| 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 | #ifndef PERL_IN_XSUB_RE |
| 78 | # include "INTERN.h" |
| 79 | #endif |
| 80 | |
| 81 | #define REG_COMP_C |
| 82 | #ifdef PERL_IN_XSUB_RE |
| 83 | # include "re_comp.h" |
| 84 | #else |
| 85 | # include "regcomp.h" |
| 86 | #endif |
| 87 | |
| 88 | #include "dquote_static.c" |
| 89 | |
| 90 | #ifdef op |
| 91 | #undef op |
| 92 | #endif /* op */ |
| 93 | |
| 94 | #ifdef MSDOS |
| 95 | # if defined(BUGGY_MSC6) |
| 96 | /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */ |
| 97 | # pragma optimize("a",off) |
| 98 | /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/ |
| 99 | # pragma optimize("w",on ) |
| 100 | # endif /* BUGGY_MSC6 */ |
| 101 | #endif /* MSDOS */ |
| 102 | |
| 103 | #ifndef STATIC |
| 104 | #define STATIC static |
| 105 | #endif |
| 106 | |
| 107 | typedef struct RExC_state_t { |
| 108 | U32 flags; /* are we folding, multilining? */ |
| 109 | char *precomp; /* uncompiled string. */ |
| 110 | REGEXP *rx_sv; /* The SV that is the regexp. */ |
| 111 | regexp *rx; /* perl core regexp structure */ |
| 112 | regexp_internal *rxi; /* internal data for regexp object pprivate field */ |
| 113 | char *start; /* Start of input for compile */ |
| 114 | char *end; /* End of input for compile */ |
| 115 | char *parse; /* Input-scan pointer. */ |
| 116 | I32 whilem_seen; /* number of WHILEM in this expr */ |
| 117 | regnode *emit_start; /* Start of emitted-code area */ |
| 118 | regnode *emit_bound; /* First regnode outside of the allocated space */ |
| 119 | regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */ |
| 120 | I32 naughty; /* How bad is this pattern? */ |
| 121 | I32 sawback; /* Did we see \1, ...? */ |
| 122 | U32 seen; |
| 123 | I32 size; /* Code size. */ |
| 124 | I32 npar; /* Capture buffer count, (OPEN). */ |
| 125 | I32 cpar; /* Capture buffer count, (CLOSE). */ |
| 126 | I32 nestroot; /* root parens we are in - used by accept */ |
| 127 | I32 extralen; |
| 128 | I32 seen_zerolen; |
| 129 | I32 seen_evals; |
| 130 | regnode **open_parens; /* pointers to open parens */ |
| 131 | regnode **close_parens; /* pointers to close parens */ |
| 132 | regnode *opend; /* END node in program */ |
| 133 | I32 utf8; /* whether the pattern is utf8 or not */ |
| 134 | I32 orig_utf8; /* whether the pattern was originally in utf8 */ |
| 135 | /* XXX use this for future optimisation of case |
| 136 | * where pattern must be upgraded to utf8. */ |
| 137 | I32 uni_semantics; /* If a d charset modifier should use unicode |
| 138 | rules, even if the pattern is not in |
| 139 | utf8 */ |
| 140 | HV *paren_names; /* Paren names */ |
| 141 | |
| 142 | regnode **recurse; /* Recurse regops */ |
| 143 | I32 recurse_count; /* Number of recurse regops */ |
| 144 | I32 in_lookbehind; |
| 145 | I32 contains_locale; |
| 146 | I32 override_recoding; |
| 147 | #if ADD_TO_REGEXEC |
| 148 | char *starttry; /* -Dr: where regtry was called. */ |
| 149 | #define RExC_starttry (pRExC_state->starttry) |
| 150 | #endif |
| 151 | #ifdef DEBUGGING |
| 152 | const char *lastparse; |
| 153 | I32 lastnum; |
| 154 | AV *paren_name_list; /* idx -> name */ |
| 155 | #define RExC_lastparse (pRExC_state->lastparse) |
| 156 | #define RExC_lastnum (pRExC_state->lastnum) |
| 157 | #define RExC_paren_name_list (pRExC_state->paren_name_list) |
| 158 | #endif |
| 159 | } RExC_state_t; |
| 160 | |
| 161 | #define RExC_flags (pRExC_state->flags) |
| 162 | #define RExC_precomp (pRExC_state->precomp) |
| 163 | #define RExC_rx_sv (pRExC_state->rx_sv) |
| 164 | #define RExC_rx (pRExC_state->rx) |
| 165 | #define RExC_rxi (pRExC_state->rxi) |
| 166 | #define RExC_start (pRExC_state->start) |
| 167 | #define RExC_end (pRExC_state->end) |
| 168 | #define RExC_parse (pRExC_state->parse) |
| 169 | #define RExC_whilem_seen (pRExC_state->whilem_seen) |
| 170 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 171 | #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */ |
| 172 | #endif |
| 173 | #define RExC_emit (pRExC_state->emit) |
| 174 | #define RExC_emit_start (pRExC_state->emit_start) |
| 175 | #define RExC_emit_bound (pRExC_state->emit_bound) |
| 176 | #define RExC_naughty (pRExC_state->naughty) |
| 177 | #define RExC_sawback (pRExC_state->sawback) |
| 178 | #define RExC_seen (pRExC_state->seen) |
| 179 | #define RExC_size (pRExC_state->size) |
| 180 | #define RExC_npar (pRExC_state->npar) |
| 181 | #define RExC_nestroot (pRExC_state->nestroot) |
| 182 | #define RExC_extralen (pRExC_state->extralen) |
| 183 | #define RExC_seen_zerolen (pRExC_state->seen_zerolen) |
| 184 | #define RExC_seen_evals (pRExC_state->seen_evals) |
| 185 | #define RExC_utf8 (pRExC_state->utf8) |
| 186 | #define RExC_uni_semantics (pRExC_state->uni_semantics) |
| 187 | #define RExC_orig_utf8 (pRExC_state->orig_utf8) |
| 188 | #define RExC_open_parens (pRExC_state->open_parens) |
| 189 | #define RExC_close_parens (pRExC_state->close_parens) |
| 190 | #define RExC_opend (pRExC_state->opend) |
| 191 | #define RExC_paren_names (pRExC_state->paren_names) |
| 192 | #define RExC_recurse (pRExC_state->recurse) |
| 193 | #define RExC_recurse_count (pRExC_state->recurse_count) |
| 194 | #define RExC_in_lookbehind (pRExC_state->in_lookbehind) |
| 195 | #define RExC_contains_locale (pRExC_state->contains_locale) |
| 196 | #define RExC_override_recoding (pRExC_state->override_recoding) |
| 197 | |
| 198 | |
| 199 | #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?') |
| 200 | #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \ |
| 201 | ((*s) == '{' && regcurly(s))) |
| 202 | |
| 203 | #ifdef SPSTART |
| 204 | #undef SPSTART /* dratted cpp namespace... */ |
| 205 | #endif |
| 206 | /* |
| 207 | * Flags to be passed up and down. |
| 208 | */ |
| 209 | #define WORST 0 /* Worst case. */ |
| 210 | #define HASWIDTH 0x01 /* Known to match non-null strings. */ |
| 211 | |
| 212 | /* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single |
| 213 | * character, and if utf8, must be invariant. Note that this is not the same thing as REGNODE_SIMPLE */ |
| 214 | #define SIMPLE 0x02 |
| 215 | #define SPSTART 0x04 /* Starts with * or +. */ |
| 216 | #define TRYAGAIN 0x08 /* Weeded out a declaration. */ |
| 217 | #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */ |
| 218 | |
| 219 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1) |
| 220 | |
| 221 | /* whether trie related optimizations are enabled */ |
| 222 | #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION |
| 223 | #define TRIE_STUDY_OPT |
| 224 | #define FULL_TRIE_STUDY |
| 225 | #define TRIE_STCLASS |
| 226 | #endif |
| 227 | |
| 228 | |
| 229 | |
| 230 | #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3] |
| 231 | #define PBITVAL(paren) (1 << ((paren) & 7)) |
| 232 | #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren)) |
| 233 | #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren) |
| 234 | #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren)) |
| 235 | |
| 236 | /* If not already in utf8, do a longjmp back to the beginning */ |
| 237 | #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */ |
| 238 | #define REQUIRE_UTF8 STMT_START { \ |
| 239 | if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \ |
| 240 | } STMT_END |
| 241 | |
| 242 | /* About scan_data_t. |
| 243 | |
| 244 | During optimisation we recurse through the regexp program performing |
| 245 | various inplace (keyhole style) optimisations. In addition study_chunk |
| 246 | and scan_commit populate this data structure with information about |
| 247 | what strings MUST appear in the pattern. We look for the longest |
| 248 | string that must appear at a fixed location, and we look for the |
| 249 | longest string that may appear at a floating location. So for instance |
| 250 | in the pattern: |
| 251 | |
| 252 | /FOO[xX]A.*B[xX]BAR/ |
| 253 | |
| 254 | Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating |
| 255 | strings (because they follow a .* construct). study_chunk will identify |
| 256 | both FOO and BAR as being the longest fixed and floating strings respectively. |
| 257 | |
| 258 | The strings can be composites, for instance |
| 259 | |
| 260 | /(f)(o)(o)/ |
| 261 | |
| 262 | will result in a composite fixed substring 'foo'. |
| 263 | |
| 264 | For each string some basic information is maintained: |
| 265 | |
| 266 | - offset or min_offset |
| 267 | This is the position the string must appear at, or not before. |
| 268 | It also implicitly (when combined with minlenp) tells us how many |
| 269 | characters must match before the string we are searching for. |
| 270 | Likewise when combined with minlenp and the length of the string it |
| 271 | tells us how many characters must appear after the string we have |
| 272 | found. |
| 273 | |
| 274 | - max_offset |
| 275 | Only used for floating strings. This is the rightmost point that |
| 276 | the string can appear at. If set to I32 max it indicates that the |
| 277 | string can occur infinitely far to the right. |
| 278 | |
| 279 | - minlenp |
| 280 | A pointer to the minimum length of the pattern that the string |
| 281 | was found inside. This is important as in the case of positive |
| 282 | lookahead or positive lookbehind we can have multiple patterns |
| 283 | involved. Consider |
| 284 | |
| 285 | /(?=FOO).*F/ |
| 286 | |
| 287 | The minimum length of the pattern overall is 3, the minimum length |
| 288 | of the lookahead part is 3, but the minimum length of the part that |
| 289 | will actually match is 1. So 'FOO's minimum length is 3, but the |
| 290 | minimum length for the F is 1. This is important as the minimum length |
| 291 | is used to determine offsets in front of and behind the string being |
| 292 | looked for. Since strings can be composites this is the length of the |
| 293 | pattern at the time it was committed with a scan_commit. Note that |
| 294 | the length is calculated by study_chunk, so that the minimum lengths |
| 295 | are not known until the full pattern has been compiled, thus the |
| 296 | pointer to the value. |
| 297 | |
| 298 | - lookbehind |
| 299 | |
| 300 | In the case of lookbehind the string being searched for can be |
| 301 | offset past the start point of the final matching string. |
| 302 | If this value was just blithely removed from the min_offset it would |
| 303 | invalidate some of the calculations for how many chars must match |
| 304 | before or after (as they are derived from min_offset and minlen and |
| 305 | the length of the string being searched for). |
| 306 | When the final pattern is compiled and the data is moved from the |
| 307 | scan_data_t structure into the regexp structure the information |
| 308 | about lookbehind is factored in, with the information that would |
| 309 | have been lost precalculated in the end_shift field for the |
| 310 | associated string. |
| 311 | |
| 312 | The fields pos_min and pos_delta are used to store the minimum offset |
| 313 | and the delta to the maximum offset at the current point in the pattern. |
| 314 | |
| 315 | */ |
| 316 | |
| 317 | typedef struct scan_data_t { |
| 318 | /*I32 len_min; unused */ |
| 319 | /*I32 len_delta; unused */ |
| 320 | I32 pos_min; |
| 321 | I32 pos_delta; |
| 322 | SV *last_found; |
| 323 | I32 last_end; /* min value, <0 unless valid. */ |
| 324 | I32 last_start_min; |
| 325 | I32 last_start_max; |
| 326 | SV **longest; /* Either &l_fixed, or &l_float. */ |
| 327 | SV *longest_fixed; /* longest fixed string found in pattern */ |
| 328 | I32 offset_fixed; /* offset where it starts */ |
| 329 | I32 *minlen_fixed; /* pointer to the minlen relevant to the string */ |
| 330 | I32 lookbehind_fixed; /* is the position of the string modfied by LB */ |
| 331 | SV *longest_float; /* longest floating string found in pattern */ |
| 332 | I32 offset_float_min; /* earliest point in string it can appear */ |
| 333 | I32 offset_float_max; /* latest point in string it can appear */ |
| 334 | I32 *minlen_float; /* pointer to the minlen relevant to the string */ |
| 335 | I32 lookbehind_float; /* is the position of the string modified by LB */ |
| 336 | I32 flags; |
| 337 | I32 whilem_c; |
| 338 | I32 *last_closep; |
| 339 | struct regnode_charclass_class *start_class; |
| 340 | } scan_data_t; |
| 341 | |
| 342 | /* |
| 343 | * Forward declarations for pregcomp()'s friends. |
| 344 | */ |
| 345 | |
| 346 | static const scan_data_t zero_scan_data = |
| 347 | { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0}; |
| 348 | |
| 349 | #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL) |
| 350 | #define SF_BEFORE_SEOL 0x0001 |
| 351 | #define SF_BEFORE_MEOL 0x0002 |
| 352 | #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL) |
| 353 | #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL) |
| 354 | |
| 355 | #ifdef NO_UNARY_PLUS |
| 356 | # define SF_FIX_SHIFT_EOL (0+2) |
| 357 | # define SF_FL_SHIFT_EOL (0+4) |
| 358 | #else |
| 359 | # define SF_FIX_SHIFT_EOL (+2) |
| 360 | # define SF_FL_SHIFT_EOL (+4) |
| 361 | #endif |
| 362 | |
| 363 | #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL) |
| 364 | #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL) |
| 365 | |
| 366 | #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL) |
| 367 | #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */ |
| 368 | #define SF_IS_INF 0x0040 |
| 369 | #define SF_HAS_PAR 0x0080 |
| 370 | #define SF_IN_PAR 0x0100 |
| 371 | #define SF_HAS_EVAL 0x0200 |
| 372 | #define SCF_DO_SUBSTR 0x0400 |
| 373 | #define SCF_DO_STCLASS_AND 0x0800 |
| 374 | #define SCF_DO_STCLASS_OR 0x1000 |
| 375 | #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR) |
| 376 | #define SCF_WHILEM_VISITED_POS 0x2000 |
| 377 | |
| 378 | #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */ |
| 379 | #define SCF_SEEN_ACCEPT 0x8000 |
| 380 | |
| 381 | #define UTF cBOOL(RExC_utf8) |
| 382 | #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET) |
| 383 | #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET) |
| 384 | #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_DEPENDS_CHARSET) |
| 385 | #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) >= REGEX_UNICODE_CHARSET) |
| 386 | #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_RESTRICTED_CHARSET) |
| 387 | #define MORE_ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_MORE_RESTRICTED_CHARSET) |
| 388 | #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) >= REGEX_ASCII_RESTRICTED_CHARSET) |
| 389 | |
| 390 | #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD) |
| 391 | |
| 392 | #define OOB_UNICODE 12345678 |
| 393 | #define OOB_NAMEDCLASS -1 |
| 394 | |
| 395 | #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv)) |
| 396 | #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b) |
| 397 | |
| 398 | |
| 399 | /* length of regex to show in messages that don't mark a position within */ |
| 400 | #define RegexLengthToShowInErrorMessages 127 |
| 401 | |
| 402 | /* |
| 403 | * If MARKER[12] are adjusted, be sure to adjust the constants at the top |
| 404 | * of t/op/regmesg.t, the tests in t/op/re_tests, and those in |
| 405 | * op/pragma/warn/regcomp. |
| 406 | */ |
| 407 | #define MARKER1 "<-- HERE" /* marker as it appears in the description */ |
| 408 | #define MARKER2 " <-- HERE " /* marker as it appears within the regex */ |
| 409 | |
| 410 | #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/" |
| 411 | |
| 412 | /* |
| 413 | * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given |
| 414 | * arg. Show regex, up to a maximum length. If it's too long, chop and add |
| 415 | * "...". |
| 416 | */ |
| 417 | #define _FAIL(code) STMT_START { \ |
| 418 | const char *ellipses = ""; \ |
| 419 | IV len = RExC_end - RExC_precomp; \ |
| 420 | \ |
| 421 | if (!SIZE_ONLY) \ |
| 422 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
| 423 | if (len > RegexLengthToShowInErrorMessages) { \ |
| 424 | /* chop 10 shorter than the max, to ensure meaning of "..." */ \ |
| 425 | len = RegexLengthToShowInErrorMessages - 10; \ |
| 426 | ellipses = "..."; \ |
| 427 | } \ |
| 428 | code; \ |
| 429 | } STMT_END |
| 430 | |
| 431 | #define FAIL(msg) _FAIL( \ |
| 432 | Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \ |
| 433 | msg, (int)len, RExC_precomp, ellipses)) |
| 434 | |
| 435 | #define FAIL2(msg,arg) _FAIL( \ |
| 436 | Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \ |
| 437 | arg, (int)len, RExC_precomp, ellipses)) |
| 438 | |
| 439 | /* |
| 440 | * Simple_vFAIL -- like FAIL, but marks the current location in the scan |
| 441 | */ |
| 442 | #define Simple_vFAIL(m) STMT_START { \ |
| 443 | const IV offset = RExC_parse - RExC_precomp; \ |
| 444 | Perl_croak(aTHX_ "%s" REPORT_LOCATION, \ |
| 445 | m, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 446 | } STMT_END |
| 447 | |
| 448 | /* |
| 449 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL() |
| 450 | */ |
| 451 | #define vFAIL(m) STMT_START { \ |
| 452 | if (!SIZE_ONLY) \ |
| 453 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
| 454 | Simple_vFAIL(m); \ |
| 455 | } STMT_END |
| 456 | |
| 457 | /* |
| 458 | * Like Simple_vFAIL(), but accepts two arguments. |
| 459 | */ |
| 460 | #define Simple_vFAIL2(m,a1) STMT_START { \ |
| 461 | const IV offset = RExC_parse - RExC_precomp; \ |
| 462 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \ |
| 463 | (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 464 | } STMT_END |
| 465 | |
| 466 | /* |
| 467 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2(). |
| 468 | */ |
| 469 | #define vFAIL2(m,a1) STMT_START { \ |
| 470 | if (!SIZE_ONLY) \ |
| 471 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
| 472 | Simple_vFAIL2(m, a1); \ |
| 473 | } STMT_END |
| 474 | |
| 475 | |
| 476 | /* |
| 477 | * Like Simple_vFAIL(), but accepts three arguments. |
| 478 | */ |
| 479 | #define Simple_vFAIL3(m, a1, a2) STMT_START { \ |
| 480 | const IV offset = RExC_parse - RExC_precomp; \ |
| 481 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \ |
| 482 | (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 483 | } STMT_END |
| 484 | |
| 485 | /* |
| 486 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3(). |
| 487 | */ |
| 488 | #define vFAIL3(m,a1,a2) STMT_START { \ |
| 489 | if (!SIZE_ONLY) \ |
| 490 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
| 491 | Simple_vFAIL3(m, a1, a2); \ |
| 492 | } STMT_END |
| 493 | |
| 494 | /* |
| 495 | * Like Simple_vFAIL(), but accepts four arguments. |
| 496 | */ |
| 497 | #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \ |
| 498 | const IV offset = RExC_parse - RExC_precomp; \ |
| 499 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \ |
| 500 | (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 501 | } STMT_END |
| 502 | |
| 503 | #define ckWARNreg(loc,m) STMT_START { \ |
| 504 | const IV offset = loc - RExC_precomp; \ |
| 505 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
| 506 | (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 507 | } STMT_END |
| 508 | |
| 509 | #define ckWARNregdep(loc,m) STMT_START { \ |
| 510 | const IV offset = loc - RExC_precomp; \ |
| 511 | Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ |
| 512 | m REPORT_LOCATION, \ |
| 513 | (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 514 | } STMT_END |
| 515 | |
| 516 | #define ckWARN2regdep(loc,m, a1) STMT_START { \ |
| 517 | const IV offset = loc - RExC_precomp; \ |
| 518 | Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ |
| 519 | m REPORT_LOCATION, \ |
| 520 | a1, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 521 | } STMT_END |
| 522 | |
| 523 | #define ckWARN2reg(loc, m, a1) STMT_START { \ |
| 524 | const IV offset = loc - RExC_precomp; \ |
| 525 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
| 526 | a1, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 527 | } STMT_END |
| 528 | |
| 529 | #define vWARN3(loc, m, a1, a2) STMT_START { \ |
| 530 | const IV offset = loc - RExC_precomp; \ |
| 531 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
| 532 | a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 533 | } STMT_END |
| 534 | |
| 535 | #define ckWARN3reg(loc, m, a1, a2) STMT_START { \ |
| 536 | const IV offset = loc - RExC_precomp; \ |
| 537 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
| 538 | a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 539 | } STMT_END |
| 540 | |
| 541 | #define vWARN4(loc, m, a1, a2, a3) STMT_START { \ |
| 542 | const IV offset = loc - RExC_precomp; \ |
| 543 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
| 544 | a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 545 | } STMT_END |
| 546 | |
| 547 | #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \ |
| 548 | const IV offset = loc - RExC_precomp; \ |
| 549 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
| 550 | a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 551 | } STMT_END |
| 552 | |
| 553 | #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \ |
| 554 | const IV offset = loc - RExC_precomp; \ |
| 555 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
| 556 | a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
| 557 | } STMT_END |
| 558 | |
| 559 | |
| 560 | /* Allow for side effects in s */ |
| 561 | #define REGC(c,s) STMT_START { \ |
| 562 | if (!SIZE_ONLY) *(s) = (c); else (void)(s); \ |
| 563 | } STMT_END |
| 564 | |
| 565 | /* Macros for recording node offsets. 20001227 mjd@plover.com |
| 566 | * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in |
| 567 | * element 2*n-1 of the array. Element #2n holds the byte length node #n. |
| 568 | * Element 0 holds the number n. |
| 569 | * Position is 1 indexed. |
| 570 | */ |
| 571 | #ifndef RE_TRACK_PATTERN_OFFSETS |
| 572 | #define Set_Node_Offset_To_R(node,byte) |
| 573 | #define Set_Node_Offset(node,byte) |
| 574 | #define Set_Cur_Node_Offset |
| 575 | #define Set_Node_Length_To_R(node,len) |
| 576 | #define Set_Node_Length(node,len) |
| 577 | #define Set_Node_Cur_Length(node) |
| 578 | #define Node_Offset(n) |
| 579 | #define Node_Length(n) |
| 580 | #define Set_Node_Offset_Length(node,offset,len) |
| 581 | #define ProgLen(ri) ri->u.proglen |
| 582 | #define SetProgLen(ri,x) ri->u.proglen = x |
| 583 | #else |
| 584 | #define ProgLen(ri) ri->u.offsets[0] |
| 585 | #define SetProgLen(ri,x) ri->u.offsets[0] = x |
| 586 | #define Set_Node_Offset_To_R(node,byte) STMT_START { \ |
| 587 | if (! SIZE_ONLY) { \ |
| 588 | MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \ |
| 589 | __LINE__, (int)(node), (int)(byte))); \ |
| 590 | if((node) < 0) { \ |
| 591 | Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \ |
| 592 | } else { \ |
| 593 | RExC_offsets[2*(node)-1] = (byte); \ |
| 594 | } \ |
| 595 | } \ |
| 596 | } STMT_END |
| 597 | |
| 598 | #define Set_Node_Offset(node,byte) \ |
| 599 | Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start) |
| 600 | #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse) |
| 601 | |
| 602 | #define Set_Node_Length_To_R(node,len) STMT_START { \ |
| 603 | if (! SIZE_ONLY) { \ |
| 604 | MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \ |
| 605 | __LINE__, (int)(node), (int)(len))); \ |
| 606 | if((node) < 0) { \ |
| 607 | Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \ |
| 608 | } else { \ |
| 609 | RExC_offsets[2*(node)] = (len); \ |
| 610 | } \ |
| 611 | } \ |
| 612 | } STMT_END |
| 613 | |
| 614 | #define Set_Node_Length(node,len) \ |
| 615 | Set_Node_Length_To_R((node)-RExC_emit_start, len) |
| 616 | #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len) |
| 617 | #define Set_Node_Cur_Length(node) \ |
| 618 | Set_Node_Length(node, RExC_parse - parse_start) |
| 619 | |
| 620 | /* Get offsets and lengths */ |
| 621 | #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1]) |
| 622 | #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)]) |
| 623 | |
| 624 | #define Set_Node_Offset_Length(node,offset,len) STMT_START { \ |
| 625 | Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \ |
| 626 | Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \ |
| 627 | } STMT_END |
| 628 | #endif |
| 629 | |
| 630 | #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS |
| 631 | #define EXPERIMENTAL_INPLACESCAN |
| 632 | #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/ |
| 633 | |
| 634 | #define DEBUG_STUDYDATA(str,data,depth) \ |
| 635 | DEBUG_OPTIMISE_MORE_r(if(data){ \ |
| 636 | PerlIO_printf(Perl_debug_log, \ |
| 637 | "%*s" str "Pos:%"IVdf"/%"IVdf \ |
| 638 | " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \ |
| 639 | (int)(depth)*2, "", \ |
| 640 | (IV)((data)->pos_min), \ |
| 641 | (IV)((data)->pos_delta), \ |
| 642 | (UV)((data)->flags), \ |
| 643 | (IV)((data)->whilem_c), \ |
| 644 | (IV)((data)->last_closep ? *((data)->last_closep) : -1), \ |
| 645 | is_inf ? "INF " : "" \ |
| 646 | ); \ |
| 647 | if ((data)->last_found) \ |
| 648 | PerlIO_printf(Perl_debug_log, \ |
| 649 | "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \ |
| 650 | " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \ |
| 651 | SvPVX_const((data)->last_found), \ |
| 652 | (IV)((data)->last_end), \ |
| 653 | (IV)((data)->last_start_min), \ |
| 654 | (IV)((data)->last_start_max), \ |
| 655 | ((data)->longest && \ |
| 656 | (data)->longest==&((data)->longest_fixed)) ? "*" : "", \ |
| 657 | SvPVX_const((data)->longest_fixed), \ |
| 658 | (IV)((data)->offset_fixed), \ |
| 659 | ((data)->longest && \ |
| 660 | (data)->longest==&((data)->longest_float)) ? "*" : "", \ |
| 661 | SvPVX_const((data)->longest_float), \ |
| 662 | (IV)((data)->offset_float_min), \ |
| 663 | (IV)((data)->offset_float_max) \ |
| 664 | ); \ |
| 665 | PerlIO_printf(Perl_debug_log,"\n"); \ |
| 666 | }); |
| 667 | |
| 668 | static void clear_re(pTHX_ void *r); |
| 669 | |
| 670 | /* Mark that we cannot extend a found fixed substring at this point. |
| 671 | Update the longest found anchored substring and the longest found |
| 672 | floating substrings if needed. */ |
| 673 | |
| 674 | STATIC void |
| 675 | S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf) |
| 676 | { |
| 677 | const STRLEN l = CHR_SVLEN(data->last_found); |
| 678 | const STRLEN old_l = CHR_SVLEN(*data->longest); |
| 679 | GET_RE_DEBUG_FLAGS_DECL; |
| 680 | |
| 681 | PERL_ARGS_ASSERT_SCAN_COMMIT; |
| 682 | |
| 683 | if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) { |
| 684 | SvSetMagicSV(*data->longest, data->last_found); |
| 685 | if (*data->longest == data->longest_fixed) { |
| 686 | data->offset_fixed = l ? data->last_start_min : data->pos_min; |
| 687 | if (data->flags & SF_BEFORE_EOL) |
| 688 | data->flags |
| 689 | |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL); |
| 690 | else |
| 691 | data->flags &= ~SF_FIX_BEFORE_EOL; |
| 692 | data->minlen_fixed=minlenp; |
| 693 | data->lookbehind_fixed=0; |
| 694 | } |
| 695 | else { /* *data->longest == data->longest_float */ |
| 696 | data->offset_float_min = l ? data->last_start_min : data->pos_min; |
| 697 | data->offset_float_max = (l |
| 698 | ? data->last_start_max |
| 699 | : data->pos_min + data->pos_delta); |
| 700 | if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX) |
| 701 | data->offset_float_max = I32_MAX; |
| 702 | if (data->flags & SF_BEFORE_EOL) |
| 703 | data->flags |
| 704 | |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL); |
| 705 | else |
| 706 | data->flags &= ~SF_FL_BEFORE_EOL; |
| 707 | data->minlen_float=minlenp; |
| 708 | data->lookbehind_float=0; |
| 709 | } |
| 710 | } |
| 711 | SvCUR_set(data->last_found, 0); |
| 712 | { |
| 713 | SV * const sv = data->last_found; |
| 714 | if (SvUTF8(sv) && SvMAGICAL(sv)) { |
| 715 | MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8); |
| 716 | if (mg) |
| 717 | mg->mg_len = 0; |
| 718 | } |
| 719 | } |
| 720 | data->last_end = -1; |
| 721 | data->flags &= ~SF_BEFORE_EOL; |
| 722 | DEBUG_STUDYDATA("commit: ",data,0); |
| 723 | } |
| 724 | |
| 725 | /* Can match anything (initialization) */ |
| 726 | STATIC void |
| 727 | S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) |
| 728 | { |
| 729 | PERL_ARGS_ASSERT_CL_ANYTHING; |
| 730 | |
| 731 | ANYOF_BITMAP_SETALL(cl); |
| 732 | cl->flags = ANYOF_CLASS|ANYOF_EOS|ANYOF_UNICODE_ALL |
| 733 | |ANYOF_LOC_NONBITMAP_FOLD|ANYOF_NON_UTF8_LATIN1_ALL; |
| 734 | |
| 735 | /* If any portion of the regex is to operate under locale rules, |
| 736 | * initialization includes it. The reason this isn't done for all regexes |
| 737 | * is that the optimizer was written under the assumption that locale was |
| 738 | * all-or-nothing. Given the complexity and lack of documentation in the |
| 739 | * optimizer, and that there are inadequate test cases for locale, so many |
| 740 | * parts of it may not work properly, it is safest to avoid locale unless |
| 741 | * necessary. */ |
| 742 | if (RExC_contains_locale) { |
| 743 | ANYOF_CLASS_SETALL(cl); /* /l uses class */ |
| 744 | cl->flags |= ANYOF_LOCALE; |
| 745 | } |
| 746 | else { |
| 747 | ANYOF_CLASS_ZERO(cl); /* Only /l uses class now */ |
| 748 | } |
| 749 | } |
| 750 | |
| 751 | /* Can match anything (initialization) */ |
| 752 | STATIC int |
| 753 | S_cl_is_anything(const struct regnode_charclass_class *cl) |
| 754 | { |
| 755 | int value; |
| 756 | |
| 757 | PERL_ARGS_ASSERT_CL_IS_ANYTHING; |
| 758 | |
| 759 | for (value = 0; value <= ANYOF_MAX; value += 2) |
| 760 | if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1)) |
| 761 | return 1; |
| 762 | if (!(cl->flags & ANYOF_UNICODE_ALL)) |
| 763 | return 0; |
| 764 | if (!ANYOF_BITMAP_TESTALLSET((const void*)cl)) |
| 765 | return 0; |
| 766 | return 1; |
| 767 | } |
| 768 | |
| 769 | /* Can match anything (initialization) */ |
| 770 | STATIC void |
| 771 | S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) |
| 772 | { |
| 773 | PERL_ARGS_ASSERT_CL_INIT; |
| 774 | |
| 775 | Zero(cl, 1, struct regnode_charclass_class); |
| 776 | cl->type = ANYOF; |
| 777 | cl_anything(pRExC_state, cl); |
| 778 | ARG_SET(cl, ANYOF_NONBITMAP_EMPTY); |
| 779 | } |
| 780 | |
| 781 | /* These two functions currently do the exact same thing */ |
| 782 | #define cl_init_zero S_cl_init |
| 783 | |
| 784 | /* 'AND' a given class with another one. Can create false positives. 'cl' |
| 785 | * should not be inverted. 'and_with->flags & ANYOF_CLASS' should be 0 if |
| 786 | * 'and_with' is a regnode_charclass instead of a regnode_charclass_class. */ |
| 787 | STATIC void |
| 788 | S_cl_and(struct regnode_charclass_class *cl, |
| 789 | const struct regnode_charclass_class *and_with) |
| 790 | { |
| 791 | PERL_ARGS_ASSERT_CL_AND; |
| 792 | |
| 793 | assert(and_with->type == ANYOF); |
| 794 | |
| 795 | /* I (khw) am not sure all these restrictions are necessary XXX */ |
| 796 | if (!(ANYOF_CLASS_TEST_ANY_SET(and_with)) |
| 797 | && !(ANYOF_CLASS_TEST_ANY_SET(cl)) |
| 798 | && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) |
| 799 | && !(and_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
| 800 | && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) { |
| 801 | int i; |
| 802 | |
| 803 | if (and_with->flags & ANYOF_INVERT) |
| 804 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) |
| 805 | cl->bitmap[i] &= ~and_with->bitmap[i]; |
| 806 | else |
| 807 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) |
| 808 | cl->bitmap[i] &= and_with->bitmap[i]; |
| 809 | } /* XXXX: logic is complicated otherwise, leave it along for a moment. */ |
| 810 | |
| 811 | if (and_with->flags & ANYOF_INVERT) { |
| 812 | |
| 813 | /* Here, the and'ed node is inverted. Get the AND of the flags that |
| 814 | * aren't affected by the inversion. Those that are affected are |
| 815 | * handled individually below */ |
| 816 | U8 affected_flags = cl->flags & ~INVERSION_UNAFFECTED_FLAGS; |
| 817 | cl->flags &= (and_with->flags & INVERSION_UNAFFECTED_FLAGS); |
| 818 | cl->flags |= affected_flags; |
| 819 | |
| 820 | /* We currently don't know how to deal with things that aren't in the |
| 821 | * bitmap, but we know that the intersection is no greater than what |
| 822 | * is already in cl, so let there be false positives that get sorted |
| 823 | * out after the synthetic start class succeeds, and the node is |
| 824 | * matched for real. */ |
| 825 | |
| 826 | /* The inversion of these two flags indicate that the resulting |
| 827 | * intersection doesn't have them */ |
| 828 | if (and_with->flags & ANYOF_UNICODE_ALL) { |
| 829 | cl->flags &= ~ANYOF_UNICODE_ALL; |
| 830 | } |
| 831 | if (and_with->flags & ANYOF_NON_UTF8_LATIN1_ALL) { |
| 832 | cl->flags &= ~ANYOF_NON_UTF8_LATIN1_ALL; |
| 833 | } |
| 834 | } |
| 835 | else { /* and'd node is not inverted */ |
| 836 | U8 outside_bitmap_but_not_utf8; /* Temp variable */ |
| 837 | |
| 838 | if (! ANYOF_NONBITMAP(and_with)) { |
| 839 | |
| 840 | /* Here 'and_with' doesn't match anything outside the bitmap |
| 841 | * (except possibly ANYOF_UNICODE_ALL), which means the |
| 842 | * intersection can't either, except for ANYOF_UNICODE_ALL, in |
| 843 | * which case we don't know what the intersection is, but it's no |
| 844 | * greater than what cl already has, so can just leave it alone, |
| 845 | * with possible false positives */ |
| 846 | if (! (and_with->flags & ANYOF_UNICODE_ALL)) { |
| 847 | ARG_SET(cl, ANYOF_NONBITMAP_EMPTY); |
| 848 | cl->flags &= ~ANYOF_NONBITMAP_NON_UTF8; |
| 849 | } |
| 850 | } |
| 851 | else if (! ANYOF_NONBITMAP(cl)) { |
| 852 | |
| 853 | /* Here, 'and_with' does match something outside the bitmap, and cl |
| 854 | * doesn't have a list of things to match outside the bitmap. If |
| 855 | * cl can match all code points above 255, the intersection will |
| 856 | * be those above-255 code points that 'and_with' matches. If cl |
| 857 | * can't match all Unicode code points, it means that it can't |
| 858 | * match anything outside the bitmap (since the 'if' that got us |
| 859 | * into this block tested for that), so we leave the bitmap empty. |
| 860 | */ |
| 861 | if (cl->flags & ANYOF_UNICODE_ALL) { |
| 862 | ARG_SET(cl, ARG(and_with)); |
| 863 | |
| 864 | /* and_with's ARG may match things that don't require UTF8. |
| 865 | * And now cl's will too, in spite of this being an 'and'. See |
| 866 | * the comments below about the kludge */ |
| 867 | cl->flags |= and_with->flags & ANYOF_NONBITMAP_NON_UTF8; |
| 868 | } |
| 869 | } |
| 870 | else { |
| 871 | /* Here, both 'and_with' and cl match something outside the |
| 872 | * bitmap. Currently we do not do the intersection, so just match |
| 873 | * whatever cl had at the beginning. */ |
| 874 | } |
| 875 | |
| 876 | |
| 877 | /* Take the intersection of the two sets of flags. However, the |
| 878 | * ANYOF_NONBITMAP_NON_UTF8 flag is treated as an 'or'. This is a |
| 879 | * kludge around the fact that this flag is not treated like the others |
| 880 | * which are initialized in cl_anything(). The way the optimizer works |
| 881 | * is that the synthetic start class (SSC) is initialized to match |
| 882 | * anything, and then the first time a real node is encountered, its |
| 883 | * values are AND'd with the SSC's with the result being the values of |
| 884 | * the real node. However, there are paths through the optimizer where |
| 885 | * the AND never gets called, so those initialized bits are set |
| 886 | * inappropriately, which is not usually a big deal, as they just cause |
| 887 | * false positives in the SSC, which will just mean a probably |
| 888 | * imperceptible slow down in execution. However this bit has a |
| 889 | * higher false positive consequence in that it can cause utf8.pm, |
| 890 | * utf8_heavy.pl ... to be loaded when not necessary, which is a much |
| 891 | * bigger slowdown and also causes significant extra memory to be used. |
| 892 | * In order to prevent this, the code now takes a different tack. The |
| 893 | * bit isn't set unless some part of the regular expression needs it, |
| 894 | * but once set it won't get cleared. This means that these extra |
| 895 | * modules won't get loaded unless there was some path through the |
| 896 | * pattern that would have required them anyway, and so any false |
| 897 | * positives that occur by not ANDing them out when they could be |
| 898 | * aren't as severe as they would be if we treated this bit like all |
| 899 | * the others */ |
| 900 | outside_bitmap_but_not_utf8 = (cl->flags | and_with->flags) |
| 901 | & ANYOF_NONBITMAP_NON_UTF8; |
| 902 | cl->flags &= and_with->flags; |
| 903 | cl->flags |= outside_bitmap_but_not_utf8; |
| 904 | } |
| 905 | } |
| 906 | |
| 907 | /* 'OR' a given class with another one. Can create false positives. 'cl' |
| 908 | * should not be inverted. 'or_with->flags & ANYOF_CLASS' should be 0 if |
| 909 | * 'or_with' is a regnode_charclass instead of a regnode_charclass_class. */ |
| 910 | STATIC void |
| 911 | S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with) |
| 912 | { |
| 913 | PERL_ARGS_ASSERT_CL_OR; |
| 914 | |
| 915 | if (or_with->flags & ANYOF_INVERT) { |
| 916 | |
| 917 | /* Here, the or'd node is to be inverted. This means we take the |
| 918 | * complement of everything not in the bitmap, but currently we don't |
| 919 | * know what that is, so give up and match anything */ |
| 920 | if (ANYOF_NONBITMAP(or_with)) { |
| 921 | cl_anything(pRExC_state, cl); |
| 922 | } |
| 923 | /* We do not use |
| 924 | * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2)) |
| 925 | * <= (B1 | !B2) | (CL1 | !CL2) |
| 926 | * which is wasteful if CL2 is small, but we ignore CL2: |
| 927 | * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1 |
| 928 | * XXXX Can we handle case-fold? Unclear: |
| 929 | * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) = |
| 930 | * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i')) |
| 931 | */ |
| 932 | else if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) |
| 933 | && !(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
| 934 | && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD) ) { |
| 935 | int i; |
| 936 | |
| 937 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) |
| 938 | cl->bitmap[i] |= ~or_with->bitmap[i]; |
| 939 | } /* XXXX: logic is complicated otherwise */ |
| 940 | else { |
| 941 | cl_anything(pRExC_state, cl); |
| 942 | } |
| 943 | |
| 944 | /* And, we can just take the union of the flags that aren't affected |
| 945 | * by the inversion */ |
| 946 | cl->flags |= or_with->flags & INVERSION_UNAFFECTED_FLAGS; |
| 947 | |
| 948 | /* For the remaining flags: |
| 949 | ANYOF_UNICODE_ALL and inverted means to not match anything above |
| 950 | 255, which means that the union with cl should just be |
| 951 | what cl has in it, so can ignore this flag |
| 952 | ANYOF_NON_UTF8_LATIN1_ALL and inverted means if not utf8 and ord |
| 953 | is 127-255 to match them, but then invert that, so the |
| 954 | union with cl should just be what cl has in it, so can |
| 955 | ignore this flag |
| 956 | */ |
| 957 | } else { /* 'or_with' is not inverted */ |
| 958 | /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */ |
| 959 | if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) |
| 960 | && (!(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
| 961 | || (cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) ) { |
| 962 | int i; |
| 963 | |
| 964 | /* OR char bitmap and class bitmap separately */ |
| 965 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) |
| 966 | cl->bitmap[i] |= or_with->bitmap[i]; |
| 967 | if (ANYOF_CLASS_TEST_ANY_SET(or_with)) { |
| 968 | for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++) |
| 969 | cl->classflags[i] |= or_with->classflags[i]; |
| 970 | cl->flags |= ANYOF_CLASS; |
| 971 | } |
| 972 | } |
| 973 | else { /* XXXX: logic is complicated, leave it along for a moment. */ |
| 974 | cl_anything(pRExC_state, cl); |
| 975 | } |
| 976 | |
| 977 | if (ANYOF_NONBITMAP(or_with)) { |
| 978 | |
| 979 | /* Use the added node's outside-the-bit-map match if there isn't a |
| 980 | * conflict. If there is a conflict (both nodes match something |
| 981 | * outside the bitmap, but what they match outside is not the same |
| 982 | * pointer, and hence not easily compared until XXX we extend |
| 983 | * inversion lists this far), give up and allow the start class to |
| 984 | * match everything outside the bitmap. If that stuff is all above |
| 985 | * 255, can just set UNICODE_ALL, otherwise caould be anything. */ |
| 986 | if (! ANYOF_NONBITMAP(cl)) { |
| 987 | ARG_SET(cl, ARG(or_with)); |
| 988 | } |
| 989 | else if (ARG(cl) != ARG(or_with)) { |
| 990 | |
| 991 | if ((or_with->flags & ANYOF_NONBITMAP_NON_UTF8)) { |
| 992 | cl_anything(pRExC_state, cl); |
| 993 | } |
| 994 | else { |
| 995 | cl->flags |= ANYOF_UNICODE_ALL; |
| 996 | } |
| 997 | } |
| 998 | } |
| 999 | |
| 1000 | /* Take the union */ |
| 1001 | cl->flags |= or_with->flags; |
| 1002 | } |
| 1003 | } |
| 1004 | |
| 1005 | #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ] |
| 1006 | #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid ) |
| 1007 | #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate ) |
| 1008 | #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 ) |
| 1009 | |
| 1010 | |
| 1011 | #ifdef DEBUGGING |
| 1012 | /* |
| 1013 | dump_trie(trie,widecharmap,revcharmap) |
| 1014 | dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc) |
| 1015 | dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc) |
| 1016 | |
| 1017 | These routines dump out a trie in a somewhat readable format. |
| 1018 | The _interim_ variants are used for debugging the interim |
| 1019 | tables that are used to generate the final compressed |
| 1020 | representation which is what dump_trie expects. |
| 1021 | |
| 1022 | Part of the reason for their existence is to provide a form |
| 1023 | of documentation as to how the different representations function. |
| 1024 | |
| 1025 | */ |
| 1026 | |
| 1027 | /* |
| 1028 | Dumps the final compressed table form of the trie to Perl_debug_log. |
| 1029 | Used for debugging make_trie(). |
| 1030 | */ |
| 1031 | |
| 1032 | STATIC void |
| 1033 | S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap, |
| 1034 | AV *revcharmap, U32 depth) |
| 1035 | { |
| 1036 | U32 state; |
| 1037 | SV *sv=sv_newmortal(); |
| 1038 | int colwidth= widecharmap ? 6 : 4; |
| 1039 | U16 word; |
| 1040 | GET_RE_DEBUG_FLAGS_DECL; |
| 1041 | |
| 1042 | PERL_ARGS_ASSERT_DUMP_TRIE; |
| 1043 | |
| 1044 | PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ", |
| 1045 | (int)depth * 2 + 2,"", |
| 1046 | "Match","Base","Ofs" ); |
| 1047 | |
| 1048 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) { |
| 1049 | SV ** const tmp = av_fetch( revcharmap, state, 0); |
| 1050 | if ( tmp ) { |
| 1051 | PerlIO_printf( Perl_debug_log, "%*s", |
| 1052 | colwidth, |
| 1053 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
| 1054 | PL_colors[0], PL_colors[1], |
| 1055 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | |
| 1056 | PERL_PV_ESCAPE_FIRSTCHAR |
| 1057 | ) |
| 1058 | ); |
| 1059 | } |
| 1060 | } |
| 1061 | PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------", |
| 1062 | (int)depth * 2 + 2,""); |
| 1063 | |
| 1064 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) |
| 1065 | PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------"); |
| 1066 | PerlIO_printf( Perl_debug_log, "\n"); |
| 1067 | |
| 1068 | for( state = 1 ; state < trie->statecount ; state++ ) { |
| 1069 | const U32 base = trie->states[ state ].trans.base; |
| 1070 | |
| 1071 | PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state); |
| 1072 | |
| 1073 | if ( trie->states[ state ].wordnum ) { |
| 1074 | PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum ); |
| 1075 | } else { |
| 1076 | PerlIO_printf( Perl_debug_log, "%6s", "" ); |
| 1077 | } |
| 1078 | |
| 1079 | PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base ); |
| 1080 | |
| 1081 | if ( base ) { |
| 1082 | U32 ofs = 0; |
| 1083 | |
| 1084 | while( ( base + ofs < trie->uniquecharcount ) || |
| 1085 | ( base + ofs - trie->uniquecharcount < trie->lasttrans |
| 1086 | && trie->trans[ base + ofs - trie->uniquecharcount ].check != state)) |
| 1087 | ofs++; |
| 1088 | |
| 1089 | PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs); |
| 1090 | |
| 1091 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { |
| 1092 | if ( ( base + ofs >= trie->uniquecharcount ) && |
| 1093 | ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && |
| 1094 | trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) |
| 1095 | { |
| 1096 | PerlIO_printf( Perl_debug_log, "%*"UVXf, |
| 1097 | colwidth, |
| 1098 | (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next ); |
| 1099 | } else { |
| 1100 | PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." ); |
| 1101 | } |
| 1102 | } |
| 1103 | |
| 1104 | PerlIO_printf( Perl_debug_log, "]"); |
| 1105 | |
| 1106 | } |
| 1107 | PerlIO_printf( Perl_debug_log, "\n" ); |
| 1108 | } |
| 1109 | PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, ""); |
| 1110 | for (word=1; word <= trie->wordcount; word++) { |
| 1111 | PerlIO_printf(Perl_debug_log, " %d:(%d,%d)", |
| 1112 | (int)word, (int)(trie->wordinfo[word].prev), |
| 1113 | (int)(trie->wordinfo[word].len)); |
| 1114 | } |
| 1115 | PerlIO_printf(Perl_debug_log, "\n" ); |
| 1116 | } |
| 1117 | /* |
| 1118 | Dumps a fully constructed but uncompressed trie in list form. |
| 1119 | List tries normally only are used for construction when the number of |
| 1120 | possible chars (trie->uniquecharcount) is very high. |
| 1121 | Used for debugging make_trie(). |
| 1122 | */ |
| 1123 | STATIC void |
| 1124 | S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie, |
| 1125 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
| 1126 | U32 depth) |
| 1127 | { |
| 1128 | U32 state; |
| 1129 | SV *sv=sv_newmortal(); |
| 1130 | int colwidth= widecharmap ? 6 : 4; |
| 1131 | GET_RE_DEBUG_FLAGS_DECL; |
| 1132 | |
| 1133 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST; |
| 1134 | |
| 1135 | /* print out the table precompression. */ |
| 1136 | PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s", |
| 1137 | (int)depth * 2 + 2,"", (int)depth * 2 + 2,"", |
| 1138 | "------:-----+-----------------\n" ); |
| 1139 | |
| 1140 | for( state=1 ; state < next_alloc ; state ++ ) { |
| 1141 | U16 charid; |
| 1142 | |
| 1143 | PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :", |
| 1144 | (int)depth * 2 + 2,"", (UV)state ); |
| 1145 | if ( ! trie->states[ state ].wordnum ) { |
| 1146 | PerlIO_printf( Perl_debug_log, "%5s| ",""); |
| 1147 | } else { |
| 1148 | PerlIO_printf( Perl_debug_log, "W%4x| ", |
| 1149 | trie->states[ state ].wordnum |
| 1150 | ); |
| 1151 | } |
| 1152 | for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) { |
| 1153 | SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0); |
| 1154 | if ( tmp ) { |
| 1155 | PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ", |
| 1156 | colwidth, |
| 1157 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
| 1158 | PL_colors[0], PL_colors[1], |
| 1159 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | |
| 1160 | PERL_PV_ESCAPE_FIRSTCHAR |
| 1161 | ) , |
| 1162 | TRIE_LIST_ITEM(state,charid).forid, |
| 1163 | (UV)TRIE_LIST_ITEM(state,charid).newstate |
| 1164 | ); |
| 1165 | if (!(charid % 10)) |
| 1166 | PerlIO_printf(Perl_debug_log, "\n%*s| ", |
| 1167 | (int)((depth * 2) + 14), ""); |
| 1168 | } |
| 1169 | } |
| 1170 | PerlIO_printf( Perl_debug_log, "\n"); |
| 1171 | } |
| 1172 | } |
| 1173 | |
| 1174 | /* |
| 1175 | Dumps a fully constructed but uncompressed trie in table form. |
| 1176 | This is the normal DFA style state transition table, with a few |
| 1177 | twists to facilitate compression later. |
| 1178 | Used for debugging make_trie(). |
| 1179 | */ |
| 1180 | STATIC void |
| 1181 | S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie, |
| 1182 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
| 1183 | U32 depth) |
| 1184 | { |
| 1185 | U32 state; |
| 1186 | U16 charid; |
| 1187 | SV *sv=sv_newmortal(); |
| 1188 | int colwidth= widecharmap ? 6 : 4; |
| 1189 | GET_RE_DEBUG_FLAGS_DECL; |
| 1190 | |
| 1191 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE; |
| 1192 | |
| 1193 | /* |
| 1194 | print out the table precompression so that we can do a visual check |
| 1195 | that they are identical. |
| 1196 | */ |
| 1197 | |
| 1198 | PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" ); |
| 1199 | |
| 1200 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { |
| 1201 | SV ** const tmp = av_fetch( revcharmap, charid, 0); |
| 1202 | if ( tmp ) { |
| 1203 | PerlIO_printf( Perl_debug_log, "%*s", |
| 1204 | colwidth, |
| 1205 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
| 1206 | PL_colors[0], PL_colors[1], |
| 1207 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | |
| 1208 | PERL_PV_ESCAPE_FIRSTCHAR |
| 1209 | ) |
| 1210 | ); |
| 1211 | } |
| 1212 | } |
| 1213 | |
| 1214 | PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" ); |
| 1215 | |
| 1216 | for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) { |
| 1217 | PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------"); |
| 1218 | } |
| 1219 | |
| 1220 | PerlIO_printf( Perl_debug_log, "\n" ); |
| 1221 | |
| 1222 | for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) { |
| 1223 | |
| 1224 | PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ", |
| 1225 | (int)depth * 2 + 2,"", |
| 1226 | (UV)TRIE_NODENUM( state ) ); |
| 1227 | |
| 1228 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { |
| 1229 | UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next ); |
| 1230 | if (v) |
| 1231 | PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v ); |
| 1232 | else |
| 1233 | PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." ); |
| 1234 | } |
| 1235 | if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) { |
| 1236 | PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check ); |
| 1237 | } else { |
| 1238 | PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check, |
| 1239 | trie->states[ TRIE_NODENUM( state ) ].wordnum ); |
| 1240 | } |
| 1241 | } |
| 1242 | } |
| 1243 | |
| 1244 | #endif |
| 1245 | |
| 1246 | |
| 1247 | /* make_trie(startbranch,first,last,tail,word_count,flags,depth) |
| 1248 | startbranch: the first branch in the whole branch sequence |
| 1249 | first : start branch of sequence of branch-exact nodes. |
| 1250 | May be the same as startbranch |
| 1251 | last : Thing following the last branch. |
| 1252 | May be the same as tail. |
| 1253 | tail : item following the branch sequence |
| 1254 | count : words in the sequence |
| 1255 | flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/ |
| 1256 | depth : indent depth |
| 1257 | |
| 1258 | Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node. |
| 1259 | |
| 1260 | A trie is an N'ary tree where the branches are determined by digital |
| 1261 | decomposition of the key. IE, at the root node you look up the 1st character and |
| 1262 | follow that branch repeat until you find the end of the branches. Nodes can be |
| 1263 | marked as "accepting" meaning they represent a complete word. Eg: |
| 1264 | |
| 1265 | /he|she|his|hers/ |
| 1266 | |
| 1267 | would convert into the following structure. Numbers represent states, letters |
| 1268 | following numbers represent valid transitions on the letter from that state, if |
| 1269 | the number is in square brackets it represents an accepting state, otherwise it |
| 1270 | will be in parenthesis. |
| 1271 | |
| 1272 | +-h->+-e->[3]-+-r->(8)-+-s->[9] |
| 1273 | | | |
| 1274 | | (2) |
| 1275 | | | |
| 1276 | (1) +-i->(6)-+-s->[7] |
| 1277 | | |
| 1278 | +-s->(3)-+-h->(4)-+-e->[5] |
| 1279 | |
| 1280 | Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers) |
| 1281 | |
| 1282 | This shows that when matching against the string 'hers' we will begin at state 1 |
| 1283 | read 'h' and move to state 2, read 'e' and move to state 3 which is accepting, |
| 1284 | then read 'r' and go to state 8 followed by 's' which takes us to state 9 which |
| 1285 | is also accepting. Thus we know that we can match both 'he' and 'hers' with a |
| 1286 | single traverse. We store a mapping from accepting to state to which word was |
| 1287 | matched, and then when we have multiple possibilities we try to complete the |
| 1288 | rest of the regex in the order in which they occured in the alternation. |
| 1289 | |
| 1290 | The only prior NFA like behaviour that would be changed by the TRIE support is |
| 1291 | the silent ignoring of duplicate alternations which are of the form: |
| 1292 | |
| 1293 | / (DUPE|DUPE) X? (?{ ... }) Y /x |
| 1294 | |
| 1295 | Thus EVAL blocks following a trie may be called a different number of times with |
| 1296 | and without the optimisation. With the optimisations dupes will be silently |
| 1297 | ignored. This inconsistent behaviour of EVAL type nodes is well established as |
| 1298 | the following demonstrates: |
| 1299 | |
| 1300 | 'words'=~/(word|word|word)(?{ print $1 })[xyz]/ |
| 1301 | |
| 1302 | which prints out 'word' three times, but |
| 1303 | |
| 1304 | 'words'=~/(word|word|word)(?{ print $1 })S/ |
| 1305 | |
| 1306 | which doesnt print it out at all. This is due to other optimisations kicking in. |
| 1307 | |
| 1308 | Example of what happens on a structural level: |
| 1309 | |
| 1310 | The regexp /(ac|ad|ab)+/ will produce the following debug output: |
| 1311 | |
| 1312 | 1: CURLYM[1] {1,32767}(18) |
| 1313 | 5: BRANCH(8) |
| 1314 | 6: EXACT <ac>(16) |
| 1315 | 8: BRANCH(11) |
| 1316 | 9: EXACT <ad>(16) |
| 1317 | 11: BRANCH(14) |
| 1318 | 12: EXACT <ab>(16) |
| 1319 | 16: SUCCEED(0) |
| 1320 | 17: NOTHING(18) |
| 1321 | 18: END(0) |
| 1322 | |
| 1323 | This would be optimizable with startbranch=5, first=5, last=16, tail=16 |
| 1324 | and should turn into: |
| 1325 | |
| 1326 | 1: CURLYM[1] {1,32767}(18) |
| 1327 | 5: TRIE(16) |
| 1328 | [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1] |
| 1329 | <ac> |
| 1330 | <ad> |
| 1331 | <ab> |
| 1332 | 16: SUCCEED(0) |
| 1333 | 17: NOTHING(18) |
| 1334 | 18: END(0) |
| 1335 | |
| 1336 | Cases where tail != last would be like /(?foo|bar)baz/: |
| 1337 | |
| 1338 | 1: BRANCH(4) |
| 1339 | 2: EXACT <foo>(8) |
| 1340 | 4: BRANCH(7) |
| 1341 | 5: EXACT <bar>(8) |
| 1342 | 7: TAIL(8) |
| 1343 | 8: EXACT <baz>(10) |
| 1344 | 10: END(0) |
| 1345 | |
| 1346 | which would be optimizable with startbranch=1, first=1, last=7, tail=8 |
| 1347 | and would end up looking like: |
| 1348 | |
| 1349 | 1: TRIE(8) |
| 1350 | [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1] |
| 1351 | <foo> |
| 1352 | <bar> |
| 1353 | 7: TAIL(8) |
| 1354 | 8: EXACT <baz>(10) |
| 1355 | 10: END(0) |
| 1356 | |
| 1357 | d = uvuni_to_utf8_flags(d, uv, 0); |
| 1358 | |
| 1359 | is the recommended Unicode-aware way of saying |
| 1360 | |
| 1361 | *(d++) = uv; |
| 1362 | */ |
| 1363 | |
| 1364 | #define TRIE_STORE_REVCHAR \ |
| 1365 | STMT_START { \ |
| 1366 | if (UTF) { \ |
| 1367 | SV *zlopp = newSV(2); \ |
| 1368 | unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \ |
| 1369 | unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \ |
| 1370 | SvCUR_set(zlopp, kapow - flrbbbbb); \ |
| 1371 | SvPOK_on(zlopp); \ |
| 1372 | SvUTF8_on(zlopp); \ |
| 1373 | av_push(revcharmap, zlopp); \ |
| 1374 | } else { \ |
| 1375 | char ooooff = (char)uvc; \ |
| 1376 | av_push(revcharmap, newSVpvn(&ooooff, 1)); \ |
| 1377 | } \ |
| 1378 | } STMT_END |
| 1379 | |
| 1380 | #define TRIE_READ_CHAR STMT_START { \ |
| 1381 | wordlen++; \ |
| 1382 | if ( UTF ) { \ |
| 1383 | if ( folder ) { \ |
| 1384 | if ( foldlen > 0 ) { \ |
| 1385 | uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \ |
| 1386 | foldlen -= len; \ |
| 1387 | scan += len; \ |
| 1388 | len = 0; \ |
| 1389 | } else { \ |
| 1390 | uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\ |
| 1391 | uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \ |
| 1392 | foldlen -= UNISKIP( uvc ); \ |
| 1393 | scan = foldbuf + UNISKIP( uvc ); \ |
| 1394 | } \ |
| 1395 | } else { \ |
| 1396 | uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\ |
| 1397 | } \ |
| 1398 | } else { \ |
| 1399 | uvc = (U32)*uc; \ |
| 1400 | len = 1; \ |
| 1401 | } \ |
| 1402 | } STMT_END |
| 1403 | |
| 1404 | |
| 1405 | |
| 1406 | #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \ |
| 1407 | if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \ |
| 1408 | U32 ging = TRIE_LIST_LEN( state ) *= 2; \ |
| 1409 | Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \ |
| 1410 | } \ |
| 1411 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \ |
| 1412 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \ |
| 1413 | TRIE_LIST_CUR( state )++; \ |
| 1414 | } STMT_END |
| 1415 | |
| 1416 | #define TRIE_LIST_NEW(state) STMT_START { \ |
| 1417 | Newxz( trie->states[ state ].trans.list, \ |
| 1418 | 4, reg_trie_trans_le ); \ |
| 1419 | TRIE_LIST_CUR( state ) = 1; \ |
| 1420 | TRIE_LIST_LEN( state ) = 4; \ |
| 1421 | } STMT_END |
| 1422 | |
| 1423 | #define TRIE_HANDLE_WORD(state) STMT_START { \ |
| 1424 | U16 dupe= trie->states[ state ].wordnum; \ |
| 1425 | regnode * const noper_next = regnext( noper ); \ |
| 1426 | \ |
| 1427 | DEBUG_r({ \ |
| 1428 | /* store the word for dumping */ \ |
| 1429 | SV* tmp; \ |
| 1430 | if (OP(noper) != NOTHING) \ |
| 1431 | tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \ |
| 1432 | else \ |
| 1433 | tmp = newSVpvn_utf8( "", 0, UTF ); \ |
| 1434 | av_push( trie_words, tmp ); \ |
| 1435 | }); \ |
| 1436 | \ |
| 1437 | curword++; \ |
| 1438 | trie->wordinfo[curword].prev = 0; \ |
| 1439 | trie->wordinfo[curword].len = wordlen; \ |
| 1440 | trie->wordinfo[curword].accept = state; \ |
| 1441 | \ |
| 1442 | if ( noper_next < tail ) { \ |
| 1443 | if (!trie->jump) \ |
| 1444 | trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \ |
| 1445 | trie->jump[curword] = (U16)(noper_next - convert); \ |
| 1446 | if (!jumper) \ |
| 1447 | jumper = noper_next; \ |
| 1448 | if (!nextbranch) \ |
| 1449 | nextbranch= regnext(cur); \ |
| 1450 | } \ |
| 1451 | \ |
| 1452 | if ( dupe ) { \ |
| 1453 | /* It's a dupe. Pre-insert into the wordinfo[].prev */\ |
| 1454 | /* chain, so that when the bits of chain are later */\ |
| 1455 | /* linked together, the dups appear in the chain */\ |
| 1456 | trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \ |
| 1457 | trie->wordinfo[dupe].prev = curword; \ |
| 1458 | } else { \ |
| 1459 | /* we haven't inserted this word yet. */ \ |
| 1460 | trie->states[ state ].wordnum = curword; \ |
| 1461 | } \ |
| 1462 | } STMT_END |
| 1463 | |
| 1464 | |
| 1465 | #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \ |
| 1466 | ( ( base + charid >= ucharcount \ |
| 1467 | && base + charid < ubound \ |
| 1468 | && state == trie->trans[ base - ucharcount + charid ].check \ |
| 1469 | && trie->trans[ base - ucharcount + charid ].next ) \ |
| 1470 | ? trie->trans[ base - ucharcount + charid ].next \ |
| 1471 | : ( state==1 ? special : 0 ) \ |
| 1472 | ) |
| 1473 | |
| 1474 | #define MADE_TRIE 1 |
| 1475 | #define MADE_JUMP_TRIE 2 |
| 1476 | #define MADE_EXACT_TRIE 4 |
| 1477 | |
| 1478 | STATIC I32 |
| 1479 | S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth) |
| 1480 | { |
| 1481 | dVAR; |
| 1482 | /* first pass, loop through and scan words */ |
| 1483 | reg_trie_data *trie; |
| 1484 | HV *widecharmap = NULL; |
| 1485 | AV *revcharmap = newAV(); |
| 1486 | regnode *cur; |
| 1487 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
| 1488 | STRLEN len = 0; |
| 1489 | UV uvc = 0; |
| 1490 | U16 curword = 0; |
| 1491 | U32 next_alloc = 0; |
| 1492 | regnode *jumper = NULL; |
| 1493 | regnode *nextbranch = NULL; |
| 1494 | regnode *convert = NULL; |
| 1495 | U32 *prev_states; /* temp array mapping each state to previous one */ |
| 1496 | /* we just use folder as a flag in utf8 */ |
| 1497 | const U8 * folder = NULL; |
| 1498 | |
| 1499 | #ifdef DEBUGGING |
| 1500 | const U32 data_slot = add_data( pRExC_state, 4, "tuuu" ); |
| 1501 | AV *trie_words = NULL; |
| 1502 | /* along with revcharmap, this only used during construction but both are |
| 1503 | * useful during debugging so we store them in the struct when debugging. |
| 1504 | */ |
| 1505 | #else |
| 1506 | const U32 data_slot = add_data( pRExC_state, 2, "tu" ); |
| 1507 | STRLEN trie_charcount=0; |
| 1508 | #endif |
| 1509 | SV *re_trie_maxbuff; |
| 1510 | GET_RE_DEBUG_FLAGS_DECL; |
| 1511 | |
| 1512 | PERL_ARGS_ASSERT_MAKE_TRIE; |
| 1513 | #ifndef DEBUGGING |
| 1514 | PERL_UNUSED_ARG(depth); |
| 1515 | #endif |
| 1516 | |
| 1517 | switch (flags) { |
| 1518 | case EXACTFA: |
| 1519 | case EXACTFU: folder = PL_fold_latin1; break; |
| 1520 | case EXACTF: folder = PL_fold; break; |
| 1521 | case EXACTFL: folder = PL_fold_locale; break; |
| 1522 | } |
| 1523 | |
| 1524 | trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) ); |
| 1525 | trie->refcount = 1; |
| 1526 | trie->startstate = 1; |
| 1527 | trie->wordcount = word_count; |
| 1528 | RExC_rxi->data->data[ data_slot ] = (void*)trie; |
| 1529 | trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) ); |
| 1530 | if (!(UTF && folder)) |
| 1531 | trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 ); |
| 1532 | trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc( |
| 1533 | trie->wordcount+1, sizeof(reg_trie_wordinfo)); |
| 1534 | |
| 1535 | DEBUG_r({ |
| 1536 | trie_words = newAV(); |
| 1537 | }); |
| 1538 | |
| 1539 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); |
| 1540 | if (!SvIOK(re_trie_maxbuff)) { |
| 1541 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); |
| 1542 | } |
| 1543 | DEBUG_OPTIMISE_r({ |
| 1544 | PerlIO_printf( Perl_debug_log, |
| 1545 | "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n", |
| 1546 | (int)depth * 2 + 2, "", |
| 1547 | REG_NODE_NUM(startbranch),REG_NODE_NUM(first), |
| 1548 | REG_NODE_NUM(last), REG_NODE_NUM(tail), |
| 1549 | (int)depth); |
| 1550 | }); |
| 1551 | |
| 1552 | /* Find the node we are going to overwrite */ |
| 1553 | if ( first == startbranch && OP( last ) != BRANCH ) { |
| 1554 | /* whole branch chain */ |
| 1555 | convert = first; |
| 1556 | } else { |
| 1557 | /* branch sub-chain */ |
| 1558 | convert = NEXTOPER( first ); |
| 1559 | } |
| 1560 | |
| 1561 | /* -- First loop and Setup -- |
| 1562 | |
| 1563 | We first traverse the branches and scan each word to determine if it |
| 1564 | contains widechars, and how many unique chars there are, this is |
| 1565 | important as we have to build a table with at least as many columns as we |
| 1566 | have unique chars. |
| 1567 | |
| 1568 | We use an array of integers to represent the character codes 0..255 |
| 1569 | (trie->charmap) and we use a an HV* to store Unicode characters. We use the |
| 1570 | native representation of the character value as the key and IV's for the |
| 1571 | coded index. |
| 1572 | |
| 1573 | *TODO* If we keep track of how many times each character is used we can |
| 1574 | remap the columns so that the table compression later on is more |
| 1575 | efficient in terms of memory by ensuring the most common value is in the |
| 1576 | middle and the least common are on the outside. IMO this would be better |
| 1577 | than a most to least common mapping as theres a decent chance the most |
| 1578 | common letter will share a node with the least common, meaning the node |
| 1579 | will not be compressible. With a middle is most common approach the worst |
| 1580 | case is when we have the least common nodes twice. |
| 1581 | |
| 1582 | */ |
| 1583 | |
| 1584 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
| 1585 | regnode * const noper = NEXTOPER( cur ); |
| 1586 | const U8 *uc = (U8*)STRING( noper ); |
| 1587 | const U8 * const e = uc + STR_LEN( noper ); |
| 1588 | STRLEN foldlen = 0; |
| 1589 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 1590 | const U8 *scan = (U8*)NULL; |
| 1591 | U32 wordlen = 0; /* required init */ |
| 1592 | STRLEN chars = 0; |
| 1593 | bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/ |
| 1594 | |
| 1595 | if (OP(noper) == NOTHING) { |
| 1596 | trie->minlen= 0; |
| 1597 | continue; |
| 1598 | } |
| 1599 | if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */ |
| 1600 | TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte |
| 1601 | regardless of encoding */ |
| 1602 | |
| 1603 | for ( ; uc < e ; uc += len ) { |
| 1604 | TRIE_CHARCOUNT(trie)++; |
| 1605 | TRIE_READ_CHAR; |
| 1606 | chars++; |
| 1607 | if ( uvc < 256 ) { |
| 1608 | if ( !trie->charmap[ uvc ] ) { |
| 1609 | trie->charmap[ uvc ]=( ++trie->uniquecharcount ); |
| 1610 | if ( folder ) |
| 1611 | trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ]; |
| 1612 | TRIE_STORE_REVCHAR; |
| 1613 | } |
| 1614 | if ( set_bit ) { |
| 1615 | /* store the codepoint in the bitmap, and its folded |
| 1616 | * equivalent. */ |
| 1617 | TRIE_BITMAP_SET(trie,uvc); |
| 1618 | |
| 1619 | /* store the folded codepoint */ |
| 1620 | if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]); |
| 1621 | |
| 1622 | if ( !UTF ) { |
| 1623 | /* store first byte of utf8 representation of |
| 1624 | variant codepoints */ |
| 1625 | if (! UNI_IS_INVARIANT(uvc)) { |
| 1626 | TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); |
| 1627 | } |
| 1628 | } |
| 1629 | set_bit = 0; /* We've done our bit :-) */ |
| 1630 | } |
| 1631 | } else { |
| 1632 | SV** svpp; |
| 1633 | if ( !widecharmap ) |
| 1634 | widecharmap = newHV(); |
| 1635 | |
| 1636 | svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 ); |
| 1637 | |
| 1638 | if ( !svpp ) |
| 1639 | Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc ); |
| 1640 | |
| 1641 | if ( !SvTRUE( *svpp ) ) { |
| 1642 | sv_setiv( *svpp, ++trie->uniquecharcount ); |
| 1643 | TRIE_STORE_REVCHAR; |
| 1644 | } |
| 1645 | } |
| 1646 | } |
| 1647 | if( cur == first ) { |
| 1648 | trie->minlen=chars; |
| 1649 | trie->maxlen=chars; |
| 1650 | } else if (chars < trie->minlen) { |
| 1651 | trie->minlen=chars; |
| 1652 | } else if (chars > trie->maxlen) { |
| 1653 | trie->maxlen=chars; |
| 1654 | } |
| 1655 | |
| 1656 | } /* end first pass */ |
| 1657 | DEBUG_TRIE_COMPILE_r( |
| 1658 | PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n", |
| 1659 | (int)depth * 2 + 2,"", |
| 1660 | ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count, |
| 1661 | (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount, |
| 1662 | (int)trie->minlen, (int)trie->maxlen ) |
| 1663 | ); |
| 1664 | |
| 1665 | /* |
| 1666 | We now know what we are dealing with in terms of unique chars and |
| 1667 | string sizes so we can calculate how much memory a naive |
| 1668 | representation using a flat table will take. If it's over a reasonable |
| 1669 | limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory |
| 1670 | conservative but potentially much slower representation using an array |
| 1671 | of lists. |
| 1672 | |
| 1673 | At the end we convert both representations into the same compressed |
| 1674 | form that will be used in regexec.c for matching with. The latter |
| 1675 | is a form that cannot be used to construct with but has memory |
| 1676 | properties similar to the list form and access properties similar |
| 1677 | to the table form making it both suitable for fast searches and |
| 1678 | small enough that its feasable to store for the duration of a program. |
| 1679 | |
| 1680 | See the comment in the code where the compressed table is produced |
| 1681 | inplace from the flat tabe representation for an explanation of how |
| 1682 | the compression works. |
| 1683 | |
| 1684 | */ |
| 1685 | |
| 1686 | |
| 1687 | Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32); |
| 1688 | prev_states[1] = 0; |
| 1689 | |
| 1690 | if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) { |
| 1691 | /* |
| 1692 | Second Pass -- Array Of Lists Representation |
| 1693 | |
| 1694 | Each state will be represented by a list of charid:state records |
| 1695 | (reg_trie_trans_le) the first such element holds the CUR and LEN |
| 1696 | points of the allocated array. (See defines above). |
| 1697 | |
| 1698 | We build the initial structure using the lists, and then convert |
| 1699 | it into the compressed table form which allows faster lookups |
| 1700 | (but cant be modified once converted). |
| 1701 | */ |
| 1702 | |
| 1703 | STRLEN transcount = 1; |
| 1704 | |
| 1705 | DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, |
| 1706 | "%*sCompiling trie using list compiler\n", |
| 1707 | (int)depth * 2 + 2, "")); |
| 1708 | |
| 1709 | trie->states = (reg_trie_state *) |
| 1710 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, |
| 1711 | sizeof(reg_trie_state) ); |
| 1712 | TRIE_LIST_NEW(1); |
| 1713 | next_alloc = 2; |
| 1714 | |
| 1715 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
| 1716 | |
| 1717 | regnode * const noper = NEXTOPER( cur ); |
| 1718 | U8 *uc = (U8*)STRING( noper ); |
| 1719 | const U8 * const e = uc + STR_LEN( noper ); |
| 1720 | U32 state = 1; /* required init */ |
| 1721 | U16 charid = 0; /* sanity init */ |
| 1722 | U8 *scan = (U8*)NULL; /* sanity init */ |
| 1723 | STRLEN foldlen = 0; /* required init */ |
| 1724 | U32 wordlen = 0; /* required init */ |
| 1725 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 1726 | |
| 1727 | if (OP(noper) != NOTHING) { |
| 1728 | for ( ; uc < e ; uc += len ) { |
| 1729 | |
| 1730 | TRIE_READ_CHAR; |
| 1731 | |
| 1732 | if ( uvc < 256 ) { |
| 1733 | charid = trie->charmap[ uvc ]; |
| 1734 | } else { |
| 1735 | SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0); |
| 1736 | if ( !svpp ) { |
| 1737 | charid = 0; |
| 1738 | } else { |
| 1739 | charid=(U16)SvIV( *svpp ); |
| 1740 | } |
| 1741 | } |
| 1742 | /* charid is now 0 if we dont know the char read, or nonzero if we do */ |
| 1743 | if ( charid ) { |
| 1744 | |
| 1745 | U16 check; |
| 1746 | U32 newstate = 0; |
| 1747 | |
| 1748 | charid--; |
| 1749 | if ( !trie->states[ state ].trans.list ) { |
| 1750 | TRIE_LIST_NEW( state ); |
| 1751 | } |
| 1752 | for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) { |
| 1753 | if ( TRIE_LIST_ITEM( state, check ).forid == charid ) { |
| 1754 | newstate = TRIE_LIST_ITEM( state, check ).newstate; |
| 1755 | break; |
| 1756 | } |
| 1757 | } |
| 1758 | if ( ! newstate ) { |
| 1759 | newstate = next_alloc++; |
| 1760 | prev_states[newstate] = state; |
| 1761 | TRIE_LIST_PUSH( state, charid, newstate ); |
| 1762 | transcount++; |
| 1763 | } |
| 1764 | state = newstate; |
| 1765 | } else { |
| 1766 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); |
| 1767 | } |
| 1768 | } |
| 1769 | } |
| 1770 | TRIE_HANDLE_WORD(state); |
| 1771 | |
| 1772 | } /* end second pass */ |
| 1773 | |
| 1774 | /* next alloc is the NEXT state to be allocated */ |
| 1775 | trie->statecount = next_alloc; |
| 1776 | trie->states = (reg_trie_state *) |
| 1777 | PerlMemShared_realloc( trie->states, |
| 1778 | next_alloc |
| 1779 | * sizeof(reg_trie_state) ); |
| 1780 | |
| 1781 | /* and now dump it out before we compress it */ |
| 1782 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap, |
| 1783 | revcharmap, next_alloc, |
| 1784 | depth+1) |
| 1785 | ); |
| 1786 | |
| 1787 | trie->trans = (reg_trie_trans *) |
| 1788 | PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) ); |
| 1789 | { |
| 1790 | U32 state; |
| 1791 | U32 tp = 0; |
| 1792 | U32 zp = 0; |
| 1793 | |
| 1794 | |
| 1795 | for( state=1 ; state < next_alloc ; state ++ ) { |
| 1796 | U32 base=0; |
| 1797 | |
| 1798 | /* |
| 1799 | DEBUG_TRIE_COMPILE_MORE_r( |
| 1800 | PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp) |
| 1801 | ); |
| 1802 | */ |
| 1803 | |
| 1804 | if (trie->states[state].trans.list) { |
| 1805 | U16 minid=TRIE_LIST_ITEM( state, 1).forid; |
| 1806 | U16 maxid=minid; |
| 1807 | U16 idx; |
| 1808 | |
| 1809 | for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) { |
| 1810 | const U16 forid = TRIE_LIST_ITEM( state, idx).forid; |
| 1811 | if ( forid < minid ) { |
| 1812 | minid=forid; |
| 1813 | } else if ( forid > maxid ) { |
| 1814 | maxid=forid; |
| 1815 | } |
| 1816 | } |
| 1817 | if ( transcount < tp + maxid - minid + 1) { |
| 1818 | transcount *= 2; |
| 1819 | trie->trans = (reg_trie_trans *) |
| 1820 | PerlMemShared_realloc( trie->trans, |
| 1821 | transcount |
| 1822 | * sizeof(reg_trie_trans) ); |
| 1823 | Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans ); |
| 1824 | } |
| 1825 | base = trie->uniquecharcount + tp - minid; |
| 1826 | if ( maxid == minid ) { |
| 1827 | U32 set = 0; |
| 1828 | for ( ; zp < tp ; zp++ ) { |
| 1829 | if ( ! trie->trans[ zp ].next ) { |
| 1830 | base = trie->uniquecharcount + zp - minid; |
| 1831 | trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate; |
| 1832 | trie->trans[ zp ].check = state; |
| 1833 | set = 1; |
| 1834 | break; |
| 1835 | } |
| 1836 | } |
| 1837 | if ( !set ) { |
| 1838 | trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate; |
| 1839 | trie->trans[ tp ].check = state; |
| 1840 | tp++; |
| 1841 | zp = tp; |
| 1842 | } |
| 1843 | } else { |
| 1844 | for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) { |
| 1845 | const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid; |
| 1846 | trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate; |
| 1847 | trie->trans[ tid ].check = state; |
| 1848 | } |
| 1849 | tp += ( maxid - minid + 1 ); |
| 1850 | } |
| 1851 | Safefree(trie->states[ state ].trans.list); |
| 1852 | } |
| 1853 | /* |
| 1854 | DEBUG_TRIE_COMPILE_MORE_r( |
| 1855 | PerlIO_printf( Perl_debug_log, " base: %d\n",base); |
| 1856 | ); |
| 1857 | */ |
| 1858 | trie->states[ state ].trans.base=base; |
| 1859 | } |
| 1860 | trie->lasttrans = tp + 1; |
| 1861 | } |
| 1862 | } else { |
| 1863 | /* |
| 1864 | Second Pass -- Flat Table Representation. |
| 1865 | |
| 1866 | we dont use the 0 slot of either trans[] or states[] so we add 1 to each. |
| 1867 | We know that we will need Charcount+1 trans at most to store the data |
| 1868 | (one row per char at worst case) So we preallocate both structures |
| 1869 | assuming worst case. |
| 1870 | |
| 1871 | We then construct the trie using only the .next slots of the entry |
| 1872 | structs. |
| 1873 | |
| 1874 | We use the .check field of the first entry of the node temporarily to |
| 1875 | make compression both faster and easier by keeping track of how many non |
| 1876 | zero fields are in the node. |
| 1877 | |
| 1878 | Since trans are numbered from 1 any 0 pointer in the table is a FAIL |
| 1879 | transition. |
| 1880 | |
| 1881 | There are two terms at use here: state as a TRIE_NODEIDX() which is a |
| 1882 | number representing the first entry of the node, and state as a |
| 1883 | TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and |
| 1884 | TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there |
| 1885 | are 2 entrys per node. eg: |
| 1886 | |
| 1887 | A B A B |
| 1888 | 1. 2 4 1. 3 7 |
| 1889 | 2. 0 3 3. 0 5 |
| 1890 | 3. 0 0 5. 0 0 |
| 1891 | 4. 0 0 7. 0 0 |
| 1892 | |
| 1893 | The table is internally in the right hand, idx form. However as we also |
| 1894 | have to deal with the states array which is indexed by nodenum we have to |
| 1895 | use TRIE_NODENUM() to convert. |
| 1896 | |
| 1897 | */ |
| 1898 | DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, |
| 1899 | "%*sCompiling trie using table compiler\n", |
| 1900 | (int)depth * 2 + 2, "")); |
| 1901 | |
| 1902 | trie->trans = (reg_trie_trans *) |
| 1903 | PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 ) |
| 1904 | * trie->uniquecharcount + 1, |
| 1905 | sizeof(reg_trie_trans) ); |
| 1906 | trie->states = (reg_trie_state *) |
| 1907 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, |
| 1908 | sizeof(reg_trie_state) ); |
| 1909 | next_alloc = trie->uniquecharcount + 1; |
| 1910 | |
| 1911 | |
| 1912 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
| 1913 | |
| 1914 | regnode * const noper = NEXTOPER( cur ); |
| 1915 | const U8 *uc = (U8*)STRING( noper ); |
| 1916 | const U8 * const e = uc + STR_LEN( noper ); |
| 1917 | |
| 1918 | U32 state = 1; /* required init */ |
| 1919 | |
| 1920 | U16 charid = 0; /* sanity init */ |
| 1921 | U32 accept_state = 0; /* sanity init */ |
| 1922 | U8 *scan = (U8*)NULL; /* sanity init */ |
| 1923 | |
| 1924 | STRLEN foldlen = 0; /* required init */ |
| 1925 | U32 wordlen = 0; /* required init */ |
| 1926 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 1927 | |
| 1928 | if ( OP(noper) != NOTHING ) { |
| 1929 | for ( ; uc < e ; uc += len ) { |
| 1930 | |
| 1931 | TRIE_READ_CHAR; |
| 1932 | |
| 1933 | if ( uvc < 256 ) { |
| 1934 | charid = trie->charmap[ uvc ]; |
| 1935 | } else { |
| 1936 | SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0); |
| 1937 | charid = svpp ? (U16)SvIV(*svpp) : 0; |
| 1938 | } |
| 1939 | if ( charid ) { |
| 1940 | charid--; |
| 1941 | if ( !trie->trans[ state + charid ].next ) { |
| 1942 | trie->trans[ state + charid ].next = next_alloc; |
| 1943 | trie->trans[ state ].check++; |
| 1944 | prev_states[TRIE_NODENUM(next_alloc)] |
| 1945 | = TRIE_NODENUM(state); |
| 1946 | next_alloc += trie->uniquecharcount; |
| 1947 | } |
| 1948 | state = trie->trans[ state + charid ].next; |
| 1949 | } else { |
| 1950 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); |
| 1951 | } |
| 1952 | /* charid is now 0 if we dont know the char read, or nonzero if we do */ |
| 1953 | } |
| 1954 | } |
| 1955 | accept_state = TRIE_NODENUM( state ); |
| 1956 | TRIE_HANDLE_WORD(accept_state); |
| 1957 | |
| 1958 | } /* end second pass */ |
| 1959 | |
| 1960 | /* and now dump it out before we compress it */ |
| 1961 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap, |
| 1962 | revcharmap, |
| 1963 | next_alloc, depth+1)); |
| 1964 | |
| 1965 | { |
| 1966 | /* |
| 1967 | * Inplace compress the table.* |
| 1968 | |
| 1969 | For sparse data sets the table constructed by the trie algorithm will |
| 1970 | be mostly 0/FAIL transitions or to put it another way mostly empty. |
| 1971 | (Note that leaf nodes will not contain any transitions.) |
| 1972 | |
| 1973 | This algorithm compresses the tables by eliminating most such |
| 1974 | transitions, at the cost of a modest bit of extra work during lookup: |
| 1975 | |
| 1976 | - Each states[] entry contains a .base field which indicates the |
| 1977 | index in the state[] array wheres its transition data is stored. |
| 1978 | |
| 1979 | - If .base is 0 there are no valid transitions from that node. |
| 1980 | |
| 1981 | - If .base is nonzero then charid is added to it to find an entry in |
| 1982 | the trans array. |
| 1983 | |
| 1984 | -If trans[states[state].base+charid].check!=state then the |
| 1985 | transition is taken to be a 0/Fail transition. Thus if there are fail |
| 1986 | transitions at the front of the node then the .base offset will point |
| 1987 | somewhere inside the previous nodes data (or maybe even into a node |
| 1988 | even earlier), but the .check field determines if the transition is |
| 1989 | valid. |
| 1990 | |
| 1991 | XXX - wrong maybe? |
| 1992 | The following process inplace converts the table to the compressed |
| 1993 | table: We first do not compress the root node 1,and mark all its |
| 1994 | .check pointers as 1 and set its .base pointer as 1 as well. This |
| 1995 | allows us to do a DFA construction from the compressed table later, |
| 1996 | and ensures that any .base pointers we calculate later are greater |
| 1997 | than 0. |
| 1998 | |
| 1999 | - We set 'pos' to indicate the first entry of the second node. |
| 2000 | |
| 2001 | - We then iterate over the columns of the node, finding the first and |
| 2002 | last used entry at l and m. We then copy l..m into pos..(pos+m-l), |
| 2003 | and set the .check pointers accordingly, and advance pos |
| 2004 | appropriately and repreat for the next node. Note that when we copy |
| 2005 | the next pointers we have to convert them from the original |
| 2006 | NODEIDX form to NODENUM form as the former is not valid post |
| 2007 | compression. |
| 2008 | |
| 2009 | - If a node has no transitions used we mark its base as 0 and do not |
| 2010 | advance the pos pointer. |
| 2011 | |
| 2012 | - If a node only has one transition we use a second pointer into the |
| 2013 | structure to fill in allocated fail transitions from other states. |
| 2014 | This pointer is independent of the main pointer and scans forward |
| 2015 | looking for null transitions that are allocated to a state. When it |
| 2016 | finds one it writes the single transition into the "hole". If the |
| 2017 | pointer doesnt find one the single transition is appended as normal. |
| 2018 | |
| 2019 | - Once compressed we can Renew/realloc the structures to release the |
| 2020 | excess space. |
| 2021 | |
| 2022 | See "Table-Compression Methods" in sec 3.9 of the Red Dragon, |
| 2023 | specifically Fig 3.47 and the associated pseudocode. |
| 2024 | |
| 2025 | demq |
| 2026 | */ |
| 2027 | const U32 laststate = TRIE_NODENUM( next_alloc ); |
| 2028 | U32 state, charid; |
| 2029 | U32 pos = 0, zp=0; |
| 2030 | trie->statecount = laststate; |
| 2031 | |
| 2032 | for ( state = 1 ; state < laststate ; state++ ) { |
| 2033 | U8 flag = 0; |
| 2034 | const U32 stateidx = TRIE_NODEIDX( state ); |
| 2035 | const U32 o_used = trie->trans[ stateidx ].check; |
| 2036 | U32 used = trie->trans[ stateidx ].check; |
| 2037 | trie->trans[ stateidx ].check = 0; |
| 2038 | |
| 2039 | for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) { |
| 2040 | if ( flag || trie->trans[ stateidx + charid ].next ) { |
| 2041 | if ( trie->trans[ stateidx + charid ].next ) { |
| 2042 | if (o_used == 1) { |
| 2043 | for ( ; zp < pos ; zp++ ) { |
| 2044 | if ( ! trie->trans[ zp ].next ) { |
| 2045 | break; |
| 2046 | } |
| 2047 | } |
| 2048 | trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ; |
| 2049 | trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); |
| 2050 | trie->trans[ zp ].check = state; |
| 2051 | if ( ++zp > pos ) pos = zp; |
| 2052 | break; |
| 2053 | } |
| 2054 | used--; |
| 2055 | } |
| 2056 | if ( !flag ) { |
| 2057 | flag = 1; |
| 2058 | trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ; |
| 2059 | } |
| 2060 | trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); |
| 2061 | trie->trans[ pos ].check = state; |
| 2062 | pos++; |
| 2063 | } |
| 2064 | } |
| 2065 | } |
| 2066 | trie->lasttrans = pos + 1; |
| 2067 | trie->states = (reg_trie_state *) |
| 2068 | PerlMemShared_realloc( trie->states, laststate |
| 2069 | * sizeof(reg_trie_state) ); |
| 2070 | DEBUG_TRIE_COMPILE_MORE_r( |
| 2071 | PerlIO_printf( Perl_debug_log, |
| 2072 | "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n", |
| 2073 | (int)depth * 2 + 2,"", |
| 2074 | (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ), |
| 2075 | (IV)next_alloc, |
| 2076 | (IV)pos, |
| 2077 | ( ( next_alloc - pos ) * 100 ) / (double)next_alloc ); |
| 2078 | ); |
| 2079 | |
| 2080 | } /* end table compress */ |
| 2081 | } |
| 2082 | DEBUG_TRIE_COMPILE_MORE_r( |
| 2083 | PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n", |
| 2084 | (int)depth * 2 + 2, "", |
| 2085 | (UV)trie->statecount, |
| 2086 | (UV)trie->lasttrans) |
| 2087 | ); |
| 2088 | /* resize the trans array to remove unused space */ |
| 2089 | trie->trans = (reg_trie_trans *) |
| 2090 | PerlMemShared_realloc( trie->trans, trie->lasttrans |
| 2091 | * sizeof(reg_trie_trans) ); |
| 2092 | |
| 2093 | { /* Modify the program and insert the new TRIE node */ |
| 2094 | U8 nodetype =(U8)(flags & 0xFF); |
| 2095 | char *str=NULL; |
| 2096 | |
| 2097 | #ifdef DEBUGGING |
| 2098 | regnode *optimize = NULL; |
| 2099 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 2100 | |
| 2101 | U32 mjd_offset = 0; |
| 2102 | U32 mjd_nodelen = 0; |
| 2103 | #endif /* RE_TRACK_PATTERN_OFFSETS */ |
| 2104 | #endif /* DEBUGGING */ |
| 2105 | /* |
| 2106 | This means we convert either the first branch or the first Exact, |
| 2107 | depending on whether the thing following (in 'last') is a branch |
| 2108 | or not and whther first is the startbranch (ie is it a sub part of |
| 2109 | the alternation or is it the whole thing.) |
| 2110 | Assuming its a sub part we convert the EXACT otherwise we convert |
| 2111 | the whole branch sequence, including the first. |
| 2112 | */ |
| 2113 | /* Find the node we are going to overwrite */ |
| 2114 | if ( first != startbranch || OP( last ) == BRANCH ) { |
| 2115 | /* branch sub-chain */ |
| 2116 | NEXT_OFF( first ) = (U16)(last - first); |
| 2117 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 2118 | DEBUG_r({ |
| 2119 | mjd_offset= Node_Offset((convert)); |
| 2120 | mjd_nodelen= Node_Length((convert)); |
| 2121 | }); |
| 2122 | #endif |
| 2123 | /* whole branch chain */ |
| 2124 | } |
| 2125 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 2126 | else { |
| 2127 | DEBUG_r({ |
| 2128 | const regnode *nop = NEXTOPER( convert ); |
| 2129 | mjd_offset= Node_Offset((nop)); |
| 2130 | mjd_nodelen= Node_Length((nop)); |
| 2131 | }); |
| 2132 | } |
| 2133 | DEBUG_OPTIMISE_r( |
| 2134 | PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n", |
| 2135 | (int)depth * 2 + 2, "", |
| 2136 | (UV)mjd_offset, (UV)mjd_nodelen) |
| 2137 | ); |
| 2138 | #endif |
| 2139 | /* But first we check to see if there is a common prefix we can |
| 2140 | split out as an EXACT and put in front of the TRIE node. */ |
| 2141 | trie->startstate= 1; |
| 2142 | if ( trie->bitmap && !widecharmap && !trie->jump ) { |
| 2143 | U32 state; |
| 2144 | for ( state = 1 ; state < trie->statecount-1 ; state++ ) { |
| 2145 | U32 ofs = 0; |
| 2146 | I32 idx = -1; |
| 2147 | U32 count = 0; |
| 2148 | const U32 base = trie->states[ state ].trans.base; |
| 2149 | |
| 2150 | if ( trie->states[state].wordnum ) |
| 2151 | count = 1; |
| 2152 | |
| 2153 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { |
| 2154 | if ( ( base + ofs >= trie->uniquecharcount ) && |
| 2155 | ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && |
| 2156 | trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) |
| 2157 | { |
| 2158 | if ( ++count > 1 ) { |
| 2159 | SV **tmp = av_fetch( revcharmap, ofs, 0); |
| 2160 | const U8 *ch = (U8*)SvPV_nolen_const( *tmp ); |
| 2161 | if ( state == 1 ) break; |
| 2162 | if ( count == 2 ) { |
| 2163 | Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char); |
| 2164 | DEBUG_OPTIMISE_r( |
| 2165 | PerlIO_printf(Perl_debug_log, |
| 2166 | "%*sNew Start State=%"UVuf" Class: [", |
| 2167 | (int)depth * 2 + 2, "", |
| 2168 | (UV)state)); |
| 2169 | if (idx >= 0) { |
| 2170 | SV ** const tmp = av_fetch( revcharmap, idx, 0); |
| 2171 | const U8 * const ch = (U8*)SvPV_nolen_const( *tmp ); |
| 2172 | |
| 2173 | TRIE_BITMAP_SET(trie,*ch); |
| 2174 | if ( folder ) |
| 2175 | TRIE_BITMAP_SET(trie, folder[ *ch ]); |
| 2176 | DEBUG_OPTIMISE_r( |
| 2177 | PerlIO_printf(Perl_debug_log, "%s", (char*)ch) |
| 2178 | ); |
| 2179 | } |
| 2180 | } |
| 2181 | TRIE_BITMAP_SET(trie,*ch); |
| 2182 | if ( folder ) |
| 2183 | TRIE_BITMAP_SET(trie,folder[ *ch ]); |
| 2184 | DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch)); |
| 2185 | } |
| 2186 | idx = ofs; |
| 2187 | } |
| 2188 | } |
| 2189 | if ( count == 1 ) { |
| 2190 | SV **tmp = av_fetch( revcharmap, idx, 0); |
| 2191 | STRLEN len; |
| 2192 | char *ch = SvPV( *tmp, len ); |
| 2193 | DEBUG_OPTIMISE_r({ |
| 2194 | SV *sv=sv_newmortal(); |
| 2195 | PerlIO_printf( Perl_debug_log, |
| 2196 | "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n", |
| 2197 | (int)depth * 2 + 2, "", |
| 2198 | (UV)state, (UV)idx, |
| 2199 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6, |
| 2200 | PL_colors[0], PL_colors[1], |
| 2201 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | |
| 2202 | PERL_PV_ESCAPE_FIRSTCHAR |
| 2203 | ) |
| 2204 | ); |
| 2205 | }); |
| 2206 | if ( state==1 ) { |
| 2207 | OP( convert ) = nodetype; |
| 2208 | str=STRING(convert); |
| 2209 | STR_LEN(convert)=0; |
| 2210 | } |
| 2211 | STR_LEN(convert) += len; |
| 2212 | while (len--) |
| 2213 | *str++ = *ch++; |
| 2214 | } else { |
| 2215 | #ifdef DEBUGGING |
| 2216 | if (state>1) |
| 2217 | DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n")); |
| 2218 | #endif |
| 2219 | break; |
| 2220 | } |
| 2221 | } |
| 2222 | trie->prefixlen = (state-1); |
| 2223 | if (str) { |
| 2224 | regnode *n = convert+NODE_SZ_STR(convert); |
| 2225 | NEXT_OFF(convert) = NODE_SZ_STR(convert); |
| 2226 | trie->startstate = state; |
| 2227 | trie->minlen -= (state - 1); |
| 2228 | trie->maxlen -= (state - 1); |
| 2229 | #ifdef DEBUGGING |
| 2230 | /* At least the UNICOS C compiler choked on this |
| 2231 | * being argument to DEBUG_r(), so let's just have |
| 2232 | * it right here. */ |
| 2233 | if ( |
| 2234 | #ifdef PERL_EXT_RE_BUILD |
| 2235 | 1 |
| 2236 | #else |
| 2237 | DEBUG_r_TEST |
| 2238 | #endif |
| 2239 | ) { |
| 2240 | regnode *fix = convert; |
| 2241 | U32 word = trie->wordcount; |
| 2242 | mjd_nodelen++; |
| 2243 | Set_Node_Offset_Length(convert, mjd_offset, state - 1); |
| 2244 | while( ++fix < n ) { |
| 2245 | Set_Node_Offset_Length(fix, 0, 0); |
| 2246 | } |
| 2247 | while (word--) { |
| 2248 | SV ** const tmp = av_fetch( trie_words, word, 0 ); |
| 2249 | if (tmp) { |
| 2250 | if ( STR_LEN(convert) <= SvCUR(*tmp) ) |
| 2251 | sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert)); |
| 2252 | else |
| 2253 | sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp)); |
| 2254 | } |
| 2255 | } |
| 2256 | } |
| 2257 | #endif |
| 2258 | if (trie->maxlen) { |
| 2259 | convert = n; |
| 2260 | } else { |
| 2261 | NEXT_OFF(convert) = (U16)(tail - convert); |
| 2262 | DEBUG_r(optimize= n); |
| 2263 | } |
| 2264 | } |
| 2265 | } |
| 2266 | if (!jumper) |
| 2267 | jumper = last; |
| 2268 | if ( trie->maxlen ) { |
| 2269 | NEXT_OFF( convert ) = (U16)(tail - convert); |
| 2270 | ARG_SET( convert, data_slot ); |
| 2271 | /* Store the offset to the first unabsorbed branch in |
| 2272 | jump[0], which is otherwise unused by the jump logic. |
| 2273 | We use this when dumping a trie and during optimisation. */ |
| 2274 | if (trie->jump) |
| 2275 | trie->jump[0] = (U16)(nextbranch - convert); |
| 2276 | |
| 2277 | /* If the start state is not accepting (meaning there is no empty string/NOTHING) |
| 2278 | * and there is a bitmap |
| 2279 | * and the first "jump target" node we found leaves enough room |
| 2280 | * then convert the TRIE node into a TRIEC node, with the bitmap |
| 2281 | * embedded inline in the opcode - this is hypothetically faster. |
| 2282 | */ |
| 2283 | if ( !trie->states[trie->startstate].wordnum |
| 2284 | && trie->bitmap |
| 2285 | && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) ) |
| 2286 | { |
| 2287 | OP( convert ) = TRIEC; |
| 2288 | Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char); |
| 2289 | PerlMemShared_free(trie->bitmap); |
| 2290 | trie->bitmap= NULL; |
| 2291 | } else |
| 2292 | OP( convert ) = TRIE; |
| 2293 | |
| 2294 | /* store the type in the flags */ |
| 2295 | convert->flags = nodetype; |
| 2296 | DEBUG_r({ |
| 2297 | optimize = convert |
| 2298 | + NODE_STEP_REGNODE |
| 2299 | + regarglen[ OP( convert ) ]; |
| 2300 | }); |
| 2301 | /* XXX We really should free up the resource in trie now, |
| 2302 | as we won't use them - (which resources?) dmq */ |
| 2303 | } |
| 2304 | /* needed for dumping*/ |
| 2305 | DEBUG_r(if (optimize) { |
| 2306 | regnode *opt = convert; |
| 2307 | |
| 2308 | while ( ++opt < optimize) { |
| 2309 | Set_Node_Offset_Length(opt,0,0); |
| 2310 | } |
| 2311 | /* |
| 2312 | Try to clean up some of the debris left after the |
| 2313 | optimisation. |
| 2314 | */ |
| 2315 | while( optimize < jumper ) { |
| 2316 | mjd_nodelen += Node_Length((optimize)); |
| 2317 | OP( optimize ) = OPTIMIZED; |
| 2318 | Set_Node_Offset_Length(optimize,0,0); |
| 2319 | optimize++; |
| 2320 | } |
| 2321 | Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen); |
| 2322 | }); |
| 2323 | } /* end node insert */ |
| 2324 | |
| 2325 | /* Finish populating the prev field of the wordinfo array. Walk back |
| 2326 | * from each accept state until we find another accept state, and if |
| 2327 | * so, point the first word's .prev field at the second word. If the |
| 2328 | * second already has a .prev field set, stop now. This will be the |
| 2329 | * case either if we've already processed that word's accept state, |
| 2330 | * or that state had multiple words, and the overspill words were |
| 2331 | * already linked up earlier. |
| 2332 | */ |
| 2333 | { |
| 2334 | U16 word; |
| 2335 | U32 state; |
| 2336 | U16 prev; |
| 2337 | |
| 2338 | for (word=1; word <= trie->wordcount; word++) { |
| 2339 | prev = 0; |
| 2340 | if (trie->wordinfo[word].prev) |
| 2341 | continue; |
| 2342 | state = trie->wordinfo[word].accept; |
| 2343 | while (state) { |
| 2344 | state = prev_states[state]; |
| 2345 | if (!state) |
| 2346 | break; |
| 2347 | prev = trie->states[state].wordnum; |
| 2348 | if (prev) |
| 2349 | break; |
| 2350 | } |
| 2351 | trie->wordinfo[word].prev = prev; |
| 2352 | } |
| 2353 | Safefree(prev_states); |
| 2354 | } |
| 2355 | |
| 2356 | |
| 2357 | /* and now dump out the compressed format */ |
| 2358 | DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1)); |
| 2359 | |
| 2360 | RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap; |
| 2361 | #ifdef DEBUGGING |
| 2362 | RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words; |
| 2363 | RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap; |
| 2364 | #else |
| 2365 | SvREFCNT_dec(revcharmap); |
| 2366 | #endif |
| 2367 | return trie->jump |
| 2368 | ? MADE_JUMP_TRIE |
| 2369 | : trie->startstate>1 |
| 2370 | ? MADE_EXACT_TRIE |
| 2371 | : MADE_TRIE; |
| 2372 | } |
| 2373 | |
| 2374 | STATIC void |
| 2375 | S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth) |
| 2376 | { |
| 2377 | /* The Trie is constructed and compressed now so we can build a fail array if it's needed |
| 2378 | |
| 2379 | This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the |
| 2380 | "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88 |
| 2381 | ISBN 0-201-10088-6 |
| 2382 | |
| 2383 | We find the fail state for each state in the trie, this state is the longest proper |
| 2384 | suffix of the current state's 'word' that is also a proper prefix of another word in our |
| 2385 | trie. State 1 represents the word '' and is thus the default fail state. This allows |
| 2386 | the DFA not to have to restart after its tried and failed a word at a given point, it |
| 2387 | simply continues as though it had been matching the other word in the first place. |
| 2388 | Consider |
| 2389 | 'abcdgu'=~/abcdefg|cdgu/ |
| 2390 | When we get to 'd' we are still matching the first word, we would encounter 'g' which would |
| 2391 | fail, which would bring us to the state representing 'd' in the second word where we would |
| 2392 | try 'g' and succeed, proceeding to match 'cdgu'. |
| 2393 | */ |
| 2394 | /* add a fail transition */ |
| 2395 | const U32 trie_offset = ARG(source); |
| 2396 | reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset]; |
| 2397 | U32 *q; |
| 2398 | const U32 ucharcount = trie->uniquecharcount; |
| 2399 | const U32 numstates = trie->statecount; |
| 2400 | const U32 ubound = trie->lasttrans + ucharcount; |
| 2401 | U32 q_read = 0; |
| 2402 | U32 q_write = 0; |
| 2403 | U32 charid; |
| 2404 | U32 base = trie->states[ 1 ].trans.base; |
| 2405 | U32 *fail; |
| 2406 | reg_ac_data *aho; |
| 2407 | const U32 data_slot = add_data( pRExC_state, 1, "T" ); |
| 2408 | GET_RE_DEBUG_FLAGS_DECL; |
| 2409 | |
| 2410 | PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE; |
| 2411 | #ifndef DEBUGGING |
| 2412 | PERL_UNUSED_ARG(depth); |
| 2413 | #endif |
| 2414 | |
| 2415 | |
| 2416 | ARG_SET( stclass, data_slot ); |
| 2417 | aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) ); |
| 2418 | RExC_rxi->data->data[ data_slot ] = (void*)aho; |
| 2419 | aho->trie=trie_offset; |
| 2420 | aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) ); |
| 2421 | Copy( trie->states, aho->states, numstates, reg_trie_state ); |
| 2422 | Newxz( q, numstates, U32); |
| 2423 | aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) ); |
| 2424 | aho->refcount = 1; |
| 2425 | fail = aho->fail; |
| 2426 | /* initialize fail[0..1] to be 1 so that we always have |
| 2427 | a valid final fail state */ |
| 2428 | fail[ 0 ] = fail[ 1 ] = 1; |
| 2429 | |
| 2430 | for ( charid = 0; charid < ucharcount ; charid++ ) { |
| 2431 | const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 ); |
| 2432 | if ( newstate ) { |
| 2433 | q[ q_write ] = newstate; |
| 2434 | /* set to point at the root */ |
| 2435 | fail[ q[ q_write++ ] ]=1; |
| 2436 | } |
| 2437 | } |
| 2438 | while ( q_read < q_write) { |
| 2439 | const U32 cur = q[ q_read++ % numstates ]; |
| 2440 | base = trie->states[ cur ].trans.base; |
| 2441 | |
| 2442 | for ( charid = 0 ; charid < ucharcount ; charid++ ) { |
| 2443 | const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 ); |
| 2444 | if (ch_state) { |
| 2445 | U32 fail_state = cur; |
| 2446 | U32 fail_base; |
| 2447 | do { |
| 2448 | fail_state = fail[ fail_state ]; |
| 2449 | fail_base = aho->states[ fail_state ].trans.base; |
| 2450 | } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) ); |
| 2451 | |
| 2452 | fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ); |
| 2453 | fail[ ch_state ] = fail_state; |
| 2454 | if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum ) |
| 2455 | { |
| 2456 | aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum; |
| 2457 | } |
| 2458 | q[ q_write++ % numstates] = ch_state; |
| 2459 | } |
| 2460 | } |
| 2461 | } |
| 2462 | /* restore fail[0..1] to 0 so that we "fall out" of the AC loop |
| 2463 | when we fail in state 1, this allows us to use the |
| 2464 | charclass scan to find a valid start char. This is based on the principle |
| 2465 | that theres a good chance the string being searched contains lots of stuff |
| 2466 | that cant be a start char. |
| 2467 | */ |
| 2468 | fail[ 0 ] = fail[ 1 ] = 0; |
| 2469 | DEBUG_TRIE_COMPILE_r({ |
| 2470 | PerlIO_printf(Perl_debug_log, |
| 2471 | "%*sStclass Failtable (%"UVuf" states): 0", |
| 2472 | (int)(depth * 2), "", (UV)numstates |
| 2473 | ); |
| 2474 | for( q_read=1; q_read<numstates; q_read++ ) { |
| 2475 | PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]); |
| 2476 | } |
| 2477 | PerlIO_printf(Perl_debug_log, "\n"); |
| 2478 | }); |
| 2479 | Safefree(q); |
| 2480 | /*RExC_seen |= REG_SEEN_TRIEDFA;*/ |
| 2481 | } |
| 2482 | |
| 2483 | |
| 2484 | /* |
| 2485 | * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2. |
| 2486 | * These need to be revisited when a newer toolchain becomes available. |
| 2487 | */ |
| 2488 | #if defined(__sparc64__) && defined(__GNUC__) |
| 2489 | # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96) |
| 2490 | # undef SPARC64_GCC_WORKAROUND |
| 2491 | # define SPARC64_GCC_WORKAROUND 1 |
| 2492 | # endif |
| 2493 | #endif |
| 2494 | |
| 2495 | #define DEBUG_PEEP(str,scan,depth) \ |
| 2496 | DEBUG_OPTIMISE_r({if (scan){ \ |
| 2497 | SV * const mysv=sv_newmortal(); \ |
| 2498 | regnode *Next = regnext(scan); \ |
| 2499 | regprop(RExC_rx, mysv, scan); \ |
| 2500 | PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \ |
| 2501 | (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\ |
| 2502 | Next ? (REG_NODE_NUM(Next)) : 0 ); \ |
| 2503 | }}); |
| 2504 | |
| 2505 | |
| 2506 | |
| 2507 | |
| 2508 | |
| 2509 | #define JOIN_EXACT(scan,min,flags) \ |
| 2510 | if (PL_regkind[OP(scan)] == EXACT) \ |
| 2511 | join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1) |
| 2512 | |
| 2513 | STATIC U32 |
| 2514 | S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) { |
| 2515 | /* Merge several consecutive EXACTish nodes into one. */ |
| 2516 | regnode *n = regnext(scan); |
| 2517 | U32 stringok = 1; |
| 2518 | regnode *next = scan + NODE_SZ_STR(scan); |
| 2519 | U32 merged = 0; |
| 2520 | U32 stopnow = 0; |
| 2521 | #ifdef DEBUGGING |
| 2522 | regnode *stop = scan; |
| 2523 | GET_RE_DEBUG_FLAGS_DECL; |
| 2524 | #else |
| 2525 | PERL_UNUSED_ARG(depth); |
| 2526 | #endif |
| 2527 | |
| 2528 | PERL_ARGS_ASSERT_JOIN_EXACT; |
| 2529 | #ifndef EXPERIMENTAL_INPLACESCAN |
| 2530 | PERL_UNUSED_ARG(flags); |
| 2531 | PERL_UNUSED_ARG(val); |
| 2532 | #endif |
| 2533 | DEBUG_PEEP("join",scan,depth); |
| 2534 | |
| 2535 | /* Skip NOTHING, merge EXACT*. */ |
| 2536 | while (n && |
| 2537 | ( PL_regkind[OP(n)] == NOTHING || |
| 2538 | (stringok && (OP(n) == OP(scan)))) |
| 2539 | && NEXT_OFF(n) |
| 2540 | && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) { |
| 2541 | |
| 2542 | if (OP(n) == TAIL || n > next) |
| 2543 | stringok = 0; |
| 2544 | if (PL_regkind[OP(n)] == NOTHING) { |
| 2545 | DEBUG_PEEP("skip:",n,depth); |
| 2546 | NEXT_OFF(scan) += NEXT_OFF(n); |
| 2547 | next = n + NODE_STEP_REGNODE; |
| 2548 | #ifdef DEBUGGING |
| 2549 | if (stringok) |
| 2550 | stop = n; |
| 2551 | #endif |
| 2552 | n = regnext(n); |
| 2553 | } |
| 2554 | else if (stringok) { |
| 2555 | const unsigned int oldl = STR_LEN(scan); |
| 2556 | regnode * const nnext = regnext(n); |
| 2557 | |
| 2558 | DEBUG_PEEP("merg",n,depth); |
| 2559 | |
| 2560 | merged++; |
| 2561 | if (oldl + STR_LEN(n) > U8_MAX) |
| 2562 | break; |
| 2563 | NEXT_OFF(scan) += NEXT_OFF(n); |
| 2564 | STR_LEN(scan) += STR_LEN(n); |
| 2565 | next = n + NODE_SZ_STR(n); |
| 2566 | /* Now we can overwrite *n : */ |
| 2567 | Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char); |
| 2568 | #ifdef DEBUGGING |
| 2569 | stop = next - 1; |
| 2570 | #endif |
| 2571 | n = nnext; |
| 2572 | if (stopnow) break; |
| 2573 | } |
| 2574 | |
| 2575 | #ifdef EXPERIMENTAL_INPLACESCAN |
| 2576 | if (flags && !NEXT_OFF(n)) { |
| 2577 | DEBUG_PEEP("atch", val, depth); |
| 2578 | if (reg_off_by_arg[OP(n)]) { |
| 2579 | ARG_SET(n, val - n); |
| 2580 | } |
| 2581 | else { |
| 2582 | NEXT_OFF(n) = val - n; |
| 2583 | } |
| 2584 | stopnow = 1; |
| 2585 | } |
| 2586 | #endif |
| 2587 | } |
| 2588 | #define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390 |
| 2589 | #define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS |
| 2590 | #define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0 |
| 2591 | #define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS |
| 2592 | |
| 2593 | if (UTF |
| 2594 | && ( OP(scan) == EXACTF || OP(scan) == EXACTFU || OP(scan) == EXACTFA) |
| 2595 | && ( STR_LEN(scan) >= 6 ) ) |
| 2596 | { |
| 2597 | /* |
| 2598 | Two problematic code points in Unicode casefolding of EXACT nodes: |
| 2599 | |
| 2600 | U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS |
| 2601 | U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS |
| 2602 | |
| 2603 | which casefold to |
| 2604 | |
| 2605 | Unicode UTF-8 |
| 2606 | |
| 2607 | U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81 |
| 2608 | U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81 |
| 2609 | |
| 2610 | This means that in case-insensitive matching (or "loose matching", |
| 2611 | as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte |
| 2612 | length of the above casefolded versions) can match a target string |
| 2613 | of length two (the byte length of UTF-8 encoded U+0390 or U+03B0). |
| 2614 | This would rather mess up the minimum length computation. |
| 2615 | |
| 2616 | What we'll do is to look for the tail four bytes, and then peek |
| 2617 | at the preceding two bytes to see whether we need to decrease |
| 2618 | the minimum length by four (six minus two). |
| 2619 | |
| 2620 | Thanks to the design of UTF-8, there cannot be false matches: |
| 2621 | A sequence of valid UTF-8 bytes cannot be a subsequence of |
| 2622 | another valid sequence of UTF-8 bytes. |
| 2623 | |
| 2624 | */ |
| 2625 | char * const s0 = STRING(scan), *s, *t; |
| 2626 | char * const s1 = s0 + STR_LEN(scan) - 1; |
| 2627 | char * const s2 = s1 - 4; |
| 2628 | #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */ |
| 2629 | const char t0[] = "\xaf\x49\xaf\x42"; |
| 2630 | #else |
| 2631 | const char t0[] = "\xcc\x88\xcc\x81"; |
| 2632 | #endif |
| 2633 | const char * const t1 = t0 + 3; |
| 2634 | |
| 2635 | for (s = s0 + 2; |
| 2636 | s < s2 && (t = ninstr(s, s1, t0, t1)); |
| 2637 | s = t + 4) { |
| 2638 | #ifdef EBCDIC |
| 2639 | if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) || |
| 2640 | ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5)) |
| 2641 | #else |
| 2642 | if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) || |
| 2643 | ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF)) |
| 2644 | #endif |
| 2645 | *min -= 4; |
| 2646 | } |
| 2647 | } |
| 2648 | |
| 2649 | #ifdef DEBUGGING |
| 2650 | /* Allow dumping */ |
| 2651 | n = scan + NODE_SZ_STR(scan); |
| 2652 | while (n <= stop) { |
| 2653 | if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) { |
| 2654 | OP(n) = OPTIMIZED; |
| 2655 | NEXT_OFF(n) = 0; |
| 2656 | } |
| 2657 | n++; |
| 2658 | } |
| 2659 | #endif |
| 2660 | DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)}); |
| 2661 | return stopnow; |
| 2662 | } |
| 2663 | |
| 2664 | /* REx optimizer. Converts nodes into quicker variants "in place". |
| 2665 | Finds fixed substrings. */ |
| 2666 | |
| 2667 | /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set |
| 2668 | to the position after last scanned or to NULL. */ |
| 2669 | |
| 2670 | #define INIT_AND_WITHP \ |
| 2671 | assert(!and_withp); \ |
| 2672 | Newx(and_withp,1,struct regnode_charclass_class); \ |
| 2673 | SAVEFREEPV(and_withp) |
| 2674 | |
| 2675 | /* this is a chain of data about sub patterns we are processing that |
| 2676 | need to be handled separately/specially in study_chunk. Its so |
| 2677 | we can simulate recursion without losing state. */ |
| 2678 | struct scan_frame; |
| 2679 | typedef struct scan_frame { |
| 2680 | regnode *last; /* last node to process in this frame */ |
| 2681 | regnode *next; /* next node to process when last is reached */ |
| 2682 | struct scan_frame *prev; /*previous frame*/ |
| 2683 | I32 stop; /* what stopparen do we use */ |
| 2684 | } scan_frame; |
| 2685 | |
| 2686 | |
| 2687 | #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf) |
| 2688 | |
| 2689 | #define CASE_SYNST_FNC(nAmE) \ |
| 2690 | case nAmE: \ |
| 2691 | if (flags & SCF_DO_STCLASS_AND) { \ |
| 2692 | for (value = 0; value < 256; value++) \ |
| 2693 | if (!is_ ## nAmE ## _cp(value)) \ |
| 2694 | ANYOF_BITMAP_CLEAR(data->start_class, value); \ |
| 2695 | } \ |
| 2696 | else { \ |
| 2697 | for (value = 0; value < 256; value++) \ |
| 2698 | if (is_ ## nAmE ## _cp(value)) \ |
| 2699 | ANYOF_BITMAP_SET(data->start_class, value); \ |
| 2700 | } \ |
| 2701 | break; \ |
| 2702 | case N ## nAmE: \ |
| 2703 | if (flags & SCF_DO_STCLASS_AND) { \ |
| 2704 | for (value = 0; value < 256; value++) \ |
| 2705 | if (is_ ## nAmE ## _cp(value)) \ |
| 2706 | ANYOF_BITMAP_CLEAR(data->start_class, value); \ |
| 2707 | } \ |
| 2708 | else { \ |
| 2709 | for (value = 0; value < 256; value++) \ |
| 2710 | if (!is_ ## nAmE ## _cp(value)) \ |
| 2711 | ANYOF_BITMAP_SET(data->start_class, value); \ |
| 2712 | } \ |
| 2713 | break |
| 2714 | |
| 2715 | |
| 2716 | |
| 2717 | STATIC I32 |
| 2718 | S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, |
| 2719 | I32 *minlenp, I32 *deltap, |
| 2720 | regnode *last, |
| 2721 | scan_data_t *data, |
| 2722 | I32 stopparen, |
| 2723 | U8* recursed, |
| 2724 | struct regnode_charclass_class *and_withp, |
| 2725 | U32 flags, U32 depth) |
| 2726 | /* scanp: Start here (read-write). */ |
| 2727 | /* deltap: Write maxlen-minlen here. */ |
| 2728 | /* last: Stop before this one. */ |
| 2729 | /* data: string data about the pattern */ |
| 2730 | /* stopparen: treat close N as END */ |
| 2731 | /* recursed: which subroutines have we recursed into */ |
| 2732 | /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */ |
| 2733 | { |
| 2734 | dVAR; |
| 2735 | I32 min = 0, pars = 0, code; |
| 2736 | regnode *scan = *scanp, *next; |
| 2737 | I32 delta = 0; |
| 2738 | int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF); |
| 2739 | int is_inf_internal = 0; /* The studied chunk is infinite */ |
| 2740 | I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0; |
| 2741 | scan_data_t data_fake; |
| 2742 | SV *re_trie_maxbuff = NULL; |
| 2743 | regnode *first_non_open = scan; |
| 2744 | I32 stopmin = I32_MAX; |
| 2745 | scan_frame *frame = NULL; |
| 2746 | GET_RE_DEBUG_FLAGS_DECL; |
| 2747 | |
| 2748 | PERL_ARGS_ASSERT_STUDY_CHUNK; |
| 2749 | |
| 2750 | #ifdef DEBUGGING |
| 2751 | StructCopy(&zero_scan_data, &data_fake, scan_data_t); |
| 2752 | #endif |
| 2753 | |
| 2754 | if ( depth == 0 ) { |
| 2755 | while (first_non_open && OP(first_non_open) == OPEN) |
| 2756 | first_non_open=regnext(first_non_open); |
| 2757 | } |
| 2758 | |
| 2759 | |
| 2760 | fake_study_recurse: |
| 2761 | while ( scan && OP(scan) != END && scan < last ){ |
| 2762 | /* Peephole optimizer: */ |
| 2763 | DEBUG_STUDYDATA("Peep:", data,depth); |
| 2764 | DEBUG_PEEP("Peep",scan,depth); |
| 2765 | JOIN_EXACT(scan,&min,0); |
| 2766 | |
| 2767 | /* Follow the next-chain of the current node and optimize |
| 2768 | away all the NOTHINGs from it. */ |
| 2769 | if (OP(scan) != CURLYX) { |
| 2770 | const int max = (reg_off_by_arg[OP(scan)] |
| 2771 | ? I32_MAX |
| 2772 | /* I32 may be smaller than U16 on CRAYs! */ |
| 2773 | : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX)); |
| 2774 | int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan)); |
| 2775 | int noff; |
| 2776 | regnode *n = scan; |
| 2777 | |
| 2778 | /* Skip NOTHING and LONGJMP. */ |
| 2779 | while ((n = regnext(n)) |
| 2780 | && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n))) |
| 2781 | || ((OP(n) == LONGJMP) && (noff = ARG(n)))) |
| 2782 | && off + noff < max) |
| 2783 | off += noff; |
| 2784 | if (reg_off_by_arg[OP(scan)]) |
| 2785 | ARG(scan) = off; |
| 2786 | else |
| 2787 | NEXT_OFF(scan) = off; |
| 2788 | } |
| 2789 | |
| 2790 | |
| 2791 | |
| 2792 | /* The principal pseudo-switch. Cannot be a switch, since we |
| 2793 | look into several different things. */ |
| 2794 | if (OP(scan) == BRANCH || OP(scan) == BRANCHJ |
| 2795 | || OP(scan) == IFTHEN) { |
| 2796 | next = regnext(scan); |
| 2797 | code = OP(scan); |
| 2798 | /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */ |
| 2799 | |
| 2800 | if (OP(next) == code || code == IFTHEN) { |
| 2801 | /* NOTE - There is similar code to this block below for handling |
| 2802 | TRIE nodes on a re-study. If you change stuff here check there |
| 2803 | too. */ |
| 2804 | I32 max1 = 0, min1 = I32_MAX, num = 0; |
| 2805 | struct regnode_charclass_class accum; |
| 2806 | regnode * const startbranch=scan; |
| 2807 | |
| 2808 | if (flags & SCF_DO_SUBSTR) |
| 2809 | SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */ |
| 2810 | if (flags & SCF_DO_STCLASS) |
| 2811 | cl_init_zero(pRExC_state, &accum); |
| 2812 | |
| 2813 | while (OP(scan) == code) { |
| 2814 | I32 deltanext, minnext, f = 0, fake; |
| 2815 | struct regnode_charclass_class this_class; |
| 2816 | |
| 2817 | num++; |
| 2818 | data_fake.flags = 0; |
| 2819 | if (data) { |
| 2820 | data_fake.whilem_c = data->whilem_c; |
| 2821 | data_fake.last_closep = data->last_closep; |
| 2822 | } |
| 2823 | else |
| 2824 | data_fake.last_closep = &fake; |
| 2825 | |
| 2826 | data_fake.pos_delta = delta; |
| 2827 | next = regnext(scan); |
| 2828 | scan = NEXTOPER(scan); |
| 2829 | if (code != BRANCH) |
| 2830 | scan = NEXTOPER(scan); |
| 2831 | if (flags & SCF_DO_STCLASS) { |
| 2832 | cl_init(pRExC_state, &this_class); |
| 2833 | data_fake.start_class = &this_class; |
| 2834 | f = SCF_DO_STCLASS_AND; |
| 2835 | } |
| 2836 | if (flags & SCF_WHILEM_VISITED_POS) |
| 2837 | f |= SCF_WHILEM_VISITED_POS; |
| 2838 | |
| 2839 | /* we suppose the run is continuous, last=next...*/ |
| 2840 | minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, |
| 2841 | next, &data_fake, |
| 2842 | stopparen, recursed, NULL, f,depth+1); |
| 2843 | if (min1 > minnext) |
| 2844 | min1 = minnext; |
| 2845 | if (max1 < minnext + deltanext) |
| 2846 | max1 = minnext + deltanext; |
| 2847 | if (deltanext == I32_MAX) |
| 2848 | is_inf = is_inf_internal = 1; |
| 2849 | scan = next; |
| 2850 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 2851 | pars++; |
| 2852 | if (data_fake.flags & SCF_SEEN_ACCEPT) { |
| 2853 | if ( stopmin > minnext) |
| 2854 | stopmin = min + min1; |
| 2855 | flags &= ~SCF_DO_SUBSTR; |
| 2856 | if (data) |
| 2857 | data->flags |= SCF_SEEN_ACCEPT; |
| 2858 | } |
| 2859 | if (data) { |
| 2860 | if (data_fake.flags & SF_HAS_EVAL) |
| 2861 | data->flags |= SF_HAS_EVAL; |
| 2862 | data->whilem_c = data_fake.whilem_c; |
| 2863 | } |
| 2864 | if (flags & SCF_DO_STCLASS) |
| 2865 | cl_or(pRExC_state, &accum, &this_class); |
| 2866 | } |
| 2867 | if (code == IFTHEN && num < 2) /* Empty ELSE branch */ |
| 2868 | min1 = 0; |
| 2869 | if (flags & SCF_DO_SUBSTR) { |
| 2870 | data->pos_min += min1; |
| 2871 | data->pos_delta += max1 - min1; |
| 2872 | if (max1 != min1 || is_inf) |
| 2873 | data->longest = &(data->longest_float); |
| 2874 | } |
| 2875 | min += min1; |
| 2876 | delta += max1 - min1; |
| 2877 | if (flags & SCF_DO_STCLASS_OR) { |
| 2878 | cl_or(pRExC_state, data->start_class, &accum); |
| 2879 | if (min1) { |
| 2880 | cl_and(data->start_class, and_withp); |
| 2881 | flags &= ~SCF_DO_STCLASS; |
| 2882 | } |
| 2883 | } |
| 2884 | else if (flags & SCF_DO_STCLASS_AND) { |
| 2885 | if (min1) { |
| 2886 | cl_and(data->start_class, &accum); |
| 2887 | flags &= ~SCF_DO_STCLASS; |
| 2888 | } |
| 2889 | else { |
| 2890 | /* Switch to OR mode: cache the old value of |
| 2891 | * data->start_class */ |
| 2892 | INIT_AND_WITHP; |
| 2893 | StructCopy(data->start_class, and_withp, |
| 2894 | struct regnode_charclass_class); |
| 2895 | flags &= ~SCF_DO_STCLASS_AND; |
| 2896 | StructCopy(&accum, data->start_class, |
| 2897 | struct regnode_charclass_class); |
| 2898 | flags |= SCF_DO_STCLASS_OR; |
| 2899 | data->start_class->flags |= ANYOF_EOS; |
| 2900 | } |
| 2901 | } |
| 2902 | |
| 2903 | if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) { |
| 2904 | /* demq. |
| 2905 | |
| 2906 | Assuming this was/is a branch we are dealing with: 'scan' now |
| 2907 | points at the item that follows the branch sequence, whatever |
| 2908 | it is. We now start at the beginning of the sequence and look |
| 2909 | for subsequences of |
| 2910 | |
| 2911 | BRANCH->EXACT=>x1 |
| 2912 | BRANCH->EXACT=>x2 |
| 2913 | tail |
| 2914 | |
| 2915 | which would be constructed from a pattern like /A|LIST|OF|WORDS/ |
| 2916 | |
| 2917 | If we can find such a subsequence we need to turn the first |
| 2918 | element into a trie and then add the subsequent branch exact |
| 2919 | strings to the trie. |
| 2920 | |
| 2921 | We have two cases |
| 2922 | |
| 2923 | 1. patterns where the whole set of branches can be converted. |
| 2924 | |
| 2925 | 2. patterns where only a subset can be converted. |
| 2926 | |
| 2927 | In case 1 we can replace the whole set with a single regop |
| 2928 | for the trie. In case 2 we need to keep the start and end |
| 2929 | branches so |
| 2930 | |
| 2931 | 'BRANCH EXACT; BRANCH EXACT; BRANCH X' |
| 2932 | becomes BRANCH TRIE; BRANCH X; |
| 2933 | |
| 2934 | There is an additional case, that being where there is a |
| 2935 | common prefix, which gets split out into an EXACT like node |
| 2936 | preceding the TRIE node. |
| 2937 | |
| 2938 | If x(1..n)==tail then we can do a simple trie, if not we make |
| 2939 | a "jump" trie, such that when we match the appropriate word |
| 2940 | we "jump" to the appropriate tail node. Essentially we turn |
| 2941 | a nested if into a case structure of sorts. |
| 2942 | |
| 2943 | */ |
| 2944 | |
| 2945 | int made=0; |
| 2946 | if (!re_trie_maxbuff) { |
| 2947 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); |
| 2948 | if (!SvIOK(re_trie_maxbuff)) |
| 2949 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); |
| 2950 | } |
| 2951 | if ( SvIV(re_trie_maxbuff)>=0 ) { |
| 2952 | regnode *cur; |
| 2953 | regnode *first = (regnode *)NULL; |
| 2954 | regnode *last = (regnode *)NULL; |
| 2955 | regnode *tail = scan; |
| 2956 | U8 optype = 0; |
| 2957 | U32 count=0; |
| 2958 | |
| 2959 | #ifdef DEBUGGING |
| 2960 | SV * const mysv = sv_newmortal(); /* for dumping */ |
| 2961 | #endif |
| 2962 | /* var tail is used because there may be a TAIL |
| 2963 | regop in the way. Ie, the exacts will point to the |
| 2964 | thing following the TAIL, but the last branch will |
| 2965 | point at the TAIL. So we advance tail. If we |
| 2966 | have nested (?:) we may have to move through several |
| 2967 | tails. |
| 2968 | */ |
| 2969 | |
| 2970 | while ( OP( tail ) == TAIL ) { |
| 2971 | /* this is the TAIL generated by (?:) */ |
| 2972 | tail = regnext( tail ); |
| 2973 | } |
| 2974 | |
| 2975 | |
| 2976 | DEBUG_OPTIMISE_r({ |
| 2977 | regprop(RExC_rx, mysv, tail ); |
| 2978 | PerlIO_printf( Perl_debug_log, "%*s%s%s\n", |
| 2979 | (int)depth * 2 + 2, "", |
| 2980 | "Looking for TRIE'able sequences. Tail node is: ", |
| 2981 | SvPV_nolen_const( mysv ) |
| 2982 | ); |
| 2983 | }); |
| 2984 | |
| 2985 | /* |
| 2986 | |
| 2987 | step through the branches, cur represents each |
| 2988 | branch, noper is the first thing to be matched |
| 2989 | as part of that branch and noper_next is the |
| 2990 | regnext() of that node. if noper is an EXACT |
| 2991 | and noper_next is the same as scan (our current |
| 2992 | position in the regex) then the EXACT branch is |
| 2993 | a possible optimization target. Once we have |
| 2994 | two or more consecutive such branches we can |
| 2995 | create a trie of the EXACT's contents and stich |
| 2996 | it in place. If the sequence represents all of |
| 2997 | the branches we eliminate the whole thing and |
| 2998 | replace it with a single TRIE. If it is a |
| 2999 | subsequence then we need to stitch it in. This |
| 3000 | means the first branch has to remain, and needs |
| 3001 | to be repointed at the item on the branch chain |
| 3002 | following the last branch optimized. This could |
| 3003 | be either a BRANCH, in which case the |
| 3004 | subsequence is internal, or it could be the |
| 3005 | item following the branch sequence in which |
| 3006 | case the subsequence is at the end. |
| 3007 | |
| 3008 | */ |
| 3009 | |
| 3010 | /* dont use tail as the end marker for this traverse */ |
| 3011 | for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) { |
| 3012 | regnode * const noper = NEXTOPER( cur ); |
| 3013 | #if defined(DEBUGGING) || defined(NOJUMPTRIE) |
| 3014 | regnode * const noper_next = regnext( noper ); |
| 3015 | #endif |
| 3016 | |
| 3017 | DEBUG_OPTIMISE_r({ |
| 3018 | regprop(RExC_rx, mysv, cur); |
| 3019 | PerlIO_printf( Perl_debug_log, "%*s- %s (%d)", |
| 3020 | (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) ); |
| 3021 | |
| 3022 | regprop(RExC_rx, mysv, noper); |
| 3023 | PerlIO_printf( Perl_debug_log, " -> %s", |
| 3024 | SvPV_nolen_const(mysv)); |
| 3025 | |
| 3026 | if ( noper_next ) { |
| 3027 | regprop(RExC_rx, mysv, noper_next ); |
| 3028 | PerlIO_printf( Perl_debug_log,"\t=> %s\t", |
| 3029 | SvPV_nolen_const(mysv)); |
| 3030 | } |
| 3031 | PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n", |
| 3032 | REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) ); |
| 3033 | }); |
| 3034 | if ( (((first && optype!=NOTHING) ? OP( noper ) == optype |
| 3035 | : PL_regkind[ OP( noper ) ] == EXACT ) |
| 3036 | || OP(noper) == NOTHING ) |
| 3037 | #ifdef NOJUMPTRIE |
| 3038 | && noper_next == tail |
| 3039 | #endif |
| 3040 | && count < U16_MAX) |
| 3041 | { |
| 3042 | count++; |
| 3043 | if ( !first || optype == NOTHING ) { |
| 3044 | if (!first) first = cur; |
| 3045 | optype = OP( noper ); |
| 3046 | } else { |
| 3047 | last = cur; |
| 3048 | } |
| 3049 | } else { |
| 3050 | /* |
| 3051 | Currently the trie logic handles case insensitive matching properly only |
| 3052 | when the pattern is UTF-8 and the node is EXACTFU (thus forcing unicode |
| 3053 | semantics). |
| 3054 | |
| 3055 | If/when this is fixed the following define can be swapped |
| 3056 | in below to fully enable trie logic. |
| 3057 | |
| 3058 | #define TRIE_TYPE_IS_SAFE 1 |
| 3059 | |
| 3060 | */ |
| 3061 | #define TRIE_TYPE_IS_SAFE ((UTF && optype == EXACTFU) || optype==EXACT) |
| 3062 | |
| 3063 | if ( last && TRIE_TYPE_IS_SAFE ) { |
| 3064 | make_trie( pRExC_state, |
| 3065 | startbranch, first, cur, tail, count, |
| 3066 | optype, depth+1 ); |
| 3067 | } |
| 3068 | if ( PL_regkind[ OP( noper ) ] == EXACT |
| 3069 | #ifdef NOJUMPTRIE |
| 3070 | && noper_next == tail |
| 3071 | #endif |
| 3072 | ){ |
| 3073 | count = 1; |
| 3074 | first = cur; |
| 3075 | optype = OP( noper ); |
| 3076 | } else { |
| 3077 | count = 0; |
| 3078 | first = NULL; |
| 3079 | optype = 0; |
| 3080 | } |
| 3081 | last = NULL; |
| 3082 | } |
| 3083 | } |
| 3084 | DEBUG_OPTIMISE_r({ |
| 3085 | regprop(RExC_rx, mysv, cur); |
| 3086 | PerlIO_printf( Perl_debug_log, |
| 3087 | "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2, |
| 3088 | "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur)); |
| 3089 | |
| 3090 | }); |
| 3091 | |
| 3092 | if ( last && TRIE_TYPE_IS_SAFE ) { |
| 3093 | made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 ); |
| 3094 | #ifdef TRIE_STUDY_OPT |
| 3095 | if ( ((made == MADE_EXACT_TRIE && |
| 3096 | startbranch == first) |
| 3097 | || ( first_non_open == first )) && |
| 3098 | depth==0 ) { |
| 3099 | flags |= SCF_TRIE_RESTUDY; |
| 3100 | if ( startbranch == first |
| 3101 | && scan == tail ) |
| 3102 | { |
| 3103 | RExC_seen &=~REG_TOP_LEVEL_BRANCHES; |
| 3104 | } |
| 3105 | } |
| 3106 | #endif |
| 3107 | } |
| 3108 | } |
| 3109 | |
| 3110 | } /* do trie */ |
| 3111 | |
| 3112 | } |
| 3113 | else if ( code == BRANCHJ ) { /* single branch is optimized. */ |
| 3114 | scan = NEXTOPER(NEXTOPER(scan)); |
| 3115 | } else /* single branch is optimized. */ |
| 3116 | scan = NEXTOPER(scan); |
| 3117 | continue; |
| 3118 | } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) { |
| 3119 | scan_frame *newframe = NULL; |
| 3120 | I32 paren; |
| 3121 | regnode *start; |
| 3122 | regnode *end; |
| 3123 | |
| 3124 | if (OP(scan) != SUSPEND) { |
| 3125 | /* set the pointer */ |
| 3126 | if (OP(scan) == GOSUB) { |
| 3127 | paren = ARG(scan); |
| 3128 | RExC_recurse[ARG2L(scan)] = scan; |
| 3129 | start = RExC_open_parens[paren-1]; |
| 3130 | end = RExC_close_parens[paren-1]; |
| 3131 | } else { |
| 3132 | paren = 0; |
| 3133 | start = RExC_rxi->program + 1; |
| 3134 | end = RExC_opend; |
| 3135 | } |
| 3136 | if (!recursed) { |
| 3137 | Newxz(recursed, (((RExC_npar)>>3) +1), U8); |
| 3138 | SAVEFREEPV(recursed); |
| 3139 | } |
| 3140 | if (!PAREN_TEST(recursed,paren+1)) { |
| 3141 | PAREN_SET(recursed,paren+1); |
| 3142 | Newx(newframe,1,scan_frame); |
| 3143 | } else { |
| 3144 | if (flags & SCF_DO_SUBSTR) { |
| 3145 | SCAN_COMMIT(pRExC_state,data,minlenp); |
| 3146 | data->longest = &(data->longest_float); |
| 3147 | } |
| 3148 | is_inf = is_inf_internal = 1; |
| 3149 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
| 3150 | cl_anything(pRExC_state, data->start_class); |
| 3151 | flags &= ~SCF_DO_STCLASS; |
| 3152 | } |
| 3153 | } else { |
| 3154 | Newx(newframe,1,scan_frame); |
| 3155 | paren = stopparen; |
| 3156 | start = scan+2; |
| 3157 | end = regnext(scan); |
| 3158 | } |
| 3159 | if (newframe) { |
| 3160 | assert(start); |
| 3161 | assert(end); |
| 3162 | SAVEFREEPV(newframe); |
| 3163 | newframe->next = regnext(scan); |
| 3164 | newframe->last = last; |
| 3165 | newframe->stop = stopparen; |
| 3166 | newframe->prev = frame; |
| 3167 | |
| 3168 | frame = newframe; |
| 3169 | scan = start; |
| 3170 | stopparen = paren; |
| 3171 | last = end; |
| 3172 | |
| 3173 | continue; |
| 3174 | } |
| 3175 | } |
| 3176 | else if (OP(scan) == EXACT) { |
| 3177 | I32 l = STR_LEN(scan); |
| 3178 | UV uc; |
| 3179 | if (UTF) { |
| 3180 | const U8 * const s = (U8*)STRING(scan); |
| 3181 | l = utf8_length(s, s + l); |
| 3182 | uc = utf8_to_uvchr(s, NULL); |
| 3183 | } else { |
| 3184 | uc = *((U8*)STRING(scan)); |
| 3185 | } |
| 3186 | min += l; |
| 3187 | if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */ |
| 3188 | /* The code below prefers earlier match for fixed |
| 3189 | offset, later match for variable offset. */ |
| 3190 | if (data->last_end == -1) { /* Update the start info. */ |
| 3191 | data->last_start_min = data->pos_min; |
| 3192 | data->last_start_max = is_inf |
| 3193 | ? I32_MAX : data->pos_min + data->pos_delta; |
| 3194 | } |
| 3195 | sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan)); |
| 3196 | if (UTF) |
| 3197 | SvUTF8_on(data->last_found); |
| 3198 | { |
| 3199 | SV * const sv = data->last_found; |
| 3200 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? |
| 3201 | mg_find(sv, PERL_MAGIC_utf8) : NULL; |
| 3202 | if (mg && mg->mg_len >= 0) |
| 3203 | mg->mg_len += utf8_length((U8*)STRING(scan), |
| 3204 | (U8*)STRING(scan)+STR_LEN(scan)); |
| 3205 | } |
| 3206 | data->last_end = data->pos_min + l; |
| 3207 | data->pos_min += l; /* As in the first entry. */ |
| 3208 | data->flags &= ~SF_BEFORE_EOL; |
| 3209 | } |
| 3210 | if (flags & SCF_DO_STCLASS_AND) { |
| 3211 | /* Check whether it is compatible with what we know already! */ |
| 3212 | int compat = 1; |
| 3213 | |
| 3214 | |
| 3215 | /* If compatible, we or it in below. It is compatible if is |
| 3216 | * in the bitmp and either 1) its bit or its fold is set, or 2) |
| 3217 | * it's for a locale. Even if there isn't unicode semantics |
| 3218 | * here, at runtime there may be because of matching against a |
| 3219 | * utf8 string, so accept a possible false positive for |
| 3220 | * latin1-range folds */ |
| 3221 | if (uc >= 0x100 || |
| 3222 | (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) |
| 3223 | && !ANYOF_BITMAP_TEST(data->start_class, uc) |
| 3224 | && (!(data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) |
| 3225 | || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc]))) |
| 3226 | ) |
| 3227 | { |
| 3228 | compat = 0; |
| 3229 | } |
| 3230 | ANYOF_CLASS_ZERO(data->start_class); |
| 3231 | ANYOF_BITMAP_ZERO(data->start_class); |
| 3232 | if (compat) |
| 3233 | ANYOF_BITMAP_SET(data->start_class, uc); |
| 3234 | else if (uc >= 0x100) { |
| 3235 | int i; |
| 3236 | |
| 3237 | /* Some Unicode code points fold to the Latin1 range; as |
| 3238 | * XXX temporary code, instead of figuring out if this is |
| 3239 | * one, just assume it is and set all the start class bits |
| 3240 | * that could be some such above 255 code point's fold |
| 3241 | * which will generate fals positives. As the code |
| 3242 | * elsewhere that does compute the fold settles down, it |
| 3243 | * can be extracted out and re-used here */ |
| 3244 | for (i = 0; i < 256; i++){ |
| 3245 | if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) { |
| 3246 | ANYOF_BITMAP_SET(data->start_class, i); |
| 3247 | } |
| 3248 | } |
| 3249 | } |
| 3250 | data->start_class->flags &= ~ANYOF_EOS; |
| 3251 | if (uc < 0x100) |
| 3252 | data->start_class->flags &= ~ANYOF_UNICODE_ALL; |
| 3253 | } |
| 3254 | else if (flags & SCF_DO_STCLASS_OR) { |
| 3255 | /* false positive possible if the class is case-folded */ |
| 3256 | if (uc < 0x100) |
| 3257 | ANYOF_BITMAP_SET(data->start_class, uc); |
| 3258 | else |
| 3259 | data->start_class->flags |= ANYOF_UNICODE_ALL; |
| 3260 | data->start_class->flags &= ~ANYOF_EOS; |
| 3261 | cl_and(data->start_class, and_withp); |
| 3262 | } |
| 3263 | flags &= ~SCF_DO_STCLASS; |
| 3264 | } |
| 3265 | else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */ |
| 3266 | I32 l = STR_LEN(scan); |
| 3267 | UV uc = *((U8*)STRING(scan)); |
| 3268 | |
| 3269 | /* Search for fixed substrings supports EXACT only. */ |
| 3270 | if (flags & SCF_DO_SUBSTR) { |
| 3271 | assert(data); |
| 3272 | SCAN_COMMIT(pRExC_state, data, minlenp); |
| 3273 | } |
| 3274 | if (UTF) { |
| 3275 | const U8 * const s = (U8 *)STRING(scan); |
| 3276 | l = utf8_length(s, s + l); |
| 3277 | uc = utf8_to_uvchr(s, NULL); |
| 3278 | } |
| 3279 | min += l; |
| 3280 | if (flags & SCF_DO_SUBSTR) |
| 3281 | data->pos_min += l; |
| 3282 | if (flags & SCF_DO_STCLASS_AND) { |
| 3283 | /* Check whether it is compatible with what we know already! */ |
| 3284 | int compat = 1; |
| 3285 | if (uc >= 0x100 || |
| 3286 | (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) |
| 3287 | && !ANYOF_BITMAP_TEST(data->start_class, uc) |
| 3288 | && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc]))) |
| 3289 | { |
| 3290 | compat = 0; |
| 3291 | } |
| 3292 | ANYOF_CLASS_ZERO(data->start_class); |
| 3293 | ANYOF_BITMAP_ZERO(data->start_class); |
| 3294 | if (compat) { |
| 3295 | ANYOF_BITMAP_SET(data->start_class, uc); |
| 3296 | data->start_class->flags &= ~ANYOF_EOS; |
| 3297 | data->start_class->flags |= ANYOF_LOC_NONBITMAP_FOLD; |
| 3298 | if (OP(scan) == EXACTFL) { |
| 3299 | /* XXX This set is probably no longer necessary, and |
| 3300 | * probably wrong as LOCALE now is on in the initial |
| 3301 | * state */ |
| 3302 | data->start_class->flags |= ANYOF_LOCALE; |
| 3303 | } |
| 3304 | else { |
| 3305 | |
| 3306 | /* Also set the other member of the fold pair. In case |
| 3307 | * that unicode semantics is called for at runtime, use |
| 3308 | * the full latin1 fold. (Can't do this for locale, |
| 3309 | * because not known until runtime */ |
| 3310 | ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]); |
| 3311 | } |
| 3312 | } |
| 3313 | else if (uc >= 0x100) { |
| 3314 | int i; |
| 3315 | for (i = 0; i < 256; i++){ |
| 3316 | if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) { |
| 3317 | ANYOF_BITMAP_SET(data->start_class, i); |
| 3318 | } |
| 3319 | } |
| 3320 | } |
| 3321 | } |
| 3322 | else if (flags & SCF_DO_STCLASS_OR) { |
| 3323 | if (data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) { |
| 3324 | /* false positive possible if the class is case-folded. |
| 3325 | Assume that the locale settings are the same... */ |
| 3326 | if (uc < 0x100) { |
| 3327 | ANYOF_BITMAP_SET(data->start_class, uc); |
| 3328 | if (OP(scan) != EXACTFL) { |
| 3329 | |
| 3330 | /* And set the other member of the fold pair, but |
| 3331 | * can't do that in locale because not known until |
| 3332 | * run-time */ |
| 3333 | ANYOF_BITMAP_SET(data->start_class, |
| 3334 | PL_fold_latin1[uc]); |
| 3335 | } |
| 3336 | } |
| 3337 | data->start_class->flags &= ~ANYOF_EOS; |
| 3338 | } |
| 3339 | cl_and(data->start_class, and_withp); |
| 3340 | } |
| 3341 | flags &= ~SCF_DO_STCLASS; |
| 3342 | } |
| 3343 | else if (REGNODE_VARIES(OP(scan))) { |
| 3344 | I32 mincount, maxcount, minnext, deltanext, fl = 0; |
| 3345 | I32 f = flags, pos_before = 0; |
| 3346 | regnode * const oscan = scan; |
| 3347 | struct regnode_charclass_class this_class; |
| 3348 | struct regnode_charclass_class *oclass = NULL; |
| 3349 | I32 next_is_eval = 0; |
| 3350 | |
| 3351 | switch (PL_regkind[OP(scan)]) { |
| 3352 | case WHILEM: /* End of (?:...)* . */ |
| 3353 | scan = NEXTOPER(scan); |
| 3354 | goto finish; |
| 3355 | case PLUS: |
| 3356 | if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) { |
| 3357 | next = NEXTOPER(scan); |
| 3358 | if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) { |
| 3359 | mincount = 1; |
| 3360 | maxcount = REG_INFTY; |
| 3361 | next = regnext(scan); |
| 3362 | scan = NEXTOPER(scan); |
| 3363 | goto do_curly; |
| 3364 | } |
| 3365 | } |
| 3366 | if (flags & SCF_DO_SUBSTR) |
| 3367 | data->pos_min++; |
| 3368 | min++; |
| 3369 | /* Fall through. */ |
| 3370 | case STAR: |
| 3371 | if (flags & SCF_DO_STCLASS) { |
| 3372 | mincount = 0; |
| 3373 | maxcount = REG_INFTY; |
| 3374 | next = regnext(scan); |
| 3375 | scan = NEXTOPER(scan); |
| 3376 | goto do_curly; |
| 3377 | } |
| 3378 | is_inf = is_inf_internal = 1; |
| 3379 | scan = regnext(scan); |
| 3380 | if (flags & SCF_DO_SUBSTR) { |
| 3381 | SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */ |
| 3382 | data->longest = &(data->longest_float); |
| 3383 | } |
| 3384 | goto optimize_curly_tail; |
| 3385 | case CURLY: |
| 3386 | if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM) |
| 3387 | && (scan->flags == stopparen)) |
| 3388 | { |
| 3389 | mincount = 1; |
| 3390 | maxcount = 1; |
| 3391 | } else { |
| 3392 | mincount = ARG1(scan); |
| 3393 | maxcount = ARG2(scan); |
| 3394 | } |
| 3395 | next = regnext(scan); |
| 3396 | if (OP(scan) == CURLYX) { |
| 3397 | I32 lp = (data ? *(data->last_closep) : 0); |
| 3398 | scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX); |
| 3399 | } |
| 3400 | scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; |
| 3401 | next_is_eval = (OP(scan) == EVAL); |
| 3402 | do_curly: |
| 3403 | if (flags & SCF_DO_SUBSTR) { |
| 3404 | if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */ |
| 3405 | pos_before = data->pos_min; |
| 3406 | } |
| 3407 | if (data) { |
| 3408 | fl = data->flags; |
| 3409 | data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL); |
| 3410 | if (is_inf) |
| 3411 | data->flags |= SF_IS_INF; |
| 3412 | } |
| 3413 | if (flags & SCF_DO_STCLASS) { |
| 3414 | cl_init(pRExC_state, &this_class); |
| 3415 | oclass = data->start_class; |
| 3416 | data->start_class = &this_class; |
| 3417 | f |= SCF_DO_STCLASS_AND; |
| 3418 | f &= ~SCF_DO_STCLASS_OR; |
| 3419 | } |
| 3420 | /* Exclude from super-linear cache processing any {n,m} |
| 3421 | regops for which the combination of input pos and regex |
| 3422 | pos is not enough information to determine if a match |
| 3423 | will be possible. |
| 3424 | |
| 3425 | For example, in the regex /foo(bar\s*){4,8}baz/ with the |
| 3426 | regex pos at the \s*, the prospects for a match depend not |
| 3427 | only on the input position but also on how many (bar\s*) |
| 3428 | repeats into the {4,8} we are. */ |
| 3429 | if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY)) |
| 3430 | f &= ~SCF_WHILEM_VISITED_POS; |
| 3431 | |
| 3432 | /* This will finish on WHILEM, setting scan, or on NULL: */ |
| 3433 | minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, |
| 3434 | last, data, stopparen, recursed, NULL, |
| 3435 | (mincount == 0 |
| 3436 | ? (f & ~SCF_DO_SUBSTR) : f),depth+1); |
| 3437 | |
| 3438 | if (flags & SCF_DO_STCLASS) |
| 3439 | data->start_class = oclass; |
| 3440 | if (mincount == 0 || minnext == 0) { |
| 3441 | if (flags & SCF_DO_STCLASS_OR) { |
| 3442 | cl_or(pRExC_state, data->start_class, &this_class); |
| 3443 | } |
| 3444 | else if (flags & SCF_DO_STCLASS_AND) { |
| 3445 | /* Switch to OR mode: cache the old value of |
| 3446 | * data->start_class */ |
| 3447 | INIT_AND_WITHP; |
| 3448 | StructCopy(data->start_class, and_withp, |
| 3449 | struct regnode_charclass_class); |
| 3450 | flags &= ~SCF_DO_STCLASS_AND; |
| 3451 | StructCopy(&this_class, data->start_class, |
| 3452 | struct regnode_charclass_class); |
| 3453 | flags |= SCF_DO_STCLASS_OR; |
| 3454 | data->start_class->flags |= ANYOF_EOS; |
| 3455 | } |
| 3456 | } else { /* Non-zero len */ |
| 3457 | if (flags & SCF_DO_STCLASS_OR) { |
| 3458 | cl_or(pRExC_state, data->start_class, &this_class); |
| 3459 | cl_and(data->start_class, and_withp); |
| 3460 | } |
| 3461 | else if (flags & SCF_DO_STCLASS_AND) |
| 3462 | cl_and(data->start_class, &this_class); |
| 3463 | flags &= ~SCF_DO_STCLASS; |
| 3464 | } |
| 3465 | if (!scan) /* It was not CURLYX, but CURLY. */ |
| 3466 | scan = next; |
| 3467 | if ( /* ? quantifier ok, except for (?{ ... }) */ |
| 3468 | (next_is_eval || !(mincount == 0 && maxcount == 1)) |
| 3469 | && (minnext == 0) && (deltanext == 0) |
| 3470 | && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 3471 | && maxcount <= REG_INFTY/3) /* Complement check for big count */ |
| 3472 | { |
| 3473 | ckWARNreg(RExC_parse, |
| 3474 | "Quantifier unexpected on zero-length expression"); |
| 3475 | } |
| 3476 | |
| 3477 | min += minnext * mincount; |
| 3478 | is_inf_internal |= ((maxcount == REG_INFTY |
| 3479 | && (minnext + deltanext) > 0) |
| 3480 | || deltanext == I32_MAX); |
| 3481 | is_inf |= is_inf_internal; |
| 3482 | delta += (minnext + deltanext) * maxcount - minnext * mincount; |
| 3483 | |
| 3484 | /* Try powerful optimization CURLYX => CURLYN. */ |
| 3485 | if ( OP(oscan) == CURLYX && data |
| 3486 | && data->flags & SF_IN_PAR |
| 3487 | && !(data->flags & SF_HAS_EVAL) |
| 3488 | && !deltanext && minnext == 1 ) { |
| 3489 | /* Try to optimize to CURLYN. */ |
| 3490 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; |
| 3491 | regnode * const nxt1 = nxt; |
| 3492 | #ifdef DEBUGGING |
| 3493 | regnode *nxt2; |
| 3494 | #endif |
| 3495 | |
| 3496 | /* Skip open. */ |
| 3497 | nxt = regnext(nxt); |
| 3498 | if (!REGNODE_SIMPLE(OP(nxt)) |
| 3499 | && !(PL_regkind[OP(nxt)] == EXACT |
| 3500 | && STR_LEN(nxt) == 1)) |
| 3501 | goto nogo; |
| 3502 | #ifdef DEBUGGING |
| 3503 | nxt2 = nxt; |
| 3504 | #endif |
| 3505 | nxt = regnext(nxt); |
| 3506 | if (OP(nxt) != CLOSE) |
| 3507 | goto nogo; |
| 3508 | if (RExC_open_parens) { |
| 3509 | RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/ |
| 3510 | RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/ |
| 3511 | } |
| 3512 | /* Now we know that nxt2 is the only contents: */ |
| 3513 | oscan->flags = (U8)ARG(nxt); |
| 3514 | OP(oscan) = CURLYN; |
| 3515 | OP(nxt1) = NOTHING; /* was OPEN. */ |
| 3516 | |
| 3517 | #ifdef DEBUGGING |
| 3518 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
| 3519 | NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */ |
| 3520 | NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */ |
| 3521 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
| 3522 | OP(nxt + 1) = OPTIMIZED; /* was count. */ |
| 3523 | NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */ |
| 3524 | #endif |
| 3525 | } |
| 3526 | nogo: |
| 3527 | |
| 3528 | /* Try optimization CURLYX => CURLYM. */ |
| 3529 | if ( OP(oscan) == CURLYX && data |
| 3530 | && !(data->flags & SF_HAS_PAR) |
| 3531 | && !(data->flags & SF_HAS_EVAL) |
| 3532 | && !deltanext /* atom is fixed width */ |
| 3533 | && minnext != 0 /* CURLYM can't handle zero width */ |
| 3534 | ) { |
| 3535 | /* XXXX How to optimize if data == 0? */ |
| 3536 | /* Optimize to a simpler form. */ |
| 3537 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */ |
| 3538 | regnode *nxt2; |
| 3539 | |
| 3540 | OP(oscan) = CURLYM; |
| 3541 | while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/ |
| 3542 | && (OP(nxt2) != WHILEM)) |
| 3543 | nxt = nxt2; |
| 3544 | OP(nxt2) = SUCCEED; /* Whas WHILEM */ |
| 3545 | /* Need to optimize away parenths. */ |
| 3546 | if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) { |
| 3547 | /* Set the parenth number. */ |
| 3548 | regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/ |
| 3549 | |
| 3550 | oscan->flags = (U8)ARG(nxt); |
| 3551 | if (RExC_open_parens) { |
| 3552 | RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/ |
| 3553 | RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/ |
| 3554 | } |
| 3555 | OP(nxt1) = OPTIMIZED; /* was OPEN. */ |
| 3556 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
| 3557 | |
| 3558 | #ifdef DEBUGGING |
| 3559 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
| 3560 | OP(nxt + 1) = OPTIMIZED; /* was count. */ |
| 3561 | NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */ |
| 3562 | NEXT_OFF(nxt + 1) = 0; /* just for consistency. */ |
| 3563 | #endif |
| 3564 | #if 0 |
| 3565 | while ( nxt1 && (OP(nxt1) != WHILEM)) { |
| 3566 | regnode *nnxt = regnext(nxt1); |
| 3567 | if (nnxt == nxt) { |
| 3568 | if (reg_off_by_arg[OP(nxt1)]) |
| 3569 | ARG_SET(nxt1, nxt2 - nxt1); |
| 3570 | else if (nxt2 - nxt1 < U16_MAX) |
| 3571 | NEXT_OFF(nxt1) = nxt2 - nxt1; |
| 3572 | else |
| 3573 | OP(nxt) = NOTHING; /* Cannot beautify */ |
| 3574 | } |
| 3575 | nxt1 = nnxt; |
| 3576 | } |
| 3577 | #endif |
| 3578 | /* Optimize again: */ |
| 3579 | study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt, |
| 3580 | NULL, stopparen, recursed, NULL, 0,depth+1); |
| 3581 | } |
| 3582 | else |
| 3583 | oscan->flags = 0; |
| 3584 | } |
| 3585 | else if ((OP(oscan) == CURLYX) |
| 3586 | && (flags & SCF_WHILEM_VISITED_POS) |
| 3587 | /* See the comment on a similar expression above. |
| 3588 | However, this time it's not a subexpression |
| 3589 | we care about, but the expression itself. */ |
| 3590 | && (maxcount == REG_INFTY) |
| 3591 | && data && ++data->whilem_c < 16) { |
| 3592 | /* This stays as CURLYX, we can put the count/of pair. */ |
| 3593 | /* Find WHILEM (as in regexec.c) */ |
| 3594 | regnode *nxt = oscan + NEXT_OFF(oscan); |
| 3595 | |
| 3596 | if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */ |
| 3597 | nxt += ARG(nxt); |
| 3598 | PREVOPER(nxt)->flags = (U8)(data->whilem_c |
| 3599 | | (RExC_whilem_seen << 4)); /* On WHILEM */ |
| 3600 | } |
| 3601 | if (data && fl & (SF_HAS_PAR|SF_IN_PAR)) |
| 3602 | pars++; |
| 3603 | if (flags & SCF_DO_SUBSTR) { |
| 3604 | SV *last_str = NULL; |
| 3605 | int counted = mincount != 0; |
| 3606 | |
| 3607 | if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */ |
| 3608 | #if defined(SPARC64_GCC_WORKAROUND) |
| 3609 | I32 b = 0; |
| 3610 | STRLEN l = 0; |
| 3611 | const char *s = NULL; |
| 3612 | I32 old = 0; |
| 3613 | |
| 3614 | if (pos_before >= data->last_start_min) |
| 3615 | b = pos_before; |
| 3616 | else |
| 3617 | b = data->last_start_min; |
| 3618 | |
| 3619 | l = 0; |
| 3620 | s = SvPV_const(data->last_found, l); |
| 3621 | old = b - data->last_start_min; |
| 3622 | |
| 3623 | #else |
| 3624 | I32 b = pos_before >= data->last_start_min |
| 3625 | ? pos_before : data->last_start_min; |
| 3626 | STRLEN l; |
| 3627 | const char * const s = SvPV_const(data->last_found, l); |
| 3628 | I32 old = b - data->last_start_min; |
| 3629 | #endif |
| 3630 | |
| 3631 | if (UTF) |
| 3632 | old = utf8_hop((U8*)s, old) - (U8*)s; |
| 3633 | l -= old; |
| 3634 | /* Get the added string: */ |
| 3635 | last_str = newSVpvn_utf8(s + old, l, UTF); |
| 3636 | if (deltanext == 0 && pos_before == b) { |
| 3637 | /* What was added is a constant string */ |
| 3638 | if (mincount > 1) { |
| 3639 | SvGROW(last_str, (mincount * l) + 1); |
| 3640 | repeatcpy(SvPVX(last_str) + l, |
| 3641 | SvPVX_const(last_str), l, mincount - 1); |
| 3642 | SvCUR_set(last_str, SvCUR(last_str) * mincount); |
| 3643 | /* Add additional parts. */ |
| 3644 | SvCUR_set(data->last_found, |
| 3645 | SvCUR(data->last_found) - l); |
| 3646 | sv_catsv(data->last_found, last_str); |
| 3647 | { |
| 3648 | SV * sv = data->last_found; |
| 3649 | MAGIC *mg = |
| 3650 | SvUTF8(sv) && SvMAGICAL(sv) ? |
| 3651 | mg_find(sv, PERL_MAGIC_utf8) : NULL; |
| 3652 | if (mg && mg->mg_len >= 0) |
| 3653 | mg->mg_len += CHR_SVLEN(last_str) - l; |
| 3654 | } |
| 3655 | data->last_end += l * (mincount - 1); |
| 3656 | } |
| 3657 | } else { |
| 3658 | /* start offset must point into the last copy */ |
| 3659 | data->last_start_min += minnext * (mincount - 1); |
| 3660 | data->last_start_max += is_inf ? I32_MAX |
| 3661 | : (maxcount - 1) * (minnext + data->pos_delta); |
| 3662 | } |
| 3663 | } |
| 3664 | /* It is counted once already... */ |
| 3665 | data->pos_min += minnext * (mincount - counted); |
| 3666 | data->pos_delta += - counted * deltanext + |
| 3667 | (minnext + deltanext) * maxcount - minnext * mincount; |
| 3668 | if (mincount != maxcount) { |
| 3669 | /* Cannot extend fixed substrings found inside |
| 3670 | the group. */ |
| 3671 | SCAN_COMMIT(pRExC_state,data,minlenp); |
| 3672 | if (mincount && last_str) { |
| 3673 | SV * const sv = data->last_found; |
| 3674 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? |
| 3675 | mg_find(sv, PERL_MAGIC_utf8) : NULL; |
| 3676 | |
| 3677 | if (mg) |
| 3678 | mg->mg_len = -1; |
| 3679 | sv_setsv(sv, last_str); |
| 3680 | data->last_end = data->pos_min; |
| 3681 | data->last_start_min = |
| 3682 | data->pos_min - CHR_SVLEN(last_str); |
| 3683 | data->last_start_max = is_inf |
| 3684 | ? I32_MAX |
| 3685 | : data->pos_min + data->pos_delta |
| 3686 | - CHR_SVLEN(last_str); |
| 3687 | } |
| 3688 | data->longest = &(data->longest_float); |
| 3689 | } |
| 3690 | SvREFCNT_dec(last_str); |
| 3691 | } |
| 3692 | if (data && (fl & SF_HAS_EVAL)) |
| 3693 | data->flags |= SF_HAS_EVAL; |
| 3694 | optimize_curly_tail: |
| 3695 | if (OP(oscan) != CURLYX) { |
| 3696 | while (PL_regkind[OP(next = regnext(oscan))] == NOTHING |
| 3697 | && NEXT_OFF(next)) |
| 3698 | NEXT_OFF(oscan) += NEXT_OFF(next); |
| 3699 | } |
| 3700 | continue; |
| 3701 | default: /* REF, ANYOFV, and CLUMP only? */ |
| 3702 | if (flags & SCF_DO_SUBSTR) { |
| 3703 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
| 3704 | data->longest = &(data->longest_float); |
| 3705 | } |
| 3706 | is_inf = is_inf_internal = 1; |
| 3707 | if (flags & SCF_DO_STCLASS_OR) |
| 3708 | cl_anything(pRExC_state, data->start_class); |
| 3709 | flags &= ~SCF_DO_STCLASS; |
| 3710 | break; |
| 3711 | } |
| 3712 | } |
| 3713 | else if (OP(scan) == LNBREAK) { |
| 3714 | if (flags & SCF_DO_STCLASS) { |
| 3715 | int value = 0; |
| 3716 | data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */ |
| 3717 | if (flags & SCF_DO_STCLASS_AND) { |
| 3718 | for (value = 0; value < 256; value++) |
| 3719 | if (!is_VERTWS_cp(value)) |
| 3720 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3721 | } |
| 3722 | else { |
| 3723 | for (value = 0; value < 256; value++) |
| 3724 | if (is_VERTWS_cp(value)) |
| 3725 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3726 | } |
| 3727 | if (flags & SCF_DO_STCLASS_OR) |
| 3728 | cl_and(data->start_class, and_withp); |
| 3729 | flags &= ~SCF_DO_STCLASS; |
| 3730 | } |
| 3731 | min += 1; |
| 3732 | delta += 1; |
| 3733 | if (flags & SCF_DO_SUBSTR) { |
| 3734 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
| 3735 | data->pos_min += 1; |
| 3736 | data->pos_delta += 1; |
| 3737 | data->longest = &(data->longest_float); |
| 3738 | } |
| 3739 | } |
| 3740 | else if (OP(scan) == FOLDCHAR) { |
| 3741 | int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2; |
| 3742 | flags &= ~SCF_DO_STCLASS; |
| 3743 | min += 1; |
| 3744 | delta += d; |
| 3745 | if (flags & SCF_DO_SUBSTR) { |
| 3746 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
| 3747 | data->pos_min += 1; |
| 3748 | data->pos_delta += d; |
| 3749 | data->longest = &(data->longest_float); |
| 3750 | } |
| 3751 | } |
| 3752 | else if (REGNODE_SIMPLE(OP(scan))) { |
| 3753 | int value = 0; |
| 3754 | |
| 3755 | if (flags & SCF_DO_SUBSTR) { |
| 3756 | SCAN_COMMIT(pRExC_state,data,minlenp); |
| 3757 | data->pos_min++; |
| 3758 | } |
| 3759 | min++; |
| 3760 | if (flags & SCF_DO_STCLASS) { |
| 3761 | data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */ |
| 3762 | |
| 3763 | /* Some of the logic below assumes that switching |
| 3764 | locale on will only add false positives. */ |
| 3765 | switch (PL_regkind[OP(scan)]) { |
| 3766 | case SANY: |
| 3767 | default: |
| 3768 | do_default: |
| 3769 | /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */ |
| 3770 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
| 3771 | cl_anything(pRExC_state, data->start_class); |
| 3772 | break; |
| 3773 | case REG_ANY: |
| 3774 | if (OP(scan) == SANY) |
| 3775 | goto do_default; |
| 3776 | if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */ |
| 3777 | value = (ANYOF_BITMAP_TEST(data->start_class,'\n') |
| 3778 | || ANYOF_CLASS_TEST_ANY_SET(data->start_class)); |
| 3779 | cl_anything(pRExC_state, data->start_class); |
| 3780 | } |
| 3781 | if (flags & SCF_DO_STCLASS_AND || !value) |
| 3782 | ANYOF_BITMAP_CLEAR(data->start_class,'\n'); |
| 3783 | break; |
| 3784 | case ANYOF: |
| 3785 | if (flags & SCF_DO_STCLASS_AND) |
| 3786 | cl_and(data->start_class, |
| 3787 | (struct regnode_charclass_class*)scan); |
| 3788 | else |
| 3789 | cl_or(pRExC_state, data->start_class, |
| 3790 | (struct regnode_charclass_class*)scan); |
| 3791 | break; |
| 3792 | case ALNUM: |
| 3793 | if (flags & SCF_DO_STCLASS_AND) { |
| 3794 | if (!(data->start_class->flags & ANYOF_LOCALE)) { |
| 3795 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM); |
| 3796 | if (OP(scan) == ALNUMU) { |
| 3797 | for (value = 0; value < 256; value++) { |
| 3798 | if (!isWORDCHAR_L1(value)) { |
| 3799 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3800 | } |
| 3801 | } |
| 3802 | } else { |
| 3803 | for (value = 0; value < 256; value++) { |
| 3804 | if (!isALNUM(value)) { |
| 3805 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3806 | } |
| 3807 | } |
| 3808 | } |
| 3809 | } |
| 3810 | } |
| 3811 | else { |
| 3812 | if (data->start_class->flags & ANYOF_LOCALE) |
| 3813 | ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM); |
| 3814 | |
| 3815 | /* Even if under locale, set the bits for non-locale |
| 3816 | * in case it isn't a true locale-node. This will |
| 3817 | * create false positives if it truly is locale */ |
| 3818 | if (OP(scan) == ALNUMU) { |
| 3819 | for (value = 0; value < 256; value++) { |
| 3820 | if (isWORDCHAR_L1(value)) { |
| 3821 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3822 | } |
| 3823 | } |
| 3824 | } else { |
| 3825 | for (value = 0; value < 256; value++) { |
| 3826 | if (isALNUM(value)) { |
| 3827 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3828 | } |
| 3829 | } |
| 3830 | } |
| 3831 | } |
| 3832 | break; |
| 3833 | case NALNUM: |
| 3834 | if (flags & SCF_DO_STCLASS_AND) { |
| 3835 | if (!(data->start_class->flags & ANYOF_LOCALE)) { |
| 3836 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM); |
| 3837 | if (OP(scan) == NALNUMU) { |
| 3838 | for (value = 0; value < 256; value++) { |
| 3839 | if (isWORDCHAR_L1(value)) { |
| 3840 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3841 | } |
| 3842 | } |
| 3843 | } else { |
| 3844 | for (value = 0; value < 256; value++) { |
| 3845 | if (isALNUM(value)) { |
| 3846 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3847 | } |
| 3848 | } |
| 3849 | } |
| 3850 | } |
| 3851 | } |
| 3852 | else { |
| 3853 | if (data->start_class->flags & ANYOF_LOCALE) |
| 3854 | ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM); |
| 3855 | |
| 3856 | /* Even if under locale, set the bits for non-locale in |
| 3857 | * case it isn't a true locale-node. This will create |
| 3858 | * false positives if it truly is locale */ |
| 3859 | if (OP(scan) == NALNUMU) { |
| 3860 | for (value = 0; value < 256; value++) { |
| 3861 | if (! isWORDCHAR_L1(value)) { |
| 3862 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3863 | } |
| 3864 | } |
| 3865 | } else { |
| 3866 | for (value = 0; value < 256; value++) { |
| 3867 | if (! isALNUM(value)) { |
| 3868 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3869 | } |
| 3870 | } |
| 3871 | } |
| 3872 | } |
| 3873 | break; |
| 3874 | case SPACE: |
| 3875 | if (flags & SCF_DO_STCLASS_AND) { |
| 3876 | if (!(data->start_class->flags & ANYOF_LOCALE)) { |
| 3877 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE); |
| 3878 | if (OP(scan) == SPACEU) { |
| 3879 | for (value = 0; value < 256; value++) { |
| 3880 | if (!isSPACE_L1(value)) { |
| 3881 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3882 | } |
| 3883 | } |
| 3884 | } else { |
| 3885 | for (value = 0; value < 256; value++) { |
| 3886 | if (!isSPACE(value)) { |
| 3887 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3888 | } |
| 3889 | } |
| 3890 | } |
| 3891 | } |
| 3892 | } |
| 3893 | else { |
| 3894 | if (data->start_class->flags & ANYOF_LOCALE) { |
| 3895 | ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE); |
| 3896 | } |
| 3897 | if (OP(scan) == SPACEU) { |
| 3898 | for (value = 0; value < 256; value++) { |
| 3899 | if (isSPACE_L1(value)) { |
| 3900 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3901 | } |
| 3902 | } |
| 3903 | } else { |
| 3904 | for (value = 0; value < 256; value++) { |
| 3905 | if (isSPACE(value)) { |
| 3906 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3907 | } |
| 3908 | } |
| 3909 | } |
| 3910 | } |
| 3911 | break; |
| 3912 | case NSPACE: |
| 3913 | if (flags & SCF_DO_STCLASS_AND) { |
| 3914 | if (!(data->start_class->flags & ANYOF_LOCALE)) { |
| 3915 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE); |
| 3916 | if (OP(scan) == NSPACEU) { |
| 3917 | for (value = 0; value < 256; value++) { |
| 3918 | if (isSPACE_L1(value)) { |
| 3919 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3920 | } |
| 3921 | } |
| 3922 | } else { |
| 3923 | for (value = 0; value < 256; value++) { |
| 3924 | if (isSPACE(value)) { |
| 3925 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3926 | } |
| 3927 | } |
| 3928 | } |
| 3929 | } |
| 3930 | } |
| 3931 | else { |
| 3932 | if (data->start_class->flags & ANYOF_LOCALE) |
| 3933 | ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE); |
| 3934 | if (OP(scan) == NSPACEU) { |
| 3935 | for (value = 0; value < 256; value++) { |
| 3936 | if (!isSPACE_L1(value)) { |
| 3937 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3938 | } |
| 3939 | } |
| 3940 | } |
| 3941 | else { |
| 3942 | for (value = 0; value < 256; value++) { |
| 3943 | if (!isSPACE(value)) { |
| 3944 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3945 | } |
| 3946 | } |
| 3947 | } |
| 3948 | } |
| 3949 | break; |
| 3950 | case DIGIT: |
| 3951 | if (flags & SCF_DO_STCLASS_AND) { |
| 3952 | if (!(data->start_class->flags & ANYOF_LOCALE)) { |
| 3953 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT); |
| 3954 | for (value = 0; value < 256; value++) |
| 3955 | if (!isDIGIT(value)) |
| 3956 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3957 | } |
| 3958 | } |
| 3959 | else { |
| 3960 | if (data->start_class->flags & ANYOF_LOCALE) |
| 3961 | ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT); |
| 3962 | for (value = 0; value < 256; value++) |
| 3963 | if (isDIGIT(value)) |
| 3964 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3965 | } |
| 3966 | break; |
| 3967 | case NDIGIT: |
| 3968 | if (flags & SCF_DO_STCLASS_AND) { |
| 3969 | if (!(data->start_class->flags & ANYOF_LOCALE)) |
| 3970 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT); |
| 3971 | for (value = 0; value < 256; value++) |
| 3972 | if (isDIGIT(value)) |
| 3973 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
| 3974 | } |
| 3975 | else { |
| 3976 | if (data->start_class->flags & ANYOF_LOCALE) |
| 3977 | ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT); |
| 3978 | for (value = 0; value < 256; value++) |
| 3979 | if (!isDIGIT(value)) |
| 3980 | ANYOF_BITMAP_SET(data->start_class, value); |
| 3981 | } |
| 3982 | break; |
| 3983 | CASE_SYNST_FNC(VERTWS); |
| 3984 | CASE_SYNST_FNC(HORIZWS); |
| 3985 | |
| 3986 | } |
| 3987 | if (flags & SCF_DO_STCLASS_OR) |
| 3988 | cl_and(data->start_class, and_withp); |
| 3989 | flags &= ~SCF_DO_STCLASS; |
| 3990 | } |
| 3991 | } |
| 3992 | else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) { |
| 3993 | data->flags |= (OP(scan) == MEOL |
| 3994 | ? SF_BEFORE_MEOL |
| 3995 | : SF_BEFORE_SEOL); |
| 3996 | } |
| 3997 | else if ( PL_regkind[OP(scan)] == BRANCHJ |
| 3998 | /* Lookbehind, or need to calculate parens/evals/stclass: */ |
| 3999 | && (scan->flags || data || (flags & SCF_DO_STCLASS)) |
| 4000 | && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) { |
| 4001 | if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY |
| 4002 | || OP(scan) == UNLESSM ) |
| 4003 | { |
| 4004 | /* Negative Lookahead/lookbehind |
| 4005 | In this case we can't do fixed string optimisation. |
| 4006 | */ |
| 4007 | |
| 4008 | I32 deltanext, minnext, fake = 0; |
| 4009 | regnode *nscan; |
| 4010 | struct regnode_charclass_class intrnl; |
| 4011 | int f = 0; |
| 4012 | |
| 4013 | data_fake.flags = 0; |
| 4014 | if (data) { |
| 4015 | data_fake.whilem_c = data->whilem_c; |
| 4016 | data_fake.last_closep = data->last_closep; |
| 4017 | } |
| 4018 | else |
| 4019 | data_fake.last_closep = &fake; |
| 4020 | data_fake.pos_delta = delta; |
| 4021 | if ( flags & SCF_DO_STCLASS && !scan->flags |
| 4022 | && OP(scan) == IFMATCH ) { /* Lookahead */ |
| 4023 | cl_init(pRExC_state, &intrnl); |
| 4024 | data_fake.start_class = &intrnl; |
| 4025 | f |= SCF_DO_STCLASS_AND; |
| 4026 | } |
| 4027 | if (flags & SCF_WHILEM_VISITED_POS) |
| 4028 | f |= SCF_WHILEM_VISITED_POS; |
| 4029 | next = regnext(scan); |
| 4030 | nscan = NEXTOPER(NEXTOPER(scan)); |
| 4031 | minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, |
| 4032 | last, &data_fake, stopparen, recursed, NULL, f, depth+1); |
| 4033 | if (scan->flags) { |
| 4034 | if (deltanext) { |
| 4035 | FAIL("Variable length lookbehind not implemented"); |
| 4036 | } |
| 4037 | else if (minnext > (I32)U8_MAX) { |
| 4038 | FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX); |
| 4039 | } |
| 4040 | scan->flags = (U8)minnext; |
| 4041 | } |
| 4042 | if (data) { |
| 4043 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 4044 | pars++; |
| 4045 | if (data_fake.flags & SF_HAS_EVAL) |
| 4046 | data->flags |= SF_HAS_EVAL; |
| 4047 | data->whilem_c = data_fake.whilem_c; |
| 4048 | } |
| 4049 | if (f & SCF_DO_STCLASS_AND) { |
| 4050 | if (flags & SCF_DO_STCLASS_OR) { |
| 4051 | /* OR before, AND after: ideally we would recurse with |
| 4052 | * data_fake to get the AND applied by study of the |
| 4053 | * remainder of the pattern, and then derecurse; |
| 4054 | * *** HACK *** for now just treat as "no information". |
| 4055 | * See [perl #56690]. |
| 4056 | */ |
| 4057 | cl_init(pRExC_state, data->start_class); |
| 4058 | } else { |
| 4059 | /* AND before and after: combine and continue */ |
| 4060 | const int was = (data->start_class->flags & ANYOF_EOS); |
| 4061 | |
| 4062 | cl_and(data->start_class, &intrnl); |
| 4063 | if (was) |
| 4064 | data->start_class->flags |= ANYOF_EOS; |
| 4065 | } |
| 4066 | } |
| 4067 | } |
| 4068 | #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY |
| 4069 | else { |
| 4070 | /* Positive Lookahead/lookbehind |
| 4071 | In this case we can do fixed string optimisation, |
| 4072 | but we must be careful about it. Note in the case of |
| 4073 | lookbehind the positions will be offset by the minimum |
| 4074 | length of the pattern, something we won't know about |
| 4075 | until after the recurse. |
| 4076 | */ |
| 4077 | I32 deltanext, fake = 0; |
| 4078 | regnode *nscan; |
| 4079 | struct regnode_charclass_class intrnl; |
| 4080 | int f = 0; |
| 4081 | /* We use SAVEFREEPV so that when the full compile |
| 4082 | is finished perl will clean up the allocated |
| 4083 | minlens when it's all done. This way we don't |
| 4084 | have to worry about freeing them when we know |
| 4085 | they wont be used, which would be a pain. |
| 4086 | */ |
| 4087 | I32 *minnextp; |
| 4088 | Newx( minnextp, 1, I32 ); |
| 4089 | SAVEFREEPV(minnextp); |
| 4090 | |
| 4091 | if (data) { |
| 4092 | StructCopy(data, &data_fake, scan_data_t); |
| 4093 | if ((flags & SCF_DO_SUBSTR) && data->last_found) { |
| 4094 | f |= SCF_DO_SUBSTR; |
| 4095 | if (scan->flags) |
| 4096 | SCAN_COMMIT(pRExC_state, &data_fake,minlenp); |
| 4097 | data_fake.last_found=newSVsv(data->last_found); |
| 4098 | } |
| 4099 | } |
| 4100 | else |
| 4101 | data_fake.last_closep = &fake; |
| 4102 | data_fake.flags = 0; |
| 4103 | data_fake.pos_delta = delta; |
| 4104 | if (is_inf) |
| 4105 | data_fake.flags |= SF_IS_INF; |
| 4106 | if ( flags & SCF_DO_STCLASS && !scan->flags |
| 4107 | && OP(scan) == IFMATCH ) { /* Lookahead */ |
| 4108 | cl_init(pRExC_state, &intrnl); |
| 4109 | data_fake.start_class = &intrnl; |
| 4110 | f |= SCF_DO_STCLASS_AND; |
| 4111 | } |
| 4112 | if (flags & SCF_WHILEM_VISITED_POS) |
| 4113 | f |= SCF_WHILEM_VISITED_POS; |
| 4114 | next = regnext(scan); |
| 4115 | nscan = NEXTOPER(NEXTOPER(scan)); |
| 4116 | |
| 4117 | *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext, |
| 4118 | last, &data_fake, stopparen, recursed, NULL, f,depth+1); |
| 4119 | if (scan->flags) { |
| 4120 | if (deltanext) { |
| 4121 | FAIL("Variable length lookbehind not implemented"); |
| 4122 | } |
| 4123 | else if (*minnextp > (I32)U8_MAX) { |
| 4124 | FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX); |
| 4125 | } |
| 4126 | scan->flags = (U8)*minnextp; |
| 4127 | } |
| 4128 | |
| 4129 | *minnextp += min; |
| 4130 | |
| 4131 | if (f & SCF_DO_STCLASS_AND) { |
| 4132 | const int was = (data->start_class->flags & ANYOF_EOS); |
| 4133 | |
| 4134 | cl_and(data->start_class, &intrnl); |
| 4135 | if (was) |
| 4136 | data->start_class->flags |= ANYOF_EOS; |
| 4137 | } |
| 4138 | if (data) { |
| 4139 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 4140 | pars++; |
| 4141 | if (data_fake.flags & SF_HAS_EVAL) |
| 4142 | data->flags |= SF_HAS_EVAL; |
| 4143 | data->whilem_c = data_fake.whilem_c; |
| 4144 | if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) { |
| 4145 | if (RExC_rx->minlen<*minnextp) |
| 4146 | RExC_rx->minlen=*minnextp; |
| 4147 | SCAN_COMMIT(pRExC_state, &data_fake, minnextp); |
| 4148 | SvREFCNT_dec(data_fake.last_found); |
| 4149 | |
| 4150 | if ( data_fake.minlen_fixed != minlenp ) |
| 4151 | { |
| 4152 | data->offset_fixed= data_fake.offset_fixed; |
| 4153 | data->minlen_fixed= data_fake.minlen_fixed; |
| 4154 | data->lookbehind_fixed+= scan->flags; |
| 4155 | } |
| 4156 | if ( data_fake.minlen_float != minlenp ) |
| 4157 | { |
| 4158 | data->minlen_float= data_fake.minlen_float; |
| 4159 | data->offset_float_min=data_fake.offset_float_min; |
| 4160 | data->offset_float_max=data_fake.offset_float_max; |
| 4161 | data->lookbehind_float+= scan->flags; |
| 4162 | } |
| 4163 | } |
| 4164 | } |
| 4165 | |
| 4166 | |
| 4167 | } |
| 4168 | #endif |
| 4169 | } |
| 4170 | else if (OP(scan) == OPEN) { |
| 4171 | if (stopparen != (I32)ARG(scan)) |
| 4172 | pars++; |
| 4173 | } |
| 4174 | else if (OP(scan) == CLOSE) { |
| 4175 | if (stopparen == (I32)ARG(scan)) { |
| 4176 | break; |
| 4177 | } |
| 4178 | if ((I32)ARG(scan) == is_par) { |
| 4179 | next = regnext(scan); |
| 4180 | |
| 4181 | if ( next && (OP(next) != WHILEM) && next < last) |
| 4182 | is_par = 0; /* Disable optimization */ |
| 4183 | } |
| 4184 | if (data) |
| 4185 | *(data->last_closep) = ARG(scan); |
| 4186 | } |
| 4187 | else if (OP(scan) == EVAL) { |
| 4188 | if (data) |
| 4189 | data->flags |= SF_HAS_EVAL; |
| 4190 | } |
| 4191 | else if ( PL_regkind[OP(scan)] == ENDLIKE ) { |
| 4192 | if (flags & SCF_DO_SUBSTR) { |
| 4193 | SCAN_COMMIT(pRExC_state,data,minlenp); |
| 4194 | flags &= ~SCF_DO_SUBSTR; |
| 4195 | } |
| 4196 | if (data && OP(scan)==ACCEPT) { |
| 4197 | data->flags |= SCF_SEEN_ACCEPT; |
| 4198 | if (stopmin > min) |
| 4199 | stopmin = min; |
| 4200 | } |
| 4201 | } |
| 4202 | else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */ |
| 4203 | { |
| 4204 | if (flags & SCF_DO_SUBSTR) { |
| 4205 | SCAN_COMMIT(pRExC_state,data,minlenp); |
| 4206 | data->longest = &(data->longest_float); |
| 4207 | } |
| 4208 | is_inf = is_inf_internal = 1; |
| 4209 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
| 4210 | cl_anything(pRExC_state, data->start_class); |
| 4211 | flags &= ~SCF_DO_STCLASS; |
| 4212 | } |
| 4213 | else if (OP(scan) == GPOS) { |
| 4214 | if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) && |
| 4215 | !(delta || is_inf || (data && data->pos_delta))) |
| 4216 | { |
| 4217 | if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR)) |
| 4218 | RExC_rx->extflags |= RXf_ANCH_GPOS; |
| 4219 | if (RExC_rx->gofs < (U32)min) |
| 4220 | RExC_rx->gofs = min; |
| 4221 | } else { |
| 4222 | RExC_rx->extflags |= RXf_GPOS_FLOAT; |
| 4223 | RExC_rx->gofs = 0; |
| 4224 | } |
| 4225 | } |
| 4226 | #ifdef TRIE_STUDY_OPT |
| 4227 | #ifdef FULL_TRIE_STUDY |
| 4228 | else if (PL_regkind[OP(scan)] == TRIE) { |
| 4229 | /* NOTE - There is similar code to this block above for handling |
| 4230 | BRANCH nodes on the initial study. If you change stuff here |
| 4231 | check there too. */ |
| 4232 | regnode *trie_node= scan; |
| 4233 | regnode *tail= regnext(scan); |
| 4234 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
| 4235 | I32 max1 = 0, min1 = I32_MAX; |
| 4236 | struct regnode_charclass_class accum; |
| 4237 | |
| 4238 | if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */ |
| 4239 | SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */ |
| 4240 | if (flags & SCF_DO_STCLASS) |
| 4241 | cl_init_zero(pRExC_state, &accum); |
| 4242 | |
| 4243 | if (!trie->jump) { |
| 4244 | min1= trie->minlen; |
| 4245 | max1= trie->maxlen; |
| 4246 | } else { |
| 4247 | const regnode *nextbranch= NULL; |
| 4248 | U32 word; |
| 4249 | |
| 4250 | for ( word=1 ; word <= trie->wordcount ; word++) |
| 4251 | { |
| 4252 | I32 deltanext=0, minnext=0, f = 0, fake; |
| 4253 | struct regnode_charclass_class this_class; |
| 4254 | |
| 4255 | data_fake.flags = 0; |
| 4256 | if (data) { |
| 4257 | data_fake.whilem_c = data->whilem_c; |
| 4258 | data_fake.last_closep = data->last_closep; |
| 4259 | } |
| 4260 | else |
| 4261 | data_fake.last_closep = &fake; |
| 4262 | data_fake.pos_delta = delta; |
| 4263 | if (flags & SCF_DO_STCLASS) { |
| 4264 | cl_init(pRExC_state, &this_class); |
| 4265 | data_fake.start_class = &this_class; |
| 4266 | f = SCF_DO_STCLASS_AND; |
| 4267 | } |
| 4268 | if (flags & SCF_WHILEM_VISITED_POS) |
| 4269 | f |= SCF_WHILEM_VISITED_POS; |
| 4270 | |
| 4271 | if (trie->jump[word]) { |
| 4272 | if (!nextbranch) |
| 4273 | nextbranch = trie_node + trie->jump[0]; |
| 4274 | scan= trie_node + trie->jump[word]; |
| 4275 | /* We go from the jump point to the branch that follows |
| 4276 | it. Note this means we need the vestigal unused branches |
| 4277 | even though they arent otherwise used. |
| 4278 | */ |
| 4279 | minnext = study_chunk(pRExC_state, &scan, minlenp, |
| 4280 | &deltanext, (regnode *)nextbranch, &data_fake, |
| 4281 | stopparen, recursed, NULL, f,depth+1); |
| 4282 | } |
| 4283 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) |
| 4284 | nextbranch= regnext((regnode*)nextbranch); |
| 4285 | |
| 4286 | if (min1 > (I32)(minnext + trie->minlen)) |
| 4287 | min1 = minnext + trie->minlen; |
| 4288 | if (max1 < (I32)(minnext + deltanext + trie->maxlen)) |
| 4289 | max1 = minnext + deltanext + trie->maxlen; |
| 4290 | if (deltanext == I32_MAX) |
| 4291 | is_inf = is_inf_internal = 1; |
| 4292 | |
| 4293 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 4294 | pars++; |
| 4295 | if (data_fake.flags & SCF_SEEN_ACCEPT) { |
| 4296 | if ( stopmin > min + min1) |
| 4297 | stopmin = min + min1; |
| 4298 | flags &= ~SCF_DO_SUBSTR; |
| 4299 | if (data) |
| 4300 | data->flags |= SCF_SEEN_ACCEPT; |
| 4301 | } |
| 4302 | if (data) { |
| 4303 | if (data_fake.flags & SF_HAS_EVAL) |
| 4304 | data->flags |= SF_HAS_EVAL; |
| 4305 | data->whilem_c = data_fake.whilem_c; |
| 4306 | } |
| 4307 | if (flags & SCF_DO_STCLASS) |
| 4308 | cl_or(pRExC_state, &accum, &this_class); |
| 4309 | } |
| 4310 | } |
| 4311 | if (flags & SCF_DO_SUBSTR) { |
| 4312 | data->pos_min += min1; |
| 4313 | data->pos_delta += max1 - min1; |
| 4314 | if (max1 != min1 || is_inf) |
| 4315 | data->longest = &(data->longest_float); |
| 4316 | } |
| 4317 | min += min1; |
| 4318 | delta += max1 - min1; |
| 4319 | if (flags & SCF_DO_STCLASS_OR) { |
| 4320 | cl_or(pRExC_state, data->start_class, &accum); |
| 4321 | if (min1) { |
| 4322 | cl_and(data->start_class, and_withp); |
| 4323 | flags &= ~SCF_DO_STCLASS; |
| 4324 | } |
| 4325 | } |
| 4326 | else if (flags & SCF_DO_STCLASS_AND) { |
| 4327 | if (min1) { |
| 4328 | cl_and(data->start_class, &accum); |
| 4329 | flags &= ~SCF_DO_STCLASS; |
| 4330 | } |
| 4331 | else { |
| 4332 | /* Switch to OR mode: cache the old value of |
| 4333 | * data->start_class */ |
| 4334 | INIT_AND_WITHP; |
| 4335 | StructCopy(data->start_class, and_withp, |
| 4336 | struct regnode_charclass_class); |
| 4337 | flags &= ~SCF_DO_STCLASS_AND; |
| 4338 | StructCopy(&accum, data->start_class, |
| 4339 | struct regnode_charclass_class); |
| 4340 | flags |= SCF_DO_STCLASS_OR; |
| 4341 | data->start_class->flags |= ANYOF_EOS; |
| 4342 | } |
| 4343 | } |
| 4344 | scan= tail; |
| 4345 | continue; |
| 4346 | } |
| 4347 | #else |
| 4348 | else if (PL_regkind[OP(scan)] == TRIE) { |
| 4349 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
| 4350 | U8*bang=NULL; |
| 4351 | |
| 4352 | min += trie->minlen; |
| 4353 | delta += (trie->maxlen - trie->minlen); |
| 4354 | flags &= ~SCF_DO_STCLASS; /* xxx */ |
| 4355 | if (flags & SCF_DO_SUBSTR) { |
| 4356 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
| 4357 | data->pos_min += trie->minlen; |
| 4358 | data->pos_delta += (trie->maxlen - trie->minlen); |
| 4359 | if (trie->maxlen != trie->minlen) |
| 4360 | data->longest = &(data->longest_float); |
| 4361 | } |
| 4362 | if (trie->jump) /* no more substrings -- for now /grr*/ |
| 4363 | flags &= ~SCF_DO_SUBSTR; |
| 4364 | } |
| 4365 | #endif /* old or new */ |
| 4366 | #endif /* TRIE_STUDY_OPT */ |
| 4367 | |
| 4368 | /* Else: zero-length, ignore. */ |
| 4369 | scan = regnext(scan); |
| 4370 | } |
| 4371 | if (frame) { |
| 4372 | last = frame->last; |
| 4373 | scan = frame->next; |
| 4374 | stopparen = frame->stop; |
| 4375 | frame = frame->prev; |
| 4376 | goto fake_study_recurse; |
| 4377 | } |
| 4378 | |
| 4379 | finish: |
| 4380 | assert(!frame); |
| 4381 | DEBUG_STUDYDATA("pre-fin:",data,depth); |
| 4382 | |
| 4383 | *scanp = scan; |
| 4384 | *deltap = is_inf_internal ? I32_MAX : delta; |
| 4385 | if (flags & SCF_DO_SUBSTR && is_inf) |
| 4386 | data->pos_delta = I32_MAX - data->pos_min; |
| 4387 | if (is_par > (I32)U8_MAX) |
| 4388 | is_par = 0; |
| 4389 | if (is_par && pars==1 && data) { |
| 4390 | data->flags |= SF_IN_PAR; |
| 4391 | data->flags &= ~SF_HAS_PAR; |
| 4392 | } |
| 4393 | else if (pars && data) { |
| 4394 | data->flags |= SF_HAS_PAR; |
| 4395 | data->flags &= ~SF_IN_PAR; |
| 4396 | } |
| 4397 | if (flags & SCF_DO_STCLASS_OR) |
| 4398 | cl_and(data->start_class, and_withp); |
| 4399 | if (flags & SCF_TRIE_RESTUDY) |
| 4400 | data->flags |= SCF_TRIE_RESTUDY; |
| 4401 | |
| 4402 | DEBUG_STUDYDATA("post-fin:",data,depth); |
| 4403 | |
| 4404 | return min < stopmin ? min : stopmin; |
| 4405 | } |
| 4406 | |
| 4407 | STATIC U32 |
| 4408 | S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s) |
| 4409 | { |
| 4410 | U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0; |
| 4411 | |
| 4412 | PERL_ARGS_ASSERT_ADD_DATA; |
| 4413 | |
| 4414 | Renewc(RExC_rxi->data, |
| 4415 | sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1), |
| 4416 | char, struct reg_data); |
| 4417 | if(count) |
| 4418 | Renew(RExC_rxi->data->what, count + n, U8); |
| 4419 | else |
| 4420 | Newx(RExC_rxi->data->what, n, U8); |
| 4421 | RExC_rxi->data->count = count + n; |
| 4422 | Copy(s, RExC_rxi->data->what + count, n, U8); |
| 4423 | return count; |
| 4424 | } |
| 4425 | |
| 4426 | /*XXX: todo make this not included in a non debugging perl */ |
| 4427 | #ifndef PERL_IN_XSUB_RE |
| 4428 | void |
| 4429 | Perl_reginitcolors(pTHX) |
| 4430 | { |
| 4431 | dVAR; |
| 4432 | const char * const s = PerlEnv_getenv("PERL_RE_COLORS"); |
| 4433 | if (s) { |
| 4434 | char *t = savepv(s); |
| 4435 | int i = 0; |
| 4436 | PL_colors[0] = t; |
| 4437 | while (++i < 6) { |
| 4438 | t = strchr(t, '\t'); |
| 4439 | if (t) { |
| 4440 | *t = '\0'; |
| 4441 | PL_colors[i] = ++t; |
| 4442 | } |
| 4443 | else |
| 4444 | PL_colors[i] = t = (char *)""; |
| 4445 | } |
| 4446 | } else { |
| 4447 | int i = 0; |
| 4448 | while (i < 6) |
| 4449 | PL_colors[i++] = (char *)""; |
| 4450 | } |
| 4451 | PL_colorset = 1; |
| 4452 | } |
| 4453 | #endif |
| 4454 | |
| 4455 | |
| 4456 | #ifdef TRIE_STUDY_OPT |
| 4457 | #define CHECK_RESTUDY_GOTO \ |
| 4458 | if ( \ |
| 4459 | (data.flags & SCF_TRIE_RESTUDY) \ |
| 4460 | && ! restudied++ \ |
| 4461 | ) goto reStudy |
| 4462 | #else |
| 4463 | #define CHECK_RESTUDY_GOTO |
| 4464 | #endif |
| 4465 | |
| 4466 | /* |
| 4467 | - pregcomp - compile a regular expression into internal code |
| 4468 | * |
| 4469 | * We can't allocate space until we know how big the compiled form will be, |
| 4470 | * but we can't compile it (and thus know how big it is) until we've got a |
| 4471 | * place to put the code. So we cheat: we compile it twice, once with code |
| 4472 | * generation turned off and size counting turned on, and once "for real". |
| 4473 | * This also means that we don't allocate space until we are sure that the |
| 4474 | * thing really will compile successfully, and we never have to move the |
| 4475 | * code and thus invalidate pointers into it. (Note that it has to be in |
| 4476 | * one piece because free() must be able to free it all.) [NB: not true in perl] |
| 4477 | * |
| 4478 | * Beware that the optimization-preparation code in here knows about some |
| 4479 | * of the structure of the compiled regexp. [I'll say.] |
| 4480 | */ |
| 4481 | |
| 4482 | |
| 4483 | |
| 4484 | #ifndef PERL_IN_XSUB_RE |
| 4485 | #define RE_ENGINE_PTR &PL_core_reg_engine |
| 4486 | #else |
| 4487 | extern const struct regexp_engine my_reg_engine; |
| 4488 | #define RE_ENGINE_PTR &my_reg_engine |
| 4489 | #endif |
| 4490 | |
| 4491 | #ifndef PERL_IN_XSUB_RE |
| 4492 | REGEXP * |
| 4493 | Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags) |
| 4494 | { |
| 4495 | dVAR; |
| 4496 | HV * const table = GvHV(PL_hintgv); |
| 4497 | |
| 4498 | PERL_ARGS_ASSERT_PREGCOMP; |
| 4499 | |
| 4500 | /* Dispatch a request to compile a regexp to correct |
| 4501 | regexp engine. */ |
| 4502 | if (table) { |
| 4503 | SV **ptr= hv_fetchs(table, "regcomp", FALSE); |
| 4504 | GET_RE_DEBUG_FLAGS_DECL; |
| 4505 | if (ptr && SvIOK(*ptr) && SvIV(*ptr)) { |
| 4506 | const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr)); |
| 4507 | DEBUG_COMPILE_r({ |
| 4508 | PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n", |
| 4509 | SvIV(*ptr)); |
| 4510 | }); |
| 4511 | return CALLREGCOMP_ENG(eng, pattern, flags); |
| 4512 | } |
| 4513 | } |
| 4514 | return Perl_re_compile(aTHX_ pattern, flags); |
| 4515 | } |
| 4516 | #endif |
| 4517 | |
| 4518 | REGEXP * |
| 4519 | Perl_re_compile(pTHX_ SV * const pattern, U32 orig_pm_flags) |
| 4520 | { |
| 4521 | dVAR; |
| 4522 | REGEXP *rx; |
| 4523 | struct regexp *r; |
| 4524 | register regexp_internal *ri; |
| 4525 | STRLEN plen; |
| 4526 | char *exp; |
| 4527 | char* xend; |
| 4528 | regnode *scan; |
| 4529 | I32 flags; |
| 4530 | I32 minlen = 0; |
| 4531 | U32 pm_flags; |
| 4532 | |
| 4533 | /* these are all flags - maybe they should be turned |
| 4534 | * into a single int with different bit masks */ |
| 4535 | I32 sawlookahead = 0; |
| 4536 | I32 sawplus = 0; |
| 4537 | I32 sawopen = 0; |
| 4538 | bool used_setjump = FALSE; |
| 4539 | regex_charset initial_charset = get_regex_charset(orig_pm_flags); |
| 4540 | |
| 4541 | U8 jump_ret = 0; |
| 4542 | dJMPENV; |
| 4543 | scan_data_t data; |
| 4544 | RExC_state_t RExC_state; |
| 4545 | RExC_state_t * const pRExC_state = &RExC_state; |
| 4546 | #ifdef TRIE_STUDY_OPT |
| 4547 | int restudied; |
| 4548 | RExC_state_t copyRExC_state; |
| 4549 | #endif |
| 4550 | GET_RE_DEBUG_FLAGS_DECL; |
| 4551 | |
| 4552 | PERL_ARGS_ASSERT_RE_COMPILE; |
| 4553 | |
| 4554 | DEBUG_r(if (!PL_colorset) reginitcolors()); |
| 4555 | |
| 4556 | RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern); |
| 4557 | RExC_uni_semantics = 0; |
| 4558 | RExC_contains_locale = 0; |
| 4559 | |
| 4560 | /****************** LONG JUMP TARGET HERE***********************/ |
| 4561 | /* Longjmp back to here if have to switch in midstream to utf8 */ |
| 4562 | if (! RExC_orig_utf8) { |
| 4563 | JMPENV_PUSH(jump_ret); |
| 4564 | used_setjump = TRUE; |
| 4565 | } |
| 4566 | |
| 4567 | if (jump_ret == 0) { /* First time through */ |
| 4568 | exp = SvPV(pattern, plen); |
| 4569 | xend = exp + plen; |
| 4570 | /* ignore the utf8ness if the pattern is 0 length */ |
| 4571 | if (plen == 0) { |
| 4572 | RExC_utf8 = RExC_orig_utf8 = 0; |
| 4573 | } |
| 4574 | |
| 4575 | DEBUG_COMPILE_r({ |
| 4576 | SV *dsv= sv_newmortal(); |
| 4577 | RE_PV_QUOTED_DECL(s, RExC_utf8, |
| 4578 | dsv, exp, plen, 60); |
| 4579 | PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n", |
| 4580 | PL_colors[4],PL_colors[5],s); |
| 4581 | }); |
| 4582 | } |
| 4583 | else { /* longjumped back */ |
| 4584 | STRLEN len = plen; |
| 4585 | |
| 4586 | /* If the cause for the longjmp was other than changing to utf8, pop |
| 4587 | * our own setjmp, and longjmp to the correct handler */ |
| 4588 | if (jump_ret != UTF8_LONGJMP) { |
| 4589 | JMPENV_POP; |
| 4590 | JMPENV_JUMP(jump_ret); |
| 4591 | } |
| 4592 | |
| 4593 | GET_RE_DEBUG_FLAGS; |
| 4594 | |
| 4595 | /* It's possible to write a regexp in ascii that represents Unicode |
| 4596 | codepoints outside of the byte range, such as via \x{100}. If we |
| 4597 | detect such a sequence we have to convert the entire pattern to utf8 |
| 4598 | and then recompile, as our sizing calculation will have been based |
| 4599 | on 1 byte == 1 character, but we will need to use utf8 to encode |
| 4600 | at least some part of the pattern, and therefore must convert the whole |
| 4601 | thing. |
| 4602 | -- dmq */ |
| 4603 | DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, |
| 4604 | "UTF8 mismatch! Converting to utf8 for resizing and compile\n")); |
| 4605 | exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len); |
| 4606 | xend = exp + len; |
| 4607 | RExC_orig_utf8 = RExC_utf8 = 1; |
| 4608 | SAVEFREEPV(exp); |
| 4609 | } |
| 4610 | |
| 4611 | #ifdef TRIE_STUDY_OPT |
| 4612 | restudied = 0; |
| 4613 | #endif |
| 4614 | |
| 4615 | pm_flags = orig_pm_flags; |
| 4616 | |
| 4617 | if (initial_charset == REGEX_LOCALE_CHARSET) { |
| 4618 | RExC_contains_locale = 1; |
| 4619 | } |
| 4620 | else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) { |
| 4621 | |
| 4622 | /* Set to use unicode semantics if the pattern is in utf8 and has the |
| 4623 | * 'depends' charset specified, as it means unicode when utf8 */ |
| 4624 | set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET); |
| 4625 | } |
| 4626 | |
| 4627 | RExC_precomp = exp; |
| 4628 | RExC_flags = pm_flags; |
| 4629 | RExC_sawback = 0; |
| 4630 | |
| 4631 | RExC_seen = 0; |
| 4632 | RExC_in_lookbehind = 0; |
| 4633 | RExC_seen_zerolen = *exp == '^' ? -1 : 0; |
| 4634 | RExC_seen_evals = 0; |
| 4635 | RExC_extralen = 0; |
| 4636 | RExC_override_recoding = 0; |
| 4637 | |
| 4638 | /* First pass: determine size, legality. */ |
| 4639 | RExC_parse = exp; |
| 4640 | RExC_start = exp; |
| 4641 | RExC_end = xend; |
| 4642 | RExC_naughty = 0; |
| 4643 | RExC_npar = 1; |
| 4644 | RExC_nestroot = 0; |
| 4645 | RExC_size = 0L; |
| 4646 | RExC_emit = &PL_regdummy; |
| 4647 | RExC_whilem_seen = 0; |
| 4648 | RExC_open_parens = NULL; |
| 4649 | RExC_close_parens = NULL; |
| 4650 | RExC_opend = NULL; |
| 4651 | RExC_paren_names = NULL; |
| 4652 | #ifdef DEBUGGING |
| 4653 | RExC_paren_name_list = NULL; |
| 4654 | #endif |
| 4655 | RExC_recurse = NULL; |
| 4656 | RExC_recurse_count = 0; |
| 4657 | |
| 4658 | #if 0 /* REGC() is (currently) a NOP at the first pass. |
| 4659 | * Clever compilers notice this and complain. --jhi */ |
| 4660 | REGC((U8)REG_MAGIC, (char*)RExC_emit); |
| 4661 | #endif |
| 4662 | DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n")); |
| 4663 | if (reg(pRExC_state, 0, &flags,1) == NULL) { |
| 4664 | RExC_precomp = NULL; |
| 4665 | return(NULL); |
| 4666 | } |
| 4667 | |
| 4668 | /* Here, finished first pass. Get rid of any added setjmp */ |
| 4669 | if (used_setjump) { |
| 4670 | JMPENV_POP; |
| 4671 | } |
| 4672 | |
| 4673 | DEBUG_PARSE_r({ |
| 4674 | PerlIO_printf(Perl_debug_log, |
| 4675 | "Required size %"IVdf" nodes\n" |
| 4676 | "Starting second pass (creation)\n", |
| 4677 | (IV)RExC_size); |
| 4678 | RExC_lastnum=0; |
| 4679 | RExC_lastparse=NULL; |
| 4680 | }); |
| 4681 | |
| 4682 | /* The first pass could have found things that force Unicode semantics */ |
| 4683 | if ((RExC_utf8 || RExC_uni_semantics) |
| 4684 | && get_regex_charset(pm_flags) == REGEX_DEPENDS_CHARSET) |
| 4685 | { |
| 4686 | set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET); |
| 4687 | } |
| 4688 | |
| 4689 | /* Small enough for pointer-storage convention? |
| 4690 | If extralen==0, this means that we will not need long jumps. */ |
| 4691 | if (RExC_size >= 0x10000L && RExC_extralen) |
| 4692 | RExC_size += RExC_extralen; |
| 4693 | else |
| 4694 | RExC_extralen = 0; |
| 4695 | if (RExC_whilem_seen > 15) |
| 4696 | RExC_whilem_seen = 15; |
| 4697 | |
| 4698 | /* Allocate space and zero-initialize. Note, the two step process |
| 4699 | of zeroing when in debug mode, thus anything assigned has to |
| 4700 | happen after that */ |
| 4701 | rx = (REGEXP*) newSV_type(SVt_REGEXP); |
| 4702 | r = (struct regexp*)SvANY(rx); |
| 4703 | Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), |
| 4704 | char, regexp_internal); |
| 4705 | if ( r == NULL || ri == NULL ) |
| 4706 | FAIL("Regexp out of space"); |
| 4707 | #ifdef DEBUGGING |
| 4708 | /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */ |
| 4709 | Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char); |
| 4710 | #else |
| 4711 | /* bulk initialize base fields with 0. */ |
| 4712 | Zero(ri, sizeof(regexp_internal), char); |
| 4713 | #endif |
| 4714 | |
| 4715 | /* non-zero initialization begins here */ |
| 4716 | RXi_SET( r, ri ); |
| 4717 | r->engine= RE_ENGINE_PTR; |
| 4718 | r->extflags = pm_flags; |
| 4719 | { |
| 4720 | bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY); |
| 4721 | bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET); |
| 4722 | |
| 4723 | /* The caret is output if there are any defaults: if not all the STD |
| 4724 | * flags are set, or if no character set specifier is needed */ |
| 4725 | bool has_default = |
| 4726 | (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD) |
| 4727 | || ! has_charset); |
| 4728 | bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT); |
| 4729 | U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) |
| 4730 | >> RXf_PMf_STD_PMMOD_SHIFT); |
| 4731 | const char *fptr = STD_PAT_MODS; /*"msix"*/ |
| 4732 | char *p; |
| 4733 | /* Allocate for the worst case, which is all the std flags are turned |
| 4734 | * on. If more precision is desired, we could do a population count of |
| 4735 | * the flags set. This could be done with a small lookup table, or by |
| 4736 | * shifting, masking and adding, or even, when available, assembly |
| 4737 | * language for a machine-language population count. |
| 4738 | * We never output a minus, as all those are defaults, so are |
| 4739 | * covered by the caret */ |
| 4740 | const STRLEN wraplen = plen + has_p + has_runon |
| 4741 | + has_default /* If needs a caret */ |
| 4742 | |
| 4743 | /* If needs a character set specifier */ |
| 4744 | + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0) |
| 4745 | + (sizeof(STD_PAT_MODS) - 1) |
| 4746 | + (sizeof("(?:)") - 1); |
| 4747 | |
| 4748 | p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */ |
| 4749 | SvPOK_on(rx); |
| 4750 | SvFLAGS(rx) |= SvUTF8(pattern); |
| 4751 | *p++='('; *p++='?'; |
| 4752 | |
| 4753 | /* If a default, cover it using the caret */ |
| 4754 | if (has_default) { |
| 4755 | *p++= DEFAULT_PAT_MOD; |
| 4756 | } |
| 4757 | if (has_charset) { |
| 4758 | STRLEN len; |
| 4759 | const char* const name = get_regex_charset_name(r->extflags, &len); |
| 4760 | Copy(name, p, len, char); |
| 4761 | p += len; |
| 4762 | } |
| 4763 | if (has_p) |
| 4764 | *p++ = KEEPCOPY_PAT_MOD; /*'p'*/ |
| 4765 | { |
| 4766 | char ch; |
| 4767 | while((ch = *fptr++)) { |
| 4768 | if(reganch & 1) |
| 4769 | *p++ = ch; |
| 4770 | reganch >>= 1; |
| 4771 | } |
| 4772 | } |
| 4773 | |
| 4774 | *p++ = ':'; |
| 4775 | Copy(RExC_precomp, p, plen, char); |
| 4776 | assert ((RX_WRAPPED(rx) - p) < 16); |
| 4777 | r->pre_prefix = p - RX_WRAPPED(rx); |
| 4778 | p += plen; |
| 4779 | if (has_runon) |
| 4780 | *p++ = '\n'; |
| 4781 | *p++ = ')'; |
| 4782 | *p = 0; |
| 4783 | SvCUR_set(rx, p - SvPVX_const(rx)); |
| 4784 | } |
| 4785 | |
| 4786 | r->intflags = 0; |
| 4787 | r->nparens = RExC_npar - 1; /* set early to validate backrefs */ |
| 4788 | |
| 4789 | if (RExC_seen & REG_SEEN_RECURSE) { |
| 4790 | Newxz(RExC_open_parens, RExC_npar,regnode *); |
| 4791 | SAVEFREEPV(RExC_open_parens); |
| 4792 | Newxz(RExC_close_parens,RExC_npar,regnode *); |
| 4793 | SAVEFREEPV(RExC_close_parens); |
| 4794 | } |
| 4795 | |
| 4796 | /* Useful during FAIL. */ |
| 4797 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 4798 | Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */ |
| 4799 | DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log, |
| 4800 | "%s %"UVuf" bytes for offset annotations.\n", |
| 4801 | ri->u.offsets ? "Got" : "Couldn't get", |
| 4802 | (UV)((2*RExC_size+1) * sizeof(U32)))); |
| 4803 | #endif |
| 4804 | SetProgLen(ri,RExC_size); |
| 4805 | RExC_rx_sv = rx; |
| 4806 | RExC_rx = r; |
| 4807 | RExC_rxi = ri; |
| 4808 | |
| 4809 | /* Second pass: emit code. */ |
| 4810 | RExC_flags = pm_flags; /* don't let top level (?i) bleed */ |
| 4811 | RExC_parse = exp; |
| 4812 | RExC_end = xend; |
| 4813 | RExC_naughty = 0; |
| 4814 | RExC_npar = 1; |
| 4815 | RExC_emit_start = ri->program; |
| 4816 | RExC_emit = ri->program; |
| 4817 | RExC_emit_bound = ri->program + RExC_size + 1; |
| 4818 | |
| 4819 | /* Store the count of eval-groups for security checks: */ |
| 4820 | RExC_rx->seen_evals = RExC_seen_evals; |
| 4821 | REGC((U8)REG_MAGIC, (char*) RExC_emit++); |
| 4822 | if (reg(pRExC_state, 0, &flags,1) == NULL) { |
| 4823 | ReREFCNT_dec(rx); |
| 4824 | return(NULL); |
| 4825 | } |
| 4826 | /* XXXX To minimize changes to RE engine we always allocate |
| 4827 | 3-units-long substrs field. */ |
| 4828 | Newx(r->substrs, 1, struct reg_substr_data); |
| 4829 | if (RExC_recurse_count) { |
| 4830 | Newxz(RExC_recurse,RExC_recurse_count,regnode *); |
| 4831 | SAVEFREEPV(RExC_recurse); |
| 4832 | } |
| 4833 | |
| 4834 | reStudy: |
| 4835 | r->minlen = minlen = sawlookahead = sawplus = sawopen = 0; |
| 4836 | Zero(r->substrs, 1, struct reg_substr_data); |
| 4837 | |
| 4838 | #ifdef TRIE_STUDY_OPT |
| 4839 | if (!restudied) { |
| 4840 | StructCopy(&zero_scan_data, &data, scan_data_t); |
| 4841 | copyRExC_state = RExC_state; |
| 4842 | } else { |
| 4843 | U32 seen=RExC_seen; |
| 4844 | DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n")); |
| 4845 | |
| 4846 | RExC_state = copyRExC_state; |
| 4847 | if (seen & REG_TOP_LEVEL_BRANCHES) |
| 4848 | RExC_seen |= REG_TOP_LEVEL_BRANCHES; |
| 4849 | else |
| 4850 | RExC_seen &= ~REG_TOP_LEVEL_BRANCHES; |
| 4851 | if (data.last_found) { |
| 4852 | SvREFCNT_dec(data.longest_fixed); |
| 4853 | SvREFCNT_dec(data.longest_float); |
| 4854 | SvREFCNT_dec(data.last_found); |
| 4855 | } |
| 4856 | StructCopy(&zero_scan_data, &data, scan_data_t); |
| 4857 | } |
| 4858 | #else |
| 4859 | StructCopy(&zero_scan_data, &data, scan_data_t); |
| 4860 | #endif |
| 4861 | |
| 4862 | /* Dig out information for optimizations. */ |
| 4863 | r->extflags = RExC_flags; /* was pm_op */ |
| 4864 | /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */ |
| 4865 | |
| 4866 | if (UTF) |
| 4867 | SvUTF8_on(rx); /* Unicode in it? */ |
| 4868 | ri->regstclass = NULL; |
| 4869 | if (RExC_naughty >= 10) /* Probably an expensive pattern. */ |
| 4870 | r->intflags |= PREGf_NAUGHTY; |
| 4871 | scan = ri->program + 1; /* First BRANCH. */ |
| 4872 | |
| 4873 | /* testing for BRANCH here tells us whether there is "must appear" |
| 4874 | data in the pattern. If there is then we can use it for optimisations */ |
| 4875 | if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */ |
| 4876 | I32 fake; |
| 4877 | STRLEN longest_float_length, longest_fixed_length; |
| 4878 | struct regnode_charclass_class ch_class; /* pointed to by data */ |
| 4879 | int stclass_flag; |
| 4880 | I32 last_close = 0; /* pointed to by data */ |
| 4881 | regnode *first= scan; |
| 4882 | regnode *first_next= regnext(first); |
| 4883 | /* |
| 4884 | * Skip introductions and multiplicators >= 1 |
| 4885 | * so that we can extract the 'meat' of the pattern that must |
| 4886 | * match in the large if() sequence following. |
| 4887 | * NOTE that EXACT is NOT covered here, as it is normally |
| 4888 | * picked up by the optimiser separately. |
| 4889 | * |
| 4890 | * This is unfortunate as the optimiser isnt handling lookahead |
| 4891 | * properly currently. |
| 4892 | * |
| 4893 | */ |
| 4894 | while ((OP(first) == OPEN && (sawopen = 1)) || |
| 4895 | /* An OR of *one* alternative - should not happen now. */ |
| 4896 | (OP(first) == BRANCH && OP(first_next) != BRANCH) || |
| 4897 | /* for now we can't handle lookbehind IFMATCH*/ |
| 4898 | (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) || |
| 4899 | (OP(first) == PLUS) || |
| 4900 | (OP(first) == MINMOD) || |
| 4901 | /* An {n,m} with n>0 */ |
| 4902 | (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) || |
| 4903 | (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END )) |
| 4904 | { |
| 4905 | /* |
| 4906 | * the only op that could be a regnode is PLUS, all the rest |
| 4907 | * will be regnode_1 or regnode_2. |
| 4908 | * |
| 4909 | */ |
| 4910 | if (OP(first) == PLUS) |
| 4911 | sawplus = 1; |
| 4912 | else |
| 4913 | first += regarglen[OP(first)]; |
| 4914 | |
| 4915 | first = NEXTOPER(first); |
| 4916 | first_next= regnext(first); |
| 4917 | } |
| 4918 | |
| 4919 | /* Starting-point info. */ |
| 4920 | again: |
| 4921 | DEBUG_PEEP("first:",first,0); |
| 4922 | /* Ignore EXACT as we deal with it later. */ |
| 4923 | if (PL_regkind[OP(first)] == EXACT) { |
| 4924 | if (OP(first) == EXACT) |
| 4925 | NOOP; /* Empty, get anchored substr later. */ |
| 4926 | else |
| 4927 | ri->regstclass = first; |
| 4928 | } |
| 4929 | #ifdef TRIE_STCLASS |
| 4930 | else if (PL_regkind[OP(first)] == TRIE && |
| 4931 | ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0) |
| 4932 | { |
| 4933 | regnode *trie_op; |
| 4934 | /* this can happen only on restudy */ |
| 4935 | if ( OP(first) == TRIE ) { |
| 4936 | struct regnode_1 *trieop = (struct regnode_1 *) |
| 4937 | PerlMemShared_calloc(1, sizeof(struct regnode_1)); |
| 4938 | StructCopy(first,trieop,struct regnode_1); |
| 4939 | trie_op=(regnode *)trieop; |
| 4940 | } else { |
| 4941 | struct regnode_charclass *trieop = (struct regnode_charclass *) |
| 4942 | PerlMemShared_calloc(1, sizeof(struct regnode_charclass)); |
| 4943 | StructCopy(first,trieop,struct regnode_charclass); |
| 4944 | trie_op=(regnode *)trieop; |
| 4945 | } |
| 4946 | OP(trie_op)+=2; |
| 4947 | make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0); |
| 4948 | ri->regstclass = trie_op; |
| 4949 | } |
| 4950 | #endif |
| 4951 | else if (REGNODE_SIMPLE(OP(first))) |
| 4952 | ri->regstclass = first; |
| 4953 | else if (PL_regkind[OP(first)] == BOUND || |
| 4954 | PL_regkind[OP(first)] == NBOUND) |
| 4955 | ri->regstclass = first; |
| 4956 | else if (PL_regkind[OP(first)] == BOL) { |
| 4957 | r->extflags |= (OP(first) == MBOL |
| 4958 | ? RXf_ANCH_MBOL |
| 4959 | : (OP(first) == SBOL |
| 4960 | ? RXf_ANCH_SBOL |
| 4961 | : RXf_ANCH_BOL)); |
| 4962 | first = NEXTOPER(first); |
| 4963 | goto again; |
| 4964 | } |
| 4965 | else if (OP(first) == GPOS) { |
| 4966 | r->extflags |= RXf_ANCH_GPOS; |
| 4967 | first = NEXTOPER(first); |
| 4968 | goto again; |
| 4969 | } |
| 4970 | else if ((!sawopen || !RExC_sawback) && |
| 4971 | (OP(first) == STAR && |
| 4972 | PL_regkind[OP(NEXTOPER(first))] == REG_ANY) && |
| 4973 | !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL)) |
| 4974 | { |
| 4975 | /* turn .* into ^.* with an implied $*=1 */ |
| 4976 | const int type = |
| 4977 | (OP(NEXTOPER(first)) == REG_ANY) |
| 4978 | ? RXf_ANCH_MBOL |
| 4979 | : RXf_ANCH_SBOL; |
| 4980 | r->extflags |= type; |
| 4981 | r->intflags |= PREGf_IMPLICIT; |
| 4982 | first = NEXTOPER(first); |
| 4983 | goto again; |
| 4984 | } |
| 4985 | if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback) |
| 4986 | && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */ |
| 4987 | /* x+ must match at the 1st pos of run of x's */ |
| 4988 | r->intflags |= PREGf_SKIP; |
| 4989 | |
| 4990 | /* Scan is after the zeroth branch, first is atomic matcher. */ |
| 4991 | #ifdef TRIE_STUDY_OPT |
| 4992 | DEBUG_PARSE_r( |
| 4993 | if (!restudied) |
| 4994 | PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n", |
| 4995 | (IV)(first - scan + 1)) |
| 4996 | ); |
| 4997 | #else |
| 4998 | DEBUG_PARSE_r( |
| 4999 | PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n", |
| 5000 | (IV)(first - scan + 1)) |
| 5001 | ); |
| 5002 | #endif |
| 5003 | |
| 5004 | |
| 5005 | /* |
| 5006 | * If there's something expensive in the r.e., find the |
| 5007 | * longest literal string that must appear and make it the |
| 5008 | * regmust. Resolve ties in favor of later strings, since |
| 5009 | * the regstart check works with the beginning of the r.e. |
| 5010 | * and avoiding duplication strengthens checking. Not a |
| 5011 | * strong reason, but sufficient in the absence of others. |
| 5012 | * [Now we resolve ties in favor of the earlier string if |
| 5013 | * it happens that c_offset_min has been invalidated, since the |
| 5014 | * earlier string may buy us something the later one won't.] |
| 5015 | */ |
| 5016 | |
| 5017 | data.longest_fixed = newSVpvs(""); |
| 5018 | data.longest_float = newSVpvs(""); |
| 5019 | data.last_found = newSVpvs(""); |
| 5020 | data.longest = &(data.longest_fixed); |
| 5021 | first = scan; |
| 5022 | if (!ri->regstclass) { |
| 5023 | cl_init(pRExC_state, &ch_class); |
| 5024 | data.start_class = &ch_class; |
| 5025 | stclass_flag = SCF_DO_STCLASS_AND; |
| 5026 | } else /* XXXX Check for BOUND? */ |
| 5027 | stclass_flag = 0; |
| 5028 | data.last_closep = &last_close; |
| 5029 | |
| 5030 | minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */ |
| 5031 | &data, -1, NULL, NULL, |
| 5032 | SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0); |
| 5033 | |
| 5034 | |
| 5035 | CHECK_RESTUDY_GOTO; |
| 5036 | |
| 5037 | |
| 5038 | if ( RExC_npar == 1 && data.longest == &(data.longest_fixed) |
| 5039 | && data.last_start_min == 0 && data.last_end > 0 |
| 5040 | && !RExC_seen_zerolen |
| 5041 | && !(RExC_seen & REG_SEEN_VERBARG) |
| 5042 | && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS))) |
| 5043 | r->extflags |= RXf_CHECK_ALL; |
| 5044 | scan_commit(pRExC_state, &data,&minlen,0); |
| 5045 | SvREFCNT_dec(data.last_found); |
| 5046 | |
| 5047 | /* Note that code very similar to this but for anchored string |
| 5048 | follows immediately below, changes may need to be made to both. |
| 5049 | Be careful. |
| 5050 | */ |
| 5051 | longest_float_length = CHR_SVLEN(data.longest_float); |
| 5052 | if (longest_float_length |
| 5053 | || (data.flags & SF_FL_BEFORE_EOL |
| 5054 | && (!(data.flags & SF_FL_BEFORE_MEOL) |
| 5055 | || (RExC_flags & RXf_PMf_MULTILINE)))) |
| 5056 | { |
| 5057 | I32 t,ml; |
| 5058 | |
| 5059 | if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */ |
| 5060 | && data.offset_fixed == data.offset_float_min |
| 5061 | && SvCUR(data.longest_fixed) == SvCUR(data.longest_float)) |
| 5062 | goto remove_float; /* As in (a)+. */ |
| 5063 | |
| 5064 | /* copy the information about the longest float from the reg_scan_data |
| 5065 | over to the program. */ |
| 5066 | if (SvUTF8(data.longest_float)) { |
| 5067 | r->float_utf8 = data.longest_float; |
| 5068 | r->float_substr = NULL; |
| 5069 | } else { |
| 5070 | r->float_substr = data.longest_float; |
| 5071 | r->float_utf8 = NULL; |
| 5072 | } |
| 5073 | /* float_end_shift is how many chars that must be matched that |
| 5074 | follow this item. We calculate it ahead of time as once the |
| 5075 | lookbehind offset is added in we lose the ability to correctly |
| 5076 | calculate it.*/ |
| 5077 | ml = data.minlen_float ? *(data.minlen_float) |
| 5078 | : (I32)longest_float_length; |
| 5079 | r->float_end_shift = ml - data.offset_float_min |
| 5080 | - longest_float_length + (SvTAIL(data.longest_float) != 0) |
| 5081 | + data.lookbehind_float; |
| 5082 | r->float_min_offset = data.offset_float_min - data.lookbehind_float; |
| 5083 | r->float_max_offset = data.offset_float_max; |
| 5084 | if (data.offset_float_max < I32_MAX) /* Don't offset infinity */ |
| 5085 | r->float_max_offset -= data.lookbehind_float; |
| 5086 | |
| 5087 | t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */ |
| 5088 | && (!(data.flags & SF_FL_BEFORE_MEOL) |
| 5089 | || (RExC_flags & RXf_PMf_MULTILINE))); |
| 5090 | fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0); |
| 5091 | } |
| 5092 | else { |
| 5093 | remove_float: |
| 5094 | r->float_substr = r->float_utf8 = NULL; |
| 5095 | SvREFCNT_dec(data.longest_float); |
| 5096 | longest_float_length = 0; |
| 5097 | } |
| 5098 | |
| 5099 | /* Note that code very similar to this but for floating string |
| 5100 | is immediately above, changes may need to be made to both. |
| 5101 | Be careful. |
| 5102 | */ |
| 5103 | longest_fixed_length = CHR_SVLEN(data.longest_fixed); |
| 5104 | if (longest_fixed_length |
| 5105 | || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */ |
| 5106 | && (!(data.flags & SF_FIX_BEFORE_MEOL) |
| 5107 | || (RExC_flags & RXf_PMf_MULTILINE)))) |
| 5108 | { |
| 5109 | I32 t,ml; |
| 5110 | |
| 5111 | /* copy the information about the longest fixed |
| 5112 | from the reg_scan_data over to the program. */ |
| 5113 | if (SvUTF8(data.longest_fixed)) { |
| 5114 | r->anchored_utf8 = data.longest_fixed; |
| 5115 | r->anchored_substr = NULL; |
| 5116 | } else { |
| 5117 | r->anchored_substr = data.longest_fixed; |
| 5118 | r->anchored_utf8 = NULL; |
| 5119 | } |
| 5120 | /* fixed_end_shift is how many chars that must be matched that |
| 5121 | follow this item. We calculate it ahead of time as once the |
| 5122 | lookbehind offset is added in we lose the ability to correctly |
| 5123 | calculate it.*/ |
| 5124 | ml = data.minlen_fixed ? *(data.minlen_fixed) |
| 5125 | : (I32)longest_fixed_length; |
| 5126 | r->anchored_end_shift = ml - data.offset_fixed |
| 5127 | - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0) |
| 5128 | + data.lookbehind_fixed; |
| 5129 | r->anchored_offset = data.offset_fixed - data.lookbehind_fixed; |
| 5130 | |
| 5131 | t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */ |
| 5132 | && (!(data.flags & SF_FIX_BEFORE_MEOL) |
| 5133 | || (RExC_flags & RXf_PMf_MULTILINE))); |
| 5134 | fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0); |
| 5135 | } |
| 5136 | else { |
| 5137 | r->anchored_substr = r->anchored_utf8 = NULL; |
| 5138 | SvREFCNT_dec(data.longest_fixed); |
| 5139 | longest_fixed_length = 0; |
| 5140 | } |
| 5141 | if (ri->regstclass |
| 5142 | && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY)) |
| 5143 | ri->regstclass = NULL; |
| 5144 | |
| 5145 | if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset) |
| 5146 | && stclass_flag |
| 5147 | && !(data.start_class->flags & ANYOF_EOS) |
| 5148 | && !cl_is_anything(data.start_class)) |
| 5149 | { |
| 5150 | const U32 n = add_data(pRExC_state, 1, "f"); |
| 5151 | data.start_class->flags |= ANYOF_IS_SYNTHETIC; |
| 5152 | |
| 5153 | Newx(RExC_rxi->data->data[n], 1, |
| 5154 | struct regnode_charclass_class); |
| 5155 | StructCopy(data.start_class, |
| 5156 | (struct regnode_charclass_class*)RExC_rxi->data->data[n], |
| 5157 | struct regnode_charclass_class); |
| 5158 | ri->regstclass = (regnode*)RExC_rxi->data->data[n]; |
| 5159 | r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
| 5160 | DEBUG_COMPILE_r({ SV *sv = sv_newmortal(); |
| 5161 | regprop(r, sv, (regnode*)data.start_class); |
| 5162 | PerlIO_printf(Perl_debug_log, |
| 5163 | "synthetic stclass \"%s\".\n", |
| 5164 | SvPVX_const(sv));}); |
| 5165 | } |
| 5166 | |
| 5167 | /* A temporary algorithm prefers floated substr to fixed one to dig more info. */ |
| 5168 | if (longest_fixed_length > longest_float_length) { |
| 5169 | r->check_end_shift = r->anchored_end_shift; |
| 5170 | r->check_substr = r->anchored_substr; |
| 5171 | r->check_utf8 = r->anchored_utf8; |
| 5172 | r->check_offset_min = r->check_offset_max = r->anchored_offset; |
| 5173 | if (r->extflags & RXf_ANCH_SINGLE) |
| 5174 | r->extflags |= RXf_NOSCAN; |
| 5175 | } |
| 5176 | else { |
| 5177 | r->check_end_shift = r->float_end_shift; |
| 5178 | r->check_substr = r->float_substr; |
| 5179 | r->check_utf8 = r->float_utf8; |
| 5180 | r->check_offset_min = r->float_min_offset; |
| 5181 | r->check_offset_max = r->float_max_offset; |
| 5182 | } |
| 5183 | /* XXXX Currently intuiting is not compatible with ANCH_GPOS. |
| 5184 | This should be changed ASAP! */ |
| 5185 | if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) { |
| 5186 | r->extflags |= RXf_USE_INTUIT; |
| 5187 | if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8)) |
| 5188 | r->extflags |= RXf_INTUIT_TAIL; |
| 5189 | } |
| 5190 | /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere) |
| 5191 | if ( (STRLEN)minlen < longest_float_length ) |
| 5192 | minlen= longest_float_length; |
| 5193 | if ( (STRLEN)minlen < longest_fixed_length ) |
| 5194 | minlen= longest_fixed_length; |
| 5195 | */ |
| 5196 | } |
| 5197 | else { |
| 5198 | /* Several toplevels. Best we can is to set minlen. */ |
| 5199 | I32 fake; |
| 5200 | struct regnode_charclass_class ch_class; |
| 5201 | I32 last_close = 0; |
| 5202 | |
| 5203 | DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n")); |
| 5204 | |
| 5205 | scan = ri->program + 1; |
| 5206 | cl_init(pRExC_state, &ch_class); |
| 5207 | data.start_class = &ch_class; |
| 5208 | data.last_closep = &last_close; |
| 5209 | |
| 5210 | |
| 5211 | minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size, |
| 5212 | &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0); |
| 5213 | |
| 5214 | CHECK_RESTUDY_GOTO; |
| 5215 | |
| 5216 | r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8 |
| 5217 | = r->float_substr = r->float_utf8 = NULL; |
| 5218 | |
| 5219 | if (!(data.start_class->flags & ANYOF_EOS) |
| 5220 | && !cl_is_anything(data.start_class)) |
| 5221 | { |
| 5222 | const U32 n = add_data(pRExC_state, 1, "f"); |
| 5223 | data.start_class->flags |= ANYOF_IS_SYNTHETIC; |
| 5224 | |
| 5225 | Newx(RExC_rxi->data->data[n], 1, |
| 5226 | struct regnode_charclass_class); |
| 5227 | StructCopy(data.start_class, |
| 5228 | (struct regnode_charclass_class*)RExC_rxi->data->data[n], |
| 5229 | struct regnode_charclass_class); |
| 5230 | ri->regstclass = (regnode*)RExC_rxi->data->data[n]; |
| 5231 | r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
| 5232 | DEBUG_COMPILE_r({ SV* sv = sv_newmortal(); |
| 5233 | regprop(r, sv, (regnode*)data.start_class); |
| 5234 | PerlIO_printf(Perl_debug_log, |
| 5235 | "synthetic stclass \"%s\".\n", |
| 5236 | SvPVX_const(sv));}); |
| 5237 | } |
| 5238 | } |
| 5239 | |
| 5240 | /* Guard against an embedded (?=) or (?<=) with a longer minlen than |
| 5241 | the "real" pattern. */ |
| 5242 | DEBUG_OPTIMISE_r({ |
| 5243 | PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n", |
| 5244 | (IV)minlen, (IV)r->minlen); |
| 5245 | }); |
| 5246 | r->minlenret = minlen; |
| 5247 | if (r->minlen < minlen) |
| 5248 | r->minlen = minlen; |
| 5249 | |
| 5250 | if (RExC_seen & REG_SEEN_GPOS) |
| 5251 | r->extflags |= RXf_GPOS_SEEN; |
| 5252 | if (RExC_seen & REG_SEEN_LOOKBEHIND) |
| 5253 | r->extflags |= RXf_LOOKBEHIND_SEEN; |
| 5254 | if (RExC_seen & REG_SEEN_EVAL) |
| 5255 | r->extflags |= RXf_EVAL_SEEN; |
| 5256 | if (RExC_seen & REG_SEEN_CANY) |
| 5257 | r->extflags |= RXf_CANY_SEEN; |
| 5258 | if (RExC_seen & REG_SEEN_VERBARG) |
| 5259 | r->intflags |= PREGf_VERBARG_SEEN; |
| 5260 | if (RExC_seen & REG_SEEN_CUTGROUP) |
| 5261 | r->intflags |= PREGf_CUTGROUP_SEEN; |
| 5262 | if (RExC_paren_names) |
| 5263 | RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names)); |
| 5264 | else |
| 5265 | RXp_PAREN_NAMES(r) = NULL; |
| 5266 | |
| 5267 | #ifdef STUPID_PATTERN_CHECKS |
| 5268 | if (RX_PRELEN(rx) == 0) |
| 5269 | r->extflags |= RXf_NULL; |
| 5270 | if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ') |
| 5271 | /* XXX: this should happen BEFORE we compile */ |
| 5272 | r->extflags |= (RXf_SKIPWHITE|RXf_WHITE); |
| 5273 | else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3)) |
| 5274 | r->extflags |= RXf_WHITE; |
| 5275 | else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^') |
| 5276 | r->extflags |= RXf_START_ONLY; |
| 5277 | #else |
| 5278 | if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ') |
| 5279 | /* XXX: this should happen BEFORE we compile */ |
| 5280 | r->extflags |= (RXf_SKIPWHITE|RXf_WHITE); |
| 5281 | else { |
| 5282 | regnode *first = ri->program + 1; |
| 5283 | U8 fop = OP(first); |
| 5284 | |
| 5285 | if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END) |
| 5286 | r->extflags |= RXf_NULL; |
| 5287 | else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END) |
| 5288 | r->extflags |= RXf_START_ONLY; |
| 5289 | else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE |
| 5290 | && OP(regnext(first)) == END) |
| 5291 | r->extflags |= RXf_WHITE; |
| 5292 | } |
| 5293 | #endif |
| 5294 | #ifdef DEBUGGING |
| 5295 | if (RExC_paren_names) { |
| 5296 | ri->name_list_idx = add_data( pRExC_state, 1, "a" ); |
| 5297 | ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list); |
| 5298 | } else |
| 5299 | #endif |
| 5300 | ri->name_list_idx = 0; |
| 5301 | |
| 5302 | if (RExC_recurse_count) { |
| 5303 | for ( ; RExC_recurse_count ; RExC_recurse_count-- ) { |
| 5304 | const regnode *scan = RExC_recurse[RExC_recurse_count-1]; |
| 5305 | ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan ); |
| 5306 | } |
| 5307 | } |
| 5308 | Newxz(r->offs, RExC_npar, regexp_paren_pair); |
| 5309 | /* assume we don't need to swap parens around before we match */ |
| 5310 | |
| 5311 | DEBUG_DUMP_r({ |
| 5312 | PerlIO_printf(Perl_debug_log,"Final program:\n"); |
| 5313 | regdump(r); |
| 5314 | }); |
| 5315 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 5316 | DEBUG_OFFSETS_r(if (ri->u.offsets) { |
| 5317 | const U32 len = ri->u.offsets[0]; |
| 5318 | U32 i; |
| 5319 | GET_RE_DEBUG_FLAGS_DECL; |
| 5320 | PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]); |
| 5321 | for (i = 1; i <= len; i++) { |
| 5322 | if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2]) |
| 5323 | PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ", |
| 5324 | (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]); |
| 5325 | } |
| 5326 | PerlIO_printf(Perl_debug_log, "\n"); |
| 5327 | }); |
| 5328 | #endif |
| 5329 | return rx; |
| 5330 | } |
| 5331 | |
| 5332 | #undef RE_ENGINE_PTR |
| 5333 | |
| 5334 | |
| 5335 | SV* |
| 5336 | Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value, |
| 5337 | const U32 flags) |
| 5338 | { |
| 5339 | PERL_ARGS_ASSERT_REG_NAMED_BUFF; |
| 5340 | |
| 5341 | PERL_UNUSED_ARG(value); |
| 5342 | |
| 5343 | if (flags & RXapif_FETCH) { |
| 5344 | return reg_named_buff_fetch(rx, key, flags); |
| 5345 | } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) { |
| 5346 | Perl_croak_no_modify(aTHX); |
| 5347 | return NULL; |
| 5348 | } else if (flags & RXapif_EXISTS) { |
| 5349 | return reg_named_buff_exists(rx, key, flags) |
| 5350 | ? &PL_sv_yes |
| 5351 | : &PL_sv_no; |
| 5352 | } else if (flags & RXapif_REGNAMES) { |
| 5353 | return reg_named_buff_all(rx, flags); |
| 5354 | } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) { |
| 5355 | return reg_named_buff_scalar(rx, flags); |
| 5356 | } else { |
| 5357 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags); |
| 5358 | return NULL; |
| 5359 | } |
| 5360 | } |
| 5361 | |
| 5362 | SV* |
| 5363 | Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey, |
| 5364 | const U32 flags) |
| 5365 | { |
| 5366 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER; |
| 5367 | PERL_UNUSED_ARG(lastkey); |
| 5368 | |
| 5369 | if (flags & RXapif_FIRSTKEY) |
| 5370 | return reg_named_buff_firstkey(rx, flags); |
| 5371 | else if (flags & RXapif_NEXTKEY) |
| 5372 | return reg_named_buff_nextkey(rx, flags); |
| 5373 | else { |
| 5374 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags); |
| 5375 | return NULL; |
| 5376 | } |
| 5377 | } |
| 5378 | |
| 5379 | SV* |
| 5380 | Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv, |
| 5381 | const U32 flags) |
| 5382 | { |
| 5383 | AV *retarray = NULL; |
| 5384 | SV *ret; |
| 5385 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5386 | |
| 5387 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH; |
| 5388 | |
| 5389 | if (flags & RXapif_ALL) |
| 5390 | retarray=newAV(); |
| 5391 | |
| 5392 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 5393 | HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 ); |
| 5394 | if (he_str) { |
| 5395 | IV i; |
| 5396 | SV* sv_dat=HeVAL(he_str); |
| 5397 | I32 *nums=(I32*)SvPVX(sv_dat); |
| 5398 | for ( i=0; i<SvIVX(sv_dat); i++ ) { |
| 5399 | if ((I32)(rx->nparens) >= nums[i] |
| 5400 | && rx->offs[nums[i]].start != -1 |
| 5401 | && rx->offs[nums[i]].end != -1) |
| 5402 | { |
| 5403 | ret = newSVpvs(""); |
| 5404 | CALLREG_NUMBUF_FETCH(r,nums[i],ret); |
| 5405 | if (!retarray) |
| 5406 | return ret; |
| 5407 | } else { |
| 5408 | ret = newSVsv(&PL_sv_undef); |
| 5409 | } |
| 5410 | if (retarray) |
| 5411 | av_push(retarray, ret); |
| 5412 | } |
| 5413 | if (retarray) |
| 5414 | return newRV_noinc(MUTABLE_SV(retarray)); |
| 5415 | } |
| 5416 | } |
| 5417 | return NULL; |
| 5418 | } |
| 5419 | |
| 5420 | bool |
| 5421 | Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key, |
| 5422 | const U32 flags) |
| 5423 | { |
| 5424 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5425 | |
| 5426 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS; |
| 5427 | |
| 5428 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 5429 | if (flags & RXapif_ALL) { |
| 5430 | return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0); |
| 5431 | } else { |
| 5432 | SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags); |
| 5433 | if (sv) { |
| 5434 | SvREFCNT_dec(sv); |
| 5435 | return TRUE; |
| 5436 | } else { |
| 5437 | return FALSE; |
| 5438 | } |
| 5439 | } |
| 5440 | } else { |
| 5441 | return FALSE; |
| 5442 | } |
| 5443 | } |
| 5444 | |
| 5445 | SV* |
| 5446 | Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags) |
| 5447 | { |
| 5448 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5449 | |
| 5450 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY; |
| 5451 | |
| 5452 | if ( rx && RXp_PAREN_NAMES(rx) ) { |
| 5453 | (void)hv_iterinit(RXp_PAREN_NAMES(rx)); |
| 5454 | |
| 5455 | return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY); |
| 5456 | } else { |
| 5457 | return FALSE; |
| 5458 | } |
| 5459 | } |
| 5460 | |
| 5461 | SV* |
| 5462 | Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags) |
| 5463 | { |
| 5464 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5465 | GET_RE_DEBUG_FLAGS_DECL; |
| 5466 | |
| 5467 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY; |
| 5468 | |
| 5469 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 5470 | HV *hv = RXp_PAREN_NAMES(rx); |
| 5471 | HE *temphe; |
| 5472 | while ( (temphe = hv_iternext_flags(hv,0)) ) { |
| 5473 | IV i; |
| 5474 | IV parno = 0; |
| 5475 | SV* sv_dat = HeVAL(temphe); |
| 5476 | I32 *nums = (I32*)SvPVX(sv_dat); |
| 5477 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { |
| 5478 | if ((I32)(rx->lastparen) >= nums[i] && |
| 5479 | rx->offs[nums[i]].start != -1 && |
| 5480 | rx->offs[nums[i]].end != -1) |
| 5481 | { |
| 5482 | parno = nums[i]; |
| 5483 | break; |
| 5484 | } |
| 5485 | } |
| 5486 | if (parno || flags & RXapif_ALL) { |
| 5487 | return newSVhek(HeKEY_hek(temphe)); |
| 5488 | } |
| 5489 | } |
| 5490 | } |
| 5491 | return NULL; |
| 5492 | } |
| 5493 | |
| 5494 | SV* |
| 5495 | Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags) |
| 5496 | { |
| 5497 | SV *ret; |
| 5498 | AV *av; |
| 5499 | I32 length; |
| 5500 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5501 | |
| 5502 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR; |
| 5503 | |
| 5504 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 5505 | if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) { |
| 5506 | return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx))); |
| 5507 | } else if (flags & RXapif_ONE) { |
| 5508 | ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES)); |
| 5509 | av = MUTABLE_AV(SvRV(ret)); |
| 5510 | length = av_len(av); |
| 5511 | SvREFCNT_dec(ret); |
| 5512 | return newSViv(length + 1); |
| 5513 | } else { |
| 5514 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags); |
| 5515 | return NULL; |
| 5516 | } |
| 5517 | } |
| 5518 | return &PL_sv_undef; |
| 5519 | } |
| 5520 | |
| 5521 | SV* |
| 5522 | Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags) |
| 5523 | { |
| 5524 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5525 | AV *av = newAV(); |
| 5526 | |
| 5527 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL; |
| 5528 | |
| 5529 | if (rx && RXp_PAREN_NAMES(rx)) { |
| 5530 | HV *hv= RXp_PAREN_NAMES(rx); |
| 5531 | HE *temphe; |
| 5532 | (void)hv_iterinit(hv); |
| 5533 | while ( (temphe = hv_iternext_flags(hv,0)) ) { |
| 5534 | IV i; |
| 5535 | IV parno = 0; |
| 5536 | SV* sv_dat = HeVAL(temphe); |
| 5537 | I32 *nums = (I32*)SvPVX(sv_dat); |
| 5538 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { |
| 5539 | if ((I32)(rx->lastparen) >= nums[i] && |
| 5540 | rx->offs[nums[i]].start != -1 && |
| 5541 | rx->offs[nums[i]].end != -1) |
| 5542 | { |
| 5543 | parno = nums[i]; |
| 5544 | break; |
| 5545 | } |
| 5546 | } |
| 5547 | if (parno || flags & RXapif_ALL) { |
| 5548 | av_push(av, newSVhek(HeKEY_hek(temphe))); |
| 5549 | } |
| 5550 | } |
| 5551 | } |
| 5552 | |
| 5553 | return newRV_noinc(MUTABLE_SV(av)); |
| 5554 | } |
| 5555 | |
| 5556 | void |
| 5557 | Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren, |
| 5558 | SV * const sv) |
| 5559 | { |
| 5560 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5561 | char *s = NULL; |
| 5562 | I32 i = 0; |
| 5563 | I32 s1, t1; |
| 5564 | |
| 5565 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH; |
| 5566 | |
| 5567 | if (!rx->subbeg) { |
| 5568 | sv_setsv(sv,&PL_sv_undef); |
| 5569 | return; |
| 5570 | } |
| 5571 | else |
| 5572 | if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) { |
| 5573 | /* $` */ |
| 5574 | i = rx->offs[0].start; |
| 5575 | s = rx->subbeg; |
| 5576 | } |
| 5577 | else |
| 5578 | if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) { |
| 5579 | /* $' */ |
| 5580 | s = rx->subbeg + rx->offs[0].end; |
| 5581 | i = rx->sublen - rx->offs[0].end; |
| 5582 | } |
| 5583 | else |
| 5584 | if ( 0 <= paren && paren <= (I32)rx->nparens && |
| 5585 | (s1 = rx->offs[paren].start) != -1 && |
| 5586 | (t1 = rx->offs[paren].end) != -1) |
| 5587 | { |
| 5588 | /* $& $1 ... */ |
| 5589 | i = t1 - s1; |
| 5590 | s = rx->subbeg + s1; |
| 5591 | } else { |
| 5592 | sv_setsv(sv,&PL_sv_undef); |
| 5593 | return; |
| 5594 | } |
| 5595 | assert(rx->sublen >= (s - rx->subbeg) + i ); |
| 5596 | if (i >= 0) { |
| 5597 | const int oldtainted = PL_tainted; |
| 5598 | TAINT_NOT; |
| 5599 | sv_setpvn(sv, s, i); |
| 5600 | PL_tainted = oldtainted; |
| 5601 | if ( (rx->extflags & RXf_CANY_SEEN) |
| 5602 | ? (RXp_MATCH_UTF8(rx) |
| 5603 | && (!i || is_utf8_string((U8*)s, i))) |
| 5604 | : (RXp_MATCH_UTF8(rx)) ) |
| 5605 | { |
| 5606 | SvUTF8_on(sv); |
| 5607 | } |
| 5608 | else |
| 5609 | SvUTF8_off(sv); |
| 5610 | if (PL_tainting) { |
| 5611 | if (RXp_MATCH_TAINTED(rx)) { |
| 5612 | if (SvTYPE(sv) >= SVt_PVMG) { |
| 5613 | MAGIC* const mg = SvMAGIC(sv); |
| 5614 | MAGIC* mgt; |
| 5615 | PL_tainted = 1; |
| 5616 | SvMAGIC_set(sv, mg->mg_moremagic); |
| 5617 | SvTAINT(sv); |
| 5618 | if ((mgt = SvMAGIC(sv))) { |
| 5619 | mg->mg_moremagic = mgt; |
| 5620 | SvMAGIC_set(sv, mg); |
| 5621 | } |
| 5622 | } else { |
| 5623 | PL_tainted = 1; |
| 5624 | SvTAINT(sv); |
| 5625 | } |
| 5626 | } else |
| 5627 | SvTAINTED_off(sv); |
| 5628 | } |
| 5629 | } else { |
| 5630 | sv_setsv(sv,&PL_sv_undef); |
| 5631 | return; |
| 5632 | } |
| 5633 | } |
| 5634 | |
| 5635 | void |
| 5636 | Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren, |
| 5637 | SV const * const value) |
| 5638 | { |
| 5639 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE; |
| 5640 | |
| 5641 | PERL_UNUSED_ARG(rx); |
| 5642 | PERL_UNUSED_ARG(paren); |
| 5643 | PERL_UNUSED_ARG(value); |
| 5644 | |
| 5645 | if (!PL_localizing) |
| 5646 | Perl_croak_no_modify(aTHX); |
| 5647 | } |
| 5648 | |
| 5649 | I32 |
| 5650 | Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv, |
| 5651 | const I32 paren) |
| 5652 | { |
| 5653 | struct regexp *const rx = (struct regexp *)SvANY(r); |
| 5654 | I32 i; |
| 5655 | I32 s1, t1; |
| 5656 | |
| 5657 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH; |
| 5658 | |
| 5659 | /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */ |
| 5660 | switch (paren) { |
| 5661 | /* $` / ${^PREMATCH} */ |
| 5662 | case RX_BUFF_IDX_PREMATCH: |
| 5663 | if (rx->offs[0].start != -1) { |
| 5664 | i = rx->offs[0].start; |
| 5665 | if (i > 0) { |
| 5666 | s1 = 0; |
| 5667 | t1 = i; |
| 5668 | goto getlen; |
| 5669 | } |
| 5670 | } |
| 5671 | return 0; |
| 5672 | /* $' / ${^POSTMATCH} */ |
| 5673 | case RX_BUFF_IDX_POSTMATCH: |
| 5674 | if (rx->offs[0].end != -1) { |
| 5675 | i = rx->sublen - rx->offs[0].end; |
| 5676 | if (i > 0) { |
| 5677 | s1 = rx->offs[0].end; |
| 5678 | t1 = rx->sublen; |
| 5679 | goto getlen; |
| 5680 | } |
| 5681 | } |
| 5682 | return 0; |
| 5683 | /* $& / ${^MATCH}, $1, $2, ... */ |
| 5684 | default: |
| 5685 | if (paren <= (I32)rx->nparens && |
| 5686 | (s1 = rx->offs[paren].start) != -1 && |
| 5687 | (t1 = rx->offs[paren].end) != -1) |
| 5688 | { |
| 5689 | i = t1 - s1; |
| 5690 | goto getlen; |
| 5691 | } else { |
| 5692 | if (ckWARN(WARN_UNINITIALIZED)) |
| 5693 | report_uninit((const SV *)sv); |
| 5694 | return 0; |
| 5695 | } |
| 5696 | } |
| 5697 | getlen: |
| 5698 | if (i > 0 && RXp_MATCH_UTF8(rx)) { |
| 5699 | const char * const s = rx->subbeg + s1; |
| 5700 | const U8 *ep; |
| 5701 | STRLEN el; |
| 5702 | |
| 5703 | i = t1 - s1; |
| 5704 | if (is_utf8_string_loclen((U8*)s, i, &ep, &el)) |
| 5705 | i = el; |
| 5706 | } |
| 5707 | return i; |
| 5708 | } |
| 5709 | |
| 5710 | SV* |
| 5711 | Perl_reg_qr_package(pTHX_ REGEXP * const rx) |
| 5712 | { |
| 5713 | PERL_ARGS_ASSERT_REG_QR_PACKAGE; |
| 5714 | PERL_UNUSED_ARG(rx); |
| 5715 | if (0) |
| 5716 | return NULL; |
| 5717 | else |
| 5718 | return newSVpvs("Regexp"); |
| 5719 | } |
| 5720 | |
| 5721 | /* Scans the name of a named buffer from the pattern. |
| 5722 | * If flags is REG_RSN_RETURN_NULL returns null. |
| 5723 | * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name |
| 5724 | * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding |
| 5725 | * to the parsed name as looked up in the RExC_paren_names hash. |
| 5726 | * If there is an error throws a vFAIL().. type exception. |
| 5727 | */ |
| 5728 | |
| 5729 | #define REG_RSN_RETURN_NULL 0 |
| 5730 | #define REG_RSN_RETURN_NAME 1 |
| 5731 | #define REG_RSN_RETURN_DATA 2 |
| 5732 | |
| 5733 | STATIC SV* |
| 5734 | S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) |
| 5735 | { |
| 5736 | char *name_start = RExC_parse; |
| 5737 | |
| 5738 | PERL_ARGS_ASSERT_REG_SCAN_NAME; |
| 5739 | |
| 5740 | if (isIDFIRST_lazy_if(RExC_parse, UTF)) { |
| 5741 | /* skip IDFIRST by using do...while */ |
| 5742 | if (UTF) |
| 5743 | do { |
| 5744 | RExC_parse += UTF8SKIP(RExC_parse); |
| 5745 | } while (isALNUM_utf8((U8*)RExC_parse)); |
| 5746 | else |
| 5747 | do { |
| 5748 | RExC_parse++; |
| 5749 | } while (isALNUM(*RExC_parse)); |
| 5750 | } |
| 5751 | |
| 5752 | if ( flags ) { |
| 5753 | SV* sv_name |
| 5754 | = newSVpvn_flags(name_start, (int)(RExC_parse - name_start), |
| 5755 | SVs_TEMP | (UTF ? SVf_UTF8 : 0)); |
| 5756 | if ( flags == REG_RSN_RETURN_NAME) |
| 5757 | return sv_name; |
| 5758 | else if (flags==REG_RSN_RETURN_DATA) { |
| 5759 | HE *he_str = NULL; |
| 5760 | SV *sv_dat = NULL; |
| 5761 | if ( ! sv_name ) /* should not happen*/ |
| 5762 | Perl_croak(aTHX_ "panic: no svname in reg_scan_name"); |
| 5763 | if (RExC_paren_names) |
| 5764 | he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 ); |
| 5765 | if ( he_str ) |
| 5766 | sv_dat = HeVAL(he_str); |
| 5767 | if ( ! sv_dat ) |
| 5768 | vFAIL("Reference to nonexistent named group"); |
| 5769 | return sv_dat; |
| 5770 | } |
| 5771 | else { |
| 5772 | Perl_croak(aTHX_ "panic: bad flag in reg_scan_name"); |
| 5773 | } |
| 5774 | /* NOT REACHED */ |
| 5775 | } |
| 5776 | return NULL; |
| 5777 | } |
| 5778 | |
| 5779 | #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \ |
| 5780 | int rem=(int)(RExC_end - RExC_parse); \ |
| 5781 | int cut; \ |
| 5782 | int num; \ |
| 5783 | int iscut=0; \ |
| 5784 | if (rem>10) { \ |
| 5785 | rem=10; \ |
| 5786 | iscut=1; \ |
| 5787 | } \ |
| 5788 | cut=10-rem; \ |
| 5789 | if (RExC_lastparse!=RExC_parse) \ |
| 5790 | PerlIO_printf(Perl_debug_log," >%.*s%-*s", \ |
| 5791 | rem, RExC_parse, \ |
| 5792 | cut + 4, \ |
| 5793 | iscut ? "..." : "<" \ |
| 5794 | ); \ |
| 5795 | else \ |
| 5796 | PerlIO_printf(Perl_debug_log,"%16s",""); \ |
| 5797 | \ |
| 5798 | if (SIZE_ONLY) \ |
| 5799 | num = RExC_size + 1; \ |
| 5800 | else \ |
| 5801 | num=REG_NODE_NUM(RExC_emit); \ |
| 5802 | if (RExC_lastnum!=num) \ |
| 5803 | PerlIO_printf(Perl_debug_log,"|%4d",num); \ |
| 5804 | else \ |
| 5805 | PerlIO_printf(Perl_debug_log,"|%4s",""); \ |
| 5806 | PerlIO_printf(Perl_debug_log,"|%*s%-4s", \ |
| 5807 | (int)((depth*2)), "", \ |
| 5808 | (funcname) \ |
| 5809 | ); \ |
| 5810 | RExC_lastnum=num; \ |
| 5811 | RExC_lastparse=RExC_parse; \ |
| 5812 | }) |
| 5813 | |
| 5814 | |
| 5815 | |
| 5816 | #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \ |
| 5817 | DEBUG_PARSE_MSG((funcname)); \ |
| 5818 | PerlIO_printf(Perl_debug_log,"%4s","\n"); \ |
| 5819 | }) |
| 5820 | #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \ |
| 5821 | DEBUG_PARSE_MSG((funcname)); \ |
| 5822 | PerlIO_printf(Perl_debug_log,fmt "\n",args); \ |
| 5823 | }) |
| 5824 | |
| 5825 | /* This section of code defines the inversion list object and its methods. The |
| 5826 | * interfaces are highly subject to change, so as much as possible is static to |
| 5827 | * this file. An inversion list is here implemented as a malloc'd C array with |
| 5828 | * some added info. More will be coming when functionality is added later. |
| 5829 | * |
| 5830 | * It is currently implemented as an HV to the outside world, but is actually |
| 5831 | * an SV pointing to an array of UVs that the SV thinks are bytes. This allows |
| 5832 | * us to have an array of UV whose memory management is automatically handled |
| 5833 | * by the existing facilities for SV's. |
| 5834 | * |
| 5835 | * Some of the methods should always be private to the implementation, and some |
| 5836 | * should eventually be made public */ |
| 5837 | |
| 5838 | #define INVLIST_INITIAL_LEN 10 |
| 5839 | |
| 5840 | PERL_STATIC_INLINE UV* |
| 5841 | S_invlist_array(pTHX_ HV* const invlist) |
| 5842 | { |
| 5843 | /* Returns the pointer to the inversion list's array. Every time the |
| 5844 | * length changes, this needs to be called in case malloc or realloc moved |
| 5845 | * it */ |
| 5846 | |
| 5847 | PERL_ARGS_ASSERT_INVLIST_ARRAY; |
| 5848 | |
| 5849 | return (UV *) SvPVX(invlist); |
| 5850 | } |
| 5851 | |
| 5852 | PERL_STATIC_INLINE UV |
| 5853 | S_invlist_len(pTHX_ HV* const invlist) |
| 5854 | { |
| 5855 | /* Returns the current number of elements in the inversion list's array */ |
| 5856 | |
| 5857 | PERL_ARGS_ASSERT_INVLIST_LEN; |
| 5858 | |
| 5859 | return SvCUR(invlist) / sizeof(UV); |
| 5860 | } |
| 5861 | |
| 5862 | PERL_STATIC_INLINE UV |
| 5863 | S_invlist_max(pTHX_ HV* const invlist) |
| 5864 | { |
| 5865 | /* Returns the maximum number of elements storable in the inversion list's |
| 5866 | * array, without having to realloc() */ |
| 5867 | |
| 5868 | PERL_ARGS_ASSERT_INVLIST_MAX; |
| 5869 | |
| 5870 | return SvLEN(invlist) / sizeof(UV); |
| 5871 | } |
| 5872 | |
| 5873 | PERL_STATIC_INLINE void |
| 5874 | S_invlist_set_len(pTHX_ HV* const invlist, const UV len) |
| 5875 | { |
| 5876 | /* Sets the current number of elements stored in the inversion list */ |
| 5877 | |
| 5878 | PERL_ARGS_ASSERT_INVLIST_SET_LEN; |
| 5879 | |
| 5880 | SvCUR_set(invlist, len * sizeof(UV)); |
| 5881 | } |
| 5882 | |
| 5883 | PERL_STATIC_INLINE void |
| 5884 | S_invlist_set_max(pTHX_ HV* const invlist, const UV max) |
| 5885 | { |
| 5886 | |
| 5887 | /* Sets the maximum number of elements storable in the inversion list |
| 5888 | * without having to realloc() */ |
| 5889 | |
| 5890 | PERL_ARGS_ASSERT_INVLIST_SET_MAX; |
| 5891 | |
| 5892 | if (max < invlist_len(invlist)) { |
| 5893 | Perl_croak(aTHX_ "panic: Can't make max size '%"UVuf"' less than current length %"UVuf" in inversion list", invlist_max(invlist), invlist_len(invlist)); |
| 5894 | } |
| 5895 | |
| 5896 | SvLEN_set(invlist, max * sizeof(UV)); |
| 5897 | } |
| 5898 | |
| 5899 | #ifndef PERL_IN_XSUB_RE |
| 5900 | HV* |
| 5901 | Perl__new_invlist(pTHX_ IV initial_size) |
| 5902 | { |
| 5903 | |
| 5904 | /* Return a pointer to a newly constructed inversion list, with enough |
| 5905 | * space to store 'initial_size' elements. If that number is negative, a |
| 5906 | * system default is used instead */ |
| 5907 | |
| 5908 | if (initial_size < 0) { |
| 5909 | initial_size = INVLIST_INITIAL_LEN; |
| 5910 | } |
| 5911 | |
| 5912 | /* Allocate the initial space */ |
| 5913 | return (HV *) newSV(initial_size * sizeof(UV)); |
| 5914 | } |
| 5915 | #endif |
| 5916 | |
| 5917 | PERL_STATIC_INLINE void |
| 5918 | S_invlist_destroy(pTHX_ HV* const invlist) |
| 5919 | { |
| 5920 | /* Inversion list destructor */ |
| 5921 | |
| 5922 | PERL_ARGS_ASSERT_INVLIST_DESTROY; |
| 5923 | |
| 5924 | SvREFCNT_dec(invlist); |
| 5925 | } |
| 5926 | |
| 5927 | STATIC void |
| 5928 | S_invlist_extend(pTHX_ HV* const invlist, const UV new_max) |
| 5929 | { |
| 5930 | /* Grow the maximum size of an inversion list */ |
| 5931 | |
| 5932 | PERL_ARGS_ASSERT_INVLIST_EXTEND; |
| 5933 | |
| 5934 | SvGROW((SV *)invlist, new_max * sizeof(UV)); |
| 5935 | } |
| 5936 | |
| 5937 | PERL_STATIC_INLINE void |
| 5938 | S_invlist_trim(pTHX_ HV* const invlist) |
| 5939 | { |
| 5940 | PERL_ARGS_ASSERT_INVLIST_TRIM; |
| 5941 | |
| 5942 | /* Change the length of the inversion list to how many entries it currently |
| 5943 | * has */ |
| 5944 | |
| 5945 | SvPV_shrink_to_cur((SV *) invlist); |
| 5946 | } |
| 5947 | |
| 5948 | /* An element is in an inversion list iff its index is even numbered: 0, 2, 4, |
| 5949 | * etc */ |
| 5950 | |
| 5951 | #define ELEMENT_IN_INVLIST_SET(i) (! ((i) & 1)) |
| 5952 | #define PREV_ELEMENT_IN_INVLIST_SET(i) ! ELEMENT_IN_INVLIST_SET(i) |
| 5953 | |
| 5954 | #ifndef PERL_IN_XSUB_RE |
| 5955 | void |
| 5956 | Perl__append_range_to_invlist(pTHX_ HV* const invlist, const UV start, const UV end) |
| 5957 | { |
| 5958 | /* Subject to change or removal. Append the range from 'start' to 'end' at |
| 5959 | * the end of the inversion list. The range must be above any existing |
| 5960 | * ones. */ |
| 5961 | |
| 5962 | UV* array = invlist_array(invlist); |
| 5963 | UV max = invlist_max(invlist); |
| 5964 | UV len = invlist_len(invlist); |
| 5965 | |
| 5966 | PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST; |
| 5967 | |
| 5968 | if (len > 0) { |
| 5969 | |
| 5970 | /* Here, the existing list is non-empty. The current max entry in the |
| 5971 | * list is generally the first value not in the set, except when the |
| 5972 | * set extends to the end of permissible values, in which case it is |
| 5973 | * the first entry in that final set, and so this call is an attempt to |
| 5974 | * append out-of-order */ |
| 5975 | |
| 5976 | UV final_element = len - 1; |
| 5977 | if (array[final_element] > start |
| 5978 | || ELEMENT_IN_INVLIST_SET(final_element)) |
| 5979 | { |
| 5980 | Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list"); |
| 5981 | } |
| 5982 | |
| 5983 | /* Here, it is a legal append. If the new range begins with the first |
| 5984 | * value not in the set, it is extending the set, so the new first |
| 5985 | * value not in the set is one greater than the newly extended range. |
| 5986 | * */ |
| 5987 | if (array[final_element] == start) { |
| 5988 | if (end != UV_MAX) { |
| 5989 | array[final_element] = end + 1; |
| 5990 | } |
| 5991 | else { |
| 5992 | /* But if the end is the maximum representable on the machine, |
| 5993 | * just let the range that this would extend have no end */ |
| 5994 | invlist_set_len(invlist, len - 1); |
| 5995 | } |
| 5996 | return; |
| 5997 | } |
| 5998 | } |
| 5999 | |
| 6000 | /* Here the new range doesn't extend any existing set. Add it */ |
| 6001 | |
| 6002 | len += 2; /* Includes an element each for the start and end of range */ |
| 6003 | |
| 6004 | /* If overflows the existing space, extend, which may cause the array to be |
| 6005 | * moved */ |
| 6006 | if (max < len) { |
| 6007 | invlist_extend(invlist, len); |
| 6008 | array = invlist_array(invlist); |
| 6009 | } |
| 6010 | |
| 6011 | invlist_set_len(invlist, len); |
| 6012 | |
| 6013 | /* The next item on the list starts the range, the one after that is |
| 6014 | * one past the new range. */ |
| 6015 | array[len - 2] = start; |
| 6016 | if (end != UV_MAX) { |
| 6017 | array[len - 1] = end + 1; |
| 6018 | } |
| 6019 | else { |
| 6020 | /* But if the end is the maximum representable on the machine, just let |
| 6021 | * the range have no end */ |
| 6022 | invlist_set_len(invlist, len - 1); |
| 6023 | } |
| 6024 | } |
| 6025 | #endif |
| 6026 | |
| 6027 | STATIC HV* |
| 6028 | S_invlist_union(pTHX_ HV* const a, HV* const b) |
| 6029 | { |
| 6030 | /* Return a new inversion list which is the union of two inversion lists. |
| 6031 | * The basis for this comes from "Unicode Demystified" Chapter 13 by |
| 6032 | * Richard Gillam, published by Addison-Wesley, and explained at some |
| 6033 | * length there. The preface says to incorporate its examples into your |
| 6034 | * code at your own risk. |
| 6035 | * |
| 6036 | * The algorithm is like a merge sort. |
| 6037 | * |
| 6038 | * XXX A potential performance improvement is to keep track as we go along |
| 6039 | * if only one of the inputs contributes to the result, meaning the other |
| 6040 | * is a subset of that one. In that case, we can skip the final copy and |
| 6041 | * return the larger of the input lists */ |
| 6042 | |
| 6043 | UV* array_a = invlist_array(a); /* a's array */ |
| 6044 | UV* array_b = invlist_array(b); |
| 6045 | UV len_a = invlist_len(a); /* length of a's array */ |
| 6046 | UV len_b = invlist_len(b); |
| 6047 | |
| 6048 | HV* u; /* the resulting union */ |
| 6049 | UV* array_u; |
| 6050 | UV len_u; |
| 6051 | |
| 6052 | UV i_a = 0; /* current index into a's array */ |
| 6053 | UV i_b = 0; |
| 6054 | UV i_u = 0; |
| 6055 | |
| 6056 | /* running count, as explained in the algorithm source book; items are |
| 6057 | * stopped accumulating and are output when the count changes to/from 0. |
| 6058 | * The count is incremented when we start a range that's in the set, and |
| 6059 | * decremented when we start a range that's not in the set. So its range |
| 6060 | * is 0 to 2. Only when the count is zero is something not in the set. |
| 6061 | */ |
| 6062 | UV count = 0; |
| 6063 | |
| 6064 | PERL_ARGS_ASSERT_INVLIST_UNION; |
| 6065 | |
| 6066 | /* Size the union for the worst case: that the sets are completely |
| 6067 | * disjoint */ |
| 6068 | u = _new_invlist(len_a + len_b); |
| 6069 | array_u = invlist_array(u); |
| 6070 | |
| 6071 | /* Go through each list item by item, stopping when exhausted one of |
| 6072 | * them */ |
| 6073 | while (i_a < len_a && i_b < len_b) { |
| 6074 | UV cp; /* The element to potentially add to the union's array */ |
| 6075 | bool cp_in_set; /* is it in the the input list's set or not */ |
| 6076 | |
| 6077 | /* We need to take one or the other of the two inputs for the union. |
| 6078 | * Since we are merging two sorted lists, we take the smaller of the |
| 6079 | * next items. In case of a tie, we take the one that is in its set |
| 6080 | * first. If we took one not in the set first, it would decrement the |
| 6081 | * count, possibly to 0 which would cause it to be output as ending the |
| 6082 | * range, and the next time through we would take the same number, and |
| 6083 | * output it again as beginning the next range. By doing it the |
| 6084 | * opposite way, there is no possibility that the count will be |
| 6085 | * momentarily decremented to 0, and thus the two adjoining ranges will |
| 6086 | * be seamlessly merged. (In a tie and both are in the set or both not |
| 6087 | * in the set, it doesn't matter which we take first.) */ |
| 6088 | if (array_a[i_a] < array_b[i_b] |
| 6089 | || (array_a[i_a] == array_b[i_b] && ELEMENT_IN_INVLIST_SET(i_a))) |
| 6090 | { |
| 6091 | cp_in_set = ELEMENT_IN_INVLIST_SET(i_a); |
| 6092 | cp= array_a[i_a++]; |
| 6093 | } |
| 6094 | else { |
| 6095 | cp_in_set = ELEMENT_IN_INVLIST_SET(i_b); |
| 6096 | cp= array_b[i_b++]; |
| 6097 | } |
| 6098 | |
| 6099 | /* Here, have chosen which of the two inputs to look at. Only output |
| 6100 | * if the running count changes to/from 0, which marks the |
| 6101 | * beginning/end of a range in that's in the set */ |
| 6102 | if (cp_in_set) { |
| 6103 | if (count == 0) { |
| 6104 | array_u[i_u++] = cp; |
| 6105 | } |
| 6106 | count++; |
| 6107 | } |
| 6108 | else { |
| 6109 | count--; |
| 6110 | if (count == 0) { |
| 6111 | array_u[i_u++] = cp; |
| 6112 | } |
| 6113 | } |
| 6114 | } |
| 6115 | |
| 6116 | /* Here, we are finished going through at least one of the lists, which |
| 6117 | * means there is something remaining in at most one. We check if the list |
| 6118 | * that hasn't been exhausted is positioned such that we are in the middle |
| 6119 | * of a range in its set or not. (i_a and i_b point to the element beyond |
| 6120 | * the one we care about.) If in the set, we decrement 'count'; if 0, there |
| 6121 | * is potentially more to output. |
| 6122 | * There are four cases: |
| 6123 | * 1) Both weren't in their sets, count is 0, and remains 0. What's left |
| 6124 | * in the union is entirely from the non-exhausted set. |
| 6125 | * 2) Both were in their sets, count is 2. Nothing further should |
| 6126 | * be output, as everything that remains will be in the exhausted |
| 6127 | * list's set, hence in the union; decrementing to 1 but not 0 insures |
| 6128 | * that |
| 6129 | * 3) the exhausted was in its set, non-exhausted isn't, count is 1. |
| 6130 | * Nothing further should be output because the union includes |
| 6131 | * everything from the exhausted set. Not decrementing ensures that. |
| 6132 | * 4) the exhausted wasn't in its set, non-exhausted is, count is 1; |
| 6133 | * decrementing to 0 insures that we look at the remainder of the |
| 6134 | * non-exhausted set */ |
| 6135 | if ((i_a != len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a)) |
| 6136 | || (i_b != len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b))) |
| 6137 | { |
| 6138 | count--; |
| 6139 | } |
| 6140 | |
| 6141 | /* The final length is what we've output so far, plus what else is about to |
| 6142 | * be output. (If 'count' is non-zero, then the input list we exhausted |
| 6143 | * has everything remaining up to the machine's limit in its set, and hence |
| 6144 | * in the union, so there will be no further output. */ |
| 6145 | len_u = i_u; |
| 6146 | if (count == 0) { |
| 6147 | /* At most one of the subexpressions will be non-zero */ |
| 6148 | len_u += (len_a - i_a) + (len_b - i_b); |
| 6149 | } |
| 6150 | |
| 6151 | /* Set result to final length, which can change the pointer to array_u, so |
| 6152 | * re-find it */ |
| 6153 | if (len_u != invlist_len(u)) { |
| 6154 | invlist_set_len(u, len_u); |
| 6155 | invlist_trim(u); |
| 6156 | array_u = invlist_array(u); |
| 6157 | } |
| 6158 | |
| 6159 | /* When 'count' is 0, the list that was exhausted (if one was shorter than |
| 6160 | * the other) ended with everything above it not in its set. That means |
| 6161 | * that the remaining part of the union is precisely the same as the |
| 6162 | * non-exhausted list, so can just copy it unchanged. (If both list were |
| 6163 | * exhausted at the same time, then the operations below will be both 0.) |
| 6164 | */ |
| 6165 | if (count == 0) { |
| 6166 | IV copy_count; /* At most one will have a non-zero copy count */ |
| 6167 | if ((copy_count = len_a - i_a) > 0) { |
| 6168 | Copy(array_a + i_a, array_u + i_u, copy_count, UV); |
| 6169 | } |
| 6170 | else if ((copy_count = len_b - i_b) > 0) { |
| 6171 | Copy(array_b + i_b, array_u + i_u, copy_count, UV); |
| 6172 | } |
| 6173 | } |
| 6174 | |
| 6175 | return u; |
| 6176 | } |
| 6177 | |
| 6178 | STATIC HV* |
| 6179 | S_invlist_intersection(pTHX_ HV* const a, HV* const b) |
| 6180 | { |
| 6181 | /* Return the intersection of two inversion lists. The basis for this |
| 6182 | * comes from "Unicode Demystified" Chapter 13 by Richard Gillam, published |
| 6183 | * by Addison-Wesley, and explained at some length there. The preface says |
| 6184 | * to incorporate its examples into your code at your own risk. In fact, |
| 6185 | * it had bugs |
| 6186 | * |
| 6187 | * The algorithm is like a merge sort, and is essentially the same as the |
| 6188 | * union above |
| 6189 | */ |
| 6190 | |
| 6191 | UV* array_a = invlist_array(a); /* a's array */ |
| 6192 | UV* array_b = invlist_array(b); |
| 6193 | UV len_a = invlist_len(a); /* length of a's array */ |
| 6194 | UV len_b = invlist_len(b); |
| 6195 | |
| 6196 | HV* r; /* the resulting intersection */ |
| 6197 | UV* array_r; |
| 6198 | UV len_r; |
| 6199 | |
| 6200 | UV i_a = 0; /* current index into a's array */ |
| 6201 | UV i_b = 0; |
| 6202 | UV i_r = 0; |
| 6203 | |
| 6204 | /* running count, as explained in the algorithm source book; items are |
| 6205 | * stopped accumulating and are output when the count changes to/from 2. |
| 6206 | * The count is incremented when we start a range that's in the set, and |
| 6207 | * decremented when we start a range that's not in the set. So its range |
| 6208 | * is 0 to 2. Only when the count is 2 is something in the intersection. |
| 6209 | */ |
| 6210 | UV count = 0; |
| 6211 | |
| 6212 | PERL_ARGS_ASSERT_INVLIST_INTERSECTION; |
| 6213 | |
| 6214 | /* Size the intersection for the worst case: that the intersection ends up |
| 6215 | * fragmenting everything to be completely disjoint */ |
| 6216 | r= _new_invlist(len_a + len_b); |
| 6217 | array_r = invlist_array(r); |
| 6218 | |
| 6219 | /* Go through each list item by item, stopping when exhausted one of |
| 6220 | * them */ |
| 6221 | while (i_a < len_a && i_b < len_b) { |
| 6222 | UV cp; /* The element to potentially add to the intersection's |
| 6223 | array */ |
| 6224 | bool cp_in_set; /* Is it in the input list's set or not */ |
| 6225 | |
| 6226 | /* We need to take one or the other of the two inputs for the |
| 6227 | * intersection. Since we are merging two sorted lists, we take the |
| 6228 | * smaller of the next items. In case of a tie, we take the one that |
| 6229 | * is not in its set first (a difference from the union algorithm). If |
| 6230 | * we took one in the set first, it would increment the count, possibly |
| 6231 | * to 2 which would cause it to be output as starting a range in the |
| 6232 | * intersection, and the next time through we would take that same |
| 6233 | * number, and output it again as ending the set. By doing it the |
| 6234 | * opposite of this, there is no possibility that the count will be |
| 6235 | * momentarily incremented to 2. (In a tie and both are in the set or |
| 6236 | * both not in the set, it doesn't matter which we take first.) */ |
| 6237 | if (array_a[i_a] < array_b[i_b] |
| 6238 | || (array_a[i_a] == array_b[i_b] && ! ELEMENT_IN_INVLIST_SET(i_a))) |
| 6239 | { |
| 6240 | cp_in_set = ELEMENT_IN_INVLIST_SET(i_a); |
| 6241 | cp= array_a[i_a++]; |
| 6242 | } |
| 6243 | else { |
| 6244 | cp_in_set = ELEMENT_IN_INVLIST_SET(i_b); |
| 6245 | cp= array_b[i_b++]; |
| 6246 | } |
| 6247 | |
| 6248 | /* Here, have chosen which of the two inputs to look at. Only output |
| 6249 | * if the running count changes to/from 2, which marks the |
| 6250 | * beginning/end of a range that's in the intersection */ |
| 6251 | if (cp_in_set) { |
| 6252 | count++; |
| 6253 | if (count == 2) { |
| 6254 | array_r[i_r++] = cp; |
| 6255 | } |
| 6256 | } |
| 6257 | else { |
| 6258 | if (count == 2) { |
| 6259 | array_r[i_r++] = cp; |
| 6260 | } |
| 6261 | count--; |
| 6262 | } |
| 6263 | } |
| 6264 | |
| 6265 | /* Here, we are finished going through at least one of the lists, which |
| 6266 | * means there is something remaining in at most one. We check if the list |
| 6267 | * that has been exhausted is positioned such that we are in the middle |
| 6268 | * of a range in its set or not. (i_a and i_b point to elements 1 beyond |
| 6269 | * the ones we care about.) There are four cases: |
| 6270 | * 1) Both weren't in their sets, count is 0, and remains 0. There's |
| 6271 | * nothing left in the intersection. |
| 6272 | * 2) Both were in their sets, count is 2 and perhaps is incremented to |
| 6273 | * above 2. What should be output is exactly that which is in the |
| 6274 | * non-exhausted set, as everything it has is also in the intersection |
| 6275 | * set, and everything it doesn't have can't be in the intersection |
| 6276 | * 3) The exhausted was in its set, non-exhausted isn't, count is 1, and |
| 6277 | * gets incremented to 2. Like the previous case, the intersection is |
| 6278 | * everything that remains in the non-exhausted set. |
| 6279 | * 4) the exhausted wasn't in its set, non-exhausted is, count is 1, and |
| 6280 | * remains 1. And the intersection has nothing more. */ |
| 6281 | if ((i_a == len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a)) |
| 6282 | || (i_b == len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b))) |
| 6283 | { |
| 6284 | count++; |
| 6285 | } |
| 6286 | |
| 6287 | /* The final length is what we've output so far plus what else is in the |
| 6288 | * intersection. At most one of the subexpressions below will be non-zero */ |
| 6289 | len_r = i_r; |
| 6290 | if (count >= 2) { |
| 6291 | len_r += (len_a - i_a) + (len_b - i_b); |
| 6292 | } |
| 6293 | |
| 6294 | /* Set result to final length, which can change the pointer to array_r, so |
| 6295 | * re-find it */ |
| 6296 | if (len_r != invlist_len(r)) { |
| 6297 | invlist_set_len(r, len_r); |
| 6298 | invlist_trim(r); |
| 6299 | array_r = invlist_array(r); |
| 6300 | } |
| 6301 | |
| 6302 | /* Finish outputting any remaining */ |
| 6303 | if (count >= 2) { /* At most one will have a non-zero copy count */ |
| 6304 | IV copy_count; |
| 6305 | if ((copy_count = len_a - i_a) > 0) { |
| 6306 | Copy(array_a + i_a, array_r + i_r, copy_count, UV); |
| 6307 | } |
| 6308 | else if ((copy_count = len_b - i_b) > 0) { |
| 6309 | Copy(array_b + i_b, array_r + i_r, copy_count, UV); |
| 6310 | } |
| 6311 | } |
| 6312 | |
| 6313 | return r; |
| 6314 | } |
| 6315 | |
| 6316 | STATIC HV* |
| 6317 | S_add_range_to_invlist(pTHX_ HV* invlist, const UV start, const UV end) |
| 6318 | { |
| 6319 | /* Add the range from 'start' to 'end' inclusive to the inversion list's |
| 6320 | * set. A pointer to the inversion list is returned. This may actually be |
| 6321 | * a new list, in which case the passed in one has been destroyed. The |
| 6322 | * passed in inversion list can be NULL, in which case a new one is created |
| 6323 | * with just the one range in it */ |
| 6324 | |
| 6325 | HV* range_invlist; |
| 6326 | HV* added_invlist; |
| 6327 | UV len; |
| 6328 | |
| 6329 | if (invlist == NULL) { |
| 6330 | invlist = _new_invlist(2); |
| 6331 | len = 0; |
| 6332 | } |
| 6333 | else { |
| 6334 | len = invlist_len(invlist); |
| 6335 | } |
| 6336 | |
| 6337 | /* If comes after the final entry, can just append it to the end */ |
| 6338 | if (len == 0 |
| 6339 | || start >= invlist_array(invlist) |
| 6340 | [invlist_len(invlist) - 1]) |
| 6341 | { |
| 6342 | _append_range_to_invlist(invlist, start, end); |
| 6343 | return invlist; |
| 6344 | } |
| 6345 | |
| 6346 | /* Here, can't just append things, create and return a new inversion list |
| 6347 | * which is the union of this range and the existing inversion list */ |
| 6348 | range_invlist = _new_invlist(2); |
| 6349 | _append_range_to_invlist(range_invlist, start, end); |
| 6350 | |
| 6351 | added_invlist = invlist_union(invlist, range_invlist); |
| 6352 | |
| 6353 | /* The passed in list can be freed, as well as our temporary */ |
| 6354 | invlist_destroy(range_invlist); |
| 6355 | if (invlist != added_invlist) { |
| 6356 | invlist_destroy(invlist); |
| 6357 | } |
| 6358 | |
| 6359 | return added_invlist; |
| 6360 | } |
| 6361 | |
| 6362 | PERL_STATIC_INLINE HV* |
| 6363 | S_add_cp_to_invlist(pTHX_ HV* invlist, const UV cp) { |
| 6364 | return add_range_to_invlist(invlist, cp, cp); |
| 6365 | } |
| 6366 | |
| 6367 | /* End of inversion list object */ |
| 6368 | |
| 6369 | /* |
| 6370 | - reg - regular expression, i.e. main body or parenthesized thing |
| 6371 | * |
| 6372 | * Caller must absorb opening parenthesis. |
| 6373 | * |
| 6374 | * Combining parenthesis handling with the base level of regular expression |
| 6375 | * is a trifle forced, but the need to tie the tails of the branches to what |
| 6376 | * follows makes it hard to avoid. |
| 6377 | */ |
| 6378 | #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1) |
| 6379 | #ifdef DEBUGGING |
| 6380 | #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1) |
| 6381 | #else |
| 6382 | #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1) |
| 6383 | #endif |
| 6384 | |
| 6385 | STATIC regnode * |
| 6386 | S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth) |
| 6387 | /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */ |
| 6388 | { |
| 6389 | dVAR; |
| 6390 | register regnode *ret; /* Will be the head of the group. */ |
| 6391 | register regnode *br; |
| 6392 | register regnode *lastbr; |
| 6393 | register regnode *ender = NULL; |
| 6394 | register I32 parno = 0; |
| 6395 | I32 flags; |
| 6396 | U32 oregflags = RExC_flags; |
| 6397 | bool have_branch = 0; |
| 6398 | bool is_open = 0; |
| 6399 | I32 freeze_paren = 0; |
| 6400 | I32 after_freeze = 0; |
| 6401 | |
| 6402 | /* for (?g), (?gc), and (?o) warnings; warning |
| 6403 | about (?c) will warn about (?g) -- japhy */ |
| 6404 | |
| 6405 | #define WASTED_O 0x01 |
| 6406 | #define WASTED_G 0x02 |
| 6407 | #define WASTED_C 0x04 |
| 6408 | #define WASTED_GC (0x02|0x04) |
| 6409 | I32 wastedflags = 0x00; |
| 6410 | |
| 6411 | char * parse_start = RExC_parse; /* MJD */ |
| 6412 | char * const oregcomp_parse = RExC_parse; |
| 6413 | |
| 6414 | GET_RE_DEBUG_FLAGS_DECL; |
| 6415 | |
| 6416 | PERL_ARGS_ASSERT_REG; |
| 6417 | DEBUG_PARSE("reg "); |
| 6418 | |
| 6419 | *flagp = 0; /* Tentatively. */ |
| 6420 | |
| 6421 | |
| 6422 | /* Make an OPEN node, if parenthesized. */ |
| 6423 | if (paren) { |
| 6424 | if ( *RExC_parse == '*') { /* (*VERB:ARG) */ |
| 6425 | char *start_verb = RExC_parse; |
| 6426 | STRLEN verb_len = 0; |
| 6427 | char *start_arg = NULL; |
| 6428 | unsigned char op = 0; |
| 6429 | int argok = 1; |
| 6430 | int internal_argval = 0; /* internal_argval is only useful if !argok */ |
| 6431 | while ( *RExC_parse && *RExC_parse != ')' ) { |
| 6432 | if ( *RExC_parse == ':' ) { |
| 6433 | start_arg = RExC_parse + 1; |
| 6434 | break; |
| 6435 | } |
| 6436 | RExC_parse++; |
| 6437 | } |
| 6438 | ++start_verb; |
| 6439 | verb_len = RExC_parse - start_verb; |
| 6440 | if ( start_arg ) { |
| 6441 | RExC_parse++; |
| 6442 | while ( *RExC_parse && *RExC_parse != ')' ) |
| 6443 | RExC_parse++; |
| 6444 | if ( *RExC_parse != ')' ) |
| 6445 | vFAIL("Unterminated verb pattern argument"); |
| 6446 | if ( RExC_parse == start_arg ) |
| 6447 | start_arg = NULL; |
| 6448 | } else { |
| 6449 | if ( *RExC_parse != ')' ) |
| 6450 | vFAIL("Unterminated verb pattern"); |
| 6451 | } |
| 6452 | |
| 6453 | switch ( *start_verb ) { |
| 6454 | case 'A': /* (*ACCEPT) */ |
| 6455 | if ( memEQs(start_verb,verb_len,"ACCEPT") ) { |
| 6456 | op = ACCEPT; |
| 6457 | internal_argval = RExC_nestroot; |
| 6458 | } |
| 6459 | break; |
| 6460 | case 'C': /* (*COMMIT) */ |
| 6461 | if ( memEQs(start_verb,verb_len,"COMMIT") ) |
| 6462 | op = COMMIT; |
| 6463 | break; |
| 6464 | case 'F': /* (*FAIL) */ |
| 6465 | if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) { |
| 6466 | op = OPFAIL; |
| 6467 | argok = 0; |
| 6468 | } |
| 6469 | break; |
| 6470 | case ':': /* (*:NAME) */ |
| 6471 | case 'M': /* (*MARK:NAME) */ |
| 6472 | if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) { |
| 6473 | op = MARKPOINT; |
| 6474 | argok = -1; |
| 6475 | } |
| 6476 | break; |
| 6477 | case 'P': /* (*PRUNE) */ |
| 6478 | if ( memEQs(start_verb,verb_len,"PRUNE") ) |
| 6479 | op = PRUNE; |
| 6480 | break; |
| 6481 | case 'S': /* (*SKIP) */ |
| 6482 | if ( memEQs(start_verb,verb_len,"SKIP") ) |
| 6483 | op = SKIP; |
| 6484 | break; |
| 6485 | case 'T': /* (*THEN) */ |
| 6486 | /* [19:06] <TimToady> :: is then */ |
| 6487 | if ( memEQs(start_verb,verb_len,"THEN") ) { |
| 6488 | op = CUTGROUP; |
| 6489 | RExC_seen |= REG_SEEN_CUTGROUP; |
| 6490 | } |
| 6491 | break; |
| 6492 | } |
| 6493 | if ( ! op ) { |
| 6494 | RExC_parse++; |
| 6495 | vFAIL3("Unknown verb pattern '%.*s'", |
| 6496 | verb_len, start_verb); |
| 6497 | } |
| 6498 | if ( argok ) { |
| 6499 | if ( start_arg && internal_argval ) { |
| 6500 | vFAIL3("Verb pattern '%.*s' may not have an argument", |
| 6501 | verb_len, start_verb); |
| 6502 | } else if ( argok < 0 && !start_arg ) { |
| 6503 | vFAIL3("Verb pattern '%.*s' has a mandatory argument", |
| 6504 | verb_len, start_verb); |
| 6505 | } else { |
| 6506 | ret = reganode(pRExC_state, op, internal_argval); |
| 6507 | if ( ! internal_argval && ! SIZE_ONLY ) { |
| 6508 | if (start_arg) { |
| 6509 | SV *sv = newSVpvn( start_arg, RExC_parse - start_arg); |
| 6510 | ARG(ret) = add_data( pRExC_state, 1, "S" ); |
| 6511 | RExC_rxi->data->data[ARG(ret)]=(void*)sv; |
| 6512 | ret->flags = 0; |
| 6513 | } else { |
| 6514 | ret->flags = 1; |
| 6515 | } |
| 6516 | } |
| 6517 | } |
| 6518 | if (!internal_argval) |
| 6519 | RExC_seen |= REG_SEEN_VERBARG; |
| 6520 | } else if ( start_arg ) { |
| 6521 | vFAIL3("Verb pattern '%.*s' may not have an argument", |
| 6522 | verb_len, start_verb); |
| 6523 | } else { |
| 6524 | ret = reg_node(pRExC_state, op); |
| 6525 | } |
| 6526 | nextchar(pRExC_state); |
| 6527 | return ret; |
| 6528 | } else |
| 6529 | if (*RExC_parse == '?') { /* (?...) */ |
| 6530 | bool is_logical = 0; |
| 6531 | const char * const seqstart = RExC_parse; |
| 6532 | bool has_use_defaults = FALSE; |
| 6533 | |
| 6534 | RExC_parse++; |
| 6535 | paren = *RExC_parse++; |
| 6536 | ret = NULL; /* For look-ahead/behind. */ |
| 6537 | switch (paren) { |
| 6538 | |
| 6539 | case 'P': /* (?P...) variants for those used to PCRE/Python */ |
| 6540 | paren = *RExC_parse++; |
| 6541 | if ( paren == '<') /* (?P<...>) named capture */ |
| 6542 | goto named_capture; |
| 6543 | else if (paren == '>') { /* (?P>name) named recursion */ |
| 6544 | goto named_recursion; |
| 6545 | } |
| 6546 | else if (paren == '=') { /* (?P=...) named backref */ |
| 6547 | /* this pretty much dupes the code for \k<NAME> in regatom(), if |
| 6548 | you change this make sure you change that */ |
| 6549 | char* name_start = RExC_parse; |
| 6550 | U32 num = 0; |
| 6551 | SV *sv_dat = reg_scan_name(pRExC_state, |
| 6552 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); |
| 6553 | if (RExC_parse == name_start || *RExC_parse != ')') |
| 6554 | vFAIL2("Sequence %.3s... not terminated",parse_start); |
| 6555 | |
| 6556 | if (!SIZE_ONLY) { |
| 6557 | num = add_data( pRExC_state, 1, "S" ); |
| 6558 | RExC_rxi->data->data[num]=(void*)sv_dat; |
| 6559 | SvREFCNT_inc_simple_void(sv_dat); |
| 6560 | } |
| 6561 | RExC_sawback = 1; |
| 6562 | ret = reganode(pRExC_state, |
| 6563 | ((! FOLD) |
| 6564 | ? NREF |
| 6565 | : (MORE_ASCII_RESTRICTED) |
| 6566 | ? NREFFA |
| 6567 | : (AT_LEAST_UNI_SEMANTICS) |
| 6568 | ? NREFFU |
| 6569 | : (LOC) |
| 6570 | ? NREFFL |
| 6571 | : NREFF), |
| 6572 | num); |
| 6573 | *flagp |= HASWIDTH; |
| 6574 | |
| 6575 | Set_Node_Offset(ret, parse_start+1); |
| 6576 | Set_Node_Cur_Length(ret); /* MJD */ |
| 6577 | |
| 6578 | nextchar(pRExC_state); |
| 6579 | return ret; |
| 6580 | } |
| 6581 | RExC_parse++; |
| 6582 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); |
| 6583 | /*NOTREACHED*/ |
| 6584 | case '<': /* (?<...) */ |
| 6585 | if (*RExC_parse == '!') |
| 6586 | paren = ','; |
| 6587 | else if (*RExC_parse != '=') |
| 6588 | named_capture: |
| 6589 | { /* (?<...>) */ |
| 6590 | char *name_start; |
| 6591 | SV *svname; |
| 6592 | paren= '>'; |
| 6593 | case '\'': /* (?'...') */ |
| 6594 | name_start= RExC_parse; |
| 6595 | svname = reg_scan_name(pRExC_state, |
| 6596 | SIZE_ONLY ? /* reverse test from the others */ |
| 6597 | REG_RSN_RETURN_NAME : |
| 6598 | REG_RSN_RETURN_NULL); |
| 6599 | if (RExC_parse == name_start) { |
| 6600 | RExC_parse++; |
| 6601 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); |
| 6602 | /*NOTREACHED*/ |
| 6603 | } |
| 6604 | if (*RExC_parse != paren) |
| 6605 | vFAIL2("Sequence (?%c... not terminated", |
| 6606 | paren=='>' ? '<' : paren); |
| 6607 | if (SIZE_ONLY) { |
| 6608 | HE *he_str; |
| 6609 | SV *sv_dat = NULL; |
| 6610 | if (!svname) /* shouldn't happen */ |
| 6611 | Perl_croak(aTHX_ |
| 6612 | "panic: reg_scan_name returned NULL"); |
| 6613 | if (!RExC_paren_names) { |
| 6614 | RExC_paren_names= newHV(); |
| 6615 | sv_2mortal(MUTABLE_SV(RExC_paren_names)); |
| 6616 | #ifdef DEBUGGING |
| 6617 | RExC_paren_name_list= newAV(); |
| 6618 | sv_2mortal(MUTABLE_SV(RExC_paren_name_list)); |
| 6619 | #endif |
| 6620 | } |
| 6621 | he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 ); |
| 6622 | if ( he_str ) |
| 6623 | sv_dat = HeVAL(he_str); |
| 6624 | if ( ! sv_dat ) { |
| 6625 | /* croak baby croak */ |
| 6626 | Perl_croak(aTHX_ |
| 6627 | "panic: paren_name hash element allocation failed"); |
| 6628 | } else if ( SvPOK(sv_dat) ) { |
| 6629 | /* (?|...) can mean we have dupes so scan to check |
| 6630 | its already been stored. Maybe a flag indicating |
| 6631 | we are inside such a construct would be useful, |
| 6632 | but the arrays are likely to be quite small, so |
| 6633 | for now we punt -- dmq */ |
| 6634 | IV count = SvIV(sv_dat); |
| 6635 | I32 *pv = (I32*)SvPVX(sv_dat); |
| 6636 | IV i; |
| 6637 | for ( i = 0 ; i < count ; i++ ) { |
| 6638 | if ( pv[i] == RExC_npar ) { |
| 6639 | count = 0; |
| 6640 | break; |
| 6641 | } |
| 6642 | } |
| 6643 | if ( count ) { |
| 6644 | pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1); |
| 6645 | SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32)); |
| 6646 | pv[count] = RExC_npar; |
| 6647 | SvIV_set(sv_dat, SvIVX(sv_dat) + 1); |
| 6648 | } |
| 6649 | } else { |
| 6650 | (void)SvUPGRADE(sv_dat,SVt_PVNV); |
| 6651 | sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32)); |
| 6652 | SvIOK_on(sv_dat); |
| 6653 | SvIV_set(sv_dat, 1); |
| 6654 | } |
| 6655 | #ifdef DEBUGGING |
| 6656 | /* Yes this does cause a memory leak in debugging Perls */ |
| 6657 | if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname))) |
| 6658 | SvREFCNT_dec(svname); |
| 6659 | #endif |
| 6660 | |
| 6661 | /*sv_dump(sv_dat);*/ |
| 6662 | } |
| 6663 | nextchar(pRExC_state); |
| 6664 | paren = 1; |
| 6665 | goto capturing_parens; |
| 6666 | } |
| 6667 | RExC_seen |= REG_SEEN_LOOKBEHIND; |
| 6668 | RExC_in_lookbehind++; |
| 6669 | RExC_parse++; |
| 6670 | case '=': /* (?=...) */ |
| 6671 | RExC_seen_zerolen++; |
| 6672 | break; |
| 6673 | case '!': /* (?!...) */ |
| 6674 | RExC_seen_zerolen++; |
| 6675 | if (*RExC_parse == ')') { |
| 6676 | ret=reg_node(pRExC_state, OPFAIL); |
| 6677 | nextchar(pRExC_state); |
| 6678 | return ret; |
| 6679 | } |
| 6680 | break; |
| 6681 | case '|': /* (?|...) */ |
| 6682 | /* branch reset, behave like a (?:...) except that |
| 6683 | buffers in alternations share the same numbers */ |
| 6684 | paren = ':'; |
| 6685 | after_freeze = freeze_paren = RExC_npar; |
| 6686 | break; |
| 6687 | case ':': /* (?:...) */ |
| 6688 | case '>': /* (?>...) */ |
| 6689 | break; |
| 6690 | case '$': /* (?$...) */ |
| 6691 | case '@': /* (?@...) */ |
| 6692 | vFAIL2("Sequence (?%c...) not implemented", (int)paren); |
| 6693 | break; |
| 6694 | case '#': /* (?#...) */ |
| 6695 | while (*RExC_parse && *RExC_parse != ')') |
| 6696 | RExC_parse++; |
| 6697 | if (*RExC_parse != ')') |
| 6698 | FAIL("Sequence (?#... not terminated"); |
| 6699 | nextchar(pRExC_state); |
| 6700 | *flagp = TRYAGAIN; |
| 6701 | return NULL; |
| 6702 | case '0' : /* (?0) */ |
| 6703 | case 'R' : /* (?R) */ |
| 6704 | if (*RExC_parse != ')') |
| 6705 | FAIL("Sequence (?R) not terminated"); |
| 6706 | ret = reg_node(pRExC_state, GOSTART); |
| 6707 | *flagp |= POSTPONED; |
| 6708 | nextchar(pRExC_state); |
| 6709 | return ret; |
| 6710 | /*notreached*/ |
| 6711 | { /* named and numeric backreferences */ |
| 6712 | I32 num; |
| 6713 | case '&': /* (?&NAME) */ |
| 6714 | parse_start = RExC_parse - 1; |
| 6715 | named_recursion: |
| 6716 | { |
| 6717 | SV *sv_dat = reg_scan_name(pRExC_state, |
| 6718 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); |
| 6719 | num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0; |
| 6720 | } |
| 6721 | goto gen_recurse_regop; |
| 6722 | /* NOT REACHED */ |
| 6723 | case '+': |
| 6724 | if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) { |
| 6725 | RExC_parse++; |
| 6726 | vFAIL("Illegal pattern"); |
| 6727 | } |
| 6728 | goto parse_recursion; |
| 6729 | /* NOT REACHED*/ |
| 6730 | case '-': /* (?-1) */ |
| 6731 | if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) { |
| 6732 | RExC_parse--; /* rewind to let it be handled later */ |
| 6733 | goto parse_flags; |
| 6734 | } |
| 6735 | /*FALLTHROUGH */ |
| 6736 | case '1': case '2': case '3': case '4': /* (?1) */ |
| 6737 | case '5': case '6': case '7': case '8': case '9': |
| 6738 | RExC_parse--; |
| 6739 | parse_recursion: |
| 6740 | num = atoi(RExC_parse); |
| 6741 | parse_start = RExC_parse - 1; /* MJD */ |
| 6742 | if (*RExC_parse == '-') |
| 6743 | RExC_parse++; |
| 6744 | while (isDIGIT(*RExC_parse)) |
| 6745 | RExC_parse++; |
| 6746 | if (*RExC_parse!=')') |
| 6747 | vFAIL("Expecting close bracket"); |
| 6748 | |
| 6749 | gen_recurse_regop: |
| 6750 | if ( paren == '-' ) { |
| 6751 | /* |
| 6752 | Diagram of capture buffer numbering. |
| 6753 | Top line is the normal capture buffer numbers |
| 6754 | Bottom line is the negative indexing as from |
| 6755 | the X (the (?-2)) |
| 6756 | |
| 6757 | + 1 2 3 4 5 X 6 7 |
| 6758 | /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/ |
| 6759 | - 5 4 3 2 1 X x x |
| 6760 | |
| 6761 | */ |
| 6762 | num = RExC_npar + num; |
| 6763 | if (num < 1) { |
| 6764 | RExC_parse++; |
| 6765 | vFAIL("Reference to nonexistent group"); |
| 6766 | } |
| 6767 | } else if ( paren == '+' ) { |
| 6768 | num = RExC_npar + num - 1; |
| 6769 | } |
| 6770 | |
| 6771 | ret = reganode(pRExC_state, GOSUB, num); |
| 6772 | if (!SIZE_ONLY) { |
| 6773 | if (num > (I32)RExC_rx->nparens) { |
| 6774 | RExC_parse++; |
| 6775 | vFAIL("Reference to nonexistent group"); |
| 6776 | } |
| 6777 | ARG2L_SET( ret, RExC_recurse_count++); |
| 6778 | RExC_emit++; |
| 6779 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, |
| 6780 | "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret))); |
| 6781 | } else { |
| 6782 | RExC_size++; |
| 6783 | } |
| 6784 | RExC_seen |= REG_SEEN_RECURSE; |
| 6785 | Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */ |
| 6786 | Set_Node_Offset(ret, parse_start); /* MJD */ |
| 6787 | |
| 6788 | *flagp |= POSTPONED; |
| 6789 | nextchar(pRExC_state); |
| 6790 | return ret; |
| 6791 | } /* named and numeric backreferences */ |
| 6792 | /* NOT REACHED */ |
| 6793 | |
| 6794 | case '?': /* (??...) */ |
| 6795 | is_logical = 1; |
| 6796 | if (*RExC_parse != '{') { |
| 6797 | RExC_parse++; |
| 6798 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); |
| 6799 | /*NOTREACHED*/ |
| 6800 | } |
| 6801 | *flagp |= POSTPONED; |
| 6802 | paren = *RExC_parse++; |
| 6803 | /* FALL THROUGH */ |
| 6804 | case '{': /* (?{...}) */ |
| 6805 | { |
| 6806 | I32 count = 1; |
| 6807 | U32 n = 0; |
| 6808 | char c; |
| 6809 | char *s = RExC_parse; |
| 6810 | |
| 6811 | RExC_seen_zerolen++; |
| 6812 | RExC_seen |= REG_SEEN_EVAL; |
| 6813 | while (count && (c = *RExC_parse)) { |
| 6814 | if (c == '\\') { |
| 6815 | if (RExC_parse[1]) |
| 6816 | RExC_parse++; |
| 6817 | } |
| 6818 | else if (c == '{') |
| 6819 | count++; |
| 6820 | else if (c == '}') |
| 6821 | count--; |
| 6822 | RExC_parse++; |
| 6823 | } |
| 6824 | if (*RExC_parse != ')') { |
| 6825 | RExC_parse = s; |
| 6826 | vFAIL("Sequence (?{...}) not terminated or not {}-balanced"); |
| 6827 | } |
| 6828 | if (!SIZE_ONLY) { |
| 6829 | PAD *pad; |
| 6830 | OP_4tree *sop, *rop; |
| 6831 | SV * const sv = newSVpvn(s, RExC_parse - 1 - s); |
| 6832 | |
| 6833 | ENTER; |
| 6834 | Perl_save_re_context(aTHX); |
| 6835 | rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad); |
| 6836 | sop->op_private |= OPpREFCOUNTED; |
| 6837 | /* re_dup will OpREFCNT_inc */ |
| 6838 | OpREFCNT_set(sop, 1); |
| 6839 | LEAVE; |
| 6840 | |
| 6841 | n = add_data(pRExC_state, 3, "nop"); |
| 6842 | RExC_rxi->data->data[n] = (void*)rop; |
| 6843 | RExC_rxi->data->data[n+1] = (void*)sop; |
| 6844 | RExC_rxi->data->data[n+2] = (void*)pad; |
| 6845 | SvREFCNT_dec(sv); |
| 6846 | } |
| 6847 | else { /* First pass */ |
| 6848 | if (PL_reginterp_cnt < ++RExC_seen_evals |
| 6849 | && IN_PERL_RUNTIME) |
| 6850 | /* No compiled RE interpolated, has runtime |
| 6851 | components ===> unsafe. */ |
| 6852 | FAIL("Eval-group not allowed at runtime, use re 'eval'"); |
| 6853 | if (PL_tainting && PL_tainted) |
| 6854 | FAIL("Eval-group in insecure regular expression"); |
| 6855 | #if PERL_VERSION > 8 |
| 6856 | if (IN_PERL_COMPILETIME) |
| 6857 | PL_cv_has_eval = 1; |
| 6858 | #endif |
| 6859 | } |
| 6860 | |
| 6861 | nextchar(pRExC_state); |
| 6862 | if (is_logical) { |
| 6863 | ret = reg_node(pRExC_state, LOGICAL); |
| 6864 | if (!SIZE_ONLY) |
| 6865 | ret->flags = 2; |
| 6866 | REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n)); |
| 6867 | /* deal with the length of this later - MJD */ |
| 6868 | return ret; |
| 6869 | } |
| 6870 | ret = reganode(pRExC_state, EVAL, n); |
| 6871 | Set_Node_Length(ret, RExC_parse - parse_start + 1); |
| 6872 | Set_Node_Offset(ret, parse_start); |
| 6873 | return ret; |
| 6874 | } |
| 6875 | case '(': /* (?(?{...})...) and (?(?=...)...) */ |
| 6876 | { |
| 6877 | int is_define= 0; |
| 6878 | if (RExC_parse[0] == '?') { /* (?(?...)) */ |
| 6879 | if (RExC_parse[1] == '=' || RExC_parse[1] == '!' |
| 6880 | || RExC_parse[1] == '<' |
| 6881 | || RExC_parse[1] == '{') { /* Lookahead or eval. */ |
| 6882 | I32 flag; |
| 6883 | |
| 6884 | ret = reg_node(pRExC_state, LOGICAL); |
| 6885 | if (!SIZE_ONLY) |
| 6886 | ret->flags = 1; |
| 6887 | REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1)); |
| 6888 | goto insert_if; |
| 6889 | } |
| 6890 | } |
| 6891 | else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */ |
| 6892 | || RExC_parse[0] == '\'' ) /* (?('NAME')...) */ |
| 6893 | { |
| 6894 | char ch = RExC_parse[0] == '<' ? '>' : '\''; |
| 6895 | char *name_start= RExC_parse++; |
| 6896 | U32 num = 0; |
| 6897 | SV *sv_dat=reg_scan_name(pRExC_state, |
| 6898 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); |
| 6899 | if (RExC_parse == name_start || *RExC_parse != ch) |
| 6900 | vFAIL2("Sequence (?(%c... not terminated", |
| 6901 | (ch == '>' ? '<' : ch)); |
| 6902 | RExC_parse++; |
| 6903 | if (!SIZE_ONLY) { |
| 6904 | num = add_data( pRExC_state, 1, "S" ); |
| 6905 | RExC_rxi->data->data[num]=(void*)sv_dat; |
| 6906 | SvREFCNT_inc_simple_void(sv_dat); |
| 6907 | } |
| 6908 | ret = reganode(pRExC_state,NGROUPP,num); |
| 6909 | goto insert_if_check_paren; |
| 6910 | } |
| 6911 | else if (RExC_parse[0] == 'D' && |
| 6912 | RExC_parse[1] == 'E' && |
| 6913 | RExC_parse[2] == 'F' && |
| 6914 | RExC_parse[3] == 'I' && |
| 6915 | RExC_parse[4] == 'N' && |
| 6916 | RExC_parse[5] == 'E') |
| 6917 | { |
| 6918 | ret = reganode(pRExC_state,DEFINEP,0); |
| 6919 | RExC_parse +=6 ; |
| 6920 | is_define = 1; |
| 6921 | goto insert_if_check_paren; |
| 6922 | } |
| 6923 | else if (RExC_parse[0] == 'R') { |
| 6924 | RExC_parse++; |
| 6925 | parno = 0; |
| 6926 | if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) { |
| 6927 | parno = atoi(RExC_parse++); |
| 6928 | while (isDIGIT(*RExC_parse)) |
| 6929 | RExC_parse++; |
| 6930 | } else if (RExC_parse[0] == '&') { |
| 6931 | SV *sv_dat; |
| 6932 | RExC_parse++; |
| 6933 | sv_dat = reg_scan_name(pRExC_state, |
| 6934 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); |
| 6935 | parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0; |
| 6936 | } |
| 6937 | ret = reganode(pRExC_state,INSUBP,parno); |
| 6938 | goto insert_if_check_paren; |
| 6939 | } |
| 6940 | else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) { |
| 6941 | /* (?(1)...) */ |
| 6942 | char c; |
| 6943 | parno = atoi(RExC_parse++); |
| 6944 | |
| 6945 | while (isDIGIT(*RExC_parse)) |
| 6946 | RExC_parse++; |
| 6947 | ret = reganode(pRExC_state, GROUPP, parno); |
| 6948 | |
| 6949 | insert_if_check_paren: |
| 6950 | if ((c = *nextchar(pRExC_state)) != ')') |
| 6951 | vFAIL("Switch condition not recognized"); |
| 6952 | insert_if: |
| 6953 | REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0)); |
| 6954 | br = regbranch(pRExC_state, &flags, 1,depth+1); |
| 6955 | if (br == NULL) |
| 6956 | br = reganode(pRExC_state, LONGJMP, 0); |
| 6957 | else |
| 6958 | REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0)); |
| 6959 | c = *nextchar(pRExC_state); |
| 6960 | if (flags&HASWIDTH) |
| 6961 | *flagp |= HASWIDTH; |
| 6962 | if (c == '|') { |
| 6963 | if (is_define) |
| 6964 | vFAIL("(?(DEFINE)....) does not allow branches"); |
| 6965 | lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */ |
| 6966 | regbranch(pRExC_state, &flags, 1,depth+1); |
| 6967 | REGTAIL(pRExC_state, ret, lastbr); |
| 6968 | if (flags&HASWIDTH) |
| 6969 | *flagp |= HASWIDTH; |
| 6970 | c = *nextchar(pRExC_state); |
| 6971 | } |
| 6972 | else |
| 6973 | lastbr = NULL; |
| 6974 | if (c != ')') |
| 6975 | vFAIL("Switch (?(condition)... contains too many branches"); |
| 6976 | ender = reg_node(pRExC_state, TAIL); |
| 6977 | REGTAIL(pRExC_state, br, ender); |
| 6978 | if (lastbr) { |
| 6979 | REGTAIL(pRExC_state, lastbr, ender); |
| 6980 | REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); |
| 6981 | } |
| 6982 | else |
| 6983 | REGTAIL(pRExC_state, ret, ender); |
| 6984 | RExC_size++; /* XXX WHY do we need this?!! |
| 6985 | For large programs it seems to be required |
| 6986 | but I can't figure out why. -- dmq*/ |
| 6987 | return ret; |
| 6988 | } |
| 6989 | else { |
| 6990 | vFAIL2("Unknown switch condition (?(%.2s", RExC_parse); |
| 6991 | } |
| 6992 | } |
| 6993 | case 0: |
| 6994 | RExC_parse--; /* for vFAIL to print correctly */ |
| 6995 | vFAIL("Sequence (? incomplete"); |
| 6996 | break; |
| 6997 | case DEFAULT_PAT_MOD: /* Use default flags with the exceptions |
| 6998 | that follow */ |
| 6999 | has_use_defaults = TRUE; |
| 7000 | STD_PMMOD_FLAGS_CLEAR(&RExC_flags); |
| 7001 | set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics) |
| 7002 | ? REGEX_UNICODE_CHARSET |
| 7003 | : REGEX_DEPENDS_CHARSET); |
| 7004 | goto parse_flags; |
| 7005 | default: |
| 7006 | --RExC_parse; |
| 7007 | parse_flags: /* (?i) */ |
| 7008 | { |
| 7009 | U32 posflags = 0, negflags = 0; |
| 7010 | U32 *flagsp = &posflags; |
| 7011 | char has_charset_modifier = '\0'; |
| 7012 | regex_charset cs = (RExC_utf8 || RExC_uni_semantics) |
| 7013 | ? REGEX_UNICODE_CHARSET |
| 7014 | : REGEX_DEPENDS_CHARSET; |
| 7015 | |
| 7016 | while (*RExC_parse) { |
| 7017 | /* && strchr("iogcmsx", *RExC_parse) */ |
| 7018 | /* (?g), (?gc) and (?o) are useless here |
| 7019 | and must be globally applied -- japhy */ |
| 7020 | switch (*RExC_parse) { |
| 7021 | CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp); |
| 7022 | case LOCALE_PAT_MOD: |
| 7023 | if (has_charset_modifier) { |
| 7024 | goto excess_modifier; |
| 7025 | } |
| 7026 | else if (flagsp == &negflags) { |
| 7027 | goto neg_modifier; |
| 7028 | } |
| 7029 | cs = REGEX_LOCALE_CHARSET; |
| 7030 | has_charset_modifier = LOCALE_PAT_MOD; |
| 7031 | RExC_contains_locale = 1; |
| 7032 | break; |
| 7033 | case UNICODE_PAT_MOD: |
| 7034 | if (has_charset_modifier) { |
| 7035 | goto excess_modifier; |
| 7036 | } |
| 7037 | else if (flagsp == &negflags) { |
| 7038 | goto neg_modifier; |
| 7039 | } |
| 7040 | cs = REGEX_UNICODE_CHARSET; |
| 7041 | has_charset_modifier = UNICODE_PAT_MOD; |
| 7042 | break; |
| 7043 | case ASCII_RESTRICT_PAT_MOD: |
| 7044 | if (flagsp == &negflags) { |
| 7045 | goto neg_modifier; |
| 7046 | } |
| 7047 | if (has_charset_modifier) { |
| 7048 | if (cs != REGEX_ASCII_RESTRICTED_CHARSET) { |
| 7049 | goto excess_modifier; |
| 7050 | } |
| 7051 | /* Doubled modifier implies more restricted */ |
| 7052 | cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET; |
| 7053 | } |
| 7054 | else { |
| 7055 | cs = REGEX_ASCII_RESTRICTED_CHARSET; |
| 7056 | } |
| 7057 | has_charset_modifier = ASCII_RESTRICT_PAT_MOD; |
| 7058 | break; |
| 7059 | case DEPENDS_PAT_MOD: |
| 7060 | if (has_use_defaults) { |
| 7061 | goto fail_modifiers; |
| 7062 | } |
| 7063 | else if (flagsp == &negflags) { |
| 7064 | goto neg_modifier; |
| 7065 | } |
| 7066 | else if (has_charset_modifier) { |
| 7067 | goto excess_modifier; |
| 7068 | } |
| 7069 | |
| 7070 | /* The dual charset means unicode semantics if the |
| 7071 | * pattern (or target, not known until runtime) are |
| 7072 | * utf8, or something in the pattern indicates unicode |
| 7073 | * semantics */ |
| 7074 | cs = (RExC_utf8 || RExC_uni_semantics) |
| 7075 | ? REGEX_UNICODE_CHARSET |
| 7076 | : REGEX_DEPENDS_CHARSET; |
| 7077 | has_charset_modifier = DEPENDS_PAT_MOD; |
| 7078 | break; |
| 7079 | excess_modifier: |
| 7080 | RExC_parse++; |
| 7081 | if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) { |
| 7082 | vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD); |
| 7083 | } |
| 7084 | else if (has_charset_modifier == *(RExC_parse - 1)) { |
| 7085 | vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1)); |
| 7086 | } |
| 7087 | else { |
| 7088 | vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1)); |
| 7089 | } |
| 7090 | /*NOTREACHED*/ |
| 7091 | neg_modifier: |
| 7092 | RExC_parse++; |
| 7093 | vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1)); |
| 7094 | /*NOTREACHED*/ |
| 7095 | case ONCE_PAT_MOD: /* 'o' */ |
| 7096 | case GLOBAL_PAT_MOD: /* 'g' */ |
| 7097 | if (SIZE_ONLY && ckWARN(WARN_REGEXP)) { |
| 7098 | const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G; |
| 7099 | if (! (wastedflags & wflagbit) ) { |
| 7100 | wastedflags |= wflagbit; |
| 7101 | vWARN5( |
| 7102 | RExC_parse + 1, |
| 7103 | "Useless (%s%c) - %suse /%c modifier", |
| 7104 | flagsp == &negflags ? "?-" : "?", |
| 7105 | *RExC_parse, |
| 7106 | flagsp == &negflags ? "don't " : "", |
| 7107 | *RExC_parse |
| 7108 | ); |
| 7109 | } |
| 7110 | } |
| 7111 | break; |
| 7112 | |
| 7113 | case CONTINUE_PAT_MOD: /* 'c' */ |
| 7114 | if (SIZE_ONLY && ckWARN(WARN_REGEXP)) { |
| 7115 | if (! (wastedflags & WASTED_C) ) { |
| 7116 | wastedflags |= WASTED_GC; |
| 7117 | vWARN3( |
| 7118 | RExC_parse + 1, |
| 7119 | "Useless (%sc) - %suse /gc modifier", |
| 7120 | flagsp == &negflags ? "?-" : "?", |
| 7121 | flagsp == &negflags ? "don't " : "" |
| 7122 | ); |
| 7123 | } |
| 7124 | } |
| 7125 | break; |
| 7126 | case KEEPCOPY_PAT_MOD: /* 'p' */ |
| 7127 | if (flagsp == &negflags) { |
| 7128 | if (SIZE_ONLY) |
| 7129 | ckWARNreg(RExC_parse + 1,"Useless use of (?-p)"); |
| 7130 | } else { |
| 7131 | *flagsp |= RXf_PMf_KEEPCOPY; |
| 7132 | } |
| 7133 | break; |
| 7134 | case '-': |
| 7135 | /* A flag is a default iff it is following a minus, so |
| 7136 | * if there is a minus, it means will be trying to |
| 7137 | * re-specify a default which is an error */ |
| 7138 | if (has_use_defaults || flagsp == &negflags) { |
| 7139 | fail_modifiers: |
| 7140 | RExC_parse++; |
| 7141 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); |
| 7142 | /*NOTREACHED*/ |
| 7143 | } |
| 7144 | flagsp = &negflags; |
| 7145 | wastedflags = 0; /* reset so (?g-c) warns twice */ |
| 7146 | break; |
| 7147 | case ':': |
| 7148 | paren = ':'; |
| 7149 | /*FALLTHROUGH*/ |
| 7150 | case ')': |
| 7151 | RExC_flags |= posflags; |
| 7152 | RExC_flags &= ~negflags; |
| 7153 | set_regex_charset(&RExC_flags, cs); |
| 7154 | if (paren != ':') { |
| 7155 | oregflags |= posflags; |
| 7156 | oregflags &= ~negflags; |
| 7157 | set_regex_charset(&oregflags, cs); |
| 7158 | } |
| 7159 | nextchar(pRExC_state); |
| 7160 | if (paren != ':') { |
| 7161 | *flagp = TRYAGAIN; |
| 7162 | return NULL; |
| 7163 | } else { |
| 7164 | ret = NULL; |
| 7165 | goto parse_rest; |
| 7166 | } |
| 7167 | /*NOTREACHED*/ |
| 7168 | default: |
| 7169 | RExC_parse++; |
| 7170 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); |
| 7171 | /*NOTREACHED*/ |
| 7172 | } |
| 7173 | ++RExC_parse; |
| 7174 | } |
| 7175 | }} /* one for the default block, one for the switch */ |
| 7176 | } |
| 7177 | else { /* (...) */ |
| 7178 | capturing_parens: |
| 7179 | parno = RExC_npar; |
| 7180 | RExC_npar++; |
| 7181 | |
| 7182 | ret = reganode(pRExC_state, OPEN, parno); |
| 7183 | if (!SIZE_ONLY ){ |
| 7184 | if (!RExC_nestroot) |
| 7185 | RExC_nestroot = parno; |
| 7186 | if (RExC_seen & REG_SEEN_RECURSE |
| 7187 | && !RExC_open_parens[parno-1]) |
| 7188 | { |
| 7189 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, |
| 7190 | "Setting open paren #%"IVdf" to %d\n", |
| 7191 | (IV)parno, REG_NODE_NUM(ret))); |
| 7192 | RExC_open_parens[parno-1]= ret; |
| 7193 | } |
| 7194 | } |
| 7195 | Set_Node_Length(ret, 1); /* MJD */ |
| 7196 | Set_Node_Offset(ret, RExC_parse); /* MJD */ |
| 7197 | is_open = 1; |
| 7198 | } |
| 7199 | } |
| 7200 | else /* ! paren */ |
| 7201 | ret = NULL; |
| 7202 | |
| 7203 | parse_rest: |
| 7204 | /* Pick up the branches, linking them together. */ |
| 7205 | parse_start = RExC_parse; /* MJD */ |
| 7206 | br = regbranch(pRExC_state, &flags, 1,depth+1); |
| 7207 | |
| 7208 | /* branch_len = (paren != 0); */ |
| 7209 | |
| 7210 | if (br == NULL) |
| 7211 | return(NULL); |
| 7212 | if (*RExC_parse == '|') { |
| 7213 | if (!SIZE_ONLY && RExC_extralen) { |
| 7214 | reginsert(pRExC_state, BRANCHJ, br, depth+1); |
| 7215 | } |
| 7216 | else { /* MJD */ |
| 7217 | reginsert(pRExC_state, BRANCH, br, depth+1); |
| 7218 | Set_Node_Length(br, paren != 0); |
| 7219 | Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start); |
| 7220 | } |
| 7221 | have_branch = 1; |
| 7222 | if (SIZE_ONLY) |
| 7223 | RExC_extralen += 1; /* For BRANCHJ-BRANCH. */ |
| 7224 | } |
| 7225 | else if (paren == ':') { |
| 7226 | *flagp |= flags&SIMPLE; |
| 7227 | } |
| 7228 | if (is_open) { /* Starts with OPEN. */ |
| 7229 | REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */ |
| 7230 | } |
| 7231 | else if (paren != '?') /* Not Conditional */ |
| 7232 | ret = br; |
| 7233 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
| 7234 | lastbr = br; |
| 7235 | while (*RExC_parse == '|') { |
| 7236 | if (!SIZE_ONLY && RExC_extralen) { |
| 7237 | ender = reganode(pRExC_state, LONGJMP,0); |
| 7238 | REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */ |
| 7239 | } |
| 7240 | if (SIZE_ONLY) |
| 7241 | RExC_extralen += 2; /* Account for LONGJMP. */ |
| 7242 | nextchar(pRExC_state); |
| 7243 | if (freeze_paren) { |
| 7244 | if (RExC_npar > after_freeze) |
| 7245 | after_freeze = RExC_npar; |
| 7246 | RExC_npar = freeze_paren; |
| 7247 | } |
| 7248 | br = regbranch(pRExC_state, &flags, 0, depth+1); |
| 7249 | |
| 7250 | if (br == NULL) |
| 7251 | return(NULL); |
| 7252 | REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */ |
| 7253 | lastbr = br; |
| 7254 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
| 7255 | } |
| 7256 | |
| 7257 | if (have_branch || paren != ':') { |
| 7258 | /* Make a closing node, and hook it on the end. */ |
| 7259 | switch (paren) { |
| 7260 | case ':': |
| 7261 | ender = reg_node(pRExC_state, TAIL); |
| 7262 | break; |
| 7263 | case 1: |
| 7264 | ender = reganode(pRExC_state, CLOSE, parno); |
| 7265 | if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) { |
| 7266 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, |
| 7267 | "Setting close paren #%"IVdf" to %d\n", |
| 7268 | (IV)parno, REG_NODE_NUM(ender))); |
| 7269 | RExC_close_parens[parno-1]= ender; |
| 7270 | if (RExC_nestroot == parno) |
| 7271 | RExC_nestroot = 0; |
| 7272 | } |
| 7273 | Set_Node_Offset(ender,RExC_parse+1); /* MJD */ |
| 7274 | Set_Node_Length(ender,1); /* MJD */ |
| 7275 | break; |
| 7276 | case '<': |
| 7277 | case ',': |
| 7278 | case '=': |
| 7279 | case '!': |
| 7280 | *flagp &= ~HASWIDTH; |
| 7281 | /* FALL THROUGH */ |
| 7282 | case '>': |
| 7283 | ender = reg_node(pRExC_state, SUCCEED); |
| 7284 | break; |
| 7285 | case 0: |
| 7286 | ender = reg_node(pRExC_state, END); |
| 7287 | if (!SIZE_ONLY) { |
| 7288 | assert(!RExC_opend); /* there can only be one! */ |
| 7289 | RExC_opend = ender; |
| 7290 | } |
| 7291 | break; |
| 7292 | } |
| 7293 | REGTAIL(pRExC_state, lastbr, ender); |
| 7294 | |
| 7295 | if (have_branch && !SIZE_ONLY) { |
| 7296 | if (depth==1) |
| 7297 | RExC_seen |= REG_TOP_LEVEL_BRANCHES; |
| 7298 | |
| 7299 | /* Hook the tails of the branches to the closing node. */ |
| 7300 | for (br = ret; br; br = regnext(br)) { |
| 7301 | const U8 op = PL_regkind[OP(br)]; |
| 7302 | if (op == BRANCH) { |
| 7303 | REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender); |
| 7304 | } |
| 7305 | else if (op == BRANCHJ) { |
| 7306 | REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender); |
| 7307 | } |
| 7308 | } |
| 7309 | } |
| 7310 | } |
| 7311 | |
| 7312 | { |
| 7313 | const char *p; |
| 7314 | static const char parens[] = "=!<,>"; |
| 7315 | |
| 7316 | if (paren && (p = strchr(parens, paren))) { |
| 7317 | U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH; |
| 7318 | int flag = (p - parens) > 1; |
| 7319 | |
| 7320 | if (paren == '>') |
| 7321 | node = SUSPEND, flag = 0; |
| 7322 | reginsert(pRExC_state, node,ret, depth+1); |
| 7323 | Set_Node_Cur_Length(ret); |
| 7324 | Set_Node_Offset(ret, parse_start + 1); |
| 7325 | ret->flags = flag; |
| 7326 | REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL)); |
| 7327 | } |
| 7328 | } |
| 7329 | |
| 7330 | /* Check for proper termination. */ |
| 7331 | if (paren) { |
| 7332 | RExC_flags = oregflags; |
| 7333 | if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') { |
| 7334 | RExC_parse = oregcomp_parse; |
| 7335 | vFAIL("Unmatched ("); |
| 7336 | } |
| 7337 | } |
| 7338 | else if (!paren && RExC_parse < RExC_end) { |
| 7339 | if (*RExC_parse == ')') { |
| 7340 | RExC_parse++; |
| 7341 | vFAIL("Unmatched )"); |
| 7342 | } |
| 7343 | else |
| 7344 | FAIL("Junk on end of regexp"); /* "Can't happen". */ |
| 7345 | /* NOTREACHED */ |
| 7346 | } |
| 7347 | |
| 7348 | if (RExC_in_lookbehind) { |
| 7349 | RExC_in_lookbehind--; |
| 7350 | } |
| 7351 | if (after_freeze > RExC_npar) |
| 7352 | RExC_npar = after_freeze; |
| 7353 | return(ret); |
| 7354 | } |
| 7355 | |
| 7356 | /* |
| 7357 | - regbranch - one alternative of an | operator |
| 7358 | * |
| 7359 | * Implements the concatenation operator. |
| 7360 | */ |
| 7361 | STATIC regnode * |
| 7362 | S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth) |
| 7363 | { |
| 7364 | dVAR; |
| 7365 | register regnode *ret; |
| 7366 | register regnode *chain = NULL; |
| 7367 | register regnode *latest; |
| 7368 | I32 flags = 0, c = 0; |
| 7369 | GET_RE_DEBUG_FLAGS_DECL; |
| 7370 | |
| 7371 | PERL_ARGS_ASSERT_REGBRANCH; |
| 7372 | |
| 7373 | DEBUG_PARSE("brnc"); |
| 7374 | |
| 7375 | if (first) |
| 7376 | ret = NULL; |
| 7377 | else { |
| 7378 | if (!SIZE_ONLY && RExC_extralen) |
| 7379 | ret = reganode(pRExC_state, BRANCHJ,0); |
| 7380 | else { |
| 7381 | ret = reg_node(pRExC_state, BRANCH); |
| 7382 | Set_Node_Length(ret, 1); |
| 7383 | } |
| 7384 | } |
| 7385 | |
| 7386 | if (!first && SIZE_ONLY) |
| 7387 | RExC_extralen += 1; /* BRANCHJ */ |
| 7388 | |
| 7389 | *flagp = WORST; /* Tentatively. */ |
| 7390 | |
| 7391 | RExC_parse--; |
| 7392 | nextchar(pRExC_state); |
| 7393 | while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') { |
| 7394 | flags &= ~TRYAGAIN; |
| 7395 | latest = regpiece(pRExC_state, &flags,depth+1); |
| 7396 | if (latest == NULL) { |
| 7397 | if (flags & TRYAGAIN) |
| 7398 | continue; |
| 7399 | return(NULL); |
| 7400 | } |
| 7401 | else if (ret == NULL) |
| 7402 | ret = latest; |
| 7403 | *flagp |= flags&(HASWIDTH|POSTPONED); |
| 7404 | if (chain == NULL) /* First piece. */ |
| 7405 | *flagp |= flags&SPSTART; |
| 7406 | else { |
| 7407 | RExC_naughty++; |
| 7408 | REGTAIL(pRExC_state, chain, latest); |
| 7409 | } |
| 7410 | chain = latest; |
| 7411 | c++; |
| 7412 | } |
| 7413 | if (chain == NULL) { /* Loop ran zero times. */ |
| 7414 | chain = reg_node(pRExC_state, NOTHING); |
| 7415 | if (ret == NULL) |
| 7416 | ret = chain; |
| 7417 | } |
| 7418 | if (c == 1) { |
| 7419 | *flagp |= flags&SIMPLE; |
| 7420 | } |
| 7421 | |
| 7422 | return ret; |
| 7423 | } |
| 7424 | |
| 7425 | /* |
| 7426 | - regpiece - something followed by possible [*+?] |
| 7427 | * |
| 7428 | * Note that the branching code sequences used for ? and the general cases |
| 7429 | * of * and + are somewhat optimized: they use the same NOTHING node as |
| 7430 | * both the endmarker for their branch list and the body of the last branch. |
| 7431 | * It might seem that this node could be dispensed with entirely, but the |
| 7432 | * endmarker role is not redundant. |
| 7433 | */ |
| 7434 | STATIC regnode * |
| 7435 | S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
| 7436 | { |
| 7437 | dVAR; |
| 7438 | register regnode *ret; |
| 7439 | register char op; |
| 7440 | register char *next; |
| 7441 | I32 flags; |
| 7442 | const char * const origparse = RExC_parse; |
| 7443 | I32 min; |
| 7444 | I32 max = REG_INFTY; |
| 7445 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 7446 | char *parse_start; |
| 7447 | #endif |
| 7448 | const char *maxpos = NULL; |
| 7449 | GET_RE_DEBUG_FLAGS_DECL; |
| 7450 | |
| 7451 | PERL_ARGS_ASSERT_REGPIECE; |
| 7452 | |
| 7453 | DEBUG_PARSE("piec"); |
| 7454 | |
| 7455 | ret = regatom(pRExC_state, &flags,depth+1); |
| 7456 | if (ret == NULL) { |
| 7457 | if (flags & TRYAGAIN) |
| 7458 | *flagp |= TRYAGAIN; |
| 7459 | return(NULL); |
| 7460 | } |
| 7461 | |
| 7462 | op = *RExC_parse; |
| 7463 | |
| 7464 | if (op == '{' && regcurly(RExC_parse)) { |
| 7465 | maxpos = NULL; |
| 7466 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 7467 | parse_start = RExC_parse; /* MJD */ |
| 7468 | #endif |
| 7469 | next = RExC_parse + 1; |
| 7470 | while (isDIGIT(*next) || *next == ',') { |
| 7471 | if (*next == ',') { |
| 7472 | if (maxpos) |
| 7473 | break; |
| 7474 | else |
| 7475 | maxpos = next; |
| 7476 | } |
| 7477 | next++; |
| 7478 | } |
| 7479 | if (*next == '}') { /* got one */ |
| 7480 | if (!maxpos) |
| 7481 | maxpos = next; |
| 7482 | RExC_parse++; |
| 7483 | min = atoi(RExC_parse); |
| 7484 | if (*maxpos == ',') |
| 7485 | maxpos++; |
| 7486 | else |
| 7487 | maxpos = RExC_parse; |
| 7488 | max = atoi(maxpos); |
| 7489 | if (!max && *maxpos != '0') |
| 7490 | max = REG_INFTY; /* meaning "infinity" */ |
| 7491 | else if (max >= REG_INFTY) |
| 7492 | vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1); |
| 7493 | RExC_parse = next; |
| 7494 | nextchar(pRExC_state); |
| 7495 | |
| 7496 | do_curly: |
| 7497 | if ((flags&SIMPLE)) { |
| 7498 | RExC_naughty += 2 + RExC_naughty / 2; |
| 7499 | reginsert(pRExC_state, CURLY, ret, depth+1); |
| 7500 | Set_Node_Offset(ret, parse_start+1); /* MJD */ |
| 7501 | Set_Node_Cur_Length(ret); |
| 7502 | } |
| 7503 | else { |
| 7504 | regnode * const w = reg_node(pRExC_state, WHILEM); |
| 7505 | |
| 7506 | w->flags = 0; |
| 7507 | REGTAIL(pRExC_state, ret, w); |
| 7508 | if (!SIZE_ONLY && RExC_extralen) { |
| 7509 | reginsert(pRExC_state, LONGJMP,ret, depth+1); |
| 7510 | reginsert(pRExC_state, NOTHING,ret, depth+1); |
| 7511 | NEXT_OFF(ret) = 3; /* Go over LONGJMP. */ |
| 7512 | } |
| 7513 | reginsert(pRExC_state, CURLYX,ret, depth+1); |
| 7514 | /* MJD hk */ |
| 7515 | Set_Node_Offset(ret, parse_start+1); |
| 7516 | Set_Node_Length(ret, |
| 7517 | op == '{' ? (RExC_parse - parse_start) : 1); |
| 7518 | |
| 7519 | if (!SIZE_ONLY && RExC_extralen) |
| 7520 | NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */ |
| 7521 | REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING)); |
| 7522 | if (SIZE_ONLY) |
| 7523 | RExC_whilem_seen++, RExC_extralen += 3; |
| 7524 | RExC_naughty += 4 + RExC_naughty; /* compound interest */ |
| 7525 | } |
| 7526 | ret->flags = 0; |
| 7527 | |
| 7528 | if (min > 0) |
| 7529 | *flagp = WORST; |
| 7530 | if (max > 0) |
| 7531 | *flagp |= HASWIDTH; |
| 7532 | if (max < min) |
| 7533 | vFAIL("Can't do {n,m} with n > m"); |
| 7534 | if (!SIZE_ONLY) { |
| 7535 | ARG1_SET(ret, (U16)min); |
| 7536 | ARG2_SET(ret, (U16)max); |
| 7537 | } |
| 7538 | |
| 7539 | goto nest_check; |
| 7540 | } |
| 7541 | } |
| 7542 | |
| 7543 | if (!ISMULT1(op)) { |
| 7544 | *flagp = flags; |
| 7545 | return(ret); |
| 7546 | } |
| 7547 | |
| 7548 | #if 0 /* Now runtime fix should be reliable. */ |
| 7549 | |
| 7550 | /* if this is reinstated, don't forget to put this back into perldiag: |
| 7551 | |
| 7552 | =item Regexp *+ operand could be empty at {#} in regex m/%s/ |
| 7553 | |
| 7554 | (F) The part of the regexp subject to either the * or + quantifier |
| 7555 | could match an empty string. The {#} shows in the regular |
| 7556 | expression about where the problem was discovered. |
| 7557 | |
| 7558 | */ |
| 7559 | |
| 7560 | if (!(flags&HASWIDTH) && op != '?') |
| 7561 | vFAIL("Regexp *+ operand could be empty"); |
| 7562 | #endif |
| 7563 | |
| 7564 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 7565 | parse_start = RExC_parse; |
| 7566 | #endif |
| 7567 | nextchar(pRExC_state); |
| 7568 | |
| 7569 | *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH); |
| 7570 | |
| 7571 | if (op == '*' && (flags&SIMPLE)) { |
| 7572 | reginsert(pRExC_state, STAR, ret, depth+1); |
| 7573 | ret->flags = 0; |
| 7574 | RExC_naughty += 4; |
| 7575 | } |
| 7576 | else if (op == '*') { |
| 7577 | min = 0; |
| 7578 | goto do_curly; |
| 7579 | } |
| 7580 | else if (op == '+' && (flags&SIMPLE)) { |
| 7581 | reginsert(pRExC_state, PLUS, ret, depth+1); |
| 7582 | ret->flags = 0; |
| 7583 | RExC_naughty += 3; |
| 7584 | } |
| 7585 | else if (op == '+') { |
| 7586 | min = 1; |
| 7587 | goto do_curly; |
| 7588 | } |
| 7589 | else if (op == '?') { |
| 7590 | min = 0; max = 1; |
| 7591 | goto do_curly; |
| 7592 | } |
| 7593 | nest_check: |
| 7594 | if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) { |
| 7595 | ckWARN3reg(RExC_parse, |
| 7596 | "%.*s matches null string many times", |
| 7597 | (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0), |
| 7598 | origparse); |
| 7599 | } |
| 7600 | |
| 7601 | if (RExC_parse < RExC_end && *RExC_parse == '?') { |
| 7602 | nextchar(pRExC_state); |
| 7603 | reginsert(pRExC_state, MINMOD, ret, depth+1); |
| 7604 | REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE); |
| 7605 | } |
| 7606 | #ifndef REG_ALLOW_MINMOD_SUSPEND |
| 7607 | else |
| 7608 | #endif |
| 7609 | if (RExC_parse < RExC_end && *RExC_parse == '+') { |
| 7610 | regnode *ender; |
| 7611 | nextchar(pRExC_state); |
| 7612 | ender = reg_node(pRExC_state, SUCCEED); |
| 7613 | REGTAIL(pRExC_state, ret, ender); |
| 7614 | reginsert(pRExC_state, SUSPEND, ret, depth+1); |
| 7615 | ret->flags = 0; |
| 7616 | ender = reg_node(pRExC_state, TAIL); |
| 7617 | REGTAIL(pRExC_state, ret, ender); |
| 7618 | /*ret= ender;*/ |
| 7619 | } |
| 7620 | |
| 7621 | if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) { |
| 7622 | RExC_parse++; |
| 7623 | vFAIL("Nested quantifiers"); |
| 7624 | } |
| 7625 | |
| 7626 | return(ret); |
| 7627 | } |
| 7628 | |
| 7629 | |
| 7630 | /* reg_namedseq(pRExC_state,UVp, UV depth) |
| 7631 | |
| 7632 | This is expected to be called by a parser routine that has |
| 7633 | recognized '\N' and needs to handle the rest. RExC_parse is |
| 7634 | expected to point at the first char following the N at the time |
| 7635 | of the call. |
| 7636 | |
| 7637 | The \N may be inside (indicated by valuep not being NULL) or outside a |
| 7638 | character class. |
| 7639 | |
| 7640 | \N may begin either a named sequence, or if outside a character class, mean |
| 7641 | to match a non-newline. For non single-quoted regexes, the tokenizer has |
| 7642 | attempted to decide which, and in the case of a named sequence converted it |
| 7643 | into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...}, |
| 7644 | where c1... are the characters in the sequence. For single-quoted regexes, |
| 7645 | the tokenizer passes the \N sequence through unchanged; this code will not |
| 7646 | attempt to determine this nor expand those. The net effect is that if the |
| 7647 | beginning of the passed-in pattern isn't '{U+' or there is no '}', it |
| 7648 | signals that this \N occurrence means to match a non-newline. |
| 7649 | |
| 7650 | Only the \N{U+...} form should occur in a character class, for the same |
| 7651 | reason that '.' inside a character class means to just match a period: it |
| 7652 | just doesn't make sense. |
| 7653 | |
| 7654 | If valuep is non-null then it is assumed that we are parsing inside |
| 7655 | of a charclass definition and the first codepoint in the resolved |
| 7656 | string is returned via *valuep and the routine will return NULL. |
| 7657 | In this mode if a multichar string is returned from the charnames |
| 7658 | handler, a warning will be issued, and only the first char in the |
| 7659 | sequence will be examined. If the string returned is zero length |
| 7660 | then the value of *valuep is undefined and NON-NULL will |
| 7661 | be returned to indicate failure. (This will NOT be a valid pointer |
| 7662 | to a regnode.) |
| 7663 | |
| 7664 | If valuep is null then it is assumed that we are parsing normal text and a |
| 7665 | new EXACT node is inserted into the program containing the resolved string, |
| 7666 | and a pointer to the new node is returned. But if the string is zero length |
| 7667 | a NOTHING node is emitted instead. |
| 7668 | |
| 7669 | On success RExC_parse is set to the char following the endbrace. |
| 7670 | Parsing failures will generate a fatal error via vFAIL(...) |
| 7671 | */ |
| 7672 | STATIC regnode * |
| 7673 | S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp, U32 depth) |
| 7674 | { |
| 7675 | char * endbrace; /* '}' following the name */ |
| 7676 | regnode *ret = NULL; |
| 7677 | char* p; |
| 7678 | |
| 7679 | GET_RE_DEBUG_FLAGS_DECL; |
| 7680 | |
| 7681 | PERL_ARGS_ASSERT_REG_NAMEDSEQ; |
| 7682 | |
| 7683 | GET_RE_DEBUG_FLAGS; |
| 7684 | |
| 7685 | /* The [^\n] meaning of \N ignores spaces and comments under the /x |
| 7686 | * modifier. The other meaning does not */ |
| 7687 | p = (RExC_flags & RXf_PMf_EXTENDED) |
| 7688 | ? regwhite( pRExC_state, RExC_parse ) |
| 7689 | : RExC_parse; |
| 7690 | |
| 7691 | /* Disambiguate between \N meaning a named character versus \N meaning |
| 7692 | * [^\n]. The former is assumed when it can't be the latter. */ |
| 7693 | if (*p != '{' || regcurly(p)) { |
| 7694 | RExC_parse = p; |
| 7695 | if (valuep) { |
| 7696 | /* no bare \N in a charclass */ |
| 7697 | vFAIL("\\N in a character class must be a named character: \\N{...}"); |
| 7698 | } |
| 7699 | nextchar(pRExC_state); |
| 7700 | ret = reg_node(pRExC_state, REG_ANY); |
| 7701 | *flagp |= HASWIDTH|SIMPLE; |
| 7702 | RExC_naughty++; |
| 7703 | RExC_parse--; |
| 7704 | Set_Node_Length(ret, 1); /* MJD */ |
| 7705 | return ret; |
| 7706 | } |
| 7707 | |
| 7708 | /* Here, we have decided it should be a named sequence */ |
| 7709 | |
| 7710 | /* The test above made sure that the next real character is a '{', but |
| 7711 | * under the /x modifier, it could be separated by space (or a comment and |
| 7712 | * \n) and this is not allowed (for consistency with \x{...} and the |
| 7713 | * tokenizer handling of \N{NAME}). */ |
| 7714 | if (*RExC_parse != '{') { |
| 7715 | vFAIL("Missing braces on \\N{}"); |
| 7716 | } |
| 7717 | |
| 7718 | RExC_parse++; /* Skip past the '{' */ |
| 7719 | |
| 7720 | if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */ |
| 7721 | || ! (endbrace == RExC_parse /* nothing between the {} */ |
| 7722 | || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */ |
| 7723 | && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */ |
| 7724 | { |
| 7725 | if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */ |
| 7726 | vFAIL("\\N{NAME} must be resolved by the lexer"); |
| 7727 | } |
| 7728 | |
| 7729 | if (endbrace == RExC_parse) { /* empty: \N{} */ |
| 7730 | if (! valuep) { |
| 7731 | RExC_parse = endbrace + 1; |
| 7732 | return reg_node(pRExC_state,NOTHING); |
| 7733 | } |
| 7734 | |
| 7735 | if (SIZE_ONLY) { |
| 7736 | ckWARNreg(RExC_parse, |
| 7737 | "Ignoring zero length \\N{} in character class" |
| 7738 | ); |
| 7739 | RExC_parse = endbrace + 1; |
| 7740 | } |
| 7741 | *valuep = 0; |
| 7742 | return (regnode *) &RExC_parse; /* Invalid regnode pointer */ |
| 7743 | } |
| 7744 | |
| 7745 | REQUIRE_UTF8; /* named sequences imply Unicode semantics */ |
| 7746 | RExC_parse += 2; /* Skip past the 'U+' */ |
| 7747 | |
| 7748 | if (valuep) { /* In a bracketed char class */ |
| 7749 | /* We only pay attention to the first char of |
| 7750 | multichar strings being returned. I kinda wonder |
| 7751 | if this makes sense as it does change the behaviour |
| 7752 | from earlier versions, OTOH that behaviour was broken |
| 7753 | as well. XXX Solution is to recharacterize as |
| 7754 | [rest-of-class]|multi1|multi2... */ |
| 7755 | |
| 7756 | STRLEN length_of_hex; |
| 7757 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES |
| 7758 | | PERL_SCAN_DISALLOW_PREFIX |
| 7759 | | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0); |
| 7760 | |
| 7761 | char * endchar = RExC_parse + strcspn(RExC_parse, ".}"); |
| 7762 | if (endchar < endbrace) { |
| 7763 | ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class"); |
| 7764 | } |
| 7765 | |
| 7766 | length_of_hex = (STRLEN)(endchar - RExC_parse); |
| 7767 | *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL); |
| 7768 | |
| 7769 | /* The tokenizer should have guaranteed validity, but it's possible to |
| 7770 | * bypass it by using single quoting, so check */ |
| 7771 | if (length_of_hex == 0 |
| 7772 | || length_of_hex != (STRLEN)(endchar - RExC_parse) ) |
| 7773 | { |
| 7774 | RExC_parse += length_of_hex; /* Includes all the valid */ |
| 7775 | RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */ |
| 7776 | ? UTF8SKIP(RExC_parse) |
| 7777 | : 1; |
| 7778 | /* Guard against malformed utf8 */ |
| 7779 | if (RExC_parse >= endchar) RExC_parse = endchar; |
| 7780 | vFAIL("Invalid hexadecimal number in \\N{U+...}"); |
| 7781 | } |
| 7782 | |
| 7783 | RExC_parse = endbrace + 1; |
| 7784 | if (endchar == endbrace) return NULL; |
| 7785 | |
| 7786 | ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */ |
| 7787 | } |
| 7788 | else { /* Not a char class */ |
| 7789 | |
| 7790 | /* What is done here is to convert this to a sub-pattern of the form |
| 7791 | * (?:\x{char1}\x{char2}...) |
| 7792 | * and then call reg recursively. That way, it retains its atomicness, |
| 7793 | * while not having to worry about special handling that some code |
| 7794 | * points may have. toke.c has converted the original Unicode values |
| 7795 | * to native, so that we can just pass on the hex values unchanged. We |
| 7796 | * do have to set a flag to keep recoding from happening in the |
| 7797 | * recursion */ |
| 7798 | |
| 7799 | SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP); |
| 7800 | STRLEN len; |
| 7801 | char *endchar; /* Points to '.' or '}' ending cur char in the input |
| 7802 | stream */ |
| 7803 | char *orig_end = RExC_end; |
| 7804 | |
| 7805 | while (RExC_parse < endbrace) { |
| 7806 | |
| 7807 | /* Code points are separated by dots. If none, there is only one |
| 7808 | * code point, and is terminated by the brace */ |
| 7809 | endchar = RExC_parse + strcspn(RExC_parse, ".}"); |
| 7810 | |
| 7811 | /* Convert to notation the rest of the code understands */ |
| 7812 | sv_catpv(substitute_parse, "\\x{"); |
| 7813 | sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse); |
| 7814 | sv_catpv(substitute_parse, "}"); |
| 7815 | |
| 7816 | /* Point to the beginning of the next character in the sequence. */ |
| 7817 | RExC_parse = endchar + 1; |
| 7818 | } |
| 7819 | sv_catpv(substitute_parse, ")"); |
| 7820 | |
| 7821 | RExC_parse = SvPV(substitute_parse, len); |
| 7822 | |
| 7823 | /* Don't allow empty number */ |
| 7824 | if (len < 8) { |
| 7825 | vFAIL("Invalid hexadecimal number in \\N{U+...}"); |
| 7826 | } |
| 7827 | RExC_end = RExC_parse + len; |
| 7828 | |
| 7829 | /* The values are Unicode, and therefore not subject to recoding */ |
| 7830 | RExC_override_recoding = 1; |
| 7831 | |
| 7832 | ret = reg(pRExC_state, 1, flagp, depth+1); |
| 7833 | |
| 7834 | RExC_parse = endbrace; |
| 7835 | RExC_end = orig_end; |
| 7836 | RExC_override_recoding = 0; |
| 7837 | |
| 7838 | nextchar(pRExC_state); |
| 7839 | } |
| 7840 | |
| 7841 | return ret; |
| 7842 | } |
| 7843 | |
| 7844 | |
| 7845 | /* |
| 7846 | * reg_recode |
| 7847 | * |
| 7848 | * It returns the code point in utf8 for the value in *encp. |
| 7849 | * value: a code value in the source encoding |
| 7850 | * encp: a pointer to an Encode object |
| 7851 | * |
| 7852 | * If the result from Encode is not a single character, |
| 7853 | * it returns U+FFFD (Replacement character) and sets *encp to NULL. |
| 7854 | */ |
| 7855 | STATIC UV |
| 7856 | S_reg_recode(pTHX_ const char value, SV **encp) |
| 7857 | { |
| 7858 | STRLEN numlen = 1; |
| 7859 | SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP); |
| 7860 | const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv); |
| 7861 | const STRLEN newlen = SvCUR(sv); |
| 7862 | UV uv = UNICODE_REPLACEMENT; |
| 7863 | |
| 7864 | PERL_ARGS_ASSERT_REG_RECODE; |
| 7865 | |
| 7866 | if (newlen) |
| 7867 | uv = SvUTF8(sv) |
| 7868 | ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT) |
| 7869 | : *(U8*)s; |
| 7870 | |
| 7871 | if (!newlen || numlen != newlen) { |
| 7872 | uv = UNICODE_REPLACEMENT; |
| 7873 | *encp = NULL; |
| 7874 | } |
| 7875 | return uv; |
| 7876 | } |
| 7877 | |
| 7878 | |
| 7879 | /* |
| 7880 | - regatom - the lowest level |
| 7881 | |
| 7882 | Try to identify anything special at the start of the pattern. If there |
| 7883 | is, then handle it as required. This may involve generating a single regop, |
| 7884 | such as for an assertion; or it may involve recursing, such as to |
| 7885 | handle a () structure. |
| 7886 | |
| 7887 | If the string doesn't start with something special then we gobble up |
| 7888 | as much literal text as we can. |
| 7889 | |
| 7890 | Once we have been able to handle whatever type of thing started the |
| 7891 | sequence, we return. |
| 7892 | |
| 7893 | Note: we have to be careful with escapes, as they can be both literal |
| 7894 | and special, and in the case of \10 and friends can either, depending |
| 7895 | on context. Specifically there are two separate switches for handling |
| 7896 | escape sequences, with the one for handling literal escapes requiring |
| 7897 | a dummy entry for all of the special escapes that are actually handled |
| 7898 | by the other. |
| 7899 | */ |
| 7900 | |
| 7901 | STATIC regnode * |
| 7902 | S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
| 7903 | { |
| 7904 | dVAR; |
| 7905 | register regnode *ret = NULL; |
| 7906 | I32 flags; |
| 7907 | char *parse_start = RExC_parse; |
| 7908 | U8 op; |
| 7909 | GET_RE_DEBUG_FLAGS_DECL; |
| 7910 | DEBUG_PARSE("atom"); |
| 7911 | *flagp = WORST; /* Tentatively. */ |
| 7912 | |
| 7913 | PERL_ARGS_ASSERT_REGATOM; |
| 7914 | |
| 7915 | tryagain: |
| 7916 | switch ((U8)*RExC_parse) { |
| 7917 | case '^': |
| 7918 | RExC_seen_zerolen++; |
| 7919 | nextchar(pRExC_state); |
| 7920 | if (RExC_flags & RXf_PMf_MULTILINE) |
| 7921 | ret = reg_node(pRExC_state, MBOL); |
| 7922 | else if (RExC_flags & RXf_PMf_SINGLELINE) |
| 7923 | ret = reg_node(pRExC_state, SBOL); |
| 7924 | else |
| 7925 | ret = reg_node(pRExC_state, BOL); |
| 7926 | Set_Node_Length(ret, 1); /* MJD */ |
| 7927 | break; |
| 7928 | case '$': |
| 7929 | nextchar(pRExC_state); |
| 7930 | if (*RExC_parse) |
| 7931 | RExC_seen_zerolen++; |
| 7932 | if (RExC_flags & RXf_PMf_MULTILINE) |
| 7933 | ret = reg_node(pRExC_state, MEOL); |
| 7934 | else if (RExC_flags & RXf_PMf_SINGLELINE) |
| 7935 | ret = reg_node(pRExC_state, SEOL); |
| 7936 | else |
| 7937 | ret = reg_node(pRExC_state, EOL); |
| 7938 | Set_Node_Length(ret, 1); /* MJD */ |
| 7939 | break; |
| 7940 | case '.': |
| 7941 | nextchar(pRExC_state); |
| 7942 | if (RExC_flags & RXf_PMf_SINGLELINE) |
| 7943 | ret = reg_node(pRExC_state, SANY); |
| 7944 | else |
| 7945 | ret = reg_node(pRExC_state, REG_ANY); |
| 7946 | *flagp |= HASWIDTH|SIMPLE; |
| 7947 | RExC_naughty++; |
| 7948 | Set_Node_Length(ret, 1); /* MJD */ |
| 7949 | break; |
| 7950 | case '[': |
| 7951 | { |
| 7952 | char * const oregcomp_parse = ++RExC_parse; |
| 7953 | ret = regclass(pRExC_state,depth+1); |
| 7954 | if (*RExC_parse != ']') { |
| 7955 | RExC_parse = oregcomp_parse; |
| 7956 | vFAIL("Unmatched ["); |
| 7957 | } |
| 7958 | nextchar(pRExC_state); |
| 7959 | *flagp |= HASWIDTH|SIMPLE; |
| 7960 | Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */ |
| 7961 | break; |
| 7962 | } |
| 7963 | case '(': |
| 7964 | nextchar(pRExC_state); |
| 7965 | ret = reg(pRExC_state, 1, &flags,depth+1); |
| 7966 | if (ret == NULL) { |
| 7967 | if (flags & TRYAGAIN) { |
| 7968 | if (RExC_parse == RExC_end) { |
| 7969 | /* Make parent create an empty node if needed. */ |
| 7970 | *flagp |= TRYAGAIN; |
| 7971 | return(NULL); |
| 7972 | } |
| 7973 | goto tryagain; |
| 7974 | } |
| 7975 | return(NULL); |
| 7976 | } |
| 7977 | *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED); |
| 7978 | break; |
| 7979 | case '|': |
| 7980 | case ')': |
| 7981 | if (flags & TRYAGAIN) { |
| 7982 | *flagp |= TRYAGAIN; |
| 7983 | return NULL; |
| 7984 | } |
| 7985 | vFAIL("Internal urp"); |
| 7986 | /* Supposed to be caught earlier. */ |
| 7987 | break; |
| 7988 | case '{': |
| 7989 | if (!regcurly(RExC_parse)) { |
| 7990 | RExC_parse++; |
| 7991 | goto defchar; |
| 7992 | } |
| 7993 | /* FALL THROUGH */ |
| 7994 | case '?': |
| 7995 | case '+': |
| 7996 | case '*': |
| 7997 | RExC_parse++; |
| 7998 | vFAIL("Quantifier follows nothing"); |
| 7999 | break; |
| 8000 | case '\\': |
| 8001 | /* Special Escapes |
| 8002 | |
| 8003 | This switch handles escape sequences that resolve to some kind |
| 8004 | of special regop and not to literal text. Escape sequnces that |
| 8005 | resolve to literal text are handled below in the switch marked |
| 8006 | "Literal Escapes". |
| 8007 | |
| 8008 | Every entry in this switch *must* have a corresponding entry |
| 8009 | in the literal escape switch. However, the opposite is not |
| 8010 | required, as the default for this switch is to jump to the |
| 8011 | literal text handling code. |
| 8012 | */ |
| 8013 | switch ((U8)*++RExC_parse) { |
| 8014 | /* Special Escapes */ |
| 8015 | case 'A': |
| 8016 | RExC_seen_zerolen++; |
| 8017 | ret = reg_node(pRExC_state, SBOL); |
| 8018 | *flagp |= SIMPLE; |
| 8019 | goto finish_meta_pat; |
| 8020 | case 'G': |
| 8021 | ret = reg_node(pRExC_state, GPOS); |
| 8022 | RExC_seen |= REG_SEEN_GPOS; |
| 8023 | *flagp |= SIMPLE; |
| 8024 | goto finish_meta_pat; |
| 8025 | case 'K': |
| 8026 | RExC_seen_zerolen++; |
| 8027 | ret = reg_node(pRExC_state, KEEPS); |
| 8028 | *flagp |= SIMPLE; |
| 8029 | /* XXX:dmq : disabling in-place substitution seems to |
| 8030 | * be necessary here to avoid cases of memory corruption, as |
| 8031 | * with: C<$_="x" x 80; s/x\K/y/> -- rgs |
| 8032 | */ |
| 8033 | RExC_seen |= REG_SEEN_LOOKBEHIND; |
| 8034 | goto finish_meta_pat; |
| 8035 | case 'Z': |
| 8036 | ret = reg_node(pRExC_state, SEOL); |
| 8037 | *flagp |= SIMPLE; |
| 8038 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
| 8039 | goto finish_meta_pat; |
| 8040 | case 'z': |
| 8041 | ret = reg_node(pRExC_state, EOS); |
| 8042 | *flagp |= SIMPLE; |
| 8043 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
| 8044 | goto finish_meta_pat; |
| 8045 | case 'C': |
| 8046 | ret = reg_node(pRExC_state, CANY); |
| 8047 | RExC_seen |= REG_SEEN_CANY; |
| 8048 | *flagp |= HASWIDTH|SIMPLE; |
| 8049 | goto finish_meta_pat; |
| 8050 | case 'X': |
| 8051 | ret = reg_node(pRExC_state, CLUMP); |
| 8052 | *flagp |= HASWIDTH; |
| 8053 | goto finish_meta_pat; |
| 8054 | case 'w': |
| 8055 | switch (get_regex_charset(RExC_flags)) { |
| 8056 | case REGEX_LOCALE_CHARSET: |
| 8057 | op = ALNUML; |
| 8058 | break; |
| 8059 | case REGEX_UNICODE_CHARSET: |
| 8060 | op = ALNUMU; |
| 8061 | break; |
| 8062 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8063 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8064 | op = ALNUMA; |
| 8065 | break; |
| 8066 | case REGEX_DEPENDS_CHARSET: |
| 8067 | op = ALNUM; |
| 8068 | break; |
| 8069 | default: |
| 8070 | goto bad_charset; |
| 8071 | } |
| 8072 | ret = reg_node(pRExC_state, op); |
| 8073 | *flagp |= HASWIDTH|SIMPLE; |
| 8074 | goto finish_meta_pat; |
| 8075 | case 'W': |
| 8076 | switch (get_regex_charset(RExC_flags)) { |
| 8077 | case REGEX_LOCALE_CHARSET: |
| 8078 | op = NALNUML; |
| 8079 | break; |
| 8080 | case REGEX_UNICODE_CHARSET: |
| 8081 | op = NALNUMU; |
| 8082 | break; |
| 8083 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8084 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8085 | op = NALNUMA; |
| 8086 | break; |
| 8087 | case REGEX_DEPENDS_CHARSET: |
| 8088 | op = NALNUM; |
| 8089 | break; |
| 8090 | default: |
| 8091 | goto bad_charset; |
| 8092 | } |
| 8093 | ret = reg_node(pRExC_state, op); |
| 8094 | *flagp |= HASWIDTH|SIMPLE; |
| 8095 | goto finish_meta_pat; |
| 8096 | case 'b': |
| 8097 | RExC_seen_zerolen++; |
| 8098 | RExC_seen |= REG_SEEN_LOOKBEHIND; |
| 8099 | switch (get_regex_charset(RExC_flags)) { |
| 8100 | case REGEX_LOCALE_CHARSET: |
| 8101 | op = BOUNDL; |
| 8102 | break; |
| 8103 | case REGEX_UNICODE_CHARSET: |
| 8104 | op = BOUNDU; |
| 8105 | break; |
| 8106 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8107 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8108 | op = BOUNDA; |
| 8109 | break; |
| 8110 | case REGEX_DEPENDS_CHARSET: |
| 8111 | op = BOUND; |
| 8112 | break; |
| 8113 | default: |
| 8114 | goto bad_charset; |
| 8115 | } |
| 8116 | ret = reg_node(pRExC_state, op); |
| 8117 | FLAGS(ret) = get_regex_charset(RExC_flags); |
| 8118 | *flagp |= SIMPLE; |
| 8119 | if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') { |
| 8120 | ckWARNregdep(RExC_parse, "\"\\b{\" is deprecated; use \"\\b\\{\" instead"); |
| 8121 | } |
| 8122 | goto finish_meta_pat; |
| 8123 | case 'B': |
| 8124 | RExC_seen_zerolen++; |
| 8125 | RExC_seen |= REG_SEEN_LOOKBEHIND; |
| 8126 | switch (get_regex_charset(RExC_flags)) { |
| 8127 | case REGEX_LOCALE_CHARSET: |
| 8128 | op = NBOUNDL; |
| 8129 | break; |
| 8130 | case REGEX_UNICODE_CHARSET: |
| 8131 | op = NBOUNDU; |
| 8132 | break; |
| 8133 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8134 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8135 | op = NBOUNDA; |
| 8136 | break; |
| 8137 | case REGEX_DEPENDS_CHARSET: |
| 8138 | op = NBOUND; |
| 8139 | break; |
| 8140 | default: |
| 8141 | goto bad_charset; |
| 8142 | } |
| 8143 | ret = reg_node(pRExC_state, op); |
| 8144 | FLAGS(ret) = get_regex_charset(RExC_flags); |
| 8145 | *flagp |= SIMPLE; |
| 8146 | if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') { |
| 8147 | ckWARNregdep(RExC_parse, "\"\\B{\" is deprecated; use \"\\B\\{\" instead"); |
| 8148 | } |
| 8149 | goto finish_meta_pat; |
| 8150 | case 's': |
| 8151 | switch (get_regex_charset(RExC_flags)) { |
| 8152 | case REGEX_LOCALE_CHARSET: |
| 8153 | op = SPACEL; |
| 8154 | break; |
| 8155 | case REGEX_UNICODE_CHARSET: |
| 8156 | op = SPACEU; |
| 8157 | break; |
| 8158 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8159 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8160 | op = SPACEA; |
| 8161 | break; |
| 8162 | case REGEX_DEPENDS_CHARSET: |
| 8163 | op = SPACE; |
| 8164 | break; |
| 8165 | default: |
| 8166 | goto bad_charset; |
| 8167 | } |
| 8168 | ret = reg_node(pRExC_state, op); |
| 8169 | *flagp |= HASWIDTH|SIMPLE; |
| 8170 | goto finish_meta_pat; |
| 8171 | case 'S': |
| 8172 | switch (get_regex_charset(RExC_flags)) { |
| 8173 | case REGEX_LOCALE_CHARSET: |
| 8174 | op = NSPACEL; |
| 8175 | break; |
| 8176 | case REGEX_UNICODE_CHARSET: |
| 8177 | op = NSPACEU; |
| 8178 | break; |
| 8179 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8180 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8181 | op = NSPACEA; |
| 8182 | break; |
| 8183 | case REGEX_DEPENDS_CHARSET: |
| 8184 | op = NSPACE; |
| 8185 | break; |
| 8186 | default: |
| 8187 | goto bad_charset; |
| 8188 | } |
| 8189 | ret = reg_node(pRExC_state, op); |
| 8190 | *flagp |= HASWIDTH|SIMPLE; |
| 8191 | goto finish_meta_pat; |
| 8192 | case 'd': |
| 8193 | switch (get_regex_charset(RExC_flags)) { |
| 8194 | case REGEX_LOCALE_CHARSET: |
| 8195 | op = DIGITL; |
| 8196 | break; |
| 8197 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8198 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8199 | op = DIGITA; |
| 8200 | break; |
| 8201 | case REGEX_DEPENDS_CHARSET: /* No difference between these */ |
| 8202 | case REGEX_UNICODE_CHARSET: |
| 8203 | op = DIGIT; |
| 8204 | break; |
| 8205 | default: |
| 8206 | goto bad_charset; |
| 8207 | } |
| 8208 | ret = reg_node(pRExC_state, op); |
| 8209 | *flagp |= HASWIDTH|SIMPLE; |
| 8210 | goto finish_meta_pat; |
| 8211 | case 'D': |
| 8212 | switch (get_regex_charset(RExC_flags)) { |
| 8213 | case REGEX_LOCALE_CHARSET: |
| 8214 | op = NDIGITL; |
| 8215 | break; |
| 8216 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 8217 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 8218 | op = NDIGITA; |
| 8219 | break; |
| 8220 | case REGEX_DEPENDS_CHARSET: /* No difference between these */ |
| 8221 | case REGEX_UNICODE_CHARSET: |
| 8222 | op = NDIGIT; |
| 8223 | break; |
| 8224 | default: |
| 8225 | goto bad_charset; |
| 8226 | } |
| 8227 | ret = reg_node(pRExC_state, op); |
| 8228 | *flagp |= HASWIDTH|SIMPLE; |
| 8229 | goto finish_meta_pat; |
| 8230 | case 'R': |
| 8231 | ret = reg_node(pRExC_state, LNBREAK); |
| 8232 | *flagp |= HASWIDTH|SIMPLE; |
| 8233 | goto finish_meta_pat; |
| 8234 | case 'h': |
| 8235 | ret = reg_node(pRExC_state, HORIZWS); |
| 8236 | *flagp |= HASWIDTH|SIMPLE; |
| 8237 | goto finish_meta_pat; |
| 8238 | case 'H': |
| 8239 | ret = reg_node(pRExC_state, NHORIZWS); |
| 8240 | *flagp |= HASWIDTH|SIMPLE; |
| 8241 | goto finish_meta_pat; |
| 8242 | case 'v': |
| 8243 | ret = reg_node(pRExC_state, VERTWS); |
| 8244 | *flagp |= HASWIDTH|SIMPLE; |
| 8245 | goto finish_meta_pat; |
| 8246 | case 'V': |
| 8247 | ret = reg_node(pRExC_state, NVERTWS); |
| 8248 | *flagp |= HASWIDTH|SIMPLE; |
| 8249 | finish_meta_pat: |
| 8250 | nextchar(pRExC_state); |
| 8251 | Set_Node_Length(ret, 2); /* MJD */ |
| 8252 | break; |
| 8253 | case 'p': |
| 8254 | case 'P': |
| 8255 | { |
| 8256 | char* const oldregxend = RExC_end; |
| 8257 | #ifdef DEBUGGING |
| 8258 | char* parse_start = RExC_parse - 2; |
| 8259 | #endif |
| 8260 | |
| 8261 | if (RExC_parse[1] == '{') { |
| 8262 | /* a lovely hack--pretend we saw [\pX] instead */ |
| 8263 | RExC_end = strchr(RExC_parse, '}'); |
| 8264 | if (!RExC_end) { |
| 8265 | const U8 c = (U8)*RExC_parse; |
| 8266 | RExC_parse += 2; |
| 8267 | RExC_end = oldregxend; |
| 8268 | vFAIL2("Missing right brace on \\%c{}", c); |
| 8269 | } |
| 8270 | RExC_end++; |
| 8271 | } |
| 8272 | else { |
| 8273 | RExC_end = RExC_parse + 2; |
| 8274 | if (RExC_end > oldregxend) |
| 8275 | RExC_end = oldregxend; |
| 8276 | } |
| 8277 | RExC_parse--; |
| 8278 | |
| 8279 | ret = regclass(pRExC_state,depth+1); |
| 8280 | |
| 8281 | RExC_end = oldregxend; |
| 8282 | RExC_parse--; |
| 8283 | |
| 8284 | Set_Node_Offset(ret, parse_start + 2); |
| 8285 | Set_Node_Cur_Length(ret); |
| 8286 | nextchar(pRExC_state); |
| 8287 | *flagp |= HASWIDTH|SIMPLE; |
| 8288 | } |
| 8289 | break; |
| 8290 | case 'N': |
| 8291 | /* Handle \N and \N{NAME} here and not below because it can be |
| 8292 | multicharacter. join_exact() will join them up later on. |
| 8293 | Also this makes sure that things like /\N{BLAH}+/ and |
| 8294 | \N{BLAH} being multi char Just Happen. dmq*/ |
| 8295 | ++RExC_parse; |
| 8296 | ret= reg_namedseq(pRExC_state, NULL, flagp, depth); |
| 8297 | break; |
| 8298 | case 'k': /* Handle \k<NAME> and \k'NAME' */ |
| 8299 | parse_named_seq: |
| 8300 | { |
| 8301 | char ch= RExC_parse[1]; |
| 8302 | if (ch != '<' && ch != '\'' && ch != '{') { |
| 8303 | RExC_parse++; |
| 8304 | vFAIL2("Sequence %.2s... not terminated",parse_start); |
| 8305 | } else { |
| 8306 | /* this pretty much dupes the code for (?P=...) in reg(), if |
| 8307 | you change this make sure you change that */ |
| 8308 | char* name_start = (RExC_parse += 2); |
| 8309 | U32 num = 0; |
| 8310 | SV *sv_dat = reg_scan_name(pRExC_state, |
| 8311 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); |
| 8312 | ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\''; |
| 8313 | if (RExC_parse == name_start || *RExC_parse != ch) |
| 8314 | vFAIL2("Sequence %.3s... not terminated",parse_start); |
| 8315 | |
| 8316 | if (!SIZE_ONLY) { |
| 8317 | num = add_data( pRExC_state, 1, "S" ); |
| 8318 | RExC_rxi->data->data[num]=(void*)sv_dat; |
| 8319 | SvREFCNT_inc_simple_void(sv_dat); |
| 8320 | } |
| 8321 | |
| 8322 | RExC_sawback = 1; |
| 8323 | ret = reganode(pRExC_state, |
| 8324 | ((! FOLD) |
| 8325 | ? NREF |
| 8326 | : (MORE_ASCII_RESTRICTED) |
| 8327 | ? NREFFA |
| 8328 | : (AT_LEAST_UNI_SEMANTICS) |
| 8329 | ? NREFFU |
| 8330 | : (LOC) |
| 8331 | ? NREFFL |
| 8332 | : NREFF), |
| 8333 | num); |
| 8334 | *flagp |= HASWIDTH; |
| 8335 | |
| 8336 | /* override incorrect value set in reganode MJD */ |
| 8337 | Set_Node_Offset(ret, parse_start+1); |
| 8338 | Set_Node_Cur_Length(ret); /* MJD */ |
| 8339 | nextchar(pRExC_state); |
| 8340 | |
| 8341 | } |
| 8342 | break; |
| 8343 | } |
| 8344 | case 'g': |
| 8345 | case '1': case '2': case '3': case '4': |
| 8346 | case '5': case '6': case '7': case '8': case '9': |
| 8347 | { |
| 8348 | I32 num; |
| 8349 | bool isg = *RExC_parse == 'g'; |
| 8350 | bool isrel = 0; |
| 8351 | bool hasbrace = 0; |
| 8352 | if (isg) { |
| 8353 | RExC_parse++; |
| 8354 | if (*RExC_parse == '{') { |
| 8355 | RExC_parse++; |
| 8356 | hasbrace = 1; |
| 8357 | } |
| 8358 | if (*RExC_parse == '-') { |
| 8359 | RExC_parse++; |
| 8360 | isrel = 1; |
| 8361 | } |
| 8362 | if (hasbrace && !isDIGIT(*RExC_parse)) { |
| 8363 | if (isrel) RExC_parse--; |
| 8364 | RExC_parse -= 2; |
| 8365 | goto parse_named_seq; |
| 8366 | } } |
| 8367 | num = atoi(RExC_parse); |
| 8368 | if (isg && num == 0) |
| 8369 | vFAIL("Reference to invalid group 0"); |
| 8370 | if (isrel) { |
| 8371 | num = RExC_npar - num; |
| 8372 | if (num < 1) |
| 8373 | vFAIL("Reference to nonexistent or unclosed group"); |
| 8374 | } |
| 8375 | if (!isg && num > 9 && num >= RExC_npar) |
| 8376 | goto defchar; |
| 8377 | else { |
| 8378 | char * const parse_start = RExC_parse - 1; /* MJD */ |
| 8379 | while (isDIGIT(*RExC_parse)) |
| 8380 | RExC_parse++; |
| 8381 | if (parse_start == RExC_parse - 1) |
| 8382 | vFAIL("Unterminated \\g... pattern"); |
| 8383 | if (hasbrace) { |
| 8384 | if (*RExC_parse != '}') |
| 8385 | vFAIL("Unterminated \\g{...} pattern"); |
| 8386 | RExC_parse++; |
| 8387 | } |
| 8388 | if (!SIZE_ONLY) { |
| 8389 | if (num > (I32)RExC_rx->nparens) |
| 8390 | vFAIL("Reference to nonexistent group"); |
| 8391 | } |
| 8392 | RExC_sawback = 1; |
| 8393 | ret = reganode(pRExC_state, |
| 8394 | ((! FOLD) |
| 8395 | ? REF |
| 8396 | : (MORE_ASCII_RESTRICTED) |
| 8397 | ? REFFA |
| 8398 | : (AT_LEAST_UNI_SEMANTICS) |
| 8399 | ? REFFU |
| 8400 | : (LOC) |
| 8401 | ? REFFL |
| 8402 | : REFF), |
| 8403 | num); |
| 8404 | *flagp |= HASWIDTH; |
| 8405 | |
| 8406 | /* override incorrect value set in reganode MJD */ |
| 8407 | Set_Node_Offset(ret, parse_start+1); |
| 8408 | Set_Node_Cur_Length(ret); /* MJD */ |
| 8409 | RExC_parse--; |
| 8410 | nextchar(pRExC_state); |
| 8411 | } |
| 8412 | } |
| 8413 | break; |
| 8414 | case '\0': |
| 8415 | if (RExC_parse >= RExC_end) |
| 8416 | FAIL("Trailing \\"); |
| 8417 | /* FALL THROUGH */ |
| 8418 | default: |
| 8419 | /* Do not generate "unrecognized" warnings here, we fall |
| 8420 | back into the quick-grab loop below */ |
| 8421 | parse_start--; |
| 8422 | goto defchar; |
| 8423 | } |
| 8424 | break; |
| 8425 | |
| 8426 | case '#': |
| 8427 | if (RExC_flags & RXf_PMf_EXTENDED) { |
| 8428 | if ( reg_skipcomment( pRExC_state ) ) |
| 8429 | goto tryagain; |
| 8430 | } |
| 8431 | /* FALL THROUGH */ |
| 8432 | |
| 8433 | default: |
| 8434 | |
| 8435 | parse_start = RExC_parse - 1; |
| 8436 | |
| 8437 | RExC_parse++; |
| 8438 | |
| 8439 | defchar: { |
| 8440 | typedef enum { |
| 8441 | generic_char = 0, |
| 8442 | char_s, |
| 8443 | upsilon_1, |
| 8444 | upsilon_2, |
| 8445 | iota_1, |
| 8446 | iota_2, |
| 8447 | } char_state; |
| 8448 | char_state latest_char_state = generic_char; |
| 8449 | register STRLEN len; |
| 8450 | register UV ender; |
| 8451 | register char *p; |
| 8452 | char *s; |
| 8453 | STRLEN foldlen; |
| 8454 | U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf; |
| 8455 | regnode * orig_emit; |
| 8456 | |
| 8457 | ender = 0; |
| 8458 | orig_emit = RExC_emit; /* Save the original output node position in |
| 8459 | case we need to output a different node |
| 8460 | type */ |
| 8461 | ret = reg_node(pRExC_state, |
| 8462 | (U8) ((! FOLD) ? EXACT |
| 8463 | : (LOC) |
| 8464 | ? EXACTFL |
| 8465 | : (MORE_ASCII_RESTRICTED) |
| 8466 | ? EXACTFA |
| 8467 | : (AT_LEAST_UNI_SEMANTICS) |
| 8468 | ? EXACTFU |
| 8469 | : EXACTF) |
| 8470 | ); |
| 8471 | s = STRING(ret); |
| 8472 | for (len = 0, p = RExC_parse - 1; |
| 8473 | len < 127 && p < RExC_end; |
| 8474 | len++) |
| 8475 | { |
| 8476 | char * const oldp = p; |
| 8477 | |
| 8478 | if (RExC_flags & RXf_PMf_EXTENDED) |
| 8479 | p = regwhite( pRExC_state, p ); |
| 8480 | switch ((U8)*p) { |
| 8481 | case '^': |
| 8482 | case '$': |
| 8483 | case '.': |
| 8484 | case '[': |
| 8485 | case '(': |
| 8486 | case ')': |
| 8487 | case '|': |
| 8488 | goto loopdone; |
| 8489 | case '\\': |
| 8490 | /* Literal Escapes Switch |
| 8491 | |
| 8492 | This switch is meant to handle escape sequences that |
| 8493 | resolve to a literal character. |
| 8494 | |
| 8495 | Every escape sequence that represents something |
| 8496 | else, like an assertion or a char class, is handled |
| 8497 | in the switch marked 'Special Escapes' above in this |
| 8498 | routine, but also has an entry here as anything that |
| 8499 | isn't explicitly mentioned here will be treated as |
| 8500 | an unescaped equivalent literal. |
| 8501 | */ |
| 8502 | |
| 8503 | switch ((U8)*++p) { |
| 8504 | /* These are all the special escapes. */ |
| 8505 | case 'A': /* Start assertion */ |
| 8506 | case 'b': case 'B': /* Word-boundary assertion*/ |
| 8507 | case 'C': /* Single char !DANGEROUS! */ |
| 8508 | case 'd': case 'D': /* digit class */ |
| 8509 | case 'g': case 'G': /* generic-backref, pos assertion */ |
| 8510 | case 'h': case 'H': /* HORIZWS */ |
| 8511 | case 'k': case 'K': /* named backref, keep marker */ |
| 8512 | case 'N': /* named char sequence */ |
| 8513 | case 'p': case 'P': /* Unicode property */ |
| 8514 | case 'R': /* LNBREAK */ |
| 8515 | case 's': case 'S': /* space class */ |
| 8516 | case 'v': case 'V': /* VERTWS */ |
| 8517 | case 'w': case 'W': /* word class */ |
| 8518 | case 'X': /* eXtended Unicode "combining character sequence" */ |
| 8519 | case 'z': case 'Z': /* End of line/string assertion */ |
| 8520 | --p; |
| 8521 | goto loopdone; |
| 8522 | |
| 8523 | /* Anything after here is an escape that resolves to a |
| 8524 | literal. (Except digits, which may or may not) |
| 8525 | */ |
| 8526 | case 'n': |
| 8527 | ender = '\n'; |
| 8528 | p++; |
| 8529 | break; |
| 8530 | case 'r': |
| 8531 | ender = '\r'; |
| 8532 | p++; |
| 8533 | break; |
| 8534 | case 't': |
| 8535 | ender = '\t'; |
| 8536 | p++; |
| 8537 | break; |
| 8538 | case 'f': |
| 8539 | ender = '\f'; |
| 8540 | p++; |
| 8541 | break; |
| 8542 | case 'e': |
| 8543 | ender = ASCII_TO_NATIVE('\033'); |
| 8544 | p++; |
| 8545 | break; |
| 8546 | case 'a': |
| 8547 | ender = ASCII_TO_NATIVE('\007'); |
| 8548 | p++; |
| 8549 | break; |
| 8550 | case 'o': |
| 8551 | { |
| 8552 | STRLEN brace_len = len; |
| 8553 | UV result; |
| 8554 | const char* error_msg; |
| 8555 | |
| 8556 | bool valid = grok_bslash_o(p, |
| 8557 | &result, |
| 8558 | &brace_len, |
| 8559 | &error_msg, |
| 8560 | 1); |
| 8561 | p += brace_len; |
| 8562 | if (! valid) { |
| 8563 | RExC_parse = p; /* going to die anyway; point |
| 8564 | to exact spot of failure */ |
| 8565 | vFAIL(error_msg); |
| 8566 | } |
| 8567 | else |
| 8568 | { |
| 8569 | ender = result; |
| 8570 | } |
| 8571 | if (PL_encoding && ender < 0x100) { |
| 8572 | goto recode_encoding; |
| 8573 | } |
| 8574 | if (ender > 0xff) { |
| 8575 | REQUIRE_UTF8; |
| 8576 | } |
| 8577 | break; |
| 8578 | } |
| 8579 | case 'x': |
| 8580 | if (*++p == '{') { |
| 8581 | char* const e = strchr(p, '}'); |
| 8582 | |
| 8583 | if (!e) { |
| 8584 | RExC_parse = p + 1; |
| 8585 | vFAIL("Missing right brace on \\x{}"); |
| 8586 | } |
| 8587 | else { |
| 8588 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES |
| 8589 | | PERL_SCAN_DISALLOW_PREFIX; |
| 8590 | STRLEN numlen = e - p - 1; |
| 8591 | ender = grok_hex(p + 1, &numlen, &flags, NULL); |
| 8592 | if (ender > 0xff) |
| 8593 | REQUIRE_UTF8; |
| 8594 | p = e + 1; |
| 8595 | } |
| 8596 | } |
| 8597 | else { |
| 8598 | I32 flags = PERL_SCAN_DISALLOW_PREFIX; |
| 8599 | STRLEN numlen = 2; |
| 8600 | ender = grok_hex(p, &numlen, &flags, NULL); |
| 8601 | p += numlen; |
| 8602 | } |
| 8603 | if (PL_encoding && ender < 0x100) |
| 8604 | goto recode_encoding; |
| 8605 | break; |
| 8606 | case 'c': |
| 8607 | p++; |
| 8608 | ender = grok_bslash_c(*p++, UTF, SIZE_ONLY); |
| 8609 | break; |
| 8610 | case '0': case '1': case '2': case '3':case '4': |
| 8611 | case '5': case '6': case '7': case '8':case '9': |
| 8612 | if (*p == '0' || |
| 8613 | (isDIGIT(p[1]) && atoi(p) >= RExC_npar)) |
| 8614 | { |
| 8615 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT; |
| 8616 | STRLEN numlen = 3; |
| 8617 | ender = grok_oct(p, &numlen, &flags, NULL); |
| 8618 | if (ender > 0xff) { |
| 8619 | REQUIRE_UTF8; |
| 8620 | } |
| 8621 | p += numlen; |
| 8622 | } |
| 8623 | else { |
| 8624 | --p; |
| 8625 | goto loopdone; |
| 8626 | } |
| 8627 | if (PL_encoding && ender < 0x100) |
| 8628 | goto recode_encoding; |
| 8629 | break; |
| 8630 | recode_encoding: |
| 8631 | if (! RExC_override_recoding) { |
| 8632 | SV* enc = PL_encoding; |
| 8633 | ender = reg_recode((const char)(U8)ender, &enc); |
| 8634 | if (!enc && SIZE_ONLY) |
| 8635 | ckWARNreg(p, "Invalid escape in the specified encoding"); |
| 8636 | REQUIRE_UTF8; |
| 8637 | } |
| 8638 | break; |
| 8639 | case '\0': |
| 8640 | if (p >= RExC_end) |
| 8641 | FAIL("Trailing \\"); |
| 8642 | /* FALL THROUGH */ |
| 8643 | default: |
| 8644 | if (!SIZE_ONLY&& isALPHA(*p)) { |
| 8645 | /* Include any { following the alpha to emphasize |
| 8646 | * that it could be part of an escape at some point |
| 8647 | * in the future */ |
| 8648 | int len = (*(p + 1) == '{') ? 2 : 1; |
| 8649 | ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p); |
| 8650 | } |
| 8651 | goto normal_default; |
| 8652 | } |
| 8653 | break; |
| 8654 | default: |
| 8655 | normal_default: |
| 8656 | if (UTF8_IS_START(*p) && UTF) { |
| 8657 | STRLEN numlen; |
| 8658 | ender = utf8n_to_uvchr((U8*)p, RExC_end - p, |
| 8659 | &numlen, UTF8_ALLOW_DEFAULT); |
| 8660 | p += numlen; |
| 8661 | } |
| 8662 | else |
| 8663 | ender = (U8) *p++; |
| 8664 | break; |
| 8665 | } /* End of switch on the literal */ |
| 8666 | |
| 8667 | /* Certain characters are problematic because their folded |
| 8668 | * length is so different from their original length that it |
| 8669 | * isn't handleable by the optimizer. They are therefore not |
| 8670 | * placed in an EXACTish node; and are here handled specially. |
| 8671 | * (Even if the optimizer handled LATIN_SMALL_LETTER_SHARP_S, |
| 8672 | * putting it in a special node keeps regexec from having to |
| 8673 | * deal with a non-utf8 multi-char fold */ |
| 8674 | if (FOLD |
| 8675 | && (ender > 255 || (! MORE_ASCII_RESTRICTED && ! LOC))) |
| 8676 | { |
| 8677 | /* We look for either side of the fold. For example \xDF |
| 8678 | * folds to 'ss'. We look for both the single character |
| 8679 | * \xDF and the sequence 'ss'. When we find something that |
| 8680 | * could be one of those, we stop and flush whatever we |
| 8681 | * have output so far into the EXACTish node that was being |
| 8682 | * built. Then restore the input pointer to what it was. |
| 8683 | * regatom will return that EXACT node, and will be called |
| 8684 | * again, positioned so the first character is the one in |
| 8685 | * question, which we return in a different node type. |
| 8686 | * The multi-char folds are a sequence, so the occurrence |
| 8687 | * of the first character in that sequence doesn't |
| 8688 | * necessarily mean that what follows is the rest of the |
| 8689 | * sequence. We keep track of that with a state machine, |
| 8690 | * with the state being set to the latest character |
| 8691 | * processed before the current one. Most characters will |
| 8692 | * set the state to 0, but if one occurs that is part of a |
| 8693 | * potential tricky fold sequence, the state is set to that |
| 8694 | * character, and the next loop iteration sees if the state |
| 8695 | * should progress towards the final folded-from character, |
| 8696 | * or if it was a false alarm. If it turns out to be a |
| 8697 | * false alarm, the character(s) will be output in a new |
| 8698 | * EXACTish node, and join_exact() will later combine them. |
| 8699 | * In the case of the 'ss' sequence, which is more common |
| 8700 | * and more easily checked, some look-ahead is done to |
| 8701 | * save time by ruling-out some false alarms */ |
| 8702 | switch (ender) { |
| 8703 | default: |
| 8704 | latest_char_state = generic_char; |
| 8705 | break; |
| 8706 | case 's': |
| 8707 | case 'S': |
| 8708 | case 0x17F: /* LATIN SMALL LETTER LONG S */ |
| 8709 | if (AT_LEAST_UNI_SEMANTICS) { |
| 8710 | if (latest_char_state == char_s) { /* 'ss' */ |
| 8711 | ender = LATIN_SMALL_LETTER_SHARP_S; |
| 8712 | goto do_tricky; |
| 8713 | } |
| 8714 | else if (p < RExC_end) { |
| 8715 | |
| 8716 | /* Look-ahead at the next character. If it |
| 8717 | * is also an s, we handle as a sharp s |
| 8718 | * tricky regnode. */ |
| 8719 | if (*p == 's' || *p == 'S') { |
| 8720 | |
| 8721 | /* But first flush anything in the |
| 8722 | * EXACTish buffer */ |
| 8723 | if (len != 0) { |
| 8724 | p = oldp; |
| 8725 | goto loopdone; |
| 8726 | } |
| 8727 | p++; /* Account for swallowing this |
| 8728 | 's' up */ |
| 8729 | ender = LATIN_SMALL_LETTER_SHARP_S; |
| 8730 | goto do_tricky; |
| 8731 | } |
| 8732 | /* Here, the next character is not a |
| 8733 | * literal 's', but still could |
| 8734 | * evaluate to one if part of a \o{}, |
| 8735 | * \x or \OCTAL-DIGIT. The minimum |
| 8736 | * length required for that is 4, eg |
| 8737 | * \x53 or \123 */ |
| 8738 | else if (*p == '\\' |
| 8739 | && p < RExC_end - 4 |
| 8740 | && (isDIGIT(*(p + 1)) |
| 8741 | || *(p + 1) == 'x' |
| 8742 | || *(p + 1) == 'o' )) |
| 8743 | { |
| 8744 | |
| 8745 | /* Here, it could be an 's', too much |
| 8746 | * bother to figure it out here. Flush |
| 8747 | * the buffer if any; when come back |
| 8748 | * here, set the state so know that the |
| 8749 | * previous char was an 's' */ |
| 8750 | if (len != 0) { |
| 8751 | latest_char_state = generic_char; |
| 8752 | p = oldp; |
| 8753 | goto loopdone; |
| 8754 | } |
| 8755 | latest_char_state = char_s; |
| 8756 | break; |
| 8757 | } |
| 8758 | } |
| 8759 | } |
| 8760 | |
| 8761 | /* Here, can't be an 'ss' sequence, or at least not |
| 8762 | * one that could fold to/from the sharp ss */ |
| 8763 | latest_char_state = generic_char; |
| 8764 | break; |
| 8765 | case 0x03C5: /* First char in upsilon series */ |
| 8766 | case 0x03A5: /* Also capital UPSILON, which folds to |
| 8767 | 03C5, and hence exhibits the same |
| 8768 | problem */ |
| 8769 | if (p < RExC_end - 4) { /* Need >= 4 bytes left */ |
| 8770 | latest_char_state = upsilon_1; |
| 8771 | if (len != 0) { |
| 8772 | p = oldp; |
| 8773 | goto loopdone; |
| 8774 | } |
| 8775 | } |
| 8776 | else { |
| 8777 | latest_char_state = generic_char; |
| 8778 | } |
| 8779 | break; |
| 8780 | case 0x03B9: /* First char in iota series */ |
| 8781 | case 0x0399: /* Also capital IOTA */ |
| 8782 | case 0x1FBE: /* GREEK PROSGEGRAMMENI folds to 3B9 */ |
| 8783 | case 0x0345: /* COMBINING GREEK YPOGEGRAMMENI folds |
| 8784 | to 3B9 */ |
| 8785 | if (p < RExC_end - 4) { |
| 8786 | latest_char_state = iota_1; |
| 8787 | if (len != 0) { |
| 8788 | p = oldp; |
| 8789 | goto loopdone; |
| 8790 | } |
| 8791 | } |
| 8792 | else { |
| 8793 | latest_char_state = generic_char; |
| 8794 | } |
| 8795 | break; |
| 8796 | case 0x0308: |
| 8797 | if (latest_char_state == upsilon_1) { |
| 8798 | latest_char_state = upsilon_2; |
| 8799 | } |
| 8800 | else if (latest_char_state == iota_1) { |
| 8801 | latest_char_state = iota_2; |
| 8802 | } |
| 8803 | else { |
| 8804 | latest_char_state = generic_char; |
| 8805 | } |
| 8806 | break; |
| 8807 | case 0x301: |
| 8808 | if (latest_char_state == upsilon_2) { |
| 8809 | ender = GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS; |
| 8810 | goto do_tricky; |
| 8811 | } |
| 8812 | else if (latest_char_state == iota_2) { |
| 8813 | ender = GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS; |
| 8814 | goto do_tricky; |
| 8815 | } |
| 8816 | latest_char_state = generic_char; |
| 8817 | break; |
| 8818 | |
| 8819 | /* These are the tricky fold characters. Flush any |
| 8820 | * buffer first. (When adding to this list, also should |
| 8821 | * add them to fold_grind.t to make sure get tested) */ |
| 8822 | case GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS: |
| 8823 | case GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS: |
| 8824 | case LATIN_SMALL_LETTER_SHARP_S: |
| 8825 | case LATIN_CAPITAL_LETTER_SHARP_S: |
| 8826 | case 0x1FD3: /* GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA */ |
| 8827 | case 0x1FE3: /* GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA */ |
| 8828 | if (len != 0) { |
| 8829 | p = oldp; |
| 8830 | goto loopdone; |
| 8831 | } |
| 8832 | /* FALL THROUGH */ |
| 8833 | do_tricky: { |
| 8834 | char* const oldregxend = RExC_end; |
| 8835 | U8 tmpbuf[UTF8_MAXBYTES+1]; |
| 8836 | |
| 8837 | /* Here, we know we need to generate a special |
| 8838 | * regnode, and 'ender' contains the tricky |
| 8839 | * character. What's done is to pretend it's in a |
| 8840 | * [bracketed] class, and let the code that deals |
| 8841 | * with those handle it, as that code has all the |
| 8842 | * intelligence necessary. First save the current |
| 8843 | * parse state, get rid of the already allocated |
| 8844 | * but empty EXACT node that the ANYOFV node will |
| 8845 | * replace, and point the parse to a buffer which |
| 8846 | * we fill with the character we want the regclass |
| 8847 | * code to think is being parsed */ |
| 8848 | RExC_emit = orig_emit; |
| 8849 | RExC_parse = (char *) tmpbuf; |
| 8850 | if (UTF) { |
| 8851 | U8 *d = uvchr_to_utf8(tmpbuf, ender); |
| 8852 | *d = '\0'; |
| 8853 | RExC_end = (char *) d; |
| 8854 | } |
| 8855 | else { /* ender above 255 already excluded */ |
| 8856 | tmpbuf[0] = (U8) ender; |
| 8857 | tmpbuf[1] = '\0'; |
| 8858 | RExC_end = RExC_parse + 1; |
| 8859 | } |
| 8860 | |
| 8861 | ret = regclass(pRExC_state,depth+1); |
| 8862 | |
| 8863 | /* Here, have parsed the buffer. Reset the parse to |
| 8864 | * the actual input, and return */ |
| 8865 | RExC_end = oldregxend; |
| 8866 | RExC_parse = p - 1; |
| 8867 | |
| 8868 | Set_Node_Offset(ret, RExC_parse); |
| 8869 | Set_Node_Cur_Length(ret); |
| 8870 | nextchar(pRExC_state); |
| 8871 | *flagp |= HASWIDTH|SIMPLE; |
| 8872 | return ret; |
| 8873 | } |
| 8874 | } |
| 8875 | } |
| 8876 | |
| 8877 | if ( RExC_flags & RXf_PMf_EXTENDED) |
| 8878 | p = regwhite( pRExC_state, p ); |
| 8879 | if (UTF && FOLD) { |
| 8880 | /* Prime the casefolded buffer. Locale rules, which apply |
| 8881 | * only to code points < 256, aren't known until execution, |
| 8882 | * so for them, just output the original character using |
| 8883 | * utf8 */ |
| 8884 | if (LOC && ender < 256) { |
| 8885 | if (UNI_IS_INVARIANT(ender)) { |
| 8886 | *tmpbuf = (U8) ender; |
| 8887 | foldlen = 1; |
| 8888 | } else { |
| 8889 | *tmpbuf = UTF8_TWO_BYTE_HI(ender); |
| 8890 | *(tmpbuf + 1) = UTF8_TWO_BYTE_LO(ender); |
| 8891 | foldlen = 2; |
| 8892 | } |
| 8893 | } |
| 8894 | else if (isASCII(ender)) { /* Note: Here can't also be LOC |
| 8895 | */ |
| 8896 | ender = toLOWER(ender); |
| 8897 | *tmpbuf = (U8) ender; |
| 8898 | foldlen = 1; |
| 8899 | } |
| 8900 | else if (! MORE_ASCII_RESTRICTED && ! LOC) { |
| 8901 | |
| 8902 | /* Locale and /aa require more selectivity about the |
| 8903 | * fold, so are handled below. Otherwise, here, just |
| 8904 | * use the fold */ |
| 8905 | ender = toFOLD_uni(ender, tmpbuf, &foldlen); |
| 8906 | } |
| 8907 | else { |
| 8908 | /* Under locale rules or /aa we are not to mix, |
| 8909 | * respectively, ords < 256 or ASCII with non-. So |
| 8910 | * reject folds that mix them, using only the |
| 8911 | * non-folded code point. So do the fold to a |
| 8912 | * temporary, and inspect each character in it. */ |
| 8913 | U8 trialbuf[UTF8_MAXBYTES_CASE+1]; |
| 8914 | U8* s = trialbuf; |
| 8915 | UV tmpender = toFOLD_uni(ender, trialbuf, &foldlen); |
| 8916 | U8* e = s + foldlen; |
| 8917 | bool fold_ok = TRUE; |
| 8918 | |
| 8919 | while (s < e) { |
| 8920 | if (isASCII(*s) |
| 8921 | || (LOC && (UTF8_IS_INVARIANT(*s) |
| 8922 | || UTF8_IS_DOWNGRADEABLE_START(*s)))) |
| 8923 | { |
| 8924 | fold_ok = FALSE; |
| 8925 | break; |
| 8926 | } |
| 8927 | s += UTF8SKIP(s); |
| 8928 | } |
| 8929 | if (fold_ok) { |
| 8930 | Copy(trialbuf, tmpbuf, foldlen, U8); |
| 8931 | ender = tmpender; |
| 8932 | } |
| 8933 | else { |
| 8934 | uvuni_to_utf8(tmpbuf, ender); |
| 8935 | foldlen = UNISKIP(ender); |
| 8936 | } |
| 8937 | } |
| 8938 | } |
| 8939 | if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */ |
| 8940 | if (len) |
| 8941 | p = oldp; |
| 8942 | else if (UTF) { |
| 8943 | if (FOLD) { |
| 8944 | /* Emit all the Unicode characters. */ |
| 8945 | STRLEN numlen; |
| 8946 | for (foldbuf = tmpbuf; |
| 8947 | foldlen; |
| 8948 | foldlen -= numlen) { |
| 8949 | ender = utf8_to_uvchr(foldbuf, &numlen); |
| 8950 | if (numlen > 0) { |
| 8951 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
| 8952 | s += unilen; |
| 8953 | len += unilen; |
| 8954 | /* In EBCDIC the numlen |
| 8955 | * and unilen can differ. */ |
| 8956 | foldbuf += numlen; |
| 8957 | if (numlen >= foldlen) |
| 8958 | break; |
| 8959 | } |
| 8960 | else |
| 8961 | break; /* "Can't happen." */ |
| 8962 | } |
| 8963 | } |
| 8964 | else { |
| 8965 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
| 8966 | if (unilen > 0) { |
| 8967 | s += unilen; |
| 8968 | len += unilen; |
| 8969 | } |
| 8970 | } |
| 8971 | } |
| 8972 | else { |
| 8973 | len++; |
| 8974 | REGC((char)ender, s++); |
| 8975 | } |
| 8976 | break; |
| 8977 | } |
| 8978 | if (UTF) { |
| 8979 | if (FOLD) { |
| 8980 | /* Emit all the Unicode characters. */ |
| 8981 | STRLEN numlen; |
| 8982 | for (foldbuf = tmpbuf; |
| 8983 | foldlen; |
| 8984 | foldlen -= numlen) { |
| 8985 | ender = utf8_to_uvchr(foldbuf, &numlen); |
| 8986 | if (numlen > 0) { |
| 8987 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
| 8988 | len += unilen; |
| 8989 | s += unilen; |
| 8990 | /* In EBCDIC the numlen |
| 8991 | * and unilen can differ. */ |
| 8992 | foldbuf += numlen; |
| 8993 | if (numlen >= foldlen) |
| 8994 | break; |
| 8995 | } |
| 8996 | else |
| 8997 | break; |
| 8998 | } |
| 8999 | } |
| 9000 | else { |
| 9001 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
| 9002 | if (unilen > 0) { |
| 9003 | s += unilen; |
| 9004 | len += unilen; |
| 9005 | } |
| 9006 | } |
| 9007 | len--; |
| 9008 | } |
| 9009 | else { |
| 9010 | REGC((char)ender, s++); |
| 9011 | } |
| 9012 | } |
| 9013 | loopdone: /* Jumped to when encounters something that shouldn't be in |
| 9014 | the node */ |
| 9015 | RExC_parse = p - 1; |
| 9016 | Set_Node_Cur_Length(ret); /* MJD */ |
| 9017 | nextchar(pRExC_state); |
| 9018 | { |
| 9019 | /* len is STRLEN which is unsigned, need to copy to signed */ |
| 9020 | IV iv = len; |
| 9021 | if (iv < 0) |
| 9022 | vFAIL("Internal disaster"); |
| 9023 | } |
| 9024 | if (len > 0) |
| 9025 | *flagp |= HASWIDTH; |
| 9026 | if (len == 1 && UNI_IS_INVARIANT(ender)) |
| 9027 | *flagp |= SIMPLE; |
| 9028 | |
| 9029 | if (SIZE_ONLY) |
| 9030 | RExC_size += STR_SZ(len); |
| 9031 | else { |
| 9032 | STR_LEN(ret) = len; |
| 9033 | RExC_emit += STR_SZ(len); |
| 9034 | } |
| 9035 | } |
| 9036 | break; |
| 9037 | } |
| 9038 | |
| 9039 | return(ret); |
| 9040 | |
| 9041 | /* Jumped to when an unrecognized character set is encountered */ |
| 9042 | bad_charset: |
| 9043 | Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags)); |
| 9044 | return(NULL); |
| 9045 | } |
| 9046 | |
| 9047 | STATIC char * |
| 9048 | S_regwhite( RExC_state_t *pRExC_state, char *p ) |
| 9049 | { |
| 9050 | const char *e = RExC_end; |
| 9051 | |
| 9052 | PERL_ARGS_ASSERT_REGWHITE; |
| 9053 | |
| 9054 | while (p < e) { |
| 9055 | if (isSPACE(*p)) |
| 9056 | ++p; |
| 9057 | else if (*p == '#') { |
| 9058 | bool ended = 0; |
| 9059 | do { |
| 9060 | if (*p++ == '\n') { |
| 9061 | ended = 1; |
| 9062 | break; |
| 9063 | } |
| 9064 | } while (p < e); |
| 9065 | if (!ended) |
| 9066 | RExC_seen |= REG_SEEN_RUN_ON_COMMENT; |
| 9067 | } |
| 9068 | else |
| 9069 | break; |
| 9070 | } |
| 9071 | return p; |
| 9072 | } |
| 9073 | |
| 9074 | /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]]. |
| 9075 | Character classes ([:foo:]) can also be negated ([:^foo:]). |
| 9076 | Returns a named class id (ANYOF_XXX) if successful, -1 otherwise. |
| 9077 | Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed, |
| 9078 | but trigger failures because they are currently unimplemented. */ |
| 9079 | |
| 9080 | #define POSIXCC_DONE(c) ((c) == ':') |
| 9081 | #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.') |
| 9082 | #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c)) |
| 9083 | |
| 9084 | STATIC I32 |
| 9085 | S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value) |
| 9086 | { |
| 9087 | dVAR; |
| 9088 | I32 namedclass = OOB_NAMEDCLASS; |
| 9089 | |
| 9090 | PERL_ARGS_ASSERT_REGPPOSIXCC; |
| 9091 | |
| 9092 | if (value == '[' && RExC_parse + 1 < RExC_end && |
| 9093 | /* I smell either [: or [= or [. -- POSIX has been here, right? */ |
| 9094 | POSIXCC(UCHARAT(RExC_parse))) { |
| 9095 | const char c = UCHARAT(RExC_parse); |
| 9096 | char* const s = RExC_parse++; |
| 9097 | |
| 9098 | while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c) |
| 9099 | RExC_parse++; |
| 9100 | if (RExC_parse == RExC_end) |
| 9101 | /* Grandfather lone [:, [=, [. */ |
| 9102 | RExC_parse = s; |
| 9103 | else { |
| 9104 | const char* const t = RExC_parse++; /* skip over the c */ |
| 9105 | assert(*t == c); |
| 9106 | |
| 9107 | if (UCHARAT(RExC_parse) == ']') { |
| 9108 | const char *posixcc = s + 1; |
| 9109 | RExC_parse++; /* skip over the ending ] */ |
| 9110 | |
| 9111 | if (*s == ':') { |
| 9112 | const I32 complement = *posixcc == '^' ? *posixcc++ : 0; |
| 9113 | const I32 skip = t - posixcc; |
| 9114 | |
| 9115 | /* Initially switch on the length of the name. */ |
| 9116 | switch (skip) { |
| 9117 | case 4: |
| 9118 | if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */ |
| 9119 | namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM; |
| 9120 | break; |
| 9121 | case 5: |
| 9122 | /* Names all of length 5. */ |
| 9123 | /* alnum alpha ascii blank cntrl digit graph lower |
| 9124 | print punct space upper */ |
| 9125 | /* Offset 4 gives the best switch position. */ |
| 9126 | switch (posixcc[4]) { |
| 9127 | case 'a': |
| 9128 | if (memEQ(posixcc, "alph", 4)) /* alpha */ |
| 9129 | namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA; |
| 9130 | break; |
| 9131 | case 'e': |
| 9132 | if (memEQ(posixcc, "spac", 4)) /* space */ |
| 9133 | namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC; |
| 9134 | break; |
| 9135 | case 'h': |
| 9136 | if (memEQ(posixcc, "grap", 4)) /* graph */ |
| 9137 | namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH; |
| 9138 | break; |
| 9139 | case 'i': |
| 9140 | if (memEQ(posixcc, "asci", 4)) /* ascii */ |
| 9141 | namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII; |
| 9142 | break; |
| 9143 | case 'k': |
| 9144 | if (memEQ(posixcc, "blan", 4)) /* blank */ |
| 9145 | namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK; |
| 9146 | break; |
| 9147 | case 'l': |
| 9148 | if (memEQ(posixcc, "cntr", 4)) /* cntrl */ |
| 9149 | namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL; |
| 9150 | break; |
| 9151 | case 'm': |
| 9152 | if (memEQ(posixcc, "alnu", 4)) /* alnum */ |
| 9153 | namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC; |
| 9154 | break; |
| 9155 | case 'r': |
| 9156 | if (memEQ(posixcc, "lowe", 4)) /* lower */ |
| 9157 | namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER; |
| 9158 | else if (memEQ(posixcc, "uppe", 4)) /* upper */ |
| 9159 | namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER; |
| 9160 | break; |
| 9161 | case 't': |
| 9162 | if (memEQ(posixcc, "digi", 4)) /* digit */ |
| 9163 | namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT; |
| 9164 | else if (memEQ(posixcc, "prin", 4)) /* print */ |
| 9165 | namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT; |
| 9166 | else if (memEQ(posixcc, "punc", 4)) /* punct */ |
| 9167 | namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT; |
| 9168 | break; |
| 9169 | } |
| 9170 | break; |
| 9171 | case 6: |
| 9172 | if (memEQ(posixcc, "xdigit", 6)) |
| 9173 | namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT; |
| 9174 | break; |
| 9175 | } |
| 9176 | |
| 9177 | if (namedclass == OOB_NAMEDCLASS) |
| 9178 | Simple_vFAIL3("POSIX class [:%.*s:] unknown", |
| 9179 | t - s - 1, s + 1); |
| 9180 | assert (posixcc[skip] == ':'); |
| 9181 | assert (posixcc[skip+1] == ']'); |
| 9182 | } else if (!SIZE_ONLY) { |
| 9183 | /* [[=foo=]] and [[.foo.]] are still future. */ |
| 9184 | |
| 9185 | /* adjust RExC_parse so the warning shows after |
| 9186 | the class closes */ |
| 9187 | while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']') |
| 9188 | RExC_parse++; |
| 9189 | Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c); |
| 9190 | } |
| 9191 | } else { |
| 9192 | /* Maternal grandfather: |
| 9193 | * "[:" ending in ":" but not in ":]" */ |
| 9194 | RExC_parse = s; |
| 9195 | } |
| 9196 | } |
| 9197 | } |
| 9198 | |
| 9199 | return namedclass; |
| 9200 | } |
| 9201 | |
| 9202 | STATIC void |
| 9203 | S_checkposixcc(pTHX_ RExC_state_t *pRExC_state) |
| 9204 | { |
| 9205 | dVAR; |
| 9206 | |
| 9207 | PERL_ARGS_ASSERT_CHECKPOSIXCC; |
| 9208 | |
| 9209 | if (POSIXCC(UCHARAT(RExC_parse))) { |
| 9210 | const char *s = RExC_parse; |
| 9211 | const char c = *s++; |
| 9212 | |
| 9213 | while (isALNUM(*s)) |
| 9214 | s++; |
| 9215 | if (*s && c == *s && s[1] == ']') { |
| 9216 | ckWARN3reg(s+2, |
| 9217 | "POSIX syntax [%c %c] belongs inside character classes", |
| 9218 | c, c); |
| 9219 | |
| 9220 | /* [[=foo=]] and [[.foo.]] are still future. */ |
| 9221 | if (POSIXCC_NOTYET(c)) { |
| 9222 | /* adjust RExC_parse so the error shows after |
| 9223 | the class closes */ |
| 9224 | while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']') |
| 9225 | NOOP; |
| 9226 | Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c); |
| 9227 | } |
| 9228 | } |
| 9229 | } |
| 9230 | } |
| 9231 | |
| 9232 | /* No locale test, and always Unicode semantics */ |
| 9233 | #define _C_C_T_NOLOC_(NAME,TEST,WORD) \ |
| 9234 | ANYOF_##NAME: \ |
| 9235 | for (value = 0; value < 256; value++) \ |
| 9236 | if (TEST) \ |
| 9237 | stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
| 9238 | yesno = '+'; \ |
| 9239 | what = WORD; \ |
| 9240 | break; \ |
| 9241 | case ANYOF_N##NAME: \ |
| 9242 | for (value = 0; value < 256; value++) \ |
| 9243 | if (!TEST) \ |
| 9244 | stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
| 9245 | yesno = '!'; \ |
| 9246 | what = WORD; \ |
| 9247 | break |
| 9248 | |
| 9249 | /* Like the above, but there are differences if we are in uni-8-bit or not, so |
| 9250 | * there are two tests passed in, to use depending on that. There aren't any |
| 9251 | * cases where the label is different from the name, so no need for that |
| 9252 | * parameter */ |
| 9253 | #define _C_C_T_(NAME, TEST_8, TEST_7, WORD) \ |
| 9254 | ANYOF_##NAME: \ |
| 9255 | if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \ |
| 9256 | else if (UNI_SEMANTICS) { \ |
| 9257 | for (value = 0; value < 256; value++) { \ |
| 9258 | if (TEST_8(value)) stored += \ |
| 9259 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
| 9260 | } \ |
| 9261 | } \ |
| 9262 | else { \ |
| 9263 | for (value = 0; value < 128; value++) { \ |
| 9264 | if (TEST_7(UNI_TO_NATIVE(value))) stored += \ |
| 9265 | set_regclass_bit(pRExC_state, ret, \ |
| 9266 | (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \ |
| 9267 | } \ |
| 9268 | } \ |
| 9269 | yesno = '+'; \ |
| 9270 | what = WORD; \ |
| 9271 | break; \ |
| 9272 | case ANYOF_N##NAME: \ |
| 9273 | if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \ |
| 9274 | else if (UNI_SEMANTICS) { \ |
| 9275 | for (value = 0; value < 256; value++) { \ |
| 9276 | if (! TEST_8(value)) stored += \ |
| 9277 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
| 9278 | } \ |
| 9279 | } \ |
| 9280 | else { \ |
| 9281 | for (value = 0; value < 128; value++) { \ |
| 9282 | if (! TEST_7(UNI_TO_NATIVE(value))) stored += set_regclass_bit( \ |
| 9283 | pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \ |
| 9284 | } \ |
| 9285 | if (AT_LEAST_ASCII_RESTRICTED) { \ |
| 9286 | for (value = 128; value < 256; value++) { \ |
| 9287 | stored += set_regclass_bit( \ |
| 9288 | pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \ |
| 9289 | } \ |
| 9290 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; \ |
| 9291 | } \ |
| 9292 | else { \ |
| 9293 | /* For a non-ut8 target string with DEPENDS semantics, all above \ |
| 9294 | * ASCII Latin1 code points match the complement of any of the \ |
| 9295 | * classes. But in utf8, they have their Unicode semantics, so \ |
| 9296 | * can't just set them in the bitmap, or else regexec.c will think \ |
| 9297 | * they matched when they shouldn't. */ \ |
| 9298 | ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; \ |
| 9299 | } \ |
| 9300 | } \ |
| 9301 | yesno = '!'; \ |
| 9302 | what = WORD; \ |
| 9303 | break |
| 9304 | |
| 9305 | STATIC U8 |
| 9306 | S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr) |
| 9307 | { |
| 9308 | |
| 9309 | /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes. |
| 9310 | * Locale folding is done at run-time, so this function should not be |
| 9311 | * called for nodes that are for locales. |
| 9312 | * |
| 9313 | * This function sets the bit corresponding to the fold of the input |
| 9314 | * 'value', if not already set. The fold of 'f' is 'F', and the fold of |
| 9315 | * 'F' is 'f'. |
| 9316 | * |
| 9317 | * It also knows about the characters that are in the bitmap that have |
| 9318 | * folds that are matchable only outside it, and sets the appropriate lists |
| 9319 | * and flags. |
| 9320 | * |
| 9321 | * It returns the number of bits that actually changed from 0 to 1 */ |
| 9322 | |
| 9323 | U8 stored = 0; |
| 9324 | U8 fold; |
| 9325 | |
| 9326 | PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD; |
| 9327 | |
| 9328 | fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value] |
| 9329 | : PL_fold[value]; |
| 9330 | |
| 9331 | /* It assumes the bit for 'value' has already been set */ |
| 9332 | if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) { |
| 9333 | ANYOF_BITMAP_SET(node, fold); |
| 9334 | stored++; |
| 9335 | } |
| 9336 | if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) { |
| 9337 | /* Certain Latin1 characters have matches outside the bitmap. To get |
| 9338 | * here, 'value' is one of those characters. None of these matches is |
| 9339 | * valid for ASCII characters under /aa, which have been excluded by |
| 9340 | * the 'if' above. The matches fall into three categories: |
| 9341 | * 1) They are singly folded-to or -from an above 255 character, as |
| 9342 | * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y |
| 9343 | * WITH DIAERESIS; |
| 9344 | * 2) They are part of a multi-char fold with another character in the |
| 9345 | * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill; |
| 9346 | * 3) They are part of a multi-char fold with a character not in the |
| 9347 | * bitmap, such as various ligatures. |
| 9348 | * We aren't dealing fully with multi-char folds, except we do deal |
| 9349 | * with the pattern containing a character that has a multi-char fold |
| 9350 | * (not so much the inverse). |
| 9351 | * For types 1) and 3), the matches only happen when the target string |
| 9352 | * is utf8; that's not true for 2), and we set a flag for it. |
| 9353 | * |
| 9354 | * The code below adds to the passed in inversion list the single fold |
| 9355 | * closures for 'value'. The values are hard-coded here so that an |
| 9356 | * innocent-looking character class, like /[ks]/i won't have to go out |
| 9357 | * to disk to find the possible matches. XXX It would be better to |
| 9358 | * generate these via regen, in case a new version of the Unicode |
| 9359 | * standard adds new mappings, though that is not really likely. */ |
| 9360 | switch (value) { |
| 9361 | case 'k': |
| 9362 | case 'K': |
| 9363 | /* KELVIN SIGN */ |
| 9364 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A); |
| 9365 | break; |
| 9366 | case 's': |
| 9367 | case 'S': |
| 9368 | /* LATIN SMALL LETTER LONG S */ |
| 9369 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F); |
| 9370 | break; |
| 9371 | case MICRO_SIGN: |
| 9372 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, |
| 9373 | GREEK_SMALL_LETTER_MU); |
| 9374 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, |
| 9375 | GREEK_CAPITAL_LETTER_MU); |
| 9376 | break; |
| 9377 | case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE: |
| 9378 | case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE: |
| 9379 | /* ANGSTROM SIGN */ |
| 9380 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B); |
| 9381 | if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */ |
| 9382 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, |
| 9383 | PL_fold_latin1[value]); |
| 9384 | } |
| 9385 | break; |
| 9386 | case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: |
| 9387 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, |
| 9388 | LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS); |
| 9389 | break; |
| 9390 | case LATIN_SMALL_LETTER_SHARP_S: |
| 9391 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, |
| 9392 | LATIN_CAPITAL_LETTER_SHARP_S); |
| 9393 | |
| 9394 | /* Under /a, /d, and /u, this can match the two chars "ss" */ |
| 9395 | if (! MORE_ASCII_RESTRICTED) { |
| 9396 | add_alternate(alternate_ptr, (U8 *) "ss", 2); |
| 9397 | |
| 9398 | /* And under /u or /a, it can match even if the target is |
| 9399 | * not utf8 */ |
| 9400 | if (AT_LEAST_UNI_SEMANTICS) { |
| 9401 | ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8; |
| 9402 | } |
| 9403 | } |
| 9404 | break; |
| 9405 | case 'F': case 'f': |
| 9406 | case 'I': case 'i': |
| 9407 | case 'L': case 'l': |
| 9408 | case 'T': case 't': |
| 9409 | case 'A': case 'a': |
| 9410 | case 'H': case 'h': |
| 9411 | case 'J': case 'j': |
| 9412 | case 'N': case 'n': |
| 9413 | case 'W': case 'w': |
| 9414 | case 'Y': case 'y': |
| 9415 | /* These all are targets of multi-character folds from code |
| 9416 | * points that require UTF8 to express, so they can't match |
| 9417 | * unless the target string is in UTF-8, so no action here is |
| 9418 | * necessary, as regexec.c properly handles the general case |
| 9419 | * for UTF-8 matching */ |
| 9420 | break; |
| 9421 | default: |
| 9422 | /* Use deprecated warning to increase the chances of this |
| 9423 | * being output */ |
| 9424 | ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value); |
| 9425 | break; |
| 9426 | } |
| 9427 | } |
| 9428 | else if (DEPENDS_SEMANTICS |
| 9429 | && ! isASCII(value) |
| 9430 | && PL_fold_latin1[value] != value) |
| 9431 | { |
| 9432 | /* Under DEPENDS rules, non-ASCII Latin1 characters match their |
| 9433 | * folds only when the target string is in UTF-8. We add the fold |
| 9434 | * here to the list of things to match outside the bitmap, which |
| 9435 | * won't be looked at unless it is UTF8 (or else if something else |
| 9436 | * says to look even if not utf8, but those things better not happen |
| 9437 | * under DEPENDS semantics. */ |
| 9438 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]); |
| 9439 | } |
| 9440 | |
| 9441 | return stored; |
| 9442 | } |
| 9443 | |
| 9444 | |
| 9445 | PERL_STATIC_INLINE U8 |
| 9446 | S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr) |
| 9447 | { |
| 9448 | /* This inline function sets a bit in the bitmap if not already set, and if |
| 9449 | * appropriate, its fold, returning the number of bits that actually |
| 9450 | * changed from 0 to 1 */ |
| 9451 | |
| 9452 | U8 stored; |
| 9453 | |
| 9454 | PERL_ARGS_ASSERT_SET_REGCLASS_BIT; |
| 9455 | |
| 9456 | if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */ |
| 9457 | return 0; |
| 9458 | } |
| 9459 | |
| 9460 | ANYOF_BITMAP_SET(node, value); |
| 9461 | stored = 1; |
| 9462 | |
| 9463 | if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */ |
| 9464 | stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr); |
| 9465 | } |
| 9466 | |
| 9467 | return stored; |
| 9468 | } |
| 9469 | |
| 9470 | STATIC void |
| 9471 | S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len) |
| 9472 | { |
| 9473 | /* Adds input 'string' with length 'len' to the ANYOF node's unicode |
| 9474 | * alternate list, pointed to by 'alternate_ptr'. This is an array of |
| 9475 | * the multi-character folds of characters in the node */ |
| 9476 | SV *sv; |
| 9477 | |
| 9478 | PERL_ARGS_ASSERT_ADD_ALTERNATE; |
| 9479 | |
| 9480 | if (! *alternate_ptr) { |
| 9481 | *alternate_ptr = newAV(); |
| 9482 | } |
| 9483 | sv = newSVpvn_utf8((char*)string, len, TRUE); |
| 9484 | av_push(*alternate_ptr, sv); |
| 9485 | return; |
| 9486 | } |
| 9487 | |
| 9488 | /* |
| 9489 | parse a class specification and produce either an ANYOF node that |
| 9490 | matches the pattern or perhaps will be optimized into an EXACTish node |
| 9491 | instead. The node contains a bit map for the first 256 characters, with the |
| 9492 | corresponding bit set if that character is in the list. For characters |
| 9493 | above 255, a range list is used */ |
| 9494 | |
| 9495 | STATIC regnode * |
| 9496 | S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth) |
| 9497 | { |
| 9498 | dVAR; |
| 9499 | register UV nextvalue; |
| 9500 | register IV prevvalue = OOB_UNICODE; |
| 9501 | register IV range = 0; |
| 9502 | UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */ |
| 9503 | register regnode *ret; |
| 9504 | STRLEN numlen; |
| 9505 | IV namedclass; |
| 9506 | char *rangebegin = NULL; |
| 9507 | bool need_class = 0; |
| 9508 | bool allow_full_fold = TRUE; /* Assume wants multi-char folding */ |
| 9509 | SV *listsv = NULL; |
| 9510 | STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more |
| 9511 | than just initialized. */ |
| 9512 | UV n; |
| 9513 | |
| 9514 | /* code points this node matches that can't be stored in the bitmap */ |
| 9515 | HV* nonbitmap = NULL; |
| 9516 | |
| 9517 | /* The items that are to match that aren't stored in the bitmap, but are a |
| 9518 | * result of things that are stored there. This is the fold closure of |
| 9519 | * such a character, either because it has DEPENDS semantics and shouldn't |
| 9520 | * be matched unless the target string is utf8, or is a code point that is |
| 9521 | * too large for the bit map, as for example, the fold of the MICRO SIGN is |
| 9522 | * above 255. This all is solely for performance reasons. By having this |
| 9523 | * code know the outside-the-bitmap folds that the bitmapped characters are |
| 9524 | * involved with, we don't have to go out to disk to find the list of |
| 9525 | * matches, unless the character class includes code points that aren't |
| 9526 | * storable in the bit map. That means that a character class with an 's' |
| 9527 | * in it, for example, doesn't need to go out to disk to find everything |
| 9528 | * that matches. A 2nd list is used so that the 'nonbitmap' list is kept |
| 9529 | * empty unless there is something whose fold we don't know about, and will |
| 9530 | * have to go out to the disk to find. */ |
| 9531 | HV* l1_fold_invlist = NULL; |
| 9532 | |
| 9533 | /* List of multi-character folds that are matched by this node */ |
| 9534 | AV* unicode_alternate = NULL; |
| 9535 | #ifdef EBCDIC |
| 9536 | UV literal_endpoint = 0; |
| 9537 | #endif |
| 9538 | UV stored = 0; /* how many chars stored in the bitmap */ |
| 9539 | |
| 9540 | regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in |
| 9541 | case we need to change the emitted regop to an EXACT. */ |
| 9542 | const char * orig_parse = RExC_parse; |
| 9543 | GET_RE_DEBUG_FLAGS_DECL; |
| 9544 | |
| 9545 | PERL_ARGS_ASSERT_REGCLASS; |
| 9546 | #ifndef DEBUGGING |
| 9547 | PERL_UNUSED_ARG(depth); |
| 9548 | #endif |
| 9549 | |
| 9550 | DEBUG_PARSE("clas"); |
| 9551 | |
| 9552 | /* Assume we are going to generate an ANYOF node. */ |
| 9553 | ret = reganode(pRExC_state, ANYOF, 0); |
| 9554 | |
| 9555 | |
| 9556 | if (!SIZE_ONLY) { |
| 9557 | ANYOF_FLAGS(ret) = 0; |
| 9558 | } |
| 9559 | |
| 9560 | if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */ |
| 9561 | RExC_naughty++; |
| 9562 | RExC_parse++; |
| 9563 | if (!SIZE_ONLY) |
| 9564 | ANYOF_FLAGS(ret) |= ANYOF_INVERT; |
| 9565 | |
| 9566 | /* We have decided to not allow multi-char folds in inverted character |
| 9567 | * classes, due to the confusion that can happen, especially with |
| 9568 | * classes that are designed for a non-Unicode world: You have the |
| 9569 | * peculiar case that: |
| 9570 | "s s" =~ /^[^\xDF]+$/i => Y |
| 9571 | "ss" =~ /^[^\xDF]+$/i => N |
| 9572 | * |
| 9573 | * See [perl #89750] */ |
| 9574 | allow_full_fold = FALSE; |
| 9575 | } |
| 9576 | |
| 9577 | if (SIZE_ONLY) { |
| 9578 | RExC_size += ANYOF_SKIP; |
| 9579 | listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */ |
| 9580 | } |
| 9581 | else { |
| 9582 | RExC_emit += ANYOF_SKIP; |
| 9583 | if (LOC) { |
| 9584 | ANYOF_FLAGS(ret) |= ANYOF_LOCALE; |
| 9585 | } |
| 9586 | ANYOF_BITMAP_ZERO(ret); |
| 9587 | listsv = newSVpvs("# comment\n"); |
| 9588 | initial_listsv_len = SvCUR(listsv); |
| 9589 | } |
| 9590 | |
| 9591 | nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0; |
| 9592 | |
| 9593 | if (!SIZE_ONLY && POSIXCC(nextvalue)) |
| 9594 | checkposixcc(pRExC_state); |
| 9595 | |
| 9596 | /* allow 1st char to be ] (allowing it to be - is dealt with later) */ |
| 9597 | if (UCHARAT(RExC_parse) == ']') |
| 9598 | goto charclassloop; |
| 9599 | |
| 9600 | parseit: |
| 9601 | while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') { |
| 9602 | |
| 9603 | charclassloop: |
| 9604 | |
| 9605 | namedclass = OOB_NAMEDCLASS; /* initialize as illegal */ |
| 9606 | |
| 9607 | if (!range) |
| 9608 | rangebegin = RExC_parse; |
| 9609 | if (UTF) { |
| 9610 | value = utf8n_to_uvchr((U8*)RExC_parse, |
| 9611 | RExC_end - RExC_parse, |
| 9612 | &numlen, UTF8_ALLOW_DEFAULT); |
| 9613 | RExC_parse += numlen; |
| 9614 | } |
| 9615 | else |
| 9616 | value = UCHARAT(RExC_parse++); |
| 9617 | |
| 9618 | nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0; |
| 9619 | if (value == '[' && POSIXCC(nextvalue)) |
| 9620 | namedclass = regpposixcc(pRExC_state, value); |
| 9621 | else if (value == '\\') { |
| 9622 | if (UTF) { |
| 9623 | value = utf8n_to_uvchr((U8*)RExC_parse, |
| 9624 | RExC_end - RExC_parse, |
| 9625 | &numlen, UTF8_ALLOW_DEFAULT); |
| 9626 | RExC_parse += numlen; |
| 9627 | } |
| 9628 | else |
| 9629 | value = UCHARAT(RExC_parse++); |
| 9630 | /* Some compilers cannot handle switching on 64-bit integer |
| 9631 | * values, therefore value cannot be an UV. Yes, this will |
| 9632 | * be a problem later if we want switch on Unicode. |
| 9633 | * A similar issue a little bit later when switching on |
| 9634 | * namedclass. --jhi */ |
| 9635 | switch ((I32)value) { |
| 9636 | case 'w': namedclass = ANYOF_ALNUM; break; |
| 9637 | case 'W': namedclass = ANYOF_NALNUM; break; |
| 9638 | case 's': namedclass = ANYOF_SPACE; break; |
| 9639 | case 'S': namedclass = ANYOF_NSPACE; break; |
| 9640 | case 'd': namedclass = ANYOF_DIGIT; break; |
| 9641 | case 'D': namedclass = ANYOF_NDIGIT; break; |
| 9642 | case 'v': namedclass = ANYOF_VERTWS; break; |
| 9643 | case 'V': namedclass = ANYOF_NVERTWS; break; |
| 9644 | case 'h': namedclass = ANYOF_HORIZWS; break; |
| 9645 | case 'H': namedclass = ANYOF_NHORIZWS; break; |
| 9646 | case 'N': /* Handle \N{NAME} in class */ |
| 9647 | { |
| 9648 | /* We only pay attention to the first char of |
| 9649 | multichar strings being returned. I kinda wonder |
| 9650 | if this makes sense as it does change the behaviour |
| 9651 | from earlier versions, OTOH that behaviour was broken |
| 9652 | as well. */ |
| 9653 | UV v; /* value is register so we cant & it /grrr */ |
| 9654 | if (reg_namedseq(pRExC_state, &v, NULL, depth)) { |
| 9655 | goto parseit; |
| 9656 | } |
| 9657 | value= v; |
| 9658 | } |
| 9659 | break; |
| 9660 | case 'p': |
| 9661 | case 'P': |
| 9662 | { |
| 9663 | char *e; |
| 9664 | if (RExC_parse >= RExC_end) |
| 9665 | vFAIL2("Empty \\%c{}", (U8)value); |
| 9666 | if (*RExC_parse == '{') { |
| 9667 | const U8 c = (U8)value; |
| 9668 | e = strchr(RExC_parse++, '}'); |
| 9669 | if (!e) |
| 9670 | vFAIL2("Missing right brace on \\%c{}", c); |
| 9671 | while (isSPACE(UCHARAT(RExC_parse))) |
| 9672 | RExC_parse++; |
| 9673 | if (e == RExC_parse) |
| 9674 | vFAIL2("Empty \\%c{}", c); |
| 9675 | n = e - RExC_parse; |
| 9676 | while (isSPACE(UCHARAT(RExC_parse + n - 1))) |
| 9677 | n--; |
| 9678 | } |
| 9679 | else { |
| 9680 | e = RExC_parse; |
| 9681 | n = 1; |
| 9682 | } |
| 9683 | if (!SIZE_ONLY) { |
| 9684 | if (UCHARAT(RExC_parse) == '^') { |
| 9685 | RExC_parse++; |
| 9686 | n--; |
| 9687 | value = value == 'p' ? 'P' : 'p'; /* toggle */ |
| 9688 | while (isSPACE(UCHARAT(RExC_parse))) { |
| 9689 | RExC_parse++; |
| 9690 | n--; |
| 9691 | } |
| 9692 | } |
| 9693 | |
| 9694 | /* Add the property name to the list. If /i matching, give |
| 9695 | * a different name which consists of the normal name |
| 9696 | * sandwiched between two underscores and '_i'. The design |
| 9697 | * is discussed in the commit message for this. */ |
| 9698 | Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s%.*s%s\n", |
| 9699 | (value=='p' ? '+' : '!'), |
| 9700 | (FOLD) ? "__" : "", |
| 9701 | (int)n, |
| 9702 | RExC_parse, |
| 9703 | (FOLD) ? "_i" : "" |
| 9704 | ); |
| 9705 | } |
| 9706 | RExC_parse = e + 1; |
| 9707 | |
| 9708 | /* The \p could match something in the Latin1 range, hence |
| 9709 | * something that isn't utf8 */ |
| 9710 | ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8; |
| 9711 | namedclass = ANYOF_MAX; /* no official name, but it's named */ |
| 9712 | |
| 9713 | /* \p means they want Unicode semantics */ |
| 9714 | RExC_uni_semantics = 1; |
| 9715 | } |
| 9716 | break; |
| 9717 | case 'n': value = '\n'; break; |
| 9718 | case 'r': value = '\r'; break; |
| 9719 | case 't': value = '\t'; break; |
| 9720 | case 'f': value = '\f'; break; |
| 9721 | case 'b': value = '\b'; break; |
| 9722 | case 'e': value = ASCII_TO_NATIVE('\033');break; |
| 9723 | case 'a': value = ASCII_TO_NATIVE('\007');break; |
| 9724 | case 'o': |
| 9725 | RExC_parse--; /* function expects to be pointed at the 'o' */ |
| 9726 | { |
| 9727 | const char* error_msg; |
| 9728 | bool valid = grok_bslash_o(RExC_parse, |
| 9729 | &value, |
| 9730 | &numlen, |
| 9731 | &error_msg, |
| 9732 | SIZE_ONLY); |
| 9733 | RExC_parse += numlen; |
| 9734 | if (! valid) { |
| 9735 | vFAIL(error_msg); |
| 9736 | } |
| 9737 | } |
| 9738 | if (PL_encoding && value < 0x100) { |
| 9739 | goto recode_encoding; |
| 9740 | } |
| 9741 | break; |
| 9742 | case 'x': |
| 9743 | if (*RExC_parse == '{') { |
| 9744 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES |
| 9745 | | PERL_SCAN_DISALLOW_PREFIX; |
| 9746 | char * const e = strchr(RExC_parse++, '}'); |
| 9747 | if (!e) |
| 9748 | vFAIL("Missing right brace on \\x{}"); |
| 9749 | |
| 9750 | numlen = e - RExC_parse; |
| 9751 | value = grok_hex(RExC_parse, &numlen, &flags, NULL); |
| 9752 | RExC_parse = e + 1; |
| 9753 | } |
| 9754 | else { |
| 9755 | I32 flags = PERL_SCAN_DISALLOW_PREFIX; |
| 9756 | numlen = 2; |
| 9757 | value = grok_hex(RExC_parse, &numlen, &flags, NULL); |
| 9758 | RExC_parse += numlen; |
| 9759 | } |
| 9760 | if (PL_encoding && value < 0x100) |
| 9761 | goto recode_encoding; |
| 9762 | break; |
| 9763 | case 'c': |
| 9764 | value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY); |
| 9765 | break; |
| 9766 | case '0': case '1': case '2': case '3': case '4': |
| 9767 | case '5': case '6': case '7': |
| 9768 | { |
| 9769 | /* Take 1-3 octal digits */ |
| 9770 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT; |
| 9771 | numlen = 3; |
| 9772 | value = grok_oct(--RExC_parse, &numlen, &flags, NULL); |
| 9773 | RExC_parse += numlen; |
| 9774 | if (PL_encoding && value < 0x100) |
| 9775 | goto recode_encoding; |
| 9776 | break; |
| 9777 | } |
| 9778 | recode_encoding: |
| 9779 | if (! RExC_override_recoding) { |
| 9780 | SV* enc = PL_encoding; |
| 9781 | value = reg_recode((const char)(U8)value, &enc); |
| 9782 | if (!enc && SIZE_ONLY) |
| 9783 | ckWARNreg(RExC_parse, |
| 9784 | "Invalid escape in the specified encoding"); |
| 9785 | break; |
| 9786 | } |
| 9787 | default: |
| 9788 | /* Allow \_ to not give an error */ |
| 9789 | if (!SIZE_ONLY && isALNUM(value) && value != '_') { |
| 9790 | ckWARN2reg(RExC_parse, |
| 9791 | "Unrecognized escape \\%c in character class passed through", |
| 9792 | (int)value); |
| 9793 | } |
| 9794 | break; |
| 9795 | } |
| 9796 | } /* end of \blah */ |
| 9797 | #ifdef EBCDIC |
| 9798 | else |
| 9799 | literal_endpoint++; |
| 9800 | #endif |
| 9801 | |
| 9802 | if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */ |
| 9803 | |
| 9804 | /* What matches in a locale is not known until runtime, so need to |
| 9805 | * (one time per class) allocate extra space to pass to regexec. |
| 9806 | * The space will contain a bit for each named class that is to be |
| 9807 | * matched against. This isn't needed for \p{} and pseudo-classes, |
| 9808 | * as they are not affected by locale, and hence are dealt with |
| 9809 | * separately */ |
| 9810 | if (LOC && namedclass < ANYOF_MAX && ! need_class) { |
| 9811 | need_class = 1; |
| 9812 | if (SIZE_ONLY) { |
| 9813 | RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP; |
| 9814 | } |
| 9815 | else { |
| 9816 | RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP; |
| 9817 | ANYOF_CLASS_ZERO(ret); |
| 9818 | } |
| 9819 | ANYOF_FLAGS(ret) |= ANYOF_CLASS; |
| 9820 | } |
| 9821 | |
| 9822 | /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a |
| 9823 | * literal, as is the character that began the false range, i.e. |
| 9824 | * the 'a' in the examples */ |
| 9825 | if (range) { |
| 9826 | if (!SIZE_ONLY) { |
| 9827 | const int w = |
| 9828 | RExC_parse >= rangebegin ? |
| 9829 | RExC_parse - rangebegin : 0; |
| 9830 | ckWARN4reg(RExC_parse, |
| 9831 | "False [] range \"%*.*s\"", |
| 9832 | w, w, rangebegin); |
| 9833 | |
| 9834 | stored += |
| 9835 | set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate); |
| 9836 | if (prevvalue < 256) { |
| 9837 | stored += |
| 9838 | set_regclass_bit(pRExC_state, ret, (U8) prevvalue, &l1_fold_invlist, &unicode_alternate); |
| 9839 | } |
| 9840 | else { |
| 9841 | nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue); |
| 9842 | } |
| 9843 | } |
| 9844 | |
| 9845 | range = 0; /* this was not a true range */ |
| 9846 | } |
| 9847 | |
| 9848 | |
| 9849 | |
| 9850 | if (!SIZE_ONLY) { |
| 9851 | const char *what = NULL; |
| 9852 | char yesno = 0; |
| 9853 | |
| 9854 | /* Possible truncation here but in some 64-bit environments |
| 9855 | * the compiler gets heartburn about switch on 64-bit values. |
| 9856 | * A similar issue a little earlier when switching on value. |
| 9857 | * --jhi */ |
| 9858 | switch ((I32)namedclass) { |
| 9859 | |
| 9860 | case _C_C_T_(ALNUMC, isALNUMC_L1, isALNUMC, "XPosixAlnum"); |
| 9861 | case _C_C_T_(ALPHA, isALPHA_L1, isALPHA, "XPosixAlpha"); |
| 9862 | case _C_C_T_(BLANK, isBLANK_L1, isBLANK, "XPosixBlank"); |
| 9863 | case _C_C_T_(CNTRL, isCNTRL_L1, isCNTRL, "XPosixCntrl"); |
| 9864 | case _C_C_T_(GRAPH, isGRAPH_L1, isGRAPH, "XPosixGraph"); |
| 9865 | case _C_C_T_(LOWER, isLOWER_L1, isLOWER, "XPosixLower"); |
| 9866 | case _C_C_T_(PRINT, isPRINT_L1, isPRINT, "XPosixPrint"); |
| 9867 | case _C_C_T_(PSXSPC, isPSXSPC_L1, isPSXSPC, "XPosixSpace"); |
| 9868 | case _C_C_T_(PUNCT, isPUNCT_L1, isPUNCT, "XPosixPunct"); |
| 9869 | case _C_C_T_(UPPER, isUPPER_L1, isUPPER, "XPosixUpper"); |
| 9870 | /* \s, \w match all unicode if utf8. */ |
| 9871 | case _C_C_T_(SPACE, isSPACE_L1, isSPACE, "SpacePerl"); |
| 9872 | case _C_C_T_(ALNUM, isWORDCHAR_L1, isALNUM, "Word"); |
| 9873 | case _C_C_T_(XDIGIT, isXDIGIT_L1, isXDIGIT, "XPosixXDigit"); |
| 9874 | case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace"); |
| 9875 | case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace"); |
| 9876 | case ANYOF_ASCII: |
| 9877 | if (LOC) |
| 9878 | ANYOF_CLASS_SET(ret, ANYOF_ASCII); |
| 9879 | else { |
| 9880 | for (value = 0; value < 128; value++) |
| 9881 | stored += |
| 9882 | set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); |
| 9883 | } |
| 9884 | yesno = '+'; |
| 9885 | what = NULL; /* Doesn't match outside ascii, so |
| 9886 | don't want to add +utf8:: */ |
| 9887 | break; |
| 9888 | case ANYOF_NASCII: |
| 9889 | if (LOC) |
| 9890 | ANYOF_CLASS_SET(ret, ANYOF_NASCII); |
| 9891 | else { |
| 9892 | for (value = 128; value < 256; value++) |
| 9893 | stored += |
| 9894 | set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); |
| 9895 | } |
| 9896 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; |
| 9897 | yesno = '!'; |
| 9898 | what = "ASCII"; |
| 9899 | break; |
| 9900 | case ANYOF_DIGIT: |
| 9901 | if (LOC) |
| 9902 | ANYOF_CLASS_SET(ret, ANYOF_DIGIT); |
| 9903 | else { |
| 9904 | /* consecutive digits assumed */ |
| 9905 | for (value = '0'; value <= '9'; value++) |
| 9906 | stored += |
| 9907 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); |
| 9908 | } |
| 9909 | yesno = '+'; |
| 9910 | what = "Digit"; |
| 9911 | break; |
| 9912 | case ANYOF_NDIGIT: |
| 9913 | if (LOC) |
| 9914 | ANYOF_CLASS_SET(ret, ANYOF_NDIGIT); |
| 9915 | else { |
| 9916 | /* consecutive digits assumed */ |
| 9917 | for (value = 0; value < '0'; value++) |
| 9918 | stored += |
| 9919 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); |
| 9920 | for (value = '9' + 1; value < 256; value++) |
| 9921 | stored += |
| 9922 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); |
| 9923 | } |
| 9924 | yesno = '!'; |
| 9925 | what = "Digit"; |
| 9926 | if (AT_LEAST_ASCII_RESTRICTED ) { |
| 9927 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; |
| 9928 | } |
| 9929 | break; |
| 9930 | case ANYOF_MAX: |
| 9931 | /* this is to handle \p and \P */ |
| 9932 | break; |
| 9933 | default: |
| 9934 | vFAIL("Invalid [::] class"); |
| 9935 | break; |
| 9936 | } |
| 9937 | if (what && ! (AT_LEAST_ASCII_RESTRICTED)) { |
| 9938 | /* Strings such as "+utf8::isWord\n" */ |
| 9939 | Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what); |
| 9940 | } |
| 9941 | |
| 9942 | continue; |
| 9943 | } |
| 9944 | } /* end of namedclass \blah */ |
| 9945 | |
| 9946 | if (range) { |
| 9947 | if (prevvalue > (IV)value) /* b-a */ { |
| 9948 | const int w = RExC_parse - rangebegin; |
| 9949 | Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin); |
| 9950 | range = 0; /* not a valid range */ |
| 9951 | } |
| 9952 | } |
| 9953 | else { |
| 9954 | prevvalue = value; /* save the beginning of the range */ |
| 9955 | if (RExC_parse+1 < RExC_end |
| 9956 | && *RExC_parse == '-' |
| 9957 | && RExC_parse[1] != ']') |
| 9958 | { |
| 9959 | RExC_parse++; |
| 9960 | |
| 9961 | /* a bad range like \w-, [:word:]- ? */ |
| 9962 | if (namedclass > OOB_NAMEDCLASS) { |
| 9963 | if (ckWARN(WARN_REGEXP)) { |
| 9964 | const int w = |
| 9965 | RExC_parse >= rangebegin ? |
| 9966 | RExC_parse - rangebegin : 0; |
| 9967 | vWARN4(RExC_parse, |
| 9968 | "False [] range \"%*.*s\"", |
| 9969 | w, w, rangebegin); |
| 9970 | } |
| 9971 | if (!SIZE_ONLY) |
| 9972 | stored += |
| 9973 | set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate); |
| 9974 | } else |
| 9975 | range = 1; /* yeah, it's a range! */ |
| 9976 | continue; /* but do it the next time */ |
| 9977 | } |
| 9978 | } |
| 9979 | |
| 9980 | /* non-Latin1 code point implies unicode semantics. Must be set in |
| 9981 | * pass1 so is there for the whole of pass 2 */ |
| 9982 | if (value > 255) { |
| 9983 | RExC_uni_semantics = 1; |
| 9984 | } |
| 9985 | |
| 9986 | /* now is the next time */ |
| 9987 | if (!SIZE_ONLY) { |
| 9988 | if (prevvalue < 256) { |
| 9989 | const IV ceilvalue = value < 256 ? value : 255; |
| 9990 | IV i; |
| 9991 | #ifdef EBCDIC |
| 9992 | /* In EBCDIC [\x89-\x91] should include |
| 9993 | * the \x8e but [i-j] should not. */ |
| 9994 | if (literal_endpoint == 2 && |
| 9995 | ((isLOWER(prevvalue) && isLOWER(ceilvalue)) || |
| 9996 | (isUPPER(prevvalue) && isUPPER(ceilvalue)))) |
| 9997 | { |
| 9998 | if (isLOWER(prevvalue)) { |
| 9999 | for (i = prevvalue; i <= ceilvalue; i++) |
| 10000 | if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) { |
| 10001 | stored += |
| 10002 | set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate); |
| 10003 | } |
| 10004 | } else { |
| 10005 | for (i = prevvalue; i <= ceilvalue; i++) |
| 10006 | if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) { |
| 10007 | stored += |
| 10008 | set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate); |
| 10009 | } |
| 10010 | } |
| 10011 | } |
| 10012 | else |
| 10013 | #endif |
| 10014 | for (i = prevvalue; i <= ceilvalue; i++) { |
| 10015 | stored += set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate); |
| 10016 | } |
| 10017 | } |
| 10018 | if (value > 255) { |
| 10019 | const UV prevnatvalue = NATIVE_TO_UNI(prevvalue); |
| 10020 | const UV natvalue = NATIVE_TO_UNI(value); |
| 10021 | nonbitmap = add_range_to_invlist(nonbitmap, prevnatvalue, natvalue); |
| 10022 | } |
| 10023 | #ifdef EBCDIC |
| 10024 | literal_endpoint = 0; |
| 10025 | #endif |
| 10026 | } |
| 10027 | |
| 10028 | range = 0; /* this range (if it was one) is done now */ |
| 10029 | } |
| 10030 | |
| 10031 | |
| 10032 | |
| 10033 | if (SIZE_ONLY) |
| 10034 | return ret; |
| 10035 | /****** !SIZE_ONLY AFTER HERE *********/ |
| 10036 | |
| 10037 | /* If folding and there are code points above 255, we calculate all |
| 10038 | * characters that could fold to or from the ones already on the list */ |
| 10039 | if (FOLD && nonbitmap) { |
| 10040 | UV i; |
| 10041 | |
| 10042 | HV* fold_intersection; |
| 10043 | UV* fold_list; |
| 10044 | |
| 10045 | /* This is a list of all the characters that participate in folds |
| 10046 | * (except marks, etc in multi-char folds */ |
| 10047 | if (! PL_utf8_foldable) { |
| 10048 | SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0); |
| 10049 | PL_utf8_foldable = _swash_to_invlist(swash); |
| 10050 | } |
| 10051 | |
| 10052 | /* This is a hash that for a particular fold gives all characters |
| 10053 | * that are involved in it */ |
| 10054 | if (! PL_utf8_foldclosures) { |
| 10055 | |
| 10056 | /* If we were unable to find any folds, then we likely won't be |
| 10057 | * able to find the closures. So just create an empty list. |
| 10058 | * Folding will effectively be restricted to the non-Unicode rules |
| 10059 | * hard-coded into Perl. (This case happens legitimately during |
| 10060 | * compilation of Perl itself before the Unicode tables are |
| 10061 | * generated) */ |
| 10062 | if (invlist_len(PL_utf8_foldable) == 0) { |
| 10063 | PL_utf8_foldclosures = _new_invlist(0); |
| 10064 | } else { |
| 10065 | /* If the folds haven't been read in, call a fold function |
| 10066 | * to force that */ |
| 10067 | if (! PL_utf8_tofold) { |
| 10068 | U8 dummy[UTF8_MAXBYTES+1]; |
| 10069 | STRLEN dummy_len; |
| 10070 | to_utf8_fold((U8*) "A", dummy, &dummy_len); |
| 10071 | } |
| 10072 | PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold); |
| 10073 | } |
| 10074 | } |
| 10075 | |
| 10076 | /* Only the characters in this class that participate in folds need |
| 10077 | * be checked. Get the intersection of this class and all the |
| 10078 | * possible characters that are foldable. This can quickly narrow |
| 10079 | * down a large class */ |
| 10080 | fold_intersection = invlist_intersection(PL_utf8_foldable, nonbitmap); |
| 10081 | |
| 10082 | /* Now look at the foldable characters in this class individually */ |
| 10083 | fold_list = invlist_array(fold_intersection); |
| 10084 | for (i = 0; i < invlist_len(fold_intersection); i++) { |
| 10085 | UV j; |
| 10086 | |
| 10087 | /* The next entry is the beginning of the range that is in the |
| 10088 | * class */ |
| 10089 | UV start = fold_list[i++]; |
| 10090 | |
| 10091 | |
| 10092 | /* The next entry is the beginning of the next range, which |
| 10093 | * isn't in the class, so the end of the current range is one |
| 10094 | * less than that */ |
| 10095 | UV end = fold_list[i] - 1; |
| 10096 | |
| 10097 | /* Look at every character in the range */ |
| 10098 | for (j = start; j <= end; j++) { |
| 10099 | |
| 10100 | /* Get its fold */ |
| 10101 | U8 foldbuf[UTF8_MAXBYTES_CASE+1]; |
| 10102 | STRLEN foldlen; |
| 10103 | const UV f = |
| 10104 | _to_uni_fold_flags(j, foldbuf, &foldlen, allow_full_fold); |
| 10105 | |
| 10106 | if (foldlen > (STRLEN)UNISKIP(f)) { |
| 10107 | |
| 10108 | /* Any multicharacter foldings (disallowed in |
| 10109 | * lookbehind patterns) require the following |
| 10110 | * transform: [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst) where |
| 10111 | * E folds into "pq" and F folds into "rst", all other |
| 10112 | * characters fold to single characters. We save away |
| 10113 | * these multicharacter foldings, to be later saved as |
| 10114 | * part of the additional "s" data. */ |
| 10115 | if (! RExC_in_lookbehind) { |
| 10116 | U8* loc = foldbuf; |
| 10117 | U8* e = foldbuf + foldlen; |
| 10118 | |
| 10119 | /* If any of the folded characters of this are in |
| 10120 | * the Latin1 range, tell the regex engine that |
| 10121 | * this can match a non-utf8 target string. The |
| 10122 | * only multi-byte fold whose source is in the |
| 10123 | * Latin1 range (U+00DF) applies only when the |
| 10124 | * target string is utf8, or under unicode rules */ |
| 10125 | if (j > 255 || AT_LEAST_UNI_SEMANTICS) { |
| 10126 | while (loc < e) { |
| 10127 | |
| 10128 | /* Can't mix ascii with non- under /aa */ |
| 10129 | if (MORE_ASCII_RESTRICTED |
| 10130 | && (isASCII(*loc) != isASCII(j))) |
| 10131 | { |
| 10132 | goto end_multi_fold; |
| 10133 | } |
| 10134 | if (UTF8_IS_INVARIANT(*loc) |
| 10135 | || UTF8_IS_DOWNGRADEABLE_START(*loc)) |
| 10136 | { |
| 10137 | /* Can't mix above and below 256 under |
| 10138 | * LOC */ |
| 10139 | if (LOC) { |
| 10140 | goto end_multi_fold; |
| 10141 | } |
| 10142 | ANYOF_FLAGS(ret) |
| 10143 | |= ANYOF_NONBITMAP_NON_UTF8; |
| 10144 | break; |
| 10145 | } |
| 10146 | loc += UTF8SKIP(loc); |
| 10147 | } |
| 10148 | } |
| 10149 | |
| 10150 | add_alternate(&unicode_alternate, foldbuf, foldlen); |
| 10151 | end_multi_fold: ; |
| 10152 | } |
| 10153 | |
| 10154 | /* This is special-cased, as it is the only letter which |
| 10155 | * has both a multi-fold and single-fold in Latin1. All |
| 10156 | * the other chars that have single and multi-folds are |
| 10157 | * always in utf8, and the utf8 folding algorithm catches |
| 10158 | * them */ |
| 10159 | if (! LOC && j == LATIN_CAPITAL_LETTER_SHARP_S) { |
| 10160 | stored += set_regclass_bit(pRExC_state, |
| 10161 | ret, |
| 10162 | LATIN_SMALL_LETTER_SHARP_S, |
| 10163 | &l1_fold_invlist, &unicode_alternate); |
| 10164 | } |
| 10165 | } |
| 10166 | else { |
| 10167 | /* Single character fold. Add everything in its fold |
| 10168 | * closure to the list that this node should match */ |
| 10169 | SV** listp; |
| 10170 | |
| 10171 | /* The fold closures data structure is a hash with the |
| 10172 | * keys being every character that is folded to, like |
| 10173 | * 'k', and the values each an array of everything that |
| 10174 | * folds to its key. e.g. [ 'k', 'K', KELVIN_SIGN ] */ |
| 10175 | if ((listp = hv_fetch(PL_utf8_foldclosures, |
| 10176 | (char *) foldbuf, foldlen, FALSE))) |
| 10177 | { |
| 10178 | AV* list = (AV*) *listp; |
| 10179 | IV k; |
| 10180 | for (k = 0; k <= av_len(list); k++) { |
| 10181 | SV** c_p = av_fetch(list, k, FALSE); |
| 10182 | UV c; |
| 10183 | if (c_p == NULL) { |
| 10184 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); |
| 10185 | } |
| 10186 | c = SvUV(*c_p); |
| 10187 | |
| 10188 | /* /aa doesn't allow folds between ASCII and |
| 10189 | * non-; /l doesn't allow them between above |
| 10190 | * and below 256 */ |
| 10191 | if ((MORE_ASCII_RESTRICTED |
| 10192 | && (isASCII(c) != isASCII(j))) |
| 10193 | || (LOC && ((c < 256) != (j < 256)))) |
| 10194 | { |
| 10195 | continue; |
| 10196 | } |
| 10197 | |
| 10198 | if (c < 256 && AT_LEAST_UNI_SEMANTICS) { |
| 10199 | stored += set_regclass_bit(pRExC_state, |
| 10200 | ret, |
| 10201 | (U8) c, |
| 10202 | &l1_fold_invlist, &unicode_alternate); |
| 10203 | } |
| 10204 | /* It may be that the code point is already |
| 10205 | * in this range or already in the bitmap, |
| 10206 | * in which case we need do nothing */ |
| 10207 | else if ((c < start || c > end) |
| 10208 | && (c > 255 |
| 10209 | || ! ANYOF_BITMAP_TEST(ret, c))) |
| 10210 | { |
| 10211 | nonbitmap = add_cp_to_invlist(nonbitmap, c); |
| 10212 | } |
| 10213 | } |
| 10214 | } |
| 10215 | } |
| 10216 | } |
| 10217 | } |
| 10218 | invlist_destroy(fold_intersection); |
| 10219 | } |
| 10220 | |
| 10221 | /* Combine the two lists into one. */ |
| 10222 | if (l1_fold_invlist) { |
| 10223 | if (nonbitmap) { |
| 10224 | HV* temp = invlist_union(nonbitmap, l1_fold_invlist); |
| 10225 | invlist_destroy(nonbitmap); |
| 10226 | nonbitmap = temp; |
| 10227 | invlist_destroy(l1_fold_invlist); |
| 10228 | } |
| 10229 | else { |
| 10230 | nonbitmap = l1_fold_invlist; |
| 10231 | } |
| 10232 | } |
| 10233 | |
| 10234 | /* Here, we have calculated what code points should be in the character |
| 10235 | * class. Now we can see about various optimizations. Fold calculation |
| 10236 | * needs to take place before inversion. Otherwise /[^k]/i would invert to |
| 10237 | * include K, which under /i would match k. */ |
| 10238 | |
| 10239 | /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't |
| 10240 | * set the FOLD flag yet, so this this does optimize those. It doesn't |
| 10241 | * optimize locale. Doing so perhaps could be done as long as there is |
| 10242 | * nothing like \w in it; some thought also would have to be given to the |
| 10243 | * interaction with above 0x100 chars */ |
| 10244 | if (! LOC |
| 10245 | && (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT |
| 10246 | && ! unicode_alternate |
| 10247 | && ! nonbitmap |
| 10248 | && SvCUR(listsv) == initial_listsv_len) |
| 10249 | { |
| 10250 | for (value = 0; value < ANYOF_BITMAP_SIZE; ++value) |
| 10251 | ANYOF_BITMAP(ret)[value] ^= 0xFF; |
| 10252 | stored = 256 - stored; |
| 10253 | |
| 10254 | /* The inversion means that everything above 255 is matched; and at the |
| 10255 | * same time we clear the invert flag */ |
| 10256 | ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL; |
| 10257 | } |
| 10258 | |
| 10259 | /* Folding in the bitmap is taken care of above, but not for locale (for |
| 10260 | * which we have to wait to see what folding is in effect at runtime), and |
| 10261 | * for things not in the bitmap. Set run-time fold flag for these */ |
| 10262 | if (FOLD && (LOC || nonbitmap || unicode_alternate)) { |
| 10263 | ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD; |
| 10264 | } |
| 10265 | |
| 10266 | /* A single character class can be "optimized" into an EXACTish node. |
| 10267 | * Note that since we don't currently count how many characters there are |
| 10268 | * outside the bitmap, we are XXX missing optimization possibilities for |
| 10269 | * them. This optimization can't happen unless this is a truly single |
| 10270 | * character class, which means that it can't be an inversion into a |
| 10271 | * many-character class, and there must be no possibility of there being |
| 10272 | * things outside the bitmap. 'stored' (only) for locales doesn't include |
| 10273 | * \w, etc, so have to make a special test that they aren't present |
| 10274 | * |
| 10275 | * Similarly A 2-character class of the very special form like [bB] can be |
| 10276 | * optimized into an EXACTFish node, but only for non-locales, and for |
| 10277 | * characters which only have the two folds; so things like 'fF' and 'Ii' |
| 10278 | * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE |
| 10279 | * FI'. */ |
| 10280 | if (! nonbitmap |
| 10281 | && ! unicode_alternate |
| 10282 | && SvCUR(listsv) == initial_listsv_len |
| 10283 | && ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL)) |
| 10284 | && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE)) |
| 10285 | || (! ANYOF_CLASS_TEST_ANY_SET(ret))))) |
| 10286 | || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE)) |
| 10287 | && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value)) |
| 10288 | /* If the latest code point has a fold whose |
| 10289 | * bit is set, it must be the only other one */ |
| 10290 | && ((prevvalue = PL_fold_latin1[value]) != (IV)value) |
| 10291 | && ANYOF_BITMAP_TEST(ret, prevvalue))))) |
| 10292 | { |
| 10293 | /* Note that the information needed to decide to do this optimization |
| 10294 | * is not currently available until the 2nd pass, and that the actually |
| 10295 | * used EXACTish node takes less space than the calculated ANYOF node, |
| 10296 | * and hence the amount of space calculated in the first pass is larger |
| 10297 | * than actually used, so this optimization doesn't gain us any space. |
| 10298 | * But an EXACT node is faster than an ANYOF node, and can be combined |
| 10299 | * with any adjacent EXACT nodes later by the optimizer for further |
| 10300 | * gains. The speed of executing an EXACTF is similar to an ANYOF |
| 10301 | * node, so the optimization advantage comes from the ability to join |
| 10302 | * it to adjacent EXACT nodes */ |
| 10303 | |
| 10304 | const char * cur_parse= RExC_parse; |
| 10305 | U8 op; |
| 10306 | RExC_emit = (regnode *)orig_emit; |
| 10307 | RExC_parse = (char *)orig_parse; |
| 10308 | |
| 10309 | if (stored == 1) { |
| 10310 | |
| 10311 | /* A locale node with one point can be folded; all the other cases |
| 10312 | * with folding will have two points, since we calculate them above |
| 10313 | */ |
| 10314 | if (ANYOF_FLAGS(ret) & ANYOF_LOC_NONBITMAP_FOLD) { |
| 10315 | op = EXACTFL; |
| 10316 | } |
| 10317 | else { |
| 10318 | op = EXACT; |
| 10319 | } |
| 10320 | } /* else 2 chars in the bit map: the folds of each other */ |
| 10321 | else if (AT_LEAST_UNI_SEMANTICS || !isASCII(value)) { |
| 10322 | |
| 10323 | /* To join adjacent nodes, they must be the exact EXACTish type. |
| 10324 | * Try to use the most likely type, by using EXACTFU if the regex |
| 10325 | * calls for them, or is required because the character is |
| 10326 | * non-ASCII */ |
| 10327 | op = EXACTFU; |
| 10328 | } |
| 10329 | else { /* Otherwise, more likely to be EXACTF type */ |
| 10330 | op = EXACTF; |
| 10331 | } |
| 10332 | |
| 10333 | ret = reg_node(pRExC_state, op); |
| 10334 | RExC_parse = (char *)cur_parse; |
| 10335 | if (UTF && ! NATIVE_IS_INVARIANT(value)) { |
| 10336 | *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value); |
| 10337 | *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value); |
| 10338 | STR_LEN(ret)= 2; |
| 10339 | RExC_emit += STR_SZ(2); |
| 10340 | } |
| 10341 | else { |
| 10342 | *STRING(ret)= (char)value; |
| 10343 | STR_LEN(ret)= 1; |
| 10344 | RExC_emit += STR_SZ(1); |
| 10345 | } |
| 10346 | SvREFCNT_dec(listsv); |
| 10347 | return ret; |
| 10348 | } |
| 10349 | |
| 10350 | if (nonbitmap) { |
| 10351 | UV* nonbitmap_array = invlist_array(nonbitmap); |
| 10352 | UV nonbitmap_len = invlist_len(nonbitmap); |
| 10353 | UV i; |
| 10354 | |
| 10355 | /* Here have the full list of items to match that aren't in the |
| 10356 | * bitmap. Convert to the structure that the rest of the code is |
| 10357 | * expecting. XXX That rest of the code should convert to this |
| 10358 | * structure */ |
| 10359 | for (i = 0; i < nonbitmap_len; i++) { |
| 10360 | |
| 10361 | /* The next entry is the beginning of the range that is in the |
| 10362 | * class */ |
| 10363 | UV start = nonbitmap_array[i++]; |
| 10364 | UV end; |
| 10365 | |
| 10366 | /* The next entry is the beginning of the next range, which isn't |
| 10367 | * in the class, so the end of the current range is one less than |
| 10368 | * that. But if there is no next range, it means that the range |
| 10369 | * begun by 'start' extends to infinity, which for this platform |
| 10370 | * ends at UV_MAX */ |
| 10371 | if (i == nonbitmap_len) { |
| 10372 | end = UV_MAX; |
| 10373 | } |
| 10374 | else { |
| 10375 | end = nonbitmap_array[i] - 1; |
| 10376 | } |
| 10377 | |
| 10378 | if (start == end) { |
| 10379 | Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", start); |
| 10380 | } |
| 10381 | else { |
| 10382 | /* The \t sets the whole range */ |
| 10383 | Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n", |
| 10384 | /* XXX EBCDIC */ |
| 10385 | start, end); |
| 10386 | } |
| 10387 | } |
| 10388 | invlist_destroy(nonbitmap); |
| 10389 | } |
| 10390 | |
| 10391 | if (SvCUR(listsv) == initial_listsv_len && ! unicode_alternate) { |
| 10392 | ARG_SET(ret, ANYOF_NONBITMAP_EMPTY); |
| 10393 | SvREFCNT_dec(listsv); |
| 10394 | SvREFCNT_dec(unicode_alternate); |
| 10395 | } |
| 10396 | else { |
| 10397 | |
| 10398 | AV * const av = newAV(); |
| 10399 | SV *rv; |
| 10400 | /* The 0th element stores the character class description |
| 10401 | * in its textual form: used later (regexec.c:Perl_regclass_swash()) |
| 10402 | * to initialize the appropriate swash (which gets stored in |
| 10403 | * the 1st element), and also useful for dumping the regnode. |
| 10404 | * The 2nd element stores the multicharacter foldings, |
| 10405 | * used later (regexec.c:S_reginclass()). */ |
| 10406 | av_store(av, 0, listsv); |
| 10407 | av_store(av, 1, NULL); |
| 10408 | |
| 10409 | /* Store any computed multi-char folds only if we are allowing |
| 10410 | * them */ |
| 10411 | if (allow_full_fold) { |
| 10412 | av_store(av, 2, MUTABLE_SV(unicode_alternate)); |
| 10413 | if (unicode_alternate) { /* This node is variable length */ |
| 10414 | OP(ret) = ANYOFV; |
| 10415 | } |
| 10416 | } |
| 10417 | else { |
| 10418 | av_store(av, 2, NULL); |
| 10419 | } |
| 10420 | rv = newRV_noinc(MUTABLE_SV(av)); |
| 10421 | n = add_data(pRExC_state, 1, "s"); |
| 10422 | RExC_rxi->data->data[n] = (void*)rv; |
| 10423 | ARG_SET(ret, n); |
| 10424 | } |
| 10425 | return ret; |
| 10426 | } |
| 10427 | #undef _C_C_T_ |
| 10428 | |
| 10429 | |
| 10430 | /* reg_skipcomment() |
| 10431 | |
| 10432 | Absorbs an /x style # comments from the input stream. |
| 10433 | Returns true if there is more text remaining in the stream. |
| 10434 | Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment |
| 10435 | terminates the pattern without including a newline. |
| 10436 | |
| 10437 | Note its the callers responsibility to ensure that we are |
| 10438 | actually in /x mode |
| 10439 | |
| 10440 | */ |
| 10441 | |
| 10442 | STATIC bool |
| 10443 | S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state) |
| 10444 | { |
| 10445 | bool ended = 0; |
| 10446 | |
| 10447 | PERL_ARGS_ASSERT_REG_SKIPCOMMENT; |
| 10448 | |
| 10449 | while (RExC_parse < RExC_end) |
| 10450 | if (*RExC_parse++ == '\n') { |
| 10451 | ended = 1; |
| 10452 | break; |
| 10453 | } |
| 10454 | if (!ended) { |
| 10455 | /* we ran off the end of the pattern without ending |
| 10456 | the comment, so we have to add an \n when wrapping */ |
| 10457 | RExC_seen |= REG_SEEN_RUN_ON_COMMENT; |
| 10458 | return 0; |
| 10459 | } else |
| 10460 | return 1; |
| 10461 | } |
| 10462 | |
| 10463 | /* nextchar() |
| 10464 | |
| 10465 | Advances the parse position, and optionally absorbs |
| 10466 | "whitespace" from the inputstream. |
| 10467 | |
| 10468 | Without /x "whitespace" means (?#...) style comments only, |
| 10469 | with /x this means (?#...) and # comments and whitespace proper. |
| 10470 | |
| 10471 | Returns the RExC_parse point from BEFORE the scan occurs. |
| 10472 | |
| 10473 | This is the /x friendly way of saying RExC_parse++. |
| 10474 | */ |
| 10475 | |
| 10476 | STATIC char* |
| 10477 | S_nextchar(pTHX_ RExC_state_t *pRExC_state) |
| 10478 | { |
| 10479 | char* const retval = RExC_parse++; |
| 10480 | |
| 10481 | PERL_ARGS_ASSERT_NEXTCHAR; |
| 10482 | |
| 10483 | for (;;) { |
| 10484 | if (*RExC_parse == '(' && RExC_parse[1] == '?' && |
| 10485 | RExC_parse[2] == '#') { |
| 10486 | while (*RExC_parse != ')') { |
| 10487 | if (RExC_parse == RExC_end) |
| 10488 | FAIL("Sequence (?#... not terminated"); |
| 10489 | RExC_parse++; |
| 10490 | } |
| 10491 | RExC_parse++; |
| 10492 | continue; |
| 10493 | } |
| 10494 | if (RExC_flags & RXf_PMf_EXTENDED) { |
| 10495 | if (isSPACE(*RExC_parse)) { |
| 10496 | RExC_parse++; |
| 10497 | continue; |
| 10498 | } |
| 10499 | else if (*RExC_parse == '#') { |
| 10500 | if ( reg_skipcomment( pRExC_state ) ) |
| 10501 | continue; |
| 10502 | } |
| 10503 | } |
| 10504 | return retval; |
| 10505 | } |
| 10506 | } |
| 10507 | |
| 10508 | /* |
| 10509 | - reg_node - emit a node |
| 10510 | */ |
| 10511 | STATIC regnode * /* Location. */ |
| 10512 | S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op) |
| 10513 | { |
| 10514 | dVAR; |
| 10515 | register regnode *ptr; |
| 10516 | regnode * const ret = RExC_emit; |
| 10517 | GET_RE_DEBUG_FLAGS_DECL; |
| 10518 | |
| 10519 | PERL_ARGS_ASSERT_REG_NODE; |
| 10520 | |
| 10521 | if (SIZE_ONLY) { |
| 10522 | SIZE_ALIGN(RExC_size); |
| 10523 | RExC_size += 1; |
| 10524 | return(ret); |
| 10525 | } |
| 10526 | if (RExC_emit >= RExC_emit_bound) |
| 10527 | Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op); |
| 10528 | |
| 10529 | NODE_ALIGN_FILL(ret); |
| 10530 | ptr = ret; |
| 10531 | FILL_ADVANCE_NODE(ptr, op); |
| 10532 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 10533 | if (RExC_offsets) { /* MJD */ |
| 10534 | MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n", |
| 10535 | "reg_node", __LINE__, |
| 10536 | PL_reg_name[op], |
| 10537 | (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] |
| 10538 | ? "Overwriting end of array!\n" : "OK", |
| 10539 | (UV)(RExC_emit - RExC_emit_start), |
| 10540 | (UV)(RExC_parse - RExC_start), |
| 10541 | (UV)RExC_offsets[0])); |
| 10542 | Set_Node_Offset(RExC_emit, RExC_parse + (op == END)); |
| 10543 | } |
| 10544 | #endif |
| 10545 | RExC_emit = ptr; |
| 10546 | return(ret); |
| 10547 | } |
| 10548 | |
| 10549 | /* |
| 10550 | - reganode - emit a node with an argument |
| 10551 | */ |
| 10552 | STATIC regnode * /* Location. */ |
| 10553 | S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg) |
| 10554 | { |
| 10555 | dVAR; |
| 10556 | register regnode *ptr; |
| 10557 | regnode * const ret = RExC_emit; |
| 10558 | GET_RE_DEBUG_FLAGS_DECL; |
| 10559 | |
| 10560 | PERL_ARGS_ASSERT_REGANODE; |
| 10561 | |
| 10562 | if (SIZE_ONLY) { |
| 10563 | SIZE_ALIGN(RExC_size); |
| 10564 | RExC_size += 2; |
| 10565 | /* |
| 10566 | We can't do this: |
| 10567 | |
| 10568 | assert(2==regarglen[op]+1); |
| 10569 | |
| 10570 | Anything larger than this has to allocate the extra amount. |
| 10571 | If we changed this to be: |
| 10572 | |
| 10573 | RExC_size += (1 + regarglen[op]); |
| 10574 | |
| 10575 | then it wouldn't matter. Its not clear what side effect |
| 10576 | might come from that so its not done so far. |
| 10577 | -- dmq |
| 10578 | */ |
| 10579 | return(ret); |
| 10580 | } |
| 10581 | if (RExC_emit >= RExC_emit_bound) |
| 10582 | Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op); |
| 10583 | |
| 10584 | NODE_ALIGN_FILL(ret); |
| 10585 | ptr = ret; |
| 10586 | FILL_ADVANCE_NODE_ARG(ptr, op, arg); |
| 10587 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 10588 | if (RExC_offsets) { /* MJD */ |
| 10589 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n", |
| 10590 | "reganode", |
| 10591 | __LINE__, |
| 10592 | PL_reg_name[op], |
| 10593 | (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ? |
| 10594 | "Overwriting end of array!\n" : "OK", |
| 10595 | (UV)(RExC_emit - RExC_emit_start), |
| 10596 | (UV)(RExC_parse - RExC_start), |
| 10597 | (UV)RExC_offsets[0])); |
| 10598 | Set_Cur_Node_Offset; |
| 10599 | } |
| 10600 | #endif |
| 10601 | RExC_emit = ptr; |
| 10602 | return(ret); |
| 10603 | } |
| 10604 | |
| 10605 | /* |
| 10606 | - reguni - emit (if appropriate) a Unicode character |
| 10607 | */ |
| 10608 | STATIC STRLEN |
| 10609 | S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s) |
| 10610 | { |
| 10611 | dVAR; |
| 10612 | |
| 10613 | PERL_ARGS_ASSERT_REGUNI; |
| 10614 | |
| 10615 | return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s); |
| 10616 | } |
| 10617 | |
| 10618 | /* |
| 10619 | - reginsert - insert an operator in front of already-emitted operand |
| 10620 | * |
| 10621 | * Means relocating the operand. |
| 10622 | */ |
| 10623 | STATIC void |
| 10624 | S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth) |
| 10625 | { |
| 10626 | dVAR; |
| 10627 | register regnode *src; |
| 10628 | register regnode *dst; |
| 10629 | register regnode *place; |
| 10630 | const int offset = regarglen[(U8)op]; |
| 10631 | const int size = NODE_STEP_REGNODE + offset; |
| 10632 | GET_RE_DEBUG_FLAGS_DECL; |
| 10633 | |
| 10634 | PERL_ARGS_ASSERT_REGINSERT; |
| 10635 | PERL_UNUSED_ARG(depth); |
| 10636 | /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */ |
| 10637 | DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]); |
| 10638 | if (SIZE_ONLY) { |
| 10639 | RExC_size += size; |
| 10640 | return; |
| 10641 | } |
| 10642 | |
| 10643 | src = RExC_emit; |
| 10644 | RExC_emit += size; |
| 10645 | dst = RExC_emit; |
| 10646 | if (RExC_open_parens) { |
| 10647 | int paren; |
| 10648 | /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/ |
| 10649 | for ( paren=0 ; paren < RExC_npar ; paren++ ) { |
| 10650 | if ( RExC_open_parens[paren] >= opnd ) { |
| 10651 | /*DEBUG_PARSE_FMT("open"," - %d",size);*/ |
| 10652 | RExC_open_parens[paren] += size; |
| 10653 | } else { |
| 10654 | /*DEBUG_PARSE_FMT("open"," - %s","ok");*/ |
| 10655 | } |
| 10656 | if ( RExC_close_parens[paren] >= opnd ) { |
| 10657 | /*DEBUG_PARSE_FMT("close"," - %d",size);*/ |
| 10658 | RExC_close_parens[paren] += size; |
| 10659 | } else { |
| 10660 | /*DEBUG_PARSE_FMT("close"," - %s","ok");*/ |
| 10661 | } |
| 10662 | } |
| 10663 | } |
| 10664 | |
| 10665 | while (src > opnd) { |
| 10666 | StructCopy(--src, --dst, regnode); |
| 10667 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 10668 | if (RExC_offsets) { /* MJD 20010112 */ |
| 10669 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n", |
| 10670 | "reg_insert", |
| 10671 | __LINE__, |
| 10672 | PL_reg_name[op], |
| 10673 | (UV)(dst - RExC_emit_start) > RExC_offsets[0] |
| 10674 | ? "Overwriting end of array!\n" : "OK", |
| 10675 | (UV)(src - RExC_emit_start), |
| 10676 | (UV)(dst - RExC_emit_start), |
| 10677 | (UV)RExC_offsets[0])); |
| 10678 | Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src)); |
| 10679 | Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src)); |
| 10680 | } |
| 10681 | #endif |
| 10682 | } |
| 10683 | |
| 10684 | |
| 10685 | place = opnd; /* Op node, where operand used to be. */ |
| 10686 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 10687 | if (RExC_offsets) { /* MJD */ |
| 10688 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n", |
| 10689 | "reginsert", |
| 10690 | __LINE__, |
| 10691 | PL_reg_name[op], |
| 10692 | (UV)(place - RExC_emit_start) > RExC_offsets[0] |
| 10693 | ? "Overwriting end of array!\n" : "OK", |
| 10694 | (UV)(place - RExC_emit_start), |
| 10695 | (UV)(RExC_parse - RExC_start), |
| 10696 | (UV)RExC_offsets[0])); |
| 10697 | Set_Node_Offset(place, RExC_parse); |
| 10698 | Set_Node_Length(place, 1); |
| 10699 | } |
| 10700 | #endif |
| 10701 | src = NEXTOPER(place); |
| 10702 | FILL_ADVANCE_NODE(place, op); |
| 10703 | Zero(src, offset, regnode); |
| 10704 | } |
| 10705 | |
| 10706 | /* |
| 10707 | - regtail - set the next-pointer at the end of a node chain of p to val. |
| 10708 | - SEE ALSO: regtail_study |
| 10709 | */ |
| 10710 | /* TODO: All three parms should be const */ |
| 10711 | STATIC void |
| 10712 | S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth) |
| 10713 | { |
| 10714 | dVAR; |
| 10715 | register regnode *scan; |
| 10716 | GET_RE_DEBUG_FLAGS_DECL; |
| 10717 | |
| 10718 | PERL_ARGS_ASSERT_REGTAIL; |
| 10719 | #ifndef DEBUGGING |
| 10720 | PERL_UNUSED_ARG(depth); |
| 10721 | #endif |
| 10722 | |
| 10723 | if (SIZE_ONLY) |
| 10724 | return; |
| 10725 | |
| 10726 | /* Find last node. */ |
| 10727 | scan = p; |
| 10728 | for (;;) { |
| 10729 | regnode * const temp = regnext(scan); |
| 10730 | DEBUG_PARSE_r({ |
| 10731 | SV * const mysv=sv_newmortal(); |
| 10732 | DEBUG_PARSE_MSG((scan==p ? "tail" : "")); |
| 10733 | regprop(RExC_rx, mysv, scan); |
| 10734 | PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n", |
| 10735 | SvPV_nolen_const(mysv), REG_NODE_NUM(scan), |
| 10736 | (temp == NULL ? "->" : ""), |
| 10737 | (temp == NULL ? PL_reg_name[OP(val)] : "") |
| 10738 | ); |
| 10739 | }); |
| 10740 | if (temp == NULL) |
| 10741 | break; |
| 10742 | scan = temp; |
| 10743 | } |
| 10744 | |
| 10745 | if (reg_off_by_arg[OP(scan)]) { |
| 10746 | ARG_SET(scan, val - scan); |
| 10747 | } |
| 10748 | else { |
| 10749 | NEXT_OFF(scan) = val - scan; |
| 10750 | } |
| 10751 | } |
| 10752 | |
| 10753 | #ifdef DEBUGGING |
| 10754 | /* |
| 10755 | - regtail_study - set the next-pointer at the end of a node chain of p to val. |
| 10756 | - Look for optimizable sequences at the same time. |
| 10757 | - currently only looks for EXACT chains. |
| 10758 | |
| 10759 | This is experimental code. The idea is to use this routine to perform |
| 10760 | in place optimizations on branches and groups as they are constructed, |
| 10761 | with the long term intention of removing optimization from study_chunk so |
| 10762 | that it is purely analytical. |
| 10763 | |
| 10764 | Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used |
| 10765 | to control which is which. |
| 10766 | |
| 10767 | */ |
| 10768 | /* TODO: All four parms should be const */ |
| 10769 | |
| 10770 | STATIC U8 |
| 10771 | S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth) |
| 10772 | { |
| 10773 | dVAR; |
| 10774 | register regnode *scan; |
| 10775 | U8 exact = PSEUDO; |
| 10776 | #ifdef EXPERIMENTAL_INPLACESCAN |
| 10777 | I32 min = 0; |
| 10778 | #endif |
| 10779 | GET_RE_DEBUG_FLAGS_DECL; |
| 10780 | |
| 10781 | PERL_ARGS_ASSERT_REGTAIL_STUDY; |
| 10782 | |
| 10783 | |
| 10784 | if (SIZE_ONLY) |
| 10785 | return exact; |
| 10786 | |
| 10787 | /* Find last node. */ |
| 10788 | |
| 10789 | scan = p; |
| 10790 | for (;;) { |
| 10791 | regnode * const temp = regnext(scan); |
| 10792 | #ifdef EXPERIMENTAL_INPLACESCAN |
| 10793 | if (PL_regkind[OP(scan)] == EXACT) |
| 10794 | if (join_exact(pRExC_state,scan,&min,1,val,depth+1)) |
| 10795 | return EXACT; |
| 10796 | #endif |
| 10797 | if ( exact ) { |
| 10798 | switch (OP(scan)) { |
| 10799 | case EXACT: |
| 10800 | case EXACTF: |
| 10801 | case EXACTFA: |
| 10802 | case EXACTFU: |
| 10803 | case EXACTFL: |
| 10804 | if( exact == PSEUDO ) |
| 10805 | exact= OP(scan); |
| 10806 | else if ( exact != OP(scan) ) |
| 10807 | exact= 0; |
| 10808 | case NOTHING: |
| 10809 | break; |
| 10810 | default: |
| 10811 | exact= 0; |
| 10812 | } |
| 10813 | } |
| 10814 | DEBUG_PARSE_r({ |
| 10815 | SV * const mysv=sv_newmortal(); |
| 10816 | DEBUG_PARSE_MSG((scan==p ? "tsdy" : "")); |
| 10817 | regprop(RExC_rx, mysv, scan); |
| 10818 | PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n", |
| 10819 | SvPV_nolen_const(mysv), |
| 10820 | REG_NODE_NUM(scan), |
| 10821 | PL_reg_name[exact]); |
| 10822 | }); |
| 10823 | if (temp == NULL) |
| 10824 | break; |
| 10825 | scan = temp; |
| 10826 | } |
| 10827 | DEBUG_PARSE_r({ |
| 10828 | SV * const mysv_val=sv_newmortal(); |
| 10829 | DEBUG_PARSE_MSG(""); |
| 10830 | regprop(RExC_rx, mysv_val, val); |
| 10831 | PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n", |
| 10832 | SvPV_nolen_const(mysv_val), |
| 10833 | (IV)REG_NODE_NUM(val), |
| 10834 | (IV)(val - scan) |
| 10835 | ); |
| 10836 | }); |
| 10837 | if (reg_off_by_arg[OP(scan)]) { |
| 10838 | ARG_SET(scan, val - scan); |
| 10839 | } |
| 10840 | else { |
| 10841 | NEXT_OFF(scan) = val - scan; |
| 10842 | } |
| 10843 | |
| 10844 | return exact; |
| 10845 | } |
| 10846 | #endif |
| 10847 | |
| 10848 | /* |
| 10849 | - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form |
| 10850 | */ |
| 10851 | #ifdef DEBUGGING |
| 10852 | static void |
| 10853 | S_regdump_extflags(pTHX_ const char *lead, const U32 flags) |
| 10854 | { |
| 10855 | int bit; |
| 10856 | int set=0; |
| 10857 | regex_charset cs; |
| 10858 | |
| 10859 | for (bit=0; bit<32; bit++) { |
| 10860 | if (flags & (1<<bit)) { |
| 10861 | if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */ |
| 10862 | continue; |
| 10863 | } |
| 10864 | if (!set++ && lead) |
| 10865 | PerlIO_printf(Perl_debug_log, "%s",lead); |
| 10866 | PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]); |
| 10867 | } |
| 10868 | } |
| 10869 | if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) { |
| 10870 | if (!set++ && lead) { |
| 10871 | PerlIO_printf(Perl_debug_log, "%s",lead); |
| 10872 | } |
| 10873 | switch (cs) { |
| 10874 | case REGEX_UNICODE_CHARSET: |
| 10875 | PerlIO_printf(Perl_debug_log, "UNICODE"); |
| 10876 | break; |
| 10877 | case REGEX_LOCALE_CHARSET: |
| 10878 | PerlIO_printf(Perl_debug_log, "LOCALE"); |
| 10879 | break; |
| 10880 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 10881 | PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED"); |
| 10882 | break; |
| 10883 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 10884 | PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED"); |
| 10885 | break; |
| 10886 | default: |
| 10887 | PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET"); |
| 10888 | break; |
| 10889 | } |
| 10890 | } |
| 10891 | if (lead) { |
| 10892 | if (set) |
| 10893 | PerlIO_printf(Perl_debug_log, "\n"); |
| 10894 | else |
| 10895 | PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead); |
| 10896 | } |
| 10897 | } |
| 10898 | #endif |
| 10899 | |
| 10900 | void |
| 10901 | Perl_regdump(pTHX_ const regexp *r) |
| 10902 | { |
| 10903 | #ifdef DEBUGGING |
| 10904 | dVAR; |
| 10905 | SV * const sv = sv_newmortal(); |
| 10906 | SV *dsv= sv_newmortal(); |
| 10907 | RXi_GET_DECL(r,ri); |
| 10908 | GET_RE_DEBUG_FLAGS_DECL; |
| 10909 | |
| 10910 | PERL_ARGS_ASSERT_REGDUMP; |
| 10911 | |
| 10912 | (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0); |
| 10913 | |
| 10914 | /* Header fields of interest. */ |
| 10915 | if (r->anchored_substr) { |
| 10916 | RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr), |
| 10917 | RE_SV_DUMPLEN(r->anchored_substr), 30); |
| 10918 | PerlIO_printf(Perl_debug_log, |
| 10919 | "anchored %s%s at %"IVdf" ", |
| 10920 | s, RE_SV_TAIL(r->anchored_substr), |
| 10921 | (IV)r->anchored_offset); |
| 10922 | } else if (r->anchored_utf8) { |
| 10923 | RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8), |
| 10924 | RE_SV_DUMPLEN(r->anchored_utf8), 30); |
| 10925 | PerlIO_printf(Perl_debug_log, |
| 10926 | "anchored utf8 %s%s at %"IVdf" ", |
| 10927 | s, RE_SV_TAIL(r->anchored_utf8), |
| 10928 | (IV)r->anchored_offset); |
| 10929 | } |
| 10930 | if (r->float_substr) { |
| 10931 | RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr), |
| 10932 | RE_SV_DUMPLEN(r->float_substr), 30); |
| 10933 | PerlIO_printf(Perl_debug_log, |
| 10934 | "floating %s%s at %"IVdf"..%"UVuf" ", |
| 10935 | s, RE_SV_TAIL(r->float_substr), |
| 10936 | (IV)r->float_min_offset, (UV)r->float_max_offset); |
| 10937 | } else if (r->float_utf8) { |
| 10938 | RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8), |
| 10939 | RE_SV_DUMPLEN(r->float_utf8), 30); |
| 10940 | PerlIO_printf(Perl_debug_log, |
| 10941 | "floating utf8 %s%s at %"IVdf"..%"UVuf" ", |
| 10942 | s, RE_SV_TAIL(r->float_utf8), |
| 10943 | (IV)r->float_min_offset, (UV)r->float_max_offset); |
| 10944 | } |
| 10945 | if (r->check_substr || r->check_utf8) |
| 10946 | PerlIO_printf(Perl_debug_log, |
| 10947 | (const char *) |
| 10948 | (r->check_substr == r->float_substr |
| 10949 | && r->check_utf8 == r->float_utf8 |
| 10950 | ? "(checking floating" : "(checking anchored")); |
| 10951 | if (r->extflags & RXf_NOSCAN) |
| 10952 | PerlIO_printf(Perl_debug_log, " noscan"); |
| 10953 | if (r->extflags & RXf_CHECK_ALL) |
| 10954 | PerlIO_printf(Perl_debug_log, " isall"); |
| 10955 | if (r->check_substr || r->check_utf8) |
| 10956 | PerlIO_printf(Perl_debug_log, ") "); |
| 10957 | |
| 10958 | if (ri->regstclass) { |
| 10959 | regprop(r, sv, ri->regstclass); |
| 10960 | PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv)); |
| 10961 | } |
| 10962 | if (r->extflags & RXf_ANCH) { |
| 10963 | PerlIO_printf(Perl_debug_log, "anchored"); |
| 10964 | if (r->extflags & RXf_ANCH_BOL) |
| 10965 | PerlIO_printf(Perl_debug_log, "(BOL)"); |
| 10966 | if (r->extflags & RXf_ANCH_MBOL) |
| 10967 | PerlIO_printf(Perl_debug_log, "(MBOL)"); |
| 10968 | if (r->extflags & RXf_ANCH_SBOL) |
| 10969 | PerlIO_printf(Perl_debug_log, "(SBOL)"); |
| 10970 | if (r->extflags & RXf_ANCH_GPOS) |
| 10971 | PerlIO_printf(Perl_debug_log, "(GPOS)"); |
| 10972 | PerlIO_putc(Perl_debug_log, ' '); |
| 10973 | } |
| 10974 | if (r->extflags & RXf_GPOS_SEEN) |
| 10975 | PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs); |
| 10976 | if (r->intflags & PREGf_SKIP) |
| 10977 | PerlIO_printf(Perl_debug_log, "plus "); |
| 10978 | if (r->intflags & PREGf_IMPLICIT) |
| 10979 | PerlIO_printf(Perl_debug_log, "implicit "); |
| 10980 | PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen); |
| 10981 | if (r->extflags & RXf_EVAL_SEEN) |
| 10982 | PerlIO_printf(Perl_debug_log, "with eval "); |
| 10983 | PerlIO_printf(Perl_debug_log, "\n"); |
| 10984 | DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags)); |
| 10985 | #else |
| 10986 | PERL_ARGS_ASSERT_REGDUMP; |
| 10987 | PERL_UNUSED_CONTEXT; |
| 10988 | PERL_UNUSED_ARG(r); |
| 10989 | #endif /* DEBUGGING */ |
| 10990 | } |
| 10991 | |
| 10992 | /* |
| 10993 | - regprop - printable representation of opcode |
| 10994 | */ |
| 10995 | #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \ |
| 10996 | STMT_START { \ |
| 10997 | if (do_sep) { \ |
| 10998 | Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \ |
| 10999 | if (flags & ANYOF_INVERT) \ |
| 11000 | /*make sure the invert info is in each */ \ |
| 11001 | sv_catpvs(sv, "^"); \ |
| 11002 | do_sep = 0; \ |
| 11003 | } \ |
| 11004 | } STMT_END |
| 11005 | |
| 11006 | void |
| 11007 | Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o) |
| 11008 | { |
| 11009 | #ifdef DEBUGGING |
| 11010 | dVAR; |
| 11011 | register int k; |
| 11012 | RXi_GET_DECL(prog,progi); |
| 11013 | GET_RE_DEBUG_FLAGS_DECL; |
| 11014 | |
| 11015 | PERL_ARGS_ASSERT_REGPROP; |
| 11016 | |
| 11017 | sv_setpvs(sv, ""); |
| 11018 | |
| 11019 | if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */ |
| 11020 | /* It would be nice to FAIL() here, but this may be called from |
| 11021 | regexec.c, and it would be hard to supply pRExC_state. */ |
| 11022 | Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX); |
| 11023 | sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */ |
| 11024 | |
| 11025 | k = PL_regkind[OP(o)]; |
| 11026 | |
| 11027 | if (k == EXACT) { |
| 11028 | sv_catpvs(sv, " "); |
| 11029 | /* Using is_utf8_string() (via PERL_PV_UNI_DETECT) |
| 11030 | * is a crude hack but it may be the best for now since |
| 11031 | * we have no flag "this EXACTish node was UTF-8" |
| 11032 | * --jhi */ |
| 11033 | pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1], |
| 11034 | PERL_PV_ESCAPE_UNI_DETECT | |
| 11035 | PERL_PV_ESCAPE_NONASCII | |
| 11036 | PERL_PV_PRETTY_ELLIPSES | |
| 11037 | PERL_PV_PRETTY_LTGT | |
| 11038 | PERL_PV_PRETTY_NOCLEAR |
| 11039 | ); |
| 11040 | } else if (k == TRIE) { |
| 11041 | /* print the details of the trie in dumpuntil instead, as |
| 11042 | * progi->data isn't available here */ |
| 11043 | const char op = OP(o); |
| 11044 | const U32 n = ARG(o); |
| 11045 | const reg_ac_data * const ac = IS_TRIE_AC(op) ? |
| 11046 | (reg_ac_data *)progi->data->data[n] : |
| 11047 | NULL; |
| 11048 | const reg_trie_data * const trie |
| 11049 | = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie]; |
| 11050 | |
| 11051 | Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]); |
| 11052 | DEBUG_TRIE_COMPILE_r( |
| 11053 | Perl_sv_catpvf(aTHX_ sv, |
| 11054 | "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">", |
| 11055 | (UV)trie->startstate, |
| 11056 | (IV)trie->statecount-1, /* -1 because of the unused 0 element */ |
| 11057 | (UV)trie->wordcount, |
| 11058 | (UV)trie->minlen, |
| 11059 | (UV)trie->maxlen, |
| 11060 | (UV)TRIE_CHARCOUNT(trie), |
| 11061 | (UV)trie->uniquecharcount |
| 11062 | ) |
| 11063 | ); |
| 11064 | if ( IS_ANYOF_TRIE(op) || trie->bitmap ) { |
| 11065 | int i; |
| 11066 | int rangestart = -1; |
| 11067 | U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie); |
| 11068 | sv_catpvs(sv, "["); |
| 11069 | for (i = 0; i <= 256; i++) { |
| 11070 | if (i < 256 && BITMAP_TEST(bitmap,i)) { |
| 11071 | if (rangestart == -1) |
| 11072 | rangestart = i; |
| 11073 | } else if (rangestart != -1) { |
| 11074 | if (i <= rangestart + 3) |
| 11075 | for (; rangestart < i; rangestart++) |
| 11076 | put_byte(sv, rangestart); |
| 11077 | else { |
| 11078 | put_byte(sv, rangestart); |
| 11079 | sv_catpvs(sv, "-"); |
| 11080 | put_byte(sv, i - 1); |
| 11081 | } |
| 11082 | rangestart = -1; |
| 11083 | } |
| 11084 | } |
| 11085 | sv_catpvs(sv, "]"); |
| 11086 | } |
| 11087 | |
| 11088 | } else if (k == CURLY) { |
| 11089 | if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX) |
| 11090 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */ |
| 11091 | Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o)); |
| 11092 | } |
| 11093 | else if (k == WHILEM && o->flags) /* Ordinal/of */ |
| 11094 | Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4); |
| 11095 | else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) { |
| 11096 | Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */ |
| 11097 | if ( RXp_PAREN_NAMES(prog) ) { |
| 11098 | if ( k != REF || (OP(o) < NREF)) { |
| 11099 | AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]); |
| 11100 | SV **name= av_fetch(list, ARG(o), 0 ); |
| 11101 | if (name) |
| 11102 | Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name)); |
| 11103 | } |
| 11104 | else { |
| 11105 | AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]); |
| 11106 | SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]); |
| 11107 | I32 *nums=(I32*)SvPVX(sv_dat); |
| 11108 | SV **name= av_fetch(list, nums[0], 0 ); |
| 11109 | I32 n; |
| 11110 | if (name) { |
| 11111 | for ( n=0; n<SvIVX(sv_dat); n++ ) { |
| 11112 | Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf, |
| 11113 | (n ? "," : ""), (IV)nums[n]); |
| 11114 | } |
| 11115 | Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name)); |
| 11116 | } |
| 11117 | } |
| 11118 | } |
| 11119 | } else if (k == GOSUB) |
| 11120 | Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */ |
| 11121 | else if (k == VERB) { |
| 11122 | if (!o->flags) |
| 11123 | Perl_sv_catpvf(aTHX_ sv, ":%"SVf, |
| 11124 | SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ])))); |
| 11125 | } else if (k == LOGICAL) |
| 11126 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */ |
| 11127 | else if (k == FOLDCHAR) |
| 11128 | Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) ); |
| 11129 | else if (k == ANYOF) { |
| 11130 | int i, rangestart = -1; |
| 11131 | const U8 flags = ANYOF_FLAGS(o); |
| 11132 | int do_sep = 0; |
| 11133 | |
| 11134 | /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */ |
| 11135 | static const char * const anyofs[] = { |
| 11136 | "\\w", |
| 11137 | "\\W", |
| 11138 | "\\s", |
| 11139 | "\\S", |
| 11140 | "\\d", |
| 11141 | "\\D", |
| 11142 | "[:alnum:]", |
| 11143 | "[:^alnum:]", |
| 11144 | "[:alpha:]", |
| 11145 | "[:^alpha:]", |
| 11146 | "[:ascii:]", |
| 11147 | "[:^ascii:]", |
| 11148 | "[:cntrl:]", |
| 11149 | "[:^cntrl:]", |
| 11150 | "[:graph:]", |
| 11151 | "[:^graph:]", |
| 11152 | "[:lower:]", |
| 11153 | "[:^lower:]", |
| 11154 | "[:print:]", |
| 11155 | "[:^print:]", |
| 11156 | "[:punct:]", |
| 11157 | "[:^punct:]", |
| 11158 | "[:upper:]", |
| 11159 | "[:^upper:]", |
| 11160 | "[:xdigit:]", |
| 11161 | "[:^xdigit:]", |
| 11162 | "[:space:]", |
| 11163 | "[:^space:]", |
| 11164 | "[:blank:]", |
| 11165 | "[:^blank:]" |
| 11166 | }; |
| 11167 | |
| 11168 | if (flags & ANYOF_LOCALE) |
| 11169 | sv_catpvs(sv, "{loc}"); |
| 11170 | if (flags & ANYOF_LOC_NONBITMAP_FOLD) |
| 11171 | sv_catpvs(sv, "{i}"); |
| 11172 | Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]); |
| 11173 | if (flags & ANYOF_INVERT) |
| 11174 | sv_catpvs(sv, "^"); |
| 11175 | |
| 11176 | /* output what the standard cp 0-255 bitmap matches */ |
| 11177 | for (i = 0; i <= 256; i++) { |
| 11178 | if (i < 256 && ANYOF_BITMAP_TEST(o,i)) { |
| 11179 | if (rangestart == -1) |
| 11180 | rangestart = i; |
| 11181 | } else if (rangestart != -1) { |
| 11182 | if (i <= rangestart + 3) |
| 11183 | for (; rangestart < i; rangestart++) |
| 11184 | put_byte(sv, rangestart); |
| 11185 | else { |
| 11186 | put_byte(sv, rangestart); |
| 11187 | sv_catpvs(sv, "-"); |
| 11188 | put_byte(sv, i - 1); |
| 11189 | } |
| 11190 | do_sep = 1; |
| 11191 | rangestart = -1; |
| 11192 | } |
| 11193 | } |
| 11194 | |
| 11195 | EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags); |
| 11196 | /* output any special charclass tests (used entirely under use locale) */ |
| 11197 | if (ANYOF_CLASS_TEST_ANY_SET(o)) |
| 11198 | for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++) |
| 11199 | if (ANYOF_CLASS_TEST(o,i)) { |
| 11200 | sv_catpv(sv, anyofs[i]); |
| 11201 | do_sep = 1; |
| 11202 | } |
| 11203 | |
| 11204 | EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags); |
| 11205 | |
| 11206 | if (flags & ANYOF_NON_UTF8_LATIN1_ALL) { |
| 11207 | sv_catpvs(sv, "{non-utf8-latin1-all}"); |
| 11208 | } |
| 11209 | |
| 11210 | /* output information about the unicode matching */ |
| 11211 | if (flags & ANYOF_UNICODE_ALL) |
| 11212 | sv_catpvs(sv, "{unicode_all}"); |
| 11213 | else if (ANYOF_NONBITMAP(o)) |
| 11214 | sv_catpvs(sv, "{unicode}"); |
| 11215 | if (flags & ANYOF_NONBITMAP_NON_UTF8) |
| 11216 | sv_catpvs(sv, "{outside bitmap}"); |
| 11217 | |
| 11218 | if (ANYOF_NONBITMAP(o)) { |
| 11219 | SV *lv; |
| 11220 | SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0); |
| 11221 | |
| 11222 | if (lv) { |
| 11223 | if (sw) { |
| 11224 | U8 s[UTF8_MAXBYTES_CASE+1]; |
| 11225 | |
| 11226 | for (i = 0; i <= 256; i++) { /* just the first 256 */ |
| 11227 | uvchr_to_utf8(s, i); |
| 11228 | |
| 11229 | if (i < 256 && swash_fetch(sw, s, TRUE)) { |
| 11230 | if (rangestart == -1) |
| 11231 | rangestart = i; |
| 11232 | } else if (rangestart != -1) { |
| 11233 | if (i <= rangestart + 3) |
| 11234 | for (; rangestart < i; rangestart++) { |
| 11235 | const U8 * const e = uvchr_to_utf8(s,rangestart); |
| 11236 | U8 *p; |
| 11237 | for(p = s; p < e; p++) |
| 11238 | put_byte(sv, *p); |
| 11239 | } |
| 11240 | else { |
| 11241 | const U8 *e = uvchr_to_utf8(s,rangestart); |
| 11242 | U8 *p; |
| 11243 | for (p = s; p < e; p++) |
| 11244 | put_byte(sv, *p); |
| 11245 | sv_catpvs(sv, "-"); |
| 11246 | e = uvchr_to_utf8(s, i-1); |
| 11247 | for (p = s; p < e; p++) |
| 11248 | put_byte(sv, *p); |
| 11249 | } |
| 11250 | rangestart = -1; |
| 11251 | } |
| 11252 | } |
| 11253 | |
| 11254 | sv_catpvs(sv, "..."); /* et cetera */ |
| 11255 | } |
| 11256 | |
| 11257 | { |
| 11258 | char *s = savesvpv(lv); |
| 11259 | char * const origs = s; |
| 11260 | |
| 11261 | while (*s && *s != '\n') |
| 11262 | s++; |
| 11263 | |
| 11264 | if (*s == '\n') { |
| 11265 | const char * const t = ++s; |
| 11266 | |
| 11267 | while (*s) { |
| 11268 | if (*s == '\n') |
| 11269 | *s = ' '; |
| 11270 | s++; |
| 11271 | } |
| 11272 | if (s[-1] == ' ') |
| 11273 | s[-1] = 0; |
| 11274 | |
| 11275 | sv_catpv(sv, t); |
| 11276 | } |
| 11277 | |
| 11278 | Safefree(origs); |
| 11279 | } |
| 11280 | } |
| 11281 | } |
| 11282 | |
| 11283 | Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]); |
| 11284 | } |
| 11285 | else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH)) |
| 11286 | Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags)); |
| 11287 | #else |
| 11288 | PERL_UNUSED_CONTEXT; |
| 11289 | PERL_UNUSED_ARG(sv); |
| 11290 | PERL_UNUSED_ARG(o); |
| 11291 | PERL_UNUSED_ARG(prog); |
| 11292 | #endif /* DEBUGGING */ |
| 11293 | } |
| 11294 | |
| 11295 | SV * |
| 11296 | Perl_re_intuit_string(pTHX_ REGEXP * const r) |
| 11297 | { /* Assume that RE_INTUIT is set */ |
| 11298 | dVAR; |
| 11299 | struct regexp *const prog = (struct regexp *)SvANY(r); |
| 11300 | GET_RE_DEBUG_FLAGS_DECL; |
| 11301 | |
| 11302 | PERL_ARGS_ASSERT_RE_INTUIT_STRING; |
| 11303 | PERL_UNUSED_CONTEXT; |
| 11304 | |
| 11305 | DEBUG_COMPILE_r( |
| 11306 | { |
| 11307 | const char * const s = SvPV_nolen_const(prog->check_substr |
| 11308 | ? prog->check_substr : prog->check_utf8); |
| 11309 | |
| 11310 | if (!PL_colorset) reginitcolors(); |
| 11311 | PerlIO_printf(Perl_debug_log, |
| 11312 | "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n", |
| 11313 | PL_colors[4], |
| 11314 | prog->check_substr ? "" : "utf8 ", |
| 11315 | PL_colors[5],PL_colors[0], |
| 11316 | s, |
| 11317 | PL_colors[1], |
| 11318 | (strlen(s) > 60 ? "..." : "")); |
| 11319 | } ); |
| 11320 | |
| 11321 | return prog->check_substr ? prog->check_substr : prog->check_utf8; |
| 11322 | } |
| 11323 | |
| 11324 | /* |
| 11325 | pregfree() |
| 11326 | |
| 11327 | handles refcounting and freeing the perl core regexp structure. When |
| 11328 | it is necessary to actually free the structure the first thing it |
| 11329 | does is call the 'free' method of the regexp_engine associated to |
| 11330 | the regexp, allowing the handling of the void *pprivate; member |
| 11331 | first. (This routine is not overridable by extensions, which is why |
| 11332 | the extensions free is called first.) |
| 11333 | |
| 11334 | See regdupe and regdupe_internal if you change anything here. |
| 11335 | */ |
| 11336 | #ifndef PERL_IN_XSUB_RE |
| 11337 | void |
| 11338 | Perl_pregfree(pTHX_ REGEXP *r) |
| 11339 | { |
| 11340 | SvREFCNT_dec(r); |
| 11341 | } |
| 11342 | |
| 11343 | void |
| 11344 | Perl_pregfree2(pTHX_ REGEXP *rx) |
| 11345 | { |
| 11346 | dVAR; |
| 11347 | struct regexp *const r = (struct regexp *)SvANY(rx); |
| 11348 | GET_RE_DEBUG_FLAGS_DECL; |
| 11349 | |
| 11350 | PERL_ARGS_ASSERT_PREGFREE2; |
| 11351 | |
| 11352 | if (r->mother_re) { |
| 11353 | ReREFCNT_dec(r->mother_re); |
| 11354 | } else { |
| 11355 | CALLREGFREE_PVT(rx); /* free the private data */ |
| 11356 | SvREFCNT_dec(RXp_PAREN_NAMES(r)); |
| 11357 | } |
| 11358 | if (r->substrs) { |
| 11359 | SvREFCNT_dec(r->anchored_substr); |
| 11360 | SvREFCNT_dec(r->anchored_utf8); |
| 11361 | SvREFCNT_dec(r->float_substr); |
| 11362 | SvREFCNT_dec(r->float_utf8); |
| 11363 | Safefree(r->substrs); |
| 11364 | } |
| 11365 | RX_MATCH_COPY_FREE(rx); |
| 11366 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 11367 | SvREFCNT_dec(r->saved_copy); |
| 11368 | #endif |
| 11369 | Safefree(r->offs); |
| 11370 | } |
| 11371 | |
| 11372 | /* reg_temp_copy() |
| 11373 | |
| 11374 | This is a hacky workaround to the structural issue of match results |
| 11375 | being stored in the regexp structure which is in turn stored in |
| 11376 | PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern |
| 11377 | could be PL_curpm in multiple contexts, and could require multiple |
| 11378 | result sets being associated with the pattern simultaneously, such |
| 11379 | as when doing a recursive match with (??{$qr}) |
| 11380 | |
| 11381 | The solution is to make a lightweight copy of the regexp structure |
| 11382 | when a qr// is returned from the code executed by (??{$qr}) this |
| 11383 | lightweight copy doesn't actually own any of its data except for |
| 11384 | the starp/end and the actual regexp structure itself. |
| 11385 | |
| 11386 | */ |
| 11387 | |
| 11388 | |
| 11389 | REGEXP * |
| 11390 | Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx) |
| 11391 | { |
| 11392 | struct regexp *ret; |
| 11393 | struct regexp *const r = (struct regexp *)SvANY(rx); |
| 11394 | register const I32 npar = r->nparens+1; |
| 11395 | |
| 11396 | PERL_ARGS_ASSERT_REG_TEMP_COPY; |
| 11397 | |
| 11398 | if (!ret_x) |
| 11399 | ret_x = (REGEXP*) newSV_type(SVt_REGEXP); |
| 11400 | ret = (struct regexp *)SvANY(ret_x); |
| 11401 | |
| 11402 | (void)ReREFCNT_inc(rx); |
| 11403 | /* We can take advantage of the existing "copied buffer" mechanism in SVs |
| 11404 | by pointing directly at the buffer, but flagging that the allocated |
| 11405 | space in the copy is zero. As we've just done a struct copy, it's now |
| 11406 | a case of zero-ing that, rather than copying the current length. */ |
| 11407 | SvPV_set(ret_x, RX_WRAPPED(rx)); |
| 11408 | SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8); |
| 11409 | memcpy(&(ret->xpv_cur), &(r->xpv_cur), |
| 11410 | sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur)); |
| 11411 | SvLEN_set(ret_x, 0); |
| 11412 | SvSTASH_set(ret_x, NULL); |
| 11413 | SvMAGIC_set(ret_x, NULL); |
| 11414 | Newx(ret->offs, npar, regexp_paren_pair); |
| 11415 | Copy(r->offs, ret->offs, npar, regexp_paren_pair); |
| 11416 | if (r->substrs) { |
| 11417 | Newx(ret->substrs, 1, struct reg_substr_data); |
| 11418 | StructCopy(r->substrs, ret->substrs, struct reg_substr_data); |
| 11419 | |
| 11420 | SvREFCNT_inc_void(ret->anchored_substr); |
| 11421 | SvREFCNT_inc_void(ret->anchored_utf8); |
| 11422 | SvREFCNT_inc_void(ret->float_substr); |
| 11423 | SvREFCNT_inc_void(ret->float_utf8); |
| 11424 | |
| 11425 | /* check_substr and check_utf8, if non-NULL, point to either their |
| 11426 | anchored or float namesakes, and don't hold a second reference. */ |
| 11427 | } |
| 11428 | RX_MATCH_COPIED_off(ret_x); |
| 11429 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 11430 | ret->saved_copy = NULL; |
| 11431 | #endif |
| 11432 | ret->mother_re = rx; |
| 11433 | |
| 11434 | return ret_x; |
| 11435 | } |
| 11436 | #endif |
| 11437 | |
| 11438 | /* regfree_internal() |
| 11439 | |
| 11440 | Free the private data in a regexp. This is overloadable by |
| 11441 | extensions. Perl takes care of the regexp structure in pregfree(), |
| 11442 | this covers the *pprivate pointer which technically perl doesn't |
| 11443 | know about, however of course we have to handle the |
| 11444 | regexp_internal structure when no extension is in use. |
| 11445 | |
| 11446 | Note this is called before freeing anything in the regexp |
| 11447 | structure. |
| 11448 | */ |
| 11449 | |
| 11450 | void |
| 11451 | Perl_regfree_internal(pTHX_ REGEXP * const rx) |
| 11452 | { |
| 11453 | dVAR; |
| 11454 | struct regexp *const r = (struct regexp *)SvANY(rx); |
| 11455 | RXi_GET_DECL(r,ri); |
| 11456 | GET_RE_DEBUG_FLAGS_DECL; |
| 11457 | |
| 11458 | PERL_ARGS_ASSERT_REGFREE_INTERNAL; |
| 11459 | |
| 11460 | DEBUG_COMPILE_r({ |
| 11461 | if (!PL_colorset) |
| 11462 | reginitcolors(); |
| 11463 | { |
| 11464 | SV *dsv= sv_newmortal(); |
| 11465 | RE_PV_QUOTED_DECL(s, RX_UTF8(rx), |
| 11466 | dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60); |
| 11467 | PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n", |
| 11468 | PL_colors[4],PL_colors[5],s); |
| 11469 | } |
| 11470 | }); |
| 11471 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 11472 | if (ri->u.offsets) |
| 11473 | Safefree(ri->u.offsets); /* 20010421 MJD */ |
| 11474 | #endif |
| 11475 | if (ri->data) { |
| 11476 | int n = ri->data->count; |
| 11477 | PAD* new_comppad = NULL; |
| 11478 | PAD* old_comppad; |
| 11479 | PADOFFSET refcnt; |
| 11480 | |
| 11481 | while (--n >= 0) { |
| 11482 | /* If you add a ->what type here, update the comment in regcomp.h */ |
| 11483 | switch (ri->data->what[n]) { |
| 11484 | case 'a': |
| 11485 | case 's': |
| 11486 | case 'S': |
| 11487 | case 'u': |
| 11488 | SvREFCNT_dec(MUTABLE_SV(ri->data->data[n])); |
| 11489 | break; |
| 11490 | case 'f': |
| 11491 | Safefree(ri->data->data[n]); |
| 11492 | break; |
| 11493 | case 'p': |
| 11494 | new_comppad = MUTABLE_AV(ri->data->data[n]); |
| 11495 | break; |
| 11496 | case 'o': |
| 11497 | if (new_comppad == NULL) |
| 11498 | Perl_croak(aTHX_ "panic: pregfree comppad"); |
| 11499 | PAD_SAVE_LOCAL(old_comppad, |
| 11500 | /* Watch out for global destruction's random ordering. */ |
| 11501 | (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL |
| 11502 | ); |
| 11503 | OP_REFCNT_LOCK; |
| 11504 | refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]); |
| 11505 | OP_REFCNT_UNLOCK; |
| 11506 | if (!refcnt) |
| 11507 | op_free((OP_4tree*)ri->data->data[n]); |
| 11508 | |
| 11509 | PAD_RESTORE_LOCAL(old_comppad); |
| 11510 | SvREFCNT_dec(MUTABLE_SV(new_comppad)); |
| 11511 | new_comppad = NULL; |
| 11512 | break; |
| 11513 | case 'n': |
| 11514 | break; |
| 11515 | case 'T': |
| 11516 | { /* Aho Corasick add-on structure for a trie node. |
| 11517 | Used in stclass optimization only */ |
| 11518 | U32 refcount; |
| 11519 | reg_ac_data *aho=(reg_ac_data*)ri->data->data[n]; |
| 11520 | OP_REFCNT_LOCK; |
| 11521 | refcount = --aho->refcount; |
| 11522 | OP_REFCNT_UNLOCK; |
| 11523 | if ( !refcount ) { |
| 11524 | PerlMemShared_free(aho->states); |
| 11525 | PerlMemShared_free(aho->fail); |
| 11526 | /* do this last!!!! */ |
| 11527 | PerlMemShared_free(ri->data->data[n]); |
| 11528 | PerlMemShared_free(ri->regstclass); |
| 11529 | } |
| 11530 | } |
| 11531 | break; |
| 11532 | case 't': |
| 11533 | { |
| 11534 | /* trie structure. */ |
| 11535 | U32 refcount; |
| 11536 | reg_trie_data *trie=(reg_trie_data*)ri->data->data[n]; |
| 11537 | OP_REFCNT_LOCK; |
| 11538 | refcount = --trie->refcount; |
| 11539 | OP_REFCNT_UNLOCK; |
| 11540 | if ( !refcount ) { |
| 11541 | PerlMemShared_free(trie->charmap); |
| 11542 | PerlMemShared_free(trie->states); |
| 11543 | PerlMemShared_free(trie->trans); |
| 11544 | if (trie->bitmap) |
| 11545 | PerlMemShared_free(trie->bitmap); |
| 11546 | if (trie->jump) |
| 11547 | PerlMemShared_free(trie->jump); |
| 11548 | PerlMemShared_free(trie->wordinfo); |
| 11549 | /* do this last!!!! */ |
| 11550 | PerlMemShared_free(ri->data->data[n]); |
| 11551 | } |
| 11552 | } |
| 11553 | break; |
| 11554 | default: |
| 11555 | Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]); |
| 11556 | } |
| 11557 | } |
| 11558 | Safefree(ri->data->what); |
| 11559 | Safefree(ri->data); |
| 11560 | } |
| 11561 | |
| 11562 | Safefree(ri); |
| 11563 | } |
| 11564 | |
| 11565 | #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t)) |
| 11566 | #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t)) |
| 11567 | #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL) |
| 11568 | |
| 11569 | /* |
| 11570 | re_dup - duplicate a regexp. |
| 11571 | |
| 11572 | This routine is expected to clone a given regexp structure. It is only |
| 11573 | compiled under USE_ITHREADS. |
| 11574 | |
| 11575 | After all of the core data stored in struct regexp is duplicated |
| 11576 | the regexp_engine.dupe method is used to copy any private data |
| 11577 | stored in the *pprivate pointer. This allows extensions to handle |
| 11578 | any duplication it needs to do. |
| 11579 | |
| 11580 | See pregfree() and regfree_internal() if you change anything here. |
| 11581 | */ |
| 11582 | #if defined(USE_ITHREADS) |
| 11583 | #ifndef PERL_IN_XSUB_RE |
| 11584 | void |
| 11585 | Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param) |
| 11586 | { |
| 11587 | dVAR; |
| 11588 | I32 npar; |
| 11589 | const struct regexp *r = (const struct regexp *)SvANY(sstr); |
| 11590 | struct regexp *ret = (struct regexp *)SvANY(dstr); |
| 11591 | |
| 11592 | PERL_ARGS_ASSERT_RE_DUP_GUTS; |
| 11593 | |
| 11594 | npar = r->nparens+1; |
| 11595 | Newx(ret->offs, npar, regexp_paren_pair); |
| 11596 | Copy(r->offs, ret->offs, npar, regexp_paren_pair); |
| 11597 | if(ret->swap) { |
| 11598 | /* no need to copy these */ |
| 11599 | Newx(ret->swap, npar, regexp_paren_pair); |
| 11600 | } |
| 11601 | |
| 11602 | if (ret->substrs) { |
| 11603 | /* Do it this way to avoid reading from *r after the StructCopy(). |
| 11604 | That way, if any of the sv_dup_inc()s dislodge *r from the L1 |
| 11605 | cache, it doesn't matter. */ |
| 11606 | const bool anchored = r->check_substr |
| 11607 | ? r->check_substr == r->anchored_substr |
| 11608 | : r->check_utf8 == r->anchored_utf8; |
| 11609 | Newx(ret->substrs, 1, struct reg_substr_data); |
| 11610 | StructCopy(r->substrs, ret->substrs, struct reg_substr_data); |
| 11611 | |
| 11612 | ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param); |
| 11613 | ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param); |
| 11614 | ret->float_substr = sv_dup_inc(ret->float_substr, param); |
| 11615 | ret->float_utf8 = sv_dup_inc(ret->float_utf8, param); |
| 11616 | |
| 11617 | /* check_substr and check_utf8, if non-NULL, point to either their |
| 11618 | anchored or float namesakes, and don't hold a second reference. */ |
| 11619 | |
| 11620 | if (ret->check_substr) { |
| 11621 | if (anchored) { |
| 11622 | assert(r->check_utf8 == r->anchored_utf8); |
| 11623 | ret->check_substr = ret->anchored_substr; |
| 11624 | ret->check_utf8 = ret->anchored_utf8; |
| 11625 | } else { |
| 11626 | assert(r->check_substr == r->float_substr); |
| 11627 | assert(r->check_utf8 == r->float_utf8); |
| 11628 | ret->check_substr = ret->float_substr; |
| 11629 | ret->check_utf8 = ret->float_utf8; |
| 11630 | } |
| 11631 | } else if (ret->check_utf8) { |
| 11632 | if (anchored) { |
| 11633 | ret->check_utf8 = ret->anchored_utf8; |
| 11634 | } else { |
| 11635 | ret->check_utf8 = ret->float_utf8; |
| 11636 | } |
| 11637 | } |
| 11638 | } |
| 11639 | |
| 11640 | RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param); |
| 11641 | |
| 11642 | if (ret->pprivate) |
| 11643 | RXi_SET(ret,CALLREGDUPE_PVT(dstr,param)); |
| 11644 | |
| 11645 | if (RX_MATCH_COPIED(dstr)) |
| 11646 | ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen); |
| 11647 | else |
| 11648 | ret->subbeg = NULL; |
| 11649 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 11650 | ret->saved_copy = NULL; |
| 11651 | #endif |
| 11652 | |
| 11653 | if (ret->mother_re) { |
| 11654 | if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) { |
| 11655 | /* Our storage points directly to our mother regexp, but that's |
| 11656 | 1: a buffer in a different thread |
| 11657 | 2: something we no longer hold a reference on |
| 11658 | so we need to copy it locally. */ |
| 11659 | /* Note we need to sue SvCUR() on our mother_re, because it, in |
| 11660 | turn, may well be pointing to its own mother_re. */ |
| 11661 | SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re), |
| 11662 | SvCUR(ret->mother_re)+1)); |
| 11663 | SvLEN_set(dstr, SvCUR(ret->mother_re)+1); |
| 11664 | } |
| 11665 | ret->mother_re = NULL; |
| 11666 | } |
| 11667 | ret->gofs = 0; |
| 11668 | } |
| 11669 | #endif /* PERL_IN_XSUB_RE */ |
| 11670 | |
| 11671 | /* |
| 11672 | regdupe_internal() |
| 11673 | |
| 11674 | This is the internal complement to regdupe() which is used to copy |
| 11675 | the structure pointed to by the *pprivate pointer in the regexp. |
| 11676 | This is the core version of the extension overridable cloning hook. |
| 11677 | The regexp structure being duplicated will be copied by perl prior |
| 11678 | to this and will be provided as the regexp *r argument, however |
| 11679 | with the /old/ structures pprivate pointer value. Thus this routine |
| 11680 | may override any copying normally done by perl. |
| 11681 | |
| 11682 | It returns a pointer to the new regexp_internal structure. |
| 11683 | */ |
| 11684 | |
| 11685 | void * |
| 11686 | Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param) |
| 11687 | { |
| 11688 | dVAR; |
| 11689 | struct regexp *const r = (struct regexp *)SvANY(rx); |
| 11690 | regexp_internal *reti; |
| 11691 | int len; |
| 11692 | RXi_GET_DECL(r,ri); |
| 11693 | |
| 11694 | PERL_ARGS_ASSERT_REGDUPE_INTERNAL; |
| 11695 | |
| 11696 | len = ProgLen(ri); |
| 11697 | |
| 11698 | Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal); |
| 11699 | Copy(ri->program, reti->program, len+1, regnode); |
| 11700 | |
| 11701 | |
| 11702 | reti->regstclass = NULL; |
| 11703 | |
| 11704 | if (ri->data) { |
| 11705 | struct reg_data *d; |
| 11706 | const int count = ri->data->count; |
| 11707 | int i; |
| 11708 | |
| 11709 | Newxc(d, sizeof(struct reg_data) + count*sizeof(void *), |
| 11710 | char, struct reg_data); |
| 11711 | Newx(d->what, count, U8); |
| 11712 | |
| 11713 | d->count = count; |
| 11714 | for (i = 0; i < count; i++) { |
| 11715 | d->what[i] = ri->data->what[i]; |
| 11716 | switch (d->what[i]) { |
| 11717 | /* legal options are one of: sSfpontTua |
| 11718 | see also regcomp.h and pregfree() */ |
| 11719 | case 'a': /* actually an AV, but the dup function is identical. */ |
| 11720 | case 's': |
| 11721 | case 'S': |
| 11722 | case 'p': /* actually an AV, but the dup function is identical. */ |
| 11723 | case 'u': /* actually an HV, but the dup function is identical. */ |
| 11724 | d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param); |
| 11725 | break; |
| 11726 | case 'f': |
| 11727 | /* This is cheating. */ |
| 11728 | Newx(d->data[i], 1, struct regnode_charclass_class); |
| 11729 | StructCopy(ri->data->data[i], d->data[i], |
| 11730 | struct regnode_charclass_class); |
| 11731 | reti->regstclass = (regnode*)d->data[i]; |
| 11732 | break; |
| 11733 | case 'o': |
| 11734 | /* Compiled op trees are readonly and in shared memory, |
| 11735 | and can thus be shared without duplication. */ |
| 11736 | OP_REFCNT_LOCK; |
| 11737 | d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]); |
| 11738 | OP_REFCNT_UNLOCK; |
| 11739 | break; |
| 11740 | case 'T': |
| 11741 | /* Trie stclasses are readonly and can thus be shared |
| 11742 | * without duplication. We free the stclass in pregfree |
| 11743 | * when the corresponding reg_ac_data struct is freed. |
| 11744 | */ |
| 11745 | reti->regstclass= ri->regstclass; |
| 11746 | /* Fall through */ |
| 11747 | case 't': |
| 11748 | OP_REFCNT_LOCK; |
| 11749 | ((reg_trie_data*)ri->data->data[i])->refcount++; |
| 11750 | OP_REFCNT_UNLOCK; |
| 11751 | /* Fall through */ |
| 11752 | case 'n': |
| 11753 | d->data[i] = ri->data->data[i]; |
| 11754 | break; |
| 11755 | default: |
| 11756 | Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]); |
| 11757 | } |
| 11758 | } |
| 11759 | |
| 11760 | reti->data = d; |
| 11761 | } |
| 11762 | else |
| 11763 | reti->data = NULL; |
| 11764 | |
| 11765 | reti->name_list_idx = ri->name_list_idx; |
| 11766 | |
| 11767 | #ifdef RE_TRACK_PATTERN_OFFSETS |
| 11768 | if (ri->u.offsets) { |
| 11769 | Newx(reti->u.offsets, 2*len+1, U32); |
| 11770 | Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32); |
| 11771 | } |
| 11772 | #else |
| 11773 | SetProgLen(reti,len); |
| 11774 | #endif |
| 11775 | |
| 11776 | return (void*)reti; |
| 11777 | } |
| 11778 | |
| 11779 | #endif /* USE_ITHREADS */ |
| 11780 | |
| 11781 | #ifndef PERL_IN_XSUB_RE |
| 11782 | |
| 11783 | /* |
| 11784 | - regnext - dig the "next" pointer out of a node |
| 11785 | */ |
| 11786 | regnode * |
| 11787 | Perl_regnext(pTHX_ register regnode *p) |
| 11788 | { |
| 11789 | dVAR; |
| 11790 | register I32 offset; |
| 11791 | |
| 11792 | if (!p) |
| 11793 | return(NULL); |
| 11794 | |
| 11795 | if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */ |
| 11796 | Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX); |
| 11797 | } |
| 11798 | |
| 11799 | offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p)); |
| 11800 | if (offset == 0) |
| 11801 | return(NULL); |
| 11802 | |
| 11803 | return(p+offset); |
| 11804 | } |
| 11805 | #endif |
| 11806 | |
| 11807 | STATIC void |
| 11808 | S_re_croak2(pTHX_ const char* pat1,const char* pat2,...) |
| 11809 | { |
| 11810 | va_list args; |
| 11811 | STRLEN l1 = strlen(pat1); |
| 11812 | STRLEN l2 = strlen(pat2); |
| 11813 | char buf[512]; |
| 11814 | SV *msv; |
| 11815 | const char *message; |
| 11816 | |
| 11817 | PERL_ARGS_ASSERT_RE_CROAK2; |
| 11818 | |
| 11819 | if (l1 > 510) |
| 11820 | l1 = 510; |
| 11821 | if (l1 + l2 > 510) |
| 11822 | l2 = 510 - l1; |
| 11823 | Copy(pat1, buf, l1 , char); |
| 11824 | Copy(pat2, buf + l1, l2 , char); |
| 11825 | buf[l1 + l2] = '\n'; |
| 11826 | buf[l1 + l2 + 1] = '\0'; |
| 11827 | #ifdef I_STDARG |
| 11828 | /* ANSI variant takes additional second argument */ |
| 11829 | va_start(args, pat2); |
| 11830 | #else |
| 11831 | va_start(args); |
| 11832 | #endif |
| 11833 | msv = vmess(buf, &args); |
| 11834 | va_end(args); |
| 11835 | message = SvPV_const(msv,l1); |
| 11836 | if (l1 > 512) |
| 11837 | l1 = 512; |
| 11838 | Copy(message, buf, l1 , char); |
| 11839 | buf[l1-1] = '\0'; /* Overwrite \n */ |
| 11840 | Perl_croak(aTHX_ "%s", buf); |
| 11841 | } |
| 11842 | |
| 11843 | /* XXX Here's a total kludge. But we need to re-enter for swash routines. */ |
| 11844 | |
| 11845 | #ifndef PERL_IN_XSUB_RE |
| 11846 | void |
| 11847 | Perl_save_re_context(pTHX) |
| 11848 | { |
| 11849 | dVAR; |
| 11850 | |
| 11851 | struct re_save_state *state; |
| 11852 | |
| 11853 | SAVEVPTR(PL_curcop); |
| 11854 | SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1); |
| 11855 | |
| 11856 | state = (struct re_save_state *)(PL_savestack + PL_savestack_ix); |
| 11857 | PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE; |
| 11858 | SSPUSHUV(SAVEt_RE_STATE); |
| 11859 | |
| 11860 | Copy(&PL_reg_state, state, 1, struct re_save_state); |
| 11861 | |
| 11862 | PL_reg_start_tmp = 0; |
| 11863 | PL_reg_start_tmpl = 0; |
| 11864 | PL_reg_oldsaved = NULL; |
| 11865 | PL_reg_oldsavedlen = 0; |
| 11866 | PL_reg_maxiter = 0; |
| 11867 | PL_reg_leftiter = 0; |
| 11868 | PL_reg_poscache = NULL; |
| 11869 | PL_reg_poscache_size = 0; |
| 11870 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 11871 | PL_nrs = NULL; |
| 11872 | #endif |
| 11873 | |
| 11874 | /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */ |
| 11875 | if (PL_curpm) { |
| 11876 | const REGEXP * const rx = PM_GETRE(PL_curpm); |
| 11877 | if (rx) { |
| 11878 | U32 i; |
| 11879 | for (i = 1; i <= RX_NPARENS(rx); i++) { |
| 11880 | char digits[TYPE_CHARS(long)]; |
| 11881 | const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i); |
| 11882 | GV *const *const gvp |
| 11883 | = (GV**)hv_fetch(PL_defstash, digits, len, 0); |
| 11884 | |
| 11885 | if (gvp) { |
| 11886 | GV * const gv = *gvp; |
| 11887 | if (SvTYPE(gv) == SVt_PVGV && GvSV(gv)) |
| 11888 | save_scalar(gv); |
| 11889 | } |
| 11890 | } |
| 11891 | } |
| 11892 | } |
| 11893 | } |
| 11894 | #endif |
| 11895 | |
| 11896 | static void |
| 11897 | clear_re(pTHX_ void *r) |
| 11898 | { |
| 11899 | dVAR; |
| 11900 | ReREFCNT_dec((REGEXP *)r); |
| 11901 | } |
| 11902 | |
| 11903 | #ifdef DEBUGGING |
| 11904 | |
| 11905 | STATIC void |
| 11906 | S_put_byte(pTHX_ SV *sv, int c) |
| 11907 | { |
| 11908 | PERL_ARGS_ASSERT_PUT_BYTE; |
| 11909 | |
| 11910 | /* Our definition of isPRINT() ignores locales, so only bytes that are |
| 11911 | not part of UTF-8 are considered printable. I assume that the same |
| 11912 | holds for UTF-EBCDIC. |
| 11913 | Also, code point 255 is not printable in either (it's E0 in EBCDIC, |
| 11914 | which Wikipedia says: |
| 11915 | |
| 11916 | EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all |
| 11917 | ones (binary 1111 1111, hexadecimal FF). It is similar, but not |
| 11918 | identical, to the ASCII delete (DEL) or rubout control character. |
| 11919 | ) So the old condition can be simplified to !isPRINT(c) */ |
| 11920 | if (!isPRINT(c)) { |
| 11921 | if (c < 256) { |
| 11922 | Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c); |
| 11923 | } |
| 11924 | else { |
| 11925 | Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c); |
| 11926 | } |
| 11927 | } |
| 11928 | else { |
| 11929 | const char string = c; |
| 11930 | if (c == '-' || c == ']' || c == '\\' || c == '^') |
| 11931 | sv_catpvs(sv, "\\"); |
| 11932 | sv_catpvn(sv, &string, 1); |
| 11933 | } |
| 11934 | } |
| 11935 | |
| 11936 | |
| 11937 | #define CLEAR_OPTSTART \ |
| 11938 | if (optstart) STMT_START { \ |
| 11939 | DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \ |
| 11940 | optstart=NULL; \ |
| 11941 | } STMT_END |
| 11942 | |
| 11943 | #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1); |
| 11944 | |
| 11945 | STATIC const regnode * |
| 11946 | S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node, |
| 11947 | const regnode *last, const regnode *plast, |
| 11948 | SV* sv, I32 indent, U32 depth) |
| 11949 | { |
| 11950 | dVAR; |
| 11951 | register U8 op = PSEUDO; /* Arbitrary non-END op. */ |
| 11952 | register const regnode *next; |
| 11953 | const regnode *optstart= NULL; |
| 11954 | |
| 11955 | RXi_GET_DECL(r,ri); |
| 11956 | GET_RE_DEBUG_FLAGS_DECL; |
| 11957 | |
| 11958 | PERL_ARGS_ASSERT_DUMPUNTIL; |
| 11959 | |
| 11960 | #ifdef DEBUG_DUMPUNTIL |
| 11961 | PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start, |
| 11962 | last ? last-start : 0,plast ? plast-start : 0); |
| 11963 | #endif |
| 11964 | |
| 11965 | if (plast && plast < last) |
| 11966 | last= plast; |
| 11967 | |
| 11968 | while (PL_regkind[op] != END && (!last || node < last)) { |
| 11969 | /* While that wasn't END last time... */ |
| 11970 | NODE_ALIGN(node); |
| 11971 | op = OP(node); |
| 11972 | if (op == CLOSE || op == WHILEM) |
| 11973 | indent--; |
| 11974 | next = regnext((regnode *)node); |
| 11975 | |
| 11976 | /* Where, what. */ |
| 11977 | if (OP(node) == OPTIMIZED) { |
| 11978 | if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) |
| 11979 | optstart = node; |
| 11980 | else |
| 11981 | goto after_print; |
| 11982 | } else |
| 11983 | CLEAR_OPTSTART; |
| 11984 | |
| 11985 | regprop(r, sv, node); |
| 11986 | PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start), |
| 11987 | (int)(2*indent + 1), "", SvPVX_const(sv)); |
| 11988 | |
| 11989 | if (OP(node) != OPTIMIZED) { |
| 11990 | if (next == NULL) /* Next ptr. */ |
| 11991 | PerlIO_printf(Perl_debug_log, " (0)"); |
| 11992 | else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH ) |
| 11993 | PerlIO_printf(Perl_debug_log, " (FAIL)"); |
| 11994 | else |
| 11995 | PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start)); |
| 11996 | (void)PerlIO_putc(Perl_debug_log, '\n'); |
| 11997 | } |
| 11998 | |
| 11999 | after_print: |
| 12000 | if (PL_regkind[(U8)op] == BRANCHJ) { |
| 12001 | assert(next); |
| 12002 | { |
| 12003 | register const regnode *nnode = (OP(next) == LONGJMP |
| 12004 | ? regnext((regnode *)next) |
| 12005 | : next); |
| 12006 | if (last && nnode > last) |
| 12007 | nnode = last; |
| 12008 | DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode); |
| 12009 | } |
| 12010 | } |
| 12011 | else if (PL_regkind[(U8)op] == BRANCH) { |
| 12012 | assert(next); |
| 12013 | DUMPUNTIL(NEXTOPER(node), next); |
| 12014 | } |
| 12015 | else if ( PL_regkind[(U8)op] == TRIE ) { |
| 12016 | const regnode *this_trie = node; |
| 12017 | const char op = OP(node); |
| 12018 | const U32 n = ARG(node); |
| 12019 | const reg_ac_data * const ac = op>=AHOCORASICK ? |
| 12020 | (reg_ac_data *)ri->data->data[n] : |
| 12021 | NULL; |
| 12022 | const reg_trie_data * const trie = |
| 12023 | (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie]; |
| 12024 | #ifdef DEBUGGING |
| 12025 | AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]); |
| 12026 | #endif |
| 12027 | const regnode *nextbranch= NULL; |
| 12028 | I32 word_idx; |
| 12029 | sv_setpvs(sv, ""); |
| 12030 | for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) { |
| 12031 | SV ** const elem_ptr = av_fetch(trie_words,word_idx,0); |
| 12032 | |
| 12033 | PerlIO_printf(Perl_debug_log, "%*s%s ", |
| 12034 | (int)(2*(indent+3)), "", |
| 12035 | elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60, |
| 12036 | PL_colors[0], PL_colors[1], |
| 12037 | (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) | |
| 12038 | PERL_PV_PRETTY_ELLIPSES | |
| 12039 | PERL_PV_PRETTY_LTGT |
| 12040 | ) |
| 12041 | : "???" |
| 12042 | ); |
| 12043 | if (trie->jump) { |
| 12044 | U16 dist= trie->jump[word_idx+1]; |
| 12045 | PerlIO_printf(Perl_debug_log, "(%"UVuf")\n", |
| 12046 | (UV)((dist ? this_trie + dist : next) - start)); |
| 12047 | if (dist) { |
| 12048 | if (!nextbranch) |
| 12049 | nextbranch= this_trie + trie->jump[0]; |
| 12050 | DUMPUNTIL(this_trie + dist, nextbranch); |
| 12051 | } |
| 12052 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) |
| 12053 | nextbranch= regnext((regnode *)nextbranch); |
| 12054 | } else { |
| 12055 | PerlIO_printf(Perl_debug_log, "\n"); |
| 12056 | } |
| 12057 | } |
| 12058 | if (last && next > last) |
| 12059 | node= last; |
| 12060 | else |
| 12061 | node= next; |
| 12062 | } |
| 12063 | else if ( op == CURLY ) { /* "next" might be very big: optimizer */ |
| 12064 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, |
| 12065 | NEXTOPER(node) + EXTRA_STEP_2ARGS + 1); |
| 12066 | } |
| 12067 | else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) { |
| 12068 | assert(next); |
| 12069 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next); |
| 12070 | } |
| 12071 | else if ( op == PLUS || op == STAR) { |
| 12072 | DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1); |
| 12073 | } |
| 12074 | else if (PL_regkind[(U8)op] == ANYOF) { |
| 12075 | /* arglen 1 + class block */ |
| 12076 | node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS) |
| 12077 | ? ANYOF_CLASS_SKIP : ANYOF_SKIP); |
| 12078 | node = NEXTOPER(node); |
| 12079 | } |
| 12080 | else if (PL_regkind[(U8)op] == EXACT) { |
| 12081 | /* Literal string, where present. */ |
| 12082 | node += NODE_SZ_STR(node) - 1; |
| 12083 | node = NEXTOPER(node); |
| 12084 | } |
| 12085 | else { |
| 12086 | node = NEXTOPER(node); |
| 12087 | node += regarglen[(U8)op]; |
| 12088 | } |
| 12089 | if (op == CURLYX || op == OPEN) |
| 12090 | indent++; |
| 12091 | } |
| 12092 | CLEAR_OPTSTART; |
| 12093 | #ifdef DEBUG_DUMPUNTIL |
| 12094 | PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent); |
| 12095 | #endif |
| 12096 | return node; |
| 12097 | } |
| 12098 | |
| 12099 | #endif /* DEBUGGING */ |
| 12100 | |
| 12101 | /* |
| 12102 | * Local variables: |
| 12103 | * c-indentation-style: bsd |
| 12104 | * c-basic-offset: 4 |
| 12105 | * indent-tabs-mode: t |
| 12106 | * End: |
| 12107 | * |
| 12108 | * ex: set ts=8 sts=4 sw=4 noet: |
| 12109 | */ |