? first_fold : remaining_folds[k-1]; |
| 10676 | |
| 10677 | /* /aa doesn't allow folds between ASCII and non- */ |
| 10678 | if ( (OP(node) == EXACTFAA || OP(node) == EXACTFAA_NO_TRIE) |
| 10679 | && isASCII(c) != isASCII(fc)) |
| 10680 | { |
| 10681 | continue; |
| 10682 | } |
| 10683 | |
| 10684 | invlist = add_cp_to_invlist(invlist, c); |
| 10685 | } |
| 10686 | |
| 10687 | if (OP(node) == EXACTFL) { |
| 10688 | |
| 10689 | /* If either [iI] are present in an EXACTFL node the above code |
| 10690 | * should have added its normal case pair, but under a Turkish |
| 10691 | * locale they could match instead the case pairs from it. Add |
| 10692 | * those as potential matches as well */ |
| 10693 | if (isALPHA_FOLD_EQ(fc, 'I')) { |
| 10694 | invlist = add_cp_to_invlist(invlist, |
| 10695 | LATIN_SMALL_LETTER_DOTLESS_I); |
| 10696 | invlist = add_cp_to_invlist(invlist, |
| 10697 | LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE); |
| 10698 | } |
| 10699 | else if (fc == LATIN_SMALL_LETTER_DOTLESS_I) { |
| 10700 | invlist = add_cp_to_invlist(invlist, 'I'); |
| 10701 | } |
| 10702 | else if (fc == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) { |
| 10703 | invlist = add_cp_to_invlist(invlist, 'i'); |
| 10704 | } |
| 10705 | } |
| 10706 | } |
| 10707 | } |
| 10708 | |
| 10709 | return invlist; |
| 10710 | } |
| 10711 | |
| 10712 | #undef HEADER_LENGTH |
| 10713 | #undef TO_INTERNAL_SIZE |
| 10714 | #undef FROM_INTERNAL_SIZE |
| 10715 | #undef INVLIST_VERSION_ID |
| 10716 | |
| 10717 | /* End of inversion list object */ |
| 10718 | |
| 10719 | STATIC void |
| 10720 | S_parse_lparen_question_flags(pTHX_ RExC_state_t *pRExC_state) |
| 10721 | { |
| 10722 | /* This parses the flags that are in either the '(?foo)' or '(?foo:bar)' |
| 10723 | * constructs, and updates RExC_flags with them. On input, RExC_parse |
| 10724 | * should point to the first flag; it is updated on output to point to the |
| 10725 | * final ')' or ':'. There needs to be at least one flag, or this will |
| 10726 | * abort */ |
| 10727 | |
| 10728 | /* for (?g), (?gc), and (?o) warnings; warning |
| 10729 | about (?c) will warn about (?g) -- japhy */ |
| 10730 | |
| 10731 | #define WASTED_O 0x01 |
| 10732 | #define WASTED_G 0x02 |
| 10733 | #define WASTED_C 0x04 |
| 10734 | #define WASTED_GC (WASTED_G|WASTED_C) |
| 10735 | I32 wastedflags = 0x00; |
| 10736 | U32 posflags = 0, negflags = 0; |
| 10737 | U32 *flagsp = &posflags; |
| 10738 | char has_charset_modifier = '\0'; |
| 10739 | regex_charset cs; |
| 10740 | bool has_use_defaults = FALSE; |
| 10741 | const char* const seqstart = RExC_parse - 1; /* Point to the '?' */ |
| 10742 | int x_mod_count = 0; |
| 10743 | |
| 10744 | PERL_ARGS_ASSERT_PARSE_LPAREN_QUESTION_FLAGS; |
| 10745 | |
| 10746 | /* '^' as an initial flag sets certain defaults */ |
| 10747 | if (UCHARAT(RExC_parse) == '^') { |
| 10748 | RExC_parse++; |
| 10749 | has_use_defaults = TRUE; |
| 10750 | STD_PMMOD_FLAGS_CLEAR(&RExC_flags); |
| 10751 | cs = (RExC_uni_semantics) |
| 10752 | ? REGEX_UNICODE_CHARSET |
| 10753 | : REGEX_DEPENDS_CHARSET; |
| 10754 | set_regex_charset(&RExC_flags, cs); |
| 10755 | } |
| 10756 | else { |
| 10757 | cs = get_regex_charset(RExC_flags); |
| 10758 | if ( cs == REGEX_DEPENDS_CHARSET |
| 10759 | && RExC_uni_semantics) |
| 10760 | { |
| 10761 | cs = REGEX_UNICODE_CHARSET; |
| 10762 | } |
| 10763 | } |
| 10764 | |
| 10765 | while (RExC_parse < RExC_end) { |
| 10766 | /* && memCHRs("iogcmsx", *RExC_parse) */ |
| 10767 | /* (?g), (?gc) and (?o) are useless here |
| 10768 | and must be globally applied -- japhy */ |
| 10769 | if ((RExC_pm_flags & PMf_WILDCARD)) { |
| 10770 | if (flagsp == & negflags) { |
| 10771 | if (*RExC_parse == 'm') { |
| 10772 | RExC_parse++; |
| 10773 | /* diag_listed_as: Use of %s is not allowed in Unicode |
| 10774 | property wildcard subpatterns in regex; marked by <-- |
| 10775 | HERE in m/%s/ */ |
| 10776 | vFAIL("Use of modifier '-m' is not allowed in Unicode" |
| 10777 | " property wildcard subpatterns"); |
| 10778 | } |
| 10779 | } |
| 10780 | else { |
| 10781 | if (*RExC_parse == 's') { |
| 10782 | goto modifier_illegal_in_wildcard; |
| 10783 | } |
| 10784 | } |
| 10785 | } |
| 10786 | |
| 10787 | switch (*RExC_parse) { |
| 10788 | |
| 10789 | /* Code for the imsxn flags */ |
| 10790 | CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp, x_mod_count); |
| 10791 | |
| 10792 | case LOCALE_PAT_MOD: |
| 10793 | if (has_charset_modifier) { |
| 10794 | goto excess_modifier; |
| 10795 | } |
| 10796 | else if (flagsp == &negflags) { |
| 10797 | goto neg_modifier; |
| 10798 | } |
| 10799 | cs = REGEX_LOCALE_CHARSET; |
| 10800 | has_charset_modifier = LOCALE_PAT_MOD; |
| 10801 | break; |
| 10802 | case UNICODE_PAT_MOD: |
| 10803 | if (has_charset_modifier) { |
| 10804 | goto excess_modifier; |
| 10805 | } |
| 10806 | else if (flagsp == &negflags) { |
| 10807 | goto neg_modifier; |
| 10808 | } |
| 10809 | cs = REGEX_UNICODE_CHARSET; |
| 10810 | has_charset_modifier = UNICODE_PAT_MOD; |
| 10811 | break; |
| 10812 | case ASCII_RESTRICT_PAT_MOD: |
| 10813 | if (flagsp == &negflags) { |
| 10814 | goto neg_modifier; |
| 10815 | } |
| 10816 | if (has_charset_modifier) { |
| 10817 | if (cs != REGEX_ASCII_RESTRICTED_CHARSET) { |
| 10818 | goto excess_modifier; |
| 10819 | } |
| 10820 | /* Doubled modifier implies more restricted */ |
| 10821 | cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET; |
| 10822 | } |
| 10823 | else { |
| 10824 | cs = REGEX_ASCII_RESTRICTED_CHARSET; |
| 10825 | } |
| 10826 | has_charset_modifier = ASCII_RESTRICT_PAT_MOD; |
| 10827 | break; |
| 10828 | case DEPENDS_PAT_MOD: |
| 10829 | if (has_use_defaults) { |
| 10830 | goto fail_modifiers; |
| 10831 | } |
| 10832 | else if (flagsp == &negflags) { |
| 10833 | goto neg_modifier; |
| 10834 | } |
| 10835 | else if (has_charset_modifier) { |
| 10836 | goto excess_modifier; |
| 10837 | } |
| 10838 | |
| 10839 | /* The dual charset means unicode semantics if the |
| 10840 | * pattern (or target, not known until runtime) are |
| 10841 | * utf8, or something in the pattern indicates unicode |
| 10842 | * semantics */ |
| 10843 | cs = (RExC_uni_semantics) |
| 10844 | ? REGEX_UNICODE_CHARSET |
| 10845 | : REGEX_DEPENDS_CHARSET; |
| 10846 | has_charset_modifier = DEPENDS_PAT_MOD; |
| 10847 | break; |
| 10848 | excess_modifier: |
| 10849 | RExC_parse++; |
| 10850 | if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) { |
| 10851 | vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD); |
| 10852 | } |
| 10853 | else if (has_charset_modifier == *(RExC_parse - 1)) { |
| 10854 | vFAIL2("Regexp modifier \"%c\" may not appear twice", |
| 10855 | *(RExC_parse - 1)); |
| 10856 | } |
| 10857 | else { |
| 10858 | vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1)); |
| 10859 | } |
| 10860 | NOT_REACHED; /*NOTREACHED*/ |
| 10861 | neg_modifier: |
| 10862 | RExC_parse++; |
| 10863 | vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", |
| 10864 | *(RExC_parse - 1)); |
| 10865 | NOT_REACHED; /*NOTREACHED*/ |
| 10866 | case GLOBAL_PAT_MOD: /* 'g' */ |
| 10867 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 10868 | goto modifier_illegal_in_wildcard; |
| 10869 | } |
| 10870 | /*FALLTHROUGH*/ |
| 10871 | case ONCE_PAT_MOD: /* 'o' */ |
| 10872 | if (ckWARN(WARN_REGEXP)) { |
| 10873 | const I32 wflagbit = *RExC_parse == 'o' |
| 10874 | ? WASTED_O |
| 10875 | : WASTED_G; |
| 10876 | if (! (wastedflags & wflagbit) ) { |
| 10877 | wastedflags |= wflagbit; |
| 10878 | /* diag_listed_as: Useless (?-%s) - don't use /%s modifier in regex; marked by <-- HERE in m/%s/ */ |
| 10879 | vWARN5( |
| 10880 | RExC_parse + 1, |
| 10881 | "Useless (%s%c) - %suse /%c modifier", |
| 10882 | flagsp == &negflags ? "?-" : "?", |
| 10883 | *RExC_parse, |
| 10884 | flagsp == &negflags ? "don't " : "", |
| 10885 | *RExC_parse |
| 10886 | ); |
| 10887 | } |
| 10888 | } |
| 10889 | break; |
| 10890 | |
| 10891 | case CONTINUE_PAT_MOD: /* 'c' */ |
| 10892 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 10893 | goto modifier_illegal_in_wildcard; |
| 10894 | } |
| 10895 | if (ckWARN(WARN_REGEXP)) { |
| 10896 | if (! (wastedflags & WASTED_C) ) { |
| 10897 | wastedflags |= WASTED_GC; |
| 10898 | /* diag_listed_as: Useless (?-%s) - don't use /%s modifier in regex; marked by <-- HERE in m/%s/ */ |
| 10899 | vWARN3( |
| 10900 | RExC_parse + 1, |
| 10901 | "Useless (%sc) - %suse /gc modifier", |
| 10902 | flagsp == &negflags ? "?-" : "?", |
| 10903 | flagsp == &negflags ? "don't " : "" |
| 10904 | ); |
| 10905 | } |
| 10906 | } |
| 10907 | break; |
| 10908 | case KEEPCOPY_PAT_MOD: /* 'p' */ |
| 10909 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 10910 | goto modifier_illegal_in_wildcard; |
| 10911 | } |
| 10912 | if (flagsp == &negflags) { |
| 10913 | ckWARNreg(RExC_parse + 1,"Useless use of (?-p)"); |
| 10914 | } else { |
| 10915 | *flagsp |= RXf_PMf_KEEPCOPY; |
| 10916 | } |
| 10917 | break; |
| 10918 | case '-': |
| 10919 | /* A flag is a default iff it is following a minus, so |
| 10920 | * if there is a minus, it means will be trying to |
| 10921 | * re-specify a default which is an error */ |
| 10922 | if (has_use_defaults || flagsp == &negflags) { |
| 10923 | goto fail_modifiers; |
| 10924 | } |
| 10925 | flagsp = &negflags; |
| 10926 | wastedflags = 0; /* reset so (?g-c) warns twice */ |
| 10927 | x_mod_count = 0; |
| 10928 | break; |
| 10929 | case ':': |
| 10930 | case ')': |
| 10931 | |
| 10932 | if ( (RExC_pm_flags & PMf_WILDCARD) |
| 10933 | && cs != REGEX_ASCII_MORE_RESTRICTED_CHARSET) |
| 10934 | { |
| 10935 | RExC_parse++; |
| 10936 | /* diag_listed_as: Use of %s is not allowed in Unicode |
| 10937 | property wildcard subpatterns in regex; marked by <-- |
| 10938 | HERE in m/%s/ */ |
| 10939 | vFAIL2("Use of modifier '%c' is not allowed in Unicode" |
| 10940 | " property wildcard subpatterns", |
| 10941 | has_charset_modifier); |
| 10942 | } |
| 10943 | |
| 10944 | if ((posflags & (RXf_PMf_EXTENDED|RXf_PMf_EXTENDED_MORE)) == RXf_PMf_EXTENDED) { |
| 10945 | negflags |= RXf_PMf_EXTENDED_MORE; |
| 10946 | } |
| 10947 | RExC_flags |= posflags; |
| 10948 | |
| 10949 | if (negflags & RXf_PMf_EXTENDED) { |
| 10950 | negflags |= RXf_PMf_EXTENDED_MORE; |
| 10951 | } |
| 10952 | RExC_flags &= ~negflags; |
| 10953 | set_regex_charset(&RExC_flags, cs); |
| 10954 | |
| 10955 | return; |
| 10956 | default: |
| 10957 | fail_modifiers: |
| 10958 | RExC_parse += SKIP_IF_CHAR(RExC_parse, RExC_end); |
| 10959 | /* diag_listed_as: Sequence (?%s...) not recognized in regex; marked by <-- HERE in m/%s/ */ |
| 10960 | vFAIL2utf8f("Sequence (%" UTF8f "...) not recognized", |
| 10961 | UTF8fARG(UTF, RExC_parse-seqstart, seqstart)); |
| 10962 | NOT_REACHED; /*NOTREACHED*/ |
| 10963 | } |
| 10964 | |
| 10965 | RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1; |
| 10966 | } |
| 10967 | |
| 10968 | vFAIL("Sequence (?... not terminated"); |
| 10969 | |
| 10970 | modifier_illegal_in_wildcard: |
| 10971 | RExC_parse++; |
| 10972 | /* diag_listed_as: Use of %s is not allowed in Unicode property wildcard |
| 10973 | subpatterns in regex; marked by <-- HERE in m/%s/ */ |
| 10974 | vFAIL2("Use of modifier '%c' is not allowed in Unicode property wildcard" |
| 10975 | " subpatterns", *(RExC_parse - 1)); |
| 10976 | } |
| 10977 | |
| 10978 | /* |
| 10979 | - reg - regular expression, i.e. main body or parenthesized thing |
| 10980 | * |
| 10981 | * Caller must absorb opening parenthesis. |
| 10982 | * |
| 10983 | * Combining parenthesis handling with the base level of regular expression |
| 10984 | * is a trifle forced, but the need to tie the tails of the branches to what |
| 10985 | * follows makes it hard to avoid. |
| 10986 | */ |
| 10987 | #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1) |
| 10988 | #ifdef DEBUGGING |
| 10989 | #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1) |
| 10990 | #else |
| 10991 | #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1) |
| 10992 | #endif |
| 10993 | |
| 10994 | PERL_STATIC_INLINE regnode_offset |
| 10995 | S_handle_named_backref(pTHX_ RExC_state_t *pRExC_state, |
| 10996 | I32 *flagp, |
| 10997 | char * parse_start, |
| 10998 | char ch |
| 10999 | ) |
| 11000 | { |
| 11001 | regnode_offset ret; |
| 11002 | char* name_start = RExC_parse; |
| 11003 | U32 num = 0; |
| 11004 | SV *sv_dat = reg_scan_name(pRExC_state, REG_RSN_RETURN_DATA); |
| 11005 | GET_RE_DEBUG_FLAGS_DECL; |
| 11006 | |
| 11007 | PERL_ARGS_ASSERT_HANDLE_NAMED_BACKREF; |
| 11008 | |
| 11009 | if (RExC_parse == name_start || *RExC_parse != ch) { |
| 11010 | /* diag_listed_as: Sequence \%s... not terminated in regex; marked by <-- HERE in m/%s/ */ |
| 11011 | vFAIL2("Sequence %.3s... not terminated", parse_start); |
| 11012 | } |
| 11013 | |
| 11014 | if (sv_dat) { |
| 11015 | num = add_data( pRExC_state, STR_WITH_LEN("S")); |
| 11016 | RExC_rxi->data->data[num]=(void*)sv_dat; |
| 11017 | SvREFCNT_inc_simple_void_NN(sv_dat); |
| 11018 | } |
| 11019 | RExC_sawback = 1; |
| 11020 | ret = reganode(pRExC_state, |
| 11021 | ((! FOLD) |
| 11022 | ? REFN |
| 11023 | : (ASCII_FOLD_RESTRICTED) |
| 11024 | ? REFFAN |
| 11025 | : (AT_LEAST_UNI_SEMANTICS) |
| 11026 | ? REFFUN |
| 11027 | : (LOC) |
| 11028 | ? REFFLN |
| 11029 | : REFFN), |
| 11030 | num); |
| 11031 | *flagp |= HASWIDTH; |
| 11032 | |
| 11033 | Set_Node_Offset(REGNODE_p(ret), parse_start+1); |
| 11034 | Set_Node_Cur_Length(REGNODE_p(ret), parse_start); |
| 11035 | |
| 11036 | nextchar(pRExC_state); |
| 11037 | return ret; |
| 11038 | } |
| 11039 | |
| 11040 | /* On success, returns the offset at which any next node should be placed into |
| 11041 | * the regex engine program being compiled. |
| 11042 | * |
| 11043 | * Returns 0 otherwise, with *flagp set to indicate why: |
| 11044 | * TRYAGAIN at the end of (?) that only sets flags. |
| 11045 | * RESTART_PARSE if the parse needs to be restarted, or'd with |
| 11046 | * NEED_UTF8 if the pattern needs to be upgraded to UTF-8. |
| 11047 | * Otherwise would only return 0 if regbranch() returns 0, which cannot |
| 11048 | * happen. */ |
| 11049 | STATIC regnode_offset |
| 11050 | S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp, U32 depth) |
| 11051 | /* paren: Parenthesized? 0=top; 1,2=inside '(': changed to letter. |
| 11052 | * 2 is like 1, but indicates that nextchar() has been called to advance |
| 11053 | * RExC_parse beyond the '('. Things like '(?' are indivisible tokens, and |
| 11054 | * this flag alerts us to the need to check for that */ |
| 11055 | { |
| 11056 | regnode_offset ret = 0; /* Will be the head of the group. */ |
| 11057 | regnode_offset br; |
| 11058 | regnode_offset lastbr; |
| 11059 | regnode_offset ender = 0; |
| 11060 | I32 parno = 0; |
| 11061 | I32 flags; |
| 11062 | U32 oregflags = RExC_flags; |
| 11063 | bool have_branch = 0; |
| 11064 | bool is_open = 0; |
| 11065 | I32 freeze_paren = 0; |
| 11066 | I32 after_freeze = 0; |
| 11067 | I32 num; /* numeric backreferences */ |
| 11068 | SV * max_open; /* Max number of unclosed parens */ |
| 11069 | |
| 11070 | char * parse_start = RExC_parse; /* MJD */ |
| 11071 | char * const oregcomp_parse = RExC_parse; |
| 11072 | |
| 11073 | GET_RE_DEBUG_FLAGS_DECL; |
| 11074 | |
| 11075 | PERL_ARGS_ASSERT_REG; |
| 11076 | DEBUG_PARSE("reg "); |
| 11077 | |
| 11078 | max_open = get_sv(RE_COMPILE_RECURSION_LIMIT, GV_ADD); |
| 11079 | assert(max_open); |
| 11080 | if (!SvIOK(max_open)) { |
| 11081 | sv_setiv(max_open, RE_COMPILE_RECURSION_INIT); |
| 11082 | } |
| 11083 | if (depth > 4 * (UV) SvIV(max_open)) { /* We increase depth by 4 for each |
| 11084 | open paren */ |
| 11085 | vFAIL("Too many nested open parens"); |
| 11086 | } |
| 11087 | |
| 11088 | *flagp = 0; /* Tentatively. */ |
| 11089 | |
| 11090 | if (RExC_in_lookbehind) { |
| 11091 | RExC_in_lookbehind++; |
| 11092 | } |
| 11093 | if (RExC_in_lookahead) { |
| 11094 | RExC_in_lookahead++; |
| 11095 | } |
| 11096 | |
| 11097 | /* Having this true makes it feasible to have a lot fewer tests for the |
| 11098 | * parse pointer being in scope. For example, we can write |
| 11099 | * while(isFOO(*RExC_parse)) RExC_parse++; |
| 11100 | * instead of |
| 11101 | * while(RExC_parse < RExC_end && isFOO(*RExC_parse)) RExC_parse++; |
| 11102 | */ |
| 11103 | assert(*RExC_end == '\0'); |
| 11104 | |
| 11105 | /* Make an OPEN node, if parenthesized. */ |
| 11106 | if (paren) { |
| 11107 | |
| 11108 | /* Under /x, space and comments can be gobbled up between the '(' and |
| 11109 | * here (if paren ==2). The forms '(*VERB' and '(?...' disallow such |
| 11110 | * intervening space, as the sequence is a token, and a token should be |
| 11111 | * indivisible */ |
| 11112 | bool has_intervening_patws = (paren == 2) |
| 11113 | && *(RExC_parse - 1) != '('; |
| 11114 | |
| 11115 | if (RExC_parse >= RExC_end) { |
| 11116 | vFAIL("Unmatched ("); |
| 11117 | } |
| 11118 | |
| 11119 | if (paren == 'r') { /* Atomic script run */ |
| 11120 | paren = '>'; |
| 11121 | goto parse_rest; |
| 11122 | } |
| 11123 | else if ( *RExC_parse == '*') { /* (*VERB:ARG), (*construct:...) */ |
| 11124 | char *start_verb = RExC_parse + 1; |
| 11125 | STRLEN verb_len; |
| 11126 | char *start_arg = NULL; |
| 11127 | unsigned char op = 0; |
| 11128 | int arg_required = 0; |
| 11129 | int internal_argval = -1; /* if >-1 we are not allowed an argument*/ |
| 11130 | bool has_upper = FALSE; |
| 11131 | |
| 11132 | if (has_intervening_patws) { |
| 11133 | RExC_parse++; /* past the '*' */ |
| 11134 | |
| 11135 | /* For strict backwards compatibility, don't change the message |
| 11136 | * now that we also have lowercase operands */ |
| 11137 | if (isUPPER(*RExC_parse)) { |
| 11138 | vFAIL("In '(*VERB...)', the '(' and '*' must be adjacent"); |
| 11139 | } |
| 11140 | else { |
| 11141 | vFAIL("In '(*...)', the '(' and '*' must be adjacent"); |
| 11142 | } |
| 11143 | } |
| 11144 | while (RExC_parse < RExC_end && *RExC_parse != ')' ) { |
| 11145 | if ( *RExC_parse == ':' ) { |
| 11146 | start_arg = RExC_parse + 1; |
| 11147 | break; |
| 11148 | } |
| 11149 | else if (! UTF) { |
| 11150 | if (isUPPER(*RExC_parse)) { |
| 11151 | has_upper = TRUE; |
| 11152 | } |
| 11153 | RExC_parse++; |
| 11154 | } |
| 11155 | else { |
| 11156 | RExC_parse += UTF8SKIP(RExC_parse); |
| 11157 | } |
| 11158 | } |
| 11159 | verb_len = RExC_parse - start_verb; |
| 11160 | if ( start_arg ) { |
| 11161 | if (RExC_parse >= RExC_end) { |
| 11162 | goto unterminated_verb_pattern; |
| 11163 | } |
| 11164 | |
| 11165 | RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1; |
| 11166 | while ( RExC_parse < RExC_end && *RExC_parse != ')' ) { |
| 11167 | RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1; |
| 11168 | } |
| 11169 | if ( RExC_parse >= RExC_end || *RExC_parse != ')' ) { |
| 11170 | unterminated_verb_pattern: |
| 11171 | if (has_upper) { |
| 11172 | vFAIL("Unterminated verb pattern argument"); |
| 11173 | } |
| 11174 | else { |
| 11175 | vFAIL("Unterminated '(*...' argument"); |
| 11176 | } |
| 11177 | } |
| 11178 | } else { |
| 11179 | if ( RExC_parse >= RExC_end || *RExC_parse != ')' ) { |
| 11180 | if (has_upper) { |
| 11181 | vFAIL("Unterminated verb pattern"); |
| 11182 | } |
| 11183 | else { |
| 11184 | vFAIL("Unterminated '(*...' construct"); |
| 11185 | } |
| 11186 | } |
| 11187 | } |
| 11188 | |
| 11189 | /* Here, we know that RExC_parse < RExC_end */ |
| 11190 | |
| 11191 | switch ( *start_verb ) { |
| 11192 | case 'A': /* (*ACCEPT) */ |
| 11193 | if ( memEQs(start_verb, verb_len,"ACCEPT") ) { |
| 11194 | op = ACCEPT; |
| 11195 | internal_argval = RExC_nestroot; |
| 11196 | } |
| 11197 | break; |
| 11198 | case 'C': /* (*COMMIT) */ |
| 11199 | if ( memEQs(start_verb, verb_len,"COMMIT") ) |
| 11200 | op = COMMIT; |
| 11201 | break; |
| 11202 | case 'F': /* (*FAIL) */ |
| 11203 | if ( verb_len==1 || memEQs(start_verb, verb_len,"FAIL") ) { |
| 11204 | op = OPFAIL; |
| 11205 | } |
| 11206 | break; |
| 11207 | case ':': /* (*:NAME) */ |
| 11208 | case 'M': /* (*MARK:NAME) */ |
| 11209 | if ( verb_len==0 || memEQs(start_verb, verb_len,"MARK") ) { |
| 11210 | op = MARKPOINT; |
| 11211 | arg_required = 1; |
| 11212 | } |
| 11213 | break; |
| 11214 | case 'P': /* (*PRUNE) */ |
| 11215 | if ( memEQs(start_verb, verb_len,"PRUNE") ) |
| 11216 | op = PRUNE; |
| 11217 | break; |
| 11218 | case 'S': /* (*SKIP) */ |
| 11219 | if ( memEQs(start_verb, verb_len,"SKIP") ) |
| 11220 | op = SKIP; |
| 11221 | break; |
| 11222 | case 'T': /* (*THEN) */ |
| 11223 | /* [19:06] <TimToady> :: is then */ |
| 11224 | if ( memEQs(start_verb, verb_len,"THEN") ) { |
| 11225 | op = CUTGROUP; |
| 11226 | RExC_seen |= REG_CUTGROUP_SEEN; |
| 11227 | } |
| 11228 | break; |
| 11229 | case 'a': |
| 11230 | if ( memEQs(start_verb, verb_len, "asr") |
| 11231 | || memEQs(start_verb, verb_len, "atomic_script_run")) |
| 11232 | { |
| 11233 | paren = 'r'; /* Mnemonic: recursed run */ |
| 11234 | goto script_run; |
| 11235 | } |
| 11236 | else if (memEQs(start_verb, verb_len, "atomic")) { |
| 11237 | paren = 't'; /* AtOMIC */ |
| 11238 | goto alpha_assertions; |
| 11239 | } |
| 11240 | break; |
| 11241 | case 'p': |
| 11242 | if ( memEQs(start_verb, verb_len, "plb") |
| 11243 | || memEQs(start_verb, verb_len, "positive_lookbehind")) |
| 11244 | { |
| 11245 | paren = 'b'; |
| 11246 | goto lookbehind_alpha_assertions; |
| 11247 | } |
| 11248 | else if ( memEQs(start_verb, verb_len, "pla") |
| 11249 | || memEQs(start_verb, verb_len, "positive_lookahead")) |
| 11250 | { |
| 11251 | paren = 'a'; |
| 11252 | goto alpha_assertions; |
| 11253 | } |
| 11254 | break; |
| 11255 | case 'n': |
| 11256 | if ( memEQs(start_verb, verb_len, "nlb") |
| 11257 | || memEQs(start_verb, verb_len, "negative_lookbehind")) |
| 11258 | { |
| 11259 | paren = 'B'; |
| 11260 | goto lookbehind_alpha_assertions; |
| 11261 | } |
| 11262 | else if ( memEQs(start_verb, verb_len, "nla") |
| 11263 | || memEQs(start_verb, verb_len, "negative_lookahead")) |
| 11264 | { |
| 11265 | paren = 'A'; |
| 11266 | goto alpha_assertions; |
| 11267 | } |
| 11268 | break; |
| 11269 | case 's': |
| 11270 | if ( memEQs(start_verb, verb_len, "sr") |
| 11271 | || memEQs(start_verb, verb_len, "script_run")) |
| 11272 | { |
| 11273 | regnode_offset atomic; |
| 11274 | |
| 11275 | paren = 's'; |
| 11276 | |
| 11277 | script_run: |
| 11278 | |
| 11279 | /* This indicates Unicode rules. */ |
| 11280 | REQUIRE_UNI_RULES(flagp, 0); |
| 11281 | |
| 11282 | if (! start_arg) { |
| 11283 | goto no_colon; |
| 11284 | } |
| 11285 | |
| 11286 | RExC_parse = start_arg; |
| 11287 | |
| 11288 | if (RExC_in_script_run) { |
| 11289 | |
| 11290 | /* Nested script runs are treated as no-ops, because |
| 11291 | * if the nested one fails, the outer one must as |
| 11292 | * well. It could fail sooner, and avoid (??{} with |
| 11293 | * side effects, but that is explicitly documented as |
| 11294 | * undefined behavior. */ |
| 11295 | |
| 11296 | ret = 0; |
| 11297 | |
| 11298 | if (paren == 's') { |
| 11299 | paren = ':'; |
| 11300 | goto parse_rest; |
| 11301 | } |
| 11302 | |
| 11303 | /* But, the atomic part of a nested atomic script run |
| 11304 | * isn't a no-op, but can be treated just like a '(?>' |
| 11305 | * */ |
| 11306 | paren = '>'; |
| 11307 | goto parse_rest; |
| 11308 | } |
| 11309 | |
| 11310 | if (paren == 's') { |
| 11311 | /* Here, we're starting a new regular script run */ |
| 11312 | ret = reg_node(pRExC_state, SROPEN); |
| 11313 | RExC_in_script_run = 1; |
| 11314 | is_open = 1; |
| 11315 | goto parse_rest; |
| 11316 | } |
| 11317 | |
| 11318 | /* Here, we are starting an atomic script run. This is |
| 11319 | * handled by recursing to deal with the atomic portion |
| 11320 | * separately, enclosed in SROPEN ... SRCLOSE nodes */ |
| 11321 | |
| 11322 | ret = reg_node(pRExC_state, SROPEN); |
| 11323 | |
| 11324 | RExC_in_script_run = 1; |
| 11325 | |
| 11326 | atomic = reg(pRExC_state, 'r', &flags, depth); |
| 11327 | if (flags & (RESTART_PARSE|NEED_UTF8)) { |
| 11328 | *flagp = flags & (RESTART_PARSE|NEED_UTF8); |
| 11329 | return 0; |
| 11330 | } |
| 11331 | |
| 11332 | if (! REGTAIL(pRExC_state, ret, atomic)) { |
| 11333 | REQUIRE_BRANCHJ(flagp, 0); |
| 11334 | } |
| 11335 | |
| 11336 | if (! REGTAIL(pRExC_state, atomic, reg_node(pRExC_state, |
| 11337 | SRCLOSE))) |
| 11338 | { |
| 11339 | REQUIRE_BRANCHJ(flagp, 0); |
| 11340 | } |
| 11341 | |
| 11342 | RExC_in_script_run = 0; |
| 11343 | return ret; |
| 11344 | } |
| 11345 | |
| 11346 | break; |
| 11347 | |
| 11348 | lookbehind_alpha_assertions: |
| 11349 | RExC_seen |= REG_LOOKBEHIND_SEEN; |
| 11350 | RExC_in_lookbehind++; |
| 11351 | /*FALLTHROUGH*/ |
| 11352 | |
| 11353 | alpha_assertions: |
| 11354 | |
| 11355 | RExC_seen_zerolen++; |
| 11356 | |
| 11357 | if (! start_arg) { |
| 11358 | goto no_colon; |
| 11359 | } |
| 11360 | |
| 11361 | /* An empty negative lookahead assertion simply is failure */ |
| 11362 | if (paren == 'A' && RExC_parse == start_arg) { |
| 11363 | ret=reganode(pRExC_state, OPFAIL, 0); |
| 11364 | nextchar(pRExC_state); |
| 11365 | return ret; |
| 11366 | } |
| 11367 | |
| 11368 | RExC_parse = start_arg; |
| 11369 | goto parse_rest; |
| 11370 | |
| 11371 | no_colon: |
| 11372 | vFAIL2utf8f( |
| 11373 | "'(*%" UTF8f "' requires a terminating ':'", |
| 11374 | UTF8fARG(UTF, verb_len, start_verb)); |
| 11375 | NOT_REACHED; /*NOTREACHED*/ |
| 11376 | |
| 11377 | } /* End of switch */ |
| 11378 | if ( ! op ) { |
| 11379 | RExC_parse += UTF |
| 11380 | ? UTF8_SAFE_SKIP(RExC_parse, RExC_end) |
| 11381 | : 1; |
| 11382 | if (has_upper || verb_len == 0) { |
| 11383 | vFAIL2utf8f( |
| 11384 | "Unknown verb pattern '%" UTF8f "'", |
| 11385 | UTF8fARG(UTF, verb_len, start_verb)); |
| 11386 | } |
| 11387 | else { |
| 11388 | vFAIL2utf8f( |
| 11389 | "Unknown '(*...)' construct '%" UTF8f "'", |
| 11390 | UTF8fARG(UTF, verb_len, start_verb)); |
| 11391 | } |
| 11392 | } |
| 11393 | if ( RExC_parse == start_arg ) { |
| 11394 | start_arg = NULL; |
| 11395 | } |
| 11396 | if ( arg_required && !start_arg ) { |
| 11397 | vFAIL3("Verb pattern '%.*s' has a mandatory argument", |
| 11398 | (int) verb_len, start_verb); |
| 11399 | } |
| 11400 | if (internal_argval == -1) { |
| 11401 | ret = reganode(pRExC_state, op, 0); |
| 11402 | } else { |
| 11403 | ret = reg2Lanode(pRExC_state, op, 0, internal_argval); |
| 11404 | } |
| 11405 | RExC_seen |= REG_VERBARG_SEEN; |
| 11406 | if (start_arg) { |
| 11407 | SV *sv = newSVpvn( start_arg, |
| 11408 | RExC_parse - start_arg); |
| 11409 | ARG(REGNODE_p(ret)) = add_data( pRExC_state, |
| 11410 | STR_WITH_LEN("S")); |
| 11411 | RExC_rxi->data->data[ARG(REGNODE_p(ret))]=(void*)sv; |
| 11412 | FLAGS(REGNODE_p(ret)) = 1; |
| 11413 | } else { |
| 11414 | FLAGS(REGNODE_p(ret)) = 0; |
| 11415 | } |
| 11416 | if ( internal_argval != -1 ) |
| 11417 | ARG2L_SET(REGNODE_p(ret), internal_argval); |
| 11418 | nextchar(pRExC_state); |
| 11419 | return ret; |
| 11420 | } |
| 11421 | else if (*RExC_parse == '?') { /* (?...) */ |
| 11422 | bool is_logical = 0; |
| 11423 | const char * const seqstart = RExC_parse; |
| 11424 | const char * endptr; |
| 11425 | if (has_intervening_patws) { |
| 11426 | RExC_parse++; |
| 11427 | vFAIL("In '(?...)', the '(' and '?' must be adjacent"); |
| 11428 | } |
| 11429 | |
| 11430 | RExC_parse++; /* past the '?' */ |
| 11431 | paren = *RExC_parse; /* might be a trailing NUL, if not |
| 11432 | well-formed */ |
| 11433 | RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1; |
| 11434 | if (RExC_parse > RExC_end) { |
| 11435 | paren = '\0'; |
| 11436 | } |
| 11437 | ret = 0; /* For look-ahead/behind. */ |
| 11438 | switch (paren) { |
| 11439 | |
| 11440 | case 'P': /* (?P...) variants for those used to PCRE/Python */ |
| 11441 | paren = *RExC_parse; |
| 11442 | if ( paren == '<') { /* (?P<...>) named capture */ |
| 11443 | RExC_parse++; |
| 11444 | if (RExC_parse >= RExC_end) { |
| 11445 | vFAIL("Sequence (?P<... not terminated"); |
| 11446 | } |
| 11447 | goto named_capture; |
| 11448 | } |
| 11449 | else if (paren == '>') { /* (?P>name) named recursion */ |
| 11450 | RExC_parse++; |
| 11451 | if (RExC_parse >= RExC_end) { |
| 11452 | vFAIL("Sequence (?P>... not terminated"); |
| 11453 | } |
| 11454 | goto named_recursion; |
| 11455 | } |
| 11456 | else if (paren == '=') { /* (?P=...) named backref */ |
| 11457 | RExC_parse++; |
| 11458 | return handle_named_backref(pRExC_state, flagp, |
| 11459 | parse_start, ')'); |
| 11460 | } |
| 11461 | RExC_parse += SKIP_IF_CHAR(RExC_parse, RExC_end); |
| 11462 | /* diag_listed_as: Sequence (?%s...) not recognized in regex; marked by <-- HERE in m/%s/ */ |
| 11463 | vFAIL3("Sequence (%.*s...) not recognized", |
| 11464 | (int) (RExC_parse - seqstart), seqstart); |
| 11465 | NOT_REACHED; /*NOTREACHED*/ |
| 11466 | case '<': /* (?<...) */ |
| 11467 | /* If you want to support (?<*...), first reconcile with GH #17363 */ |
| 11468 | if (*RExC_parse == '!') |
| 11469 | paren = ','; |
| 11470 | else if (*RExC_parse != '=') |
| 11471 | named_capture: |
| 11472 | { /* (?<...>) */ |
| 11473 | char *name_start; |
| 11474 | SV *svname; |
| 11475 | paren= '>'; |
| 11476 | /* FALLTHROUGH */ |
| 11477 | case '\'': /* (?'...') */ |
| 11478 | name_start = RExC_parse; |
| 11479 | svname = reg_scan_name(pRExC_state, REG_RSN_RETURN_NAME); |
| 11480 | if ( RExC_parse == name_start |
| 11481 | || RExC_parse >= RExC_end |
| 11482 | || *RExC_parse != paren) |
| 11483 | { |
| 11484 | vFAIL2("Sequence (?%c... not terminated", |
| 11485 | paren=='>' ? '<' : (char) paren); |
| 11486 | } |
| 11487 | { |
| 11488 | HE *he_str; |
| 11489 | SV *sv_dat = NULL; |
| 11490 | if (!svname) /* shouldn't happen */ |
| 11491 | Perl_croak(aTHX_ |
| 11492 | "panic: reg_scan_name returned NULL"); |
| 11493 | if (!RExC_paren_names) { |
| 11494 | RExC_paren_names= newHV(); |
| 11495 | sv_2mortal(MUTABLE_SV(RExC_paren_names)); |
| 11496 | #ifdef DEBUGGING |
| 11497 | RExC_paren_name_list= newAV(); |
| 11498 | sv_2mortal(MUTABLE_SV(RExC_paren_name_list)); |
| 11499 | #endif |
| 11500 | } |
| 11501 | he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 ); |
| 11502 | if ( he_str ) |
| 11503 | sv_dat = HeVAL(he_str); |
| 11504 | if ( ! sv_dat ) { |
| 11505 | /* croak baby croak */ |
| 11506 | Perl_croak(aTHX_ |
| 11507 | "panic: paren_name hash element allocation failed"); |
| 11508 | } else if ( SvPOK(sv_dat) ) { |
| 11509 | /* (?|...) can mean we have dupes so scan to check |
| 11510 | its already been stored. Maybe a flag indicating |
| 11511 | we are inside such a construct would be useful, |
| 11512 | but the arrays are likely to be quite small, so |
| 11513 | for now we punt -- dmq */ |
| 11514 | IV count = SvIV(sv_dat); |
| 11515 | I32 *pv = (I32*)SvPVX(sv_dat); |
| 11516 | IV i; |
| 11517 | for ( i = 0 ; i < count ; i++ ) { |
| 11518 | if ( pv[i] == RExC_npar ) { |
| 11519 | count = 0; |
| 11520 | break; |
| 11521 | } |
| 11522 | } |
| 11523 | if ( count ) { |
| 11524 | pv = (I32*)SvGROW(sv_dat, |
| 11525 | SvCUR(sv_dat) + sizeof(I32)+1); |
| 11526 | SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32)); |
| 11527 | pv[count] = RExC_npar; |
| 11528 | SvIV_set(sv_dat, SvIVX(sv_dat) + 1); |
| 11529 | } |
| 11530 | } else { |
| 11531 | (void)SvUPGRADE(sv_dat, SVt_PVNV); |
| 11532 | sv_setpvn(sv_dat, (char *)&(RExC_npar), |
| 11533 | sizeof(I32)); |
| 11534 | SvIOK_on(sv_dat); |
| 11535 | SvIV_set(sv_dat, 1); |
| 11536 | } |
| 11537 | #ifdef DEBUGGING |
| 11538 | /* Yes this does cause a memory leak in debugging Perls |
| 11539 | * */ |
| 11540 | if (!av_store(RExC_paren_name_list, |
| 11541 | RExC_npar, SvREFCNT_inc_NN(svname))) |
| 11542 | SvREFCNT_dec_NN(svname); |
| 11543 | #endif |
| 11544 | |
| 11545 | /*sv_dump(sv_dat);*/ |
| 11546 | } |
| 11547 | nextchar(pRExC_state); |
| 11548 | paren = 1; |
| 11549 | goto capturing_parens; |
| 11550 | } |
| 11551 | |
| 11552 | RExC_seen |= REG_LOOKBEHIND_SEEN; |
| 11553 | RExC_in_lookbehind++; |
| 11554 | RExC_parse++; |
| 11555 | if (RExC_parse >= RExC_end) { |
| 11556 | vFAIL("Sequence (?... not terminated"); |
| 11557 | } |
| 11558 | RExC_seen_zerolen++; |
| 11559 | break; |
| 11560 | case '=': /* (?=...) */ |
| 11561 | RExC_seen_zerolen++; |
| 11562 | RExC_in_lookahead++; |
| 11563 | break; |
| 11564 | case '!': /* (?!...) */ |
| 11565 | RExC_seen_zerolen++; |
| 11566 | /* check if we're really just a "FAIL" assertion */ |
| 11567 | skip_to_be_ignored_text(pRExC_state, &RExC_parse, |
| 11568 | FALSE /* Don't force to /x */ ); |
| 11569 | if (*RExC_parse == ')') { |
| 11570 | ret=reganode(pRExC_state, OPFAIL, 0); |
| 11571 | nextchar(pRExC_state); |
| 11572 | return ret; |
| 11573 | } |
| 11574 | break; |
| 11575 | case '|': /* (?|...) */ |
| 11576 | /* branch reset, behave like a (?:...) except that |
| 11577 | buffers in alternations share the same numbers */ |
| 11578 | paren = ':'; |
| 11579 | after_freeze = freeze_paren = RExC_npar; |
| 11580 | |
| 11581 | /* XXX This construct currently requires an extra pass. |
| 11582 | * Investigation would be required to see if that could be |
| 11583 | * changed */ |
| 11584 | REQUIRE_PARENS_PASS; |
| 11585 | break; |
| 11586 | case ':': /* (?:...) */ |
| 11587 | case '>': /* (?>...) */ |
| 11588 | break; |
| 11589 | case '$': /* (?$...) */ |
| 11590 | case '@': /* (?@...) */ |
| 11591 | vFAIL2("Sequence (?%c...) not implemented", (int)paren); |
| 11592 | break; |
| 11593 | case '0' : /* (?0) */ |
| 11594 | case 'R' : /* (?R) */ |
| 11595 | if (RExC_parse == RExC_end || *RExC_parse != ')') |
| 11596 | FAIL("Sequence (?R) not terminated"); |
| 11597 | num = 0; |
| 11598 | RExC_seen |= REG_RECURSE_SEEN; |
| 11599 | |
| 11600 | /* XXX These constructs currently require an extra pass. |
| 11601 | * It probably could be changed */ |
| 11602 | REQUIRE_PARENS_PASS; |
| 11603 | |
| 11604 | *flagp |= POSTPONED; |
| 11605 | goto gen_recurse_regop; |
| 11606 | /*notreached*/ |
| 11607 | /* named and numeric backreferences */ |
| 11608 | case '&': /* (?&NAME) */ |
| 11609 | parse_start = RExC_parse - 1; |
| 11610 | named_recursion: |
| 11611 | { |
| 11612 | SV *sv_dat = reg_scan_name(pRExC_state, |
| 11613 | REG_RSN_RETURN_DATA); |
| 11614 | num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0; |
| 11615 | } |
| 11616 | if (RExC_parse >= RExC_end || *RExC_parse != ')') |
| 11617 | vFAIL("Sequence (?&... not terminated"); |
| 11618 | goto gen_recurse_regop; |
| 11619 | /* NOTREACHED */ |
| 11620 | case '+': |
| 11621 | if (! inRANGE(RExC_parse[0], '1', '9')) { |
| 11622 | RExC_parse++; |
| 11623 | vFAIL("Illegal pattern"); |
| 11624 | } |
| 11625 | goto parse_recursion; |
| 11626 | /* NOTREACHED*/ |
| 11627 | case '-': /* (?-1) */ |
| 11628 | if (! inRANGE(RExC_parse[0], '1', '9')) { |
| 11629 | RExC_parse--; /* rewind to let it be handled later */ |
| 11630 | goto parse_flags; |
| 11631 | } |
| 11632 | /* FALLTHROUGH */ |
| 11633 | case '1': case '2': case '3': case '4': /* (?1) */ |
| 11634 | case '5': case '6': case '7': case '8': case '9': |
| 11635 | RExC_parse = (char *) seqstart + 1; /* Point to the digit */ |
| 11636 | parse_recursion: |
| 11637 | { |
| 11638 | bool is_neg = FALSE; |
| 11639 | UV unum; |
| 11640 | parse_start = RExC_parse - 1; /* MJD */ |
| 11641 | if (*RExC_parse == '-') { |
| 11642 | RExC_parse++; |
| 11643 | is_neg = TRUE; |
| 11644 | } |
| 11645 | endptr = RExC_end; |
| 11646 | if (grok_atoUV(RExC_parse, &unum, &endptr) |
| 11647 | && unum <= I32_MAX |
| 11648 | ) { |
| 11649 | num = (I32)unum; |
| 11650 | RExC_parse = (char*)endptr; |
| 11651 | } else |
| 11652 | num = I32_MAX; |
| 11653 | if (is_neg) { |
| 11654 | /* Some limit for num? */ |
| 11655 | num = -num; |
| 11656 | } |
| 11657 | } |
| 11658 | if (*RExC_parse!=')') |
| 11659 | vFAIL("Expecting close bracket"); |
| 11660 | |
| 11661 | gen_recurse_regop: |
| 11662 | if ( paren == '-' ) { |
| 11663 | /* |
| 11664 | Diagram of capture buffer numbering. |
| 11665 | Top line is the normal capture buffer numbers |
| 11666 | Bottom line is the negative indexing as from |
| 11667 | the X (the (?-2)) |
| 11668 | |
| 11669 | + 1 2 3 4 5 X 6 7 |
| 11670 | /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/ |
| 11671 | - 5 4 3 2 1 X x x |
| 11672 | |
| 11673 | */ |
| 11674 | num = RExC_npar + num; |
| 11675 | if (num < 1) { |
| 11676 | |
| 11677 | /* It might be a forward reference; we can't fail until |
| 11678 | * we know, by completing the parse to get all the |
| 11679 | * groups, and then reparsing */ |
| 11680 | if (ALL_PARENS_COUNTED) { |
| 11681 | RExC_parse++; |
| 11682 | vFAIL("Reference to nonexistent group"); |
| 11683 | } |
| 11684 | else { |
| 11685 | REQUIRE_PARENS_PASS; |
| 11686 | } |
| 11687 | } |
| 11688 | } else if ( paren == '+' ) { |
| 11689 | num = RExC_npar + num - 1; |
| 11690 | } |
| 11691 | /* We keep track how many GOSUB items we have produced. |
| 11692 | To start off the ARG2L() of the GOSUB holds its "id", |
| 11693 | which is used later in conjunction with RExC_recurse |
| 11694 | to calculate the offset we need to jump for the GOSUB, |
| 11695 | which it will store in the final representation. |
| 11696 | We have to defer the actual calculation until much later |
| 11697 | as the regop may move. |
| 11698 | */ |
| 11699 | |
| 11700 | ret = reg2Lanode(pRExC_state, GOSUB, num, RExC_recurse_count); |
| 11701 | if (num >= RExC_npar) { |
| 11702 | |
| 11703 | /* It might be a forward reference; we can't fail until we |
| 11704 | * know, by completing the parse to get all the groups, and |
| 11705 | * then reparsing */ |
| 11706 | if (ALL_PARENS_COUNTED) { |
| 11707 | if (num >= RExC_total_parens) { |
| 11708 | RExC_parse++; |
| 11709 | vFAIL("Reference to nonexistent group"); |
| 11710 | } |
| 11711 | } |
| 11712 | else { |
| 11713 | REQUIRE_PARENS_PASS; |
| 11714 | } |
| 11715 | } |
| 11716 | RExC_recurse_count++; |
| 11717 | DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_ |
| 11718 | "%*s%*s Recurse #%" UVuf " to %" IVdf "\n", |
| 11719 | 22, "| |", (int)(depth * 2 + 1), "", |
| 11720 | (UV)ARG(REGNODE_p(ret)), |
| 11721 | (IV)ARG2L(REGNODE_p(ret)))); |
| 11722 | RExC_seen |= REG_RECURSE_SEEN; |
| 11723 | |
| 11724 | Set_Node_Length(REGNODE_p(ret), |
| 11725 | 1 + regarglen[OP(REGNODE_p(ret))]); /* MJD */ |
| 11726 | Set_Node_Offset(REGNODE_p(ret), parse_start); /* MJD */ |
| 11727 | |
| 11728 | *flagp |= POSTPONED; |
| 11729 | assert(*RExC_parse == ')'); |
| 11730 | nextchar(pRExC_state); |
| 11731 | return ret; |
| 11732 | |
| 11733 | /* NOTREACHED */ |
| 11734 | |
| 11735 | case '?': /* (??...) */ |
| 11736 | is_logical = 1; |
| 11737 | if (*RExC_parse != '{') { |
| 11738 | RExC_parse += SKIP_IF_CHAR(RExC_parse, RExC_end); |
| 11739 | /* diag_listed_as: Sequence (?%s...) not recognized in regex; marked by <-- HERE in m/%s/ */ |
| 11740 | vFAIL2utf8f( |
| 11741 | "Sequence (%" UTF8f "...) not recognized", |
| 11742 | UTF8fARG(UTF, RExC_parse-seqstart, seqstart)); |
| 11743 | NOT_REACHED; /*NOTREACHED*/ |
| 11744 | } |
| 11745 | *flagp |= POSTPONED; |
| 11746 | paren = '{'; |
| 11747 | RExC_parse++; |
| 11748 | /* FALLTHROUGH */ |
| 11749 | case '{': /* (?{...}) */ |
| 11750 | { |
| 11751 | U32 n = 0; |
| 11752 | struct reg_code_block *cb; |
| 11753 | OP * o; |
| 11754 | |
| 11755 | RExC_seen_zerolen++; |
| 11756 | |
| 11757 | if ( !pRExC_state->code_blocks |
| 11758 | || pRExC_state->code_index |
| 11759 | >= pRExC_state->code_blocks->count |
| 11760 | || pRExC_state->code_blocks->cb[pRExC_state->code_index].start |
| 11761 | != (STRLEN)((RExC_parse -3 - (is_logical ? 1 : 0)) |
| 11762 | - RExC_start) |
| 11763 | ) { |
| 11764 | if (RExC_pm_flags & PMf_USE_RE_EVAL) |
| 11765 | FAIL("panic: Sequence (?{...}): no code block found\n"); |
| 11766 | FAIL("Eval-group not allowed at runtime, use re 'eval'"); |
| 11767 | } |
| 11768 | /* this is a pre-compiled code block (?{...}) */ |
| 11769 | cb = &pRExC_state->code_blocks->cb[pRExC_state->code_index]; |
| 11770 | RExC_parse = RExC_start + cb->end; |
| 11771 | o = cb->block; |
| 11772 | if (cb->src_regex) { |
| 11773 | n = add_data(pRExC_state, STR_WITH_LEN("rl")); |
| 11774 | RExC_rxi->data->data[n] = |
| 11775 | (void*)SvREFCNT_inc((SV*)cb->src_regex); |
| 11776 | RExC_rxi->data->data[n+1] = (void*)o; |
| 11777 | } |
| 11778 | else { |
| 11779 | n = add_data(pRExC_state, |
| 11780 | (RExC_pm_flags & PMf_HAS_CV) ? "L" : "l", 1); |
| 11781 | RExC_rxi->data->data[n] = (void*)o; |
| 11782 | } |
| 11783 | pRExC_state->code_index++; |
| 11784 | nextchar(pRExC_state); |
| 11785 | |
| 11786 | if (is_logical) { |
| 11787 | regnode_offset eval; |
| 11788 | ret = reg_node(pRExC_state, LOGICAL); |
| 11789 | |
| 11790 | eval = reg2Lanode(pRExC_state, EVAL, |
| 11791 | n, |
| 11792 | |
| 11793 | /* for later propagation into (??{}) |
| 11794 | * return value */ |
| 11795 | RExC_flags & RXf_PMf_COMPILETIME |
| 11796 | ); |
| 11797 | FLAGS(REGNODE_p(ret)) = 2; |
| 11798 | if (! REGTAIL(pRExC_state, ret, eval)) { |
| 11799 | REQUIRE_BRANCHJ(flagp, 0); |
| 11800 | } |
| 11801 | /* deal with the length of this later - MJD */ |
| 11802 | return ret; |
| 11803 | } |
| 11804 | ret = reg2Lanode(pRExC_state, EVAL, n, 0); |
| 11805 | Set_Node_Length(REGNODE_p(ret), RExC_parse - parse_start + 1); |
| 11806 | Set_Node_Offset(REGNODE_p(ret), parse_start); |
| 11807 | return ret; |
| 11808 | } |
| 11809 | case '(': /* (?(?{...})...) and (?(?=...)...) */ |
| 11810 | { |
| 11811 | int is_define= 0; |
| 11812 | const int DEFINE_len = sizeof("DEFINE") - 1; |
| 11813 | if ( RExC_parse < RExC_end - 1 |
| 11814 | && ( ( RExC_parse[0] == '?' /* (?(?...)) */ |
| 11815 | && ( RExC_parse[1] == '=' |
| 11816 | || RExC_parse[1] == '!' |
| 11817 | || RExC_parse[1] == '<' |
| 11818 | || RExC_parse[1] == '{')) |
| 11819 | || ( RExC_parse[0] == '*' /* (?(*...)) */ |
| 11820 | && ( memBEGINs(RExC_parse + 1, |
| 11821 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11822 | "pla:") |
| 11823 | || memBEGINs(RExC_parse + 1, |
| 11824 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11825 | "plb:") |
| 11826 | || memBEGINs(RExC_parse + 1, |
| 11827 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11828 | "nla:") |
| 11829 | || memBEGINs(RExC_parse + 1, |
| 11830 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11831 | "nlb:") |
| 11832 | || memBEGINs(RExC_parse + 1, |
| 11833 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11834 | "positive_lookahead:") |
| 11835 | || memBEGINs(RExC_parse + 1, |
| 11836 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11837 | "positive_lookbehind:") |
| 11838 | || memBEGINs(RExC_parse + 1, |
| 11839 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11840 | "negative_lookahead:") |
| 11841 | || memBEGINs(RExC_parse + 1, |
| 11842 | (Size_t) (RExC_end - (RExC_parse + 1)), |
| 11843 | "negative_lookbehind:")))) |
| 11844 | ) { /* Lookahead or eval. */ |
| 11845 | I32 flag; |
| 11846 | regnode_offset tail; |
| 11847 | |
| 11848 | ret = reg_node(pRExC_state, LOGICAL); |
| 11849 | FLAGS(REGNODE_p(ret)) = 1; |
| 11850 | |
| 11851 | tail = reg(pRExC_state, 1, &flag, depth+1); |
| 11852 | RETURN_FAIL_ON_RESTART(flag, flagp); |
| 11853 | if (! REGTAIL(pRExC_state, ret, tail)) { |
| 11854 | REQUIRE_BRANCHJ(flagp, 0); |
| 11855 | } |
| 11856 | goto insert_if; |
| 11857 | } |
| 11858 | else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */ |
| 11859 | || RExC_parse[0] == '\'' ) /* (?('NAME')...) */ |
| 11860 | { |
| 11861 | char ch = RExC_parse[0] == '<' ? '>' : '\''; |
| 11862 | char *name_start= RExC_parse++; |
| 11863 | U32 num = 0; |
| 11864 | SV *sv_dat=reg_scan_name(pRExC_state, REG_RSN_RETURN_DATA); |
| 11865 | if ( RExC_parse == name_start |
| 11866 | || RExC_parse >= RExC_end |
| 11867 | || *RExC_parse != ch) |
| 11868 | { |
| 11869 | vFAIL2("Sequence (?(%c... not terminated", |
| 11870 | (ch == '>' ? '<' : ch)); |
| 11871 | } |
| 11872 | RExC_parse++; |
| 11873 | if (sv_dat) { |
| 11874 | num = add_data( pRExC_state, STR_WITH_LEN("S")); |
| 11875 | RExC_rxi->data->data[num]=(void*)sv_dat; |
| 11876 | SvREFCNT_inc_simple_void_NN(sv_dat); |
| 11877 | } |
| 11878 | ret = reganode(pRExC_state, GROUPPN, num); |
| 11879 | goto insert_if_check_paren; |
| 11880 | } |
| 11881 | else if (memBEGINs(RExC_parse, |
| 11882 | (STRLEN) (RExC_end - RExC_parse), |
| 11883 | "DEFINE")) |
| 11884 | { |
| 11885 | ret = reganode(pRExC_state, DEFINEP, 0); |
| 11886 | RExC_parse += DEFINE_len; |
| 11887 | is_define = 1; |
| 11888 | goto insert_if_check_paren; |
| 11889 | } |
| 11890 | else if (RExC_parse[0] == 'R') { |
| 11891 | RExC_parse++; |
| 11892 | /* parno == 0 => /(?(R)YES|NO)/ "in any form of recursion OR eval" |
| 11893 | * parno == 1 => /(?(R0)YES|NO)/ "in GOSUB (?0) / (?R)" |
| 11894 | * parno == 2 => /(?(R1)YES|NO)/ "in GOSUB (?1) (parno-1)" |
| 11895 | */ |
| 11896 | parno = 0; |
| 11897 | if (RExC_parse[0] == '0') { |
| 11898 | parno = 1; |
| 11899 | RExC_parse++; |
| 11900 | } |
| 11901 | else if (inRANGE(RExC_parse[0], '1', '9')) { |
| 11902 | UV uv; |
| 11903 | endptr = RExC_end; |
| 11904 | if (grok_atoUV(RExC_parse, &uv, &endptr) |
| 11905 | && uv <= I32_MAX |
| 11906 | ) { |
| 11907 | parno = (I32)uv + 1; |
| 11908 | RExC_parse = (char*)endptr; |
| 11909 | } |
| 11910 | /* else "Switch condition not recognized" below */ |
| 11911 | } else if (RExC_parse[0] == '&') { |
| 11912 | SV *sv_dat; |
| 11913 | RExC_parse++; |
| 11914 | sv_dat = reg_scan_name(pRExC_state, |
| 11915 | REG_RSN_RETURN_DATA); |
| 11916 | if (sv_dat) |
| 11917 | parno = 1 + *((I32 *)SvPVX(sv_dat)); |
| 11918 | } |
| 11919 | ret = reganode(pRExC_state, INSUBP, parno); |
| 11920 | goto insert_if_check_paren; |
| 11921 | } |
| 11922 | else if (inRANGE(RExC_parse[0], '1', '9')) { |
| 11923 | /* (?(1)...) */ |
| 11924 | char c; |
| 11925 | UV uv; |
| 11926 | endptr = RExC_end; |
| 11927 | if (grok_atoUV(RExC_parse, &uv, &endptr) |
| 11928 | && uv <= I32_MAX |
| 11929 | ) { |
| 11930 | parno = (I32)uv; |
| 11931 | RExC_parse = (char*)endptr; |
| 11932 | } |
| 11933 | else { |
| 11934 | vFAIL("panic: grok_atoUV returned FALSE"); |
| 11935 | } |
| 11936 | ret = reganode(pRExC_state, GROUPP, parno); |
| 11937 | |
| 11938 | insert_if_check_paren: |
| 11939 | if (UCHARAT(RExC_parse) != ')') { |
| 11940 | RExC_parse += UTF |
| 11941 | ? UTF8_SAFE_SKIP(RExC_parse, RExC_end) |
| 11942 | : 1; |
| 11943 | vFAIL("Switch condition not recognized"); |
| 11944 | } |
| 11945 | nextchar(pRExC_state); |
| 11946 | insert_if: |
| 11947 | if (! REGTAIL(pRExC_state, ret, reganode(pRExC_state, |
| 11948 | IFTHEN, 0))) |
| 11949 | { |
| 11950 | REQUIRE_BRANCHJ(flagp, 0); |
| 11951 | } |
| 11952 | br = regbranch(pRExC_state, &flags, 1, depth+1); |
| 11953 | if (br == 0) { |
| 11954 | RETURN_FAIL_ON_RESTART(flags,flagp); |
| 11955 | FAIL2("panic: regbranch returned failure, flags=%#" UVxf, |
| 11956 | (UV) flags); |
| 11957 | } else |
| 11958 | if (! REGTAIL(pRExC_state, br, reganode(pRExC_state, |
| 11959 | LONGJMP, 0))) |
| 11960 | { |
| 11961 | REQUIRE_BRANCHJ(flagp, 0); |
| 11962 | } |
| 11963 | c = UCHARAT(RExC_parse); |
| 11964 | nextchar(pRExC_state); |
| 11965 | if (flags&HASWIDTH) |
| 11966 | *flagp |= HASWIDTH; |
| 11967 | if (c == '|') { |
| 11968 | if (is_define) |
| 11969 | vFAIL("(?(DEFINE)....) does not allow branches"); |
| 11970 | |
| 11971 | /* Fake one for optimizer. */ |
| 11972 | lastbr = reganode(pRExC_state, IFTHEN, 0); |
| 11973 | |
| 11974 | if (!regbranch(pRExC_state, &flags, 1, depth+1)) { |
| 11975 | RETURN_FAIL_ON_RESTART(flags, flagp); |
| 11976 | FAIL2("panic: regbranch returned failure, flags=%#" UVxf, |
| 11977 | (UV) flags); |
| 11978 | } |
| 11979 | if (! REGTAIL(pRExC_state, ret, lastbr)) { |
| 11980 | REQUIRE_BRANCHJ(flagp, 0); |
| 11981 | } |
| 11982 | if (flags&HASWIDTH) |
| 11983 | *flagp |= HASWIDTH; |
| 11984 | c = UCHARAT(RExC_parse); |
| 11985 | nextchar(pRExC_state); |
| 11986 | } |
| 11987 | else |
| 11988 | lastbr = 0; |
| 11989 | if (c != ')') { |
| 11990 | if (RExC_parse >= RExC_end) |
| 11991 | vFAIL("Switch (?(condition)... not terminated"); |
| 11992 | else |
| 11993 | vFAIL("Switch (?(condition)... contains too many branches"); |
| 11994 | } |
| 11995 | ender = reg_node(pRExC_state, TAIL); |
| 11996 | if (! REGTAIL(pRExC_state, br, ender)) { |
| 11997 | REQUIRE_BRANCHJ(flagp, 0); |
| 11998 | } |
| 11999 | if (lastbr) { |
| 12000 | if (! REGTAIL(pRExC_state, lastbr, ender)) { |
| 12001 | REQUIRE_BRANCHJ(flagp, 0); |
| 12002 | } |
| 12003 | if (! REGTAIL(pRExC_state, |
| 12004 | REGNODE_OFFSET( |
| 12005 | NEXTOPER( |
| 12006 | NEXTOPER(REGNODE_p(lastbr)))), |
| 12007 | ender)) |
| 12008 | { |
| 12009 | REQUIRE_BRANCHJ(flagp, 0); |
| 12010 | } |
| 12011 | } |
| 12012 | else |
| 12013 | if (! REGTAIL(pRExC_state, ret, ender)) { |
| 12014 | REQUIRE_BRANCHJ(flagp, 0); |
| 12015 | } |
| 12016 | #if 0 /* Removing this doesn't cause failures in the test suite -- khw */ |
| 12017 | RExC_size++; /* XXX WHY do we need this?!! |
| 12018 | For large programs it seems to be required |
| 12019 | but I can't figure out why. -- dmq*/ |
| 12020 | #endif |
| 12021 | return ret; |
| 12022 | } |
| 12023 | RExC_parse += UTF |
| 12024 | ? UTF8_SAFE_SKIP(RExC_parse, RExC_end) |
| 12025 | : 1; |
| 12026 | vFAIL("Unknown switch condition (?(...))"); |
| 12027 | } |
| 12028 | case '[': /* (?[ ... ]) */ |
| 12029 | return handle_regex_sets(pRExC_state, NULL, flagp, depth+1, |
| 12030 | oregcomp_parse); |
| 12031 | case 0: /* A NUL */ |
| 12032 | RExC_parse--; /* for vFAIL to print correctly */ |
| 12033 | vFAIL("Sequence (? incomplete"); |
| 12034 | break; |
| 12035 | |
| 12036 | case ')': |
| 12037 | if (RExC_strict) { /* [perl #132851] */ |
| 12038 | ckWARNreg(RExC_parse, "Empty (?) without any modifiers"); |
| 12039 | } |
| 12040 | /* FALLTHROUGH */ |
| 12041 | case '*': /* If you want to support (?*...), first reconcile with GH #17363 */ |
| 12042 | /* FALLTHROUGH */ |
| 12043 | default: /* e.g., (?i) */ |
| 12044 | RExC_parse = (char *) seqstart + 1; |
| 12045 | parse_flags: |
| 12046 | parse_lparen_question_flags(pRExC_state); |
| 12047 | if (UCHARAT(RExC_parse) != ':') { |
| 12048 | if (RExC_parse < RExC_end) |
| 12049 | nextchar(pRExC_state); |
| 12050 | *flagp = TRYAGAIN; |
| 12051 | return 0; |
| 12052 | } |
| 12053 | paren = ':'; |
| 12054 | nextchar(pRExC_state); |
| 12055 | ret = 0; |
| 12056 | goto parse_rest; |
| 12057 | } /* end switch */ |
| 12058 | } |
| 12059 | else if (!(RExC_flags & RXf_PMf_NOCAPTURE)) { /* (...) */ |
| 12060 | capturing_parens: |
| 12061 | parno = RExC_npar; |
| 12062 | RExC_npar++; |
| 12063 | if (! ALL_PARENS_COUNTED) { |
| 12064 | /* If we are in our first pass through (and maybe only pass), |
| 12065 | * we need to allocate memory for the capturing parentheses |
| 12066 | * data structures. |
| 12067 | */ |
| 12068 | |
| 12069 | if (!RExC_parens_buf_size) { |
| 12070 | /* first guess at number of parens we might encounter */ |
| 12071 | RExC_parens_buf_size = 10; |
| 12072 | |
| 12073 | /* setup RExC_open_parens, which holds the address of each |
| 12074 | * OPEN tag, and to make things simpler for the 0 index the |
| 12075 | * start of the program - this is used later for offsets */ |
| 12076 | Newxz(RExC_open_parens, RExC_parens_buf_size, |
| 12077 | regnode_offset); |
| 12078 | RExC_open_parens[0] = 1; /* +1 for REG_MAGIC */ |
| 12079 | |
| 12080 | /* setup RExC_close_parens, which holds the address of each |
| 12081 | * CLOSE tag, and to make things simpler for the 0 index |
| 12082 | * the end of the program - this is used later for offsets |
| 12083 | * */ |
| 12084 | Newxz(RExC_close_parens, RExC_parens_buf_size, |
| 12085 | regnode_offset); |
| 12086 | /* we dont know where end op starts yet, so we dont need to |
| 12087 | * set RExC_close_parens[0] like we do RExC_open_parens[0] |
| 12088 | * above */ |
| 12089 | } |
| 12090 | else if (RExC_npar > RExC_parens_buf_size) { |
| 12091 | I32 old_size = RExC_parens_buf_size; |
| 12092 | |
| 12093 | RExC_parens_buf_size *= 2; |
| 12094 | |
| 12095 | Renew(RExC_open_parens, RExC_parens_buf_size, |
| 12096 | regnode_offset); |
| 12097 | Zero(RExC_open_parens + old_size, |
| 12098 | RExC_parens_buf_size - old_size, regnode_offset); |
| 12099 | |
| 12100 | Renew(RExC_close_parens, RExC_parens_buf_size, |
| 12101 | regnode_offset); |
| 12102 | Zero(RExC_close_parens + old_size, |
| 12103 | RExC_parens_buf_size - old_size, regnode_offset); |
| 12104 | } |
| 12105 | } |
| 12106 | |
| 12107 | ret = reganode(pRExC_state, OPEN, parno); |
| 12108 | if (!RExC_nestroot) |
| 12109 | RExC_nestroot = parno; |
| 12110 | if (RExC_open_parens && !RExC_open_parens[parno]) |
| 12111 | { |
| 12112 | DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_ |
| 12113 | "%*s%*s Setting open paren #%" IVdf " to %zu\n", |
| 12114 | 22, "| |", (int)(depth * 2 + 1), "", |
| 12115 | (IV)parno, ret)); |
| 12116 | RExC_open_parens[parno]= ret; |
| 12117 | } |
| 12118 | |
| 12119 | Set_Node_Length(REGNODE_p(ret), 1); /* MJD */ |
| 12120 | Set_Node_Offset(REGNODE_p(ret), RExC_parse); /* MJD */ |
| 12121 | is_open = 1; |
| 12122 | } else { |
| 12123 | /* with RXf_PMf_NOCAPTURE treat (...) as (?:...) */ |
| 12124 | paren = ':'; |
| 12125 | ret = 0; |
| 12126 | } |
| 12127 | } |
| 12128 | else /* ! paren */ |
| 12129 | ret = 0; |
| 12130 | |
| 12131 | parse_rest: |
| 12132 | /* Pick up the branches, linking them together. */ |
| 12133 | parse_start = RExC_parse; /* MJD */ |
| 12134 | br = regbranch(pRExC_state, &flags, 1, depth+1); |
| 12135 | |
| 12136 | /* branch_len = (paren != 0); */ |
| 12137 | |
| 12138 | if (br == 0) { |
| 12139 | RETURN_FAIL_ON_RESTART(flags, flagp); |
| 12140 | FAIL2("panic: regbranch returned failure, flags=%#" UVxf, (UV) flags); |
| 12141 | } |
| 12142 | if (*RExC_parse == '|') { |
| 12143 | if (RExC_use_BRANCHJ) { |
| 12144 | reginsert(pRExC_state, BRANCHJ, br, depth+1); |
| 12145 | } |
| 12146 | else { /* MJD */ |
| 12147 | reginsert(pRExC_state, BRANCH, br, depth+1); |
| 12148 | Set_Node_Length(REGNODE_p(br), paren != 0); |
| 12149 | Set_Node_Offset_To_R(br, parse_start-RExC_start); |
| 12150 | } |
| 12151 | have_branch = 1; |
| 12152 | } |
| 12153 | else if (paren == ':') { |
| 12154 | *flagp |= flags&SIMPLE; |
| 12155 | } |
| 12156 | if (is_open) { /* Starts with OPEN. */ |
| 12157 | if (! REGTAIL(pRExC_state, ret, br)) { /* OPEN -> first. */ |
| 12158 | REQUIRE_BRANCHJ(flagp, 0); |
| 12159 | } |
| 12160 | } |
| 12161 | else if (paren != '?') /* Not Conditional */ |
| 12162 | ret = br; |
| 12163 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
| 12164 | lastbr = br; |
| 12165 | while (*RExC_parse == '|') { |
| 12166 | if (RExC_use_BRANCHJ) { |
| 12167 | bool shut_gcc_up; |
| 12168 | |
| 12169 | ender = reganode(pRExC_state, LONGJMP, 0); |
| 12170 | |
| 12171 | /* Append to the previous. */ |
| 12172 | shut_gcc_up = REGTAIL(pRExC_state, |
| 12173 | REGNODE_OFFSET(NEXTOPER(NEXTOPER(REGNODE_p(lastbr)))), |
| 12174 | ender); |
| 12175 | PERL_UNUSED_VAR(shut_gcc_up); |
| 12176 | } |
| 12177 | nextchar(pRExC_state); |
| 12178 | if (freeze_paren) { |
| 12179 | if (RExC_npar > after_freeze) |
| 12180 | after_freeze = RExC_npar; |
| 12181 | RExC_npar = freeze_paren; |
| 12182 | } |
| 12183 | br = regbranch(pRExC_state, &flags, 0, depth+1); |
| 12184 | |
| 12185 | if (br == 0) { |
| 12186 | RETURN_FAIL_ON_RESTART(flags, flagp); |
| 12187 | FAIL2("panic: regbranch returned failure, flags=%#" UVxf, (UV) flags); |
| 12188 | } |
| 12189 | if (! REGTAIL(pRExC_state, lastbr, br)) { /* BRANCH -> BRANCH. */ |
| 12190 | REQUIRE_BRANCHJ(flagp, 0); |
| 12191 | } |
| 12192 | lastbr = br; |
| 12193 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
| 12194 | } |
| 12195 | |
| 12196 | if (have_branch || paren != ':') { |
| 12197 | regnode * br; |
| 12198 | |
| 12199 | /* Make a closing node, and hook it on the end. */ |
| 12200 | switch (paren) { |
| 12201 | case ':': |
| 12202 | ender = reg_node(pRExC_state, TAIL); |
| 12203 | break; |
| 12204 | case 1: case 2: |
| 12205 | ender = reganode(pRExC_state, CLOSE, parno); |
| 12206 | if ( RExC_close_parens ) { |
| 12207 | DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_ |
| 12208 | "%*s%*s Setting close paren #%" IVdf " to %zu\n", |
| 12209 | 22, "| |", (int)(depth * 2 + 1), "", |
| 12210 | (IV)parno, ender)); |
| 12211 | RExC_close_parens[parno]= ender; |
| 12212 | if (RExC_nestroot == parno) |
| 12213 | RExC_nestroot = 0; |
| 12214 | } |
| 12215 | Set_Node_Offset(REGNODE_p(ender), RExC_parse+1); /* MJD */ |
| 12216 | Set_Node_Length(REGNODE_p(ender), 1); /* MJD */ |
| 12217 | break; |
| 12218 | case 's': |
| 12219 | ender = reg_node(pRExC_state, SRCLOSE); |
| 12220 | RExC_in_script_run = 0; |
| 12221 | break; |
| 12222 | case '<': |
| 12223 | case 'a': |
| 12224 | case 'A': |
| 12225 | case 'b': |
| 12226 | case 'B': |
| 12227 | case ',': |
| 12228 | case '=': |
| 12229 | case '!': |
| 12230 | *flagp &= ~HASWIDTH; |
| 12231 | /* FALLTHROUGH */ |
| 12232 | case 't': /* aTomic */ |
| 12233 | case '>': |
| 12234 | ender = reg_node(pRExC_state, SUCCEED); |
| 12235 | break; |
| 12236 | case 0: |
| 12237 | ender = reg_node(pRExC_state, END); |
| 12238 | assert(!RExC_end_op); /* there can only be one! */ |
| 12239 | RExC_end_op = REGNODE_p(ender); |
| 12240 | if (RExC_close_parens) { |
| 12241 | DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_ |
| 12242 | "%*s%*s Setting close paren #0 (END) to %zu\n", |
| 12243 | 22, "| |", (int)(depth * 2 + 1), "", |
| 12244 | ender)); |
| 12245 | |
| 12246 | RExC_close_parens[0]= ender; |
| 12247 | } |
| 12248 | break; |
| 12249 | } |
| 12250 | DEBUG_PARSE_r({ |
| 12251 | DEBUG_PARSE_MSG("lsbr"); |
| 12252 | regprop(RExC_rx, RExC_mysv1, REGNODE_p(lastbr), NULL, pRExC_state); |
| 12253 | regprop(RExC_rx, RExC_mysv2, REGNODE_p(ender), NULL, pRExC_state); |
| 12254 | Perl_re_printf( aTHX_ "~ tying lastbr %s (%" IVdf ") to ender %s (%" IVdf ") offset %" IVdf "\n", |
| 12255 | SvPV_nolen_const(RExC_mysv1), |
| 12256 | (IV)lastbr, |
| 12257 | SvPV_nolen_const(RExC_mysv2), |
| 12258 | (IV)ender, |
| 12259 | (IV)(ender - lastbr) |
| 12260 | ); |
| 12261 | }); |
| 12262 | if (! REGTAIL(pRExC_state, lastbr, ender)) { |
| 12263 | REQUIRE_BRANCHJ(flagp, 0); |
| 12264 | } |
| 12265 | |
| 12266 | if (have_branch) { |
| 12267 | char is_nothing= 1; |
| 12268 | if (depth==1) |
| 12269 | RExC_seen |= REG_TOP_LEVEL_BRANCHES_SEEN; |
| 12270 | |
| 12271 | /* Hook the tails of the branches to the closing node. */ |
| 12272 | for (br = REGNODE_p(ret); br; br = regnext(br)) { |
| 12273 | const U8 op = PL_regkind[OP(br)]; |
| 12274 | if (op == BRANCH) { |
| 12275 | if (! REGTAIL_STUDY(pRExC_state, |
| 12276 | REGNODE_OFFSET(NEXTOPER(br)), |
| 12277 | ender)) |
| 12278 | { |
| 12279 | REQUIRE_BRANCHJ(flagp, 0); |
| 12280 | } |
| 12281 | if ( OP(NEXTOPER(br)) != NOTHING |
| 12282 | || regnext(NEXTOPER(br)) != REGNODE_p(ender)) |
| 12283 | is_nothing= 0; |
| 12284 | } |
| 12285 | else if (op == BRANCHJ) { |
| 12286 | bool shut_gcc_up = REGTAIL_STUDY(pRExC_state, |
| 12287 | REGNODE_OFFSET(NEXTOPER(NEXTOPER(br))), |
| 12288 | ender); |
| 12289 | PERL_UNUSED_VAR(shut_gcc_up); |
| 12290 | /* for now we always disable this optimisation * / |
| 12291 | if ( OP(NEXTOPER(NEXTOPER(br))) != NOTHING |
| 12292 | || regnext(NEXTOPER(NEXTOPER(br))) != REGNODE_p(ender)) |
| 12293 | */ |
| 12294 | is_nothing= 0; |
| 12295 | } |
| 12296 | } |
| 12297 | if (is_nothing) { |
| 12298 | regnode * ret_as_regnode = REGNODE_p(ret); |
| 12299 | br= PL_regkind[OP(ret_as_regnode)] != BRANCH |
| 12300 | ? regnext(ret_as_regnode) |
| 12301 | : ret_as_regnode; |
| 12302 | DEBUG_PARSE_r({ |
| 12303 | DEBUG_PARSE_MSG("NADA"); |
| 12304 | regprop(RExC_rx, RExC_mysv1, ret_as_regnode, |
| 12305 | NULL, pRExC_state); |
| 12306 | regprop(RExC_rx, RExC_mysv2, REGNODE_p(ender), |
| 12307 | NULL, pRExC_state); |
| 12308 | Perl_re_printf( aTHX_ "~ converting ret %s (%" IVdf ") to ender %s (%" IVdf ") offset %" IVdf "\n", |
| 12309 | SvPV_nolen_const(RExC_mysv1), |
| 12310 | (IV)REG_NODE_NUM(ret_as_regnode), |
| 12311 | SvPV_nolen_const(RExC_mysv2), |
| 12312 | (IV)ender, |
| 12313 | (IV)(ender - ret) |
| 12314 | ); |
| 12315 | }); |
| 12316 | OP(br)= NOTHING; |
| 12317 | if (OP(REGNODE_p(ender)) == TAIL) { |
| 12318 | NEXT_OFF(br)= 0; |
| 12319 | RExC_emit= REGNODE_OFFSET(br) + 1; |
| 12320 | } else { |
| 12321 | regnode *opt; |
| 12322 | for ( opt= br + 1; opt < REGNODE_p(ender) ; opt++ ) |
| 12323 | OP(opt)= OPTIMIZED; |
| 12324 | NEXT_OFF(br)= REGNODE_p(ender) - br; |
| 12325 | } |
| 12326 | } |
| 12327 | } |
| 12328 | } |
| 12329 | |
| 12330 | { |
| 12331 | const char *p; |
| 12332 | /* Even/odd or x=don't care: 010101x10x */ |
| 12333 | static const char parens[] = "=!aA<,>Bbt"; |
| 12334 | /* flag below is set to 0 up through 'A'; 1 for larger */ |
| 12335 | |
| 12336 | if (paren && (p = strchr(parens, paren))) { |
| 12337 | U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH; |
| 12338 | int flag = (p - parens) > 3; |
| 12339 | |
| 12340 | if (paren == '>' || paren == 't') { |
| 12341 | node = SUSPEND, flag = 0; |
| 12342 | } |
| 12343 | |
| 12344 | reginsert(pRExC_state, node, ret, depth+1); |
| 12345 | Set_Node_Cur_Length(REGNODE_p(ret), parse_start); |
| 12346 | Set_Node_Offset(REGNODE_p(ret), parse_start + 1); |
| 12347 | FLAGS(REGNODE_p(ret)) = flag; |
| 12348 | if (! REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL))) |
| 12349 | { |
| 12350 | REQUIRE_BRANCHJ(flagp, 0); |
| 12351 | } |
| 12352 | } |
| 12353 | } |
| 12354 | |
| 12355 | /* Check for proper termination. */ |
| 12356 | if (paren) { |
| 12357 | /* restore original flags, but keep (?p) and, if we've encountered |
| 12358 | * something in the parse that changes /d rules into /u, keep the /u */ |
| 12359 | RExC_flags = oregflags | (RExC_flags & RXf_PMf_KEEPCOPY); |
| 12360 | if (DEPENDS_SEMANTICS && RExC_uni_semantics) { |
| 12361 | set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); |
| 12362 | } |
| 12363 | if (RExC_parse >= RExC_end || UCHARAT(RExC_parse) != ')') { |
| 12364 | RExC_parse = oregcomp_parse; |
| 12365 | vFAIL("Unmatched ("); |
| 12366 | } |
| 12367 | nextchar(pRExC_state); |
| 12368 | } |
| 12369 | else if (!paren && RExC_parse < RExC_end) { |
| 12370 | if (*RExC_parse == ')') { |
| 12371 | RExC_parse++; |
| 12372 | vFAIL("Unmatched )"); |
| 12373 | } |
| 12374 | else |
| 12375 | FAIL("Junk on end of regexp"); /* "Can't happen". */ |
| 12376 | NOT_REACHED; /* NOTREACHED */ |
| 12377 | } |
| 12378 | |
| 12379 | if (RExC_in_lookbehind) { |
| 12380 | RExC_in_lookbehind--; |
| 12381 | } |
| 12382 | if (RExC_in_lookahead) { |
| 12383 | RExC_in_lookahead--; |
| 12384 | } |
| 12385 | if (after_freeze > RExC_npar) |
| 12386 | RExC_npar = after_freeze; |
| 12387 | return(ret); |
| 12388 | } |
| 12389 | |
| 12390 | /* |
| 12391 | - regbranch - one alternative of an | operator |
| 12392 | * |
| 12393 | * Implements the concatenation operator. |
| 12394 | * |
| 12395 | * On success, returns the offset at which any next node should be placed into |
| 12396 | * the regex engine program being compiled. |
| 12397 | * |
| 12398 | * Returns 0 otherwise, setting flagp to RESTART_PARSE if the parse needs |
| 12399 | * to be restarted, or'd with NEED_UTF8 if the pattern needs to be upgraded to |
| 12400 | * UTF-8 |
| 12401 | */ |
| 12402 | STATIC regnode_offset |
| 12403 | S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth) |
| 12404 | { |
| 12405 | regnode_offset ret; |
| 12406 | regnode_offset chain = 0; |
| 12407 | regnode_offset latest; |
| 12408 | I32 flags = 0, c = 0; |
| 12409 | GET_RE_DEBUG_FLAGS_DECL; |
| 12410 | |
| 12411 | PERL_ARGS_ASSERT_REGBRANCH; |
| 12412 | |
| 12413 | DEBUG_PARSE("brnc"); |
| 12414 | |
| 12415 | if (first) |
| 12416 | ret = 0; |
| 12417 | else { |
| 12418 | if (RExC_use_BRANCHJ) |
| 12419 | ret = reganode(pRExC_state, BRANCHJ, 0); |
| 12420 | else { |
| 12421 | ret = reg_node(pRExC_state, BRANCH); |
| 12422 | Set_Node_Length(REGNODE_p(ret), 1); |
| 12423 | } |
| 12424 | } |
| 12425 | |
| 12426 | *flagp = WORST; /* Tentatively. */ |
| 12427 | |
| 12428 | skip_to_be_ignored_text(pRExC_state, &RExC_parse, |
| 12429 | FALSE /* Don't force to /x */ ); |
| 12430 | while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') { |
| 12431 | flags &= ~TRYAGAIN; |
| 12432 | latest = regpiece(pRExC_state, &flags, depth+1); |
| 12433 | if (latest == 0) { |
| 12434 | if (flags & TRYAGAIN) |
| 12435 | continue; |
| 12436 | RETURN_FAIL_ON_RESTART(flags, flagp); |
| 12437 | FAIL2("panic: regpiece returned failure, flags=%#" UVxf, (UV) flags); |
| 12438 | } |
| 12439 | else if (ret == 0) |
| 12440 | ret = latest; |
| 12441 | *flagp |= flags&(HASWIDTH|POSTPONED); |
| 12442 | if (chain == 0) /* First piece. */ |
| 12443 | *flagp |= flags&SPSTART; |
| 12444 | else { |
| 12445 | /* FIXME adding one for every branch after the first is probably |
| 12446 | * excessive now we have TRIE support. (hv) */ |
| 12447 | MARK_NAUGHTY(1); |
| 12448 | if (! REGTAIL(pRExC_state, chain, latest)) { |
| 12449 | /* XXX We could just redo this branch, but figuring out what |
| 12450 | * bookkeeping needs to be reset is a pain, and it's likely |
| 12451 | * that other branches that goto END will also be too large */ |
| 12452 | REQUIRE_BRANCHJ(flagp, 0); |
| 12453 | } |
| 12454 | } |
| 12455 | chain = latest; |
| 12456 | c++; |
| 12457 | } |
| 12458 | if (chain == 0) { /* Loop ran zero times. */ |
| 12459 | chain = reg_node(pRExC_state, NOTHING); |
| 12460 | if (ret == 0) |
| 12461 | ret = chain; |
| 12462 | } |
| 12463 | if (c == 1) { |
| 12464 | *flagp |= flags&SIMPLE; |
| 12465 | } |
| 12466 | |
| 12467 | return ret; |
| 12468 | } |
| 12469 | |
| 12470 | /* |
| 12471 | - regpiece - something followed by possible quantifier * + ? {n,m} |
| 12472 | * |
| 12473 | * Note that the branching code sequences used for ? and the general cases |
| 12474 | * of * and + are somewhat optimized: they use the same NOTHING node as |
| 12475 | * both the endmarker for their branch list and the body of the last branch. |
| 12476 | * It might seem that this node could be dispensed with entirely, but the |
| 12477 | * endmarker role is not redundant. |
| 12478 | * |
| 12479 | * On success, returns the offset at which any next node should be placed into |
| 12480 | * the regex engine program being compiled. |
| 12481 | * |
| 12482 | * Returns 0 otherwise, with *flagp set to indicate why: |
| 12483 | * TRYAGAIN if regatom() returns 0 with TRYAGAIN. |
| 12484 | * RESTART_PARSE if the parse needs to be restarted, or'd with |
| 12485 | * NEED_UTF8 if the pattern needs to be upgraded to UTF-8. |
| 12486 | */ |
| 12487 | STATIC regnode_offset |
| 12488 | S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
| 12489 | { |
| 12490 | regnode_offset ret; |
| 12491 | char op; |
| 12492 | char *next; |
| 12493 | I32 flags; |
| 12494 | const char * const origparse = RExC_parse; |
| 12495 | I32 min; |
| 12496 | I32 max = REG_INFTY; |
| 12497 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 12498 | char *parse_start; |
| 12499 | #endif |
| 12500 | const char *maxpos = NULL; |
| 12501 | UV uv; |
| 12502 | |
| 12503 | /* Save the original in case we change the emitted regop to a FAIL. */ |
| 12504 | const regnode_offset orig_emit = RExC_emit; |
| 12505 | |
| 12506 | GET_RE_DEBUG_FLAGS_DECL; |
| 12507 | |
| 12508 | PERL_ARGS_ASSERT_REGPIECE; |
| 12509 | |
| 12510 | DEBUG_PARSE("piec"); |
| 12511 | |
| 12512 | ret = regatom(pRExC_state, &flags, depth+1); |
| 12513 | if (ret == 0) { |
| 12514 | RETURN_FAIL_ON_RESTART_OR_FLAGS(flags, flagp, TRYAGAIN); |
| 12515 | FAIL2("panic: regatom returned failure, flags=%#" UVxf, (UV) flags); |
| 12516 | } |
| 12517 | |
| 12518 | op = *RExC_parse; |
| 12519 | |
| 12520 | if (op == '{' && regcurly(RExC_parse)) { |
| 12521 | maxpos = NULL; |
| 12522 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 12523 | parse_start = RExC_parse; /* MJD */ |
| 12524 | #endif |
| 12525 | next = RExC_parse + 1; |
| 12526 | while (isDIGIT(*next) || *next == ',') { |
| 12527 | if (*next == ',') { |
| 12528 | if (maxpos) |
| 12529 | break; |
| 12530 | else |
| 12531 | maxpos = next; |
| 12532 | } |
| 12533 | next++; |
| 12534 | } |
| 12535 | if (*next == '}') { /* got one */ |
| 12536 | const char* endptr; |
| 12537 | if (!maxpos) |
| 12538 | maxpos = next; |
| 12539 | RExC_parse++; |
| 12540 | if (isDIGIT(*RExC_parse)) { |
| 12541 | endptr = RExC_end; |
| 12542 | if (!grok_atoUV(RExC_parse, &uv, &endptr)) |
| 12543 | vFAIL("Invalid quantifier in {,}"); |
| 12544 | if (uv >= REG_INFTY) |
| 12545 | vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1); |
| 12546 | min = (I32)uv; |
| 12547 | } else { |
| 12548 | min = 0; |
| 12549 | } |
| 12550 | if (*maxpos == ',') |
| 12551 | maxpos++; |
| 12552 | else |
| 12553 | maxpos = RExC_parse; |
| 12554 | if (isDIGIT(*maxpos)) { |
| 12555 | endptr = RExC_end; |
| 12556 | if (!grok_atoUV(maxpos, &uv, &endptr)) |
| 12557 | vFAIL("Invalid quantifier in {,}"); |
| 12558 | if (uv >= REG_INFTY) |
| 12559 | vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1); |
| 12560 | max = (I32)uv; |
| 12561 | } else { |
| 12562 | max = REG_INFTY; /* meaning "infinity" */ |
| 12563 | } |
| 12564 | RExC_parse = next; |
| 12565 | nextchar(pRExC_state); |
| 12566 | if (max < min) { /* If can't match, warn and optimize to fail |
| 12567 | unconditionally */ |
| 12568 | reginsert(pRExC_state, OPFAIL, orig_emit, depth+1); |
| 12569 | ckWARNreg(RExC_parse, "Quantifier {n,m} with n > m can't match"); |
| 12570 | NEXT_OFF(REGNODE_p(orig_emit)) = |
| 12571 | regarglen[OPFAIL] + NODE_STEP_REGNODE; |
| 12572 | return ret; |
| 12573 | } |
| 12574 | else if (min == max && *RExC_parse == '?') |
| 12575 | { |
| 12576 | ckWARN2reg(RExC_parse + 1, |
| 12577 | "Useless use of greediness modifier '%c'", |
| 12578 | *RExC_parse); |
| 12579 | } |
| 12580 | |
| 12581 | do_curly: |
| 12582 | if ((flags&SIMPLE)) { |
| 12583 | if (min == 0 && max == REG_INFTY) { |
| 12584 | |
| 12585 | /* Going from 0..inf is currently forbidden in wildcard |
| 12586 | * subpatterns. The only reason is to make it harder to |
| 12587 | * write patterns that take a long long time to halt, and |
| 12588 | * because the use of this construct isn't necessary in |
| 12589 | * matching Unicode property values */ |
| 12590 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 12591 | RExC_parse++; |
| 12592 | /* diag_listed_as: Use of %s is not allowed in Unicode |
| 12593 | property wildcard subpatterns in regex; marked by |
| 12594 | <-- HERE in m/%s/ */ |
| 12595 | vFAIL("Use of quantifier '*' is not allowed in" |
| 12596 | " Unicode property wildcard subpatterns"); |
| 12597 | /* Note, don't need to worry about {0,}, as a '}' isn't |
| 12598 | * legal at all in wildcards, so wouldn't get this far |
| 12599 | * */ |
| 12600 | } |
| 12601 | reginsert(pRExC_state, STAR, ret, depth+1); |
| 12602 | MARK_NAUGHTY(4); |
| 12603 | RExC_seen |= REG_UNBOUNDED_QUANTIFIER_SEEN; |
| 12604 | goto nest_check; |
| 12605 | } |
| 12606 | if (min == 1 && max == REG_INFTY) { |
| 12607 | reginsert(pRExC_state, PLUS, ret, depth+1); |
| 12608 | MARK_NAUGHTY(3); |
| 12609 | RExC_seen |= REG_UNBOUNDED_QUANTIFIER_SEEN; |
| 12610 | goto nest_check; |
| 12611 | } |
| 12612 | MARK_NAUGHTY_EXP(2, 2); |
| 12613 | reginsert(pRExC_state, CURLY, ret, depth+1); |
| 12614 | Set_Node_Offset(REGNODE_p(ret), parse_start+1); /* MJD */ |
| 12615 | Set_Node_Cur_Length(REGNODE_p(ret), parse_start); |
| 12616 | } |
| 12617 | else { |
| 12618 | const regnode_offset w = reg_node(pRExC_state, WHILEM); |
| 12619 | |
| 12620 | FLAGS(REGNODE_p(w)) = 0; |
| 12621 | if (! REGTAIL(pRExC_state, ret, w)) { |
| 12622 | REQUIRE_BRANCHJ(flagp, 0); |
| 12623 | } |
| 12624 | if (RExC_use_BRANCHJ) { |
| 12625 | reginsert(pRExC_state, LONGJMP, ret, depth+1); |
| 12626 | reginsert(pRExC_state, NOTHING, ret, depth+1); |
| 12627 | NEXT_OFF(REGNODE_p(ret)) = 3; /* Go over LONGJMP. */ |
| 12628 | } |
| 12629 | reginsert(pRExC_state, CURLYX, ret, depth+1); |
| 12630 | /* MJD hk */ |
| 12631 | Set_Node_Offset(REGNODE_p(ret), parse_start+1); |
| 12632 | Set_Node_Length(REGNODE_p(ret), |
| 12633 | op == '{' ? (RExC_parse - parse_start) : 1); |
| 12634 | |
| 12635 | if (RExC_use_BRANCHJ) |
| 12636 | NEXT_OFF(REGNODE_p(ret)) = 3; /* Go over NOTHING to |
| 12637 | LONGJMP. */ |
| 12638 | if (! REGTAIL(pRExC_state, ret, reg_node(pRExC_state, |
| 12639 | NOTHING))) |
| 12640 | { |
| 12641 | REQUIRE_BRANCHJ(flagp, 0); |
| 12642 | } |
| 12643 | RExC_whilem_seen++; |
| 12644 | MARK_NAUGHTY_EXP(1, 4); /* compound interest */ |
| 12645 | } |
| 12646 | FLAGS(REGNODE_p(ret)) = 0; |
| 12647 | |
| 12648 | if (min > 0) |
| 12649 | *flagp = WORST; |
| 12650 | if (max > 0) |
| 12651 | *flagp |= HASWIDTH; |
| 12652 | ARG1_SET(REGNODE_p(ret), (U16)min); |
| 12653 | ARG2_SET(REGNODE_p(ret), (U16)max); |
| 12654 | if (max == REG_INFTY) |
| 12655 | RExC_seen |= REG_UNBOUNDED_QUANTIFIER_SEEN; |
| 12656 | |
| 12657 | goto nest_check; |
| 12658 | } |
| 12659 | } |
| 12660 | |
| 12661 | if (!ISMULT1(op)) { |
| 12662 | *flagp = flags; |
| 12663 | return(ret); |
| 12664 | } |
| 12665 | |
| 12666 | #if 0 /* Now runtime fix should be reliable. */ |
| 12667 | |
| 12668 | /* if this is reinstated, don't forget to put this back into perldiag: |
| 12669 | |
| 12670 | =item Regexp *+ operand could be empty at {#} in regex m/%s/ |
| 12671 | |
| 12672 | (F) The part of the regexp subject to either the * or + quantifier |
| 12673 | could match an empty string. The {#} shows in the regular |
| 12674 | expression about where the problem was discovered. |
| 12675 | |
| 12676 | */ |
| 12677 | |
| 12678 | if (!(flags&HASWIDTH) && op != '?') |
| 12679 | vFAIL("Regexp *+ operand could be empty"); |
| 12680 | #endif |
| 12681 | |
| 12682 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 12683 | parse_start = RExC_parse; |
| 12684 | #endif |
| 12685 | nextchar(pRExC_state); |
| 12686 | |
| 12687 | *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH); |
| 12688 | |
| 12689 | if (op == '*') { |
| 12690 | min = 0; |
| 12691 | goto do_curly; |
| 12692 | } |
| 12693 | else if (op == '+') { |
| 12694 | min = 1; |
| 12695 | goto do_curly; |
| 12696 | } |
| 12697 | else if (op == '?') { |
| 12698 | min = 0; max = 1; |
| 12699 | goto do_curly; |
| 12700 | } |
| 12701 | nest_check: |
| 12702 | if (!(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) { |
| 12703 | if (origparse[0] == '\\' && origparse[1] == 'K') { |
| 12704 | vFAIL2utf8f( |
| 12705 | "%" UTF8f " is forbidden - matches null string many times", |
| 12706 | UTF8fARG(UTF, (RExC_parse >= origparse |
| 12707 | ? RExC_parse - origparse |
| 12708 | : 0), |
| 12709 | origparse)); |
| 12710 | /* NOT-REACHED */ |
| 12711 | } else { |
| 12712 | ckWARN2reg(RExC_parse, |
| 12713 | "%" UTF8f " matches null string many times", |
| 12714 | UTF8fARG(UTF, (RExC_parse >= origparse |
| 12715 | ? RExC_parse - origparse |
| 12716 | : 0), |
| 12717 | origparse)); |
| 12718 | } |
| 12719 | } |
| 12720 | |
| 12721 | if (*RExC_parse == '?') { |
| 12722 | nextchar(pRExC_state); |
| 12723 | reginsert(pRExC_state, MINMOD, ret, depth+1); |
| 12724 | if (! REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE)) { |
| 12725 | REQUIRE_BRANCHJ(flagp, 0); |
| 12726 | } |
| 12727 | } |
| 12728 | else if (*RExC_parse == '+') { |
| 12729 | regnode_offset ender; |
| 12730 | nextchar(pRExC_state); |
| 12731 | ender = reg_node(pRExC_state, SUCCEED); |
| 12732 | if (! REGTAIL(pRExC_state, ret, ender)) { |
| 12733 | REQUIRE_BRANCHJ(flagp, 0); |
| 12734 | } |
| 12735 | reginsert(pRExC_state, SUSPEND, ret, depth+1); |
| 12736 | ender = reg_node(pRExC_state, TAIL); |
| 12737 | if (! REGTAIL(pRExC_state, ret, ender)) { |
| 12738 | REQUIRE_BRANCHJ(flagp, 0); |
| 12739 | } |
| 12740 | } |
| 12741 | |
| 12742 | if (ISMULT2(RExC_parse)) { |
| 12743 | RExC_parse++; |
| 12744 | vFAIL("Nested quantifiers"); |
| 12745 | } |
| 12746 | |
| 12747 | return(ret); |
| 12748 | } |
| 12749 | |
| 12750 | STATIC bool |
| 12751 | S_grok_bslash_N(pTHX_ RExC_state_t *pRExC_state, |
| 12752 | regnode_offset * node_p, |
| 12753 | UV * code_point_p, |
| 12754 | int * cp_count, |
| 12755 | I32 * flagp, |
| 12756 | const bool strict, |
| 12757 | const U32 depth |
| 12758 | ) |
| 12759 | { |
| 12760 | /* This routine teases apart the various meanings of \N and returns |
| 12761 | * accordingly. The input parameters constrain which meaning(s) is/are valid |
| 12762 | * in the current context. |
| 12763 | * |
| 12764 | * Exactly one of <node_p> and <code_point_p> must be non-NULL. |
| 12765 | * |
| 12766 | * If <code_point_p> is not NULL, the context is expecting the result to be a |
| 12767 | * single code point. If this \N instance turns out to a single code point, |
| 12768 | * the function returns TRUE and sets *code_point_p to that code point. |
| 12769 | * |
| 12770 | * If <node_p> is not NULL, the context is expecting the result to be one of |
| 12771 | * the things representable by a regnode. If this \N instance turns out to be |
| 12772 | * one such, the function generates the regnode, returns TRUE and sets *node_p |
| 12773 | * to point to the offset of that regnode into the regex engine program being |
| 12774 | * compiled. |
| 12775 | * |
| 12776 | * If this instance of \N isn't legal in any context, this function will |
| 12777 | * generate a fatal error and not return. |
| 12778 | * |
| 12779 | * On input, RExC_parse should point to the first char following the \N at the |
| 12780 | * time of the call. On successful return, RExC_parse will have been updated |
| 12781 | * to point to just after the sequence identified by this routine. Also |
| 12782 | * *flagp has been updated as needed. |
| 12783 | * |
| 12784 | * When there is some problem with the current context and this \N instance, |
| 12785 | * the function returns FALSE, without advancing RExC_parse, nor setting |
| 12786 | * *node_p, nor *code_point_p, nor *flagp. |
| 12787 | * |
| 12788 | * If <cp_count> is not NULL, the caller wants to know the length (in code |
| 12789 | * points) that this \N sequence matches. This is set, and the input is |
| 12790 | * parsed for errors, even if the function returns FALSE, as detailed below. |
| 12791 | * |
| 12792 | * There are 6 possibilities here, as detailed in the next 6 paragraphs. |
| 12793 | * |
| 12794 | * Probably the most common case is for the \N to specify a single code point. |
| 12795 | * *cp_count will be set to 1, and *code_point_p will be set to that code |
| 12796 | * point. |
| 12797 | * |
| 12798 | * Another possibility is for the input to be an empty \N{}. This is no |
| 12799 | * longer accepted, and will generate a fatal error. |
| 12800 | * |
| 12801 | * Another possibility is for a custom charnames handler to be in effect which |
| 12802 | * translates the input name to an empty string. *cp_count will be set to 0. |
| 12803 | * *node_p will be set to a generated NOTHING node. |
| 12804 | * |
| 12805 | * Still another possibility is for the \N to mean [^\n]. *cp_count will be |
| 12806 | * set to 0. *node_p will be set to a generated REG_ANY node. |
| 12807 | * |
| 12808 | * The fifth possibility is that \N resolves to a sequence of more than one |
| 12809 | * code points. *cp_count will be set to the number of code points in the |
| 12810 | * sequence. *node_p will be set to a generated node returned by this |
| 12811 | * function calling S_reg(). |
| 12812 | * |
| 12813 | * The final possibility is that it is premature to be calling this function; |
| 12814 | * the parse needs to be restarted. This can happen when this changes from |
| 12815 | * /d to /u rules, or when the pattern needs to be upgraded to UTF-8. The |
| 12816 | * latter occurs only when the fifth possibility would otherwise be in |
| 12817 | * effect, and is because one of those code points requires the pattern to be |
| 12818 | * recompiled as UTF-8. The function returns FALSE, and sets the |
| 12819 | * RESTART_PARSE and NEED_UTF8 flags in *flagp, as appropriate. When this |
| 12820 | * happens, the caller needs to desist from continuing parsing, and return |
| 12821 | * this information to its caller. This is not set for when there is only one |
| 12822 | * code point, as this can be called as part of an ANYOF node, and they can |
| 12823 | * store above-Latin1 code points without the pattern having to be in UTF-8. |
| 12824 | * |
| 12825 | * For non-single-quoted regexes, the tokenizer has resolved character and |
| 12826 | * sequence names inside \N{...} into their Unicode values, normalizing the |
| 12827 | * result into what we should see here: '\N{U+c1.c2...}', where c1... are the |
| 12828 | * hex-represented code points in the sequence. This is done there because |
| 12829 | * the names can vary based on what charnames pragma is in scope at the time, |
| 12830 | * so we need a way to take a snapshot of what they resolve to at the time of |
| 12831 | * the original parse. [perl #56444]. |
| 12832 | * |
| 12833 | * That parsing is skipped for single-quoted regexes, so here we may get |
| 12834 | * '\N{NAME}', which is parsed now. If the single-quoted regex is something |
| 12835 | * like '\N{U+41}', that code point is Unicode, and has to be translated into |
| 12836 | * the native character set for non-ASCII platforms. The other possibilities |
| 12837 | * are already native, so no translation is done. */ |
| 12838 | |
| 12839 | char * endbrace; /* points to '}' following the name */ |
| 12840 | char* p = RExC_parse; /* Temporary */ |
| 12841 | |
| 12842 | SV * substitute_parse = NULL; |
| 12843 | char *orig_end; |
| 12844 | char *save_start; |
| 12845 | I32 flags; |
| 12846 | |
| 12847 | GET_RE_DEBUG_FLAGS_DECL; |
| 12848 | |
| 12849 | PERL_ARGS_ASSERT_GROK_BSLASH_N; |
| 12850 | |
| 12851 | GET_RE_DEBUG_FLAGS; |
| 12852 | |
| 12853 | assert(cBOOL(node_p) ^ cBOOL(code_point_p)); /* Exactly one should be set */ |
| 12854 | assert(! (node_p && cp_count)); /* At most 1 should be set */ |
| 12855 | |
| 12856 | if (cp_count) { /* Initialize return for the most common case */ |
| 12857 | *cp_count = 1; |
| 12858 | } |
| 12859 | |
| 12860 | /* The [^\n] meaning of \N ignores spaces and comments under the /x |
| 12861 | * modifier. The other meanings do not, so use a temporary until we find |
| 12862 | * out which we are being called with */ |
| 12863 | skip_to_be_ignored_text(pRExC_state, &p, |
| 12864 | FALSE /* Don't force to /x */ ); |
| 12865 | |
| 12866 | /* Disambiguate between \N meaning a named character versus \N meaning |
| 12867 | * [^\n]. The latter is assumed when the {...} following the \N is a legal |
| 12868 | * quantifier, or if there is no '{' at all */ |
| 12869 | if (*p != '{' || regcurly(p)) { |
| 12870 | RExC_parse = p; |
| 12871 | if (cp_count) { |
| 12872 | *cp_count = -1; |
| 12873 | } |
| 12874 | |
| 12875 | if (! node_p) { |
| 12876 | return FALSE; |
| 12877 | } |
| 12878 | |
| 12879 | *node_p = reg_node(pRExC_state, REG_ANY); |
| 12880 | *flagp |= HASWIDTH|SIMPLE; |
| 12881 | MARK_NAUGHTY(1); |
| 12882 | Set_Node_Length(REGNODE_p(*(node_p)), 1); /* MJD */ |
| 12883 | return TRUE; |
| 12884 | } |
| 12885 | |
| 12886 | /* The test above made sure that the next real character is a '{', but |
| 12887 | * under the /x modifier, it could be separated by space (or a comment and |
| 12888 | * \n) and this is not allowed (for consistency with \x{...} and the |
| 12889 | * tokenizer handling of \N{NAME}). */ |
| 12890 | if (*RExC_parse != '{') { |
| 12891 | vFAIL("Missing braces on \\N{}"); |
| 12892 | } |
| 12893 | |
| 12894 | RExC_parse++; /* Skip past the '{' */ |
| 12895 | |
| 12896 | endbrace = (char *) memchr(RExC_parse, '}', RExC_end - RExC_parse); |
| 12897 | if (! endbrace) { /* no trailing brace */ |
| 12898 | vFAIL2("Missing right brace on \\%c{}", 'N'); |
| 12899 | } |
| 12900 | |
| 12901 | /* Here, we have decided it should be a named character or sequence. These |
| 12902 | * imply Unicode semantics */ |
| 12903 | REQUIRE_UNI_RULES(flagp, FALSE); |
| 12904 | |
| 12905 | /* \N{_} is what toke.c returns to us to indicate a name that evaluates to |
| 12906 | * nothing at all (not allowed under strict) */ |
| 12907 | if (endbrace - RExC_parse == 1 && *RExC_parse == '_') { |
| 12908 | RExC_parse = endbrace; |
| 12909 | if (strict) { |
| 12910 | RExC_parse++; /* Position after the "}" */ |
| 12911 | vFAIL("Zero length \\N{}"); |
| 12912 | } |
| 12913 | |
| 12914 | if (cp_count) { |
| 12915 | *cp_count = 0; |
| 12916 | } |
| 12917 | nextchar(pRExC_state); |
| 12918 | if (! node_p) { |
| 12919 | return FALSE; |
| 12920 | } |
| 12921 | |
| 12922 | *node_p = reg_node(pRExC_state, NOTHING); |
| 12923 | return TRUE; |
| 12924 | } |
| 12925 | |
| 12926 | if (endbrace - RExC_parse < 2 || ! strBEGINs(RExC_parse, "U+")) { |
| 12927 | |
| 12928 | /* Here, the name isn't of the form U+.... This can happen if the |
| 12929 | * pattern is single-quoted, so didn't get evaluated in toke.c. Now |
| 12930 | * is the time to find out what the name means */ |
| 12931 | |
| 12932 | const STRLEN name_len = endbrace - RExC_parse; |
| 12933 | SV * value_sv; /* What does this name evaluate to */ |
| 12934 | SV ** value_svp; |
| 12935 | const U8 * value; /* string of name's value */ |
| 12936 | STRLEN value_len; /* and its length */ |
| 12937 | |
| 12938 | /* RExC_unlexed_names is a hash of names that weren't evaluated by |
| 12939 | * toke.c, and their values. Make sure is initialized */ |
| 12940 | if (! RExC_unlexed_names) { |
| 12941 | RExC_unlexed_names = newHV(); |
| 12942 | } |
| 12943 | |
| 12944 | /* If we have already seen this name in this pattern, use that. This |
| 12945 | * allows us to only call the charnames handler once per name per |
| 12946 | * pattern. A broken or malicious handler could return something |
| 12947 | * different each time, which could cause the results to vary depending |
| 12948 | * on if something gets added or subtracted from the pattern that |
| 12949 | * causes the number of passes to change, for example */ |
| 12950 | if ((value_svp = hv_fetch(RExC_unlexed_names, RExC_parse, |
| 12951 | name_len, 0))) |
| 12952 | { |
| 12953 | value_sv = *value_svp; |
| 12954 | } |
| 12955 | else { /* Otherwise we have to go out and get the name */ |
| 12956 | const char * error_msg = NULL; |
| 12957 | value_sv = get_and_check_backslash_N_name(RExC_parse, endbrace, |
| 12958 | UTF, |
| 12959 | &error_msg); |
| 12960 | if (error_msg) { |
| 12961 | RExC_parse = endbrace; |
| 12962 | vFAIL(error_msg); |
| 12963 | } |
| 12964 | |
| 12965 | /* If no error message, should have gotten a valid return */ |
| 12966 | assert (value_sv); |
| 12967 | |
| 12968 | /* Save the name's meaning for later use */ |
| 12969 | if (! hv_store(RExC_unlexed_names, RExC_parse, name_len, |
| 12970 | value_sv, 0)) |
| 12971 | { |
| 12972 | Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed"); |
| 12973 | } |
| 12974 | } |
| 12975 | |
| 12976 | /* Here, we have the value the name evaluates to in 'value_sv' */ |
| 12977 | value = (U8 *) SvPV(value_sv, value_len); |
| 12978 | |
| 12979 | /* See if the result is one code point vs 0 or multiple */ |
| 12980 | if (inRANGE(value_len, 1, ((UV) SvUTF8(value_sv) |
| 12981 | ? UTF8SKIP(value) |
| 12982 | : 1))) |
| 12983 | { |
| 12984 | /* Here, exactly one code point. If that isn't what is wanted, |
| 12985 | * fail */ |
| 12986 | if (! code_point_p) { |
| 12987 | RExC_parse = p; |
| 12988 | return FALSE; |
| 12989 | } |
| 12990 | |
| 12991 | /* Convert from string to numeric code point */ |
| 12992 | *code_point_p = (SvUTF8(value_sv)) |
| 12993 | ? valid_utf8_to_uvchr(value, NULL) |
| 12994 | : *value; |
| 12995 | |
| 12996 | /* Have parsed this entire single code point \N{...}. *cp_count |
| 12997 | * has already been set to 1, so don't do it again. */ |
| 12998 | RExC_parse = endbrace; |
| 12999 | nextchar(pRExC_state); |
| 13000 | return TRUE; |
| 13001 | } /* End of is a single code point */ |
| 13002 | |
| 13003 | /* Count the code points, if caller desires. The API says to do this |
| 13004 | * even if we will later return FALSE */ |
| 13005 | if (cp_count) { |
| 13006 | *cp_count = 0; |
| 13007 | |
| 13008 | *cp_count = (SvUTF8(value_sv)) |
| 13009 | ? utf8_length(value, value + value_len) |
| 13010 | : value_len; |
| 13011 | } |
| 13012 | |
| 13013 | /* Fail if caller doesn't want to handle a multi-code-point sequence. |
| 13014 | * But don't back the pointer up if the caller wants to know how many |
| 13015 | * code points there are (they need to handle it themselves in this |
| 13016 | * case). */ |
| 13017 | if (! node_p) { |
| 13018 | if (! cp_count) { |
| 13019 | RExC_parse = p; |
| 13020 | } |
| 13021 | return FALSE; |
| 13022 | } |
| 13023 | |
| 13024 | /* Convert this to a sub-pattern of the form "(?: ... )", and then call |
| 13025 | * reg recursively to parse it. That way, it retains its atomicness, |
| 13026 | * while not having to worry about any special handling that some code |
| 13027 | * points may have. */ |
| 13028 | |
| 13029 | substitute_parse = newSVpvs("?:"); |
| 13030 | sv_catsv(substitute_parse, value_sv); |
| 13031 | sv_catpv(substitute_parse, ")"); |
| 13032 | |
| 13033 | /* The value should already be native, so no need to convert on EBCDIC |
| 13034 | * platforms.*/ |
| 13035 | assert(! RExC_recode_x_to_native); |
| 13036 | |
| 13037 | } |
| 13038 | else { /* \N{U+...} */ |
| 13039 | Size_t count = 0; /* code point count kept internally */ |
| 13040 | |
| 13041 | /* We can get to here when the input is \N{U+...} or when toke.c has |
| 13042 | * converted a name to the \N{U+...} form. This include changing a |
| 13043 | * name that evaluates to multiple code points to \N{U+c1.c2.c3 ...} */ |
| 13044 | |
| 13045 | RExC_parse += 2; /* Skip past the 'U+' */ |
| 13046 | |
| 13047 | /* Code points are separated by dots. The '}' terminates the whole |
| 13048 | * thing. */ |
| 13049 | |
| 13050 | do { /* Loop until the ending brace */ |
| 13051 | I32 flags = PERL_SCAN_SILENT_OVERFLOW |
| 13052 | | PERL_SCAN_SILENT_ILLDIGIT |
| 13053 | | PERL_SCAN_NOTIFY_ILLDIGIT |
| 13054 | | PERL_SCAN_ALLOW_MEDIAL_UNDERSCORES |
| 13055 | | PERL_SCAN_DISALLOW_PREFIX; |
| 13056 | STRLEN len = endbrace - RExC_parse; |
| 13057 | NV overflow_value; |
| 13058 | char * start_digit = RExC_parse; |
| 13059 | UV cp = grok_hex(RExC_parse, &len, &flags, &overflow_value); |
| 13060 | |
| 13061 | if (len == 0) { |
| 13062 | RExC_parse++; |
| 13063 | bad_NU: |
| 13064 | vFAIL("Invalid hexadecimal number in \\N{U+...}"); |
| 13065 | } |
| 13066 | |
| 13067 | RExC_parse += len; |
| 13068 | |
| 13069 | if (cp > MAX_LEGAL_CP) { |
| 13070 | vFAIL(form_cp_too_large_msg(16, start_digit, len, 0)); |
| 13071 | } |
| 13072 | |
| 13073 | if (RExC_parse >= endbrace) { /* Got to the closing '}' */ |
| 13074 | if (count) { |
| 13075 | goto do_concat; |
| 13076 | } |
| 13077 | |
| 13078 | /* Here, is a single code point; fail if doesn't want that */ |
| 13079 | if (! code_point_p) { |
| 13080 | RExC_parse = p; |
| 13081 | return FALSE; |
| 13082 | } |
| 13083 | |
| 13084 | /* A single code point is easy to handle; just return it */ |
| 13085 | *code_point_p = UNI_TO_NATIVE(cp); |
| 13086 | RExC_parse = endbrace; |
| 13087 | nextchar(pRExC_state); |
| 13088 | return TRUE; |
| 13089 | } |
| 13090 | |
| 13091 | /* Here, the parse stopped bfore the ending brace. This is legal |
| 13092 | * only if that character is a dot separating code points, like a |
| 13093 | * multiple character sequence (of the form "\N{U+c1.c2. ... }". |
| 13094 | * So the next character must be a dot (and the one after that |
| 13095 | * can't be the endbrace, or we'd have something like \N{U+100.} ) |
| 13096 | * */ |
| 13097 | if (*RExC_parse != '.' || RExC_parse + 1 >= endbrace) { |
| 13098 | RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */ |
| 13099 | ? UTF8SKIP(RExC_parse) |
| 13100 | : 1; |
| 13101 | RExC_parse = MIN(endbrace, RExC_parse);/* Guard against |
| 13102 | malformed utf8 */ |
| 13103 | goto bad_NU; |
| 13104 | } |
| 13105 | |
| 13106 | /* Here, looks like its really a multiple character sequence. Fail |
| 13107 | * if that's not what the caller wants. But continue with counting |
| 13108 | * and error checking if they still want a count */ |
| 13109 | if (! node_p && ! cp_count) { |
| 13110 | return FALSE; |
| 13111 | } |
| 13112 | |
| 13113 | /* What is done here is to convert this to a sub-pattern of the |
| 13114 | * form \x{char1}\x{char2}... and then call reg recursively to |
| 13115 | * parse it (enclosing in "(?: ... )" ). That way, it retains its |
| 13116 | * atomicness, while not having to worry about special handling |
| 13117 | * that some code points may have. We don't create a subpattern, |
| 13118 | * but go through the motions of code point counting and error |
| 13119 | * checking, if the caller doesn't want a node returned. */ |
| 13120 | |
| 13121 | if (node_p && ! substitute_parse) { |
| 13122 | substitute_parse = newSVpvs("?:"); |
| 13123 | } |
| 13124 | |
| 13125 | do_concat: |
| 13126 | |
| 13127 | if (node_p) { |
| 13128 | /* Convert to notation the rest of the code understands */ |
| 13129 | sv_catpvs(substitute_parse, "\\x{"); |
| 13130 | sv_catpvn(substitute_parse, start_digit, |
| 13131 | RExC_parse - start_digit); |
| 13132 | sv_catpvs(substitute_parse, "}"); |
| 13133 | } |
| 13134 | |
| 13135 | /* Move to after the dot (or ending brace the final time through.) |
| 13136 | * */ |
| 13137 | RExC_parse++; |
| 13138 | count++; |
| 13139 | |
| 13140 | } while (RExC_parse < endbrace); |
| 13141 | |
| 13142 | if (! node_p) { /* Doesn't want the node */ |
| 13143 | assert (cp_count); |
| 13144 | |
| 13145 | *cp_count = count; |
| 13146 | return FALSE; |
| 13147 | } |
| 13148 | |
| 13149 | sv_catpvs(substitute_parse, ")"); |
| 13150 | |
| 13151 | /* The values are Unicode, and therefore have to be converted to native |
| 13152 | * on a non-Unicode (meaning non-ASCII) platform. */ |
| 13153 | SET_recode_x_to_native(1); |
| 13154 | } |
| 13155 | |
| 13156 | /* Here, we have the string the name evaluates to, ready to be parsed, |
| 13157 | * stored in 'substitute_parse' as a series of valid "\x{...}\x{...}" |
| 13158 | * constructs. This can be called from within a substitute parse already. |
| 13159 | * The error reporting mechanism doesn't work for 2 levels of this, but the |
| 13160 | * code above has validated this new construct, so there should be no |
| 13161 | * errors generated by the below. And this isn' an exact copy, so the |
| 13162 | * mechanism to seamlessly deal with this won't work, so turn off warnings |
| 13163 | * during it */ |
| 13164 | save_start = RExC_start; |
| 13165 | orig_end = RExC_end; |
| 13166 | |
| 13167 | RExC_parse = RExC_start = SvPVX(substitute_parse); |
| 13168 | RExC_end = RExC_parse + SvCUR(substitute_parse); |
| 13169 | TURN_OFF_WARNINGS_IN_SUBSTITUTE_PARSE; |
| 13170 | |
| 13171 | *node_p = reg(pRExC_state, 1, &flags, depth+1); |
| 13172 | |
| 13173 | /* Restore the saved values */ |
| 13174 | RESTORE_WARNINGS; |
| 13175 | RExC_start = save_start; |
| 13176 | RExC_parse = endbrace; |
| 13177 | RExC_end = orig_end; |
| 13178 | SET_recode_x_to_native(0); |
| 13179 | |
| 13180 | SvREFCNT_dec_NN(substitute_parse); |
| 13181 | |
| 13182 | if (! *node_p) { |
| 13183 | RETURN_FAIL_ON_RESTART(flags, flagp); |
| 13184 | FAIL2("panic: reg returned failure to grok_bslash_N, flags=%#" UVxf, |
| 13185 | (UV) flags); |
| 13186 | } |
| 13187 | *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED); |
| 13188 | |
| 13189 | nextchar(pRExC_state); |
| 13190 | |
| 13191 | return TRUE; |
| 13192 | } |
| 13193 | |
| 13194 | |
| 13195 | PERL_STATIC_INLINE U8 |
| 13196 | S_compute_EXACTish(RExC_state_t *pRExC_state) |
| 13197 | { |
| 13198 | U8 op; |
| 13199 | |
| 13200 | PERL_ARGS_ASSERT_COMPUTE_EXACTISH; |
| 13201 | |
| 13202 | if (! FOLD) { |
| 13203 | return (LOC) |
| 13204 | ? EXACTL |
| 13205 | : EXACT; |
| 13206 | } |
| 13207 | |
| 13208 | op = get_regex_charset(RExC_flags); |
| 13209 | if (op >= REGEX_ASCII_RESTRICTED_CHARSET) { |
| 13210 | op--; /* /a is same as /u, and map /aa's offset to what /a's would have |
| 13211 | been, so there is no hole */ |
| 13212 | } |
| 13213 | |
| 13214 | return op + EXACTF; |
| 13215 | } |
| 13216 | |
| 13217 | STATIC bool |
| 13218 | S_new_regcurly(const char *s, const char *e) |
| 13219 | { |
| 13220 | /* This is a temporary function designed to match the most lenient form of |
| 13221 | * a {m,n} quantifier we ever envision, with either number omitted, and |
| 13222 | * spaces anywhere between/before/after them. |
| 13223 | * |
| 13224 | * If this function fails, then the string it matches is very unlikely to |
| 13225 | * ever be considered a valid quantifier, so we can allow the '{' that |
| 13226 | * begins it to be considered as a literal */ |
| 13227 | |
| 13228 | bool has_min = FALSE; |
| 13229 | bool has_max = FALSE; |
| 13230 | |
| 13231 | PERL_ARGS_ASSERT_NEW_REGCURLY; |
| 13232 | |
| 13233 | if (s >= e || *s++ != '{') |
| 13234 | return FALSE; |
| 13235 | |
| 13236 | while (s < e && isSPACE(*s)) { |
| 13237 | s++; |
| 13238 | } |
| 13239 | while (s < e && isDIGIT(*s)) { |
| 13240 | has_min = TRUE; |
| 13241 | s++; |
| 13242 | } |
| 13243 | while (s < e && isSPACE(*s)) { |
| 13244 | s++; |
| 13245 | } |
| 13246 | |
| 13247 | if (*s == ',') { |
| 13248 | s++; |
| 13249 | while (s < e && isSPACE(*s)) { |
| 13250 | s++; |
| 13251 | } |
| 13252 | while (s < e && isDIGIT(*s)) { |
| 13253 | has_max = TRUE; |
| 13254 | s++; |
| 13255 | } |
| 13256 | while (s < e && isSPACE(*s)) { |
| 13257 | s++; |
| 13258 | } |
| 13259 | } |
| 13260 | |
| 13261 | return s < e && *s == '}' && (has_min || has_max); |
| 13262 | } |
| 13263 | |
| 13264 | /* Parse backref decimal value, unless it's too big to sensibly be a backref, |
| 13265 | * in which case return I32_MAX (rather than possibly 32-bit wrapping) */ |
| 13266 | |
| 13267 | static I32 |
| 13268 | S_backref_value(char *p, char *e) |
| 13269 | { |
| 13270 | const char* endptr = e; |
| 13271 | UV val; |
| 13272 | if (grok_atoUV(p, &val, &endptr) && val <= I32_MAX) |
| 13273 | return (I32)val; |
| 13274 | return I32_MAX; |
| 13275 | } |
| 13276 | |
| 13277 | |
| 13278 | /* |
| 13279 | - regatom - the lowest level |
| 13280 | |
| 13281 | Try to identify anything special at the start of the current parse position. |
| 13282 | If there is, then handle it as required. This may involve generating a |
| 13283 | single regop, such as for an assertion; or it may involve recursing, such as |
| 13284 | to handle a () structure. |
| 13285 | |
| 13286 | If the string doesn't start with something special then we gobble up |
| 13287 | as much literal text as we can. If we encounter a quantifier, we have to |
| 13288 | back off the final literal character, as that quantifier applies to just it |
| 13289 | and not to the whole string of literals. |
| 13290 | |
| 13291 | Once we have been able to handle whatever type of thing started the |
| 13292 | sequence, we return the offset into the regex engine program being compiled |
| 13293 | at which any next regnode should be placed. |
| 13294 | |
| 13295 | Returns 0, setting *flagp to TRYAGAIN if reg() returns 0 with TRYAGAIN. |
| 13296 | Returns 0, setting *flagp to RESTART_PARSE if the parse needs to be |
| 13297 | restarted, or'd with NEED_UTF8 if the pattern needs to be upgraded to UTF-8 |
| 13298 | Otherwise does not return 0. |
| 13299 | |
| 13300 | Note: we have to be careful with escapes, as they can be both literal |
| 13301 | and special, and in the case of \10 and friends, context determines which. |
| 13302 | |
| 13303 | A summary of the code structure is: |
| 13304 | |
| 13305 | switch (first_byte) { |
| 13306 | cases for each special: |
| 13307 | handle this special; |
| 13308 | break; |
| 13309 | case '\\': |
| 13310 | switch (2nd byte) { |
| 13311 | cases for each unambiguous special: |
| 13312 | handle this special; |
| 13313 | break; |
| 13314 | cases for each ambigous special/literal: |
| 13315 | disambiguate; |
| 13316 | if (special) handle here |
| 13317 | else goto defchar; |
| 13318 | default: // unambiguously literal: |
| 13319 | goto defchar; |
| 13320 | } |
| 13321 | default: // is a literal char |
| 13322 | // FALL THROUGH |
| 13323 | defchar: |
| 13324 | create EXACTish node for literal; |
| 13325 | while (more input and node isn't full) { |
| 13326 | switch (input_byte) { |
| 13327 | cases for each special; |
| 13328 | make sure parse pointer is set so that the next call to |
| 13329 | regatom will see this special first |
| 13330 | goto loopdone; // EXACTish node terminated by prev. char |
| 13331 | default: |
| 13332 | append char to EXACTISH node; |
| 13333 | } |
| 13334 | get next input byte; |
| 13335 | } |
| 13336 | loopdone: |
| 13337 | } |
| 13338 | return the generated node; |
| 13339 | |
| 13340 | Specifically there are two separate switches for handling |
| 13341 | escape sequences, with the one for handling literal escapes requiring |
| 13342 | a dummy entry for all of the special escapes that are actually handled |
| 13343 | by the other. |
| 13344 | |
| 13345 | */ |
| 13346 | |
| 13347 | STATIC regnode_offset |
| 13348 | S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
| 13349 | { |
| 13350 | dVAR; |
| 13351 | regnode_offset ret = 0; |
| 13352 | I32 flags = 0; |
| 13353 | char *parse_start; |
| 13354 | U8 op; |
| 13355 | int invert = 0; |
| 13356 | |
| 13357 | GET_RE_DEBUG_FLAGS_DECL; |
| 13358 | |
| 13359 | *flagp = WORST; /* Tentatively. */ |
| 13360 | |
| 13361 | DEBUG_PARSE("atom"); |
| 13362 | |
| 13363 | PERL_ARGS_ASSERT_REGATOM; |
| 13364 | |
| 13365 | tryagain: |
| 13366 | parse_start = RExC_parse; |
| 13367 | assert(RExC_parse < RExC_end); |
| 13368 | switch ((U8)*RExC_parse) { |
| 13369 | case '^': |
| 13370 | RExC_seen_zerolen++; |
| 13371 | nextchar(pRExC_state); |
| 13372 | if (RExC_flags & RXf_PMf_MULTILINE) |
| 13373 | ret = reg_node(pRExC_state, MBOL); |
| 13374 | else |
| 13375 | ret = reg_node(pRExC_state, SBOL); |
| 13376 | Set_Node_Length(REGNODE_p(ret), 1); /* MJD */ |
| 13377 | break; |
| 13378 | case '$': |
| 13379 | nextchar(pRExC_state); |
| 13380 | if (*RExC_parse) |
| 13381 | RExC_seen_zerolen++; |
| 13382 | if (RExC_flags & RXf_PMf_MULTILINE) |
| 13383 | ret = reg_node(pRExC_state, MEOL); |
| 13384 | else |
| 13385 | ret = reg_node(pRExC_state, SEOL); |
| 13386 | Set_Node_Length(REGNODE_p(ret), 1); /* MJD */ |
| 13387 | break; |
| 13388 | case '.': |
| 13389 | nextchar(pRExC_state); |
| 13390 | if (RExC_flags & RXf_PMf_SINGLELINE) |
| 13391 | ret = reg_node(pRExC_state, SANY); |
| 13392 | else |
| 13393 | ret = reg_node(pRExC_state, REG_ANY); |
| 13394 | *flagp |= HASWIDTH|SIMPLE; |
| 13395 | MARK_NAUGHTY(1); |
| 13396 | Set_Node_Length(REGNODE_p(ret), 1); /* MJD */ |
| 13397 | break; |
| 13398 | case '[': |
| 13399 | { |
| 13400 | char * const oregcomp_parse = ++RExC_parse; |
| 13401 | ret = regclass(pRExC_state, flagp, depth+1, |
| 13402 | FALSE, /* means parse the whole char class */ |
| 13403 | TRUE, /* allow multi-char folds */ |
| 13404 | FALSE, /* don't silence non-portable warnings. */ |
| 13405 | (bool) RExC_strict, |
| 13406 | TRUE, /* Allow an optimized regnode result */ |
| 13407 | NULL); |
| 13408 | if (ret == 0) { |
| 13409 | RETURN_FAIL_ON_RESTART_FLAGP(flagp); |
| 13410 | FAIL2("panic: regclass returned failure to regatom, flags=%#" UVxf, |
| 13411 | (UV) *flagp); |
| 13412 | } |
| 13413 | if (*RExC_parse != ']') { |
| 13414 | RExC_parse = oregcomp_parse; |
| 13415 | vFAIL("Unmatched ["); |
| 13416 | } |
| 13417 | nextchar(pRExC_state); |
| 13418 | Set_Node_Length(REGNODE_p(ret), RExC_parse - oregcomp_parse + 1); /* MJD */ |
| 13419 | break; |
| 13420 | } |
| 13421 | case '(': |
| 13422 | nextchar(pRExC_state); |
| 13423 | ret = reg(pRExC_state, 2, &flags, depth+1); |
| 13424 | if (ret == 0) { |
| 13425 | if (flags & TRYAGAIN) { |
| 13426 | if (RExC_parse >= RExC_end) { |
| 13427 | /* Make parent create an empty node if needed. */ |
| 13428 | *flagp |= TRYAGAIN; |
| 13429 | return(0); |
| 13430 | } |
| 13431 | goto tryagain; |
| 13432 | } |
| 13433 | RETURN_FAIL_ON_RESTART(flags, flagp); |
| 13434 | FAIL2("panic: reg returned failure to regatom, flags=%#" UVxf, |
| 13435 | (UV) flags); |
| 13436 | } |
| 13437 | *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED); |
| 13438 | break; |
| 13439 | case '|': |
| 13440 | case ')': |
| 13441 | if (flags & TRYAGAIN) { |
| 13442 | *flagp |= TRYAGAIN; |
| 13443 | return 0; |
| 13444 | } |
| 13445 | vFAIL("Internal urp"); |
| 13446 | /* Supposed to be caught earlier. */ |
| 13447 | break; |
| 13448 | case '?': |
| 13449 | case '+': |
| 13450 | case '*': |
| 13451 | RExC_parse++; |
| 13452 | vFAIL("Quantifier follows nothing"); |
| 13453 | break; |
| 13454 | case '\\': |
| 13455 | /* Special Escapes |
| 13456 | |
| 13457 | This switch handles escape sequences that resolve to some kind |
| 13458 | of special regop and not to literal text. Escape sequences that |
| 13459 | resolve to literal text are handled below in the switch marked |
| 13460 | "Literal Escapes". |
| 13461 | |
| 13462 | Every entry in this switch *must* have a corresponding entry |
| 13463 | in the literal escape switch. However, the opposite is not |
| 13464 | required, as the default for this switch is to jump to the |
| 13465 | literal text handling code. |
| 13466 | */ |
| 13467 | RExC_parse++; |
| 13468 | switch ((U8)*RExC_parse) { |
| 13469 | /* Special Escapes */ |
| 13470 | case 'A': |
| 13471 | RExC_seen_zerolen++; |
| 13472 | /* Under wildcards, this is changed to match \n; should be |
| 13473 | * invisible to the user, as they have to compile under /m */ |
| 13474 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 13475 | ret = reg_node(pRExC_state, MBOL); |
| 13476 | } |
| 13477 | else { |
| 13478 | ret = reg_node(pRExC_state, SBOL); |
| 13479 | /* SBOL is shared with /^/ so we set the flags so we can tell |
| 13480 | * /\A/ from /^/ in split. */ |
| 13481 | FLAGS(REGNODE_p(ret)) = 1; |
| 13482 | } |
| 13483 | *flagp |= SIMPLE; |
| 13484 | goto finish_meta_pat; |
| 13485 | case 'G': |
| 13486 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 13487 | RExC_parse++; |
| 13488 | /* diag_listed_as: Use of %s is not allowed in Unicode property |
| 13489 | wildcard subpatterns in regex; marked by <-- HERE in m/%s/ |
| 13490 | */ |
| 13491 | vFAIL("Use of '\\G' is not allowed in Unicode property" |
| 13492 | " wildcard subpatterns"); |
| 13493 | } |
| 13494 | ret = reg_node(pRExC_state, GPOS); |
| 13495 | RExC_seen |= REG_GPOS_SEEN; |
| 13496 | *flagp |= SIMPLE; |
| 13497 | goto finish_meta_pat; |
| 13498 | case 'K': |
| 13499 | if (!RExC_in_lookbehind && !RExC_in_lookahead) { |
| 13500 | RExC_seen_zerolen++; |
| 13501 | ret = reg_node(pRExC_state, KEEPS); |
| 13502 | *flagp |= SIMPLE; |
| 13503 | /* XXX:dmq : disabling in-place substitution seems to |
| 13504 | * be necessary here to avoid cases of memory corruption, as |
| 13505 | * with: C<$_="x" x 80; s/x\K/y/> -- rgs |
| 13506 | */ |
| 13507 | RExC_seen |= REG_LOOKBEHIND_SEEN; |
| 13508 | goto finish_meta_pat; |
| 13509 | } |
| 13510 | else { |
| 13511 | ++RExC_parse; /* advance past the 'K' */ |
| 13512 | vFAIL("\\K not permitted in lookahead/lookbehind"); |
| 13513 | } |
| 13514 | case 'Z': |
| 13515 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 13516 | /* See comment under \A above */ |
| 13517 | ret = reg_node(pRExC_state, MEOL); |
| 13518 | } |
| 13519 | else { |
| 13520 | ret = reg_node(pRExC_state, SEOL); |
| 13521 | } |
| 13522 | *flagp |= SIMPLE; |
| 13523 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
| 13524 | goto finish_meta_pat; |
| 13525 | case 'z': |
| 13526 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 13527 | /* See comment under \A above */ |
| 13528 | ret = reg_node(pRExC_state, MEOL); |
| 13529 | } |
| 13530 | else { |
| 13531 | ret = reg_node(pRExC_state, EOS); |
| 13532 | } |
| 13533 | *flagp |= SIMPLE; |
| 13534 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
| 13535 | goto finish_meta_pat; |
| 13536 | case 'C': |
| 13537 | vFAIL("\\C no longer supported"); |
| 13538 | case 'X': |
| 13539 | ret = reg_node(pRExC_state, CLUMP); |
| 13540 | *flagp |= HASWIDTH; |
| 13541 | goto finish_meta_pat; |
| 13542 | |
| 13543 | case 'B': |
| 13544 | invert = 1; |
| 13545 | /* FALLTHROUGH */ |
| 13546 | case 'b': |
| 13547 | { |
| 13548 | U8 flags = 0; |
| 13549 | regex_charset charset = get_regex_charset(RExC_flags); |
| 13550 | |
| 13551 | RExC_seen_zerolen++; |
| 13552 | RExC_seen |= REG_LOOKBEHIND_SEEN; |
| 13553 | op = BOUND + charset; |
| 13554 | |
| 13555 | if (RExC_parse >= RExC_end || *(RExC_parse + 1) != '{') { |
| 13556 | flags = TRADITIONAL_BOUND; |
| 13557 | if (op > BOUNDA) { /* /aa is same as /a */ |
| 13558 | op = BOUNDA; |
| 13559 | } |
| 13560 | } |
| 13561 | else { |
| 13562 | STRLEN length; |
| 13563 | char name = *RExC_parse; |
| 13564 | char * endbrace = NULL; |
| 13565 | RExC_parse += 2; |
| 13566 | endbrace = (char *) memchr(RExC_parse, '}', RExC_end - RExC_parse); |
| 13567 | |
| 13568 | if (! endbrace) { |
| 13569 | vFAIL2("Missing right brace on \\%c{}", name); |
| 13570 | } |
| 13571 | /* XXX Need to decide whether to take spaces or not. Should be |
| 13572 | * consistent with \p{}, but that currently is SPACE, which |
| 13573 | * means vertical too, which seems wrong |
| 13574 | * while (isBLANK(*RExC_parse)) { |
| 13575 | RExC_parse++; |
| 13576 | }*/ |
| 13577 | if (endbrace == RExC_parse) { |
| 13578 | RExC_parse++; /* After the '}' */ |
| 13579 | vFAIL2("Empty \\%c{}", name); |
| 13580 | } |
| 13581 | length = endbrace - RExC_parse; |
| 13582 | /*while (isBLANK(*(RExC_parse + length - 1))) { |
| 13583 | length--; |
| 13584 | }*/ |
| 13585 | switch (*RExC_parse) { |
| 13586 | case 'g': |
| 13587 | if ( length != 1 |
| 13588 | && (memNEs(RExC_parse + 1, length - 1, "cb"))) |
| 13589 | { |
| 13590 | goto bad_bound_type; |
| 13591 | } |
| 13592 | flags = GCB_BOUND; |
| 13593 | break; |
| 13594 | case 'l': |
| 13595 | if (length != 2 || *(RExC_parse + 1) != 'b') { |
| 13596 | goto bad_bound_type; |
| 13597 | } |
| 13598 | flags = LB_BOUND; |
| 13599 | break; |
| 13600 | case 's': |
| 13601 | if (length != 2 || *(RExC_parse + 1) != 'b') { |
| 13602 | goto bad_bound_type; |
| 13603 | } |
| 13604 | flags = SB_BOUND; |
| 13605 | break; |
| 13606 | case 'w': |
| 13607 | if (length != 2 || *(RExC_parse + 1) != 'b') { |
| 13608 | goto bad_bound_type; |
| 13609 | } |
| 13610 | flags = WB_BOUND; |
| 13611 | break; |
| 13612 | default: |
| 13613 | bad_bound_type: |
| 13614 | RExC_parse = endbrace; |
| 13615 | vFAIL2utf8f( |
| 13616 | "'%" UTF8f "' is an unknown bound type", |
| 13617 | UTF8fARG(UTF, length, endbrace - length)); |
| 13618 | NOT_REACHED; /*NOTREACHED*/ |
| 13619 | } |
| 13620 | RExC_parse = endbrace; |
| 13621 | REQUIRE_UNI_RULES(flagp, 0); |
| 13622 | |
| 13623 | if (op == BOUND) { |
| 13624 | op = BOUNDU; |
| 13625 | } |
| 13626 | else if (op >= BOUNDA) { /* /aa is same as /a */ |
| 13627 | op = BOUNDU; |
| 13628 | length += 4; |
| 13629 | |
| 13630 | /* Don't have to worry about UTF-8, in this message because |
| 13631 | * to get here the contents of the \b must be ASCII */ |
| 13632 | ckWARN4reg(RExC_parse + 1, /* Include the '}' in msg */ |
| 13633 | "Using /u for '%.*s' instead of /%s", |
| 13634 | (unsigned) length, |
| 13635 | endbrace - length + 1, |
| 13636 | (charset == REGEX_ASCII_RESTRICTED_CHARSET) |
| 13637 | ? ASCII_RESTRICT_PAT_MODS |
| 13638 | : ASCII_MORE_RESTRICT_PAT_MODS); |
| 13639 | } |
| 13640 | } |
| 13641 | |
| 13642 | if (op == BOUND) { |
| 13643 | RExC_seen_d_op = TRUE; |
| 13644 | } |
| 13645 | else if (op == BOUNDL) { |
| 13646 | RExC_contains_locale = 1; |
| 13647 | } |
| 13648 | |
| 13649 | if (invert) { |
| 13650 | op += NBOUND - BOUND; |
| 13651 | } |
| 13652 | |
| 13653 | ret = reg_node(pRExC_state, op); |
| 13654 | FLAGS(REGNODE_p(ret)) = flags; |
| 13655 | |
| 13656 | *flagp |= SIMPLE; |
| 13657 | |
| 13658 | goto finish_meta_pat; |
| 13659 | } |
| 13660 | |
| 13661 | case 'R': |
| 13662 | ret = reg_node(pRExC_state, LNBREAK); |
| 13663 | *flagp |= HASWIDTH|SIMPLE; |
| 13664 | goto finish_meta_pat; |
| 13665 | |
| 13666 | case 'd': |
| 13667 | case 'D': |
| 13668 | case 'h': |
| 13669 | case 'H': |
| 13670 | case 'p': |
| 13671 | case 'P': |
| 13672 | case 's': |
| 13673 | case 'S': |
| 13674 | case 'v': |
| 13675 | case 'V': |
| 13676 | case 'w': |
| 13677 | case 'W': |
| 13678 | /* These all have the same meaning inside [brackets], and it knows |
| 13679 | * how to do the best optimizations for them. So, pretend we found |
| 13680 | * these within brackets, and let it do the work */ |
| 13681 | RExC_parse--; |
| 13682 | |
| 13683 | ret = regclass(pRExC_state, flagp, depth+1, |
| 13684 | TRUE, /* means just parse this element */ |
| 13685 | FALSE, /* don't allow multi-char folds */ |
| 13686 | FALSE, /* don't silence non-portable warnings. It |
| 13687 | would be a bug if these returned |
| 13688 | non-portables */ |
| 13689 | (bool) RExC_strict, |
| 13690 | TRUE, /* Allow an optimized regnode result */ |
| 13691 | NULL); |
| 13692 | RETURN_FAIL_ON_RESTART_FLAGP(flagp); |
| 13693 | /* regclass() can only return RESTART_PARSE and NEED_UTF8 if |
| 13694 | * multi-char folds are allowed. */ |
| 13695 | if (!ret) |
| 13696 | FAIL2("panic: regclass returned failure to regatom, flags=%#" UVxf, |
| 13697 | (UV) *flagp); |
| 13698 | |
| 13699 | RExC_parse--; /* regclass() leaves this one too far ahead */ |
| 13700 | |
| 13701 | finish_meta_pat: |
| 13702 | /* The escapes above that don't take a parameter can't be |
| 13703 | * followed by a '{'. But 'pX', 'p{foo}' and |
| 13704 | * correspondingly 'P' can be */ |
| 13705 | if ( RExC_parse - parse_start == 1 |
| 13706 | && UCHARAT(RExC_parse + 1) == '{' |
| 13707 | && UNLIKELY(! new_regcurly(RExC_parse + 1, RExC_end))) |
| 13708 | { |
| 13709 | RExC_parse += 2; |
| 13710 | vFAIL("Unescaped left brace in regex is illegal here"); |
| 13711 | } |
| 13712 | Set_Node_Offset(REGNODE_p(ret), parse_start); |
| 13713 | Set_Node_Length(REGNODE_p(ret), RExC_parse - parse_start + 1); /* MJD */ |
| 13714 | nextchar(pRExC_state); |
| 13715 | break; |
| 13716 | case 'N': |
| 13717 | /* Handle \N, \N{} and \N{NAMED SEQUENCE} (the latter meaning the |
| 13718 | * \N{...} evaluates to a sequence of more than one code points). |
| 13719 | * The function call below returns a regnode, which is our result. |
| 13720 | * The parameters cause it to fail if the \N{} evaluates to a |
| 13721 | * single code point; we handle those like any other literal. The |
| 13722 | * reason that the multicharacter case is handled here and not as |
| 13723 | * part of the EXACtish code is because of quantifiers. In |
| 13724 | * /\N{BLAH}+/, the '+' applies to the whole thing, and doing it |
| 13725 | * this way makes that Just Happen. dmq. |
| 13726 | * join_exact() will join this up with adjacent EXACTish nodes |
| 13727 | * later on, if appropriate. */ |
| 13728 | ++RExC_parse; |
| 13729 | if (grok_bslash_N(pRExC_state, |
| 13730 | &ret, /* Want a regnode returned */ |
| 13731 | NULL, /* Fail if evaluates to a single code |
| 13732 | point */ |
| 13733 | NULL, /* Don't need a count of how many code |
| 13734 | points */ |
| 13735 | flagp, |
| 13736 | RExC_strict, |
| 13737 | depth) |
| 13738 | ) { |
| 13739 | break; |
| 13740 | } |
| 13741 | |
| 13742 | RETURN_FAIL_ON_RESTART_FLAGP(flagp); |
| 13743 | |
| 13744 | /* Here, evaluates to a single code point. Go get that */ |
| 13745 | RExC_parse = parse_start; |
| 13746 | goto defchar; |
| 13747 | |
| 13748 | case 'k': /* Handle \k<NAME> and \k'NAME' */ |
| 13749 | parse_named_seq: |
| 13750 | { |
| 13751 | char ch; |
| 13752 | if ( RExC_parse >= RExC_end - 1 |
| 13753 | || (( ch = RExC_parse[1]) != '<' |
| 13754 | && ch != '\'' |
| 13755 | && ch != '{')) |
| 13756 | { |
| 13757 | RExC_parse++; |
| 13758 | /* diag_listed_as: Sequence \%s... not terminated in regex; marked by <-- HERE in m/%s/ */ |
| 13759 | vFAIL2("Sequence %.2s... not terminated", parse_start); |
| 13760 | } else { |
| 13761 | RExC_parse += 2; |
| 13762 | ret = handle_named_backref(pRExC_state, |
| 13763 | flagp, |
| 13764 | parse_start, |
| 13765 | (ch == '<') |
| 13766 | ? '>' |
| 13767 | : (ch == '{') |
| 13768 | ? '}' |
| 13769 | : '\''); |
| 13770 | } |
| 13771 | break; |
| 13772 | } |
| 13773 | case 'g': |
| 13774 | case '1': case '2': case '3': case '4': |
| 13775 | case '5': case '6': case '7': case '8': case '9': |
| 13776 | { |
| 13777 | I32 num; |
| 13778 | bool hasbrace = 0; |
| 13779 | |
| 13780 | if (*RExC_parse == 'g') { |
| 13781 | bool isrel = 0; |
| 13782 | |
| 13783 | RExC_parse++; |
| 13784 | if (*RExC_parse == '{') { |
| 13785 | RExC_parse++; |
| 13786 | hasbrace = 1; |
| 13787 | } |
| 13788 | if (*RExC_parse == '-') { |
| 13789 | RExC_parse++; |
| 13790 | isrel = 1; |
| 13791 | } |
| 13792 | if (hasbrace && !isDIGIT(*RExC_parse)) { |
| 13793 | if (isrel) RExC_parse--; |
| 13794 | RExC_parse -= 2; |
| 13795 | goto parse_named_seq; |
| 13796 | } |
| 13797 | |
| 13798 | if (RExC_parse >= RExC_end) { |
| 13799 | goto unterminated_g; |
| 13800 | } |
| 13801 | num = S_backref_value(RExC_parse, RExC_end); |
| 13802 | if (num == 0) |
| 13803 | vFAIL("Reference to invalid group 0"); |
| 13804 | else if (num == I32_MAX) { |
| 13805 | if (isDIGIT(*RExC_parse)) |
| 13806 | vFAIL("Reference to nonexistent group"); |
| 13807 | else |
| 13808 | unterminated_g: |
| 13809 | vFAIL("Unterminated \\g... pattern"); |
| 13810 | } |
| 13811 | |
| 13812 | if (isrel) { |
| 13813 | num = RExC_npar - num; |
| 13814 | if (num < 1) |
| 13815 | vFAIL("Reference to nonexistent or unclosed group"); |
| 13816 | } |
| 13817 | } |
| 13818 | else { |
| 13819 | num = S_backref_value(RExC_parse, RExC_end); |
| 13820 | /* bare \NNN might be backref or octal - if it is larger |
| 13821 | * than or equal RExC_npar then it is assumed to be an |
| 13822 | * octal escape. Note RExC_npar is +1 from the actual |
| 13823 | * number of parens. */ |
| 13824 | /* Note we do NOT check if num == I32_MAX here, as that is |
| 13825 | * handled by the RExC_npar check */ |
| 13826 | |
| 13827 | if ( |
| 13828 | /* any numeric escape < 10 is always a backref */ |
| 13829 | num > 9 |
| 13830 | /* any numeric escape < RExC_npar is a backref */ |
| 13831 | && num >= RExC_npar |
| 13832 | /* cannot be an octal escape if it starts with 8 */ |
| 13833 | && *RExC_parse != '8' |
| 13834 | /* cannot be an octal escape if it starts with 9 */ |
| 13835 | && *RExC_parse != '9' |
| 13836 | ) { |
| 13837 | /* Probably not meant to be a backref, instead likely |
| 13838 | * to be an octal character escape, e.g. \35 or \777. |
| 13839 | * The above logic should make it obvious why using |
| 13840 | * octal escapes in patterns is problematic. - Yves */ |
| 13841 | RExC_parse = parse_start; |
| 13842 | goto defchar; |
| 13843 | } |
| 13844 | } |
| 13845 | |
| 13846 | /* At this point RExC_parse points at a numeric escape like |
| 13847 | * \12 or \88 or something similar, which we should NOT treat |
| 13848 | * as an octal escape. It may or may not be a valid backref |
| 13849 | * escape. For instance \88888888 is unlikely to be a valid |
| 13850 | * backref. */ |
| 13851 | while (isDIGIT(*RExC_parse)) |
| 13852 | RExC_parse++; |
| 13853 | if (hasbrace) { |
| 13854 | if (*RExC_parse != '}') |
| 13855 | vFAIL("Unterminated \\g{...} pattern"); |
| 13856 | RExC_parse++; |
| 13857 | } |
| 13858 | if (num >= (I32)RExC_npar) { |
| 13859 | |
| 13860 | /* It might be a forward reference; we can't fail until we |
| 13861 | * know, by completing the parse to get all the groups, and |
| 13862 | * then reparsing */ |
| 13863 | if (ALL_PARENS_COUNTED) { |
| 13864 | if (num >= RExC_total_parens) { |
| 13865 | vFAIL("Reference to nonexistent group"); |
| 13866 | } |
| 13867 | } |
| 13868 | else { |
| 13869 | REQUIRE_PARENS_PASS; |
| 13870 | } |
| 13871 | } |
| 13872 | RExC_sawback = 1; |
| 13873 | ret = reganode(pRExC_state, |
| 13874 | ((! FOLD) |
| 13875 | ? REF |
| 13876 | : (ASCII_FOLD_RESTRICTED) |
| 13877 | ? REFFA |
| 13878 | : (AT_LEAST_UNI_SEMANTICS) |
| 13879 | ? REFFU |
| 13880 | : (LOC) |
| 13881 | ? REFFL |
| 13882 | : REFF), |
| 13883 | num); |
| 13884 | if (OP(REGNODE_p(ret)) == REFF) { |
| 13885 | RExC_seen_d_op = TRUE; |
| 13886 | } |
| 13887 | *flagp |= HASWIDTH; |
| 13888 | |
| 13889 | /* override incorrect value set in reganode MJD */ |
| 13890 | Set_Node_Offset(REGNODE_p(ret), parse_start); |
| 13891 | Set_Node_Cur_Length(REGNODE_p(ret), parse_start-1); |
| 13892 | skip_to_be_ignored_text(pRExC_state, &RExC_parse, |
| 13893 | FALSE /* Don't force to /x */ ); |
| 13894 | } |
| 13895 | break; |
| 13896 | case '\0': |
| 13897 | if (RExC_parse >= RExC_end) |
| 13898 | FAIL("Trailing \\"); |
| 13899 | /* FALLTHROUGH */ |
| 13900 | default: |
| 13901 | /* Do not generate "unrecognized" warnings here, we fall |
| 13902 | back into the quick-grab loop below */ |
| 13903 | RExC_parse = parse_start; |
| 13904 | goto defchar; |
| 13905 | } /* end of switch on a \foo sequence */ |
| 13906 | break; |
| 13907 | |
| 13908 | case '#': |
| 13909 | |
| 13910 | /* '#' comments should have been spaced over before this function was |
| 13911 | * called */ |
| 13912 | assert((RExC_flags & RXf_PMf_EXTENDED) == 0); |
| 13913 | /* |
| 13914 | if (RExC_flags & RXf_PMf_EXTENDED) { |
| 13915 | RExC_parse = reg_skipcomment( pRExC_state, RExC_parse ); |
| 13916 | if (RExC_parse < RExC_end) |
| 13917 | goto tryagain; |
| 13918 | } |
| 13919 | */ |
| 13920 | |
| 13921 | /* FALLTHROUGH */ |
| 13922 | |
| 13923 | default: |
| 13924 | defchar: { |
| 13925 | |
| 13926 | /* Here, we have determined that the next thing is probably a |
| 13927 | * literal character. RExC_parse points to the first byte of its |
| 13928 | * definition. (It still may be an escape sequence that evaluates |
| 13929 | * to a single character) */ |
| 13930 | |
| 13931 | STRLEN len = 0; |
| 13932 | UV ender = 0; |
| 13933 | char *p; |
| 13934 | char *s, *old_s = NULL, *old_old_s = NULL; |
| 13935 | char *s0; |
| 13936 | U32 max_string_len = 255; |
| 13937 | |
| 13938 | /* We may have to reparse the node, artificially stopping filling |
| 13939 | * it early, based on info gleaned in the first parse. This |
| 13940 | * variable gives where we stop. Make it above the normal stopping |
| 13941 | * place first time through; otherwise it would stop too early */ |
| 13942 | U32 upper_fill = max_string_len + 1; |
| 13943 | |
| 13944 | /* We start out as an EXACT node, even if under /i, until we find a |
| 13945 | * character which is in a fold. The algorithm now segregates into |
| 13946 | * separate nodes, characters that fold from those that don't under |
| 13947 | * /i. (This hopefully will create nodes that are fixed strings |
| 13948 | * even under /i, giving the optimizer something to grab on to.) |
| 13949 | * So, if a node has something in it and the next character is in |
| 13950 | * the opposite category, that node is closed up, and the function |
| 13951 | * returns. Then regatom is called again, and a new node is |
| 13952 | * created for the new category. */ |
| 13953 | U8 node_type = EXACT; |
| 13954 | |
| 13955 | /* Assume the node will be fully used; the excess is given back at |
| 13956 | * the end. Under /i, we may need to temporarily add the fold of |
| 13957 | * an extra character or two at the end to check for splitting |
| 13958 | * multi-char folds, so allocate extra space for that. We can't |
| 13959 | * make any other length assumptions, as a byte input sequence |
| 13960 | * could shrink down. */ |
| 13961 | Ptrdiff_t current_string_nodes = STR_SZ(max_string_len |
| 13962 | + ((! FOLD) |
| 13963 | ? 0 |
| 13964 | : 2 * ((UTF) |
| 13965 | ? UTF8_MAXBYTES_CASE |
| 13966 | /* Max non-UTF-8 expansion is 2 */ : 2))); |
| 13967 | |
| 13968 | bool next_is_quantifier; |
| 13969 | char * oldp = NULL; |
| 13970 | |
| 13971 | /* We can convert EXACTF nodes to EXACTFU if they contain only |
| 13972 | * characters that match identically regardless of the target |
| 13973 | * string's UTF8ness. The reason to do this is that EXACTF is not |
| 13974 | * trie-able, EXACTFU is, and EXACTFU requires fewer operations at |
| 13975 | * runtime. |
| 13976 | * |
| 13977 | * Similarly, we can convert EXACTFL nodes to EXACTFLU8 if they |
| 13978 | * contain only above-Latin1 characters (hence must be in UTF8), |
| 13979 | * which don't participate in folds with Latin1-range characters, |
| 13980 | * as the latter's folds aren't known until runtime. */ |
| 13981 | bool maybe_exactfu = FOLD && (DEPENDS_SEMANTICS || LOC); |
| 13982 | |
| 13983 | /* Single-character EXACTish nodes are almost always SIMPLE. This |
| 13984 | * allows us to override this as encountered */ |
| 13985 | U8 maybe_SIMPLE = SIMPLE; |
| 13986 | |
| 13987 | /* Does this node contain something that can't match unless the |
| 13988 | * target string is (also) in UTF-8 */ |
| 13989 | bool requires_utf8_target = FALSE; |
| 13990 | |
| 13991 | /* The sequence 'ss' is problematic in non-UTF-8 patterns. */ |
| 13992 | bool has_ss = FALSE; |
| 13993 | |
| 13994 | /* So is the MICRO SIGN */ |
| 13995 | bool has_micro_sign = FALSE; |
| 13996 | |
| 13997 | /* Set when we fill up the current node and there is still more |
| 13998 | * text to process */ |
| 13999 | bool overflowed; |
| 14000 | |
| 14001 | /* Allocate an EXACT node. The node_type may change below to |
| 14002 | * another EXACTish node, but since the size of the node doesn't |
| 14003 | * change, it works */ |
| 14004 | ret = regnode_guts(pRExC_state, node_type, current_string_nodes, |
| 14005 | "exact"); |
| 14006 | FILL_NODE(ret, node_type); |
| 14007 | RExC_emit++; |
| 14008 | |
| 14009 | s = STRING(REGNODE_p(ret)); |
| 14010 | |
| 14011 | s0 = s; |
| 14012 | |
| 14013 | reparse: |
| 14014 | |
| 14015 | p = RExC_parse; |
| 14016 | len = 0; |
| 14017 | s = s0; |
| 14018 | node_type = EXACT; |
| 14019 | oldp = NULL; |
| 14020 | maybe_exactfu = FOLD && (DEPENDS_SEMANTICS || LOC); |
| 14021 | maybe_SIMPLE = SIMPLE; |
| 14022 | requires_utf8_target = FALSE; |
| 14023 | has_ss = FALSE; |
| 14024 | has_micro_sign = FALSE; |
| 14025 | |
| 14026 | continue_parse: |
| 14027 | |
| 14028 | /* This breaks under rare circumstances. If folding, we do not |
| 14029 | * want to split a node at a character that is a non-final in a |
| 14030 | * multi-char fold, as an input string could just happen to want to |
| 14031 | * match across the node boundary. The code at the end of the loop |
| 14032 | * looks for this, and backs off until it finds not such a |
| 14033 | * character, but it is possible (though extremely, extremely |
| 14034 | * unlikely) for all characters in the node to be non-final fold |
| 14035 | * ones, in which case we just leave the node fully filled, and |
| 14036 | * hope that it doesn't match the string in just the wrong place */ |
| 14037 | |
| 14038 | assert( ! UTF /* Is at the beginning of a character */ |
| 14039 | || UTF8_IS_INVARIANT(UCHARAT(RExC_parse)) |
| 14040 | || UTF8_IS_START(UCHARAT(RExC_parse))); |
| 14041 | |
| 14042 | overflowed = FALSE; |
| 14043 | |
| 14044 | /* Here, we have a literal character. Find the maximal string of |
| 14045 | * them in the input that we can fit into a single EXACTish node. |
| 14046 | * We quit at the first non-literal or when the node gets full, or |
| 14047 | * under /i the categorization of folding/non-folding character |
| 14048 | * changes */ |
| 14049 | while (p < RExC_end && len < upper_fill) { |
| 14050 | |
| 14051 | /* In most cases each iteration adds one byte to the output. |
| 14052 | * The exceptions override this */ |
| 14053 | Size_t added_len = 1; |
| 14054 | |
| 14055 | oldp = p; |
| 14056 | old_old_s = old_s; |
| 14057 | old_s = s; |
| 14058 | |
| 14059 | /* White space has already been ignored */ |
| 14060 | assert( (RExC_flags & RXf_PMf_EXTENDED) == 0 |
| 14061 | || ! is_PATWS_safe((p), RExC_end, UTF)); |
| 14062 | |
| 14063 | switch ((U8)*p) { |
| 14064 | const char* message; |
| 14065 | U32 packed_warn; |
| 14066 | U8 grok_c_char; |
| 14067 | |
| 14068 | case '^': |
| 14069 | case '$': |
| 14070 | case '.': |
| 14071 | case '[': |
| 14072 | case '(': |
| 14073 | case ')': |
| 14074 | case '|': |
| 14075 | goto loopdone; |
| 14076 | case '\\': |
| 14077 | /* Literal Escapes Switch |
| 14078 | |
| 14079 | This switch is meant to handle escape sequences that |
| 14080 | resolve to a literal character. |
| 14081 | |
| 14082 | Every escape sequence that represents something |
| 14083 | else, like an assertion or a char class, is handled |
| 14084 | in the switch marked 'Special Escapes' above in this |
| 14085 | routine, but also has an entry here as anything that |
| 14086 | isn't explicitly mentioned here will be treated as |
| 14087 | an unescaped equivalent literal. |
| 14088 | */ |
| 14089 | |
| 14090 | switch ((U8)*++p) { |
| 14091 | |
| 14092 | /* These are all the special escapes. */ |
| 14093 | case 'A': /* Start assertion */ |
| 14094 | case 'b': case 'B': /* Word-boundary assertion*/ |
| 14095 | case 'C': /* Single char !DANGEROUS! */ |
| 14096 | case 'd': case 'D': /* digit class */ |
| 14097 | case 'g': case 'G': /* generic-backref, pos assertion */ |
| 14098 | case 'h': case 'H': /* HORIZWS */ |
| 14099 | case 'k': case 'K': /* named backref, keep marker */ |
| 14100 | case 'p': case 'P': /* Unicode property */ |
| 14101 | case 'R': /* LNBREAK */ |
| 14102 | case 's': case 'S': /* space class */ |
| 14103 | case 'v': case 'V': /* VERTWS */ |
| 14104 | case 'w': case 'W': /* word class */ |
| 14105 | case 'X': /* eXtended Unicode "combining |
| 14106 | character sequence" */ |
| 14107 | case 'z': case 'Z': /* End of line/string assertion */ |
| 14108 | --p; |
| 14109 | goto loopdone; |
| 14110 | |
| 14111 | /* Anything after here is an escape that resolves to a |
| 14112 | literal. (Except digits, which may or may not) |
| 14113 | */ |
| 14114 | case 'n': |
| 14115 | ender = '\n'; |
| 14116 | p++; |
| 14117 | break; |
| 14118 | case 'N': /* Handle a single-code point named character. */ |
| 14119 | RExC_parse = p + 1; |
| 14120 | if (! grok_bslash_N(pRExC_state, |
| 14121 | NULL, /* Fail if evaluates to |
| 14122 | anything other than a |
| 14123 | single code point */ |
| 14124 | &ender, /* The returned single code |
| 14125 | point */ |
| 14126 | NULL, /* Don't need a count of |
| 14127 | how many code points */ |
| 14128 | flagp, |
| 14129 | RExC_strict, |
| 14130 | depth) |
| 14131 | ) { |
| 14132 | if (*flagp & NEED_UTF8) |
| 14133 | FAIL("panic: grok_bslash_N set NEED_UTF8"); |
| 14134 | RETURN_FAIL_ON_RESTART_FLAGP(flagp); |
| 14135 | |
| 14136 | /* Here, it wasn't a single code point. Go close |
| 14137 | * up this EXACTish node. The switch() prior to |
| 14138 | * this switch handles the other cases */ |
| 14139 | RExC_parse = p = oldp; |
| 14140 | goto loopdone; |
| 14141 | } |
| 14142 | p = RExC_parse; |
| 14143 | RExC_parse = parse_start; |
| 14144 | |
| 14145 | /* The \N{} means the pattern, if previously /d, |
| 14146 | * becomes /u. That means it can't be an EXACTF node, |
| 14147 | * but an EXACTFU */ |
| 14148 | if (node_type == EXACTF) { |
| 14149 | node_type = EXACTFU; |
| 14150 | |
| 14151 | /* If the node already contains something that |
| 14152 | * differs between EXACTF and EXACTFU, reparse it |
| 14153 | * as EXACTFU */ |
| 14154 | if (! maybe_exactfu) { |
| 14155 | len = 0; |
| 14156 | s = s0; |
| 14157 | goto reparse; |
| 14158 | } |
| 14159 | } |
| 14160 | |
| 14161 | break; |
| 14162 | case 'r': |
| 14163 | ender = '\r'; |
| 14164 | p++; |
| 14165 | break; |
| 14166 | case 't': |
| 14167 | ender = '\t'; |
| 14168 | p++; |
| 14169 | break; |
| 14170 | case 'f': |
| 14171 | ender = '\f'; |
| 14172 | p++; |
| 14173 | break; |
| 14174 | case 'e': |
| 14175 | ender = ESC_NATIVE; |
| 14176 | p++; |
| 14177 | break; |
| 14178 | case 'a': |
| 14179 | ender = '\a'; |
| 14180 | p++; |
| 14181 | break; |
| 14182 | case 'o': |
| 14183 | if (! grok_bslash_o(&p, |
| 14184 | RExC_end, |
| 14185 | &ender, |
| 14186 | &message, |
| 14187 | &packed_warn, |
| 14188 | (bool) RExC_strict, |
| 14189 | FALSE, /* No illegal cp's */ |
| 14190 | UTF)) |
| 14191 | { |
| 14192 | RExC_parse = p; /* going to die anyway; point to |
| 14193 | exact spot of failure */ |
| 14194 | vFAIL(message); |
| 14195 | } |
| 14196 | |
| 14197 | if (message && TO_OUTPUT_WARNINGS(p)) { |
| 14198 | warn_non_literal_string(p, packed_warn, message); |
| 14199 | } |
| 14200 | break; |
| 14201 | case 'x': |
| 14202 | if (! grok_bslash_x(&p, |
| 14203 | RExC_end, |
| 14204 | &ender, |
| 14205 | &message, |
| 14206 | &packed_warn, |
| 14207 | (bool) RExC_strict, |
| 14208 | FALSE, /* No illegal cp's */ |
| 14209 | UTF)) |
| 14210 | { |
| 14211 | RExC_parse = p; /* going to die anyway; point |
| 14212 | to exact spot of failure */ |
| 14213 | vFAIL(message); |
| 14214 | } |
| 14215 | |
| 14216 | if (message && TO_OUTPUT_WARNINGS(p)) { |
| 14217 | warn_non_literal_string(p, packed_warn, message); |
| 14218 | } |
| 14219 | |
| 14220 | #ifdef EBCDIC |
| 14221 | if (ender < 0x100) { |
| 14222 | if (RExC_recode_x_to_native) { |
| 14223 | ender = LATIN1_TO_NATIVE(ender); |
| 14224 | } |
| 14225 | } |
| 14226 | #endif |
| 14227 | break; |
| 14228 | case 'c': |
| 14229 | p++; |
| 14230 | if (! grok_bslash_c(*p, &grok_c_char, |
| 14231 | &message, &packed_warn)) |
| 14232 | { |
| 14233 | /* going to die anyway; point to exact spot of |
| 14234 | * failure */ |
| 14235 | RExC_parse = p + ((UTF) |
| 14236 | ? UTF8_SAFE_SKIP(p, RExC_end) |
| 14237 | : 1); |
| 14238 | vFAIL(message); |
| 14239 | } |
| 14240 | |
| 14241 | ender = grok_c_char; |
| 14242 | p++; |
| 14243 | if (message && TO_OUTPUT_WARNINGS(p)) { |
| 14244 | warn_non_literal_string(p, packed_warn, message); |
| 14245 | } |
| 14246 | |
| 14247 | break; |
| 14248 | case '8': case '9': /* must be a backreference */ |
| 14249 | --p; |
| 14250 | /* we have an escape like \8 which cannot be an octal escape |
| 14251 | * so we exit the loop, and let the outer loop handle this |
| 14252 | * escape which may or may not be a legitimate backref. */ |
| 14253 | goto loopdone; |
| 14254 | case '1': case '2': case '3':case '4': |
| 14255 | case '5': case '6': case '7': |
| 14256 | /* When we parse backslash escapes there is ambiguity |
| 14257 | * between backreferences and octal escapes. Any escape |
| 14258 | * from \1 - \9 is a backreference, any multi-digit |
| 14259 | * escape which does not start with 0 and which when |
| 14260 | * evaluated as decimal could refer to an already |
| 14261 | * parsed capture buffer is a back reference. Anything |
| 14262 | * else is octal. |
| 14263 | * |
| 14264 | * Note this implies that \118 could be interpreted as |
| 14265 | * 118 OR as "\11" . "8" depending on whether there |
| 14266 | * were 118 capture buffers defined already in the |
| 14267 | * pattern. */ |
| 14268 | |
| 14269 | /* NOTE, RExC_npar is 1 more than the actual number of |
| 14270 | * parens we have seen so far, hence the "<" as opposed |
| 14271 | * to "<=" */ |
| 14272 | if ( !isDIGIT(p[1]) || S_backref_value(p, RExC_end) < RExC_npar) |
| 14273 | { /* Not to be treated as an octal constant, go |
| 14274 | find backref */ |
| 14275 | --p; |
| 14276 | goto loopdone; |
| 14277 | } |
| 14278 | /* FALLTHROUGH */ |
| 14279 | case '0': |
| 14280 | { |
| 14281 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT |
| 14282 | | PERL_SCAN_NOTIFY_ILLDIGIT; |
| 14283 | STRLEN numlen = 3; |
| 14284 | ender = grok_oct(p, &numlen, &flags, NULL); |
| 14285 | p += numlen; |
| 14286 | if ( (flags & PERL_SCAN_NOTIFY_ILLDIGIT) |
| 14287 | && isDIGIT(*p) /* like \08, \178 */ |
| 14288 | && ckWARN(WARN_REGEXP)) |
| 14289 | { |
| 14290 | reg_warn_non_literal_string( |
| 14291 | p + 1, |
| 14292 | form_alien_digit_msg(8, numlen, p, |
| 14293 | RExC_end, UTF, FALSE)); |
| 14294 | } |
| 14295 | } |
| 14296 | break; |
| 14297 | case '\0': |
| 14298 | if (p >= RExC_end) |
| 14299 | FAIL("Trailing \\"); |
| 14300 | /* FALLTHROUGH */ |
| 14301 | default: |
| 14302 | if (isALPHANUMERIC(*p)) { |
| 14303 | /* An alpha followed by '{' is going to fail next |
| 14304 | * iteration, so don't output this warning in that |
| 14305 | * case */ |
| 14306 | if (! isALPHA(*p) || *(p + 1) != '{') { |
| 14307 | ckWARN2reg(p + 1, "Unrecognized escape \\%.1s" |
| 14308 | " passed through", p); |
| 14309 | } |
| 14310 | } |
| 14311 | goto normal_default; |
| 14312 | } /* End of switch on '\' */ |
| 14313 | break; |
| 14314 | case '{': |
| 14315 | /* Trying to gain new uses for '{' without breaking too |
| 14316 | * much existing code is hard. The solution currently |
| 14317 | * adopted is: |
| 14318 | * 1) If there is no ambiguity that a '{' should always |
| 14319 | * be taken literally, at the start of a construct, we |
| 14320 | * just do so. |
| 14321 | * 2) If the literal '{' conflicts with our desired use |
| 14322 | * of it as a metacharacter, we die. The deprecation |
| 14323 | * cycles for this have come and gone. |
| 14324 | * 3) If there is ambiguity, we raise a simple warning. |
| 14325 | * This could happen, for example, if the user |
| 14326 | * intended it to introduce a quantifier, but slightly |
| 14327 | * misspelled the quantifier. Without this warning, |
| 14328 | * the quantifier would silently be taken as a literal |
| 14329 | * string of characters instead of a meta construct */ |
| 14330 | if (len || (p > RExC_start && isALPHA_A(*(p - 1)))) { |
| 14331 | if ( RExC_strict |
| 14332 | || ( p > parse_start + 1 |
| 14333 | && isALPHA_A(*(p - 1)) |
| 14334 | && *(p - 2) == '\\') |
| 14335 | || new_regcurly(p, RExC_end)) |
| 14336 | { |
| 14337 | RExC_parse = p + 1; |
| 14338 | vFAIL("Unescaped left brace in regex is " |
| 14339 | "illegal here"); |
| 14340 | } |
| 14341 | ckWARNreg(p + 1, "Unescaped left brace in regex is" |
| 14342 | " passed through"); |
| 14343 | } |
| 14344 | goto normal_default; |
| 14345 | case '}': |
| 14346 | case ']': |
| 14347 | if (p > RExC_parse && RExC_strict) { |
| 14348 | ckWARN2reg(p + 1, "Unescaped literal '%c'", *p); |
| 14349 | } |
| 14350 | /*FALLTHROUGH*/ |
| 14351 | default: /* A literal character */ |
| 14352 | normal_default: |
| 14353 | if (! UTF8_IS_INVARIANT(*p) && UTF) { |
| 14354 | STRLEN numlen; |
| 14355 | ender = utf8n_to_uvchr((U8*)p, RExC_end - p, |
| 14356 | &numlen, UTF8_ALLOW_DEFAULT); |
| 14357 | p += numlen; |
| 14358 | } |
| 14359 | else |
| 14360 | ender = (U8) *p++; |
| 14361 | break; |
| 14362 | } /* End of switch on the literal */ |
| 14363 | |
| 14364 | /* Here, have looked at the literal character, and <ender> |
| 14365 | * contains its ordinal; <p> points to the character after it. |
| 14366 | * */ |
| 14367 | |
| 14368 | if (ender > 255) { |
| 14369 | REQUIRE_UTF8(flagp); |
| 14370 | if ( UNICODE_IS_PERL_EXTENDED(ender) |
| 14371 | && TO_OUTPUT_WARNINGS(p)) |
| 14372 | { |
| 14373 | ckWARN2_non_literal_string(p, |
| 14374 | packWARN(WARN_PORTABLE), |
| 14375 | PL_extended_cp_format, |
| 14376 | ender); |
| 14377 | } |
| 14378 | } |
| 14379 | |
| 14380 | /* We need to check if the next non-ignored thing is a |
| 14381 | * quantifier. Move <p> to after anything that should be |
| 14382 | * ignored, which, as a side effect, positions <p> for the next |
| 14383 | * loop iteration */ |
| 14384 | skip_to_be_ignored_text(pRExC_state, &p, |
| 14385 | FALSE /* Don't force to /x */ ); |
| 14386 | |
| 14387 | /* If the next thing is a quantifier, it applies to this |
| 14388 | * character only, which means that this character has to be in |
| 14389 | * its own node and can't just be appended to the string in an |
| 14390 | * existing node, so if there are already other characters in |
| 14391 | * the node, close the node with just them, and set up to do |
| 14392 | * this character again next time through, when it will be the |
| 14393 | * only thing in its new node */ |
| 14394 | |
| 14395 | next_is_quantifier = LIKELY(p < RExC_end) |
| 14396 | && UNLIKELY(ISMULT2(p)); |
| 14397 | |
| 14398 | if (next_is_quantifier && LIKELY(len)) { |
| 14399 | p = oldp; |
| 14400 | goto loopdone; |
| 14401 | } |
| 14402 | |
| 14403 | /* Ready to add 'ender' to the node */ |
| 14404 | |
| 14405 | if (! FOLD) { /* The simple case, just append the literal */ |
| 14406 | not_fold_common: |
| 14407 | |
| 14408 | /* Don't output if it would overflow */ |
| 14409 | if (UNLIKELY(len > max_string_len - ((UTF) |
| 14410 | ? UVCHR_SKIP(ender) |
| 14411 | : 1))) |
| 14412 | { |
| 14413 | overflowed = TRUE; |
| 14414 | break; |
| 14415 | } |
| 14416 | |
| 14417 | if (UVCHR_IS_INVARIANT(ender) || ! UTF) { |
| 14418 | *(s++) = (char) ender; |
| 14419 | } |
| 14420 | else { |
| 14421 | U8 * new_s = uvchr_to_utf8((U8*)s, ender); |
| 14422 | added_len = (char *) new_s - s; |
| 14423 | s = (char *) new_s; |
| 14424 | |
| 14425 | if (ender > 255) { |
| 14426 | requires_utf8_target = TRUE; |
| 14427 | } |
| 14428 | } |
| 14429 | } |
| 14430 | else if (LOC && is_PROBLEMATIC_LOCALE_FOLD_cp(ender)) { |
| 14431 | |
| 14432 | /* Here are folding under /l, and the code point is |
| 14433 | * problematic. If this is the first character in the |
| 14434 | * node, change the node type to folding. Otherwise, if |
| 14435 | * this is the first problematic character, close up the |
| 14436 | * existing node, so can start a new node with this one */ |
| 14437 | if (! len) { |
| 14438 | node_type = EXACTFL; |
| 14439 | RExC_contains_locale = 1; |
| 14440 | } |
| 14441 | else if (node_type == EXACT) { |
| 14442 | p = oldp; |
| 14443 | goto loopdone; |
| 14444 | } |
| 14445 | |
| 14446 | /* This problematic code point means we can't simplify |
| 14447 | * things */ |
| 14448 | maybe_exactfu = FALSE; |
| 14449 | |
| 14450 | /* Here, we are adding a problematic fold character. |
| 14451 | * "Problematic" in this context means that its fold isn't |
| 14452 | * known until runtime. (The non-problematic code points |
| 14453 | * are the above-Latin1 ones that fold to also all |
| 14454 | * above-Latin1. Their folds don't vary no matter what the |
| 14455 | * locale is.) But here we have characters whose fold |
| 14456 | * depends on the locale. We just add in the unfolded |
| 14457 | * character, and wait until runtime to fold it */ |
| 14458 | goto not_fold_common; |
| 14459 | } |
| 14460 | else /* regular fold; see if actually is in a fold */ |
| 14461 | if ( (ender < 256 && ! IS_IN_SOME_FOLD_L1(ender)) |
| 14462 | || (ender > 255 |
| 14463 | && ! _invlist_contains_cp(PL_in_some_fold, ender))) |
| 14464 | { |
| 14465 | /* Here, folding, but the character isn't in a fold. |
| 14466 | * |
| 14467 | * Start a new node if previous characters in the node were |
| 14468 | * folded */ |
| 14469 | if (len && node_type != EXACT) { |
| 14470 | p = oldp; |
| 14471 | goto loopdone; |
| 14472 | } |
| 14473 | |
| 14474 | /* Here, continuing a node with non-folded characters. Add |
| 14475 | * this one */ |
| 14476 | goto not_fold_common; |
| 14477 | } |
| 14478 | else { /* Here, does participate in some fold */ |
| 14479 | |
| 14480 | /* If this is the first character in the node, change its |
| 14481 | * type to folding. Otherwise, if this is the first |
| 14482 | * folding character in the node, close up the existing |
| 14483 | * node, so can start a new node with this one. */ |
| 14484 | if (! len) { |
| 14485 | node_type = compute_EXACTish(pRExC_state); |
| 14486 | } |
| 14487 | else if (node_type == EXACT) { |
| 14488 | p = oldp; |
| 14489 | goto loopdone; |
| 14490 | } |
| 14491 | |
| 14492 | if (UTF) { /* Alway use the folded value for UTF-8 |
| 14493 | patterns */ |
| 14494 | if (UVCHR_IS_INVARIANT(ender)) { |
| 14495 | if (UNLIKELY(len + 1 > max_string_len)) { |
| 14496 | overflowed = TRUE; |
| 14497 | break; |
| 14498 | } |
| 14499 | |
| 14500 | *(s)++ = (U8) toFOLD(ender); |
| 14501 | } |
| 14502 | else { |
| 14503 | UV folded = _to_uni_fold_flags( |
| 14504 | ender, |
| 14505 | (U8 *) s, /* We have allocated extra space |
| 14506 | in 's' so can't run off the |
| 14507 | end */ |
| 14508 | &added_len, |
| 14509 | FOLD_FLAGS_FULL | ((ASCII_FOLD_RESTRICTED) |
| 14510 | ? FOLD_FLAGS_NOMIX_ASCII |
| 14511 | : 0)); |
| 14512 | if (UNLIKELY(len + added_len > max_string_len)) { |
| 14513 | overflowed = TRUE; |
| 14514 | break; |
| 14515 | } |
| 14516 | |
| 14517 | s += added_len; |
| 14518 | |
| 14519 | if ( folded > 255 |
| 14520 | && LIKELY(folded != GREEK_SMALL_LETTER_MU)) |
| 14521 | { |
| 14522 | /* U+B5 folds to the MU, so its possible for a |
| 14523 | * non-UTF-8 target to match it */ |
| 14524 | requires_utf8_target = TRUE; |
| 14525 | } |
| 14526 | } |
| 14527 | } |
| 14528 | else { /* Here is non-UTF8. */ |
| 14529 | |
| 14530 | /* The fold will be one or (rarely) two characters. |
| 14531 | * Check that there's room for at least a single one |
| 14532 | * before setting any flags, etc. Because otherwise an |
| 14533 | * overflowing character could cause a flag to be set |
| 14534 | * even though it doesn't end up in this node. (For |
| 14535 | * the two character fold, we check again, before |
| 14536 | * setting any flags) */ |
| 14537 | if (UNLIKELY(len + 1 > max_string_len)) { |
| 14538 | overflowed = TRUE; |
| 14539 | break; |
| 14540 | } |
| 14541 | |
| 14542 | #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \ |
| 14543 | || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \ |
| 14544 | || UNICODE_DOT_DOT_VERSION > 0) |
| 14545 | |
| 14546 | /* On non-ancient Unicodes, check for the only possible |
| 14547 | * multi-char fold */ |
| 14548 | if (UNLIKELY(ender == LATIN_SMALL_LETTER_SHARP_S)) { |
| 14549 | |
| 14550 | /* This potential multi-char fold means the node |
| 14551 | * can't be simple (because it could match more |
| 14552 | * than a single char). And in some cases it will |
| 14553 | * match 'ss', so set that flag */ |
| 14554 | maybe_SIMPLE = 0; |
| 14555 | has_ss = TRUE; |
| 14556 | |
| 14557 | /* It can't change to be an EXACTFU (unless already |
| 14558 | * is one). We fold it iff under /u rules. */ |
| 14559 | if (node_type != EXACTFU) { |
| 14560 | maybe_exactfu = FALSE; |
| 14561 | } |
| 14562 | else { |
| 14563 | if (UNLIKELY(len + 2 > max_string_len)) { |
| 14564 | overflowed = TRUE; |
| 14565 | break; |
| 14566 | } |
| 14567 | |
| 14568 | *(s++) = 's'; |
| 14569 | *(s++) = 's'; |
| 14570 | added_len = 2; |
| 14571 | |
| 14572 | goto done_with_this_char; |
| 14573 | } |
| 14574 | } |
| 14575 | else if ( UNLIKELY(isALPHA_FOLD_EQ(ender, 's')) |
| 14576 | && LIKELY(len > 0) |
| 14577 | && UNLIKELY(isALPHA_FOLD_EQ(*(s-1), 's'))) |
| 14578 | { |
| 14579 | /* Also, the sequence 'ss' is special when not |
| 14580 | * under /u. If the target string is UTF-8, it |
| 14581 | * should match SHARP S; otherwise it won't. So, |
| 14582 | * here we have to exclude the possibility of this |
| 14583 | * node moving to /u.*/ |
| 14584 | has_ss = TRUE; |
| 14585 | maybe_exactfu = FALSE; |
| 14586 | } |
| 14587 | #endif |
| 14588 | /* Here, the fold will be a single character */ |
| 14589 | |
| 14590 | if (UNLIKELY(ender == MICRO_SIGN)) { |
| 14591 | has_micro_sign = TRUE; |
| 14592 | } |
| 14593 | else if (PL_fold[ender] != PL_fold_latin1[ender]) { |
| 14594 | |
| 14595 | /* If the character's fold differs between /d and |
| 14596 | * /u, this can't change to be an EXACTFU node */ |
| 14597 | maybe_exactfu = FALSE; |
| 14598 | } |
| 14599 | |
| 14600 | *(s++) = (DEPENDS_SEMANTICS) |
| 14601 | ? (char) toFOLD(ender) |
| 14602 | |
| 14603 | /* Under /u, the fold of any character in |
| 14604 | * the 0-255 range happens to be its |
| 14605 | * lowercase equivalent, except for LATIN |
| 14606 | * SMALL LETTER SHARP S, which was handled |
| 14607 | * above, and the MICRO SIGN, whose fold |
| 14608 | * requires UTF-8 to represent. */ |
| 14609 | : (char) toLOWER_L1(ender); |
| 14610 | } |
| 14611 | } /* End of adding current character to the node */ |
| 14612 | |
| 14613 | done_with_this_char: |
| 14614 | |
| 14615 | len += added_len; |
| 14616 | |
| 14617 | if (next_is_quantifier) { |
| 14618 | |
| 14619 | /* Here, the next input is a quantifier, and to get here, |
| 14620 | * the current character is the only one in the node. */ |
| 14621 | goto loopdone; |
| 14622 | } |
| 14623 | |
| 14624 | } /* End of loop through literal characters */ |
| 14625 | |
| 14626 | /* Here we have either exhausted the input or run out of room in |
| 14627 | * the node. If the former, we are done. (If we encountered a |
| 14628 | * character that can't be in the node, transfer is made directly |
| 14629 | * to <loopdone>, and so we wouldn't have fallen off the end of the |
| 14630 | * loop.) */ |
| 14631 | if (LIKELY(! overflowed)) { |
| 14632 | goto loopdone; |
| 14633 | } |
| 14634 | |
| 14635 | /* Here we have run out of room. We can grow plain EXACT and |
| 14636 | * LEXACT nodes. If the pattern is gigantic enough, though, |
| 14637 | * eventually we'll have to artificially chunk the pattern into |
| 14638 | * multiple nodes. */ |
| 14639 | if (! LOC && (node_type == EXACT || node_type == LEXACT)) { |
| 14640 | Size_t overhead = 1 + regarglen[OP(REGNODE_p(ret))]; |
| 14641 | Size_t overhead_expansion = 0; |
| 14642 | char temp[256]; |
| 14643 | Size_t max_nodes_for_string; |
| 14644 | Size_t achievable; |
| 14645 | SSize_t delta; |
| 14646 | |
| 14647 | /* Here we couldn't fit the final character in the current |
| 14648 | * node, so it will have to be reparsed, no matter what else we |
| 14649 | * do */ |
| 14650 | p = oldp; |
| 14651 | |
| 14652 | /* If would have overflowed a regular EXACT node, switch |
| 14653 | * instead to an LEXACT. The code below is structured so that |
| 14654 | * the actual growing code is common to changing from an EXACT |
| 14655 | * or just increasing the LEXACT size. This means that we have |
| 14656 | * to save the string in the EXACT case before growing, and |
| 14657 | * then copy it afterwards to its new location */ |
| 14658 | if (node_type == EXACT) { |
| 14659 | overhead_expansion = regarglen[LEXACT] - regarglen[EXACT]; |
| 14660 | RExC_emit += overhead_expansion; |
| 14661 | Copy(s0, temp, len, char); |
| 14662 | } |
| 14663 | |
| 14664 | /* Ready to grow. If it was a plain EXACT, the string was |
| 14665 | * saved, and the first few bytes of it overwritten by adding |
| 14666 | * an argument field. We assume, as we do elsewhere in this |
| 14667 | * file, that one byte of remaining input will translate into |
| 14668 | * one byte of output, and if that's too small, we grow again, |
| 14669 | * if too large the excess memory is freed at the end */ |
| 14670 | |
| 14671 | max_nodes_for_string = U16_MAX - overhead - overhead_expansion; |
| 14672 | achievable = MIN(max_nodes_for_string, |
| 14673 | current_string_nodes + STR_SZ(RExC_end - p)); |
| 14674 | delta = achievable - current_string_nodes; |
| 14675 | |
| 14676 | /* If there is just no more room, go finish up this chunk of |
| 14677 | * the pattern. */ |
| 14678 | if (delta <= 0) { |
| 14679 | goto loopdone; |
| 14680 | } |
| 14681 | |
| 14682 | change_engine_size(pRExC_state, delta + overhead_expansion); |
| 14683 | current_string_nodes += delta; |
| 14684 | max_string_len |
| 14685 | = sizeof(struct regnode) * current_string_nodes; |
| 14686 | upper_fill = max_string_len + 1; |
| 14687 | |
| 14688 | /* If the length was small, we know this was originally an |
| 14689 | * EXACT node now converted to LEXACT, and the string has to be |
| 14690 | * restored. Otherwise the string was untouched. 260 is just |
| 14691 | * a number safely above 255 so don't have to worry about |
| 14692 | * getting it precise */ |
| 14693 | if (len < 260) { |
| 14694 | node_type = LEXACT; |
| 14695 | FILL_NODE(ret, node_type); |
| 14696 | s0 = STRING(REGNODE_p(ret)); |
| 14697 | Copy(temp, s0, len, char); |
| 14698 | s = s0 + len; |
| 14699 | } |
| 14700 | |
| 14701 | goto continue_parse; |
| 14702 | } |
| 14703 | else if (FOLD) { |
| 14704 | bool splittable = FALSE; |
| 14705 | bool backed_up = FALSE; |
| 14706 | char * e; /* should this be U8? */ |
| 14707 | char * s_start; /* should this be U8? */ |
| 14708 | |
| 14709 | /* Here is /i. Running out of room creates a problem if we are |
| 14710 | * folding, and the split happens in the middle of a |
| 14711 | * multi-character fold, as a match that should have occurred, |
| 14712 | * won't, due to the way nodes are matched, and our artificial |
| 14713 | * boundary. So back off until we aren't splitting such a |
| 14714 | * fold. If there is no such place to back off to, we end up |
| 14715 | * taking the entire node as-is. This can happen if the node |
| 14716 | * consists entirely of 'f' or entirely of 's' characters (or |
| 14717 | * things that fold to them) as 'ff' and 'ss' are |
| 14718 | * multi-character folds. |
| 14719 | * |
| 14720 | * The Unicode standard says that multi character folds consist |
| 14721 | * of either two or three characters. That means we would be |
| 14722 | * splitting one if the final character in the node is at the |
| 14723 | * beginning of either type, or is the second of a three |
| 14724 | * character fold. |
| 14725 | * |
| 14726 | * At this point: |
| 14727 | * ender is the code point of the character that won't fit |
| 14728 | * in the node |
| 14729 | * s points to just beyond the final byte in the node. |
| 14730 | * It's where we would place ender if there were |
| 14731 | * room, and where in fact we do place ender's fold |
| 14732 | * in the code below, as we've over-allocated space |
| 14733 | * for s0 (hence s) to allow for this |
| 14734 | * e starts at 's' and advances as we append things. |
| 14735 | * old_s is the same as 's'. (If ender had fit, 's' would |
| 14736 | * have been advanced to beyond it). |
| 14737 | * old_old_s points to the beginning byte of the final |
| 14738 | * character in the node |
| 14739 | * p points to the beginning byte in the input of the |
| 14740 | * character beyond 'ender'. |
| 14741 | * oldp points to the beginning byte in the input of |
| 14742 | * 'ender'. |
| 14743 | * |
| 14744 | * In the case of /il, we haven't folded anything that could be |
| 14745 | * affected by the locale. That means only above-Latin1 |
| 14746 | * characters that fold to other above-latin1 characters get |
| 14747 | * folded at compile time. To check where a good place to |
| 14748 | * split nodes is, everything in it will have to be folded. |
| 14749 | * The boolean 'maybe_exactfu' keeps track in /il if there are |
| 14750 | * any unfolded characters in the node. */ |
| 14751 | bool need_to_fold_loc = LOC && ! maybe_exactfu; |
| 14752 | |
| 14753 | /* If we do need to fold the node, we need a place to store the |
| 14754 | * folded copy, and a way to map back to the unfolded original |
| 14755 | * */ |
| 14756 | char * locfold_buf = NULL; |
| 14757 | Size_t * loc_correspondence = NULL; |
| 14758 | |
| 14759 | if (! need_to_fold_loc) { /* The normal case. Just |
| 14760 | initialize to the actual node */ |
| 14761 | e = s; |
| 14762 | s_start = s0; |
| 14763 | s = old_old_s; /* Point to the beginning of the final char |
| 14764 | that fits in the node */ |
| 14765 | } |
| 14766 | else { |
| 14767 | |
| 14768 | /* Here, we have filled a /il node, and there are unfolded |
| 14769 | * characters in it. If the runtime locale turns out to be |
| 14770 | * UTF-8, there are possible multi-character folds, just |
| 14771 | * like when not under /l. The node hence can't terminate |
| 14772 | * in the middle of such a fold. To determine this, we |
| 14773 | * have to create a folded copy of this node. That means |
| 14774 | * reparsing the node, folding everything assuming a UTF-8 |
| 14775 | * locale. (If at runtime it isn't such a locale, the |
| 14776 | * actions here wouldn't have been necessary, but we have |
| 14777 | * to assume the worst case.) If we find we need to back |
| 14778 | * off the folded string, we do so, and then map that |
| 14779 | * position back to the original unfolded node, which then |
| 14780 | * gets output, truncated at that spot */ |
| 14781 | |
| 14782 | char * redo_p = RExC_parse; |
| 14783 | char * redo_e; |
| 14784 | char * old_redo_e; |
| 14785 | |
| 14786 | /* Allow enough space assuming a single byte input folds to |
| 14787 | * a single byte output, plus assume that the two unparsed |
| 14788 | * characters (that we may need) fold to the largest number |
| 14789 | * of bytes possible, plus extra for one more worst case |
| 14790 | * scenario. In the loop below, if we start eating into |
| 14791 | * that final spare space, we enlarge this initial space */ |
| 14792 | Size_t size = max_string_len + (3 * UTF8_MAXBYTES_CASE) + 1; |
| 14793 | |
| 14794 | Newxz(locfold_buf, size, char); |
| 14795 | Newxz(loc_correspondence, size, Size_t); |
| 14796 | |
| 14797 | /* Redo this node's parse, folding into 'locfold_buf' */ |
| 14798 | redo_p = RExC_parse; |
| 14799 | old_redo_e = redo_e = locfold_buf; |
| 14800 | while (redo_p <= oldp) { |
| 14801 | |
| 14802 | old_redo_e = redo_e; |
| 14803 | loc_correspondence[redo_e - locfold_buf] |
| 14804 | = redo_p - RExC_parse; |
| 14805 | |
| 14806 | if (UTF) { |
| 14807 | Size_t added_len; |
| 14808 | |
| 14809 | (void) _to_utf8_fold_flags((U8 *) redo_p, |
| 14810 | (U8 *) RExC_end, |
| 14811 | (U8 *) redo_e, |
| 14812 | &added_len, |
| 14813 | FOLD_FLAGS_FULL); |
| 14814 | redo_e += added_len; |
| 14815 | redo_p += UTF8SKIP(redo_p); |
| 14816 | } |
| 14817 | else { |
| 14818 | |
| 14819 | /* Note that if this code is run on some ancient |
| 14820 | * Unicode versions, SHARP S doesn't fold to 'ss', |
| 14821 | * but rather than clutter the code with #ifdef's, |
| 14822 | * as is done above, we ignore that possibility. |
| 14823 | * This is ok because this code doesn't affect what |
| 14824 | * gets matched, but merely where the node gets |
| 14825 | * split */ |
| 14826 | if (UCHARAT(redo_p) != LATIN_SMALL_LETTER_SHARP_S) { |
| 14827 | *redo_e++ = toLOWER_L1(UCHARAT(redo_p)); |
| 14828 | } |
| 14829 | else { |
| 14830 | *redo_e++ = 's'; |
| 14831 | *redo_e++ = 's'; |
| 14832 | } |
| 14833 | redo_p++; |
| 14834 | } |
| 14835 | |
| 14836 | |
| 14837 | /* If we're getting so close to the end that a |
| 14838 | * worst-case fold in the next character would cause us |
| 14839 | * to overflow, increase, assuming one byte output byte |
| 14840 | * per one byte input one, plus room for another worst |
| 14841 | * case fold */ |
| 14842 | if ( redo_p <= oldp |
| 14843 | && redo_e > locfold_buf + size |
| 14844 | - (UTF8_MAXBYTES_CASE + 1)) |
| 14845 | { |
| 14846 | Size_t new_size = size |
| 14847 | + (oldp - redo_p) |
| 14848 | + UTF8_MAXBYTES_CASE + 1; |
| 14849 | Ptrdiff_t e_offset = redo_e - locfold_buf; |
| 14850 | |
| 14851 | Renew(locfold_buf, new_size, char); |
| 14852 | Renew(loc_correspondence, new_size, Size_t); |
| 14853 | size = new_size; |
| 14854 | |
| 14855 | redo_e = locfold_buf + e_offset; |
| 14856 | } |
| 14857 | } |
| 14858 | |
| 14859 | /* Set so that things are in terms of the folded, temporary |
| 14860 | * string */ |
| 14861 | s = old_redo_e; |
| 14862 | s_start = locfold_buf; |
| 14863 | e = redo_e; |
| 14864 | |
| 14865 | } |
| 14866 | |
| 14867 | /* Here, we have 's', 's_start' and 'e' set up to point to the |
| 14868 | * input that goes into the node, folded. |
| 14869 | * |
| 14870 | * If the final character of the node and the fold of ender |
| 14871 | * form the first two characters of a three character fold, we |
| 14872 | * need to peek ahead at the next (unparsed) character in the |
| 14873 | * input to determine if the three actually do form such a |
| 14874 | * fold. Just looking at that character is not generally |
| 14875 | * sufficient, as it could be, for example, an escape sequence |
| 14876 | * that evaluates to something else, and it needs to be folded. |
| 14877 | * |
| 14878 | * khw originally thought to just go through the parse loop one |
| 14879 | * extra time, but that doesn't work easily as that iteration |
| 14880 | * could cause things to think that the parse is over and to |
| 14881 | * goto loopdone. The character could be a '$' for example, or |
| 14882 | * the character beyond could be a quantifier, and other |
| 14883 | * glitches as well. |
| 14884 | * |
| 14885 | * The solution used here for peeking ahead is to look at that |
| 14886 | * next character. If it isn't ASCII punctuation, then it will |
| 14887 | * be something that continues in an EXACTish node if there |
| 14888 | * were space. We append the fold of it to s, having reserved |
| 14889 | * enough room in s0 for the purpose. If we can't reasonably |
| 14890 | * peek ahead, we instead assume the worst case: that it is |
| 14891 | * something that would form the completion of a multi-char |
| 14892 | * fold. |
| 14893 | * |
| 14894 | * If we can't split between s and ender, we work backwards |
| 14895 | * character-by-character down to s0. At each current point |
| 14896 | * see if we are at the beginning of a multi-char fold. If so, |
| 14897 | * that means we would be splitting the fold across nodes, and |
| 14898 | * so we back up one and try again. |
| 14899 | * |
| 14900 | * If we're not at the beginning, we still could be at the |
| 14901 | * final two characters of a (rare) three character fold. We |
| 14902 | * check if the sequence starting at the character before the |
| 14903 | * current position (and including the current and next |
| 14904 | * characters) is a three character fold. If not, the node can |
| 14905 | * be split here. If it is, we have to backup two characters |
| 14906 | * and try again. |
| 14907 | * |
| 14908 | * Otherwise, the node can be split at the current position. |
| 14909 | * |
| 14910 | * The same logic is used for UTF-8 patterns and not */ |
| 14911 | if (UTF) { |
| 14912 | Size_t added_len; |
| 14913 | |
| 14914 | /* Append the fold of ender */ |
| 14915 | (void) _to_uni_fold_flags( |
| 14916 | ender, |
| 14917 | (U8 *) e, |
| 14918 | &added_len, |
| 14919 | FOLD_FLAGS_FULL | ((ASCII_FOLD_RESTRICTED) |
| 14920 | ? FOLD_FLAGS_NOMIX_ASCII |
| 14921 | : 0)); |
| 14922 | e += added_len; |
| 14923 | |
| 14924 | /* 's' and the character folded to by ender may be the |
| 14925 | * first two of a three-character fold, in which case the |
| 14926 | * node should not be split here. That may mean examining |
| 14927 | * the so-far unparsed character starting at 'p'. But if |
| 14928 | * ender folded to more than one character, we already have |
| 14929 | * three characters to look at. Also, we first check if |
| 14930 | * the sequence consisting of s and the next character form |
| 14931 | * the first two of some three character fold. If not, |
| 14932 | * there's no need to peek ahead. */ |
| 14933 | if ( added_len <= UTF8SKIP(e - added_len) |
| 14934 | && UNLIKELY(is_THREE_CHAR_FOLD_HEAD_utf8_safe(s, e))) |
| 14935 | { |
| 14936 | /* Here, the two do form the beginning of a potential |
| 14937 | * three character fold. The unexamined character may |
| 14938 | * or may not complete it. Peek at it. It might be |
| 14939 | * something that ends the node or an escape sequence, |
| 14940 | * in which case we don't know without a lot of work |
| 14941 | * what it evaluates to, so we have to assume the worst |
| 14942 | * case: that it does complete the fold, and so we |
| 14943 | * can't split here. All such instances will have |
| 14944 | * that character be an ASCII punctuation character, |
| 14945 | * like a backslash. So, for that case, backup one and |
| 14946 | * drop down to try at that position */ |
| 14947 | if (isPUNCT(*p)) { |
| 14948 | s = (char *) utf8_hop_back((U8 *) s, -1, |
| 14949 | (U8 *) s_start); |
| 14950 | backed_up = TRUE; |
| 14951 | } |
| 14952 | else { |
| 14953 | /* Here, since it's not punctuation, it must be a |
| 14954 | * real character, and we can append its fold to |
| 14955 | * 'e' (having deliberately reserved enough space |
| 14956 | * for this eventuality) and drop down to check if |
| 14957 | * the three actually do form a folded sequence */ |
| 14958 | (void) _to_utf8_fold_flags( |
| 14959 | (U8 *) p, (U8 *) RExC_end, |
| 14960 | (U8 *) e, |
| 14961 | &added_len, |
| 14962 | FOLD_FLAGS_FULL | ((ASCII_FOLD_RESTRICTED) |
| 14963 | ? FOLD_FLAGS_NOMIX_ASCII |
| 14964 | : 0)); |
| 14965 | e += added_len; |
| 14966 | } |
| 14967 | } |
| 14968 | |
| 14969 | /* Here, we either have three characters available in |
| 14970 | * sequence starting at 's', or we have two characters and |
| 14971 | * know that the following one can't possibly be part of a |
| 14972 | * three character fold. We go through the node backwards |
| 14973 | * until we find a place where we can split it without |
| 14974 | * breaking apart a multi-character fold. At any given |
| 14975 | * point we have to worry about if such a fold begins at |
| 14976 | * the current 's', and also if a three-character fold |
| 14977 | * begins at s-1, (containing s and s+1). Splitting in |
| 14978 | * either case would break apart a fold */ |
| 14979 | do { |
| 14980 | char *prev_s = (char *) utf8_hop_back((U8 *) s, -1, |
| 14981 | (U8 *) s_start); |
| 14982 | |
| 14983 | /* If is a multi-char fold, can't split here. Backup |
| 14984 | * one char and try again */ |
| 14985 | if (UNLIKELY(is_MULTI_CHAR_FOLD_utf8_safe(s, e))) { |
| 14986 | s = prev_s; |
| 14987 | backed_up = TRUE; |
| 14988 | continue; |
| 14989 | } |
| 14990 | |
| 14991 | /* If the two characters beginning at 's' are part of a |
| 14992 | * three character fold starting at the character |
| 14993 | * before s, we can't split either before or after s. |
| 14994 | * Backup two chars and try again */ |
| 14995 | if ( LIKELY(s > s_start) |
| 14996 | && UNLIKELY(is_THREE_CHAR_FOLD_utf8_safe(prev_s, e))) |
| 14997 | { |
| 14998 | s = prev_s; |
| 14999 | s = (char *) utf8_hop_back((U8 *) s, -1, (U8 *) s_start); |
| 15000 | backed_up = TRUE; |
| 15001 | continue; |
| 15002 | } |
| 15003 | |
| 15004 | /* Here there's no multi-char fold between s and the |
| 15005 | * next character following it. We can split */ |
| 15006 | splittable = TRUE; |
| 15007 | break; |
| 15008 | |
| 15009 | } while (s > s_start); /* End of loops backing up through the node */ |
| 15010 | |
| 15011 | /* Here we either couldn't find a place to split the node, |
| 15012 | * or else we broke out of the loop setting 'splittable' to |
| 15013 | * true. In the latter case, the place to split is between |
| 15014 | * the first and second characters in the sequence starting |
| 15015 | * at 's' */ |
| 15016 | if (splittable) { |
| 15017 | s += UTF8SKIP(s); |
| 15018 | } |
| 15019 | } |
| 15020 | else { /* Pattern not UTF-8 */ |
| 15021 | if ( ender != LATIN_SMALL_LETTER_SHARP_S |
| 15022 | || ASCII_FOLD_RESTRICTED) |
| 15023 | { |
| 15024 | assert( toLOWER_L1(ender) < 256 ); |
| 15025 | *e++ = (char)(toLOWER_L1(ender)); /* should e and the cast be U8? */ |
| 15026 | } |
| 15027 | else { |
| 15028 | *e++ = 's'; |
| 15029 | *e++ = 's'; |
| 15030 | } |
| 15031 | |
| 15032 | if ( e - s <= 1 |
| 15033 | && UNLIKELY(is_THREE_CHAR_FOLD_HEAD_latin1_safe(s, e))) |
| 15034 | { |
| 15035 | if (isPUNCT(*p)) { |
| 15036 | s--; |
| 15037 | backed_up = TRUE; |
| 15038 | } |
| 15039 | else { |
| 15040 | if ( UCHARAT(p) != LATIN_SMALL_LETTER_SHARP_S |
| 15041 | || ASCII_FOLD_RESTRICTED) |
| 15042 | { |
| 15043 | assert( toLOWER_L1(ender) < 256 ); |
| 15044 | *e++ = (char)(toLOWER_L1(ender)); /* should e and the cast be U8? */ |
| 15045 | } |
| 15046 | else { |
| 15047 | *e++ = 's'; |
| 15048 | *e++ = 's'; |
| 15049 | } |
| 15050 | } |
| 15051 | } |
| 15052 | |
| 15053 | do { |
| 15054 | if (UNLIKELY(is_MULTI_CHAR_FOLD_latin1_safe(s, e))) { |
| 15055 | s--; |
| 15056 | backed_up = TRUE; |
| 15057 | continue; |
| 15058 | } |
| 15059 | |
| 15060 | if ( LIKELY(s > s_start) |
| 15061 | && UNLIKELY(is_THREE_CHAR_FOLD_latin1_safe(s - 1, e))) |
| 15062 | { |
| 15063 | s -= 2; |
| 15064 | backed_up = TRUE; |
| 15065 | continue; |
| 15066 | } |
| 15067 | |
| 15068 | splittable = TRUE; |
| 15069 | break; |
| 15070 | |
| 15071 | } while (s > s_start); |
| 15072 | |
| 15073 | if (splittable) { |
| 15074 | s++; |
| 15075 | } |
| 15076 | } |
| 15077 | |
| 15078 | /* Here, we are done backing up. If we didn't backup at all |
| 15079 | * (the likely case), just proceed */ |
| 15080 | if (backed_up) { |
| 15081 | |
| 15082 | /* If we did find a place to split, reparse the entire node |
| 15083 | * stopping where we have calculated. */ |
| 15084 | if (splittable) { |
| 15085 | |
| 15086 | /* If we created a temporary folded string under /l, we |
| 15087 | * have to map that back to the original */ |
| 15088 | if (need_to_fold_loc) { |
| 15089 | upper_fill = loc_correspondence[s - s_start]; |
| 15090 | Safefree(locfold_buf); |
| 15091 | Safefree(loc_correspondence); |
| 15092 | |
| 15093 | if (upper_fill == 0) { |
| 15094 | FAIL2("panic: loc_correspondence[%d] is 0", |
| 15095 | (int) (s - s_start)); |
| 15096 | } |
| 15097 | } |
| 15098 | else { |
| 15099 | upper_fill = s - s0; |
| 15100 | } |
| 15101 | goto reparse; |
| 15102 | } |
| 15103 | else if (need_to_fold_loc) { |
| 15104 | Safefree(locfold_buf); |
| 15105 | Safefree(loc_correspondence); |
| 15106 | } |
| 15107 | |
| 15108 | /* Here the node consists entirely of non-final multi-char |
| 15109 | * folds. (Likely it is all 'f's or all 's's.) There's no |
| 15110 | * decent place to split it, so give up and just take the |
| 15111 | * whole thing */ |
| 15112 | len = old_s - s0; |
| 15113 | } |
| 15114 | } /* End of verifying node ends with an appropriate char */ |
| 15115 | |
| 15116 | /* We need to start the next node at the character that didn't fit |
| 15117 | * in this one */ |
| 15118 | p = oldp; |
| 15119 | |
| 15120 | loopdone: /* Jumped to when encounters something that shouldn't be |
| 15121 | in the node */ |
| 15122 | |
| 15123 | /* Free up any over-allocated space; cast is to silence bogus |
| 15124 | * warning in MS VC */ |
| 15125 | change_engine_size(pRExC_state, |
| 15126 | - (Ptrdiff_t) (current_string_nodes - STR_SZ(len))); |
| 15127 | |
| 15128 | /* I (khw) don't know if you can get here with zero length, but the |
| 15129 | * old code handled this situation by creating a zero-length EXACT |
| 15130 | * node. Might as well be NOTHING instead */ |
| 15131 | if (len == 0) { |
| 15132 | OP(REGNODE_p(ret)) = NOTHING; |
| 15133 | } |
| 15134 | else { |
| 15135 | |
| 15136 | /* If the node type is EXACT here, check to see if it |
| 15137 | * should be EXACTL, or EXACT_REQ8. */ |
| 15138 | if (node_type == EXACT) { |
| 15139 | if (LOC) { |
| 15140 | node_type = EXACTL; |
| 15141 | } |
| 15142 | else if (requires_utf8_target) { |
| 15143 | node_type = EXACT_REQ8; |
| 15144 | } |
| 15145 | } |
| 15146 | else if (node_type == LEXACT) { |
| 15147 | if (requires_utf8_target) { |
| 15148 | node_type = LEXACT_REQ8; |
| 15149 | } |
| 15150 | } |
| 15151 | else if (FOLD) { |
| 15152 | if ( UNLIKELY(has_micro_sign || has_ss) |
| 15153 | && (node_type == EXACTFU || ( node_type == EXACTF |
| 15154 | && maybe_exactfu))) |
| 15155 | { /* These two conditions are problematic in non-UTF-8 |
| 15156 | EXACTFU nodes. */ |
| 15157 | assert(! UTF); |
| 15158 | node_type = EXACTFUP; |
| 15159 | } |
| 15160 | else if (node_type == EXACTFL) { |
| 15161 | |
| 15162 | /* 'maybe_exactfu' is deliberately set above to |
| 15163 | * indicate this node type, where all code points in it |
| 15164 | * are above 255 */ |
| 15165 | if (maybe_exactfu) { |
| 15166 | node_type = EXACTFLU8; |
| 15167 | } |
| 15168 | else if (UNLIKELY( |
| 15169 | _invlist_contains_cp(PL_HasMultiCharFold, ender))) |
| 15170 | { |
| 15171 | /* A character that folds to more than one will |
| 15172 | * match multiple characters, so can't be SIMPLE. |
| 15173 | * We don't have to worry about this with EXACTFLU8 |
| 15174 | * nodes just above, as they have already been |
| 15175 | * folded (since the fold doesn't vary at run |
| 15176 | * time). Here, if the final character in the node |
| 15177 | * folds to multiple, it can't be simple. (This |
| 15178 | * only has an effect if the node has only a single |
| 15179 | * character, hence the final one, as elsewhere we |
| 15180 | * turn off simple for nodes whose length > 1 */ |
| 15181 | maybe_SIMPLE = 0; |
| 15182 | } |
| 15183 | } |
| 15184 | else if (node_type == EXACTF) { /* Means is /di */ |
| 15185 | |
| 15186 | /* This intermediate variable is needed solely because |
| 15187 | * the asserts in the macro where used exceed Win32's |
| 15188 | * literal string capacity */ |
| 15189 | char first_char = * STRING(REGNODE_p(ret)); |
| 15190 | |
| 15191 | /* If 'maybe_exactfu' is clear, then we need to stay |
| 15192 | * /di. If it is set, it means there are no code |
| 15193 | * points that match differently depending on UTF8ness |
| 15194 | * of the target string, so it can become an EXACTFU |
| 15195 | * node */ |
| 15196 | if (! maybe_exactfu) { |
| 15197 | RExC_seen_d_op = TRUE; |
| 15198 | } |
| 15199 | else if ( isALPHA_FOLD_EQ(first_char, 's') |
| 15200 | || isALPHA_FOLD_EQ(ender, 's')) |
| 15201 | { |
| 15202 | /* But, if the node begins or ends in an 's' we |
| 15203 | * have to defer changing it into an EXACTFU, as |
| 15204 | * the node could later get joined with another one |
| 15205 | * that ends or begins with 's' creating an 'ss' |
| 15206 | * sequence which would then wrongly match the |
| 15207 | * sharp s without the target being UTF-8. We |
| 15208 | * create a special node that we resolve later when |
| 15209 | * we join nodes together */ |
| 15210 | |
| 15211 | node_type = EXACTFU_S_EDGE; |
| 15212 | } |
| 15213 | else { |
| 15214 | node_type = EXACTFU; |
| 15215 | } |
| 15216 | } |
| 15217 | |
| 15218 | if (requires_utf8_target && node_type == EXACTFU) { |
| 15219 | node_type = EXACTFU_REQ8; |
| 15220 | } |
| 15221 | } |
| 15222 | |
| 15223 | OP(REGNODE_p(ret)) = node_type; |
| 15224 | setSTR_LEN(REGNODE_p(ret), len); |
| 15225 | RExC_emit += STR_SZ(len); |
| 15226 | |
| 15227 | /* If the node isn't a single character, it can't be SIMPLE */ |
| 15228 | if (len > (Size_t) ((UTF) ? UTF8SKIP(STRING(REGNODE_p(ret))) : 1)) { |
| 15229 | maybe_SIMPLE = 0; |
| 15230 | } |
| 15231 | |
| 15232 | *flagp |= HASWIDTH | maybe_SIMPLE; |
| 15233 | } |
| 15234 | |
| 15235 | Set_Node_Length(REGNODE_p(ret), p - parse_start - 1); |
| 15236 | RExC_parse = p; |
| 15237 | |
| 15238 | { |
| 15239 | /* len is STRLEN which is unsigned, need to copy to signed */ |
| 15240 | IV iv = len; |
| 15241 | if (iv < 0) |
| 15242 | vFAIL("Internal disaster"); |
| 15243 | } |
| 15244 | |
| 15245 | } /* End of label 'defchar:' */ |
| 15246 | break; |
| 15247 | } /* End of giant switch on input character */ |
| 15248 | |
| 15249 | /* Position parse to next real character */ |
| 15250 | skip_to_be_ignored_text(pRExC_state, &RExC_parse, |
| 15251 | FALSE /* Don't force to /x */ ); |
| 15252 | if ( *RExC_parse == '{' |
| 15253 | && OP(REGNODE_p(ret)) != SBOL && ! regcurly(RExC_parse)) |
| 15254 | { |
| 15255 | if (RExC_strict || new_regcurly(RExC_parse, RExC_end)) { |
| 15256 | RExC_parse++; |
| 15257 | vFAIL("Unescaped left brace in regex is illegal here"); |
| 15258 | } |
| 15259 | ckWARNreg(RExC_parse + 1, "Unescaped left brace in regex is" |
| 15260 | " passed through"); |
| 15261 | } |
| 15262 | |
| 15263 | return(ret); |
| 15264 | } |
| 15265 | |
| 15266 | |
| 15267 | STATIC void |
| 15268 | S_populate_ANYOF_from_invlist(pTHX_ regnode *node, SV** invlist_ptr) |
| 15269 | { |
| 15270 | /* Uses the inversion list '*invlist_ptr' to populate the ANYOF 'node'. It |
| 15271 | * sets up the bitmap and any flags, removing those code points from the |
| 15272 | * inversion list, setting it to NULL should it become completely empty */ |
| 15273 | |
| 15274 | dVAR; |
| 15275 | |
| 15276 | PERL_ARGS_ASSERT_POPULATE_ANYOF_FROM_INVLIST; |
| 15277 | assert(PL_regkind[OP(node)] == ANYOF); |
| 15278 | |
| 15279 | /* There is no bitmap for this node type */ |
| 15280 | if (inRANGE(OP(node), ANYOFH, ANYOFRb)) { |
| 15281 | return; |
| 15282 | } |
| 15283 | |
| 15284 | ANYOF_BITMAP_ZERO(node); |
| 15285 | if (*invlist_ptr) { |
| 15286 | |
| 15287 | /* This gets set if we actually need to modify things */ |
| 15288 | bool change_invlist = FALSE; |
| 15289 | |
| 15290 | UV start, end; |
| 15291 | |
| 15292 | /* Start looking through *invlist_ptr */ |
| 15293 | invlist_iterinit(*invlist_ptr); |
| 15294 | while (invlist_iternext(*invlist_ptr, &start, &end)) { |
| 15295 | UV high; |
| 15296 | int i; |
| 15297 | |
| 15298 | if (end == UV_MAX && start <= NUM_ANYOF_CODE_POINTS) { |
| 15299 | ANYOF_FLAGS(node) |= ANYOF_MATCHES_ALL_ABOVE_BITMAP; |
| 15300 | } |
| 15301 | |
| 15302 | /* Quit if are above what we should change */ |
| 15303 | if (start >= NUM_ANYOF_CODE_POINTS) { |
| 15304 | break; |
| 15305 | } |
| 15306 | |
| 15307 | change_invlist = TRUE; |
| 15308 | |
| 15309 | /* Set all the bits in the range, up to the max that we are doing */ |
| 15310 | high = (end < NUM_ANYOF_CODE_POINTS - 1) |
| 15311 | ? end |
| 15312 | : NUM_ANYOF_CODE_POINTS - 1; |
| 15313 | for (i = start; i <= (int) high; i++) { |
| 15314 | if (! ANYOF_BITMAP_TEST(node, i)) { |
| 15315 | ANYOF_BITMAP_SET(node, i); |
| 15316 | } |
| 15317 | } |
| 15318 | } |
| 15319 | invlist_iterfinish(*invlist_ptr); |
| 15320 | |
| 15321 | /* Done with loop; remove any code points that are in the bitmap from |
| 15322 | * *invlist_ptr; similarly for code points above the bitmap if we have |
| 15323 | * a flag to match all of them anyways */ |
| 15324 | if (change_invlist) { |
| 15325 | _invlist_subtract(*invlist_ptr, PL_InBitmap, invlist_ptr); |
| 15326 | } |
| 15327 | if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) { |
| 15328 | _invlist_intersection(*invlist_ptr, PL_InBitmap, invlist_ptr); |
| 15329 | } |
| 15330 | |
| 15331 | /* If have completely emptied it, remove it completely */ |
| 15332 | if (_invlist_len(*invlist_ptr) == 0) { |
| 15333 | SvREFCNT_dec_NN(*invlist_ptr); |
| 15334 | *invlist_ptr = NULL; |
| 15335 | } |
| 15336 | } |
| 15337 | } |
| 15338 | |
| 15339 | /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]]. |
| 15340 | Character classes ([:foo:]) can also be negated ([:^foo:]). |
| 15341 | Returns a named class id (ANYOF_XXX) if successful, -1 otherwise. |
| 15342 | Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed, |
| 15343 | but trigger failures because they are currently unimplemented. */ |
| 15344 | |
| 15345 | #define POSIXCC_DONE(c) ((c) == ':') |
| 15346 | #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.') |
| 15347 | #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c)) |
| 15348 | #define MAYBE_POSIXCC(c) (POSIXCC(c) || (c) == '^' || (c) == ';') |
| 15349 | |
| 15350 | #define WARNING_PREFIX "Assuming NOT a POSIX class since " |
| 15351 | #define NO_BLANKS_POSIX_WARNING "no blanks are allowed in one" |
| 15352 | #define SEMI_COLON_POSIX_WARNING "a semi-colon was found instead of a colon" |
| 15353 | |
| 15354 | #define NOT_MEANT_TO_BE_A_POSIX_CLASS (OOB_NAMEDCLASS - 1) |
| 15355 | |
| 15356 | /* 'posix_warnings' and 'warn_text' are names of variables in the following |
| 15357 | * routine. q.v. */ |
| 15358 | #define ADD_POSIX_WARNING(p, text) STMT_START { \ |
| 15359 | if (posix_warnings) { \ |
| 15360 | if (! RExC_warn_text ) RExC_warn_text = \ |
| 15361 | (AV *) sv_2mortal((SV *) newAV()); \ |
| 15362 | av_push(RExC_warn_text, Perl_newSVpvf(aTHX_ \ |
| 15363 | WARNING_PREFIX \ |
| 15364 | text \ |
| 15365 | REPORT_LOCATION, \ |
| 15366 | REPORT_LOCATION_ARGS(p))); \ |
| 15367 | } \ |
| 15368 | } STMT_END |
| 15369 | #define CLEAR_POSIX_WARNINGS() \ |
| 15370 | STMT_START { \ |
| 15371 | if (posix_warnings && RExC_warn_text) \ |
| 15372 | av_clear(RExC_warn_text); \ |
| 15373 | } STMT_END |
| 15374 | |
| 15375 | #define CLEAR_POSIX_WARNINGS_AND_RETURN(ret) \ |
| 15376 | STMT_START { \ |
| 15377 | CLEAR_POSIX_WARNINGS(); \ |
| 15378 | return ret; \ |
| 15379 | } STMT_END |
| 15380 | |
| 15381 | STATIC int |
| 15382 | S_handle_possible_posix(pTHX_ RExC_state_t *pRExC_state, |
| 15383 | |
| 15384 | const char * const s, /* Where the putative posix class begins. |
| 15385 | Normally, this is one past the '['. This |
| 15386 | parameter exists so it can be somewhere |
| 15387 | besides RExC_parse. */ |
| 15388 | char ** updated_parse_ptr, /* Where to set the updated parse pointer, or |
| 15389 | NULL */ |
| 15390 | AV ** posix_warnings, /* Where to place any generated warnings, or |
| 15391 | NULL */ |
| 15392 | const bool check_only /* Don't die if error */ |
| 15393 | ) |
| 15394 | { |
| 15395 | /* This parses what the caller thinks may be one of the three POSIX |
| 15396 | * constructs: |
| 15397 | * 1) a character class, like [:blank:] |
| 15398 | * 2) a collating symbol, like [. .] |
| 15399 | * 3) an equivalence class, like [= =] |
| 15400 | * In the latter two cases, it croaks if it finds a syntactically legal |
| 15401 | * one, as these are not handled by Perl. |
| 15402 | * |
| 15403 | * The main purpose is to look for a POSIX character class. It returns: |
| 15404 | * a) the class number |
| 15405 | * if it is a completely syntactically and semantically legal class. |
| 15406 | * 'updated_parse_ptr', if not NULL, is set to point to just after the |
| 15407 | * closing ']' of the class |
| 15408 | * b) OOB_NAMEDCLASS |
| 15409 | * if it appears that one of the three POSIX constructs was meant, but |
| 15410 | * its specification was somehow defective. 'updated_parse_ptr', if |
| 15411 | * not NULL, is set to point to the character just after the end |
| 15412 | * character of the class. See below for handling of warnings. |
| 15413 | * c) NOT_MEANT_TO_BE_A_POSIX_CLASS |
| 15414 | * if it doesn't appear that a POSIX construct was intended. |
| 15415 | * 'updated_parse_ptr' is not changed. No warnings nor errors are |
| 15416 | * raised. |
| 15417 | * |
| 15418 | * In b) there may be errors or warnings generated. If 'check_only' is |
| 15419 | * TRUE, then any errors are discarded. Warnings are returned to the |
| 15420 | * caller via an AV* created into '*posix_warnings' if it is not NULL. If |
| 15421 | * instead it is NULL, warnings are suppressed. |
| 15422 | * |
| 15423 | * The reason for this function, and its complexity is that a bracketed |
| 15424 | * character class can contain just about anything. But it's easy to |
| 15425 | * mistype the very specific posix class syntax but yielding a valid |
| 15426 | * regular bracketed class, so it silently gets compiled into something |
| 15427 | * quite unintended. |
| 15428 | * |
| 15429 | * The solution adopted here maintains backward compatibility except that |
| 15430 | * it adds a warning if it looks like a posix class was intended but |
| 15431 | * improperly specified. The warning is not raised unless what is input |
| 15432 | * very closely resembles one of the 14 legal posix classes. To do this, |
| 15433 | * it uses fuzzy parsing. It calculates how many single-character edits it |
| 15434 | * would take to transform what was input into a legal posix class. Only |
| 15435 | * if that number is quite small does it think that the intention was a |
| 15436 | * posix class. Obviously these are heuristics, and there will be cases |
| 15437 | * where it errs on one side or another, and they can be tweaked as |
| 15438 | * experience informs. |
| 15439 | * |
| 15440 | * The syntax for a legal posix class is: |
| 15441 | * |
| 15442 | * qr/(?xa: \[ : \^? [[:lower:]]{4,6} : \] )/ |
| 15443 | * |
| 15444 | * What this routine considers syntactically to be an intended posix class |
| 15445 | * is this (the comments indicate some restrictions that the pattern |
| 15446 | * doesn't show): |
| 15447 | * |
| 15448 | * qr/(?x: \[? # The left bracket, possibly |
| 15449 | * # omitted |
| 15450 | * \h* # possibly followed by blanks |
| 15451 | * (?: \^ \h* )? # possibly a misplaced caret |
| 15452 | * [:;]? # The opening class character, |
| 15453 | * # possibly omitted. A typo |
| 15454 | * # semi-colon can also be used. |
| 15455 | * \h* |
| 15456 | * \^? # possibly a correctly placed |
| 15457 | * # caret, but not if there was also |
| 15458 | * # a misplaced one |
| 15459 | * \h* |
| 15460 | * .{3,15} # The class name. If there are |
| 15461 | * # deviations from the legal syntax, |
| 15462 | * # its edit distance must be close |
| 15463 | * # to a real class name in order |
| 15464 | * # for it to be considered to be |
| 15465 | * # an intended posix class. |
| 15466 | * \h* |
| 15467 | * [[:punct:]]? # The closing class character, |
| 15468 | * # possibly omitted. If not a colon |
| 15469 | * # nor semi colon, the class name |
| 15470 | * # must be even closer to a valid |
| 15471 | * # one |
| 15472 | * \h* |
| 15473 | * \]? # The right bracket, possibly |
| 15474 | * # omitted. |
| 15475 | * )/ |
| 15476 | * |
| 15477 | * In the above, \h must be ASCII-only. |
| 15478 | * |
| 15479 | * These are heuristics, and can be tweaked as field experience dictates. |
| 15480 | * There will be cases when someone didn't intend to specify a posix class |
| 15481 | * that this warns as being so. The goal is to minimize these, while |
| 15482 | * maximizing the catching of things intended to be a posix class that |
| 15483 | * aren't parsed as such. |
| 15484 | */ |
| 15485 | |
| 15486 | const char* p = s; |
| 15487 | const char * const e = RExC_end; |
| 15488 | unsigned complement = 0; /* If to complement the class */ |
| 15489 | bool found_problem = FALSE; /* Assume OK until proven otherwise */ |
| 15490 | bool has_opening_bracket = FALSE; |
| 15491 | bool has_opening_colon = FALSE; |
| 15492 | int class_number = OOB_NAMEDCLASS; /* Out-of-bounds until find |
| 15493 | valid class */ |
| 15494 | const char * possible_end = NULL; /* used for a 2nd parse pass */ |
| 15495 | const char* name_start; /* ptr to class name first char */ |
| 15496 | |
| 15497 | /* If the number of single-character typos the input name is away from a |
| 15498 | * legal name is no more than this number, it is considered to have meant |
| 15499 | * the legal name */ |
| 15500 | int max_distance = 2; |
| 15501 | |
| 15502 | /* to store the name. The size determines the maximum length before we |
| 15503 | * decide that no posix class was intended. Should be at least |
| 15504 | * sizeof("alphanumeric") */ |
| 15505 | UV input_text[15]; |
| 15506 | STATIC_ASSERT_DECL(C_ARRAY_LENGTH(input_text) >= sizeof "alphanumeric"); |
| 15507 | |
| 15508 | PERL_ARGS_ASSERT_HANDLE_POSSIBLE_POSIX; |
| 15509 | |
| 15510 | CLEAR_POSIX_WARNINGS(); |
| 15511 | |
| 15512 | if (p >= e) { |
| 15513 | return NOT_MEANT_TO_BE_A_POSIX_CLASS; |
| 15514 | } |
| 15515 | |
| 15516 | if (*(p - 1) != '[') { |
| 15517 | ADD_POSIX_WARNING(p, "it doesn't start with a '['"); |
| 15518 | found_problem = TRUE; |
| 15519 | } |
| 15520 | else { |
| 15521 | has_opening_bracket = TRUE; |
| 15522 | } |
| 15523 | |
| 15524 | /* They could be confused and think you can put spaces between the |
| 15525 | * components */ |
| 15526 | if (isBLANK(*p)) { |
| 15527 | found_problem = TRUE; |
| 15528 | |
| 15529 | do { |
| 15530 | p++; |
| 15531 | } while (p < e && isBLANK(*p)); |
| 15532 | |
| 15533 | ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING); |
| 15534 | } |
| 15535 | |
| 15536 | /* For [. .] and [= =]. These are quite different internally from [: :], |
| 15537 | * so they are handled separately. */ |
| 15538 | if (POSIXCC_NOTYET(*p) && p < e - 3) /* 1 for the close, and 1 for the ']' |
| 15539 | and 1 for at least one char in it |
| 15540 | */ |
| 15541 | { |
| 15542 | const char open_char = *p; |
| 15543 | const char * temp_ptr = p + 1; |
| 15544 | |
| 15545 | /* These two constructs are not handled by perl, and if we find a |
| 15546 | * syntactically valid one, we croak. khw, who wrote this code, finds |
| 15547 | * this explanation of them very unclear: |
| 15548 | * http://pubs.opengroup.org/onlinepubs/009696899/basedefs/xbd_chap09.html |
| 15549 | * And searching the rest of the internet wasn't very helpful either. |
| 15550 | * It looks like just about any byte can be in these constructs, |
| 15551 | * depending on the locale. But unless the pattern is being compiled |
| 15552 | * under /l, which is very rare, Perl runs under the C or POSIX locale. |
| 15553 | * In that case, it looks like [= =] isn't allowed at all, and that |
| 15554 | * [. .] could be any single code point, but for longer strings the |
| 15555 | * constituent characters would have to be the ASCII alphabetics plus |
| 15556 | * the minus-hyphen. Any sensible locale definition would limit itself |
| 15557 | * to these. And any portable one definitely should. Trying to parse |
| 15558 | * the general case is a nightmare (see [perl #127604]). So, this code |
| 15559 | * looks only for interiors of these constructs that match: |
| 15560 | * qr/.|[-\w]{2,}/ |
| 15561 | * Using \w relaxes the apparent rules a little, without adding much |
| 15562 | * danger of mistaking something else for one of these constructs. |
| 15563 | * |
| 15564 | * [. .] in some implementations described on the internet is usable to |
| 15565 | * escape a character that otherwise is special in bracketed character |
| 15566 | * classes. For example [.].] means a literal right bracket instead of |
| 15567 | * the ending of the class |
| 15568 | * |
| 15569 | * [= =] can legitimately contain a [. .] construct, but we don't |
| 15570 | * handle this case, as that [. .] construct will later get parsed |
| 15571 | * itself and croak then. And [= =] is checked for even when not under |
| 15572 | * /l, as Perl has long done so. |
| 15573 | * |
| 15574 | * The code below relies on there being a trailing NUL, so it doesn't |
| 15575 | * have to keep checking if the parse ptr < e. |
| 15576 | */ |
| 15577 | if (temp_ptr[1] == open_char) { |
| 15578 | temp_ptr++; |
| 15579 | } |
| 15580 | else while ( temp_ptr < e |
| 15581 | && (isWORDCHAR(*temp_ptr) || *temp_ptr == '-')) |
| 15582 | { |
| 15583 | temp_ptr++; |
| 15584 | } |
| 15585 | |
| 15586 | if (*temp_ptr == open_char) { |
| 15587 | temp_ptr++; |
| 15588 | if (*temp_ptr == ']') { |
| 15589 | temp_ptr++; |
| 15590 | if (! found_problem && ! check_only) { |
| 15591 | RExC_parse = (char *) temp_ptr; |
| 15592 | vFAIL3("POSIX syntax [%c %c] is reserved for future " |
| 15593 | "extensions", open_char, open_char); |
| 15594 | } |
| 15595 | |
| 15596 | /* Here, the syntax wasn't completely valid, or else the call |
| 15597 | * is to check-only */ |
| 15598 | if (updated_parse_ptr) { |
| 15599 | *updated_parse_ptr = (char *) temp_ptr; |
| 15600 | } |
| 15601 | |
| 15602 | CLEAR_POSIX_WARNINGS_AND_RETURN(OOB_NAMEDCLASS); |
| 15603 | } |
| 15604 | } |
| 15605 | |
| 15606 | /* If we find something that started out to look like one of these |
| 15607 | * constructs, but isn't, we continue below so that it can be checked |
| 15608 | * for being a class name with a typo of '.' or '=' instead of a colon. |
| 15609 | * */ |
| 15610 | } |
| 15611 | |
| 15612 | /* Here, we think there is a possibility that a [: :] class was meant, and |
| 15613 | * we have the first real character. It could be they think the '^' comes |
| 15614 | * first */ |
| 15615 | if (*p == '^') { |
| 15616 | found_problem = TRUE; |
| 15617 | ADD_POSIX_WARNING(p + 1, "the '^' must come after the colon"); |
| 15618 | complement = 1; |
| 15619 | p++; |
| 15620 | |
| 15621 | if (isBLANK(*p)) { |
| 15622 | found_problem = TRUE; |
| 15623 | |
| 15624 | do { |
| 15625 | p++; |
| 15626 | } while (p < e && isBLANK(*p)); |
| 15627 | |
| 15628 | ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING); |
| 15629 | } |
| 15630 | } |
| 15631 | |
| 15632 | /* But the first character should be a colon, which they could have easily |
| 15633 | * mistyped on a qwerty keyboard as a semi-colon (and which may be hard to |
| 15634 | * distinguish from a colon, so treat that as a colon). */ |
| 15635 | if (*p == ':') { |
| 15636 | p++; |
| 15637 | has_opening_colon = TRUE; |
| 15638 | } |
| 15639 | else if (*p == ';') { |
| 15640 | found_problem = TRUE; |
| 15641 | p++; |
| 15642 | ADD_POSIX_WARNING(p, SEMI_COLON_POSIX_WARNING); |
| 15643 | has_opening_colon = TRUE; |
| 15644 | } |
| 15645 | else { |
| 15646 | found_problem = TRUE; |
| 15647 | ADD_POSIX_WARNING(p, "there must be a starting ':'"); |
| 15648 | |
| 15649 | /* Consider an initial punctuation (not one of the recognized ones) to |
| 15650 | * be a left terminator */ |
| 15651 | if (*p != '^' && *p != ']' && isPUNCT(*p)) { |
| 15652 | p++; |
| 15653 | } |
| 15654 | } |
| 15655 | |
| 15656 | /* They may think that you can put spaces between the components */ |
| 15657 | if (isBLANK(*p)) { |
| 15658 | found_problem = TRUE; |
| 15659 | |
| 15660 | do { |
| 15661 | p++; |
| 15662 | } while (p < e && isBLANK(*p)); |
| 15663 | |
| 15664 | ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING); |
| 15665 | } |
| 15666 | |
| 15667 | if (*p == '^') { |
| 15668 | |
| 15669 | /* We consider something like [^:^alnum:]] to not have been intended to |
| 15670 | * be a posix class, but XXX maybe we should */ |
| 15671 | if (complement) { |
| 15672 | CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS); |
| 15673 | } |
| 15674 | |
| 15675 | complement = 1; |
| 15676 | p++; |
| 15677 | } |
| 15678 | |
| 15679 | /* Again, they may think that you can put spaces between the components */ |
| 15680 | if (isBLANK(*p)) { |
| 15681 | found_problem = TRUE; |
| 15682 | |
| 15683 | do { |
| 15684 | p++; |
| 15685 | } while (p < e && isBLANK(*p)); |
| 15686 | |
| 15687 | ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING); |
| 15688 | } |
| 15689 | |
| 15690 | if (*p == ']') { |
| 15691 | |
| 15692 | /* XXX This ']' may be a typo, and something else was meant. But |
| 15693 | * treating it as such creates enough complications, that that |
| 15694 | * possibility isn't currently considered here. So we assume that the |
| 15695 | * ']' is what is intended, and if we've already found an initial '[', |
| 15696 | * this leaves this construct looking like [:] or [:^], which almost |
| 15697 | * certainly weren't intended to be posix classes */ |
| 15698 | if (has_opening_bracket) { |
| 15699 | CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS); |
| 15700 | } |
| 15701 | |
| 15702 | /* But this function can be called when we parse the colon for |
| 15703 | * something like qr/[alpha:]]/, so we back up to look for the |
| 15704 | * beginning */ |
| 15705 | p--; |
| 15706 | |
| 15707 | if (*p == ';') { |
| 15708 | found_problem = TRUE; |
| 15709 | ADD_POSIX_WARNING(p, SEMI_COLON_POSIX_WARNING); |
| 15710 | } |
| 15711 | else if (*p != ':') { |
| 15712 | |
| 15713 | /* XXX We are currently very restrictive here, so this code doesn't |
| 15714 | * consider the possibility that, say, /[alpha.]]/ was intended to |
| 15715 | * be a posix class. */ |
| 15716 | CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS); |
| 15717 | } |
| 15718 | |
| 15719 | /* Here we have something like 'foo:]'. There was no initial colon, |
| 15720 | * and we back up over 'foo. XXX Unlike the going forward case, we |
| 15721 | * don't handle typos of non-word chars in the middle */ |
| 15722 | has_opening_colon = FALSE; |
| 15723 | p--; |
| 15724 | |
| 15725 | while (p > RExC_start && isWORDCHAR(*p)) { |
| 15726 | p--; |
| 15727 | } |
| 15728 | p++; |
| 15729 | |
| 15730 | /* Here, we have positioned ourselves to where we think the first |
| 15731 | * character in the potential class is */ |
| 15732 | } |
| 15733 | |
| 15734 | /* Now the interior really starts. There are certain key characters that |
| 15735 | * can end the interior, or these could just be typos. To catch both |
| 15736 | * cases, we may have to do two passes. In the first pass, we keep on |
| 15737 | * going unless we come to a sequence that matches |
| 15738 | * qr/ [[:punct:]] [[:blank:]]* \] /xa |
| 15739 | * This means it takes a sequence to end the pass, so two typos in a row if |
| 15740 | * that wasn't what was intended. If the class is perfectly formed, just |
| 15741 | * this one pass is needed. We also stop if there are too many characters |
| 15742 | * being accumulated, but this number is deliberately set higher than any |
| 15743 | * real class. It is set high enough so that someone who thinks that |
| 15744 | * 'alphanumeric' is a correct name would get warned that it wasn't. |
| 15745 | * While doing the pass, we keep track of where the key characters were in |
| 15746 | * it. If we don't find an end to the class, and one of the key characters |
| 15747 | * was found, we redo the pass, but stop when we get to that character. |
| 15748 | * Thus the key character was considered a typo in the first pass, but a |
| 15749 | * terminator in the second. If two key characters are found, we stop at |
| 15750 | * the second one in the first pass. Again this can miss two typos, but |
| 15751 | * catches a single one |
| 15752 | * |
| 15753 | * In the first pass, 'possible_end' starts as NULL, and then gets set to |
| 15754 | * point to the first key character. For the second pass, it starts as -1. |
| 15755 | * */ |
| 15756 | |
| 15757 | name_start = p; |
| 15758 | parse_name: |
| 15759 | { |
| 15760 | bool has_blank = FALSE; |
| 15761 | bool has_upper = FALSE; |
| 15762 | bool has_terminating_colon = FALSE; |
| 15763 | bool has_terminating_bracket = FALSE; |
| 15764 | bool has_semi_colon = FALSE; |
| 15765 | unsigned int name_len = 0; |
| 15766 | int punct_count = 0; |
| 15767 | |
| 15768 | while (p < e) { |
| 15769 | |
| 15770 | /* Squeeze out blanks when looking up the class name below */ |
| 15771 | if (isBLANK(*p) ) { |
| 15772 | has_blank = TRUE; |
| 15773 | found_problem = TRUE; |
| 15774 | p++; |
| 15775 | continue; |
| 15776 | } |
| 15777 | |
| 15778 | /* The name will end with a punctuation */ |
| 15779 | if (isPUNCT(*p)) { |
| 15780 | const char * peek = p + 1; |
| 15781 | |
| 15782 | /* Treat any non-']' punctuation followed by a ']' (possibly |
| 15783 | * with intervening blanks) as trying to terminate the class. |
| 15784 | * ']]' is very likely to mean a class was intended (but |
| 15785 | * missing the colon), but the warning message that gets |
| 15786 | * generated shows the error position better if we exit the |
| 15787 | * loop at the bottom (eventually), so skip it here. */ |
| 15788 | if (*p != ']') { |
| 15789 | if (peek < e && isBLANK(*peek)) { |
| 15790 | has_blank = TRUE; |
| 15791 | found_problem = TRUE; |
| 15792 | do { |
| 15793 | peek++; |
| 15794 | } while (peek < e && isBLANK(*peek)); |
| 15795 | } |
| 15796 | |
| 15797 | if (peek < e && *peek == ']') { |
| 15798 | has_terminating_bracket = TRUE; |
| 15799 | if (*p == ':') { |
| 15800 | has_terminating_colon = TRUE; |
| 15801 | } |
| 15802 | else if (*p == ';') { |
| 15803 | has_semi_colon = TRUE; |
| 15804 | has_terminating_colon = TRUE; |
| 15805 | } |
| 15806 | else { |
| 15807 | found_problem = TRUE; |
| 15808 | } |
| 15809 | p = peek + 1; |
| 15810 | goto try_posix; |
| 15811 | } |
| 15812 | } |
| 15813 | |
| 15814 | /* Here we have punctuation we thought didn't end the class. |
| 15815 | * Keep track of the position of the key characters that are |
| 15816 | * more likely to have been class-enders */ |
| 15817 | if (*p == ']' || *p == '[' || *p == ':' || *p == ';') { |
| 15818 | |
| 15819 | /* Allow just one such possible class-ender not actually |
| 15820 | * ending the class. */ |
| 15821 | if (possible_end) { |
| 15822 | break; |
| 15823 | } |
| 15824 | possible_end = p; |
| 15825 | } |
| 15826 | |
| 15827 | /* If we have too many punctuation characters, no use in |
| 15828 | * keeping going */ |
| 15829 | if (++punct_count > max_distance) { |
| 15830 | break; |
| 15831 | } |
| 15832 | |
| 15833 | /* Treat the punctuation as a typo. */ |
| 15834 | input_text[name_len++] = *p; |
| 15835 | p++; |
| 15836 | } |
| 15837 | else if (isUPPER(*p)) { /* Use lowercase for lookup */ |
| 15838 | input_text[name_len++] = toLOWER(*p); |
| 15839 | has_upper = TRUE; |
| 15840 | found_problem = TRUE; |
| 15841 | p++; |
| 15842 | } else if (! UTF || UTF8_IS_INVARIANT(*p)) { |
| 15843 | input_text[name_len++] = *p; |
| 15844 | p++; |
| 15845 | } |
| 15846 | else { |
| 15847 | input_text[name_len++] = utf8_to_uvchr_buf((U8 *) p, e, NULL); |
| 15848 | p+= UTF8SKIP(p); |
| 15849 | } |
| 15850 | |
| 15851 | /* The declaration of 'input_text' is how long we allow a potential |
| 15852 | * class name to be, before saying they didn't mean a class name at |
| 15853 | * all */ |
| 15854 | if (name_len >= C_ARRAY_LENGTH(input_text)) { |
| 15855 | break; |
| 15856 | } |
| 15857 | } |
| 15858 | |
| 15859 | /* We get to here when the possible class name hasn't been properly |
| 15860 | * terminated before: |
| 15861 | * 1) we ran off the end of the pattern; or |
| 15862 | * 2) found two characters, each of which might have been intended to |
| 15863 | * be the name's terminator |
| 15864 | * 3) found so many punctuation characters in the purported name, |
| 15865 | * that the edit distance to a valid one is exceeded |
| 15866 | * 4) we decided it was more characters than anyone could have |
| 15867 | * intended to be one. */ |
| 15868 | |
| 15869 | found_problem = TRUE; |
| 15870 | |
| 15871 | /* In the final two cases, we know that looking up what we've |
| 15872 | * accumulated won't lead to a match, even a fuzzy one. */ |
| 15873 | if ( name_len >= C_ARRAY_LENGTH(input_text) |
| 15874 | || punct_count > max_distance) |
| 15875 | { |
| 15876 | /* If there was an intermediate key character that could have been |
| 15877 | * an intended end, redo the parse, but stop there */ |
| 15878 | if (possible_end && possible_end != (char *) -1) { |
| 15879 | possible_end = (char *) -1; /* Special signal value to say |
| 15880 | we've done a first pass */ |
| 15881 | p = name_start; |
| 15882 | goto parse_name; |
| 15883 | } |
| 15884 | |
| 15885 | /* Otherwise, it can't have meant to have been a class */ |
| 15886 | CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS); |
| 15887 | } |
| 15888 | |
| 15889 | /* If we ran off the end, and the final character was a punctuation |
| 15890 | * one, back up one, to look at that final one just below. Later, we |
| 15891 | * will restore the parse pointer if appropriate */ |
| 15892 | if (name_len && p == e && isPUNCT(*(p-1))) { |
| 15893 | p--; |
| 15894 | name_len--; |
| 15895 | } |
| 15896 | |
| 15897 | if (p < e && isPUNCT(*p)) { |
| 15898 | if (*p == ']') { |
| 15899 | has_terminating_bracket = TRUE; |
| 15900 | |
| 15901 | /* If this is a 2nd ']', and the first one is just below this |
| 15902 | * one, consider that to be the real terminator. This gives a |
| 15903 | * uniform and better positioning for the warning message */ |
| 15904 | if ( possible_end |
| 15905 | && possible_end != (char *) -1 |
| 15906 | && *possible_end == ']' |
| 15907 | && name_len && input_text[name_len - 1] == ']') |
| 15908 | { |
| 15909 | name_len--; |
| 15910 | p = possible_end; |
| 15911 | |
| 15912 | /* And this is actually equivalent to having done the 2nd |
| 15913 | * pass now, so set it to not try again */ |
| 15914 | possible_end = (char *) -1; |
| 15915 | } |
| 15916 | } |
| 15917 | else { |
| 15918 | if (*p == ':') { |
| 15919 | has_terminating_colon = TRUE; |
| 15920 | } |
| 15921 | else if (*p == ';') { |
| 15922 | has_semi_colon = TRUE; |
| 15923 | has_terminating_colon = TRUE; |
| 15924 | } |
| 15925 | p++; |
| 15926 | } |
| 15927 | } |
| 15928 | |
| 15929 | try_posix: |
| 15930 | |
| 15931 | /* Here, we have a class name to look up. We can short circuit the |
| 15932 | * stuff below for short names that can't possibly be meant to be a |
| 15933 | * class name. (We can do this on the first pass, as any second pass |
| 15934 | * will yield an even shorter name) */ |
| 15935 | if (name_len < 3) { |
| 15936 | CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS); |
| 15937 | } |
| 15938 | |
| 15939 | /* Find which class it is. Initially switch on the length of the name. |
| 15940 | * */ |
| 15941 | switch (name_len) { |
| 15942 | case 4: |
| 15943 | if (memEQs(name_start, 4, "word")) { |
| 15944 | /* this is not POSIX, this is the Perl \w */ |
| 15945 | class_number = ANYOF_WORDCHAR; |
| 15946 | } |
| 15947 | break; |
| 15948 | case 5: |
| 15949 | /* Names all of length 5: alnum alpha ascii blank cntrl digit |
| 15950 | * graph lower print punct space upper |
| 15951 | * Offset 4 gives the best switch position. */ |
| 15952 | switch (name_start[4]) { |
| 15953 | case 'a': |
| 15954 | if (memBEGINs(name_start, 5, "alph")) /* alpha */ |
| 15955 | class_number = ANYOF_ALPHA; |
| 15956 | break; |
| 15957 | case 'e': |
| 15958 | if (memBEGINs(name_start, 5, "spac")) /* space */ |
| 15959 | class_number = ANYOF_SPACE; |
| 15960 | break; |
| 15961 | case 'h': |
| 15962 | if (memBEGINs(name_start, 5, "grap")) /* graph */ |
| 15963 | class_number = ANYOF_GRAPH; |
| 15964 | break; |
| 15965 | case 'i': |
| 15966 | if (memBEGINs(name_start, 5, "asci")) /* ascii */ |
| 15967 | class_number = ANYOF_ASCII; |
| 15968 | break; |
| 15969 | case 'k': |
| 15970 | if (memBEGINs(name_start, 5, "blan")) /* blank */ |
| 15971 | class_number = ANYOF_BLANK; |
| 15972 | break; |
| 15973 | case 'l': |
| 15974 | if (memBEGINs(name_start, 5, "cntr")) /* cntrl */ |
| 15975 | class_number = ANYOF_CNTRL; |
| 15976 | break; |
| 15977 | case 'm': |
| 15978 | if (memBEGINs(name_start, 5, "alnu")) /* alnum */ |
| 15979 | class_number = ANYOF_ALPHANUMERIC; |
| 15980 | break; |
| 15981 | case 'r': |
| 15982 | if (memBEGINs(name_start, 5, "lowe")) /* lower */ |
| 15983 | class_number = (FOLD) ? ANYOF_CASED : ANYOF_LOWER; |
| 15984 | else if (memBEGINs(name_start, 5, "uppe")) /* upper */ |
| 15985 | class_number = (FOLD) ? ANYOF_CASED : ANYOF_UPPER; |
| 15986 | break; |
| 15987 | case 't': |
| 15988 | if (memBEGINs(name_start, 5, "digi")) /* digit */ |
| 15989 | class_number = ANYOF_DIGIT; |
| 15990 | else if (memBEGINs(name_start, 5, "prin")) /* print */ |
| 15991 | class_number = ANYOF_PRINT; |
| 15992 | else if (memBEGINs(name_start, 5, "punc")) /* punct */ |
| 15993 | class_number = ANYOF_PUNCT; |
| 15994 | break; |
| 15995 | } |
| 15996 | break; |
| 15997 | case 6: |
| 15998 | if (memEQs(name_start, 6, "xdigit")) |
| 15999 | class_number = ANYOF_XDIGIT; |
| 16000 | break; |
| 16001 | } |
| 16002 | |
| 16003 | /* If the name exactly matches a posix class name the class number will |
| 16004 | * here be set to it, and the input almost certainly was meant to be a |
| 16005 | * posix class, so we can skip further checking. If instead the syntax |
| 16006 | * is exactly correct, but the name isn't one of the legal ones, we |
| 16007 | * will return that as an error below. But if neither of these apply, |
| 16008 | * it could be that no posix class was intended at all, or that one |
| 16009 | * was, but there was a typo. We tease these apart by doing fuzzy |
| 16010 | * matching on the name */ |
| 16011 | if (class_number == OOB_NAMEDCLASS && found_problem) { |
| 16012 | const UV posix_names[][6] = { |
| 16013 | { 'a', 'l', 'n', 'u', 'm' }, |
| 16014 | { 'a', 'l', 'p', 'h', 'a' }, |
| 16015 | { 'a', 's', 'c', 'i', 'i' }, |
| 16016 | { 'b', 'l', 'a', 'n', 'k' }, |
| 16017 | { 'c', 'n', 't', 'r', 'l' }, |
| 16018 | { 'd', 'i', 'g', 'i', 't' }, |
| 16019 | { 'g', 'r', 'a', 'p', 'h' }, |
| 16020 | { 'l', 'o', 'w', 'e', 'r' }, |
| 16021 | { 'p', 'r', 'i', 'n', 't' }, |
| 16022 | { 'p', 'u', 'n', 'c', 't' }, |
| 16023 | { 's', 'p', 'a', 'c', 'e' }, |
| 16024 | { 'u', 'p', 'p', 'e', 'r' }, |
| 16025 | { 'w', 'o', 'r', 'd' }, |
| 16026 | { 'x', 'd', 'i', 'g', 'i', 't' } |
| 16027 | }; |
| 16028 | /* The names of the above all have added NULs to make them the same |
| 16029 | * size, so we need to also have the real lengths */ |
| 16030 | const UV posix_name_lengths[] = { |
| 16031 | sizeof("alnum") - 1, |
| 16032 | sizeof("alpha") - 1, |
| 16033 | sizeof("ascii") - 1, |
| 16034 | sizeof("blank") - 1, |
| 16035 | sizeof("cntrl") - 1, |
| 16036 | sizeof("digit") - 1, |
| 16037 | sizeof("graph") - 1, |
| 16038 | sizeof("lower") - 1, |
| 16039 | sizeof("print") - 1, |
| 16040 | sizeof("punct") - 1, |
| 16041 | sizeof("space") - 1, |
| 16042 | sizeof("upper") - 1, |
| 16043 | sizeof("word") - 1, |
| 16044 | sizeof("xdigit")- 1 |
| 16045 | }; |
| 16046 | unsigned int i; |
| 16047 | int temp_max = max_distance; /* Use a temporary, so if we |
| 16048 | reparse, we haven't changed the |
| 16049 | outer one */ |
| 16050 | |
| 16051 | /* Use a smaller max edit distance if we are missing one of the |
| 16052 | * delimiters */ |
| 16053 | if ( has_opening_bracket + has_opening_colon < 2 |
| 16054 | || has_terminating_bracket + has_terminating_colon < 2) |
| 16055 | { |
| 16056 | temp_max--; |
| 16057 | } |
| 16058 | |
| 16059 | /* See if the input name is close to a legal one */ |
| 16060 | for (i = 0; i < C_ARRAY_LENGTH(posix_names); i++) { |
| 16061 | |
| 16062 | /* Short circuit call if the lengths are too far apart to be |
| 16063 | * able to match */ |
| 16064 | if (abs( (int) (name_len - posix_name_lengths[i])) |
| 16065 | > temp_max) |
| 16066 | { |
| 16067 | continue; |
| 16068 | } |
| 16069 | |
| 16070 | if (edit_distance(input_text, |
| 16071 | posix_names[i], |
| 16072 | name_len, |
| 16073 | posix_name_lengths[i], |
| 16074 | temp_max |
| 16075 | ) |
| 16076 | > -1) |
| 16077 | { /* If it is close, it probably was intended to be a class */ |
| 16078 | goto probably_meant_to_be; |
| 16079 | } |
| 16080 | } |
| 16081 | |
| 16082 | /* Here the input name is not close enough to a valid class name |
| 16083 | * for us to consider it to be intended to be a posix class. If |
| 16084 | * we haven't already done so, and the parse found a character that |
| 16085 | * could have been terminators for the name, but which we absorbed |
| 16086 | * as typos during the first pass, repeat the parse, signalling it |
| 16087 | * to stop at that character */ |
| 16088 | if (possible_end && possible_end != (char *) -1) { |
| 16089 | possible_end = (char *) -1; |
| 16090 | p = name_start; |
| 16091 | goto parse_name; |
| 16092 | } |
| 16093 | |
| 16094 | /* Here neither pass found a close-enough class name */ |
| 16095 | CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS); |
| 16096 | } |
| 16097 | |
| 16098 | probably_meant_to_be: |
| 16099 | |
| 16100 | /* Here we think that a posix specification was intended. Update any |
| 16101 | * parse pointer */ |
| 16102 | if (updated_parse_ptr) { |
| 16103 | *updated_parse_ptr = (char *) p; |
| 16104 | } |
| 16105 | |
| 16106 | /* If a posix class name was intended but incorrectly specified, we |
| 16107 | * output or return the warnings */ |
| 16108 | if (found_problem) { |
| 16109 | |
| 16110 | /* We set flags for these issues in the parse loop above instead of |
| 16111 | * adding them to the list of warnings, because we can parse it |
| 16112 | * twice, and we only want one warning instance */ |
| 16113 | if (has_upper) { |
| 16114 | ADD_POSIX_WARNING(p, "the name must be all lowercase letters"); |
| 16115 | } |
| 16116 | if (has_blank) { |
| 16117 | ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING); |
| 16118 | } |
| 16119 | if (has_semi_colon) { |
| 16120 | ADD_POSIX_WARNING(p, SEMI_COLON_POSIX_WARNING); |
| 16121 | } |
| 16122 | else if (! has_terminating_colon) { |
| 16123 | ADD_POSIX_WARNING(p, "there is no terminating ':'"); |
| 16124 | } |
| 16125 | if (! has_terminating_bracket) { |
| 16126 | ADD_POSIX_WARNING(p, "there is no terminating ']'"); |
| 16127 | } |
| 16128 | |
| 16129 | if ( posix_warnings |
| 16130 | && RExC_warn_text |
| 16131 | && av_top_index(RExC_warn_text) > -1) |
| 16132 | { |
| 16133 | *posix_warnings = RExC_warn_text; |
| 16134 | } |
| 16135 | } |
| 16136 | else if (class_number != OOB_NAMEDCLASS) { |
| 16137 | /* If it is a known class, return the class. The class number |
| 16138 | * #defines are structured so each complement is +1 to the normal |
| 16139 | * one */ |
| 16140 | CLEAR_POSIX_WARNINGS_AND_RETURN(class_number + complement); |
| 16141 | } |
| 16142 | else if (! check_only) { |
| 16143 | |
| 16144 | /* Here, it is an unrecognized class. This is an error (unless the |
| 16145 | * call is to check only, which we've already handled above) */ |
| 16146 | const char * const complement_string = (complement) |
| 16147 | ? "^" |
| 16148 | : ""; |
| 16149 | RExC_parse = (char *) p; |
| 16150 | vFAIL3utf8f("POSIX class [:%s%" UTF8f ":] unknown", |
| 16151 | complement_string, |
| 16152 | UTF8fARG(UTF, RExC_parse - name_start - 2, name_start)); |
| 16153 | } |
| 16154 | } |
| 16155 | |
| 16156 | return OOB_NAMEDCLASS; |
| 16157 | } |
| 16158 | #undef ADD_POSIX_WARNING |
| 16159 | |
| 16160 | STATIC unsigned int |
| 16161 | S_regex_set_precedence(const U8 my_operator) { |
| 16162 | |
| 16163 | /* Returns the precedence in the (?[...]) construct of the input operator, |
| 16164 | * specified by its character representation. The precedence follows |
| 16165 | * general Perl rules, but it extends this so that ')' and ']' have (low) |
| 16166 | * precedence even though they aren't really operators */ |
| 16167 | |
| 16168 | switch (my_operator) { |
| 16169 | case '!': |
| 16170 | return 5; |
| 16171 | case '&': |
| 16172 | return 4; |
| 16173 | case '^': |
| 16174 | case '|': |
| 16175 | case '+': |
| 16176 | case '-': |
| 16177 | return 3; |
| 16178 | case ')': |
| 16179 | return 2; |
| 16180 | case ']': |
| 16181 | return 1; |
| 16182 | } |
| 16183 | |
| 16184 | NOT_REACHED; /* NOTREACHED */ |
| 16185 | return 0; /* Silence compiler warning */ |
| 16186 | } |
| 16187 | |
| 16188 | STATIC regnode_offset |
| 16189 | S_handle_regex_sets(pTHX_ RExC_state_t *pRExC_state, SV** return_invlist, |
| 16190 | I32 *flagp, U32 depth, |
| 16191 | char * const oregcomp_parse) |
| 16192 | { |
| 16193 | /* Handle the (?[...]) construct to do set operations */ |
| 16194 | |
| 16195 | U8 curchar; /* Current character being parsed */ |
| 16196 | UV start, end; /* End points of code point ranges */ |
| 16197 | SV* final = NULL; /* The end result inversion list */ |
| 16198 | SV* result_string; /* 'final' stringified */ |
| 16199 | AV* stack; /* stack of operators and operands not yet |
| 16200 | resolved */ |
| 16201 | AV* fence_stack = NULL; /* A stack containing the positions in |
| 16202 | 'stack' of where the undealt-with left |
| 16203 | parens would be if they were actually |
| 16204 | put there */ |
| 16205 | /* The 'volatile' is a workaround for an optimiser bug |
| 16206 | * in Solaris Studio 12.3. See RT #127455 */ |
| 16207 | volatile IV fence = 0; /* Position of where most recent undealt- |
| 16208 | with left paren in stack is; -1 if none. |
| 16209 | */ |
| 16210 | STRLEN len; /* Temporary */ |
| 16211 | regnode_offset node; /* Temporary, and final regnode returned by |
| 16212 | this function */ |
| 16213 | const bool save_fold = FOLD; /* Temporary */ |
| 16214 | char *save_end, *save_parse; /* Temporaries */ |
| 16215 | const bool in_locale = LOC; /* we turn off /l during processing */ |
| 16216 | |
| 16217 | GET_RE_DEBUG_FLAGS_DECL; |
| 16218 | |
| 16219 | PERL_ARGS_ASSERT_HANDLE_REGEX_SETS; |
| 16220 | PERL_UNUSED_ARG(oregcomp_parse); /* Only for Set_Node_Length */ |
| 16221 | |
| 16222 | DEBUG_PARSE("xcls"); |
| 16223 | |
| 16224 | if (in_locale) { |
| 16225 | set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); |
| 16226 | } |
| 16227 | |
| 16228 | /* The use of this operator implies /u. This is required so that the |
| 16229 | * compile time values are valid in all runtime cases */ |
| 16230 | REQUIRE_UNI_RULES(flagp, 0); |
| 16231 | |
| 16232 | ckWARNexperimental(RExC_parse, |
| 16233 | WARN_EXPERIMENTAL__REGEX_SETS, |
| 16234 | "The regex_sets feature is experimental"); |
| 16235 | |
| 16236 | /* Everything in this construct is a metacharacter. Operands begin with |
| 16237 | * either a '\' (for an escape sequence), or a '[' for a bracketed |
| 16238 | * character class. Any other character should be an operator, or |
| 16239 | * parenthesis for grouping. Both types of operands are handled by calling |
| 16240 | * regclass() to parse them. It is called with a parameter to indicate to |
| 16241 | * return the computed inversion list. The parsing here is implemented via |
| 16242 | * a stack. Each entry on the stack is a single character representing one |
| 16243 | * of the operators; or else a pointer to an operand inversion list. */ |
| 16244 | |
| 16245 | #define IS_OPERATOR(a) SvIOK(a) |
| 16246 | #define IS_OPERAND(a) (! IS_OPERATOR(a)) |
| 16247 | |
| 16248 | /* The stack is kept in Łukasiewicz order. (That's pronounced similar |
| 16249 | * to luke-a-shave-itch (or -itz), but people who didn't want to bother |
| 16250 | * with pronouncing it called it Reverse Polish instead, but now that YOU |
| 16251 | * know how to pronounce it you can use the correct term, thus giving due |
| 16252 | * credit to the person who invented it, and impressing your geek friends. |
| 16253 | * Wikipedia says that the pronounciation of "Ł" has been changing so that |
| 16254 | * it is now more like an English initial W (as in wonk) than an L.) |
| 16255 | * |
| 16256 | * This means that, for example, 'a | b & c' is stored on the stack as |
| 16257 | * |
| 16258 | * c [4] |
| 16259 | * b [3] |
| 16260 | * & [2] |
| 16261 | * a [1] |
| 16262 | * | [0] |
| 16263 | * |
| 16264 | * where the numbers in brackets give the stack [array] element number. |
| 16265 | * In this implementation, parentheses are not stored on the stack. |
| 16266 | * Instead a '(' creates a "fence" so that the part of the stack below the |
| 16267 | * fence is invisible except to the corresponding ')' (this allows us to |
| 16268 | * replace testing for parens, by using instead subtraction of the fence |
| 16269 | * position). As new operands are processed they are pushed onto the stack |
| 16270 | * (except as noted in the next paragraph). New operators of higher |
| 16271 | * precedence than the current final one are inserted on the stack before |
| 16272 | * the lhs operand (so that when the rhs is pushed next, everything will be |
| 16273 | * in the correct positions shown above. When an operator of equal or |
| 16274 | * lower precedence is encountered in parsing, all the stacked operations |
| 16275 | * of equal or higher precedence are evaluated, leaving the result as the |
| 16276 | * top entry on the stack. This makes higher precedence operations |
| 16277 | * evaluate before lower precedence ones, and causes operations of equal |
| 16278 | * precedence to left associate. |
| 16279 | * |
| 16280 | * The only unary operator '!' is immediately pushed onto the stack when |
| 16281 | * encountered. When an operand is encountered, if the top of the stack is |
| 16282 | * a '!", the complement is immediately performed, and the '!' popped. The |
| 16283 | * resulting value is treated as a new operand, and the logic in the |
| 16284 | * previous paragraph is executed. Thus in the expression |
| 16285 | * [a] + ! [b] |
| 16286 | * the stack looks like |
| 16287 | * |
| 16288 | * ! |
| 16289 | * a |
| 16290 | * + |
| 16291 | * |
| 16292 | * as 'b' gets parsed, the latter gets evaluated to '!b', and the stack |
| 16293 | * becomes |
| 16294 | * |
| 16295 | * !b |
| 16296 | * a |
| 16297 | * + |
| 16298 | * |
| 16299 | * A ')' is treated as an operator with lower precedence than all the |
| 16300 | * aforementioned ones, which causes all operations on the stack above the |
| 16301 | * corresponding '(' to be evaluated down to a single resultant operand. |
| 16302 | * Then the fence for the '(' is removed, and the operand goes through the |
| 16303 | * algorithm above, without the fence. |
| 16304 | * |
| 16305 | * A separate stack is kept of the fence positions, so that the position of |
| 16306 | * the latest so-far unbalanced '(' is at the top of it. |
| 16307 | * |
| 16308 | * The ']' ending the construct is treated as the lowest operator of all, |
| 16309 | * so that everything gets evaluated down to a single operand, which is the |
| 16310 | * result */ |
| 16311 | |
| 16312 | sv_2mortal((SV *)(stack = newAV())); |
| 16313 | sv_2mortal((SV *)(fence_stack = newAV())); |
| 16314 | |
| 16315 | while (RExC_parse < RExC_end) { |
| 16316 | I32 top_index; /* Index of top-most element in 'stack' */ |
| 16317 | SV** top_ptr; /* Pointer to top 'stack' element */ |
| 16318 | SV* current = NULL; /* To contain the current inversion list |
| 16319 | operand */ |
| 16320 | SV* only_to_avoid_leaks; |
| 16321 | |
| 16322 | skip_to_be_ignored_text(pRExC_state, &RExC_parse, |
| 16323 | TRUE /* Force /x */ ); |
| 16324 | if (RExC_parse >= RExC_end) { /* Fail */ |
| 16325 | break; |
| 16326 | } |
| 16327 | |
| 16328 | curchar = UCHARAT(RExC_parse); |
| 16329 | |
| 16330 | redo_curchar: |
| 16331 | |
| 16332 | #ifdef ENABLE_REGEX_SETS_DEBUGGING |
| 16333 | /* Enable with -Accflags=-DENABLE_REGEX_SETS_DEBUGGING */ |
| 16334 | DEBUG_U(dump_regex_sets_structures(pRExC_state, |
| 16335 | stack, fence, fence_stack)); |
| 16336 | #endif |
| 16337 | |
| 16338 | top_index = av_tindex_skip_len_mg(stack); |
| 16339 | |
| 16340 | switch (curchar) { |
| 16341 | SV** stacked_ptr; /* Ptr to something already on 'stack' */ |
| 16342 | char stacked_operator; /* The topmost operator on the 'stack'. */ |
| 16343 | SV* lhs; /* Operand to the left of the operator */ |
| 16344 | SV* rhs; /* Operand to the right of the operator */ |
| 16345 | SV* fence_ptr; /* Pointer to top element of the fence |
| 16346 | stack */ |
| 16347 | case '(': |
| 16348 | |
| 16349 | if ( RExC_parse < RExC_end - 2 |
| 16350 | && UCHARAT(RExC_parse + 1) == '?' |
| 16351 | && UCHARAT(RExC_parse + 2) == '^') |
| 16352 | { |
| 16353 | const regnode_offset orig_emit = RExC_emit; |
| 16354 | SV * resultant_invlist; |
| 16355 | |
| 16356 | /* If is a '(?^', could be an embedded '(?^flags:(?[...])'. |
| 16357 | * This happens when we have some thing like |
| 16358 | * |
| 16359 | * my $thai_or_lao = qr/(?[ \p{Thai} + \p{Lao} ])/; |
| 16360 | * ... |
| 16361 | * qr/(?[ \p{Digit} & $thai_or_lao ])/; |
| 16362 | * |
| 16363 | * Here we would be handling the interpolated |
| 16364 | * '$thai_or_lao'. We handle this by a recursive call to |
| 16365 | * reg which returns the inversion list the |
| 16366 | * interpolated expression evaluates to. Actually, the |
| 16367 | * return is a special regnode containing a pointer to that |
| 16368 | * inversion list. If the return isn't that regnode alone, |
| 16369 | * we know that this wasn't such an interpolation, which is |
| 16370 | * an error: we need to get a single inversion list back |
| 16371 | * from the recursion */ |
| 16372 | |
| 16373 | RExC_parse++; |
| 16374 | RExC_sets_depth++; |
| 16375 | |
| 16376 | node = reg(pRExC_state, 2, flagp, depth+1); |
| 16377 | RETURN_FAIL_ON_RESTART(*flagp, flagp); |
| 16378 | |
| 16379 | if ( OP(REGNODE_p(node)) != REGEX_SET |
| 16380 | /* If more than a single node returned, the nested |
| 16381 | * parens evaluated to more than just a (?[...]), |
| 16382 | * which isn't legal */ |
| 16383 | || node != 1) { |
| 16384 | vFAIL("Expecting interpolated extended charclass"); |
| 16385 | } |
| 16386 | resultant_invlist = (SV *) ARGp(REGNODE_p(node)); |
| 16387 | current = invlist_clone(resultant_invlist, NULL); |
| 16388 | SvREFCNT_dec(resultant_invlist); |
| 16389 | |
| 16390 | RExC_sets_depth--; |
| 16391 | RExC_emit = orig_emit; |
| 16392 | goto handle_operand; |
| 16393 | } |
| 16394 | |
| 16395 | /* A regular '('. Look behind for illegal syntax */ |
| 16396 | if (top_index - fence >= 0) { |
| 16397 | /* If the top entry on the stack is an operator, it had |
| 16398 | * better be a '!', otherwise the entry below the top |
| 16399 | * operand should be an operator */ |
| 16400 | if ( ! (top_ptr = av_fetch(stack, top_index, FALSE)) |
| 16401 | || (IS_OPERATOR(*top_ptr) && SvUV(*top_ptr) != '!') |
| 16402 | || ( IS_OPERAND(*top_ptr) |
| 16403 | && ( top_index - fence < 1 |
| 16404 | || ! (stacked_ptr = av_fetch(stack, |
| 16405 | top_index - 1, |
| 16406 | FALSE)) |
| 16407 | || ! IS_OPERATOR(*stacked_ptr)))) |
| 16408 | { |
| 16409 | RExC_parse++; |
| 16410 | vFAIL("Unexpected '(' with no preceding operator"); |
| 16411 | } |
| 16412 | } |
| 16413 | |
| 16414 | /* Stack the position of this undealt-with left paren */ |
| 16415 | av_push(fence_stack, newSViv(fence)); |
| 16416 | fence = top_index + 1; |
| 16417 | break; |
| 16418 | |
| 16419 | case '\\': |
| 16420 | /* regclass() can only return RESTART_PARSE and NEED_UTF8 if |
| 16421 | * multi-char folds are allowed. */ |
| 16422 | if (!regclass(pRExC_state, flagp, depth+1, |
| 16423 | TRUE, /* means parse just the next thing */ |
| 16424 | FALSE, /* don't allow multi-char folds */ |
| 16425 | FALSE, /* don't silence non-portable warnings. */ |
| 16426 | TRUE, /* strict */ |
| 16427 | FALSE, /* Require return to be an ANYOF */ |
| 16428 | ¤t)) |
| 16429 | { |
| 16430 | RETURN_FAIL_ON_RESTART(*flagp, flagp); |
| 16431 | goto regclass_failed; |
| 16432 | } |
| 16433 | |
| 16434 | /* regclass() will return with parsing just the \ sequence, |
| 16435 | * leaving the parse pointer at the next thing to parse */ |
| 16436 | RExC_parse--; |
| 16437 | goto handle_operand; |
| 16438 | |
| 16439 | case '[': /* Is a bracketed character class */ |
| 16440 | { |
| 16441 | /* See if this is a [:posix:] class. */ |
| 16442 | bool is_posix_class = (OOB_NAMEDCLASS |
| 16443 | < handle_possible_posix(pRExC_state, |
| 16444 | RExC_parse + 1, |
| 16445 | NULL, |
| 16446 | NULL, |
| 16447 | TRUE /* checking only */)); |
| 16448 | /* If it is a posix class, leave the parse pointer at the '[' |
| 16449 | * to fool regclass() into thinking it is part of a |
| 16450 | * '[[:posix:]]'. */ |
| 16451 | if (! is_posix_class) { |
| 16452 | RExC_parse++; |
| 16453 | } |
| 16454 | |
| 16455 | /* regclass() can only return RESTART_PARSE and NEED_UTF8 if |
| 16456 | * multi-char folds are allowed. */ |
| 16457 | if (!regclass(pRExC_state, flagp, depth+1, |
| 16458 | is_posix_class, /* parse the whole char |
| 16459 | class only if not a |
| 16460 | posix class */ |
| 16461 | FALSE, /* don't allow multi-char folds */ |
| 16462 | TRUE, /* silence non-portable warnings. */ |
| 16463 | TRUE, /* strict */ |
| 16464 | FALSE, /* Require return to be an ANYOF */ |
| 16465 | ¤t)) |
| 16466 | { |
| 16467 | RETURN_FAIL_ON_RESTART(*flagp, flagp); |
| 16468 | goto regclass_failed; |
| 16469 | } |
| 16470 | |
| 16471 | if (! current) { |
| 16472 | break; |
| 16473 | } |
| 16474 | |
| 16475 | /* function call leaves parse pointing to the ']', except if we |
| 16476 | * faked it */ |
| 16477 | if (is_posix_class) { |
| 16478 | RExC_parse--; |
| 16479 | } |
| 16480 | |
| 16481 | goto handle_operand; |
| 16482 | } |
| 16483 | |
| 16484 | case ']': |
| 16485 | if (top_index >= 1) { |
| 16486 | goto join_operators; |
| 16487 | } |
| 16488 | |
| 16489 | /* Only a single operand on the stack: are done */ |
| 16490 | goto done; |
| 16491 | |
| 16492 | case ')': |
| 16493 | if (av_tindex_skip_len_mg(fence_stack) < 0) { |
| 16494 | if (UCHARAT(RExC_parse - 1) == ']') { |
| 16495 | break; |
| 16496 | } |
| 16497 | RExC_parse++; |
| 16498 | vFAIL("Unexpected ')'"); |
| 16499 | } |
| 16500 | |
| 16501 | /* If nothing after the fence, is missing an operand */ |
| 16502 | if (top_index - fence < 0) { |
| 16503 | RExC_parse++; |
| 16504 | goto bad_syntax; |
| 16505 | } |
| 16506 | /* If at least two things on the stack, treat this as an |
| 16507 | * operator */ |
| 16508 | if (top_index - fence >= 1) { |
| 16509 | goto join_operators; |
| 16510 | } |
| 16511 | |
| 16512 | /* Here only a single thing on the fenced stack, and there is a |
| 16513 | * fence. Get rid of it */ |
| 16514 | fence_ptr = av_pop(fence_stack); |
| 16515 | assert(fence_ptr); |
| 16516 | fence = SvIV(fence_ptr); |
| 16517 | SvREFCNT_dec_NN(fence_ptr); |
| 16518 | fence_ptr = NULL; |
| 16519 | |
| 16520 | if (fence < 0) { |
| 16521 | fence = 0; |
| 16522 | } |
| 16523 | |
| 16524 | /* Having gotten rid of the fence, we pop the operand at the |
| 16525 | * stack top and process it as a newly encountered operand */ |
| 16526 | current = av_pop(stack); |
| 16527 | if (IS_OPERAND(current)) { |
| 16528 | goto handle_operand; |
| 16529 | } |
| 16530 | |
| 16531 | RExC_parse++; |
| 16532 | goto bad_syntax; |
| 16533 | |
| 16534 | case '&': |
| 16535 | case '|': |
| 16536 | case '+': |
| 16537 | case '-': |
| 16538 | case '^': |
| 16539 | |
| 16540 | /* These binary operators should have a left operand already |
| 16541 | * parsed */ |
| 16542 | if ( top_index - fence < 0 |
| 16543 | || top_index - fence == 1 |
| 16544 | || ( ! (top_ptr = av_fetch(stack, top_index, FALSE))) |
| 16545 | || ! IS_OPERAND(*top_ptr)) |
| 16546 | { |
| 16547 | goto unexpected_binary; |
| 16548 | } |
| 16549 | |
| 16550 | /* If only the one operand is on the part of the stack visible |
| 16551 | * to us, we just place this operator in the proper position */ |
| 16552 | if (top_index - fence < 2) { |
| 16553 | |
| 16554 | /* Place the operator before the operand */ |
| 16555 | |
| 16556 | SV* lhs = av_pop(stack); |
| 16557 | av_push(stack, newSVuv(curchar)); |
| 16558 | av_push(stack, lhs); |
| 16559 | break; |
| 16560 | } |
| 16561 | |
| 16562 | /* But if there is something else on the stack, we need to |
| 16563 | * process it before this new operator if and only if the |
| 16564 | * stacked operation has equal or higher precedence than the |
| 16565 | * new one */ |
| 16566 | |
| 16567 | join_operators: |
| 16568 | |
| 16569 | /* The operator on the stack is supposed to be below both its |
| 16570 | * operands */ |
| 16571 | if ( ! (stacked_ptr = av_fetch(stack, top_index - 2, FALSE)) |
| 16572 | || IS_OPERAND(*stacked_ptr)) |
| 16573 | { |
| 16574 | /* But if not, it's legal and indicates we are completely |
| 16575 | * done if and only if we're currently processing a ']', |
| 16576 | * which should be the final thing in the expression */ |
| 16577 | if (curchar == ']') { |
| 16578 | goto done; |
| 16579 | } |
| 16580 | |
| 16581 | unexpected_binary: |
| 16582 | RExC_parse++; |
| 16583 | vFAIL2("Unexpected binary operator '%c' with no " |
| 16584 | "preceding operand", curchar); |
| 16585 | } |
| 16586 | stacked_operator = (char) SvUV(*stacked_ptr); |
| 16587 | |
| 16588 | if (regex_set_precedence(curchar) |
| 16589 | > regex_set_precedence(stacked_operator)) |
| 16590 | { |
| 16591 | /* Here, the new operator has higher precedence than the |
| 16592 | * stacked one. This means we need to add the new one to |
| 16593 | * the stack to await its rhs operand (and maybe more |
| 16594 | * stuff). We put it before the lhs operand, leaving |
| 16595 | * untouched the stacked operator and everything below it |
| 16596 | * */ |
| 16597 | lhs = av_pop(stack); |
| 16598 | assert(IS_OPERAND(lhs)); |
| 16599 | |
| 16600 | av_push(stack, newSVuv(curchar)); |
| 16601 | av_push(stack, lhs); |
| 16602 | break; |
| 16603 | } |
| 16604 | |
| 16605 | /* Here, the new operator has equal or lower precedence than |
| 16606 | * what's already there. This means the operation already |
| 16607 | * there should be performed now, before the new one. */ |
| 16608 | |
| 16609 | rhs = av_pop(stack); |
| 16610 | if (! IS_OPERAND(rhs)) { |
| 16611 | |
| 16612 | /* This can happen when a ! is not followed by an operand, |
| 16613 | * like in /(?[\t &!])/ */ |
| 16614 | goto bad_syntax; |
| 16615 | } |
| 16616 | |
| 16617 | lhs = av_pop(stack); |
| 16618 | |
| 16619 | if (! IS_OPERAND(lhs)) { |
| 16620 | |
| 16621 | /* This can happen when there is an empty (), like in |
| 16622 | * /(?[[0]+()+])/ */ |
| 16623 | goto bad_syntax; |
| 16624 | } |
| 16625 | |
| 16626 | switch (stacked_operator) { |
| 16627 | case '&': |
| 16628 | _invlist_intersection(lhs, rhs, &rhs); |
| 16629 | break; |
| 16630 | |
| 16631 | case '|': |
| 16632 | case '+': |
| 16633 | _invlist_union(lhs, rhs, &rhs); |
| 16634 | break; |
| 16635 | |
| 16636 | case '-': |
| 16637 | _invlist_subtract(lhs, rhs, &rhs); |
| 16638 | break; |
| 16639 | |
| 16640 | case '^': /* The union minus the intersection */ |
| 16641 | { |
| 16642 | SV* i = NULL; |
| 16643 | SV* u = NULL; |
| 16644 | |
| 16645 | _invlist_union(lhs, rhs, &u); |
| 16646 | _invlist_intersection(lhs, rhs, &i); |
| 16647 | _invlist_subtract(u, i, &rhs); |
| 16648 | SvREFCNT_dec_NN(i); |
| 16649 | SvREFCNT_dec_NN(u); |
| 16650 | break; |
| 16651 | } |
| 16652 | } |
| 16653 | SvREFCNT_dec(lhs); |
| 16654 | |
| 16655 | /* Here, the higher precedence operation has been done, and the |
| 16656 | * result is in 'rhs'. We overwrite the stacked operator with |
| 16657 | * the result. Then we redo this code to either push the new |
| 16658 | * operator onto the stack or perform any higher precedence |
| 16659 | * stacked operation */ |
| 16660 | only_to_avoid_leaks = av_pop(stack); |
| 16661 | SvREFCNT_dec(only_to_avoid_leaks); |
| 16662 | av_push(stack, rhs); |
| 16663 | goto redo_curchar; |
| 16664 | |
| 16665 | case '!': /* Highest priority, right associative */ |
| 16666 | |
| 16667 | /* If what's already at the top of the stack is another '!", |
| 16668 | * they just cancel each other out */ |
| 16669 | if ( (top_ptr = av_fetch(stack, top_index, FALSE)) |
| 16670 | && (IS_OPERATOR(*top_ptr) && SvUV(*top_ptr) == '!')) |
| 16671 | { |
| 16672 | only_to_avoid_leaks = av_pop(stack); |
| 16673 | SvREFCNT_dec(only_to_avoid_leaks); |
| 16674 | } |
| 16675 | else { /* Otherwise, since it's right associative, just push |
| 16676 | onto the stack */ |
| 16677 | av_push(stack, newSVuv(curchar)); |
| 16678 | } |
| 16679 | break; |
| 16680 | |
| 16681 | default: |
| 16682 | RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1; |
| 16683 | if (RExC_parse >= RExC_end) { |
| 16684 | break; |
| 16685 | } |
| 16686 | vFAIL("Unexpected character"); |
| 16687 | |
| 16688 | handle_operand: |
| 16689 | |
| 16690 | /* Here 'current' is the operand. If something is already on the |
| 16691 | * stack, we have to check if it is a !. But first, the code above |
| 16692 | * may have altered the stack in the time since we earlier set |
| 16693 | * 'top_index'. */ |
| 16694 | |
| 16695 | top_index = av_tindex_skip_len_mg(stack); |
| 16696 | if (top_index - fence >= 0) { |
| 16697 | /* If the top entry on the stack is an operator, it had better |
| 16698 | * be a '!', otherwise the entry below the top operand should |
| 16699 | * be an operator */ |
| 16700 | top_ptr = av_fetch(stack, top_index, FALSE); |
| 16701 | assert(top_ptr); |
| 16702 | if (IS_OPERATOR(*top_ptr)) { |
| 16703 | |
| 16704 | /* The only permissible operator at the top of the stack is |
| 16705 | * '!', which is applied immediately to this operand. */ |
| 16706 | curchar = (char) SvUV(*top_ptr); |
| 16707 | if (curchar != '!') { |
| 16708 | SvREFCNT_dec(current); |
| 16709 | vFAIL2("Unexpected binary operator '%c' with no " |
| 16710 | "preceding operand", curchar); |
| 16711 | } |
| 16712 | |
| 16713 | _invlist_invert(current); |
| 16714 | |
| 16715 | only_to_avoid_leaks = av_pop(stack); |
| 16716 | SvREFCNT_dec(only_to_avoid_leaks); |
| 16717 | |
| 16718 | /* And we redo with the inverted operand. This allows |
| 16719 | * handling multiple ! in a row */ |
| 16720 | goto handle_operand; |
| 16721 | } |
| 16722 | /* Single operand is ok only for the non-binary ')' |
| 16723 | * operator */ |
| 16724 | else if ((top_index - fence == 0 && curchar != ')') |
| 16725 | || (top_index - fence > 0 |
| 16726 | && (! (stacked_ptr = av_fetch(stack, |
| 16727 | top_index - 1, |
| 16728 | FALSE)) |
| 16729 | || IS_OPERAND(*stacked_ptr)))) |
| 16730 | { |
| 16731 | SvREFCNT_dec(current); |
| 16732 | vFAIL("Operand with no preceding operator"); |
| 16733 | } |
| 16734 | } |
| 16735 | |
| 16736 | /* Here there was nothing on the stack or the top element was |
| 16737 | * another operand. Just add this new one */ |
| 16738 | av_push(stack, current); |
| 16739 | |
| 16740 | } /* End of switch on next parse token */ |
| 16741 | |
| 16742 | RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1; |
| 16743 | } /* End of loop parsing through the construct */ |
| 16744 | |
| 16745 | vFAIL("Syntax error in (?[...])"); |
| 16746 | |
| 16747 | done: |
| 16748 | |
| 16749 | if (RExC_parse >= RExC_end || RExC_parse[1] != ')') { |
| 16750 | if (RExC_parse < RExC_end) { |
| 16751 | RExC_parse++; |
| 16752 | } |
| 16753 | |
| 16754 | vFAIL("Unexpected ']' with no following ')' in (?[..."); |
| 16755 | } |
| 16756 | |
| 16757 | if (av_tindex_skip_len_mg(fence_stack) >= 0) { |
| 16758 | vFAIL("Unmatched ("); |
| 16759 | } |
| 16760 | |
| 16761 | if (av_tindex_skip_len_mg(stack) < 0 /* Was empty */ |
| 16762 | || ((final = av_pop(stack)) == NULL) |
| 16763 | || ! IS_OPERAND(final) |
| 16764 | || ! is_invlist(final) |
| 16765 | || av_tindex_skip_len_mg(stack) >= 0) /* More left on stack */ |
| 16766 | { |
| 16767 | bad_syntax: |
| 16768 | SvREFCNT_dec(final); |
| 16769 | vFAIL("Incomplete expression within '(?[ ])'"); |
| 16770 | } |
| 16771 | |
| 16772 | /* Here, 'final' is the resultant inversion list from evaluating the |
| 16773 | * expression. Return it if so requested */ |
| 16774 | if (return_invlist) { |
| 16775 | *return_invlist = final; |
| 16776 | return END; |
| 16777 | } |
| 16778 | |
| 16779 | if (RExC_sets_depth) { /* If within a recursive call, return in a special |
| 16780 | regnode */ |
| 16781 | RExC_parse++; |
| 16782 | node = regpnode(pRExC_state, REGEX_SET, (void *) final); |
| 16783 | } |
| 16784 | else { |
| 16785 | |
| 16786 | /* Otherwise generate a resultant node, based on 'final'. regclass() |
| 16787 | * is expecting a string of ranges and individual code points */ |
| 16788 | invlist_iterinit(final); |
| 16789 | result_string = newSVpvs(""); |
| 16790 | while (invlist_iternext(final, &start, &end)) { |
| 16791 | if (start == end) { |
| 16792 | Perl_sv_catpvf(aTHX_ result_string, "\\x{%" UVXf "}", start); |
| 16793 | } |
| 16794 | else { |
| 16795 | Perl_sv_catpvf(aTHX_ result_string, "\\x{%" UVXf "}-\\x{%" |
| 16796 | UVXf "}", start, end); |
| 16797 | } |
| 16798 | } |
| 16799 | |
| 16800 | /* About to generate an ANYOF (or similar) node from the inversion list |
| 16801 | * we have calculated */ |
| 16802 | save_parse = RExC_parse; |
| 16803 | RExC_parse = SvPV(result_string, len); |
| 16804 | save_end = RExC_end; |
| 16805 | RExC_end = RExC_parse + len; |
| 16806 | TURN_OFF_WARNINGS_IN_SUBSTITUTE_PARSE; |
| 16807 | |
| 16808 | /* We turn off folding around the call, as the class we have |
| 16809 | * constructed already has all folding taken into consideration, and we |
| 16810 | * don't want regclass() to add to that */ |
| 16811 | RExC_flags &= ~RXf_PMf_FOLD; |
| 16812 | /* regclass() can only return RESTART_PARSE and NEED_UTF8 if multi-char |
| 16813 | * folds are allowed. */ |
| 16814 | node = regclass(pRExC_state, flagp, depth+1, |
| 16815 | FALSE, /* means parse the whole char class */ |
| 16816 | FALSE, /* don't allow multi-char folds */ |
| 16817 | TRUE, /* silence non-portable warnings. The above may |
| 16818 | very well have generated non-portable code |
| 16819 | points, but they're valid on this machine */ |
| 16820 | FALSE, /* similarly, no need for strict */ |
| 16821 | |
| 16822 | /* We can optimize into something besides an ANYOF, |
| 16823 | * except under /l, which needs to be ANYOF because of |
| 16824 | * runtime checks for locale sanity, etc */ |
| 16825 | ! in_locale, |
| 16826 | NULL |
| 16827 | ); |
| 16828 | |
| 16829 | RESTORE_WARNINGS; |
| 16830 | RExC_parse = save_parse + 1; |
| 16831 | RExC_end = save_end; |
| 16832 | SvREFCNT_dec_NN(final); |
| 16833 | SvREFCNT_dec_NN(result_string); |
| 16834 | |
| 16835 | if (save_fold) { |
| 16836 | RExC_flags |= RXf_PMf_FOLD; |
| 16837 | } |
| 16838 | |
| 16839 | if (!node) { |
| 16840 | RETURN_FAIL_ON_RESTART(*flagp, flagp); |
| 16841 | goto regclass_failed; |
| 16842 | } |
| 16843 | |
| 16844 | /* Fix up the node type if we are in locale. (We have pretended we are |
| 16845 | * under /u for the purposes of regclass(), as this construct will only |
| 16846 | * work under UTF-8 locales. But now we change the opcode to be ANYOFL |
| 16847 | * (so as to cause any warnings about bad locales to be output in |
| 16848 | * regexec.c), and add the flag that indicates to check if not in a |
| 16849 | * UTF-8 locale. The reason we above forbid optimization into |
| 16850 | * something other than an ANYOF node is simply to minimize the number |
| 16851 | * of code changes in regexec.c. Otherwise we would have to create new |
| 16852 | * EXACTish node types and deal with them. This decision could be |
| 16853 | * revisited should this construct become popular. |
| 16854 | * |
| 16855 | * (One might think we could look at the resulting ANYOF node and |
| 16856 | * suppress the flag if everything is above 255, as those would be |
| 16857 | * UTF-8 only, but this isn't true, as the components that led to that |
| 16858 | * result could have been locale-affected, and just happen to cancel |
| 16859 | * each other out under UTF-8 locales.) */ |
| 16860 | if (in_locale) { |
| 16861 | set_regex_charset(&RExC_flags, REGEX_LOCALE_CHARSET); |
| 16862 | |
| 16863 | assert(OP(REGNODE_p(node)) == ANYOF); |
| 16864 | |
| 16865 | OP(REGNODE_p(node)) = ANYOFL; |
| 16866 | ANYOF_FLAGS(REGNODE_p(node)) |
| 16867 | |= ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD; |
| 16868 | } |
| 16869 | } |
| 16870 | |
| 16871 | nextchar(pRExC_state); |
| 16872 | Set_Node_Length(REGNODE_p(node), RExC_parse - oregcomp_parse + 1); /* MJD */ |
| 16873 | return node; |
| 16874 | |
| 16875 | regclass_failed: |
| 16876 | FAIL2("panic: regclass returned failure to handle_sets, " "flags=%#" UVxf, |
| 16877 | (UV) *flagp); |
| 16878 | } |
| 16879 | |
| 16880 | #ifdef ENABLE_REGEX_SETS_DEBUGGING |
| 16881 | |
| 16882 | STATIC void |
| 16883 | S_dump_regex_sets_structures(pTHX_ RExC_state_t *pRExC_state, |
| 16884 | AV * stack, const IV fence, AV * fence_stack) |
| 16885 | { /* Dumps the stacks in handle_regex_sets() */ |
| 16886 | |
| 16887 | const SSize_t stack_top = av_tindex_skip_len_mg(stack); |
| 16888 | const SSize_t fence_stack_top = av_tindex_skip_len_mg(fence_stack); |
| 16889 | SSize_t i; |
| 16890 | |
| 16891 | PERL_ARGS_ASSERT_DUMP_REGEX_SETS_STRUCTURES; |
| 16892 | |
| 16893 | PerlIO_printf(Perl_debug_log, "\nParse position is:%s\n", RExC_parse); |
| 16894 | |
| 16895 | if (stack_top < 0) { |
| 16896 | PerlIO_printf(Perl_debug_log, "Nothing on stack\n"); |
| 16897 | } |
| 16898 | else { |
| 16899 | PerlIO_printf(Perl_debug_log, "Stack: (fence=%d)\n", (int) fence); |
| 16900 | for (i = stack_top; i >= 0; i--) { |
| 16901 | SV ** element_ptr = av_fetch(stack, i, FALSE); |
| 16902 | if (! element_ptr) { |
| 16903 | } |
| 16904 | |
| 16905 | if (IS_OPERATOR(*element_ptr)) { |
| 16906 | PerlIO_printf(Perl_debug_log, "[%d]: %c\n", |
| 16907 | (int) i, (int) SvIV(*element_ptr)); |
| 16908 | } |
| 16909 | else { |
| 16910 | PerlIO_printf(Perl_debug_log, "[%d] ", (int) i); |
| 16911 | sv_dump(*element_ptr); |
| 16912 | } |
| 16913 | } |
| 16914 | } |
| 16915 | |
| 16916 | if (fence_stack_top < 0) { |
| 16917 | PerlIO_printf(Perl_debug_log, "Nothing on fence_stack\n"); |
| 16918 | } |
| 16919 | else { |
| 16920 | PerlIO_printf(Perl_debug_log, "Fence_stack: \n"); |
| 16921 | for (i = fence_stack_top; i >= 0; i--) { |
| 16922 | SV ** element_ptr = av_fetch(fence_stack, i, FALSE); |
| 16923 | if (! element_ptr) { |
| 16924 | } |
| 16925 | |
| 16926 | PerlIO_printf(Perl_debug_log, "[%d]: %d\n", |
| 16927 | (int) i, (int) SvIV(*element_ptr)); |
| 16928 | } |
| 16929 | } |
| 16930 | } |
| 16931 | |
| 16932 | #endif |
| 16933 | |
| 16934 | #undef IS_OPERATOR |
| 16935 | #undef IS_OPERAND |
| 16936 | |
| 16937 | STATIC void |
| 16938 | S_add_above_Latin1_folds(pTHX_ RExC_state_t *pRExC_state, const U8 cp, SV** invlist) |
| 16939 | { |
| 16940 | /* This adds the Latin1/above-Latin1 folding rules. |
| 16941 | * |
| 16942 | * This should be called only for a Latin1-range code points, cp, which is |
| 16943 | * known to be involved in a simple fold with other code points above |
| 16944 | * Latin1. It would give false results if /aa has been specified. |
| 16945 | * Multi-char folds are outside the scope of this, and must be handled |
| 16946 | * specially. */ |
| 16947 | |
| 16948 | PERL_ARGS_ASSERT_ADD_ABOVE_LATIN1_FOLDS; |
| 16949 | |
| 16950 | assert(HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(cp)); |
| 16951 | |
| 16952 | /* The rules that are valid for all Unicode versions are hard-coded in */ |
| 16953 | switch (cp) { |
| 16954 | case 'k': |
| 16955 | case 'K': |
| 16956 | *invlist = |
| 16957 | add_cp_to_invlist(*invlist, KELVIN_SIGN); |
| 16958 | break; |
| 16959 | case 's': |
| 16960 | case 'S': |
| 16961 | *invlist = add_cp_to_invlist(*invlist, LATIN_SMALL_LETTER_LONG_S); |
| 16962 | break; |
| 16963 | case MICRO_SIGN: |
| 16964 | *invlist = add_cp_to_invlist(*invlist, GREEK_CAPITAL_LETTER_MU); |
| 16965 | *invlist = add_cp_to_invlist(*invlist, GREEK_SMALL_LETTER_MU); |
| 16966 | break; |
| 16967 | case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE: |
| 16968 | case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE: |
| 16969 | *invlist = add_cp_to_invlist(*invlist, ANGSTROM_SIGN); |
| 16970 | break; |
| 16971 | case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: |
| 16972 | *invlist = add_cp_to_invlist(*invlist, |
| 16973 | LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS); |
| 16974 | break; |
| 16975 | |
| 16976 | default: /* Other code points are checked against the data for the |
| 16977 | current Unicode version */ |
| 16978 | { |
| 16979 | Size_t folds_count; |
| 16980 | U32 first_fold; |
| 16981 | const U32 * remaining_folds; |
| 16982 | UV folded_cp; |
| 16983 | |
| 16984 | if (isASCII(cp)) { |
| 16985 | folded_cp = toFOLD(cp); |
| 16986 | } |
| 16987 | else { |
| 16988 | U8 dummy_fold[UTF8_MAXBYTES_CASE+1]; |
| 16989 | Size_t dummy_len; |
| 16990 | folded_cp = _to_fold_latin1(cp, dummy_fold, &dummy_len, 0); |
| 16991 | } |
| 16992 | |
| 16993 | if (folded_cp > 255) { |
| 16994 | *invlist = add_cp_to_invlist(*invlist, folded_cp); |
| 16995 | } |
| 16996 | |
| 16997 | folds_count = _inverse_folds(folded_cp, &first_fold, |
| 16998 | &remaining_folds); |
| 16999 | if (folds_count == 0) { |
| 17000 | |
| 17001 | /* Use deprecated warning to increase the chances of this being |
| 17002 | * output */ |
| 17003 | ckWARN2reg_d(RExC_parse, |
| 17004 | "Perl folding rules are not up-to-date for 0x%02X;" |
| 17005 | " please use the perlbug utility to report;", cp); |
| 17006 | } |
| 17007 | else { |
| 17008 | unsigned int i; |
| 17009 | |
| 17010 | if (first_fold > 255) { |
| 17011 | *invlist = add_cp_to_invlist(*invlist, first_fold); |
| 17012 | } |
| 17013 | for (i = 0; i < folds_count - 1; i++) { |
| 17014 | if (remaining_folds[i] > 255) { |
| 17015 | *invlist = add_cp_to_invlist(*invlist, |
| 17016 | remaining_folds[i]); |
| 17017 | } |
| 17018 | } |
| 17019 | } |
| 17020 | break; |
| 17021 | } |
| 17022 | } |
| 17023 | } |
| 17024 | |
| 17025 | STATIC void |
| 17026 | S_output_posix_warnings(pTHX_ RExC_state_t *pRExC_state, AV* posix_warnings) |
| 17027 | { |
| 17028 | /* Output the elements of the array given by '*posix_warnings' as REGEXP |
| 17029 | * warnings. */ |
| 17030 | |
| 17031 | SV * msg; |
| 17032 | const bool first_is_fatal = ckDEAD(packWARN(WARN_REGEXP)); |
| 17033 | |
| 17034 | PERL_ARGS_ASSERT_OUTPUT_POSIX_WARNINGS; |
| 17035 | |
| 17036 | if (! TO_OUTPUT_WARNINGS(RExC_parse)) { |
| 17037 | CLEAR_POSIX_WARNINGS(); |
| 17038 | return; |
| 17039 | } |
| 17040 | |
| 17041 | while ((msg = av_shift(posix_warnings)) != &PL_sv_undef) { |
| 17042 | if (first_is_fatal) { /* Avoid leaking this */ |
| 17043 | av_undef(posix_warnings); /* This isn't necessary if the |
| 17044 | array is mortal, but is a |
| 17045 | fail-safe */ |
| 17046 | (void) sv_2mortal(msg); |
| 17047 | PREPARE_TO_DIE; |
| 17048 | } |
| 17049 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s", SvPVX(msg)); |
| 17050 | SvREFCNT_dec_NN(msg); |
| 17051 | } |
| 17052 | |
| 17053 | UPDATE_WARNINGS_LOC(RExC_parse); |
| 17054 | } |
| 17055 | |
| 17056 | PERL_STATIC_INLINE Size_t |
| 17057 | S_find_first_differing_byte_pos(const U8 * s1, const U8 * s2, const Size_t max) |
| 17058 | { |
| 17059 | const U8 * const start = s1; |
| 17060 | const U8 * const send = start + max; |
| 17061 | |
| 17062 | PERL_ARGS_ASSERT_FIND_FIRST_DIFFERING_BYTE_POS; |
| 17063 | |
| 17064 | while (s1 < send && *s1 == *s2) { |
| 17065 | s1++; s2++; |
| 17066 | } |
| 17067 | |
| 17068 | return s1 - start; |
| 17069 | } |
| 17070 | |
| 17071 | |
| 17072 | STATIC AV * |
| 17073 | S_add_multi_match(pTHX_ AV* multi_char_matches, SV* multi_string, const STRLEN cp_count) |
| 17074 | { |
| 17075 | /* This adds the string scalar <multi_string> to the array |
| 17076 | * <multi_char_matches>. <multi_string> is known to have exactly |
| 17077 | * <cp_count> code points in it. This is used when constructing a |
| 17078 | * bracketed character class and we find something that needs to match more |
| 17079 | * than a single character. |
| 17080 | * |
| 17081 | * <multi_char_matches> is actually an array of arrays. Each top-level |
| 17082 | * element is an array that contains all the strings known so far that are |
| 17083 | * the same length. And that length (in number of code points) is the same |
| 17084 | * as the index of the top-level array. Hence, the [2] element is an |
| 17085 | * array, each element thereof is a string containing TWO code points; |
| 17086 | * while element [3] is for strings of THREE characters, and so on. Since |
| 17087 | * this is for multi-char strings there can never be a [0] nor [1] element. |
| 17088 | * |
| 17089 | * When we rewrite the character class below, we will do so such that the |
| 17090 | * longest strings are written first, so that it prefers the longest |
| 17091 | * matching strings first. This is done even if it turns out that any |
| 17092 | * quantifier is non-greedy, out of this programmer's (khw) laziness. Tom |
| 17093 | * Christiansen has agreed that this is ok. This makes the test for the |
| 17094 | * ligature 'ffi' come before the test for 'ff', for example */ |
| 17095 | |
| 17096 | AV* this_array; |
| 17097 | AV** this_array_ptr; |
| 17098 | |
| 17099 | PERL_ARGS_ASSERT_ADD_MULTI_MATCH; |
| 17100 | |
| 17101 | if (! multi_char_matches) { |
| 17102 | multi_char_matches = newAV(); |
| 17103 | } |
| 17104 | |
| 17105 | if (av_exists(multi_char_matches, cp_count)) { |
| 17106 | this_array_ptr = (AV**) av_fetch(multi_char_matches, cp_count, FALSE); |
| 17107 | this_array = *this_array_ptr; |
| 17108 | } |
| 17109 | else { |
| 17110 | this_array = newAV(); |
| 17111 | av_store(multi_char_matches, cp_count, |
| 17112 | (SV*) this_array); |
| 17113 | } |
| 17114 | av_push(this_array, multi_string); |
| 17115 | |
| 17116 | return multi_char_matches; |
| 17117 | } |
| 17118 | |
| 17119 | /* The names of properties whose definitions are not known at compile time are |
| 17120 | * stored in this SV, after a constant heading. So if the length has been |
| 17121 | * changed since initialization, then there is a run-time definition. */ |
| 17122 | #define HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION \ |
| 17123 | (SvCUR(listsv) != initial_listsv_len) |
| 17124 | |
| 17125 | /* There is a restricted set of white space characters that are legal when |
| 17126 | * ignoring white space in a bracketed character class. This generates the |
| 17127 | * code to skip them. |
| 17128 | * |
| 17129 | * There is a line below that uses the same white space criteria but is outside |
| 17130 | * this macro. Both here and there must use the same definition */ |
| 17131 | #define SKIP_BRACKETED_WHITE_SPACE(do_skip, p) \ |
| 17132 | STMT_START { \ |
| 17133 | if (do_skip) { \ |
| 17134 | while (isBLANK_A(UCHARAT(p))) \ |
| 17135 | { \ |
| 17136 | p++; \ |
| 17137 | } \ |
| 17138 | } \ |
| 17139 | } STMT_END |
| 17140 | |
| 17141 | STATIC regnode_offset |
| 17142 | S_regclass(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth, |
| 17143 | const bool stop_at_1, /* Just parse the next thing, don't |
| 17144 | look for a full character class */ |
| 17145 | bool allow_mutiple_chars, |
| 17146 | const bool silence_non_portable, /* Don't output warnings |
| 17147 | about too large |
| 17148 | characters */ |
| 17149 | const bool strict, |
| 17150 | bool optimizable, /* ? Allow a non-ANYOF return |
| 17151 | node */ |
| 17152 | SV** ret_invlist /* Return an inversion list, not a node */ |
| 17153 | ) |
| 17154 | { |
| 17155 | /* parse a bracketed class specification. Most of these will produce an |
| 17156 | * ANYOF node; but something like [a] will produce an EXACT node; [aA], an |
| 17157 | * EXACTFish node; [[:ascii:]], a POSIXA node; etc. It is more complex |
| 17158 | * under /i with multi-character folds: it will be rewritten following the |
| 17159 | * paradigm of this example, where the <multi-fold>s are characters which |
| 17160 | * fold to multiple character sequences: |
| 17161 | * /[abc\x{multi-fold1}def\x{multi-fold2}ghi]/i |
| 17162 | * gets effectively rewritten as: |
| 17163 | * /(?:\x{multi-fold1}|\x{multi-fold2}|[abcdefghi]/i |
| 17164 | * reg() gets called (recursively) on the rewritten version, and this |
| 17165 | * function will return what it constructs. (Actually the <multi-fold>s |
| 17166 | * aren't physically removed from the [abcdefghi], it's just that they are |
| 17167 | * ignored in the recursion by means of a flag: |
| 17168 | * <RExC_in_multi_char_class>.) |
| 17169 | * |
| 17170 | * ANYOF nodes contain a bit map for the first NUM_ANYOF_CODE_POINTS |
| 17171 | * characters, with the corresponding bit set if that character is in the |
| 17172 | * list. For characters above this, an inversion list is used. There |
| 17173 | * are extra bits for \w, etc. in locale ANYOFs, as what these match is not |
| 17174 | * determinable at compile time |
| 17175 | * |
| 17176 | * On success, returns the offset at which any next node should be placed |
| 17177 | * into the regex engine program being compiled. |
| 17178 | * |
| 17179 | * Returns 0 otherwise, setting flagp to RESTART_PARSE if the parse needs |
| 17180 | * to be restarted, or'd with NEED_UTF8 if the pattern needs to be upgraded to |
| 17181 | * UTF-8 |
| 17182 | */ |
| 17183 | |
| 17184 | dVAR; |
| 17185 | UV prevvalue = OOB_UNICODE, save_prevvalue = OOB_UNICODE; |
| 17186 | IV range = 0; |
| 17187 | UV value = OOB_UNICODE, save_value = OOB_UNICODE; |
| 17188 | regnode_offset ret = -1; /* Initialized to an illegal value */ |
| 17189 | STRLEN numlen; |
| 17190 | int namedclass = OOB_NAMEDCLASS; |
| 17191 | char *rangebegin = NULL; |
| 17192 | SV *listsv = NULL; /* List of \p{user-defined} whose definitions |
| 17193 | aren't available at the time this was called */ |
| 17194 | STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more |
| 17195 | than just initialized. */ |
| 17196 | SV* properties = NULL; /* Code points that match \p{} \P{} */ |
| 17197 | SV* posixes = NULL; /* Code points that match classes like [:word:], |
| 17198 | extended beyond the Latin1 range. These have to |
| 17199 | be kept separate from other code points for much |
| 17200 | of this function because their handling is |
| 17201 | different under /i, and for most classes under |
| 17202 | /d as well */ |
| 17203 | SV* nposixes = NULL; /* Similarly for [:^word:]. These are kept |
| 17204 | separate for a while from the non-complemented |
| 17205 | versions because of complications with /d |
| 17206 | matching */ |
| 17207 | SV* simple_posixes = NULL; /* But under some conditions, the classes can be |
| 17208 | treated more simply than the general case, |
| 17209 | leading to less compilation and execution |
| 17210 | work */ |
| 17211 | UV element_count = 0; /* Number of distinct elements in the class. |
| 17212 | Optimizations may be possible if this is tiny */ |
| 17213 | AV * multi_char_matches = NULL; /* Code points that fold to more than one |
| 17214 | character; used under /i */ |
| 17215 | UV n; |
| 17216 | char * stop_ptr = RExC_end; /* where to stop parsing */ |
| 17217 | |
| 17218 | /* ignore unescaped whitespace? */ |
| 17219 | const bool skip_white = cBOOL( ret_invlist |
| 17220 | || (RExC_flags & RXf_PMf_EXTENDED_MORE)); |
| 17221 | |
| 17222 | /* inversion list of code points this node matches only when the target |
| 17223 | * string is in UTF-8. These are all non-ASCII, < 256. (Because is under |
| 17224 | * /d) */ |
| 17225 | SV* upper_latin1_only_utf8_matches = NULL; |
| 17226 | |
| 17227 | /* Inversion list of code points this node matches regardless of things |
| 17228 | * like locale, folding, utf8ness of the target string */ |
| 17229 | SV* cp_list = NULL; |
| 17230 | |
| 17231 | /* Like cp_list, but code points on this list need to be checked for things |
| 17232 | * that fold to/from them under /i */ |
| 17233 | SV* cp_foldable_list = NULL; |
| 17234 | |
| 17235 | /* Like cp_list, but code points on this list are valid only when the |
| 17236 | * runtime locale is UTF-8 */ |
| 17237 | SV* only_utf8_locale_list = NULL; |
| 17238 | |
| 17239 | /* In a range, if one of the endpoints is non-character-set portable, |
| 17240 | * meaning that it hard-codes a code point that may mean a different |
| 17241 | * charactger in ASCII vs. EBCDIC, as opposed to, say, a literal 'A' or a |
| 17242 | * mnemonic '\t' which each mean the same character no matter which |
| 17243 | * character set the platform is on. */ |
| 17244 | unsigned int non_portable_endpoint = 0; |
| 17245 | |
| 17246 | /* Is the range unicode? which means on a platform that isn't 1-1 native |
| 17247 | * to Unicode (i.e. non-ASCII), each code point in it should be considered |
| 17248 | * to be a Unicode value. */ |
| 17249 | bool unicode_range = FALSE; |
| 17250 | bool invert = FALSE; /* Is this class to be complemented */ |
| 17251 | |
| 17252 | bool warn_super = ALWAYS_WARN_SUPER; |
| 17253 | |
| 17254 | const char * orig_parse = RExC_parse; |
| 17255 | |
| 17256 | /* This variable is used to mark where the end in the input is of something |
| 17257 | * that looks like a POSIX construct but isn't. During the parse, when |
| 17258 | * something looks like it could be such a construct is encountered, it is |
| 17259 | * checked for being one, but not if we've already checked this area of the |
| 17260 | * input. Only after this position is reached do we check again */ |
| 17261 | char *not_posix_region_end = RExC_parse - 1; |
| 17262 | |
| 17263 | AV* posix_warnings = NULL; |
| 17264 | const bool do_posix_warnings = ckWARN(WARN_REGEXP); |
| 17265 | U8 op = END; /* The returned node-type, initialized to an impossible |
| 17266 | one. */ |
| 17267 | U8 anyof_flags = 0; /* flag bits if the node is an ANYOF-type */ |
| 17268 | U32 posixl = 0; /* bit field of posix classes matched under /l */ |
| 17269 | |
| 17270 | |
| 17271 | /* Flags as to what things aren't knowable until runtime. (Note that these are |
| 17272 | * mutually exclusive.) */ |
| 17273 | #define HAS_USER_DEFINED_PROPERTY 0x01 /* /u any user-defined properties that |
| 17274 | haven't been defined as of yet */ |
| 17275 | #define HAS_D_RUNTIME_DEPENDENCY 0x02 /* /d if the target being matched is |
| 17276 | UTF-8 or not */ |
| 17277 | #define HAS_L_RUNTIME_DEPENDENCY 0x04 /* /l what the posix classes match and |
| 17278 | what gets folded */ |
| 17279 | U32 has_runtime_dependency = 0; /* OR of the above flags */ |
| 17280 | |
| 17281 | GET_RE_DEBUG_FLAGS_DECL; |
| 17282 | |
| 17283 | PERL_ARGS_ASSERT_REGCLASS; |
| 17284 | #ifndef DEBUGGING |
| 17285 | PERL_UNUSED_ARG(depth); |
| 17286 | #endif |
| 17287 | |
| 17288 | |
| 17289 | /* If wants an inversion list returned, we can't optimize to something |
| 17290 | * else. */ |
| 17291 | if (ret_invlist) { |
| 17292 | optimizable = FALSE; |
| 17293 | } |
| 17294 | |
| 17295 | DEBUG_PARSE("clas"); |
| 17296 | |
| 17297 | #if UNICODE_MAJOR_VERSION < 3 /* no multifolds in early Unicode */ \ |
| 17298 | || (UNICODE_MAJOR_VERSION == 3 && UNICODE_DOT_VERSION == 0 \ |
| 17299 | && UNICODE_DOT_DOT_VERSION == 0) |
| 17300 | allow_mutiple_chars = FALSE; |
| 17301 | #endif |
| 17302 | |
| 17303 | /* We include the /i status at the beginning of this so that we can |
| 17304 | * know it at runtime */ |
| 17305 | listsv = sv_2mortal(Perl_newSVpvf(aTHX_ "#%d\n", cBOOL(FOLD))); |
| 17306 | initial_listsv_len = SvCUR(listsv); |
| 17307 | SvTEMP_off(listsv); /* Grr, TEMPs and mortals are conflated. */ |
| 17308 | |
| 17309 | SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse); |
| 17310 | |
| 17311 | assert(RExC_parse <= RExC_end); |
| 17312 | |
| 17313 | if (UCHARAT(RExC_parse) == '^') { /* Complement the class */ |
| 17314 | RExC_parse++; |
| 17315 | invert = TRUE; |
| 17316 | allow_mutiple_chars = FALSE; |
| 17317 | MARK_NAUGHTY(1); |
| 17318 | SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse); |
| 17319 | } |
| 17320 | |
| 17321 | /* Check that they didn't say [:posix:] instead of [[:posix:]] */ |
| 17322 | if (! ret_invlist && MAYBE_POSIXCC(UCHARAT(RExC_parse))) { |
| 17323 | int maybe_class = handle_possible_posix(pRExC_state, |
| 17324 | RExC_parse, |
| 17325 | ¬_posix_region_end, |
| 17326 | NULL, |
| 17327 | TRUE /* checking only */); |
| 17328 | if (maybe_class >= OOB_NAMEDCLASS && do_posix_warnings) { |
| 17329 | ckWARN4reg(not_posix_region_end, |
| 17330 | "POSIX syntax [%c %c] belongs inside character classes%s", |
| 17331 | *RExC_parse, *RExC_parse, |
| 17332 | (maybe_class == OOB_NAMEDCLASS) |
| 17333 | ? ((POSIXCC_NOTYET(*RExC_parse)) |
| 17334 | ? " (but this one isn't implemented)" |
| 17335 | : " (but this one isn't fully valid)") |
| 17336 | : "" |
| 17337 | ); |
| 17338 | } |
| 17339 | } |
| 17340 | |
| 17341 | /* If the caller wants us to just parse a single element, accomplish this |
| 17342 | * by faking the loop ending condition */ |
| 17343 | if (stop_at_1 && RExC_end > RExC_parse) { |
| 17344 | stop_ptr = RExC_parse + 1; |
| 17345 | } |
| 17346 | |
| 17347 | /* allow 1st char to be ']' (allowing it to be '-' is dealt with later) */ |
| 17348 | if (UCHARAT(RExC_parse) == ']') |
| 17349 | goto charclassloop; |
| 17350 | |
| 17351 | while (1) { |
| 17352 | |
| 17353 | if ( posix_warnings |
| 17354 | && av_tindex_skip_len_mg(posix_warnings) >= 0 |
| 17355 | && RExC_parse > not_posix_region_end) |
| 17356 | { |
| 17357 | /* Warnings about posix class issues are considered tentative until |
| 17358 | * we are far enough along in the parse that we can no longer |
| 17359 | * change our mind, at which point we output them. This is done |
| 17360 | * each time through the loop so that a later class won't zap them |
| 17361 | * before they have been dealt with. */ |
| 17362 | output_posix_warnings(pRExC_state, posix_warnings); |
| 17363 | } |
| 17364 | |
| 17365 | if (RExC_parse >= stop_ptr) { |
| 17366 | break; |
| 17367 | } |
| 17368 | |
| 17369 | SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse); |
| 17370 | |
| 17371 | if (UCHARAT(RExC_parse) == ']') { |
| 17372 | break; |
| 17373 | } |
| 17374 | |
| 17375 | charclassloop: |
| 17376 | |
| 17377 | namedclass = OOB_NAMEDCLASS; /* initialize as illegal */ |
| 17378 | save_value = value; |
| 17379 | save_prevvalue = prevvalue; |
| 17380 | |
| 17381 | if (!range) { |
| 17382 | rangebegin = RExC_parse; |
| 17383 | element_count++; |
| 17384 | non_portable_endpoint = 0; |
| 17385 | } |
| 17386 | if (UTF && ! UTF8_IS_INVARIANT(* RExC_parse)) { |
| 17387 | value = utf8n_to_uvchr((U8*)RExC_parse, |
| 17388 | RExC_end - RExC_parse, |
| 17389 | &numlen, UTF8_ALLOW_DEFAULT); |
| 17390 | RExC_parse += numlen; |
| 17391 | } |
| 17392 | else |
| 17393 | value = UCHARAT(RExC_parse++); |
| 17394 | |
| 17395 | if (value == '[') { |
| 17396 | char * posix_class_end; |
| 17397 | namedclass = handle_possible_posix(pRExC_state, |
| 17398 | RExC_parse, |
| 17399 | &posix_class_end, |
| 17400 | do_posix_warnings ? &posix_warnings : NULL, |
| 17401 | FALSE /* die if error */); |
| 17402 | if (namedclass > OOB_NAMEDCLASS) { |
| 17403 | |
| 17404 | /* If there was an earlier attempt to parse this particular |
| 17405 | * posix class, and it failed, it was a false alarm, as this |
| 17406 | * successful one proves */ |
| 17407 | if ( posix_warnings |
| 17408 | && av_tindex_skip_len_mg(posix_warnings) >= 0 |
| 17409 | && not_posix_region_end >= RExC_parse |
| 17410 | && not_posix_region_end <= posix_class_end) |
| 17411 | { |
| 17412 | av_undef(posix_warnings); |
| 17413 | } |
| 17414 | |
| 17415 | RExC_parse = posix_class_end; |
| 17416 | } |
| 17417 | else if (namedclass == OOB_NAMEDCLASS) { |
| 17418 | not_posix_region_end = posix_class_end; |
| 17419 | } |
| 17420 | else { |
| 17421 | namedclass = OOB_NAMEDCLASS; |
| 17422 | } |
| 17423 | } |
| 17424 | else if ( RExC_parse - 1 > not_posix_region_end |
| 17425 | && MAYBE_POSIXCC(value)) |
| 17426 | { |
| 17427 | (void) handle_possible_posix( |
| 17428 | pRExC_state, |
| 17429 | RExC_parse - 1, /* -1 because parse has already been |
| 17430 | advanced */ |
| 17431 | ¬_posix_region_end, |
| 17432 | do_posix_warnings ? &posix_warnings : NULL, |
| 17433 | TRUE /* checking only */); |
| 17434 | } |
| 17435 | else if ( strict && ! skip_white |
| 17436 | && ( _generic_isCC(value, _CC_VERTSPACE) |
| 17437 | || is_VERTWS_cp_high(value))) |
| 17438 | { |
| 17439 | vFAIL("Literal vertical space in [] is illegal except under /x"); |
| 17440 | } |
| 17441 | else if (value == '\\') { |
| 17442 | /* Is a backslash; get the code point of the char after it */ |
| 17443 | |
| 17444 | if (RExC_parse >= RExC_end) { |
| 17445 | vFAIL("Unmatched ["); |
| 17446 | } |
| 17447 | |
| 17448 | if (UTF && ! UTF8_IS_INVARIANT(UCHARAT(RExC_parse))) { |
| 17449 | value = utf8n_to_uvchr((U8*)RExC_parse, |
| 17450 | RExC_end - RExC_parse, |
| 17451 | &numlen, UTF8_ALLOW_DEFAULT); |
| 17452 | RExC_parse += numlen; |
| 17453 | } |
| 17454 | else |
| 17455 | value = UCHARAT(RExC_parse++); |
| 17456 | |
| 17457 | /* Some compilers cannot handle switching on 64-bit integer |
| 17458 | * values, therefore value cannot be an UV. Yes, this will |
| 17459 | * be a problem later if we want switch on Unicode. |
| 17460 | * A similar issue a little bit later when switching on |
| 17461 | * namedclass. --jhi */ |
| 17462 | |
| 17463 | /* If the \ is escaping white space when white space is being |
| 17464 | * skipped, it means that that white space is wanted literally, and |
| 17465 | * is already in 'value'. Otherwise, need to translate the escape |
| 17466 | * into what it signifies. */ |
| 17467 | if (! skip_white || ! isBLANK_A(value)) switch ((I32)value) { |
| 17468 | const char * message; |
| 17469 | U32 packed_warn; |
| 17470 | U8 grok_c_char; |
| 17471 | |
| 17472 | case 'w': namedclass = ANYOF_WORDCHAR; break; |
| 17473 | case 'W': namedclass = ANYOF_NWORDCHAR; break; |
| 17474 | case 's': namedclass = ANYOF_SPACE; break; |
| 17475 | case 'S': namedclass = ANYOF_NSPACE; break; |
| 17476 | case 'd': namedclass = ANYOF_DIGIT; break; |
| 17477 | case 'D': namedclass = ANYOF_NDIGIT; break; |
| 17478 | case 'v': namedclass = ANYOF_VERTWS; break; |
| 17479 | case 'V': namedclass = ANYOF_NVERTWS; break; |
| 17480 | case 'h': namedclass = ANYOF_HORIZWS; break; |
| 17481 | case 'H': namedclass = ANYOF_NHORIZWS; break; |
| 17482 | case 'N': /* Handle \N{NAME} in class */ |
| 17483 | { |
| 17484 | const char * const backslash_N_beg = RExC_parse - 2; |
| 17485 | int cp_count; |
| 17486 | |
| 17487 | if (! grok_bslash_N(pRExC_state, |
| 17488 | NULL, /* No regnode */ |
| 17489 | &value, /* Yes single value */ |
| 17490 | &cp_count, /* Multiple code pt count */ |
| 17491 | flagp, |
| 17492 | strict, |
| 17493 | depth) |
| 17494 | ) { |
| 17495 | |
| 17496 | if (*flagp & NEED_UTF8) |
| 17497 | FAIL("panic: grok_bslash_N set NEED_UTF8"); |
| 17498 | |
| 17499 | RETURN_FAIL_ON_RESTART_FLAGP(flagp); |
| 17500 | |
| 17501 | if (cp_count < 0) { |
| 17502 | vFAIL("\\N in a character class must be a named character: \\N{...}"); |
| 17503 | } |
| 17504 | else if (cp_count == 0) { |
| 17505 | ckWARNreg(RExC_parse, |
| 17506 | "Ignoring zero length \\N{} in character class"); |
| 17507 | } |
| 17508 | else { /* cp_count > 1 */ |
| 17509 | assert(cp_count > 1); |
| 17510 | if (! RExC_in_multi_char_class) { |
| 17511 | if ( ! allow_mutiple_chars |
| 17512 | || invert |
| 17513 | || range |
| 17514 | || *RExC_parse == '-') |
| 17515 | { |
| 17516 | if (strict) { |
| 17517 | RExC_parse--; |
| 17518 | vFAIL("\\N{} here is restricted to one character"); |
| 17519 | } |
| 17520 | ckWARNreg(RExC_parse, "Using just the first character returned by \\N{} in character class"); |
| 17521 | break; /* <value> contains the first code |
| 17522 | point. Drop out of the switch to |
| 17523 | process it */ |
| 17524 | } |
| 17525 | else { |
| 17526 | SV * multi_char_N = newSVpvn(backslash_N_beg, |
| 17527 | RExC_parse - backslash_N_beg); |
| 17528 | multi_char_matches |
| 17529 | = add_multi_match(multi_char_matches, |
| 17530 | multi_char_N, |
| 17531 | cp_count); |
| 17532 | } |
| 17533 | } |
| 17534 | } /* End of cp_count != 1 */ |
| 17535 | |
| 17536 | /* This element should not be processed further in this |
| 17537 | * class */ |
| 17538 | element_count--; |
| 17539 | value = save_value; |
| 17540 | prevvalue = save_prevvalue; |
| 17541 | continue; /* Back to top of loop to get next char */ |
| 17542 | } |
| 17543 | |
| 17544 | /* Here, is a single code point, and <value> contains it */ |
| 17545 | unicode_range = TRUE; /* \N{} are Unicode */ |
| 17546 | } |
| 17547 | break; |
| 17548 | case 'p': |
| 17549 | case 'P': |
| 17550 | { |
| 17551 | char *e; |
| 17552 | |
| 17553 | if (RExC_pm_flags & PMf_WILDCARD) { |
| 17554 | RExC_parse++; |
| 17555 | /* diag_listed_as: Use of %s is not allowed in Unicode |
| 17556 | property wildcard subpatterns in regex; marked by <-- |
| 17557 | HERE in m/%s/ */ |
| 17558 | vFAIL3("Use of '\\%c%c' is not allowed in Unicode property" |
| 17559 | " wildcard subpatterns", (char) value, *(RExC_parse - 1)); |
| 17560 | } |
| 17561 | |
| 17562 | /* \p means they want Unicode semantics */ |
| 17563 | REQUIRE_UNI_RULES(flagp, 0); |
| 17564 | |
| 17565 | if (RExC_parse >= RExC_end) |
| 17566 | vFAIL2("Empty \\%c", (U8)value); |
| 17567 | if (*RExC_parse == '{') { |
| 17568 | const U8 c = (U8)value; |
| 17569 | e = (char *) memchr(RExC_parse, '}', RExC_end - RExC_parse); |
| 17570 | if (!e) { |
| 17571 | RExC_parse++; |
| 17572 | vFAIL2("Missing right brace on \\%c{}", c); |
| 17573 | } |
| 17574 | |
| 17575 | RExC_parse++; |
| 17576 | |
| 17577 | /* White space is allowed adjacent to the braces and after |
| 17578 | * any '^', even when not under /x */ |
| 17579 | while (isSPACE(*RExC_parse)) { |
| 17580 | RExC_parse++; |
| 17581 | } |
| 17582 | |
| 17583 | if (UCHARAT(RExC_parse) == '^') { |
| 17584 | |
| 17585 | /* toggle. (The rhs xor gets the single bit that |
| 17586 | * differs between P and p; the other xor inverts just |
| 17587 | * that bit) */ |
| 17588 | value ^= 'P' ^ 'p'; |
| 17589 | |
| 17590 | RExC_parse++; |
| 17591 | while (isSPACE(*RExC_parse)) { |
| 17592 | RExC_parse++; |
| 17593 | } |
| 17594 | } |
| 17595 | |
| 17596 | if (e == RExC_parse) |
| 17597 | vFAIL2("Empty \\%c{}", c); |
| 17598 | |
| 17599 | n = e - RExC_parse; |
| 17600 | while (isSPACE(*(RExC_parse + n - 1))) |
| 17601 | n--; |
| 17602 | |
| 17603 | } /* The \p isn't immediately followed by a '{' */ |
| 17604 | else if (! isALPHA(*RExC_parse)) { |
| 17605 | RExC_parse += (UTF) |
| 17606 | ? UTF8_SAFE_SKIP(RExC_parse, RExC_end) |
| 17607 | : 1; |
| 17608 | vFAIL2("Character following \\%c must be '{' or a " |
| 17609 | "single-character Unicode property name", |
| 17610 | (U8) value); |
| 17611 | } |
| 17612 | else { |
| 17613 | e = RExC_parse; |
| 17614 | n = 1; |
| 17615 | } |
| 17616 | { |
| 17617 | char* name = RExC_parse; |
| 17618 | |
| 17619 | /* Any message returned about expanding the definition */ |
| 17620 | SV* msg = newSVpvs_flags("", SVs_TEMP); |
| 17621 | |
| 17622 | /* If set TRUE, the property is user-defined as opposed to |
| 17623 | * official Unicode */ |
| 17624 | bool user_defined = FALSE; |
| 17625 | |
| 17626 | SV * prop_definition = parse_uniprop_string( |
| 17627 | name, n, UTF, FOLD, |
| 17628 | FALSE, /* This is compile-time */ |
| 17629 | |
| 17630 | /* We can't defer this defn when |
| 17631 | * the full result is required in |
| 17632 | * this call */ |
| 17633 | ! cBOOL(ret_invlist), |
| 17634 | |
| 17635 | &user_defined, |
| 17636 | msg, |
| 17637 | 0 /* Base level */ |
| 17638 | ); |
| 17639 | if (SvCUR(msg)) { /* Assumes any error causes a msg */ |
| 17640 | assert(prop_definition == NULL); |
| 17641 | RExC_parse = e + 1; |
| 17642 | if (SvUTF8(msg)) { /* msg being UTF-8 makes the whole |
| 17643 | thing so, or else the display is |
| 17644 | mojibake */ |
| 17645 | RExC_utf8 = TRUE; |
| 17646 | } |
| 17647 | /* diag_listed_as: Can't find Unicode property definition "%s" in regex; marked by <-- HERE in m/%s/ */ |
| 17648 | vFAIL2utf8f("%" UTF8f, UTF8fARG(SvUTF8(msg), |
| 17649 | SvCUR(msg), SvPVX(msg))); |
| 17650 | } |
| 17651 | |
| 17652 | if (! is_invlist(prop_definition)) { |
| 17653 | |
| 17654 | /* Here, the definition isn't known, so we have gotten |
| 17655 | * returned a string that will be evaluated if and when |
| 17656 | * encountered at runtime. We add it to the list of |
| 17657 | * such properties, along with whether it should be |
| 17658 | * complemented or not */ |
| 17659 | if (value == 'P') { |
| 17660 | sv_catpvs(listsv, "!"); |
| 17661 | } |
| 17662 | else { |
| 17663 | sv_catpvs(listsv, "+"); |
| 17664 | } |
| 17665 | sv_catsv(listsv, prop_definition); |
| 17666 | |
| 17667 | has_runtime_dependency |= HAS_USER_DEFINED_PROPERTY; |
| 17668 | |
| 17669 | /* We don't know yet what this matches, so have to flag |
| 17670 | * it */ |
| 17671 | anyof_flags |= ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP; |
| 17672 | } |
| 17673 | else { |
| 17674 | assert (prop_definition && is_invlist(prop_definition)); |
| 17675 | |
| 17676 | /* Here we do have the complete property definition |
| 17677 | * |
| 17678 | * Temporary workaround for [perl #133136]. For this |
| 17679 | * precise input that is in the .t that is failing, |
| 17680 | * load utf8.pm, which is what the test wants, so that |
| 17681 | * that .t passes */ |
| 17682 | if ( memEQs(RExC_start, e + 1 - RExC_start, |
| 17683 | "foo\\p{Alnum}") |
| 17684 | && ! hv_common(GvHVn(PL_incgv), |
| 17685 | NULL, |
| 17686 | "utf8.pm", sizeof("utf8.pm") - 1, |
| 17687 | 0, HV_FETCH_ISEXISTS, NULL, 0)) |
| 17688 | { |
| 17689 | require_pv("utf8.pm"); |
| 17690 | } |
| 17691 | |
| 17692 | if (! user_defined && |
| 17693 | /* We warn on matching an above-Unicode code point |
| 17694 | * if the match would return true, except don't |
| 17695 | * warn for \p{All}, which has exactly one element |
| 17696 | * = 0 */ |
| 17697 | (_invlist_contains_cp(prop_definition, 0x110000) |
| 17698 | && (! (_invlist_len(prop_definition) == 1 |
| 17699 | && *invlist_array(prop_definition) == 0)))) |
| 17700 | { |
| 17701 | warn_super = TRUE; |
| 17702 | } |
| 17703 | |
| 17704 | /* Invert if asking for the complement */ |
| 17705 | if (value == 'P') { |
| 17706 | _invlist_union_complement_2nd(properties, |
| 17707 | prop_definition, |
| 17708 | &properties); |
| 17709 | } |
| 17710 | else { |
| 17711 | _invlist_union(properties, prop_definition, &properties); |
| 17712 | } |
| 17713 | } |
| 17714 | } |
| 17715 | |
| 17716 | RExC_parse = e + 1; |
| 17717 | namedclass = ANYOF_UNIPROP; /* no official name, but it's |
| 17718 | named */ |
| 17719 | } |
| 17720 | break; |
| 17721 | case 'n': value = '\n'; break; |
| 17722 | case 'r': value = '\r'; break; |
| 17723 | case 't': value = '\t'; break; |
| 17724 | case 'f': value = '\f'; break; |
| 17725 | case 'b': value = '\b'; break; |
| 17726 | case 'e': value = ESC_NATIVE; break; |
| 17727 | case 'a': value = '\a'; break; |
| 17728 | case 'o': |
| 17729 | RExC_parse--; /* function expects to be pointed at the 'o' */ |
| 17730 | if (! grok_bslash_o(&RExC_parse, |
| 17731 | RExC_end, |
| 17732 | &value, |
| 17733 | &message, |
| 17734 | &packed_warn, |
| 17735 | strict, |
| 17736 | cBOOL(range), /* MAX_UV allowed for range |
| 17737 | upper limit */ |
| 17738 | UTF)) |
| 17739 | { |
| 17740 | vFAIL(message); |
| 17741 | } |
| 17742 | else if (message && TO_OUTPUT_WARNINGS(RExC_parse)) { |
| 17743 | warn_non_literal_string(RExC_parse, packed_warn, message); |
| 17744 | } |
| 17745 | |
| 17746 | if (value < 256) { |
| 17747 | non_portable_endpoint++; |
| 17748 | } |
| 17749 | break; |
| 17750 | case 'x': |
| 17751 | RExC_parse--; /* function expects to be pointed at the 'x' */ |
| 17752 | if (! grok_bslash_x(&RExC_parse, |
| 17753 | RExC_end, |
| 17754 | &value, |
| 17755 | &message, |
| 17756 | &packed_warn, |
| 17757 | strict, |
| 17758 | cBOOL(range), /* MAX_UV allowed for range |
| 17759 | upper limit */ |
| 17760 | UTF)) |
| 17761 | { |
| 17762 | vFAIL(message); |
| 17763 | } |
| 17764 | else if (message && TO_OUTPUT_WARNINGS(RExC_parse)) { |
| 17765 | warn_non_literal_string(RExC_parse, packed_warn, message); |
| 17766 | } |
| 17767 | |
| 17768 | if (value < 256) { |
| 17769 | non_portable_endpoint++; |
| 17770 | } |
| 17771 | break; |
| 17772 | case 'c': |
| 17773 | if (! grok_bslash_c(*RExC_parse, &grok_c_char, &message, |
| 17774 | &packed_warn)) |
| 17775 | { |
| 17776 | /* going to die anyway; point to exact spot of |
| 17777 | * failure */ |
| 17778 | RExC_parse += (UTF) |
| 17779 | ? UTF8_SAFE_SKIP(RExC_parse, RExC_end) |
| 17780 | : 1; |
| 17781 | vFAIL(message); |
| 17782 | } |
| 17783 | |
| 17784 | value = grok_c_char; |
| 17785 | RExC_parse++; |
| 17786 | if (message && TO_OUTPUT_WARNINGS(RExC_parse)) { |
| 17787 | warn_non_literal_string(RExC_parse, packed_warn, message); |
| 17788 | } |
| 17789 | |
| 17790 | non_portable_endpoint++; |
| 17791 | break; |
| 17792 | case '0': case '1': case '2': case '3': case '4': |
| 17793 | case '5': case '6': case '7': |
| 17794 | { |
| 17795 | /* Take 1-3 octal digits */ |
| 17796 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT |
| 17797 | | PERL_SCAN_NOTIFY_ILLDIGIT; |
| 17798 | numlen = (strict) ? 4 : 3; |
| 17799 | value = grok_oct(--RExC_parse, &numlen, &flags, NULL); |
| 17800 | RExC_parse += numlen; |
| 17801 | if (numlen != 3) { |
| 17802 | if (strict) { |
| 17803 | RExC_parse += (UTF) |
| 17804 | ? UTF8_SAFE_SKIP(RExC_parse, RExC_end) |
| 17805 | : 1; |
| 17806 | vFAIL("Need exactly 3 octal digits"); |
| 17807 | } |
| 17808 | else if ( (flags & PERL_SCAN_NOTIFY_ILLDIGIT) |
| 17809 | && RExC_parse < RExC_end |
| 17810 | && isDIGIT(*RExC_parse) |
| 17811 | && ckWARN(WARN_REGEXP)) |
| 17812 | { |
| 17813 | reg_warn_non_literal_string( |
| 17814 | RExC_parse + 1, |
| 17815 | form_alien_digit_msg(8, numlen, RExC_parse, |
| 17816 | RExC_end, UTF, FALSE)); |
| 17817 | } |
| 17818 | } |
| 17819 | if (value < 256) { |
| 17820 | non_portable_endpoint++; |
| 17821 | } |
| 17822 | break; |
| 17823 | } |
| 17824 | default: |
| 17825 | /* Allow \_ to not give an error */ |
| 17826 | if (isWORDCHAR(value) && value != '_') { |
| 17827 | if (strict) { |
| 17828 | vFAIL2("Unrecognized escape \\%c in character class", |
| 17829 | (int)value); |
| 17830 | } |
| 17831 | else { |
| 17832 | ckWARN2reg(RExC_parse, |
| 17833 | "Unrecognized escape \\%c in character class passed through", |
| 17834 | (int)value); |
| 17835 | } |
| 17836 | } |
| 17837 | break; |
| 17838 | } /* End of switch on char following backslash */ |
| 17839 | } /* end of handling backslash escape sequences */ |
| 17840 | |
| 17841 | /* Here, we have the current token in 'value' */ |
| 17842 | |
| 17843 | if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */ |
| 17844 | U8 classnum; |
| 17845 | |
| 17846 | /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a |
| 17847 | * literal, as is the character that began the false range, i.e. |
| 17848 | * the 'a' in the examples */ |
| 17849 | if (range) { |
| 17850 | const int w = (RExC_parse >= rangebegin) |
| 17851 | ? RExC_parse - rangebegin |
| 17852 | : 0; |
| 17853 | if (strict) { |
| 17854 | vFAIL2utf8f( |
| 17855 | "False [] range \"%" UTF8f "\"", |
| 17856 | UTF8fARG(UTF, w, rangebegin)); |
| 17857 | } |
| 17858 | else { |
| 17859 | ckWARN2reg(RExC_parse, |
| 17860 | "False [] range \"%" UTF8f "\"", |
| 17861 | UTF8fARG(UTF, w, rangebegin)); |
| 17862 | cp_list = add_cp_to_invlist(cp_list, '-'); |
| 17863 | cp_foldable_list = add_cp_to_invlist(cp_foldable_list, |
| 17864 | prevvalue); |
| 17865 | } |
| 17866 | |
| 17867 | range = 0; /* this was not a true range */ |
| 17868 | element_count += 2; /* So counts for three values */ |
| 17869 | } |
| 17870 | |
| 17871 | classnum = namedclass_to_classnum(namedclass); |
| 17872 | |
| 17873 | if (LOC && namedclass < ANYOF_POSIXL_MAX |
| 17874 | #ifndef HAS_ISASCII |
| 17875 | && classnum != _CC_ASCII |
| 17876 | #endif |
| 17877 | ) { |
| 17878 | SV* scratch_list = NULL; |
| 17879 | |
| 17880 | /* What the Posix classes (like \w, [:space:]) match isn't |
| 17881 | * generally knowable under locale until actual match time. A |
| 17882 | * special node is used for these which has extra space for a |
| 17883 | * bitmap, with a bit reserved for each named class that is to |
| 17884 | * be matched against. (This isn't needed for \p{} and |
| 17885 | * pseudo-classes, as they are not affected by locale, and |
| 17886 | * hence are dealt with separately.) However, if a named class |
| 17887 | * and its complement are both present, then it matches |
| 17888 | * everything, and there is no runtime dependency. Odd numbers |
| 17889 | * are the complements of the next lower number, so xor works. |
| 17890 | * (Note that something like [\w\D] should match everything, |
| 17891 | * because \d should be a proper subset of \w. But rather than |
| 17892 | * trust that the locale is well behaved, we leave this to |
| 17893 | * runtime to sort out) */ |
| 17894 | if (POSIXL_TEST(posixl, namedclass ^ 1)) { |
| 17895 | cp_list = _add_range_to_invlist(cp_list, 0, UV_MAX); |
| 17896 | POSIXL_ZERO(posixl); |
| 17897 | has_runtime_dependency &= ~HAS_L_RUNTIME_DEPENDENCY; |
| 17898 | anyof_flags &= ~ANYOF_MATCHES_POSIXL; |
| 17899 | continue; /* We could ignore the rest of the class, but |
| 17900 | best to parse it for any errors */ |
| 17901 | } |
| 17902 | else { /* Here, isn't the complement of any already parsed |
| 17903 | class */ |
| 17904 | POSIXL_SET(posixl, namedclass); |
| 17905 | has_runtime_dependency |= HAS_L_RUNTIME_DEPENDENCY; |
| 17906 | anyof_flags |= ANYOF_MATCHES_POSIXL; |
| 17907 | |
| 17908 | /* The above-Latin1 characters are not subject to locale |
| 17909 | * rules. Just add them to the unconditionally-matched |
| 17910 | * list */ |
| 17911 | |
| 17912 | /* Get the list of the above-Latin1 code points this |
| 17913 | * matches */ |
| 17914 | _invlist_intersection_maybe_complement_2nd(PL_AboveLatin1, |
| 17915 | PL_XPosix_ptrs[classnum], |
| 17916 | |
| 17917 | /* Odd numbers are complements, |
| 17918 | * like NDIGIT, NASCII, ... */ |
| 17919 | namedclass % 2 != 0, |
| 17920 | &scratch_list); |
| 17921 | /* Checking if 'cp_list' is NULL first saves an extra |
| 17922 | * clone. Its reference count will be decremented at the |
| 17923 | * next union, etc, or if this is the only instance, at the |
| 17924 | * end of the routine */ |
| 17925 | if (! cp_list) { |
| 17926 | cp_list = scratch_list; |
| 17927 | } |
| 17928 | else { |
| 17929 | _invlist_union(cp_list, scratch_list, &cp_list); |
| 17930 | SvREFCNT_dec_NN(scratch_list); |
| 17931 | } |
| 17932 | continue; /* Go get next character */ |
| 17933 | } |
| 17934 | } |
| 17935 | else { |
| 17936 | |
| 17937 | /* Here, is not /l, or is a POSIX class for which /l doesn't |
| 17938 | * matter (or is a Unicode property, which is skipped here). */ |
| 17939 | if (namedclass >= ANYOF_POSIXL_MAX) { /* If a special class */ |
| 17940 | if (namedclass != ANYOF_UNIPROP) { /* UNIPROP = \p and \P */ |
| 17941 | |
| 17942 | /* Here, should be \h, \H, \v, or \V. None of /d, /i |
| 17943 | * nor /l make a difference in what these match, |
| 17944 | * therefore we just add what they match to cp_list. */ |
| 17945 | if (classnum != _CC_VERTSPACE) { |
| 17946 | assert( namedclass == ANYOF_HORIZWS |
| 17947 | || namedclass == ANYOF_NHORIZWS); |
| 17948 | |
| 17949 | /* It turns out that \h is just a synonym for |
| 17950 | * XPosixBlank */ |
| 17951 | classnum = _CC_BLANK; |
| 17952 | } |
| 17953 | |
| 17954 | _invlist_union_maybe_complement_2nd( |
| 17955 | cp_list, |
| 17956 | PL_XPosix_ptrs[classnum], |
| 17957 | namedclass % 2 != 0, /* Complement if odd |
| 17958 | (NHORIZWS, NVERTWS) |
| 17959 | */ |
| 17960 | &cp_list); |
| 17961 | } |
| 17962 | } |
| 17963 | else if ( AT_LEAST_UNI_SEMANTICS |
| 17964 | || classnum == _CC_ASCII |
| 17965 | || (DEPENDS_SEMANTICS && ( classnum == _CC_DIGIT |
| 17966 | || classnum == _CC_XDIGIT))) |
| 17967 | { |
| 17968 | /* We usually have to worry about /d affecting what POSIX |
| 17969 | * classes match, with special code needed because we won't |
| 17970 | * know until runtime what all matches. But there is no |
| 17971 | * extra work needed under /u and /a; and [:ascii:] is |
| 17972 | * unaffected by /d; and :digit: and :xdigit: don't have |
| 17973 | * runtime differences under /d. So we can special case |
| 17974 | * these, and avoid some extra work below, and at runtime. |
| 17975 | * */ |
| 17976 | _invlist_union_maybe_complement_2nd( |
| 17977 | simple_posixes, |
| 17978 | ((AT_LEAST_ASCII_RESTRICTED) |
| 17979 | ? PL_Posix_ptrs[classnum] |
| 17980 | : PL_XPosix_ptrs[classnum]), |
| 17981 | namedclass % 2 != 0, |
| 17982 | &simple_posixes); |
| 17983 | } |
| 17984 | else { /* Garden variety class. If is NUPPER, NALPHA, ... |
| 17985 | complement and use nposixes */ |
| 17986 | SV** posixes_ptr = namedclass % 2 == 0 |
| 17987 | ? &posixes |
| 17988 | : &nposixes; |
| 17989 | _invlist_union_maybe_complement_2nd( |
| 17990 | *posixes_ptr, |
| 17991 | PL_XPosix_ptrs[classnum], |
| 17992 | namedclass % 2 != 0, |
| 17993 | posixes_ptr); |
| 17994 | } |
| 17995 | } |
| 17996 | } /* end of namedclass \blah */ |
| 17997 | |
| 17998 | SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse); |
| 17999 | |
| 18000 | /* If 'range' is set, 'value' is the ending of a range--check its |
| 18001 | * validity. (If value isn't a single code point in the case of a |
| 18002 | * range, we should have figured that out above in the code that |
| 18003 | * catches false ranges). Later, we will handle each individual code |
| 18004 | * point in the range. If 'range' isn't set, this could be the |
| 18005 | * beginning of a range, so check for that by looking ahead to see if |
| 18006 | * the next real character to be processed is the range indicator--the |
| 18007 | * minus sign */ |
| 18008 | |
| 18009 | if (range) { |
| 18010 | #ifdef EBCDIC |
| 18011 | /* For unicode ranges, we have to test that the Unicode as opposed |
| 18012 | * to the native values are not decreasing. (Above 255, there is |
| 18013 | * no difference between native and Unicode) */ |
| 18014 | if (unicode_range && prevvalue < 255 && value < 255) { |
| 18015 | if (NATIVE_TO_LATIN1(prevvalue) > NATIVE_TO_LATIN1(value)) { |
| 18016 | goto backwards_range; |
| 18017 | } |
| 18018 | } |
| 18019 | else |
| 18020 | #endif |
| 18021 | if (prevvalue > value) /* b-a */ { |
| 18022 | int w; |
| 18023 | #ifdef EBCDIC |
| 18024 | backwards_range: |
| 18025 | #endif |
| 18026 | w = RExC_parse - rangebegin; |
| 18027 | vFAIL2utf8f( |
| 18028 | "Invalid [] range \"%" UTF8f "\"", |
| 18029 | UTF8fARG(UTF, w, rangebegin)); |
| 18030 | NOT_REACHED; /* NOTREACHED */ |
| 18031 | } |
| 18032 | } |
| 18033 | else { |
| 18034 | prevvalue = value; /* save the beginning of the potential range */ |
| 18035 | if (! stop_at_1 /* Can't be a range if parsing just one thing */ |
| 18036 | && *RExC_parse == '-') |
| 18037 | { |
| 18038 | char* next_char_ptr = RExC_parse + 1; |
| 18039 | |
| 18040 | /* Get the next real char after the '-' */ |
| 18041 | SKIP_BRACKETED_WHITE_SPACE(skip_white, next_char_ptr); |
| 18042 | |
| 18043 | /* If the '-' is at the end of the class (just before the ']', |
| 18044 | * it is a literal minus; otherwise it is a range */ |
| 18045 | if (next_char_ptr < RExC_end && *next_char_ptr != ']') { |
| 18046 | RExC_parse = next_char_ptr; |
| 18047 | |
| 18048 | /* a bad range like \w-, [:word:]- ? */ |
| 18049 | if (namedclass > OOB_NAMEDCLASS) { |
| 18050 | if (strict || ckWARN(WARN_REGEXP)) { |
| 18051 | const int w = RExC_parse >= rangebegin |
| 18052 | ? RExC_parse - rangebegin |
| 18053 | : 0; |
| 18054 | if (strict) { |
| 18055 | vFAIL4("False [] range \"%*.*s\"", |
| 18056 | w, w, rangebegin); |
| 18057 | } |
| 18058 | else { |
| 18059 | vWARN4(RExC_parse, |
| 18060 | "False [] range \"%*.*s\"", |
| 18061 | w, w, rangebegin); |
| 18062 | } |
| 18063 | } |
| 18064 | cp_list = add_cp_to_invlist(cp_list, '-'); |
| 18065 | element_count++; |
| 18066 | } else |
| 18067 | range = 1; /* yeah, it's a range! */ |
| 18068 | continue; /* but do it the next time */ |
| 18069 | } |
| 18070 | } |
| 18071 | } |
| 18072 | |
| 18073 | if (namedclass > OOB_NAMEDCLASS) { |
| 18074 | continue; |
| 18075 | } |
| 18076 | |
| 18077 | /* Here, we have a single value this time through the loop, and |
| 18078 | * <prevvalue> is the beginning of the range, if any; or <value> if |
| 18079 | * not. */ |
| 18080 | |
| 18081 | /* non-Latin1 code point implies unicode semantics. */ |
| 18082 | if (value > 255) { |
| 18083 | if (value > MAX_LEGAL_CP && ( value != UV_MAX |
| 18084 | || prevvalue > MAX_LEGAL_CP)) |
| 18085 | { |
| 18086 | vFAIL(form_cp_too_large_msg(16, NULL, 0, value)); |
| 18087 | } |
| 18088 | REQUIRE_UNI_RULES(flagp, 0); |
| 18089 | if ( ! silence_non_portable |
| 18090 | && UNICODE_IS_PERL_EXTENDED(value) |
| 18091 | && TO_OUTPUT_WARNINGS(RExC_parse)) |
| 18092 | { |
| 18093 | ckWARN2_non_literal_string(RExC_parse, |
| 18094 | packWARN(WARN_PORTABLE), |
| 18095 | PL_extended_cp_format, |
| 18096 | value); |
| 18097 | } |
| 18098 | } |
| 18099 | |
| 18100 | /* Ready to process either the single value, or the completed range. |
| 18101 | * For single-valued non-inverted ranges, we consider the possibility |
| 18102 | * of multi-char folds. (We made a conscious decision to not do this |
| 18103 | * for the other cases because it can often lead to non-intuitive |
| 18104 | * results. For example, you have the peculiar case that: |
| 18105 | * "s s" =~ /^[^\xDF]+$/i => Y |
| 18106 | * "ss" =~ /^[^\xDF]+$/i => N |
| 18107 | * |
| 18108 | * See [perl #89750] */ |
| 18109 | if (FOLD && allow_mutiple_chars && value == prevvalue) { |
| 18110 | if ( value == LATIN_SMALL_LETTER_SHARP_S |
| 18111 | || (value > 255 && _invlist_contains_cp(PL_HasMultiCharFold, |
| 18112 | value))) |
| 18113 | { |
| 18114 | /* Here <value> is indeed a multi-char fold. Get what it is */ |
| 18115 | |
| 18116 | U8 foldbuf[UTF8_MAXBYTES_CASE+1]; |
| 18117 | STRLEN foldlen; |
| 18118 | |
| 18119 | UV folded = _to_uni_fold_flags( |
| 18120 | value, |
| 18121 | foldbuf, |
| 18122 | &foldlen, |
| 18123 | FOLD_FLAGS_FULL | (ASCII_FOLD_RESTRICTED |
| 18124 | ? FOLD_FLAGS_NOMIX_ASCII |
| 18125 | : 0) |
| 18126 | ); |
| 18127 | |
| 18128 | /* Here, <folded> should be the first character of the |
| 18129 | * multi-char fold of <value>, with <foldbuf> containing the |
| 18130 | * whole thing. But, if this fold is not allowed (because of |
| 18131 | * the flags), <fold> will be the same as <value>, and should |
| 18132 | * be processed like any other character, so skip the special |
| 18133 | * handling */ |
| 18134 | if (folded != value) { |
| 18135 | |
| 18136 | /* Skip if we are recursed, currently parsing the class |
| 18137 | * again. Otherwise add this character to the list of |
| 18138 | * multi-char folds. */ |
| 18139 | if (! RExC_in_multi_char_class) { |
| 18140 | STRLEN cp_count = utf8_length(foldbuf, |
| 18141 | foldbuf + foldlen); |
| 18142 | SV* multi_fold = sv_2mortal(newSVpvs("")); |
| 18143 | |
| 18144 | Perl_sv_catpvf(aTHX_ multi_fold, "\\x{%" UVXf "}", value); |
| 18145 | |
| 18146 | multi_char_matches |
| 18147 | = add_multi_match(multi_char_matches, |
| 18148 | multi_fold, |
| 18149 | cp_count); |
| 18150 | |
| 18151 | } |
| 18152 | |
| 18153 | /* This element should not be processed further in this |
| 18154 | * class */ |
| 18155 | element_count--; |
| 18156 | value = save_value; |
| 18157 | prevvalue = save_prevvalue; |
| 18158 | continue; |
| 18159 | } |
| 18160 | } |
| 18161 | } |
| 18162 | |
| 18163 | if (strict && ckWARN(WARN_REGEXP)) { |
| 18164 | if (range) { |
| 18165 | |
| 18166 | /* If the range starts above 255, everything is portable and |
| 18167 | * likely to be so for any forseeable character set, so don't |
| 18168 | * warn. */ |
| 18169 | if (unicode_range && non_portable_endpoint && prevvalue < 256) { |
| 18170 | vWARN(RExC_parse, "Both or neither range ends should be Unicode"); |
| 18171 | } |
| 18172 | else if (prevvalue != value) { |
| 18173 | |
| 18174 | /* Under strict, ranges that stop and/or end in an ASCII |
| 18175 | * printable should have each end point be a portable value |
| 18176 | * for it (preferably like 'A', but we don't warn if it is |
| 18177 | * a (portable) Unicode name or code point), and the range |
| 18178 | * must be be all digits or all letters of the same case. |
| 18179 | * Otherwise, the range is non-portable and unclear as to |
| 18180 | * what it contains */ |
| 18181 | if ( (isPRINT_A(prevvalue) || isPRINT_A(value)) |
| 18182 | && ( non_portable_endpoint |
| 18183 | || ! ( (isDIGIT_A(prevvalue) && isDIGIT_A(value)) |
| 18184 | || (isLOWER_A(prevvalue) && isLOWER_A(value)) |
| 18185 | || (isUPPER_A(prevvalue) && isUPPER_A(value)) |
| 18186 | ))) { |
| 18187 | vWARN(RExC_parse, "Ranges of ASCII printables should" |
| 18188 | " be some subset of \"0-9\"," |
| 18189 | " \"A-Z\", or \"a-z\""); |
| 18190 | } |
| 18191 | else if (prevvalue >= FIRST_NON_ASCII_DECIMAL_DIGIT) { |
| 18192 | SSize_t index_start; |
| 18193 | SSize_t index_final; |
| 18194 | |
| 18195 | /* But the nature of Unicode and languages mean we |
| 18196 | * can't do the same checks for above-ASCII ranges, |
| 18197 | * except in the case of digit ones. These should |
| 18198 | * contain only digits from the same group of 10. The |
| 18199 | * ASCII case is handled just above. Hence here, the |
| 18200 | * range could be a range of digits. First some |
| 18201 | * unlikely special cases. Grandfather in that a range |
| 18202 | * ending in 19DA (NEW TAI LUE THAM DIGIT ONE) is bad |
| 18203 | * if its starting value is one of the 10 digits prior |
| 18204 | * to it. This is because it is an alternate way of |
| 18205 | * writing 19D1, and some people may expect it to be in |
| 18206 | * that group. But it is bad, because it won't give |
| 18207 | * the expected results. In Unicode 5.2 it was |
| 18208 | * considered to be in that group (of 11, hence), but |
| 18209 | * this was fixed in the next version */ |
| 18210 | |
| 18211 | if (UNLIKELY(value == 0x19DA && prevvalue >= 0x19D0)) { |
| 18212 | goto warn_bad_digit_range; |
| 18213 | } |
| 18214 | else if (UNLIKELY( prevvalue >= 0x1D7CE |
| 18215 | && value <= 0x1D7FF)) |
| 18216 | { |
| 18217 | /* This is the only other case currently in Unicode |
| 18218 | * where the algorithm below fails. The code |
| 18219 | * points just above are the end points of a single |
| 18220 | * range containing only decimal digits. It is 5 |
| 18221 | * different series of 0-9. All other ranges of |
| 18222 | * digits currently in Unicode are just a single |
| 18223 | * series. (And mktables will notify us if a later |
| 18224 | * Unicode version breaks this.) |
| 18225 | * |
| 18226 | * If the range being checked is at most 9 long, |
| 18227 | * and the digit values represented are in |
| 18228 | * numerical order, they are from the same series. |
| 18229 | * */ |
| 18230 | if ( value - prevvalue > 9 |
| 18231 | || ((( value - 0x1D7CE) % 10) |
| 18232 | <= (prevvalue - 0x1D7CE) % 10)) |
| 18233 | { |
| 18234 | goto warn_bad_digit_range; |
| 18235 | } |
| 18236 | } |
| 18237 | else { |
| 18238 | |
| 18239 | /* For all other ranges of digits in Unicode, the |
| 18240 | * algorithm is just to check if both end points |
| 18241 | * are in the same series, which is the same range. |
| 18242 | * */ |
| 18243 | index_start = _invlist_search( |
| 18244 | PL_XPosix_ptrs[_CC_DIGIT], |
| 18245 | prevvalue); |
| 18246 | |
| 18247 | /* Warn if the range starts and ends with a digit, |
| 18248 | * and they are not in the same group of 10. */ |
| 18249 | if ( index_start >= 0 |
| 18250 | && ELEMENT_RANGE_MATCHES_INVLIST(index_start) |
| 18251 | && (index_final = |
| 18252 | _invlist_search(PL_XPosix_ptrs[_CC_DIGIT], |
| 18253 | value)) != index_start |
| 18254 | && index_final >= 0 |
| 18255 | && ELEMENT_RANGE_MATCHES_INVLIST(index_final)) |
| 18256 | { |
| 18257 | warn_bad_digit_range: |
| 18258 | vWARN(RExC_parse, "Ranges of digits should be" |
| 18259 | " from the same group of" |
| 18260 | " 10"); |
| 18261 | } |
| 18262 | } |
| 18263 | } |
| 18264 | } |
| 18265 | } |
| 18266 | if ((! range || prevvalue == value) && non_portable_endpoint) { |
| 18267 | if (isPRINT_A(value)) { |
| 18268 | char literal[3]; |
| 18269 | unsigned d = 0; |
| 18270 | if (isBACKSLASHED_PUNCT(value)) { |
| 18271 | literal[d++] = '\\'; |
| 18272 | } |
| 18273 | literal[d++] = (char) value; |
| 18274 | literal[d++] = '\0'; |
| 18275 | |
| 18276 | vWARN4(RExC_parse, |
| 18277 | "\"%.*s\" is more clearly written simply as \"%s\"", |
| 18278 | (int) (RExC_parse - rangebegin), |
| 18279 | rangebegin, |
| 18280 | literal |
| 18281 | ); |
| 18282 | } |
| 18283 | else if (isMNEMONIC_CNTRL(value)) { |
| 18284 | vWARN4(RExC_parse, |
| 18285 | "\"%.*s\" is more clearly written simply as \"%s\"", |
| 18286 | (int) (RExC_parse - rangebegin), |
| 18287 | rangebegin, |
| 18288 | cntrl_to_mnemonic((U8) value) |
| 18289 | ); |
| 18290 | } |
| 18291 | } |
| 18292 | } |
| 18293 | |
| 18294 | /* Deal with this element of the class */ |
| 18295 | |
| 18296 | #ifndef EBCDIC |
| 18297 | cp_foldable_list = _add_range_to_invlist(cp_foldable_list, |
| 18298 | prevvalue, value); |
| 18299 | #else |
| 18300 | /* On non-ASCII platforms, for ranges that span all of 0..255, and ones |
| 18301 | * that don't require special handling, we can just add the range like |
| 18302 | * we do for ASCII platforms */ |
| 18303 | if ((UNLIKELY(prevvalue == 0) && value >= 255) |
| 18304 | || ! (prevvalue < 256 |
| 18305 | && (unicode_range |
| 18306 | || (! non_portable_endpoint |
| 18307 | && ((isLOWER_A(prevvalue) && isLOWER_A(value)) |
| 18308 | || (isUPPER_A(prevvalue) |
| 18309 | && isUPPER_A(value))))))) |
| 18310 | { |
| 18311 | cp_foldable_list = _add_range_to_invlist(cp_foldable_list, |
| 18312 | prevvalue, value); |
| 18313 | } |
| 18314 | else { |
| 18315 | /* Here, requires special handling. This can be because it is a |
| 18316 | * range whose code points are considered to be Unicode, and so |
| 18317 | * must be individually translated into native, or because its a |
| 18318 | * subrange of 'A-Z' or 'a-z' which each aren't contiguous in |
| 18319 | * EBCDIC, but we have defined them to include only the "expected" |
| 18320 | * upper or lower case ASCII alphabetics. Subranges above 255 are |
| 18321 | * the same in native and Unicode, so can be added as a range */ |
| 18322 | U8 start = NATIVE_TO_LATIN1(prevvalue); |
| 18323 | unsigned j; |
| 18324 | U8 end = (value < 256) ? NATIVE_TO_LATIN1(value) : 255; |
| 18325 | for (j = start; j <= end; j++) { |
| 18326 | cp_foldable_list = add_cp_to_invlist(cp_foldable_list, LATIN1_TO_NATIVE(j)); |
| 18327 | } |
| 18328 | if (value > 255) { |
| 18329 | cp_foldable_list = _add_range_to_invlist(cp_foldable_list, |
| 18330 | 256, value); |
| 18331 | } |
| 18332 | } |
| 18333 | #endif |
| 18334 | |
| 18335 | range = 0; /* this range (if it was one) is done now */ |
| 18336 | } /* End of loop through all the text within the brackets */ |
| 18337 | |
| 18338 | if ( posix_warnings && av_tindex_skip_len_mg(posix_warnings) >= 0) { |
| 18339 | output_posix_warnings(pRExC_state, posix_warnings); |
| 18340 | } |
| 18341 | |
| 18342 | /* If anything in the class expands to more than one character, we have to |
| 18343 | * deal with them by building up a substitute parse string, and recursively |
| 18344 | * calling reg() on it, instead of proceeding */ |
| 18345 | if (multi_char_matches) { |
| 18346 | SV * substitute_parse = newSVpvn_flags("?:", 2, SVs_TEMP); |
| 18347 | I32 cp_count; |
| 18348 | STRLEN len; |
| 18349 | char *save_end = RExC_end; |
| 18350 | char *save_parse = RExC_parse; |
| 18351 | char *save_start = RExC_start; |
| 18352 | Size_t constructed_prefix_len = 0; /* This gives the length of the |
| 18353 | constructed portion of the |
| 18354 | substitute parse. */ |
| 18355 | bool first_time = TRUE; /* First multi-char occurrence doesn't get |
| 18356 | a "|" */ |
| 18357 | I32 reg_flags; |
| 18358 | |
| 18359 | assert(! invert); |
| 18360 | /* Only one level of recursion allowed */ |
| 18361 | assert(RExC_copy_start_in_constructed == RExC_precomp); |
| 18362 | |
| 18363 | #if 0 /* Have decided not to deal with multi-char folds in inverted classes, |
| 18364 | because too confusing */ |
| 18365 | if (invert) { |
| 18366 | sv_catpvs(substitute_parse, "(?:"); |
| 18367 | } |
| 18368 | #endif |
| 18369 | |
| 18370 | /* Look at the longest folds first */ |
| 18371 | for (cp_count = av_tindex_skip_len_mg(multi_char_matches); |
| 18372 | cp_count > 0; |
| 18373 | cp_count--) |
| 18374 | { |
| 18375 | |
| 18376 | if (av_exists(multi_char_matches, cp_count)) { |
| 18377 | AV** this_array_ptr; |
| 18378 | SV* this_sequence; |
| 18379 | |
| 18380 | this_array_ptr = (AV**) av_fetch(multi_char_matches, |
| 18381 | cp_count, FALSE); |
| 18382 | while ((this_sequence = av_pop(*this_array_ptr)) != |
| 18383 | &PL_sv_undef) |
| 18384 | { |
| 18385 | if (! first_time) { |
| 18386 | sv_catpvs(substitute_parse, "|"); |
| 18387 | } |
| 18388 | first_time = FALSE; |
| 18389 | |
| 18390 | sv_catpv(substitute_parse, SvPVX(this_sequence)); |
| 18391 | } |
| 18392 | } |
| 18393 | } |
| 18394 | |
| 18395 | /* If the character class contains anything else besides these |
| 18396 | * multi-character folds, have to include it in recursive parsing */ |
| 18397 | if (element_count) { |
| 18398 | sv_catpvs(substitute_parse, "|["); |
| 18399 | constructed_prefix_len = SvCUR(substitute_parse); |
| 18400 | sv_catpvn(substitute_parse, orig_parse, RExC_parse - orig_parse); |
| 18401 | |
| 18402 | /* Put in a closing ']' only if not going off the end, as otherwise |
| 18403 | * we are adding something that really isn't there */ |
| 18404 | if (RExC_parse < RExC_end) { |
| 18405 | sv_catpvs(substitute_parse, "]"); |
| 18406 | } |
| 18407 | } |
| 18408 | |
| 18409 | sv_catpvs(substitute_parse, ")"); |
| 18410 | #if 0 |
| 18411 | if (invert) { |
| 18412 | /* This is a way to get the parse to skip forward a whole named |
| 18413 | * sequence instead of matching the 2nd character when it fails the |
| 18414 | * first */ |
| 18415 | sv_catpvs(substitute_parse, "(*THEN)(*SKIP)(*FAIL)|.)"); |
| 18416 | } |
| 18417 | #endif |
| 18418 | |
| 18419 | /* Set up the data structure so that any errors will be properly |
| 18420 | * reported. See the comments at the definition of |
| 18421 | * REPORT_LOCATION_ARGS for details */ |
| 18422 | RExC_copy_start_in_input = (char *) orig_parse; |
| 18423 | RExC_start = RExC_parse = SvPV(substitute_parse, len); |
| 18424 | RExC_copy_start_in_constructed = RExC_start + constructed_prefix_len; |
| 18425 | RExC_end = RExC_parse + len; |
| 18426 | RExC_in_multi_char_class = 1; |
| 18427 | |
| 18428 | ret = reg(pRExC_state, 1, ®_flags, depth+1); |
| 18429 | |
| 18430 | *flagp |= reg_flags & (HASWIDTH|SIMPLE|SPSTART|POSTPONED|RESTART_PARSE|NEED_UTF8); |
| 18431 | |
| 18432 | /* And restore so can parse the rest of the pattern */ |
| 18433 | RExC_parse = save_parse; |
| 18434 | RExC_start = RExC_copy_start_in_constructed = RExC_copy_start_in_input = save_start; |
| 18435 | RExC_end = save_end; |
| 18436 | RExC_in_multi_char_class = 0; |
| 18437 | SvREFCNT_dec_NN(multi_char_matches); |
| 18438 | return ret; |
| 18439 | } |
| 18440 | |
| 18441 | /* If folding, we calculate all characters that could fold to or from the |
| 18442 | * ones already on the list */ |
| 18443 | if (cp_foldable_list) { |
| 18444 | if (FOLD) { |
| 18445 | UV start, end; /* End points of code point ranges */ |
| 18446 | |
| 18447 | SV* fold_intersection = NULL; |
| 18448 | SV** use_list; |
| 18449 | |
| 18450 | /* Our calculated list will be for Unicode rules. For locale |
| 18451 | * matching, we have to keep a separate list that is consulted at |
| 18452 | * runtime only when the locale indicates Unicode rules (and we |
| 18453 | * don't include potential matches in the ASCII/Latin1 range, as |
| 18454 | * any code point could fold to any other, based on the run-time |
| 18455 | * locale). For non-locale, we just use the general list */ |
| 18456 | if (LOC) { |
| 18457 | use_list = &only_utf8_locale_list; |
| 18458 | } |
| 18459 | else { |
| 18460 | use_list = &cp_list; |
| 18461 | } |
| 18462 | |
| 18463 | /* Only the characters in this class that participate in folds need |
| 18464 | * be checked. Get the intersection of this class and all the |
| 18465 | * possible characters that are foldable. This can quickly narrow |
| 18466 | * down a large class */ |
| 18467 | _invlist_intersection(PL_in_some_fold, cp_foldable_list, |
| 18468 | &fold_intersection); |
| 18469 | |
| 18470 | /* Now look at the foldable characters in this class individually */ |
| 18471 | invlist_iterinit(fold_intersection); |
| 18472 | while (invlist_iternext(fold_intersection, &start, &end)) { |
| 18473 | UV j; |
| 18474 | UV folded; |
| 18475 | |
| 18476 | /* Look at every character in the range */ |
| 18477 | for (j = start; j <= end; j++) { |
| 18478 | U8 foldbuf[UTF8_MAXBYTES_CASE+1]; |
| 18479 | STRLEN foldlen; |
| 18480 | unsigned int k; |
| 18481 | Size_t folds_count; |
| 18482 | U32 first_fold; |
| 18483 | const U32 * remaining_folds; |
| 18484 | |
| 18485 | if (j < 256) { |
| 18486 | |
| 18487 | /* Under /l, we don't know what code points below 256 |
| 18488 | * fold to, except we do know the MICRO SIGN folds to |
| 18489 | * an above-255 character if the locale is UTF-8, so we |
| 18490 | * add it to the special list (in *use_list) Otherwise |
| 18491 | * we know now what things can match, though some folds |
| 18492 | * are valid under /d only if the target is UTF-8. |
| 18493 | * Those go in a separate list */ |
| 18494 | if ( IS_IN_SOME_FOLD_L1(j) |
| 18495 | && ! (LOC && j != MICRO_SIGN)) |
| 18496 | { |
| 18497 | |
| 18498 | /* ASCII is always matched; non-ASCII is matched |
| 18499 | * only under Unicode rules (which could happen |
| 18500 | * under /l if the locale is a UTF-8 one */ |
| 18501 | if (isASCII(j) || ! DEPENDS_SEMANTICS) { |
| 18502 | *use_list = add_cp_to_invlist(*use_list, |
| 18503 | PL_fold_latin1[j]); |
| 18504 | } |
| 18505 | else if (j != PL_fold_latin1[j]) { |
| 18506 | upper_latin1_only_utf8_matches |
| 18507 | = add_cp_to_invlist( |
| 18508 | upper_latin1_only_utf8_matches, |
| 18509 | PL_fold_latin1[j]); |
| 18510 | } |
| 18511 | } |
| 18512 | |
| 18513 | if (HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(j) |
| 18514 | && (! isASCII(j) || ! ASCII_FOLD_RESTRICTED)) |
| 18515 | { |
| 18516 | add_above_Latin1_folds(pRExC_state, |
| 18517 | (U8) j, |
| 18518 | use_list); |
| 18519 | } |
| 18520 | continue; |
| 18521 | } |
| 18522 | |
| 18523 | /* Here is an above Latin1 character. We don't have the |
| 18524 | * rules hard-coded for it. First, get its fold. This is |
| 18525 | * the simple fold, as the multi-character folds have been |
| 18526 | * handled earlier and separated out */ |
| 18527 | folded = _to_uni_fold_flags(j, foldbuf, &foldlen, |
| 18528 | (ASCII_FOLD_RESTRICTED) |
| 18529 | ? FOLD_FLAGS_NOMIX_ASCII |
| 18530 | : 0); |
| 18531 | |
| 18532 | /* Single character fold of above Latin1. Add everything |
| 18533 | * in its fold closure to the list that this node should |
| 18534 | * match. */ |
| 18535 | folds_count = _inverse_folds(folded, &first_fold, |
| 18536 | &remaining_folds); |
| 18537 | for (k = 0; k <= folds_count; k++) { |
| 18538 | UV c = (k == 0) /* First time through use itself */ |
| 18539 | ? folded |
| 18540 | : (k == 1) /* 2nd time use, the first fold */ |
| 18541 | ? first_fold |
| 18542 | |
| 18543 | /* Then the remaining ones */ |
| 18544 | : remaining_folds[k-2]; |
| 18545 | |
| 18546 | /* /aa doesn't allow folds between ASCII and non- */ |
| 18547 | if (( ASCII_FOLD_RESTRICTED |
| 18548 | && (isASCII(c) != isASCII(j)))) |
| 18549 | { |
| 18550 | continue; |
| 18551 | } |
| 18552 | |
| 18553 | /* Folds under /l which cross the 255/256 boundary are |
| 18554 | * added to a separate list. (These are valid only |
| 18555 | * when the locale is UTF-8.) */ |
| 18556 | if (c < 256 && LOC) { |
| 18557 | *use_list = add_cp_to_invlist(*use_list, c); |
| 18558 | continue; |
| 18559 | } |
| 18560 | |
| 18561 | if (isASCII(c) || c > 255 || AT_LEAST_UNI_SEMANTICS) |
| 18562 | { |
| 18563 | cp_list = add_cp_to_invlist(cp_list, c); |
| 18564 | } |
| 18565 | else { |
| 18566 | /* Similarly folds involving non-ascii Latin1 |
| 18567 | * characters under /d are added to their list */ |
| 18568 | upper_latin1_only_utf8_matches |
| 18569 | = add_cp_to_invlist( |
| 18570 | upper_latin1_only_utf8_matches, |
| 18571 | c); |
| 18572 | } |
| 18573 | } |
| 18574 | } |
| 18575 | } |
| 18576 | SvREFCNT_dec_NN(fold_intersection); |
| 18577 | } |
| 18578 | |
| 18579 | /* Now that we have finished adding all the folds, there is no reason |
| 18580 | * to keep the foldable list separate */ |
| 18581 | _invlist_union(cp_list, cp_foldable_list, &cp_list); |
| 18582 | SvREFCNT_dec_NN(cp_foldable_list); |
| 18583 | } |
| 18584 | |
| 18585 | /* And combine the result (if any) with any inversion lists from posix |
| 18586 | * classes. The lists are kept separate up to now because we don't want to |
| 18587 | * fold the classes */ |
| 18588 | if (simple_posixes) { /* These are the classes known to be unaffected by |
| 18589 | /a, /aa, and /d */ |
| 18590 | if (cp_list) { |
| 18591 | _invlist_union(cp_list, simple_posixes, &cp_list); |
| 18592 | SvREFCNT_dec_NN(simple_posixes); |
| 18593 | } |
| 18594 | else { |
| 18595 | cp_list = simple_posixes; |
| 18596 | } |
| 18597 | } |
| 18598 | if (posixes || nposixes) { |
| 18599 | if (! DEPENDS_SEMANTICS) { |
| 18600 | |
| 18601 | /* For everything but /d, we can just add the current 'posixes' and |
| 18602 | * 'nposixes' to the main list */ |
| 18603 | if (posixes) { |
| 18604 | if (cp_list) { |
| 18605 | _invlist_union(cp_list, posixes, &cp_list); |
| 18606 | SvREFCNT_dec_NN(posixes); |
| 18607 | } |
| 18608 | else { |
| 18609 | cp_list = posixes; |
| 18610 | } |
| 18611 | } |
| 18612 | if (nposixes) { |
| 18613 | if (cp_list) { |
| 18614 | _invlist_union(cp_list, nposixes, &cp_list); |
| 18615 | SvREFCNT_dec_NN(nposixes); |
| 18616 | } |
| 18617 | else { |
| 18618 | cp_list = nposixes; |
| 18619 | } |
| 18620 | } |
| 18621 | } |
| 18622 | else { |
| 18623 | /* Under /d, things like \w match upper Latin1 characters only if |
| 18624 | * the target string is in UTF-8. But things like \W match all the |
| 18625 | * upper Latin1 characters if the target string is not in UTF-8. |
| 18626 | * |
| 18627 | * Handle the case with something like \W separately */ |
| 18628 | if (nposixes) { |
| 18629 | SV* only_non_utf8_list = invlist_clone(PL_UpperLatin1, NULL); |
| 18630 | |
| 18631 | /* A complemented posix class matches all upper Latin1 |
| 18632 | * characters if not in UTF-8. And it matches just certain |
| 18633 | * ones when in UTF-8. That means those certain ones are |
| 18634 | * matched regardless, so can just be added to the |
| 18635 | * unconditional list */ |
| 18636 | if (cp_list) { |
| 18637 | _invlist_union(cp_list, nposixes, &cp_list); |
| 18638 | SvREFCNT_dec_NN(nposixes); |
| 18639 | nposixes = NULL; |
| 18640 | } |
| 18641 | else { |
| 18642 | cp_list = nposixes; |
| 18643 | } |
| 18644 | |
| 18645 | /* Likewise for 'posixes' */ |
| 18646 | _invlist_union(posixes, cp_list, &cp_list); |
| 18647 | SvREFCNT_dec(posixes); |
| 18648 | |
| 18649 | /* Likewise for anything else in the range that matched only |
| 18650 | * under UTF-8 */ |
| 18651 | if (upper_latin1_only_utf8_matches) { |
| 18652 | _invlist_union(cp_list, |
| 18653 | upper_latin1_only_utf8_matches, |
| 18654 | &cp_list); |
| 18655 | SvREFCNT_dec_NN(upper_latin1_only_utf8_matches); |
| 18656 | upper_latin1_only_utf8_matches = NULL; |
| 18657 | } |
| 18658 | |
| 18659 | /* If we don't match all the upper Latin1 characters regardless |
| 18660 | * of UTF-8ness, we have to set a flag to match the rest when |
| 18661 | * not in UTF-8 */ |
| 18662 | _invlist_subtract(only_non_utf8_list, cp_list, |
| 18663 | &only_non_utf8_list); |
| 18664 | if (_invlist_len(only_non_utf8_list) != 0) { |
| 18665 | anyof_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER; |
| 18666 | } |
| 18667 | SvREFCNT_dec_NN(only_non_utf8_list); |
| 18668 | } |
| 18669 | else { |
| 18670 | /* Here there were no complemented posix classes. That means |
| 18671 | * the upper Latin1 characters in 'posixes' match only when the |
| 18672 | * target string is in UTF-8. So we have to add them to the |
| 18673 | * list of those types of code points, while adding the |
| 18674 | * remainder to the unconditional list. |
| 18675 | * |
| 18676 | * First calculate what they are */ |
| 18677 | SV* nonascii_but_latin1_properties = NULL; |
| 18678 | _invlist_intersection(posixes, PL_UpperLatin1, |
| 18679 | &nonascii_but_latin1_properties); |
| 18680 | |
| 18681 | /* And add them to the final list of such characters. */ |
| 18682 | _invlist_union(upper_latin1_only_utf8_matches, |
| 18683 | nonascii_but_latin1_properties, |
| 18684 | &upper_latin1_only_utf8_matches); |
| 18685 | |
| 18686 | /* Remove them from what now becomes the unconditional list */ |
| 18687 | _invlist_subtract(posixes, nonascii_but_latin1_properties, |
| 18688 | &posixes); |
| 18689 | |
| 18690 | /* And add those unconditional ones to the final list */ |
| 18691 | if (cp_list) { |
| 18692 | _invlist_union(cp_list, posixes, &cp_list); |
| 18693 | SvREFCNT_dec_NN(posixes); |
| 18694 | posixes = NULL; |
| 18695 | } |
| 18696 | else { |
| 18697 | cp_list = posixes; |
| 18698 | } |
| 18699 | |
| 18700 | SvREFCNT_dec(nonascii_but_latin1_properties); |
| 18701 | |
| 18702 | /* Get rid of any characters from the conditional list that we |
| 18703 | * now know are matched unconditionally, which may make that |
| 18704 | * list empty */ |
| 18705 | _invlist_subtract(upper_latin1_only_utf8_matches, |
| 18706 | cp_list, |
| 18707 | &upper_latin1_only_utf8_matches); |
| 18708 | if (_invlist_len(upper_latin1_only_utf8_matches) == 0) { |
| 18709 | SvREFCNT_dec_NN(upper_latin1_only_utf8_matches); |
| 18710 | upper_latin1_only_utf8_matches = NULL; |
| 18711 | } |
| 18712 | } |
| 18713 | } |
| 18714 | } |
| 18715 | |
| 18716 | /* And combine the result (if any) with any inversion list from properties. |
| 18717 | * The lists are kept separate up to now so that we can distinguish the two |
| 18718 | * in regards to matching above-Unicode. A run-time warning is generated |
| 18719 | * if a Unicode property is matched against a non-Unicode code point. But, |
| 18720 | * we allow user-defined properties to match anything, without any warning, |
| 18721 | * and we also suppress the warning if there is a portion of the character |
| 18722 | * class that isn't a Unicode property, and which matches above Unicode, \W |
| 18723 | * or [\x{110000}] for example. |
| 18724 | * (Note that in this case, unlike the Posix one above, there is no |
| 18725 | * <upper_latin1_only_utf8_matches>, because having a Unicode property |
| 18726 | * forces Unicode semantics */ |
| 18727 | if (properties) { |
| 18728 | if (cp_list) { |
| 18729 | |
| 18730 | /* If it matters to the final outcome, see if a non-property |
| 18731 | * component of the class matches above Unicode. If so, the |
| 18732 | * warning gets suppressed. This is true even if just a single |
| 18733 | * such code point is specified, as, though not strictly correct if |
| 18734 | * another such code point is matched against, the fact that they |
| 18735 | * are using above-Unicode code points indicates they should know |
| 18736 | * the issues involved */ |
| 18737 | if (warn_super) { |
| 18738 | warn_super = ! (invert |
| 18739 | ^ (invlist_highest(cp_list) > PERL_UNICODE_MAX)); |
| 18740 | } |
| 18741 | |
| 18742 | _invlist_union(properties, cp_list, &cp_list); |
| 18743 | SvREFCNT_dec_NN(properties); |
| 18744 | } |
| 18745 | else { |
| 18746 | cp_list = properties; |
| 18747 | } |
| 18748 | |
| 18749 | if (warn_super) { |
| 18750 | anyof_flags |
| 18751 | |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER; |
| 18752 | |
| 18753 | /* Because an ANYOF node is the only one that warns, this node |
| 18754 | * can't be optimized into something else */ |
| 18755 | optimizable = FALSE; |
| 18756 | } |
| 18757 | } |
| 18758 | |
| 18759 | /* Here, we have calculated what code points should be in the character |
| 18760 | * class. |
| 18761 | * |
| 18762 | * Now we can see about various optimizations. Fold calculation (which we |
| 18763 | * did above) needs to take place before inversion. Otherwise /[^k]/i |
| 18764 | * would invert to include K, which under /i would match k, which it |
| 18765 | * shouldn't. Therefore we can't invert folded locale now, as it won't be |
| 18766 | * folded until runtime */ |
| 18767 | |
| 18768 | /* If we didn't do folding, it's because some information isn't available |
| 18769 | * until runtime; set the run-time fold flag for these We know to set the |
| 18770 | * flag if we have a non-NULL list for UTF-8 locales, or the class matches |
| 18771 | * at least one 0-255 range code point */ |
| 18772 | if (LOC && FOLD) { |
| 18773 | |
| 18774 | /* Some things on the list might be unconditionally included because of |
| 18775 | * other components. Remove them, and clean up the list if it goes to |
| 18776 | * 0 elements */ |
| 18777 | if (only_utf8_locale_list && cp_list) { |
| 18778 | _invlist_subtract(only_utf8_locale_list, cp_list, |
| 18779 | &only_utf8_locale_list); |
| 18780 | |
| 18781 | if (_invlist_len(only_utf8_locale_list) == 0) { |
| 18782 | SvREFCNT_dec_NN(only_utf8_locale_list); |
| 18783 | only_utf8_locale_list = NULL; |
| 18784 | } |
| 18785 | } |
| 18786 | if ( only_utf8_locale_list |
| 18787 | || (cp_list && ( _invlist_contains_cp(cp_list, LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) |
| 18788 | || _invlist_contains_cp(cp_list, LATIN_SMALL_LETTER_DOTLESS_I)))) |
| 18789 | { |
| 18790 | has_runtime_dependency |= HAS_L_RUNTIME_DEPENDENCY; |
| 18791 | anyof_flags |
| 18792 | |= ANYOFL_FOLD |
| 18793 | | ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD; |
| 18794 | } |
| 18795 | else if (cp_list && invlist_lowest(cp_list) < 256) { |
| 18796 | /* If nothing is below 256, has no locale dependency; otherwise it |
| 18797 | * does */ |
| 18798 | anyof_flags |= ANYOFL_FOLD; |
| 18799 | has_runtime_dependency |= HAS_L_RUNTIME_DEPENDENCY; |
| 18800 | } |
| 18801 | } |
| 18802 | else if ( DEPENDS_SEMANTICS |
| 18803 | && ( upper_latin1_only_utf8_matches |
| 18804 | || (anyof_flags & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))) |
| 18805 | { |
| 18806 | RExC_seen_d_op = TRUE; |
| 18807 | has_runtime_dependency |= HAS_D_RUNTIME_DEPENDENCY; |
| 18808 | } |
| 18809 | |
| 18810 | /* Optimize inverted patterns (e.g. [^a-z]) when everything is known at |
| 18811 | * compile time. */ |
| 18812 | if ( cp_list |
| 18813 | && invert |
| 18814 | && ! has_runtime_dependency) |
| 18815 | { |
| 18816 | _invlist_invert(cp_list); |
| 18817 | |
| 18818 | /* Clear the invert flag since have just done it here */ |
| 18819 | invert = FALSE; |
| 18820 | } |
| 18821 | |
| 18822 | /* All possible optimizations below still have these characteristics. |
| 18823 | * (Multi-char folds aren't SIMPLE, but they don't get this far in this |
| 18824 | * routine) */ |
| 18825 | *flagp |= HASWIDTH|SIMPLE; |
| 18826 | |
| 18827 | if (ret_invlist) { |
| 18828 | *ret_invlist = cp_list; |
| 18829 | |
| 18830 | return RExC_emit; |
| 18831 | } |
| 18832 | |
| 18833 | if (anyof_flags & ANYOF_LOCALE_FLAGS) { |
| 18834 | RExC_contains_locale = 1; |
| 18835 | } |
| 18836 | |
| 18837 | /* Some character classes are equivalent to other nodes. Such nodes take |
| 18838 | * up less room, and some nodes require fewer operations to execute, than |
| 18839 | * ANYOF nodes. EXACTish nodes may be joinable with adjacent nodes to |
| 18840 | * improve efficiency. */ |
| 18841 | |
| 18842 | if (optimizable) { |
| 18843 | PERL_UINT_FAST8_T i; |
| 18844 | UV partial_cp_count = 0; |
| 18845 | UV start[MAX_FOLD_FROMS+1] = { 0 }; /* +1 for the folded-to char */ |
| 18846 | UV end[MAX_FOLD_FROMS+1] = { 0 }; |
| 18847 | bool single_range = FALSE; |
| 18848 | |
| 18849 | if (cp_list) { /* Count the code points in enough ranges that we would |
| 18850 | see all the ones possible in any fold in this version |
| 18851 | of Unicode */ |
| 18852 | |
| 18853 | invlist_iterinit(cp_list); |
| 18854 | for (i = 0; i <= MAX_FOLD_FROMS; i++) { |
| 18855 | if (! invlist_iternext(cp_list, &start[i], &end[i])) { |
| 18856 | break; |
| 18857 | } |
| 18858 | partial_cp_count += end[i] - start[i] + 1; |
| 18859 | } |
| 18860 | |
| 18861 | if (i == 1) { |
| 18862 | single_range = TRUE; |
| 18863 | } |
| 18864 | invlist_iterfinish(cp_list); |
| 18865 | } |
| 18866 | |
| 18867 | /* If we know at compile time that this matches every possible code |
| 18868 | * point, any run-time dependencies don't matter */ |
| 18869 | if (start[0] == 0 && end[0] == UV_MAX) { |
| 18870 | if (invert) { |
| 18871 | ret = reganode(pRExC_state, OPFAIL, 0); |
| 18872 | } |
| 18873 | else { |
| 18874 | ret = reg_node(pRExC_state, SANY); |
| 18875 | MARK_NAUGHTY(1); |
| 18876 | } |
| 18877 | goto not_anyof; |
| 18878 | } |
| 18879 | |
| 18880 | /* Similarly, for /l posix classes, if both a class and its |
| 18881 | * complement match, any run-time dependencies don't matter */ |
| 18882 | if (posixl) { |
| 18883 | for (namedclass = 0; namedclass < ANYOF_POSIXL_MAX; |
| 18884 | namedclass += 2) |
| 18885 | { |
| 18886 | if ( POSIXL_TEST(posixl, namedclass) /* class */ |
| 18887 | && POSIXL_TEST(posixl, namedclass + 1)) /* its complement */ |
| 18888 | { |
| 18889 | if (invert) { |
| 18890 | ret = reganode(pRExC_state, OPFAIL, 0); |
| 18891 | } |
| 18892 | else { |
| 18893 | ret = reg_node(pRExC_state, SANY); |
| 18894 | MARK_NAUGHTY(1); |
| 18895 | } |
| 18896 | goto not_anyof; |
| 18897 | } |
| 18898 | } |
| 18899 | |
| 18900 | /* For well-behaved locales, some classes are subsets of others, |
| 18901 | * so complementing the subset and including the non-complemented |
| 18902 | * superset should match everything, like [\D[:alnum:]], and |
| 18903 | * [[:^alpha:][:alnum:]], but some implementations of locales are |
| 18904 | * buggy, and khw thinks its a bad idea to have optimization change |
| 18905 | * behavior, even if it avoids an OS bug in a given case */ |
| 18906 | |
| 18907 | #define isSINGLE_BIT_SET(n) isPOWER_OF_2(n) |
| 18908 | |
| 18909 | /* If is a single posix /l class, can optimize to just that op. |
| 18910 | * Such a node will not match anything in the Latin1 range, as that |
| 18911 | * is not determinable until runtime, but will match whatever the |
| 18912 | * class does outside that range. (Note that some classes won't |
| 18913 | * match anything outside the range, like [:ascii:]) */ |
| 18914 | if ( isSINGLE_BIT_SET(posixl) |
| 18915 | && (partial_cp_count == 0 || start[0] > 255)) |
| 18916 | { |
| 18917 | U8 classnum; |
| 18918 | SV * class_above_latin1 = NULL; |
| 18919 | bool already_inverted; |
| 18920 | bool are_equivalent; |
| 18921 | |
| 18922 | /* Compute which bit is set, which is the same thing as, e.g., |
| 18923 | * ANYOF_CNTRL. From |
| 18924 | * https://graphics.stanford.edu/~seander/bithacks.html#IntegerLogDeBruijn |
| 18925 | * */ |
| 18926 | static const int MultiplyDeBruijnBitPosition2[32] = |
| 18927 | { |
| 18928 | 0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8, |
| 18929 | 31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9 |
| 18930 | }; |
| 18931 | |
| 18932 | namedclass = MultiplyDeBruijnBitPosition2[(posixl |
| 18933 | * 0x077CB531U) >> 27]; |
| 18934 | classnum = namedclass_to_classnum(namedclass); |
| 18935 | |
| 18936 | /* The named classes are such that the inverted number is one |
| 18937 | * larger than the non-inverted one */ |
| 18938 | already_inverted = namedclass |
| 18939 | - classnum_to_namedclass(classnum); |
| 18940 | |
| 18941 | /* Create an inversion list of the official property, inverted |
| 18942 | * if the constructed node list is inverted, and restricted to |
| 18943 | * only the above latin1 code points, which are the only ones |
| 18944 | * known at compile time */ |
| 18945 | _invlist_intersection_maybe_complement_2nd( |
| 18946 | PL_AboveLatin1, |
| 18947 | PL_XPosix_ptrs[classnum], |
| 18948 | already_inverted, |
| 18949 | &class_above_latin1); |
| 18950 | are_equivalent = _invlistEQ(class_above_latin1, cp_list, |
| 18951 | FALSE); |
| 18952 | SvREFCNT_dec_NN(class_above_latin1); |
| 18953 | |
| 18954 | if (are_equivalent) { |
| 18955 | |
| 18956 | /* Resolve the run-time inversion flag with this possibly |
| 18957 | * inverted class */ |
| 18958 | invert = invert ^ already_inverted; |
| 18959 | |
| 18960 | ret = reg_node(pRExC_state, |
| 18961 | POSIXL + invert * (NPOSIXL - POSIXL)); |
| 18962 | FLAGS(REGNODE_p(ret)) = classnum; |
| 18963 | goto not_anyof; |
| 18964 | } |
| 18965 | } |
| 18966 | } |
| 18967 | |
| 18968 | /* khw can't think of any other possible transformation involving |
| 18969 | * these. */ |
| 18970 | if (has_runtime_dependency & HAS_USER_DEFINED_PROPERTY) { |
| 18971 | goto is_anyof; |
| 18972 | } |
| 18973 | |
| 18974 | if (! has_runtime_dependency) { |
| 18975 | |
| 18976 | /* If the list is empty, nothing matches. This happens, for |
| 18977 | * example, when a Unicode property that doesn't match anything is |
| 18978 | * the only element in the character class (perluniprops.pod notes |
| 18979 | * such properties). */ |
| 18980 | if (partial_cp_count == 0) { |
| 18981 | if (invert) { |
| 18982 | ret = reg_node(pRExC_state, SANY); |
| 18983 | } |
| 18984 | else { |
| 18985 | ret = reganode(pRExC_state, OPFAIL, 0); |
| 18986 | } |
| 18987 | |
| 18988 | goto not_anyof; |
| 18989 | } |
| 18990 | |
| 18991 | /* If matches everything but \n */ |
| 18992 | if ( start[0] == 0 && end[0] == '\n' - 1 |
| 18993 | && start[1] == '\n' + 1 && end[1] == UV_MAX) |
| 18994 | { |
| 18995 | assert (! invert); |
| 18996 | ret = reg_node(pRExC_state, REG_ANY); |
| 18997 | MARK_NAUGHTY(1); |
| 18998 | goto not_anyof; |
| 18999 | } |
| 19000 | } |
| 19001 | |
| 19002 | /* Next see if can optimize classes that contain just a few code points |
| 19003 | * into an EXACTish node. The reason to do this is to let the |
| 19004 | * optimizer join this node with adjacent EXACTish ones, and ANYOF |
| 19005 | * nodes require conversion to code point from UTF-8. |
| 19006 | * |
| 19007 | * An EXACTFish node can be generated even if not under /i, and vice |
| 19008 | * versa. But care must be taken. An EXACTFish node has to be such |
| 19009 | * that it only matches precisely the code points in the class, but we |
| 19010 | * want to generate the least restrictive one that does that, to |
| 19011 | * increase the odds of being able to join with an adjacent node. For |
| 19012 | * example, if the class contains [kK], we have to make it an EXACTFAA |
| 19013 | * node to prevent the KELVIN SIGN from matching. Whether we are under |
| 19014 | * /i or not is irrelevant in this case. Less obvious is the pattern |
| 19015 | * qr/[\x{02BC}]n/i. U+02BC is MODIFIER LETTER APOSTROPHE. That is |
| 19016 | * supposed to match the single character U+0149 LATIN SMALL LETTER N |
| 19017 | * PRECEDED BY APOSTROPHE. And so even though there is no simple fold |
| 19018 | * that includes \X{02BC}, there is a multi-char fold that does, and so |
| 19019 | * the node generated for it must be an EXACTFish one. On the other |
| 19020 | * hand qr/:/i should generate a plain EXACT node since the colon |
| 19021 | * participates in no fold whatsoever, and having it EXACT tells the |
| 19022 | * optimizer the target string cannot match unless it has a colon in |
| 19023 | * it. |
| 19024 | */ |
| 19025 | if ( ! posixl |
| 19026 | && ! invert |
| 19027 | |
| 19028 | /* Only try if there are no more code points in the class than |
| 19029 | * in the max possible fold */ |
| 19030 | && inRANGE(partial_cp_count, 1, MAX_FOLD_FROMS + 1)) |
| 19031 | { |
| 19032 | if (partial_cp_count == 1 && ! upper_latin1_only_utf8_matches) |
| 19033 | { |
| 19034 | /* We can always make a single code point class into an |
| 19035 | * EXACTish node. */ |
| 19036 | |
| 19037 | if (LOC) { |
| 19038 | |
| 19039 | /* Here is /l: Use EXACTL, except if there is a fold not |
| 19040 | * known until runtime so shows as only a single code point |
| 19041 | * here. For code points above 255, we know which can |
| 19042 | * cause problems by having a potential fold to the Latin1 |
| 19043 | * range. */ |
| 19044 | if ( ! FOLD |
| 19045 | || ( start[0] > 255 |
| 19046 | && ! is_PROBLEMATIC_LOCALE_FOLD_cp(start[0]))) |
| 19047 | { |
| 19048 | op = EXACTL; |
| 19049 | } |
| 19050 | else { |
| 19051 | op = EXACTFL; |
| 19052 | } |
| 19053 | } |
| 19054 | else if (! FOLD) { /* Not /l and not /i */ |
| 19055 | op = (start[0] < 256) ? EXACT : EXACT_REQ8; |
| 19056 | } |
| 19057 | else if (start[0] < 256) { /* /i, not /l, and the code point is |
| 19058 | small */ |
| 19059 | |
| 19060 | /* Under /i, it gets a little tricky. A code point that |
| 19061 | * doesn't participate in a fold should be an EXACT node. |
| 19062 | * We know this one isn't the result of a simple fold, or |
| 19063 | * there'd be more than one code point in the list, but it |
| 19064 | * could be part of a multi- character fold. In that case |
| 19065 | * we better not create an EXACT node, as we would wrongly |
| 19066 | * be telling the optimizer that this code point must be in |
| 19067 | * the target string, and that is wrong. This is because |
| 19068 | * if the sequence around this code point forms a |
| 19069 | * multi-char fold, what needs to be in the string could be |
| 19070 | * the code point that folds to the sequence. |
| 19071 | * |
| 19072 | * This handles the case of below-255 code points, as we |
| 19073 | * have an easy look up for those. The next clause handles |
| 19074 | * the above-256 one */ |
| 19075 | op = IS_IN_SOME_FOLD_L1(start[0]) |
| 19076 | ? EXACTFU |
| 19077 | : EXACT; |
| 19078 | } |
| 19079 | else { /* /i, larger code point. Since we are under /i, and |
| 19080 | have just this code point, we know that it can't |
| 19081 | fold to something else, so PL_InMultiCharFold |
| 19082 | applies to it */ |
| 19083 | op = _invlist_contains_cp(PL_InMultiCharFold, |
| 19084 | start[0]) |
| 19085 | ? EXACTFU_REQ8 |
| 19086 | : EXACT_REQ8; |
| 19087 | } |
| 19088 | |
| 19089 | value = start[0]; |
| 19090 | } |
| 19091 | else if ( ! (has_runtime_dependency & ~HAS_D_RUNTIME_DEPENDENCY) |
| 19092 | && _invlist_contains_cp(PL_in_some_fold, start[0])) |
| 19093 | { |
| 19094 | /* Here, the only runtime dependency, if any, is from /d, and |
| 19095 | * the class matches more than one code point, and the lowest |
| 19096 | * code point participates in some fold. It might be that the |
| 19097 | * other code points are /i equivalent to this one, and hence |
| 19098 | * they would representable by an EXACTFish node. Above, we |
| 19099 | * eliminated classes that contain too many code points to be |
| 19100 | * EXACTFish, with the test for MAX_FOLD_FROMS |
| 19101 | * |
| 19102 | * First, special case the ASCII fold pairs, like 'B' and 'b'. |
| 19103 | * We do this because we have EXACTFAA at our disposal for the |
| 19104 | * ASCII range */ |
| 19105 | if (partial_cp_count == 2 && isASCII(start[0])) { |
| 19106 | |
| 19107 | /* The only ASCII characters that participate in folds are |
| 19108 | * alphabetics */ |
| 19109 | assert(isALPHA(start[0])); |
| 19110 | if ( end[0] == start[0] /* First range is a single |
| 19111 | character, so 2nd exists */ |
| 19112 | && isALPHA_FOLD_EQ(start[0], start[1])) |
| 19113 | { |
| 19114 | |
| 19115 | /* Here, is part of an ASCII fold pair */ |
| 19116 | |
| 19117 | if ( ASCII_FOLD_RESTRICTED |
| 19118 | || HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(start[0])) |
| 19119 | { |
| 19120 | /* If the second clause just above was true, it |
| 19121 | * means we can't be under /i, or else the list |
| 19122 | * would have included more than this fold pair. |
| 19123 | * Therefore we have to exclude the possibility of |
| 19124 | * whatever else it is that folds to these, by |
| 19125 | * using EXACTFAA */ |
| 19126 | op = EXACTFAA; |
| 19127 | } |
| 19128 | else if (HAS_NONLATIN1_FOLD_CLOSURE(start[0])) { |
| 19129 | |
| 19130 | /* Here, there's no simple fold that start[0] is part |
| 19131 | * of, but there is a multi-character one. If we |
| 19132 | * are not under /i, we want to exclude that |
| 19133 | * possibility; if under /i, we want to include it |
| 19134 | * */ |
| 19135 | op = (FOLD) ? EXACTFU : EXACTFAA; |
| 19136 | } |
| 19137 | else { |
| 19138 | |
| 19139 | /* Here, the only possible fold start[0] particpates in |
| 19140 | * is with start[1]. /i or not isn't relevant */ |
| 19141 | op = EXACTFU; |
| 19142 | } |
| 19143 | |
| 19144 | value = toFOLD(start[0]); |
| 19145 | } |
| 19146 | } |
| 19147 | else if ( ! upper_latin1_only_utf8_matches |
| 19148 | || ( _invlist_len(upper_latin1_only_utf8_matches) |
| 19149 | == 2 |
| 19150 | && PL_fold_latin1[ |
| 19151 | invlist_highest(upper_latin1_only_utf8_matches)] |
| 19152 | == start[0])) |
| 19153 | { |
| 19154 | /* Here, the smallest character is non-ascii or there are |
| 19155 | * more than 2 code points matched by this node. Also, we |
| 19156 | * either don't have /d UTF-8 dependent matches, or if we |
| 19157 | * do, they look like they could be a single character that |
| 19158 | * is the fold of the lowest one in the always-match list. |
| 19159 | * This test quickly excludes most of the false positives |
| 19160 | * when there are /d UTF-8 depdendent matches. These are |
| 19161 | * like LATIN CAPITAL LETTER A WITH GRAVE matching LATIN |
| 19162 | * SMALL LETTER A WITH GRAVE iff the target string is |
| 19163 | * UTF-8. (We don't have to worry above about exceeding |
| 19164 | * the array bounds of PL_fold_latin1[] because any code |
| 19165 | * point in 'upper_latin1_only_utf8_matches' is below 256.) |
| 19166 | * |
| 19167 | * EXACTFAA would apply only to pairs (hence exactly 2 code |
| 19168 | * points) in the ASCII range, so we can't use it here to |
| 19169 | * artificially restrict the fold domain, so we check if |
| 19170 | * the class does or does not match some EXACTFish node. |
| 19171 | * Further, if we aren't under /i, and and the folded-to |
| 19172 | * character is part of a multi-character fold, we can't do |
| 19173 | * this optimization, as the sequence around it could be |
| 19174 | * that multi-character fold, and we don't here know the |
| 19175 | * context, so we have to assume it is that multi-char |
| 19176 | * fold, to prevent potential bugs. |
| 19177 | * |
| 19178 | * To do the general case, we first find the fold of the |
| 19179 | * lowest code point (which may be higher than the lowest |
| 19180 | * one), then find everything that folds to it. (The data |
| 19181 | * structure we have only maps from the folded code points, |
| 19182 | * so we have to do the earlier step.) */ |
| 19183 | |
| 19184 | Size_t foldlen; |
| 19185 | U8 foldbuf[UTF8_MAXBYTES_CASE]; |
| 19186 | UV folded = _to_uni_fold_flags(start[0], |
| 19187 | foldbuf, &foldlen, 0); |
| 19188 | U32 first_fold; |
| 19189 | const U32 * remaining_folds; |
| 19190 | Size_t folds_to_this_cp_count = _inverse_folds( |
| 19191 | folded, |
| 19192 | &first_fold, |
| 19193 | &remaining_folds); |
| 19194 | Size_t folds_count = folds_to_this_cp_count + 1; |
| 19195 | SV * fold_list = _new_invlist(folds_count); |
| 19196 | unsigned int i; |
| 19197 | |
| 19198 | /* If there are UTF-8 dependent matches, create a temporary |
| 19199 | * list of what this node matches, including them. */ |
| 19200 | SV * all_cp_list = NULL; |
| 19201 | SV ** use_this_list = &cp_list; |
| 19202 | |
| 19203 | if (upper_latin1_only_utf8_matches) { |
| 19204 | all_cp_list = _new_invlist(0); |
| 19205 | use_this_list = &all_cp_list; |
| 19206 | _invlist_union(cp_list, |
| 19207 | upper_latin1_only_utf8_matches, |
| 19208 | use_this_list); |
| 19209 | } |
| 19210 | |
| 19211 | /* Having gotten everything that participates in the fold |
| 19212 | * containing the lowest code point, we turn that into an |
| 19213 | * inversion list, making sure everything is included. */ |
| 19214 | fold_list = add_cp_to_invlist(fold_list, start[0]); |
| 19215 | fold_list = add_cp_to_invlist(fold_list, folded); |
| 19216 | if (folds_to_this_cp_count > 0) { |
| 19217 | fold_list = add_cp_to_invlist(fold_list, first_fold); |
| 19218 | for (i = 0; i + 1 < folds_to_this_cp_count; i++) { |
| 19219 | fold_list = add_cp_to_invlist(fold_list, |
| 19220 | remaining_folds[i]); |
| 19221 | } |
| 19222 | } |
| 19223 | |
| 19224 | /* If the fold list is identical to what's in this ANYOF |
| 19225 | * node, the node can be represented by an EXACTFish one |
| 19226 | * instead */ |
| 19227 | if (_invlistEQ(*use_this_list, fold_list, |
| 19228 | 0 /* Don't complement */ ) |
| 19229 | ) { |
| 19230 | |
| 19231 | /* But, we have to be careful, as mentioned above. |
| 19232 | * Just the right sequence of characters could match |
| 19233 | * this if it is part of a multi-character fold. That |
| 19234 | * IS what we want if we are under /i. But it ISN'T |
| 19235 | * what we want if not under /i, as it could match when |
| 19236 | * it shouldn't. So, when we aren't under /i and this |
| 19237 | * character participates in a multi-char fold, we |
| 19238 | * don't optimize into an EXACTFish node. So, for each |
| 19239 | * case below we have to check if we are folding |
| 19240 | * and if not, if it is not part of a multi-char fold. |
| 19241 | * */ |
| 19242 | if (start[0] > 255) { /* Highish code point */ |
| 19243 | if (FOLD || ! _invlist_contains_cp( |
| 19244 | PL_InMultiCharFold, folded)) |
| 19245 | { |
| 19246 | op = (LOC) |
| 19247 | ? EXACTFLU8 |
| 19248 | : (ASCII_FOLD_RESTRICTED) |
| 19249 | ? EXACTFAA |
| 19250 | : EXACTFU_REQ8; |
| 19251 | value = folded; |
| 19252 | } |
| 19253 | } /* Below, the lowest code point < 256 */ |
| 19254 | else if ( FOLD |
| 19255 | && folded == 's' |
| 19256 | && DEPENDS_SEMANTICS) |
| 19257 | { /* An EXACTF node containing a single character |
| 19258 | 's', can be an EXACTFU if it doesn't get |
| 19259 | joined with an adjacent 's' */ |
| 19260 | op = EXACTFU_S_EDGE; |
| 19261 | value = folded; |
| 19262 | } |
| 19263 | else if ( FOLD |
| 19264 | || ! HAS_NONLATIN1_FOLD_CLOSURE(start[0])) |
| 19265 | { |
| 19266 | if (upper_latin1_only_utf8_matches) { |
| 19267 | op = EXACTF; |
| 19268 | |
| 19269 | /* We can't use the fold, as that only matches |
| 19270 | * under UTF-8 */ |
| 19271 | value = start[0]; |
| 19272 | } |
| 19273 | else if ( UNLIKELY(start[0] == MICRO_SIGN) |
| 19274 | && ! UTF) |
| 19275 | { /* EXACTFUP is a special node for this |
| 19276 | character */ |
| 19277 | op = (ASCII_FOLD_RESTRICTED) |
| 19278 | ? EXACTFAA |
| 19279 | : EXACTFUP; |
| 19280 | value = MICRO_SIGN; |
| 19281 | } |
| 19282 | else if ( ASCII_FOLD_RESTRICTED |
| 19283 | && ! isASCII(start[0])) |
| 19284 | { /* For ASCII under /iaa, we can use EXACTFU |
| 19285 | below */ |
| 19286 | op = EXACTFAA; |
| 19287 | value = folded; |
| 19288 | } |
| 19289 | else { |
| 19290 | op = EXACTFU; |
| 19291 | value = folded; |
| 19292 | } |
| 19293 | } |
| 19294 | } |
| 19295 | |
| 19296 | SvREFCNT_dec_NN(fold_list); |
| 19297 | SvREFCNT_dec(all_cp_list); |
| 19298 | } |
| 19299 | } |
| 19300 | |
| 19301 | if (op != END) { |
| 19302 | U8 len; |
| 19303 | |
| 19304 | /* Here, we have calculated what EXACTish node to use. Have to |
| 19305 | * convert to UTF-8 if not already there */ |
| 19306 | if (value > 255) { |
| 19307 | if (! UTF) { |
| 19308 | SvREFCNT_dec(cp_list);; |
| 19309 | REQUIRE_UTF8(flagp); |
| 19310 | } |
| 19311 | |
| 19312 | /* This is a kludge to the special casing issues with this |
| 19313 | * ligature under /aa. FB05 should fold to FB06, but the |
| 19314 | * call above to _to_uni_fold_flags() didn't find this, as |
| 19315 | * it didn't use the /aa restriction in order to not miss |
| 19316 | * other folds that would be affected. This is the only |
| 19317 | * instance likely to ever be a problem in all of Unicode. |
| 19318 | * So special case it. */ |
| 19319 | if ( value == LATIN_SMALL_LIGATURE_LONG_S_T |
| 19320 | && ASCII_FOLD_RESTRICTED) |
| 19321 | { |
| 19322 | value = LATIN_SMALL_LIGATURE_ST; |
| 19323 | } |
| 19324 | } |
| 19325 | |
| 19326 | len = (UTF) ? UVCHR_SKIP(value) : 1; |
| 19327 | |
| 19328 | ret = regnode_guts(pRExC_state, op, len, "exact"); |
| 19329 | FILL_NODE(ret, op); |
| 19330 | RExC_emit += 1 + STR_SZ(len); |
| 19331 | setSTR_LEN(REGNODE_p(ret), len); |
| 19332 | if (len == 1) { |
| 19333 | *STRINGs(REGNODE_p(ret)) = (U8) value; |
| 19334 | } |
| 19335 | else { |
| 19336 | uvchr_to_utf8((U8 *) STRINGs(REGNODE_p(ret)), value); |
| 19337 | } |
| 19338 | goto not_anyof; |
| 19339 | } |
| 19340 | } |
| 19341 | |
| 19342 | if (! has_runtime_dependency) { |
| 19343 | |
| 19344 | /* See if this can be turned into an ANYOFM node. Think about the |
| 19345 | * bit patterns in two different bytes. In some positions, the |
| 19346 | * bits in each will be 1; and in other positions both will be 0; |
| 19347 | * and in some positions the bit will be 1 in one byte, and 0 in |
| 19348 | * the other. Let 'n' be the number of positions where the bits |
| 19349 | * differ. We create a mask which has exactly 'n' 0 bits, each in |
| 19350 | * a position where the two bytes differ. Now take the set of all |
| 19351 | * bytes that when ANDed with the mask yield the same result. That |
| 19352 | * set has 2**n elements, and is representable by just two 8 bit |
| 19353 | * numbers: the result and the mask. Importantly, matching the set |
| 19354 | * can be vectorized by creating a word full of the result bytes, |
| 19355 | * and a word full of the mask bytes, yielding a significant speed |
| 19356 | * up. Here, see if this node matches such a set. As a concrete |
| 19357 | * example consider [01], and the byte representing '0' which is |
| 19358 | * 0x30 on ASCII machines. It has the bits 0011 0000. Take the |
| 19359 | * mask 1111 1110. If we AND 0x31 and 0x30 with that mask we get |
| 19360 | * 0x30. Any other bytes ANDed yield something else. So [01], |
| 19361 | * which is a common usage, is optimizable into ANYOFM, and can |
| 19362 | * benefit from the speed up. We can only do this on UTF-8 |
| 19363 | * invariant bytes, because they have the same bit patterns under |
| 19364 | * UTF-8 as not. */ |
| 19365 | PERL_UINT_FAST8_T inverted = 0; |
| 19366 | #ifdef EBCDIC |
| 19367 | const PERL_UINT_FAST8_T max_permissible = 0xFF; |
| 19368 | #else |
| 19369 | const PERL_UINT_FAST8_T max_permissible = 0x7F; |
| 19370 | #endif |
| 19371 | /* If doesn't fit the criteria for ANYOFM, invert and try again. |
| 19372 | * If that works we will instead later generate an NANYOFM, and |
| 19373 | * invert back when through */ |
| 19374 | if (invlist_highest(cp_list) > max_permissible) { |
| 19375 | _invlist_invert(cp_list); |
| 19376 | inverted = 1; |
| 19377 | } |
| 19378 | |
| 19379 | if (invlist_highest(cp_list) <= max_permissible) { |
| 19380 | UV this_start, this_end; |
| 19381 | UV lowest_cp = UV_MAX; /* init'ed to suppress compiler warn */ |
| 19382 | U8 bits_differing = 0; |
| 19383 | Size_t full_cp_count = 0; |
| 19384 | bool first_time = TRUE; |
| 19385 | |
| 19386 | /* Go through the bytes and find the bit positions that differ |
| 19387 | * */ |
| 19388 | invlist_iterinit(cp_list); |
| 19389 | while (invlist_iternext(cp_list, &this_start, &this_end)) { |
| 19390 | unsigned int i = this_start; |
| 19391 | |
| 19392 | if (first_time) { |
| 19393 | if (! UVCHR_IS_INVARIANT(i)) { |
| 19394 | goto done_anyofm; |
| 19395 | } |
| 19396 | |
| 19397 | first_time = FALSE; |
| 19398 | lowest_cp = this_start; |
| 19399 | |
| 19400 | /* We have set up the code point to compare with. |
| 19401 | * Don't compare it with itself */ |
| 19402 | i++; |
| 19403 | } |
| 19404 | |
| 19405 | /* Find the bit positions that differ from the lowest code |
| 19406 | * point in the node. Keep track of all such positions by |
| 19407 | * OR'ing */ |
| 19408 | for (; i <= this_end; i++) { |
| 19409 | if (! UVCHR_IS_INVARIANT(i)) { |
| 19410 | goto done_anyofm; |
| 19411 | } |
| 19412 | |
| 19413 | bits_differing |= i ^ lowest_cp; |
| 19414 | } |
| 19415 | |
| 19416 | full_cp_count += this_end - this_start + 1; |
| 19417 | } |
| 19418 | |
| 19419 | /* At the end of the loop, we count how many bits differ from |
| 19420 | * the bits in lowest code point, call the count 'd'. If the |
| 19421 | * set we found contains 2**d elements, it is the closure of |
| 19422 | * all code points that differ only in those bit positions. To |
| 19423 | * convince yourself of that, first note that the number in the |
| 19424 | * closure must be a power of 2, which we test for. The only |
| 19425 | * way we could have that count and it be some differing set, |
| 19426 | * is if we got some code points that don't differ from the |
| 19427 | * lowest code point in any position, but do differ from each |
| 19428 | * other in some other position. That means one code point has |
| 19429 | * a 1 in that position, and another has a 0. But that would |
| 19430 | * mean that one of them differs from the lowest code point in |
| 19431 | * that position, which possibility we've already excluded. */ |
| 19432 | if ( (inverted || full_cp_count > 1) |
| 19433 | && full_cp_count == 1U << PL_bitcount[bits_differing]) |
| 19434 | { |
| 19435 | U8 ANYOFM_mask; |
| 19436 | |
| 19437 | op = ANYOFM + inverted;; |
| 19438 | |
| 19439 | /* We need to make the bits that differ be 0's */ |
| 19440 | ANYOFM_mask = ~ bits_differing; /* This goes into FLAGS */ |
| 19441 | |
| 19442 | /* The argument is the lowest code point */ |
| 19443 | ret = reganode(pRExC_state, op, lowest_cp); |
| 19444 | FLAGS(REGNODE_p(ret)) = ANYOFM_mask; |
| 19445 | } |
| 19446 | |
| 19447 | done_anyofm: |
| 19448 | invlist_iterfinish(cp_list); |
| 19449 | } |
| 19450 | |
| 19451 | if (inverted) { |
| 19452 | _invlist_invert(cp_list); |
| 19453 | } |
| 19454 | |
| 19455 | if (op != END) { |
| 19456 | goto not_anyof; |
| 19457 | } |
| 19458 | |
| 19459 | /* XXX We could create an ANYOFR_LOW node here if we saved above if |
| 19460 | * all were invariants, it wasn't inverted, and there is a single |
| 19461 | * range. This would be faster than some of the posix nodes we |
| 19462 | * create below like /\d/a, but would be twice the size. Without |
| 19463 | * having actually measured the gain, khw doesn't think the |
| 19464 | * tradeoff is really worth it */ |
| 19465 | } |
| 19466 | |
| 19467 | if (! (anyof_flags & ANYOF_LOCALE_FLAGS)) { |
| 19468 | PERL_UINT_FAST8_T type; |
| 19469 | SV * intersection = NULL; |
| 19470 | SV* d_invlist = NULL; |
| 19471 | |
| 19472 | /* See if this matches any of the POSIX classes. The POSIXA and |
| 19473 | * POSIXD ones are about the same speed as ANYOF ops, but take less |
| 19474 | * room; the ones that have above-Latin1 code point matches are |
| 19475 | * somewhat faster than ANYOF. */ |
| 19476 | |
| 19477 | for (type = POSIXA; type >= POSIXD; type--) { |
| 19478 | int posix_class; |
| 19479 | |
| 19480 | if (type == POSIXL) { /* But not /l posix classes */ |
| 19481 | continue; |
| 19482 | } |
| 19483 | |
| 19484 | for (posix_class = 0; |
| 19485 | posix_class <= _HIGHEST_REGCOMP_DOT_H_SYNC; |
| 19486 | posix_class++) |
| 19487 | { |
| 19488 | SV** our_code_points = &cp_list; |
| 19489 | SV** official_code_points; |
| 19490 | int try_inverted; |
| 19491 | |
| 19492 | if (type == POSIXA) { |
| 19493 | official_code_points = &PL_Posix_ptrs[posix_class]; |
| 19494 | } |
| 19495 | else { |
| 19496 | official_code_points = &PL_XPosix_ptrs[posix_class]; |
| 19497 | } |
| 19498 | |
| 19499 | /* Skip non-existent classes of this type. e.g. \v only |
| 19500 | * has an entry in PL_XPosix_ptrs */ |
| 19501 | if (! *official_code_points) { |
| 19502 | continue; |
| 19503 | } |
| 19504 | |
| 19505 | /* Try both the regular class, and its inversion */ |
| 19506 | for (try_inverted = 0; try_inverted < 2; try_inverted++) { |
| 19507 | bool this_inverted = invert ^ try_inverted; |
| 19508 | |
| 19509 | if (type != POSIXD) { |
| 19510 | |
| 19511 | /* This class that isn't /d can't match if we have |
| 19512 | * /d dependencies */ |
| 19513 | if (has_runtime_dependency |
| 19514 | & HAS_D_RUNTIME_DEPENDENCY) |
| 19515 | { |
| 19516 | continue; |
| 19517 | } |
| 19518 | } |
| 19519 | else /* is /d */ if (! this_inverted) { |
| 19520 | |
| 19521 | /* /d classes don't match anything non-ASCII below |
| 19522 | * 256 unconditionally (which cp_list contains) */ |
| 19523 | _invlist_intersection(cp_list, PL_UpperLatin1, |
| 19524 | &intersection); |
| 19525 | if (_invlist_len(intersection) != 0) { |
| 19526 | continue; |
| 19527 | } |
| 19528 | |
| 19529 | SvREFCNT_dec(d_invlist); |
| 19530 | d_invlist = invlist_clone(cp_list, NULL); |
| 19531 | |
| 19532 | /* But under UTF-8 it turns into using /u rules. |
| 19533 | * Add the things it matches under these conditions |
| 19534 | * so that we check below that these are identical |
| 19535 | * to what the tested class should match */ |
| 19536 | if (upper_latin1_only_utf8_matches) { |
| 19537 | _invlist_union( |
| 19538 | d_invlist, |
| 19539 | upper_latin1_only_utf8_matches, |
| 19540 | &d_invlist); |
| 19541 | } |
| 19542 | our_code_points = &d_invlist; |
| 19543 | } |
| 19544 | else { /* POSIXD, inverted. If this doesn't have this |
| 19545 | flag set, it isn't /d. */ |
| 19546 | if (! (anyof_flags & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)) |
| 19547 | { |
| 19548 | continue; |
| 19549 | } |
| 19550 | our_code_points = &cp_list; |
| 19551 | } |
| 19552 | |
| 19553 | /* Here, have weeded out some things. We want to see |
| 19554 | * if the list of characters this node contains |
| 19555 | * ('*our_code_points') precisely matches those of the |
| 19556 | * class we are currently checking against |
| 19557 | * ('*official_code_points'). */ |
| 19558 | if (_invlistEQ(*our_code_points, |
| 19559 | *official_code_points, |
| 19560 | try_inverted)) |
| 19561 | { |
| 19562 | /* Here, they precisely match. Optimize this ANYOF |
| 19563 | * node into its equivalent POSIX one of the |
| 19564 | * correct type, possibly inverted */ |
| 19565 | ret = reg_node(pRExC_state, (try_inverted) |
| 19566 | ? type + NPOSIXA |
| 19567 | - POSIXA |
| 19568 | : type); |
| 19569 | FLAGS(REGNODE_p(ret)) = posix_class; |
| 19570 | SvREFCNT_dec(d_invlist); |
| 19571 | SvREFCNT_dec(intersection); |
| 19572 | goto not_anyof; |
| 19573 | } |
| 19574 | } |
| 19575 | } |
| 19576 | } |
| 19577 | SvREFCNT_dec(d_invlist); |
| 19578 | SvREFCNT_dec(intersection); |
| 19579 | } |
| 19580 | |
| 19581 | /* If it is a single contiguous range, ANYOFR is an efficient regnode, |
| 19582 | * both in size and speed. Currently, a 20 bit range base (smallest |
| 19583 | * code point in the range), and a 12 bit maximum delta are packed into |
| 19584 | * a 32 bit word. This allows for using it on all of the Unicode code |
| 19585 | * points except for the highest plane, which is only for private use |
| 19586 | * code points. khw doubts that a bigger delta is likely in real world |
| 19587 | * applications */ |
| 19588 | if ( single_range |
| 19589 | && ! has_runtime_dependency |
| 19590 | && anyof_flags == 0 |
| 19591 | && start[0] < (1 << ANYOFR_BASE_BITS) |
| 19592 | && end[0] - start[0] |
| 19593 | < ((1U << (sizeof(((struct regnode_1 *)NULL)->arg1) |
| 19594 | * CHARBITS - ANYOFR_BASE_BITS)))) |
| 19595 | |
| 19596 | { |
| 19597 | U8 low_utf8[UTF8_MAXBYTES+1]; |
| 19598 | U8 high_utf8[UTF8_MAXBYTES+1]; |
| 19599 | |
| 19600 | ret = reganode(pRExC_state, ANYOFR, |
| 19601 | (start[0] | (end[0] - start[0]) << ANYOFR_BASE_BITS)); |
| 19602 | |
| 19603 | /* Place the lowest UTF-8 start byte in the flags field, so as to |
| 19604 | * allow efficient ruling out at run time of many possible inputs. |
| 19605 | * */ |
| 19606 | (void) uvchr_to_utf8(low_utf8, start[0]); |
| 19607 | (void) uvchr_to_utf8(high_utf8, end[0]); |
| 19608 | |
| 19609 | /* If all code points share the same first byte, this can be an |
| 19610 | * ANYOFRb. Otherwise store the lowest UTF-8 start byte which can |
| 19611 | * quickly rule out many inputs at run-time without having to |
| 19612 | * compute the code point from UTF-8. For EBCDIC, we use I8, as |
| 19613 | * not doing that transformation would not rule out nearly so many |
| 19614 | * things */ |
| 19615 | if (low_utf8[0] == high_utf8[0]) { |
| 19616 | OP(REGNODE_p(ret)) = ANYOFRb; |
| 19617 | ANYOF_FLAGS(REGNODE_p(ret)) = low_utf8[0]; |
| 19618 | } |
| 19619 | else { |
| 19620 | ANYOF_FLAGS(REGNODE_p(ret)) |
| 19621 | = NATIVE_UTF8_TO_I8(low_utf8[0]); |
| 19622 | } |
| 19623 | |
| 19624 | goto not_anyof; |
| 19625 | } |
| 19626 | |
| 19627 | /* If didn't find an optimization and there is no need for a bitmap, |
| 19628 | * optimize to indicate that */ |
| 19629 | if ( start[0] >= NUM_ANYOF_CODE_POINTS |
| 19630 | && ! LOC |
| 19631 | && ! upper_latin1_only_utf8_matches |
| 19632 | && anyof_flags == 0) |
| 19633 | { |
| 19634 | U8 low_utf8[UTF8_MAXBYTES+1]; |
| 19635 | UV highest_cp = invlist_highest(cp_list); |
| 19636 | |
| 19637 | /* Currently the maximum allowed code point by the system is |
| 19638 | * IV_MAX. Higher ones are reserved for future internal use. This |
| 19639 | * particular regnode can be used for higher ones, but we can't |
| 19640 | * calculate the code point of those. IV_MAX suffices though, as |
| 19641 | * it will be a large first byte */ |
| 19642 | Size_t low_len = uvchr_to_utf8(low_utf8, MIN(start[0], IV_MAX)) |
| 19643 | - low_utf8; |
| 19644 | |
| 19645 | /* We store the lowest possible first byte of the UTF-8 |
| 19646 | * representation, using the flags field. This allows for quick |
| 19647 | * ruling out of some inputs without having to convert from UTF-8 |
| 19648 | * to code point. For EBCDIC, we use I8, as not doing that |
| 19649 | * transformation would not rule out nearly so many things */ |
| 19650 | anyof_flags = NATIVE_UTF8_TO_I8(low_utf8[0]); |
| 19651 | |
| 19652 | op = ANYOFH; |
| 19653 | |
| 19654 | /* If the first UTF-8 start byte for the highest code point in the |
| 19655 | * range is suitably small, we may be able to get an upper bound as |
| 19656 | * well */ |
| 19657 | if (highest_cp <= IV_MAX) { |
| 19658 | U8 high_utf8[UTF8_MAXBYTES+1]; |
| 19659 | Size_t high_len = uvchr_to_utf8(high_utf8, highest_cp) |
| 19660 | - high_utf8; |
| 19661 | |
| 19662 | /* If the lowest and highest are the same, we can get an exact |
| 19663 | * first byte instead of a just minimum or even a sequence of |
| 19664 | * exact leading bytes. We signal these with different |
| 19665 | * regnodes */ |
| 19666 | if (low_utf8[0] == high_utf8[0]) { |
| 19667 | Size_t len = find_first_differing_byte_pos(low_utf8, |
| 19668 | high_utf8, |
| 19669 | MIN(low_len, high_len)); |
| 19670 | |
| 19671 | if (len == 1) { |
| 19672 | |
| 19673 | /* No need to convert to I8 for EBCDIC as this is an |
| 19674 | * exact match */ |
| 19675 | anyof_flags = low_utf8[0]; |
| 19676 | op = ANYOFHb; |
| 19677 | } |
| 19678 | else { |
| 19679 | op = ANYOFHs; |
| 19680 | ret = regnode_guts(pRExC_state, op, |
| 19681 | regarglen[op] + STR_SZ(len), |
| 19682 | "anyofhs"); |
| 19683 | FILL_NODE(ret, op); |
| 19684 | ((struct regnode_anyofhs *) REGNODE_p(ret))->str_len |
| 19685 | = len; |
| 19686 | Copy(low_utf8, /* Add the common bytes */ |
| 19687 | ((struct regnode_anyofhs *) REGNODE_p(ret))->string, |
| 19688 | len, U8); |
| 19689 | RExC_emit += NODE_SZ_STR(REGNODE_p(ret)); |
| 19690 | set_ANYOF_arg(pRExC_state, REGNODE_p(ret), cp_list, |
| 19691 | NULL, only_utf8_locale_list); |
| 19692 | goto not_anyof; |
| 19693 | } |
| 19694 | } |
| 19695 | else if (NATIVE_UTF8_TO_I8(high_utf8[0]) <= MAX_ANYOF_HRx_BYTE) |
| 19696 | { |
| 19697 | |
| 19698 | /* Here, the high byte is not the same as the low, but is |
| 19699 | * small enough that its reasonable to have a loose upper |
| 19700 | * bound, which is packed in with the strict lower bound. |
| 19701 | * See comments at the definition of MAX_ANYOF_HRx_BYTE. |
| 19702 | * On EBCDIC platforms, I8 is used. On ASCII platforms I8 |
| 19703 | * is the same thing as UTF-8 */ |
| 19704 | |
| 19705 | U8 bits = 0; |
| 19706 | U8 max_range_diff = MAX_ANYOF_HRx_BYTE - anyof_flags; |
| 19707 | U8 range_diff = NATIVE_UTF8_TO_I8(high_utf8[0]) |
| 19708 | - anyof_flags; |
| 19709 | |
| 19710 | if (range_diff <= max_range_diff / 8) { |
| 19711 | bits = 3; |
| 19712 | } |
| 19713 | else if (range_diff <= max_range_diff / 4) { |
| 19714 | bits = 2; |
| 19715 | } |
| 19716 | else if (range_diff <= max_range_diff / 2) { |
| 19717 | bits = 1; |
| 19718 | } |
| 19719 | anyof_flags = (anyof_flags - 0xC0) << 2 | bits; |
| 19720 | op = ANYOFHr; |
| 19721 | } |
| 19722 | } |
| 19723 | |
| 19724 | goto done_finding_op; |
| 19725 | } |
| 19726 | } /* End of seeing if can optimize it into a different node */ |
| 19727 | |
| 19728 | is_anyof: /* It's going to be an ANYOF node. */ |
| 19729 | op = (has_runtime_dependency & HAS_D_RUNTIME_DEPENDENCY) |
| 19730 | ? ANYOFD |
| 19731 | : ((posixl) |
| 19732 | ? ANYOFPOSIXL |
| 19733 | : ((LOC) |
| 19734 | ? ANYOFL |
| 19735 | : ANYOF)); |
| 19736 | |
| 19737 | done_finding_op: |
| 19738 | |
| 19739 | ret = regnode_guts(pRExC_state, op, regarglen[op], "anyof"); |
| 19740 | FILL_NODE(ret, op); /* We set the argument later */ |
| 19741 | RExC_emit += 1 + regarglen[op]; |
| 19742 | ANYOF_FLAGS(REGNODE_p(ret)) = anyof_flags; |
| 19743 | |
| 19744 | /* Here, <cp_list> contains all the code points we can determine at |
| 19745 | * compile time that match under all conditions. Go through it, and |
| 19746 | * for things that belong in the bitmap, put them there, and delete from |
| 19747 | * <cp_list>. While we are at it, see if everything above 255 is in the |
| 19748 | * list, and if so, set a flag to speed up execution */ |
| 19749 | |
| 19750 | populate_ANYOF_from_invlist(REGNODE_p(ret), &cp_list); |
| 19751 | |
| 19752 | if (posixl) { |
| 19753 | ANYOF_POSIXL_SET_TO_BITMAP(REGNODE_p(ret), posixl); |
| 19754 | } |
| 19755 | |
| 19756 | if (invert) { |
| 19757 | ANYOF_FLAGS(REGNODE_p(ret)) |= ANYOF_INVERT; |
| 19758 | } |
| 19759 | |
| 19760 | /* Here, the bitmap has been populated with all the Latin1 code points that |
| 19761 | * always match. Can now add to the overall list those that match only |
| 19762 | * when the target string is UTF-8 (<upper_latin1_only_utf8_matches>). |
| 19763 | * */ |
| 19764 | if (upper_latin1_only_utf8_matches) { |
| 19765 | if (cp_list) { |
| 19766 | _invlist_union(cp_list, |
| 19767 | upper_latin1_only_utf8_matches, |
| 19768 | &cp_list); |
| 19769 | SvREFCNT_dec_NN(upper_latin1_only_utf8_matches); |
| 19770 | } |
| 19771 | else { |
| 19772 | cp_list = upper_latin1_only_utf8_matches; |
| 19773 | } |
| 19774 | ANYOF_FLAGS(REGNODE_p(ret)) |= ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP; |
| 19775 | } |
| 19776 | |
| 19777 | set_ANYOF_arg(pRExC_state, REGNODE_p(ret), cp_list, |
| 19778 | (HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION) |
| 19779 | ? listsv |
| 19780 | : NULL, |
| 19781 | only_utf8_locale_list); |
| 19782 | SvREFCNT_dec(cp_list);; |
| 19783 | SvREFCNT_dec(only_utf8_locale_list); |
| 19784 | return ret; |
| 19785 | |
| 19786 | not_anyof: |
| 19787 | |
| 19788 | /* Here, the node is getting optimized into something that's not an ANYOF |
| 19789 | * one. Finish up. */ |
| 19790 | |
| 19791 | Set_Node_Offset_Length(REGNODE_p(ret), orig_parse - RExC_start, |
| 19792 | RExC_parse - orig_parse);; |
| 19793 | SvREFCNT_dec(cp_list);; |
| 19794 | SvREFCNT_dec(only_utf8_locale_list); |
| 19795 | return ret; |
| 19796 | } |
| 19797 | |
| 19798 | #undef HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION |
| 19799 | |
| 19800 | STATIC void |
| 19801 | S_set_ANYOF_arg(pTHX_ RExC_state_t* const pRExC_state, |
| 19802 | regnode* const node, |
| 19803 | SV* const cp_list, |
| 19804 | SV* const runtime_defns, |
| 19805 | SV* const only_utf8_locale_list) |
| 19806 | { |
| 19807 | /* Sets the arg field of an ANYOF-type node 'node', using information about |
| 19808 | * the node passed-in. If there is nothing outside the node's bitmap, the |
| 19809 | * arg is set to ANYOF_ONLY_HAS_BITMAP. Otherwise, it sets the argument to |
| 19810 | * the count returned by add_data(), having allocated and stored an array, |
| 19811 | * av, as follows: |
| 19812 | * |
| 19813 | * av[0] stores the inversion list defining this class as far as known at |
| 19814 | * this time, or PL_sv_undef if nothing definite is now known. |
| 19815 | * av[1] stores the inversion list of code points that match only if the |
| 19816 | * current locale is UTF-8, or if none, PL_sv_undef if there is an |
| 19817 | * av[2], or no entry otherwise. |
| 19818 | * av[2] stores the list of user-defined properties whose subroutine |
| 19819 | * definitions aren't known at this time, or no entry if none. */ |
| 19820 | |
| 19821 | UV n; |
| 19822 | |
| 19823 | PERL_ARGS_ASSERT_SET_ANYOF_ARG; |
| 19824 | |
| 19825 | if (! cp_list && ! runtime_defns && ! only_utf8_locale_list) { |
| 19826 | assert(! (ANYOF_FLAGS(node) |
| 19827 | & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)); |
| 19828 | ARG_SET(node, ANYOF_ONLY_HAS_BITMAP); |
| 19829 | } |
| 19830 | else { |
| 19831 | AV * const av = newAV(); |
| 19832 | SV *rv; |
| 19833 | |
| 19834 | if (cp_list) { |
| 19835 | av_store(av, INVLIST_INDEX, SvREFCNT_inc_NN(cp_list)); |
| 19836 | } |
| 19837 | |
| 19838 | if (only_utf8_locale_list) { |
| 19839 | av_store(av, ONLY_LOCALE_MATCHES_INDEX, |
| 19840 | SvREFCNT_inc_NN(only_utf8_locale_list)); |
| 19841 | } |
| 19842 | |
| 19843 | if (runtime_defns) { |
| 19844 | av_store(av, DEFERRED_USER_DEFINED_INDEX, |
| 19845 | SvREFCNT_inc_NN(runtime_defns)); |
| 19846 | } |
| 19847 | |
| 19848 | rv = newRV_noinc(MUTABLE_SV(av)); |
| 19849 | n = add_data(pRExC_state, STR_WITH_LEN("s")); |
| 19850 | RExC_rxi->data->data[n] = (void*)rv; |
| 19851 | ARG_SET(node, n); |
| 19852 | } |
| 19853 | } |
| 19854 | |
| 19855 | #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) |
| 19856 | SV * |
| 19857 | Perl__get_regclass_nonbitmap_data(pTHX_ const regexp *prog, |
| 19858 | const regnode* node, |
| 19859 | bool doinit, |
| 19860 | SV** listsvp, |
| 19861 | SV** only_utf8_locale_ptr, |
| 19862 | SV** output_invlist) |
| 19863 | |
| 19864 | { |
| 19865 | /* For internal core use only. |
| 19866 | * Returns the inversion list for the input 'node' in the regex 'prog'. |
| 19867 | * If <doinit> is 'true', will attempt to create the inversion list if not |
| 19868 | * already done. |
| 19869 | * If <listsvp> is non-null, will return the printable contents of the |
| 19870 | * property definition. This can be used to get debugging information |
| 19871 | * even before the inversion list exists, by calling this function with |
| 19872 | * 'doinit' set to false, in which case the components that will be used |
| 19873 | * to eventually create the inversion list are returned (in a printable |
| 19874 | * form). |
| 19875 | * If <only_utf8_locale_ptr> is not NULL, it is where this routine is to |
| 19876 | * store an inversion list of code points that should match only if the |
| 19877 | * execution-time locale is a UTF-8 one. |
| 19878 | * If <output_invlist> is not NULL, it is where this routine is to store an |
| 19879 | * inversion list of the code points that would be instead returned in |
| 19880 | * <listsvp> if this were NULL. Thus, what gets output in <listsvp> |
| 19881 | * when this parameter is used, is just the non-code point data that |
| 19882 | * will go into creating the inversion list. This currently should be just |
| 19883 | * user-defined properties whose definitions were not known at compile |
| 19884 | * time. Using this parameter allows for easier manipulation of the |
| 19885 | * inversion list's data by the caller. It is illegal to call this |
| 19886 | * function with this parameter set, but not <listsvp> |
| 19887 | * |
| 19888 | * Tied intimately to how S_set_ANYOF_arg sets up the data structure. Note |
| 19889 | * that, in spite of this function's name, the inversion list it returns |
| 19890 | * may include the bitmap data as well */ |
| 19891 | |
| 19892 | SV *si = NULL; /* Input initialization string */ |
| 19893 | SV* invlist = NULL; |
| 19894 | |
| 19895 | RXi_GET_DECL(prog, progi); |
| 19896 | const struct reg_data * const data = prog ? progi->data : NULL; |
| 19897 | |
| 19898 | PERL_ARGS_ASSERT__GET_REGCLASS_NONBITMAP_DATA; |
| 19899 | assert(! output_invlist || listsvp); |
| 19900 | |
| 19901 | if (data && data->count) { |
| 19902 | const U32 n = ARG(node); |
| 19903 | |
| 19904 | if (data->what[n] == 's') { |
| 19905 | SV * const rv = MUTABLE_SV(data->data[n]); |
| 19906 | AV * const av = MUTABLE_AV(SvRV(rv)); |
| 19907 | SV **const ary = AvARRAY(av); |
| 19908 | |
| 19909 | invlist = ary[INVLIST_INDEX]; |
| 19910 | |
| 19911 | if (av_tindex_skip_len_mg(av) >= ONLY_LOCALE_MATCHES_INDEX) { |
| 19912 | *only_utf8_locale_ptr = ary[ONLY_LOCALE_MATCHES_INDEX]; |
| 19913 | } |
| 19914 | |
| 19915 | if (av_tindex_skip_len_mg(av) >= DEFERRED_USER_DEFINED_INDEX) { |
| 19916 | si = ary[DEFERRED_USER_DEFINED_INDEX]; |
| 19917 | } |
| 19918 | |
| 19919 | if (doinit && (si || invlist)) { |
| 19920 | if (si) { |
| 19921 | bool user_defined; |
| 19922 | SV * msg = newSVpvs_flags("", SVs_TEMP); |
| 19923 | |
| 19924 | SV * prop_definition = handle_user_defined_property( |
| 19925 | "", 0, FALSE, /* There is no \p{}, \P{} */ |
| 19926 | SvPVX_const(si)[1] - '0', /* /i or not has been |
| 19927 | stored here for just |
| 19928 | this occasion */ |
| 19929 | TRUE, /* run time */ |
| 19930 | FALSE, /* This call must find the defn */ |
| 19931 | si, /* The property definition */ |
| 19932 | &user_defined, |
| 19933 | msg, |
| 19934 | 0 /* base level call */ |
| 19935 | ); |
| 19936 | |
| 19937 | if (SvCUR(msg)) { |
| 19938 | assert(prop_definition == NULL); |
| 19939 | |
| 19940 | Perl_croak(aTHX_ "%" UTF8f, |
| 19941 | UTF8fARG(SvUTF8(msg), SvCUR(msg), SvPVX(msg))); |
| 19942 | } |
| 19943 | |
| 19944 | if (invlist) { |
| 19945 | _invlist_union(invlist, prop_definition, &invlist); |
| 19946 | SvREFCNT_dec_NN(prop_definition); |
| 19947 | } |
| 19948 | else { |
| 19949 | invlist = prop_definition; |
| 19950 | } |
| 19951 | |
| 19952 | STATIC_ASSERT_STMT(ONLY_LOCALE_MATCHES_INDEX == 1 + INVLIST_INDEX); |
| 19953 | STATIC_ASSERT_STMT(DEFERRED_USER_DEFINED_INDEX == 1 + ONLY_LOCALE_MATCHES_INDEX); |
| 19954 | |
| 19955 | ary[INVLIST_INDEX] = invlist; |
| 19956 | av_fill(av, (ary[ONLY_LOCALE_MATCHES_INDEX]) |
| 19957 | ? ONLY_LOCALE_MATCHES_INDEX |
| 19958 | : INVLIST_INDEX); |
| 19959 | si = NULL; |
| 19960 | } |
| 19961 | } |
| 19962 | } |
| 19963 | } |
| 19964 | |
| 19965 | /* If requested, return a printable version of what this ANYOF node matches |
| 19966 | * */ |
| 19967 | if (listsvp) { |
| 19968 | SV* matches_string = NULL; |
| 19969 | |
| 19970 | /* This function can be called at compile-time, before everything gets |
| 19971 | * resolved, in which case we return the currently best available |
| 19972 | * information, which is the string that will eventually be used to do |
| 19973 | * that resolving, 'si' */ |
| 19974 | if (si) { |
| 19975 | /* Here, we only have 'si' (and possibly some passed-in data in |
| 19976 | * 'invlist', which is handled below) If the caller only wants |
| 19977 | * 'si', use that. */ |
| 19978 | if (! output_invlist) { |
| 19979 | matches_string = newSVsv(si); |
| 19980 | } |
| 19981 | else { |
| 19982 | /* But if the caller wants an inversion list of the node, we |
| 19983 | * need to parse 'si' and place as much as possible in the |
| 19984 | * desired output inversion list, making 'matches_string' only |
| 19985 | * contain the currently unresolvable things */ |
| 19986 | const char *si_string = SvPVX(si); |
| 19987 | STRLEN remaining = SvCUR(si); |
| 19988 | UV prev_cp = 0; |
| 19989 | U8 count = 0; |
| 19990 | |
| 19991 | /* Ignore everything before and including the first new-line */ |
| 19992 | si_string = (const char *) memchr(si_string, '\n', SvCUR(si)); |
| 19993 | assert (si_string != NULL); |
| 19994 | si_string++; |
| 19995 | remaining = SvPVX(si) + SvCUR(si) - si_string; |
| 19996 | |
| 19997 | while (remaining > 0) { |
| 19998 | |
| 19999 | /* The data consists of just strings defining user-defined |
| 20000 | * property names, but in prior incarnations, and perhaps |
| 20001 | * somehow from pluggable regex engines, it could still |
| 20002 | * hold hex code point definitions, all of which should be |
| 20003 | * legal (or it wouldn't have gotten this far). Each |
| 20004 | * component of a range would be separated by a tab, and |
| 20005 | * each range by a new-line. If these are found, instead |
| 20006 | * add them to the inversion list */ |
| 20007 | I32 grok_flags = PERL_SCAN_SILENT_ILLDIGIT |
| 20008 | |PERL_SCAN_SILENT_NON_PORTABLE; |
| 20009 | STRLEN len = remaining; |
| 20010 | UV cp = grok_hex(si_string, &len, &grok_flags, NULL); |
| 20011 | |
| 20012 | /* If the hex decode routine found something, it should go |
| 20013 | * up to the next \n */ |
| 20014 | if ( *(si_string + len) == '\n') { |
| 20015 | if (count) { /* 2nd code point on line */ |
| 20016 | *output_invlist = _add_range_to_invlist(*output_invlist, prev_cp, cp); |
| 20017 | } |
| 20018 | else { |
| 20019 | *output_invlist = add_cp_to_invlist(*output_invlist, cp); |
| 20020 | } |
| 20021 | count = 0; |
| 20022 | goto prepare_for_next_iteration; |
| 20023 | } |
| 20024 | |
| 20025 | /* If the hex decode was instead for the lower range limit, |
| 20026 | * save it, and go parse the upper range limit */ |
| 20027 | if (*(si_string + len) == '\t') { |
| 20028 | assert(count == 0); |
| 20029 | |
| 20030 | prev_cp = cp; |
| 20031 | count = 1; |
| 20032 | prepare_for_next_iteration: |
| 20033 | si_string += len + 1; |
| 20034 | remaining -= len + 1; |
| 20035 | continue; |
| 20036 | } |
| 20037 | |
| 20038 | /* Here, didn't find a legal hex number. Just add the text |
| 20039 | * from here up to the next \n, omitting any trailing |
| 20040 | * markers. */ |
| 20041 | |
| 20042 | remaining -= len; |
| 20043 | len = strcspn(si_string, |
| 20044 | DEFERRED_COULD_BE_OFFICIAL_MARKERs "\n"); |
| 20045 | remaining -= len; |
| 20046 | if (matches_string) { |
| 20047 | sv_catpvn(matches_string, si_string, len); |
| 20048 | } |
| 20049 | else { |
| 20050 | matches_string = newSVpvn(si_string, len); |
| 20051 | } |
| 20052 | sv_catpvs(matches_string, " "); |
| 20053 | |
| 20054 | si_string += len; |
| 20055 | if ( remaining |
| 20056 | && UCHARAT(si_string) |
| 20057 | == DEFERRED_COULD_BE_OFFICIAL_MARKERc) |
| 20058 | { |
| 20059 | si_string++; |
| 20060 | remaining--; |
| 20061 | } |
| 20062 | if (remaining && UCHARAT(si_string) == '\n') { |
| 20063 | si_string++; |
| 20064 | remaining--; |
| 20065 | } |
| 20066 | } /* end of loop through the text */ |
| 20067 | |
| 20068 | assert(matches_string); |
| 20069 | if (SvCUR(matches_string)) { /* Get rid of trailing blank */ |
| 20070 | SvCUR_set(matches_string, SvCUR(matches_string) - 1); |
| 20071 | } |
| 20072 | } /* end of has an 'si' */ |
| 20073 | } |
| 20074 | |
| 20075 | /* Add the stuff that's already known */ |
| 20076 | if (invlist) { |
| 20077 | |
| 20078 | /* Again, if the caller doesn't want the output inversion list, put |
| 20079 | * everything in 'matches-string' */ |
| 20080 | if (! output_invlist) { |
| 20081 | if ( ! matches_string) { |
| 20082 | matches_string = newSVpvs("\n"); |
| 20083 | } |
| 20084 | sv_catsv(matches_string, invlist_contents(invlist, |
| 20085 | TRUE /* traditional style */ |
| 20086 | )); |
| 20087 | } |
| 20088 | else if (! *output_invlist) { |
| 20089 | *output_invlist = invlist_clone(invlist, NULL); |
| 20090 | } |
| 20091 | else { |
| 20092 | _invlist_union(*output_invlist, invlist, output_invlist); |
| 20093 | } |
| 20094 | } |
| 20095 | |
| 20096 | *listsvp = matches_string; |
| 20097 | } |
| 20098 | |
| 20099 | return invlist; |
| 20100 | } |
| 20101 | #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */ |
| 20102 | |
| 20103 | /* reg_skipcomment() |
| 20104 | |
| 20105 | Absorbs an /x style # comment from the input stream, |
| 20106 | returning a pointer to the first character beyond the comment, or if the |
| 20107 | comment terminates the pattern without anything following it, this returns |
| 20108 | one past the final character of the pattern (in other words, RExC_end) and |
| 20109 | sets the REG_RUN_ON_COMMENT_SEEN flag. |
| 20110 | |
| 20111 | Note it's the callers responsibility to ensure that we are |
| 20112 | actually in /x mode |
| 20113 | |
| 20114 | */ |
| 20115 | |
| 20116 | PERL_STATIC_INLINE char* |
| 20117 | S_reg_skipcomment(RExC_state_t *pRExC_state, char* p) |
| 20118 | { |
| 20119 | PERL_ARGS_ASSERT_REG_SKIPCOMMENT; |
| 20120 | |
| 20121 | assert(*p == '#'); |
| 20122 | |
| 20123 | while (p < RExC_end) { |
| 20124 | if (*(++p) == '\n') { |
| 20125 | return p+1; |
| 20126 | } |
| 20127 | } |
| 20128 | |
| 20129 | /* we ran off the end of the pattern without ending the comment, so we have |
| 20130 | * to add an \n when wrapping */ |
| 20131 | RExC_seen |= REG_RUN_ON_COMMENT_SEEN; |
| 20132 | return p; |
| 20133 | } |
| 20134 | |
| 20135 | STATIC void |
| 20136 | S_skip_to_be_ignored_text(pTHX_ RExC_state_t *pRExC_state, |
| 20137 | char ** p, |
| 20138 | const bool force_to_xmod |
| 20139 | ) |
| 20140 | { |
| 20141 | /* If the text at the current parse position '*p' is a '(?#...)' comment, |
| 20142 | * or if we are under /x or 'force_to_xmod' is TRUE, and the text at '*p' |
| 20143 | * is /x whitespace, advance '*p' so that on exit it points to the first |
| 20144 | * byte past all such white space and comments */ |
| 20145 | |
| 20146 | const bool use_xmod = force_to_xmod || (RExC_flags & RXf_PMf_EXTENDED); |
| 20147 | |
| 20148 | PERL_ARGS_ASSERT_SKIP_TO_BE_IGNORED_TEXT; |
| 20149 | |
| 20150 | assert( ! UTF || UTF8_IS_INVARIANT(**p) || UTF8_IS_START(**p)); |
| 20151 | |
| 20152 | for (;;) { |
| 20153 | if (RExC_end - (*p) >= 3 |
| 20154 | && *(*p) == '(' |
| 20155 | && *(*p + 1) == '?' |
| 20156 | && *(*p + 2) == '#') |
| 20157 | { |
| 20158 | while (*(*p) != ')') { |
| 20159 | if ((*p) == RExC_end) |
| 20160 | FAIL("Sequence (?#... not terminated"); |
| 20161 | (*p)++; |
| 20162 | } |
| 20163 | (*p)++; |
| 20164 | continue; |
| 20165 | } |
| 20166 | |
| 20167 | if (use_xmod) { |
| 20168 | const char * save_p = *p; |
| 20169 | while ((*p) < RExC_end) { |
| 20170 | STRLEN len; |
| 20171 | if ((len = is_PATWS_safe((*p), RExC_end, UTF))) { |
| 20172 | (*p) += len; |
| 20173 | } |
| 20174 | else if (*(*p) == '#') { |
| 20175 | (*p) = reg_skipcomment(pRExC_state, (*p)); |
| 20176 | } |
| 20177 | else { |
| 20178 | break; |
| 20179 | } |
| 20180 | } |
| 20181 | if (*p != save_p) { |
| 20182 | continue; |
| 20183 | } |
| 20184 | } |
| 20185 | |
| 20186 | break; |
| 20187 | } |
| 20188 | |
| 20189 | return; |
| 20190 | } |
| 20191 | |
| 20192 | /* nextchar() |
| 20193 | |
| 20194 | Advances the parse position by one byte, unless that byte is the beginning |
| 20195 | of a '(?#...)' style comment, or is /x whitespace and /x is in effect. In |
| 20196 | those two cases, the parse position is advanced beyond all such comments and |
| 20197 | white space. |
| 20198 | |
| 20199 | This is the UTF, (?#...), and /x friendly way of saying RExC_parse++. |
| 20200 | */ |
| 20201 | |
| 20202 | STATIC void |
| 20203 | S_nextchar(pTHX_ RExC_state_t *pRExC_state) |
| 20204 | { |
| 20205 | PERL_ARGS_ASSERT_NEXTCHAR; |
| 20206 | |
| 20207 | if (RExC_parse < RExC_end) { |
| 20208 | assert( ! UTF |
| 20209 | || UTF8_IS_INVARIANT(*RExC_parse) |
| 20210 | || UTF8_IS_START(*RExC_parse)); |
| 20211 | |
| 20212 | RExC_parse += (UTF) |
| 20213 | ? UTF8_SAFE_SKIP(RExC_parse, RExC_end) |
| 20214 | : 1; |
| 20215 | |
| 20216 | skip_to_be_ignored_text(pRExC_state, &RExC_parse, |
| 20217 | FALSE /* Don't force /x */ ); |
| 20218 | } |
| 20219 | } |
| 20220 | |
| 20221 | STATIC void |
| 20222 | S_change_engine_size(pTHX_ RExC_state_t *pRExC_state, const Ptrdiff_t size) |
| 20223 | { |
| 20224 | /* 'size' is the delta number of smallest regnode equivalents to add or |
| 20225 | * subtract from the current memory allocated to the regex engine being |
| 20226 | * constructed. */ |
| 20227 | |
| 20228 | PERL_ARGS_ASSERT_CHANGE_ENGINE_SIZE; |
| 20229 | |
| 20230 | RExC_size += size; |
| 20231 | |
| 20232 | Renewc(RExC_rxi, |
| 20233 | sizeof(regexp_internal) + (RExC_size + 1) * sizeof(regnode), |
| 20234 | /* +1 for REG_MAGIC */ |
| 20235 | char, |
| 20236 | regexp_internal); |
| 20237 | if ( RExC_rxi == NULL ) |
| 20238 | FAIL("Regexp out of space"); |
| 20239 | RXi_SET(RExC_rx, RExC_rxi); |
| 20240 | |
| 20241 | RExC_emit_start = RExC_rxi->program; |
| 20242 | if (size > 0) { |
| 20243 | Zero(REGNODE_p(RExC_emit), size, regnode); |
| 20244 | } |
| 20245 | |
| 20246 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 20247 | Renew(RExC_offsets, 2*RExC_size+1, U32); |
| 20248 | if (size > 0) { |
| 20249 | Zero(RExC_offsets + 2*(RExC_size - size) + 1, 2 * size, U32); |
| 20250 | } |
| 20251 | RExC_offsets[0] = RExC_size; |
| 20252 | #endif |
| 20253 | } |
| 20254 | |
| 20255 | STATIC regnode_offset |
| 20256 | S_regnode_guts(pTHX_ RExC_state_t *pRExC_state, const U8 op, const STRLEN extra_size, const char* const name) |
| 20257 | { |
| 20258 | /* Allocate a regnode for 'op', with 'extra_size' extra (smallest) regnode |
| 20259 | * equivalents space. It aligns and increments RExC_size |
| 20260 | * |
| 20261 | * It returns the regnode's offset into the regex engine program */ |
| 20262 | |
| 20263 | const regnode_offset ret = RExC_emit; |
| 20264 | |
| 20265 | GET_RE_DEBUG_FLAGS_DECL; |
| 20266 | |
| 20267 | PERL_ARGS_ASSERT_REGNODE_GUTS; |
| 20268 | |
| 20269 | SIZE_ALIGN(RExC_size); |
| 20270 | change_engine_size(pRExC_state, (Ptrdiff_t) 1 + extra_size); |
| 20271 | NODE_ALIGN_FILL(REGNODE_p(ret)); |
| 20272 | #ifndef RE_TRACK_PATTERN_OFFSETS |
| 20273 | PERL_UNUSED_ARG(name); |
| 20274 | PERL_UNUSED_ARG(op); |
| 20275 | #else |
| 20276 | assert(extra_size >= regarglen[op] || PL_regkind[op] == ANYOF); |
| 20277 | |
| 20278 | if (RExC_offsets) { /* MJD */ |
| 20279 | MJD_OFFSET_DEBUG( |
| 20280 | ("%s:%d: (op %s) %s %" UVuf " (len %" UVuf ") (max %" UVuf ").\n", |
| 20281 | name, __LINE__, |
| 20282 | PL_reg_name[op], |
| 20283 | (UV)(RExC_emit) > RExC_offsets[0] |
| 20284 | ? "Overwriting end of array!\n" : "OK", |
| 20285 | (UV)(RExC_emit), |
| 20286 | (UV)(RExC_parse - RExC_start), |
| 20287 | (UV)RExC_offsets[0])); |
| 20288 | Set_Node_Offset(REGNODE_p(RExC_emit), RExC_parse + (op == END)); |
| 20289 | } |
| 20290 | #endif |
| 20291 | return(ret); |
| 20292 | } |
| 20293 | |
| 20294 | /* |
| 20295 | - reg_node - emit a node |
| 20296 | */ |
| 20297 | STATIC regnode_offset /* Location. */ |
| 20298 | S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op) |
| 20299 | { |
| 20300 | const regnode_offset ret = regnode_guts(pRExC_state, op, regarglen[op], "reg_node"); |
| 20301 | regnode_offset ptr = ret; |
| 20302 | |
| 20303 | PERL_ARGS_ASSERT_REG_NODE; |
| 20304 | |
| 20305 | assert(regarglen[op] == 0); |
| 20306 | |
| 20307 | FILL_ADVANCE_NODE(ptr, op); |
| 20308 | RExC_emit = ptr; |
| 20309 | return(ret); |
| 20310 | } |
| 20311 | |
| 20312 | /* |
| 20313 | - reganode - emit a node with an argument |
| 20314 | */ |
| 20315 | STATIC regnode_offset /* Location. */ |
| 20316 | S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg) |
| 20317 | { |
| 20318 | const regnode_offset ret = regnode_guts(pRExC_state, op, regarglen[op], "reganode"); |
| 20319 | regnode_offset ptr = ret; |
| 20320 | |
| 20321 | PERL_ARGS_ASSERT_REGANODE; |
| 20322 | |
| 20323 | /* ANYOF are special cased to allow non-length 1 args */ |
| 20324 | assert(regarglen[op] == 1); |
| 20325 | |
| 20326 | FILL_ADVANCE_NODE_ARG(ptr, op, arg); |
| 20327 | RExC_emit = ptr; |
| 20328 | return(ret); |
| 20329 | } |
| 20330 | |
| 20331 | /* |
| 20332 | - regpnode - emit a temporary node with a void* argument |
| 20333 | */ |
| 20334 | STATIC regnode_offset /* Location. */ |
| 20335 | S_regpnode(pTHX_ RExC_state_t *pRExC_state, U8 op, void * arg) |
| 20336 | { |
| 20337 | const regnode_offset ret = regnode_guts(pRExC_state, op, regarglen[op], "regvnode"); |
| 20338 | regnode_offset ptr = ret; |
| 20339 | |
| 20340 | PERL_ARGS_ASSERT_REGPNODE; |
| 20341 | |
| 20342 | FILL_ADVANCE_NODE_ARGp(ptr, op, arg); |
| 20343 | RExC_emit = ptr; |
| 20344 | return(ret); |
| 20345 | } |
| 20346 | |
| 20347 | STATIC regnode_offset |
| 20348 | S_reg2Lanode(pTHX_ RExC_state_t *pRExC_state, const U8 op, const U32 arg1, const I32 arg2) |
| 20349 | { |
| 20350 | /* emit a node with U32 and I32 arguments */ |
| 20351 | |
| 20352 | const regnode_offset ret = regnode_guts(pRExC_state, op, regarglen[op], "reg2Lanode"); |
| 20353 | regnode_offset ptr = ret; |
| 20354 | |
| 20355 | PERL_ARGS_ASSERT_REG2LANODE; |
| 20356 | |
| 20357 | assert(regarglen[op] == 2); |
| 20358 | |
| 20359 | FILL_ADVANCE_NODE_2L_ARG(ptr, op, arg1, arg2); |
| 20360 | RExC_emit = ptr; |
| 20361 | return(ret); |
| 20362 | } |
| 20363 | |
| 20364 | /* |
| 20365 | - reginsert - insert an operator in front of already-emitted operand |
| 20366 | * |
| 20367 | * That means that on exit 'operand' is the offset of the newly inserted |
| 20368 | * operator, and the original operand has been relocated. |
| 20369 | * |
| 20370 | * IMPORTANT NOTE - it is the *callers* responsibility to correctly |
| 20371 | * set up NEXT_OFF() of the inserted node if needed. Something like this: |
| 20372 | * |
| 20373 | * reginsert(pRExC, OPFAIL, orig_emit, depth+1); |
| 20374 | * NEXT_OFF(orig_emit) = regarglen[OPFAIL] + NODE_STEP_REGNODE; |
| 20375 | * |
| 20376 | * ALSO NOTE - FLAGS(newly-inserted-operator) will be set to 0 as well. |
| 20377 | */ |
| 20378 | STATIC void |
| 20379 | S_reginsert(pTHX_ RExC_state_t *pRExC_state, const U8 op, |
| 20380 | const regnode_offset operand, const U32 depth) |
| 20381 | { |
| 20382 | regnode *src; |
| 20383 | regnode *dst; |
| 20384 | regnode *place; |
| 20385 | const int offset = regarglen[(U8)op]; |
| 20386 | const int size = NODE_STEP_REGNODE + offset; |
| 20387 | GET_RE_DEBUG_FLAGS_DECL; |
| 20388 | |
| 20389 | PERL_ARGS_ASSERT_REGINSERT; |
| 20390 | PERL_UNUSED_CONTEXT; |
| 20391 | PERL_UNUSED_ARG(depth); |
| 20392 | /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */ |
| 20393 | DEBUG_PARSE_FMT("inst"," - %s", PL_reg_name[op]); |
| 20394 | assert(!RExC_study_started); /* I believe we should never use reginsert once we have started |
| 20395 | studying. If this is wrong then we need to adjust RExC_recurse |
| 20396 | below like we do with RExC_open_parens/RExC_close_parens. */ |
| 20397 | change_engine_size(pRExC_state, (Ptrdiff_t) size); |
| 20398 | src = REGNODE_p(RExC_emit); |
| 20399 | RExC_emit += size; |
| 20400 | dst = REGNODE_p(RExC_emit); |
| 20401 | |
| 20402 | /* If we are in a "count the parentheses" pass, the numbers are unreliable, |
| 20403 | * and [perl #133871] shows this can lead to problems, so skip this |
| 20404 | * realignment of parens until a later pass when they are reliable */ |
| 20405 | if (! IN_PARENS_PASS && RExC_open_parens) { |
| 20406 | int paren; |
| 20407 | /*DEBUG_PARSE_FMT("inst"," - %" IVdf, (IV)RExC_npar);*/ |
| 20408 | /* remember that RExC_npar is rex->nparens + 1, |
| 20409 | * iow it is 1 more than the number of parens seen in |
| 20410 | * the pattern so far. */ |
| 20411 | for ( paren=0 ; paren < RExC_npar ; paren++ ) { |
| 20412 | /* note, RExC_open_parens[0] is the start of the |
| 20413 | * regex, it can't move. RExC_close_parens[0] is the end |
| 20414 | * of the regex, it *can* move. */ |
| 20415 | if ( paren && RExC_open_parens[paren] >= operand ) { |
| 20416 | /*DEBUG_PARSE_FMT("open"," - %d", size);*/ |
| 20417 | RExC_open_parens[paren] += size; |
| 20418 | } else { |
| 20419 | /*DEBUG_PARSE_FMT("open"," - %s","ok");*/ |
| 20420 | } |
| 20421 | if ( RExC_close_parens[paren] >= operand ) { |
| 20422 | /*DEBUG_PARSE_FMT("close"," - %d", size);*/ |
| 20423 | RExC_close_parens[paren] += size; |
| 20424 | } else { |
| 20425 | /*DEBUG_PARSE_FMT("close"," - %s","ok");*/ |
| 20426 | } |
| 20427 | } |
| 20428 | } |
| 20429 | if (RExC_end_op) |
| 20430 | RExC_end_op += size; |
| 20431 | |
| 20432 | while (src > REGNODE_p(operand)) { |
| 20433 | StructCopy(--src, --dst, regnode); |
| 20434 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 20435 | if (RExC_offsets) { /* MJD 20010112 */ |
| 20436 | MJD_OFFSET_DEBUG( |
| 20437 | ("%s(%d): (op %s) %s copy %" UVuf " -> %" UVuf " (max %" UVuf ").\n", |
| 20438 | "reginsert", |
| 20439 | __LINE__, |
| 20440 | PL_reg_name[op], |
| 20441 | (UV)(REGNODE_OFFSET(dst)) > RExC_offsets[0] |
| 20442 | ? "Overwriting end of array!\n" : "OK", |
| 20443 | (UV)REGNODE_OFFSET(src), |
| 20444 | (UV)REGNODE_OFFSET(dst), |
| 20445 | (UV)RExC_offsets[0])); |
| 20446 | Set_Node_Offset_To_R(REGNODE_OFFSET(dst), Node_Offset(src)); |
| 20447 | Set_Node_Length_To_R(REGNODE_OFFSET(dst), Node_Length(src)); |
| 20448 | } |
| 20449 | #endif |
| 20450 | } |
| 20451 | |
| 20452 | place = REGNODE_p(operand); /* Op node, where operand used to be. */ |
| 20453 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 20454 | if (RExC_offsets) { /* MJD */ |
| 20455 | MJD_OFFSET_DEBUG( |
| 20456 | ("%s(%d): (op %s) %s %" UVuf " <- %" UVuf " (max %" UVuf ").\n", |
| 20457 | "reginsert", |
| 20458 | __LINE__, |
| 20459 | PL_reg_name[op], |
| 20460 | (UV)REGNODE_OFFSET(place) > RExC_offsets[0] |
| 20461 | ? "Overwriting end of array!\n" : "OK", |
| 20462 | (UV)REGNODE_OFFSET(place), |
| 20463 | (UV)(RExC_parse - RExC_start), |
| 20464 | (UV)RExC_offsets[0])); |
| 20465 | Set_Node_Offset(place, RExC_parse); |
| 20466 | Set_Node_Length(place, 1); |
| 20467 | } |
| 20468 | #endif |
| 20469 | src = NEXTOPER(place); |
| 20470 | FLAGS(place) = 0; |
| 20471 | FILL_NODE(operand, op); |
| 20472 | |
| 20473 | /* Zero out any arguments in the new node */ |
| 20474 | Zero(src, offset, regnode); |
| 20475 | } |
| 20476 | |
| 20477 | /* |
| 20478 | - regtail - set the next-pointer at the end of a node chain of p to val. If |
| 20479 | that value won't fit in the space available, instead returns FALSE. |
| 20480 | (Except asserts if we can't fit in the largest space the regex |
| 20481 | engine is designed for.) |
| 20482 | - SEE ALSO: regtail_study |
| 20483 | */ |
| 20484 | STATIC bool |
| 20485 | S_regtail(pTHX_ RExC_state_t * pRExC_state, |
| 20486 | const regnode_offset p, |
| 20487 | const regnode_offset val, |
| 20488 | const U32 depth) |
| 20489 | { |
| 20490 | regnode_offset scan; |
| 20491 | GET_RE_DEBUG_FLAGS_DECL; |
| 20492 | |
| 20493 | PERL_ARGS_ASSERT_REGTAIL; |
| 20494 | #ifndef DEBUGGING |
| 20495 | PERL_UNUSED_ARG(depth); |
| 20496 | #endif |
| 20497 | |
| 20498 | /* Find last node. */ |
| 20499 | scan = (regnode_offset) p; |
| 20500 | for (;;) { |
| 20501 | regnode * const temp = regnext(REGNODE_p(scan)); |
| 20502 | DEBUG_PARSE_r({ |
| 20503 | DEBUG_PARSE_MSG((scan==p ? "tail" : "")); |
| 20504 | regprop(RExC_rx, RExC_mysv, REGNODE_p(scan), NULL, pRExC_state); |
| 20505 | Perl_re_printf( aTHX_ "~ %s (%zu) %s %s\n", |
| 20506 | SvPV_nolen_const(RExC_mysv), scan, |
| 20507 | (temp == NULL ? "->" : ""), |
| 20508 | (temp == NULL ? PL_reg_name[OP(REGNODE_p(val))] : "") |
| 20509 | ); |
| 20510 | }); |
| 20511 | if (temp == NULL) |
| 20512 | break; |
| 20513 | scan = REGNODE_OFFSET(temp); |
| 20514 | } |
| 20515 | |
| 20516 | assert(val >= scan); |
| 20517 | if (reg_off_by_arg[OP(REGNODE_p(scan))]) { |
| 20518 | assert((UV) (val - scan) <= U32_MAX); |
| 20519 | ARG_SET(REGNODE_p(scan), val - scan); |
| 20520 | } |
| 20521 | else { |
| 20522 | if (val - scan > U16_MAX) { |
| 20523 | /* Populate this with something that won't loop and will likely |
| 20524 | * lead to a crash if the caller ignores the failure return, and |
| 20525 | * execution continues */ |
| 20526 | NEXT_OFF(REGNODE_p(scan)) = U16_MAX; |
| 20527 | return FALSE; |
| 20528 | } |
| 20529 | NEXT_OFF(REGNODE_p(scan)) = val - scan; |
| 20530 | } |
| 20531 | |
| 20532 | return TRUE; |
| 20533 | } |
| 20534 | |
| 20535 | #ifdef DEBUGGING |
| 20536 | /* |
| 20537 | - regtail_study - set the next-pointer at the end of a node chain of p to val. |
| 20538 | - Look for optimizable sequences at the same time. |
| 20539 | - currently only looks for EXACT chains. |
| 20540 | |
| 20541 | This is experimental code. The idea is to use this routine to perform |
| 20542 | in place optimizations on branches and groups as they are constructed, |
| 20543 | with the long term intention of removing optimization from study_chunk so |
| 20544 | that it is purely analytical. |
| 20545 | |
| 20546 | Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used |
| 20547 | to control which is which. |
| 20548 | |
| 20549 | This used to return a value that was ignored. It was a problem that it is |
| 20550 | #ifdef'd to be another function that didn't return a value. khw has changed it |
| 20551 | so both currently return a pass/fail return. |
| 20552 | |
| 20553 | */ |
| 20554 | /* TODO: All four parms should be const */ |
| 20555 | |
| 20556 | STATIC bool |
| 20557 | S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode_offset p, |
| 20558 | const regnode_offset val, U32 depth) |
| 20559 | { |
| 20560 | regnode_offset scan; |
| 20561 | U8 exact = PSEUDO; |
| 20562 | #ifdef EXPERIMENTAL_INPLACESCAN |
| 20563 | I32 min = 0; |
| 20564 | #endif |
| 20565 | GET_RE_DEBUG_FLAGS_DECL; |
| 20566 | |
| 20567 | PERL_ARGS_ASSERT_REGTAIL_STUDY; |
| 20568 | |
| 20569 | |
| 20570 | /* Find last node. */ |
| 20571 | |
| 20572 | scan = p; |
| 20573 | for (;;) { |
| 20574 | regnode * const temp = regnext(REGNODE_p(scan)); |
| 20575 | #ifdef EXPERIMENTAL_INPLACESCAN |
| 20576 | if (PL_regkind[OP(REGNODE_p(scan))] == EXACT) { |
| 20577 | bool unfolded_multi_char; /* Unexamined in this routine */ |
| 20578 | if (join_exact(pRExC_state, scan, &min, |
| 20579 | &unfolded_multi_char, 1, REGNODE_p(val), depth+1)) |
| 20580 | return TRUE; /* Was return EXACT */ |
| 20581 | } |
| 20582 | #endif |
| 20583 | if ( exact ) { |
| 20584 | switch (OP(REGNODE_p(scan))) { |
| 20585 | case LEXACT: |
| 20586 | case EXACT: |
| 20587 | case LEXACT_REQ8: |
| 20588 | case EXACT_REQ8: |
| 20589 | case EXACTL: |
| 20590 | case EXACTF: |
| 20591 | case EXACTFU_S_EDGE: |
| 20592 | case EXACTFAA_NO_TRIE: |
| 20593 | case EXACTFAA: |
| 20594 | case EXACTFU: |
| 20595 | case EXACTFU_REQ8: |
| 20596 | case EXACTFLU8: |
| 20597 | case EXACTFUP: |
| 20598 | case EXACTFL: |
| 20599 | if( exact == PSEUDO ) |
| 20600 | exact= OP(REGNODE_p(scan)); |
| 20601 | else if ( exact != OP(REGNODE_p(scan)) ) |
| 20602 | exact= 0; |
| 20603 | case NOTHING: |
| 20604 | break; |
| 20605 | default: |
| 20606 | exact= 0; |
| 20607 | } |
| 20608 | } |
| 20609 | DEBUG_PARSE_r({ |
| 20610 | DEBUG_PARSE_MSG((scan==p ? "tsdy" : "")); |
| 20611 | regprop(RExC_rx, RExC_mysv, REGNODE_p(scan), NULL, pRExC_state); |
| 20612 | Perl_re_printf( aTHX_ "~ %s (%zu) -> %s\n", |
| 20613 | SvPV_nolen_const(RExC_mysv), |
| 20614 | scan, |
| 20615 | PL_reg_name[exact]); |
| 20616 | }); |
| 20617 | if (temp == NULL) |
| 20618 | break; |
| 20619 | scan = REGNODE_OFFSET(temp); |
| 20620 | } |
| 20621 | DEBUG_PARSE_r({ |
| 20622 | DEBUG_PARSE_MSG(""); |
| 20623 | regprop(RExC_rx, RExC_mysv, REGNODE_p(val), NULL, pRExC_state); |
| 20624 | Perl_re_printf( aTHX_ |
| 20625 | "~ attach to %s (%" IVdf ") offset to %" IVdf "\n", |
| 20626 | SvPV_nolen_const(RExC_mysv), |
| 20627 | (IV)val, |
| 20628 | (IV)(val - scan) |
| 20629 | ); |
| 20630 | }); |
| 20631 | if (reg_off_by_arg[OP(REGNODE_p(scan))]) { |
| 20632 | assert((UV) (val - scan) <= U32_MAX); |
| 20633 | ARG_SET(REGNODE_p(scan), val - scan); |
| 20634 | } |
| 20635 | else { |
| 20636 | if (val - scan > U16_MAX) { |
| 20637 | /* Populate this with something that won't loop and will likely |
| 20638 | * lead to a crash if the caller ignores the failure return, and |
| 20639 | * execution continues */ |
| 20640 | NEXT_OFF(REGNODE_p(scan)) = U16_MAX; |
| 20641 | return FALSE; |
| 20642 | } |
| 20643 | NEXT_OFF(REGNODE_p(scan)) = val - scan; |
| 20644 | } |
| 20645 | |
| 20646 | return TRUE; /* Was 'return exact' */ |
| 20647 | } |
| 20648 | #endif |
| 20649 | |
| 20650 | STATIC SV* |
| 20651 | S_get_ANYOFM_contents(pTHX_ const regnode * n) { |
| 20652 | |
| 20653 | /* Returns an inversion list of all the code points matched by the |
| 20654 | * ANYOFM/NANYOFM node 'n' */ |
| 20655 | |
| 20656 | SV * cp_list = _new_invlist(-1); |
| 20657 | const U8 lowest = (U8) ARG(n); |
| 20658 | unsigned int i; |
| 20659 | U8 count = 0; |
| 20660 | U8 needed = 1U << PL_bitcount[ (U8) ~ FLAGS(n)]; |
| 20661 | |
| 20662 | PERL_ARGS_ASSERT_GET_ANYOFM_CONTENTS; |
| 20663 | |
| 20664 | /* Starting with the lowest code point, any code point that ANDed with the |
| 20665 | * mask yields the lowest code point is in the set */ |
| 20666 | for (i = lowest; i <= 0xFF; i++) { |
| 20667 | if ((i & FLAGS(n)) == ARG(n)) { |
| 20668 | cp_list = add_cp_to_invlist(cp_list, i); |
| 20669 | count++; |
| 20670 | |
| 20671 | /* We know how many code points (a power of two) that are in the |
| 20672 | * set. No use looking once we've got that number */ |
| 20673 | if (count >= needed) break; |
| 20674 | } |
| 20675 | } |
| 20676 | |
| 20677 | if (OP(n) == NANYOFM) { |
| 20678 | _invlist_invert(cp_list); |
| 20679 | } |
| 20680 | return cp_list; |
| 20681 | } |
| 20682 | |
| 20683 | /* |
| 20684 | - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form |
| 20685 | */ |
| 20686 | #ifdef DEBUGGING |
| 20687 | |
| 20688 | static void |
| 20689 | S_regdump_intflags(pTHX_ const char *lead, const U32 flags) |
| 20690 | { |
| 20691 | int bit; |
| 20692 | int set=0; |
| 20693 | |
| 20694 | ASSUME(REG_INTFLAGS_NAME_SIZE <= sizeof(flags)*8); |
| 20695 | |
| 20696 | for (bit=0; bit<REG_INTFLAGS_NAME_SIZE; bit++) { |
| 20697 | if (flags & (1<<bit)) { |
| 20698 | if (!set++ && lead) |
| 20699 | Perl_re_printf( aTHX_ "%s", lead); |
| 20700 | Perl_re_printf( aTHX_ "%s ", PL_reg_intflags_name[bit]); |
| 20701 | } |
| 20702 | } |
| 20703 | if (lead) { |
| 20704 | if (set) |
| 20705 | Perl_re_printf( aTHX_ "\n"); |
| 20706 | else |
| 20707 | Perl_re_printf( aTHX_ "%s[none-set]\n", lead); |
| 20708 | } |
| 20709 | } |
| 20710 | |
| 20711 | static void |
| 20712 | S_regdump_extflags(pTHX_ const char *lead, const U32 flags) |
| 20713 | { |
| 20714 | int bit; |
| 20715 | int set=0; |
| 20716 | regex_charset cs; |
| 20717 | |
| 20718 | ASSUME(REG_EXTFLAGS_NAME_SIZE <= sizeof(flags)*8); |
| 20719 | |
| 20720 | for (bit=0; bit<REG_EXTFLAGS_NAME_SIZE; bit++) { |
| 20721 | if (flags & (1<<bit)) { |
| 20722 | if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */ |
| 20723 | continue; |
| 20724 | } |
| 20725 | if (!set++ && lead) |
| 20726 | Perl_re_printf( aTHX_ "%s", lead); |
| 20727 | Perl_re_printf( aTHX_ "%s ", PL_reg_extflags_name[bit]); |
| 20728 | } |
| 20729 | } |
| 20730 | if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) { |
| 20731 | if (!set++ && lead) { |
| 20732 | Perl_re_printf( aTHX_ "%s", lead); |
| 20733 | } |
| 20734 | switch (cs) { |
| 20735 | case REGEX_UNICODE_CHARSET: |
| 20736 | Perl_re_printf( aTHX_ "UNICODE"); |
| 20737 | break; |
| 20738 | case REGEX_LOCALE_CHARSET: |
| 20739 | Perl_re_printf( aTHX_ "LOCALE"); |
| 20740 | break; |
| 20741 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 20742 | Perl_re_printf( aTHX_ "ASCII-RESTRICTED"); |
| 20743 | break; |
| 20744 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 20745 | Perl_re_printf( aTHX_ "ASCII-MORE_RESTRICTED"); |
| 20746 | break; |
| 20747 | default: |
| 20748 | Perl_re_printf( aTHX_ "UNKNOWN CHARACTER SET"); |
| 20749 | break; |
| 20750 | } |
| 20751 | } |
| 20752 | if (lead) { |
| 20753 | if (set) |
| 20754 | Perl_re_printf( aTHX_ "\n"); |
| 20755 | else |
| 20756 | Perl_re_printf( aTHX_ "%s[none-set]\n", lead); |
| 20757 | } |
| 20758 | } |
| 20759 | #endif |
| 20760 | |
| 20761 | void |
| 20762 | Perl_regdump(pTHX_ const regexp *r) |
| 20763 | { |
| 20764 | #ifdef DEBUGGING |
| 20765 | int i; |
| 20766 | SV * const sv = sv_newmortal(); |
| 20767 | SV *dsv= sv_newmortal(); |
| 20768 | RXi_GET_DECL(r, ri); |
| 20769 | GET_RE_DEBUG_FLAGS_DECL; |
| 20770 | |
| 20771 | PERL_ARGS_ASSERT_REGDUMP; |
| 20772 | |
| 20773 | (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0); |
| 20774 | |
| 20775 | /* Header fields of interest. */ |
| 20776 | for (i = 0; i < 2; i++) { |
| 20777 | if (r->substrs->data[i].substr) { |
| 20778 | RE_PV_QUOTED_DECL(s, 0, dsv, |
| 20779 | SvPVX_const(r->substrs->data[i].substr), |
| 20780 | RE_SV_DUMPLEN(r->substrs->data[i].substr), |
| 20781 | PL_dump_re_max_len); |
| 20782 | Perl_re_printf( aTHX_ |
| 20783 | "%s %s%s at %" IVdf "..%" UVuf " ", |
| 20784 | i ? "floating" : "anchored", |
| 20785 | s, |
| 20786 | RE_SV_TAIL(r->substrs->data[i].substr), |
| 20787 | (IV)r->substrs->data[i].min_offset, |
| 20788 | (UV)r->substrs->data[i].max_offset); |
| 20789 | } |
| 20790 | else if (r->substrs->data[i].utf8_substr) { |
| 20791 | RE_PV_QUOTED_DECL(s, 1, dsv, |
| 20792 | SvPVX_const(r->substrs->data[i].utf8_substr), |
| 20793 | RE_SV_DUMPLEN(r->substrs->data[i].utf8_substr), |
| 20794 | 30); |
| 20795 | Perl_re_printf( aTHX_ |
| 20796 | "%s utf8 %s%s at %" IVdf "..%" UVuf " ", |
| 20797 | i ? "floating" : "anchored", |
| 20798 | s, |
| 20799 | RE_SV_TAIL(r->substrs->data[i].utf8_substr), |
| 20800 | (IV)r->substrs->data[i].min_offset, |
| 20801 | (UV)r->substrs->data[i].max_offset); |
| 20802 | } |
| 20803 | } |
| 20804 | |
| 20805 | if (r->check_substr || r->check_utf8) |
| 20806 | Perl_re_printf( aTHX_ |
| 20807 | (const char *) |
| 20808 | ( r->check_substr == r->substrs->data[1].substr |
| 20809 | && r->check_utf8 == r->substrs->data[1].utf8_substr |
| 20810 | ? "(checking floating" : "(checking anchored")); |
| 20811 | if (r->intflags & PREGf_NOSCAN) |
| 20812 | Perl_re_printf( aTHX_ " noscan"); |
| 20813 | if (r->extflags & RXf_CHECK_ALL) |
| 20814 | Perl_re_printf( aTHX_ " isall"); |
| 20815 | if (r->check_substr || r->check_utf8) |
| 20816 | Perl_re_printf( aTHX_ ") "); |
| 20817 | |
| 20818 | if (ri->regstclass) { |
| 20819 | regprop(r, sv, ri->regstclass, NULL, NULL); |
| 20820 | Perl_re_printf( aTHX_ "stclass %s ", SvPVX_const(sv)); |
| 20821 | } |
| 20822 | if (r->intflags & PREGf_ANCH) { |
| 20823 | Perl_re_printf( aTHX_ "anchored"); |
| 20824 | if (r->intflags & PREGf_ANCH_MBOL) |
| 20825 | Perl_re_printf( aTHX_ "(MBOL)"); |
| 20826 | if (r->intflags & PREGf_ANCH_SBOL) |
| 20827 | Perl_re_printf( aTHX_ "(SBOL)"); |
| 20828 | if (r->intflags & PREGf_ANCH_GPOS) |
| 20829 | Perl_re_printf( aTHX_ "(GPOS)"); |
| 20830 | Perl_re_printf( aTHX_ " "); |
| 20831 | } |
| 20832 | if (r->intflags & PREGf_GPOS_SEEN) |
| 20833 | Perl_re_printf( aTHX_ "GPOS:%" UVuf " ", (UV)r->gofs); |
| 20834 | if (r->intflags & PREGf_SKIP) |
| 20835 | Perl_re_printf( aTHX_ "plus "); |
| 20836 | if (r->intflags & PREGf_IMPLICIT) |
| 20837 | Perl_re_printf( aTHX_ "implicit "); |
| 20838 | Perl_re_printf( aTHX_ "minlen %" IVdf " ", (IV)r->minlen); |
| 20839 | if (r->extflags & RXf_EVAL_SEEN) |
| 20840 | Perl_re_printf( aTHX_ "with eval "); |
| 20841 | Perl_re_printf( aTHX_ "\n"); |
| 20842 | DEBUG_FLAGS_r({ |
| 20843 | regdump_extflags("r->extflags: ", r->extflags); |
| 20844 | regdump_intflags("r->intflags: ", r->intflags); |
| 20845 | }); |
| 20846 | #else |
| 20847 | PERL_ARGS_ASSERT_REGDUMP; |
| 20848 | PERL_UNUSED_CONTEXT; |
| 20849 | PERL_UNUSED_ARG(r); |
| 20850 | #endif /* DEBUGGING */ |
| 20851 | } |
| 20852 | |
| 20853 | /* Should be synchronized with ANYOF_ #defines in regcomp.h */ |
| 20854 | #ifdef DEBUGGING |
| 20855 | |
| 20856 | # if _CC_WORDCHAR != 0 || _CC_DIGIT != 1 || _CC_ALPHA != 2 \ |
| 20857 | || _CC_LOWER != 3 || _CC_UPPER != 4 || _CC_PUNCT != 5 \ |
| 20858 | || _CC_PRINT != 6 || _CC_ALPHANUMERIC != 7 || _CC_GRAPH != 8 \ |
| 20859 | || _CC_CASED != 9 || _CC_SPACE != 10 || _CC_BLANK != 11 \ |
| 20860 | || _CC_XDIGIT != 12 || _CC_CNTRL != 13 || _CC_ASCII != 14 \ |
| 20861 | || _CC_VERTSPACE != 15 |
| 20862 | # error Need to adjust order of anyofs[] |
| 20863 | # endif |
| 20864 | static const char * const anyofs[] = { |
| 20865 | "\\w", |
| 20866 | "\\W", |
| 20867 | "\\d", |
| 20868 | "\\D", |
| 20869 | "[:alpha:]", |
| 20870 | "[:^alpha:]", |
| 20871 | "[:lower:]", |
| 20872 | "[:^lower:]", |
| 20873 | "[:upper:]", |
| 20874 | "[:^upper:]", |
| 20875 | "[:punct:]", |
| 20876 | "[:^punct:]", |
| 20877 | "[:print:]", |
| 20878 | "[:^print:]", |
| 20879 | "[:alnum:]", |
| 20880 | "[:^alnum:]", |
| 20881 | "[:graph:]", |
| 20882 | "[:^graph:]", |
| 20883 | "[:cased:]", |
| 20884 | "[:^cased:]", |
| 20885 | "\\s", |
| 20886 | "\\S", |
| 20887 | "[:blank:]", |
| 20888 | "[:^blank:]", |
| 20889 | "[:xdigit:]", |
| 20890 | "[:^xdigit:]", |
| 20891 | "[:cntrl:]", |
| 20892 | "[:^cntrl:]", |
| 20893 | "[:ascii:]", |
| 20894 | "[:^ascii:]", |
| 20895 | "\\v", |
| 20896 | "\\V" |
| 20897 | }; |
| 20898 | #endif |
| 20899 | |
| 20900 | /* |
| 20901 | - regprop - printable representation of opcode, with run time support |
| 20902 | */ |
| 20903 | |
| 20904 | void |
| 20905 | Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o, const regmatch_info *reginfo, const RExC_state_t *pRExC_state) |
| 20906 | { |
| 20907 | #ifdef DEBUGGING |
| 20908 | dVAR; |
| 20909 | int k; |
| 20910 | RXi_GET_DECL(prog, progi); |
| 20911 | GET_RE_DEBUG_FLAGS_DECL; |
| 20912 | |
| 20913 | PERL_ARGS_ASSERT_REGPROP; |
| 20914 | |
| 20915 | SvPVCLEAR(sv); |
| 20916 | |
| 20917 | if (OP(o) > REGNODE_MAX) { /* regnode.type is unsigned */ |
| 20918 | if (pRExC_state) { /* This gives more info, if we have it */ |
| 20919 | FAIL3("panic: corrupted regexp opcode %d > %d", |
| 20920 | (int)OP(o), (int)REGNODE_MAX); |
| 20921 | } |
| 20922 | else { |
| 20923 | Perl_croak(aTHX_ "panic: corrupted regexp opcode %d > %d", |
| 20924 | (int)OP(o), (int)REGNODE_MAX); |
| 20925 | } |
| 20926 | } |
| 20927 | sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */ |
| 20928 | |
| 20929 | k = PL_regkind[OP(o)]; |
| 20930 | |
| 20931 | if (k == EXACT) { |
| 20932 | sv_catpvs(sv, " "); |
| 20933 | /* Using is_utf8_string() (via PERL_PV_UNI_DETECT) |
| 20934 | * is a crude hack but it may be the best for now since |
| 20935 | * we have no flag "this EXACTish node was UTF-8" |
| 20936 | * --jhi */ |
| 20937 | pv_pretty(sv, STRING(o), STR_LEN(o), PL_dump_re_max_len, |
| 20938 | PL_colors[0], PL_colors[1], |
| 20939 | PERL_PV_ESCAPE_UNI_DETECT | |
| 20940 | PERL_PV_ESCAPE_NONASCII | |
| 20941 | PERL_PV_PRETTY_ELLIPSES | |
| 20942 | PERL_PV_PRETTY_LTGT | |
| 20943 | PERL_PV_PRETTY_NOCLEAR |
| 20944 | ); |
| 20945 | } else if (k == TRIE) { |
| 20946 | /* print the details of the trie in dumpuntil instead, as |
| 20947 | * progi->data isn't available here */ |
| 20948 | const char op = OP(o); |
| 20949 | const U32 n = ARG(o); |
| 20950 | const reg_ac_data * const ac = IS_TRIE_AC(op) ? |
| 20951 | (reg_ac_data *)progi->data->data[n] : |
| 20952 | NULL; |
| 20953 | const reg_trie_data * const trie |
| 20954 | = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie]; |
| 20955 | |
| 20956 | Perl_sv_catpvf(aTHX_ sv, "-%s", PL_reg_name[o->flags]); |
| 20957 | DEBUG_TRIE_COMPILE_r({ |
| 20958 | if (trie->jump) |
| 20959 | sv_catpvs(sv, "(JUMP)"); |
| 20960 | Perl_sv_catpvf(aTHX_ sv, |
| 20961 | "<S:%" UVuf "/%" IVdf " W:%" UVuf " L:%" UVuf "/%" UVuf " C:%" UVuf "/%" UVuf ">", |
| 20962 | (UV)trie->startstate, |
| 20963 | (IV)trie->statecount-1, /* -1 because of the unused 0 element */ |
| 20964 | (UV)trie->wordcount, |
| 20965 | (UV)trie->minlen, |
| 20966 | (UV)trie->maxlen, |
| 20967 | (UV)TRIE_CHARCOUNT(trie), |
| 20968 | (UV)trie->uniquecharcount |
| 20969 | ); |
| 20970 | }); |
| 20971 | if ( IS_ANYOF_TRIE(op) || trie->bitmap ) { |
| 20972 | sv_catpvs(sv, "["); |
| 20973 | (void) put_charclass_bitmap_innards(sv, |
| 20974 | ((IS_ANYOF_TRIE(op)) |
| 20975 | ? ANYOF_BITMAP(o) |
| 20976 | : TRIE_BITMAP(trie)), |
| 20977 | NULL, |
| 20978 | NULL, |
| 20979 | NULL, |
| 20980 | 0, |
| 20981 | FALSE |
| 20982 | ); |
| 20983 | sv_catpvs(sv, "]"); |
| 20984 | } |
| 20985 | } else if (k == CURLY) { |
| 20986 | U32 lo = ARG1(o), hi = ARG2(o); |
| 20987 | if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX) |
| 20988 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */ |
| 20989 | Perl_sv_catpvf(aTHX_ sv, "{%u,", (unsigned) lo); |
| 20990 | if (hi == REG_INFTY) |
| 20991 | sv_catpvs(sv, "INFTY"); |
| 20992 | else |
| 20993 | Perl_sv_catpvf(aTHX_ sv, "%u", (unsigned) hi); |
| 20994 | sv_catpvs(sv, "}"); |
| 20995 | } |
| 20996 | else if (k == WHILEM && o->flags) /* Ordinal/of */ |
| 20997 | Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4); |
| 20998 | else if (k == REF || k == OPEN || k == CLOSE |
| 20999 | || k == GROUPP || OP(o)==ACCEPT) |
| 21000 | { |
| 21001 | AV *name_list= NULL; |
| 21002 | U32 parno= OP(o) == ACCEPT ? (U32)ARG2L(o) : ARG(o); |
| 21003 | Perl_sv_catpvf(aTHX_ sv, "%" UVuf, (UV)parno); /* Parenth number */ |
| 21004 | if ( RXp_PAREN_NAMES(prog) ) { |
| 21005 | name_list= MUTABLE_AV(progi->data->data[progi->name_list_idx]); |
| 21006 | } else if ( pRExC_state ) { |
| 21007 | name_list= RExC_paren_name_list; |
| 21008 | } |
| 21009 | if (name_list) { |
| 21010 | if ( k != REF || (OP(o) < REFN)) { |
| 21011 | SV **name= av_fetch(name_list, parno, 0 ); |
| 21012 | if (name) |
| 21013 | Perl_sv_catpvf(aTHX_ sv, " '%" SVf "'", SVfARG(*name)); |
| 21014 | } |
| 21015 | else { |
| 21016 | SV *sv_dat= MUTABLE_SV(progi->data->data[ parno ]); |
| 21017 | I32 *nums=(I32*)SvPVX(sv_dat); |
| 21018 | SV **name= av_fetch(name_list, nums[0], 0 ); |
| 21019 | I32 n; |
| 21020 | if (name) { |
| 21021 | for ( n=0; n<SvIVX(sv_dat); n++ ) { |
| 21022 | Perl_sv_catpvf(aTHX_ sv, "%s%" IVdf, |
| 21023 | (n ? "," : ""), (IV)nums[n]); |
| 21024 | } |
| 21025 | Perl_sv_catpvf(aTHX_ sv, " '%" SVf "'", SVfARG(*name)); |
| 21026 | } |
| 21027 | } |
| 21028 | } |
| 21029 | if ( k == REF && reginfo) { |
| 21030 | U32 n = ARG(o); /* which paren pair */ |
| 21031 | I32 ln = prog->offs[n].start; |
| 21032 | if (prog->lastparen < n || ln == -1 || prog->offs[n].end == -1) |
| 21033 | Perl_sv_catpvf(aTHX_ sv, ": FAIL"); |
| 21034 | else if (ln == prog->offs[n].end) |
| 21035 | Perl_sv_catpvf(aTHX_ sv, ": ACCEPT - EMPTY STRING"); |
| 21036 | else { |
| 21037 | const char *s = reginfo->strbeg + ln; |
| 21038 | Perl_sv_catpvf(aTHX_ sv, ": "); |
| 21039 | Perl_pv_pretty( aTHX_ sv, s, prog->offs[n].end - prog->offs[n].start, 32, 0, 0, |
| 21040 | PERL_PV_ESCAPE_UNI_DETECT|PERL_PV_PRETTY_NOCLEAR|PERL_PV_PRETTY_ELLIPSES|PERL_PV_PRETTY_QUOTE ); |
| 21041 | } |
| 21042 | } |
| 21043 | } else if (k == GOSUB) { |
| 21044 | AV *name_list= NULL; |
| 21045 | if ( RXp_PAREN_NAMES(prog) ) { |
| 21046 | name_list= MUTABLE_AV(progi->data->data[progi->name_list_idx]); |
| 21047 | } else if ( pRExC_state ) { |
| 21048 | name_list= RExC_paren_name_list; |
| 21049 | } |
| 21050 | |
| 21051 | /* Paren and offset */ |
| 21052 | Perl_sv_catpvf(aTHX_ sv, "%d[%+d:%d]", (int)ARG(o),(int)ARG2L(o), |
| 21053 | (int)((o + (int)ARG2L(o)) - progi->program) ); |
| 21054 | if (name_list) { |
| 21055 | SV **name= av_fetch(name_list, ARG(o), 0 ); |
| 21056 | if (name) |
| 21057 | Perl_sv_catpvf(aTHX_ sv, " '%" SVf "'", SVfARG(*name)); |
| 21058 | } |
| 21059 | } |
| 21060 | else if (k == LOGICAL) |
| 21061 | /* 2: embedded, otherwise 1 */ |
| 21062 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); |
| 21063 | else if (k == ANYOF || k == ANYOFR) { |
| 21064 | U8 flags; |
| 21065 | char * bitmap; |
| 21066 | U32 arg; |
| 21067 | bool do_sep = FALSE; /* Do we need to separate various components of |
| 21068 | the output? */ |
| 21069 | /* Set if there is still an unresolved user-defined property */ |
| 21070 | SV *unresolved = NULL; |
| 21071 | |
| 21072 | /* Things that are ignored except when the runtime locale is UTF-8 */ |
| 21073 | SV *only_utf8_locale_invlist = NULL; |
| 21074 | |
| 21075 | /* Code points that don't fit in the bitmap */ |
| 21076 | SV *nonbitmap_invlist = NULL; |
| 21077 | |
| 21078 | /* And things that aren't in the bitmap, but are small enough to be */ |
| 21079 | SV* bitmap_range_not_in_bitmap = NULL; |
| 21080 | |
| 21081 | bool inverted; |
| 21082 | |
| 21083 | if (inRANGE(OP(o), ANYOFH, ANYOFRb)) { |
| 21084 | flags = 0; |
| 21085 | bitmap = NULL; |
| 21086 | arg = 0; |
| 21087 | } |
| 21088 | else { |
| 21089 | flags = ANYOF_FLAGS(o); |
| 21090 | bitmap = ANYOF_BITMAP(o); |
| 21091 | arg = ARG(o); |
| 21092 | } |
| 21093 | |
| 21094 | if (OP(o) == ANYOFL || OP(o) == ANYOFPOSIXL) { |
| 21095 | if (ANYOFL_UTF8_LOCALE_REQD(flags)) { |
| 21096 | sv_catpvs(sv, "{utf8-locale-reqd}"); |
| 21097 | } |
| 21098 | if (flags & ANYOFL_FOLD) { |
| 21099 | sv_catpvs(sv, "{i}"); |
| 21100 | } |
| 21101 | } |
| 21102 | |
| 21103 | inverted = flags & ANYOF_INVERT; |
| 21104 | |
| 21105 | /* If there is stuff outside the bitmap, get it */ |
| 21106 | if (arg != ANYOF_ONLY_HAS_BITMAP) { |
| 21107 | if (inRANGE(OP(o), ANYOFR, ANYOFRb)) { |
| 21108 | nonbitmap_invlist = _add_range_to_invlist(nonbitmap_invlist, |
| 21109 | ANYOFRbase(o), |
| 21110 | ANYOFRbase(o) + ANYOFRdelta(o)); |
| 21111 | } |
| 21112 | else { |
| 21113 | (void) _get_regclass_nonbitmap_data(prog, o, FALSE, |
| 21114 | &unresolved, |
| 21115 | &only_utf8_locale_invlist, |
| 21116 | &nonbitmap_invlist); |
| 21117 | } |
| 21118 | |
| 21119 | /* The non-bitmap data may contain stuff that could fit in the |
| 21120 | * bitmap. This could come from a user-defined property being |
| 21121 | * finally resolved when this call was done; or much more likely |
| 21122 | * because there are matches that require UTF-8 to be valid, and so |
| 21123 | * aren't in the bitmap (or ANYOFR). This is teased apart later */ |
| 21124 | _invlist_intersection(nonbitmap_invlist, |
| 21125 | PL_InBitmap, |
| 21126 | &bitmap_range_not_in_bitmap); |
| 21127 | /* Leave just the things that don't fit into the bitmap */ |
| 21128 | _invlist_subtract(nonbitmap_invlist, |
| 21129 | PL_InBitmap, |
| 21130 | &nonbitmap_invlist); |
| 21131 | } |
| 21132 | |
| 21133 | /* Obey this flag to add all above-the-bitmap code points */ |
| 21134 | if (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP) { |
| 21135 | nonbitmap_invlist = _add_range_to_invlist(nonbitmap_invlist, |
| 21136 | NUM_ANYOF_CODE_POINTS, |
| 21137 | UV_MAX); |
| 21138 | } |
| 21139 | |
| 21140 | /* Ready to start outputting. First, the initial left bracket */ |
| 21141 | Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]); |
| 21142 | |
| 21143 | /* ANYOFH by definition doesn't have anything that will fit inside the |
| 21144 | * bitmap; ANYOFR may or may not. */ |
| 21145 | if ( ! inRANGE(OP(o), ANYOFH, ANYOFHr) |
| 21146 | && ( ! inRANGE(OP(o), ANYOFR, ANYOFRb) |
| 21147 | || ANYOFRbase(o) < NUM_ANYOF_CODE_POINTS)) |
| 21148 | { |
| 21149 | /* Then all the things that could fit in the bitmap */ |
| 21150 | do_sep = put_charclass_bitmap_innards(sv, |
| 21151 | bitmap, |
| 21152 | bitmap_range_not_in_bitmap, |
| 21153 | only_utf8_locale_invlist, |
| 21154 | o, |
| 21155 | flags, |
| 21156 | |
| 21157 | /* Can't try inverting for a |
| 21158 | * better display if there |
| 21159 | * are things that haven't |
| 21160 | * been resolved */ |
| 21161 | unresolved != NULL |
| 21162 | || inRANGE(OP(o), ANYOFR, ANYOFRb)); |
| 21163 | SvREFCNT_dec(bitmap_range_not_in_bitmap); |
| 21164 | |
| 21165 | /* If there are user-defined properties which haven't been defined |
| 21166 | * yet, output them. If the result is not to be inverted, it is |
| 21167 | * clearest to output them in a separate [] from the bitmap range |
| 21168 | * stuff. If the result is to be complemented, we have to show |
| 21169 | * everything in one [], as the inversion applies to the whole |
| 21170 | * thing. Use {braces} to separate them from anything in the |
| 21171 | * bitmap and anything above the bitmap. */ |
| 21172 | if (unresolved) { |
| 21173 | if (inverted) { |
| 21174 | if (! do_sep) { /* If didn't output anything in the bitmap |
| 21175 | */ |
| 21176 | sv_catpvs(sv, "^"); |
| 21177 | } |
| 21178 | sv_catpvs(sv, "{"); |
| 21179 | } |
| 21180 | else if (do_sep) { |
| 21181 | Perl_sv_catpvf(aTHX_ sv,"%s][%s", PL_colors[1], |
| 21182 | PL_colors[0]); |
| 21183 | } |
| 21184 | sv_catsv(sv, unresolved); |
| 21185 | if (inverted) { |
| 21186 | sv_catpvs(sv, "}"); |
| 21187 | } |
| 21188 | do_sep = ! inverted; |
| 21189 | } |
| 21190 | } |
| 21191 | |
| 21192 | /* And, finally, add the above-the-bitmap stuff */ |
| 21193 | if (nonbitmap_invlist && _invlist_len(nonbitmap_invlist)) { |
| 21194 | SV* contents; |
| 21195 | |
| 21196 | /* See if truncation size is overridden */ |
| 21197 | const STRLEN dump_len = (PL_dump_re_max_len > 256) |
| 21198 | ? PL_dump_re_max_len |
| 21199 | : 256; |
| 21200 | |
| 21201 | /* This is output in a separate [] */ |
| 21202 | if (do_sep) { |
| 21203 | Perl_sv_catpvf(aTHX_ sv,"%s][%s", PL_colors[1], PL_colors[0]); |
| 21204 | } |
| 21205 | |
| 21206 | /* And, for easy of understanding, it is shown in the |
| 21207 | * uncomplemented form if possible. The one exception being if |
| 21208 | * there are unresolved items, where the inversion has to be |
| 21209 | * delayed until runtime */ |
| 21210 | if (inverted && ! unresolved) { |
| 21211 | _invlist_invert(nonbitmap_invlist); |
| 21212 | _invlist_subtract(nonbitmap_invlist, PL_InBitmap, &nonbitmap_invlist); |
| 21213 | } |
| 21214 | |
| 21215 | contents = invlist_contents(nonbitmap_invlist, |
| 21216 | FALSE /* output suitable for catsv */ |
| 21217 | ); |
| 21218 | |
| 21219 | /* If the output is shorter than the permissible maximum, just do it. */ |
| 21220 | if (SvCUR(contents) <= dump_len) { |
| 21221 | sv_catsv(sv, contents); |
| 21222 | } |
| 21223 | else { |
| 21224 | const char * contents_string = SvPVX(contents); |
| 21225 | STRLEN i = dump_len; |
| 21226 | |
| 21227 | /* Otherwise, start at the permissible max and work back to the |
| 21228 | * first break possibility */ |
| 21229 | while (i > 0 && contents_string[i] != ' ') { |
| 21230 | i--; |
| 21231 | } |
| 21232 | if (i == 0) { /* Fail-safe. Use the max if we couldn't |
| 21233 | find a legal break */ |
| 21234 | i = dump_len; |
| 21235 | } |
| 21236 | |
| 21237 | sv_catpvn(sv, contents_string, i); |
| 21238 | sv_catpvs(sv, "..."); |
| 21239 | } |
| 21240 | |
| 21241 | SvREFCNT_dec_NN(contents); |
| 21242 | SvREFCNT_dec_NN(nonbitmap_invlist); |
| 21243 | } |
| 21244 | |
| 21245 | /* And finally the matching, closing ']' */ |
| 21246 | Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]); |
| 21247 | |
| 21248 | if (OP(o) == ANYOFHs) { |
| 21249 | Perl_sv_catpvf(aTHX_ sv, " (Leading UTF-8 bytes=%s", _byte_dump_string((U8 *) ((struct regnode_anyofhs *) o)->string, FLAGS(o), 1)); |
| 21250 | } |
| 21251 | else if (inRANGE(OP(o), ANYOFH, ANYOFRb)) { |
| 21252 | U8 lowest = (OP(o) != ANYOFHr) |
| 21253 | ? FLAGS(o) |
| 21254 | : LOWEST_ANYOF_HRx_BYTE(FLAGS(o)); |
| 21255 | U8 highest = (OP(o) == ANYOFHr) |
| 21256 | ? HIGHEST_ANYOF_HRx_BYTE(FLAGS(o)) |
| 21257 | : (OP(o) == ANYOFH || OP(o) == ANYOFR) |
| 21258 | ? 0xFF |
| 21259 | : lowest; |
| 21260 | Perl_sv_catpvf(aTHX_ sv, " (First UTF-8 byte=%02X", lowest); |
| 21261 | if (lowest != highest) { |
| 21262 | Perl_sv_catpvf(aTHX_ sv, "-%02X", highest); |
| 21263 | } |
| 21264 | Perl_sv_catpvf(aTHX_ sv, ")"); |
| 21265 | } |
| 21266 | |
| 21267 | SvREFCNT_dec(unresolved); |
| 21268 | } |
| 21269 | else if (k == ANYOFM) { |
| 21270 | SV * cp_list = get_ANYOFM_contents(o); |
| 21271 | |
| 21272 | Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]); |
| 21273 | if (OP(o) == NANYOFM) { |
| 21274 | _invlist_invert(cp_list); |
| 21275 | } |
| 21276 | |
| 21277 | put_charclass_bitmap_innards(sv, NULL, cp_list, NULL, NULL, 0, TRUE); |
| 21278 | Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]); |
| 21279 | |
| 21280 | SvREFCNT_dec(cp_list); |
| 21281 | } |
| 21282 | else if (k == POSIXD || k == NPOSIXD) { |
| 21283 | U8 index = FLAGS(o) * 2; |
| 21284 | if (index < C_ARRAY_LENGTH(anyofs)) { |
| 21285 | if (*anyofs[index] != '[') { |
| 21286 | sv_catpvs(sv, "["); |
| 21287 | } |
| 21288 | sv_catpv(sv, anyofs[index]); |
| 21289 | if (*anyofs[index] != '[') { |
| 21290 | sv_catpvs(sv, "]"); |
| 21291 | } |
| 21292 | } |
| 21293 | else { |
| 21294 | Perl_sv_catpvf(aTHX_ sv, "[illegal type=%d])", index); |
| 21295 | } |
| 21296 | } |
| 21297 | else if (k == BOUND || k == NBOUND) { |
| 21298 | /* Must be synced with order of 'bound_type' in regcomp.h */ |
| 21299 | const char * const bounds[] = { |
| 21300 | "", /* Traditional */ |
| 21301 | "{gcb}", |
| 21302 | "{lb}", |
| 21303 | "{sb}", |
| 21304 | "{wb}" |
| 21305 | }; |
| 21306 | assert(FLAGS(o) < C_ARRAY_LENGTH(bounds)); |
| 21307 | sv_catpv(sv, bounds[FLAGS(o)]); |
| 21308 | } |
| 21309 | else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH)) { |
| 21310 | Perl_sv_catpvf(aTHX_ sv, "[%d", -(o->flags)); |
| 21311 | if (o->next_off) { |
| 21312 | Perl_sv_catpvf(aTHX_ sv, "..-%d", o->flags - o->next_off); |
| 21313 | } |
| 21314 | Perl_sv_catpvf(aTHX_ sv, "]"); |
| 21315 | } |
| 21316 | else if (OP(o) == SBOL) |
| 21317 | Perl_sv_catpvf(aTHX_ sv, " /%s/", o->flags ? "\\A" : "^"); |
| 21318 | |
| 21319 | /* add on the verb argument if there is one */ |
| 21320 | if ( ( k == VERB || OP(o) == ACCEPT || OP(o) == OPFAIL ) && o->flags) { |
| 21321 | if ( ARG(o) ) |
| 21322 | Perl_sv_catpvf(aTHX_ sv, ":%" SVf, |
| 21323 | SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ])))); |
| 21324 | else |
| 21325 | sv_catpvs(sv, ":NULL"); |
| 21326 | } |
| 21327 | #else |
| 21328 | PERL_UNUSED_CONTEXT; |
| 21329 | PERL_UNUSED_ARG(sv); |
| 21330 | PERL_UNUSED_ARG(o); |
| 21331 | PERL_UNUSED_ARG(prog); |
| 21332 | PERL_UNUSED_ARG(reginfo); |
| 21333 | PERL_UNUSED_ARG(pRExC_state); |
| 21334 | #endif /* DEBUGGING */ |
| 21335 | } |
| 21336 | |
| 21337 | |
| 21338 | |
| 21339 | SV * |
| 21340 | Perl_re_intuit_string(pTHX_ REGEXP * const r) |
| 21341 | { /* Assume that RE_INTUIT is set */ |
| 21342 | /* Returns an SV containing a string that must appear in the target for it |
| 21343 | * to match */ |
| 21344 | |
| 21345 | struct regexp *const prog = ReANY(r); |
| 21346 | GET_RE_DEBUG_FLAGS_DECL; |
| 21347 | |
| 21348 | PERL_ARGS_ASSERT_RE_INTUIT_STRING; |
| 21349 | PERL_UNUSED_CONTEXT; |
| 21350 | |
| 21351 | DEBUG_COMPILE_r( |
| 21352 | { |
| 21353 | if (prog->maxlen > 0) { |
| 21354 | const char * const s = SvPV_nolen_const(RX_UTF8(r) |
| 21355 | ? prog->check_utf8 : prog->check_substr); |
| 21356 | |
| 21357 | if (!PL_colorset) reginitcolors(); |
| 21358 | Perl_re_printf( aTHX_ |
| 21359 | "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n", |
| 21360 | PL_colors[4], |
| 21361 | RX_UTF8(r) ? "utf8 " : "", |
| 21362 | PL_colors[5], PL_colors[0], |
| 21363 | s, |
| 21364 | PL_colors[1], |
| 21365 | (strlen(s) > PL_dump_re_max_len ? "..." : "")); |
| 21366 | } |
| 21367 | } ); |
| 21368 | |
| 21369 | /* use UTF8 check substring if regexp pattern itself is in UTF8 */ |
| 21370 | return RX_UTF8(r) ? prog->check_utf8 : prog->check_substr; |
| 21371 | } |
| 21372 | |
| 21373 | /* |
| 21374 | pregfree() |
| 21375 | |
| 21376 | handles refcounting and freeing the perl core regexp structure. When |
| 21377 | it is necessary to actually free the structure the first thing it |
| 21378 | does is call the 'free' method of the regexp_engine associated to |
| 21379 | the regexp, allowing the handling of the void *pprivate; member |
| 21380 | first. (This routine is not overridable by extensions, which is why |
| 21381 | the extensions free is called first.) |
| 21382 | |
| 21383 | See regdupe and regdupe_internal if you change anything here. |
| 21384 | */ |
| 21385 | #ifndef PERL_IN_XSUB_RE |
| 21386 | void |
| 21387 | Perl_pregfree(pTHX_ REGEXP *r) |
| 21388 | { |
| 21389 | SvREFCNT_dec(r); |
| 21390 | } |
| 21391 | |
| 21392 | void |
| 21393 | Perl_pregfree2(pTHX_ REGEXP *rx) |
| 21394 | { |
| 21395 | struct regexp *const r = ReANY(rx); |
| 21396 | GET_RE_DEBUG_FLAGS_DECL; |
| 21397 | |
| 21398 | PERL_ARGS_ASSERT_PREGFREE2; |
| 21399 | |
| 21400 | if (! r) |
| 21401 | return; |
| 21402 | |
| 21403 | if (r->mother_re) { |
| 21404 | ReREFCNT_dec(r->mother_re); |
| 21405 | } else { |
| 21406 | CALLREGFREE_PVT(rx); /* free the private data */ |
| 21407 | SvREFCNT_dec(RXp_PAREN_NAMES(r)); |
| 21408 | } |
| 21409 | if (r->substrs) { |
| 21410 | int i; |
| 21411 | for (i = 0; i < 2; i++) { |
| 21412 | SvREFCNT_dec(r->substrs->data[i].substr); |
| 21413 | SvREFCNT_dec(r->substrs->data[i].utf8_substr); |
| 21414 | } |
| 21415 | Safefree(r->substrs); |
| 21416 | } |
| 21417 | RX_MATCH_COPY_FREE(rx); |
| 21418 | #ifdef PERL_ANY_COW |
| 21419 | SvREFCNT_dec(r->saved_copy); |
| 21420 | #endif |
| 21421 | Safefree(r->offs); |
| 21422 | SvREFCNT_dec(r->qr_anoncv); |
| 21423 | if (r->recurse_locinput) |
| 21424 | Safefree(r->recurse_locinput); |
| 21425 | } |
| 21426 | |
| 21427 | |
| 21428 | /* reg_temp_copy() |
| 21429 | |
| 21430 | Copy ssv to dsv, both of which should of type SVt_REGEXP or SVt_PVLV, |
| 21431 | except that dsv will be created if NULL. |
| 21432 | |
| 21433 | This function is used in two main ways. First to implement |
| 21434 | $r = qr/....; $s = $$r; |
| 21435 | |
| 21436 | Secondly, it is used as a hacky workaround to the structural issue of |
| 21437 | match results |
| 21438 | being stored in the regexp structure which is in turn stored in |
| 21439 | PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern |
| 21440 | could be PL_curpm in multiple contexts, and could require multiple |
| 21441 | result sets being associated with the pattern simultaneously, such |
| 21442 | as when doing a recursive match with (??{$qr}) |
| 21443 | |
| 21444 | The solution is to make a lightweight copy of the regexp structure |
| 21445 | when a qr// is returned from the code executed by (??{$qr}) this |
| 21446 | lightweight copy doesn't actually own any of its data except for |
| 21447 | the starp/end and the actual regexp structure itself. |
| 21448 | |
| 21449 | */ |
| 21450 | |
| 21451 | |
| 21452 | REGEXP * |
| 21453 | Perl_reg_temp_copy(pTHX_ REGEXP *dsv, REGEXP *ssv) |
| 21454 | { |
| 21455 | struct regexp *drx; |
| 21456 | struct regexp *const srx = ReANY(ssv); |
| 21457 | const bool islv = dsv && SvTYPE(dsv) == SVt_PVLV; |
| 21458 | |
| 21459 | PERL_ARGS_ASSERT_REG_TEMP_COPY; |
| 21460 | |
| 21461 | if (!dsv) |
| 21462 | dsv = (REGEXP*) newSV_type(SVt_REGEXP); |
| 21463 | else { |
| 21464 | assert(SvTYPE(dsv) == SVt_REGEXP || (SvTYPE(dsv) == SVt_PVLV)); |
| 21465 | |
| 21466 | /* our only valid caller, sv_setsv_flags(), should have done |
| 21467 | * a SV_CHECK_THINKFIRST_COW_DROP() by now */ |
| 21468 | assert(!SvOOK(dsv)); |
| 21469 | assert(!SvIsCOW(dsv)); |
| 21470 | assert(!SvROK(dsv)); |
| 21471 | |
| 21472 | if (SvPVX_const(dsv)) { |
| 21473 | if (SvLEN(dsv)) |
| 21474 | Safefree(SvPVX(dsv)); |
| 21475 | SvPVX(dsv) = NULL; |
| 21476 | } |
| 21477 | SvLEN_set(dsv, 0); |
| 21478 | SvCUR_set(dsv, 0); |
| 21479 | SvOK_off((SV *)dsv); |
| 21480 | |
| 21481 | if (islv) { |
| 21482 | /* For PVLVs, the head (sv_any) points to an XPVLV, while |
| 21483 | * the LV's xpvlenu_rx will point to a regexp body, which |
| 21484 | * we allocate here */ |
| 21485 | REGEXP *temp = (REGEXP *)newSV_type(SVt_REGEXP); |
| 21486 | assert(!SvPVX(dsv)); |
| 21487 | ((XPV*)SvANY(dsv))->xpv_len_u.xpvlenu_rx = temp->sv_any; |
| 21488 | temp->sv_any = NULL; |
| 21489 | SvFLAGS(temp) = (SvFLAGS(temp) & ~SVTYPEMASK) | SVt_NULL; |
| 21490 | SvREFCNT_dec_NN(temp); |
| 21491 | /* SvCUR still resides in the xpvlv struct, so the regexp copy- |
| 21492 | ing below will not set it. */ |
| 21493 | SvCUR_set(dsv, SvCUR(ssv)); |
| 21494 | } |
| 21495 | } |
| 21496 | /* This ensures that SvTHINKFIRST(sv) is true, and hence that |
| 21497 | sv_force_normal(sv) is called. */ |
| 21498 | SvFAKE_on(dsv); |
| 21499 | drx = ReANY(dsv); |
| 21500 | |
| 21501 | SvFLAGS(dsv) |= SvFLAGS(ssv) & (SVf_POK|SVp_POK|SVf_UTF8); |
| 21502 | SvPV_set(dsv, RX_WRAPPED(ssv)); |
| 21503 | /* We share the same string buffer as the original regexp, on which we |
| 21504 | hold a reference count, incremented when mother_re is set below. |
| 21505 | The string pointer is copied here, being part of the regexp struct. |
| 21506 | */ |
| 21507 | memcpy(&(drx->xpv_cur), &(srx->xpv_cur), |
| 21508 | sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur)); |
| 21509 | if (!islv) |
| 21510 | SvLEN_set(dsv, 0); |
| 21511 | if (srx->offs) { |
| 21512 | const I32 npar = srx->nparens+1; |
| 21513 | Newx(drx->offs, npar, regexp_paren_pair); |
| 21514 | Copy(srx->offs, drx->offs, npar, regexp_paren_pair); |
| 21515 | } |
| 21516 | if (srx->substrs) { |
| 21517 | int i; |
| 21518 | Newx(drx->substrs, 1, struct reg_substr_data); |
| 21519 | StructCopy(srx->substrs, drx->substrs, struct reg_substr_data); |
| 21520 | |
| 21521 | for (i = 0; i < 2; i++) { |
| 21522 | SvREFCNT_inc_void(drx->substrs->data[i].substr); |
| 21523 | SvREFCNT_inc_void(drx->substrs->data[i].utf8_substr); |
| 21524 | } |
| 21525 | |
| 21526 | /* check_substr and check_utf8, if non-NULL, point to either their |
| 21527 | anchored or float namesakes, and don't hold a second reference. */ |
| 21528 | } |
| 21529 | RX_MATCH_COPIED_off(dsv); |
| 21530 | #ifdef PERL_ANY_COW |
| 21531 | drx->saved_copy = NULL; |
| 21532 | #endif |
| 21533 | drx->mother_re = ReREFCNT_inc(srx->mother_re ? srx->mother_re : ssv); |
| 21534 | SvREFCNT_inc_void(drx->qr_anoncv); |
| 21535 | if (srx->recurse_locinput) |
| 21536 | Newx(drx->recurse_locinput, srx->nparens + 1, char *); |
| 21537 | |
| 21538 | return dsv; |
| 21539 | } |
| 21540 | #endif |
| 21541 | |
| 21542 | |
| 21543 | /* regfree_internal() |
| 21544 | |
| 21545 | Free the private data in a regexp. This is overloadable by |
| 21546 | extensions. Perl takes care of the regexp structure in pregfree(), |
| 21547 | this covers the *pprivate pointer which technically perl doesn't |
| 21548 | know about, however of course we have to handle the |
| 21549 | regexp_internal structure when no extension is in use. |
| 21550 | |
| 21551 | Note this is called before freeing anything in the regexp |
| 21552 | structure. |
| 21553 | */ |
| 21554 | |
| 21555 | void |
| 21556 | Perl_regfree_internal(pTHX_ REGEXP * const rx) |
| 21557 | { |
| 21558 | struct regexp *const r = ReANY(rx); |
| 21559 | RXi_GET_DECL(r, ri); |
| 21560 | GET_RE_DEBUG_FLAGS_DECL; |
| 21561 | |
| 21562 | PERL_ARGS_ASSERT_REGFREE_INTERNAL; |
| 21563 | |
| 21564 | if (! ri) { |
| 21565 | return; |
| 21566 | } |
| 21567 | |
| 21568 | DEBUG_COMPILE_r({ |
| 21569 | if (!PL_colorset) |
| 21570 | reginitcolors(); |
| 21571 | { |
| 21572 | SV *dsv= sv_newmortal(); |
| 21573 | RE_PV_QUOTED_DECL(s, RX_UTF8(rx), |
| 21574 | dsv, RX_PRECOMP(rx), RX_PRELEN(rx), PL_dump_re_max_len); |
| 21575 | Perl_re_printf( aTHX_ "%sFreeing REx:%s %s\n", |
| 21576 | PL_colors[4], PL_colors[5], s); |
| 21577 | } |
| 21578 | }); |
| 21579 | |
| 21580 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 21581 | if (ri->u.offsets) |
| 21582 | Safefree(ri->u.offsets); /* 20010421 MJD */ |
| 21583 | #endif |
| 21584 | if (ri->code_blocks) |
| 21585 | S_free_codeblocks(aTHX_ ri->code_blocks); |
| 21586 | |
| 21587 | if (ri->data) { |
| 21588 | int n = ri->data->count; |
| 21589 | |
| 21590 | while (--n >= 0) { |
| 21591 | /* If you add a ->what type here, update the comment in regcomp.h */ |
| 21592 | switch (ri->data->what[n]) { |
| 21593 | case 'a': |
| 21594 | case 'r': |
| 21595 | case 's': |
| 21596 | case 'S': |
| 21597 | case 'u': |
| 21598 | SvREFCNT_dec(MUTABLE_SV(ri->data->data[n])); |
| 21599 | break; |
| 21600 | case 'f': |
| 21601 | Safefree(ri->data->data[n]); |
| 21602 | break; |
| 21603 | case 'l': |
| 21604 | case 'L': |
| 21605 | break; |
| 21606 | case 'T': |
| 21607 | { /* Aho Corasick add-on structure for a trie node. |
| 21608 | Used in stclass optimization only */ |
| 21609 | U32 refcount; |
| 21610 | reg_ac_data *aho=(reg_ac_data*)ri->data->data[n]; |
| 21611 | #ifdef USE_ITHREADS |
| 21612 | dVAR; |
| 21613 | #endif |
| 21614 | OP_REFCNT_LOCK; |
| 21615 | refcount = --aho->refcount; |
| 21616 | OP_REFCNT_UNLOCK; |
| 21617 | if ( !refcount ) { |
| 21618 | PerlMemShared_free(aho->states); |
| 21619 | PerlMemShared_free(aho->fail); |
| 21620 | /* do this last!!!! */ |
| 21621 | PerlMemShared_free(ri->data->data[n]); |
| 21622 | /* we should only ever get called once, so |
| 21623 | * assert as much, and also guard the free |
| 21624 | * which /might/ happen twice. At the least |
| 21625 | * it will make code anlyzers happy and it |
| 21626 | * doesn't cost much. - Yves */ |
| 21627 | assert(ri->regstclass); |
| 21628 | if (ri->regstclass) { |
| 21629 | PerlMemShared_free(ri->regstclass); |
| 21630 | ri->regstclass = 0; |
| 21631 | } |
| 21632 | } |
| 21633 | } |
| 21634 | break; |
| 21635 | case 't': |
| 21636 | { |
| 21637 | /* trie structure. */ |
| 21638 | U32 refcount; |
| 21639 | reg_trie_data *trie=(reg_trie_data*)ri->data->data[n]; |
| 21640 | #ifdef USE_ITHREADS |
| 21641 | dVAR; |
| 21642 | #endif |
| 21643 | OP_REFCNT_LOCK; |
| 21644 | refcount = --trie->refcount; |
| 21645 | OP_REFCNT_UNLOCK; |
| 21646 | if ( !refcount ) { |
| 21647 | PerlMemShared_free(trie->charmap); |
| 21648 | PerlMemShared_free(trie->states); |
| 21649 | PerlMemShared_free(trie->trans); |
| 21650 | if (trie->bitmap) |
| 21651 | PerlMemShared_free(trie->bitmap); |
| 21652 | if (trie->jump) |
| 21653 | PerlMemShared_free(trie->jump); |
| 21654 | PerlMemShared_free(trie->wordinfo); |
| 21655 | /* do this last!!!! */ |
| 21656 | PerlMemShared_free(ri->data->data[n]); |
| 21657 | } |
| 21658 | } |
| 21659 | break; |
| 21660 | default: |
| 21661 | Perl_croak(aTHX_ "panic: regfree data code '%c'", |
| 21662 | ri->data->what[n]); |
| 21663 | } |
| 21664 | } |
| 21665 | Safefree(ri->data->what); |
| 21666 | Safefree(ri->data); |
| 21667 | } |
| 21668 | |
| 21669 | Safefree(ri); |
| 21670 | } |
| 21671 | |
| 21672 | #define av_dup_inc(s, t) MUTABLE_AV(sv_dup_inc((const SV *)s, t)) |
| 21673 | #define hv_dup_inc(s, t) MUTABLE_HV(sv_dup_inc((const SV *)s, t)) |
| 21674 | #define SAVEPVN(p, n) ((p) ? savepvn(p, n) : NULL) |
| 21675 | |
| 21676 | /* |
| 21677 | re_dup_guts - duplicate a regexp. |
| 21678 | |
| 21679 | This routine is expected to clone a given regexp structure. It is only |
| 21680 | compiled under USE_ITHREADS. |
| 21681 | |
| 21682 | After all of the core data stored in struct regexp is duplicated |
| 21683 | the regexp_engine.dupe method is used to copy any private data |
| 21684 | stored in the *pprivate pointer. This allows extensions to handle |
| 21685 | any duplication it needs to do. |
| 21686 | |
| 21687 | See pregfree() and regfree_internal() if you change anything here. |
| 21688 | */ |
| 21689 | #if defined(USE_ITHREADS) |
| 21690 | #ifndef PERL_IN_XSUB_RE |
| 21691 | void |
| 21692 | Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param) |
| 21693 | { |
| 21694 | dVAR; |
| 21695 | I32 npar; |
| 21696 | const struct regexp *r = ReANY(sstr); |
| 21697 | struct regexp *ret = ReANY(dstr); |
| 21698 | |
| 21699 | PERL_ARGS_ASSERT_RE_DUP_GUTS; |
| 21700 | |
| 21701 | npar = r->nparens+1; |
| 21702 | Newx(ret->offs, npar, regexp_paren_pair); |
| 21703 | Copy(r->offs, ret->offs, npar, regexp_paren_pair); |
| 21704 | |
| 21705 | if (ret->substrs) { |
| 21706 | /* Do it this way to avoid reading from *r after the StructCopy(). |
| 21707 | That way, if any of the sv_dup_inc()s dislodge *r from the L1 |
| 21708 | cache, it doesn't matter. */ |
| 21709 | int i; |
| 21710 | const bool anchored = r->check_substr |
| 21711 | ? r->check_substr == r->substrs->data[0].substr |
| 21712 | : r->check_utf8 == r->substrs->data[0].utf8_substr; |
| 21713 | Newx(ret->substrs, 1, struct reg_substr_data); |
| 21714 | StructCopy(r->substrs, ret->substrs, struct reg_substr_data); |
| 21715 | |
| 21716 | for (i = 0; i < 2; i++) { |
| 21717 | ret->substrs->data[i].substr = |
| 21718 | sv_dup_inc(ret->substrs->data[i].substr, param); |
| 21719 | ret->substrs->data[i].utf8_substr = |
| 21720 | sv_dup_inc(ret->substrs->data[i].utf8_substr, param); |
| 21721 | } |
| 21722 | |
| 21723 | /* check_substr and check_utf8, if non-NULL, point to either their |
| 21724 | anchored or float namesakes, and don't hold a second reference. */ |
| 21725 | |
| 21726 | if (ret->check_substr) { |
| 21727 | if (anchored) { |
| 21728 | assert(r->check_utf8 == r->substrs->data[0].utf8_substr); |
| 21729 | |
| 21730 | ret->check_substr = ret->substrs->data[0].substr; |
| 21731 | ret->check_utf8 = ret->substrs->data[0].utf8_substr; |
| 21732 | } else { |
| 21733 | assert(r->check_substr == r->substrs->data[1].substr); |
| 21734 | assert(r->check_utf8 == r->substrs->data[1].utf8_substr); |
| 21735 | |
| 21736 | ret->check_substr = ret->substrs->data[1].substr; |
| 21737 | ret->check_utf8 = ret->substrs->data[1].utf8_substr; |
| 21738 | } |
| 21739 | } else if (ret->check_utf8) { |
| 21740 | if (anchored) { |
| 21741 | ret->check_utf8 = ret->substrs->data[0].utf8_substr; |
| 21742 | } else { |
| 21743 | ret->check_utf8 = ret->substrs->data[1].utf8_substr; |
| 21744 | } |
| 21745 | } |
| 21746 | } |
| 21747 | |
| 21748 | RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param); |
| 21749 | ret->qr_anoncv = MUTABLE_CV(sv_dup_inc((const SV *)ret->qr_anoncv, param)); |
| 21750 | if (r->recurse_locinput) |
| 21751 | Newx(ret->recurse_locinput, r->nparens + 1, char *); |
| 21752 | |
| 21753 | if (ret->pprivate) |
| 21754 | RXi_SET(ret, CALLREGDUPE_PVT(dstr, param)); |
| 21755 | |
| 21756 | if (RX_MATCH_COPIED(dstr)) |
| 21757 | ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen); |
| 21758 | else |
| 21759 | ret->subbeg = NULL; |
| 21760 | #ifdef PERL_ANY_COW |
| 21761 | ret->saved_copy = NULL; |
| 21762 | #endif |
| 21763 | |
| 21764 | /* Whether mother_re be set or no, we need to copy the string. We |
| 21765 | cannot refrain from copying it when the storage points directly to |
| 21766 | our mother regexp, because that's |
| 21767 | 1: a buffer in a different thread |
| 21768 | 2: something we no longer hold a reference on |
| 21769 | so we need to copy it locally. */ |
| 21770 | RX_WRAPPED(dstr) = SAVEPVN(RX_WRAPPED_const(sstr), SvCUR(sstr)+1); |
| 21771 | /* set malloced length to a non-zero value so it will be freed |
| 21772 | * (otherwise in combination with SVf_FAKE it looks like an alien |
| 21773 | * buffer). It doesn't have to be the actual malloced size, since it |
| 21774 | * should never be grown */ |
| 21775 | SvLEN_set(dstr, SvCUR(sstr)+1); |
| 21776 | ret->mother_re = NULL; |
| 21777 | } |
| 21778 | #endif /* PERL_IN_XSUB_RE */ |
| 21779 | |
| 21780 | /* |
| 21781 | regdupe_internal() |
| 21782 | |
| 21783 | This is the internal complement to regdupe() which is used to copy |
| 21784 | the structure pointed to by the *pprivate pointer in the regexp. |
| 21785 | This is the core version of the extension overridable cloning hook. |
| 21786 | The regexp structure being duplicated will be copied by perl prior |
| 21787 | to this and will be provided as the regexp *r argument, however |
| 21788 | with the /old/ structures pprivate pointer value. Thus this routine |
| 21789 | may override any copying normally done by perl. |
| 21790 | |
| 21791 | It returns a pointer to the new regexp_internal structure. |
| 21792 | */ |
| 21793 | |
| 21794 | void * |
| 21795 | Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param) |
| 21796 | { |
| 21797 | dVAR; |
| 21798 | struct regexp *const r = ReANY(rx); |
| 21799 | regexp_internal *reti; |
| 21800 | int len; |
| 21801 | RXi_GET_DECL(r, ri); |
| 21802 | |
| 21803 | PERL_ARGS_ASSERT_REGDUPE_INTERNAL; |
| 21804 | |
| 21805 | len = ProgLen(ri); |
| 21806 | |
| 21807 | Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), |
| 21808 | char, regexp_internal); |
| 21809 | Copy(ri->program, reti->program, len+1, regnode); |
| 21810 | |
| 21811 | |
| 21812 | if (ri->code_blocks) { |
| 21813 | int n; |
| 21814 | Newx(reti->code_blocks, 1, struct reg_code_blocks); |
| 21815 | Newx(reti->code_blocks->cb, ri->code_blocks->count, |
| 21816 | struct reg_code_block); |
| 21817 | Copy(ri->code_blocks->cb, reti->code_blocks->cb, |
| 21818 | ri->code_blocks->count, struct reg_code_block); |
| 21819 | for (n = 0; n < ri->code_blocks->count; n++) |
| 21820 | reti->code_blocks->cb[n].src_regex = (REGEXP*) |
| 21821 | sv_dup_inc((SV*)(ri->code_blocks->cb[n].src_regex), param); |
| 21822 | reti->code_blocks->count = ri->code_blocks->count; |
| 21823 | reti->code_blocks->refcnt = 1; |
| 21824 | } |
| 21825 | else |
| 21826 | reti->code_blocks = NULL; |
| 21827 | |
| 21828 | reti->regstclass = NULL; |
| 21829 | |
| 21830 | if (ri->data) { |
| 21831 | struct reg_data *d; |
| 21832 | const int count = ri->data->count; |
| 21833 | int i; |
| 21834 | |
| 21835 | Newxc(d, sizeof(struct reg_data) + count*sizeof(void *), |
| 21836 | char, struct reg_data); |
| 21837 | Newx(d->what, count, U8); |
| 21838 | |
| 21839 | d->count = count; |
| 21840 | for (i = 0; i < count; i++) { |
| 21841 | d->what[i] = ri->data->what[i]; |
| 21842 | switch (d->what[i]) { |
| 21843 | /* see also regcomp.h and regfree_internal() */ |
| 21844 | case 'a': /* actually an AV, but the dup function is identical. |
| 21845 | values seem to be "plain sv's" generally. */ |
| 21846 | case 'r': /* a compiled regex (but still just another SV) */ |
| 21847 | case 's': /* an RV (currently only used for an RV to an AV by the ANYOF code) |
| 21848 | this use case should go away, the code could have used |
| 21849 | 'a' instead - see S_set_ANYOF_arg() for array contents. */ |
| 21850 | case 'S': /* actually an SV, but the dup function is identical. */ |
| 21851 | case 'u': /* actually an HV, but the dup function is identical. |
| 21852 | values are "plain sv's" */ |
| 21853 | d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param); |
| 21854 | break; |
| 21855 | case 'f': |
| 21856 | /* Synthetic Start Class - "Fake" charclass we generate to optimize |
| 21857 | * patterns which could start with several different things. Pre-TRIE |
| 21858 | * this was more important than it is now, however this still helps |
| 21859 | * in some places, for instance /x?a+/ might produce a SSC equivalent |
| 21860 | * to [xa]. This is used by Perl_re_intuit_start() and S_find_byclass() |
| 21861 | * in regexec.c |
| 21862 | */ |
| 21863 | /* This is cheating. */ |
| 21864 | Newx(d->data[i], 1, regnode_ssc); |
| 21865 | StructCopy(ri->data->data[i], d->data[i], regnode_ssc); |
| 21866 | reti->regstclass = (regnode*)d->data[i]; |
| 21867 | break; |
| 21868 | case 'T': |
| 21869 | /* AHO-CORASICK fail table */ |
| 21870 | /* Trie stclasses are readonly and can thus be shared |
| 21871 | * without duplication. We free the stclass in pregfree |
| 21872 | * when the corresponding reg_ac_data struct is freed. |
| 21873 | */ |
| 21874 | reti->regstclass= ri->regstclass; |
| 21875 | /* FALLTHROUGH */ |
| 21876 | case 't': |
| 21877 | /* TRIE transition table */ |
| 21878 | OP_REFCNT_LOCK; |
| 21879 | ((reg_trie_data*)ri->data->data[i])->refcount++; |
| 21880 | OP_REFCNT_UNLOCK; |
| 21881 | /* FALLTHROUGH */ |
| 21882 | case 'l': /* (?{...}) or (??{ ... }) code (cb->block) */ |
| 21883 | case 'L': /* same when RExC_pm_flags & PMf_HAS_CV and code |
| 21884 | is not from another regexp */ |
| 21885 | d->data[i] = ri->data->data[i]; |
| 21886 | break; |
| 21887 | default: |
| 21888 | Perl_croak(aTHX_ "panic: re_dup_guts unknown data code '%c'", |
| 21889 | ri->data->what[i]); |
| 21890 | } |
| 21891 | } |
| 21892 | |
| 21893 | reti->data = d; |
| 21894 | } |
| 21895 | else |
| 21896 | reti->data = NULL; |
| 21897 | |
| 21898 | reti->name_list_idx = ri->name_list_idx; |
| 21899 | |
| 21900 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 21901 | if (ri->u.offsets) { |
| 21902 | Newx(reti->u.offsets, 2*len+1, U32); |
| 21903 | Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32); |
| 21904 | } |
| 21905 | #else |
| 21906 | SetProgLen(reti, len); |
| 21907 | #endif |
| 21908 | |
| 21909 | return (void*)reti; |
| 21910 | } |
| 21911 | |
| 21912 | #endif /* USE_ITHREADS */ |
| 21913 | |
| 21914 | #ifndef PERL_IN_XSUB_RE |
| 21915 | |
| 21916 | /* |
| 21917 | - regnext - dig the "next" pointer out of a node |
| 21918 | */ |
| 21919 | regnode * |
| 21920 | Perl_regnext(pTHX_ regnode *p) |
| 21921 | { |
| 21922 | I32 offset; |
| 21923 | |
| 21924 | if (!p) |
| 21925 | return(NULL); |
| 21926 | |
| 21927 | if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */ |
| 21928 | Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", |
| 21929 | (int)OP(p), (int)REGNODE_MAX); |
| 21930 | } |
| 21931 | |
| 21932 | offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p)); |
| 21933 | if (offset == 0) |
| 21934 | return(NULL); |
| 21935 | |
| 21936 | return(p+offset); |
| 21937 | } |
| 21938 | |
| 21939 | #endif |
| 21940 | |
| 21941 | STATIC void |
| 21942 | S_re_croak(pTHX_ bool utf8, const char* pat,...) |
| 21943 | { |
| 21944 | va_list args; |
| 21945 | STRLEN len = strlen(pat); |
| 21946 | char buf[512]; |
| 21947 | SV *msv; |
| 21948 | const char *message; |
| 21949 | |
| 21950 | PERL_ARGS_ASSERT_RE_CROAK; |
| 21951 | |
| 21952 | if (len > 510) |
| 21953 | len = 510; |
| 21954 | Copy(pat, buf, len , char); |
| 21955 | buf[len] = '\n'; |
| 21956 | buf[len + 1] = '\0'; |
| 21957 | va_start(args, pat); |
| 21958 | msv = vmess(buf, &args); |
| 21959 | va_end(args); |
| 21960 | message = SvPV_const(msv, len); |
| 21961 | if (len > 512) |
| 21962 | len = 512; |
| 21963 | Copy(message, buf, len , char); |
| 21964 | /* len-1 to avoid \n */ |
| 21965 | Perl_croak(aTHX_ "%" UTF8f, UTF8fARG(utf8, len-1, buf)); |
| 21966 | } |
| 21967 | |
| 21968 | /* XXX Here's a total kludge. But we need to re-enter for swash routines. */ |
| 21969 | |
| 21970 | #ifndef PERL_IN_XSUB_RE |
| 21971 | void |
| 21972 | Perl_save_re_context(pTHX) |
| 21973 | { |
| 21974 | I32 nparens = -1; |
| 21975 | I32 i; |
| 21976 | |
| 21977 | /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */ |
| 21978 | |
| 21979 | if (PL_curpm) { |
| 21980 | const REGEXP * const rx = PM_GETRE(PL_curpm); |
| 21981 | if (rx) |
| 21982 | nparens = RX_NPARENS(rx); |
| 21983 | } |
| 21984 | |
| 21985 | /* RT #124109. This is a complete hack; in the SWASHNEW case we know |
| 21986 | * that PL_curpm will be null, but that utf8.pm and the modules it |
| 21987 | * loads will only use $1..$3. |
| 21988 | * The t/porting/re_context.t test file checks this assumption. |
| 21989 | */ |
| 21990 | if (nparens == -1) |
| 21991 | nparens = 3; |
| 21992 | |
| 21993 | for (i = 1; i <= nparens; i++) { |
| 21994 | char digits[TYPE_CHARS(long)]; |
| 21995 | const STRLEN len = my_snprintf(digits, sizeof(digits), |
| 21996 | "%lu", (long)i); |
| 21997 | GV *const *const gvp |
| 21998 | = (GV**)hv_fetch(PL_defstash, digits, len, 0); |
| 21999 | |
| 22000 | if (gvp) { |
| 22001 | GV * const gv = *gvp; |
| 22002 | if (SvTYPE(gv) == SVt_PVGV && GvSV(gv)) |
| 22003 | save_scalar(gv); |
| 22004 | } |
| 22005 | } |
| 22006 | } |
| 22007 | #endif |
| 22008 | |
| 22009 | #ifdef DEBUGGING |
| 22010 | |
| 22011 | STATIC void |
| 22012 | S_put_code_point(pTHX_ SV *sv, UV c) |
| 22013 | { |
| 22014 | PERL_ARGS_ASSERT_PUT_CODE_POINT; |
| 22015 | |
| 22016 | if (c > 255) { |
| 22017 | Perl_sv_catpvf(aTHX_ sv, "\\x{%04" UVXf "}", c); |
| 22018 | } |
| 22019 | else if (isPRINT(c)) { |
| 22020 | const char string = (char) c; |
| 22021 | |
| 22022 | /* We use {phrase} as metanotation in the class, so also escape literal |
| 22023 | * braces */ |
| 22024 | if (isBACKSLASHED_PUNCT(c) || c == '{' || c == '}') |
| 22025 | sv_catpvs(sv, "\\"); |
| 22026 | sv_catpvn(sv, &string, 1); |
| 22027 | } |
| 22028 | else if (isMNEMONIC_CNTRL(c)) { |
| 22029 | Perl_sv_catpvf(aTHX_ sv, "%s", cntrl_to_mnemonic((U8) c)); |
| 22030 | } |
| 22031 | else { |
| 22032 | Perl_sv_catpvf(aTHX_ sv, "\\x%02X", (U8) c); |
| 22033 | } |
| 22034 | } |
| 22035 | |
| 22036 | #define MAX_PRINT_A MAX_PRINT_A_FOR_USE_ONLY_BY_REGCOMP_DOT_C |
| 22037 | |
| 22038 | STATIC void |
| 22039 | S_put_range(pTHX_ SV *sv, UV start, const UV end, const bool allow_literals) |
| 22040 | { |
| 22041 | /* Appends to 'sv' a displayable version of the range of code points from |
| 22042 | * 'start' to 'end'. Mnemonics (like '\r') are used for the few controls |
| 22043 | * that have them, when they occur at the beginning or end of the range. |
| 22044 | * It uses hex to output the remaining code points, unless 'allow_literals' |
| 22045 | * is true, in which case the printable ASCII ones are output as-is (though |
| 22046 | * some of these will be escaped by put_code_point()). |
| 22047 | * |
| 22048 | * NOTE: This is designed only for printing ranges of code points that fit |
| 22049 | * inside an ANYOF bitmap. Higher code points are simply suppressed |
| 22050 | */ |
| 22051 | |
| 22052 | const unsigned int min_range_count = 3; |
| 22053 | |
| 22054 | assert(start <= end); |
| 22055 | |
| 22056 | PERL_ARGS_ASSERT_PUT_RANGE; |
| 22057 | |
| 22058 | while (start <= end) { |
| 22059 | UV this_end; |
| 22060 | const char * format; |
| 22061 | |
| 22062 | if (end - start < min_range_count) { |
| 22063 | |
| 22064 | /* Output chars individually when they occur in short ranges */ |
| 22065 | for (; start <= end; start++) { |
| 22066 | put_code_point(sv, start); |
| 22067 | } |
| 22068 | break; |
| 22069 | } |
| 22070 | |
| 22071 | /* If permitted by the input options, and there is a possibility that |
| 22072 | * this range contains a printable literal, look to see if there is |
| 22073 | * one. */ |
| 22074 | if (allow_literals && start <= MAX_PRINT_A) { |
| 22075 | |
| 22076 | /* If the character at the beginning of the range isn't an ASCII |
| 22077 | * printable, effectively split the range into two parts: |
| 22078 | * 1) the portion before the first such printable, |
| 22079 | * 2) the rest |
| 22080 | * and output them separately. */ |
| 22081 | if (! isPRINT_A(start)) { |
| 22082 | UV temp_end = start + 1; |
| 22083 | |
| 22084 | /* There is no point looking beyond the final possible |
| 22085 | * printable, in MAX_PRINT_A */ |
| 22086 | UV max = MIN(end, MAX_PRINT_A); |
| 22087 | |
| 22088 | while (temp_end <= max && ! isPRINT_A(temp_end)) { |
| 22089 | temp_end++; |
| 22090 | } |
| 22091 | |
| 22092 | /* Here, temp_end points to one beyond the first printable if |
| 22093 | * found, or to one beyond 'max' if not. If none found, make |
| 22094 | * sure that we use the entire range */ |
| 22095 | if (temp_end > MAX_PRINT_A) { |
| 22096 | temp_end = end + 1; |
| 22097 | } |
| 22098 | |
| 22099 | /* Output the first part of the split range: the part that |
| 22100 | * doesn't have printables, with the parameter set to not look |
| 22101 | * for literals (otherwise we would infinitely recurse) */ |
| 22102 | put_range(sv, start, temp_end - 1, FALSE); |
| 22103 | |
| 22104 | /* The 2nd part of the range (if any) starts here. */ |
| 22105 | start = temp_end; |
| 22106 | |
| 22107 | /* We do a continue, instead of dropping down, because even if |
| 22108 | * the 2nd part is non-empty, it could be so short that we want |
| 22109 | * to output it as individual characters, as tested for at the |
| 22110 | * top of this loop. */ |
| 22111 | continue; |
| 22112 | } |
| 22113 | |
| 22114 | /* Here, 'start' is a printable ASCII. If it is an alphanumeric, |
| 22115 | * output a sub-range of just the digits or letters, then process |
| 22116 | * the remaining portion as usual. */ |
| 22117 | if (isALPHANUMERIC_A(start)) { |
| 22118 | UV mask = (isDIGIT_A(start)) |
| 22119 | ? _CC_DIGIT |
| 22120 | : isUPPER_A(start) |
| 22121 | ? _CC_UPPER |
| 22122 | : _CC_LOWER; |
| 22123 | UV temp_end = start + 1; |
| 22124 | |
| 22125 | /* Find the end of the sub-range that includes just the |
| 22126 | * characters in the same class as the first character in it */ |
| 22127 | while (temp_end <= end && _generic_isCC_A(temp_end, mask)) { |
| 22128 | temp_end++; |
| 22129 | } |
| 22130 | temp_end--; |
| 22131 | |
| 22132 | /* For short ranges, don't duplicate the code above to output |
| 22133 | * them; just call recursively */ |
| 22134 | if (temp_end - start < min_range_count) { |
| 22135 | put_range(sv, start, temp_end, FALSE); |
| 22136 | } |
| 22137 | else { /* Output as a range */ |
| 22138 | put_code_point(sv, start); |
| 22139 | sv_catpvs(sv, "-"); |
| 22140 | put_code_point(sv, temp_end); |
| 22141 | } |
| 22142 | start = temp_end + 1; |
| 22143 | continue; |
| 22144 | } |
| 22145 | |
| 22146 | /* We output any other printables as individual characters */ |
| 22147 | if (isPUNCT_A(start) || isSPACE_A(start)) { |
| 22148 | while (start <= end && (isPUNCT_A(start) |
| 22149 | || isSPACE_A(start))) |
| 22150 | { |
| 22151 | put_code_point(sv, start); |
| 22152 | start++; |
| 22153 | } |
| 22154 | continue; |
| 22155 | } |
| 22156 | } /* End of looking for literals */ |
| 22157 | |
| 22158 | /* Here is not to output as a literal. Some control characters have |
| 22159 | * mnemonic names. Split off any of those at the beginning and end of |
| 22160 | * the range to print mnemonically. It isn't possible for many of |
| 22161 | * these to be in a row, so this won't overwhelm with output */ |
| 22162 | if ( start <= end |
| 22163 | && (isMNEMONIC_CNTRL(start) || isMNEMONIC_CNTRL(end))) |
| 22164 | { |
| 22165 | while (isMNEMONIC_CNTRL(start) && start <= end) { |
| 22166 | put_code_point(sv, start); |
| 22167 | start++; |
| 22168 | } |
| 22169 | |
| 22170 | /* If this didn't take care of the whole range ... */ |
| 22171 | if (start <= end) { |
| 22172 | |
| 22173 | /* Look backwards from the end to find the final non-mnemonic |
| 22174 | * */ |
| 22175 | UV temp_end = end; |
| 22176 | while (isMNEMONIC_CNTRL(temp_end)) { |
| 22177 | temp_end--; |
| 22178 | } |
| 22179 | |
| 22180 | /* And separately output the interior range that doesn't start |
| 22181 | * or end with mnemonics */ |
| 22182 | put_range(sv, start, temp_end, FALSE); |
| 22183 | |
| 22184 | /* Then output the mnemonic trailing controls */ |
| 22185 | start = temp_end + 1; |
| 22186 | while (start <= end) { |
| 22187 | put_code_point(sv, start); |
| 22188 | start++; |
| 22189 | } |
| 22190 | break; |
| 22191 | } |
| 22192 | } |
| 22193 | |
| 22194 | /* As a final resort, output the range or subrange as hex. */ |
| 22195 | |
| 22196 | if (start >= NUM_ANYOF_CODE_POINTS) { |
| 22197 | this_end = end; |
| 22198 | } |
| 22199 | else { /* Have to split range at the bitmap boundary */ |
| 22200 | this_end = (end < NUM_ANYOF_CODE_POINTS) |
| 22201 | ? end |
| 22202 | : NUM_ANYOF_CODE_POINTS - 1; |
| 22203 | } |
| 22204 | #if NUM_ANYOF_CODE_POINTS > 256 |
| 22205 | format = (this_end < 256) |
| 22206 | ? "\\x%02" UVXf "-\\x%02" UVXf |
| 22207 | : "\\x{%04" UVXf "}-\\x{%04" UVXf "}"; |
| 22208 | #else |
| 22209 | format = "\\x%02" UVXf "-\\x%02" UVXf; |
| 22210 | #endif |
| 22211 | GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral); |
| 22212 | Perl_sv_catpvf(aTHX_ sv, format, start, this_end); |
| 22213 | GCC_DIAG_RESTORE_STMT; |
| 22214 | break; |
| 22215 | } |
| 22216 | } |
| 22217 | |
| 22218 | STATIC void |
| 22219 | S_put_charclass_bitmap_innards_invlist(pTHX_ SV *sv, SV* invlist) |
| 22220 | { |
| 22221 | /* Concatenate onto the PV in 'sv' a displayable form of the inversion list |
| 22222 | * 'invlist' */ |
| 22223 | |
| 22224 | UV start, end; |
| 22225 | bool allow_literals = TRUE; |
| 22226 | |
| 22227 | PERL_ARGS_ASSERT_PUT_CHARCLASS_BITMAP_INNARDS_INVLIST; |
| 22228 | |
| 22229 | /* Generally, it is more readable if printable characters are output as |
| 22230 | * literals, but if a range (nearly) spans all of them, it's best to output |
| 22231 | * it as a single range. This code will use a single range if all but 2 |
| 22232 | * ASCII printables are in it */ |
| 22233 | invlist_iterinit(invlist); |
| 22234 | while (invlist_iternext(invlist, &start, &end)) { |
| 22235 | |
| 22236 | /* If the range starts beyond the final printable, it doesn't have any |
| 22237 | * in it */ |
| 22238 | if (start > MAX_PRINT_A) { |
| 22239 | break; |
| 22240 | } |
| 22241 | |
| 22242 | /* In both ASCII and EBCDIC, a SPACE is the lowest printable. To span |
| 22243 | * all but two, the range must start and end no later than 2 from |
| 22244 | * either end */ |
| 22245 | if (start < ' ' + 2 && end > MAX_PRINT_A - 2) { |
| 22246 | if (end > MAX_PRINT_A) { |
| 22247 | end = MAX_PRINT_A; |
| 22248 | } |
| 22249 | if (start < ' ') { |
| 22250 | start = ' '; |
| 22251 | } |
| 22252 | if (end - start >= MAX_PRINT_A - ' ' - 2) { |
| 22253 | allow_literals = FALSE; |
| 22254 | } |
| 22255 | break; |
| 22256 | } |
| 22257 | } |
| 22258 | invlist_iterfinish(invlist); |
| 22259 | |
| 22260 | /* Here we have figured things out. Output each range */ |
| 22261 | invlist_iterinit(invlist); |
| 22262 | while (invlist_iternext(invlist, &start, &end)) { |
| 22263 | if (start >= NUM_ANYOF_CODE_POINTS) { |
| 22264 | break; |
| 22265 | } |
| 22266 | put_range(sv, start, end, allow_literals); |
| 22267 | } |
| 22268 | invlist_iterfinish(invlist); |
| 22269 | |
| 22270 | return; |
| 22271 | } |
| 22272 | |
| 22273 | STATIC SV* |
| 22274 | S_put_charclass_bitmap_innards_common(pTHX_ |
| 22275 | SV* invlist, /* The bitmap */ |
| 22276 | SV* posixes, /* Under /l, things like [:word:], \S */ |
| 22277 | SV* only_utf8, /* Under /d, matches iff the target is UTF-8 */ |
| 22278 | SV* not_utf8, /* /d, matches iff the target isn't UTF-8 */ |
| 22279 | SV* only_utf8_locale, /* Under /l, matches if the locale is UTF-8 */ |
| 22280 | const bool invert /* Is the result to be inverted? */ |
| 22281 | ) |
| 22282 | { |
| 22283 | /* Create and return an SV containing a displayable version of the bitmap |
| 22284 | * and associated information determined by the input parameters. If the |
| 22285 | * output would have been only the inversion indicator '^', NULL is instead |
| 22286 | * returned. */ |
| 22287 | |
| 22288 | dVAR; |
| 22289 | SV * output; |
| 22290 | |
| 22291 | PERL_ARGS_ASSERT_PUT_CHARCLASS_BITMAP_INNARDS_COMMON; |
| 22292 | |
| 22293 | if (invert) { |
| 22294 | output = newSVpvs("^"); |
| 22295 | } |
| 22296 | else { |
| 22297 | output = newSVpvs(""); |
| 22298 | } |
| 22299 | |
| 22300 | /* First, the code points in the bitmap that are unconditionally there */ |
| 22301 | put_charclass_bitmap_innards_invlist(output, invlist); |
| 22302 | |
| 22303 | /* Traditionally, these have been placed after the main code points */ |
| 22304 | if (posixes) { |
| 22305 | sv_catsv(output, posixes); |
| 22306 | } |
| 22307 | |
| 22308 | if (only_utf8 && _invlist_len(only_utf8)) { |
| 22309 | Perl_sv_catpvf(aTHX_ output, "%s{utf8}%s", PL_colors[1], PL_colors[0]); |
| 22310 | put_charclass_bitmap_innards_invlist(output, only_utf8); |
| 22311 | } |
| 22312 | |
| 22313 | if (not_utf8 && _invlist_len(not_utf8)) { |
| 22314 | Perl_sv_catpvf(aTHX_ output, "%s{not utf8}%s", PL_colors[1], PL_colors[0]); |
| 22315 | put_charclass_bitmap_innards_invlist(output, not_utf8); |
| 22316 | } |
| 22317 | |
| 22318 | if (only_utf8_locale && _invlist_len(only_utf8_locale)) { |
| 22319 | Perl_sv_catpvf(aTHX_ output, "%s{utf8 locale}%s", PL_colors[1], PL_colors[0]); |
| 22320 | put_charclass_bitmap_innards_invlist(output, only_utf8_locale); |
| 22321 | |
| 22322 | /* This is the only list in this routine that can legally contain code |
| 22323 | * points outside the bitmap range. The call just above to |
| 22324 | * 'put_charclass_bitmap_innards_invlist' will simply suppress them, so |
| 22325 | * output them here. There's about a half-dozen possible, and none in |
| 22326 | * contiguous ranges longer than 2 */ |
| 22327 | if (invlist_highest(only_utf8_locale) >= NUM_ANYOF_CODE_POINTS) { |
| 22328 | UV start, end; |
| 22329 | SV* above_bitmap = NULL; |
| 22330 | |
| 22331 | _invlist_subtract(only_utf8_locale, PL_InBitmap, &above_bitmap); |
| 22332 | |
| 22333 | invlist_iterinit(above_bitmap); |
| 22334 | while (invlist_iternext(above_bitmap, &start, &end)) { |
| 22335 | UV i; |
| 22336 | |
| 22337 | for (i = start; i <= end; i++) { |
| 22338 | put_code_point(output, i); |
| 22339 | } |
| 22340 | } |
| 22341 | invlist_iterfinish(above_bitmap); |
| 22342 | SvREFCNT_dec_NN(above_bitmap); |
| 22343 | } |
| 22344 | } |
| 22345 | |
| 22346 | if (invert && SvCUR(output) == 1) { |
| 22347 | return NULL; |
| 22348 | } |
| 22349 | |
| 22350 | return output; |
| 22351 | } |
| 22352 | |
| 22353 | STATIC bool |
| 22354 | S_put_charclass_bitmap_innards(pTHX_ SV *sv, |
| 22355 | char *bitmap, |
| 22356 | SV *nonbitmap_invlist, |
| 22357 | SV *only_utf8_locale_invlist, |
| 22358 | const regnode * const node, |
| 22359 | const U8 flags, |
| 22360 | const bool force_as_is_display) |
| 22361 | { |
| 22362 | /* Appends to 'sv' a displayable version of the innards of the bracketed |
| 22363 | * character class defined by the other arguments: |
| 22364 | * 'bitmap' points to the bitmap, or NULL if to ignore that. |
| 22365 | * 'nonbitmap_invlist' is an inversion list of the code points that are in |
| 22366 | * the bitmap range, but for some reason aren't in the bitmap; NULL if |
| 22367 | * none. The reasons for this could be that they require some |
| 22368 | * condition such as the target string being or not being in UTF-8 |
| 22369 | * (under /d), or because they came from a user-defined property that |
| 22370 | * was not resolved at the time of the regex compilation (under /u) |
| 22371 | * 'only_utf8_locale_invlist' is an inversion list of the code points that |
| 22372 | * are valid only if the runtime locale is a UTF-8 one; NULL if none |
| 22373 | * 'node' is the regex pattern ANYOF node. It is needed only when the |
| 22374 | * above two parameters are not null, and is passed so that this |
| 22375 | * routine can tease apart the various reasons for them. |
| 22376 | * 'flags' is the flags field of 'node' |
| 22377 | * 'force_as_is_display' is TRUE if this routine should definitely NOT try |
| 22378 | * to invert things to see if that leads to a cleaner display. If |
| 22379 | * FALSE, this routine is free to use its judgment about doing this. |
| 22380 | * |
| 22381 | * It returns TRUE if there was actually something output. (It may be that |
| 22382 | * the bitmap, etc is empty.) |
| 22383 | * |
| 22384 | * When called for outputting the bitmap of a non-ANYOF node, just pass the |
| 22385 | * bitmap, with the succeeding parameters set to NULL, and the final one to |
| 22386 | * FALSE. |
| 22387 | */ |
| 22388 | |
| 22389 | /* In general, it tries to display the 'cleanest' representation of the |
| 22390 | * innards, choosing whether to display them inverted or not, regardless of |
| 22391 | * whether the class itself is to be inverted. However, there are some |
| 22392 | * cases where it can't try inverting, as what actually matches isn't known |
| 22393 | * until runtime, and hence the inversion isn't either. */ |
| 22394 | |
| 22395 | dVAR; |
| 22396 | bool inverting_allowed = ! force_as_is_display; |
| 22397 | |
| 22398 | int i; |
| 22399 | STRLEN orig_sv_cur = SvCUR(sv); |
| 22400 | |
| 22401 | SV* invlist; /* Inversion list we accumulate of code points that |
| 22402 | are unconditionally matched */ |
| 22403 | SV* only_utf8 = NULL; /* Under /d, list of matches iff the target is |
| 22404 | UTF-8 */ |
| 22405 | SV* not_utf8 = NULL; /* /d, list of matches iff the target isn't UTF-8 |
| 22406 | */ |
| 22407 | SV* posixes = NULL; /* Under /l, string of things like [:word:], \D */ |
| 22408 | SV* only_utf8_locale = NULL; /* Under /l, list of matches if the locale |
| 22409 | is UTF-8 */ |
| 22410 | |
| 22411 | SV* as_is_display; /* The output string when we take the inputs |
| 22412 | literally */ |
| 22413 | SV* inverted_display; /* The output string when we invert the inputs */ |
| 22414 | |
| 22415 | bool invert = cBOOL(flags & ANYOF_INVERT); /* Is the input to be inverted |
| 22416 | to match? */ |
| 22417 | /* We are biased in favor of displaying things without them being inverted, |
| 22418 | * as that is generally easier to understand */ |
| 22419 | const int bias = 5; |
| 22420 | |
| 22421 | PERL_ARGS_ASSERT_PUT_CHARCLASS_BITMAP_INNARDS; |
| 22422 | |
| 22423 | /* Start off with whatever code points are passed in. (We clone, so we |
| 22424 | * don't change the caller's list) */ |
| 22425 | if (nonbitmap_invlist) { |
| 22426 | assert(invlist_highest(nonbitmap_invlist) < NUM_ANYOF_CODE_POINTS); |
| 22427 | invlist = invlist_clone(nonbitmap_invlist, NULL); |
| 22428 | } |
| 22429 | else { /* Worst case size is every other code point is matched */ |
| 22430 | invlist = _new_invlist(NUM_ANYOF_CODE_POINTS / 2); |
| 22431 | } |
| 22432 | |
| 22433 | if (flags) { |
| 22434 | if (OP(node) == ANYOFD) { |
| 22435 | |
| 22436 | /* This flag indicates that the code points below 0x100 in the |
| 22437 | * nonbitmap list are precisely the ones that match only when the |
| 22438 | * target is UTF-8 (they should all be non-ASCII). */ |
| 22439 | if (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP) |
| 22440 | { |
| 22441 | _invlist_intersection(invlist, PL_UpperLatin1, &only_utf8); |
| 22442 | _invlist_subtract(invlist, only_utf8, &invlist); |
| 22443 | } |
| 22444 | |
| 22445 | /* And this flag for matching all non-ASCII 0xFF and below */ |
| 22446 | if (flags & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER) |
| 22447 | { |
| 22448 | not_utf8 = invlist_clone(PL_UpperLatin1, NULL); |
| 22449 | } |
| 22450 | } |
| 22451 | else if (OP(node) == ANYOFL || OP(node) == ANYOFPOSIXL) { |
| 22452 | |
| 22453 | /* If either of these flags are set, what matches isn't |
| 22454 | * determinable except during execution, so don't know enough here |
| 22455 | * to invert */ |
| 22456 | if (flags & (ANYOFL_FOLD|ANYOF_MATCHES_POSIXL)) { |
| 22457 | inverting_allowed = FALSE; |
| 22458 | } |
| 22459 | |
| 22460 | /* What the posix classes match also varies at runtime, so these |
| 22461 | * will be output symbolically. */ |
| 22462 | if (ANYOF_POSIXL_TEST_ANY_SET(node)) { |
| 22463 | int i; |
| 22464 | |
| 22465 | posixes = newSVpvs(""); |
| 22466 | for (i = 0; i < ANYOF_POSIXL_MAX; i++) { |
| 22467 | if (ANYOF_POSIXL_TEST(node, i)) { |
| 22468 | sv_catpv(posixes, anyofs[i]); |
| 22469 | } |
| 22470 | } |
| 22471 | } |
| 22472 | } |
| 22473 | } |
| 22474 | |
| 22475 | /* Accumulate the bit map into the unconditional match list */ |
| 22476 | if (bitmap) { |
| 22477 | for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) { |
| 22478 | if (BITMAP_TEST(bitmap, i)) { |
| 22479 | int start = i++; |
| 22480 | for (; |
| 22481 | i < NUM_ANYOF_CODE_POINTS && BITMAP_TEST(bitmap, i); |
| 22482 | i++) |
| 22483 | { /* empty */ } |
| 22484 | invlist = _add_range_to_invlist(invlist, start, i-1); |
| 22485 | } |
| 22486 | } |
| 22487 | } |
| 22488 | |
| 22489 | /* Make sure that the conditional match lists don't have anything in them |
| 22490 | * that match unconditionally; otherwise the output is quite confusing. |
| 22491 | * This could happen if the code that populates these misses some |
| 22492 | * duplication. */ |
| 22493 | if (only_utf8) { |
| 22494 | _invlist_subtract(only_utf8, invlist, &only_utf8); |
| 22495 | } |
| 22496 | if (not_utf8) { |
| 22497 | _invlist_subtract(not_utf8, invlist, ¬_utf8); |
| 22498 | } |
| 22499 | |
| 22500 | if (only_utf8_locale_invlist) { |
| 22501 | |
| 22502 | /* Since this list is passed in, we have to make a copy before |
| 22503 | * modifying it */ |
| 22504 | only_utf8_locale = invlist_clone(only_utf8_locale_invlist, NULL); |
| 22505 | |
| 22506 | _invlist_subtract(only_utf8_locale, invlist, &only_utf8_locale); |
| 22507 | |
| 22508 | /* And, it can get really weird for us to try outputting an inverted |
| 22509 | * form of this list when it has things above the bitmap, so don't even |
| 22510 | * try */ |
| 22511 | if (invlist_highest(only_utf8_locale) >= NUM_ANYOF_CODE_POINTS) { |
| 22512 | inverting_allowed = FALSE; |
| 22513 | } |
| 22514 | } |
| 22515 | |
| 22516 | /* Calculate what the output would be if we take the input as-is */ |
| 22517 | as_is_display = put_charclass_bitmap_innards_common(invlist, |
| 22518 | posixes, |
| 22519 | only_utf8, |
| 22520 | not_utf8, |
| 22521 | only_utf8_locale, |
| 22522 | invert); |
| 22523 | |
| 22524 | /* If have to take the output as-is, just do that */ |
| 22525 | if (! inverting_allowed) { |
| 22526 | if (as_is_display) { |
| 22527 | sv_catsv(sv, as_is_display); |
| 22528 | SvREFCNT_dec_NN(as_is_display); |
| 22529 | } |
| 22530 | } |
| 22531 | else { /* But otherwise, create the output again on the inverted input, and |
| 22532 | use whichever version is shorter */ |
| 22533 | |
| 22534 | int inverted_bias, as_is_bias; |
| 22535 | |
| 22536 | /* We will apply our bias to whichever of the the results doesn't have |
| 22537 | * the '^' */ |
| 22538 | if (invert) { |
| 22539 | invert = FALSE; |
| 22540 | as_is_bias = bias; |
| 22541 | inverted_bias = 0; |
| 22542 | } |
| 22543 | else { |
| 22544 | invert = TRUE; |
| 22545 | as_is_bias = 0; |
| 22546 | inverted_bias = bias; |
| 22547 | } |
| 22548 | |
| 22549 | /* Now invert each of the lists that contribute to the output, |
| 22550 | * excluding from the result things outside the possible range */ |
| 22551 | |
| 22552 | /* For the unconditional inversion list, we have to add in all the |
| 22553 | * conditional code points, so that when inverted, they will be gone |
| 22554 | * from it */ |
| 22555 | _invlist_union(only_utf8, invlist, &invlist); |
| 22556 | _invlist_union(not_utf8, invlist, &invlist); |
| 22557 | _invlist_union(only_utf8_locale, invlist, &invlist); |
| 22558 | _invlist_invert(invlist); |
| 22559 | _invlist_intersection(invlist, PL_InBitmap, &invlist); |
| 22560 | |
| 22561 | if (only_utf8) { |
| 22562 | _invlist_invert(only_utf8); |
| 22563 | _invlist_intersection(only_utf8, PL_UpperLatin1, &only_utf8); |
| 22564 | } |
| 22565 | else if (not_utf8) { |
| 22566 | |
| 22567 | /* If a code point matches iff the target string is not in UTF-8, |
| 22568 | * then complementing the result has it not match iff not in UTF-8, |
| 22569 | * which is the same thing as matching iff it is UTF-8. */ |
| 22570 | only_utf8 = not_utf8; |
| 22571 | not_utf8 = NULL; |
| 22572 | } |
| 22573 | |
| 22574 | if (only_utf8_locale) { |
| 22575 | _invlist_invert(only_utf8_locale); |
| 22576 | _invlist_intersection(only_utf8_locale, |
| 22577 | PL_InBitmap, |
| 22578 | &only_utf8_locale); |
| 22579 | } |
| 22580 | |
| 22581 | inverted_display = put_charclass_bitmap_innards_common( |
| 22582 | invlist, |
| 22583 | posixes, |
| 22584 | only_utf8, |
| 22585 | not_utf8, |
| 22586 | only_utf8_locale, invert); |
| 22587 | |
| 22588 | /* Use the shortest representation, taking into account our bias |
| 22589 | * against showing it inverted */ |
| 22590 | if ( inverted_display |
| 22591 | && ( ! as_is_display |
| 22592 | || ( SvCUR(inverted_display) + inverted_bias |
| 22593 | < SvCUR(as_is_display) + as_is_bias))) |
| 22594 | { |
| 22595 | sv_catsv(sv, inverted_display); |
| 22596 | } |
| 22597 | else if (as_is_display) { |
| 22598 | sv_catsv(sv, as_is_display); |
| 22599 | } |
| 22600 | |
| 22601 | SvREFCNT_dec(as_is_display); |
| 22602 | SvREFCNT_dec(inverted_display); |
| 22603 | } |
| 22604 | |
| 22605 | SvREFCNT_dec_NN(invlist); |
| 22606 | SvREFCNT_dec(only_utf8); |
| 22607 | SvREFCNT_dec(not_utf8); |
| 22608 | SvREFCNT_dec(posixes); |
| 22609 | SvREFCNT_dec(only_utf8_locale); |
| 22610 | |
| 22611 | return SvCUR(sv) > orig_sv_cur; |
| 22612 | } |
| 22613 | |
| 22614 | #define CLEAR_OPTSTART \ |
| 22615 | if (optstart) STMT_START { \ |
| 22616 | DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ \ |
| 22617 | " (%" IVdf " nodes)\n", (IV)(node - optstart))); \ |
| 22618 | optstart=NULL; \ |
| 22619 | } STMT_END |
| 22620 | |
| 22621 | #define DUMPUNTIL(b,e) \ |
| 22622 | CLEAR_OPTSTART; \ |
| 22623 | node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1); |
| 22624 | |
| 22625 | STATIC const regnode * |
| 22626 | S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node, |
| 22627 | const regnode *last, const regnode *plast, |
| 22628 | SV* sv, I32 indent, U32 depth) |
| 22629 | { |
| 22630 | U8 op = PSEUDO; /* Arbitrary non-END op. */ |
| 22631 | const regnode *next; |
| 22632 | const regnode *optstart= NULL; |
| 22633 | |
| 22634 | RXi_GET_DECL(r, ri); |
| 22635 | GET_RE_DEBUG_FLAGS_DECL; |
| 22636 | |
| 22637 | PERL_ARGS_ASSERT_DUMPUNTIL; |
| 22638 | |
| 22639 | #ifdef DEBUG_DUMPUNTIL |
| 22640 | Perl_re_printf( aTHX_ "--- %d : %d - %d - %d\n", indent, node-start, |
| 22641 | last ? last-start : 0, plast ? plast-start : 0); |
| 22642 | #endif |
| 22643 | |
| 22644 | if (plast && plast < last) |
| 22645 | last= plast; |
| 22646 | |
| 22647 | while (PL_regkind[op] != END && (!last || node < last)) { |
| 22648 | assert(node); |
| 22649 | /* While that wasn't END last time... */ |
| 22650 | NODE_ALIGN(node); |
| 22651 | op = OP(node); |
| 22652 | if (op == CLOSE || op == SRCLOSE || op == WHILEM) |
| 22653 | indent--; |
| 22654 | next = regnext((regnode *)node); |
| 22655 | |
| 22656 | /* Where, what. */ |
| 22657 | if (OP(node) == OPTIMIZED) { |
| 22658 | if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) |
| 22659 | optstart = node; |
| 22660 | else |
| 22661 | goto after_print; |
| 22662 | } else |
| 22663 | CLEAR_OPTSTART; |
| 22664 | |
| 22665 | regprop(r, sv, node, NULL, NULL); |
| 22666 | Perl_re_printf( aTHX_ "%4" IVdf ":%*s%s", (IV)(node - start), |
| 22667 | (int)(2*indent + 1), "", SvPVX_const(sv)); |
| 22668 | |
| 22669 | if (OP(node) != OPTIMIZED) { |
| 22670 | if (next == NULL) /* Next ptr. */ |
| 22671 | Perl_re_printf( aTHX_ " (0)"); |
| 22672 | else if (PL_regkind[(U8)op] == BRANCH |
| 22673 | && PL_regkind[OP(next)] != BRANCH ) |
| 22674 | Perl_re_printf( aTHX_ " (FAIL)"); |
| 22675 | else |
| 22676 | Perl_re_printf( aTHX_ " (%" IVdf ")", (IV)(next - start)); |
| 22677 | Perl_re_printf( aTHX_ "\n"); |
| 22678 | } |
| 22679 | |
| 22680 | after_print: |
| 22681 | if (PL_regkind[(U8)op] == BRANCHJ) { |
| 22682 | assert(next); |
| 22683 | { |
| 22684 | const regnode *nnode = (OP(next) == LONGJMP |
| 22685 | ? regnext((regnode *)next) |
| 22686 | : next); |
| 22687 | if (last && nnode > last) |
| 22688 | nnode = last; |
| 22689 | DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode); |
| 22690 | } |
| 22691 | } |
| 22692 | else if (PL_regkind[(U8)op] == BRANCH) { |
| 22693 | assert(next); |
| 22694 | DUMPUNTIL(NEXTOPER(node), next); |
| 22695 | } |
| 22696 | else if ( PL_regkind[(U8)op] == TRIE ) { |
| 22697 | const regnode *this_trie = node; |
| 22698 | const char op = OP(node); |
| 22699 | const U32 n = ARG(node); |
| 22700 | const reg_ac_data * const ac = op>=AHOCORASICK ? |
| 22701 | (reg_ac_data *)ri->data->data[n] : |
| 22702 | NULL; |
| 22703 | const reg_trie_data * const trie = |
| 22704 | (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie]; |
| 22705 | #ifdef DEBUGGING |
| 22706 | AV *const trie_words |
| 22707 | = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]); |
| 22708 | #endif |
| 22709 | const regnode *nextbranch= NULL; |
| 22710 | I32 word_idx; |
| 22711 | SvPVCLEAR(sv); |
| 22712 | for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) { |
| 22713 | SV ** const elem_ptr = av_fetch(trie_words, word_idx, 0); |
| 22714 | |
| 22715 | Perl_re_indentf( aTHX_ "%s ", |
| 22716 | indent+3, |
| 22717 | elem_ptr |
| 22718 | ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), |
| 22719 | SvCUR(*elem_ptr), PL_dump_re_max_len, |
| 22720 | PL_colors[0], PL_colors[1], |
| 22721 | (SvUTF8(*elem_ptr) |
| 22722 | ? PERL_PV_ESCAPE_UNI |
| 22723 | : 0) |
| 22724 | | PERL_PV_PRETTY_ELLIPSES |
| 22725 | | PERL_PV_PRETTY_LTGT |
| 22726 | ) |
| 22727 | : "???" |
| 22728 | ); |
| 22729 | if (trie->jump) { |
| 22730 | U16 dist= trie->jump[word_idx+1]; |
| 22731 | Perl_re_printf( aTHX_ "(%" UVuf ")\n", |
| 22732 | (UV)((dist ? this_trie + dist : next) - start)); |
| 22733 | if (dist) { |
| 22734 | if (!nextbranch) |
| 22735 | nextbranch= this_trie + trie->jump[0]; |
| 22736 | DUMPUNTIL(this_trie + dist, nextbranch); |
| 22737 | } |
| 22738 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) |
| 22739 | nextbranch= regnext((regnode *)nextbranch); |
| 22740 | } else { |
| 22741 | Perl_re_printf( aTHX_ "\n"); |
| 22742 | } |
| 22743 | } |
| 22744 | if (last && next > last) |
| 22745 | node= last; |
| 22746 | else |
| 22747 | node= next; |
| 22748 | } |
| 22749 | else if ( op == CURLY ) { /* "next" might be very big: optimizer */ |
| 22750 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, |
| 22751 | NEXTOPER(node) + EXTRA_STEP_2ARGS + 1); |
| 22752 | } |
| 22753 | else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) { |
| 22754 | assert(next); |
| 22755 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next); |
| 22756 | } |
| 22757 | else if ( op == PLUS || op == STAR) { |
| 22758 | DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1); |
| 22759 | } |
| 22760 | else if (PL_regkind[(U8)op] == EXACT || op == ANYOFHs) { |
| 22761 | /* Literal string, where present. */ |
| 22762 | node += NODE_SZ_STR(node) - 1; |
| 22763 | node = NEXTOPER(node); |
| 22764 | } |
| 22765 | else { |
| 22766 | node = NEXTOPER(node); |
| 22767 | node += regarglen[(U8)op]; |
| 22768 | } |
| 22769 | if (op == CURLYX || op == OPEN || op == SROPEN) |
| 22770 | indent++; |
| 22771 | } |
| 22772 | CLEAR_OPTSTART; |
| 22773 | #ifdef DEBUG_DUMPUNTIL |
| 22774 | Perl_re_printf( aTHX_ "--- %d\n", (int)indent); |
| 22775 | #endif |
| 22776 | return node; |
| 22777 | } |
| 22778 | |
| 22779 | #endif /* DEBUGGING */ |
| 22780 | |
| 22781 | #ifndef PERL_IN_XSUB_RE |
| 22782 | |
| 22783 | # include "uni_keywords.h" |
| 22784 | |
| 22785 | void |
| 22786 | Perl_init_uniprops(pTHX) |
| 22787 | { |
| 22788 | dVAR; |
| 22789 | |
| 22790 | # ifdef DEBUGGING |
| 22791 | char * dump_len_string; |
| 22792 | |
| 22793 | dump_len_string = PerlEnv_getenv("PERL_DUMP_RE_MAX_LEN"); |
| 22794 | if ( ! dump_len_string |
| 22795 | || ! grok_atoUV(dump_len_string, (UV *)&PL_dump_re_max_len, NULL)) |
| 22796 | { |
| 22797 | PL_dump_re_max_len = 60; /* A reasonable default */ |
| 22798 | } |
| 22799 | # endif |
| 22800 | |
| 22801 | PL_user_def_props = newHV(); |
| 22802 | |
| 22803 | # ifdef USE_ITHREADS |
| 22804 | |
| 22805 | HvSHAREKEYS_off(PL_user_def_props); |
| 22806 | PL_user_def_props_aTHX = aTHX; |
| 22807 | |
| 22808 | # endif |
| 22809 | |
| 22810 | /* Set up the inversion list interpreter-level variables */ |
| 22811 | |
| 22812 | PL_XPosix_ptrs[_CC_ASCII] = _new_invlist_C_array(uni_prop_ptrs[UNI_ASCII]); |
| 22813 | PL_XPosix_ptrs[_CC_ALPHANUMERIC] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXALNUM]); |
| 22814 | PL_XPosix_ptrs[_CC_ALPHA] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXALPHA]); |
| 22815 | PL_XPosix_ptrs[_CC_BLANK] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXBLANK]); |
| 22816 | PL_XPosix_ptrs[_CC_CASED] = _new_invlist_C_array(uni_prop_ptrs[UNI_CASED]); |
| 22817 | PL_XPosix_ptrs[_CC_CNTRL] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXCNTRL]); |
| 22818 | PL_XPosix_ptrs[_CC_DIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXDIGIT]); |
| 22819 | PL_XPosix_ptrs[_CC_GRAPH] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXGRAPH]); |
| 22820 | PL_XPosix_ptrs[_CC_LOWER] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXLOWER]); |
| 22821 | PL_XPosix_ptrs[_CC_PRINT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXPRINT]); |
| 22822 | PL_XPosix_ptrs[_CC_PUNCT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXPUNCT]); |
| 22823 | PL_XPosix_ptrs[_CC_SPACE] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXSPACE]); |
| 22824 | PL_XPosix_ptrs[_CC_UPPER] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXUPPER]); |
| 22825 | PL_XPosix_ptrs[_CC_VERTSPACE] = _new_invlist_C_array(uni_prop_ptrs[UNI_VERTSPACE]); |
| 22826 | PL_XPosix_ptrs[_CC_WORDCHAR] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXWORD]); |
| 22827 | PL_XPosix_ptrs[_CC_XDIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXXDIGIT]); |
| 22828 | |
| 22829 | PL_Posix_ptrs[_CC_ASCII] = _new_invlist_C_array(uni_prop_ptrs[UNI_ASCII]); |
| 22830 | PL_Posix_ptrs[_CC_ALPHANUMERIC] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXALNUM]); |
| 22831 | PL_Posix_ptrs[_CC_ALPHA] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXALPHA]); |
| 22832 | PL_Posix_ptrs[_CC_BLANK] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXBLANK]); |
| 22833 | PL_Posix_ptrs[_CC_CASED] = PL_Posix_ptrs[_CC_ALPHA]; |
| 22834 | PL_Posix_ptrs[_CC_CNTRL] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXCNTRL]); |
| 22835 | PL_Posix_ptrs[_CC_DIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXDIGIT]); |
| 22836 | PL_Posix_ptrs[_CC_GRAPH] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXGRAPH]); |
| 22837 | PL_Posix_ptrs[_CC_LOWER] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXLOWER]); |
| 22838 | PL_Posix_ptrs[_CC_PRINT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXPRINT]); |
| 22839 | PL_Posix_ptrs[_CC_PUNCT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXPUNCT]); |
| 22840 | PL_Posix_ptrs[_CC_SPACE] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXSPACE]); |
| 22841 | PL_Posix_ptrs[_CC_UPPER] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXUPPER]); |
| 22842 | PL_Posix_ptrs[_CC_VERTSPACE] = NULL; |
| 22843 | PL_Posix_ptrs[_CC_WORDCHAR] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXWORD]); |
| 22844 | PL_Posix_ptrs[_CC_XDIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXXDIGIT]); |
| 22845 | |
| 22846 | PL_GCB_invlist = _new_invlist_C_array(_Perl_GCB_invlist); |
| 22847 | PL_SB_invlist = _new_invlist_C_array(_Perl_SB_invlist); |
| 22848 | PL_WB_invlist = _new_invlist_C_array(_Perl_WB_invlist); |
| 22849 | PL_LB_invlist = _new_invlist_C_array(_Perl_LB_invlist); |
| 22850 | PL_SCX_invlist = _new_invlist_C_array(_Perl_SCX_invlist); |
| 22851 | |
| 22852 | PL_InBitmap = _new_invlist_C_array(InBitmap_invlist); |
| 22853 | PL_AboveLatin1 = _new_invlist_C_array(AboveLatin1_invlist); |
| 22854 | PL_Latin1 = _new_invlist_C_array(Latin1_invlist); |
| 22855 | PL_UpperLatin1 = _new_invlist_C_array(UpperLatin1_invlist); |
| 22856 | |
| 22857 | PL_Assigned_invlist = _new_invlist_C_array(uni_prop_ptrs[UNI_ASSIGNED]); |
| 22858 | |
| 22859 | PL_utf8_perl_idstart = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_IDSTART]); |
| 22860 | PL_utf8_perl_idcont = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_IDCONT]); |
| 22861 | |
| 22862 | PL_utf8_charname_begin = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_CHARNAME_BEGIN]); |
| 22863 | PL_utf8_charname_continue = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_CHARNAME_CONTINUE]); |
| 22864 | |
| 22865 | PL_in_some_fold = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_ANY_FOLDS]); |
| 22866 | PL_HasMultiCharFold = _new_invlist_C_array(uni_prop_ptrs[ |
| 22867 | UNI__PERL_FOLDS_TO_MULTI_CHAR]); |
| 22868 | PL_InMultiCharFold = _new_invlist_C_array(uni_prop_ptrs[ |
| 22869 | UNI__PERL_IS_IN_MULTI_CHAR_FOLD]); |
| 22870 | PL_utf8_toupper = _new_invlist_C_array(Uppercase_Mapping_invlist); |
| 22871 | PL_utf8_tolower = _new_invlist_C_array(Lowercase_Mapping_invlist); |
| 22872 | PL_utf8_totitle = _new_invlist_C_array(Titlecase_Mapping_invlist); |
| 22873 | PL_utf8_tofold = _new_invlist_C_array(Case_Folding_invlist); |
| 22874 | PL_utf8_tosimplefold = _new_invlist_C_array(Simple_Case_Folding_invlist); |
| 22875 | PL_utf8_foldclosures = _new_invlist_C_array(_Perl_IVCF_invlist); |
| 22876 | PL_utf8_mark = _new_invlist_C_array(uni_prop_ptrs[UNI_M]); |
| 22877 | PL_CCC_non0_non230 = _new_invlist_C_array(_Perl_CCC_non0_non230_invlist); |
| 22878 | PL_Private_Use = _new_invlist_C_array(uni_prop_ptrs[UNI_CO]); |
| 22879 | |
| 22880 | # ifdef UNI_XIDC |
| 22881 | /* The below are used only by deprecated functions. They could be removed */ |
| 22882 | PL_utf8_xidcont = _new_invlist_C_array(uni_prop_ptrs[UNI_XIDC]); |
| 22883 | PL_utf8_idcont = _new_invlist_C_array(uni_prop_ptrs[UNI_IDC]); |
| 22884 | PL_utf8_xidstart = _new_invlist_C_array(uni_prop_ptrs[UNI_XIDS]); |
| 22885 | # endif |
| 22886 | } |
| 22887 | |
| 22888 | # if 0 |
| 22889 | |
| 22890 | This code was mainly added for backcompat to give a warning for non-portable |
| 22891 | code points in user-defined properties. But experiments showed that the |
| 22892 | warning in earlier perls were only omitted on overflow, which should be an |
| 22893 | error, so there really isnt a backcompat issue, and actually adding the |
| 22894 | warning when none was present before might cause breakage, for little gain. So |
| 22895 | khw left this code in, but not enabled. Tests were never added. |
| 22896 | |
| 22897 | embed.fnc entry: |
| 22898 | Ei |const char *|get_extended_utf8_msg|const UV cp |
| 22899 | |
| 22900 | PERL_STATIC_INLINE const char * |
| 22901 | S_get_extended_utf8_msg(pTHX_ const UV cp) |
| 22902 | { |
| 22903 | U8 dummy[UTF8_MAXBYTES + 1]; |
| 22904 | HV *msgs; |
| 22905 | SV **msg; |
| 22906 | |
| 22907 | uvchr_to_utf8_flags_msgs(dummy, cp, UNICODE_WARN_PERL_EXTENDED, |
| 22908 | &msgs); |
| 22909 | |
| 22910 | msg = hv_fetchs(msgs, "text", 0); |
| 22911 | assert(msg); |
| 22912 | |
| 22913 | (void) sv_2mortal((SV *) msgs); |
| 22914 | |
| 22915 | return SvPVX(*msg); |
| 22916 | } |
| 22917 | |
| 22918 | # endif |
| 22919 | |
| 22920 | STATIC REGEXP * |
| 22921 | S_compile_wildcard(pTHX_ const char * name, const STRLEN len, |
| 22922 | const bool ignore_case) |
| 22923 | { |
| 22924 | U32 flags = PMf_MULTILINE|PMf_WILDCARD; |
| 22925 | REGEXP * subpattern_re; |
| 22926 | |
| 22927 | PERL_ARGS_ASSERT_COMPILE_WILDCARD; |
| 22928 | |
| 22929 | if (ignore_case) { |
| 22930 | flags |= PMf_FOLD; |
| 22931 | } |
| 22932 | set_regex_charset(&flags, REGEX_ASCII_MORE_RESTRICTED_CHARSET); |
| 22933 | |
| 22934 | subpattern_re = re_op_compile_wrapper(sv_2mortal(newSVpvn(name, len)), |
| 22935 | /* Like in op.c, we copy the compile |
| 22936 | * time pm flags to the rx ones */ |
| 22937 | (flags & RXf_PMf_COMPILETIME), flags); |
| 22938 | |
| 22939 | assert(subpattern_re); /* Should have died if didn't compile successfully */ |
| 22940 | return subpattern_re; |
| 22941 | } |
| 22942 | |
| 22943 | STATIC I32 |
| 22944 | S_execute_wildcard(pTHX_ REGEXP * const prog, char* stringarg, char *strend, |
| 22945 | char *strbeg, SSize_t minend, SV *screamer, U32 nosave) |
| 22946 | { |
| 22947 | I32 result; |
| 22948 | |
| 22949 | PERL_ARGS_ASSERT_EXECUTE_WILDCARD; |
| 22950 | |
| 22951 | result = pregexec(prog, stringarg, strend, strbeg, minend, screamer, nosave); |
| 22952 | |
| 22953 | return result; |
| 22954 | } |
| 22955 | |
| 22956 | SV * |
| 22957 | Perl_handle_user_defined_property(pTHX_ |
| 22958 | |
| 22959 | /* Parses the contents of a user-defined property definition; returning the |
| 22960 | * expanded definition if possible. If so, the return is an inversion |
| 22961 | * list. |
| 22962 | * |
| 22963 | * If there are subroutines that are part of the expansion and which aren't |
| 22964 | * known at the time of the call to this function, this returns what |
| 22965 | * parse_uniprop_string() returned for the first one encountered. |
| 22966 | * |
| 22967 | * If an error was found, NULL is returned, and 'msg' gets a suitable |
| 22968 | * message appended to it. (Appending allows the back trace of how we got |
| 22969 | * to the faulty definition to be displayed through nested calls of |
| 22970 | * user-defined subs.) |
| 22971 | * |
| 22972 | * The caller IS responsible for freeing any returned SV. |
| 22973 | * |
| 22974 | * The syntax of the contents is pretty much described in perlunicode.pod, |
| 22975 | * but we also allow comments on each line */ |
| 22976 | |
| 22977 | const char * name, /* Name of property */ |
| 22978 | const STRLEN name_len, /* The name's length in bytes */ |
| 22979 | const bool is_utf8, /* ? Is 'name' encoded in UTF-8 */ |
| 22980 | const bool to_fold, /* ? Is this under /i */ |
| 22981 | const bool runtime, /* ? Are we in compile- or run-time */ |
| 22982 | const bool deferrable, /* Is it ok for this property's full definition |
| 22983 | to be deferred until later? */ |
| 22984 | SV* contents, /* The property's definition */ |
| 22985 | bool *user_defined_ptr, /* This will be set TRUE as we wouldn't be |
| 22986 | getting called unless this is thought to be |
| 22987 | a user-defined property */ |
| 22988 | SV * msg, /* Any error or warning msg(s) are appended to |
| 22989 | this */ |
| 22990 | const STRLEN level) /* Recursion level of this call */ |
| 22991 | { |
| 22992 | STRLEN len; |
| 22993 | const char * string = SvPV_const(contents, len); |
| 22994 | const char * const e = string + len; |
| 22995 | const bool is_contents_utf8 = cBOOL(SvUTF8(contents)); |
| 22996 | const STRLEN msgs_length_on_entry = SvCUR(msg); |
| 22997 | |
| 22998 | const char * s0 = string; /* Points to first byte in the current line |
| 22999 | being parsed in 'string' */ |
| 23000 | const char overflow_msg[] = "Code point too large in \""; |
| 23001 | SV* running_definition = NULL; |
| 23002 | |
| 23003 | PERL_ARGS_ASSERT_HANDLE_USER_DEFINED_PROPERTY; |
| 23004 | |
| 23005 | *user_defined_ptr = TRUE; |
| 23006 | |
| 23007 | /* Look at each line */ |
| 23008 | while (s0 < e) { |
| 23009 | const char * s; /* Current byte */ |
| 23010 | char op = '+'; /* Default operation is 'union' */ |
| 23011 | IV min = 0; /* range begin code point */ |
| 23012 | IV max = -1; /* and range end */ |
| 23013 | SV* this_definition; |
| 23014 | |
| 23015 | /* Skip comment lines */ |
| 23016 | if (*s0 == '#') { |
| 23017 | s0 = strchr(s0, '\n'); |
| 23018 | if (s0 == NULL) { |
| 23019 | break; |
| 23020 | } |
| 23021 | s0++; |
| 23022 | continue; |
| 23023 | } |
| 23024 | |
| 23025 | /* For backcompat, allow an empty first line */ |
| 23026 | if (*s0 == '\n') { |
| 23027 | s0++; |
| 23028 | continue; |
| 23029 | } |
| 23030 | |
| 23031 | /* First character in the line may optionally be the operation */ |
| 23032 | if ( *s0 == '+' |
| 23033 | || *s0 == '!' |
| 23034 | || *s0 == '-' |
| 23035 | || *s0 == '&') |
| 23036 | { |
| 23037 | op = *s0++; |
| 23038 | } |
| 23039 | |
| 23040 | /* If the line is one or two hex digits separated by blank space, its |
| 23041 | * a range; otherwise it is either another user-defined property or an |
| 23042 | * error */ |
| 23043 | |
| 23044 | s = s0; |
| 23045 | |
| 23046 | if (! isXDIGIT(*s)) { |
| 23047 | goto check_if_property; |
| 23048 | } |
| 23049 | |
| 23050 | do { /* Each new hex digit will add 4 bits. */ |
| 23051 | if (min > ( (IV) MAX_LEGAL_CP >> 4)) { |
| 23052 | s = strchr(s, '\n'); |
| 23053 | if (s == NULL) { |
| 23054 | s = e; |
| 23055 | } |
| 23056 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 23057 | sv_catpv(msg, overflow_msg); |
| 23058 | Perl_sv_catpvf(aTHX_ msg, "%" UTF8f, |
| 23059 | UTF8fARG(is_contents_utf8, s - s0, s0)); |
| 23060 | sv_catpvs(msg, "\""); |
| 23061 | goto return_failure; |
| 23062 | } |
| 23063 | |
| 23064 | /* Accumulate this digit into the value */ |
| 23065 | min = (min << 4) + READ_XDIGIT(s); |
| 23066 | } while (isXDIGIT(*s)); |
| 23067 | |
| 23068 | while (isBLANK(*s)) { s++; } |
| 23069 | |
| 23070 | /* We allow comments at the end of the line */ |
| 23071 | if (*s == '#') { |
| 23072 | s = strchr(s, '\n'); |
| 23073 | if (s == NULL) { |
| 23074 | s = e; |
| 23075 | } |
| 23076 | s++; |
| 23077 | } |
| 23078 | else if (s < e && *s != '\n') { |
| 23079 | if (! isXDIGIT(*s)) { |
| 23080 | goto check_if_property; |
| 23081 | } |
| 23082 | |
| 23083 | /* Look for the high point of the range */ |
| 23084 | max = 0; |
| 23085 | do { |
| 23086 | if (max > ( (IV) MAX_LEGAL_CP >> 4)) { |
| 23087 | s = strchr(s, '\n'); |
| 23088 | if (s == NULL) { |
| 23089 | s = e; |
| 23090 | } |
| 23091 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 23092 | sv_catpv(msg, overflow_msg); |
| 23093 | Perl_sv_catpvf(aTHX_ msg, "%" UTF8f, |
| 23094 | UTF8fARG(is_contents_utf8, s - s0, s0)); |
| 23095 | sv_catpvs(msg, "\""); |
| 23096 | goto return_failure; |
| 23097 | } |
| 23098 | |
| 23099 | max = (max << 4) + READ_XDIGIT(s); |
| 23100 | } while (isXDIGIT(*s)); |
| 23101 | |
| 23102 | while (isBLANK(*s)) { s++; } |
| 23103 | |
| 23104 | if (*s == '#') { |
| 23105 | s = strchr(s, '\n'); |
| 23106 | if (s == NULL) { |
| 23107 | s = e; |
| 23108 | } |
| 23109 | } |
| 23110 | else if (s < e && *s != '\n') { |
| 23111 | goto check_if_property; |
| 23112 | } |
| 23113 | } |
| 23114 | |
| 23115 | if (max == -1) { /* The line only had one entry */ |
| 23116 | max = min; |
| 23117 | } |
| 23118 | else if (max < min) { |
| 23119 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 23120 | sv_catpvs(msg, "Illegal range in \""); |
| 23121 | Perl_sv_catpvf(aTHX_ msg, "%" UTF8f, |
| 23122 | UTF8fARG(is_contents_utf8, s - s0, s0)); |
| 23123 | sv_catpvs(msg, "\""); |
| 23124 | goto return_failure; |
| 23125 | } |
| 23126 | |
| 23127 | # if 0 /* See explanation at definition above of get_extended_utf8_msg() */ |
| 23128 | |
| 23129 | if ( UNICODE_IS_PERL_EXTENDED(min) |
| 23130 | || UNICODE_IS_PERL_EXTENDED(max)) |
| 23131 | { |
| 23132 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 23133 | |
| 23134 | /* If both code points are non-portable, warn only on the lower |
| 23135 | * one. */ |
| 23136 | sv_catpv(msg, get_extended_utf8_msg( |
| 23137 | (UNICODE_IS_PERL_EXTENDED(min)) |
| 23138 | ? min : max)); |
| 23139 | sv_catpvs(msg, " in \""); |
| 23140 | Perl_sv_catpvf(aTHX_ msg, "%" UTF8f, |
| 23141 | UTF8fARG(is_contents_utf8, s - s0, s0)); |
| 23142 | sv_catpvs(msg, "\""); |
| 23143 | } |
| 23144 | |
| 23145 | # endif |
| 23146 | |
| 23147 | /* Here, this line contains a legal range */ |
| 23148 | this_definition = sv_2mortal(_new_invlist(2)); |
| 23149 | this_definition = _add_range_to_invlist(this_definition, min, max); |
| 23150 | goto calculate; |
| 23151 | |
| 23152 | check_if_property: |
| 23153 | |
| 23154 | /* Here it isn't a legal range line. See if it is a legal property |
| 23155 | * line. First find the end of the meat of the line */ |
| 23156 | s = strpbrk(s, "#\n"); |
| 23157 | if (s == NULL) { |
| 23158 | s = e; |
| 23159 | } |
| 23160 | |
| 23161 | /* Ignore trailing blanks in keeping with the requirements of |
| 23162 | * parse_uniprop_string() */ |
| 23163 | s--; |
| 23164 | while (s > s0 && isBLANK_A(*s)) { |
| 23165 | s--; |
| 23166 | } |
| 23167 | s++; |
| 23168 | |
| 23169 | this_definition = parse_uniprop_string(s0, s - s0, |
| 23170 | is_utf8, to_fold, runtime, |
| 23171 | deferrable, |
| 23172 | user_defined_ptr, msg, |
| 23173 | (name_len == 0) |
| 23174 | ? level /* Don't increase level |
| 23175 | if input is empty */ |
| 23176 | : level + 1 |
| 23177 | ); |
| 23178 | if (this_definition == NULL) { |
| 23179 | goto return_failure; /* 'msg' should have had the reason |
| 23180 | appended to it by the above call */ |
| 23181 | } |
| 23182 | |
| 23183 | if (! is_invlist(this_definition)) { /* Unknown at this time */ |
| 23184 | return newSVsv(this_definition); |
| 23185 | } |
| 23186 | |
| 23187 | if (*s != '\n') { |
| 23188 | s = strchr(s, '\n'); |
| 23189 | if (s == NULL) { |
| 23190 | s = e; |
| 23191 | } |
| 23192 | } |
| 23193 | |
| 23194 | calculate: |
| 23195 | |
| 23196 | switch (op) { |
| 23197 | case '+': |
| 23198 | _invlist_union(running_definition, this_definition, |
| 23199 | &running_definition); |
| 23200 | break; |
| 23201 | case '-': |
| 23202 | _invlist_subtract(running_definition, this_definition, |
| 23203 | &running_definition); |
| 23204 | break; |
| 23205 | case '&': |
| 23206 | _invlist_intersection(running_definition, this_definition, |
| 23207 | &running_definition); |
| 23208 | break; |
| 23209 | case '!': |
| 23210 | _invlist_union_complement_2nd(running_definition, |
| 23211 | this_definition, &running_definition); |
| 23212 | break; |
| 23213 | default: |
| 23214 | Perl_croak(aTHX_ "panic: %s: %d: Unexpected operation %d", |
| 23215 | __FILE__, __LINE__, op); |
| 23216 | break; |
| 23217 | } |
| 23218 | |
| 23219 | /* Position past the '\n' */ |
| 23220 | s0 = s + 1; |
| 23221 | } /* End of loop through the lines of 'contents' */ |
| 23222 | |
| 23223 | /* Here, we processed all the lines in 'contents' without error. If we |
| 23224 | * didn't add any warnings, simply return success */ |
| 23225 | if (msgs_length_on_entry == SvCUR(msg)) { |
| 23226 | |
| 23227 | /* If the expansion was empty, the answer isn't nothing: its an empty |
| 23228 | * inversion list */ |
| 23229 | if (running_definition == NULL) { |
| 23230 | running_definition = _new_invlist(1); |
| 23231 | } |
| 23232 | |
| 23233 | return running_definition; |
| 23234 | } |
| 23235 | |
| 23236 | /* Otherwise, add some explanatory text, but we will return success */ |
| 23237 | goto return_msg; |
| 23238 | |
| 23239 | return_failure: |
| 23240 | running_definition = NULL; |
| 23241 | |
| 23242 | return_msg: |
| 23243 | |
| 23244 | if (name_len > 0) { |
| 23245 | sv_catpvs(msg, " in expansion of "); |
| 23246 | Perl_sv_catpvf(aTHX_ msg, "%" UTF8f, UTF8fARG(is_utf8, name_len, name)); |
| 23247 | } |
| 23248 | |
| 23249 | return running_definition; |
| 23250 | } |
| 23251 | |
| 23252 | /* As explained below, certain operations need to take place in the first |
| 23253 | * thread created. These macros switch contexts */ |
| 23254 | # ifdef USE_ITHREADS |
| 23255 | # define DECLARATION_FOR_GLOBAL_CONTEXT \ |
| 23256 | PerlInterpreter * save_aTHX = aTHX; |
| 23257 | # define SWITCH_TO_GLOBAL_CONTEXT \ |
| 23258 | PERL_SET_CONTEXT((aTHX = PL_user_def_props_aTHX)) |
| 23259 | # define RESTORE_CONTEXT PERL_SET_CONTEXT((aTHX = save_aTHX)); |
| 23260 | # define CUR_CONTEXT aTHX |
| 23261 | # define ORIGINAL_CONTEXT save_aTHX |
| 23262 | # else |
| 23263 | # define DECLARATION_FOR_GLOBAL_CONTEXT |
| 23264 | # define SWITCH_TO_GLOBAL_CONTEXT NOOP |
| 23265 | # define RESTORE_CONTEXT NOOP |
| 23266 | # define CUR_CONTEXT NULL |
| 23267 | # define ORIGINAL_CONTEXT NULL |
| 23268 | # endif |
| 23269 | |
| 23270 | STATIC void |
| 23271 | S_delete_recursion_entry(pTHX_ void *key) |
| 23272 | { |
| 23273 | /* Deletes the entry used to detect recursion when expanding user-defined |
| 23274 | * properties. This is a function so it can be set up to be called even if |
| 23275 | * the program unexpectedly quits */ |
| 23276 | |
| 23277 | dVAR; |
| 23278 | SV ** current_entry; |
| 23279 | const STRLEN key_len = strlen((const char *) key); |
| 23280 | DECLARATION_FOR_GLOBAL_CONTEXT; |
| 23281 | |
| 23282 | SWITCH_TO_GLOBAL_CONTEXT; |
| 23283 | |
| 23284 | /* If the entry is one of these types, it is a permanent entry, and not the |
| 23285 | * one used to detect recursions. This function should delete only the |
| 23286 | * recursion entry */ |
| 23287 | current_entry = hv_fetch(PL_user_def_props, (const char *) key, key_len, 0); |
| 23288 | if ( current_entry |
| 23289 | && ! is_invlist(*current_entry) |
| 23290 | && ! SvPOK(*current_entry)) |
| 23291 | { |
| 23292 | (void) hv_delete(PL_user_def_props, (const char *) key, key_len, |
| 23293 | G_DISCARD); |
| 23294 | } |
| 23295 | |
| 23296 | RESTORE_CONTEXT; |
| 23297 | } |
| 23298 | |
| 23299 | STATIC SV * |
| 23300 | S_get_fq_name(pTHX_ |
| 23301 | const char * const name, /* The first non-blank in the \p{}, \P{} */ |
| 23302 | const Size_t name_len, /* Its length in bytes, not including any trailing space */ |
| 23303 | const bool is_utf8, /* ? Is 'name' encoded in UTF-8 */ |
| 23304 | const bool has_colon_colon |
| 23305 | ) |
| 23306 | { |
| 23307 | /* Returns a mortal SV containing the fully qualified version of the input |
| 23308 | * name */ |
| 23309 | |
| 23310 | SV * fq_name; |
| 23311 | |
| 23312 | fq_name = newSVpvs_flags("", SVs_TEMP); |
| 23313 | |
| 23314 | /* Use the current package if it wasn't included in our input */ |
| 23315 | if (! has_colon_colon) { |
| 23316 | const HV * pkg = (IN_PERL_COMPILETIME) |
| 23317 | ? PL_curstash |
| 23318 | : CopSTASH(PL_curcop); |
| 23319 | const char* pkgname = HvNAME(pkg); |
| 23320 | |
| 23321 | Perl_sv_catpvf(aTHX_ fq_name, "%" UTF8f, |
| 23322 | UTF8fARG(is_utf8, strlen(pkgname), pkgname)); |
| 23323 | sv_catpvs(fq_name, "::"); |
| 23324 | } |
| 23325 | |
| 23326 | Perl_sv_catpvf(aTHX_ fq_name, "%" UTF8f, |
| 23327 | UTF8fARG(is_utf8, name_len, name)); |
| 23328 | return fq_name; |
| 23329 | } |
| 23330 | |
| 23331 | SV * |
| 23332 | Perl_parse_uniprop_string(pTHX_ |
| 23333 | |
| 23334 | /* Parse the interior of a \p{}, \P{}. Returns its definition if knowable |
| 23335 | * now. If so, the return is an inversion list. |
| 23336 | * |
| 23337 | * If the property is user-defined, it is a subroutine, which in turn |
| 23338 | * may call other subroutines. This function will call the whole nest of |
| 23339 | * them to get the definition they return; if some aren't known at the time |
| 23340 | * of the call to this function, the fully qualified name of the highest |
| 23341 | * level sub is returned. It is an error to call this function at runtime |
| 23342 | * without every sub defined. |
| 23343 | * |
| 23344 | * If an error was found, NULL is returned, and 'msg' gets a suitable |
| 23345 | * message appended to it. (Appending allows the back trace of how we got |
| 23346 | * to the faulty definition to be displayed through nested calls of |
| 23347 | * user-defined subs.) |
| 23348 | * |
| 23349 | * The caller should NOT try to free any returned inversion list. |
| 23350 | * |
| 23351 | * Other parameters will be set on return as described below */ |
| 23352 | |
| 23353 | const char * const name, /* The first non-blank in the \p{}, \P{} */ |
| 23354 | Size_t name_len, /* Its length in bytes, not including any |
| 23355 | trailing space */ |
| 23356 | const bool is_utf8, /* ? Is 'name' encoded in UTF-8 */ |
| 23357 | const bool to_fold, /* ? Is this under /i */ |
| 23358 | const bool runtime, /* TRUE if this is being called at run time */ |
| 23359 | const bool deferrable, /* TRUE if it's ok for the definition to not be |
| 23360 | known at this call */ |
| 23361 | bool *user_defined_ptr, /* Upon return from this function it will be |
| 23362 | set to TRUE if any component is a |
| 23363 | user-defined property */ |
| 23364 | SV * msg, /* Any error or warning msg(s) are appended to |
| 23365 | this */ |
| 23366 | const STRLEN level) /* Recursion level of this call */ |
| 23367 | { |
| 23368 | dVAR; |
| 23369 | char* lookup_name; /* normalized name for lookup in our tables */ |
| 23370 | unsigned lookup_len; /* Its length */ |
| 23371 | enum { Not_Strict = 0, /* Some properties have stricter name */ |
| 23372 | Strict, /* normalization rules, which we decide */ |
| 23373 | As_Is /* upon based on parsing */ |
| 23374 | } stricter = Not_Strict; |
| 23375 | |
| 23376 | /* nv= or numeric_value=, or possibly one of the cjk numeric properties |
| 23377 | * (though it requires extra effort to download them from Unicode and |
| 23378 | * compile perl to know about them) */ |
| 23379 | bool is_nv_type = FALSE; |
| 23380 | |
| 23381 | unsigned int i, j = 0; |
| 23382 | int equals_pos = -1; /* Where the '=' is found, or negative if none */ |
| 23383 | int slash_pos = -1; /* Where the '/' is found, or negative if none */ |
| 23384 | int table_index = 0; /* The entry number for this property in the table |
| 23385 | of all Unicode property names */ |
| 23386 | bool starts_with_Is = FALSE; /* ? Does the name start with 'Is' */ |
| 23387 | Size_t lookup_offset = 0; /* Used to ignore the first few characters of |
| 23388 | the normalized name in certain situations */ |
| 23389 | Size_t non_pkg_begin = 0; /* Offset of first byte in 'name' that isn't |
| 23390 | part of a package name */ |
| 23391 | Size_t lun_non_pkg_begin = 0; /* Similarly for 'lookup_name' */ |
| 23392 | bool could_be_user_defined = TRUE; /* ? Could this be a user-defined |
| 23393 | property rather than a Unicode |
| 23394 | one. */ |
| 23395 | SV * prop_definition = NULL; /* The returned definition of 'name' or NULL |
| 23396 | if an error. If it is an inversion list, |
| 23397 | it is the definition. Otherwise it is a |
| 23398 | string containing the fully qualified sub |
| 23399 | name of 'name' */ |
| 23400 | SV * fq_name = NULL; /* For user-defined properties, the fully |
| 23401 | qualified name */ |
| 23402 | bool invert_return = FALSE; /* ? Do we need to complement the result before |
| 23403 | returning it */ |
| 23404 | bool stripped_utf8_pkg = FALSE; /* Set TRUE if the input includes an |
| 23405 | explicit utf8:: package that we strip |
| 23406 | off */ |
| 23407 | /* The expansion of properties that could be either user-defined or |
| 23408 | * official unicode ones is deferred until runtime, including a marker for |
| 23409 | * those that might be in the latter category. This boolean indicates if |
| 23410 | * we've seen that marker. If not, what we're parsing can't be such an |
| 23411 | * official Unicode property whose expansion was deferred */ |
| 23412 | bool could_be_deferred_official = FALSE; |
| 23413 | |
| 23414 | PERL_ARGS_ASSERT_PARSE_UNIPROP_STRING; |
| 23415 | |
| 23416 | /* The input will be normalized into 'lookup_name' */ |
| 23417 | Newx(lookup_name, name_len, char); |
| 23418 | SAVEFREEPV(lookup_name); |
| 23419 | |
| 23420 | /* Parse the input. */ |
| 23421 | for (i = 0; i < name_len; i++) { |
| 23422 | char cur = name[i]; |
| 23423 | |
| 23424 | /* Most of the characters in the input will be of this ilk, being parts |
| 23425 | * of a name */ |
| 23426 | if (isIDCONT_A(cur)) { |
| 23427 | |
| 23428 | /* Case differences are ignored. Our lookup routine assumes |
| 23429 | * everything is lowercase, so normalize to that */ |
| 23430 | if (isUPPER_A(cur)) { |
| 23431 | lookup_name[j++] = toLOWER_A(cur); |
| 23432 | continue; |
| 23433 | } |
| 23434 | |
| 23435 | if (cur == '_') { /* Don't include these in the normalized name */ |
| 23436 | continue; |
| 23437 | } |
| 23438 | |
| 23439 | lookup_name[j++] = cur; |
| 23440 | |
| 23441 | /* The first character in a user-defined name must be of this type. |
| 23442 | * */ |
| 23443 | if (i - non_pkg_begin == 0 && ! isIDFIRST_A(cur)) { |
| 23444 | could_be_user_defined = FALSE; |
| 23445 | } |
| 23446 | |
| 23447 | continue; |
| 23448 | } |
| 23449 | |
| 23450 | /* Here, the character is not something typically in a name, But these |
| 23451 | * two types of characters (and the '_' above) can be freely ignored in |
| 23452 | * most situations. Later it may turn out we shouldn't have ignored |
| 23453 | * them, and we have to reparse, but we don't have enough information |
| 23454 | * yet to make that decision */ |
| 23455 | if (cur == '-' || isSPACE_A(cur)) { |
| 23456 | could_be_user_defined = FALSE; |
| 23457 | continue; |
| 23458 | } |
| 23459 | |
| 23460 | /* An equals sign or single colon mark the end of the first part of |
| 23461 | * the property name */ |
| 23462 | if ( cur == '=' |
| 23463 | || (cur == ':' && (i >= name_len - 1 || name[i+1] != ':'))) |
| 23464 | { |
| 23465 | lookup_name[j++] = '='; /* Treat the colon as an '=' */ |
| 23466 | equals_pos = j; /* Note where it occurred in the input */ |
| 23467 | could_be_user_defined = FALSE; |
| 23468 | break; |
| 23469 | } |
| 23470 | |
| 23471 | /* If this looks like it is a marker we inserted at compile time, |
| 23472 | * set a flag and otherwise ignore it. If it isn't in the final |
| 23473 | * position, keep it as it would have been user input. */ |
| 23474 | if ( UNLIKELY(cur == DEFERRED_COULD_BE_OFFICIAL_MARKERc) |
| 23475 | && ! deferrable |
| 23476 | && could_be_user_defined |
| 23477 | && i == name_len - 1) |
| 23478 | { |
| 23479 | name_len--; |
| 23480 | could_be_deferred_official = TRUE; |
| 23481 | continue; |
| 23482 | } |
| 23483 | |
| 23484 | /* Otherwise, this character is part of the name. */ |
| 23485 | lookup_name[j++] = cur; |
| 23486 | |
| 23487 | /* Here it isn't a single colon, so if it is a colon, it must be a |
| 23488 | * double colon */ |
| 23489 | if (cur == ':') { |
| 23490 | |
| 23491 | /* A double colon should be a package qualifier. We note its |
| 23492 | * position and continue. Note that one could have |
| 23493 | * pkg1::pkg2::...::foo |
| 23494 | * so that the position at the end of the loop will be just after |
| 23495 | * the final qualifier */ |
| 23496 | |
| 23497 | i++; |
| 23498 | non_pkg_begin = i + 1; |
| 23499 | lookup_name[j++] = ':'; |
| 23500 | lun_non_pkg_begin = j; |
| 23501 | } |
| 23502 | else { /* Only word chars (and '::') can be in a user-defined name */ |
| 23503 | could_be_user_defined = FALSE; |
| 23504 | } |
| 23505 | } /* End of parsing through the lhs of the property name (or all of it if |
| 23506 | no rhs) */ |
| 23507 | |
| 23508 | # define STRLENs(s) (sizeof("" s "") - 1) |
| 23509 | |
| 23510 | /* If there is a single package name 'utf8::', it is ambiguous. It could |
| 23511 | * be for a user-defined property, or it could be a Unicode property, as |
| 23512 | * all of them are considered to be for that package. For the purposes of |
| 23513 | * parsing the rest of the property, strip it off */ |
| 23514 | if (non_pkg_begin == STRLENs("utf8::") && memBEGINPs(name, name_len, "utf8::")) { |
| 23515 | lookup_name += STRLENs("utf8::"); |
| 23516 | j -= STRLENs("utf8::"); |
| 23517 | equals_pos -= STRLENs("utf8::"); |
| 23518 | stripped_utf8_pkg = TRUE; |
| 23519 | } |
| 23520 | |
| 23521 | /* Here, we are either done with the whole property name, if it was simple; |
| 23522 | * or are positioned just after the '=' if it is compound. */ |
| 23523 | |
| 23524 | if (equals_pos >= 0) { |
| 23525 | assert(stricter == Not_Strict); /* We shouldn't have set this yet */ |
| 23526 | |
| 23527 | /* Space immediately after the '=' is ignored */ |
| 23528 | i++; |
| 23529 | for (; i < name_len; i++) { |
| 23530 | if (! isSPACE_A(name[i])) { |
| 23531 | break; |
| 23532 | } |
| 23533 | } |
| 23534 | |
| 23535 | /* Most punctuation after the equals indicates a subpattern, like |
| 23536 | * \p{foo=/bar/} */ |
| 23537 | if ( isPUNCT_A(name[i]) |
| 23538 | && name[i] != '-' |
| 23539 | && name[i] != '+' |
| 23540 | && name[i] != '_' |
| 23541 | && name[i] != '{' |
| 23542 | /* A backslash means the real delimitter is the next character, |
| 23543 | * but it must be punctuation */ |
| 23544 | && (name[i] != '\\' || (i < name_len && isPUNCT_A(name[i+1])))) |
| 23545 | { |
| 23546 | /* Find the property. The table includes the equals sign, so we |
| 23547 | * use 'j' as-is */ |
| 23548 | table_index = match_uniprop((U8 *) lookup_name, j); |
| 23549 | if (table_index) { |
| 23550 | const char * const * prop_values |
| 23551 | = UNI_prop_value_ptrs[table_index]; |
| 23552 | REGEXP * subpattern_re; |
| 23553 | char open = name[i++]; |
| 23554 | char close; |
| 23555 | const char * pos_in_brackets; |
| 23556 | bool escaped = 0; |
| 23557 | |
| 23558 | /* Backslash => delimitter is the character following. We |
| 23559 | * already checked that it is punctuation */ |
| 23560 | if (open == '\\') { |
| 23561 | open = name[i++]; |
| 23562 | escaped = 1; |
| 23563 | } |
| 23564 | |
| 23565 | /* This data structure is constructed so that the matching |
| 23566 | * closing bracket is 3 past its matching opening. The second |
| 23567 | * set of closing is so that if the opening is something like |
| 23568 | * ']', the closing will be that as well. Something similar is |
| 23569 | * done in toke.c */ |
| 23570 | pos_in_brackets = memCHRs("([<)]>)]>", open); |
| 23571 | close = (pos_in_brackets) ? pos_in_brackets[3] : open; |
| 23572 | |
| 23573 | if ( i >= name_len |
| 23574 | || name[name_len-1] != close |
| 23575 | || (escaped && name[name_len-2] != '\\') |
| 23576 | /* Also make sure that there are enough characters. |
| 23577 | * e.g., '\\\' would show up incorrectly as legal even |
| 23578 | * though it is too short */ |
| 23579 | || (SSize_t) (name_len - i - 1 - escaped) < 0) |
| 23580 | { |
| 23581 | sv_catpvs(msg, "Unicode property wildcard not terminated"); |
| 23582 | goto append_name_to_msg; |
| 23583 | } |
| 23584 | |
| 23585 | Perl_ck_warner_d(aTHX_ |
| 23586 | packWARN(WARN_EXPERIMENTAL__UNIPROP_WILDCARDS), |
| 23587 | "The Unicode property wildcards feature is experimental"); |
| 23588 | |
| 23589 | /* Now create and compile the wildcard subpattern. Use /iaa |
| 23590 | * because nothing outside of ASCII will match, and it the |
| 23591 | * property values should all match /i. Note that when the |
| 23592 | * pattern fails to compile, our added text to the user's |
| 23593 | * pattern will be displayed to the user, which is not so |
| 23594 | * desirable. */ |
| 23595 | subpattern_re = compile_wildcard(name + i, |
| 23596 | name_len - i - 1 - escaped, |
| 23597 | TRUE /* /i */ |
| 23598 | ); |
| 23599 | |
| 23600 | /* For each legal property value, see if the supplied pattern |
| 23601 | * matches it. */ |
| 23602 | while (*prop_values) { |
| 23603 | const char * const entry = *prop_values; |
| 23604 | const Size_t len = strlen(entry); |
| 23605 | SV* entry_sv = newSVpvn_flags(entry, len, SVs_TEMP); |
| 23606 | |
| 23607 | if (execute_wildcard(subpattern_re, |
| 23608 | (char *) entry, |
| 23609 | (char *) entry + len, |
| 23610 | (char *) entry, 0, |
| 23611 | entry_sv, |
| 23612 | 0)) |
| 23613 | { /* Here, matched. Add to the returned list */ |
| 23614 | Size_t total_len = j + len; |
| 23615 | SV * sub_invlist = NULL; |
| 23616 | char * this_string; |
| 23617 | |
| 23618 | /* We know this is a legal \p{property=value}. Call |
| 23619 | * the function to return the list of code points that |
| 23620 | * match it */ |
| 23621 | Newxz(this_string, total_len + 1, char); |
| 23622 | Copy(lookup_name, this_string, j, char); |
| 23623 | my_strlcat(this_string, entry, total_len + 1); |
| 23624 | SAVEFREEPV(this_string); |
| 23625 | sub_invlist = parse_uniprop_string(this_string, |
| 23626 | total_len, |
| 23627 | is_utf8, |
| 23628 | to_fold, |
| 23629 | runtime, |
| 23630 | deferrable, |
| 23631 | user_defined_ptr, |
| 23632 | msg, |
| 23633 | level + 1); |
| 23634 | _invlist_union(prop_definition, sub_invlist, |
| 23635 | &prop_definition); |
| 23636 | } |
| 23637 | |
| 23638 | prop_values++; /* Next iteration, look at next propvalue */ |
| 23639 | } /* End of looking through property values; (the data |
| 23640 | structure is terminated by a NULL ptr) */ |
| 23641 | |
| 23642 | SvREFCNT_dec_NN(subpattern_re); |
| 23643 | |
| 23644 | if (prop_definition) { |
| 23645 | return prop_definition; |
| 23646 | } |
| 23647 | |
| 23648 | sv_catpvs(msg, "No Unicode property value wildcard matches:"); |
| 23649 | goto append_name_to_msg; |
| 23650 | } |
| 23651 | |
| 23652 | /* Here's how khw thinks we should proceed to handle the properties |
| 23653 | * not yet done: Bidi Mirroring Glyph |
| 23654 | Bidi Paired Bracket |
| 23655 | Case Folding (both full and simple) |
| 23656 | Decomposition Mapping |
| 23657 | Equivalent Unified Ideograph |
| 23658 | Name |
| 23659 | Name Alias |
| 23660 | Lowercase Mapping (both full and simple) |
| 23661 | NFKC Case Fold |
| 23662 | Titlecase Mapping (both full and simple) |
| 23663 | Uppercase Mapping (both full and simple) |
| 23664 | * Move the part that looks at the property values into a perl |
| 23665 | * script, like utf8_heavy.pl was done. This makes things somewhat |
| 23666 | * easier, but most importantly, it avoids always adding all these |
| 23667 | * strings to the memory usage when the feature is little-used. |
| 23668 | * |
| 23669 | * The property values would all be concatenated into a single |
| 23670 | * string per property with each value on a separate line, and the |
| 23671 | * code point it's for on alternating lines. Then we match the |
| 23672 | * user's input pattern m//mg, without having to worry about their |
| 23673 | * uses of '^' and '$'. Only the values that aren't the default |
| 23674 | * would be in the strings. Code points would be in UTF-8. The |
| 23675 | * search pattern that we would construct would look like |
| 23676 | * (?: \n (code-point_re) \n (?aam: user-re ) \n ) |
| 23677 | * And so $1 would contain the code point that matched the user-re. |
| 23678 | * For properties where the default is the code point itself, such |
| 23679 | * as any of the case changing mappings, the string would otherwise |
| 23680 | * consist of all Unicode code points in UTF-8 strung together. |
| 23681 | * This would be impractical. So instead, examine their compiled |
| 23682 | * pattern, looking at the ssc. If none, reject the pattern as an |
| 23683 | * error. Otherwise run the pattern against every code point in |
| 23684 | * the ssc. The ssc is kind of like tr18's 3.9 Possible Match Sets |
| 23685 | * And it might be good to create an API to return the ssc. |
| 23686 | * |
| 23687 | * For the name properties, a new function could be created in |
| 23688 | * charnames which essentially does the same thing as above, |
| 23689 | * sharing Name.pl with the other charname functions. Don't know |
| 23690 | * about loose name matching, or algorithmically determined names. |
| 23691 | * Decomposition.pl similarly. |
| 23692 | * |
| 23693 | * It might be that a new pattern modifier would have to be |
| 23694 | * created, like /t for resTricTed, which changed the behavior of |
| 23695 | * some constructs in their subpattern, like \A. */ |
| 23696 | } /* End of is a wildcard subppattern */ |
| 23697 | |
| 23698 | /* \p{name=...} is handled specially. Instead of using the normal |
| 23699 | * mechanism involving charclass_invlists.h, it uses _charnames.pm |
| 23700 | * which has the necessary (huge) data accessible to it, and which |
| 23701 | * doesn't get loaded unless necessary. The legal syntax for names is |
| 23702 | * somewhat different than other properties due both to the vagaries of |
| 23703 | * a few outlier official names, and the fact that only a few ASCII |
| 23704 | * characters are permitted in them */ |
| 23705 | if ( memEQs(lookup_name, j - 1, "name") |
| 23706 | || memEQs(lookup_name, j - 1, "na")) |
| 23707 | { |
| 23708 | dSP; |
| 23709 | HV * table; |
| 23710 | SV * character; |
| 23711 | const char * error_msg; |
| 23712 | CV* lookup_loose; |
| 23713 | SV * character_name; |
| 23714 | STRLEN character_len; |
| 23715 | UV cp; |
| 23716 | |
| 23717 | stricter = As_Is; |
| 23718 | |
| 23719 | /* Since the RHS (after skipping initial space) is passed unchanged |
| 23720 | * to charnames, and there are different criteria for what are |
| 23721 | * legal characters in the name, just parse it here. A character |
| 23722 | * name must begin with an ASCII alphabetic */ |
| 23723 | if (! isALPHA(name[i])) { |
| 23724 | goto failed; |
| 23725 | } |
| 23726 | lookup_name[j++] = name[i]; |
| 23727 | |
| 23728 | for (++i; i < name_len; i++) { |
| 23729 | /* Official names can only be in the ASCII range, and only |
| 23730 | * certain characters */ |
| 23731 | if (! isASCII(name[i]) || ! isCHARNAME_CONT(name[i])) { |
| 23732 | goto failed; |
| 23733 | } |
| 23734 | lookup_name[j++] = name[i]; |
| 23735 | } |
| 23736 | |
| 23737 | /* Finished parsing, save the name into an SV */ |
| 23738 | character_name = newSVpvn(lookup_name + equals_pos, j - equals_pos); |
| 23739 | |
| 23740 | /* Make sure _charnames is loaded. (The parameters give context |
| 23741 | * for any errors generated */ |
| 23742 | table = load_charnames(character_name, name, name_len, &error_msg); |
| 23743 | if (table == NULL) { |
| 23744 | sv_catpv(msg, error_msg); |
| 23745 | goto append_name_to_msg; |
| 23746 | } |
| 23747 | |
| 23748 | lookup_loose = get_cv("_charnames::_loose_regcomp_lookup", 0); |
| 23749 | if (! lookup_loose) { |
| 23750 | Perl_croak(aTHX_ |
| 23751 | "panic: Can't find '_charnames::_loose_regcomp_lookup"); |
| 23752 | } |
| 23753 | |
| 23754 | PUSHSTACKi(PERLSI_OVERLOAD); |
| 23755 | ENTER ; |
| 23756 | SAVETMPS; |
| 23757 | save_re_context(); |
| 23758 | |
| 23759 | PUSHMARK(SP) ; |
| 23760 | XPUSHs(character_name); |
| 23761 | PUTBACK; |
| 23762 | call_sv(MUTABLE_SV(lookup_loose), G_SCALAR); |
| 23763 | |
| 23764 | SPAGAIN ; |
| 23765 | |
| 23766 | character = POPs; |
| 23767 | SvREFCNT_inc_simple_void_NN(character); |
| 23768 | |
| 23769 | PUTBACK ; |
| 23770 | FREETMPS ; |
| 23771 | LEAVE ; |
| 23772 | POPSTACK; |
| 23773 | |
| 23774 | if (! SvOK(character)) { |
| 23775 | goto failed; |
| 23776 | } |
| 23777 | |
| 23778 | cp = valid_utf8_to_uvchr((U8 *) SvPVX(character), &character_len); |
| 23779 | if (character_len < SvCUR(character)) { |
| 23780 | goto failed; |
| 23781 | } |
| 23782 | |
| 23783 | prop_definition = add_cp_to_invlist(NULL, cp); |
| 23784 | return prop_definition; |
| 23785 | } |
| 23786 | |
| 23787 | /* Certain properties whose values are numeric need special handling. |
| 23788 | * They may optionally be prefixed by 'is'. Ignore that prefix for the |
| 23789 | * purposes of checking if this is one of those properties */ |
| 23790 | if (memBEGINPs(lookup_name, j, "is")) { |
| 23791 | lookup_offset = 2; |
| 23792 | } |
| 23793 | |
| 23794 | /* Then check if it is one of these specially-handled properties. The |
| 23795 | * possibilities are hard-coded because easier this way, and the list |
| 23796 | * is unlikely to change. |
| 23797 | * |
| 23798 | * All numeric value type properties are of this ilk, and are also |
| 23799 | * special in a different way later on. So find those first. There |
| 23800 | * are several numeric value type properties in the Unihan DB (which is |
| 23801 | * unlikely to be compiled with perl, but we handle it here in case it |
| 23802 | * does get compiled). They all end with 'numeric'. The interiors |
| 23803 | * aren't checked for the precise property. This would stop working if |
| 23804 | * a cjk property were to be created that ended with 'numeric' and |
| 23805 | * wasn't a numeric type */ |
| 23806 | is_nv_type = memEQs(lookup_name + lookup_offset, |
| 23807 | j - 1 - lookup_offset, "numericvalue") |
| 23808 | || memEQs(lookup_name + lookup_offset, |
| 23809 | j - 1 - lookup_offset, "nv") |
| 23810 | || ( memENDPs(lookup_name + lookup_offset, |
| 23811 | j - 1 - lookup_offset, "numeric") |
| 23812 | && ( memBEGINPs(lookup_name + lookup_offset, |
| 23813 | j - 1 - lookup_offset, "cjk") |
| 23814 | || memBEGINPs(lookup_name + lookup_offset, |
| 23815 | j - 1 - lookup_offset, "k"))); |
| 23816 | if ( is_nv_type |
| 23817 | || memEQs(lookup_name + lookup_offset, |
| 23818 | j - 1 - lookup_offset, "canonicalcombiningclass") |
| 23819 | || memEQs(lookup_name + lookup_offset, |
| 23820 | j - 1 - lookup_offset, "ccc") |
| 23821 | || memEQs(lookup_name + lookup_offset, |
| 23822 | j - 1 - lookup_offset, "age") |
| 23823 | || memEQs(lookup_name + lookup_offset, |
| 23824 | j - 1 - lookup_offset, "in") |
| 23825 | || memEQs(lookup_name + lookup_offset, |
| 23826 | j - 1 - lookup_offset, "presentin")) |
| 23827 | { |
| 23828 | unsigned int k; |
| 23829 | |
| 23830 | /* Since the stuff after the '=' is a number, we can't throw away |
| 23831 | * '-' willy-nilly, as those could be a minus sign. Other stricter |
| 23832 | * rules also apply. However, these properties all can have the |
| 23833 | * rhs not be a number, in which case they contain at least one |
| 23834 | * alphabetic. In those cases, the stricter rules don't apply. |
| 23835 | * But the numeric type properties can have the alphas [Ee] to |
| 23836 | * signify an exponent, and it is still a number with stricter |
| 23837 | * rules. So look for an alpha that signifies not-strict */ |
| 23838 | stricter = Strict; |
| 23839 | for (k = i; k < name_len; k++) { |
| 23840 | if ( isALPHA_A(name[k]) |
| 23841 | && (! is_nv_type || ! isALPHA_FOLD_EQ(name[k], 'E'))) |
| 23842 | { |
| 23843 | stricter = Not_Strict; |
| 23844 | break; |
| 23845 | } |
| 23846 | } |
| 23847 | } |
| 23848 | |
| 23849 | if (stricter) { |
| 23850 | |
| 23851 | /* A number may have a leading '+' or '-'. The latter is retained |
| 23852 | * */ |
| 23853 | if (name[i] == '+') { |
| 23854 | i++; |
| 23855 | } |
| 23856 | else if (name[i] == '-') { |
| 23857 | lookup_name[j++] = '-'; |
| 23858 | i++; |
| 23859 | } |
| 23860 | |
| 23861 | /* Skip leading zeros including single underscores separating the |
| 23862 | * zeros, or between the final leading zero and the first other |
| 23863 | * digit */ |
| 23864 | for (; i < name_len - 1; i++) { |
| 23865 | if ( name[i] != '0' |
| 23866 | && (name[i] != '_' || ! isDIGIT_A(name[i+1]))) |
| 23867 | { |
| 23868 | break; |
| 23869 | } |
| 23870 | } |
| 23871 | } |
| 23872 | } |
| 23873 | else { /* No '=' */ |
| 23874 | |
| 23875 | /* Only a few properties without an '=' should be parsed with stricter |
| 23876 | * rules. The list is unlikely to change. */ |
| 23877 | if ( memBEGINPs(lookup_name, j, "perl") |
| 23878 | && memNEs(lookup_name + 4, j - 4, "space") |
| 23879 | && memNEs(lookup_name + 4, j - 4, "word")) |
| 23880 | { |
| 23881 | stricter = Strict; |
| 23882 | |
| 23883 | /* We set the inputs back to 0 and the code below will reparse, |
| 23884 | * using strict */ |
| 23885 | i = j = 0; |
| 23886 | } |
| 23887 | } |
| 23888 | |
| 23889 | /* Here, we have either finished the property, or are positioned to parse |
| 23890 | * the remainder, and we know if stricter rules apply. Finish out, if not |
| 23891 | * already done */ |
| 23892 | for (; i < name_len; i++) { |
| 23893 | char cur = name[i]; |
| 23894 | |
| 23895 | /* In all instances, case differences are ignored, and we normalize to |
| 23896 | * lowercase */ |
| 23897 | if (isUPPER_A(cur)) { |
| 23898 | lookup_name[j++] = toLOWER(cur); |
| 23899 | continue; |
| 23900 | } |
| 23901 | |
| 23902 | /* An underscore is skipped, but not under strict rules unless it |
| 23903 | * separates two digits */ |
| 23904 | if (cur == '_') { |
| 23905 | if ( stricter |
| 23906 | && ( i == 0 || (int) i == equals_pos || i == name_len- 1 |
| 23907 | || ! isDIGIT_A(name[i-1]) || ! isDIGIT_A(name[i+1]))) |
| 23908 | { |
| 23909 | lookup_name[j++] = '_'; |
| 23910 | } |
| 23911 | continue; |
| 23912 | } |
| 23913 | |
| 23914 | /* Hyphens are skipped except under strict */ |
| 23915 | if (cur == '-' && ! stricter) { |
| 23916 | continue; |
| 23917 | } |
| 23918 | |
| 23919 | /* XXX Bug in documentation. It says white space skipped adjacent to |
| 23920 | * non-word char. Maybe we should, but shouldn't skip it next to a dot |
| 23921 | * in a number */ |
| 23922 | if (isSPACE_A(cur) && ! stricter) { |
| 23923 | continue; |
| 23924 | } |
| 23925 | |
| 23926 | lookup_name[j++] = cur; |
| 23927 | |
| 23928 | /* Unless this is a non-trailing slash, we are done with it */ |
| 23929 | if (i >= name_len - 1 || cur != '/') { |
| 23930 | continue; |
| 23931 | } |
| 23932 | |
| 23933 | slash_pos = j; |
| 23934 | |
| 23935 | /* A slash in the 'numeric value' property indicates that what follows |
| 23936 | * is a denominator. It can have a leading '+' and '0's that should be |
| 23937 | * skipped. But we have never allowed a negative denominator, so treat |
| 23938 | * a minus like every other character. (No need to rule out a second |
| 23939 | * '/', as that won't match anything anyway */ |
| 23940 | if (is_nv_type) { |
| 23941 | i++; |
| 23942 | if (i < name_len && name[i] == '+') { |
| 23943 | i++; |
| 23944 | } |
| 23945 | |
| 23946 | /* Skip leading zeros including underscores separating digits */ |
| 23947 | for (; i < name_len - 1; i++) { |
| 23948 | if ( name[i] != '0' |
| 23949 | && (name[i] != '_' || ! isDIGIT_A(name[i+1]))) |
| 23950 | { |
| 23951 | break; |
| 23952 | } |
| 23953 | } |
| 23954 | |
| 23955 | /* Store the first real character in the denominator */ |
| 23956 | if (i < name_len) { |
| 23957 | lookup_name[j++] = name[i]; |
| 23958 | } |
| 23959 | } |
| 23960 | } |
| 23961 | |
| 23962 | /* Here are completely done parsing the input 'name', and 'lookup_name' |
| 23963 | * contains a copy, normalized. |
| 23964 | * |
| 23965 | * This special case is grandfathered in: 'L_' and 'GC=L_' are accepted and |
| 23966 | * different from without the underscores. */ |
| 23967 | if ( ( UNLIKELY(memEQs(lookup_name, j, "l")) |
| 23968 | || UNLIKELY(memEQs(lookup_name, j, "gc=l"))) |
| 23969 | && UNLIKELY(name[name_len-1] == '_')) |
| 23970 | { |
| 23971 | lookup_name[j++] = '&'; |
| 23972 | } |
| 23973 | |
| 23974 | /* If the original input began with 'In' or 'Is', it could be a subroutine |
| 23975 | * call to a user-defined property instead of a Unicode property name. */ |
| 23976 | if ( name_len - non_pkg_begin > 2 |
| 23977 | && name[non_pkg_begin+0] == 'I' |
| 23978 | && (name[non_pkg_begin+1] == 'n' || name[non_pkg_begin+1] == 's')) |
| 23979 | { |
| 23980 | /* Names that start with In have different characterstics than those |
| 23981 | * that start with Is */ |
| 23982 | if (name[non_pkg_begin+1] == 's') { |
| 23983 | starts_with_Is = TRUE; |
| 23984 | } |
| 23985 | } |
| 23986 | else { |
| 23987 | could_be_user_defined = FALSE; |
| 23988 | } |
| 23989 | |
| 23990 | if (could_be_user_defined) { |
| 23991 | CV* user_sub; |
| 23992 | |
| 23993 | /* If the user defined property returns the empty string, it could |
| 23994 | * easily be because the pattern is being compiled before the data it |
| 23995 | * actually needs to compile is available. This could be argued to be |
| 23996 | * a bug in the perl code, but this is a change of behavior for Perl, |
| 23997 | * so we handle it. This means that intentionally returning nothing |
| 23998 | * will not be resolved until runtime */ |
| 23999 | bool empty_return = FALSE; |
| 24000 | |
| 24001 | /* Here, the name could be for a user defined property, which are |
| 24002 | * implemented as subs. */ |
| 24003 | user_sub = get_cvn_flags(name, name_len, 0); |
| 24004 | if (! user_sub) { |
| 24005 | |
| 24006 | /* Here, the property name could be a user-defined one, but there |
| 24007 | * is no subroutine to handle it (as of now). Defer handling it |
| 24008 | * until runtime. Otherwise, a block defined by Unicode in a later |
| 24009 | * release would get the synonym InFoo added for it, and existing |
| 24010 | * code that used that name would suddenly break if it referred to |
| 24011 | * the property before the sub was declared. See [perl #134146] */ |
| 24012 | if (deferrable) { |
| 24013 | goto definition_deferred; |
| 24014 | } |
| 24015 | |
| 24016 | /* Here, we are at runtime, and didn't find the user property. It |
| 24017 | * could be an official property, but only if no package was |
| 24018 | * specified, or just the utf8:: package. */ |
| 24019 | if (could_be_deferred_official) { |
| 24020 | lookup_name += lun_non_pkg_begin; |
| 24021 | j -= lun_non_pkg_begin; |
| 24022 | } |
| 24023 | else if (! stripped_utf8_pkg) { |
| 24024 | goto unknown_user_defined; |
| 24025 | } |
| 24026 | |
| 24027 | /* Drop down to look up in the official properties */ |
| 24028 | } |
| 24029 | else { |
| 24030 | const char insecure[] = "Insecure user-defined property"; |
| 24031 | |
| 24032 | /* Here, there is a sub by the correct name. Normally we call it |
| 24033 | * to get the property definition */ |
| 24034 | dSP; |
| 24035 | SV * user_sub_sv = MUTABLE_SV(user_sub); |
| 24036 | SV * error; /* Any error returned by calling 'user_sub' */ |
| 24037 | SV * key; /* The key into the hash of user defined sub names |
| 24038 | */ |
| 24039 | SV * placeholder; |
| 24040 | SV ** saved_user_prop_ptr; /* Hash entry for this property */ |
| 24041 | |
| 24042 | /* How many times to retry when another thread is in the middle of |
| 24043 | * expanding the same definition we want */ |
| 24044 | PERL_INT_FAST8_T retry_countdown = 10; |
| 24045 | |
| 24046 | DECLARATION_FOR_GLOBAL_CONTEXT; |
| 24047 | |
| 24048 | /* If we get here, we know this property is user-defined */ |
| 24049 | *user_defined_ptr = TRUE; |
| 24050 | |
| 24051 | /* We refuse to call a potentially tainted subroutine; returning an |
| 24052 | * error instead */ |
| 24053 | if (TAINT_get) { |
| 24054 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24055 | sv_catpvn(msg, insecure, sizeof(insecure) - 1); |
| 24056 | goto append_name_to_msg; |
| 24057 | } |
| 24058 | |
| 24059 | /* In principal, we only call each subroutine property definition |
| 24060 | * once during the life of the program. This guarantees that the |
| 24061 | * property definition never changes. The results of the single |
| 24062 | * sub call are stored in a hash, which is used instead for future |
| 24063 | * references to this property. The property definition is thus |
| 24064 | * immutable. But, to allow the user to have a /i-dependent |
| 24065 | * definition, we call the sub once for non-/i, and once for /i, |
| 24066 | * should the need arise, passing the /i status as a parameter. |
| 24067 | * |
| 24068 | * We start by constructing the hash key name, consisting of the |
| 24069 | * fully qualified subroutine name, preceded by the /i status, so |
| 24070 | * that there is a key for /i and a different key for non-/i */ |
| 24071 | key = newSVpvn(((to_fold) ? "1" : "0"), 1); |
| 24072 | fq_name = S_get_fq_name(aTHX_ name, name_len, is_utf8, |
| 24073 | non_pkg_begin != 0); |
| 24074 | sv_catsv(key, fq_name); |
| 24075 | sv_2mortal(key); |
| 24076 | |
| 24077 | /* We only call the sub once throughout the life of the program |
| 24078 | * (with the /i, non-/i exception noted above). That means the |
| 24079 | * hash must be global and accessible to all threads. It is |
| 24080 | * created at program start-up, before any threads are created, so |
| 24081 | * is accessible to all children. But this creates some |
| 24082 | * complications. |
| 24083 | * |
| 24084 | * 1) The keys can't be shared, or else problems arise; sharing is |
| 24085 | * turned off at hash creation time |
| 24086 | * 2) All SVs in it are there for the remainder of the life of the |
| 24087 | * program, and must be created in the same interpreter context |
| 24088 | * as the hash, or else they will be freed from the wrong pool |
| 24089 | * at global destruction time. This is handled by switching to |
| 24090 | * the hash's context to create each SV going into it, and then |
| 24091 | * immediately switching back |
| 24092 | * 3) All accesses to the hash must be controlled by a mutex, to |
| 24093 | * prevent two threads from getting an unstable state should |
| 24094 | * they simultaneously be accessing it. The code below is |
| 24095 | * crafted so that the mutex is locked whenever there is an |
| 24096 | * access and unlocked only when the next stable state is |
| 24097 | * achieved. |
| 24098 | * |
| 24099 | * The hash stores either the definition of the property if it was |
| 24100 | * valid, or, if invalid, the error message that was raised. We |
| 24101 | * use the type of SV to distinguish. |
| 24102 | * |
| 24103 | * There's also the need to guard against the definition expansion |
| 24104 | * from infinitely recursing. This is handled by storing the aTHX |
| 24105 | * of the expanding thread during the expansion. Again the SV type |
| 24106 | * is used to distinguish this from the other two cases. If we |
| 24107 | * come to here and the hash entry for this property is our aTHX, |
| 24108 | * it means we have recursed, and the code assumes that we would |
| 24109 | * infinitely recurse, so instead stops and raises an error. |
| 24110 | * (Any recursion has always been treated as infinite recursion in |
| 24111 | * this feature.) |
| 24112 | * |
| 24113 | * If instead, the entry is for a different aTHX, it means that |
| 24114 | * that thread has gotten here first, and hasn't finished expanding |
| 24115 | * the definition yet. We just have to wait until it is done. We |
| 24116 | * sleep and retry a few times, returning an error if the other |
| 24117 | * thread doesn't complete. */ |
| 24118 | |
| 24119 | re_fetch: |
| 24120 | USER_PROP_MUTEX_LOCK; |
| 24121 | |
| 24122 | /* If we have an entry for this key, the subroutine has already |
| 24123 | * been called once with this /i status. */ |
| 24124 | saved_user_prop_ptr = hv_fetch(PL_user_def_props, |
| 24125 | SvPVX(key), SvCUR(key), 0); |
| 24126 | if (saved_user_prop_ptr) { |
| 24127 | |
| 24128 | /* If the saved result is an inversion list, it is the valid |
| 24129 | * definition of this property */ |
| 24130 | if (is_invlist(*saved_user_prop_ptr)) { |
| 24131 | prop_definition = *saved_user_prop_ptr; |
| 24132 | |
| 24133 | /* The SV in the hash won't be removed until global |
| 24134 | * destruction, so it is stable and we can unlock */ |
| 24135 | USER_PROP_MUTEX_UNLOCK; |
| 24136 | |
| 24137 | /* The caller shouldn't try to free this SV */ |
| 24138 | return prop_definition; |
| 24139 | } |
| 24140 | |
| 24141 | /* Otherwise, if it is a string, it is the error message |
| 24142 | * that was returned when we first tried to evaluate this |
| 24143 | * property. Fail, and append the message */ |
| 24144 | if (SvPOK(*saved_user_prop_ptr)) { |
| 24145 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24146 | sv_catsv(msg, *saved_user_prop_ptr); |
| 24147 | |
| 24148 | /* The SV in the hash won't be removed until global |
| 24149 | * destruction, so it is stable and we can unlock */ |
| 24150 | USER_PROP_MUTEX_UNLOCK; |
| 24151 | |
| 24152 | return NULL; |
| 24153 | } |
| 24154 | |
| 24155 | assert(SvIOK(*saved_user_prop_ptr)); |
| 24156 | |
| 24157 | /* Here, we have an unstable entry in the hash. Either another |
| 24158 | * thread is in the middle of expanding the property's |
| 24159 | * definition, or we are ourselves recursing. We use the aTHX |
| 24160 | * in it to distinguish */ |
| 24161 | if (SvIV(*saved_user_prop_ptr) != PTR2IV(CUR_CONTEXT)) { |
| 24162 | |
| 24163 | /* Here, it's another thread doing the expanding. We've |
| 24164 | * looked as much as we are going to at the contents of the |
| 24165 | * hash entry. It's safe to unlock. */ |
| 24166 | USER_PROP_MUTEX_UNLOCK; |
| 24167 | |
| 24168 | /* Retry a few times */ |
| 24169 | if (retry_countdown-- > 0) { |
| 24170 | PerlProc_sleep(1); |
| 24171 | goto re_fetch; |
| 24172 | } |
| 24173 | |
| 24174 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24175 | sv_catpvs(msg, "Timeout waiting for another thread to " |
| 24176 | "define"); |
| 24177 | goto append_name_to_msg; |
| 24178 | } |
| 24179 | |
| 24180 | /* Here, we are recursing; don't dig any deeper */ |
| 24181 | USER_PROP_MUTEX_UNLOCK; |
| 24182 | |
| 24183 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24184 | sv_catpvs(msg, |
| 24185 | "Infinite recursion in user-defined property"); |
| 24186 | goto append_name_to_msg; |
| 24187 | } |
| 24188 | |
| 24189 | /* Here, this thread has exclusive control, and there is no entry |
| 24190 | * for this property in the hash. So we have the go ahead to |
| 24191 | * expand the definition ourselves. */ |
| 24192 | |
| 24193 | PUSHSTACKi(PERLSI_MAGIC); |
| 24194 | ENTER; |
| 24195 | |
| 24196 | /* Create a temporary placeholder in the hash to detect recursion |
| 24197 | * */ |
| 24198 | SWITCH_TO_GLOBAL_CONTEXT; |
| 24199 | placeholder= newSVuv(PTR2IV(ORIGINAL_CONTEXT)); |
| 24200 | (void) hv_store_ent(PL_user_def_props, key, placeholder, 0); |
| 24201 | RESTORE_CONTEXT; |
| 24202 | |
| 24203 | /* Now that we have a placeholder, we can let other threads |
| 24204 | * continue */ |
| 24205 | USER_PROP_MUTEX_UNLOCK; |
| 24206 | |
| 24207 | /* Make sure the placeholder always gets destroyed */ |
| 24208 | SAVEDESTRUCTOR_X(S_delete_recursion_entry, SvPVX(key)); |
| 24209 | |
| 24210 | PUSHMARK(SP); |
| 24211 | SAVETMPS; |
| 24212 | |
| 24213 | /* Call the user's function, with the /i status as a parameter. |
| 24214 | * Note that we have gone to a lot of trouble to keep this call |
| 24215 | * from being within the locked mutex region. */ |
| 24216 | XPUSHs(boolSV(to_fold)); |
| 24217 | PUTBACK; |
| 24218 | |
| 24219 | /* The following block was taken from swash_init(). Presumably |
| 24220 | * they apply to here as well, though we no longer use a swash -- |
| 24221 | * khw */ |
| 24222 | SAVEHINTS(); |
| 24223 | save_re_context(); |
| 24224 | /* We might get here via a subroutine signature which uses a utf8 |
| 24225 | * parameter name, at which point PL_subname will have been set |
| 24226 | * but not yet used. */ |
| 24227 | save_item(PL_subname); |
| 24228 | |
| 24229 | /* G_SCALAR guarantees a single return value */ |
| 24230 | (void) call_sv(user_sub_sv, G_EVAL|G_SCALAR); |
| 24231 | |
| 24232 | SPAGAIN; |
| 24233 | |
| 24234 | error = ERRSV; |
| 24235 | if (TAINT_get || SvTRUE(error)) { |
| 24236 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24237 | if (SvTRUE(error)) { |
| 24238 | sv_catpvs(msg, "Error \""); |
| 24239 | sv_catsv(msg, error); |
| 24240 | sv_catpvs(msg, "\""); |
| 24241 | } |
| 24242 | if (TAINT_get) { |
| 24243 | if (SvTRUE(error)) sv_catpvs(msg, "; "); |
| 24244 | sv_catpvn(msg, insecure, sizeof(insecure) - 1); |
| 24245 | } |
| 24246 | |
| 24247 | if (name_len > 0) { |
| 24248 | sv_catpvs(msg, " in expansion of "); |
| 24249 | Perl_sv_catpvf(aTHX_ msg, "%" UTF8f, UTF8fARG(is_utf8, |
| 24250 | name_len, |
| 24251 | name)); |
| 24252 | } |
| 24253 | |
| 24254 | (void) POPs; |
| 24255 | prop_definition = NULL; |
| 24256 | } |
| 24257 | else { |
| 24258 | SV * contents = POPs; |
| 24259 | |
| 24260 | /* The contents is supposed to be the expansion of the property |
| 24261 | * definition. If the definition is deferrable, and we got an |
| 24262 | * empty string back, set a flag to later defer it (after clean |
| 24263 | * up below). */ |
| 24264 | if ( deferrable |
| 24265 | && (! SvPOK(contents) || SvCUR(contents) == 0)) |
| 24266 | { |
| 24267 | empty_return = TRUE; |
| 24268 | } |
| 24269 | else { /* Otherwise, call a function to check for valid syntax, |
| 24270 | and handle it */ |
| 24271 | |
| 24272 | prop_definition = handle_user_defined_property( |
| 24273 | name, name_len, |
| 24274 | is_utf8, to_fold, runtime, |
| 24275 | deferrable, |
| 24276 | contents, user_defined_ptr, |
| 24277 | msg, |
| 24278 | level); |
| 24279 | } |
| 24280 | } |
| 24281 | |
| 24282 | /* Here, we have the results of the expansion. Delete the |
| 24283 | * placeholder, and if the definition is now known, replace it with |
| 24284 | * that definition. We need exclusive access to the hash, and we |
| 24285 | * can't let anyone else in, between when we delete the placeholder |
| 24286 | * and add the permanent entry */ |
| 24287 | USER_PROP_MUTEX_LOCK; |
| 24288 | |
| 24289 | S_delete_recursion_entry(aTHX_ SvPVX(key)); |
| 24290 | |
| 24291 | if ( ! empty_return |
| 24292 | && (! prop_definition || is_invlist(prop_definition))) |
| 24293 | { |
| 24294 | /* If we got success we use the inversion list defining the |
| 24295 | * property; otherwise use the error message */ |
| 24296 | SWITCH_TO_GLOBAL_CONTEXT; |
| 24297 | (void) hv_store_ent(PL_user_def_props, |
| 24298 | key, |
| 24299 | ((prop_definition) |
| 24300 | ? newSVsv(prop_definition) |
| 24301 | : newSVsv(msg)), |
| 24302 | 0); |
| 24303 | RESTORE_CONTEXT; |
| 24304 | } |
| 24305 | |
| 24306 | /* All done, and the hash now has a permanent entry for this |
| 24307 | * property. Give up exclusive control */ |
| 24308 | USER_PROP_MUTEX_UNLOCK; |
| 24309 | |
| 24310 | FREETMPS; |
| 24311 | LEAVE; |
| 24312 | POPSTACK; |
| 24313 | |
| 24314 | if (empty_return) { |
| 24315 | goto definition_deferred; |
| 24316 | } |
| 24317 | |
| 24318 | if (prop_definition) { |
| 24319 | |
| 24320 | /* If the definition is for something not known at this time, |
| 24321 | * we toss it, and go return the main property name, as that's |
| 24322 | * the one the user will be aware of */ |
| 24323 | if (! is_invlist(prop_definition)) { |
| 24324 | SvREFCNT_dec_NN(prop_definition); |
| 24325 | goto definition_deferred; |
| 24326 | } |
| 24327 | |
| 24328 | sv_2mortal(prop_definition); |
| 24329 | } |
| 24330 | |
| 24331 | /* And return */ |
| 24332 | return prop_definition; |
| 24333 | |
| 24334 | } /* End of calling the subroutine for the user-defined property */ |
| 24335 | } /* End of it could be a user-defined property */ |
| 24336 | |
| 24337 | /* Here it wasn't a user-defined property that is known at this time. See |
| 24338 | * if it is a Unicode property */ |
| 24339 | |
| 24340 | lookup_len = j; /* This is a more mnemonic name than 'j' */ |
| 24341 | |
| 24342 | /* Get the index into our pointer table of the inversion list corresponding |
| 24343 | * to the property */ |
| 24344 | table_index = match_uniprop((U8 *) lookup_name, lookup_len); |
| 24345 | |
| 24346 | /* If it didn't find the property ... */ |
| 24347 | if (table_index == 0) { |
| 24348 | |
| 24349 | /* Try again stripping off any initial 'Is'. This is because we |
| 24350 | * promise that an initial Is is optional. The same isn't true of |
| 24351 | * names that start with 'In'. Those can match only blocks, and the |
| 24352 | * lookup table already has those accounted for. */ |
| 24353 | if (starts_with_Is) { |
| 24354 | lookup_name += 2; |
| 24355 | lookup_len -= 2; |
| 24356 | equals_pos -= 2; |
| 24357 | slash_pos -= 2; |
| 24358 | |
| 24359 | table_index = match_uniprop((U8 *) lookup_name, lookup_len); |
| 24360 | } |
| 24361 | |
| 24362 | if (table_index == 0) { |
| 24363 | char * canonical; |
| 24364 | |
| 24365 | /* Here, we didn't find it. If not a numeric type property, and |
| 24366 | * can't be a user-defined one, it isn't a legal property */ |
| 24367 | if (! is_nv_type) { |
| 24368 | if (! could_be_user_defined) { |
| 24369 | goto failed; |
| 24370 | } |
| 24371 | |
| 24372 | /* Here, the property name is legal as a user-defined one. At |
| 24373 | * compile time, it might just be that the subroutine for that |
| 24374 | * property hasn't been encountered yet, but at runtime, it's |
| 24375 | * an error to try to use an undefined one */ |
| 24376 | if (! deferrable) { |
| 24377 | goto unknown_user_defined;; |
| 24378 | } |
| 24379 | |
| 24380 | goto definition_deferred; |
| 24381 | } /* End of isn't a numeric type property */ |
| 24382 | |
| 24383 | /* The numeric type properties need more work to decide. What we |
| 24384 | * do is make sure we have the number in canonical form and look |
| 24385 | * that up. */ |
| 24386 | |
| 24387 | if (slash_pos < 0) { /* No slash */ |
| 24388 | |
| 24389 | /* When it isn't a rational, take the input, convert it to a |
| 24390 | * NV, then create a canonical string representation of that |
| 24391 | * NV. */ |
| 24392 | |
| 24393 | NV value; |
| 24394 | SSize_t value_len = lookup_len - equals_pos; |
| 24395 | |
| 24396 | /* Get the value */ |
| 24397 | if ( value_len <= 0 |
| 24398 | || my_atof3(lookup_name + equals_pos, &value, |
| 24399 | value_len) |
| 24400 | != lookup_name + lookup_len) |
| 24401 | { |
| 24402 | goto failed; |
| 24403 | } |
| 24404 | |
| 24405 | /* If the value is an integer, the canonical value is integral |
| 24406 | * */ |
| 24407 | if (Perl_ceil(value) == value) { |
| 24408 | canonical = Perl_form(aTHX_ "%.*s%.0" NVff, |
| 24409 | equals_pos, lookup_name, value); |
| 24410 | } |
| 24411 | else { /* Otherwise, it is %e with a known precision */ |
| 24412 | char * exp_ptr; |
| 24413 | |
| 24414 | canonical = Perl_form(aTHX_ "%.*s%.*" NVef, |
| 24415 | equals_pos, lookup_name, |
| 24416 | PL_E_FORMAT_PRECISION, value); |
| 24417 | |
| 24418 | /* The exponent generated is expecting two digits, whereas |
| 24419 | * %e on some systems will generate three. Remove leading |
| 24420 | * zeros in excess of 2 from the exponent. We start |
| 24421 | * looking for them after the '=' */ |
| 24422 | exp_ptr = strchr(canonical + equals_pos, 'e'); |
| 24423 | if (exp_ptr) { |
| 24424 | char * cur_ptr = exp_ptr + 2; /* past the 'e[+-]' */ |
| 24425 | SSize_t excess_exponent_len = strlen(cur_ptr) - 2; |
| 24426 | |
| 24427 | assert(*(cur_ptr - 1) == '-' || *(cur_ptr - 1) == '+'); |
| 24428 | |
| 24429 | if (excess_exponent_len > 0) { |
| 24430 | SSize_t leading_zeros = strspn(cur_ptr, "0"); |
| 24431 | SSize_t excess_leading_zeros |
| 24432 | = MIN(leading_zeros, excess_exponent_len); |
| 24433 | if (excess_leading_zeros > 0) { |
| 24434 | Move(cur_ptr + excess_leading_zeros, |
| 24435 | cur_ptr, |
| 24436 | strlen(cur_ptr) - excess_leading_zeros |
| 24437 | + 1, /* Copy the NUL as well */ |
| 24438 | char); |
| 24439 | } |
| 24440 | } |
| 24441 | } |
| 24442 | } |
| 24443 | } |
| 24444 | else { /* Has a slash. Create a rational in canonical form */ |
| 24445 | UV numerator, denominator, gcd, trial; |
| 24446 | const char * end_ptr; |
| 24447 | const char * sign = ""; |
| 24448 | |
| 24449 | /* We can't just find the numerator, denominator, and do the |
| 24450 | * division, then use the method above, because that is |
| 24451 | * inexact. And the input could be a rational that is within |
| 24452 | * epsilon (given our precision) of a valid rational, and would |
| 24453 | * then incorrectly compare valid. |
| 24454 | * |
| 24455 | * We're only interested in the part after the '=' */ |
| 24456 | const char * this_lookup_name = lookup_name + equals_pos; |
| 24457 | lookup_len -= equals_pos; |
| 24458 | slash_pos -= equals_pos; |
| 24459 | |
| 24460 | /* Handle any leading minus */ |
| 24461 | if (this_lookup_name[0] == '-') { |
| 24462 | sign = "-"; |
| 24463 | this_lookup_name++; |
| 24464 | lookup_len--; |
| 24465 | slash_pos--; |
| 24466 | } |
| 24467 | |
| 24468 | /* Convert the numerator to numeric */ |
| 24469 | end_ptr = this_lookup_name + slash_pos; |
| 24470 | if (! grok_atoUV(this_lookup_name, &numerator, &end_ptr)) { |
| 24471 | goto failed; |
| 24472 | } |
| 24473 | |
| 24474 | /* It better have included all characters before the slash */ |
| 24475 | if (*end_ptr != '/') { |
| 24476 | goto failed; |
| 24477 | } |
| 24478 | |
| 24479 | /* Set to look at just the denominator */ |
| 24480 | this_lookup_name += slash_pos; |
| 24481 | lookup_len -= slash_pos; |
| 24482 | end_ptr = this_lookup_name + lookup_len; |
| 24483 | |
| 24484 | /* Convert the denominator to numeric */ |
| 24485 | if (! grok_atoUV(this_lookup_name, &denominator, &end_ptr)) { |
| 24486 | goto failed; |
| 24487 | } |
| 24488 | |
| 24489 | /* It better be the rest of the characters, and don't divide by |
| 24490 | * 0 */ |
| 24491 | if ( end_ptr != this_lookup_name + lookup_len |
| 24492 | || denominator == 0) |
| 24493 | { |
| 24494 | goto failed; |
| 24495 | } |
| 24496 | |
| 24497 | /* Get the greatest common denominator using |
| 24498 | http://en.wikipedia.org/wiki/Euclidean_algorithm */ |
| 24499 | gcd = numerator; |
| 24500 | trial = denominator; |
| 24501 | while (trial != 0) { |
| 24502 | UV temp = trial; |
| 24503 | trial = gcd % trial; |
| 24504 | gcd = temp; |
| 24505 | } |
| 24506 | |
| 24507 | /* If already in lowest possible terms, we have already tried |
| 24508 | * looking this up */ |
| 24509 | if (gcd == 1) { |
| 24510 | goto failed; |
| 24511 | } |
| 24512 | |
| 24513 | /* Reduce the rational, which should put it in canonical form |
| 24514 | * */ |
| 24515 | numerator /= gcd; |
| 24516 | denominator /= gcd; |
| 24517 | |
| 24518 | canonical = Perl_form(aTHX_ "%.*s%s%" UVuf "/%" UVuf, |
| 24519 | equals_pos, lookup_name, sign, numerator, denominator); |
| 24520 | } |
| 24521 | |
| 24522 | /* Here, we have the number in canonical form. Try that */ |
| 24523 | table_index = match_uniprop((U8 *) canonical, strlen(canonical)); |
| 24524 | if (table_index == 0) { |
| 24525 | goto failed; |
| 24526 | } |
| 24527 | } /* End of still didn't find the property in our table */ |
| 24528 | } /* End of didn't find the property in our table */ |
| 24529 | |
| 24530 | /* Here, we have a non-zero return, which is an index into a table of ptrs. |
| 24531 | * A negative return signifies that the real index is the absolute value, |
| 24532 | * but the result needs to be inverted */ |
| 24533 | if (table_index < 0) { |
| 24534 | invert_return = TRUE; |
| 24535 | table_index = -table_index; |
| 24536 | } |
| 24537 | |
| 24538 | /* Out-of band indices indicate a deprecated property. The proper index is |
| 24539 | * modulo it with the table size. And dividing by the table size yields |
| 24540 | * an offset into a table constructed by regen/mk_invlists.pl to contain |
| 24541 | * the corresponding warning message */ |
| 24542 | if (table_index > MAX_UNI_KEYWORD_INDEX) { |
| 24543 | Size_t warning_offset = table_index / MAX_UNI_KEYWORD_INDEX; |
| 24544 | table_index %= MAX_UNI_KEYWORD_INDEX; |
| 24545 | Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), |
| 24546 | "Use of '%.*s' in \\p{} or \\P{} is deprecated because: %s", |
| 24547 | (int) name_len, name, deprecated_property_msgs[warning_offset]); |
| 24548 | } |
| 24549 | |
| 24550 | /* In a few properties, a different property is used under /i. These are |
| 24551 | * unlikely to change, so are hard-coded here. */ |
| 24552 | if (to_fold) { |
| 24553 | if ( table_index == UNI_XPOSIXUPPER |
| 24554 | || table_index == UNI_XPOSIXLOWER |
| 24555 | || table_index == UNI_TITLE) |
| 24556 | { |
| 24557 | table_index = UNI_CASED; |
| 24558 | } |
| 24559 | else if ( table_index == UNI_UPPERCASELETTER |
| 24560 | || table_index == UNI_LOWERCASELETTER |
| 24561 | # ifdef UNI_TITLECASELETTER /* Missing from early Unicodes */ |
| 24562 | || table_index == UNI_TITLECASELETTER |
| 24563 | # endif |
| 24564 | ) { |
| 24565 | table_index = UNI_CASEDLETTER; |
| 24566 | } |
| 24567 | else if ( table_index == UNI_POSIXUPPER |
| 24568 | || table_index == UNI_POSIXLOWER) |
| 24569 | { |
| 24570 | table_index = UNI_POSIXALPHA; |
| 24571 | } |
| 24572 | } |
| 24573 | |
| 24574 | /* Create and return the inversion list */ |
| 24575 | prop_definition =_new_invlist_C_array(uni_prop_ptrs[table_index]); |
| 24576 | sv_2mortal(prop_definition); |
| 24577 | |
| 24578 | |
| 24579 | /* See if there is a private use override to add to this definition */ |
| 24580 | { |
| 24581 | COPHH * hinthash = (IN_PERL_COMPILETIME) |
| 24582 | ? CopHINTHASH_get(&PL_compiling) |
| 24583 | : CopHINTHASH_get(PL_curcop); |
| 24584 | SV * pu_overrides = cophh_fetch_pv(hinthash, "private_use", 0, 0); |
| 24585 | |
| 24586 | if (UNLIKELY(pu_overrides && SvPOK(pu_overrides))) { |
| 24587 | |
| 24588 | /* See if there is an element in the hints hash for this table */ |
| 24589 | SV * pu_lookup = Perl_newSVpvf(aTHX_ "%d=", table_index); |
| 24590 | const char * pos = strstr(SvPVX(pu_overrides), SvPVX(pu_lookup)); |
| 24591 | |
| 24592 | if (pos) { |
| 24593 | bool dummy; |
| 24594 | SV * pu_definition; |
| 24595 | SV * pu_invlist; |
| 24596 | SV * expanded_prop_definition = |
| 24597 | sv_2mortal(invlist_clone(prop_definition, NULL)); |
| 24598 | |
| 24599 | /* If so, it's definition is the string from here to the next |
| 24600 | * \a character. And its format is the same as a user-defined |
| 24601 | * property */ |
| 24602 | pos += SvCUR(pu_lookup); |
| 24603 | pu_definition = newSVpvn(pos, strchr(pos, '\a') - pos); |
| 24604 | pu_invlist = handle_user_defined_property(lookup_name, |
| 24605 | lookup_len, |
| 24606 | 0, /* Not UTF-8 */ |
| 24607 | 0, /* Not folded */ |
| 24608 | runtime, |
| 24609 | deferrable, |
| 24610 | pu_definition, |
| 24611 | &dummy, |
| 24612 | msg, |
| 24613 | level); |
| 24614 | if (TAINT_get) { |
| 24615 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24616 | sv_catpvs(msg, "Insecure private-use override"); |
| 24617 | goto append_name_to_msg; |
| 24618 | } |
| 24619 | |
| 24620 | /* For now, as a safety measure, make sure that it doesn't |
| 24621 | * override non-private use code points */ |
| 24622 | _invlist_intersection(pu_invlist, PL_Private_Use, &pu_invlist); |
| 24623 | |
| 24624 | /* Add it to the list to be returned */ |
| 24625 | _invlist_union(prop_definition, pu_invlist, |
| 24626 | &expanded_prop_definition); |
| 24627 | prop_definition = expanded_prop_definition; |
| 24628 | Perl_ck_warner_d(aTHX_ packWARN(WARN_EXPERIMENTAL__PRIVATE_USE), "The private_use feature is experimental"); |
| 24629 | } |
| 24630 | } |
| 24631 | } |
| 24632 | |
| 24633 | if (invert_return) { |
| 24634 | _invlist_invert(prop_definition); |
| 24635 | } |
| 24636 | return prop_definition; |
| 24637 | |
| 24638 | unknown_user_defined: |
| 24639 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24640 | sv_catpvs(msg, "Unknown user-defined property name"); |
| 24641 | goto append_name_to_msg; |
| 24642 | |
| 24643 | failed: |
| 24644 | if (non_pkg_begin != 0) { |
| 24645 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24646 | sv_catpvs(msg, "Illegal user-defined property name"); |
| 24647 | } |
| 24648 | else { |
| 24649 | if (SvCUR(msg) > 0) sv_catpvs(msg, "; "); |
| 24650 | sv_catpvs(msg, "Can't find Unicode property definition"); |
| 24651 | } |
| 24652 | /* FALLTHROUGH */ |
| 24653 | |
| 24654 | append_name_to_msg: |
| 24655 | { |
| 24656 | const char * prefix = (runtime && level == 0) ? " \\p{" : " \""; |
| 24657 | const char * suffix = (runtime && level == 0) ? "}" : "\""; |
| 24658 | |
| 24659 | sv_catpv(msg, prefix); |
| 24660 | Perl_sv_catpvf(aTHX_ msg, "%" UTF8f, UTF8fARG(is_utf8, name_len, name)); |
| 24661 | sv_catpv(msg, suffix); |
| 24662 | } |
| 24663 | |
| 24664 | return NULL; |
| 24665 | |
| 24666 | definition_deferred: |
| 24667 | |
| 24668 | { |
| 24669 | bool is_qualified = non_pkg_begin != 0; /* If has "::" */ |
| 24670 | |
| 24671 | /* Here it could yet to be defined, so defer evaluation of this until |
| 24672 | * its needed at runtime. We need the fully qualified property name to |
| 24673 | * avoid ambiguity */ |
| 24674 | if (! fq_name) { |
| 24675 | fq_name = S_get_fq_name(aTHX_ name, name_len, is_utf8, |
| 24676 | is_qualified); |
| 24677 | } |
| 24678 | |
| 24679 | /* If it didn't come with a package, or the package is utf8::, this |
| 24680 | * actually could be an official Unicode property whose inclusion we |
| 24681 | * are deferring until runtime to make sure that it isn't overridden by |
| 24682 | * a user-defined property of the same name (which we haven't |
| 24683 | * encountered yet). Add a marker to indicate this possibility, for |
| 24684 | * use at such time when we first need the definition during pattern |
| 24685 | * matching execution */ |
| 24686 | if (! is_qualified || memBEGINPs(name, non_pkg_begin, "utf8::")) { |
| 24687 | sv_catpvs(fq_name, DEFERRED_COULD_BE_OFFICIAL_MARKERs); |
| 24688 | } |
| 24689 | |
| 24690 | /* We also need a trailing newline */ |
| 24691 | sv_catpvs(fq_name, "\n"); |
| 24692 | |
| 24693 | *user_defined_ptr = TRUE; |
| 24694 | return fq_name; |
| 24695 | } |
| 24696 | } |
| 24697 | |
| 24698 | #endif /* end of ! PERL_IN_XSUB_RE */ |
| 24699 | |
| 24700 | /* |
| 24701 | * ex: set ts=8 sts=4 sw=4 et: |
| 24702 | */ |