| 1 | /* regexec.c |
| 2 | */ |
| 3 | |
| 4 | /* |
| 5 | * One Ring to rule them all, One Ring to find them |
| 6 | * |
| 7 | * [p.v of _The Lord of the Rings_, opening poem] |
| 8 | * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"] |
| 9 | * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"] |
| 10 | */ |
| 11 | |
| 12 | /* This file contains functions for executing a regular expression. See |
| 13 | * also regcomp.c which funnily enough, contains functions for compiling |
| 14 | * a regular expression. |
| 15 | * |
| 16 | * This file is also copied at build time to ext/re/re_exec.c, where |
| 17 | * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT. |
| 18 | * This causes the main functions to be compiled under new names and with |
| 19 | * debugging support added, which makes "use re 'debug'" work. |
| 20 | */ |
| 21 | |
| 22 | /* NOTE: this is derived from Henry Spencer's regexp code, and should not |
| 23 | * confused with the original package (see point 3 below). Thanks, Henry! |
| 24 | */ |
| 25 | |
| 26 | /* Additional note: this code is very heavily munged from Henry's version |
| 27 | * in places. In some spots I've traded clarity for efficiency, so don't |
| 28 | * blame Henry for some of the lack of readability. |
| 29 | */ |
| 30 | |
| 31 | /* The names of the functions have been changed from regcomp and |
| 32 | * regexec to pregcomp and pregexec in order to avoid conflicts |
| 33 | * with the POSIX routines of the same names. |
| 34 | */ |
| 35 | |
| 36 | #ifdef PERL_EXT_RE_BUILD |
| 37 | #include "re_top.h" |
| 38 | #endif |
| 39 | |
| 40 | /* |
| 41 | * pregcomp and pregexec -- regsub and regerror are not used in perl |
| 42 | * |
| 43 | * Copyright (c) 1986 by University of Toronto. |
| 44 | * Written by Henry Spencer. Not derived from licensed software. |
| 45 | * |
| 46 | * Permission is granted to anyone to use this software for any |
| 47 | * purpose on any computer system, and to redistribute it freely, |
| 48 | * subject to the following restrictions: |
| 49 | * |
| 50 | * 1. The author is not responsible for the consequences of use of |
| 51 | * this software, no matter how awful, even if they arise |
| 52 | * from defects in it. |
| 53 | * |
| 54 | * 2. The origin of this software must not be misrepresented, either |
| 55 | * by explicit claim or by omission. |
| 56 | * |
| 57 | * 3. Altered versions must be plainly marked as such, and must not |
| 58 | * be misrepresented as being the original software. |
| 59 | * |
| 60 | **** Alterations to Henry's code are... |
| 61 | **** |
| 62 | **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
| 63 | **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
| 64 | **** by Larry Wall and others |
| 65 | **** |
| 66 | **** You may distribute under the terms of either the GNU General Public |
| 67 | **** License or the Artistic License, as specified in the README file. |
| 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_REGEXEC_C |
| 75 | #include "perl.h" |
| 76 | |
| 77 | #ifdef PERL_IN_XSUB_RE |
| 78 | # include "re_comp.h" |
| 79 | #else |
| 80 | # include "regcomp.h" |
| 81 | #endif |
| 82 | |
| 83 | #include "invlist_inline.h" |
| 84 | #include "unicode_constants.h" |
| 85 | |
| 86 | #define B_ON_NON_UTF8_LOCALE_IS_WRONG \ |
| 87 | "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale" |
| 88 | |
| 89 | static const char utf8_locale_required[] = |
| 90 | "Use of (?[ ]) for non-UTF-8 locale is wrong. Assuming a UTF-8 locale"; |
| 91 | |
| 92 | #ifdef DEBUGGING |
| 93 | /* At least one required character in the target string is expressible only in |
| 94 | * UTF-8. */ |
| 95 | static const char* const non_utf8_target_but_utf8_required |
| 96 | = "Can't match, because target string needs to be in UTF-8\n"; |
| 97 | #endif |
| 98 | |
| 99 | #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \ |
| 100 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%s", non_utf8_target_but_utf8_required));\ |
| 101 | goto target; \ |
| 102 | } STMT_END |
| 103 | |
| 104 | #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i) |
| 105 | |
| 106 | #ifndef STATIC |
| 107 | #define STATIC static |
| 108 | #endif |
| 109 | |
| 110 | /* Valid only if 'c', the character being looke-up, is an invariant under |
| 111 | * UTF-8: it avoids the reginclass call if there are no complications: i.e., if |
| 112 | * everything matchable is straight forward in the bitmap */ |
| 113 | #define REGINCLASS(prog,p,c,u) (ANYOF_FLAGS(p) \ |
| 114 | ? reginclass(prog,p,c,c+1,u) \ |
| 115 | : ANYOF_BITMAP_TEST(p,*(c))) |
| 116 | |
| 117 | /* |
| 118 | * Forwards. |
| 119 | */ |
| 120 | |
| 121 | #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv)) |
| 122 | #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b) |
| 123 | |
| 124 | #define HOPc(pos,off) \ |
| 125 | (char *)(reginfo->is_utf8_target \ |
| 126 | ? reghop3((U8*)pos, off, \ |
| 127 | (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \ |
| 128 | : (U8*)(pos + off)) |
| 129 | |
| 130 | #define HOPBACKc(pos, off) \ |
| 131 | (char*)(reginfo->is_utf8_target \ |
| 132 | ? reghopmaybe3((U8*)pos, (SSize_t)0-off, (U8*)(reginfo->strbeg)) \ |
| 133 | : (pos - off >= reginfo->strbeg) \ |
| 134 | ? (U8*)pos - off \ |
| 135 | : NULL) |
| 136 | |
| 137 | #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off)) |
| 138 | #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) |
| 139 | |
| 140 | /* lim must be +ve. Returns NULL on overshoot */ |
| 141 | #define HOPMAYBE3(pos,off,lim) \ |
| 142 | (reginfo->is_utf8_target \ |
| 143 | ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \ |
| 144 | : ((U8*)pos + off <= lim) \ |
| 145 | ? (U8*)pos + off \ |
| 146 | : NULL) |
| 147 | |
| 148 | /* like HOP3, but limits the result to <= lim even for the non-utf8 case. |
| 149 | * off must be >=0; args should be vars rather than expressions */ |
| 150 | #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \ |
| 151 | ? reghop3((U8*)(pos), off, (U8*)(lim)) \ |
| 152 | : (U8*)((pos + off) > lim ? lim : (pos + off))) |
| 153 | |
| 154 | #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \ |
| 155 | ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \ |
| 156 | : (U8*)(pos + off)) |
| 157 | #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim)) |
| 158 | |
| 159 | #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */ |
| 160 | #define NEXTCHR_IS_EOS (nextchr < 0) |
| 161 | |
| 162 | #define SET_nextchr \ |
| 163 | nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS) |
| 164 | |
| 165 | #define SET_locinput(p) \ |
| 166 | locinput = (p); \ |
| 167 | SET_nextchr |
| 168 | |
| 169 | |
| 170 | #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \ |
| 171 | if (!swash_ptr) { \ |
| 172 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \ |
| 173 | swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \ |
| 174 | 1, 0, invlist, &flags); \ |
| 175 | assert(swash_ptr); \ |
| 176 | } \ |
| 177 | } STMT_END |
| 178 | |
| 179 | /* If in debug mode, we test that a known character properly matches */ |
| 180 | #ifdef DEBUGGING |
| 181 | # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \ |
| 182 | property_name, \ |
| 183 | invlist, \ |
| 184 | utf8_char_in_property) \ |
| 185 | LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \ |
| 186 | assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE)); |
| 187 | #else |
| 188 | # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \ |
| 189 | property_name, \ |
| 190 | invlist, \ |
| 191 | utf8_char_in_property) \ |
| 192 | LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) |
| 193 | #endif |
| 194 | |
| 195 | #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \ |
| 196 | PL_utf8_swash_ptrs[_CC_WORDCHAR], \ |
| 197 | "", \ |
| 198 | PL_XPosix_ptrs[_CC_WORDCHAR], \ |
| 199 | LATIN_SMALL_LIGATURE_LONG_S_T_UTF8); |
| 200 | |
| 201 | #define PLACEHOLDER /* Something for the preprocessor to grab onto */ |
| 202 | /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */ |
| 203 | |
| 204 | /* for use after a quantifier and before an EXACT-like node -- japhy */ |
| 205 | /* it would be nice to rework regcomp.sym to generate this stuff. sigh |
| 206 | * |
| 207 | * NOTE that *nothing* that affects backtracking should be in here, specifically |
| 208 | * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a |
| 209 | * node that is in between two EXACT like nodes when ascertaining what the required |
| 210 | * "follow" character is. This should probably be moved to regex compile time |
| 211 | * although it may be done at run time beause of the REF possibility - more |
| 212 | * investigation required. -- demerphq |
| 213 | */ |
| 214 | #define JUMPABLE(rn) ( \ |
| 215 | OP(rn) == OPEN || \ |
| 216 | (OP(rn) == CLOSE && \ |
| 217 | !EVAL_CLOSE_PAREN_IS(cur_eval,ARG(rn)) ) || \ |
| 218 | OP(rn) == EVAL || \ |
| 219 | OP(rn) == SUSPEND || OP(rn) == IFMATCH || \ |
| 220 | OP(rn) == PLUS || OP(rn) == MINMOD || \ |
| 221 | OP(rn) == KEEPS || \ |
| 222 | (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \ |
| 223 | ) |
| 224 | #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT) |
| 225 | |
| 226 | #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF ) |
| 227 | |
| 228 | #if 0 |
| 229 | /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so |
| 230 | we don't need this definition. XXX These are now out-of-sync*/ |
| 231 | #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF ) |
| 232 | #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF ) |
| 233 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL ) |
| 234 | |
| 235 | #else |
| 236 | /* ... so we use this as its faster. */ |
| 237 | #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==EXACTL ) |
| 238 | #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFLU8 || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE) |
| 239 | #define IS_TEXTF(rn) ( OP(rn)==EXACTF ) |
| 240 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL ) |
| 241 | |
| 242 | #endif |
| 243 | |
| 244 | /* |
| 245 | Search for mandatory following text node; for lookahead, the text must |
| 246 | follow but for lookbehind (rn->flags != 0) we skip to the next step. |
| 247 | */ |
| 248 | #define FIND_NEXT_IMPT(rn) STMT_START { \ |
| 249 | while (JUMPABLE(rn)) { \ |
| 250 | const OPCODE type = OP(rn); \ |
| 251 | if (type == SUSPEND || PL_regkind[type] == CURLY) \ |
| 252 | rn = NEXTOPER(NEXTOPER(rn)); \ |
| 253 | else if (type == PLUS) \ |
| 254 | rn = NEXTOPER(rn); \ |
| 255 | else if (type == IFMATCH) \ |
| 256 | rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \ |
| 257 | else rn += NEXT_OFF(rn); \ |
| 258 | } \ |
| 259 | } STMT_END |
| 260 | |
| 261 | #define SLAB_FIRST(s) (&(s)->states[0]) |
| 262 | #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1]) |
| 263 | |
| 264 | static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo); |
| 265 | static void S_cleanup_regmatch_info_aux(pTHX_ void *arg); |
| 266 | static regmatch_state * S_push_slab(pTHX); |
| 267 | |
| 268 | #define REGCP_PAREN_ELEMS 3 |
| 269 | #define REGCP_OTHER_ELEMS 3 |
| 270 | #define REGCP_FRAME_ELEMS 1 |
| 271 | /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and |
| 272 | * are needed for the regexp context stack bookkeeping. */ |
| 273 | |
| 274 | STATIC CHECKPOINT |
| 275 | S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen) |
| 276 | { |
| 277 | const int retval = PL_savestack_ix; |
| 278 | const int paren_elems_to_push = |
| 279 | (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS; |
| 280 | const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS; |
| 281 | const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT; |
| 282 | I32 p; |
| 283 | GET_RE_DEBUG_FLAGS_DECL; |
| 284 | |
| 285 | PERL_ARGS_ASSERT_REGCPPUSH; |
| 286 | |
| 287 | if (paren_elems_to_push < 0) |
| 288 | Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u", |
| 289 | (int)paren_elems_to_push, (int)maxopenparen, |
| 290 | (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS); |
| 291 | |
| 292 | if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems) |
| 293 | Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf |
| 294 | " out of range (%lu-%ld)", |
| 295 | total_elems, |
| 296 | (unsigned long)maxopenparen, |
| 297 | (long)parenfloor); |
| 298 | |
| 299 | SSGROW(total_elems + REGCP_FRAME_ELEMS); |
| 300 | |
| 301 | DEBUG_BUFFERS_r( |
| 302 | if ((int)maxopenparen > (int)parenfloor) |
| 303 | Perl_re_printf( aTHX_ |
| 304 | "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n", |
| 305 | PTR2UV(rex), |
| 306 | PTR2UV(rex->offs) |
| 307 | ); |
| 308 | ); |
| 309 | for (p = parenfloor+1; p <= (I32)maxopenparen; p++) { |
| 310 | /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */ |
| 311 | SSPUSHIV(rex->offs[p].end); |
| 312 | SSPUSHIV(rex->offs[p].start); |
| 313 | SSPUSHINT(rex->offs[p].start_tmp); |
| 314 | DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ |
| 315 | " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n", |
| 316 | (UV)p, |
| 317 | (IV)rex->offs[p].start, |
| 318 | (IV)rex->offs[p].start_tmp, |
| 319 | (IV)rex->offs[p].end |
| 320 | )); |
| 321 | } |
| 322 | /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */ |
| 323 | SSPUSHINT(maxopenparen); |
| 324 | SSPUSHINT(rex->lastparen); |
| 325 | SSPUSHINT(rex->lastcloseparen); |
| 326 | SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */ |
| 327 | |
| 328 | return retval; |
| 329 | } |
| 330 | |
| 331 | /* These are needed since we do not localize EVAL nodes: */ |
| 332 | #define REGCP_SET(cp) \ |
| 333 | DEBUG_STATE_r( \ |
| 334 | Perl_re_exec_indentf( aTHX_ \ |
| 335 | "Setting an EVAL scope, savestack=%"IVdf",\n", \ |
| 336 | depth, (IV)PL_savestack_ix \ |
| 337 | ) \ |
| 338 | ); \ |
| 339 | cp = PL_savestack_ix |
| 340 | |
| 341 | #define REGCP_UNWIND(cp) \ |
| 342 | DEBUG_STATE_r( \ |
| 343 | if (cp != PL_savestack_ix) \ |
| 344 | Perl_re_exec_indentf( aTHX_ \ |
| 345 | "Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n",\ |
| 346 | depth, (IV)(cp), (IV)PL_savestack_ix \ |
| 347 | ) \ |
| 348 | ); \ |
| 349 | regcpblow(cp) |
| 350 | |
| 351 | #define UNWIND_PAREN(lp, lcp) \ |
| 352 | for (n = rex->lastparen; n > lp; n--) \ |
| 353 | rex->offs[n].end = -1; \ |
| 354 | rex->lastparen = n; \ |
| 355 | rex->lastcloseparen = lcp; |
| 356 | |
| 357 | |
| 358 | STATIC void |
| 359 | S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p) |
| 360 | { |
| 361 | UV i; |
| 362 | U32 paren; |
| 363 | GET_RE_DEBUG_FLAGS_DECL; |
| 364 | |
| 365 | PERL_ARGS_ASSERT_REGCPPOP; |
| 366 | |
| 367 | /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */ |
| 368 | i = SSPOPUV; |
| 369 | assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */ |
| 370 | i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */ |
| 371 | rex->lastcloseparen = SSPOPINT; |
| 372 | rex->lastparen = SSPOPINT; |
| 373 | *maxopenparen_p = SSPOPINT; |
| 374 | |
| 375 | i -= REGCP_OTHER_ELEMS; |
| 376 | /* Now restore the parentheses context. */ |
| 377 | DEBUG_BUFFERS_r( |
| 378 | if (i || rex->lastparen + 1 <= rex->nparens) |
| 379 | Perl_re_printf( aTHX_ |
| 380 | "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n", |
| 381 | PTR2UV(rex), |
| 382 | PTR2UV(rex->offs) |
| 383 | ); |
| 384 | ); |
| 385 | paren = *maxopenparen_p; |
| 386 | for ( ; i > 0; i -= REGCP_PAREN_ELEMS) { |
| 387 | SSize_t tmps; |
| 388 | rex->offs[paren].start_tmp = SSPOPINT; |
| 389 | rex->offs[paren].start = SSPOPIV; |
| 390 | tmps = SSPOPIV; |
| 391 | if (paren <= rex->lastparen) |
| 392 | rex->offs[paren].end = tmps; |
| 393 | DEBUG_BUFFERS_r( Perl_re_printf( aTHX_ |
| 394 | " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n", |
| 395 | (UV)paren, |
| 396 | (IV)rex->offs[paren].start, |
| 397 | (IV)rex->offs[paren].start_tmp, |
| 398 | (IV)rex->offs[paren].end, |
| 399 | (paren > rex->lastparen ? "(skipped)" : "")); |
| 400 | ); |
| 401 | paren--; |
| 402 | } |
| 403 | #if 1 |
| 404 | /* It would seem that the similar code in regtry() |
| 405 | * already takes care of this, and in fact it is in |
| 406 | * a better location to since this code can #if 0-ed out |
| 407 | * but the code in regtry() is needed or otherwise tests |
| 408 | * requiring null fields (pat.t#187 and split.t#{13,14} |
| 409 | * (as of patchlevel 7877) will fail. Then again, |
| 410 | * this code seems to be necessary or otherwise |
| 411 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ |
| 412 | * --jhi updated by dapm */ |
| 413 | for (i = rex->lastparen + 1; i <= rex->nparens; i++) { |
| 414 | if (i > *maxopenparen_p) |
| 415 | rex->offs[i].start = -1; |
| 416 | rex->offs[i].end = -1; |
| 417 | DEBUG_BUFFERS_r( Perl_re_printf( aTHX_ |
| 418 | " \\%"UVuf": %s ..-1 undeffing\n", |
| 419 | (UV)i, |
| 420 | (i > *maxopenparen_p) ? "-1" : " " |
| 421 | )); |
| 422 | } |
| 423 | #endif |
| 424 | } |
| 425 | |
| 426 | /* restore the parens and associated vars at savestack position ix, |
| 427 | * but without popping the stack */ |
| 428 | |
| 429 | STATIC void |
| 430 | S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p) |
| 431 | { |
| 432 | I32 tmpix = PL_savestack_ix; |
| 433 | PL_savestack_ix = ix; |
| 434 | regcppop(rex, maxopenparen_p); |
| 435 | PL_savestack_ix = tmpix; |
| 436 | } |
| 437 | |
| 438 | #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */ |
| 439 | |
| 440 | STATIC bool |
| 441 | S_isFOO_lc(pTHX_ const U8 classnum, const U8 character) |
| 442 | { |
| 443 | /* Returns a boolean as to whether or not 'character' is a member of the |
| 444 | * Posix character class given by 'classnum' that should be equivalent to a |
| 445 | * value in the typedef '_char_class_number'. |
| 446 | * |
| 447 | * Ideally this could be replaced by a just an array of function pointers |
| 448 | * to the C library functions that implement the macros this calls. |
| 449 | * However, to compile, the precise function signatures are required, and |
| 450 | * these may vary from platform to to platform. To avoid having to figure |
| 451 | * out what those all are on each platform, I (khw) am using this method, |
| 452 | * which adds an extra layer of function call overhead (unless the C |
| 453 | * optimizer strips it away). But we don't particularly care about |
| 454 | * performance with locales anyway. */ |
| 455 | |
| 456 | switch ((_char_class_number) classnum) { |
| 457 | case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character); |
| 458 | case _CC_ENUM_ALPHA: return isALPHA_LC(character); |
| 459 | case _CC_ENUM_ASCII: return isASCII_LC(character); |
| 460 | case _CC_ENUM_BLANK: return isBLANK_LC(character); |
| 461 | case _CC_ENUM_CASED: return isLOWER_LC(character) |
| 462 | || isUPPER_LC(character); |
| 463 | case _CC_ENUM_CNTRL: return isCNTRL_LC(character); |
| 464 | case _CC_ENUM_DIGIT: return isDIGIT_LC(character); |
| 465 | case _CC_ENUM_GRAPH: return isGRAPH_LC(character); |
| 466 | case _CC_ENUM_LOWER: return isLOWER_LC(character); |
| 467 | case _CC_ENUM_PRINT: return isPRINT_LC(character); |
| 468 | case _CC_ENUM_PUNCT: return isPUNCT_LC(character); |
| 469 | case _CC_ENUM_SPACE: return isSPACE_LC(character); |
| 470 | case _CC_ENUM_UPPER: return isUPPER_LC(character); |
| 471 | case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character); |
| 472 | case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character); |
| 473 | default: /* VERTSPACE should never occur in locales */ |
| 474 | Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum); |
| 475 | } |
| 476 | |
| 477 | NOT_REACHED; /* NOTREACHED */ |
| 478 | return FALSE; |
| 479 | } |
| 480 | |
| 481 | STATIC bool |
| 482 | S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character) |
| 483 | { |
| 484 | /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded |
| 485 | * 'character' is a member of the Posix character class given by 'classnum' |
| 486 | * that should be equivalent to a value in the typedef |
| 487 | * '_char_class_number'. |
| 488 | * |
| 489 | * This just calls isFOO_lc on the code point for the character if it is in |
| 490 | * the range 0-255. Outside that range, all characters use Unicode |
| 491 | * rules, ignoring any locale. So use the Unicode function if this class |
| 492 | * requires a swash, and use the Unicode macro otherwise. */ |
| 493 | |
| 494 | PERL_ARGS_ASSERT_ISFOO_UTF8_LC; |
| 495 | |
| 496 | if (UTF8_IS_INVARIANT(*character)) { |
| 497 | return isFOO_lc(classnum, *character); |
| 498 | } |
| 499 | else if (UTF8_IS_DOWNGRADEABLE_START(*character)) { |
| 500 | return isFOO_lc(classnum, |
| 501 | EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1))); |
| 502 | } |
| 503 | |
| 504 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, character + UTF8SKIP(character)); |
| 505 | |
| 506 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 507 | |
| 508 | /* Initialize the swash unless done already */ |
| 509 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 510 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 511 | PL_utf8_swash_ptrs[classnum] = |
| 512 | _core_swash_init("utf8", |
| 513 | "", |
| 514 | &PL_sv_undef, 1, 0, |
| 515 | PL_XPosix_ptrs[classnum], &flags); |
| 516 | } |
| 517 | |
| 518 | return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *) |
| 519 | character, |
| 520 | TRUE /* is UTF */ )); |
| 521 | } |
| 522 | |
| 523 | switch ((_char_class_number) classnum) { |
| 524 | case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character); |
| 525 | case _CC_ENUM_BLANK: return is_HORIZWS_high(character); |
| 526 | case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character); |
| 527 | case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character); |
| 528 | default: break; |
| 529 | } |
| 530 | |
| 531 | return FALSE; /* Things like CNTRL are always below 256 */ |
| 532 | } |
| 533 | |
| 534 | /* |
| 535 | * pregexec and friends |
| 536 | */ |
| 537 | |
| 538 | #ifndef PERL_IN_XSUB_RE |
| 539 | /* |
| 540 | - pregexec - match a regexp against a string |
| 541 | */ |
| 542 | I32 |
| 543 | Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend, |
| 544 | char *strbeg, SSize_t minend, SV *screamer, U32 nosave) |
| 545 | /* stringarg: the point in the string at which to begin matching */ |
| 546 | /* strend: pointer to null at end of string */ |
| 547 | /* strbeg: real beginning of string */ |
| 548 | /* minend: end of match must be >= minend bytes after stringarg. */ |
| 549 | /* screamer: SV being matched: only used for utf8 flag, pos() etc; string |
| 550 | * itself is accessed via the pointers above */ |
| 551 | /* nosave: For optimizations. */ |
| 552 | { |
| 553 | PERL_ARGS_ASSERT_PREGEXEC; |
| 554 | |
| 555 | return |
| 556 | regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL, |
| 557 | nosave ? 0 : REXEC_COPY_STR); |
| 558 | } |
| 559 | #endif |
| 560 | |
| 561 | |
| 562 | |
| 563 | /* re_intuit_start(): |
| 564 | * |
| 565 | * Based on some optimiser hints, try to find the earliest position in the |
| 566 | * string where the regex could match. |
| 567 | * |
| 568 | * rx: the regex to match against |
| 569 | * sv: the SV being matched: only used for utf8 flag; the string |
| 570 | * itself is accessed via the pointers below. Note that on |
| 571 | * something like an overloaded SV, SvPOK(sv) may be false |
| 572 | * and the string pointers may point to something unrelated to |
| 573 | * the SV itself. |
| 574 | * strbeg: real beginning of string |
| 575 | * strpos: the point in the string at which to begin matching |
| 576 | * strend: pointer to the byte following the last char of the string |
| 577 | * flags currently unused; set to 0 |
| 578 | * data: currently unused; set to NULL |
| 579 | * |
| 580 | * The basic idea of re_intuit_start() is to use some known information |
| 581 | * about the pattern, namely: |
| 582 | * |
| 583 | * a) the longest known anchored substring (i.e. one that's at a |
| 584 | * constant offset from the beginning of the pattern; but not |
| 585 | * necessarily at a fixed offset from the beginning of the |
| 586 | * string); |
| 587 | * b) the longest floating substring (i.e. one that's not at a constant |
| 588 | * offset from the beginning of the pattern); |
| 589 | * c) Whether the pattern is anchored to the string; either |
| 590 | * an absolute anchor: /^../, or anchored to \n: /^.../m, |
| 591 | * or anchored to pos(): /\G/; |
| 592 | * d) A start class: a real or synthetic character class which |
| 593 | * represents which characters are legal at the start of the pattern; |
| 594 | * |
| 595 | * to either quickly reject the match, or to find the earliest position |
| 596 | * within the string at which the pattern might match, thus avoiding |
| 597 | * running the full NFA engine at those earlier locations, only to |
| 598 | * eventually fail and retry further along. |
| 599 | * |
| 600 | * Returns NULL if the pattern can't match, or returns the address within |
| 601 | * the string which is the earliest place the match could occur. |
| 602 | * |
| 603 | * The longest of the anchored and floating substrings is called 'check' |
| 604 | * and is checked first. The other is called 'other' and is checked |
| 605 | * second. The 'other' substring may not be present. For example, |
| 606 | * |
| 607 | * /(abc|xyz)ABC\d{0,3}DEFG/ |
| 608 | * |
| 609 | * will have |
| 610 | * |
| 611 | * check substr (float) = "DEFG", offset 6..9 chars |
| 612 | * other substr (anchored) = "ABC", offset 3..3 chars |
| 613 | * stclass = [ax] |
| 614 | * |
| 615 | * Be aware that during the course of this function, sometimes 'anchored' |
| 616 | * refers to a substring being anchored relative to the start of the |
| 617 | * pattern, and sometimes to the pattern itself being anchored relative to |
| 618 | * the string. For example: |
| 619 | * |
| 620 | * /\dabc/: "abc" is anchored to the pattern; |
| 621 | * /^\dabc/: "abc" is anchored to the pattern and the string; |
| 622 | * /\d+abc/: "abc" is anchored to neither the pattern nor the string; |
| 623 | * /^\d+abc/: "abc" is anchored to neither the pattern nor the string, |
| 624 | * but the pattern is anchored to the string. |
| 625 | */ |
| 626 | |
| 627 | char * |
| 628 | Perl_re_intuit_start(pTHX_ |
| 629 | REGEXP * const rx, |
| 630 | SV *sv, |
| 631 | const char * const strbeg, |
| 632 | char *strpos, |
| 633 | char *strend, |
| 634 | const U32 flags, |
| 635 | re_scream_pos_data *data) |
| 636 | { |
| 637 | struct regexp *const prog = ReANY(rx); |
| 638 | SSize_t start_shift = prog->check_offset_min; |
| 639 | /* Should be nonnegative! */ |
| 640 | SSize_t end_shift = 0; |
| 641 | /* current lowest pos in string where the regex can start matching */ |
| 642 | char *rx_origin = strpos; |
| 643 | SV *check; |
| 644 | const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */ |
| 645 | U8 other_ix = 1 - prog->substrs->check_ix; |
| 646 | bool ml_anch = 0; |
| 647 | char *other_last = strpos;/* latest pos 'other' substr already checked to */ |
| 648 | char *check_at = NULL; /* check substr found at this pos */ |
| 649 | const I32 multiline = prog->extflags & RXf_PMf_MULTILINE; |
| 650 | RXi_GET_DECL(prog,progi); |
| 651 | regmatch_info reginfo_buf; /* create some info to pass to find_byclass */ |
| 652 | regmatch_info *const reginfo = ®info_buf; |
| 653 | GET_RE_DEBUG_FLAGS_DECL; |
| 654 | |
| 655 | PERL_ARGS_ASSERT_RE_INTUIT_START; |
| 656 | PERL_UNUSED_ARG(flags); |
| 657 | PERL_UNUSED_ARG(data); |
| 658 | |
| 659 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 660 | "Intuit: trying to determine minimum start position...\n")); |
| 661 | |
| 662 | /* for now, assume that all substr offsets are positive. If at some point |
| 663 | * in the future someone wants to do clever things with lookbehind and |
| 664 | * -ve offsets, they'll need to fix up any code in this function |
| 665 | * which uses these offsets. See the thread beginning |
| 666 | * <20140113145929.GF27210@iabyn.com> |
| 667 | */ |
| 668 | assert(prog->substrs->data[0].min_offset >= 0); |
| 669 | assert(prog->substrs->data[0].max_offset >= 0); |
| 670 | assert(prog->substrs->data[1].min_offset >= 0); |
| 671 | assert(prog->substrs->data[1].max_offset >= 0); |
| 672 | assert(prog->substrs->data[2].min_offset >= 0); |
| 673 | assert(prog->substrs->data[2].max_offset >= 0); |
| 674 | |
| 675 | /* for now, assume that if both present, that the floating substring |
| 676 | * doesn't start before the anchored substring. |
| 677 | * If you break this assumption (e.g. doing better optimisations |
| 678 | * with lookahead/behind), then you'll need to audit the code in this |
| 679 | * function carefully first |
| 680 | */ |
| 681 | assert( |
| 682 | ! ( (prog->anchored_utf8 || prog->anchored_substr) |
| 683 | && (prog->float_utf8 || prog->float_substr)) |
| 684 | || (prog->float_min_offset >= prog->anchored_offset)); |
| 685 | |
| 686 | /* byte rather than char calculation for efficiency. It fails |
| 687 | * to quickly reject some cases that can't match, but will reject |
| 688 | * them later after doing full char arithmetic */ |
| 689 | if (prog->minlen > strend - strpos) { |
| 690 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 691 | " String too short...\n")); |
| 692 | goto fail; |
| 693 | } |
| 694 | |
| 695 | RX_MATCH_UTF8_set(rx,utf8_target); |
| 696 | reginfo->is_utf8_target = cBOOL(utf8_target); |
| 697 | reginfo->info_aux = NULL; |
| 698 | reginfo->strbeg = strbeg; |
| 699 | reginfo->strend = strend; |
| 700 | reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx)); |
| 701 | reginfo->intuit = 1; |
| 702 | /* not actually used within intuit, but zero for safety anyway */ |
| 703 | reginfo->poscache_maxiter = 0; |
| 704 | |
| 705 | if (utf8_target) { |
| 706 | if (!prog->check_utf8 && prog->check_substr) |
| 707 | to_utf8_substr(prog); |
| 708 | check = prog->check_utf8; |
| 709 | } else { |
| 710 | if (!prog->check_substr && prog->check_utf8) { |
| 711 | if (! to_byte_substr(prog)) { |
| 712 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail); |
| 713 | } |
| 714 | } |
| 715 | check = prog->check_substr; |
| 716 | } |
| 717 | |
| 718 | /* dump the various substring data */ |
| 719 | DEBUG_OPTIMISE_MORE_r({ |
| 720 | int i; |
| 721 | for (i=0; i<=2; i++) { |
| 722 | SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr |
| 723 | : prog->substrs->data[i].substr); |
| 724 | if (!sv) |
| 725 | continue; |
| 726 | |
| 727 | Perl_re_printf( aTHX_ |
| 728 | " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf |
| 729 | " useful=%"IVdf" utf8=%d [%s]\n", |
| 730 | i, |
| 731 | (IV)prog->substrs->data[i].min_offset, |
| 732 | (IV)prog->substrs->data[i].max_offset, |
| 733 | (IV)prog->substrs->data[i].end_shift, |
| 734 | BmUSEFUL(sv), |
| 735 | utf8_target ? 1 : 0, |
| 736 | SvPEEK(sv)); |
| 737 | } |
| 738 | }); |
| 739 | |
| 740 | if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */ |
| 741 | |
| 742 | /* ml_anch: check after \n? |
| 743 | * |
| 744 | * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning |
| 745 | * with /.*.../, these flags will have been added by the |
| 746 | * compiler: |
| 747 | * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL |
| 748 | * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL |
| 749 | */ |
| 750 | ml_anch = (prog->intflags & PREGf_ANCH_MBOL) |
| 751 | && !(prog->intflags & PREGf_IMPLICIT); |
| 752 | |
| 753 | if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) { |
| 754 | /* we are only allowed to match at BOS or \G */ |
| 755 | |
| 756 | /* trivially reject if there's a BOS anchor and we're not at BOS. |
| 757 | * |
| 758 | * Note that we don't try to do a similar quick reject for |
| 759 | * \G, since generally the caller will have calculated strpos |
| 760 | * based on pos() and gofs, so the string is already correctly |
| 761 | * anchored by definition; and handling the exceptions would |
| 762 | * be too fiddly (e.g. REXEC_IGNOREPOS). |
| 763 | */ |
| 764 | if ( strpos != strbeg |
| 765 | && (prog->intflags & PREGf_ANCH_SBOL)) |
| 766 | { |
| 767 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 768 | " Not at start...\n")); |
| 769 | goto fail; |
| 770 | } |
| 771 | |
| 772 | /* in the presence of an anchor, the anchored (relative to the |
| 773 | * start of the regex) substr must also be anchored relative |
| 774 | * to strpos. So quickly reject if substr isn't found there. |
| 775 | * This works for \G too, because the caller will already have |
| 776 | * subtracted gofs from pos, and gofs is the offset from the |
| 777 | * \G to the start of the regex. For example, in /.abc\Gdef/, |
| 778 | * where substr="abcdef", pos()=3, gofs=4, offset_min=1: |
| 779 | * caller will have set strpos=pos()-4; we look for the substr |
| 780 | * at position pos()-4+1, which lines up with the "a" */ |
| 781 | |
| 782 | if (prog->check_offset_min == prog->check_offset_max) { |
| 783 | /* Substring at constant offset from beg-of-str... */ |
| 784 | SSize_t slen = SvCUR(check); |
| 785 | char *s = HOP3c(strpos, prog->check_offset_min, strend); |
| 786 | |
| 787 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 788 | " Looking for check substr at fixed offset %"IVdf"...\n", |
| 789 | (IV)prog->check_offset_min)); |
| 790 | |
| 791 | if (SvTAIL(check)) { |
| 792 | /* In this case, the regex is anchored at the end too. |
| 793 | * Unless it's a multiline match, the lengths must match |
| 794 | * exactly, give or take a \n. NB: slen >= 1 since |
| 795 | * the last char of check is \n */ |
| 796 | if (!multiline |
| 797 | && ( strend - s > slen |
| 798 | || strend - s < slen - 1 |
| 799 | || (strend - s == slen && strend[-1] != '\n'))) |
| 800 | { |
| 801 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 802 | " String too long...\n")); |
| 803 | goto fail_finish; |
| 804 | } |
| 805 | /* Now should match s[0..slen-2] */ |
| 806 | slen--; |
| 807 | } |
| 808 | if (slen && (*SvPVX_const(check) != *s |
| 809 | || (slen > 1 && memNE(SvPVX_const(check), s, slen)))) |
| 810 | { |
| 811 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 812 | " String not equal...\n")); |
| 813 | goto fail_finish; |
| 814 | } |
| 815 | |
| 816 | check_at = s; |
| 817 | goto success_at_start; |
| 818 | } |
| 819 | } |
| 820 | } |
| 821 | |
| 822 | end_shift = prog->check_end_shift; |
| 823 | |
| 824 | #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */ |
| 825 | if (end_shift < 0) |
| 826 | Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ", |
| 827 | (IV)end_shift, RX_PRECOMP(prog)); |
| 828 | #endif |
| 829 | |
| 830 | restart: |
| 831 | |
| 832 | /* This is the (re)entry point of the main loop in this function. |
| 833 | * The goal of this loop is to: |
| 834 | * 1) find the "check" substring in the region rx_origin..strend |
| 835 | * (adjusted by start_shift / end_shift). If not found, reject |
| 836 | * immediately. |
| 837 | * 2) If it exists, look for the "other" substr too if defined; for |
| 838 | * example, if the check substr maps to the anchored substr, then |
| 839 | * check the floating substr, and vice-versa. If not found, go |
| 840 | * back to (1) with rx_origin suitably incremented. |
| 841 | * 3) If we find an rx_origin position that doesn't contradict |
| 842 | * either of the substrings, then check the possible additional |
| 843 | * constraints on rx_origin of /^.../m or a known start class. |
| 844 | * If these fail, then depending on which constraints fail, jump |
| 845 | * back to here, or to various other re-entry points further along |
| 846 | * that skip some of the first steps. |
| 847 | * 4) If we pass all those tests, update the BmUSEFUL() count on the |
| 848 | * substring. If the start position was determined to be at the |
| 849 | * beginning of the string - so, not rejected, but not optimised, |
| 850 | * since we have to run regmatch from position 0 - decrement the |
| 851 | * BmUSEFUL() count. Otherwise increment it. |
| 852 | */ |
| 853 | |
| 854 | |
| 855 | /* first, look for the 'check' substring */ |
| 856 | |
| 857 | { |
| 858 | U8* start_point; |
| 859 | U8* end_point; |
| 860 | |
| 861 | DEBUG_OPTIMISE_MORE_r({ |
| 862 | Perl_re_printf( aTHX_ |
| 863 | " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf |
| 864 | " Start shift: %"IVdf" End shift %"IVdf |
| 865 | " Real end Shift: %"IVdf"\n", |
| 866 | (IV)(rx_origin - strbeg), |
| 867 | (IV)prog->check_offset_min, |
| 868 | (IV)start_shift, |
| 869 | (IV)end_shift, |
| 870 | (IV)prog->check_end_shift); |
| 871 | }); |
| 872 | |
| 873 | end_point = HOP3(strend, -end_shift, strbeg); |
| 874 | start_point = HOPMAYBE3(rx_origin, start_shift, end_point); |
| 875 | if (!start_point) |
| 876 | goto fail_finish; |
| 877 | |
| 878 | |
| 879 | /* If the regex is absolutely anchored to either the start of the |
| 880 | * string (SBOL) or to pos() (ANCH_GPOS), then |
| 881 | * check_offset_max represents an upper bound on the string where |
| 882 | * the substr could start. For the ANCH_GPOS case, we assume that |
| 883 | * the caller of intuit will have already set strpos to |
| 884 | * pos()-gofs, so in this case strpos + offset_max will still be |
| 885 | * an upper bound on the substr. |
| 886 | */ |
| 887 | if (!ml_anch |
| 888 | && prog->intflags & PREGf_ANCH |
| 889 | && prog->check_offset_max != SSize_t_MAX) |
| 890 | { |
| 891 | SSize_t len = SvCUR(check) - !!SvTAIL(check); |
| 892 | const char * const anchor = |
| 893 | (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg); |
| 894 | |
| 895 | /* do a bytes rather than chars comparison. It's conservative; |
| 896 | * so it skips doing the HOP if the result can't possibly end |
| 897 | * up earlier than the old value of end_point. |
| 898 | */ |
| 899 | if ((char*)end_point - anchor > prog->check_offset_max) { |
| 900 | end_point = HOP3lim((U8*)anchor, |
| 901 | prog->check_offset_max, |
| 902 | end_point -len) |
| 903 | + len; |
| 904 | } |
| 905 | } |
| 906 | |
| 907 | check_at = fbm_instr( start_point, end_point, |
| 908 | check, multiline ? FBMrf_MULTILINE : 0); |
| 909 | |
| 910 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 911 | " doing 'check' fbm scan, [%"IVdf"..%"IVdf"] gave %"IVdf"\n", |
| 912 | (IV)((char*)start_point - strbeg), |
| 913 | (IV)((char*)end_point - strbeg), |
| 914 | (IV)(check_at ? check_at - strbeg : -1) |
| 915 | )); |
| 916 | |
| 917 | /* Update the count-of-usability, remove useless subpatterns, |
| 918 | unshift s. */ |
| 919 | |
| 920 | DEBUG_EXECUTE_r({ |
| 921 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 922 | SvPVX_const(check), RE_SV_DUMPLEN(check), 30); |
| 923 | Perl_re_printf( aTHX_ " %s %s substr %s%s%s", |
| 924 | (check_at ? "Found" : "Did not find"), |
| 925 | (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) |
| 926 | ? "anchored" : "floating"), |
| 927 | quoted, |
| 928 | RE_SV_TAIL(check), |
| 929 | (check_at ? " at offset " : "...\n") ); |
| 930 | }); |
| 931 | |
| 932 | if (!check_at) |
| 933 | goto fail_finish; |
| 934 | /* set rx_origin to the minimum position where the regex could start |
| 935 | * matching, given the constraint of the just-matched check substring. |
| 936 | * But don't set it lower than previously. |
| 937 | */ |
| 938 | |
| 939 | if (check_at - rx_origin > prog->check_offset_max) |
| 940 | rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin); |
| 941 | /* Finish the diagnostic message */ |
| 942 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 943 | "%ld (rx_origin now %"IVdf")...\n", |
| 944 | (long)(check_at - strbeg), |
| 945 | (IV)(rx_origin - strbeg) |
| 946 | )); |
| 947 | } |
| 948 | |
| 949 | |
| 950 | /* now look for the 'other' substring if defined */ |
| 951 | |
| 952 | if (utf8_target ? prog->substrs->data[other_ix].utf8_substr |
| 953 | : prog->substrs->data[other_ix].substr) |
| 954 | { |
| 955 | /* Take into account the "other" substring. */ |
| 956 | char *last, *last1; |
| 957 | char *s; |
| 958 | SV* must; |
| 959 | struct reg_substr_datum *other; |
| 960 | |
| 961 | do_other_substr: |
| 962 | other = &prog->substrs->data[other_ix]; |
| 963 | |
| 964 | /* if "other" is anchored: |
| 965 | * we've previously found a floating substr starting at check_at. |
| 966 | * This means that the regex origin must lie somewhere |
| 967 | * between min (rx_origin): HOP3(check_at, -check_offset_max) |
| 968 | * and max: HOP3(check_at, -check_offset_min) |
| 969 | * (except that min will be >= strpos) |
| 970 | * So the fixed substr must lie somewhere between |
| 971 | * HOP3(min, anchored_offset) |
| 972 | * HOP3(max, anchored_offset) + SvCUR(substr) |
| 973 | */ |
| 974 | |
| 975 | /* if "other" is floating |
| 976 | * Calculate last1, the absolute latest point where the |
| 977 | * floating substr could start in the string, ignoring any |
| 978 | * constraints from the earlier fixed match. It is calculated |
| 979 | * as follows: |
| 980 | * |
| 981 | * strend - prog->minlen (in chars) is the absolute latest |
| 982 | * position within the string where the origin of the regex |
| 983 | * could appear. The latest start point for the floating |
| 984 | * substr is float_min_offset(*) on from the start of the |
| 985 | * regex. last1 simply combines thee two offsets. |
| 986 | * |
| 987 | * (*) You might think the latest start point should be |
| 988 | * float_max_offset from the regex origin, and technically |
| 989 | * you'd be correct. However, consider |
| 990 | * /a\d{2,4}bcd\w/ |
| 991 | * Here, float min, max are 3,5 and minlen is 7. |
| 992 | * This can match either |
| 993 | * /a\d\dbcd\w/ |
| 994 | * /a\d\d\dbcd\w/ |
| 995 | * /a\d\d\d\dbcd\w/ |
| 996 | * In the first case, the regex matches minlen chars; in the |
| 997 | * second, minlen+1, in the third, minlen+2. |
| 998 | * In the first case, the floating offset is 3 (which equals |
| 999 | * float_min), in the second, 4, and in the third, 5 (which |
| 1000 | * equals float_max). In all cases, the floating string bcd |
| 1001 | * can never start more than 4 chars from the end of the |
| 1002 | * string, which equals minlen - float_min. As the substring |
| 1003 | * starts to match more than float_min from the start of the |
| 1004 | * regex, it makes the regex match more than minlen chars, |
| 1005 | * and the two cancel each other out. So we can always use |
| 1006 | * float_min - minlen, rather than float_max - minlen for the |
| 1007 | * latest position in the string. |
| 1008 | * |
| 1009 | * Note that -minlen + float_min_offset is equivalent (AFAIKT) |
| 1010 | * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift |
| 1011 | */ |
| 1012 | |
| 1013 | assert(prog->minlen >= other->min_offset); |
| 1014 | last1 = HOP3c(strend, |
| 1015 | other->min_offset - prog->minlen, strbeg); |
| 1016 | |
| 1017 | if (other_ix) {/* i.e. if (other-is-float) */ |
| 1018 | /* last is the latest point where the floating substr could |
| 1019 | * start, *given* any constraints from the earlier fixed |
| 1020 | * match. This constraint is that the floating string starts |
| 1021 | * <= float_max_offset chars from the regex origin (rx_origin). |
| 1022 | * If this value is less than last1, use it instead. |
| 1023 | */ |
| 1024 | assert(rx_origin <= last1); |
| 1025 | last = |
| 1026 | /* this condition handles the offset==infinity case, and |
| 1027 | * is a short-cut otherwise. Although it's comparing a |
| 1028 | * byte offset to a char length, it does so in a safe way, |
| 1029 | * since 1 char always occupies 1 or more bytes, |
| 1030 | * so if a string range is (last1 - rx_origin) bytes, |
| 1031 | * it will be less than or equal to (last1 - rx_origin) |
| 1032 | * chars; meaning it errs towards doing the accurate HOP3 |
| 1033 | * rather than just using last1 as a short-cut */ |
| 1034 | (last1 - rx_origin) < other->max_offset |
| 1035 | ? last1 |
| 1036 | : (char*)HOP3lim(rx_origin, other->max_offset, last1); |
| 1037 | } |
| 1038 | else { |
| 1039 | assert(strpos + start_shift <= check_at); |
| 1040 | last = HOP4c(check_at, other->min_offset - start_shift, |
| 1041 | strbeg, strend); |
| 1042 | } |
| 1043 | |
| 1044 | s = HOP3c(rx_origin, other->min_offset, strend); |
| 1045 | if (s < other_last) /* These positions already checked */ |
| 1046 | s = other_last; |
| 1047 | |
| 1048 | must = utf8_target ? other->utf8_substr : other->substr; |
| 1049 | assert(SvPOK(must)); |
| 1050 | { |
| 1051 | char *from = s; |
| 1052 | char *to = last + SvCUR(must) - (SvTAIL(must)!=0); |
| 1053 | |
| 1054 | if (from > to) { |
| 1055 | s = NULL; |
| 1056 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1057 | " skipping 'other' fbm scan: %"IVdf" > %"IVdf"\n", |
| 1058 | (IV)(from - strbeg), |
| 1059 | (IV)(to - strbeg) |
| 1060 | )); |
| 1061 | } |
| 1062 | else { |
| 1063 | s = fbm_instr( |
| 1064 | (unsigned char*)from, |
| 1065 | (unsigned char*)to, |
| 1066 | must, |
| 1067 | multiline ? FBMrf_MULTILINE : 0 |
| 1068 | ); |
| 1069 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1070 | " doing 'other' fbm scan, [%"IVdf"..%"IVdf"] gave %"IVdf"\n", |
| 1071 | (IV)(from - strbeg), |
| 1072 | (IV)(to - strbeg), |
| 1073 | (IV)(s ? s - strbeg : -1) |
| 1074 | )); |
| 1075 | } |
| 1076 | } |
| 1077 | |
| 1078 | DEBUG_EXECUTE_r({ |
| 1079 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 1080 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 1081 | Perl_re_printf( aTHX_ " %s %s substr %s%s", |
| 1082 | s ? "Found" : "Contradicts", |
| 1083 | other_ix ? "floating" : "anchored", |
| 1084 | quoted, RE_SV_TAIL(must)); |
| 1085 | }); |
| 1086 | |
| 1087 | |
| 1088 | if (!s) { |
| 1089 | /* last1 is latest possible substr location. If we didn't |
| 1090 | * find it before there, we never will */ |
| 1091 | if (last >= last1) { |
| 1092 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1093 | "; giving up...\n")); |
| 1094 | goto fail_finish; |
| 1095 | } |
| 1096 | |
| 1097 | /* try to find the check substr again at a later |
| 1098 | * position. Maybe next time we'll find the "other" substr |
| 1099 | * in range too */ |
| 1100 | other_last = HOP3c(last, 1, strend) /* highest failure */; |
| 1101 | rx_origin = |
| 1102 | other_ix /* i.e. if other-is-float */ |
| 1103 | ? HOP3c(rx_origin, 1, strend) |
| 1104 | : HOP4c(last, 1 - other->min_offset, strbeg, strend); |
| 1105 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1106 | "; about to retry %s at offset %ld (rx_origin now %"IVdf")...\n", |
| 1107 | (other_ix ? "floating" : "anchored"), |
| 1108 | (long)(HOP3c(check_at, 1, strend) - strbeg), |
| 1109 | (IV)(rx_origin - strbeg) |
| 1110 | )); |
| 1111 | goto restart; |
| 1112 | } |
| 1113 | else { |
| 1114 | if (other_ix) { /* if (other-is-float) */ |
| 1115 | /* other_last is set to s, not s+1, since its possible for |
| 1116 | * a floating substr to fail first time, then succeed |
| 1117 | * second time at the same floating position; e.g.: |
| 1118 | * "-AB--AABZ" =~ /\wAB\d*Z/ |
| 1119 | * The first time round, anchored and float match at |
| 1120 | * "-(AB)--AAB(Z)" then fail on the initial \w character |
| 1121 | * class. Second time round, they match at "-AB--A(AB)(Z)". |
| 1122 | */ |
| 1123 | other_last = s; |
| 1124 | } |
| 1125 | else { |
| 1126 | rx_origin = HOP3c(s, -other->min_offset, strbeg); |
| 1127 | other_last = HOP3c(s, 1, strend); |
| 1128 | } |
| 1129 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1130 | " at offset %ld (rx_origin now %"IVdf")...\n", |
| 1131 | (long)(s - strbeg), |
| 1132 | (IV)(rx_origin - strbeg) |
| 1133 | )); |
| 1134 | |
| 1135 | } |
| 1136 | } |
| 1137 | else { |
| 1138 | DEBUG_OPTIMISE_MORE_r( |
| 1139 | Perl_re_printf( aTHX_ |
| 1140 | " Check-only match: offset min:%"IVdf" max:%"IVdf |
| 1141 | " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf |
| 1142 | " strend:%"IVdf"\n", |
| 1143 | (IV)prog->check_offset_min, |
| 1144 | (IV)prog->check_offset_max, |
| 1145 | (IV)(check_at-strbeg), |
| 1146 | (IV)(rx_origin-strbeg), |
| 1147 | (IV)(rx_origin-check_at), |
| 1148 | (IV)(strend-strbeg) |
| 1149 | ) |
| 1150 | ); |
| 1151 | } |
| 1152 | |
| 1153 | postprocess_substr_matches: |
| 1154 | |
| 1155 | /* handle the extra constraint of /^.../m if present */ |
| 1156 | |
| 1157 | if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') { |
| 1158 | char *s; |
| 1159 | |
| 1160 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1161 | " looking for /^/m anchor")); |
| 1162 | |
| 1163 | /* we have failed the constraint of a \n before rx_origin. |
| 1164 | * Find the next \n, if any, even if it's beyond the current |
| 1165 | * anchored and/or floating substrings. Whether we should be |
| 1166 | * scanning ahead for the next \n or the next substr is debatable. |
| 1167 | * On the one hand you'd expect rare substrings to appear less |
| 1168 | * often than \n's. On the other hand, searching for \n means |
| 1169 | * we're effectively flipping between check_substr and "\n" on each |
| 1170 | * iteration as the current "rarest" string candidate, which |
| 1171 | * means for example that we'll quickly reject the whole string if |
| 1172 | * hasn't got a \n, rather than trying every substr position |
| 1173 | * first |
| 1174 | */ |
| 1175 | |
| 1176 | s = HOP3c(strend, - prog->minlen, strpos); |
| 1177 | if (s <= rx_origin || |
| 1178 | ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin))) |
| 1179 | { |
| 1180 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1181 | " Did not find /%s^%s/m...\n", |
| 1182 | PL_colors[0], PL_colors[1])); |
| 1183 | goto fail_finish; |
| 1184 | } |
| 1185 | |
| 1186 | /* earliest possible origin is 1 char after the \n. |
| 1187 | * (since *rx_origin == '\n', it's safe to ++ here rather than |
| 1188 | * HOP(rx_origin, 1)) */ |
| 1189 | rx_origin++; |
| 1190 | |
| 1191 | if (prog->substrs->check_ix == 0 /* check is anchored */ |
| 1192 | || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos)) |
| 1193 | { |
| 1194 | /* Position contradicts check-string; either because |
| 1195 | * check was anchored (and thus has no wiggle room), |
| 1196 | * or check was float and rx_origin is above the float range */ |
| 1197 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1198 | " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n", |
| 1199 | PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg))); |
| 1200 | goto restart; |
| 1201 | } |
| 1202 | |
| 1203 | /* if we get here, the check substr must have been float, |
| 1204 | * is in range, and we may or may not have had an anchored |
| 1205 | * "other" substr which still contradicts */ |
| 1206 | assert(prog->substrs->check_ix); /* check is float */ |
| 1207 | |
| 1208 | if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) { |
| 1209 | /* whoops, the anchored "other" substr exists, so we still |
| 1210 | * contradict. On the other hand, the float "check" substr |
| 1211 | * didn't contradict, so just retry the anchored "other" |
| 1212 | * substr */ |
| 1213 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1214 | " Found /%s^%s/m, rescanning for anchored from offset %"IVdf" (rx_origin now %"IVdf")...\n", |
| 1215 | PL_colors[0], PL_colors[1], |
| 1216 | (IV)(rx_origin - strbeg + prog->anchored_offset), |
| 1217 | (IV)(rx_origin - strbeg) |
| 1218 | )); |
| 1219 | goto do_other_substr; |
| 1220 | } |
| 1221 | |
| 1222 | /* success: we don't contradict the found floating substring |
| 1223 | * (and there's no anchored substr). */ |
| 1224 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1225 | " Found /%s^%s/m with rx_origin %ld...\n", |
| 1226 | PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg))); |
| 1227 | } |
| 1228 | else { |
| 1229 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1230 | " (multiline anchor test skipped)\n")); |
| 1231 | } |
| 1232 | |
| 1233 | success_at_start: |
| 1234 | |
| 1235 | |
| 1236 | /* if we have a starting character class, then test that extra constraint. |
| 1237 | * (trie stclasses are too expensive to use here, we are better off to |
| 1238 | * leave it to regmatch itself) */ |
| 1239 | |
| 1240 | if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) { |
| 1241 | const U8* const str = (U8*)STRING(progi->regstclass); |
| 1242 | |
| 1243 | /* XXX this value could be pre-computed */ |
| 1244 | const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT |
| 1245 | ? (reginfo->is_utf8_pat |
| 1246 | ? utf8_distance(str + STR_LEN(progi->regstclass), str) |
| 1247 | : STR_LEN(progi->regstclass)) |
| 1248 | : 1); |
| 1249 | char * endpos; |
| 1250 | char *s; |
| 1251 | /* latest pos that a matching float substr constrains rx start to */ |
| 1252 | char *rx_max_float = NULL; |
| 1253 | |
| 1254 | /* if the current rx_origin is anchored, either by satisfying an |
| 1255 | * anchored substring constraint, or a /^.../m constraint, then we |
| 1256 | * can reject the current origin if the start class isn't found |
| 1257 | * at the current position. If we have a float-only match, then |
| 1258 | * rx_origin is constrained to a range; so look for the start class |
| 1259 | * in that range. if neither, then look for the start class in the |
| 1260 | * whole rest of the string */ |
| 1261 | |
| 1262 | /* XXX DAPM it's not clear what the minlen test is for, and why |
| 1263 | * it's not used in the floating case. Nothing in the test suite |
| 1264 | * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>. |
| 1265 | * Here are some old comments, which may or may not be correct: |
| 1266 | * |
| 1267 | * minlen == 0 is possible if regstclass is \b or \B, |
| 1268 | * and the fixed substr is ''$. |
| 1269 | * Since minlen is already taken into account, rx_origin+1 is |
| 1270 | * before strend; accidentally, minlen >= 1 guaranties no false |
| 1271 | * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 : |
| 1272 | * 0) below assumes that regstclass does not come from lookahead... |
| 1273 | * If regstclass takes bytelength more than 1: If charlength==1, OK. |
| 1274 | * This leaves EXACTF-ish only, which are dealt with in |
| 1275 | * find_byclass(). |
| 1276 | */ |
| 1277 | |
| 1278 | if (prog->anchored_substr || prog->anchored_utf8 || ml_anch) |
| 1279 | endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend); |
| 1280 | else if (prog->float_substr || prog->float_utf8) { |
| 1281 | rx_max_float = HOP3c(check_at, -start_shift, strbeg); |
| 1282 | endpos= HOP3c(rx_max_float, cl_l, strend); |
| 1283 | } |
| 1284 | else |
| 1285 | endpos= strend; |
| 1286 | |
| 1287 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1288 | " looking for class: start_shift: %"IVdf" check_at: %"IVdf |
| 1289 | " rx_origin: %"IVdf" endpos: %"IVdf"\n", |
| 1290 | (IV)start_shift, (IV)(check_at - strbeg), |
| 1291 | (IV)(rx_origin - strbeg), (IV)(endpos - strbeg))); |
| 1292 | |
| 1293 | s = find_byclass(prog, progi->regstclass, rx_origin, endpos, |
| 1294 | reginfo); |
| 1295 | if (!s) { |
| 1296 | if (endpos == strend) { |
| 1297 | DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ |
| 1298 | " Could not match STCLASS...\n") ); |
| 1299 | goto fail; |
| 1300 | } |
| 1301 | DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ |
| 1302 | " This position contradicts STCLASS...\n") ); |
| 1303 | if ((prog->intflags & PREGf_ANCH) && !ml_anch |
| 1304 | && !(prog->intflags & PREGf_IMPLICIT)) |
| 1305 | goto fail; |
| 1306 | |
| 1307 | /* Contradict one of substrings */ |
| 1308 | if (prog->anchored_substr || prog->anchored_utf8) { |
| 1309 | if (prog->substrs->check_ix == 1) { /* check is float */ |
| 1310 | /* Have both, check_string is floating */ |
| 1311 | assert(rx_origin + start_shift <= check_at); |
| 1312 | if (rx_origin + start_shift != check_at) { |
| 1313 | /* not at latest position float substr could match: |
| 1314 | * Recheck anchored substring, but not floating. |
| 1315 | * The condition above is in bytes rather than |
| 1316 | * chars for efficiency. It's conservative, in |
| 1317 | * that it errs on the side of doing 'goto |
| 1318 | * do_other_substr'. In this case, at worst, |
| 1319 | * an extra anchored search may get done, but in |
| 1320 | * practice the extra fbm_instr() is likely to |
| 1321 | * get skipped anyway. */ |
| 1322 | DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ |
| 1323 | " about to retry anchored at offset %ld (rx_origin now %"IVdf")...\n", |
| 1324 | (long)(other_last - strbeg), |
| 1325 | (IV)(rx_origin - strbeg) |
| 1326 | )); |
| 1327 | goto do_other_substr; |
| 1328 | } |
| 1329 | } |
| 1330 | } |
| 1331 | else { |
| 1332 | /* float-only */ |
| 1333 | |
| 1334 | if (ml_anch) { |
| 1335 | /* In the presence of ml_anch, we might be able to |
| 1336 | * find another \n without breaking the current float |
| 1337 | * constraint. */ |
| 1338 | |
| 1339 | /* strictly speaking this should be HOP3c(..., 1, ...), |
| 1340 | * but since we goto a block of code that's going to |
| 1341 | * search for the next \n if any, its safe here */ |
| 1342 | rx_origin++; |
| 1343 | DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ |
| 1344 | " about to look for /%s^%s/m starting at rx_origin %ld...\n", |
| 1345 | PL_colors[0], PL_colors[1], |
| 1346 | (long)(rx_origin - strbeg)) ); |
| 1347 | goto postprocess_substr_matches; |
| 1348 | } |
| 1349 | |
| 1350 | /* strictly speaking this can never be true; but might |
| 1351 | * be if we ever allow intuit without substrings */ |
| 1352 | if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) |
| 1353 | goto fail; |
| 1354 | |
| 1355 | rx_origin = rx_max_float; |
| 1356 | } |
| 1357 | |
| 1358 | /* at this point, any matching substrings have been |
| 1359 | * contradicted. Start again... */ |
| 1360 | |
| 1361 | rx_origin = HOP3c(rx_origin, 1, strend); |
| 1362 | |
| 1363 | /* uses bytes rather than char calculations for efficiency. |
| 1364 | * It's conservative: it errs on the side of doing 'goto restart', |
| 1365 | * where there is code that does a proper char-based test */ |
| 1366 | if (rx_origin + start_shift + end_shift > strend) { |
| 1367 | DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ |
| 1368 | " Could not match STCLASS...\n") ); |
| 1369 | goto fail; |
| 1370 | } |
| 1371 | DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ |
| 1372 | " about to look for %s substr starting at offset %ld (rx_origin now %"IVdf")...\n", |
| 1373 | (prog->substrs->check_ix ? "floating" : "anchored"), |
| 1374 | (long)(rx_origin + start_shift - strbeg), |
| 1375 | (IV)(rx_origin - strbeg) |
| 1376 | )); |
| 1377 | goto restart; |
| 1378 | } |
| 1379 | |
| 1380 | /* Success !!! */ |
| 1381 | |
| 1382 | if (rx_origin != s) { |
| 1383 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1384 | " By STCLASS: moving %ld --> %ld\n", |
| 1385 | (long)(rx_origin - strbeg), (long)(s - strbeg)) |
| 1386 | ); |
| 1387 | } |
| 1388 | else { |
| 1389 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1390 | " Does not contradict STCLASS...\n"); |
| 1391 | ); |
| 1392 | } |
| 1393 | } |
| 1394 | |
| 1395 | /* Decide whether using the substrings helped */ |
| 1396 | |
| 1397 | if (rx_origin != strpos) { |
| 1398 | /* Fixed substring is found far enough so that the match |
| 1399 | cannot start at strpos. */ |
| 1400 | |
| 1401 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " try at offset...\n")); |
| 1402 | ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */ |
| 1403 | } |
| 1404 | else { |
| 1405 | /* The found rx_origin position does not prohibit matching at |
| 1406 | * strpos, so calling intuit didn't gain us anything. Decrement |
| 1407 | * the BmUSEFUL() count on the check substring, and if we reach |
| 1408 | * zero, free it. */ |
| 1409 | if (!(prog->intflags & PREGf_NAUGHTY) |
| 1410 | && (utf8_target ? ( |
| 1411 | prog->check_utf8 /* Could be deleted already */ |
| 1412 | && --BmUSEFUL(prog->check_utf8) < 0 |
| 1413 | && (prog->check_utf8 == prog->float_utf8) |
| 1414 | ) : ( |
| 1415 | prog->check_substr /* Could be deleted already */ |
| 1416 | && --BmUSEFUL(prog->check_substr) < 0 |
| 1417 | && (prog->check_substr == prog->float_substr) |
| 1418 | ))) |
| 1419 | { |
| 1420 | /* If flags & SOMETHING - do not do it many times on the same match */ |
| 1421 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " ... Disabling check substring...\n")); |
| 1422 | /* XXX Does the destruction order has to change with utf8_target? */ |
| 1423 | SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr); |
| 1424 | SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8); |
| 1425 | prog->check_substr = prog->check_utf8 = NULL; /* disable */ |
| 1426 | prog->float_substr = prog->float_utf8 = NULL; /* clear */ |
| 1427 | check = NULL; /* abort */ |
| 1428 | /* XXXX This is a remnant of the old implementation. It |
| 1429 | looks wasteful, since now INTUIT can use many |
| 1430 | other heuristics. */ |
| 1431 | prog->extflags &= ~RXf_USE_INTUIT; |
| 1432 | } |
| 1433 | } |
| 1434 | |
| 1435 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 1436 | "Intuit: %sSuccessfully guessed:%s match at offset %ld\n", |
| 1437 | PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) ); |
| 1438 | |
| 1439 | return rx_origin; |
| 1440 | |
| 1441 | fail_finish: /* Substring not found */ |
| 1442 | if (prog->check_substr || prog->check_utf8) /* could be removed already */ |
| 1443 | BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */ |
| 1444 | fail: |
| 1445 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch rejected by optimizer%s\n", |
| 1446 | PL_colors[4], PL_colors[5])); |
| 1447 | return NULL; |
| 1448 | } |
| 1449 | |
| 1450 | |
| 1451 | #define DECL_TRIE_TYPE(scan) \ |
| 1452 | const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \ |
| 1453 | trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \ |
| 1454 | trie_utf8l, trie_flu8 } \ |
| 1455 | trie_type = ((scan->flags == EXACT) \ |
| 1456 | ? (utf8_target ? trie_utf8 : trie_plain) \ |
| 1457 | : (scan->flags == EXACTL) \ |
| 1458 | ? (utf8_target ? trie_utf8l : trie_plain) \ |
| 1459 | : (scan->flags == EXACTFA) \ |
| 1460 | ? (utf8_target \ |
| 1461 | ? trie_utf8_exactfa_fold \ |
| 1462 | : trie_latin_utf8_exactfa_fold) \ |
| 1463 | : (scan->flags == EXACTFLU8 \ |
| 1464 | ? trie_flu8 \ |
| 1465 | : (utf8_target \ |
| 1466 | ? trie_utf8_fold \ |
| 1467 | : trie_latin_utf8_fold))) |
| 1468 | |
| 1469 | #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \ |
| 1470 | STMT_START { \ |
| 1471 | STRLEN skiplen; \ |
| 1472 | U8 flags = FOLD_FLAGS_FULL; \ |
| 1473 | switch (trie_type) { \ |
| 1474 | case trie_flu8: \ |
| 1475 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ |
| 1476 | if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \ |
| 1477 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \ |
| 1478 | } \ |
| 1479 | goto do_trie_utf8_fold; \ |
| 1480 | case trie_utf8_exactfa_fold: \ |
| 1481 | flags |= FOLD_FLAGS_NOMIX_ASCII; \ |
| 1482 | /* FALLTHROUGH */ \ |
| 1483 | case trie_utf8_fold: \ |
| 1484 | do_trie_utf8_fold: \ |
| 1485 | if ( foldlen>0 ) { \ |
| 1486 | uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ |
| 1487 | foldlen -= len; \ |
| 1488 | uscan += len; \ |
| 1489 | len=0; \ |
| 1490 | } else { \ |
| 1491 | uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \ |
| 1492 | len = UTF8SKIP(uc); \ |
| 1493 | skiplen = UVCHR_SKIP( uvc ); \ |
| 1494 | foldlen -= skiplen; \ |
| 1495 | uscan = foldbuf + skiplen; \ |
| 1496 | } \ |
| 1497 | break; \ |
| 1498 | case trie_latin_utf8_exactfa_fold: \ |
| 1499 | flags |= FOLD_FLAGS_NOMIX_ASCII; \ |
| 1500 | /* FALLTHROUGH */ \ |
| 1501 | case trie_latin_utf8_fold: \ |
| 1502 | if ( foldlen>0 ) { \ |
| 1503 | uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ |
| 1504 | foldlen -= len; \ |
| 1505 | uscan += len; \ |
| 1506 | len=0; \ |
| 1507 | } else { \ |
| 1508 | len = 1; \ |
| 1509 | uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \ |
| 1510 | skiplen = UVCHR_SKIP( uvc ); \ |
| 1511 | foldlen -= skiplen; \ |
| 1512 | uscan = foldbuf + skiplen; \ |
| 1513 | } \ |
| 1514 | break; \ |
| 1515 | case trie_utf8l: \ |
| 1516 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ |
| 1517 | if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \ |
| 1518 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \ |
| 1519 | } \ |
| 1520 | /* FALLTHROUGH */ \ |
| 1521 | case trie_utf8: \ |
| 1522 | uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \ |
| 1523 | break; \ |
| 1524 | case trie_plain: \ |
| 1525 | uvc = (UV)*uc; \ |
| 1526 | len = 1; \ |
| 1527 | } \ |
| 1528 | if (uvc < 256) { \ |
| 1529 | charid = trie->charmap[ uvc ]; \ |
| 1530 | } \ |
| 1531 | else { \ |
| 1532 | charid = 0; \ |
| 1533 | if (widecharmap) { \ |
| 1534 | SV** const svpp = hv_fetch(widecharmap, \ |
| 1535 | (char*)&uvc, sizeof(UV), 0); \ |
| 1536 | if (svpp) \ |
| 1537 | charid = (U16)SvIV(*svpp); \ |
| 1538 | } \ |
| 1539 | } \ |
| 1540 | } STMT_END |
| 1541 | |
| 1542 | #define DUMP_EXEC_POS(li,s,doutf8,depth) \ |
| 1543 | dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \ |
| 1544 | startpos, doutf8, depth) |
| 1545 | |
| 1546 | #define REXEC_FBC_EXACTISH_SCAN(COND) \ |
| 1547 | STMT_START { \ |
| 1548 | while (s <= e) { \ |
| 1549 | if ( (COND) \ |
| 1550 | && (ln == 1 || folder(s, pat_string, ln)) \ |
| 1551 | && (reginfo->intuit || regtry(reginfo, &s)) )\ |
| 1552 | goto got_it; \ |
| 1553 | s++; \ |
| 1554 | } \ |
| 1555 | } STMT_END |
| 1556 | |
| 1557 | #define REXEC_FBC_UTF8_SCAN(CODE) \ |
| 1558 | STMT_START { \ |
| 1559 | while (s < strend) { \ |
| 1560 | CODE \ |
| 1561 | s += UTF8SKIP(s); \ |
| 1562 | } \ |
| 1563 | } STMT_END |
| 1564 | |
| 1565 | #define REXEC_FBC_SCAN(CODE) \ |
| 1566 | STMT_START { \ |
| 1567 | while (s < strend) { \ |
| 1568 | CODE \ |
| 1569 | s++; \ |
| 1570 | } \ |
| 1571 | } STMT_END |
| 1572 | |
| 1573 | #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \ |
| 1574 | REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \ |
| 1575 | if (COND) { \ |
| 1576 | if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ |
| 1577 | goto got_it; \ |
| 1578 | else \ |
| 1579 | tmp = doevery; \ |
| 1580 | } \ |
| 1581 | else \ |
| 1582 | tmp = 1; \ |
| 1583 | ) |
| 1584 | |
| 1585 | #define REXEC_FBC_CLASS_SCAN(COND) \ |
| 1586 | REXEC_FBC_SCAN( /* Loops while (s < strend) */ \ |
| 1587 | if (COND) { \ |
| 1588 | if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ |
| 1589 | goto got_it; \ |
| 1590 | else \ |
| 1591 | tmp = doevery; \ |
| 1592 | } \ |
| 1593 | else \ |
| 1594 | tmp = 1; \ |
| 1595 | ) |
| 1596 | |
| 1597 | #define REXEC_FBC_CSCAN(CONDUTF8,COND) \ |
| 1598 | if (utf8_target) { \ |
| 1599 | REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \ |
| 1600 | } \ |
| 1601 | else { \ |
| 1602 | REXEC_FBC_CLASS_SCAN(COND); \ |
| 1603 | } |
| 1604 | |
| 1605 | /* The three macros below are slightly different versions of the same logic. |
| 1606 | * |
| 1607 | * The first is for /a and /aa when the target string is UTF-8. This can only |
| 1608 | * match ascii, but it must advance based on UTF-8. The other two handle the |
| 1609 | * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking |
| 1610 | * for the boundary (or non-boundary) between a word and non-word character. |
| 1611 | * The utf8 and non-utf8 cases have the same logic, but the details must be |
| 1612 | * different. Find the "wordness" of the character just prior to this one, and |
| 1613 | * compare it with the wordness of this one. If they differ, we have a |
| 1614 | * boundary. At the beginning of the string, pretend that the previous |
| 1615 | * character was a new-line. |
| 1616 | * |
| 1617 | * All these macros uncleanly have side-effects with each other and outside |
| 1618 | * variables. So far it's been too much trouble to clean-up |
| 1619 | * |
| 1620 | * TEST_NON_UTF8 is the macro or function to call to test if its byte input is |
| 1621 | * a word character or not. |
| 1622 | * IF_SUCCESS is code to do if it finds that we are at a boundary between |
| 1623 | * word/non-word |
| 1624 | * IF_FAIL is code to do if we aren't at a boundary between word/non-word |
| 1625 | * |
| 1626 | * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we |
| 1627 | * are looking for a boundary or for a non-boundary. If we are looking for a |
| 1628 | * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and |
| 1629 | * see if this tentative match actually works, and if so, to quit the loop |
| 1630 | * here. And vice-versa if we are looking for a non-boundary. |
| 1631 | * |
| 1632 | * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and |
| 1633 | * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of |
| 1634 | * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be |
| 1635 | * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal |
| 1636 | * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that |
| 1637 | * complement. But in that branch we complement tmp, meaning that at the |
| 1638 | * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s), |
| 1639 | * which means at the top of the loop in the next iteration, it is |
| 1640 | * TEST_NON_UTF8(s-1) */ |
| 1641 | #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
| 1642 | tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ |
| 1643 | tmp = TEST_NON_UTF8(tmp); \ |
| 1644 | REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \ |
| 1645 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ |
| 1646 | tmp = !tmp; \ |
| 1647 | IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \ |
| 1648 | } \ |
| 1649 | else { \ |
| 1650 | IF_FAIL; \ |
| 1651 | } \ |
| 1652 | ); \ |
| 1653 | |
| 1654 | /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and |
| 1655 | * TEST_UTF8 is a macro that for the same input code points returns identically |
| 1656 | * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */ |
| 1657 | #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \ |
| 1658 | if (s == reginfo->strbeg) { \ |
| 1659 | tmp = '\n'; \ |
| 1660 | } \ |
| 1661 | else { /* Back-up to the start of the previous character */ \ |
| 1662 | U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \ |
| 1663 | tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \ |
| 1664 | 0, UTF8_ALLOW_DEFAULT); \ |
| 1665 | } \ |
| 1666 | tmp = TEST_UV(tmp); \ |
| 1667 | LOAD_UTF8_CHARCLASS_ALNUM(); \ |
| 1668 | REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \ |
| 1669 | if (tmp == ! (TEST_UTF8((U8 *) s))) { \ |
| 1670 | tmp = !tmp; \ |
| 1671 | IF_SUCCESS; \ |
| 1672 | } \ |
| 1673 | else { \ |
| 1674 | IF_FAIL; \ |
| 1675 | } \ |
| 1676 | ); |
| 1677 | |
| 1678 | /* Like the above two macros. UTF8_CODE is the complete code for handling |
| 1679 | * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc |
| 1680 | * macros below */ |
| 1681 | #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
| 1682 | if (utf8_target) { \ |
| 1683 | UTF8_CODE \ |
| 1684 | } \ |
| 1685 | else { /* Not utf8 */ \ |
| 1686 | tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ |
| 1687 | tmp = TEST_NON_UTF8(tmp); \ |
| 1688 | REXEC_FBC_SCAN( /* advances s while s < strend */ \ |
| 1689 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ |
| 1690 | IF_SUCCESS; \ |
| 1691 | tmp = !tmp; \ |
| 1692 | } \ |
| 1693 | else { \ |
| 1694 | IF_FAIL; \ |
| 1695 | } \ |
| 1696 | ); \ |
| 1697 | } \ |
| 1698 | /* Here, things have been set up by the previous code so that tmp is the \ |
| 1699 | * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \ |
| 1700 | * utf8ness of the target). We also have to check if this matches against \ |
| 1701 | * the EOS, which we treat as a \n (which is the same value in both UTF-8 \ |
| 1702 | * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \ |
| 1703 | * string */ \ |
| 1704 | if (tmp == ! TEST_NON_UTF8('\n')) { \ |
| 1705 | IF_SUCCESS; \ |
| 1706 | } \ |
| 1707 | else { \ |
| 1708 | IF_FAIL; \ |
| 1709 | } |
| 1710 | |
| 1711 | /* This is the macro to use when we want to see if something that looks like it |
| 1712 | * could match, actually does, and if so exits the loop */ |
| 1713 | #define REXEC_FBC_TRYIT \ |
| 1714 | if ((reginfo->intuit || regtry(reginfo, &s))) \ |
| 1715 | goto got_it |
| 1716 | |
| 1717 | /* The only difference between the BOUND and NBOUND cases is that |
| 1718 | * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in |
| 1719 | * NBOUND. This is accomplished by passing it as either the if or else clause, |
| 1720 | * with the other one being empty (PLACEHOLDER is defined as empty). |
| 1721 | * |
| 1722 | * The TEST_FOO parameters are for operating on different forms of input, but |
| 1723 | * all should be ones that return identically for the same underlying code |
| 1724 | * points */ |
| 1725 | #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ |
| 1726 | FBC_BOUND_COMMON( \ |
| 1727 | FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \ |
| 1728 | TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) |
| 1729 | |
| 1730 | #define FBC_BOUND_A(TEST_NON_UTF8) \ |
| 1731 | FBC_BOUND_COMMON( \ |
| 1732 | FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \ |
| 1733 | TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) |
| 1734 | |
| 1735 | #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ |
| 1736 | FBC_BOUND_COMMON( \ |
| 1737 | FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \ |
| 1738 | TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) |
| 1739 | |
| 1740 | #define FBC_NBOUND_A(TEST_NON_UTF8) \ |
| 1741 | FBC_BOUND_COMMON( \ |
| 1742 | FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \ |
| 1743 | TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) |
| 1744 | |
| 1745 | #ifdef DEBUGGING |
| 1746 | static IV |
| 1747 | S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) { |
| 1748 | IV cp_out = Perl__invlist_search(invlist, cp_in); |
| 1749 | assert(cp_out >= 0); |
| 1750 | return cp_out; |
| 1751 | } |
| 1752 | # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \ |
| 1753 | invmap[S_get_break_val_cp_checked(invlist, cp)] |
| 1754 | #else |
| 1755 | # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \ |
| 1756 | invmap[_invlist_search(invlist, cp)] |
| 1757 | #endif |
| 1758 | |
| 1759 | /* Takes a pointer to an inversion list, a pointer to its corresponding |
| 1760 | * inversion map, and a code point, and returns the code point's value |
| 1761 | * according to the two arrays. It assumes that all code points have a value. |
| 1762 | * This is used as the base macro for macros for particular properties */ |
| 1763 | #define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \ |
| 1764 | _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) |
| 1765 | |
| 1766 | /* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead |
| 1767 | * of a code point, returning the value for the first code point in the string. |
| 1768 | * And it takes the particular macro name that finds the desired value given a |
| 1769 | * code point. Merely convert the UTF-8 to code point and call the cp macro */ |
| 1770 | #define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \ |
| 1771 | (__ASSERT_(pos < strend) \ |
| 1772 | /* Note assumes is valid UTF-8 */ \ |
| 1773 | (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL)))) |
| 1774 | |
| 1775 | /* Returns the GCB value for the input code point */ |
| 1776 | #define getGCB_VAL_CP(cp) \ |
| 1777 | _generic_GET_BREAK_VAL_CP( \ |
| 1778 | PL_GCB_invlist, \ |
| 1779 | _Perl_GCB_invmap, \ |
| 1780 | (cp)) |
| 1781 | |
| 1782 | /* Returns the GCB value for the first code point in the UTF-8 encoded string |
| 1783 | * bounded by pos and strend */ |
| 1784 | #define getGCB_VAL_UTF8(pos, strend) \ |
| 1785 | _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend) |
| 1786 | |
| 1787 | /* Returns the LB value for the input code point */ |
| 1788 | #define getLB_VAL_CP(cp) \ |
| 1789 | _generic_GET_BREAK_VAL_CP( \ |
| 1790 | PL_LB_invlist, \ |
| 1791 | _Perl_LB_invmap, \ |
| 1792 | (cp)) |
| 1793 | |
| 1794 | /* Returns the LB value for the first code point in the UTF-8 encoded string |
| 1795 | * bounded by pos and strend */ |
| 1796 | #define getLB_VAL_UTF8(pos, strend) \ |
| 1797 | _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend) |
| 1798 | |
| 1799 | |
| 1800 | /* Returns the SB value for the input code point */ |
| 1801 | #define getSB_VAL_CP(cp) \ |
| 1802 | _generic_GET_BREAK_VAL_CP( \ |
| 1803 | PL_SB_invlist, \ |
| 1804 | _Perl_SB_invmap, \ |
| 1805 | (cp)) |
| 1806 | |
| 1807 | /* Returns the SB value for the first code point in the UTF-8 encoded string |
| 1808 | * bounded by pos and strend */ |
| 1809 | #define getSB_VAL_UTF8(pos, strend) \ |
| 1810 | _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend) |
| 1811 | |
| 1812 | /* Returns the WB value for the input code point */ |
| 1813 | #define getWB_VAL_CP(cp) \ |
| 1814 | _generic_GET_BREAK_VAL_CP( \ |
| 1815 | PL_WB_invlist, \ |
| 1816 | _Perl_WB_invmap, \ |
| 1817 | (cp)) |
| 1818 | |
| 1819 | /* Returns the WB value for the first code point in the UTF-8 encoded string |
| 1820 | * bounded by pos and strend */ |
| 1821 | #define getWB_VAL_UTF8(pos, strend) \ |
| 1822 | _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend) |
| 1823 | |
| 1824 | /* We know what class REx starts with. Try to find this position... */ |
| 1825 | /* if reginfo->intuit, its a dryrun */ |
| 1826 | /* annoyingly all the vars in this routine have different names from their counterparts |
| 1827 | in regmatch. /grrr */ |
| 1828 | STATIC char * |
| 1829 | S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, |
| 1830 | const char *strend, regmatch_info *reginfo) |
| 1831 | { |
| 1832 | dVAR; |
| 1833 | const I32 doevery = (prog->intflags & PREGf_SKIP) == 0; |
| 1834 | char *pat_string; /* The pattern's exactish string */ |
| 1835 | char *pat_end; /* ptr to end char of pat_string */ |
| 1836 | re_fold_t folder; /* Function for computing non-utf8 folds */ |
| 1837 | const U8 *fold_array; /* array for folding ords < 256 */ |
| 1838 | STRLEN ln; |
| 1839 | STRLEN lnc; |
| 1840 | U8 c1; |
| 1841 | U8 c2; |
| 1842 | char *e; |
| 1843 | I32 tmp = 1; /* Scratch variable? */ |
| 1844 | const bool utf8_target = reginfo->is_utf8_target; |
| 1845 | UV utf8_fold_flags = 0; |
| 1846 | const bool is_utf8_pat = reginfo->is_utf8_pat; |
| 1847 | bool to_complement = FALSE; /* Invert the result? Taking the xor of this |
| 1848 | with a result inverts that result, as 0^1 = |
| 1849 | 1 and 1^1 = 0 */ |
| 1850 | _char_class_number classnum; |
| 1851 | |
| 1852 | RXi_GET_DECL(prog,progi); |
| 1853 | |
| 1854 | PERL_ARGS_ASSERT_FIND_BYCLASS; |
| 1855 | |
| 1856 | /* We know what class it must start with. */ |
| 1857 | switch (OP(c)) { |
| 1858 | case ANYOFL: |
| 1859 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 1860 | |
| 1861 | if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) { |
| 1862 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required); |
| 1863 | } |
| 1864 | |
| 1865 | /* FALLTHROUGH */ |
| 1866 | case ANYOFD: |
| 1867 | case ANYOF: |
| 1868 | if (utf8_target) { |
| 1869 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1870 | reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target)); |
| 1871 | } |
| 1872 | else { |
| 1873 | REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s, 0)); |
| 1874 | } |
| 1875 | break; |
| 1876 | |
| 1877 | case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ |
| 1878 | assert(! is_utf8_pat); |
| 1879 | /* FALLTHROUGH */ |
| 1880 | case EXACTFA: |
| 1881 | if (is_utf8_pat || utf8_target) { |
| 1882 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 1883 | goto do_exactf_utf8; |
| 1884 | } |
| 1885 | fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */ |
| 1886 | folder = foldEQ_latin1; /* /a, except the sharp s one which */ |
| 1887 | goto do_exactf_non_utf8; /* isn't dealt with by these */ |
| 1888 | |
| 1889 | case EXACTF: /* This node only generated for non-utf8 patterns */ |
| 1890 | assert(! is_utf8_pat); |
| 1891 | if (utf8_target) { |
| 1892 | utf8_fold_flags = 0; |
| 1893 | goto do_exactf_utf8; |
| 1894 | } |
| 1895 | fold_array = PL_fold; |
| 1896 | folder = foldEQ; |
| 1897 | goto do_exactf_non_utf8; |
| 1898 | |
| 1899 | case EXACTFL: |
| 1900 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 1901 | if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) { |
| 1902 | utf8_fold_flags = FOLDEQ_LOCALE; |
| 1903 | goto do_exactf_utf8; |
| 1904 | } |
| 1905 | fold_array = PL_fold_locale; |
| 1906 | folder = foldEQ_locale; |
| 1907 | goto do_exactf_non_utf8; |
| 1908 | |
| 1909 | case EXACTFU_SS: |
| 1910 | if (is_utf8_pat) { |
| 1911 | utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED; |
| 1912 | } |
| 1913 | goto do_exactf_utf8; |
| 1914 | |
| 1915 | case EXACTFLU8: |
| 1916 | if (! utf8_target) { /* All code points in this node require |
| 1917 | UTF-8 to express. */ |
| 1918 | break; |
| 1919 | } |
| 1920 | utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED |
| 1921 | | FOLDEQ_S2_FOLDS_SANE; |
| 1922 | goto do_exactf_utf8; |
| 1923 | |
| 1924 | case EXACTFU: |
| 1925 | if (is_utf8_pat || utf8_target) { |
| 1926 | utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; |
| 1927 | goto do_exactf_utf8; |
| 1928 | } |
| 1929 | |
| 1930 | /* Any 'ss' in the pattern should have been replaced by regcomp, |
| 1931 | * so we don't have to worry here about this single special case |
| 1932 | * in the Latin1 range */ |
| 1933 | fold_array = PL_fold_latin1; |
| 1934 | folder = foldEQ_latin1; |
| 1935 | |
| 1936 | /* FALLTHROUGH */ |
| 1937 | |
| 1938 | do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there |
| 1939 | are no glitches with fold-length differences |
| 1940 | between the target string and pattern */ |
| 1941 | |
| 1942 | /* The idea in the non-utf8 EXACTF* cases is to first find the |
| 1943 | * first character of the EXACTF* node and then, if necessary, |
| 1944 | * case-insensitively compare the full text of the node. c1 is the |
| 1945 | * first character. c2 is its fold. This logic will not work for |
| 1946 | * Unicode semantics and the german sharp ss, which hence should |
| 1947 | * not be compiled into a node that gets here. */ |
| 1948 | pat_string = STRING(c); |
| 1949 | ln = STR_LEN(c); /* length to match in octets/bytes */ |
| 1950 | |
| 1951 | /* We know that we have to match at least 'ln' bytes (which is the |
| 1952 | * same as characters, since not utf8). If we have to match 3 |
| 1953 | * characters, and there are only 2 availabe, we know without |
| 1954 | * trying that it will fail; so don't start a match past the |
| 1955 | * required minimum number from the far end */ |
| 1956 | e = HOP3c(strend, -((SSize_t)ln), s); |
| 1957 | |
| 1958 | if (reginfo->intuit && e < s) { |
| 1959 | e = s; /* Due to minlen logic of intuit() */ |
| 1960 | } |
| 1961 | |
| 1962 | c1 = *pat_string; |
| 1963 | c2 = fold_array[c1]; |
| 1964 | if (c1 == c2) { /* If char and fold are the same */ |
| 1965 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1); |
| 1966 | } |
| 1967 | else { |
| 1968 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2); |
| 1969 | } |
| 1970 | break; |
| 1971 | |
| 1972 | do_exactf_utf8: |
| 1973 | { |
| 1974 | unsigned expansion; |
| 1975 | |
| 1976 | /* If one of the operands is in utf8, we can't use the simpler folding |
| 1977 | * above, due to the fact that many different characters can have the |
| 1978 | * same fold, or portion of a fold, or different- length fold */ |
| 1979 | pat_string = STRING(c); |
| 1980 | ln = STR_LEN(c); /* length to match in octets/bytes */ |
| 1981 | pat_end = pat_string + ln; |
| 1982 | lnc = is_utf8_pat /* length to match in characters */ |
| 1983 | ? utf8_length((U8 *) pat_string, (U8 *) pat_end) |
| 1984 | : ln; |
| 1985 | |
| 1986 | /* We have 'lnc' characters to match in the pattern, but because of |
| 1987 | * multi-character folding, each character in the target can match |
| 1988 | * up to 3 characters (Unicode guarantees it will never exceed |
| 1989 | * this) if it is utf8-encoded; and up to 2 if not (based on the |
| 1990 | * fact that the Latin 1 folds are already determined, and the |
| 1991 | * only multi-char fold in that range is the sharp-s folding to |
| 1992 | * 'ss'. Thus, a pattern character can match as little as 1/3 of a |
| 1993 | * string character. Adjust lnc accordingly, rounding up, so that |
| 1994 | * if we need to match at least 4+1/3 chars, that really is 5. */ |
| 1995 | expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2; |
| 1996 | lnc = (lnc + expansion - 1) / expansion; |
| 1997 | |
| 1998 | /* As in the non-UTF8 case, if we have to match 3 characters, and |
| 1999 | * only 2 are left, it's guaranteed to fail, so don't start a |
| 2000 | * match that would require us to go beyond the end of the string |
| 2001 | */ |
| 2002 | e = HOP3c(strend, -((SSize_t)lnc), s); |
| 2003 | |
| 2004 | if (reginfo->intuit && e < s) { |
| 2005 | e = s; /* Due to minlen logic of intuit() */ |
| 2006 | } |
| 2007 | |
| 2008 | /* XXX Note that we could recalculate e to stop the loop earlier, |
| 2009 | * as the worst case expansion above will rarely be met, and as we |
| 2010 | * go along we would usually find that e moves further to the left. |
| 2011 | * This would happen only after we reached the point in the loop |
| 2012 | * where if there were no expansion we should fail. Unclear if |
| 2013 | * worth the expense */ |
| 2014 | |
| 2015 | while (s <= e) { |
| 2016 | char *my_strend= (char *)strend; |
| 2017 | if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target, |
| 2018 | pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags) |
| 2019 | && (reginfo->intuit || regtry(reginfo, &s)) ) |
| 2020 | { |
| 2021 | goto got_it; |
| 2022 | } |
| 2023 | s += (utf8_target) ? UTF8SKIP(s) : 1; |
| 2024 | } |
| 2025 | break; |
| 2026 | } |
| 2027 | |
| 2028 | case BOUNDL: |
| 2029 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2030 | if (FLAGS(c) != TRADITIONAL_BOUND) { |
| 2031 | if (! IN_UTF8_CTYPE_LOCALE) { |
| 2032 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 2033 | B_ON_NON_UTF8_LOCALE_IS_WRONG); |
| 2034 | } |
| 2035 | goto do_boundu; |
| 2036 | } |
| 2037 | |
| 2038 | FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8); |
| 2039 | break; |
| 2040 | |
| 2041 | case NBOUNDL: |
| 2042 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2043 | if (FLAGS(c) != TRADITIONAL_BOUND) { |
| 2044 | if (! IN_UTF8_CTYPE_LOCALE) { |
| 2045 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 2046 | B_ON_NON_UTF8_LOCALE_IS_WRONG); |
| 2047 | } |
| 2048 | goto do_nboundu; |
| 2049 | } |
| 2050 | |
| 2051 | FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8); |
| 2052 | break; |
| 2053 | |
| 2054 | case BOUND: /* regcomp.c makes sure that this only has the traditional \b |
| 2055 | meaning */ |
| 2056 | assert(FLAGS(c) == TRADITIONAL_BOUND); |
| 2057 | |
| 2058 | FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8); |
| 2059 | break; |
| 2060 | |
| 2061 | case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b |
| 2062 | meaning */ |
| 2063 | assert(FLAGS(c) == TRADITIONAL_BOUND); |
| 2064 | |
| 2065 | FBC_BOUND_A(isWORDCHAR_A); |
| 2066 | break; |
| 2067 | |
| 2068 | case NBOUND: /* regcomp.c makes sure that this only has the traditional \b |
| 2069 | meaning */ |
| 2070 | assert(FLAGS(c) == TRADITIONAL_BOUND); |
| 2071 | |
| 2072 | FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8); |
| 2073 | break; |
| 2074 | |
| 2075 | case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b |
| 2076 | meaning */ |
| 2077 | assert(FLAGS(c) == TRADITIONAL_BOUND); |
| 2078 | |
| 2079 | FBC_NBOUND_A(isWORDCHAR_A); |
| 2080 | break; |
| 2081 | |
| 2082 | case NBOUNDU: |
| 2083 | if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) { |
| 2084 | FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8); |
| 2085 | break; |
| 2086 | } |
| 2087 | |
| 2088 | do_nboundu: |
| 2089 | |
| 2090 | to_complement = 1; |
| 2091 | /* FALLTHROUGH */ |
| 2092 | |
| 2093 | case BOUNDU: |
| 2094 | do_boundu: |
| 2095 | switch((bound_type) FLAGS(c)) { |
| 2096 | case TRADITIONAL_BOUND: |
| 2097 | FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8); |
| 2098 | break; |
| 2099 | case GCB_BOUND: |
| 2100 | if (s == reginfo->strbeg) { |
| 2101 | if (reginfo->intuit || regtry(reginfo, &s)) |
| 2102 | { |
| 2103 | goto got_it; |
| 2104 | } |
| 2105 | |
| 2106 | /* Didn't match. Try at the next position (if there is one) */ |
| 2107 | s += (utf8_target) ? UTF8SKIP(s) : 1; |
| 2108 | if (UNLIKELY(s >= reginfo->strend)) { |
| 2109 | break; |
| 2110 | } |
| 2111 | } |
| 2112 | |
| 2113 | if (utf8_target) { |
| 2114 | GCB_enum before = getGCB_VAL_UTF8( |
| 2115 | reghop3((U8*)s, -1, |
| 2116 | (U8*)(reginfo->strbeg)), |
| 2117 | (U8*) reginfo->strend); |
| 2118 | while (s < strend) { |
| 2119 | GCB_enum after = getGCB_VAL_UTF8((U8*) s, |
| 2120 | (U8*) reginfo->strend); |
| 2121 | if ( (to_complement ^ isGCB(before, after)) |
| 2122 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2123 | { |
| 2124 | goto got_it; |
| 2125 | } |
| 2126 | before = after; |
| 2127 | s += UTF8SKIP(s); |
| 2128 | } |
| 2129 | } |
| 2130 | else { /* Not utf8. Everything is a GCB except between CR and |
| 2131 | LF */ |
| 2132 | while (s < strend) { |
| 2133 | if ((to_complement ^ ( UCHARAT(s - 1) != '\r' |
| 2134 | || UCHARAT(s) != '\n')) |
| 2135 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2136 | { |
| 2137 | goto got_it; |
| 2138 | } |
| 2139 | s++; |
| 2140 | } |
| 2141 | } |
| 2142 | |
| 2143 | /* And, since this is a bound, it can match after the final |
| 2144 | * character in the string */ |
| 2145 | if ((reginfo->intuit || regtry(reginfo, &s))) { |
| 2146 | goto got_it; |
| 2147 | } |
| 2148 | break; |
| 2149 | |
| 2150 | case LB_BOUND: |
| 2151 | if (s == reginfo->strbeg) { |
| 2152 | if (reginfo->intuit || regtry(reginfo, &s)) { |
| 2153 | goto got_it; |
| 2154 | } |
| 2155 | s += (utf8_target) ? UTF8SKIP(s) : 1; |
| 2156 | if (UNLIKELY(s >= reginfo->strend)) { |
| 2157 | break; |
| 2158 | } |
| 2159 | } |
| 2160 | |
| 2161 | if (utf8_target) { |
| 2162 | LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s, |
| 2163 | -1, |
| 2164 | (U8*)(reginfo->strbeg)), |
| 2165 | (U8*) reginfo->strend); |
| 2166 | while (s < strend) { |
| 2167 | LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend); |
| 2168 | if (to_complement ^ isLB(before, |
| 2169 | after, |
| 2170 | (U8*) reginfo->strbeg, |
| 2171 | (U8*) s, |
| 2172 | (U8*) reginfo->strend, |
| 2173 | utf8_target) |
| 2174 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2175 | { |
| 2176 | goto got_it; |
| 2177 | } |
| 2178 | before = after; |
| 2179 | s += UTF8SKIP(s); |
| 2180 | } |
| 2181 | } |
| 2182 | else { /* Not utf8. */ |
| 2183 | LB_enum before = getLB_VAL_CP((U8) *(s -1)); |
| 2184 | while (s < strend) { |
| 2185 | LB_enum after = getLB_VAL_CP((U8) *s); |
| 2186 | if (to_complement ^ isLB(before, |
| 2187 | after, |
| 2188 | (U8*) reginfo->strbeg, |
| 2189 | (U8*) s, |
| 2190 | (U8*) reginfo->strend, |
| 2191 | utf8_target) |
| 2192 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2193 | { |
| 2194 | goto got_it; |
| 2195 | } |
| 2196 | before = after; |
| 2197 | s++; |
| 2198 | } |
| 2199 | } |
| 2200 | |
| 2201 | if (reginfo->intuit || regtry(reginfo, &s)) { |
| 2202 | goto got_it; |
| 2203 | } |
| 2204 | |
| 2205 | break; |
| 2206 | |
| 2207 | case SB_BOUND: |
| 2208 | if (s == reginfo->strbeg) { |
| 2209 | if (reginfo->intuit || regtry(reginfo, &s)) { |
| 2210 | goto got_it; |
| 2211 | } |
| 2212 | s += (utf8_target) ? UTF8SKIP(s) : 1; |
| 2213 | if (UNLIKELY(s >= reginfo->strend)) { |
| 2214 | break; |
| 2215 | } |
| 2216 | } |
| 2217 | |
| 2218 | if (utf8_target) { |
| 2219 | SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s, |
| 2220 | -1, |
| 2221 | (U8*)(reginfo->strbeg)), |
| 2222 | (U8*) reginfo->strend); |
| 2223 | while (s < strend) { |
| 2224 | SB_enum after = getSB_VAL_UTF8((U8*) s, |
| 2225 | (U8*) reginfo->strend); |
| 2226 | if ((to_complement ^ isSB(before, |
| 2227 | after, |
| 2228 | (U8*) reginfo->strbeg, |
| 2229 | (U8*) s, |
| 2230 | (U8*) reginfo->strend, |
| 2231 | utf8_target)) |
| 2232 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2233 | { |
| 2234 | goto got_it; |
| 2235 | } |
| 2236 | before = after; |
| 2237 | s += UTF8SKIP(s); |
| 2238 | } |
| 2239 | } |
| 2240 | else { /* Not utf8. */ |
| 2241 | SB_enum before = getSB_VAL_CP((U8) *(s -1)); |
| 2242 | while (s < strend) { |
| 2243 | SB_enum after = getSB_VAL_CP((U8) *s); |
| 2244 | if ((to_complement ^ isSB(before, |
| 2245 | after, |
| 2246 | (U8*) reginfo->strbeg, |
| 2247 | (U8*) s, |
| 2248 | (U8*) reginfo->strend, |
| 2249 | utf8_target)) |
| 2250 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2251 | { |
| 2252 | goto got_it; |
| 2253 | } |
| 2254 | before = after; |
| 2255 | s++; |
| 2256 | } |
| 2257 | } |
| 2258 | |
| 2259 | /* Here are at the final position in the target string. The SB |
| 2260 | * value is always true here, so matches, depending on other |
| 2261 | * constraints */ |
| 2262 | if (reginfo->intuit || regtry(reginfo, &s)) { |
| 2263 | goto got_it; |
| 2264 | } |
| 2265 | |
| 2266 | break; |
| 2267 | |
| 2268 | case WB_BOUND: |
| 2269 | if (s == reginfo->strbeg) { |
| 2270 | if (reginfo->intuit || regtry(reginfo, &s)) { |
| 2271 | goto got_it; |
| 2272 | } |
| 2273 | s += (utf8_target) ? UTF8SKIP(s) : 1; |
| 2274 | if (UNLIKELY(s >= reginfo->strend)) { |
| 2275 | break; |
| 2276 | } |
| 2277 | } |
| 2278 | |
| 2279 | if (utf8_target) { |
| 2280 | /* We are at a boundary between char_sub_0 and char_sub_1. |
| 2281 | * We also keep track of the value for char_sub_-1 as we |
| 2282 | * loop through the line. Context may be needed to make a |
| 2283 | * determination, and if so, this can save having to |
| 2284 | * recalculate it */ |
| 2285 | WB_enum previous = WB_UNKNOWN; |
| 2286 | WB_enum before = getWB_VAL_UTF8( |
| 2287 | reghop3((U8*)s, |
| 2288 | -1, |
| 2289 | (U8*)(reginfo->strbeg)), |
| 2290 | (U8*) reginfo->strend); |
| 2291 | while (s < strend) { |
| 2292 | WB_enum after = getWB_VAL_UTF8((U8*) s, |
| 2293 | (U8*) reginfo->strend); |
| 2294 | if ((to_complement ^ isWB(previous, |
| 2295 | before, |
| 2296 | after, |
| 2297 | (U8*) reginfo->strbeg, |
| 2298 | (U8*) s, |
| 2299 | (U8*) reginfo->strend, |
| 2300 | utf8_target)) |
| 2301 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2302 | { |
| 2303 | goto got_it; |
| 2304 | } |
| 2305 | previous = before; |
| 2306 | before = after; |
| 2307 | s += UTF8SKIP(s); |
| 2308 | } |
| 2309 | } |
| 2310 | else { /* Not utf8. */ |
| 2311 | WB_enum previous = WB_UNKNOWN; |
| 2312 | WB_enum before = getWB_VAL_CP((U8) *(s -1)); |
| 2313 | while (s < strend) { |
| 2314 | WB_enum after = getWB_VAL_CP((U8) *s); |
| 2315 | if ((to_complement ^ isWB(previous, |
| 2316 | before, |
| 2317 | after, |
| 2318 | (U8*) reginfo->strbeg, |
| 2319 | (U8*) s, |
| 2320 | (U8*) reginfo->strend, |
| 2321 | utf8_target)) |
| 2322 | && (reginfo->intuit || regtry(reginfo, &s))) |
| 2323 | { |
| 2324 | goto got_it; |
| 2325 | } |
| 2326 | previous = before; |
| 2327 | before = after; |
| 2328 | s++; |
| 2329 | } |
| 2330 | } |
| 2331 | |
| 2332 | if (reginfo->intuit || regtry(reginfo, &s)) { |
| 2333 | goto got_it; |
| 2334 | } |
| 2335 | } |
| 2336 | break; |
| 2337 | |
| 2338 | case LNBREAK: |
| 2339 | REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend), |
| 2340 | is_LNBREAK_latin1_safe(s, strend) |
| 2341 | ); |
| 2342 | break; |
| 2343 | |
| 2344 | /* The argument to all the POSIX node types is the class number to pass to |
| 2345 | * _generic_isCC() to build a mask for searching in PL_charclass[] */ |
| 2346 | |
| 2347 | case NPOSIXL: |
| 2348 | to_complement = 1; |
| 2349 | /* FALLTHROUGH */ |
| 2350 | |
| 2351 | case POSIXL: |
| 2352 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 2353 | REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)), |
| 2354 | to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s))); |
| 2355 | break; |
| 2356 | |
| 2357 | case NPOSIXD: |
| 2358 | to_complement = 1; |
| 2359 | /* FALLTHROUGH */ |
| 2360 | |
| 2361 | case POSIXD: |
| 2362 | if (utf8_target) { |
| 2363 | goto posix_utf8; |
| 2364 | } |
| 2365 | goto posixa; |
| 2366 | |
| 2367 | case NPOSIXA: |
| 2368 | if (utf8_target) { |
| 2369 | /* The complement of something that matches only ASCII matches all |
| 2370 | * non-ASCII, plus everything in ASCII that isn't in the class. */ |
| 2371 | REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s) |
| 2372 | || ! _generic_isCC_A(*s, FLAGS(c))); |
| 2373 | break; |
| 2374 | } |
| 2375 | |
| 2376 | to_complement = 1; |
| 2377 | /* FALLTHROUGH */ |
| 2378 | |
| 2379 | case POSIXA: |
| 2380 | posixa: |
| 2381 | /* Don't need to worry about utf8, as it can match only a single |
| 2382 | * byte invariant character. */ |
| 2383 | REXEC_FBC_CLASS_SCAN( |
| 2384 | to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c)))); |
| 2385 | break; |
| 2386 | |
| 2387 | case NPOSIXU: |
| 2388 | to_complement = 1; |
| 2389 | /* FALLTHROUGH */ |
| 2390 | |
| 2391 | case POSIXU: |
| 2392 | if (! utf8_target) { |
| 2393 | REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s, |
| 2394 | FLAGS(c)))); |
| 2395 | } |
| 2396 | else { |
| 2397 | |
| 2398 | posix_utf8: |
| 2399 | classnum = (_char_class_number) FLAGS(c); |
| 2400 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 2401 | while (s < strend) { |
| 2402 | |
| 2403 | /* We avoid loading in the swash as long as possible, but |
| 2404 | * should we have to, we jump to a separate loop. This |
| 2405 | * extra 'if' statement is what keeps this code from being |
| 2406 | * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */ |
| 2407 | if (UTF8_IS_ABOVE_LATIN1(*s)) { |
| 2408 | goto found_above_latin1; |
| 2409 | } |
| 2410 | if ((UTF8_IS_INVARIANT(*s) |
| 2411 | && to_complement ^ cBOOL(_generic_isCC((U8) *s, |
| 2412 | classnum))) |
| 2413 | || (UTF8_IS_DOWNGRADEABLE_START(*s) |
| 2414 | && to_complement ^ cBOOL( |
| 2415 | _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*s, |
| 2416 | *(s + 1)), |
| 2417 | classnum)))) |
| 2418 | { |
| 2419 | if (tmp && (reginfo->intuit || regtry(reginfo, &s))) |
| 2420 | goto got_it; |
| 2421 | else { |
| 2422 | tmp = doevery; |
| 2423 | } |
| 2424 | } |
| 2425 | else { |
| 2426 | tmp = 1; |
| 2427 | } |
| 2428 | s += UTF8SKIP(s); |
| 2429 | } |
| 2430 | } |
| 2431 | else switch (classnum) { /* These classes are implemented as |
| 2432 | macros */ |
| 2433 | case _CC_ENUM_SPACE: |
| 2434 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 2435 | to_complement ^ cBOOL(isSPACE_utf8(s))); |
| 2436 | break; |
| 2437 | |
| 2438 | case _CC_ENUM_BLANK: |
| 2439 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 2440 | to_complement ^ cBOOL(isBLANK_utf8(s))); |
| 2441 | break; |
| 2442 | |
| 2443 | case _CC_ENUM_XDIGIT: |
| 2444 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 2445 | to_complement ^ cBOOL(isXDIGIT_utf8(s))); |
| 2446 | break; |
| 2447 | |
| 2448 | case _CC_ENUM_VERTSPACE: |
| 2449 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 2450 | to_complement ^ cBOOL(isVERTWS_utf8(s))); |
| 2451 | break; |
| 2452 | |
| 2453 | case _CC_ENUM_CNTRL: |
| 2454 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 2455 | to_complement ^ cBOOL(isCNTRL_utf8(s))); |
| 2456 | break; |
| 2457 | |
| 2458 | default: |
| 2459 | Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum); |
| 2460 | NOT_REACHED; /* NOTREACHED */ |
| 2461 | } |
| 2462 | } |
| 2463 | break; |
| 2464 | |
| 2465 | found_above_latin1: /* Here we have to load a swash to get the result |
| 2466 | for the current code point */ |
| 2467 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 2468 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 2469 | PL_utf8_swash_ptrs[classnum] = |
| 2470 | _core_swash_init("utf8", |
| 2471 | "", |
| 2472 | &PL_sv_undef, 1, 0, |
| 2473 | PL_XPosix_ptrs[classnum], &flags); |
| 2474 | } |
| 2475 | |
| 2476 | /* This is a copy of the loop above for swash classes, though using the |
| 2477 | * FBC macro instead of being expanded out. Since we've loaded the |
| 2478 | * swash, we don't have to check for that each time through the loop */ |
| 2479 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 2480 | to_complement ^ cBOOL(_generic_utf8( |
| 2481 | classnum, |
| 2482 | s, |
| 2483 | swash_fetch(PL_utf8_swash_ptrs[classnum], |
| 2484 | (U8 *) s, TRUE)))); |
| 2485 | break; |
| 2486 | |
| 2487 | case AHOCORASICKC: |
| 2488 | case AHOCORASICK: |
| 2489 | { |
| 2490 | DECL_TRIE_TYPE(c); |
| 2491 | /* what trie are we using right now */ |
| 2492 | reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ]; |
| 2493 | reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ]; |
| 2494 | HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]); |
| 2495 | |
| 2496 | const char *last_start = strend - trie->minlen; |
| 2497 | #ifdef DEBUGGING |
| 2498 | const char *real_start = s; |
| 2499 | #endif |
| 2500 | STRLEN maxlen = trie->maxlen; |
| 2501 | SV *sv_points; |
| 2502 | U8 **points; /* map of where we were in the input string |
| 2503 | when reading a given char. For ASCII this |
| 2504 | is unnecessary overhead as the relationship |
| 2505 | is always 1:1, but for Unicode, especially |
| 2506 | case folded Unicode this is not true. */ |
| 2507 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 2508 | U8 *bitmap=NULL; |
| 2509 | |
| 2510 | |
| 2511 | GET_RE_DEBUG_FLAGS_DECL; |
| 2512 | |
| 2513 | /* We can't just allocate points here. We need to wrap it in |
| 2514 | * an SV so it gets freed properly if there is a croak while |
| 2515 | * running the match */ |
| 2516 | ENTER; |
| 2517 | SAVETMPS; |
| 2518 | sv_points=newSV(maxlen * sizeof(U8 *)); |
| 2519 | SvCUR_set(sv_points, |
| 2520 | maxlen * sizeof(U8 *)); |
| 2521 | SvPOK_on(sv_points); |
| 2522 | sv_2mortal(sv_points); |
| 2523 | points=(U8**)SvPV_nolen(sv_points ); |
| 2524 | if ( trie_type != trie_utf8_fold |
| 2525 | && (trie->bitmap || OP(c)==AHOCORASICKC) ) |
| 2526 | { |
| 2527 | if (trie->bitmap) |
| 2528 | bitmap=(U8*)trie->bitmap; |
| 2529 | else |
| 2530 | bitmap=(U8*)ANYOF_BITMAP(c); |
| 2531 | } |
| 2532 | /* this is the Aho-Corasick algorithm modified a touch |
| 2533 | to include special handling for long "unknown char" sequences. |
| 2534 | The basic idea being that we use AC as long as we are dealing |
| 2535 | with a possible matching char, when we encounter an unknown char |
| 2536 | (and we have not encountered an accepting state) we scan forward |
| 2537 | until we find a legal starting char. |
| 2538 | AC matching is basically that of trie matching, except that when |
| 2539 | we encounter a failing transition, we fall back to the current |
| 2540 | states "fail state", and try the current char again, a process |
| 2541 | we repeat until we reach the root state, state 1, or a legal |
| 2542 | transition. If we fail on the root state then we can either |
| 2543 | terminate if we have reached an accepting state previously, or |
| 2544 | restart the entire process from the beginning if we have not. |
| 2545 | |
| 2546 | */ |
| 2547 | while (s <= last_start) { |
| 2548 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
| 2549 | U8 *uc = (U8*)s; |
| 2550 | U16 charid = 0; |
| 2551 | U32 base = 1; |
| 2552 | U32 state = 1; |
| 2553 | UV uvc = 0; |
| 2554 | STRLEN len = 0; |
| 2555 | STRLEN foldlen = 0; |
| 2556 | U8 *uscan = (U8*)NULL; |
| 2557 | U8 *leftmost = NULL; |
| 2558 | #ifdef DEBUGGING |
| 2559 | U32 accepted_word= 0; |
| 2560 | #endif |
| 2561 | U32 pointpos = 0; |
| 2562 | |
| 2563 | while ( state && uc <= (U8*)strend ) { |
| 2564 | int failed=0; |
| 2565 | U32 word = aho->states[ state ].wordnum; |
| 2566 | |
| 2567 | if( state==1 ) { |
| 2568 | if ( bitmap ) { |
| 2569 | DEBUG_TRIE_EXECUTE_r( |
| 2570 | if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 2571 | dump_exec_pos( (char *)uc, c, strend, real_start, |
| 2572 | (char *)uc, utf8_target, 0 ); |
| 2573 | Perl_re_printf( aTHX_ |
| 2574 | " Scanning for legal start char...\n"); |
| 2575 | } |
| 2576 | ); |
| 2577 | if (utf8_target) { |
| 2578 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 2579 | uc += UTF8SKIP(uc); |
| 2580 | } |
| 2581 | } else { |
| 2582 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 2583 | uc++; |
| 2584 | } |
| 2585 | } |
| 2586 | s= (char *)uc; |
| 2587 | } |
| 2588 | if (uc >(U8*)last_start) break; |
| 2589 | } |
| 2590 | |
| 2591 | if ( word ) { |
| 2592 | U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ]; |
| 2593 | if (!leftmost || lpos < leftmost) { |
| 2594 | DEBUG_r(accepted_word=word); |
| 2595 | leftmost= lpos; |
| 2596 | } |
| 2597 | if (base==0) break; |
| 2598 | |
| 2599 | } |
| 2600 | points[pointpos++ % maxlen]= uc; |
| 2601 | if (foldlen || uc < (U8*)strend) { |
| 2602 | REXEC_TRIE_READ_CHAR(trie_type, trie, |
| 2603 | widecharmap, uc, |
| 2604 | uscan, len, uvc, charid, foldlen, |
| 2605 | foldbuf, uniflags); |
| 2606 | DEBUG_TRIE_EXECUTE_r({ |
| 2607 | dump_exec_pos( (char *)uc, c, strend, |
| 2608 | real_start, s, utf8_target, 0); |
| 2609 | Perl_re_printf( aTHX_ |
| 2610 | " Charid:%3u CP:%4"UVxf" ", |
| 2611 | charid, uvc); |
| 2612 | }); |
| 2613 | } |
| 2614 | else { |
| 2615 | len = 0; |
| 2616 | charid = 0; |
| 2617 | } |
| 2618 | |
| 2619 | |
| 2620 | do { |
| 2621 | #ifdef DEBUGGING |
| 2622 | word = aho->states[ state ].wordnum; |
| 2623 | #endif |
| 2624 | base = aho->states[ state ].trans.base; |
| 2625 | |
| 2626 | DEBUG_TRIE_EXECUTE_r({ |
| 2627 | if (failed) |
| 2628 | dump_exec_pos( (char *)uc, c, strend, real_start, |
| 2629 | s, utf8_target, 0 ); |
| 2630 | Perl_re_printf( aTHX_ |
| 2631 | "%sState: %4"UVxf", word=%"UVxf, |
| 2632 | failed ? " Fail transition to " : "", |
| 2633 | (UV)state, (UV)word); |
| 2634 | }); |
| 2635 | if ( base ) { |
| 2636 | U32 tmp; |
| 2637 | I32 offset; |
| 2638 | if (charid && |
| 2639 | ( ((offset = base + charid |
| 2640 | - 1 - trie->uniquecharcount)) >= 0) |
| 2641 | && ((U32)offset < trie->lasttrans) |
| 2642 | && trie->trans[offset].check == state |
| 2643 | && (tmp=trie->trans[offset].next)) |
| 2644 | { |
| 2645 | DEBUG_TRIE_EXECUTE_r( |
| 2646 | Perl_re_printf( aTHX_ " - legal\n")); |
| 2647 | state = tmp; |
| 2648 | break; |
| 2649 | } |
| 2650 | else { |
| 2651 | DEBUG_TRIE_EXECUTE_r( |
| 2652 | Perl_re_printf( aTHX_ " - fail\n")); |
| 2653 | failed = 1; |
| 2654 | state = aho->fail[state]; |
| 2655 | } |
| 2656 | } |
| 2657 | else { |
| 2658 | /* we must be accepting here */ |
| 2659 | DEBUG_TRIE_EXECUTE_r( |
| 2660 | Perl_re_printf( aTHX_ " - accepting\n")); |
| 2661 | failed = 1; |
| 2662 | break; |
| 2663 | } |
| 2664 | } while(state); |
| 2665 | uc += len; |
| 2666 | if (failed) { |
| 2667 | if (leftmost) |
| 2668 | break; |
| 2669 | if (!state) state = 1; |
| 2670 | } |
| 2671 | } |
| 2672 | if ( aho->states[ state ].wordnum ) { |
| 2673 | U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ]; |
| 2674 | if (!leftmost || lpos < leftmost) { |
| 2675 | DEBUG_r(accepted_word=aho->states[ state ].wordnum); |
| 2676 | leftmost = lpos; |
| 2677 | } |
| 2678 | } |
| 2679 | if (leftmost) { |
| 2680 | s = (char*)leftmost; |
| 2681 | DEBUG_TRIE_EXECUTE_r({ |
| 2682 | Perl_re_printf( aTHX_ "Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n", |
| 2683 | (UV)accepted_word, (IV)(s - real_start) |
| 2684 | ); |
| 2685 | }); |
| 2686 | if (reginfo->intuit || regtry(reginfo, &s)) { |
| 2687 | FREETMPS; |
| 2688 | LEAVE; |
| 2689 | goto got_it; |
| 2690 | } |
| 2691 | s = HOPc(s,1); |
| 2692 | DEBUG_TRIE_EXECUTE_r({ |
| 2693 | Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n"); |
| 2694 | }); |
| 2695 | } else { |
| 2696 | DEBUG_TRIE_EXECUTE_r( |
| 2697 | Perl_re_printf( aTHX_ "No match.\n")); |
| 2698 | break; |
| 2699 | } |
| 2700 | } |
| 2701 | FREETMPS; |
| 2702 | LEAVE; |
| 2703 | } |
| 2704 | break; |
| 2705 | default: |
| 2706 | Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c)); |
| 2707 | } |
| 2708 | return 0; |
| 2709 | got_it: |
| 2710 | return s; |
| 2711 | } |
| 2712 | |
| 2713 | /* set RX_SAVED_COPY, RX_SUBBEG etc. |
| 2714 | * flags have same meanings as with regexec_flags() */ |
| 2715 | |
| 2716 | static void |
| 2717 | S_reg_set_capture_string(pTHX_ REGEXP * const rx, |
| 2718 | char *strbeg, |
| 2719 | char *strend, |
| 2720 | SV *sv, |
| 2721 | U32 flags, |
| 2722 | bool utf8_target) |
| 2723 | { |
| 2724 | struct regexp *const prog = ReANY(rx); |
| 2725 | |
| 2726 | if (flags & REXEC_COPY_STR) { |
| 2727 | #ifdef PERL_ANY_COW |
| 2728 | if (SvCANCOW(sv)) { |
| 2729 | DEBUG_C(Perl_re_printf( aTHX_ |
| 2730 | "Copy on write: regexp capture, type %d\n", |
| 2731 | (int) SvTYPE(sv))); |
| 2732 | /* Create a new COW SV to share the match string and store |
| 2733 | * in saved_copy, unless the current COW SV in saved_copy |
| 2734 | * is valid and suitable for our purpose */ |
| 2735 | if (( prog->saved_copy |
| 2736 | && SvIsCOW(prog->saved_copy) |
| 2737 | && SvPOKp(prog->saved_copy) |
| 2738 | && SvIsCOW(sv) |
| 2739 | && SvPOKp(sv) |
| 2740 | && SvPVX(sv) == SvPVX(prog->saved_copy))) |
| 2741 | { |
| 2742 | /* just reuse saved_copy SV */ |
| 2743 | if (RXp_MATCH_COPIED(prog)) { |
| 2744 | Safefree(prog->subbeg); |
| 2745 | RXp_MATCH_COPIED_off(prog); |
| 2746 | } |
| 2747 | } |
| 2748 | else { |
| 2749 | /* create new COW SV to share string */ |
| 2750 | RX_MATCH_COPY_FREE(rx); |
| 2751 | prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv); |
| 2752 | } |
| 2753 | prog->subbeg = (char *)SvPVX_const(prog->saved_copy); |
| 2754 | assert (SvPOKp(prog->saved_copy)); |
| 2755 | prog->sublen = strend - strbeg; |
| 2756 | prog->suboffset = 0; |
| 2757 | prog->subcoffset = 0; |
| 2758 | } else |
| 2759 | #endif |
| 2760 | { |
| 2761 | SSize_t min = 0; |
| 2762 | SSize_t max = strend - strbeg; |
| 2763 | SSize_t sublen; |
| 2764 | |
| 2765 | if ( (flags & REXEC_COPY_SKIP_POST) |
| 2766 | && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ |
| 2767 | && !(PL_sawampersand & SAWAMPERSAND_RIGHT) |
| 2768 | ) { /* don't copy $' part of string */ |
| 2769 | U32 n = 0; |
| 2770 | max = -1; |
| 2771 | /* calculate the right-most part of the string covered |
| 2772 | * by a capture. Due to lookahead, this may be to |
| 2773 | * the right of $&, so we have to scan all captures */ |
| 2774 | while (n <= prog->lastparen) { |
| 2775 | if (prog->offs[n].end > max) |
| 2776 | max = prog->offs[n].end; |
| 2777 | n++; |
| 2778 | } |
| 2779 | if (max == -1) |
| 2780 | max = (PL_sawampersand & SAWAMPERSAND_LEFT) |
| 2781 | ? prog->offs[0].start |
| 2782 | : 0; |
| 2783 | assert(max >= 0 && max <= strend - strbeg); |
| 2784 | } |
| 2785 | |
| 2786 | if ( (flags & REXEC_COPY_SKIP_PRE) |
| 2787 | && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ |
| 2788 | && !(PL_sawampersand & SAWAMPERSAND_LEFT) |
| 2789 | ) { /* don't copy $` part of string */ |
| 2790 | U32 n = 0; |
| 2791 | min = max; |
| 2792 | /* calculate the left-most part of the string covered |
| 2793 | * by a capture. Due to lookbehind, this may be to |
| 2794 | * the left of $&, so we have to scan all captures */ |
| 2795 | while (min && n <= prog->lastparen) { |
| 2796 | if ( prog->offs[n].start != -1 |
| 2797 | && prog->offs[n].start < min) |
| 2798 | { |
| 2799 | min = prog->offs[n].start; |
| 2800 | } |
| 2801 | n++; |
| 2802 | } |
| 2803 | if ((PL_sawampersand & SAWAMPERSAND_RIGHT) |
| 2804 | && min > prog->offs[0].end |
| 2805 | ) |
| 2806 | min = prog->offs[0].end; |
| 2807 | |
| 2808 | } |
| 2809 | |
| 2810 | assert(min >= 0 && min <= max && min <= strend - strbeg); |
| 2811 | sublen = max - min; |
| 2812 | |
| 2813 | if (RX_MATCH_COPIED(rx)) { |
| 2814 | if (sublen > prog->sublen) |
| 2815 | prog->subbeg = |
| 2816 | (char*)saferealloc(prog->subbeg, sublen+1); |
| 2817 | } |
| 2818 | else |
| 2819 | prog->subbeg = (char*)safemalloc(sublen+1); |
| 2820 | Copy(strbeg + min, prog->subbeg, sublen, char); |
| 2821 | prog->subbeg[sublen] = '\0'; |
| 2822 | prog->suboffset = min; |
| 2823 | prog->sublen = sublen; |
| 2824 | RX_MATCH_COPIED_on(rx); |
| 2825 | } |
| 2826 | prog->subcoffset = prog->suboffset; |
| 2827 | if (prog->suboffset && utf8_target) { |
| 2828 | /* Convert byte offset to chars. |
| 2829 | * XXX ideally should only compute this if @-/@+ |
| 2830 | * has been seen, a la PL_sawampersand ??? */ |
| 2831 | |
| 2832 | /* If there's a direct correspondence between the |
| 2833 | * string which we're matching and the original SV, |
| 2834 | * then we can use the utf8 len cache associated with |
| 2835 | * the SV. In particular, it means that under //g, |
| 2836 | * sv_pos_b2u() will use the previously cached |
| 2837 | * position to speed up working out the new length of |
| 2838 | * subcoffset, rather than counting from the start of |
| 2839 | * the string each time. This stops |
| 2840 | * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g; |
| 2841 | * from going quadratic */ |
| 2842 | if (SvPOKp(sv) && SvPVX(sv) == strbeg) |
| 2843 | prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset, |
| 2844 | SV_GMAGIC|SV_CONST_RETURN); |
| 2845 | else |
| 2846 | prog->subcoffset = utf8_length((U8*)strbeg, |
| 2847 | (U8*)(strbeg+prog->suboffset)); |
| 2848 | } |
| 2849 | } |
| 2850 | else { |
| 2851 | RX_MATCH_COPY_FREE(rx); |
| 2852 | prog->subbeg = strbeg; |
| 2853 | prog->suboffset = 0; |
| 2854 | prog->subcoffset = 0; |
| 2855 | prog->sublen = strend - strbeg; |
| 2856 | } |
| 2857 | } |
| 2858 | |
| 2859 | |
| 2860 | |
| 2861 | |
| 2862 | /* |
| 2863 | - regexec_flags - match a regexp against a string |
| 2864 | */ |
| 2865 | I32 |
| 2866 | Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend, |
| 2867 | char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags) |
| 2868 | /* stringarg: the point in the string at which to begin matching */ |
| 2869 | /* strend: pointer to null at end of string */ |
| 2870 | /* strbeg: real beginning of string */ |
| 2871 | /* minend: end of match must be >= minend bytes after stringarg. */ |
| 2872 | /* sv: SV being matched: only used for utf8 flag, pos() etc; string |
| 2873 | * itself is accessed via the pointers above */ |
| 2874 | /* data: May be used for some additional optimizations. |
| 2875 | Currently unused. */ |
| 2876 | /* flags: For optimizations. See REXEC_* in regexp.h */ |
| 2877 | |
| 2878 | { |
| 2879 | struct regexp *const prog = ReANY(rx); |
| 2880 | char *s; |
| 2881 | regnode *c; |
| 2882 | char *startpos; |
| 2883 | SSize_t minlen; /* must match at least this many chars */ |
| 2884 | SSize_t dontbother = 0; /* how many characters not to try at end */ |
| 2885 | const bool utf8_target = cBOOL(DO_UTF8(sv)); |
| 2886 | I32 multiline; |
| 2887 | RXi_GET_DECL(prog,progi); |
| 2888 | regmatch_info reginfo_buf; /* create some info to pass to regtry etc */ |
| 2889 | regmatch_info *const reginfo = ®info_buf; |
| 2890 | regexp_paren_pair *swap = NULL; |
| 2891 | I32 oldsave; |
| 2892 | GET_RE_DEBUG_FLAGS_DECL; |
| 2893 | |
| 2894 | PERL_ARGS_ASSERT_REGEXEC_FLAGS; |
| 2895 | PERL_UNUSED_ARG(data); |
| 2896 | |
| 2897 | /* Be paranoid... */ |
| 2898 | if (prog == NULL) { |
| 2899 | Perl_croak(aTHX_ "NULL regexp parameter"); |
| 2900 | } |
| 2901 | |
| 2902 | DEBUG_EXECUTE_r( |
| 2903 | debug_start_match(rx, utf8_target, stringarg, strend, |
| 2904 | "Matching"); |
| 2905 | ); |
| 2906 | |
| 2907 | startpos = stringarg; |
| 2908 | |
| 2909 | /* set these early as they may be used by the HOP macros below */ |
| 2910 | reginfo->strbeg = strbeg; |
| 2911 | reginfo->strend = strend; |
| 2912 | reginfo->is_utf8_target = cBOOL(utf8_target); |
| 2913 | |
| 2914 | if (prog->intflags & PREGf_GPOS_SEEN) { |
| 2915 | MAGIC *mg; |
| 2916 | |
| 2917 | /* set reginfo->ganch, the position where \G can match */ |
| 2918 | |
| 2919 | reginfo->ganch = |
| 2920 | (flags & REXEC_IGNOREPOS) |
| 2921 | ? stringarg /* use start pos rather than pos() */ |
| 2922 | : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0) |
| 2923 | /* Defined pos(): */ |
| 2924 | ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg) |
| 2925 | : strbeg; /* pos() not defined; use start of string */ |
| 2926 | |
| 2927 | DEBUG_GPOS_r(Perl_re_printf( aTHX_ |
| 2928 | "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg))); |
| 2929 | |
| 2930 | /* in the presence of \G, we may need to start looking earlier in |
| 2931 | * the string than the suggested start point of stringarg: |
| 2932 | * if prog->gofs is set, then that's a known, fixed minimum |
| 2933 | * offset, such as |
| 2934 | * /..\G/: gofs = 2 |
| 2935 | * /ab|c\G/: gofs = 1 |
| 2936 | * or if the minimum offset isn't known, then we have to go back |
| 2937 | * to the start of the string, e.g. /w+\G/ |
| 2938 | */ |
| 2939 | |
| 2940 | if (prog->intflags & PREGf_ANCH_GPOS) { |
| 2941 | if (prog->gofs) { |
| 2942 | startpos = HOPBACKc(reginfo->ganch, prog->gofs); |
| 2943 | if (!startpos || |
| 2944 | ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg)) |
| 2945 | { |
| 2946 | DEBUG_r(Perl_re_printf( aTHX_ |
| 2947 | "fail: ganch-gofs before earliest possible start\n")); |
| 2948 | return 0; |
| 2949 | } |
| 2950 | } |
| 2951 | else |
| 2952 | startpos = reginfo->ganch; |
| 2953 | } |
| 2954 | else if (prog->gofs) { |
| 2955 | startpos = HOPBACKc(startpos, prog->gofs); |
| 2956 | if (!startpos) |
| 2957 | startpos = strbeg; |
| 2958 | } |
| 2959 | else if (prog->intflags & PREGf_GPOS_FLOAT) |
| 2960 | startpos = strbeg; |
| 2961 | } |
| 2962 | |
| 2963 | minlen = prog->minlen; |
| 2964 | if ((startpos + minlen) > strend || startpos < strbeg) { |
| 2965 | DEBUG_r(Perl_re_printf( aTHX_ |
| 2966 | "Regex match can't succeed, so not even tried\n")); |
| 2967 | return 0; |
| 2968 | } |
| 2969 | |
| 2970 | /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave), |
| 2971 | * which will call destuctors to reset PL_regmatch_state, free higher |
| 2972 | * PL_regmatch_slabs, and clean up regmatch_info_aux and |
| 2973 | * regmatch_info_aux_eval */ |
| 2974 | |
| 2975 | oldsave = PL_savestack_ix; |
| 2976 | |
| 2977 | s = startpos; |
| 2978 | |
| 2979 | if ((prog->extflags & RXf_USE_INTUIT) |
| 2980 | && !(flags & REXEC_CHECKED)) |
| 2981 | { |
| 2982 | s = re_intuit_start(rx, sv, strbeg, startpos, strend, |
| 2983 | flags, NULL); |
| 2984 | if (!s) |
| 2985 | return 0; |
| 2986 | |
| 2987 | if (prog->extflags & RXf_CHECK_ALL) { |
| 2988 | /* we can match based purely on the result of INTUIT. |
| 2989 | * Set up captures etc just for $& and $-[0] |
| 2990 | * (an intuit-only match wont have $1,$2,..) */ |
| 2991 | assert(!prog->nparens); |
| 2992 | |
| 2993 | /* s/// doesn't like it if $& is earlier than where we asked it to |
| 2994 | * start searching (which can happen on something like /.\G/) */ |
| 2995 | if ( (flags & REXEC_FAIL_ON_UNDERFLOW) |
| 2996 | && (s < stringarg)) |
| 2997 | { |
| 2998 | /* this should only be possible under \G */ |
| 2999 | assert(prog->intflags & PREGf_GPOS_SEEN); |
| 3000 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 3001 | "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); |
| 3002 | goto phooey; |
| 3003 | } |
| 3004 | |
| 3005 | /* match via INTUIT shouldn't have any captures. |
| 3006 | * Let @-, @+, $^N know */ |
| 3007 | prog->lastparen = prog->lastcloseparen = 0; |
| 3008 | RX_MATCH_UTF8_set(rx, utf8_target); |
| 3009 | prog->offs[0].start = s - strbeg; |
| 3010 | prog->offs[0].end = utf8_target |
| 3011 | ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg |
| 3012 | : s - strbeg + prog->minlenret; |
| 3013 | if ( !(flags & REXEC_NOT_FIRST) ) |
| 3014 | S_reg_set_capture_string(aTHX_ rx, |
| 3015 | strbeg, strend, |
| 3016 | sv, flags, utf8_target); |
| 3017 | |
| 3018 | return 1; |
| 3019 | } |
| 3020 | } |
| 3021 | |
| 3022 | multiline = prog->extflags & RXf_PMf_MULTILINE; |
| 3023 | |
| 3024 | if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { |
| 3025 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 3026 | "String too short [regexec_flags]...\n")); |
| 3027 | goto phooey; |
| 3028 | } |
| 3029 | |
| 3030 | /* Check validity of program. */ |
| 3031 | if (UCHARAT(progi->program) != REG_MAGIC) { |
| 3032 | Perl_croak(aTHX_ "corrupted regexp program"); |
| 3033 | } |
| 3034 | |
| 3035 | RX_MATCH_TAINTED_off(rx); |
| 3036 | RX_MATCH_UTF8_set(rx, utf8_target); |
| 3037 | |
| 3038 | reginfo->prog = rx; /* Yes, sorry that this is confusing. */ |
| 3039 | reginfo->intuit = 0; |
| 3040 | reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx)); |
| 3041 | reginfo->warned = FALSE; |
| 3042 | reginfo->sv = sv; |
| 3043 | reginfo->poscache_maxiter = 0; /* not yet started a countdown */ |
| 3044 | /* see how far we have to get to not match where we matched before */ |
| 3045 | reginfo->till = stringarg + minend; |
| 3046 | |
| 3047 | if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) { |
| 3048 | /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after |
| 3049 | S_cleanup_regmatch_info_aux has executed (registered by |
| 3050 | SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies |
| 3051 | magic belonging to this SV. |
| 3052 | Not newSVsv, either, as it does not COW. |
| 3053 | */ |
| 3054 | reginfo->sv = newSV(0); |
| 3055 | SvSetSV_nosteal(reginfo->sv, sv); |
| 3056 | SAVEFREESV(reginfo->sv); |
| 3057 | } |
| 3058 | |
| 3059 | /* reserve next 2 or 3 slots in PL_regmatch_state: |
| 3060 | * slot N+0: may currently be in use: skip it |
| 3061 | * slot N+1: use for regmatch_info_aux struct |
| 3062 | * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s |
| 3063 | * slot N+3: ready for use by regmatch() |
| 3064 | */ |
| 3065 | |
| 3066 | { |
| 3067 | regmatch_state *old_regmatch_state; |
| 3068 | regmatch_slab *old_regmatch_slab; |
| 3069 | int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1; |
| 3070 | |
| 3071 | /* on first ever match, allocate first slab */ |
| 3072 | if (!PL_regmatch_slab) { |
| 3073 | Newx(PL_regmatch_slab, 1, regmatch_slab); |
| 3074 | PL_regmatch_slab->prev = NULL; |
| 3075 | PL_regmatch_slab->next = NULL; |
| 3076 | PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab); |
| 3077 | } |
| 3078 | |
| 3079 | old_regmatch_state = PL_regmatch_state; |
| 3080 | old_regmatch_slab = PL_regmatch_slab; |
| 3081 | |
| 3082 | for (i=0; i <= max; i++) { |
| 3083 | if (i == 1) |
| 3084 | reginfo->info_aux = &(PL_regmatch_state->u.info_aux); |
| 3085 | else if (i ==2) |
| 3086 | reginfo->info_aux_eval = |
| 3087 | reginfo->info_aux->info_aux_eval = |
| 3088 | &(PL_regmatch_state->u.info_aux_eval); |
| 3089 | |
| 3090 | if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab)) |
| 3091 | PL_regmatch_state = S_push_slab(aTHX); |
| 3092 | } |
| 3093 | |
| 3094 | /* note initial PL_regmatch_state position; at end of match we'll |
| 3095 | * pop back to there and free any higher slabs */ |
| 3096 | |
| 3097 | reginfo->info_aux->old_regmatch_state = old_regmatch_state; |
| 3098 | reginfo->info_aux->old_regmatch_slab = old_regmatch_slab; |
| 3099 | reginfo->info_aux->poscache = NULL; |
| 3100 | |
| 3101 | SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux); |
| 3102 | |
| 3103 | if ((prog->extflags & RXf_EVAL_SEEN)) |
| 3104 | S_setup_eval_state(aTHX_ reginfo); |
| 3105 | else |
| 3106 | reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL; |
| 3107 | } |
| 3108 | |
| 3109 | /* If there is a "must appear" string, look for it. */ |
| 3110 | |
| 3111 | if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) { |
| 3112 | /* We have to be careful. If the previous successful match |
| 3113 | was from this regex we don't want a subsequent partially |
| 3114 | successful match to clobber the old results. |
| 3115 | So when we detect this possibility we add a swap buffer |
| 3116 | to the re, and switch the buffer each match. If we fail, |
| 3117 | we switch it back; otherwise we leave it swapped. |
| 3118 | */ |
| 3119 | swap = prog->offs; |
| 3120 | /* do we need a save destructor here for eval dies? */ |
| 3121 | Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair); |
| 3122 | DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ |
| 3123 | "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n", |
| 3124 | PTR2UV(prog), |
| 3125 | PTR2UV(swap), |
| 3126 | PTR2UV(prog->offs) |
| 3127 | )); |
| 3128 | } |
| 3129 | |
| 3130 | if (prog->recurse_locinput) |
| 3131 | Zero(prog->recurse_locinput,prog->nparens + 1, char *); |
| 3132 | |
| 3133 | /* Simplest case: anchored match need be tried only once, or with |
| 3134 | * MBOL, only at the beginning of each line. |
| 3135 | * |
| 3136 | * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets |
| 3137 | * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't |
| 3138 | * match at the start of the string then it won't match anywhere else |
| 3139 | * either; while with /.*.../, if it doesn't match at the beginning, |
| 3140 | * the earliest it could match is at the start of the next line */ |
| 3141 | |
| 3142 | if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) { |
| 3143 | char *end; |
| 3144 | |
| 3145 | if (regtry(reginfo, &s)) |
| 3146 | goto got_it; |
| 3147 | |
| 3148 | if (!(prog->intflags & PREGf_ANCH_MBOL)) |
| 3149 | goto phooey; |
| 3150 | |
| 3151 | /* didn't match at start, try at other newline positions */ |
| 3152 | |
| 3153 | if (minlen) |
| 3154 | dontbother = minlen - 1; |
| 3155 | end = HOP3c(strend, -dontbother, strbeg) - 1; |
| 3156 | |
| 3157 | /* skip to next newline */ |
| 3158 | |
| 3159 | while (s <= end) { /* note it could be possible to match at the end of the string */ |
| 3160 | /* NB: newlines are the same in unicode as they are in latin */ |
| 3161 | if (*s++ != '\n') |
| 3162 | continue; |
| 3163 | if (prog->check_substr || prog->check_utf8) { |
| 3164 | /* note that with PREGf_IMPLICIT, intuit can only fail |
| 3165 | * or return the start position, so it's of limited utility. |
| 3166 | * Nevertheless, I made the decision that the potential for |
| 3167 | * quick fail was still worth it - DAPM */ |
| 3168 | s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL); |
| 3169 | if (!s) |
| 3170 | goto phooey; |
| 3171 | } |
| 3172 | if (regtry(reginfo, &s)) |
| 3173 | goto got_it; |
| 3174 | } |
| 3175 | goto phooey; |
| 3176 | } /* end anchored search */ |
| 3177 | |
| 3178 | if (prog->intflags & PREGf_ANCH_GPOS) |
| 3179 | { |
| 3180 | /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */ |
| 3181 | assert(prog->intflags & PREGf_GPOS_SEEN); |
| 3182 | /* For anchored \G, the only position it can match from is |
| 3183 | * (ganch-gofs); we already set startpos to this above; if intuit |
| 3184 | * moved us on from there, we can't possibly succeed */ |
| 3185 | assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs)); |
| 3186 | if (s == startpos && regtry(reginfo, &s)) |
| 3187 | goto got_it; |
| 3188 | goto phooey; |
| 3189 | } |
| 3190 | |
| 3191 | /* Messy cases: unanchored match. */ |
| 3192 | if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) { |
| 3193 | /* we have /x+whatever/ */ |
| 3194 | /* it must be a one character string (XXXX Except is_utf8_pat?) */ |
| 3195 | char ch; |
| 3196 | #ifdef DEBUGGING |
| 3197 | int did_match = 0; |
| 3198 | #endif |
| 3199 | if (utf8_target) { |
| 3200 | if (! prog->anchored_utf8) { |
| 3201 | to_utf8_substr(prog); |
| 3202 | } |
| 3203 | ch = SvPVX_const(prog->anchored_utf8)[0]; |
| 3204 | REXEC_FBC_SCAN( |
| 3205 | if (*s == ch) { |
| 3206 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 3207 | if (regtry(reginfo, &s)) goto got_it; |
| 3208 | s += UTF8SKIP(s); |
| 3209 | while (s < strend && *s == ch) |
| 3210 | s += UTF8SKIP(s); |
| 3211 | } |
| 3212 | ); |
| 3213 | |
| 3214 | } |
| 3215 | else { |
| 3216 | if (! prog->anchored_substr) { |
| 3217 | if (! to_byte_substr(prog)) { |
| 3218 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 3219 | } |
| 3220 | } |
| 3221 | ch = SvPVX_const(prog->anchored_substr)[0]; |
| 3222 | REXEC_FBC_SCAN( |
| 3223 | if (*s == ch) { |
| 3224 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 3225 | if (regtry(reginfo, &s)) goto got_it; |
| 3226 | s++; |
| 3227 | while (s < strend && *s == ch) |
| 3228 | s++; |
| 3229 | } |
| 3230 | ); |
| 3231 | } |
| 3232 | DEBUG_EXECUTE_r(if (!did_match) |
| 3233 | Perl_re_printf( aTHX_ |
| 3234 | "Did not find anchored character...\n") |
| 3235 | ); |
| 3236 | } |
| 3237 | else if (prog->anchored_substr != NULL |
| 3238 | || prog->anchored_utf8 != NULL |
| 3239 | || ((prog->float_substr != NULL || prog->float_utf8 != NULL) |
| 3240 | && prog->float_max_offset < strend - s)) { |
| 3241 | SV *must; |
| 3242 | SSize_t back_max; |
| 3243 | SSize_t back_min; |
| 3244 | char *last; |
| 3245 | char *last1; /* Last position checked before */ |
| 3246 | #ifdef DEBUGGING |
| 3247 | int did_match = 0; |
| 3248 | #endif |
| 3249 | if (prog->anchored_substr || prog->anchored_utf8) { |
| 3250 | if (utf8_target) { |
| 3251 | if (! prog->anchored_utf8) { |
| 3252 | to_utf8_substr(prog); |
| 3253 | } |
| 3254 | must = prog->anchored_utf8; |
| 3255 | } |
| 3256 | else { |
| 3257 | if (! prog->anchored_substr) { |
| 3258 | if (! to_byte_substr(prog)) { |
| 3259 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 3260 | } |
| 3261 | } |
| 3262 | must = prog->anchored_substr; |
| 3263 | } |
| 3264 | back_max = back_min = prog->anchored_offset; |
| 3265 | } else { |
| 3266 | if (utf8_target) { |
| 3267 | if (! prog->float_utf8) { |
| 3268 | to_utf8_substr(prog); |
| 3269 | } |
| 3270 | must = prog->float_utf8; |
| 3271 | } |
| 3272 | else { |
| 3273 | if (! prog->float_substr) { |
| 3274 | if (! to_byte_substr(prog)) { |
| 3275 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 3276 | } |
| 3277 | } |
| 3278 | must = prog->float_substr; |
| 3279 | } |
| 3280 | back_max = prog->float_max_offset; |
| 3281 | back_min = prog->float_min_offset; |
| 3282 | } |
| 3283 | |
| 3284 | if (back_min<0) { |
| 3285 | last = strend; |
| 3286 | } else { |
| 3287 | last = HOP3c(strend, /* Cannot start after this */ |
| 3288 | -(SSize_t)(CHR_SVLEN(must) |
| 3289 | - (SvTAIL(must) != 0) + back_min), strbeg); |
| 3290 | } |
| 3291 | if (s > reginfo->strbeg) |
| 3292 | last1 = HOPc(s, -1); |
| 3293 | else |
| 3294 | last1 = s - 1; /* bogus */ |
| 3295 | |
| 3296 | /* XXXX check_substr already used to find "s", can optimize if |
| 3297 | check_substr==must. */ |
| 3298 | dontbother = 0; |
| 3299 | strend = HOPc(strend, -dontbother); |
| 3300 | while ( (s <= last) && |
| 3301 | (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend), |
| 3302 | (unsigned char*)strend, must, |
| 3303 | multiline ? FBMrf_MULTILINE : 0)) ) { |
| 3304 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 3305 | if (HOPc(s, -back_max) > last1) { |
| 3306 | last1 = HOPc(s, -back_min); |
| 3307 | s = HOPc(s, -back_max); |
| 3308 | } |
| 3309 | else { |
| 3310 | char * const t = (last1 >= reginfo->strbeg) |
| 3311 | ? HOPc(last1, 1) : last1 + 1; |
| 3312 | |
| 3313 | last1 = HOPc(s, -back_min); |
| 3314 | s = t; |
| 3315 | } |
| 3316 | if (utf8_target) { |
| 3317 | while (s <= last1) { |
| 3318 | if (regtry(reginfo, &s)) |
| 3319 | goto got_it; |
| 3320 | if (s >= last1) { |
| 3321 | s++; /* to break out of outer loop */ |
| 3322 | break; |
| 3323 | } |
| 3324 | s += UTF8SKIP(s); |
| 3325 | } |
| 3326 | } |
| 3327 | else { |
| 3328 | while (s <= last1) { |
| 3329 | if (regtry(reginfo, &s)) |
| 3330 | goto got_it; |
| 3331 | s++; |
| 3332 | } |
| 3333 | } |
| 3334 | } |
| 3335 | DEBUG_EXECUTE_r(if (!did_match) { |
| 3336 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 3337 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 3338 | Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n", |
| 3339 | ((must == prog->anchored_substr || must == prog->anchored_utf8) |
| 3340 | ? "anchored" : "floating"), |
| 3341 | quoted, RE_SV_TAIL(must)); |
| 3342 | }); |
| 3343 | goto phooey; |
| 3344 | } |
| 3345 | else if ( (c = progi->regstclass) ) { |
| 3346 | if (minlen) { |
| 3347 | const OPCODE op = OP(progi->regstclass); |
| 3348 | /* don't bother with what can't match */ |
| 3349 | if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE) |
| 3350 | strend = HOPc(strend, -(minlen - 1)); |
| 3351 | } |
| 3352 | DEBUG_EXECUTE_r({ |
| 3353 | SV * const prop = sv_newmortal(); |
| 3354 | regprop(prog, prop, c, reginfo, NULL); |
| 3355 | { |
| 3356 | RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1), |
| 3357 | s,strend-s,60); |
| 3358 | Perl_re_printf( aTHX_ |
| 3359 | "Matching stclass %.*s against %s (%d bytes)\n", |
| 3360 | (int)SvCUR(prop), SvPVX_const(prop), |
| 3361 | quoted, (int)(strend - s)); |
| 3362 | } |
| 3363 | }); |
| 3364 | if (find_byclass(prog, c, s, strend, reginfo)) |
| 3365 | goto got_it; |
| 3366 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n")); |
| 3367 | } |
| 3368 | else { |
| 3369 | dontbother = 0; |
| 3370 | if (prog->float_substr != NULL || prog->float_utf8 != NULL) { |
| 3371 | /* Trim the end. */ |
| 3372 | char *last= NULL; |
| 3373 | SV* float_real; |
| 3374 | STRLEN len; |
| 3375 | const char *little; |
| 3376 | |
| 3377 | if (utf8_target) { |
| 3378 | if (! prog->float_utf8) { |
| 3379 | to_utf8_substr(prog); |
| 3380 | } |
| 3381 | float_real = prog->float_utf8; |
| 3382 | } |
| 3383 | else { |
| 3384 | if (! prog->float_substr) { |
| 3385 | if (! to_byte_substr(prog)) { |
| 3386 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 3387 | } |
| 3388 | } |
| 3389 | float_real = prog->float_substr; |
| 3390 | } |
| 3391 | |
| 3392 | little = SvPV_const(float_real, len); |
| 3393 | if (SvTAIL(float_real)) { |
| 3394 | /* This means that float_real contains an artificial \n on |
| 3395 | * the end due to the presence of something like this: |
| 3396 | * /foo$/ where we can match both "foo" and "foo\n" at the |
| 3397 | * end of the string. So we have to compare the end of the |
| 3398 | * string first against the float_real without the \n and |
| 3399 | * then against the full float_real with the string. We |
| 3400 | * have to watch out for cases where the string might be |
| 3401 | * smaller than the float_real or the float_real without |
| 3402 | * the \n. */ |
| 3403 | char *checkpos= strend - len; |
| 3404 | DEBUG_OPTIMISE_r( |
| 3405 | Perl_re_printf( aTHX_ |
| 3406 | "%sChecking for float_real.%s\n", |
| 3407 | PL_colors[4], PL_colors[5])); |
| 3408 | if (checkpos + 1 < strbeg) { |
| 3409 | /* can't match, even if we remove the trailing \n |
| 3410 | * string is too short to match */ |
| 3411 | DEBUG_EXECUTE_r( |
| 3412 | Perl_re_printf( aTHX_ |
| 3413 | "%sString shorter than required trailing substring, cannot match.%s\n", |
| 3414 | PL_colors[4], PL_colors[5])); |
| 3415 | goto phooey; |
| 3416 | } else if (memEQ(checkpos + 1, little, len - 1)) { |
| 3417 | /* can match, the end of the string matches without the |
| 3418 | * "\n" */ |
| 3419 | last = checkpos + 1; |
| 3420 | } else if (checkpos < strbeg) { |
| 3421 | /* cant match, string is too short when the "\n" is |
| 3422 | * included */ |
| 3423 | DEBUG_EXECUTE_r( |
| 3424 | Perl_re_printf( aTHX_ |
| 3425 | "%sString does not contain required trailing substring, cannot match.%s\n", |
| 3426 | PL_colors[4], PL_colors[5])); |
| 3427 | goto phooey; |
| 3428 | } else if (!multiline) { |
| 3429 | /* non multiline match, so compare with the "\n" at the |
| 3430 | * end of the string */ |
| 3431 | if (memEQ(checkpos, little, len)) { |
| 3432 | last= checkpos; |
| 3433 | } else { |
| 3434 | DEBUG_EXECUTE_r( |
| 3435 | Perl_re_printf( aTHX_ |
| 3436 | "%sString does not contain required trailing substring, cannot match.%s\n", |
| 3437 | PL_colors[4], PL_colors[5])); |
| 3438 | goto phooey; |
| 3439 | } |
| 3440 | } else { |
| 3441 | /* multiline match, so we have to search for a place |
| 3442 | * where the full string is located */ |
| 3443 | goto find_last; |
| 3444 | } |
| 3445 | } else { |
| 3446 | find_last: |
| 3447 | if (len) |
| 3448 | last = rninstr(s, strend, little, little + len); |
| 3449 | else |
| 3450 | last = strend; /* matching "$" */ |
| 3451 | } |
| 3452 | if (!last) { |
| 3453 | /* at one point this block contained a comment which was |
| 3454 | * probably incorrect, which said that this was a "should not |
| 3455 | * happen" case. Even if it was true when it was written I am |
| 3456 | * pretty sure it is not anymore, so I have removed the comment |
| 3457 | * and replaced it with this one. Yves */ |
| 3458 | DEBUG_EXECUTE_r( |
| 3459 | Perl_re_printf( aTHX_ |
| 3460 | "%sString does not contain required substring, cannot match.%s\n", |
| 3461 | PL_colors[4], PL_colors[5] |
| 3462 | )); |
| 3463 | goto phooey; |
| 3464 | } |
| 3465 | dontbother = strend - last + prog->float_min_offset; |
| 3466 | } |
| 3467 | if (minlen && (dontbother < minlen)) |
| 3468 | dontbother = minlen - 1; |
| 3469 | strend -= dontbother; /* this one's always in bytes! */ |
| 3470 | /* We don't know much -- general case. */ |
| 3471 | if (utf8_target) { |
| 3472 | for (;;) { |
| 3473 | if (regtry(reginfo, &s)) |
| 3474 | goto got_it; |
| 3475 | if (s >= strend) |
| 3476 | break; |
| 3477 | s += UTF8SKIP(s); |
| 3478 | }; |
| 3479 | } |
| 3480 | else { |
| 3481 | do { |
| 3482 | if (regtry(reginfo, &s)) |
| 3483 | goto got_it; |
| 3484 | } while (s++ < strend); |
| 3485 | } |
| 3486 | } |
| 3487 | |
| 3488 | /* Failure. */ |
| 3489 | goto phooey; |
| 3490 | |
| 3491 | got_it: |
| 3492 | /* s/// doesn't like it if $& is earlier than where we asked it to |
| 3493 | * start searching (which can happen on something like /.\G/) */ |
| 3494 | if ( (flags & REXEC_FAIL_ON_UNDERFLOW) |
| 3495 | && (prog->offs[0].start < stringarg - strbeg)) |
| 3496 | { |
| 3497 | /* this should only be possible under \G */ |
| 3498 | assert(prog->intflags & PREGf_GPOS_SEEN); |
| 3499 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 3500 | "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); |
| 3501 | goto phooey; |
| 3502 | } |
| 3503 | |
| 3504 | DEBUG_BUFFERS_r( |
| 3505 | if (swap) |
| 3506 | Perl_re_printf( aTHX_ |
| 3507 | "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n", |
| 3508 | PTR2UV(prog), |
| 3509 | PTR2UV(swap) |
| 3510 | ); |
| 3511 | ); |
| 3512 | Safefree(swap); |
| 3513 | |
| 3514 | /* clean up; this will trigger destructors that will free all slabs |
| 3515 | * above the current one, and cleanup the regmatch_info_aux |
| 3516 | * and regmatch_info_aux_eval sructs */ |
| 3517 | |
| 3518 | LEAVE_SCOPE(oldsave); |
| 3519 | |
| 3520 | if (RXp_PAREN_NAMES(prog)) |
| 3521 | (void)hv_iterinit(RXp_PAREN_NAMES(prog)); |
| 3522 | |
| 3523 | /* make sure $`, $&, $', and $digit will work later */ |
| 3524 | if ( !(flags & REXEC_NOT_FIRST) ) |
| 3525 | S_reg_set_capture_string(aTHX_ rx, |
| 3526 | strbeg, reginfo->strend, |
| 3527 | sv, flags, utf8_target); |
| 3528 | |
| 3529 | return 1; |
| 3530 | |
| 3531 | phooey: |
| 3532 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n", |
| 3533 | PL_colors[4], PL_colors[5])); |
| 3534 | |
| 3535 | /* clean up; this will trigger destructors that will free all slabs |
| 3536 | * above the current one, and cleanup the regmatch_info_aux |
| 3537 | * and regmatch_info_aux_eval sructs */ |
| 3538 | |
| 3539 | LEAVE_SCOPE(oldsave); |
| 3540 | |
| 3541 | if (swap) { |
| 3542 | /* we failed :-( roll it back */ |
| 3543 | DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ |
| 3544 | "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n", |
| 3545 | PTR2UV(prog), |
| 3546 | PTR2UV(prog->offs), |
| 3547 | PTR2UV(swap) |
| 3548 | )); |
| 3549 | Safefree(prog->offs); |
| 3550 | prog->offs = swap; |
| 3551 | } |
| 3552 | return 0; |
| 3553 | } |
| 3554 | |
| 3555 | |
| 3556 | /* Set which rex is pointed to by PL_reg_curpm, handling ref counting. |
| 3557 | * Do inc before dec, in case old and new rex are the same */ |
| 3558 | #define SET_reg_curpm(Re2) \ |
| 3559 | if (reginfo->info_aux_eval) { \ |
| 3560 | (void)ReREFCNT_inc(Re2); \ |
| 3561 | ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \ |
| 3562 | PM_SETRE((PL_reg_curpm), (Re2)); \ |
| 3563 | } |
| 3564 | |
| 3565 | |
| 3566 | /* |
| 3567 | - regtry - try match at specific point |
| 3568 | */ |
| 3569 | STATIC bool /* 0 failure, 1 success */ |
| 3570 | S_regtry(pTHX_ regmatch_info *reginfo, char **startposp) |
| 3571 | { |
| 3572 | CHECKPOINT lastcp; |
| 3573 | REGEXP *const rx = reginfo->prog; |
| 3574 | regexp *const prog = ReANY(rx); |
| 3575 | SSize_t result; |
| 3576 | #ifdef DEBUGGING |
| 3577 | U32 depth = 0; /* used by REGCP_SET */ |
| 3578 | #endif |
| 3579 | RXi_GET_DECL(prog,progi); |
| 3580 | GET_RE_DEBUG_FLAGS_DECL; |
| 3581 | |
| 3582 | PERL_ARGS_ASSERT_REGTRY; |
| 3583 | |
| 3584 | reginfo->cutpoint=NULL; |
| 3585 | |
| 3586 | prog->offs[0].start = *startposp - reginfo->strbeg; |
| 3587 | prog->lastparen = 0; |
| 3588 | prog->lastcloseparen = 0; |
| 3589 | |
| 3590 | /* XXXX What this code is doing here?!!! There should be no need |
| 3591 | to do this again and again, prog->lastparen should take care of |
| 3592 | this! --ilya*/ |
| 3593 | |
| 3594 | /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code. |
| 3595 | * Actually, the code in regcppop() (which Ilya may be meaning by |
| 3596 | * prog->lastparen), is not needed at all by the test suite |
| 3597 | * (op/regexp, op/pat, op/split), but that code is needed otherwise |
| 3598 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ |
| 3599 | * Meanwhile, this code *is* needed for the |
| 3600 | * above-mentioned test suite tests to succeed. The common theme |
| 3601 | * on those tests seems to be returning null fields from matches. |
| 3602 | * --jhi updated by dapm */ |
| 3603 | #if 1 |
| 3604 | if (prog->nparens) { |
| 3605 | regexp_paren_pair *pp = prog->offs; |
| 3606 | I32 i; |
| 3607 | for (i = prog->nparens; i > (I32)prog->lastparen; i--) { |
| 3608 | ++pp; |
| 3609 | pp->start = -1; |
| 3610 | pp->end = -1; |
| 3611 | } |
| 3612 | } |
| 3613 | #endif |
| 3614 | REGCP_SET(lastcp); |
| 3615 | result = regmatch(reginfo, *startposp, progi->program + 1); |
| 3616 | if (result != -1) { |
| 3617 | prog->offs[0].end = result; |
| 3618 | return 1; |
| 3619 | } |
| 3620 | if (reginfo->cutpoint) |
| 3621 | *startposp= reginfo->cutpoint; |
| 3622 | REGCP_UNWIND(lastcp); |
| 3623 | return 0; |
| 3624 | } |
| 3625 | |
| 3626 | |
| 3627 | #define sayYES goto yes |
| 3628 | #define sayNO goto no |
| 3629 | #define sayNO_SILENT goto no_silent |
| 3630 | |
| 3631 | /* we dont use STMT_START/END here because it leads to |
| 3632 | "unreachable code" warnings, which are bogus, but distracting. */ |
| 3633 | #define CACHEsayNO \ |
| 3634 | if (ST.cache_mask) \ |
| 3635 | reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \ |
| 3636 | sayNO |
| 3637 | |
| 3638 | /* this is used to determine how far from the left messages like |
| 3639 | 'failed...' are printed in regexec.c. It should be set such that |
| 3640 | messages are inline with the regop output that created them. |
| 3641 | */ |
| 3642 | #define REPORT_CODE_OFF 29 |
| 3643 | #define INDENT_CHARS(depth) ((int)(depth) % 20) |
| 3644 | #ifdef DEBUGGING |
| 3645 | int |
| 3646 | Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...) |
| 3647 | { |
| 3648 | va_list ap; |
| 3649 | int result; |
| 3650 | PerlIO *f= Perl_debug_log; |
| 3651 | PERL_ARGS_ASSERT_RE_EXEC_INDENTF; |
| 3652 | va_start(ap, depth); |
| 3653 | PerlIO_printf(f, "%*s|%4"UVuf"| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" ); |
| 3654 | result = PerlIO_vprintf(f, fmt, ap); |
| 3655 | va_end(ap); |
| 3656 | return result; |
| 3657 | } |
| 3658 | #endif /* DEBUGGING */ |
| 3659 | |
| 3660 | |
| 3661 | #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */ |
| 3662 | #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */ |
| 3663 | #define CHRTEST_NOT_A_CP_1 -999 |
| 3664 | #define CHRTEST_NOT_A_CP_2 -998 |
| 3665 | |
| 3666 | /* grab a new slab and return the first slot in it */ |
| 3667 | |
| 3668 | STATIC regmatch_state * |
| 3669 | S_push_slab(pTHX) |
| 3670 | { |
| 3671 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
| 3672 | dMY_CXT; |
| 3673 | #endif |
| 3674 | regmatch_slab *s = PL_regmatch_slab->next; |
| 3675 | if (!s) { |
| 3676 | Newx(s, 1, regmatch_slab); |
| 3677 | s->prev = PL_regmatch_slab; |
| 3678 | s->next = NULL; |
| 3679 | PL_regmatch_slab->next = s; |
| 3680 | } |
| 3681 | PL_regmatch_slab = s; |
| 3682 | return SLAB_FIRST(s); |
| 3683 | } |
| 3684 | |
| 3685 | |
| 3686 | /* push a new state then goto it */ |
| 3687 | |
| 3688 | #define PUSH_STATE_GOTO(state, node, input) \ |
| 3689 | pushinput = input; \ |
| 3690 | scan = node; \ |
| 3691 | st->resume_state = state; \ |
| 3692 | goto push_state; |
| 3693 | |
| 3694 | /* push a new state with success backtracking, then goto it */ |
| 3695 | |
| 3696 | #define PUSH_YES_STATE_GOTO(state, node, input) \ |
| 3697 | pushinput = input; \ |
| 3698 | scan = node; \ |
| 3699 | st->resume_state = state; \ |
| 3700 | goto push_yes_state; |
| 3701 | |
| 3702 | |
| 3703 | |
| 3704 | |
| 3705 | /* |
| 3706 | |
| 3707 | regmatch() - main matching routine |
| 3708 | |
| 3709 | This is basically one big switch statement in a loop. We execute an op, |
| 3710 | set 'next' to point the next op, and continue. If we come to a point which |
| 3711 | we may need to backtrack to on failure such as (A|B|C), we push a |
| 3712 | backtrack state onto the backtrack stack. On failure, we pop the top |
| 3713 | state, and re-enter the loop at the state indicated. If there are no more |
| 3714 | states to pop, we return failure. |
| 3715 | |
| 3716 | Sometimes we also need to backtrack on success; for example /A+/, where |
| 3717 | after successfully matching one A, we need to go back and try to |
| 3718 | match another one; similarly for lookahead assertions: if the assertion |
| 3719 | completes successfully, we backtrack to the state just before the assertion |
| 3720 | and then carry on. In these cases, the pushed state is marked as |
| 3721 | 'backtrack on success too'. This marking is in fact done by a chain of |
| 3722 | pointers, each pointing to the previous 'yes' state. On success, we pop to |
| 3723 | the nearest yes state, discarding any intermediate failure-only states. |
| 3724 | Sometimes a yes state is pushed just to force some cleanup code to be |
| 3725 | called at the end of a successful match or submatch; e.g. (??{$re}) uses |
| 3726 | it to free the inner regex. |
| 3727 | |
| 3728 | Note that failure backtracking rewinds the cursor position, while |
| 3729 | success backtracking leaves it alone. |
| 3730 | |
| 3731 | A pattern is complete when the END op is executed, while a subpattern |
| 3732 | such as (?=foo) is complete when the SUCCESS op is executed. Both of these |
| 3733 | ops trigger the "pop to last yes state if any, otherwise return true" |
| 3734 | behaviour. |
| 3735 | |
| 3736 | A common convention in this function is to use A and B to refer to the two |
| 3737 | subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is |
| 3738 | the subpattern to be matched possibly multiple times, while B is the entire |
| 3739 | rest of the pattern. Variable and state names reflect this convention. |
| 3740 | |
| 3741 | The states in the main switch are the union of ops and failure/success of |
| 3742 | substates associated with with that op. For example, IFMATCH is the op |
| 3743 | that does lookahead assertions /(?=A)B/ and so the IFMATCH state means |
| 3744 | 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just |
| 3745 | successfully matched A and IFMATCH_A_fail is a state saying that we have |
| 3746 | just failed to match A. Resume states always come in pairs. The backtrack |
| 3747 | state we push is marked as 'IFMATCH_A', but when that is popped, we resume |
| 3748 | at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking |
| 3749 | on success or failure. |
| 3750 | |
| 3751 | The struct that holds a backtracking state is actually a big union, with |
| 3752 | one variant for each major type of op. The variable st points to the |
| 3753 | top-most backtrack struct. To make the code clearer, within each |
| 3754 | block of code we #define ST to alias the relevant union. |
| 3755 | |
| 3756 | Here's a concrete example of a (vastly oversimplified) IFMATCH |
| 3757 | implementation: |
| 3758 | |
| 3759 | switch (state) { |
| 3760 | .... |
| 3761 | |
| 3762 | #define ST st->u.ifmatch |
| 3763 | |
| 3764 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B |
| 3765 | ST.foo = ...; // some state we wish to save |
| 3766 | ... |
| 3767 | // push a yes backtrack state with a resume value of |
| 3768 | // IFMATCH_A/IFMATCH_A_fail, then continue execution at the |
| 3769 | // first node of A: |
| 3770 | PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput); |
| 3771 | // NOTREACHED |
| 3772 | |
| 3773 | case IFMATCH_A: // we have successfully executed A; now continue with B |
| 3774 | next = B; |
| 3775 | bar = ST.foo; // do something with the preserved value |
| 3776 | break; |
| 3777 | |
| 3778 | case IFMATCH_A_fail: // A failed, so the assertion failed |
| 3779 | ...; // do some housekeeping, then ... |
| 3780 | sayNO; // propagate the failure |
| 3781 | |
| 3782 | #undef ST |
| 3783 | |
| 3784 | ... |
| 3785 | } |
| 3786 | |
| 3787 | For any old-timers reading this who are familiar with the old recursive |
| 3788 | approach, the code above is equivalent to: |
| 3789 | |
| 3790 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B |
| 3791 | { |
| 3792 | int foo = ... |
| 3793 | ... |
| 3794 | if (regmatch(A)) { |
| 3795 | next = B; |
| 3796 | bar = foo; |
| 3797 | break; |
| 3798 | } |
| 3799 | ...; // do some housekeeping, then ... |
| 3800 | sayNO; // propagate the failure |
| 3801 | } |
| 3802 | |
| 3803 | The topmost backtrack state, pointed to by st, is usually free. If you |
| 3804 | want to claim it, populate any ST.foo fields in it with values you wish to |
| 3805 | save, then do one of |
| 3806 | |
| 3807 | PUSH_STATE_GOTO(resume_state, node, newinput); |
| 3808 | PUSH_YES_STATE_GOTO(resume_state, node, newinput); |
| 3809 | |
| 3810 | which sets that backtrack state's resume value to 'resume_state', pushes a |
| 3811 | new free entry to the top of the backtrack stack, then goes to 'node'. |
| 3812 | On backtracking, the free slot is popped, and the saved state becomes the |
| 3813 | new free state. An ST.foo field in this new top state can be temporarily |
| 3814 | accessed to retrieve values, but once the main loop is re-entered, it |
| 3815 | becomes available for reuse. |
| 3816 | |
| 3817 | Note that the depth of the backtrack stack constantly increases during the |
| 3818 | left-to-right execution of the pattern, rather than going up and down with |
| 3819 | the pattern nesting. For example the stack is at its maximum at Z at the |
| 3820 | end of the pattern, rather than at X in the following: |
| 3821 | |
| 3822 | /(((X)+)+)+....(Y)+....Z/ |
| 3823 | |
| 3824 | The only exceptions to this are lookahead/behind assertions and the cut, |
| 3825 | (?>A), which pop all the backtrack states associated with A before |
| 3826 | continuing. |
| 3827 | |
| 3828 | Backtrack state structs are allocated in slabs of about 4K in size. |
| 3829 | PL_regmatch_state and st always point to the currently active state, |
| 3830 | and PL_regmatch_slab points to the slab currently containing |
| 3831 | PL_regmatch_state. The first time regmatch() is called, the first slab is |
| 3832 | allocated, and is never freed until interpreter destruction. When the slab |
| 3833 | is full, a new one is allocated and chained to the end. At exit from |
| 3834 | regmatch(), slabs allocated since entry are freed. |
| 3835 | |
| 3836 | */ |
| 3837 | |
| 3838 | |
| 3839 | #define DEBUG_STATE_pp(pp) \ |
| 3840 | DEBUG_STATE_r({ \ |
| 3841 | DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \ |
| 3842 | Perl_re_printf( aTHX_ \ |
| 3843 | "%*s" pp " %s%s%s%s%s\n", \ |
| 3844 | INDENT_CHARS(depth), "", \ |
| 3845 | PL_reg_name[st->resume_state], \ |
| 3846 | ((st==yes_state||st==mark_state) ? "[" : ""), \ |
| 3847 | ((st==yes_state) ? "Y" : ""), \ |
| 3848 | ((st==mark_state) ? "M" : ""), \ |
| 3849 | ((st==yes_state||st==mark_state) ? "]" : "") \ |
| 3850 | ); \ |
| 3851 | }); |
| 3852 | |
| 3853 | |
| 3854 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) |
| 3855 | |
| 3856 | #ifdef DEBUGGING |
| 3857 | |
| 3858 | STATIC void |
| 3859 | S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target, |
| 3860 | const char *start, const char *end, const char *blurb) |
| 3861 | { |
| 3862 | const bool utf8_pat = RX_UTF8(prog) ? 1 : 0; |
| 3863 | |
| 3864 | PERL_ARGS_ASSERT_DEBUG_START_MATCH; |
| 3865 | |
| 3866 | if (!PL_colorset) |
| 3867 | reginitcolors(); |
| 3868 | { |
| 3869 | RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0), |
| 3870 | RX_PRECOMP_const(prog), RX_PRELEN(prog), 60); |
| 3871 | |
| 3872 | RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1), |
| 3873 | start, end - start, 60); |
| 3874 | |
| 3875 | Perl_re_printf( aTHX_ |
| 3876 | "%s%s REx%s %s against %s\n", |
| 3877 | PL_colors[4], blurb, PL_colors[5], s0, s1); |
| 3878 | |
| 3879 | if (utf8_target||utf8_pat) |
| 3880 | Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n", |
| 3881 | utf8_pat ? "pattern" : "", |
| 3882 | utf8_pat && utf8_target ? " and " : "", |
| 3883 | utf8_target ? "string" : "" |
| 3884 | ); |
| 3885 | } |
| 3886 | } |
| 3887 | |
| 3888 | STATIC void |
| 3889 | S_dump_exec_pos(pTHX_ const char *locinput, |
| 3890 | const regnode *scan, |
| 3891 | const char *loc_regeol, |
| 3892 | const char *loc_bostr, |
| 3893 | const char *loc_reg_starttry, |
| 3894 | const bool utf8_target, |
| 3895 | const U32 depth |
| 3896 | ) |
| 3897 | { |
| 3898 | const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; |
| 3899 | const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ |
| 3900 | int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); |
| 3901 | /* The part of the string before starttry has one color |
| 3902 | (pref0_len chars), between starttry and current |
| 3903 | position another one (pref_len - pref0_len chars), |
| 3904 | after the current position the third one. |
| 3905 | We assume that pref0_len <= pref_len, otherwise we |
| 3906 | decrease pref0_len. */ |
| 3907 | int pref_len = (locinput - loc_bostr) > (5 + taill) - l |
| 3908 | ? (5 + taill) - l : locinput - loc_bostr; |
| 3909 | int pref0_len; |
| 3910 | |
| 3911 | PERL_ARGS_ASSERT_DUMP_EXEC_POS; |
| 3912 | |
| 3913 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) |
| 3914 | pref_len++; |
| 3915 | pref0_len = pref_len - (locinput - loc_reg_starttry); |
| 3916 | if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) |
| 3917 | l = ( loc_regeol - locinput > (5 + taill) - pref_len |
| 3918 | ? (5 + taill) - pref_len : loc_regeol - locinput); |
| 3919 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) |
| 3920 | l--; |
| 3921 | if (pref0_len < 0) |
| 3922 | pref0_len = 0; |
| 3923 | if (pref0_len > pref_len) |
| 3924 | pref0_len = pref_len; |
| 3925 | { |
| 3926 | const int is_uni = utf8_target ? 1 : 0; |
| 3927 | |
| 3928 | RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0), |
| 3929 | (locinput - pref_len),pref0_len, 60, 4, 5); |
| 3930 | |
| 3931 | RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1), |
| 3932 | (locinput - pref_len + pref0_len), |
| 3933 | pref_len - pref0_len, 60, 2, 3); |
| 3934 | |
| 3935 | RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), |
| 3936 | locinput, loc_regeol - locinput, 10, 0, 1); |
| 3937 | |
| 3938 | const STRLEN tlen=len0+len1+len2; |
| 3939 | Perl_re_printf( aTHX_ |
| 3940 | "%4"IVdf" <%.*s%.*s%s%.*s>%*s|%4u| ", |
| 3941 | (IV)(locinput - loc_bostr), |
| 3942 | len0, s0, |
| 3943 | len1, s1, |
| 3944 | (docolor ? "" : "> <"), |
| 3945 | len2, s2, |
| 3946 | (int)(tlen > 19 ? 0 : 19 - tlen), |
| 3947 | "", |
| 3948 | depth); |
| 3949 | } |
| 3950 | } |
| 3951 | |
| 3952 | #endif |
| 3953 | |
| 3954 | /* reg_check_named_buff_matched() |
| 3955 | * Checks to see if a named buffer has matched. The data array of |
| 3956 | * buffer numbers corresponding to the buffer is expected to reside |
| 3957 | * in the regexp->data->data array in the slot stored in the ARG() of |
| 3958 | * node involved. Note that this routine doesn't actually care about the |
| 3959 | * name, that information is not preserved from compilation to execution. |
| 3960 | * Returns the index of the leftmost defined buffer with the given name |
| 3961 | * or 0 if non of the buffers matched. |
| 3962 | */ |
| 3963 | STATIC I32 |
| 3964 | S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan) |
| 3965 | { |
| 3966 | I32 n; |
| 3967 | RXi_GET_DECL(rex,rexi); |
| 3968 | SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 3969 | I32 *nums=(I32*)SvPVX(sv_dat); |
| 3970 | |
| 3971 | PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED; |
| 3972 | |
| 3973 | for ( n=0; n<SvIVX(sv_dat); n++ ) { |
| 3974 | if ((I32)rex->lastparen >= nums[n] && |
| 3975 | rex->offs[nums[n]].end != -1) |
| 3976 | { |
| 3977 | return nums[n]; |
| 3978 | } |
| 3979 | } |
| 3980 | return 0; |
| 3981 | } |
| 3982 | |
| 3983 | |
| 3984 | static bool |
| 3985 | S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, |
| 3986 | U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo) |
| 3987 | { |
| 3988 | /* This function determines if there are one or two characters that match |
| 3989 | * the first character of the passed-in EXACTish node <text_node>, and if |
| 3990 | * so, returns them in the passed-in pointers. |
| 3991 | * |
| 3992 | * If it determines that no possible character in the target string can |
| 3993 | * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if |
| 3994 | * the first character in <text_node> requires UTF-8 to represent, and the |
| 3995 | * target string isn't in UTF-8.) |
| 3996 | * |
| 3997 | * If there are more than two characters that could match the beginning of |
| 3998 | * <text_node>, or if more context is required to determine a match or not, |
| 3999 | * it sets both *<c1p> and *<c2p> to CHRTEST_VOID. |
| 4000 | * |
| 4001 | * The motiviation behind this function is to allow the caller to set up |
| 4002 | * tight loops for matching. If <text_node> is of type EXACT, there is |
| 4003 | * only one possible character that can match its first character, and so |
| 4004 | * the situation is quite simple. But things get much more complicated if |
| 4005 | * folding is involved. It may be that the first character of an EXACTFish |
| 4006 | * node doesn't participate in any possible fold, e.g., punctuation, so it |
| 4007 | * can be matched only by itself. The vast majority of characters that are |
| 4008 | * in folds match just two things, their lower and upper-case equivalents. |
| 4009 | * But not all are like that; some have multiple possible matches, or match |
| 4010 | * sequences of more than one character. This function sorts all that out. |
| 4011 | * |
| 4012 | * Consider the patterns A*B or A*?B where A and B are arbitrary. In a |
| 4013 | * loop of trying to match A*, we know we can't exit where the thing |
| 4014 | * following it isn't a B. And something can't be a B unless it is the |
| 4015 | * beginning of B. By putting a quick test for that beginning in a tight |
| 4016 | * loop, we can rule out things that can't possibly be B without having to |
| 4017 | * break out of the loop, thus avoiding work. Similarly, if A is a single |
| 4018 | * character, we can make a tight loop matching A*, using the outputs of |
| 4019 | * this function. |
| 4020 | * |
| 4021 | * If the target string to match isn't in UTF-8, and there aren't |
| 4022 | * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to |
| 4023 | * the one or two possible octets (which are characters in this situation) |
| 4024 | * that can match. In all cases, if there is only one character that can |
| 4025 | * match, *<c1p> and *<c2p> will be identical. |
| 4026 | * |
| 4027 | * If the target string is in UTF-8, the buffers pointed to by <c1_utf8> |
| 4028 | * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that |
| 4029 | * can match the beginning of <text_node>. They should be declared with at |
| 4030 | * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is |
| 4031 | * undefined what these contain.) If one or both of the buffers are |
| 4032 | * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the |
| 4033 | * corresponding invariant. If variant, the corresponding *<c1p> and/or |
| 4034 | * *<c2p> will be set to a negative number(s) that shouldn't match any code |
| 4035 | * point (unless inappropriately coerced to unsigned). *<c1p> will equal |
| 4036 | * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */ |
| 4037 | |
| 4038 | const bool utf8_target = reginfo->is_utf8_target; |
| 4039 | |
| 4040 | UV c1 = (UV)CHRTEST_NOT_A_CP_1; |
| 4041 | UV c2 = (UV)CHRTEST_NOT_A_CP_2; |
| 4042 | bool use_chrtest_void = FALSE; |
| 4043 | const bool is_utf8_pat = reginfo->is_utf8_pat; |
| 4044 | |
| 4045 | /* Used when we have both utf8 input and utf8 output, to avoid converting |
| 4046 | * to/from code points */ |
| 4047 | bool utf8_has_been_setup = FALSE; |
| 4048 | |
| 4049 | dVAR; |
| 4050 | |
| 4051 | U8 *pat = (U8*)STRING(text_node); |
| 4052 | U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' }; |
| 4053 | |
| 4054 | if (OP(text_node) == EXACT || OP(text_node) == EXACTL) { |
| 4055 | |
| 4056 | /* In an exact node, only one thing can be matched, that first |
| 4057 | * character. If both the pat and the target are UTF-8, we can just |
| 4058 | * copy the input to the output, avoiding finding the code point of |
| 4059 | * that character */ |
| 4060 | if (!is_utf8_pat) { |
| 4061 | c2 = c1 = *pat; |
| 4062 | } |
| 4063 | else if (utf8_target) { |
| 4064 | Copy(pat, c1_utf8, UTF8SKIP(pat), U8); |
| 4065 | Copy(pat, c2_utf8, UTF8SKIP(pat), U8); |
| 4066 | utf8_has_been_setup = TRUE; |
| 4067 | } |
| 4068 | else { |
| 4069 | c2 = c1 = valid_utf8_to_uvchr(pat, NULL); |
| 4070 | } |
| 4071 | } |
| 4072 | else { /* an EXACTFish node */ |
| 4073 | U8 *pat_end = pat + STR_LEN(text_node); |
| 4074 | |
| 4075 | /* An EXACTFL node has at least some characters unfolded, because what |
| 4076 | * they match is not known until now. So, now is the time to fold |
| 4077 | * the first few of them, as many as are needed to determine 'c1' and |
| 4078 | * 'c2' later in the routine. If the pattern isn't UTF-8, we only need |
| 4079 | * to fold if in a UTF-8 locale, and then only the Sharp S; everything |
| 4080 | * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we |
| 4081 | * need to fold as many characters as a single character can fold to, |
| 4082 | * so that later we can check if the first ones are such a multi-char |
| 4083 | * fold. But, in such a pattern only locale-problematic characters |
| 4084 | * aren't folded, so we can skip this completely if the first character |
| 4085 | * in the node isn't one of the tricky ones */ |
| 4086 | if (OP(text_node) == EXACTFL) { |
| 4087 | |
| 4088 | if (! is_utf8_pat) { |
| 4089 | if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S) |
| 4090 | { |
| 4091 | folded[0] = folded[1] = 's'; |
| 4092 | pat = folded; |
| 4093 | pat_end = folded + 2; |
| 4094 | } |
| 4095 | } |
| 4096 | else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) { |
| 4097 | U8 *s = pat; |
| 4098 | U8 *d = folded; |
| 4099 | int i; |
| 4100 | |
| 4101 | for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) { |
| 4102 | if (isASCII(*s)) { |
| 4103 | *(d++) = (U8) toFOLD_LC(*s); |
| 4104 | s++; |
| 4105 | } |
| 4106 | else { |
| 4107 | STRLEN len; |
| 4108 | _to_utf8_fold_flags(s, |
| 4109 | d, |
| 4110 | &len, |
| 4111 | FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE); |
| 4112 | d += len; |
| 4113 | s += UTF8SKIP(s); |
| 4114 | } |
| 4115 | } |
| 4116 | |
| 4117 | pat = folded; |
| 4118 | pat_end = d; |
| 4119 | } |
| 4120 | } |
| 4121 | |
| 4122 | if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end)) |
| 4123 | || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end))) |
| 4124 | { |
| 4125 | /* Multi-character folds require more context to sort out. Also |
| 4126 | * PL_utf8_foldclosures used below doesn't handle them, so have to |
| 4127 | * be handled outside this routine */ |
| 4128 | use_chrtest_void = TRUE; |
| 4129 | } |
| 4130 | else { /* an EXACTFish node which doesn't begin with a multi-char fold */ |
| 4131 | c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat; |
| 4132 | if (c1 > 255) { |
| 4133 | /* Load the folds hash, if not already done */ |
| 4134 | SV** listp; |
| 4135 | if (! PL_utf8_foldclosures) { |
| 4136 | _load_PL_utf8_foldclosures(); |
| 4137 | } |
| 4138 | |
| 4139 | /* The fold closures data structure is a hash with the keys |
| 4140 | * being the UTF-8 of every character that is folded to, like |
| 4141 | * 'k', and the values each an array of all code points that |
| 4142 | * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ]. |
| 4143 | * Multi-character folds are not included */ |
| 4144 | if ((! (listp = hv_fetch(PL_utf8_foldclosures, |
| 4145 | (char *) pat, |
| 4146 | UTF8SKIP(pat), |
| 4147 | FALSE)))) |
| 4148 | { |
| 4149 | /* Not found in the hash, therefore there are no folds |
| 4150 | * containing it, so there is only a single character that |
| 4151 | * could match */ |
| 4152 | c2 = c1; |
| 4153 | } |
| 4154 | else { /* Does participate in folds */ |
| 4155 | AV* list = (AV*) *listp; |
| 4156 | if (av_tindex_nomg(list) != 1) { |
| 4157 | |
| 4158 | /* If there aren't exactly two folds to this, it is |
| 4159 | * outside the scope of this function */ |
| 4160 | use_chrtest_void = TRUE; |
| 4161 | } |
| 4162 | else { /* There are two. Get them */ |
| 4163 | SV** c_p = av_fetch(list, 0, FALSE); |
| 4164 | if (c_p == NULL) { |
| 4165 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); |
| 4166 | } |
| 4167 | c1 = SvUV(*c_p); |
| 4168 | |
| 4169 | c_p = av_fetch(list, 1, FALSE); |
| 4170 | if (c_p == NULL) { |
| 4171 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); |
| 4172 | } |
| 4173 | c2 = SvUV(*c_p); |
| 4174 | |
| 4175 | /* Folds that cross the 255/256 boundary are forbidden |
| 4176 | * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and |
| 4177 | * one is ASCIII. Since the pattern character is above |
| 4178 | * 255, and its only other match is below 256, the only |
| 4179 | * legal match will be to itself. We have thrown away |
| 4180 | * the original, so have to compute which is the one |
| 4181 | * above 255. */ |
| 4182 | if ((c1 < 256) != (c2 < 256)) { |
| 4183 | if ((OP(text_node) == EXACTFL |
| 4184 | && ! IN_UTF8_CTYPE_LOCALE) |
| 4185 | || ((OP(text_node) == EXACTFA |
| 4186 | || OP(text_node) == EXACTFA_NO_TRIE) |
| 4187 | && (isASCII(c1) || isASCII(c2)))) |
| 4188 | { |
| 4189 | if (c1 < 256) { |
| 4190 | c1 = c2; |
| 4191 | } |
| 4192 | else { |
| 4193 | c2 = c1; |
| 4194 | } |
| 4195 | } |
| 4196 | } |
| 4197 | } |
| 4198 | } |
| 4199 | } |
| 4200 | else /* Here, c1 is <= 255 */ |
| 4201 | if (utf8_target |
| 4202 | && HAS_NONLATIN1_FOLD_CLOSURE(c1) |
| 4203 | && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE)) |
| 4204 | && ((OP(text_node) != EXACTFA |
| 4205 | && OP(text_node) != EXACTFA_NO_TRIE) |
| 4206 | || ! isASCII(c1))) |
| 4207 | { |
| 4208 | /* Here, there could be something above Latin1 in the target |
| 4209 | * which folds to this character in the pattern. All such |
| 4210 | * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more |
| 4211 | * than two characters involved in their folds, so are outside |
| 4212 | * the scope of this function */ |
| 4213 | if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { |
| 4214 | c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; |
| 4215 | } |
| 4216 | else { |
| 4217 | use_chrtest_void = TRUE; |
| 4218 | } |
| 4219 | } |
| 4220 | else { /* Here nothing above Latin1 can fold to the pattern |
| 4221 | character */ |
| 4222 | switch (OP(text_node)) { |
| 4223 | |
| 4224 | case EXACTFL: /* /l rules */ |
| 4225 | c2 = PL_fold_locale[c1]; |
| 4226 | break; |
| 4227 | |
| 4228 | case EXACTF: /* This node only generated for non-utf8 |
| 4229 | patterns */ |
| 4230 | assert(! is_utf8_pat); |
| 4231 | if (! utf8_target) { /* /d rules */ |
| 4232 | c2 = PL_fold[c1]; |
| 4233 | break; |
| 4234 | } |
| 4235 | /* FALLTHROUGH */ |
| 4236 | /* /u rules for all these. This happens to work for |
| 4237 | * EXACTFA as nothing in Latin1 folds to ASCII */ |
| 4238 | case EXACTFA_NO_TRIE: /* This node only generated for |
| 4239 | non-utf8 patterns */ |
| 4240 | assert(! is_utf8_pat); |
| 4241 | /* FALLTHROUGH */ |
| 4242 | case EXACTFA: |
| 4243 | case EXACTFU_SS: |
| 4244 | case EXACTFU: |
| 4245 | c2 = PL_fold_latin1[c1]; |
| 4246 | break; |
| 4247 | |
| 4248 | default: |
| 4249 | Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node)); |
| 4250 | NOT_REACHED; /* NOTREACHED */ |
| 4251 | } |
| 4252 | } |
| 4253 | } |
| 4254 | } |
| 4255 | |
| 4256 | /* Here have figured things out. Set up the returns */ |
| 4257 | if (use_chrtest_void) { |
| 4258 | *c2p = *c1p = CHRTEST_VOID; |
| 4259 | } |
| 4260 | else if (utf8_target) { |
| 4261 | if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */ |
| 4262 | uvchr_to_utf8(c1_utf8, c1); |
| 4263 | uvchr_to_utf8(c2_utf8, c2); |
| 4264 | } |
| 4265 | |
| 4266 | /* Invariants are stored in both the utf8 and byte outputs; Use |
| 4267 | * negative numbers otherwise for the byte ones. Make sure that the |
| 4268 | * byte ones are the same iff the utf8 ones are the same */ |
| 4269 | *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1; |
| 4270 | *c2p = (UTF8_IS_INVARIANT(*c2_utf8)) |
| 4271 | ? *c2_utf8 |
| 4272 | : (c1 == c2) |
| 4273 | ? CHRTEST_NOT_A_CP_1 |
| 4274 | : CHRTEST_NOT_A_CP_2; |
| 4275 | } |
| 4276 | else if (c1 > 255) { |
| 4277 | if (c2 > 255) { /* both possibilities are above what a non-utf8 string |
| 4278 | can represent */ |
| 4279 | return FALSE; |
| 4280 | } |
| 4281 | |
| 4282 | *c1p = *c2p = c2; /* c2 is the only representable value */ |
| 4283 | } |
| 4284 | else { /* c1 is representable; see about c2 */ |
| 4285 | *c1p = c1; |
| 4286 | *c2p = (c2 < 256) ? c2 : c1; |
| 4287 | } |
| 4288 | |
| 4289 | return TRUE; |
| 4290 | } |
| 4291 | |
| 4292 | PERL_STATIC_INLINE bool |
| 4293 | S_isGCB(const GCB_enum before, const GCB_enum after) |
| 4294 | { |
| 4295 | /* returns a boolean indicating if there is a Grapheme Cluster Boundary |
| 4296 | * between the inputs. See http://www.unicode.org/reports/tr29/ */ |
| 4297 | |
| 4298 | return GCB_table[before][after]; |
| 4299 | } |
| 4300 | |
| 4301 | /* Combining marks attach to most classes that precede them, but this defines |
| 4302 | * the exceptions (from TR14) */ |
| 4303 | #define LB_CM_ATTACHES_TO(prev) ( ! ( prev == LB_EDGE \ |
| 4304 | || prev == LB_Mandatory_Break \ |
| 4305 | || prev == LB_Carriage_Return \ |
| 4306 | || prev == LB_Line_Feed \ |
| 4307 | || prev == LB_Next_Line \ |
| 4308 | || prev == LB_Space \ |
| 4309 | || prev == LB_ZWSpace)) |
| 4310 | |
| 4311 | STATIC bool |
| 4312 | S_isLB(pTHX_ LB_enum before, |
| 4313 | LB_enum after, |
| 4314 | const U8 * const strbeg, |
| 4315 | const U8 * const curpos, |
| 4316 | const U8 * const strend, |
| 4317 | const bool utf8_target) |
| 4318 | { |
| 4319 | U8 * temp_pos = (U8 *) curpos; |
| 4320 | LB_enum prev = before; |
| 4321 | |
| 4322 | /* Is the boundary between 'before' and 'after' line-breakable? |
| 4323 | * Most of this is just a table lookup of a generated table from Unicode |
| 4324 | * rules. But some rules require context to decide, and so have to be |
| 4325 | * implemented in code */ |
| 4326 | |
| 4327 | PERL_ARGS_ASSERT_ISLB; |
| 4328 | |
| 4329 | /* Rule numbers in the comments below are as of Unicode 8.0 */ |
| 4330 | |
| 4331 | redo: |
| 4332 | before = prev; |
| 4333 | switch (LB_table[before][after]) { |
| 4334 | case LB_BREAKABLE: |
| 4335 | return TRUE; |
| 4336 | |
| 4337 | case LB_NOBREAK: |
| 4338 | case LB_NOBREAK_EVEN_WITH_SP_BETWEEN: |
| 4339 | return FALSE; |
| 4340 | |
| 4341 | case LB_SP_foo + LB_BREAKABLE: |
| 4342 | case LB_SP_foo + LB_NOBREAK: |
| 4343 | case LB_SP_foo + LB_NOBREAK_EVEN_WITH_SP_BETWEEN: |
| 4344 | |
| 4345 | /* When we have something following a SP, we have to look at the |
| 4346 | * context in order to know what to do. |
| 4347 | * |
| 4348 | * SP SP should not reach here because LB7: Do not break before |
| 4349 | * spaces. (For two spaces in a row there is nothing that |
| 4350 | * overrides that) */ |
| 4351 | assert(after != LB_Space); |
| 4352 | |
| 4353 | /* Here we have a space followed by a non-space. Mostly this is a |
| 4354 | * case of LB18: "Break after spaces". But there are complications |
| 4355 | * as the handling of spaces is somewhat tricky. They are in a |
| 4356 | * number of rules, which have to be applied in priority order, but |
| 4357 | * something earlier in the string can cause a rule to be skipped |
| 4358 | * and a lower priority rule invoked. A prime example is LB7 which |
| 4359 | * says don't break before a space. But rule LB8 (lower priority) |
| 4360 | * says that the first break opportunity after a ZW is after any |
| 4361 | * span of spaces immediately after it. If a ZW comes before a SP |
| 4362 | * in the input, rule LB8 applies, and not LB7. Other such rules |
| 4363 | * involve combining marks which are rules 9 and 10, but they may |
| 4364 | * override higher priority rules if they come earlier in the |
| 4365 | * string. Since we're doing random access into the middle of the |
| 4366 | * string, we have to look for rules that should get applied based |
| 4367 | * on both string position and priority. Combining marks do not |
| 4368 | * attach to either ZW nor SP, so we don't have to consider them |
| 4369 | * until later. |
| 4370 | * |
| 4371 | * To check for LB8, we have to find the first non-space character |
| 4372 | * before this span of spaces */ |
| 4373 | do { |
| 4374 | prev = backup_one_LB(strbeg, &temp_pos, utf8_target); |
| 4375 | } |
| 4376 | while (prev == LB_Space); |
| 4377 | |
| 4378 | /* LB8 Break before any character following a zero-width space, |
| 4379 | * even if one or more spaces intervene. |
| 4380 | * ZW SP* ÷ |
| 4381 | * So if we have a ZW just before this span, and to get here this |
| 4382 | * is the final space in the span. */ |
| 4383 | if (prev == LB_ZWSpace) { |
| 4384 | return TRUE; |
| 4385 | } |
| 4386 | |
| 4387 | /* Here, not ZW SP+. There are several rules that have higher |
| 4388 | * priority than LB18 and can be resolved now, as they don't depend |
| 4389 | * on anything earlier in the string (except ZW, which we have |
| 4390 | * already handled). One of these rules is LB11 Do not break |
| 4391 | * before Word joiner, but we have specially encoded that in the |
| 4392 | * lookup table so it is caught by the single test below which |
| 4393 | * catches the other ones. */ |
| 4394 | if (LB_table[LB_Space][after] - LB_SP_foo |
| 4395 | == LB_NOBREAK_EVEN_WITH_SP_BETWEEN) |
| 4396 | { |
| 4397 | return FALSE; |
| 4398 | } |
| 4399 | |
| 4400 | /* If we get here, we have to XXX consider combining marks. */ |
| 4401 | if (prev == LB_Combining_Mark) { |
| 4402 | |
| 4403 | /* What happens with these depends on the character they |
| 4404 | * follow. */ |
| 4405 | do { |
| 4406 | prev = backup_one_LB(strbeg, &temp_pos, utf8_target); |
| 4407 | } |
| 4408 | while (prev == LB_Combining_Mark); |
| 4409 | |
| 4410 | /* Most times these attach to and inherit the characteristics |
| 4411 | * of that character, but not always, and when not, they are to |
| 4412 | * be treated as AL by rule LB10. */ |
| 4413 | if (! LB_CM_ATTACHES_TO(prev)) { |
| 4414 | prev = LB_Alphabetic; |
| 4415 | } |
| 4416 | } |
| 4417 | |
| 4418 | /* Here, we have the character preceding the span of spaces all set |
| 4419 | * up. We follow LB18: "Break after spaces" unless the table shows |
| 4420 | * that is overriden */ |
| 4421 | return LB_table[prev][after] != LB_NOBREAK_EVEN_WITH_SP_BETWEEN; |
| 4422 | |
| 4423 | case LB_CM_foo: |
| 4424 | |
| 4425 | /* We don't know how to treat the CM except by looking at the first |
| 4426 | * non-CM character preceding it */ |
| 4427 | do { |
| 4428 | prev = backup_one_LB(strbeg, &temp_pos, utf8_target); |
| 4429 | } |
| 4430 | while (prev == LB_Combining_Mark); |
| 4431 | |
| 4432 | /* Here, 'prev' is that first earlier non-CM character. If the CM |
| 4433 | * attatches to it, then it inherits the behavior of 'prev'. If it |
| 4434 | * doesn't attach, it is to be treated as an AL */ |
| 4435 | if (! LB_CM_ATTACHES_TO(prev)) { |
| 4436 | prev = LB_Alphabetic; |
| 4437 | } |
| 4438 | |
| 4439 | goto redo; |
| 4440 | |
| 4441 | case LB_HY_or_BA_then_foo + LB_BREAKABLE: |
| 4442 | case LB_HY_or_BA_then_foo + LB_NOBREAK: |
| 4443 | |
| 4444 | /* LB21a Don't break after Hebrew + Hyphen. |
| 4445 | * HL (HY | BA) × */ |
| 4446 | |
| 4447 | if (backup_one_LB(strbeg, &temp_pos, utf8_target) |
| 4448 | == LB_Hebrew_Letter) |
| 4449 | { |
| 4450 | return FALSE; |
| 4451 | } |
| 4452 | |
| 4453 | return LB_table[prev][after] - LB_HY_or_BA_then_foo == LB_BREAKABLE; |
| 4454 | |
| 4455 | case LB_PR_or_PO_then_OP_or_HY + LB_BREAKABLE: |
| 4456 | case LB_PR_or_PO_then_OP_or_HY + LB_NOBREAK: |
| 4457 | |
| 4458 | /* LB25a (PR | PO) × ( OP | HY )? NU */ |
| 4459 | if (advance_one_LB(&temp_pos, strend, utf8_target) == LB_Numeric) { |
| 4460 | return FALSE; |
| 4461 | } |
| 4462 | |
| 4463 | return LB_table[prev][after] - LB_PR_or_PO_then_OP_or_HY |
| 4464 | == LB_BREAKABLE; |
| 4465 | |
| 4466 | case LB_SY_or_IS_then_various + LB_BREAKABLE: |
| 4467 | case LB_SY_or_IS_then_various + LB_NOBREAK: |
| 4468 | { |
| 4469 | /* LB25d NU (SY | IS)* × (NU | SY | IS | CL | CP ) */ |
| 4470 | |
| 4471 | LB_enum temp = prev; |
| 4472 | do { |
| 4473 | temp = backup_one_LB(strbeg, &temp_pos, utf8_target); |
| 4474 | } |
| 4475 | while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric); |
| 4476 | if (temp == LB_Numeric) { |
| 4477 | return FALSE; |
| 4478 | } |
| 4479 | |
| 4480 | return LB_table[prev][after] - LB_SY_or_IS_then_various |
| 4481 | == LB_BREAKABLE; |
| 4482 | } |
| 4483 | |
| 4484 | case LB_various_then_PO_or_PR + LB_BREAKABLE: |
| 4485 | case LB_various_then_PO_or_PR + LB_NOBREAK: |
| 4486 | { |
| 4487 | /* LB25e NU (SY | IS)* (CL | CP)? × (PO | PR) */ |
| 4488 | |
| 4489 | LB_enum temp = prev; |
| 4490 | if (temp == LB_Close_Punctuation || temp == LB_Close_Parenthesis) |
| 4491 | { |
| 4492 | temp = backup_one_LB(strbeg, &temp_pos, utf8_target); |
| 4493 | } |
| 4494 | while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric) { |
| 4495 | temp = backup_one_LB(strbeg, &temp_pos, utf8_target); |
| 4496 | } |
| 4497 | if (temp == LB_Numeric) { |
| 4498 | return FALSE; |
| 4499 | } |
| 4500 | return LB_various_then_PO_or_PR; |
| 4501 | } |
| 4502 | |
| 4503 | default: |
| 4504 | break; |
| 4505 | } |
| 4506 | |
| 4507 | #ifdef DEBUGGING |
| 4508 | Perl_re_printf( aTHX_ "Unhandled LB pair: LB_table[%d, %d] = %d\n", |
| 4509 | before, after, LB_table[before][after]); |
| 4510 | assert(0); |
| 4511 | #endif |
| 4512 | return TRUE; |
| 4513 | } |
| 4514 | |
| 4515 | STATIC LB_enum |
| 4516 | S_advance_one_LB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target) |
| 4517 | { |
| 4518 | LB_enum lb; |
| 4519 | |
| 4520 | PERL_ARGS_ASSERT_ADVANCE_ONE_LB; |
| 4521 | |
| 4522 | if (*curpos >= strend) { |
| 4523 | return LB_EDGE; |
| 4524 | } |
| 4525 | |
| 4526 | if (utf8_target) { |
| 4527 | *curpos += UTF8SKIP(*curpos); |
| 4528 | if (*curpos >= strend) { |
| 4529 | return LB_EDGE; |
| 4530 | } |
| 4531 | lb = getLB_VAL_UTF8(*curpos, strend); |
| 4532 | } |
| 4533 | else { |
| 4534 | (*curpos)++; |
| 4535 | if (*curpos >= strend) { |
| 4536 | return LB_EDGE; |
| 4537 | } |
| 4538 | lb = getLB_VAL_CP(**curpos); |
| 4539 | } |
| 4540 | |
| 4541 | return lb; |
| 4542 | } |
| 4543 | |
| 4544 | STATIC LB_enum |
| 4545 | S_backup_one_LB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target) |
| 4546 | { |
| 4547 | LB_enum lb; |
| 4548 | |
| 4549 | PERL_ARGS_ASSERT_BACKUP_ONE_LB; |
| 4550 | |
| 4551 | if (*curpos < strbeg) { |
| 4552 | return LB_EDGE; |
| 4553 | } |
| 4554 | |
| 4555 | if (utf8_target) { |
| 4556 | U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg); |
| 4557 | U8 * prev_prev_char_pos; |
| 4558 | |
| 4559 | if (! prev_char_pos) { |
| 4560 | return LB_EDGE; |
| 4561 | } |
| 4562 | |
| 4563 | if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) { |
| 4564 | lb = getLB_VAL_UTF8(prev_prev_char_pos, prev_char_pos); |
| 4565 | *curpos = prev_char_pos; |
| 4566 | prev_char_pos = prev_prev_char_pos; |
| 4567 | } |
| 4568 | else { |
| 4569 | *curpos = (U8 *) strbeg; |
| 4570 | return LB_EDGE; |
| 4571 | } |
| 4572 | } |
| 4573 | else { |
| 4574 | if (*curpos - 2 < strbeg) { |
| 4575 | *curpos = (U8 *) strbeg; |
| 4576 | return LB_EDGE; |
| 4577 | } |
| 4578 | (*curpos)--; |
| 4579 | lb = getLB_VAL_CP(*(*curpos - 1)); |
| 4580 | } |
| 4581 | |
| 4582 | return lb; |
| 4583 | } |
| 4584 | |
| 4585 | STATIC bool |
| 4586 | S_isSB(pTHX_ SB_enum before, |
| 4587 | SB_enum after, |
| 4588 | const U8 * const strbeg, |
| 4589 | const U8 * const curpos, |
| 4590 | const U8 * const strend, |
| 4591 | const bool utf8_target) |
| 4592 | { |
| 4593 | /* returns a boolean indicating if there is a Sentence Boundary Break |
| 4594 | * between the inputs. See http://www.unicode.org/reports/tr29/ */ |
| 4595 | |
| 4596 | U8 * lpos = (U8 *) curpos; |
| 4597 | bool has_para_sep = FALSE; |
| 4598 | bool has_sp = FALSE; |
| 4599 | |
| 4600 | PERL_ARGS_ASSERT_ISSB; |
| 4601 | |
| 4602 | /* Break at the start and end of text. |
| 4603 | SB1. sot ÷ |
| 4604 | SB2. ÷ eot |
| 4605 | But unstated in Unicode is don't break if the text is empty */ |
| 4606 | if (before == SB_EDGE || after == SB_EDGE) { |
| 4607 | return before != after; |
| 4608 | } |
| 4609 | |
| 4610 | /* SB 3: Do not break within CRLF. */ |
| 4611 | if (before == SB_CR && after == SB_LF) { |
| 4612 | return FALSE; |
| 4613 | } |
| 4614 | |
| 4615 | /* Break after paragraph separators. CR and LF are considered |
| 4616 | * so because Unicode views text as like word processing text where there |
| 4617 | * are no newlines except between paragraphs, and the word processor takes |
| 4618 | * care of wrapping without there being hard line-breaks in the text *./ |
| 4619 | SB4. Sep | CR | LF ÷ */ |
| 4620 | if (before == SB_Sep || before == SB_CR || before == SB_LF) { |
| 4621 | return TRUE; |
| 4622 | } |
| 4623 | |
| 4624 | /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF. |
| 4625 | * (See Section 6.2, Replacing Ignore Rules.) |
| 4626 | SB5. X (Extend | Format)* → X */ |
| 4627 | if (after == SB_Extend || after == SB_Format) { |
| 4628 | |
| 4629 | /* Implied is that the these characters attach to everything |
| 4630 | * immediately prior to them except for those separator-type |
| 4631 | * characters. And the rules earlier have already handled the case |
| 4632 | * when one of those immediately precedes the extend char */ |
| 4633 | return FALSE; |
| 4634 | } |
| 4635 | |
| 4636 | if (before == SB_Extend || before == SB_Format) { |
| 4637 | U8 * temp_pos = lpos; |
| 4638 | const SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target); |
| 4639 | if ( backup != SB_EDGE |
| 4640 | && backup != SB_Sep |
| 4641 | && backup != SB_CR |
| 4642 | && backup != SB_LF) |
| 4643 | { |
| 4644 | before = backup; |
| 4645 | lpos = temp_pos; |
| 4646 | } |
| 4647 | |
| 4648 | /* Here, both 'before' and 'backup' are these types; implied is that we |
| 4649 | * don't break between them */ |
| 4650 | if (backup == SB_Extend || backup == SB_Format) { |
| 4651 | return FALSE; |
| 4652 | } |
| 4653 | } |
| 4654 | |
| 4655 | /* Do not break after ambiguous terminators like period, if they are |
| 4656 | * immediately followed by a number or lowercase letter, if they are |
| 4657 | * between uppercase letters, if the first following letter (optionally |
| 4658 | * after certain punctuation) is lowercase, or if they are followed by |
| 4659 | * "continuation" punctuation such as comma, colon, or semicolon. For |
| 4660 | * example, a period may be an abbreviation or numeric period, and thus may |
| 4661 | * not mark the end of a sentence. |
| 4662 | |
| 4663 | * SB6. ATerm × Numeric */ |
| 4664 | if (before == SB_ATerm && after == SB_Numeric) { |
| 4665 | return FALSE; |
| 4666 | } |
| 4667 | |
| 4668 | /* SB7. (Upper | Lower) ATerm × Upper */ |
| 4669 | if (before == SB_ATerm && after == SB_Upper) { |
| 4670 | U8 * temp_pos = lpos; |
| 4671 | SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target); |
| 4672 | if (backup == SB_Upper || backup == SB_Lower) { |
| 4673 | return FALSE; |
| 4674 | } |
| 4675 | } |
| 4676 | |
| 4677 | /* The remaining rules that aren't the final one, all require an STerm or |
| 4678 | * an ATerm after having backed up over some Close* Sp*, and in one case an |
| 4679 | * optional Paragraph separator, although one rule doesn't have any Sp's in it. |
| 4680 | * So do that backup now, setting flags if either Sp or a paragraph |
| 4681 | * separator are found */ |
| 4682 | |
| 4683 | if (before == SB_Sep || before == SB_CR || before == SB_LF) { |
| 4684 | has_para_sep = TRUE; |
| 4685 | before = backup_one_SB(strbeg, &lpos, utf8_target); |
| 4686 | } |
| 4687 | |
| 4688 | if (before == SB_Sp) { |
| 4689 | has_sp = TRUE; |
| 4690 | do { |
| 4691 | before = backup_one_SB(strbeg, &lpos, utf8_target); |
| 4692 | } |
| 4693 | while (before == SB_Sp); |
| 4694 | } |
| 4695 | |
| 4696 | while (before == SB_Close) { |
| 4697 | before = backup_one_SB(strbeg, &lpos, utf8_target); |
| 4698 | } |
| 4699 | |
| 4700 | /* The next few rules apply only when the backed-up-to is an ATerm, and in |
| 4701 | * most cases an STerm */ |
| 4702 | if (before == SB_STerm || before == SB_ATerm) { |
| 4703 | |
| 4704 | /* So, here the lhs matches |
| 4705 | * (STerm | ATerm) Close* Sp* (Sep | CR | LF)? |
| 4706 | * and we have set flags if we found an Sp, or the optional Sep,CR,LF. |
| 4707 | * The rules that apply here are: |
| 4708 | * |
| 4709 | * SB8 ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR |
| 4710 | | LF | STerm | ATerm) )* Lower |
| 4711 | SB8a (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm) |
| 4712 | SB9 (STerm | ATerm) Close* × (Close | Sp | Sep | CR | LF) |
| 4713 | SB10 (STerm | ATerm) Close* Sp* × (Sp | Sep | CR | LF) |
| 4714 | SB11 (STerm | ATerm) Close* Sp* (Sep | CR | LF)? ÷ |
| 4715 | */ |
| 4716 | |
| 4717 | /* And all but SB11 forbid having seen a paragraph separator */ |
| 4718 | if (! has_para_sep) { |
| 4719 | if (before == SB_ATerm) { /* SB8 */ |
| 4720 | U8 * rpos = (U8 *) curpos; |
| 4721 | SB_enum later = after; |
| 4722 | |
| 4723 | while ( later != SB_OLetter |
| 4724 | && later != SB_Upper |
| 4725 | && later != SB_Lower |
| 4726 | && later != SB_Sep |
| 4727 | && later != SB_CR |
| 4728 | && later != SB_LF |
| 4729 | && later != SB_STerm |
| 4730 | && later != SB_ATerm |
| 4731 | && later != SB_EDGE) |
| 4732 | { |
| 4733 | later = advance_one_SB(&rpos, strend, utf8_target); |
| 4734 | } |
| 4735 | if (later == SB_Lower) { |
| 4736 | return FALSE; |
| 4737 | } |
| 4738 | } |
| 4739 | |
| 4740 | if ( after == SB_SContinue /* SB8a */ |
| 4741 | || after == SB_STerm |
| 4742 | || after == SB_ATerm) |
| 4743 | { |
| 4744 | return FALSE; |
| 4745 | } |
| 4746 | |
| 4747 | if (! has_sp) { /* SB9 applies only if there was no Sp* */ |
| 4748 | if ( after == SB_Close |
| 4749 | || after == SB_Sp |
| 4750 | || after == SB_Sep |
| 4751 | || after == SB_CR |
| 4752 | || after == SB_LF) |
| 4753 | { |
| 4754 | return FALSE; |
| 4755 | } |
| 4756 | } |
| 4757 | |
| 4758 | /* SB10. This and SB9 could probably be combined some way, but khw |
| 4759 | * has decided to follow the Unicode rule book precisely for |
| 4760 | * simplified maintenance */ |
| 4761 | if ( after == SB_Sp |
| 4762 | || after == SB_Sep |
| 4763 | || after == SB_CR |
| 4764 | || after == SB_LF) |
| 4765 | { |
| 4766 | return FALSE; |
| 4767 | } |
| 4768 | } |
| 4769 | |
| 4770 | /* SB11. */ |
| 4771 | return TRUE; |
| 4772 | } |
| 4773 | |
| 4774 | /* Otherwise, do not break. |
| 4775 | SB12. Any × Any */ |
| 4776 | |
| 4777 | return FALSE; |
| 4778 | } |
| 4779 | |
| 4780 | STATIC SB_enum |
| 4781 | S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target) |
| 4782 | { |
| 4783 | SB_enum sb; |
| 4784 | |
| 4785 | PERL_ARGS_ASSERT_ADVANCE_ONE_SB; |
| 4786 | |
| 4787 | if (*curpos >= strend) { |
| 4788 | return SB_EDGE; |
| 4789 | } |
| 4790 | |
| 4791 | if (utf8_target) { |
| 4792 | do { |
| 4793 | *curpos += UTF8SKIP(*curpos); |
| 4794 | if (*curpos >= strend) { |
| 4795 | return SB_EDGE; |
| 4796 | } |
| 4797 | sb = getSB_VAL_UTF8(*curpos, strend); |
| 4798 | } while (sb == SB_Extend || sb == SB_Format); |
| 4799 | } |
| 4800 | else { |
| 4801 | do { |
| 4802 | (*curpos)++; |
| 4803 | if (*curpos >= strend) { |
| 4804 | return SB_EDGE; |
| 4805 | } |
| 4806 | sb = getSB_VAL_CP(**curpos); |
| 4807 | } while (sb == SB_Extend || sb == SB_Format); |
| 4808 | } |
| 4809 | |
| 4810 | return sb; |
| 4811 | } |
| 4812 | |
| 4813 | STATIC SB_enum |
| 4814 | S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target) |
| 4815 | { |
| 4816 | SB_enum sb; |
| 4817 | |
| 4818 | PERL_ARGS_ASSERT_BACKUP_ONE_SB; |
| 4819 | |
| 4820 | if (*curpos < strbeg) { |
| 4821 | return SB_EDGE; |
| 4822 | } |
| 4823 | |
| 4824 | if (utf8_target) { |
| 4825 | U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg); |
| 4826 | if (! prev_char_pos) { |
| 4827 | return SB_EDGE; |
| 4828 | } |
| 4829 | |
| 4830 | /* Back up over Extend and Format. curpos is always just to the right |
| 4831 | * of the characater whose value we are getting */ |
| 4832 | do { |
| 4833 | U8 * prev_prev_char_pos; |
| 4834 | if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, |
| 4835 | strbeg))) |
| 4836 | { |
| 4837 | sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos); |
| 4838 | *curpos = prev_char_pos; |
| 4839 | prev_char_pos = prev_prev_char_pos; |
| 4840 | } |
| 4841 | else { |
| 4842 | *curpos = (U8 *) strbeg; |
| 4843 | return SB_EDGE; |
| 4844 | } |
| 4845 | } while (sb == SB_Extend || sb == SB_Format); |
| 4846 | } |
| 4847 | else { |
| 4848 | do { |
| 4849 | if (*curpos - 2 < strbeg) { |
| 4850 | *curpos = (U8 *) strbeg; |
| 4851 | return SB_EDGE; |
| 4852 | } |
| 4853 | (*curpos)--; |
| 4854 | sb = getSB_VAL_CP(*(*curpos - 1)); |
| 4855 | } while (sb == SB_Extend || sb == SB_Format); |
| 4856 | } |
| 4857 | |
| 4858 | return sb; |
| 4859 | } |
| 4860 | |
| 4861 | STATIC bool |
| 4862 | S_isWB(pTHX_ WB_enum previous, |
| 4863 | WB_enum before, |
| 4864 | WB_enum after, |
| 4865 | const U8 * const strbeg, |
| 4866 | const U8 * const curpos, |
| 4867 | const U8 * const strend, |
| 4868 | const bool utf8_target) |
| 4869 | { |
| 4870 | /* Return a boolean as to if the boundary between 'before' and 'after' is |
| 4871 | * a Unicode word break, using their published algorithm, but tailored for |
| 4872 | * Perl by treating spans of white space as one unit. Context may be |
| 4873 | * needed to make this determination. If the value for the character |
| 4874 | * before 'before' is known, it is passed as 'previous'; otherwise that |
| 4875 | * should be set to WB_UNKNOWN. The other input parameters give the |
| 4876 | * boundaries and current position in the matching of the string. That |
| 4877 | * is, 'curpos' marks the position where the character whose wb value is |
| 4878 | * 'after' begins. See http://www.unicode.org/reports/tr29/ */ |
| 4879 | |
| 4880 | U8 * before_pos = (U8 *) curpos; |
| 4881 | U8 * after_pos = (U8 *) curpos; |
| 4882 | WB_enum prev = before; |
| 4883 | WB_enum next; |
| 4884 | |
| 4885 | PERL_ARGS_ASSERT_ISWB; |
| 4886 | |
| 4887 | /* Rule numbers in the comments below are as of Unicode 8.0 */ |
| 4888 | |
| 4889 | redo: |
| 4890 | before = prev; |
| 4891 | switch (WB_table[before][after]) { |
| 4892 | case WB_BREAKABLE: |
| 4893 | return TRUE; |
| 4894 | |
| 4895 | case WB_NOBREAK: |
| 4896 | return FALSE; |
| 4897 | |
| 4898 | case WB_hs_then_hs: /* 2 horizontal spaces in a row */ |
| 4899 | next = advance_one_WB(&after_pos, strend, utf8_target, |
| 4900 | FALSE /* Don't skip Extend nor Format */ ); |
| 4901 | /* A space immediately preceeding an Extend or Format is attached |
| 4902 | * to by them, and hence gets separated from previous spaces. |
| 4903 | * Otherwise don't break between horizontal white space */ |
| 4904 | return next == WB_Extend || next == WB_Format; |
| 4905 | |
| 4906 | /* WB4 Ignore Format and Extend characters, except when they appear at |
| 4907 | * the beginning of a region of text. This code currently isn't |
| 4908 | * general purpose, but it works as the rules are currently and likely |
| 4909 | * to be laid out. The reason it works is that when 'they appear at |
| 4910 | * the beginning of a region of text', the rule is to break before |
| 4911 | * them, just like any other character. Therefore, the default rule |
| 4912 | * applies and we don't have to look in more depth. Should this ever |
| 4913 | * change, we would have to have 2 'case' statements, like in the |
| 4914 | * rules below, and backup a single character (not spacing over the |
| 4915 | * extend ones) and then see if that is one of the region-end |
| 4916 | * characters and go from there */ |
| 4917 | case WB_Ex_or_FO_then_foo: |
| 4918 | prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target); |
| 4919 | goto redo; |
| 4920 | |
| 4921 | case WB_DQ_then_HL + WB_BREAKABLE: |
| 4922 | case WB_DQ_then_HL + WB_NOBREAK: |
| 4923 | |
| 4924 | /* WB7c Hebrew_Letter Double_Quote × Hebrew_Letter */ |
| 4925 | |
| 4926 | if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target) |
| 4927 | == WB_Hebrew_Letter) |
| 4928 | { |
| 4929 | return FALSE; |
| 4930 | } |
| 4931 | |
| 4932 | return WB_table[before][after] - WB_DQ_then_HL == WB_BREAKABLE; |
| 4933 | |
| 4934 | case WB_HL_then_DQ + WB_BREAKABLE: |
| 4935 | case WB_HL_then_DQ + WB_NOBREAK: |
| 4936 | |
| 4937 | /* WB7b Hebrew_Letter × Double_Quote Hebrew_Letter */ |
| 4938 | |
| 4939 | if (advance_one_WB(&after_pos, strend, utf8_target, |
| 4940 | TRUE /* Do skip Extend and Format */ ) |
| 4941 | == WB_Hebrew_Letter) |
| 4942 | { |
| 4943 | return FALSE; |
| 4944 | } |
| 4945 | |
| 4946 | return WB_table[before][after] - WB_HL_then_DQ == WB_BREAKABLE; |
| 4947 | |
| 4948 | case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_NOBREAK: |
| 4949 | case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_BREAKABLE: |
| 4950 | |
| 4951 | /* WB6 (ALetter | Hebrew_Letter) × (MidLetter | MidNumLet |
| 4952 | * | Single_Quote) (ALetter | Hebrew_Letter) */ |
| 4953 | |
| 4954 | next = advance_one_WB(&after_pos, strend, utf8_target, |
| 4955 | TRUE /* Do skip Extend and Format */ ); |
| 4956 | |
| 4957 | if (next == WB_ALetter || next == WB_Hebrew_Letter) |
| 4958 | { |
| 4959 | return FALSE; |
| 4960 | } |
| 4961 | |
| 4962 | return WB_table[before][after] |
| 4963 | - WB_LE_or_HL_then_MB_or_ML_or_SQ == WB_BREAKABLE; |
| 4964 | |
| 4965 | case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_NOBREAK: |
| 4966 | case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_BREAKABLE: |
| 4967 | |
| 4968 | /* WB7 (ALetter | Hebrew_Letter) (MidLetter | MidNumLet |
| 4969 | * | Single_Quote) × (ALetter | Hebrew_Letter) */ |
| 4970 | |
| 4971 | prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target); |
| 4972 | if (prev == WB_ALetter || prev == WB_Hebrew_Letter) |
| 4973 | { |
| 4974 | return FALSE; |
| 4975 | } |
| 4976 | |
| 4977 | return WB_table[before][after] |
| 4978 | - WB_MB_or_ML_or_SQ_then_LE_or_HL == WB_BREAKABLE; |
| 4979 | |
| 4980 | case WB_MB_or_MN_or_SQ_then_NU + WB_NOBREAK: |
| 4981 | case WB_MB_or_MN_or_SQ_then_NU + WB_BREAKABLE: |
| 4982 | |
| 4983 | /* WB11 Numeric (MidNum | (MidNumLet | Single_Quote)) × Numeric |
| 4984 | * */ |
| 4985 | |
| 4986 | if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target) |
| 4987 | == WB_Numeric) |
| 4988 | { |
| 4989 | return FALSE; |
| 4990 | } |
| 4991 | |
| 4992 | return WB_table[before][after] |
| 4993 | - WB_MB_or_MN_or_SQ_then_NU == WB_BREAKABLE; |
| 4994 | |
| 4995 | case WB_NU_then_MB_or_MN_or_SQ + WB_NOBREAK: |
| 4996 | case WB_NU_then_MB_or_MN_or_SQ + WB_BREAKABLE: |
| 4997 | |
| 4998 | /* WB12 Numeric × (MidNum | MidNumLet | Single_Quote) Numeric */ |
| 4999 | |
| 5000 | if (advance_one_WB(&after_pos, strend, utf8_target, |
| 5001 | TRUE /* Do skip Extend and Format */ ) |
| 5002 | == WB_Numeric) |
| 5003 | { |
| 5004 | return FALSE; |
| 5005 | } |
| 5006 | |
| 5007 | return WB_table[before][after] |
| 5008 | - WB_NU_then_MB_or_MN_or_SQ == WB_BREAKABLE; |
| 5009 | |
| 5010 | default: |
| 5011 | break; |
| 5012 | } |
| 5013 | |
| 5014 | #ifdef DEBUGGING |
| 5015 | Perl_re_printf( aTHX_ "Unhandled WB pair: WB_table[%d, %d] = %d\n", |
| 5016 | before, after, WB_table[before][after]); |
| 5017 | assert(0); |
| 5018 | #endif |
| 5019 | return TRUE; |
| 5020 | } |
| 5021 | |
| 5022 | STATIC WB_enum |
| 5023 | S_advance_one_WB(pTHX_ U8 ** curpos, |
| 5024 | const U8 * const strend, |
| 5025 | const bool utf8_target, |
| 5026 | const bool skip_Extend_Format) |
| 5027 | { |
| 5028 | WB_enum wb; |
| 5029 | |
| 5030 | PERL_ARGS_ASSERT_ADVANCE_ONE_WB; |
| 5031 | |
| 5032 | if (*curpos >= strend) { |
| 5033 | return WB_EDGE; |
| 5034 | } |
| 5035 | |
| 5036 | if (utf8_target) { |
| 5037 | |
| 5038 | /* Advance over Extend and Format */ |
| 5039 | do { |
| 5040 | *curpos += UTF8SKIP(*curpos); |
| 5041 | if (*curpos >= strend) { |
| 5042 | return WB_EDGE; |
| 5043 | } |
| 5044 | wb = getWB_VAL_UTF8(*curpos, strend); |
| 5045 | } while ( skip_Extend_Format |
| 5046 | && (wb == WB_Extend || wb == WB_Format)); |
| 5047 | } |
| 5048 | else { |
| 5049 | do { |
| 5050 | (*curpos)++; |
| 5051 | if (*curpos >= strend) { |
| 5052 | return WB_EDGE; |
| 5053 | } |
| 5054 | wb = getWB_VAL_CP(**curpos); |
| 5055 | } while ( skip_Extend_Format |
| 5056 | && (wb == WB_Extend || wb == WB_Format)); |
| 5057 | } |
| 5058 | |
| 5059 | return wb; |
| 5060 | } |
| 5061 | |
| 5062 | STATIC WB_enum |
| 5063 | S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target) |
| 5064 | { |
| 5065 | WB_enum wb; |
| 5066 | |
| 5067 | PERL_ARGS_ASSERT_BACKUP_ONE_WB; |
| 5068 | |
| 5069 | /* If we know what the previous character's break value is, don't have |
| 5070 | * to look it up */ |
| 5071 | if (*previous != WB_UNKNOWN) { |
| 5072 | wb = *previous; |
| 5073 | |
| 5074 | /* But we need to move backwards by one */ |
| 5075 | if (utf8_target) { |
| 5076 | *curpos = reghopmaybe3(*curpos, -1, strbeg); |
| 5077 | if (! *curpos) { |
| 5078 | *previous = WB_EDGE; |
| 5079 | *curpos = (U8 *) strbeg; |
| 5080 | } |
| 5081 | else { |
| 5082 | *previous = WB_UNKNOWN; |
| 5083 | } |
| 5084 | } |
| 5085 | else { |
| 5086 | (*curpos)--; |
| 5087 | *previous = (*curpos <= strbeg) ? WB_EDGE : WB_UNKNOWN; |
| 5088 | } |
| 5089 | |
| 5090 | /* And we always back up over these two types */ |
| 5091 | if (wb != WB_Extend && wb != WB_Format) { |
| 5092 | return wb; |
| 5093 | } |
| 5094 | } |
| 5095 | |
| 5096 | if (*curpos < strbeg) { |
| 5097 | return WB_EDGE; |
| 5098 | } |
| 5099 | |
| 5100 | if (utf8_target) { |
| 5101 | U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg); |
| 5102 | if (! prev_char_pos) { |
| 5103 | return WB_EDGE; |
| 5104 | } |
| 5105 | |
| 5106 | /* Back up over Extend and Format. curpos is always just to the right |
| 5107 | * of the characater whose value we are getting */ |
| 5108 | do { |
| 5109 | U8 * prev_prev_char_pos; |
| 5110 | if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, |
| 5111 | -1, |
| 5112 | strbeg))) |
| 5113 | { |
| 5114 | wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos); |
| 5115 | *curpos = prev_char_pos; |
| 5116 | prev_char_pos = prev_prev_char_pos; |
| 5117 | } |
| 5118 | else { |
| 5119 | *curpos = (U8 *) strbeg; |
| 5120 | return WB_EDGE; |
| 5121 | } |
| 5122 | } while (wb == WB_Extend || wb == WB_Format); |
| 5123 | } |
| 5124 | else { |
| 5125 | do { |
| 5126 | if (*curpos - 2 < strbeg) { |
| 5127 | *curpos = (U8 *) strbeg; |
| 5128 | return WB_EDGE; |
| 5129 | } |
| 5130 | (*curpos)--; |
| 5131 | wb = getWB_VAL_CP(*(*curpos - 1)); |
| 5132 | } while (wb == WB_Extend || wb == WB_Format); |
| 5133 | } |
| 5134 | |
| 5135 | return wb; |
| 5136 | } |
| 5137 | |
| 5138 | #define EVAL_CLOSE_PAREN_IS(st,expr) \ |
| 5139 | ( \ |
| 5140 | ( ( st ) ) && \ |
| 5141 | ( ( st )->u.eval.close_paren ) && \ |
| 5142 | ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \ |
| 5143 | ) |
| 5144 | |
| 5145 | #define EVAL_CLOSE_PAREN_IS_TRUE(st,expr) \ |
| 5146 | ( \ |
| 5147 | ( ( st ) ) && \ |
| 5148 | ( ( st )->u.eval.close_paren ) && \ |
| 5149 | ( ( expr ) ) && \ |
| 5150 | ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \ |
| 5151 | ) |
| 5152 | |
| 5153 | |
| 5154 | #define EVAL_CLOSE_PAREN_SET(st,expr) \ |
| 5155 | (st)->u.eval.close_paren = ( (expr) + 1 ) |
| 5156 | |
| 5157 | #define EVAL_CLOSE_PAREN_CLEAR(st) \ |
| 5158 | (st)->u.eval.close_paren = 0 |
| 5159 | |
| 5160 | /* returns -1 on failure, $+[0] on success */ |
| 5161 | STATIC SSize_t |
| 5162 | S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog) |
| 5163 | { |
| 5164 | |
| 5165 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
| 5166 | dMY_CXT; |
| 5167 | #endif |
| 5168 | dVAR; |
| 5169 | const bool utf8_target = reginfo->is_utf8_target; |
| 5170 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
| 5171 | REGEXP *rex_sv = reginfo->prog; |
| 5172 | regexp *rex = ReANY(rex_sv); |
| 5173 | RXi_GET_DECL(rex,rexi); |
| 5174 | /* the current state. This is a cached copy of PL_regmatch_state */ |
| 5175 | regmatch_state *st; |
| 5176 | /* cache heavy used fields of st in registers */ |
| 5177 | regnode *scan; |
| 5178 | regnode *next; |
| 5179 | U32 n = 0; /* general value; init to avoid compiler warning */ |
| 5180 | SSize_t ln = 0; /* len or last; init to avoid compiler warning */ |
| 5181 | char *locinput = startpos; |
| 5182 | char *pushinput; /* where to continue after a PUSH */ |
| 5183 | I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */ |
| 5184 | |
| 5185 | bool result = 0; /* return value of S_regmatch */ |
| 5186 | int depth = 0; /* depth of backtrack stack */ |
| 5187 | U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */ |
| 5188 | const U32 max_nochange_depth = |
| 5189 | (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ? |
| 5190 | 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH; |
| 5191 | regmatch_state *yes_state = NULL; /* state to pop to on success of |
| 5192 | subpattern */ |
| 5193 | /* mark_state piggy backs on the yes_state logic so that when we unwind |
| 5194 | the stack on success we can update the mark_state as we go */ |
| 5195 | regmatch_state *mark_state = NULL; /* last mark state we have seen */ |
| 5196 | regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ |
| 5197 | struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ |
| 5198 | U32 state_num; |
| 5199 | bool no_final = 0; /* prevent failure from backtracking? */ |
| 5200 | bool do_cutgroup = 0; /* no_final only until next branch/trie entry */ |
| 5201 | char *startpoint = locinput; |
| 5202 | SV *popmark = NULL; /* are we looking for a mark? */ |
| 5203 | SV *sv_commit = NULL; /* last mark name seen in failure */ |
| 5204 | SV *sv_yes_mark = NULL; /* last mark name we have seen |
| 5205 | during a successful match */ |
| 5206 | U32 lastopen = 0; /* last open we saw */ |
| 5207 | bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0; |
| 5208 | SV* const oreplsv = GvSVn(PL_replgv); |
| 5209 | /* these three flags are set by various ops to signal information to |
| 5210 | * the very next op. They have a useful lifetime of exactly one loop |
| 5211 | * iteration, and are not preserved or restored by state pushes/pops |
| 5212 | */ |
| 5213 | bool sw = 0; /* the condition value in (?(cond)a|b) */ |
| 5214 | bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ |
| 5215 | int logical = 0; /* the following EVAL is: |
| 5216 | 0: (?{...}) |
| 5217 | 1: (?(?{...})X|Y) |
| 5218 | 2: (??{...}) |
| 5219 | or the following IFMATCH/UNLESSM is: |
| 5220 | false: plain (?=foo) |
| 5221 | true: used as a condition: (?(?=foo)) |
| 5222 | */ |
| 5223 | PAD* last_pad = NULL; |
| 5224 | dMULTICALL; |
| 5225 | U8 gimme = G_SCALAR; |
| 5226 | CV *caller_cv = NULL; /* who called us */ |
| 5227 | CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */ |
| 5228 | CHECKPOINT runops_cp; /* savestack position before executing EVAL */ |
| 5229 | U32 maxopenparen = 0; /* max '(' index seen so far */ |
| 5230 | int to_complement; /* Invert the result? */ |
| 5231 | _char_class_number classnum; |
| 5232 | bool is_utf8_pat = reginfo->is_utf8_pat; |
| 5233 | bool match = FALSE; |
| 5234 | |
| 5235 | /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */ |
| 5236 | #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL)) |
| 5237 | # define SOLARIS_BAD_OPTIMIZER |
| 5238 | const U32 *pl_charclass_dup = PL_charclass; |
| 5239 | # define PL_charclass pl_charclass_dup |
| 5240 | #endif |
| 5241 | |
| 5242 | #ifdef DEBUGGING |
| 5243 | GET_RE_DEBUG_FLAGS_DECL; |
| 5244 | #endif |
| 5245 | |
| 5246 | /* protect against undef(*^R) */ |
| 5247 | SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv)); |
| 5248 | |
| 5249 | /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */ |
| 5250 | multicall_oldcatch = 0; |
| 5251 | PERL_UNUSED_VAR(multicall_cop); |
| 5252 | |
| 5253 | PERL_ARGS_ASSERT_REGMATCH; |
| 5254 | |
| 5255 | DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({ |
| 5256 | Perl_re_printf( aTHX_ "regmatch start\n"); |
| 5257 | })); |
| 5258 | |
| 5259 | st = PL_regmatch_state; |
| 5260 | |
| 5261 | /* Note that nextchr is a byte even in UTF */ |
| 5262 | SET_nextchr; |
| 5263 | scan = prog; |
| 5264 | while (scan != NULL) { |
| 5265 | |
| 5266 | |
| 5267 | next = scan + NEXT_OFF(scan); |
| 5268 | if (next == scan) |
| 5269 | next = NULL; |
| 5270 | state_num = OP(scan); |
| 5271 | |
| 5272 | reenter_switch: |
| 5273 | DEBUG_EXECUTE_r( |
| 5274 | if (state_num <= REGNODE_MAX) { |
| 5275 | SV * const prop = sv_newmortal(); |
| 5276 | regnode *rnext = regnext(scan); |
| 5277 | |
| 5278 | DUMP_EXEC_POS( locinput, scan, utf8_target, depth ); |
| 5279 | regprop(rex, prop, scan, reginfo, NULL); |
| 5280 | Perl_re_printf( aTHX_ |
| 5281 | "%*s%"IVdf":%s(%"IVdf")\n", |
| 5282 | INDENT_CHARS(depth), "", |
| 5283 | (IV)(scan - rexi->program), |
| 5284 | SvPVX_const(prop), |
| 5285 | (PL_regkind[OP(scan)] == END || !rnext) ? |
| 5286 | 0 : (IV)(rnext - rexi->program)); |
| 5287 | } |
| 5288 | ); |
| 5289 | |
| 5290 | to_complement = 0; |
| 5291 | |
| 5292 | SET_nextchr; |
| 5293 | assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS)); |
| 5294 | |
| 5295 | switch (state_num) { |
| 5296 | case SBOL: /* /^../ and /\A../ */ |
| 5297 | if (locinput == reginfo->strbeg) |
| 5298 | break; |
| 5299 | sayNO; |
| 5300 | |
| 5301 | case MBOL: /* /^../m */ |
| 5302 | if (locinput == reginfo->strbeg || |
| 5303 | (!NEXTCHR_IS_EOS && locinput[-1] == '\n')) |
| 5304 | { |
| 5305 | break; |
| 5306 | } |
| 5307 | sayNO; |
| 5308 | |
| 5309 | case GPOS: /* \G */ |
| 5310 | if (locinput == reginfo->ganch) |
| 5311 | break; |
| 5312 | sayNO; |
| 5313 | |
| 5314 | case KEEPS: /* \K */ |
| 5315 | /* update the startpoint */ |
| 5316 | st->u.keeper.val = rex->offs[0].start; |
| 5317 | rex->offs[0].start = locinput - reginfo->strbeg; |
| 5318 | PUSH_STATE_GOTO(KEEPS_next, next, locinput); |
| 5319 | NOT_REACHED; /* NOTREACHED */ |
| 5320 | |
| 5321 | case KEEPS_next_fail: |
| 5322 | /* rollback the start point change */ |
| 5323 | rex->offs[0].start = st->u.keeper.val; |
| 5324 | sayNO_SILENT; |
| 5325 | NOT_REACHED; /* NOTREACHED */ |
| 5326 | |
| 5327 | case MEOL: /* /..$/m */ |
| 5328 | if (!NEXTCHR_IS_EOS && nextchr != '\n') |
| 5329 | sayNO; |
| 5330 | break; |
| 5331 | |
| 5332 | case SEOL: /* /..$/ */ |
| 5333 | if (!NEXTCHR_IS_EOS && nextchr != '\n') |
| 5334 | sayNO; |
| 5335 | if (reginfo->strend - locinput > 1) |
| 5336 | sayNO; |
| 5337 | break; |
| 5338 | |
| 5339 | case EOS: /* \z */ |
| 5340 | if (!NEXTCHR_IS_EOS) |
| 5341 | sayNO; |
| 5342 | break; |
| 5343 | |
| 5344 | case SANY: /* /./s */ |
| 5345 | if (NEXTCHR_IS_EOS) |
| 5346 | sayNO; |
| 5347 | goto increment_locinput; |
| 5348 | |
| 5349 | case REG_ANY: /* /./ */ |
| 5350 | if ((NEXTCHR_IS_EOS) || nextchr == '\n') |
| 5351 | sayNO; |
| 5352 | goto increment_locinput; |
| 5353 | |
| 5354 | |
| 5355 | #undef ST |
| 5356 | #define ST st->u.trie |
| 5357 | case TRIEC: /* (ab|cd) with known charclass */ |
| 5358 | /* In this case the charclass data is available inline so |
| 5359 | we can fail fast without a lot of extra overhead. |
| 5360 | */ |
| 5361 | if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) { |
| 5362 | DEBUG_EXECUTE_r( |
| 5363 | Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n", |
| 5364 | depth, PL_colors[4], PL_colors[5]) |
| 5365 | ); |
| 5366 | sayNO_SILENT; |
| 5367 | NOT_REACHED; /* NOTREACHED */ |
| 5368 | } |
| 5369 | /* FALLTHROUGH */ |
| 5370 | case TRIE: /* (ab|cd) */ |
| 5371 | /* the basic plan of execution of the trie is: |
| 5372 | * At the beginning, run though all the states, and |
| 5373 | * find the longest-matching word. Also remember the position |
| 5374 | * of the shortest matching word. For example, this pattern: |
| 5375 | * 1 2 3 4 5 |
| 5376 | * ab|a|x|abcd|abc |
| 5377 | * when matched against the string "abcde", will generate |
| 5378 | * accept states for all words except 3, with the longest |
| 5379 | * matching word being 4, and the shortest being 2 (with |
| 5380 | * the position being after char 1 of the string). |
| 5381 | * |
| 5382 | * Then for each matching word, in word order (i.e. 1,2,4,5), |
| 5383 | * we run the remainder of the pattern; on each try setting |
| 5384 | * the current position to the character following the word, |
| 5385 | * returning to try the next word on failure. |
| 5386 | * |
| 5387 | * We avoid having to build a list of words at runtime by |
| 5388 | * using a compile-time structure, wordinfo[].prev, which |
| 5389 | * gives, for each word, the previous accepting word (if any). |
| 5390 | * In the case above it would contain the mappings 1->2, 2->0, |
| 5391 | * 3->0, 4->5, 5->1. We can use this table to generate, from |
| 5392 | * the longest word (4 above), a list of all words, by |
| 5393 | * following the list of prev pointers; this gives us the |
| 5394 | * unordered list 4,5,1,2. Then given the current word we have |
| 5395 | * just tried, we can go through the list and find the |
| 5396 | * next-biggest word to try (so if we just failed on word 2, |
| 5397 | * the next in the list is 4). |
| 5398 | * |
| 5399 | * Since at runtime we don't record the matching position in |
| 5400 | * the string for each word, we have to work that out for |
| 5401 | * each word we're about to process. The wordinfo table holds |
| 5402 | * the character length of each word; given that we recorded |
| 5403 | * at the start: the position of the shortest word and its |
| 5404 | * length in chars, we just need to move the pointer the |
| 5405 | * difference between the two char lengths. Depending on |
| 5406 | * Unicode status and folding, that's cheap or expensive. |
| 5407 | * |
| 5408 | * This algorithm is optimised for the case where are only a |
| 5409 | * small number of accept states, i.e. 0,1, or maybe 2. |
| 5410 | * With lots of accepts states, and having to try all of them, |
| 5411 | * it becomes quadratic on number of accept states to find all |
| 5412 | * the next words. |
| 5413 | */ |
| 5414 | |
| 5415 | { |
| 5416 | /* what type of TRIE am I? (utf8 makes this contextual) */ |
| 5417 | DECL_TRIE_TYPE(scan); |
| 5418 | |
| 5419 | /* what trie are we using right now */ |
| 5420 | reg_trie_data * const trie |
| 5421 | = (reg_trie_data*)rexi->data->data[ ARG( scan ) ]; |
| 5422 | HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]); |
| 5423 | U32 state = trie->startstate; |
| 5424 | |
| 5425 | if (scan->flags == EXACTL || scan->flags == EXACTFLU8) { |
| 5426 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 5427 | if (utf8_target |
| 5428 | && UTF8_IS_ABOVE_LATIN1(nextchr) |
| 5429 | && scan->flags == EXACTL) |
| 5430 | { |
| 5431 | /* We only output for EXACTL, as we let the folder |
| 5432 | * output this message for EXACTFLU8 to avoid |
| 5433 | * duplication */ |
| 5434 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, |
| 5435 | reginfo->strend); |
| 5436 | } |
| 5437 | } |
| 5438 | if ( trie->bitmap |
| 5439 | && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr))) |
| 5440 | { |
| 5441 | if (trie->states[ state ].wordnum) { |
| 5442 | DEBUG_EXECUTE_r( |
| 5443 | Perl_re_exec_indentf( aTHX_ "%smatched empty string...%s\n", |
| 5444 | depth, PL_colors[4], PL_colors[5]) |
| 5445 | ); |
| 5446 | if (!trie->jump) |
| 5447 | break; |
| 5448 | } else { |
| 5449 | DEBUG_EXECUTE_r( |
| 5450 | Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n", |
| 5451 | depth, PL_colors[4], PL_colors[5]) |
| 5452 | ); |
| 5453 | sayNO_SILENT; |
| 5454 | } |
| 5455 | } |
| 5456 | |
| 5457 | { |
| 5458 | U8 *uc = ( U8* )locinput; |
| 5459 | |
| 5460 | STRLEN len = 0; |
| 5461 | STRLEN foldlen = 0; |
| 5462 | U8 *uscan = (U8*)NULL; |
| 5463 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 5464 | U32 charcount = 0; /* how many input chars we have matched */ |
| 5465 | U32 accepted = 0; /* have we seen any accepting states? */ |
| 5466 | |
| 5467 | ST.jump = trie->jump; |
| 5468 | ST.me = scan; |
| 5469 | ST.firstpos = NULL; |
| 5470 | ST.longfold = FALSE; /* char longer if folded => it's harder */ |
| 5471 | ST.nextword = 0; |
| 5472 | |
| 5473 | /* fully traverse the TRIE; note the position of the |
| 5474 | shortest accept state and the wordnum of the longest |
| 5475 | accept state */ |
| 5476 | |
| 5477 | while ( state && uc <= (U8*)(reginfo->strend) ) { |
| 5478 | U32 base = trie->states[ state ].trans.base; |
| 5479 | UV uvc = 0; |
| 5480 | U16 charid = 0; |
| 5481 | U16 wordnum; |
| 5482 | wordnum = trie->states[ state ].wordnum; |
| 5483 | |
| 5484 | if (wordnum) { /* it's an accept state */ |
| 5485 | if (!accepted) { |
| 5486 | accepted = 1; |
| 5487 | /* record first match position */ |
| 5488 | if (ST.longfold) { |
| 5489 | ST.firstpos = (U8*)locinput; |
| 5490 | ST.firstchars = 0; |
| 5491 | } |
| 5492 | else { |
| 5493 | ST.firstpos = uc; |
| 5494 | ST.firstchars = charcount; |
| 5495 | } |
| 5496 | } |
| 5497 | if (!ST.nextword || wordnum < ST.nextword) |
| 5498 | ST.nextword = wordnum; |
| 5499 | ST.topword = wordnum; |
| 5500 | } |
| 5501 | |
| 5502 | DEBUG_TRIE_EXECUTE_r({ |
| 5503 | DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth ); |
| 5504 | Perl_re_exec_indentf( aTHX_ |
| 5505 | "%sState: %4"UVxf" Accepted: %c ", |
| 5506 | depth, PL_colors[4], |
| 5507 | (UV)state, (accepted ? 'Y' : 'N')); |
| 5508 | }); |
| 5509 | |
| 5510 | /* read a char and goto next state */ |
| 5511 | if ( base && (foldlen || uc < (U8*)(reginfo->strend))) { |
| 5512 | I32 offset; |
| 5513 | REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, |
| 5514 | uscan, len, uvc, charid, foldlen, |
| 5515 | foldbuf, uniflags); |
| 5516 | charcount++; |
| 5517 | if (foldlen>0) |
| 5518 | ST.longfold = TRUE; |
| 5519 | if (charid && |
| 5520 | ( ((offset = |
| 5521 | base + charid - 1 - trie->uniquecharcount)) >= 0) |
| 5522 | |
| 5523 | && ((U32)offset < trie->lasttrans) |
| 5524 | && trie->trans[offset].check == state) |
| 5525 | { |
| 5526 | state = trie->trans[offset].next; |
| 5527 | } |
| 5528 | else { |
| 5529 | state = 0; |
| 5530 | } |
| 5531 | uc += len; |
| 5532 | |
| 5533 | } |
| 5534 | else { |
| 5535 | state = 0; |
| 5536 | } |
| 5537 | DEBUG_TRIE_EXECUTE_r( |
| 5538 | Perl_re_printf( aTHX_ |
| 5539 | "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", |
| 5540 | charid, uvc, (UV)state, PL_colors[5] ); |
| 5541 | ); |
| 5542 | } |
| 5543 | if (!accepted) |
| 5544 | sayNO; |
| 5545 | |
| 5546 | /* calculate total number of accept states */ |
| 5547 | { |
| 5548 | U16 w = ST.topword; |
| 5549 | accepted = 0; |
| 5550 | while (w) { |
| 5551 | w = trie->wordinfo[w].prev; |
| 5552 | accepted++; |
| 5553 | } |
| 5554 | ST.accepted = accepted; |
| 5555 | } |
| 5556 | |
| 5557 | DEBUG_EXECUTE_r( |
| 5558 | Perl_re_exec_indentf( aTHX_ "%sgot %"IVdf" possible matches%s\n", |
| 5559 | depth, |
| 5560 | PL_colors[4], (IV)ST.accepted, PL_colors[5] ); |
| 5561 | ); |
| 5562 | goto trie_first_try; /* jump into the fail handler */ |
| 5563 | }} |
| 5564 | NOT_REACHED; /* NOTREACHED */ |
| 5565 | |
| 5566 | case TRIE_next_fail: /* we failed - try next alternative */ |
| 5567 | { |
| 5568 | U8 *uc; |
| 5569 | if ( ST.jump) { |
| 5570 | REGCP_UNWIND(ST.cp); |
| 5571 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 5572 | } |
| 5573 | if (!--ST.accepted) { |
| 5574 | DEBUG_EXECUTE_r({ |
| 5575 | Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n", |
| 5576 | depth, |
| 5577 | PL_colors[4], |
| 5578 | PL_colors[5] ); |
| 5579 | }); |
| 5580 | sayNO_SILENT; |
| 5581 | } |
| 5582 | { |
| 5583 | /* Find next-highest word to process. Note that this code |
| 5584 | * is O(N^2) per trie run (O(N) per branch), so keep tight */ |
| 5585 | U16 min = 0; |
| 5586 | U16 word; |
| 5587 | U16 const nextword = ST.nextword; |
| 5588 | reg_trie_wordinfo * const wordinfo |
| 5589 | = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo; |
| 5590 | for (word=ST.topword; word; word=wordinfo[word].prev) { |
| 5591 | if (word > nextword && (!min || word < min)) |
| 5592 | min = word; |
| 5593 | } |
| 5594 | ST.nextword = min; |
| 5595 | } |
| 5596 | |
| 5597 | trie_first_try: |
| 5598 | if (do_cutgroup) { |
| 5599 | do_cutgroup = 0; |
| 5600 | no_final = 0; |
| 5601 | } |
| 5602 | |
| 5603 | if ( ST.jump) { |
| 5604 | ST.lastparen = rex->lastparen; |
| 5605 | ST.lastcloseparen = rex->lastcloseparen; |
| 5606 | REGCP_SET(ST.cp); |
| 5607 | } |
| 5608 | |
| 5609 | /* find start char of end of current word */ |
| 5610 | { |
| 5611 | U32 chars; /* how many chars to skip */ |
| 5612 | reg_trie_data * const trie |
| 5613 | = (reg_trie_data*)rexi->data->data[ARG(ST.me)]; |
| 5614 | |
| 5615 | assert((trie->wordinfo[ST.nextword].len - trie->prefixlen) |
| 5616 | >= ST.firstchars); |
| 5617 | chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen) |
| 5618 | - ST.firstchars; |
| 5619 | uc = ST.firstpos; |
| 5620 | |
| 5621 | if (ST.longfold) { |
| 5622 | /* the hard option - fold each char in turn and find |
| 5623 | * its folded length (which may be different */ |
| 5624 | U8 foldbuf[UTF8_MAXBYTES_CASE + 1]; |
| 5625 | STRLEN foldlen; |
| 5626 | STRLEN len; |
| 5627 | UV uvc; |
| 5628 | U8 *uscan; |
| 5629 | |
| 5630 | while (chars) { |
| 5631 | if (utf8_target) { |
| 5632 | uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len, |
| 5633 | uniflags); |
| 5634 | uc += len; |
| 5635 | } |
| 5636 | else { |
| 5637 | uvc = *uc; |
| 5638 | uc++; |
| 5639 | } |
| 5640 | uvc = to_uni_fold(uvc, foldbuf, &foldlen); |
| 5641 | uscan = foldbuf; |
| 5642 | while (foldlen) { |
| 5643 | if (!--chars) |
| 5644 | break; |
| 5645 | uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len, |
| 5646 | uniflags); |
| 5647 | uscan += len; |
| 5648 | foldlen -= len; |
| 5649 | } |
| 5650 | } |
| 5651 | } |
| 5652 | else { |
| 5653 | if (utf8_target) |
| 5654 | while (chars--) |
| 5655 | uc += UTF8SKIP(uc); |
| 5656 | else |
| 5657 | uc += chars; |
| 5658 | } |
| 5659 | } |
| 5660 | |
| 5661 | scan = ST.me + ((ST.jump && ST.jump[ST.nextword]) |
| 5662 | ? ST.jump[ST.nextword] |
| 5663 | : NEXT_OFF(ST.me)); |
| 5664 | |
| 5665 | DEBUG_EXECUTE_r({ |
| 5666 | Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n", |
| 5667 | depth, |
| 5668 | PL_colors[4], |
| 5669 | ST.nextword, |
| 5670 | PL_colors[5] |
| 5671 | ); |
| 5672 | }); |
| 5673 | |
| 5674 | if (ST.accepted > 1 || has_cutgroup) { |
| 5675 | PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc); |
| 5676 | NOT_REACHED; /* NOTREACHED */ |
| 5677 | } |
| 5678 | /* only one choice left - just continue */ |
| 5679 | DEBUG_EXECUTE_r({ |
| 5680 | AV *const trie_words |
| 5681 | = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]); |
| 5682 | SV ** const tmp = trie_words |
| 5683 | ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL; |
| 5684 | SV *sv= tmp ? sv_newmortal() : NULL; |
| 5685 | |
| 5686 | Perl_re_exec_indentf( aTHX_ "%sonly one match left, short-circuiting: #%d <%s>%s\n", |
| 5687 | depth, PL_colors[4], |
| 5688 | ST.nextword, |
| 5689 | tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0, |
| 5690 | PL_colors[0], PL_colors[1], |
| 5691 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII |
| 5692 | ) |
| 5693 | : "not compiled under -Dr", |
| 5694 | PL_colors[5] ); |
| 5695 | }); |
| 5696 | |
| 5697 | locinput = (char*)uc; |
| 5698 | continue; /* execute rest of RE */ |
| 5699 | /* NOTREACHED */ |
| 5700 | } |
| 5701 | #undef ST |
| 5702 | |
| 5703 | case EXACTL: /* /abc/l */ |
| 5704 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 5705 | |
| 5706 | /* Complete checking would involve going through every character |
| 5707 | * matched by the string to see if any is above latin1. But the |
| 5708 | * comparision otherwise might very well be a fast assembly |
| 5709 | * language routine, and I (khw) don't think slowing things down |
| 5710 | * just to check for this warning is worth it. So this just checks |
| 5711 | * the first character */ |
| 5712 | if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) { |
| 5713 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend); |
| 5714 | } |
| 5715 | /* FALLTHROUGH */ |
| 5716 | case EXACT: { /* /abc/ */ |
| 5717 | char *s = STRING(scan); |
| 5718 | ln = STR_LEN(scan); |
| 5719 | if (utf8_target != is_utf8_pat) { |
| 5720 | /* The target and the pattern have differing utf8ness. */ |
| 5721 | char *l = locinput; |
| 5722 | const char * const e = s + ln; |
| 5723 | |
| 5724 | if (utf8_target) { |
| 5725 | /* The target is utf8, the pattern is not utf8. |
| 5726 | * Above-Latin1 code points can't match the pattern; |
| 5727 | * invariants match exactly, and the other Latin1 ones need |
| 5728 | * to be downgraded to a single byte in order to do the |
| 5729 | * comparison. (If we could be confident that the target |
| 5730 | * is not malformed, this could be refactored to have fewer |
| 5731 | * tests by just assuming that if the first bytes match, it |
| 5732 | * is an invariant, but there are tests in the test suite |
| 5733 | * dealing with (??{...}) which violate this) */ |
| 5734 | while (s < e) { |
| 5735 | if (l >= reginfo->strend |
| 5736 | || UTF8_IS_ABOVE_LATIN1(* (U8*) l)) |
| 5737 | { |
| 5738 | sayNO; |
| 5739 | } |
| 5740 | if (UTF8_IS_INVARIANT(*(U8*)l)) { |
| 5741 | if (*l != *s) { |
| 5742 | sayNO; |
| 5743 | } |
| 5744 | l++; |
| 5745 | } |
| 5746 | else { |
| 5747 | if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s) |
| 5748 | { |
| 5749 | sayNO; |
| 5750 | } |
| 5751 | l += 2; |
| 5752 | } |
| 5753 | s++; |
| 5754 | } |
| 5755 | } |
| 5756 | else { |
| 5757 | /* The target is not utf8, the pattern is utf8. */ |
| 5758 | while (s < e) { |
| 5759 | if (l >= reginfo->strend |
| 5760 | || UTF8_IS_ABOVE_LATIN1(* (U8*) s)) |
| 5761 | { |
| 5762 | sayNO; |
| 5763 | } |
| 5764 | if (UTF8_IS_INVARIANT(*(U8*)s)) { |
| 5765 | if (*s != *l) { |
| 5766 | sayNO; |
| 5767 | } |
| 5768 | s++; |
| 5769 | } |
| 5770 | else { |
| 5771 | if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l) |
| 5772 | { |
| 5773 | sayNO; |
| 5774 | } |
| 5775 | s += 2; |
| 5776 | } |
| 5777 | l++; |
| 5778 | } |
| 5779 | } |
| 5780 | locinput = l; |
| 5781 | } |
| 5782 | else { |
| 5783 | /* The target and the pattern have the same utf8ness. */ |
| 5784 | /* Inline the first character, for speed. */ |
| 5785 | if (reginfo->strend - locinput < ln |
| 5786 | || UCHARAT(s) != nextchr |
| 5787 | || (ln > 1 && memNE(s, locinput, ln))) |
| 5788 | { |
| 5789 | sayNO; |
| 5790 | } |
| 5791 | locinput += ln; |
| 5792 | } |
| 5793 | break; |
| 5794 | } |
| 5795 | |
| 5796 | case EXACTFL: { /* /abc/il */ |
| 5797 | re_fold_t folder; |
| 5798 | const U8 * fold_array; |
| 5799 | const char * s; |
| 5800 | U32 fold_utf8_flags; |
| 5801 | |
| 5802 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 5803 | folder = foldEQ_locale; |
| 5804 | fold_array = PL_fold_locale; |
| 5805 | fold_utf8_flags = FOLDEQ_LOCALE; |
| 5806 | goto do_exactf; |
| 5807 | |
| 5808 | case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so |
| 5809 | is effectively /u; hence to match, target |
| 5810 | must be UTF-8. */ |
| 5811 | if (! utf8_target) { |
| 5812 | sayNO; |
| 5813 | } |
| 5814 | fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S1_ALREADY_FOLDED |
| 5815 | | FOLDEQ_S1_FOLDS_SANE; |
| 5816 | folder = foldEQ_latin1; |
| 5817 | fold_array = PL_fold_latin1; |
| 5818 | goto do_exactf; |
| 5819 | |
| 5820 | case EXACTFU_SS: /* /\x{df}/iu */ |
| 5821 | case EXACTFU: /* /abc/iu */ |
| 5822 | folder = foldEQ_latin1; |
| 5823 | fold_array = PL_fold_latin1; |
| 5824 | fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0; |
| 5825 | goto do_exactf; |
| 5826 | |
| 5827 | case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 |
| 5828 | patterns */ |
| 5829 | assert(! is_utf8_pat); |
| 5830 | /* FALLTHROUGH */ |
| 5831 | case EXACTFA: /* /abc/iaa */ |
| 5832 | folder = foldEQ_latin1; |
| 5833 | fold_array = PL_fold_latin1; |
| 5834 | fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 5835 | goto do_exactf; |
| 5836 | |
| 5837 | case EXACTF: /* /abc/i This node only generated for |
| 5838 | non-utf8 patterns */ |
| 5839 | assert(! is_utf8_pat); |
| 5840 | folder = foldEQ; |
| 5841 | fold_array = PL_fold; |
| 5842 | fold_utf8_flags = 0; |
| 5843 | |
| 5844 | do_exactf: |
| 5845 | s = STRING(scan); |
| 5846 | ln = STR_LEN(scan); |
| 5847 | |
| 5848 | if (utf8_target |
| 5849 | || is_utf8_pat |
| 5850 | || state_num == EXACTFU_SS |
| 5851 | || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE)) |
| 5852 | { |
| 5853 | /* Either target or the pattern are utf8, or has the issue where |
| 5854 | * the fold lengths may differ. */ |
| 5855 | const char * const l = locinput; |
| 5856 | char *e = reginfo->strend; |
| 5857 | |
| 5858 | if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat, |
| 5859 | l, &e, 0, utf8_target, fold_utf8_flags)) |
| 5860 | { |
| 5861 | sayNO; |
| 5862 | } |
| 5863 | locinput = e; |
| 5864 | break; |
| 5865 | } |
| 5866 | |
| 5867 | /* Neither the target nor the pattern are utf8 */ |
| 5868 | if (UCHARAT(s) != nextchr |
| 5869 | && !NEXTCHR_IS_EOS |
| 5870 | && UCHARAT(s) != fold_array[nextchr]) |
| 5871 | { |
| 5872 | sayNO; |
| 5873 | } |
| 5874 | if (reginfo->strend - locinput < ln) |
| 5875 | sayNO; |
| 5876 | if (ln > 1 && ! folder(s, locinput, ln)) |
| 5877 | sayNO; |
| 5878 | locinput += ln; |
| 5879 | break; |
| 5880 | } |
| 5881 | |
| 5882 | case NBOUNDL: /* /\B/l */ |
| 5883 | to_complement = 1; |
| 5884 | /* FALLTHROUGH */ |
| 5885 | |
| 5886 | case BOUNDL: /* /\b/l */ |
| 5887 | { |
| 5888 | bool b1, b2; |
| 5889 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 5890 | |
| 5891 | if (FLAGS(scan) != TRADITIONAL_BOUND) { |
| 5892 | if (! IN_UTF8_CTYPE_LOCALE) { |
| 5893 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
| 5894 | B_ON_NON_UTF8_LOCALE_IS_WRONG); |
| 5895 | } |
| 5896 | goto boundu; |
| 5897 | } |
| 5898 | |
| 5899 | if (utf8_target) { |
| 5900 | if (locinput == reginfo->strbeg) |
| 5901 | b1 = isWORDCHAR_LC('\n'); |
| 5902 | else { |
| 5903 | b1 = isWORDCHAR_LC_utf8(reghop3((U8*)locinput, -1, |
| 5904 | (U8*)(reginfo->strbeg))); |
| 5905 | } |
| 5906 | b2 = (NEXTCHR_IS_EOS) |
| 5907 | ? isWORDCHAR_LC('\n') |
| 5908 | : isWORDCHAR_LC_utf8((U8*)locinput); |
| 5909 | } |
| 5910 | else { /* Here the string isn't utf8 */ |
| 5911 | b1 = (locinput == reginfo->strbeg) |
| 5912 | ? isWORDCHAR_LC('\n') |
| 5913 | : isWORDCHAR_LC(UCHARAT(locinput - 1)); |
| 5914 | b2 = (NEXTCHR_IS_EOS) |
| 5915 | ? isWORDCHAR_LC('\n') |
| 5916 | : isWORDCHAR_LC(nextchr); |
| 5917 | } |
| 5918 | if (to_complement ^ (b1 == b2)) { |
| 5919 | sayNO; |
| 5920 | } |
| 5921 | break; |
| 5922 | } |
| 5923 | |
| 5924 | case NBOUND: /* /\B/ */ |
| 5925 | to_complement = 1; |
| 5926 | /* FALLTHROUGH */ |
| 5927 | |
| 5928 | case BOUND: /* /\b/ */ |
| 5929 | if (utf8_target) { |
| 5930 | goto bound_utf8; |
| 5931 | } |
| 5932 | goto bound_ascii_match_only; |
| 5933 | |
| 5934 | case NBOUNDA: /* /\B/a */ |
| 5935 | to_complement = 1; |
| 5936 | /* FALLTHROUGH */ |
| 5937 | |
| 5938 | case BOUNDA: /* /\b/a */ |
| 5939 | { |
| 5940 | bool b1, b2; |
| 5941 | |
| 5942 | bound_ascii_match_only: |
| 5943 | /* Here the string isn't utf8, or is utf8 and only ascii characters |
| 5944 | * are to match \w. In the latter case looking at the byte just |
| 5945 | * prior to the current one may be just the final byte of a |
| 5946 | * multi-byte character. This is ok. There are two cases: |
| 5947 | * 1) it is a single byte character, and then the test is doing |
| 5948 | * just what it's supposed to. |
| 5949 | * 2) it is a multi-byte character, in which case the final byte is |
| 5950 | * never mistakable for ASCII, and so the test will say it is |
| 5951 | * not a word character, which is the correct answer. */ |
| 5952 | b1 = (locinput == reginfo->strbeg) |
| 5953 | ? isWORDCHAR_A('\n') |
| 5954 | : isWORDCHAR_A(UCHARAT(locinput - 1)); |
| 5955 | b2 = (NEXTCHR_IS_EOS) |
| 5956 | ? isWORDCHAR_A('\n') |
| 5957 | : isWORDCHAR_A(nextchr); |
| 5958 | if (to_complement ^ (b1 == b2)) { |
| 5959 | sayNO; |
| 5960 | } |
| 5961 | break; |
| 5962 | } |
| 5963 | |
| 5964 | case NBOUNDU: /* /\B/u */ |
| 5965 | to_complement = 1; |
| 5966 | /* FALLTHROUGH */ |
| 5967 | |
| 5968 | case BOUNDU: /* /\b/u */ |
| 5969 | |
| 5970 | boundu: |
| 5971 | if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) { |
| 5972 | match = FALSE; |
| 5973 | } |
| 5974 | else if (utf8_target) { |
| 5975 | bound_utf8: |
| 5976 | switch((bound_type) FLAGS(scan)) { |
| 5977 | case TRADITIONAL_BOUND: |
| 5978 | { |
| 5979 | bool b1, b2; |
| 5980 | b1 = (locinput == reginfo->strbeg) |
| 5981 | ? 0 /* isWORDCHAR_L1('\n') */ |
| 5982 | : isWORDCHAR_utf8(reghop3((U8*)locinput, -1, |
| 5983 | (U8*)(reginfo->strbeg))); |
| 5984 | b2 = (NEXTCHR_IS_EOS) |
| 5985 | ? 0 /* isWORDCHAR_L1('\n') */ |
| 5986 | : isWORDCHAR_utf8((U8*)locinput); |
| 5987 | match = cBOOL(b1 != b2); |
| 5988 | break; |
| 5989 | } |
| 5990 | case GCB_BOUND: |
| 5991 | if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { |
| 5992 | match = TRUE; /* GCB always matches at begin and |
| 5993 | end */ |
| 5994 | } |
| 5995 | else { |
| 5996 | /* Find the gcb values of previous and current |
| 5997 | * chars, then see if is a break point */ |
| 5998 | match = isGCB(getGCB_VAL_UTF8( |
| 5999 | reghop3((U8*)locinput, |
| 6000 | -1, |
| 6001 | (U8*)(reginfo->strbeg)), |
| 6002 | (U8*) reginfo->strend), |
| 6003 | getGCB_VAL_UTF8((U8*) locinput, |
| 6004 | (U8*) reginfo->strend)); |
| 6005 | } |
| 6006 | break; |
| 6007 | |
| 6008 | case LB_BOUND: |
| 6009 | if (locinput == reginfo->strbeg) { |
| 6010 | match = FALSE; |
| 6011 | } |
| 6012 | else if (NEXTCHR_IS_EOS) { |
| 6013 | match = TRUE; |
| 6014 | } |
| 6015 | else { |
| 6016 | match = isLB(getLB_VAL_UTF8( |
| 6017 | reghop3((U8*)locinput, |
| 6018 | -1, |
| 6019 | (U8*)(reginfo->strbeg)), |
| 6020 | (U8*) reginfo->strend), |
| 6021 | getLB_VAL_UTF8((U8*) locinput, |
| 6022 | (U8*) reginfo->strend), |
| 6023 | (U8*) reginfo->strbeg, |
| 6024 | (U8*) locinput, |
| 6025 | (U8*) reginfo->strend, |
| 6026 | utf8_target); |
| 6027 | } |
| 6028 | break; |
| 6029 | |
| 6030 | case SB_BOUND: /* Always matches at begin and end */ |
| 6031 | if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { |
| 6032 | match = TRUE; |
| 6033 | } |
| 6034 | else { |
| 6035 | match = isSB(getSB_VAL_UTF8( |
| 6036 | reghop3((U8*)locinput, |
| 6037 | -1, |
| 6038 | (U8*)(reginfo->strbeg)), |
| 6039 | (U8*) reginfo->strend), |
| 6040 | getSB_VAL_UTF8((U8*) locinput, |
| 6041 | (U8*) reginfo->strend), |
| 6042 | (U8*) reginfo->strbeg, |
| 6043 | (U8*) locinput, |
| 6044 | (U8*) reginfo->strend, |
| 6045 | utf8_target); |
| 6046 | } |
| 6047 | break; |
| 6048 | |
| 6049 | case WB_BOUND: |
| 6050 | if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { |
| 6051 | match = TRUE; |
| 6052 | } |
| 6053 | else { |
| 6054 | match = isWB(WB_UNKNOWN, |
| 6055 | getWB_VAL_UTF8( |
| 6056 | reghop3((U8*)locinput, |
| 6057 | -1, |
| 6058 | (U8*)(reginfo->strbeg)), |
| 6059 | (U8*) reginfo->strend), |
| 6060 | getWB_VAL_UTF8((U8*) locinput, |
| 6061 | (U8*) reginfo->strend), |
| 6062 | (U8*) reginfo->strbeg, |
| 6063 | (U8*) locinput, |
| 6064 | (U8*) reginfo->strend, |
| 6065 | utf8_target); |
| 6066 | } |
| 6067 | break; |
| 6068 | } |
| 6069 | } |
| 6070 | else { /* Not utf8 target */ |
| 6071 | switch((bound_type) FLAGS(scan)) { |
| 6072 | case TRADITIONAL_BOUND: |
| 6073 | { |
| 6074 | bool b1, b2; |
| 6075 | b1 = (locinput == reginfo->strbeg) |
| 6076 | ? 0 /* isWORDCHAR_L1('\n') */ |
| 6077 | : isWORDCHAR_L1(UCHARAT(locinput - 1)); |
| 6078 | b2 = (NEXTCHR_IS_EOS) |
| 6079 | ? 0 /* isWORDCHAR_L1('\n') */ |
| 6080 | : isWORDCHAR_L1(nextchr); |
| 6081 | match = cBOOL(b1 != b2); |
| 6082 | break; |
| 6083 | } |
| 6084 | |
| 6085 | case GCB_BOUND: |
| 6086 | if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { |
| 6087 | match = TRUE; /* GCB always matches at begin and |
| 6088 | end */ |
| 6089 | } |
| 6090 | else { /* Only CR-LF combo isn't a GCB in 0-255 |
| 6091 | range */ |
| 6092 | match = UCHARAT(locinput - 1) != '\r' |
| 6093 | || UCHARAT(locinput) != '\n'; |
| 6094 | } |
| 6095 | break; |
| 6096 | |
| 6097 | case LB_BOUND: |
| 6098 | if (locinput == reginfo->strbeg) { |
| 6099 | match = FALSE; |
| 6100 | } |
| 6101 | else if (NEXTCHR_IS_EOS) { |
| 6102 | match = TRUE; |
| 6103 | } |
| 6104 | else { |
| 6105 | match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)), |
| 6106 | getLB_VAL_CP(UCHARAT(locinput)), |
| 6107 | (U8*) reginfo->strbeg, |
| 6108 | (U8*) locinput, |
| 6109 | (U8*) reginfo->strend, |
| 6110 | utf8_target); |
| 6111 | } |
| 6112 | break; |
| 6113 | |
| 6114 | case SB_BOUND: /* Always matches at begin and end */ |
| 6115 | if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { |
| 6116 | match = TRUE; |
| 6117 | } |
| 6118 | else { |
| 6119 | match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)), |
| 6120 | getSB_VAL_CP(UCHARAT(locinput)), |
| 6121 | (U8*) reginfo->strbeg, |
| 6122 | (U8*) locinput, |
| 6123 | (U8*) reginfo->strend, |
| 6124 | utf8_target); |
| 6125 | } |
| 6126 | break; |
| 6127 | |
| 6128 | case WB_BOUND: |
| 6129 | if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { |
| 6130 | match = TRUE; |
| 6131 | } |
| 6132 | else { |
| 6133 | match = isWB(WB_UNKNOWN, |
| 6134 | getWB_VAL_CP(UCHARAT(locinput -1)), |
| 6135 | getWB_VAL_CP(UCHARAT(locinput)), |
| 6136 | (U8*) reginfo->strbeg, |
| 6137 | (U8*) locinput, |
| 6138 | (U8*) reginfo->strend, |
| 6139 | utf8_target); |
| 6140 | } |
| 6141 | break; |
| 6142 | } |
| 6143 | } |
| 6144 | |
| 6145 | if (to_complement ^ ! match) { |
| 6146 | sayNO; |
| 6147 | } |
| 6148 | break; |
| 6149 | |
| 6150 | case ANYOFL: /* /[abc]/l */ |
| 6151 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 6152 | |
| 6153 | if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE) |
| 6154 | { |
| 6155 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required); |
| 6156 | } |
| 6157 | /* FALLTHROUGH */ |
| 6158 | case ANYOFD: /* /[abc]/d */ |
| 6159 | case ANYOF: /* /[abc]/ */ |
| 6160 | if (NEXTCHR_IS_EOS) |
| 6161 | sayNO; |
| 6162 | if (utf8_target && ! UTF8_IS_INVARIANT(*locinput)) { |
| 6163 | if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend, |
| 6164 | utf8_target)) |
| 6165 | sayNO; |
| 6166 | locinput += UTF8SKIP(locinput); |
| 6167 | } |
| 6168 | else { |
| 6169 | if (!REGINCLASS(rex, scan, (U8*)locinput, utf8_target)) |
| 6170 | sayNO; |
| 6171 | locinput++; |
| 6172 | } |
| 6173 | break; |
| 6174 | |
| 6175 | /* The argument (FLAGS) to all the POSIX node types is the class number |
| 6176 | * */ |
| 6177 | |
| 6178 | case NPOSIXL: /* \W or [:^punct:] etc. under /l */ |
| 6179 | to_complement = 1; |
| 6180 | /* FALLTHROUGH */ |
| 6181 | |
| 6182 | case POSIXL: /* \w or [:punct:] etc. under /l */ |
| 6183 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 6184 | if (NEXTCHR_IS_EOS) |
| 6185 | sayNO; |
| 6186 | |
| 6187 | /* Use isFOO_lc() for characters within Latin1. (Note that |
| 6188 | * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else |
| 6189 | * wouldn't be invariant) */ |
| 6190 | if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { |
| 6191 | if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) { |
| 6192 | sayNO; |
| 6193 | } |
| 6194 | } |
| 6195 | else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { |
| 6196 | if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), |
| 6197 | EIGHT_BIT_UTF8_TO_NATIVE(nextchr, |
| 6198 | *(locinput + 1)))))) |
| 6199 | { |
| 6200 | sayNO; |
| 6201 | } |
| 6202 | } |
| 6203 | else { /* Here, must be an above Latin-1 code point */ |
| 6204 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend); |
| 6205 | goto utf8_posix_above_latin1; |
| 6206 | } |
| 6207 | |
| 6208 | /* Here, must be utf8 */ |
| 6209 | locinput += UTF8SKIP(locinput); |
| 6210 | break; |
| 6211 | |
| 6212 | case NPOSIXD: /* \W or [:^punct:] etc. under /d */ |
| 6213 | to_complement = 1; |
| 6214 | /* FALLTHROUGH */ |
| 6215 | |
| 6216 | case POSIXD: /* \w or [:punct:] etc. under /d */ |
| 6217 | if (utf8_target) { |
| 6218 | goto utf8_posix; |
| 6219 | } |
| 6220 | goto posixa; |
| 6221 | |
| 6222 | case NPOSIXA: /* \W or [:^punct:] etc. under /a */ |
| 6223 | |
| 6224 | if (NEXTCHR_IS_EOS) { |
| 6225 | sayNO; |
| 6226 | } |
| 6227 | |
| 6228 | /* All UTF-8 variants match */ |
| 6229 | if (! UTF8_IS_INVARIANT(nextchr)) { |
| 6230 | goto increment_locinput; |
| 6231 | } |
| 6232 | |
| 6233 | to_complement = 1; |
| 6234 | goto join_nposixa; |
| 6235 | |
| 6236 | case POSIXA: /* \w or [:punct:] etc. under /a */ |
| 6237 | |
| 6238 | posixa: |
| 6239 | /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in |
| 6240 | * UTF-8, and also from NPOSIXA even in UTF-8 when the current |
| 6241 | * character is a single byte */ |
| 6242 | |
| 6243 | if (NEXTCHR_IS_EOS) { |
| 6244 | sayNO; |
| 6245 | } |
| 6246 | |
| 6247 | join_nposixa: |
| 6248 | |
| 6249 | if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr, |
| 6250 | FLAGS(scan))))) |
| 6251 | { |
| 6252 | sayNO; |
| 6253 | } |
| 6254 | |
| 6255 | /* Here we are either not in utf8, or we matched a utf8-invariant, |
| 6256 | * so the next char is the next byte */ |
| 6257 | locinput++; |
| 6258 | break; |
| 6259 | |
| 6260 | case NPOSIXU: /* \W or [:^punct:] etc. under /u */ |
| 6261 | to_complement = 1; |
| 6262 | /* FALLTHROUGH */ |
| 6263 | |
| 6264 | case POSIXU: /* \w or [:punct:] etc. under /u */ |
| 6265 | utf8_posix: |
| 6266 | if (NEXTCHR_IS_EOS) { |
| 6267 | sayNO; |
| 6268 | } |
| 6269 | |
| 6270 | /* Use _generic_isCC() for characters within Latin1. (Note that |
| 6271 | * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else |
| 6272 | * wouldn't be invariant) */ |
| 6273 | if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { |
| 6274 | if (! (to_complement ^ cBOOL(_generic_isCC(nextchr, |
| 6275 | FLAGS(scan))))) |
| 6276 | { |
| 6277 | sayNO; |
| 6278 | } |
| 6279 | locinput++; |
| 6280 | } |
| 6281 | else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { |
| 6282 | if (! (to_complement |
| 6283 | ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr, |
| 6284 | *(locinput + 1)), |
| 6285 | FLAGS(scan))))) |
| 6286 | { |
| 6287 | sayNO; |
| 6288 | } |
| 6289 | locinput += 2; |
| 6290 | } |
| 6291 | else { /* Handle above Latin-1 code points */ |
| 6292 | utf8_posix_above_latin1: |
| 6293 | classnum = (_char_class_number) FLAGS(scan); |
| 6294 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 6295 | |
| 6296 | /* Here, uses a swash to find such code points. Load if if |
| 6297 | * not done already */ |
| 6298 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 6299 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 6300 | PL_utf8_swash_ptrs[classnum] |
| 6301 | = _core_swash_init("utf8", |
| 6302 | "", |
| 6303 | &PL_sv_undef, 1, 0, |
| 6304 | PL_XPosix_ptrs[classnum], &flags); |
| 6305 | } |
| 6306 | if (! (to_complement |
| 6307 | ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], |
| 6308 | (U8 *) locinput, TRUE)))) |
| 6309 | { |
| 6310 | sayNO; |
| 6311 | } |
| 6312 | } |
| 6313 | else { /* Here, uses macros to find above Latin-1 code points */ |
| 6314 | switch (classnum) { |
| 6315 | case _CC_ENUM_SPACE: |
| 6316 | if (! (to_complement |
| 6317 | ^ cBOOL(is_XPERLSPACE_high(locinput)))) |
| 6318 | { |
| 6319 | sayNO; |
| 6320 | } |
| 6321 | break; |
| 6322 | case _CC_ENUM_BLANK: |
| 6323 | if (! (to_complement |
| 6324 | ^ cBOOL(is_HORIZWS_high(locinput)))) |
| 6325 | { |
| 6326 | sayNO; |
| 6327 | } |
| 6328 | break; |
| 6329 | case _CC_ENUM_XDIGIT: |
| 6330 | if (! (to_complement |
| 6331 | ^ cBOOL(is_XDIGIT_high(locinput)))) |
| 6332 | { |
| 6333 | sayNO; |
| 6334 | } |
| 6335 | break; |
| 6336 | case _CC_ENUM_VERTSPACE: |
| 6337 | if (! (to_complement |
| 6338 | ^ cBOOL(is_VERTWS_high(locinput)))) |
| 6339 | { |
| 6340 | sayNO; |
| 6341 | } |
| 6342 | break; |
| 6343 | default: /* The rest, e.g. [:cntrl:], can't match |
| 6344 | above Latin1 */ |
| 6345 | if (! to_complement) { |
| 6346 | sayNO; |
| 6347 | } |
| 6348 | break; |
| 6349 | } |
| 6350 | } |
| 6351 | locinput += UTF8SKIP(locinput); |
| 6352 | } |
| 6353 | break; |
| 6354 | |
| 6355 | case CLUMP: /* Match \X: logical Unicode character. This is defined as |
| 6356 | a Unicode extended Grapheme Cluster */ |
| 6357 | if (NEXTCHR_IS_EOS) |
| 6358 | sayNO; |
| 6359 | if (! utf8_target) { |
| 6360 | |
| 6361 | /* Match either CR LF or '.', as all the other possibilities |
| 6362 | * require utf8 */ |
| 6363 | locinput++; /* Match the . or CR */ |
| 6364 | if (nextchr == '\r' /* And if it was CR, and the next is LF, |
| 6365 | match the LF */ |
| 6366 | && locinput < reginfo->strend |
| 6367 | && UCHARAT(locinput) == '\n') |
| 6368 | { |
| 6369 | locinput++; |
| 6370 | } |
| 6371 | } |
| 6372 | else { |
| 6373 | |
| 6374 | /* Get the gcb type for the current character */ |
| 6375 | GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput, |
| 6376 | (U8*) reginfo->strend); |
| 6377 | |
| 6378 | /* Then scan through the input until we get to the first |
| 6379 | * character whose type is supposed to be a gcb with the |
| 6380 | * current character. (There is always a break at the |
| 6381 | * end-of-input) */ |
| 6382 | locinput += UTF8SKIP(locinput); |
| 6383 | while (locinput < reginfo->strend) { |
| 6384 | GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput, |
| 6385 | (U8*) reginfo->strend); |
| 6386 | if (isGCB(prev_gcb, cur_gcb)) { |
| 6387 | break; |
| 6388 | } |
| 6389 | |
| 6390 | prev_gcb = cur_gcb; |
| 6391 | locinput += UTF8SKIP(locinput); |
| 6392 | } |
| 6393 | |
| 6394 | |
| 6395 | } |
| 6396 | break; |
| 6397 | |
| 6398 | case NREFFL: /* /\g{name}/il */ |
| 6399 | { /* The capture buffer cases. The ones beginning with N for the |
| 6400 | named buffers just convert to the equivalent numbered and |
| 6401 | pretend they were called as the corresponding numbered buffer |
| 6402 | op. */ |
| 6403 | /* don't initialize these in the declaration, it makes C++ |
| 6404 | unhappy */ |
| 6405 | const char *s; |
| 6406 | char type; |
| 6407 | re_fold_t folder; |
| 6408 | const U8 *fold_array; |
| 6409 | UV utf8_fold_flags; |
| 6410 | |
| 6411 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 6412 | folder = foldEQ_locale; |
| 6413 | fold_array = PL_fold_locale; |
| 6414 | type = REFFL; |
| 6415 | utf8_fold_flags = FOLDEQ_LOCALE; |
| 6416 | goto do_nref; |
| 6417 | |
| 6418 | case NREFFA: /* /\g{name}/iaa */ |
| 6419 | folder = foldEQ_latin1; |
| 6420 | fold_array = PL_fold_latin1; |
| 6421 | type = REFFA; |
| 6422 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 6423 | goto do_nref; |
| 6424 | |
| 6425 | case NREFFU: /* /\g{name}/iu */ |
| 6426 | folder = foldEQ_latin1; |
| 6427 | fold_array = PL_fold_latin1; |
| 6428 | type = REFFU; |
| 6429 | utf8_fold_flags = 0; |
| 6430 | goto do_nref; |
| 6431 | |
| 6432 | case NREFF: /* /\g{name}/i */ |
| 6433 | folder = foldEQ; |
| 6434 | fold_array = PL_fold; |
| 6435 | type = REFF; |
| 6436 | utf8_fold_flags = 0; |
| 6437 | goto do_nref; |
| 6438 | |
| 6439 | case NREF: /* /\g{name}/ */ |
| 6440 | type = REF; |
| 6441 | folder = NULL; |
| 6442 | fold_array = NULL; |
| 6443 | utf8_fold_flags = 0; |
| 6444 | do_nref: |
| 6445 | |
| 6446 | /* For the named back references, find the corresponding buffer |
| 6447 | * number */ |
| 6448 | n = reg_check_named_buff_matched(rex,scan); |
| 6449 | |
| 6450 | if ( ! n ) { |
| 6451 | sayNO; |
| 6452 | } |
| 6453 | goto do_nref_ref_common; |
| 6454 | |
| 6455 | case REFFL: /* /\1/il */ |
| 6456 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 6457 | folder = foldEQ_locale; |
| 6458 | fold_array = PL_fold_locale; |
| 6459 | utf8_fold_flags = FOLDEQ_LOCALE; |
| 6460 | goto do_ref; |
| 6461 | |
| 6462 | case REFFA: /* /\1/iaa */ |
| 6463 | folder = foldEQ_latin1; |
| 6464 | fold_array = PL_fold_latin1; |
| 6465 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 6466 | goto do_ref; |
| 6467 | |
| 6468 | case REFFU: /* /\1/iu */ |
| 6469 | folder = foldEQ_latin1; |
| 6470 | fold_array = PL_fold_latin1; |
| 6471 | utf8_fold_flags = 0; |
| 6472 | goto do_ref; |
| 6473 | |
| 6474 | case REFF: /* /\1/i */ |
| 6475 | folder = foldEQ; |
| 6476 | fold_array = PL_fold; |
| 6477 | utf8_fold_flags = 0; |
| 6478 | goto do_ref; |
| 6479 | |
| 6480 | case REF: /* /\1/ */ |
| 6481 | folder = NULL; |
| 6482 | fold_array = NULL; |
| 6483 | utf8_fold_flags = 0; |
| 6484 | |
| 6485 | do_ref: |
| 6486 | type = OP(scan); |
| 6487 | n = ARG(scan); /* which paren pair */ |
| 6488 | |
| 6489 | do_nref_ref_common: |
| 6490 | ln = rex->offs[n].start; |
| 6491 | reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ |
| 6492 | if (rex->lastparen < n || ln == -1) |
| 6493 | sayNO; /* Do not match unless seen CLOSEn. */ |
| 6494 | if (ln == rex->offs[n].end) |
| 6495 | break; |
| 6496 | |
| 6497 | s = reginfo->strbeg + ln; |
| 6498 | if (type != REF /* REF can do byte comparison */ |
| 6499 | && (utf8_target || type == REFFU || type == REFFL)) |
| 6500 | { |
| 6501 | char * limit = reginfo->strend; |
| 6502 | |
| 6503 | /* This call case insensitively compares the entire buffer |
| 6504 | * at s, with the current input starting at locinput, but |
| 6505 | * not going off the end given by reginfo->strend, and |
| 6506 | * returns in <limit> upon success, how much of the |
| 6507 | * current input was matched */ |
| 6508 | if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target, |
| 6509 | locinput, &limit, 0, utf8_target, utf8_fold_flags)) |
| 6510 | { |
| 6511 | sayNO; |
| 6512 | } |
| 6513 | locinput = limit; |
| 6514 | break; |
| 6515 | } |
| 6516 | |
| 6517 | /* Not utf8: Inline the first character, for speed. */ |
| 6518 | if (!NEXTCHR_IS_EOS && |
| 6519 | UCHARAT(s) != nextchr && |
| 6520 | (type == REF || |
| 6521 | UCHARAT(s) != fold_array[nextchr])) |
| 6522 | sayNO; |
| 6523 | ln = rex->offs[n].end - ln; |
| 6524 | if (locinput + ln > reginfo->strend) |
| 6525 | sayNO; |
| 6526 | if (ln > 1 && (type == REF |
| 6527 | ? memNE(s, locinput, ln) |
| 6528 | : ! folder(s, locinput, ln))) |
| 6529 | sayNO; |
| 6530 | locinput += ln; |
| 6531 | break; |
| 6532 | } |
| 6533 | |
| 6534 | case NOTHING: /* null op; e.g. the 'nothing' following |
| 6535 | * the '*' in m{(a+|b)*}' */ |
| 6536 | break; |
| 6537 | case TAIL: /* placeholder while compiling (A|B|C) */ |
| 6538 | break; |
| 6539 | |
| 6540 | #undef ST |
| 6541 | #define ST st->u.eval |
| 6542 | #define CUR_EVAL cur_eval->u.eval |
| 6543 | |
| 6544 | { |
| 6545 | SV *ret; |
| 6546 | REGEXP *re_sv; |
| 6547 | regexp *re; |
| 6548 | regexp_internal *rei; |
| 6549 | regnode *startpoint; |
| 6550 | U32 arg; |
| 6551 | |
| 6552 | case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */ |
| 6553 | arg= (U32)ARG(scan); |
| 6554 | if (cur_eval && cur_eval->locinput == locinput) { |
| 6555 | if ( ++nochange_depth > max_nochange_depth ) |
| 6556 | Perl_croak(aTHX_ |
| 6557 | "Pattern subroutine nesting without pos change" |
| 6558 | " exceeded limit in regex"); |
| 6559 | } else { |
| 6560 | nochange_depth = 0; |
| 6561 | } |
| 6562 | re_sv = rex_sv; |
| 6563 | re = rex; |
| 6564 | rei = rexi; |
| 6565 | startpoint = scan + ARG2L(scan); |
| 6566 | EVAL_CLOSE_PAREN_SET( st, arg ); |
| 6567 | /* Detect infinite recursion |
| 6568 | * |
| 6569 | * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/ |
| 6570 | * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever. |
| 6571 | * So we track the position in the string we are at each time |
| 6572 | * we recurse and if we try to enter the same routine twice from |
| 6573 | * the same position we throw an error. |
| 6574 | */ |
| 6575 | if ( rex->recurse_locinput[arg] == locinput ) { |
| 6576 | /* FIXME: we should show the regop that is failing as part |
| 6577 | * of the error message. */ |
| 6578 | Perl_croak(aTHX_ "Infinite recursion in regex"); |
| 6579 | } else { |
| 6580 | ST.prev_recurse_locinput= rex->recurse_locinput[arg]; |
| 6581 | rex->recurse_locinput[arg]= locinput; |
| 6582 | |
| 6583 | DEBUG_r({ |
| 6584 | GET_RE_DEBUG_FLAGS_DECL; |
| 6585 | DEBUG_STACK_r({ |
| 6586 | Perl_re_exec_indentf( aTHX_ |
| 6587 | "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n", |
| 6588 | depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg] |
| 6589 | ); |
| 6590 | }); |
| 6591 | }); |
| 6592 | } |
| 6593 | |
| 6594 | /* Save all the positions seen so far. */ |
| 6595 | ST.cp = regcppush(rex, 0, maxopenparen); |
| 6596 | REGCP_SET(ST.lastcp); |
| 6597 | |
| 6598 | /* and then jump to the code we share with EVAL */ |
| 6599 | goto eval_recurse_doit; |
| 6600 | /* NOTREACHED */ |
| 6601 | |
| 6602 | case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ |
| 6603 | if (cur_eval && cur_eval->locinput==locinput) { |
| 6604 | if ( ++nochange_depth > max_nochange_depth ) |
| 6605 | Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex"); |
| 6606 | } else { |
| 6607 | nochange_depth = 0; |
| 6608 | } |
| 6609 | { |
| 6610 | /* execute the code in the {...} */ |
| 6611 | |
| 6612 | dSP; |
| 6613 | IV before; |
| 6614 | OP * const oop = PL_op; |
| 6615 | COP * const ocurcop = PL_curcop; |
| 6616 | OP *nop; |
| 6617 | CV *newcv; |
| 6618 | |
| 6619 | /* save *all* paren positions */ |
| 6620 | regcppush(rex, 0, maxopenparen); |
| 6621 | REGCP_SET(runops_cp); |
| 6622 | |
| 6623 | if (!caller_cv) |
| 6624 | caller_cv = find_runcv(NULL); |
| 6625 | |
| 6626 | n = ARG(scan); |
| 6627 | |
| 6628 | if (rexi->data->what[n] == 'r') { /* code from an external qr */ |
| 6629 | newcv = (ReANY( |
| 6630 | (REGEXP*)(rexi->data->data[n]) |
| 6631 | ))->qr_anoncv; |
| 6632 | nop = (OP*)rexi->data->data[n+1]; |
| 6633 | } |
| 6634 | else if (rexi->data->what[n] == 'l') { /* literal code */ |
| 6635 | newcv = caller_cv; |
| 6636 | nop = (OP*)rexi->data->data[n]; |
| 6637 | assert(CvDEPTH(newcv)); |
| 6638 | } |
| 6639 | else { |
| 6640 | /* literal with own CV */ |
| 6641 | assert(rexi->data->what[n] == 'L'); |
| 6642 | newcv = rex->qr_anoncv; |
| 6643 | nop = (OP*)rexi->data->data[n]; |
| 6644 | } |
| 6645 | |
| 6646 | /* normally if we're about to execute code from the same |
| 6647 | * CV that we used previously, we just use the existing |
| 6648 | * CX stack entry. However, its possible that in the |
| 6649 | * meantime we may have backtracked, popped from the save |
| 6650 | * stack, and undone the SAVECOMPPAD(s) associated with |
| 6651 | * PUSH_MULTICALL; in which case PL_comppad no longer |
| 6652 | * points to newcv's pad. */ |
| 6653 | if (newcv != last_pushed_cv || PL_comppad != last_pad) |
| 6654 | { |
| 6655 | U8 flags = (CXp_SUB_RE | |
| 6656 | ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0)); |
| 6657 | if (last_pushed_cv) { |
| 6658 | /* PUSH/POP_MULTICALL save and restore the |
| 6659 | * caller's PL_comppad; if we call multiple subs |
| 6660 | * using the same CX block, we have to save and |
| 6661 | * unwind the varying PL_comppad's ourselves, |
| 6662 | * especially restoring the right PL_comppad on |
| 6663 | * backtrack - so save it on the save stack */ |
| 6664 | SAVECOMPPAD(); |
| 6665 | CHANGE_MULTICALL_FLAGS(newcv, flags); |
| 6666 | } |
| 6667 | else { |
| 6668 | PUSH_MULTICALL_FLAGS(newcv, flags); |
| 6669 | } |
| 6670 | last_pushed_cv = newcv; |
| 6671 | } |
| 6672 | else { |
| 6673 | /* these assignments are just to silence compiler |
| 6674 | * warnings */ |
| 6675 | multicall_cop = NULL; |
| 6676 | } |
| 6677 | last_pad = PL_comppad; |
| 6678 | |
| 6679 | /* the initial nextstate you would normally execute |
| 6680 | * at the start of an eval (which would cause error |
| 6681 | * messages to come from the eval), may be optimised |
| 6682 | * away from the execution path in the regex code blocks; |
| 6683 | * so manually set PL_curcop to it initially */ |
| 6684 | { |
| 6685 | OP *o = cUNOPx(nop)->op_first; |
| 6686 | assert(o->op_type == OP_NULL); |
| 6687 | if (o->op_targ == OP_SCOPE) { |
| 6688 | o = cUNOPo->op_first; |
| 6689 | } |
| 6690 | else { |
| 6691 | assert(o->op_targ == OP_LEAVE); |
| 6692 | o = cUNOPo->op_first; |
| 6693 | assert(o->op_type == OP_ENTER); |
| 6694 | o = OpSIBLING(o); |
| 6695 | } |
| 6696 | |
| 6697 | if (o->op_type != OP_STUB) { |
| 6698 | assert( o->op_type == OP_NEXTSTATE |
| 6699 | || o->op_type == OP_DBSTATE |
| 6700 | || (o->op_type == OP_NULL |
| 6701 | && ( o->op_targ == OP_NEXTSTATE |
| 6702 | || o->op_targ == OP_DBSTATE |
| 6703 | ) |
| 6704 | ) |
| 6705 | ); |
| 6706 | PL_curcop = (COP*)o; |
| 6707 | } |
| 6708 | } |
| 6709 | nop = nop->op_next; |
| 6710 | |
| 6711 | DEBUG_STATE_r( Perl_re_printf( aTHX_ |
| 6712 | " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) ); |
| 6713 | |
| 6714 | rex->offs[0].end = locinput - reginfo->strbeg; |
| 6715 | if (reginfo->info_aux_eval->pos_magic) |
| 6716 | MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic, |
| 6717 | reginfo->sv, reginfo->strbeg, |
| 6718 | locinput - reginfo->strbeg); |
| 6719 | |
| 6720 | if (sv_yes_mark) { |
| 6721 | SV *sv_mrk = get_sv("REGMARK", 1); |
| 6722 | sv_setsv(sv_mrk, sv_yes_mark); |
| 6723 | } |
| 6724 | |
| 6725 | /* we don't use MULTICALL here as we want to call the |
| 6726 | * first op of the block of interest, rather than the |
| 6727 | * first op of the sub. Also, we don't want to free |
| 6728 | * the savestack frame */ |
| 6729 | before = (IV)(SP-PL_stack_base); |
| 6730 | PL_op = nop; |
| 6731 | CALLRUNOPS(aTHX); /* Scalar context. */ |
| 6732 | SPAGAIN; |
| 6733 | if ((IV)(SP-PL_stack_base) == before) |
| 6734 | ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */ |
| 6735 | else { |
| 6736 | ret = POPs; |
| 6737 | PUTBACK; |
| 6738 | } |
| 6739 | |
| 6740 | /* before restoring everything, evaluate the returned |
| 6741 | * value, so that 'uninit' warnings don't use the wrong |
| 6742 | * PL_op or pad. Also need to process any magic vars |
| 6743 | * (e.g. $1) *before* parentheses are restored */ |
| 6744 | |
| 6745 | PL_op = NULL; |
| 6746 | |
| 6747 | re_sv = NULL; |
| 6748 | if (logical == 0) /* (?{})/ */ |
| 6749 | sv_setsv(save_scalar(PL_replgv), ret); /* $^R */ |
| 6750 | else if (logical == 1) { /* /(?(?{...})X|Y)/ */ |
| 6751 | sw = cBOOL(SvTRUE(ret)); |
| 6752 | logical = 0; |
| 6753 | } |
| 6754 | else { /* /(??{}) */ |
| 6755 | /* if its overloaded, let the regex compiler handle |
| 6756 | * it; otherwise extract regex, or stringify */ |
| 6757 | if (SvGMAGICAL(ret)) |
| 6758 | ret = sv_mortalcopy(ret); |
| 6759 | if (!SvAMAGIC(ret)) { |
| 6760 | SV *sv = ret; |
| 6761 | if (SvROK(sv)) |
| 6762 | sv = SvRV(sv); |
| 6763 | if (SvTYPE(sv) == SVt_REGEXP) |
| 6764 | re_sv = (REGEXP*) sv; |
| 6765 | else if (SvSMAGICAL(ret)) { |
| 6766 | MAGIC *mg = mg_find(ret, PERL_MAGIC_qr); |
| 6767 | if (mg) |
| 6768 | re_sv = (REGEXP *) mg->mg_obj; |
| 6769 | } |
| 6770 | |
| 6771 | /* force any undef warnings here */ |
| 6772 | if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) { |
| 6773 | ret = sv_mortalcopy(ret); |
| 6774 | (void) SvPV_force_nolen(ret); |
| 6775 | } |
| 6776 | } |
| 6777 | |
| 6778 | } |
| 6779 | |
| 6780 | /* *** Note that at this point we don't restore |
| 6781 | * PL_comppad, (or pop the CxSUB) on the assumption it may |
| 6782 | * be used again soon. This is safe as long as nothing |
| 6783 | * in the regexp code uses the pad ! */ |
| 6784 | PL_op = oop; |
| 6785 | PL_curcop = ocurcop; |
| 6786 | S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen); |
| 6787 | PL_curpm = PL_reg_curpm; |
| 6788 | |
| 6789 | if (logical != 2) |
| 6790 | break; |
| 6791 | } |
| 6792 | |
| 6793 | /* only /(??{})/ from now on */ |
| 6794 | logical = 0; |
| 6795 | { |
| 6796 | /* extract RE object from returned value; compiling if |
| 6797 | * necessary */ |
| 6798 | |
| 6799 | if (re_sv) { |
| 6800 | re_sv = reg_temp_copy(NULL, re_sv); |
| 6801 | } |
| 6802 | else { |
| 6803 | U32 pm_flags = 0; |
| 6804 | |
| 6805 | if (SvUTF8(ret) && IN_BYTES) { |
| 6806 | /* In use 'bytes': make a copy of the octet |
| 6807 | * sequence, but without the flag on */ |
| 6808 | STRLEN len; |
| 6809 | const char *const p = SvPV(ret, len); |
| 6810 | ret = newSVpvn_flags(p, len, SVs_TEMP); |
| 6811 | } |
| 6812 | if (rex->intflags & PREGf_USE_RE_EVAL) |
| 6813 | pm_flags |= PMf_USE_RE_EVAL; |
| 6814 | |
| 6815 | /* if we got here, it should be an engine which |
| 6816 | * supports compiling code blocks and stuff */ |
| 6817 | assert(rex->engine && rex->engine->op_comp); |
| 6818 | assert(!(scan->flags & ~RXf_PMf_COMPILETIME)); |
| 6819 | re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL, |
| 6820 | rex->engine, NULL, NULL, |
| 6821 | /* copy /msixn etc to inner pattern */ |
| 6822 | ARG2L(scan), |
| 6823 | pm_flags); |
| 6824 | |
| 6825 | if (!(SvFLAGS(ret) |
| 6826 | & (SVs_TEMP | SVs_GMG | SVf_ROK)) |
| 6827 | && (!SvPADTMP(ret) || SvREADONLY(ret))) { |
| 6828 | /* This isn't a first class regexp. Instead, it's |
| 6829 | caching a regexp onto an existing, Perl visible |
| 6830 | scalar. */ |
| 6831 | sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0); |
| 6832 | } |
| 6833 | } |
| 6834 | SAVEFREESV(re_sv); |
| 6835 | re = ReANY(re_sv); |
| 6836 | } |
| 6837 | RXp_MATCH_COPIED_off(re); |
| 6838 | re->subbeg = rex->subbeg; |
| 6839 | re->sublen = rex->sublen; |
| 6840 | re->suboffset = rex->suboffset; |
| 6841 | re->subcoffset = rex->subcoffset; |
| 6842 | re->lastparen = 0; |
| 6843 | re->lastcloseparen = 0; |
| 6844 | rei = RXi_GET(re); |
| 6845 | DEBUG_EXECUTE_r( |
| 6846 | debug_start_match(re_sv, utf8_target, locinput, |
| 6847 | reginfo->strend, "Matching embedded"); |
| 6848 | ); |
| 6849 | startpoint = rei->program + 1; |
| 6850 | EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0; |
| 6851 | * close_paren only for GOSUB */ |
| 6852 | ST.prev_recurse_locinput= NULL; /* only used for GOSUB */ |
| 6853 | /* Save all the seen positions so far. */ |
| 6854 | ST.cp = regcppush(rex, 0, maxopenparen); |
| 6855 | REGCP_SET(ST.lastcp); |
| 6856 | /* and set maxopenparen to 0, since we are starting a "fresh" match */ |
| 6857 | maxopenparen = 0; |
| 6858 | /* run the pattern returned from (??{...}) */ |
| 6859 | |
| 6860 | eval_recurse_doit: /* Share code with GOSUB below this line |
| 6861 | * At this point we expect the stack context to be |
| 6862 | * set up correctly */ |
| 6863 | |
| 6864 | /* invalidate the S-L poscache. We're now executing a |
| 6865 | * different set of WHILEM ops (and their associated |
| 6866 | * indexes) against the same string, so the bits in the |
| 6867 | * cache are meaningless. Setting maxiter to zero forces |
| 6868 | * the cache to be invalidated and zeroed before reuse. |
| 6869 | * XXX This is too dramatic a measure. Ideally we should |
| 6870 | * save the old cache and restore when running the outer |
| 6871 | * pattern again */ |
| 6872 | reginfo->poscache_maxiter = 0; |
| 6873 | |
| 6874 | /* the new regexp might have a different is_utf8_pat than we do */ |
| 6875 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv)); |
| 6876 | |
| 6877 | ST.prev_rex = rex_sv; |
| 6878 | ST.prev_curlyx = cur_curlyx; |
| 6879 | rex_sv = re_sv; |
| 6880 | SET_reg_curpm(rex_sv); |
| 6881 | rex = re; |
| 6882 | rexi = rei; |
| 6883 | cur_curlyx = NULL; |
| 6884 | ST.B = next; |
| 6885 | ST.prev_eval = cur_eval; |
| 6886 | cur_eval = st; |
| 6887 | /* now continue from first node in postoned RE */ |
| 6888 | PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput); |
| 6889 | NOT_REACHED; /* NOTREACHED */ |
| 6890 | } |
| 6891 | |
| 6892 | case EVAL_AB: /* cleanup after a successful (??{A})B */ |
| 6893 | /* note: this is called twice; first after popping B, then A */ |
| 6894 | DEBUG_STACK_r({ |
| 6895 | Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n", |
| 6896 | depth, cur_eval, ST.prev_eval); |
| 6897 | }); |
| 6898 | |
| 6899 | #define SET_RECURSE_LOCINPUT(STR,VAL)\ |
| 6900 | if ( cur_eval && CUR_EVAL.close_paren ) {\ |
| 6901 | DEBUG_STACK_r({ \ |
| 6902 | Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\ |
| 6903 | depth, \ |
| 6904 | CUR_EVAL.close_paren - 1,\ |
| 6905 | cur_eval, \ |
| 6906 | VAL); \ |
| 6907 | }); \ |
| 6908 | rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\ |
| 6909 | } |
| 6910 | |
| 6911 | SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput); |
| 6912 | |
| 6913 | rex_sv = ST.prev_rex; |
| 6914 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); |
| 6915 | SET_reg_curpm(rex_sv); |
| 6916 | rex = ReANY(rex_sv); |
| 6917 | rexi = RXi_GET(rex); |
| 6918 | { |
| 6919 | /* preserve $^R across LEAVE's. See Bug 121070. */ |
| 6920 | SV *save_sv= GvSV(PL_replgv); |
| 6921 | SvREFCNT_inc(save_sv); |
| 6922 | regcpblow(ST.cp); /* LEAVE in disguise */ |
| 6923 | sv_setsv(GvSV(PL_replgv), save_sv); |
| 6924 | SvREFCNT_dec(save_sv); |
| 6925 | } |
| 6926 | cur_eval = ST.prev_eval; |
| 6927 | cur_curlyx = ST.prev_curlyx; |
| 6928 | |
| 6929 | /* Invalidate cache. See "invalidate" comment above. */ |
| 6930 | reginfo->poscache_maxiter = 0; |
| 6931 | if ( nochange_depth ) |
| 6932 | nochange_depth--; |
| 6933 | |
| 6934 | SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput); |
| 6935 | sayYES; |
| 6936 | |
| 6937 | |
| 6938 | case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ |
| 6939 | /* note: this is called twice; first after popping B, then A */ |
| 6940 | DEBUG_STACK_r({ |
| 6941 | Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n", |
| 6942 | depth, cur_eval, ST.prev_eval); |
| 6943 | }); |
| 6944 | |
| 6945 | SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput); |
| 6946 | |
| 6947 | rex_sv = ST.prev_rex; |
| 6948 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); |
| 6949 | SET_reg_curpm(rex_sv); |
| 6950 | rex = ReANY(rex_sv); |
| 6951 | rexi = RXi_GET(rex); |
| 6952 | |
| 6953 | REGCP_UNWIND(ST.lastcp); |
| 6954 | regcppop(rex, &maxopenparen); |
| 6955 | cur_eval = ST.prev_eval; |
| 6956 | cur_curlyx = ST.prev_curlyx; |
| 6957 | |
| 6958 | /* Invalidate cache. See "invalidate" comment above. */ |
| 6959 | reginfo->poscache_maxiter = 0; |
| 6960 | if ( nochange_depth ) |
| 6961 | nochange_depth--; |
| 6962 | |
| 6963 | SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput); |
| 6964 | sayNO_SILENT; |
| 6965 | #undef ST |
| 6966 | |
| 6967 | case OPEN: /* ( */ |
| 6968 | n = ARG(scan); /* which paren pair */ |
| 6969 | rex->offs[n].start_tmp = locinput - reginfo->strbeg; |
| 6970 | if (n > maxopenparen) |
| 6971 | maxopenparen = n; |
| 6972 | DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ |
| 6973 | "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n", |
| 6974 | PTR2UV(rex), |
| 6975 | PTR2UV(rex->offs), |
| 6976 | (UV)n, |
| 6977 | (IV)rex->offs[n].start_tmp, |
| 6978 | (UV)maxopenparen |
| 6979 | )); |
| 6980 | lastopen = n; |
| 6981 | break; |
| 6982 | |
| 6983 | /* XXX really need to log other places start/end are set too */ |
| 6984 | #define CLOSE_CAPTURE \ |
| 6985 | rex->offs[n].start = rex->offs[n].start_tmp; \ |
| 6986 | rex->offs[n].end = locinput - reginfo->strbeg; \ |
| 6987 | DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ \ |
| 6988 | "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \ |
| 6989 | PTR2UV(rex), \ |
| 6990 | PTR2UV(rex->offs), \ |
| 6991 | (UV)n, \ |
| 6992 | (IV)rex->offs[n].start, \ |
| 6993 | (IV)rex->offs[n].end \ |
| 6994 | )) |
| 6995 | |
| 6996 | case CLOSE: /* ) */ |
| 6997 | n = ARG(scan); /* which paren pair */ |
| 6998 | CLOSE_CAPTURE; |
| 6999 | if (n > rex->lastparen) |
| 7000 | rex->lastparen = n; |
| 7001 | rex->lastcloseparen = n; |
| 7002 | if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) ) |
| 7003 | goto fake_end; |
| 7004 | |
| 7005 | break; |
| 7006 | |
| 7007 | case ACCEPT: /* (*ACCEPT) */ |
| 7008 | if (scan->flags) |
| 7009 | sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 7010 | if (ARG2L(scan)){ |
| 7011 | regnode *cursor; |
| 7012 | for (cursor=scan; |
| 7013 | cursor && OP(cursor)!=END; |
| 7014 | cursor=regnext(cursor)) |
| 7015 | { |
| 7016 | if ( OP(cursor)==CLOSE ){ |
| 7017 | n = ARG(cursor); |
| 7018 | if ( n <= lastopen ) { |
| 7019 | CLOSE_CAPTURE; |
| 7020 | if (n > rex->lastparen) |
| 7021 | rex->lastparen = n; |
| 7022 | rex->lastcloseparen = n; |
| 7023 | if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) ) |
| 7024 | break; |
| 7025 | } |
| 7026 | } |
| 7027 | } |
| 7028 | } |
| 7029 | goto fake_end; |
| 7030 | /* NOTREACHED */ |
| 7031 | |
| 7032 | case GROUPP: /* (?(1)) */ |
| 7033 | n = ARG(scan); /* which paren pair */ |
| 7034 | sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1); |
| 7035 | break; |
| 7036 | |
| 7037 | case NGROUPP: /* (?(<name>)) */ |
| 7038 | /* reg_check_named_buff_matched returns 0 for no match */ |
| 7039 | sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan)); |
| 7040 | break; |
| 7041 | |
| 7042 | case INSUBP: /* (?(R)) */ |
| 7043 | n = ARG(scan); |
| 7044 | /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg |
| 7045 | * of SCAN is already set up as matches a eval.close_paren */ |
| 7046 | sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n); |
| 7047 | break; |
| 7048 | |
| 7049 | case DEFINEP: /* (?(DEFINE)) */ |
| 7050 | sw = 0; |
| 7051 | break; |
| 7052 | |
| 7053 | case IFTHEN: /* (?(cond)A|B) */ |
| 7054 | reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ |
| 7055 | if (sw) |
| 7056 | next = NEXTOPER(NEXTOPER(scan)); |
| 7057 | else { |
| 7058 | next = scan + ARG(scan); |
| 7059 | if (OP(next) == IFTHEN) /* Fake one. */ |
| 7060 | next = NEXTOPER(NEXTOPER(next)); |
| 7061 | } |
| 7062 | break; |
| 7063 | |
| 7064 | case LOGICAL: /* modifier for EVAL and IFMATCH */ |
| 7065 | logical = scan->flags; |
| 7066 | break; |
| 7067 | |
| 7068 | /******************************************************************* |
| 7069 | |
| 7070 | The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ |
| 7071 | pattern, where A and B are subpatterns. (For simple A, CURLYM or |
| 7072 | STAR/PLUS/CURLY/CURLYN are used instead.) |
| 7073 | |
| 7074 | A*B is compiled as <CURLYX><A><WHILEM><B> |
| 7075 | |
| 7076 | On entry to the subpattern, CURLYX is called. This pushes a CURLYX |
| 7077 | state, which contains the current count, initialised to -1. It also sets |
| 7078 | cur_curlyx to point to this state, with any previous value saved in the |
| 7079 | state block. |
| 7080 | |
| 7081 | CURLYX then jumps straight to the WHILEM op, rather than executing A, |
| 7082 | since the pattern may possibly match zero times (i.e. it's a while {} loop |
| 7083 | rather than a do {} while loop). |
| 7084 | |
| 7085 | Each entry to WHILEM represents a successful match of A. The count in the |
| 7086 | CURLYX block is incremented, another WHILEM state is pushed, and execution |
| 7087 | passes to A or B depending on greediness and the current count. |
| 7088 | |
| 7089 | For example, if matching against the string a1a2a3b (where the aN are |
| 7090 | substrings that match /A/), then the match progresses as follows: (the |
| 7091 | pushed states are interspersed with the bits of strings matched so far): |
| 7092 | |
| 7093 | <CURLYX cnt=-1> |
| 7094 | <CURLYX cnt=0><WHILEM> |
| 7095 | <CURLYX cnt=1><WHILEM> a1 <WHILEM> |
| 7096 | <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM> |
| 7097 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> |
| 7098 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b |
| 7099 | |
| 7100 | (Contrast this with something like CURLYM, which maintains only a single |
| 7101 | backtrack state: |
| 7102 | |
| 7103 | <CURLYM cnt=0> a1 |
| 7104 | a1 <CURLYM cnt=1> a2 |
| 7105 | a1 a2 <CURLYM cnt=2> a3 |
| 7106 | a1 a2 a3 <CURLYM cnt=3> b |
| 7107 | ) |
| 7108 | |
| 7109 | Each WHILEM state block marks a point to backtrack to upon partial failure |
| 7110 | of A or B, and also contains some minor state data related to that |
| 7111 | iteration. The CURLYX block, pointed to by cur_curlyx, contains the |
| 7112 | overall state, such as the count, and pointers to the A and B ops. |
| 7113 | |
| 7114 | This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx |
| 7115 | must always point to the *current* CURLYX block, the rules are: |
| 7116 | |
| 7117 | When executing CURLYX, save the old cur_curlyx in the CURLYX state block, |
| 7118 | and set cur_curlyx to point the new block. |
| 7119 | |
| 7120 | When popping the CURLYX block after a successful or unsuccessful match, |
| 7121 | restore the previous cur_curlyx. |
| 7122 | |
| 7123 | When WHILEM is about to execute B, save the current cur_curlyx, and set it |
| 7124 | to the outer one saved in the CURLYX block. |
| 7125 | |
| 7126 | When popping the WHILEM block after a successful or unsuccessful B match, |
| 7127 | restore the previous cur_curlyx. |
| 7128 | |
| 7129 | Here's an example for the pattern (AI* BI)*BO |
| 7130 | I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: |
| 7131 | |
| 7132 | cur_ |
| 7133 | curlyx backtrack stack |
| 7134 | ------ --------------- |
| 7135 | NULL |
| 7136 | CO <CO prev=NULL> <WO> |
| 7137 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai |
| 7138 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi |
| 7139 | NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo |
| 7140 | |
| 7141 | At this point the pattern succeeds, and we work back down the stack to |
| 7142 | clean up, restoring as we go: |
| 7143 | |
| 7144 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi |
| 7145 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai |
| 7146 | CO <CO prev=NULL> <WO> |
| 7147 | NULL |
| 7148 | |
| 7149 | *******************************************************************/ |
| 7150 | |
| 7151 | #define ST st->u.curlyx |
| 7152 | |
| 7153 | case CURLYX: /* start of /A*B/ (for complex A) */ |
| 7154 | { |
| 7155 | /* No need to save/restore up to this paren */ |
| 7156 | I32 parenfloor = scan->flags; |
| 7157 | |
| 7158 | assert(next); /* keep Coverity happy */ |
| 7159 | if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ |
| 7160 | next += ARG(next); |
| 7161 | |
| 7162 | /* XXXX Probably it is better to teach regpush to support |
| 7163 | parenfloor > maxopenparen ... */ |
| 7164 | if (parenfloor > (I32)rex->lastparen) |
| 7165 | parenfloor = rex->lastparen; /* Pessimization... */ |
| 7166 | |
| 7167 | ST.prev_curlyx= cur_curlyx; |
| 7168 | cur_curlyx = st; |
| 7169 | ST.cp = PL_savestack_ix; |
| 7170 | |
| 7171 | /* these fields contain the state of the current curly. |
| 7172 | * they are accessed by subsequent WHILEMs */ |
| 7173 | ST.parenfloor = parenfloor; |
| 7174 | ST.me = scan; |
| 7175 | ST.B = next; |
| 7176 | ST.minmod = minmod; |
| 7177 | minmod = 0; |
| 7178 | ST.count = -1; /* this will be updated by WHILEM */ |
| 7179 | ST.lastloc = NULL; /* this will be updated by WHILEM */ |
| 7180 | |
| 7181 | PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput); |
| 7182 | NOT_REACHED; /* NOTREACHED */ |
| 7183 | } |
| 7184 | |
| 7185 | case CURLYX_end: /* just finished matching all of A*B */ |
| 7186 | cur_curlyx = ST.prev_curlyx; |
| 7187 | sayYES; |
| 7188 | NOT_REACHED; /* NOTREACHED */ |
| 7189 | |
| 7190 | case CURLYX_end_fail: /* just failed to match all of A*B */ |
| 7191 | regcpblow(ST.cp); |
| 7192 | cur_curlyx = ST.prev_curlyx; |
| 7193 | sayNO; |
| 7194 | NOT_REACHED; /* NOTREACHED */ |
| 7195 | |
| 7196 | |
| 7197 | #undef ST |
| 7198 | #define ST st->u.whilem |
| 7199 | |
| 7200 | case WHILEM: /* just matched an A in /A*B/ (for complex A) */ |
| 7201 | { |
| 7202 | /* see the discussion above about CURLYX/WHILEM */ |
| 7203 | I32 n; |
| 7204 | int min, max; |
| 7205 | regnode *A; |
| 7206 | |
| 7207 | assert(cur_curlyx); /* keep Coverity happy */ |
| 7208 | |
| 7209 | min = ARG1(cur_curlyx->u.curlyx.me); |
| 7210 | max = ARG2(cur_curlyx->u.curlyx.me); |
| 7211 | A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS; |
| 7212 | n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ |
| 7213 | ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; |
| 7214 | ST.cache_offset = 0; |
| 7215 | ST.cache_mask = 0; |
| 7216 | |
| 7217 | |
| 7218 | DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: matched %ld out of %d..%d\n", |
| 7219 | depth, (long)n, min, max) |
| 7220 | ); |
| 7221 | |
| 7222 | /* First just match a string of min A's. */ |
| 7223 | |
| 7224 | if (n < min) { |
| 7225 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
| 7226 | maxopenparen); |
| 7227 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 7228 | REGCP_SET(ST.lastcp); |
| 7229 | |
| 7230 | PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput); |
| 7231 | NOT_REACHED; /* NOTREACHED */ |
| 7232 | } |
| 7233 | |
| 7234 | /* If degenerate A matches "", assume A done. */ |
| 7235 | |
| 7236 | if (locinput == cur_curlyx->u.curlyx.lastloc) { |
| 7237 | DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: empty match detected, trying continuation...\n", |
| 7238 | depth) |
| 7239 | ); |
| 7240 | goto do_whilem_B_max; |
| 7241 | } |
| 7242 | |
| 7243 | /* super-linear cache processing. |
| 7244 | * |
| 7245 | * The idea here is that for certain types of CURLYX/WHILEM - |
| 7246 | * principally those whose upper bound is infinity (and |
| 7247 | * excluding regexes that have things like \1 and other very |
| 7248 | * non-regular expresssiony things), then if a pattern like |
| 7249 | * /....A*.../ fails and we backtrack to the WHILEM, then we |
| 7250 | * make a note that this particular WHILEM op was at string |
| 7251 | * position 47 (say) when the rest of pattern failed. Then, if |
| 7252 | * we ever find ourselves back at that WHILEM, and at string |
| 7253 | * position 47 again, we can just fail immediately rather than |
| 7254 | * running the rest of the pattern again. |
| 7255 | * |
| 7256 | * This is very handy when patterns start to go |
| 7257 | * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up |
| 7258 | * with a combinatorial explosion of backtracking. |
| 7259 | * |
| 7260 | * The cache is implemented as a bit array, with one bit per |
| 7261 | * string byte position per WHILEM op (up to 16) - so its |
| 7262 | * between 0.25 and 2x the string size. |
| 7263 | * |
| 7264 | * To avoid allocating a poscache buffer every time, we do an |
| 7265 | * initially countdown; only after we have executed a WHILEM |
| 7266 | * op (string-length x #WHILEMs) times do we allocate the |
| 7267 | * cache. |
| 7268 | * |
| 7269 | * The top 4 bits of scan->flags byte say how many different |
| 7270 | * relevant CURLLYX/WHILEM op pairs there are, while the |
| 7271 | * bottom 4-bits is the identifying index number of this |
| 7272 | * WHILEM. |
| 7273 | */ |
| 7274 | |
| 7275 | if (scan->flags) { |
| 7276 | |
| 7277 | if (!reginfo->poscache_maxiter) { |
| 7278 | /* start the countdown: Postpone detection until we |
| 7279 | * know the match is not *that* much linear. */ |
| 7280 | reginfo->poscache_maxiter |
| 7281 | = (reginfo->strend - reginfo->strbeg + 1) |
| 7282 | * (scan->flags>>4); |
| 7283 | /* possible overflow for long strings and many CURLYX's */ |
| 7284 | if (reginfo->poscache_maxiter < 0) |
| 7285 | reginfo->poscache_maxiter = I32_MAX; |
| 7286 | reginfo->poscache_iter = reginfo->poscache_maxiter; |
| 7287 | } |
| 7288 | |
| 7289 | if (reginfo->poscache_iter-- == 0) { |
| 7290 | /* initialise cache */ |
| 7291 | const SSize_t size = (reginfo->poscache_maxiter + 7)/8; |
| 7292 | regmatch_info_aux *const aux = reginfo->info_aux; |
| 7293 | if (aux->poscache) { |
| 7294 | if ((SSize_t)reginfo->poscache_size < size) { |
| 7295 | Renew(aux->poscache, size, char); |
| 7296 | reginfo->poscache_size = size; |
| 7297 | } |
| 7298 | Zero(aux->poscache, size, char); |
| 7299 | } |
| 7300 | else { |
| 7301 | reginfo->poscache_size = size; |
| 7302 | Newxz(aux->poscache, size, char); |
| 7303 | } |
| 7304 | DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ |
| 7305 | "%swhilem: Detected a super-linear match, switching on caching%s...\n", |
| 7306 | PL_colors[4], PL_colors[5]) |
| 7307 | ); |
| 7308 | } |
| 7309 | |
| 7310 | if (reginfo->poscache_iter < 0) { |
| 7311 | /* have we already failed at this position? */ |
| 7312 | SSize_t offset, mask; |
| 7313 | |
| 7314 | reginfo->poscache_iter = -1; /* stop eventual underflow */ |
| 7315 | offset = (scan->flags & 0xf) - 1 |
| 7316 | + (locinput - reginfo->strbeg) |
| 7317 | * (scan->flags>>4); |
| 7318 | mask = 1 << (offset % 8); |
| 7319 | offset /= 8; |
| 7320 | if (reginfo->info_aux->poscache[offset] & mask) { |
| 7321 | DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: (cache) already tried at this position...\n", |
| 7322 | depth) |
| 7323 | ); |
| 7324 | sayNO; /* cache records failure */ |
| 7325 | } |
| 7326 | ST.cache_offset = offset; |
| 7327 | ST.cache_mask = mask; |
| 7328 | } |
| 7329 | } |
| 7330 | |
| 7331 | /* Prefer B over A for minimal matching. */ |
| 7332 | |
| 7333 | if (cur_curlyx->u.curlyx.minmod) { |
| 7334 | ST.save_curlyx = cur_curlyx; |
| 7335 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; |
| 7336 | ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor, |
| 7337 | maxopenparen); |
| 7338 | REGCP_SET(ST.lastcp); |
| 7339 | PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B, |
| 7340 | locinput); |
| 7341 | NOT_REACHED; /* NOTREACHED */ |
| 7342 | } |
| 7343 | |
| 7344 | /* Prefer A over B for maximal matching. */ |
| 7345 | |
| 7346 | if (n < max) { /* More greed allowed? */ |
| 7347 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
| 7348 | maxopenparen); |
| 7349 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 7350 | REGCP_SET(ST.lastcp); |
| 7351 | PUSH_STATE_GOTO(WHILEM_A_max, A, locinput); |
| 7352 | NOT_REACHED; /* NOTREACHED */ |
| 7353 | } |
| 7354 | goto do_whilem_B_max; |
| 7355 | } |
| 7356 | NOT_REACHED; /* NOTREACHED */ |
| 7357 | |
| 7358 | case WHILEM_B_min: /* just matched B in a minimal match */ |
| 7359 | case WHILEM_B_max: /* just matched B in a maximal match */ |
| 7360 | cur_curlyx = ST.save_curlyx; |
| 7361 | sayYES; |
| 7362 | NOT_REACHED; /* NOTREACHED */ |
| 7363 | |
| 7364 | case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ |
| 7365 | cur_curlyx = ST.save_curlyx; |
| 7366 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; |
| 7367 | cur_curlyx->u.curlyx.count--; |
| 7368 | CACHEsayNO; |
| 7369 | NOT_REACHED; /* NOTREACHED */ |
| 7370 | |
| 7371 | case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ |
| 7372 | /* FALLTHROUGH */ |
| 7373 | case WHILEM_A_pre_fail: /* just failed to match even minimal A */ |
| 7374 | REGCP_UNWIND(ST.lastcp); |
| 7375 | regcppop(rex, &maxopenparen); |
| 7376 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; |
| 7377 | cur_curlyx->u.curlyx.count--; |
| 7378 | CACHEsayNO; |
| 7379 | NOT_REACHED; /* NOTREACHED */ |
| 7380 | |
| 7381 | case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ |
| 7382 | REGCP_UNWIND(ST.lastcp); |
| 7383 | regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */ |
| 7384 | DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "whilem: failed, trying continuation...\n", |
| 7385 | depth) |
| 7386 | ); |
| 7387 | do_whilem_B_max: |
| 7388 | if (cur_curlyx->u.curlyx.count >= REG_INFTY |
| 7389 | && ckWARN(WARN_REGEXP) |
| 7390 | && !reginfo->warned) |
| 7391 | { |
| 7392 | reginfo->warned = TRUE; |
| 7393 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), |
| 7394 | "Complex regular subexpression recursion limit (%d) " |
| 7395 | "exceeded", |
| 7396 | REG_INFTY - 1); |
| 7397 | } |
| 7398 | |
| 7399 | /* now try B */ |
| 7400 | ST.save_curlyx = cur_curlyx; |
| 7401 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; |
| 7402 | PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B, |
| 7403 | locinput); |
| 7404 | NOT_REACHED; /* NOTREACHED */ |
| 7405 | |
| 7406 | case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ |
| 7407 | cur_curlyx = ST.save_curlyx; |
| 7408 | REGCP_UNWIND(ST.lastcp); |
| 7409 | regcppop(rex, &maxopenparen); |
| 7410 | |
| 7411 | if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) { |
| 7412 | /* Maximum greed exceeded */ |
| 7413 | if (cur_curlyx->u.curlyx.count >= REG_INFTY |
| 7414 | && ckWARN(WARN_REGEXP) |
| 7415 | && !reginfo->warned) |
| 7416 | { |
| 7417 | reginfo->warned = TRUE; |
| 7418 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), |
| 7419 | "Complex regular subexpression recursion " |
| 7420 | "limit (%d) exceeded", |
| 7421 | REG_INFTY - 1); |
| 7422 | } |
| 7423 | cur_curlyx->u.curlyx.count--; |
| 7424 | CACHEsayNO; |
| 7425 | } |
| 7426 | |
| 7427 | DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "trying longer...\n", depth) |
| 7428 | ); |
| 7429 | /* Try grabbing another A and see if it helps. */ |
| 7430 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 7431 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
| 7432 | maxopenparen); |
| 7433 | REGCP_SET(ST.lastcp); |
| 7434 | PUSH_STATE_GOTO(WHILEM_A_min, |
| 7435 | /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS, |
| 7436 | locinput); |
| 7437 | NOT_REACHED; /* NOTREACHED */ |
| 7438 | |
| 7439 | #undef ST |
| 7440 | #define ST st->u.branch |
| 7441 | |
| 7442 | case BRANCHJ: /* /(...|A|...)/ with long next pointer */ |
| 7443 | next = scan + ARG(scan); |
| 7444 | if (next == scan) |
| 7445 | next = NULL; |
| 7446 | scan = NEXTOPER(scan); |
| 7447 | /* FALLTHROUGH */ |
| 7448 | |
| 7449 | case BRANCH: /* /(...|A|...)/ */ |
| 7450 | scan = NEXTOPER(scan); /* scan now points to inner node */ |
| 7451 | ST.lastparen = rex->lastparen; |
| 7452 | ST.lastcloseparen = rex->lastcloseparen; |
| 7453 | ST.next_branch = next; |
| 7454 | REGCP_SET(ST.cp); |
| 7455 | |
| 7456 | /* Now go into the branch */ |
| 7457 | if (has_cutgroup) { |
| 7458 | PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput); |
| 7459 | } else { |
| 7460 | PUSH_STATE_GOTO(BRANCH_next, scan, locinput); |
| 7461 | } |
| 7462 | NOT_REACHED; /* NOTREACHED */ |
| 7463 | |
| 7464 | case CUTGROUP: /* /(*THEN)/ */ |
| 7465 | sv_yes_mark = st->u.mark.mark_name = scan->flags |
| 7466 | ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ]) |
| 7467 | : NULL; |
| 7468 | PUSH_STATE_GOTO(CUTGROUP_next, next, locinput); |
| 7469 | NOT_REACHED; /* NOTREACHED */ |
| 7470 | |
| 7471 | case CUTGROUP_next_fail: |
| 7472 | do_cutgroup = 1; |
| 7473 | no_final = 1; |
| 7474 | if (st->u.mark.mark_name) |
| 7475 | sv_commit = st->u.mark.mark_name; |
| 7476 | sayNO; |
| 7477 | NOT_REACHED; /* NOTREACHED */ |
| 7478 | |
| 7479 | case BRANCH_next: |
| 7480 | sayYES; |
| 7481 | NOT_REACHED; /* NOTREACHED */ |
| 7482 | |
| 7483 | case BRANCH_next_fail: /* that branch failed; try the next, if any */ |
| 7484 | if (do_cutgroup) { |
| 7485 | do_cutgroup = 0; |
| 7486 | no_final = 0; |
| 7487 | } |
| 7488 | REGCP_UNWIND(ST.cp); |
| 7489 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 7490 | scan = ST.next_branch; |
| 7491 | /* no more branches? */ |
| 7492 | if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) { |
| 7493 | DEBUG_EXECUTE_r({ |
| 7494 | Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n", |
| 7495 | depth, |
| 7496 | PL_colors[4], |
| 7497 | PL_colors[5] ); |
| 7498 | }); |
| 7499 | sayNO_SILENT; |
| 7500 | } |
| 7501 | continue; /* execute next BRANCH[J] op */ |
| 7502 | /* NOTREACHED */ |
| 7503 | |
| 7504 | case MINMOD: /* next op will be non-greedy, e.g. A*? */ |
| 7505 | minmod = 1; |
| 7506 | break; |
| 7507 | |
| 7508 | #undef ST |
| 7509 | #define ST st->u.curlym |
| 7510 | |
| 7511 | case CURLYM: /* /A{m,n}B/ where A is fixed-length */ |
| 7512 | |
| 7513 | /* This is an optimisation of CURLYX that enables us to push |
| 7514 | * only a single backtracking state, no matter how many matches |
| 7515 | * there are in {m,n}. It relies on the pattern being constant |
| 7516 | * length, with no parens to influence future backrefs |
| 7517 | */ |
| 7518 | |
| 7519 | ST.me = scan; |
| 7520 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; |
| 7521 | |
| 7522 | ST.lastparen = rex->lastparen; |
| 7523 | ST.lastcloseparen = rex->lastcloseparen; |
| 7524 | |
| 7525 | /* if paren positive, emulate an OPEN/CLOSE around A */ |
| 7526 | if (ST.me->flags) { |
| 7527 | U32 paren = ST.me->flags; |
| 7528 | if (paren > maxopenparen) |
| 7529 | maxopenparen = paren; |
| 7530 | scan += NEXT_OFF(scan); /* Skip former OPEN. */ |
| 7531 | } |
| 7532 | ST.A = scan; |
| 7533 | ST.B = next; |
| 7534 | ST.alen = 0; |
| 7535 | ST.count = 0; |
| 7536 | ST.minmod = minmod; |
| 7537 | minmod = 0; |
| 7538 | ST.c1 = CHRTEST_UNINIT; |
| 7539 | REGCP_SET(ST.cp); |
| 7540 | |
| 7541 | if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */ |
| 7542 | goto curlym_do_B; |
| 7543 | |
| 7544 | curlym_do_A: /* execute the A in /A{m,n}B/ */ |
| 7545 | PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */ |
| 7546 | NOT_REACHED; /* NOTREACHED */ |
| 7547 | |
| 7548 | case CURLYM_A: /* we've just matched an A */ |
| 7549 | ST.count++; |
| 7550 | /* after first match, determine A's length: u.curlym.alen */ |
| 7551 | if (ST.count == 1) { |
| 7552 | if (reginfo->is_utf8_target) { |
| 7553 | char *s = st->locinput; |
| 7554 | while (s < locinput) { |
| 7555 | ST.alen++; |
| 7556 | s += UTF8SKIP(s); |
| 7557 | } |
| 7558 | } |
| 7559 | else { |
| 7560 | ST.alen = locinput - st->locinput; |
| 7561 | } |
| 7562 | if (ST.alen == 0) |
| 7563 | ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me); |
| 7564 | } |
| 7565 | DEBUG_EXECUTE_r( |
| 7566 | Perl_re_exec_indentf( aTHX_ "CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", |
| 7567 | depth, (IV) ST.count, (IV)ST.alen) |
| 7568 | ); |
| 7569 | |
| 7570 | if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags)) |
| 7571 | goto fake_end; |
| 7572 | |
| 7573 | { |
| 7574 | I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)); |
| 7575 | if ( max == REG_INFTY || ST.count < max ) |
| 7576 | goto curlym_do_A; /* try to match another A */ |
| 7577 | } |
| 7578 | goto curlym_do_B; /* try to match B */ |
| 7579 | |
| 7580 | case CURLYM_A_fail: /* just failed to match an A */ |
| 7581 | REGCP_UNWIND(ST.cp); |
| 7582 | |
| 7583 | |
| 7584 | if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ |
| 7585 | || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags)) |
| 7586 | sayNO; |
| 7587 | |
| 7588 | curlym_do_B: /* execute the B in /A{m,n}B/ */ |
| 7589 | if (ST.c1 == CHRTEST_UNINIT) { |
| 7590 | /* calculate c1 and c2 for possible match of 1st char |
| 7591 | * following curly */ |
| 7592 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 7593 | assert(ST.B); |
| 7594 | if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) { |
| 7595 | regnode *text_node = ST.B; |
| 7596 | if (! HAS_TEXT(text_node)) |
| 7597 | FIND_NEXT_IMPT(text_node); |
| 7598 | /* this used to be |
| 7599 | |
| 7600 | (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT) |
| 7601 | |
| 7602 | But the former is redundant in light of the latter. |
| 7603 | |
| 7604 | if this changes back then the macro for |
| 7605 | IS_TEXT and friends need to change. |
| 7606 | */ |
| 7607 | if (PL_regkind[OP(text_node)] == EXACT) { |
| 7608 | if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ |
| 7609 | text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, |
| 7610 | reginfo)) |
| 7611 | { |
| 7612 | sayNO; |
| 7613 | } |
| 7614 | } |
| 7615 | } |
| 7616 | } |
| 7617 | |
| 7618 | DEBUG_EXECUTE_r( |
| 7619 | Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%"IVdf"...\n", |
| 7620 | depth, (IV)ST.count) |
| 7621 | ); |
| 7622 | if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) { |
| 7623 | if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) { |
| 7624 | if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) |
| 7625 | && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) |
| 7626 | { |
| 7627 | /* simulate B failing */ |
| 7628 | DEBUG_OPTIMISE_r( |
| 7629 | Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n", |
| 7630 | depth, |
| 7631 | valid_utf8_to_uvchr((U8 *) locinput, NULL), |
| 7632 | valid_utf8_to_uvchr(ST.c1_utf8, NULL), |
| 7633 | valid_utf8_to_uvchr(ST.c2_utf8, NULL)) |
| 7634 | ); |
| 7635 | state_num = CURLYM_B_fail; |
| 7636 | goto reenter_switch; |
| 7637 | } |
| 7638 | } |
| 7639 | else if (nextchr != ST.c1 && nextchr != ST.c2) { |
| 7640 | /* simulate B failing */ |
| 7641 | DEBUG_OPTIMISE_r( |
| 7642 | Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n", |
| 7643 | depth, |
| 7644 | (int) nextchr, ST.c1, ST.c2) |
| 7645 | ); |
| 7646 | state_num = CURLYM_B_fail; |
| 7647 | goto reenter_switch; |
| 7648 | } |
| 7649 | } |
| 7650 | |
| 7651 | if (ST.me->flags) { |
| 7652 | /* emulate CLOSE: mark current A as captured */ |
| 7653 | I32 paren = ST.me->flags; |
| 7654 | if (ST.count) { |
| 7655 | rex->offs[paren].start |
| 7656 | = HOPc(locinput, -ST.alen) - reginfo->strbeg; |
| 7657 | rex->offs[paren].end = locinput - reginfo->strbeg; |
| 7658 | if ((U32)paren > rex->lastparen) |
| 7659 | rex->lastparen = paren; |
| 7660 | rex->lastcloseparen = paren; |
| 7661 | } |
| 7662 | else |
| 7663 | rex->offs[paren].end = -1; |
| 7664 | |
| 7665 | if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags)) |
| 7666 | { |
| 7667 | if (ST.count) |
| 7668 | goto fake_end; |
| 7669 | else |
| 7670 | sayNO; |
| 7671 | } |
| 7672 | } |
| 7673 | |
| 7674 | PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */ |
| 7675 | NOT_REACHED; /* NOTREACHED */ |
| 7676 | |
| 7677 | case CURLYM_B_fail: /* just failed to match a B */ |
| 7678 | REGCP_UNWIND(ST.cp); |
| 7679 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 7680 | if (ST.minmod) { |
| 7681 | I32 max = ARG2(ST.me); |
| 7682 | if (max != REG_INFTY && ST.count == max) |
| 7683 | sayNO; |
| 7684 | goto curlym_do_A; /* try to match a further A */ |
| 7685 | } |
| 7686 | /* backtrack one A */ |
| 7687 | if (ST.count == ARG1(ST.me) /* min */) |
| 7688 | sayNO; |
| 7689 | ST.count--; |
| 7690 | SET_locinput(HOPc(locinput, -ST.alen)); |
| 7691 | goto curlym_do_B; /* try to match B */ |
| 7692 | |
| 7693 | #undef ST |
| 7694 | #define ST st->u.curly |
| 7695 | |
| 7696 | #define CURLY_SETPAREN(paren, success) \ |
| 7697 | if (paren) { \ |
| 7698 | if (success) { \ |
| 7699 | rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \ |
| 7700 | rex->offs[paren].end = locinput - reginfo->strbeg; \ |
| 7701 | if (paren > rex->lastparen) \ |
| 7702 | rex->lastparen = paren; \ |
| 7703 | rex->lastcloseparen = paren; \ |
| 7704 | } \ |
| 7705 | else { \ |
| 7706 | rex->offs[paren].end = -1; \ |
| 7707 | rex->lastparen = ST.lastparen; \ |
| 7708 | rex->lastcloseparen = ST.lastcloseparen; \ |
| 7709 | } \ |
| 7710 | } |
| 7711 | |
| 7712 | case STAR: /* /A*B/ where A is width 1 char */ |
| 7713 | ST.paren = 0; |
| 7714 | ST.min = 0; |
| 7715 | ST.max = REG_INFTY; |
| 7716 | scan = NEXTOPER(scan); |
| 7717 | goto repeat; |
| 7718 | |
| 7719 | case PLUS: /* /A+B/ where A is width 1 char */ |
| 7720 | ST.paren = 0; |
| 7721 | ST.min = 1; |
| 7722 | ST.max = REG_INFTY; |
| 7723 | scan = NEXTOPER(scan); |
| 7724 | goto repeat; |
| 7725 | |
| 7726 | case CURLYN: /* /(A){m,n}B/ where A is width 1 char */ |
| 7727 | ST.paren = scan->flags; /* Which paren to set */ |
| 7728 | ST.lastparen = rex->lastparen; |
| 7729 | ST.lastcloseparen = rex->lastcloseparen; |
| 7730 | if (ST.paren > maxopenparen) |
| 7731 | maxopenparen = ST.paren; |
| 7732 | ST.min = ARG1(scan); /* min to match */ |
| 7733 | ST.max = ARG2(scan); /* max to match */ |
| 7734 | if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) |
| 7735 | { |
| 7736 | ST.min=1; |
| 7737 | ST.max=1; |
| 7738 | } |
| 7739 | scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); |
| 7740 | goto repeat; |
| 7741 | |
| 7742 | case CURLY: /* /A{m,n}B/ where A is width 1 char */ |
| 7743 | ST.paren = 0; |
| 7744 | ST.min = ARG1(scan); /* min to match */ |
| 7745 | ST.max = ARG2(scan); /* max to match */ |
| 7746 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; |
| 7747 | repeat: |
| 7748 | /* |
| 7749 | * Lookahead to avoid useless match attempts |
| 7750 | * when we know what character comes next. |
| 7751 | * |
| 7752 | * Used to only do .*x and .*?x, but now it allows |
| 7753 | * for )'s, ('s and (?{ ... })'s to be in the way |
| 7754 | * of the quantifier and the EXACT-like node. -- japhy |
| 7755 | */ |
| 7756 | |
| 7757 | assert(ST.min <= ST.max); |
| 7758 | if (! HAS_TEXT(next) && ! JUMPABLE(next)) { |
| 7759 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 7760 | } |
| 7761 | else { |
| 7762 | regnode *text_node = next; |
| 7763 | |
| 7764 | if (! HAS_TEXT(text_node)) |
| 7765 | FIND_NEXT_IMPT(text_node); |
| 7766 | |
| 7767 | if (! HAS_TEXT(text_node)) |
| 7768 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 7769 | else { |
| 7770 | if ( PL_regkind[OP(text_node)] != EXACT ) { |
| 7771 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 7772 | } |
| 7773 | else { |
| 7774 | |
| 7775 | /* Currently we only get here when |
| 7776 | |
| 7777 | PL_rekind[OP(text_node)] == EXACT |
| 7778 | |
| 7779 | if this changes back then the macro for IS_TEXT and |
| 7780 | friends need to change. */ |
| 7781 | if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ |
| 7782 | text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, |
| 7783 | reginfo)) |
| 7784 | { |
| 7785 | sayNO; |
| 7786 | } |
| 7787 | } |
| 7788 | } |
| 7789 | } |
| 7790 | |
| 7791 | ST.A = scan; |
| 7792 | ST.B = next; |
| 7793 | if (minmod) { |
| 7794 | char *li = locinput; |
| 7795 | minmod = 0; |
| 7796 | if (ST.min && |
| 7797 | regrepeat(rex, &li, ST.A, reginfo, ST.min, depth) |
| 7798 | < ST.min) |
| 7799 | sayNO; |
| 7800 | SET_locinput(li); |
| 7801 | ST.count = ST.min; |
| 7802 | REGCP_SET(ST.cp); |
| 7803 | if (ST.c1 == CHRTEST_VOID) |
| 7804 | goto curly_try_B_min; |
| 7805 | |
| 7806 | ST.oldloc = locinput; |
| 7807 | |
| 7808 | /* set ST.maxpos to the furthest point along the |
| 7809 | * string that could possibly match */ |
| 7810 | if (ST.max == REG_INFTY) { |
| 7811 | ST.maxpos = reginfo->strend - 1; |
| 7812 | if (utf8_target) |
| 7813 | while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos)) |
| 7814 | ST.maxpos--; |
| 7815 | } |
| 7816 | else if (utf8_target) { |
| 7817 | int m = ST.max - ST.min; |
| 7818 | for (ST.maxpos = locinput; |
| 7819 | m >0 && ST.maxpos < reginfo->strend; m--) |
| 7820 | ST.maxpos += UTF8SKIP(ST.maxpos); |
| 7821 | } |
| 7822 | else { |
| 7823 | ST.maxpos = locinput + ST.max - ST.min; |
| 7824 | if (ST.maxpos >= reginfo->strend) |
| 7825 | ST.maxpos = reginfo->strend - 1; |
| 7826 | } |
| 7827 | goto curly_try_B_min_known; |
| 7828 | |
| 7829 | } |
| 7830 | else { |
| 7831 | /* avoid taking address of locinput, so it can remain |
| 7832 | * a register var */ |
| 7833 | char *li = locinput; |
| 7834 | ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth); |
| 7835 | if (ST.count < ST.min) |
| 7836 | sayNO; |
| 7837 | SET_locinput(li); |
| 7838 | if ((ST.count > ST.min) |
| 7839 | && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL)) |
| 7840 | { |
| 7841 | /* A{m,n} must come at the end of the string, there's |
| 7842 | * no point in backing off ... */ |
| 7843 | ST.min = ST.count; |
| 7844 | /* ...except that $ and \Z can match before *and* after |
| 7845 | newline at the end. Consider "\n\n" =~ /\n+\Z\n/. |
| 7846 | We may back off by one in this case. */ |
| 7847 | if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS) |
| 7848 | ST.min--; |
| 7849 | } |
| 7850 | REGCP_SET(ST.cp); |
| 7851 | goto curly_try_B_max; |
| 7852 | } |
| 7853 | NOT_REACHED; /* NOTREACHED */ |
| 7854 | |
| 7855 | case CURLY_B_min_known_fail: |
| 7856 | /* failed to find B in a non-greedy match where c1,c2 valid */ |
| 7857 | |
| 7858 | REGCP_UNWIND(ST.cp); |
| 7859 | if (ST.paren) { |
| 7860 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 7861 | } |
| 7862 | /* Couldn't or didn't -- move forward. */ |
| 7863 | ST.oldloc = locinput; |
| 7864 | if (utf8_target) |
| 7865 | locinput += UTF8SKIP(locinput); |
| 7866 | else |
| 7867 | locinput++; |
| 7868 | ST.count++; |
| 7869 | curly_try_B_min_known: |
| 7870 | /* find the next place where 'B' could work, then call B */ |
| 7871 | { |
| 7872 | int n; |
| 7873 | if (utf8_target) { |
| 7874 | n = (ST.oldloc == locinput) ? 0 : 1; |
| 7875 | if (ST.c1 == ST.c2) { |
| 7876 | /* set n to utf8_distance(oldloc, locinput) */ |
| 7877 | while (locinput <= ST.maxpos |
| 7878 | && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))) |
| 7879 | { |
| 7880 | locinput += UTF8SKIP(locinput); |
| 7881 | n++; |
| 7882 | } |
| 7883 | } |
| 7884 | else { |
| 7885 | /* set n to utf8_distance(oldloc, locinput) */ |
| 7886 | while (locinput <= ST.maxpos |
| 7887 | && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) |
| 7888 | && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) |
| 7889 | { |
| 7890 | locinput += UTF8SKIP(locinput); |
| 7891 | n++; |
| 7892 | } |
| 7893 | } |
| 7894 | } |
| 7895 | else { /* Not utf8_target */ |
| 7896 | if (ST.c1 == ST.c2) { |
| 7897 | while (locinput <= ST.maxpos && |
| 7898 | UCHARAT(locinput) != ST.c1) |
| 7899 | locinput++; |
| 7900 | } |
| 7901 | else { |
| 7902 | while (locinput <= ST.maxpos |
| 7903 | && UCHARAT(locinput) != ST.c1 |
| 7904 | && UCHARAT(locinput) != ST.c2) |
| 7905 | locinput++; |
| 7906 | } |
| 7907 | n = locinput - ST.oldloc; |
| 7908 | } |
| 7909 | if (locinput > ST.maxpos) |
| 7910 | sayNO; |
| 7911 | if (n) { |
| 7912 | /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is |
| 7913 | * at b; check that everything between oldloc and |
| 7914 | * locinput matches */ |
| 7915 | char *li = ST.oldloc; |
| 7916 | ST.count += n; |
| 7917 | if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n) |
| 7918 | sayNO; |
| 7919 | assert(n == REG_INFTY || locinput == li); |
| 7920 | } |
| 7921 | CURLY_SETPAREN(ST.paren, ST.count); |
| 7922 | if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) |
| 7923 | goto fake_end; |
| 7924 | PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput); |
| 7925 | } |
| 7926 | NOT_REACHED; /* NOTREACHED */ |
| 7927 | |
| 7928 | case CURLY_B_min_fail: |
| 7929 | /* failed to find B in a non-greedy match where c1,c2 invalid */ |
| 7930 | |
| 7931 | REGCP_UNWIND(ST.cp); |
| 7932 | if (ST.paren) { |
| 7933 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 7934 | } |
| 7935 | /* failed -- move forward one */ |
| 7936 | { |
| 7937 | char *li = locinput; |
| 7938 | if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) { |
| 7939 | sayNO; |
| 7940 | } |
| 7941 | locinput = li; |
| 7942 | } |
| 7943 | { |
| 7944 | ST.count++; |
| 7945 | if (ST.count <= ST.max || (ST.max == REG_INFTY && |
| 7946 | ST.count > 0)) /* count overflow ? */ |
| 7947 | { |
| 7948 | curly_try_B_min: |
| 7949 | CURLY_SETPAREN(ST.paren, ST.count); |
| 7950 | if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) |
| 7951 | goto fake_end; |
| 7952 | PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput); |
| 7953 | } |
| 7954 | } |
| 7955 | sayNO; |
| 7956 | NOT_REACHED; /* NOTREACHED */ |
| 7957 | |
| 7958 | curly_try_B_max: |
| 7959 | /* a successful greedy match: now try to match B */ |
| 7960 | if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) |
| 7961 | goto fake_end; |
| 7962 | { |
| 7963 | bool could_match = locinput < reginfo->strend; |
| 7964 | |
| 7965 | /* If it could work, try it. */ |
| 7966 | if (ST.c1 != CHRTEST_VOID && could_match) { |
| 7967 | if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target) |
| 7968 | { |
| 7969 | could_match = memEQ(locinput, |
| 7970 | ST.c1_utf8, |
| 7971 | UTF8SKIP(locinput)) |
| 7972 | || memEQ(locinput, |
| 7973 | ST.c2_utf8, |
| 7974 | UTF8SKIP(locinput)); |
| 7975 | } |
| 7976 | else { |
| 7977 | could_match = UCHARAT(locinput) == ST.c1 |
| 7978 | || UCHARAT(locinput) == ST.c2; |
| 7979 | } |
| 7980 | } |
| 7981 | if (ST.c1 == CHRTEST_VOID || could_match) { |
| 7982 | CURLY_SETPAREN(ST.paren, ST.count); |
| 7983 | PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput); |
| 7984 | NOT_REACHED; /* NOTREACHED */ |
| 7985 | } |
| 7986 | } |
| 7987 | /* FALLTHROUGH */ |
| 7988 | |
| 7989 | case CURLY_B_max_fail: |
| 7990 | /* failed to find B in a greedy match */ |
| 7991 | |
| 7992 | REGCP_UNWIND(ST.cp); |
| 7993 | if (ST.paren) { |
| 7994 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 7995 | } |
| 7996 | /* back up. */ |
| 7997 | if (--ST.count < ST.min) |
| 7998 | sayNO; |
| 7999 | locinput = HOPc(locinput, -1); |
| 8000 | goto curly_try_B_max; |
| 8001 | |
| 8002 | #undef ST |
| 8003 | |
| 8004 | case END: /* last op of main pattern */ |
| 8005 | fake_end: |
| 8006 | if (cur_eval) { |
| 8007 | /* we've just finished A in /(??{A})B/; now continue with B */ |
| 8008 | SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput); |
| 8009 | st->u.eval.prev_rex = rex_sv; /* inner */ |
| 8010 | |
| 8011 | /* Save *all* the positions. */ |
| 8012 | st->u.eval.cp = regcppush(rex, 0, maxopenparen); |
| 8013 | rex_sv = CUR_EVAL.prev_rex; |
| 8014 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); |
| 8015 | SET_reg_curpm(rex_sv); |
| 8016 | rex = ReANY(rex_sv); |
| 8017 | rexi = RXi_GET(rex); |
| 8018 | |
| 8019 | st->u.eval.prev_curlyx = cur_curlyx; |
| 8020 | cur_curlyx = CUR_EVAL.prev_curlyx; |
| 8021 | |
| 8022 | REGCP_SET(st->u.eval.lastcp); |
| 8023 | |
| 8024 | /* Restore parens of the outer rex without popping the |
| 8025 | * savestack */ |
| 8026 | S_regcp_restore(aTHX_ rex, CUR_EVAL.lastcp, |
| 8027 | &maxopenparen); |
| 8028 | |
| 8029 | st->u.eval.prev_eval = cur_eval; |
| 8030 | cur_eval = CUR_EVAL.prev_eval; |
| 8031 | DEBUG_EXECUTE_r( |
| 8032 | Perl_re_exec_indentf( aTHX_ "EVAL trying tail ... (cur_eval=%p)\n", |
| 8033 | depth, cur_eval);); |
| 8034 | if ( nochange_depth ) |
| 8035 | nochange_depth--; |
| 8036 | |
| 8037 | SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput); |
| 8038 | |
| 8039 | PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B, |
| 8040 | locinput); /* match B */ |
| 8041 | } |
| 8042 | |
| 8043 | if (locinput < reginfo->till) { |
| 8044 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ |
| 8045 | "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", |
| 8046 | PL_colors[4], |
| 8047 | (long)(locinput - startpos), |
| 8048 | (long)(reginfo->till - startpos), |
| 8049 | PL_colors[5])); |
| 8050 | |
| 8051 | sayNO_SILENT; /* Cannot match: too short. */ |
| 8052 | } |
| 8053 | sayYES; /* Success! */ |
| 8054 | |
| 8055 | case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ |
| 8056 | DEBUG_EXECUTE_r( |
| 8057 | Perl_re_exec_indentf( aTHX_ "%ssubpattern success...%s\n", |
| 8058 | depth, PL_colors[4], PL_colors[5])); |
| 8059 | sayYES; /* Success! */ |
| 8060 | |
| 8061 | #undef ST |
| 8062 | #define ST st->u.ifmatch |
| 8063 | |
| 8064 | { |
| 8065 | char *newstart; |
| 8066 | |
| 8067 | case SUSPEND: /* (?>A) */ |
| 8068 | ST.wanted = 1; |
| 8069 | newstart = locinput; |
| 8070 | goto do_ifmatch; |
| 8071 | |
| 8072 | case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */ |
| 8073 | ST.wanted = 0; |
| 8074 | goto ifmatch_trivial_fail_test; |
| 8075 | |
| 8076 | case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */ |
| 8077 | ST.wanted = 1; |
| 8078 | ifmatch_trivial_fail_test: |
| 8079 | if (scan->flags) { |
| 8080 | char * const s = HOPBACKc(locinput, scan->flags); |
| 8081 | if (!s) { |
| 8082 | /* trivial fail */ |
| 8083 | if (logical) { |
| 8084 | logical = 0; |
| 8085 | sw = 1 - cBOOL(ST.wanted); |
| 8086 | } |
| 8087 | else if (ST.wanted) |
| 8088 | sayNO; |
| 8089 | next = scan + ARG(scan); |
| 8090 | if (next == scan) |
| 8091 | next = NULL; |
| 8092 | break; |
| 8093 | } |
| 8094 | newstart = s; |
| 8095 | } |
| 8096 | else |
| 8097 | newstart = locinput; |
| 8098 | |
| 8099 | do_ifmatch: |
| 8100 | ST.me = scan; |
| 8101 | ST.logical = logical; |
| 8102 | logical = 0; /* XXX: reset state of logical once it has been saved into ST */ |
| 8103 | |
| 8104 | /* execute body of (?...A) */ |
| 8105 | PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart); |
| 8106 | NOT_REACHED; /* NOTREACHED */ |
| 8107 | } |
| 8108 | |
| 8109 | case IFMATCH_A_fail: /* body of (?...A) failed */ |
| 8110 | ST.wanted = !ST.wanted; |
| 8111 | /* FALLTHROUGH */ |
| 8112 | |
| 8113 | case IFMATCH_A: /* body of (?...A) succeeded */ |
| 8114 | if (ST.logical) { |
| 8115 | sw = cBOOL(ST.wanted); |
| 8116 | } |
| 8117 | else if (!ST.wanted) |
| 8118 | sayNO; |
| 8119 | |
| 8120 | if (OP(ST.me) != SUSPEND) { |
| 8121 | /* restore old position except for (?>...) */ |
| 8122 | locinput = st->locinput; |
| 8123 | } |
| 8124 | scan = ST.me + ARG(ST.me); |
| 8125 | if (scan == ST.me) |
| 8126 | scan = NULL; |
| 8127 | continue; /* execute B */ |
| 8128 | |
| 8129 | #undef ST |
| 8130 | |
| 8131 | case LONGJMP: /* alternative with many branches compiles to |
| 8132 | * (BRANCHJ; EXACT ...; LONGJMP ) x N */ |
| 8133 | next = scan + ARG(scan); |
| 8134 | if (next == scan) |
| 8135 | next = NULL; |
| 8136 | break; |
| 8137 | |
| 8138 | case COMMIT: /* (*COMMIT) */ |
| 8139 | reginfo->cutpoint = reginfo->strend; |
| 8140 | /* FALLTHROUGH */ |
| 8141 | |
| 8142 | case PRUNE: /* (*PRUNE) */ |
| 8143 | if (scan->flags) |
| 8144 | sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 8145 | PUSH_STATE_GOTO(COMMIT_next, next, locinput); |
| 8146 | NOT_REACHED; /* NOTREACHED */ |
| 8147 | |
| 8148 | case COMMIT_next_fail: |
| 8149 | no_final = 1; |
| 8150 | /* FALLTHROUGH */ |
| 8151 | sayNO; |
| 8152 | NOT_REACHED; /* NOTREACHED */ |
| 8153 | |
| 8154 | case OPFAIL: /* (*FAIL) */ |
| 8155 | if (scan->flags) |
| 8156 | sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 8157 | if (logical) { |
| 8158 | /* deal with (?(?!)X|Y) properly, |
| 8159 | * make sure we trigger the no branch |
| 8160 | * of the trailing IFTHEN structure*/ |
| 8161 | sw= 0; |
| 8162 | break; |
| 8163 | } else { |
| 8164 | sayNO; |
| 8165 | } |
| 8166 | NOT_REACHED; /* NOTREACHED */ |
| 8167 | |
| 8168 | #define ST st->u.mark |
| 8169 | case MARKPOINT: /* (*MARK:foo) */ |
| 8170 | ST.prev_mark = mark_state; |
| 8171 | ST.mark_name = sv_commit = sv_yes_mark |
| 8172 | = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 8173 | mark_state = st; |
| 8174 | ST.mark_loc = locinput; |
| 8175 | PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput); |
| 8176 | NOT_REACHED; /* NOTREACHED */ |
| 8177 | |
| 8178 | case MARKPOINT_next: |
| 8179 | mark_state = ST.prev_mark; |
| 8180 | sayYES; |
| 8181 | NOT_REACHED; /* NOTREACHED */ |
| 8182 | |
| 8183 | case MARKPOINT_next_fail: |
| 8184 | if (popmark && sv_eq(ST.mark_name,popmark)) |
| 8185 | { |
| 8186 | if (ST.mark_loc > startpoint) |
| 8187 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); |
| 8188 | popmark = NULL; /* we found our mark */ |
| 8189 | sv_commit = ST.mark_name; |
| 8190 | |
| 8191 | DEBUG_EXECUTE_r({ |
| 8192 | Perl_re_exec_indentf( aTHX_ "%ssetting cutpoint to mark:%"SVf"...%s\n", |
| 8193 | depth, |
| 8194 | PL_colors[4], SVfARG(sv_commit), PL_colors[5]); |
| 8195 | }); |
| 8196 | } |
| 8197 | mark_state = ST.prev_mark; |
| 8198 | sv_yes_mark = mark_state ? |
| 8199 | mark_state->u.mark.mark_name : NULL; |
| 8200 | sayNO; |
| 8201 | NOT_REACHED; /* NOTREACHED */ |
| 8202 | |
| 8203 | case SKIP: /* (*SKIP) */ |
| 8204 | if (!scan->flags) { |
| 8205 | /* (*SKIP) : if we fail we cut here*/ |
| 8206 | ST.mark_name = NULL; |
| 8207 | ST.mark_loc = locinput; |
| 8208 | PUSH_STATE_GOTO(SKIP_next,next, locinput); |
| 8209 | } else { |
| 8210 | /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was, |
| 8211 | otherwise do nothing. Meaning we need to scan |
| 8212 | */ |
| 8213 | regmatch_state *cur = mark_state; |
| 8214 | SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 8215 | |
| 8216 | while (cur) { |
| 8217 | if ( sv_eq( cur->u.mark.mark_name, |
| 8218 | find ) ) |
| 8219 | { |
| 8220 | ST.mark_name = find; |
| 8221 | PUSH_STATE_GOTO( SKIP_next, next, locinput); |
| 8222 | } |
| 8223 | cur = cur->u.mark.prev_mark; |
| 8224 | } |
| 8225 | } |
| 8226 | /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */ |
| 8227 | break; |
| 8228 | |
| 8229 | case SKIP_next_fail: |
| 8230 | if (ST.mark_name) { |
| 8231 | /* (*CUT:NAME) - Set up to search for the name as we |
| 8232 | collapse the stack*/ |
| 8233 | popmark = ST.mark_name; |
| 8234 | } else { |
| 8235 | /* (*CUT) - No name, we cut here.*/ |
| 8236 | if (ST.mark_loc > startpoint) |
| 8237 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); |
| 8238 | /* but we set sv_commit to latest mark_name if there |
| 8239 | is one so they can test to see how things lead to this |
| 8240 | cut */ |
| 8241 | if (mark_state) |
| 8242 | sv_commit=mark_state->u.mark.mark_name; |
| 8243 | } |
| 8244 | no_final = 1; |
| 8245 | sayNO; |
| 8246 | NOT_REACHED; /* NOTREACHED */ |
| 8247 | #undef ST |
| 8248 | |
| 8249 | case LNBREAK: /* \R */ |
| 8250 | if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) { |
| 8251 | locinput += n; |
| 8252 | } else |
| 8253 | sayNO; |
| 8254 | break; |
| 8255 | |
| 8256 | default: |
| 8257 | PerlIO_printf(Perl_error_log, "%"UVxf" %d\n", |
| 8258 | PTR2UV(scan), OP(scan)); |
| 8259 | Perl_croak(aTHX_ "regexp memory corruption"); |
| 8260 | |
| 8261 | /* this is a point to jump to in order to increment |
| 8262 | * locinput by one character */ |
| 8263 | increment_locinput: |
| 8264 | assert(!NEXTCHR_IS_EOS); |
| 8265 | if (utf8_target) { |
| 8266 | locinput += PL_utf8skip[nextchr]; |
| 8267 | /* locinput is allowed to go 1 char off the end, but not 2+ */ |
| 8268 | if (locinput > reginfo->strend) |
| 8269 | sayNO; |
| 8270 | } |
| 8271 | else |
| 8272 | locinput++; |
| 8273 | break; |
| 8274 | |
| 8275 | } /* end switch */ |
| 8276 | |
| 8277 | /* switch break jumps here */ |
| 8278 | scan = next; /* prepare to execute the next op and ... */ |
| 8279 | continue; /* ... jump back to the top, reusing st */ |
| 8280 | /* NOTREACHED */ |
| 8281 | |
| 8282 | push_yes_state: |
| 8283 | /* push a state that backtracks on success */ |
| 8284 | st->u.yes.prev_yes_state = yes_state; |
| 8285 | yes_state = st; |
| 8286 | /* FALLTHROUGH */ |
| 8287 | push_state: |
| 8288 | /* push a new regex state, then continue at scan */ |
| 8289 | { |
| 8290 | regmatch_state *newst; |
| 8291 | |
| 8292 | DEBUG_STACK_r({ |
| 8293 | regmatch_state *cur = st; |
| 8294 | regmatch_state *curyes = yes_state; |
| 8295 | int curd = depth; |
| 8296 | regmatch_slab *slab = PL_regmatch_slab; |
| 8297 | for (;curd > -1 && (depth-curd < 3);cur--,curd--) { |
| 8298 | if (cur < SLAB_FIRST(slab)) { |
| 8299 | slab = slab->prev; |
| 8300 | cur = SLAB_LAST(slab); |
| 8301 | } |
| 8302 | Perl_re_exec_indentf( aTHX_ "#%-3d %-10s %s\n", |
| 8303 | depth, |
| 8304 | curd, PL_reg_name[cur->resume_state], |
| 8305 | (curyes == cur) ? "yes" : "" |
| 8306 | ); |
| 8307 | if (curyes == cur) |
| 8308 | curyes = cur->u.yes.prev_yes_state; |
| 8309 | } |
| 8310 | } else |
| 8311 | DEBUG_STATE_pp("push") |
| 8312 | ); |
| 8313 | depth++; |
| 8314 | st->locinput = locinput; |
| 8315 | newst = st+1; |
| 8316 | if (newst > SLAB_LAST(PL_regmatch_slab)) |
| 8317 | newst = S_push_slab(aTHX); |
| 8318 | PL_regmatch_state = newst; |
| 8319 | |
| 8320 | locinput = pushinput; |
| 8321 | st = newst; |
| 8322 | continue; |
| 8323 | /* NOTREACHED */ |
| 8324 | } |
| 8325 | } |
| 8326 | #ifdef SOLARIS_BAD_OPTIMIZER |
| 8327 | # undef PL_charclass |
| 8328 | #endif |
| 8329 | |
| 8330 | /* |
| 8331 | * We get here only if there's trouble -- normally "case END" is |
| 8332 | * the terminating point. |
| 8333 | */ |
| 8334 | Perl_croak(aTHX_ "corrupted regexp pointers"); |
| 8335 | NOT_REACHED; /* NOTREACHED */ |
| 8336 | |
| 8337 | yes: |
| 8338 | if (yes_state) { |
| 8339 | /* we have successfully completed a subexpression, but we must now |
| 8340 | * pop to the state marked by yes_state and continue from there */ |
| 8341 | assert(st != yes_state); |
| 8342 | #ifdef DEBUGGING |
| 8343 | while (st != yes_state) { |
| 8344 | st--; |
| 8345 | if (st < SLAB_FIRST(PL_regmatch_slab)) { |
| 8346 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 8347 | st = SLAB_LAST(PL_regmatch_slab); |
| 8348 | } |
| 8349 | DEBUG_STATE_r({ |
| 8350 | if (no_final) { |
| 8351 | DEBUG_STATE_pp("pop (no final)"); |
| 8352 | } else { |
| 8353 | DEBUG_STATE_pp("pop (yes)"); |
| 8354 | } |
| 8355 | }); |
| 8356 | depth--; |
| 8357 | } |
| 8358 | #else |
| 8359 | while (yes_state < SLAB_FIRST(PL_regmatch_slab) |
| 8360 | || yes_state > SLAB_LAST(PL_regmatch_slab)) |
| 8361 | { |
| 8362 | /* not in this slab, pop slab */ |
| 8363 | depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1); |
| 8364 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 8365 | st = SLAB_LAST(PL_regmatch_slab); |
| 8366 | } |
| 8367 | depth -= (st - yes_state); |
| 8368 | #endif |
| 8369 | st = yes_state; |
| 8370 | yes_state = st->u.yes.prev_yes_state; |
| 8371 | PL_regmatch_state = st; |
| 8372 | |
| 8373 | if (no_final) |
| 8374 | locinput= st->locinput; |
| 8375 | state_num = st->resume_state + no_final; |
| 8376 | goto reenter_switch; |
| 8377 | } |
| 8378 | |
| 8379 | DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n", |
| 8380 | PL_colors[4], PL_colors[5])); |
| 8381 | |
| 8382 | if (reginfo->info_aux_eval) { |
| 8383 | /* each successfully executed (?{...}) block does the equivalent of |
| 8384 | * local $^R = do {...} |
| 8385 | * When popping the save stack, all these locals would be undone; |
| 8386 | * bypass this by setting the outermost saved $^R to the latest |
| 8387 | * value */ |
| 8388 | /* I dont know if this is needed or works properly now. |
| 8389 | * see code related to PL_replgv elsewhere in this file. |
| 8390 | * Yves |
| 8391 | */ |
| 8392 | if (oreplsv != GvSV(PL_replgv)) |
| 8393 | sv_setsv(oreplsv, GvSV(PL_replgv)); |
| 8394 | } |
| 8395 | result = 1; |
| 8396 | goto final_exit; |
| 8397 | |
| 8398 | no: |
| 8399 | DEBUG_EXECUTE_r( |
| 8400 | Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n", |
| 8401 | depth, |
| 8402 | PL_colors[4], PL_colors[5]) |
| 8403 | ); |
| 8404 | |
| 8405 | no_silent: |
| 8406 | if (no_final) { |
| 8407 | if (yes_state) { |
| 8408 | goto yes; |
| 8409 | } else { |
| 8410 | goto final_exit; |
| 8411 | } |
| 8412 | } |
| 8413 | if (depth) { |
| 8414 | /* there's a previous state to backtrack to */ |
| 8415 | st--; |
| 8416 | if (st < SLAB_FIRST(PL_regmatch_slab)) { |
| 8417 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 8418 | st = SLAB_LAST(PL_regmatch_slab); |
| 8419 | } |
| 8420 | PL_regmatch_state = st; |
| 8421 | locinput= st->locinput; |
| 8422 | |
| 8423 | DEBUG_STATE_pp("pop"); |
| 8424 | depth--; |
| 8425 | if (yes_state == st) |
| 8426 | yes_state = st->u.yes.prev_yes_state; |
| 8427 | |
| 8428 | state_num = st->resume_state + 1; /* failure = success + 1 */ |
| 8429 | PERL_ASYNC_CHECK(); |
| 8430 | goto reenter_switch; |
| 8431 | } |
| 8432 | result = 0; |
| 8433 | |
| 8434 | final_exit: |
| 8435 | if (rex->intflags & PREGf_VERBARG_SEEN) { |
| 8436 | SV *sv_err = get_sv("REGERROR", 1); |
| 8437 | SV *sv_mrk = get_sv("REGMARK", 1); |
| 8438 | if (result) { |
| 8439 | sv_commit = &PL_sv_no; |
| 8440 | if (!sv_yes_mark) |
| 8441 | sv_yes_mark = &PL_sv_yes; |
| 8442 | } else { |
| 8443 | if (!sv_commit) |
| 8444 | sv_commit = &PL_sv_yes; |
| 8445 | sv_yes_mark = &PL_sv_no; |
| 8446 | } |
| 8447 | assert(sv_err); |
| 8448 | assert(sv_mrk); |
| 8449 | sv_setsv(sv_err, sv_commit); |
| 8450 | sv_setsv(sv_mrk, sv_yes_mark); |
| 8451 | } |
| 8452 | |
| 8453 | |
| 8454 | if (last_pushed_cv) { |
| 8455 | dSP; |
| 8456 | POP_MULTICALL; |
| 8457 | PERL_UNUSED_VAR(SP); |
| 8458 | } |
| 8459 | |
| 8460 | assert(!result || locinput - reginfo->strbeg >= 0); |
| 8461 | return result ? locinput - reginfo->strbeg : -1; |
| 8462 | } |
| 8463 | |
| 8464 | /* |
| 8465 | - regrepeat - repeatedly match something simple, report how many |
| 8466 | * |
| 8467 | * What 'simple' means is a node which can be the operand of a quantifier like |
| 8468 | * '+', or {1,3} |
| 8469 | * |
| 8470 | * startposp - pointer a pointer to the start position. This is updated |
| 8471 | * to point to the byte following the highest successful |
| 8472 | * match. |
| 8473 | * p - the regnode to be repeatedly matched against. |
| 8474 | * reginfo - struct holding match state, such as strend |
| 8475 | * max - maximum number of things to match. |
| 8476 | * depth - (for debugging) backtracking depth. |
| 8477 | */ |
| 8478 | STATIC I32 |
| 8479 | S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p, |
| 8480 | regmatch_info *const reginfo, I32 max, int depth) |
| 8481 | { |
| 8482 | char *scan; /* Pointer to current position in target string */ |
| 8483 | I32 c; |
| 8484 | char *loceol = reginfo->strend; /* local version */ |
| 8485 | I32 hardcount = 0; /* How many matches so far */ |
| 8486 | bool utf8_target = reginfo->is_utf8_target; |
| 8487 | unsigned int to_complement = 0; /* Invert the result? */ |
| 8488 | UV utf8_flags; |
| 8489 | _char_class_number classnum; |
| 8490 | #ifndef DEBUGGING |
| 8491 | PERL_UNUSED_ARG(depth); |
| 8492 | #endif |
| 8493 | |
| 8494 | PERL_ARGS_ASSERT_REGREPEAT; |
| 8495 | |
| 8496 | scan = *startposp; |
| 8497 | if (max == REG_INFTY) |
| 8498 | max = I32_MAX; |
| 8499 | else if (! utf8_target && loceol - scan > max) |
| 8500 | loceol = scan + max; |
| 8501 | |
| 8502 | /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down |
| 8503 | * to the maximum of how far we should go in it (leaving it set to the real |
| 8504 | * end, if the maximum permissible would take us beyond that). This allows |
| 8505 | * us to make the loop exit condition that we haven't gone past <loceol> to |
| 8506 | * also mean that we haven't exceeded the max permissible count, saving a |
| 8507 | * test each time through the loop. But it assumes that the OP matches a |
| 8508 | * single byte, which is true for most of the OPs below when applied to a |
| 8509 | * non-UTF-8 target. Those relatively few OPs that don't have this |
| 8510 | * characteristic will have to compensate. |
| 8511 | * |
| 8512 | * There is no adjustment for UTF-8 targets, as the number of bytes per |
| 8513 | * character varies. OPs will have to test both that the count is less |
| 8514 | * than the max permissible (using <hardcount> to keep track), and that we |
| 8515 | * are still within the bounds of the string (using <loceol>. A few OPs |
| 8516 | * match a single byte no matter what the encoding. They can omit the max |
| 8517 | * test if, for the UTF-8 case, they do the adjustment that was skipped |
| 8518 | * above. |
| 8519 | * |
| 8520 | * Thus, the code above sets things up for the common case; and exceptional |
| 8521 | * cases need extra work; the common case is to make sure <scan> doesn't |
| 8522 | * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the |
| 8523 | * count doesn't exceed the maximum permissible */ |
| 8524 | |
| 8525 | switch (OP(p)) { |
| 8526 | case REG_ANY: |
| 8527 | if (utf8_target) { |
| 8528 | while (scan < loceol && hardcount < max && *scan != '\n') { |
| 8529 | scan += UTF8SKIP(scan); |
| 8530 | hardcount++; |
| 8531 | } |
| 8532 | } else { |
| 8533 | while (scan < loceol && *scan != '\n') |
| 8534 | scan++; |
| 8535 | } |
| 8536 | break; |
| 8537 | case SANY: |
| 8538 | if (utf8_target) { |
| 8539 | while (scan < loceol && hardcount < max) { |
| 8540 | scan += UTF8SKIP(scan); |
| 8541 | hardcount++; |
| 8542 | } |
| 8543 | } |
| 8544 | else |
| 8545 | scan = loceol; |
| 8546 | break; |
| 8547 | case EXACTL: |
| 8548 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 8549 | if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) { |
| 8550 | _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol); |
| 8551 | } |
| 8552 | /* FALLTHROUGH */ |
| 8553 | case EXACT: |
| 8554 | assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); |
| 8555 | |
| 8556 | c = (U8)*STRING(p); |
| 8557 | |
| 8558 | /* Can use a simple loop if the pattern char to match on is invariant |
| 8559 | * under UTF-8, or both target and pattern aren't UTF-8. Note that we |
| 8560 | * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's |
| 8561 | * true iff it doesn't matter if the argument is in UTF-8 or not */ |
| 8562 | if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) { |
| 8563 | if (utf8_target && loceol - scan > max) { |
| 8564 | /* We didn't adjust <loceol> because is UTF-8, but ok to do so, |
| 8565 | * since here, to match at all, 1 char == 1 byte */ |
| 8566 | loceol = scan + max; |
| 8567 | } |
| 8568 | while (scan < loceol && UCHARAT(scan) == c) { |
| 8569 | scan++; |
| 8570 | } |
| 8571 | } |
| 8572 | else if (reginfo->is_utf8_pat) { |
| 8573 | if (utf8_target) { |
| 8574 | STRLEN scan_char_len; |
| 8575 | |
| 8576 | /* When both target and pattern are UTF-8, we have to do |
| 8577 | * string EQ */ |
| 8578 | while (hardcount < max |
| 8579 | && scan < loceol |
| 8580 | && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p) |
| 8581 | && memEQ(scan, STRING(p), scan_char_len)) |
| 8582 | { |
| 8583 | scan += scan_char_len; |
| 8584 | hardcount++; |
| 8585 | } |
| 8586 | } |
| 8587 | else if (! UTF8_IS_ABOVE_LATIN1(c)) { |
| 8588 | |
| 8589 | /* Target isn't utf8; convert the character in the UTF-8 |
| 8590 | * pattern to non-UTF8, and do a simple loop */ |
| 8591 | c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1)); |
| 8592 | while (scan < loceol && UCHARAT(scan) == c) { |
| 8593 | scan++; |
| 8594 | } |
| 8595 | } /* else pattern char is above Latin1, can't possibly match the |
| 8596 | non-UTF-8 target */ |
| 8597 | } |
| 8598 | else { |
| 8599 | |
| 8600 | /* Here, the string must be utf8; pattern isn't, and <c> is |
| 8601 | * different in utf8 than not, so can't compare them directly. |
| 8602 | * Outside the loop, find the two utf8 bytes that represent c, and |
| 8603 | * then look for those in sequence in the utf8 string */ |
| 8604 | U8 high = UTF8_TWO_BYTE_HI(c); |
| 8605 | U8 low = UTF8_TWO_BYTE_LO(c); |
| 8606 | |
| 8607 | while (hardcount < max |
| 8608 | && scan + 1 < loceol |
| 8609 | && UCHARAT(scan) == high |
| 8610 | && UCHARAT(scan + 1) == low) |
| 8611 | { |
| 8612 | scan += 2; |
| 8613 | hardcount++; |
| 8614 | } |
| 8615 | } |
| 8616 | break; |
| 8617 | |
| 8618 | case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ |
| 8619 | assert(! reginfo->is_utf8_pat); |
| 8620 | /* FALLTHROUGH */ |
| 8621 | case EXACTFA: |
| 8622 | utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 8623 | goto do_exactf; |
| 8624 | |
| 8625 | case EXACTFL: |
| 8626 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 8627 | utf8_flags = FOLDEQ_LOCALE; |
| 8628 | goto do_exactf; |
| 8629 | |
| 8630 | case EXACTF: /* This node only generated for non-utf8 patterns */ |
| 8631 | assert(! reginfo->is_utf8_pat); |
| 8632 | utf8_flags = 0; |
| 8633 | goto do_exactf; |
| 8634 | |
| 8635 | case EXACTFLU8: |
| 8636 | if (! utf8_target) { |
| 8637 | break; |
| 8638 | } |
| 8639 | utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED |
| 8640 | | FOLDEQ_S2_FOLDS_SANE; |
| 8641 | goto do_exactf; |
| 8642 | |
| 8643 | case EXACTFU_SS: |
| 8644 | case EXACTFU: |
| 8645 | utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; |
| 8646 | |
| 8647 | do_exactf: { |
| 8648 | int c1, c2; |
| 8649 | U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1]; |
| 8650 | |
| 8651 | assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); |
| 8652 | |
| 8653 | if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8, |
| 8654 | reginfo)) |
| 8655 | { |
| 8656 | if (c1 == CHRTEST_VOID) { |
| 8657 | /* Use full Unicode fold matching */ |
| 8658 | char *tmpeol = reginfo->strend; |
| 8659 | STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1; |
| 8660 | while (hardcount < max |
| 8661 | && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target, |
| 8662 | STRING(p), NULL, pat_len, |
| 8663 | reginfo->is_utf8_pat, utf8_flags)) |
| 8664 | { |
| 8665 | scan = tmpeol; |
| 8666 | tmpeol = reginfo->strend; |
| 8667 | hardcount++; |
| 8668 | } |
| 8669 | } |
| 8670 | else if (utf8_target) { |
| 8671 | if (c1 == c2) { |
| 8672 | while (scan < loceol |
| 8673 | && hardcount < max |
| 8674 | && memEQ(scan, c1_utf8, UTF8SKIP(scan))) |
| 8675 | { |
| 8676 | scan += UTF8SKIP(scan); |
| 8677 | hardcount++; |
| 8678 | } |
| 8679 | } |
| 8680 | else { |
| 8681 | while (scan < loceol |
| 8682 | && hardcount < max |
| 8683 | && (memEQ(scan, c1_utf8, UTF8SKIP(scan)) |
| 8684 | || memEQ(scan, c2_utf8, UTF8SKIP(scan)))) |
| 8685 | { |
| 8686 | scan += UTF8SKIP(scan); |
| 8687 | hardcount++; |
| 8688 | } |
| 8689 | } |
| 8690 | } |
| 8691 | else if (c1 == c2) { |
| 8692 | while (scan < loceol && UCHARAT(scan) == c1) { |
| 8693 | scan++; |
| 8694 | } |
| 8695 | } |
| 8696 | else { |
| 8697 | while (scan < loceol && |
| 8698 | (UCHARAT(scan) == c1 || UCHARAT(scan) == c2)) |
| 8699 | { |
| 8700 | scan++; |
| 8701 | } |
| 8702 | } |
| 8703 | } |
| 8704 | break; |
| 8705 | } |
| 8706 | case ANYOFL: |
| 8707 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 8708 | |
| 8709 | if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) { |
| 8710 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required); |
| 8711 | } |
| 8712 | /* FALLTHROUGH */ |
| 8713 | case ANYOFD: |
| 8714 | case ANYOF: |
| 8715 | if (utf8_target) { |
| 8716 | while (hardcount < max |
| 8717 | && scan < loceol |
| 8718 | && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target)) |
| 8719 | { |
| 8720 | scan += UTF8SKIP(scan); |
| 8721 | hardcount++; |
| 8722 | } |
| 8723 | } else { |
| 8724 | while (scan < loceol && REGINCLASS(prog, p, (U8*)scan, 0)) |
| 8725 | scan++; |
| 8726 | } |
| 8727 | break; |
| 8728 | |
| 8729 | /* The argument (FLAGS) to all the POSIX node types is the class number */ |
| 8730 | |
| 8731 | case NPOSIXL: |
| 8732 | to_complement = 1; |
| 8733 | /* FALLTHROUGH */ |
| 8734 | |
| 8735 | case POSIXL: |
| 8736 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 8737 | if (! utf8_target) { |
| 8738 | while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p), |
| 8739 | *scan))) |
| 8740 | { |
| 8741 | scan++; |
| 8742 | } |
| 8743 | } else { |
| 8744 | while (hardcount < max && scan < loceol |
| 8745 | && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p), |
| 8746 | (U8 *) scan))) |
| 8747 | { |
| 8748 | scan += UTF8SKIP(scan); |
| 8749 | hardcount++; |
| 8750 | } |
| 8751 | } |
| 8752 | break; |
| 8753 | |
| 8754 | case POSIXD: |
| 8755 | if (utf8_target) { |
| 8756 | goto utf8_posix; |
| 8757 | } |
| 8758 | /* FALLTHROUGH */ |
| 8759 | |
| 8760 | case POSIXA: |
| 8761 | if (utf8_target && loceol - scan > max) { |
| 8762 | |
| 8763 | /* We didn't adjust <loceol> at the beginning of this routine |
| 8764 | * because is UTF-8, but it is actually ok to do so, since here, to |
| 8765 | * match, 1 char == 1 byte. */ |
| 8766 | loceol = scan + max; |
| 8767 | } |
| 8768 | while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) { |
| 8769 | scan++; |
| 8770 | } |
| 8771 | break; |
| 8772 | |
| 8773 | case NPOSIXD: |
| 8774 | if (utf8_target) { |
| 8775 | to_complement = 1; |
| 8776 | goto utf8_posix; |
| 8777 | } |
| 8778 | /* FALLTHROUGH */ |
| 8779 | |
| 8780 | case NPOSIXA: |
| 8781 | if (! utf8_target) { |
| 8782 | while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) { |
| 8783 | scan++; |
| 8784 | } |
| 8785 | } |
| 8786 | else { |
| 8787 | |
| 8788 | /* The complement of something that matches only ASCII matches all |
| 8789 | * non-ASCII, plus everything in ASCII that isn't in the class. */ |
| 8790 | while (hardcount < max && scan < loceol |
| 8791 | && (! isASCII_utf8(scan) |
| 8792 | || ! _generic_isCC_A((U8) *scan, FLAGS(p)))) |
| 8793 | { |
| 8794 | scan += UTF8SKIP(scan); |
| 8795 | hardcount++; |
| 8796 | } |
| 8797 | } |
| 8798 | break; |
| 8799 | |
| 8800 | case NPOSIXU: |
| 8801 | to_complement = 1; |
| 8802 | /* FALLTHROUGH */ |
| 8803 | |
| 8804 | case POSIXU: |
| 8805 | if (! utf8_target) { |
| 8806 | while (scan < loceol && to_complement |
| 8807 | ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p)))) |
| 8808 | { |
| 8809 | scan++; |
| 8810 | } |
| 8811 | } |
| 8812 | else { |
| 8813 | utf8_posix: |
| 8814 | classnum = (_char_class_number) FLAGS(p); |
| 8815 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 8816 | |
| 8817 | /* Here, a swash is needed for above-Latin1 code points. |
| 8818 | * Process as many Latin1 code points using the built-in rules. |
| 8819 | * Go to another loop to finish processing upon encountering |
| 8820 | * the first Latin1 code point. We could do that in this loop |
| 8821 | * as well, but the other way saves having to test if the swash |
| 8822 | * has been loaded every time through the loop: extra space to |
| 8823 | * save a test. */ |
| 8824 | while (hardcount < max && scan < loceol) { |
| 8825 | if (UTF8_IS_INVARIANT(*scan)) { |
| 8826 | if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan, |
| 8827 | classnum)))) |
| 8828 | { |
| 8829 | break; |
| 8830 | } |
| 8831 | scan++; |
| 8832 | } |
| 8833 | else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) { |
| 8834 | if (! (to_complement |
| 8835 | ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*scan, |
| 8836 | *(scan + 1)), |
| 8837 | classnum)))) |
| 8838 | { |
| 8839 | break; |
| 8840 | } |
| 8841 | scan += 2; |
| 8842 | } |
| 8843 | else { |
| 8844 | goto found_above_latin1; |
| 8845 | } |
| 8846 | |
| 8847 | hardcount++; |
| 8848 | } |
| 8849 | } |
| 8850 | else { |
| 8851 | /* For these character classes, the knowledge of how to handle |
| 8852 | * every code point is compiled in to Perl via a macro. This |
| 8853 | * code is written for making the loops as tight as possible. |
| 8854 | * It could be refactored to save space instead */ |
| 8855 | switch (classnum) { |
| 8856 | case _CC_ENUM_SPACE: |
| 8857 | while (hardcount < max |
| 8858 | && scan < loceol |
| 8859 | && (to_complement ^ cBOOL(isSPACE_utf8(scan)))) |
| 8860 | { |
| 8861 | scan += UTF8SKIP(scan); |
| 8862 | hardcount++; |
| 8863 | } |
| 8864 | break; |
| 8865 | case _CC_ENUM_BLANK: |
| 8866 | while (hardcount < max |
| 8867 | && scan < loceol |
| 8868 | && (to_complement ^ cBOOL(isBLANK_utf8(scan)))) |
| 8869 | { |
| 8870 | scan += UTF8SKIP(scan); |
| 8871 | hardcount++; |
| 8872 | } |
| 8873 | break; |
| 8874 | case _CC_ENUM_XDIGIT: |
| 8875 | while (hardcount < max |
| 8876 | && scan < loceol |
| 8877 | && (to_complement ^ cBOOL(isXDIGIT_utf8(scan)))) |
| 8878 | { |
| 8879 | scan += UTF8SKIP(scan); |
| 8880 | hardcount++; |
| 8881 | } |
| 8882 | break; |
| 8883 | case _CC_ENUM_VERTSPACE: |
| 8884 | while (hardcount < max |
| 8885 | && scan < loceol |
| 8886 | && (to_complement ^ cBOOL(isVERTWS_utf8(scan)))) |
| 8887 | { |
| 8888 | scan += UTF8SKIP(scan); |
| 8889 | hardcount++; |
| 8890 | } |
| 8891 | break; |
| 8892 | case _CC_ENUM_CNTRL: |
| 8893 | while (hardcount < max |
| 8894 | && scan < loceol |
| 8895 | && (to_complement ^ cBOOL(isCNTRL_utf8(scan)))) |
| 8896 | { |
| 8897 | scan += UTF8SKIP(scan); |
| 8898 | hardcount++; |
| 8899 | } |
| 8900 | break; |
| 8901 | default: |
| 8902 | Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum); |
| 8903 | } |
| 8904 | } |
| 8905 | } |
| 8906 | break; |
| 8907 | |
| 8908 | found_above_latin1: /* Continuation of POSIXU and NPOSIXU */ |
| 8909 | |
| 8910 | /* Load the swash if not already present */ |
| 8911 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 8912 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 8913 | PL_utf8_swash_ptrs[classnum] = _core_swash_init( |
| 8914 | "utf8", |
| 8915 | "", |
| 8916 | &PL_sv_undef, 1, 0, |
| 8917 | PL_XPosix_ptrs[classnum], &flags); |
| 8918 | } |
| 8919 | |
| 8920 | while (hardcount < max && scan < loceol |
| 8921 | && to_complement ^ cBOOL(_generic_utf8( |
| 8922 | classnum, |
| 8923 | scan, |
| 8924 | swash_fetch(PL_utf8_swash_ptrs[classnum], |
| 8925 | (U8 *) scan, |
| 8926 | TRUE)))) |
| 8927 | { |
| 8928 | scan += UTF8SKIP(scan); |
| 8929 | hardcount++; |
| 8930 | } |
| 8931 | break; |
| 8932 | |
| 8933 | case LNBREAK: |
| 8934 | if (utf8_target) { |
| 8935 | while (hardcount < max && scan < loceol && |
| 8936 | (c=is_LNBREAK_utf8_safe(scan, loceol))) { |
| 8937 | scan += c; |
| 8938 | hardcount++; |
| 8939 | } |
| 8940 | } else { |
| 8941 | /* LNBREAK can match one or two latin chars, which is ok, but we |
| 8942 | * have to use hardcount in this situation, and throw away the |
| 8943 | * adjustment to <loceol> done before the switch statement */ |
| 8944 | loceol = reginfo->strend; |
| 8945 | while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) { |
| 8946 | scan+=c; |
| 8947 | hardcount++; |
| 8948 | } |
| 8949 | } |
| 8950 | break; |
| 8951 | |
| 8952 | case BOUNDL: |
| 8953 | case NBOUNDL: |
| 8954 | _CHECK_AND_WARN_PROBLEMATIC_LOCALE; |
| 8955 | /* FALLTHROUGH */ |
| 8956 | case BOUND: |
| 8957 | case BOUNDA: |
| 8958 | case BOUNDU: |
| 8959 | case EOS: |
| 8960 | case GPOS: |
| 8961 | case KEEPS: |
| 8962 | case NBOUND: |
| 8963 | case NBOUNDA: |
| 8964 | case NBOUNDU: |
| 8965 | case OPFAIL: |
| 8966 | case SBOL: |
| 8967 | case SEOL: |
| 8968 | /* These are all 0 width, so match right here or not at all. */ |
| 8969 | break; |
| 8970 | |
| 8971 | default: |
| 8972 | Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]); |
| 8973 | NOT_REACHED; /* NOTREACHED */ |
| 8974 | |
| 8975 | } |
| 8976 | |
| 8977 | if (hardcount) |
| 8978 | c = hardcount; |
| 8979 | else |
| 8980 | c = scan - *startposp; |
| 8981 | *startposp = scan; |
| 8982 | |
| 8983 | DEBUG_r({ |
| 8984 | GET_RE_DEBUG_FLAGS_DECL; |
| 8985 | DEBUG_EXECUTE_r({ |
| 8986 | SV * const prop = sv_newmortal(); |
| 8987 | regprop(prog, prop, p, reginfo, NULL); |
| 8988 | Perl_re_exec_indentf( aTHX_ "%s can match %"IVdf" times out of %"IVdf"...\n", |
| 8989 | depth, SvPVX_const(prop),(IV)c,(IV)max); |
| 8990 | }); |
| 8991 | }); |
| 8992 | |
| 8993 | return(c); |
| 8994 | } |
| 8995 | |
| 8996 | |
| 8997 | #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) |
| 8998 | /* |
| 8999 | - regclass_swash - prepare the utf8 swash. Wraps the shared core version to |
| 9000 | create a copy so that changes the caller makes won't change the shared one. |
| 9001 | If <altsvp> is non-null, will return NULL in it, for back-compat. |
| 9002 | */ |
| 9003 | SV * |
| 9004 | Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp) |
| 9005 | { |
| 9006 | PERL_ARGS_ASSERT_REGCLASS_SWASH; |
| 9007 | |
| 9008 | if (altsvp) { |
| 9009 | *altsvp = NULL; |
| 9010 | } |
| 9011 | |
| 9012 | return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL)); |
| 9013 | } |
| 9014 | |
| 9015 | #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */ |
| 9016 | |
| 9017 | /* |
| 9018 | - reginclass - determine if a character falls into a character class |
| 9019 | |
| 9020 | n is the ANYOF-type regnode |
| 9021 | p is the target string |
| 9022 | p_end points to one byte beyond the end of the target string |
| 9023 | utf8_target tells whether p is in UTF-8. |
| 9024 | |
| 9025 | Returns true if matched; false otherwise. |
| 9026 | |
| 9027 | Note that this can be a synthetic start class, a combination of various |
| 9028 | nodes, so things you think might be mutually exclusive, such as locale, |
| 9029 | aren't. It can match both locale and non-locale |
| 9030 | |
| 9031 | */ |
| 9032 | |
| 9033 | STATIC bool |
| 9034 | S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target) |
| 9035 | { |
| 9036 | dVAR; |
| 9037 | const char flags = ANYOF_FLAGS(n); |
| 9038 | bool match = FALSE; |
| 9039 | UV c = *p; |
| 9040 | |
| 9041 | PERL_ARGS_ASSERT_REGINCLASS; |
| 9042 | |
| 9043 | /* If c is not already the code point, get it. Note that |
| 9044 | * UTF8_IS_INVARIANT() works even if not in UTF-8 */ |
| 9045 | if (! UTF8_IS_INVARIANT(c) && utf8_target) { |
| 9046 | STRLEN c_len = 0; |
| 9047 | c = utf8n_to_uvchr(p, p_end - p, &c_len, |
| 9048 | (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) |
| 9049 | | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY); |
| 9050 | /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for |
| 9051 | * UTF8_ALLOW_FFFF */ |
| 9052 | if (c_len == (STRLEN)-1) |
| 9053 | Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); |
| 9054 | if (c > 255 && OP(n) == ANYOFL && ! ANYOFL_UTF8_LOCALE_REQD(flags)) { |
| 9055 | _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c); |
| 9056 | } |
| 9057 | } |
| 9058 | |
| 9059 | /* If this character is potentially in the bitmap, check it */ |
| 9060 | if (c < NUM_ANYOF_CODE_POINTS) { |
| 9061 | if (ANYOF_BITMAP_TEST(n, c)) |
| 9062 | match = TRUE; |
| 9063 | else if ((flags |
| 9064 | & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER) |
| 9065 | && OP(n) == ANYOFD |
| 9066 | && ! utf8_target |
| 9067 | && ! isASCII(c)) |
| 9068 | { |
| 9069 | match = TRUE; |
| 9070 | } |
| 9071 | else if (flags & ANYOF_LOCALE_FLAGS) { |
| 9072 | if ((flags & ANYOFL_FOLD) |
| 9073 | && c < 256 |
| 9074 | && ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) |
| 9075 | { |
| 9076 | match = TRUE; |
| 9077 | } |
| 9078 | else if (ANYOF_POSIXL_TEST_ANY_SET(n) |
| 9079 | && c < 256 |
| 9080 | ) { |
| 9081 | |
| 9082 | /* The data structure is arranged so bits 0, 2, 4, ... are set |
| 9083 | * if the class includes the Posix character class given by |
| 9084 | * bit/2; and 1, 3, 5, ... are set if the class includes the |
| 9085 | * complemented Posix class given by int(bit/2). So we loop |
| 9086 | * through the bits, each time changing whether we complement |
| 9087 | * the result or not. Suppose for the sake of illustration |
| 9088 | * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0 |
| 9089 | * is set, it means there is a match for this ANYOF node if the |
| 9090 | * character is in the class given by the expression (0 / 2 = 0 |
| 9091 | * = \w). If it is in that class, isFOO_lc() will return 1, |
| 9092 | * and since 'to_complement' is 0, the result will stay TRUE, |
| 9093 | * and we exit the loop. Suppose instead that bit 0 is 0, but |
| 9094 | * bit 1 is 1. That means there is a match if the character |
| 9095 | * matches \W. We won't bother to call isFOO_lc() on bit 0, |
| 9096 | * but will on bit 1. On the second iteration 'to_complement' |
| 9097 | * will be 1, so the exclusive or will reverse things, so we |
| 9098 | * are testing for \W. On the third iteration, 'to_complement' |
| 9099 | * will be 0, and we would be testing for \s; the fourth |
| 9100 | * iteration would test for \S, etc. |
| 9101 | * |
| 9102 | * Note that this code assumes that all the classes are closed |
| 9103 | * under folding. For example, if a character matches \w, then |
| 9104 | * its fold does too; and vice versa. This should be true for |
| 9105 | * any well-behaved locale for all the currently defined Posix |
| 9106 | * classes, except for :lower: and :upper:, which are handled |
| 9107 | * by the pseudo-class :cased: which matches if either of the |
| 9108 | * other two does. To get rid of this assumption, an outer |
| 9109 | * loop could be used below to iterate over both the source |
| 9110 | * character, and its fold (if different) */ |
| 9111 | |
| 9112 | int count = 0; |
| 9113 | int to_complement = 0; |
| 9114 | |
| 9115 | while (count < ANYOF_MAX) { |
| 9116 | if (ANYOF_POSIXL_TEST(n, count) |
| 9117 | && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c))) |
| 9118 | { |
| 9119 | match = TRUE; |
| 9120 | break; |
| 9121 | } |
| 9122 | count++; |
| 9123 | to_complement ^= 1; |
| 9124 | } |
| 9125 | } |
| 9126 | } |
| 9127 | } |
| 9128 | |
| 9129 | |
| 9130 | /* If the bitmap didn't (or couldn't) match, and something outside the |
| 9131 | * bitmap could match, try that. */ |
| 9132 | if (!match) { |
| 9133 | if (c >= NUM_ANYOF_CODE_POINTS |
| 9134 | && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP)) |
| 9135 | { |
| 9136 | match = TRUE; /* Everything above the bitmap matches */ |
| 9137 | } |
| 9138 | /* Here doesn't match everything above the bitmap. If there is |
| 9139 | * some information available beyond the bitmap, we may find a |
| 9140 | * match in it. If so, this is most likely because the code point |
| 9141 | * is outside the bitmap range. But rarely, it could be because of |
| 9142 | * some other reason. If so, various flags are set to indicate |
| 9143 | * this possibility. On ANYOFD nodes, there may be matches that |
| 9144 | * happen only when the target string is UTF-8; or for other node |
| 9145 | * types, because runtime lookup is needed, regardless of the |
| 9146 | * UTF-8ness of the target string. Finally, under /il, there may |
| 9147 | * be some matches only possible if the locale is a UTF-8 one. */ |
| 9148 | else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP |
| 9149 | && ( c >= NUM_ANYOF_CODE_POINTS |
| 9150 | || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP) |
| 9151 | && ( UNLIKELY(OP(n) != ANYOFD) |
| 9152 | || (utf8_target && ! isASCII_uni(c) |
| 9153 | # if NUM_ANYOF_CODE_POINTS > 256 |
| 9154 | && c < 256 |
| 9155 | # endif |
| 9156 | ))) |
| 9157 | || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) |
| 9158 | && IN_UTF8_CTYPE_LOCALE))) |
| 9159 | { |
| 9160 | SV* only_utf8_locale = NULL; |
| 9161 | SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0, |
| 9162 | &only_utf8_locale, NULL); |
| 9163 | if (sw) { |
| 9164 | U8 utf8_buffer[2]; |
| 9165 | U8 * utf8_p; |
| 9166 | if (utf8_target) { |
| 9167 | utf8_p = (U8 *) p; |
| 9168 | } else { /* Convert to utf8 */ |
| 9169 | utf8_p = utf8_buffer; |
| 9170 | append_utf8_from_native_byte(*p, &utf8_p); |
| 9171 | utf8_p = utf8_buffer; |
| 9172 | } |
| 9173 | |
| 9174 | if (swash_fetch(sw, utf8_p, TRUE)) { |
| 9175 | match = TRUE; |
| 9176 | } |
| 9177 | } |
| 9178 | if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) { |
| 9179 | match = _invlist_contains_cp(only_utf8_locale, c); |
| 9180 | } |
| 9181 | } |
| 9182 | |
| 9183 | if (UNICODE_IS_SUPER(c) |
| 9184 | && (flags |
| 9185 | & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER) |
| 9186 | && OP(n) != ANYOFD |
| 9187 | && ckWARN_d(WARN_NON_UNICODE)) |
| 9188 | { |
| 9189 | Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), |
| 9190 | "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c); |
| 9191 | } |
| 9192 | } |
| 9193 | |
| 9194 | #if ANYOF_INVERT != 1 |
| 9195 | /* Depending on compiler optimization cBOOL takes time, so if don't have to |
| 9196 | * use it, don't */ |
| 9197 | # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below, |
| 9198 | #endif |
| 9199 | |
| 9200 | /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */ |
| 9201 | return (flags & ANYOF_INVERT) ^ match; |
| 9202 | } |
| 9203 | |
| 9204 | STATIC U8 * |
| 9205 | S_reghop3(U8 *s, SSize_t off, const U8* lim) |
| 9206 | { |
| 9207 | /* return the position 'off' UTF-8 characters away from 's', forward if |
| 9208 | * 'off' >= 0, backwards if negative. But don't go outside of position |
| 9209 | * 'lim', which better be < s if off < 0 */ |
| 9210 | |
| 9211 | PERL_ARGS_ASSERT_REGHOP3; |
| 9212 | |
| 9213 | if (off >= 0) { |
| 9214 | while (off-- && s < lim) { |
| 9215 | /* XXX could check well-formedness here */ |
| 9216 | s += UTF8SKIP(s); |
| 9217 | } |
| 9218 | } |
| 9219 | else { |
| 9220 | while (off++ && s > lim) { |
| 9221 | s--; |
| 9222 | if (UTF8_IS_CONTINUED(*s)) { |
| 9223 | while (s > lim && UTF8_IS_CONTINUATION(*s)) |
| 9224 | s--; |
| 9225 | if (! UTF8_IS_START(*s)) { |
| 9226 | Perl_croak_nocontext("Malformed UTF-8 character (fatal)"); |
| 9227 | } |
| 9228 | } |
| 9229 | /* XXX could check well-formedness here */ |
| 9230 | } |
| 9231 | } |
| 9232 | return s; |
| 9233 | } |
| 9234 | |
| 9235 | STATIC U8 * |
| 9236 | S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim) |
| 9237 | { |
| 9238 | PERL_ARGS_ASSERT_REGHOP4; |
| 9239 | |
| 9240 | if (off >= 0) { |
| 9241 | while (off-- && s < rlim) { |
| 9242 | /* XXX could check well-formedness here */ |
| 9243 | s += UTF8SKIP(s); |
| 9244 | } |
| 9245 | } |
| 9246 | else { |
| 9247 | while (off++ && s > llim) { |
| 9248 | s--; |
| 9249 | if (UTF8_IS_CONTINUED(*s)) { |
| 9250 | while (s > llim && UTF8_IS_CONTINUATION(*s)) |
| 9251 | s--; |
| 9252 | if (! UTF8_IS_START(*s)) { |
| 9253 | Perl_croak_nocontext("Malformed UTF-8 character (fatal)"); |
| 9254 | } |
| 9255 | } |
| 9256 | /* XXX could check well-formedness here */ |
| 9257 | } |
| 9258 | } |
| 9259 | return s; |
| 9260 | } |
| 9261 | |
| 9262 | /* like reghop3, but returns NULL on overrun, rather than returning last |
| 9263 | * char pos */ |
| 9264 | |
| 9265 | STATIC U8 * |
| 9266 | S_reghopmaybe3(U8* s, SSize_t off, const U8* lim) |
| 9267 | { |
| 9268 | PERL_ARGS_ASSERT_REGHOPMAYBE3; |
| 9269 | |
| 9270 | if (off >= 0) { |
| 9271 | while (off-- && s < lim) { |
| 9272 | /* XXX could check well-formedness here */ |
| 9273 | s += UTF8SKIP(s); |
| 9274 | } |
| 9275 | if (off >= 0) |
| 9276 | return NULL; |
| 9277 | } |
| 9278 | else { |
| 9279 | while (off++ && s > lim) { |
| 9280 | s--; |
| 9281 | if (UTF8_IS_CONTINUED(*s)) { |
| 9282 | while (s > lim && UTF8_IS_CONTINUATION(*s)) |
| 9283 | s--; |
| 9284 | if (! UTF8_IS_START(*s)) { |
| 9285 | Perl_croak_nocontext("Malformed UTF-8 character (fatal)"); |
| 9286 | } |
| 9287 | } |
| 9288 | /* XXX could check well-formedness here */ |
| 9289 | } |
| 9290 | if (off <= 0) |
| 9291 | return NULL; |
| 9292 | } |
| 9293 | return s; |
| 9294 | } |
| 9295 | |
| 9296 | |
| 9297 | /* when executing a regex that may have (?{}), extra stuff needs setting |
| 9298 | up that will be visible to the called code, even before the current |
| 9299 | match has finished. In particular: |
| 9300 | |
| 9301 | * $_ is localised to the SV currently being matched; |
| 9302 | * pos($_) is created if necessary, ready to be updated on each call-out |
| 9303 | to code; |
| 9304 | * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm |
| 9305 | isn't set until the current pattern is successfully finished), so that |
| 9306 | $1 etc of the match-so-far can be seen; |
| 9307 | * save the old values of subbeg etc of the current regex, and set then |
| 9308 | to the current string (again, this is normally only done at the end |
| 9309 | of execution) |
| 9310 | */ |
| 9311 | |
| 9312 | static void |
| 9313 | S_setup_eval_state(pTHX_ regmatch_info *const reginfo) |
| 9314 | { |
| 9315 | MAGIC *mg; |
| 9316 | regexp *const rex = ReANY(reginfo->prog); |
| 9317 | regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval; |
| 9318 | |
| 9319 | eval_state->rex = rex; |
| 9320 | |
| 9321 | if (reginfo->sv) { |
| 9322 | /* Make $_ available to executed code. */ |
| 9323 | if (reginfo->sv != DEFSV) { |
| 9324 | SAVE_DEFSV; |
| 9325 | DEFSV_set(reginfo->sv); |
| 9326 | } |
| 9327 | |
| 9328 | if (!(mg = mg_find_mglob(reginfo->sv))) { |
| 9329 | /* prepare for quick setting of pos */ |
| 9330 | mg = sv_magicext_mglob(reginfo->sv); |
| 9331 | mg->mg_len = -1; |
| 9332 | } |
| 9333 | eval_state->pos_magic = mg; |
| 9334 | eval_state->pos = mg->mg_len; |
| 9335 | eval_state->pos_flags = mg->mg_flags; |
| 9336 | } |
| 9337 | else |
| 9338 | eval_state->pos_magic = NULL; |
| 9339 | |
| 9340 | if (!PL_reg_curpm) { |
| 9341 | /* PL_reg_curpm is a fake PMOP that we can attach the current |
| 9342 | * regex to and point PL_curpm at, so that $1 et al are visible |
| 9343 | * within a /(?{})/. It's just allocated once per interpreter the |
| 9344 | * first time its needed */ |
| 9345 | Newxz(PL_reg_curpm, 1, PMOP); |
| 9346 | #ifdef USE_ITHREADS |
| 9347 | { |
| 9348 | SV* const repointer = &PL_sv_undef; |
| 9349 | /* this regexp is also owned by the new PL_reg_curpm, which |
| 9350 | will try to free it. */ |
| 9351 | av_push(PL_regex_padav, repointer); |
| 9352 | PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav); |
| 9353 | PL_regex_pad = AvARRAY(PL_regex_padav); |
| 9354 | } |
| 9355 | #endif |
| 9356 | } |
| 9357 | SET_reg_curpm(reginfo->prog); |
| 9358 | eval_state->curpm = PL_curpm; |
| 9359 | PL_curpm = PL_reg_curpm; |
| 9360 | if (RXp_MATCH_COPIED(rex)) { |
| 9361 | /* Here is a serious problem: we cannot rewrite subbeg, |
| 9362 | since it may be needed if this match fails. Thus |
| 9363 | $` inside (?{}) could fail... */ |
| 9364 | eval_state->subbeg = rex->subbeg; |
| 9365 | eval_state->sublen = rex->sublen; |
| 9366 | eval_state->suboffset = rex->suboffset; |
| 9367 | eval_state->subcoffset = rex->subcoffset; |
| 9368 | #ifdef PERL_ANY_COW |
| 9369 | eval_state->saved_copy = rex->saved_copy; |
| 9370 | #endif |
| 9371 | RXp_MATCH_COPIED_off(rex); |
| 9372 | } |
| 9373 | else |
| 9374 | eval_state->subbeg = NULL; |
| 9375 | rex->subbeg = (char *)reginfo->strbeg; |
| 9376 | rex->suboffset = 0; |
| 9377 | rex->subcoffset = 0; |
| 9378 | rex->sublen = reginfo->strend - reginfo->strbeg; |
| 9379 | } |
| 9380 | |
| 9381 | |
| 9382 | /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */ |
| 9383 | |
| 9384 | static void |
| 9385 | S_cleanup_regmatch_info_aux(pTHX_ void *arg) |
| 9386 | { |
| 9387 | regmatch_info_aux *aux = (regmatch_info_aux *) arg; |
| 9388 | regmatch_info_aux_eval *eval_state = aux->info_aux_eval; |
| 9389 | regmatch_slab *s; |
| 9390 | |
| 9391 | Safefree(aux->poscache); |
| 9392 | |
| 9393 | if (eval_state) { |
| 9394 | |
| 9395 | /* undo the effects of S_setup_eval_state() */ |
| 9396 | |
| 9397 | if (eval_state->subbeg) { |
| 9398 | regexp * const rex = eval_state->rex; |
| 9399 | rex->subbeg = eval_state->subbeg; |
| 9400 | rex->sublen = eval_state->sublen; |
| 9401 | rex->suboffset = eval_state->suboffset; |
| 9402 | rex->subcoffset = eval_state->subcoffset; |
| 9403 | #ifdef PERL_ANY_COW |
| 9404 | rex->saved_copy = eval_state->saved_copy; |
| 9405 | #endif |
| 9406 | RXp_MATCH_COPIED_on(rex); |
| 9407 | } |
| 9408 | if (eval_state->pos_magic) |
| 9409 | { |
| 9410 | eval_state->pos_magic->mg_len = eval_state->pos; |
| 9411 | eval_state->pos_magic->mg_flags = |
| 9412 | (eval_state->pos_magic->mg_flags & ~MGf_BYTES) |
| 9413 | | (eval_state->pos_flags & MGf_BYTES); |
| 9414 | } |
| 9415 | |
| 9416 | PL_curpm = eval_state->curpm; |
| 9417 | } |
| 9418 | |
| 9419 | PL_regmatch_state = aux->old_regmatch_state; |
| 9420 | PL_regmatch_slab = aux->old_regmatch_slab; |
| 9421 | |
| 9422 | /* free all slabs above current one - this must be the last action |
| 9423 | * of this function, as aux and eval_state are allocated within |
| 9424 | * slabs and may be freed here */ |
| 9425 | |
| 9426 | s = PL_regmatch_slab->next; |
| 9427 | if (s) { |
| 9428 | PL_regmatch_slab->next = NULL; |
| 9429 | while (s) { |
| 9430 | regmatch_slab * const osl = s; |
| 9431 | s = s->next; |
| 9432 | Safefree(osl); |
| 9433 | } |
| 9434 | } |
| 9435 | } |
| 9436 | |
| 9437 | |
| 9438 | STATIC void |
| 9439 | S_to_utf8_substr(pTHX_ regexp *prog) |
| 9440 | { |
| 9441 | /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile |
| 9442 | * on the converted value */ |
| 9443 | |
| 9444 | int i = 1; |
| 9445 | |
| 9446 | PERL_ARGS_ASSERT_TO_UTF8_SUBSTR; |
| 9447 | |
| 9448 | do { |
| 9449 | if (prog->substrs->data[i].substr |
| 9450 | && !prog->substrs->data[i].utf8_substr) { |
| 9451 | SV* const sv = newSVsv(prog->substrs->data[i].substr); |
| 9452 | prog->substrs->data[i].utf8_substr = sv; |
| 9453 | sv_utf8_upgrade(sv); |
| 9454 | if (SvVALID(prog->substrs->data[i].substr)) { |
| 9455 | if (SvTAIL(prog->substrs->data[i].substr)) { |
| 9456 | /* Trim the trailing \n that fbm_compile added last |
| 9457 | time. */ |
| 9458 | SvCUR_set(sv, SvCUR(sv) - 1); |
| 9459 | /* Whilst this makes the SV technically "invalid" (as its |
| 9460 | buffer is no longer followed by "\0") when fbm_compile() |
| 9461 | adds the "\n" back, a "\0" is restored. */ |
| 9462 | fbm_compile(sv, FBMcf_TAIL); |
| 9463 | } else |
| 9464 | fbm_compile(sv, 0); |
| 9465 | } |
| 9466 | if (prog->substrs->data[i].substr == prog->check_substr) |
| 9467 | prog->check_utf8 = sv; |
| 9468 | } |
| 9469 | } while (i--); |
| 9470 | } |
| 9471 | |
| 9472 | STATIC bool |
| 9473 | S_to_byte_substr(pTHX_ regexp *prog) |
| 9474 | { |
| 9475 | /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile |
| 9476 | * on the converted value; returns FALSE if can't be converted. */ |
| 9477 | |
| 9478 | int i = 1; |
| 9479 | |
| 9480 | PERL_ARGS_ASSERT_TO_BYTE_SUBSTR; |
| 9481 | |
| 9482 | do { |
| 9483 | if (prog->substrs->data[i].utf8_substr |
| 9484 | && !prog->substrs->data[i].substr) { |
| 9485 | SV* sv = newSVsv(prog->substrs->data[i].utf8_substr); |
| 9486 | if (! sv_utf8_downgrade(sv, TRUE)) { |
| 9487 | return FALSE; |
| 9488 | } |
| 9489 | if (SvVALID(prog->substrs->data[i].utf8_substr)) { |
| 9490 | if (SvTAIL(prog->substrs->data[i].utf8_substr)) { |
| 9491 | /* Trim the trailing \n that fbm_compile added last |
| 9492 | time. */ |
| 9493 | SvCUR_set(sv, SvCUR(sv) - 1); |
| 9494 | fbm_compile(sv, FBMcf_TAIL); |
| 9495 | } else |
| 9496 | fbm_compile(sv, 0); |
| 9497 | } |
| 9498 | prog->substrs->data[i].substr = sv; |
| 9499 | if (prog->substrs->data[i].utf8_substr == prog->check_utf8) |
| 9500 | prog->check_substr = sv; |
| 9501 | } |
| 9502 | } while (i--); |
| 9503 | |
| 9504 | return TRUE; |
| 9505 | } |
| 9506 | |
| 9507 | /* |
| 9508 | * ex: set ts=8 sts=4 sw=4 et: |
| 9509 | */ |