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, trie_flu8_latin } \ 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 ? (utf8_target \ 1465 ? trie_flu8 \ 1466 : trie_flu8_latin) \ 1467 : (utf8_target \ 1468 ? trie_utf8_fold \ 1469 : trie_latin_utf8_fold))) 1470 1471 /* 'uscan' is set to foldbuf, and incremented, so below the end of uscan is 1472 * 'foldbuf+sizeof(foldbuf)' */ 1473 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uc_end, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \ 1474 STMT_START { \ 1475 STRLEN skiplen; \ 1476 U8 flags = FOLD_FLAGS_FULL; \ 1477 switch (trie_type) { \ 1478 case trie_flu8: \ 1479 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ 1480 if (UTF8_IS_ABOVE_LATIN1(*uc)) { \ 1481 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end - uc); \ 1482 } \ 1483 goto do_trie_utf8_fold; \ 1484 case trie_utf8_exactfa_fold: \ 1485 flags |= FOLD_FLAGS_NOMIX_ASCII; \ 1486 /* FALLTHROUGH */ \ 1487 case trie_utf8_fold: \ 1488 do_trie_utf8_fold: \ 1489 if ( foldlen>0 ) { \ 1490 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \ 1491 foldlen -= len; \ 1492 uscan += len; \ 1493 len=0; \ 1494 } else { \ 1495 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \ 1496 len = UTF8SKIP(uc); \ 1497 skiplen = UVCHR_SKIP( uvc ); \ 1498 foldlen -= skiplen; \ 1499 uscan = foldbuf + skiplen; \ 1500 } \ 1501 break; \ 1502 case trie_flu8_latin: \ 1503 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ 1504 goto do_trie_latin_utf8_fold; \ 1505 case trie_latin_utf8_exactfa_fold: \ 1506 flags |= FOLD_FLAGS_NOMIX_ASCII; \ 1507 /* FALLTHROUGH */ \ 1508 case trie_latin_utf8_fold: \ 1509 do_trie_latin_utf8_fold: \ 1510 if ( foldlen>0 ) { \ 1511 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \ 1512 foldlen -= len; \ 1513 uscan += len; \ 1514 len=0; \ 1515 } else { \ 1516 len = 1; \ 1517 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \ 1518 skiplen = UVCHR_SKIP( uvc ); \ 1519 foldlen -= skiplen; \ 1520 uscan = foldbuf + skiplen; \ 1521 } \ 1522 break; \ 1523 case trie_utf8l: \ 1524 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \ 1525 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \ 1526 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \ 1527 } \ 1528 /* FALLTHROUGH */ \ 1529 case trie_utf8: \ 1530 uvc = utf8n_to_uvchr( (const U8*) uc, uc_end - uc, &len, uniflags ); \ 1531 break; \ 1532 case trie_plain: \ 1533 uvc = (UV)*uc; \ 1534 len = 1; \ 1535 } \ 1536 if (uvc < 256) { \ 1537 charid = trie->charmap[ uvc ]; \ 1538 } \ 1539 else { \ 1540 charid = 0; \ 1541 if (widecharmap) { \ 1542 SV** const svpp = hv_fetch(widecharmap, \ 1543 (char*)&uvc, sizeof(UV), 0); \ 1544 if (svpp) \ 1545 charid = (U16)SvIV(*svpp); \ 1546 } \ 1547 } \ 1548 } STMT_END 1549 1550 #define DUMP_EXEC_POS(li,s,doutf8,depth) \ 1551 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \ 1552 startpos, doutf8, depth) 1553 1554 #define REXEC_FBC_EXACTISH_SCAN(COND) \ 1555 STMT_START { \ 1556 while (s <= e) { \ 1557 if ( (COND) \ 1558 && (ln == 1 || folder(s, pat_string, ln)) \ 1559 && (reginfo->intuit || regtry(reginfo, &s)) )\ 1560 goto got_it; \ 1561 s++; \ 1562 } \ 1563 } STMT_END 1564 1565 #define REXEC_FBC_UTF8_SCAN(CODE) \ 1566 STMT_START { \ 1567 while (s < strend) { \ 1568 CODE \ 1569 s += UTF8SKIP(s); \ 1570 } \ 1571 } STMT_END 1572 1573 #define REXEC_FBC_SCAN(CODE) \ 1574 STMT_START { \ 1575 while (s < strend) { \ 1576 CODE \ 1577 s++; \ 1578 } \ 1579 } STMT_END 1580 1581 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \ 1582 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \ 1583 if (COND) { \ 1584 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ 1585 goto got_it; \ 1586 else \ 1587 tmp = doevery; \ 1588 } \ 1589 else \ 1590 tmp = 1; \ 1591 ) 1592 1593 #define REXEC_FBC_CLASS_SCAN(COND) \ 1594 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \ 1595 if (COND) { \ 1596 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ 1597 goto got_it; \ 1598 else \ 1599 tmp = doevery; \ 1600 } \ 1601 else \ 1602 tmp = 1; \ 1603 ) 1604 1605 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \ 1606 if (utf8_target) { \ 1607 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \ 1608 } \ 1609 else { \ 1610 REXEC_FBC_CLASS_SCAN(COND); \ 1611 } 1612 1613 /* The three macros below are slightly different versions of the same logic. 1614 * 1615 * The first is for /a and /aa when the target string is UTF-8. This can only 1616 * match ascii, but it must advance based on UTF-8. The other two handle the 1617 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking 1618 * for the boundary (or non-boundary) between a word and non-word character. 1619 * The utf8 and non-utf8 cases have the same logic, but the details must be 1620 * different. Find the "wordness" of the character just prior to this one, and 1621 * compare it with the wordness of this one. If they differ, we have a 1622 * boundary. At the beginning of the string, pretend that the previous 1623 * character was a new-line. 1624 * 1625 * All these macros uncleanly have side-effects with each other and outside 1626 * variables. So far it's been too much trouble to clean-up 1627 * 1628 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is 1629 * a word character or not. 1630 * IF_SUCCESS is code to do if it finds that we are at a boundary between 1631 * word/non-word 1632 * IF_FAIL is code to do if we aren't at a boundary between word/non-word 1633 * 1634 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we 1635 * are looking for a boundary or for a non-boundary. If we are looking for a 1636 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and 1637 * see if this tentative match actually works, and if so, to quit the loop 1638 * here. And vice-versa if we are looking for a non-boundary. 1639 * 1640 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and 1641 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of 1642 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be 1643 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal 1644 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that 1645 * complement. But in that branch we complement tmp, meaning that at the 1646 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s), 1647 * which means at the top of the loop in the next iteration, it is 1648 * TEST_NON_UTF8(s-1) */ 1649 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ 1650 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ 1651 tmp = TEST_NON_UTF8(tmp); \ 1652 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \ 1653 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ 1654 tmp = !tmp; \ 1655 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \ 1656 } \ 1657 else { \ 1658 IF_FAIL; \ 1659 } \ 1660 ); \ 1661 1662 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and 1663 * TEST_UTF8 is a macro that for the same input code points returns identically 1664 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */ 1665 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \ 1666 if (s == reginfo->strbeg) { \ 1667 tmp = '\n'; \ 1668 } \ 1669 else { /* Back-up to the start of the previous character */ \ 1670 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \ 1671 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \ 1672 0, UTF8_ALLOW_DEFAULT); \ 1673 } \ 1674 tmp = TEST_UV(tmp); \ 1675 LOAD_UTF8_CHARCLASS_ALNUM(); \ 1676 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \ 1677 if (tmp == ! (TEST_UTF8((U8 *) s))) { \ 1678 tmp = !tmp; \ 1679 IF_SUCCESS; \ 1680 } \ 1681 else { \ 1682 IF_FAIL; \ 1683 } \ 1684 ); 1685 1686 /* Like the above two macros. UTF8_CODE is the complete code for handling 1687 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc 1688 * macros below */ 1689 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ 1690 if (utf8_target) { \ 1691 UTF8_CODE \ 1692 } \ 1693 else { /* Not utf8 */ \ 1694 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ 1695 tmp = TEST_NON_UTF8(tmp); \ 1696 REXEC_FBC_SCAN( /* advances s while s < strend */ \ 1697 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ 1698 IF_SUCCESS; \ 1699 tmp = !tmp; \ 1700 } \ 1701 else { \ 1702 IF_FAIL; \ 1703 } \ 1704 ); \ 1705 } \ 1706 /* Here, things have been set up by the previous code so that tmp is the \ 1707 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \ 1708 * utf8ness of the target). We also have to check if this matches against \ 1709 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \ 1710 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \ 1711 * string */ \ 1712 if (tmp == ! TEST_NON_UTF8('\n')) { \ 1713 IF_SUCCESS; \ 1714 } \ 1715 else { \ 1716 IF_FAIL; \ 1717 } 1718 1719 /* This is the macro to use when we want to see if something that looks like it 1720 * could match, actually does, and if so exits the loop */ 1721 #define REXEC_FBC_TRYIT \ 1722 if ((reginfo->intuit || regtry(reginfo, &s))) \ 1723 goto got_it 1724 1725 /* The only difference between the BOUND and NBOUND cases is that 1726 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in 1727 * NBOUND. This is accomplished by passing it as either the if or else clause, 1728 * with the other one being empty (PLACEHOLDER is defined as empty). 1729 * 1730 * The TEST_FOO parameters are for operating on different forms of input, but 1731 * all should be ones that return identically for the same underlying code 1732 * points */ 1733 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ 1734 FBC_BOUND_COMMON( \ 1735 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \ 1736 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) 1737 1738 #define FBC_BOUND_A(TEST_NON_UTF8) \ 1739 FBC_BOUND_COMMON( \ 1740 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \ 1741 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) 1742 1743 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ 1744 FBC_BOUND_COMMON( \ 1745 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \ 1746 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) 1747 1748 #define FBC_NBOUND_A(TEST_NON_UTF8) \ 1749 FBC_BOUND_COMMON( \ 1750 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \ 1751 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) 1752 1753 #ifdef DEBUGGING 1754 static IV 1755 S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) { 1756 IV cp_out = Perl__invlist_search(invlist, cp_in); 1757 assert(cp_out >= 0); 1758 return cp_out; 1759 } 1760 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \ 1761 invmap[S_get_break_val_cp_checked(invlist, cp)] 1762 #else 1763 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \ 1764 invmap[_invlist_search(invlist, cp)] 1765 #endif 1766 1767 /* Takes a pointer to an inversion list, a pointer to its corresponding 1768 * inversion map, and a code point, and returns the code point's value 1769 * according to the two arrays. It assumes that all code points have a value. 1770 * This is used as the base macro for macros for particular properties */ 1771 #define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \ 1772 _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) 1773 1774 /* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead 1775 * of a code point, returning the value for the first code point in the string. 1776 * And it takes the particular macro name that finds the desired value given a 1777 * code point. Merely convert the UTF-8 to code point and call the cp macro */ 1778 #define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \ 1779 (__ASSERT_(pos < strend) \ 1780 /* Note assumes is valid UTF-8 */ \ 1781 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL)))) 1782 1783 /* Returns the GCB value for the input code point */ 1784 #define getGCB_VAL_CP(cp) \ 1785 _generic_GET_BREAK_VAL_CP( \ 1786 PL_GCB_invlist, \ 1787 _Perl_GCB_invmap, \ 1788 (cp)) 1789 1790 /* Returns the GCB value for the first code point in the UTF-8 encoded string 1791 * bounded by pos and strend */ 1792 #define getGCB_VAL_UTF8(pos, strend) \ 1793 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend) 1794 1795 /* Returns the LB value for the input code point */ 1796 #define getLB_VAL_CP(cp) \ 1797 _generic_GET_BREAK_VAL_CP( \ 1798 PL_LB_invlist, \ 1799 _Perl_LB_invmap, \ 1800 (cp)) 1801 1802 /* Returns the LB value for the first code point in the UTF-8 encoded string 1803 * bounded by pos and strend */ 1804 #define getLB_VAL_UTF8(pos, strend) \ 1805 _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend) 1806 1807 1808 /* Returns the SB value for the input code point */ 1809 #define getSB_VAL_CP(cp) \ 1810 _generic_GET_BREAK_VAL_CP( \ 1811 PL_SB_invlist, \ 1812 _Perl_SB_invmap, \ 1813 (cp)) 1814 1815 /* Returns the SB value for the first code point in the UTF-8 encoded string 1816 * bounded by pos and strend */ 1817 #define getSB_VAL_UTF8(pos, strend) \ 1818 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend) 1819 1820 /* Returns the WB value for the input code point */ 1821 #define getWB_VAL_CP(cp) \ 1822 _generic_GET_BREAK_VAL_CP( \ 1823 PL_WB_invlist, \ 1824 _Perl_WB_invmap, \ 1825 (cp)) 1826 1827 /* Returns the WB value for the first code point in the UTF-8 encoded string 1828 * bounded by pos and strend */ 1829 #define getWB_VAL_UTF8(pos, strend) \ 1830 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend) 1831 1832 /* We know what class REx starts with. Try to find this position... */ 1833 /* if reginfo->intuit, its a dryrun */ 1834 /* annoyingly all the vars in this routine have different names from their counterparts 1835 in regmatch. /grrr */ 1836 STATIC char * 1837 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, 1838 const char *strend, regmatch_info *reginfo) 1839 { 1840 dVAR; 1841 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0; 1842 char *pat_string; /* The pattern's exactish string */ 1843 char *pat_end; /* ptr to end char of pat_string */ 1844 re_fold_t folder; /* Function for computing non-utf8 folds */ 1845 const U8 *fold_array; /* array for folding ords < 256 */ 1846 STRLEN ln; 1847 STRLEN lnc; 1848 U8 c1; 1849 U8 c2; 1850 char *e; 1851 I32 tmp = 1; /* Scratch variable? */ 1852 const bool utf8_target = reginfo->is_utf8_target; 1853 UV utf8_fold_flags = 0; 1854 const bool is_utf8_pat = reginfo->is_utf8_pat; 1855 bool to_complement = FALSE; /* Invert the result? Taking the xor of this 1856 with a result inverts that result, as 0^1 = 1857 1 and 1^1 = 0 */ 1858 _char_class_number classnum; 1859 1860 RXi_GET_DECL(prog,progi); 1861 1862 PERL_ARGS_ASSERT_FIND_BYCLASS; 1863 1864 /* We know what class it must start with. */ 1865 switch (OP(c)) { 1866 case ANYOFL: 1867 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 1868 1869 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) { 1870 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required); 1871 } 1872 1873 /* FALLTHROUGH */ 1874 case ANYOFD: 1875 case ANYOF: 1876 if (utf8_target) { 1877 REXEC_FBC_UTF8_CLASS_SCAN( 1878 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target)); 1879 } 1880 else { 1881 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s, 0)); 1882 } 1883 break; 1884 1885 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ 1886 assert(! is_utf8_pat); 1887 /* FALLTHROUGH */ 1888 case EXACTFA: 1889 if (is_utf8_pat || utf8_target) { 1890 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; 1891 goto do_exactf_utf8; 1892 } 1893 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */ 1894 folder = foldEQ_latin1; /* /a, except the sharp s one which */ 1895 goto do_exactf_non_utf8; /* isn't dealt with by these */ 1896 1897 case EXACTF: /* This node only generated for non-utf8 patterns */ 1898 assert(! is_utf8_pat); 1899 if (utf8_target) { 1900 utf8_fold_flags = 0; 1901 goto do_exactf_utf8; 1902 } 1903 fold_array = PL_fold; 1904 folder = foldEQ; 1905 goto do_exactf_non_utf8; 1906 1907 case EXACTFL: 1908 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 1909 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) { 1910 utf8_fold_flags = FOLDEQ_LOCALE; 1911 goto do_exactf_utf8; 1912 } 1913 fold_array = PL_fold_locale; 1914 folder = foldEQ_locale; 1915 goto do_exactf_non_utf8; 1916 1917 case EXACTFU_SS: 1918 if (is_utf8_pat) { 1919 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED; 1920 } 1921 goto do_exactf_utf8; 1922 1923 case EXACTFLU8: 1924 if (! utf8_target) { /* All code points in this node require 1925 UTF-8 to express. */ 1926 break; 1927 } 1928 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED 1929 | FOLDEQ_S2_FOLDS_SANE; 1930 goto do_exactf_utf8; 1931 1932 case EXACTFU: 1933 if (is_utf8_pat || utf8_target) { 1934 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; 1935 goto do_exactf_utf8; 1936 } 1937 1938 /* Any 'ss' in the pattern should have been replaced by regcomp, 1939 * so we don't have to worry here about this single special case 1940 * in the Latin1 range */ 1941 fold_array = PL_fold_latin1; 1942 folder = foldEQ_latin1; 1943 1944 /* FALLTHROUGH */ 1945 1946 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there 1947 are no glitches with fold-length differences 1948 between the target string and pattern */ 1949 1950 /* The idea in the non-utf8 EXACTF* cases is to first find the 1951 * first character of the EXACTF* node and then, if necessary, 1952 * case-insensitively compare the full text of the node. c1 is the 1953 * first character. c2 is its fold. This logic will not work for 1954 * Unicode semantics and the german sharp ss, which hence should 1955 * not be compiled into a node that gets here. */ 1956 pat_string = STRING(c); 1957 ln = STR_LEN(c); /* length to match in octets/bytes */ 1958 1959 /* We know that we have to match at least 'ln' bytes (which is the 1960 * same as characters, since not utf8). If we have to match 3 1961 * characters, and there are only 2 availabe, we know without 1962 * trying that it will fail; so don't start a match past the 1963 * required minimum number from the far end */ 1964 e = HOP3c(strend, -((SSize_t)ln), s); 1965 1966 if (reginfo->intuit && e < s) { 1967 e = s; /* Due to minlen logic of intuit() */ 1968 } 1969 1970 c1 = *pat_string; 1971 c2 = fold_array[c1]; 1972 if (c1 == c2) { /* If char and fold are the same */ 1973 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1); 1974 } 1975 else { 1976 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2); 1977 } 1978 break; 1979 1980 do_exactf_utf8: 1981 { 1982 unsigned expansion; 1983 1984 /* If one of the operands is in utf8, we can't use the simpler folding 1985 * above, due to the fact that many different characters can have the 1986 * same fold, or portion of a fold, or different- length fold */ 1987 pat_string = STRING(c); 1988 ln = STR_LEN(c); /* length to match in octets/bytes */ 1989 pat_end = pat_string + ln; 1990 lnc = is_utf8_pat /* length to match in characters */ 1991 ? utf8_length((U8 *) pat_string, (U8 *) pat_end) 1992 : ln; 1993 1994 /* We have 'lnc' characters to match in the pattern, but because of 1995 * multi-character folding, each character in the target can match 1996 * up to 3 characters (Unicode guarantees it will never exceed 1997 * this) if it is utf8-encoded; and up to 2 if not (based on the 1998 * fact that the Latin 1 folds are already determined, and the 1999 * only multi-char fold in that range is the sharp-s folding to 2000 * 'ss'. Thus, a pattern character can match as little as 1/3 of a 2001 * string character. Adjust lnc accordingly, rounding up, so that 2002 * if we need to match at least 4+1/3 chars, that really is 5. */ 2003 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2; 2004 lnc = (lnc + expansion - 1) / expansion; 2005 2006 /* As in the non-UTF8 case, if we have to match 3 characters, and 2007 * only 2 are left, it's guaranteed to fail, so don't start a 2008 * match that would require us to go beyond the end of the string 2009 */ 2010 e = HOP3c(strend, -((SSize_t)lnc), s); 2011 2012 if (reginfo->intuit && e < s) { 2013 e = s; /* Due to minlen logic of intuit() */ 2014 } 2015 2016 /* XXX Note that we could recalculate e to stop the loop earlier, 2017 * as the worst case expansion above will rarely be met, and as we 2018 * go along we would usually find that e moves further to the left. 2019 * This would happen only after we reached the point in the loop 2020 * where if there were no expansion we should fail. Unclear if 2021 * worth the expense */ 2022 2023 while (s <= e) { 2024 char *my_strend= (char *)strend; 2025 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target, 2026 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags) 2027 && (reginfo->intuit || regtry(reginfo, &s)) ) 2028 { 2029 goto got_it; 2030 } 2031 s += (utf8_target) ? UTF8SKIP(s) : 1; 2032 } 2033 break; 2034 } 2035 2036 case BOUNDL: 2037 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 2038 if (FLAGS(c) != TRADITIONAL_BOUND) { 2039 if (! IN_UTF8_CTYPE_LOCALE) { 2040 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 2041 B_ON_NON_UTF8_LOCALE_IS_WRONG); 2042 } 2043 goto do_boundu; 2044 } 2045 2046 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8); 2047 break; 2048 2049 case NBOUNDL: 2050 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 2051 if (FLAGS(c) != TRADITIONAL_BOUND) { 2052 if (! IN_UTF8_CTYPE_LOCALE) { 2053 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 2054 B_ON_NON_UTF8_LOCALE_IS_WRONG); 2055 } 2056 goto do_nboundu; 2057 } 2058 2059 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8); 2060 break; 2061 2062 case BOUND: /* regcomp.c makes sure that this only has the traditional \b 2063 meaning */ 2064 assert(FLAGS(c) == TRADITIONAL_BOUND); 2065 2066 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8); 2067 break; 2068 2069 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b 2070 meaning */ 2071 assert(FLAGS(c) == TRADITIONAL_BOUND); 2072 2073 FBC_BOUND_A(isWORDCHAR_A); 2074 break; 2075 2076 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b 2077 meaning */ 2078 assert(FLAGS(c) == TRADITIONAL_BOUND); 2079 2080 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8); 2081 break; 2082 2083 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b 2084 meaning */ 2085 assert(FLAGS(c) == TRADITIONAL_BOUND); 2086 2087 FBC_NBOUND_A(isWORDCHAR_A); 2088 break; 2089 2090 case NBOUNDU: 2091 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) { 2092 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8); 2093 break; 2094 } 2095 2096 do_nboundu: 2097 2098 to_complement = 1; 2099 /* FALLTHROUGH */ 2100 2101 case BOUNDU: 2102 do_boundu: 2103 switch((bound_type) FLAGS(c)) { 2104 case TRADITIONAL_BOUND: 2105 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8); 2106 break; 2107 case GCB_BOUND: 2108 if (s == reginfo->strbeg) { 2109 if (reginfo->intuit || regtry(reginfo, &s)) 2110 { 2111 goto got_it; 2112 } 2113 2114 /* Didn't match. Try at the next position (if there is one) */ 2115 s += (utf8_target) ? UTF8SKIP(s) : 1; 2116 if (UNLIKELY(s >= reginfo->strend)) { 2117 break; 2118 } 2119 } 2120 2121 if (utf8_target) { 2122 GCB_enum before = getGCB_VAL_UTF8( 2123 reghop3((U8*)s, -1, 2124 (U8*)(reginfo->strbeg)), 2125 (U8*) reginfo->strend); 2126 while (s < strend) { 2127 GCB_enum after = getGCB_VAL_UTF8((U8*) s, 2128 (U8*) reginfo->strend); 2129 if ( (to_complement ^ isGCB(before, after)) 2130 && (reginfo->intuit || regtry(reginfo, &s))) 2131 { 2132 goto got_it; 2133 } 2134 before = after; 2135 s += UTF8SKIP(s); 2136 } 2137 } 2138 else { /* Not utf8. Everything is a GCB except between CR and 2139 LF */ 2140 while (s < strend) { 2141 if ((to_complement ^ ( UCHARAT(s - 1) != '\r' 2142 || UCHARAT(s) != '\n')) 2143 && (reginfo->intuit || regtry(reginfo, &s))) 2144 { 2145 goto got_it; 2146 } 2147 s++; 2148 } 2149 } 2150 2151 /* And, since this is a bound, it can match after the final 2152 * character in the string */ 2153 if ((reginfo->intuit || regtry(reginfo, &s))) { 2154 goto got_it; 2155 } 2156 break; 2157 2158 case LB_BOUND: 2159 if (s == reginfo->strbeg) { 2160 if (reginfo->intuit || regtry(reginfo, &s)) { 2161 goto got_it; 2162 } 2163 s += (utf8_target) ? UTF8SKIP(s) : 1; 2164 if (UNLIKELY(s >= reginfo->strend)) { 2165 break; 2166 } 2167 } 2168 2169 if (utf8_target) { 2170 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s, 2171 -1, 2172 (U8*)(reginfo->strbeg)), 2173 (U8*) reginfo->strend); 2174 while (s < strend) { 2175 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend); 2176 if (to_complement ^ isLB(before, 2177 after, 2178 (U8*) reginfo->strbeg, 2179 (U8*) s, 2180 (U8*) reginfo->strend, 2181 utf8_target) 2182 && (reginfo->intuit || regtry(reginfo, &s))) 2183 { 2184 goto got_it; 2185 } 2186 before = after; 2187 s += UTF8SKIP(s); 2188 } 2189 } 2190 else { /* Not utf8. */ 2191 LB_enum before = getLB_VAL_CP((U8) *(s -1)); 2192 while (s < strend) { 2193 LB_enum after = getLB_VAL_CP((U8) *s); 2194 if (to_complement ^ isLB(before, 2195 after, 2196 (U8*) reginfo->strbeg, 2197 (U8*) s, 2198 (U8*) reginfo->strend, 2199 utf8_target) 2200 && (reginfo->intuit || regtry(reginfo, &s))) 2201 { 2202 goto got_it; 2203 } 2204 before = after; 2205 s++; 2206 } 2207 } 2208 2209 if (reginfo->intuit || regtry(reginfo, &s)) { 2210 goto got_it; 2211 } 2212 2213 break; 2214 2215 case SB_BOUND: 2216 if (s == reginfo->strbeg) { 2217 if (reginfo->intuit || regtry(reginfo, &s)) { 2218 goto got_it; 2219 } 2220 s += (utf8_target) ? UTF8SKIP(s) : 1; 2221 if (UNLIKELY(s >= reginfo->strend)) { 2222 break; 2223 } 2224 } 2225 2226 if (utf8_target) { 2227 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s, 2228 -1, 2229 (U8*)(reginfo->strbeg)), 2230 (U8*) reginfo->strend); 2231 while (s < strend) { 2232 SB_enum after = getSB_VAL_UTF8((U8*) s, 2233 (U8*) reginfo->strend); 2234 if ((to_complement ^ isSB(before, 2235 after, 2236 (U8*) reginfo->strbeg, 2237 (U8*) s, 2238 (U8*) reginfo->strend, 2239 utf8_target)) 2240 && (reginfo->intuit || regtry(reginfo, &s))) 2241 { 2242 goto got_it; 2243 } 2244 before = after; 2245 s += UTF8SKIP(s); 2246 } 2247 } 2248 else { /* Not utf8. */ 2249 SB_enum before = getSB_VAL_CP((U8) *(s -1)); 2250 while (s < strend) { 2251 SB_enum after = getSB_VAL_CP((U8) *s); 2252 if ((to_complement ^ isSB(before, 2253 after, 2254 (U8*) reginfo->strbeg, 2255 (U8*) s, 2256 (U8*) reginfo->strend, 2257 utf8_target)) 2258 && (reginfo->intuit || regtry(reginfo, &s))) 2259 { 2260 goto got_it; 2261 } 2262 before = after; 2263 s++; 2264 } 2265 } 2266 2267 /* Here are at the final position in the target string. The SB 2268 * value is always true here, so matches, depending on other 2269 * constraints */ 2270 if (reginfo->intuit || regtry(reginfo, &s)) { 2271 goto got_it; 2272 } 2273 2274 break; 2275 2276 case WB_BOUND: 2277 if (s == reginfo->strbeg) { 2278 if (reginfo->intuit || regtry(reginfo, &s)) { 2279 goto got_it; 2280 } 2281 s += (utf8_target) ? UTF8SKIP(s) : 1; 2282 if (UNLIKELY(s >= reginfo->strend)) { 2283 break; 2284 } 2285 } 2286 2287 if (utf8_target) { 2288 /* We are at a boundary between char_sub_0 and char_sub_1. 2289 * We also keep track of the value for char_sub_-1 as we 2290 * loop through the line. Context may be needed to make a 2291 * determination, and if so, this can save having to 2292 * recalculate it */ 2293 WB_enum previous = WB_UNKNOWN; 2294 WB_enum before = getWB_VAL_UTF8( 2295 reghop3((U8*)s, 2296 -1, 2297 (U8*)(reginfo->strbeg)), 2298 (U8*) reginfo->strend); 2299 while (s < strend) { 2300 WB_enum after = getWB_VAL_UTF8((U8*) s, 2301 (U8*) reginfo->strend); 2302 if ((to_complement ^ isWB(previous, 2303 before, 2304 after, 2305 (U8*) reginfo->strbeg, 2306 (U8*) s, 2307 (U8*) reginfo->strend, 2308 utf8_target)) 2309 && (reginfo->intuit || regtry(reginfo, &s))) 2310 { 2311 goto got_it; 2312 } 2313 previous = before; 2314 before = after; 2315 s += UTF8SKIP(s); 2316 } 2317 } 2318 else { /* Not utf8. */ 2319 WB_enum previous = WB_UNKNOWN; 2320 WB_enum before = getWB_VAL_CP((U8) *(s -1)); 2321 while (s < strend) { 2322 WB_enum after = getWB_VAL_CP((U8) *s); 2323 if ((to_complement ^ isWB(previous, 2324 before, 2325 after, 2326 (U8*) reginfo->strbeg, 2327 (U8*) s, 2328 (U8*) reginfo->strend, 2329 utf8_target)) 2330 && (reginfo->intuit || regtry(reginfo, &s))) 2331 { 2332 goto got_it; 2333 } 2334 previous = before; 2335 before = after; 2336 s++; 2337 } 2338 } 2339 2340 if (reginfo->intuit || regtry(reginfo, &s)) { 2341 goto got_it; 2342 } 2343 } 2344 break; 2345 2346 case LNBREAK: 2347 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend), 2348 is_LNBREAK_latin1_safe(s, strend) 2349 ); 2350 break; 2351 2352 /* The argument to all the POSIX node types is the class number to pass to 2353 * _generic_isCC() to build a mask for searching in PL_charclass[] */ 2354 2355 case NPOSIXL: 2356 to_complement = 1; 2357 /* FALLTHROUGH */ 2358 2359 case POSIXL: 2360 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 2361 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)), 2362 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s))); 2363 break; 2364 2365 case NPOSIXD: 2366 to_complement = 1; 2367 /* FALLTHROUGH */ 2368 2369 case POSIXD: 2370 if (utf8_target) { 2371 goto posix_utf8; 2372 } 2373 goto posixa; 2374 2375 case NPOSIXA: 2376 if (utf8_target) { 2377 /* The complement of something that matches only ASCII matches all 2378 * non-ASCII, plus everything in ASCII that isn't in the class. */ 2379 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s) 2380 || ! _generic_isCC_A(*s, FLAGS(c))); 2381 break; 2382 } 2383 2384 to_complement = 1; 2385 /* FALLTHROUGH */ 2386 2387 case POSIXA: 2388 posixa: 2389 /* Don't need to worry about utf8, as it can match only a single 2390 * byte invariant character. */ 2391 REXEC_FBC_CLASS_SCAN( 2392 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c)))); 2393 break; 2394 2395 case NPOSIXU: 2396 to_complement = 1; 2397 /* FALLTHROUGH */ 2398 2399 case POSIXU: 2400 if (! utf8_target) { 2401 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s, 2402 FLAGS(c)))); 2403 } 2404 else { 2405 2406 posix_utf8: 2407 classnum = (_char_class_number) FLAGS(c); 2408 if (classnum < _FIRST_NON_SWASH_CC) { 2409 while (s < strend) { 2410 2411 /* We avoid loading in the swash as long as possible, but 2412 * should we have to, we jump to a separate loop. This 2413 * extra 'if' statement is what keeps this code from being 2414 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */ 2415 if (UTF8_IS_ABOVE_LATIN1(*s)) { 2416 goto found_above_latin1; 2417 } 2418 if ((UTF8_IS_INVARIANT(*s) 2419 && to_complement ^ cBOOL(_generic_isCC((U8) *s, 2420 classnum))) 2421 || (UTF8_IS_DOWNGRADEABLE_START(*s) 2422 && to_complement ^ cBOOL( 2423 _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*s, 2424 *(s + 1)), 2425 classnum)))) 2426 { 2427 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) 2428 goto got_it; 2429 else { 2430 tmp = doevery; 2431 } 2432 } 2433 else { 2434 tmp = 1; 2435 } 2436 s += UTF8SKIP(s); 2437 } 2438 } 2439 else switch (classnum) { /* These classes are implemented as 2440 macros */ 2441 case _CC_ENUM_SPACE: 2442 REXEC_FBC_UTF8_CLASS_SCAN( 2443 to_complement ^ cBOOL(isSPACE_utf8(s))); 2444 break; 2445 2446 case _CC_ENUM_BLANK: 2447 REXEC_FBC_UTF8_CLASS_SCAN( 2448 to_complement ^ cBOOL(isBLANK_utf8(s))); 2449 break; 2450 2451 case _CC_ENUM_XDIGIT: 2452 REXEC_FBC_UTF8_CLASS_SCAN( 2453 to_complement ^ cBOOL(isXDIGIT_utf8(s))); 2454 break; 2455 2456 case _CC_ENUM_VERTSPACE: 2457 REXEC_FBC_UTF8_CLASS_SCAN( 2458 to_complement ^ cBOOL(isVERTWS_utf8(s))); 2459 break; 2460 2461 case _CC_ENUM_CNTRL: 2462 REXEC_FBC_UTF8_CLASS_SCAN( 2463 to_complement ^ cBOOL(isCNTRL_utf8(s))); 2464 break; 2465 2466 default: 2467 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum); 2468 NOT_REACHED; /* NOTREACHED */ 2469 } 2470 } 2471 break; 2472 2473 found_above_latin1: /* Here we have to load a swash to get the result 2474 for the current code point */ 2475 if (! PL_utf8_swash_ptrs[classnum]) { 2476 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 2477 PL_utf8_swash_ptrs[classnum] = 2478 _core_swash_init("utf8", 2479 "", 2480 &PL_sv_undef, 1, 0, 2481 PL_XPosix_ptrs[classnum], &flags); 2482 } 2483 2484 /* This is a copy of the loop above for swash classes, though using the 2485 * FBC macro instead of being expanded out. Since we've loaded the 2486 * swash, we don't have to check for that each time through the loop */ 2487 REXEC_FBC_UTF8_CLASS_SCAN( 2488 to_complement ^ cBOOL(_generic_utf8( 2489 classnum, 2490 s, 2491 swash_fetch(PL_utf8_swash_ptrs[classnum], 2492 (U8 *) s, TRUE)))); 2493 break; 2494 2495 case AHOCORASICKC: 2496 case AHOCORASICK: 2497 { 2498 DECL_TRIE_TYPE(c); 2499 /* what trie are we using right now */ 2500 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ]; 2501 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ]; 2502 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]); 2503 2504 const char *last_start = strend - trie->minlen; 2505 #ifdef DEBUGGING 2506 const char *real_start = s; 2507 #endif 2508 STRLEN maxlen = trie->maxlen; 2509 SV *sv_points; 2510 U8 **points; /* map of where we were in the input string 2511 when reading a given char. For ASCII this 2512 is unnecessary overhead as the relationship 2513 is always 1:1, but for Unicode, especially 2514 case folded Unicode this is not true. */ 2515 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; 2516 U8 *bitmap=NULL; 2517 2518 2519 GET_RE_DEBUG_FLAGS_DECL; 2520 2521 /* We can't just allocate points here. We need to wrap it in 2522 * an SV so it gets freed properly if there is a croak while 2523 * running the match */ 2524 ENTER; 2525 SAVETMPS; 2526 sv_points=newSV(maxlen * sizeof(U8 *)); 2527 SvCUR_set(sv_points, 2528 maxlen * sizeof(U8 *)); 2529 SvPOK_on(sv_points); 2530 sv_2mortal(sv_points); 2531 points=(U8**)SvPV_nolen(sv_points ); 2532 if ( trie_type != trie_utf8_fold 2533 && (trie->bitmap || OP(c)==AHOCORASICKC) ) 2534 { 2535 if (trie->bitmap) 2536 bitmap=(U8*)trie->bitmap; 2537 else 2538 bitmap=(U8*)ANYOF_BITMAP(c); 2539 } 2540 /* this is the Aho-Corasick algorithm modified a touch 2541 to include special handling for long "unknown char" sequences. 2542 The basic idea being that we use AC as long as we are dealing 2543 with a possible matching char, when we encounter an unknown char 2544 (and we have not encountered an accepting state) we scan forward 2545 until we find a legal starting char. 2546 AC matching is basically that of trie matching, except that when 2547 we encounter a failing transition, we fall back to the current 2548 states "fail state", and try the current char again, a process 2549 we repeat until we reach the root state, state 1, or a legal 2550 transition. If we fail on the root state then we can either 2551 terminate if we have reached an accepting state previously, or 2552 restart the entire process from the beginning if we have not. 2553 2554 */ 2555 while (s <= last_start) { 2556 const U32 uniflags = UTF8_ALLOW_DEFAULT; 2557 U8 *uc = (U8*)s; 2558 U16 charid = 0; 2559 U32 base = 1; 2560 U32 state = 1; 2561 UV uvc = 0; 2562 STRLEN len = 0; 2563 STRLEN foldlen = 0; 2564 U8 *uscan = (U8*)NULL; 2565 U8 *leftmost = NULL; 2566 #ifdef DEBUGGING 2567 U32 accepted_word= 0; 2568 #endif 2569 U32 pointpos = 0; 2570 2571 while ( state && uc <= (U8*)strend ) { 2572 int failed=0; 2573 U32 word = aho->states[ state ].wordnum; 2574 2575 if( state==1 ) { 2576 if ( bitmap ) { 2577 DEBUG_TRIE_EXECUTE_r( 2578 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { 2579 dump_exec_pos( (char *)uc, c, strend, real_start, 2580 (char *)uc, utf8_target, 0 ); 2581 Perl_re_printf( aTHX_ 2582 " Scanning for legal start char...\n"); 2583 } 2584 ); 2585 if (utf8_target) { 2586 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { 2587 uc += UTF8SKIP(uc); 2588 } 2589 } else { 2590 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { 2591 uc++; 2592 } 2593 } 2594 s= (char *)uc; 2595 } 2596 if (uc >(U8*)last_start) break; 2597 } 2598 2599 if ( word ) { 2600 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ]; 2601 if (!leftmost || lpos < leftmost) { 2602 DEBUG_r(accepted_word=word); 2603 leftmost= lpos; 2604 } 2605 if (base==0) break; 2606 2607 } 2608 points[pointpos++ % maxlen]= uc; 2609 if (foldlen || uc < (U8*)strend) { 2610 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, 2611 (U8 *) strend, uscan, len, uvc, 2612 charid, foldlen, foldbuf, 2613 uniflags); 2614 DEBUG_TRIE_EXECUTE_r({ 2615 dump_exec_pos( (char *)uc, c, strend, 2616 real_start, s, utf8_target, 0); 2617 Perl_re_printf( aTHX_ 2618 " Charid:%3u CP:%4"UVxf" ", 2619 charid, uvc); 2620 }); 2621 } 2622 else { 2623 len = 0; 2624 charid = 0; 2625 } 2626 2627 2628 do { 2629 #ifdef DEBUGGING 2630 word = aho->states[ state ].wordnum; 2631 #endif 2632 base = aho->states[ state ].trans.base; 2633 2634 DEBUG_TRIE_EXECUTE_r({ 2635 if (failed) 2636 dump_exec_pos( (char *)uc, c, strend, real_start, 2637 s, utf8_target, 0 ); 2638 Perl_re_printf( aTHX_ 2639 "%sState: %4"UVxf", word=%"UVxf, 2640 failed ? " Fail transition to " : "", 2641 (UV)state, (UV)word); 2642 }); 2643 if ( base ) { 2644 U32 tmp; 2645 I32 offset; 2646 if (charid && 2647 ( ((offset = base + charid 2648 - 1 - trie->uniquecharcount)) >= 0) 2649 && ((U32)offset < trie->lasttrans) 2650 && trie->trans[offset].check == state 2651 && (tmp=trie->trans[offset].next)) 2652 { 2653 DEBUG_TRIE_EXECUTE_r( 2654 Perl_re_printf( aTHX_ " - legal\n")); 2655 state = tmp; 2656 break; 2657 } 2658 else { 2659 DEBUG_TRIE_EXECUTE_r( 2660 Perl_re_printf( aTHX_ " - fail\n")); 2661 failed = 1; 2662 state = aho->fail[state]; 2663 } 2664 } 2665 else { 2666 /* we must be accepting here */ 2667 DEBUG_TRIE_EXECUTE_r( 2668 Perl_re_printf( aTHX_ " - accepting\n")); 2669 failed = 1; 2670 break; 2671 } 2672 } while(state); 2673 uc += len; 2674 if (failed) { 2675 if (leftmost) 2676 break; 2677 if (!state) state = 1; 2678 } 2679 } 2680 if ( aho->states[ state ].wordnum ) { 2681 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ]; 2682 if (!leftmost || lpos < leftmost) { 2683 DEBUG_r(accepted_word=aho->states[ state ].wordnum); 2684 leftmost = lpos; 2685 } 2686 } 2687 if (leftmost) { 2688 s = (char*)leftmost; 2689 DEBUG_TRIE_EXECUTE_r({ 2690 Perl_re_printf( aTHX_ "Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n", 2691 (UV)accepted_word, (IV)(s - real_start) 2692 ); 2693 }); 2694 if (reginfo->intuit || regtry(reginfo, &s)) { 2695 FREETMPS; 2696 LEAVE; 2697 goto got_it; 2698 } 2699 s = HOPc(s,1); 2700 DEBUG_TRIE_EXECUTE_r({ 2701 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n"); 2702 }); 2703 } else { 2704 DEBUG_TRIE_EXECUTE_r( 2705 Perl_re_printf( aTHX_ "No match.\n")); 2706 break; 2707 } 2708 } 2709 FREETMPS; 2710 LEAVE; 2711 } 2712 break; 2713 default: 2714 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c)); 2715 } 2716 return 0; 2717 got_it: 2718 return s; 2719 } 2720 2721 /* set RX_SAVED_COPY, RX_SUBBEG etc. 2722 * flags have same meanings as with regexec_flags() */ 2723 2724 static void 2725 S_reg_set_capture_string(pTHX_ REGEXP * const rx, 2726 char *strbeg, 2727 char *strend, 2728 SV *sv, 2729 U32 flags, 2730 bool utf8_target) 2731 { 2732 struct regexp *const prog = ReANY(rx); 2733 2734 if (flags & REXEC_COPY_STR) { 2735 #ifdef PERL_ANY_COW 2736 if (SvCANCOW(sv)) { 2737 DEBUG_C(Perl_re_printf( aTHX_ 2738 "Copy on write: regexp capture, type %d\n", 2739 (int) SvTYPE(sv))); 2740 /* Create a new COW SV to share the match string and store 2741 * in saved_copy, unless the current COW SV in saved_copy 2742 * is valid and suitable for our purpose */ 2743 if (( prog->saved_copy 2744 && SvIsCOW(prog->saved_copy) 2745 && SvPOKp(prog->saved_copy) 2746 && SvIsCOW(sv) 2747 && SvPOKp(sv) 2748 && SvPVX(sv) == SvPVX(prog->saved_copy))) 2749 { 2750 /* just reuse saved_copy SV */ 2751 if (RXp_MATCH_COPIED(prog)) { 2752 Safefree(prog->subbeg); 2753 RXp_MATCH_COPIED_off(prog); 2754 } 2755 } 2756 else { 2757 /* create new COW SV to share string */ 2758 RX_MATCH_COPY_FREE(rx); 2759 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv); 2760 } 2761 prog->subbeg = (char *)SvPVX_const(prog->saved_copy); 2762 assert (SvPOKp(prog->saved_copy)); 2763 prog->sublen = strend - strbeg; 2764 prog->suboffset = 0; 2765 prog->subcoffset = 0; 2766 } else 2767 #endif 2768 { 2769 SSize_t min = 0; 2770 SSize_t max = strend - strbeg; 2771 SSize_t sublen; 2772 2773 if ( (flags & REXEC_COPY_SKIP_POST) 2774 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ 2775 && !(PL_sawampersand & SAWAMPERSAND_RIGHT) 2776 ) { /* don't copy $' part of string */ 2777 U32 n = 0; 2778 max = -1; 2779 /* calculate the right-most part of the string covered 2780 * by a capture. Due to lookahead, this may be to 2781 * the right of $&, so we have to scan all captures */ 2782 while (n <= prog->lastparen) { 2783 if (prog->offs[n].end > max) 2784 max = prog->offs[n].end; 2785 n++; 2786 } 2787 if (max == -1) 2788 max = (PL_sawampersand & SAWAMPERSAND_LEFT) 2789 ? prog->offs[0].start 2790 : 0; 2791 assert(max >= 0 && max <= strend - strbeg); 2792 } 2793 2794 if ( (flags & REXEC_COPY_SKIP_PRE) 2795 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ 2796 && !(PL_sawampersand & SAWAMPERSAND_LEFT) 2797 ) { /* don't copy $` part of string */ 2798 U32 n = 0; 2799 min = max; 2800 /* calculate the left-most part of the string covered 2801 * by a capture. Due to lookbehind, this may be to 2802 * the left of $&, so we have to scan all captures */ 2803 while (min && n <= prog->lastparen) { 2804 if ( prog->offs[n].start != -1 2805 && prog->offs[n].start < min) 2806 { 2807 min = prog->offs[n].start; 2808 } 2809 n++; 2810 } 2811 if ((PL_sawampersand & SAWAMPERSAND_RIGHT) 2812 && min > prog->offs[0].end 2813 ) 2814 min = prog->offs[0].end; 2815 2816 } 2817 2818 assert(min >= 0 && min <= max && min <= strend - strbeg); 2819 sublen = max - min; 2820 2821 if (RX_MATCH_COPIED(rx)) { 2822 if (sublen > prog->sublen) 2823 prog->subbeg = 2824 (char*)saferealloc(prog->subbeg, sublen+1); 2825 } 2826 else 2827 prog->subbeg = (char*)safemalloc(sublen+1); 2828 Copy(strbeg + min, prog->subbeg, sublen, char); 2829 prog->subbeg[sublen] = '\0'; 2830 prog->suboffset = min; 2831 prog->sublen = sublen; 2832 RX_MATCH_COPIED_on(rx); 2833 } 2834 prog->subcoffset = prog->suboffset; 2835 if (prog->suboffset && utf8_target) { 2836 /* Convert byte offset to chars. 2837 * XXX ideally should only compute this if @-/@+ 2838 * has been seen, a la PL_sawampersand ??? */ 2839 2840 /* If there's a direct correspondence between the 2841 * string which we're matching and the original SV, 2842 * then we can use the utf8 len cache associated with 2843 * the SV. In particular, it means that under //g, 2844 * sv_pos_b2u() will use the previously cached 2845 * position to speed up working out the new length of 2846 * subcoffset, rather than counting from the start of 2847 * the string each time. This stops 2848 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g; 2849 * from going quadratic */ 2850 if (SvPOKp(sv) && SvPVX(sv) == strbeg) 2851 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset, 2852 SV_GMAGIC|SV_CONST_RETURN); 2853 else 2854 prog->subcoffset = utf8_length((U8*)strbeg, 2855 (U8*)(strbeg+prog->suboffset)); 2856 } 2857 } 2858 else { 2859 RX_MATCH_COPY_FREE(rx); 2860 prog->subbeg = strbeg; 2861 prog->suboffset = 0; 2862 prog->subcoffset = 0; 2863 prog->sublen = strend - strbeg; 2864 } 2865 } 2866 2867 2868 2869 2870 /* 2871 - regexec_flags - match a regexp against a string 2872 */ 2873 I32 2874 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend, 2875 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags) 2876 /* stringarg: the point in the string at which to begin matching */ 2877 /* strend: pointer to null at end of string */ 2878 /* strbeg: real beginning of string */ 2879 /* minend: end of match must be >= minend bytes after stringarg. */ 2880 /* sv: SV being matched: only used for utf8 flag, pos() etc; string 2881 * itself is accessed via the pointers above */ 2882 /* data: May be used for some additional optimizations. 2883 Currently unused. */ 2884 /* flags: For optimizations. See REXEC_* in regexp.h */ 2885 2886 { 2887 struct regexp *const prog = ReANY(rx); 2888 char *s; 2889 regnode *c; 2890 char *startpos; 2891 SSize_t minlen; /* must match at least this many chars */ 2892 SSize_t dontbother = 0; /* how many characters not to try at end */ 2893 const bool utf8_target = cBOOL(DO_UTF8(sv)); 2894 I32 multiline; 2895 RXi_GET_DECL(prog,progi); 2896 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */ 2897 regmatch_info *const reginfo = ®info_buf; 2898 regexp_paren_pair *swap = NULL; 2899 I32 oldsave; 2900 GET_RE_DEBUG_FLAGS_DECL; 2901 2902 PERL_ARGS_ASSERT_REGEXEC_FLAGS; 2903 PERL_UNUSED_ARG(data); 2904 2905 /* Be paranoid... */ 2906 if (prog == NULL) { 2907 Perl_croak(aTHX_ "NULL regexp parameter"); 2908 } 2909 2910 DEBUG_EXECUTE_r( 2911 debug_start_match(rx, utf8_target, stringarg, strend, 2912 "Matching"); 2913 ); 2914 2915 startpos = stringarg; 2916 2917 /* set these early as they may be used by the HOP macros below */ 2918 reginfo->strbeg = strbeg; 2919 reginfo->strend = strend; 2920 reginfo->is_utf8_target = cBOOL(utf8_target); 2921 2922 if (prog->intflags & PREGf_GPOS_SEEN) { 2923 MAGIC *mg; 2924 2925 /* set reginfo->ganch, the position where \G can match */ 2926 2927 reginfo->ganch = 2928 (flags & REXEC_IGNOREPOS) 2929 ? stringarg /* use start pos rather than pos() */ 2930 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0) 2931 /* Defined pos(): */ 2932 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg) 2933 : strbeg; /* pos() not defined; use start of string */ 2934 2935 DEBUG_GPOS_r(Perl_re_printf( aTHX_ 2936 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg))); 2937 2938 /* in the presence of \G, we may need to start looking earlier in 2939 * the string than the suggested start point of stringarg: 2940 * if prog->gofs is set, then that's a known, fixed minimum 2941 * offset, such as 2942 * /..\G/: gofs = 2 2943 * /ab|c\G/: gofs = 1 2944 * or if the minimum offset isn't known, then we have to go back 2945 * to the start of the string, e.g. /w+\G/ 2946 */ 2947 2948 if (prog->intflags & PREGf_ANCH_GPOS) { 2949 if (prog->gofs) { 2950 startpos = HOPBACKc(reginfo->ganch, prog->gofs); 2951 if (!startpos || 2952 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg)) 2953 { 2954 DEBUG_r(Perl_re_printf( aTHX_ 2955 "fail: ganch-gofs before earliest possible start\n")); 2956 return 0; 2957 } 2958 } 2959 else 2960 startpos = reginfo->ganch; 2961 } 2962 else if (prog->gofs) { 2963 startpos = HOPBACKc(startpos, prog->gofs); 2964 if (!startpos) 2965 startpos = strbeg; 2966 } 2967 else if (prog->intflags & PREGf_GPOS_FLOAT) 2968 startpos = strbeg; 2969 } 2970 2971 minlen = prog->minlen; 2972 if ((startpos + minlen) > strend || startpos < strbeg) { 2973 DEBUG_r(Perl_re_printf( aTHX_ 2974 "Regex match can't succeed, so not even tried\n")); 2975 return 0; 2976 } 2977 2978 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave), 2979 * which will call destuctors to reset PL_regmatch_state, free higher 2980 * PL_regmatch_slabs, and clean up regmatch_info_aux and 2981 * regmatch_info_aux_eval */ 2982 2983 oldsave = PL_savestack_ix; 2984 2985 s = startpos; 2986 2987 if ((prog->extflags & RXf_USE_INTUIT) 2988 && !(flags & REXEC_CHECKED)) 2989 { 2990 s = re_intuit_start(rx, sv, strbeg, startpos, strend, 2991 flags, NULL); 2992 if (!s) 2993 return 0; 2994 2995 if (prog->extflags & RXf_CHECK_ALL) { 2996 /* we can match based purely on the result of INTUIT. 2997 * Set up captures etc just for $& and $-[0] 2998 * (an intuit-only match wont have $1,$2,..) */ 2999 assert(!prog->nparens); 3000 3001 /* s/// doesn't like it if $& is earlier than where we asked it to 3002 * start searching (which can happen on something like /.\G/) */ 3003 if ( (flags & REXEC_FAIL_ON_UNDERFLOW) 3004 && (s < stringarg)) 3005 { 3006 /* this should only be possible under \G */ 3007 assert(prog->intflags & PREGf_GPOS_SEEN); 3008 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ 3009 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); 3010 goto phooey; 3011 } 3012 3013 /* match via INTUIT shouldn't have any captures. 3014 * Let @-, @+, $^N know */ 3015 prog->lastparen = prog->lastcloseparen = 0; 3016 RX_MATCH_UTF8_set(rx, utf8_target); 3017 prog->offs[0].start = s - strbeg; 3018 prog->offs[0].end = utf8_target 3019 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg 3020 : s - strbeg + prog->minlenret; 3021 if ( !(flags & REXEC_NOT_FIRST) ) 3022 S_reg_set_capture_string(aTHX_ rx, 3023 strbeg, strend, 3024 sv, flags, utf8_target); 3025 3026 return 1; 3027 } 3028 } 3029 3030 multiline = prog->extflags & RXf_PMf_MULTILINE; 3031 3032 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { 3033 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ 3034 "String too short [regexec_flags]...\n")); 3035 goto phooey; 3036 } 3037 3038 /* Check validity of program. */ 3039 if (UCHARAT(progi->program) != REG_MAGIC) { 3040 Perl_croak(aTHX_ "corrupted regexp program"); 3041 } 3042 3043 RX_MATCH_TAINTED_off(rx); 3044 RX_MATCH_UTF8_set(rx, utf8_target); 3045 3046 reginfo->prog = rx; /* Yes, sorry that this is confusing. */ 3047 reginfo->intuit = 0; 3048 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx)); 3049 reginfo->warned = FALSE; 3050 reginfo->sv = sv; 3051 reginfo->poscache_maxiter = 0; /* not yet started a countdown */ 3052 /* see how far we have to get to not match where we matched before */ 3053 reginfo->till = stringarg + minend; 3054 3055 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) { 3056 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after 3057 S_cleanup_regmatch_info_aux has executed (registered by 3058 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies 3059 magic belonging to this SV. 3060 Not newSVsv, either, as it does not COW. 3061 */ 3062 reginfo->sv = newSV(0); 3063 SvSetSV_nosteal(reginfo->sv, sv); 3064 SAVEFREESV(reginfo->sv); 3065 } 3066 3067 /* reserve next 2 or 3 slots in PL_regmatch_state: 3068 * slot N+0: may currently be in use: skip it 3069 * slot N+1: use for regmatch_info_aux struct 3070 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s 3071 * slot N+3: ready for use by regmatch() 3072 */ 3073 3074 { 3075 regmatch_state *old_regmatch_state; 3076 regmatch_slab *old_regmatch_slab; 3077 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1; 3078 3079 /* on first ever match, allocate first slab */ 3080 if (!PL_regmatch_slab) { 3081 Newx(PL_regmatch_slab, 1, regmatch_slab); 3082 PL_regmatch_slab->prev = NULL; 3083 PL_regmatch_slab->next = NULL; 3084 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab); 3085 } 3086 3087 old_regmatch_state = PL_regmatch_state; 3088 old_regmatch_slab = PL_regmatch_slab; 3089 3090 for (i=0; i <= max; i++) { 3091 if (i == 1) 3092 reginfo->info_aux = &(PL_regmatch_state->u.info_aux); 3093 else if (i ==2) 3094 reginfo->info_aux_eval = 3095 reginfo->info_aux->info_aux_eval = 3096 &(PL_regmatch_state->u.info_aux_eval); 3097 3098 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab)) 3099 PL_regmatch_state = S_push_slab(aTHX); 3100 } 3101 3102 /* note initial PL_regmatch_state position; at end of match we'll 3103 * pop back to there and free any higher slabs */ 3104 3105 reginfo->info_aux->old_regmatch_state = old_regmatch_state; 3106 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab; 3107 reginfo->info_aux->poscache = NULL; 3108 3109 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux); 3110 3111 if ((prog->extflags & RXf_EVAL_SEEN)) 3112 S_setup_eval_state(aTHX_ reginfo); 3113 else 3114 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL; 3115 } 3116 3117 /* If there is a "must appear" string, look for it. */ 3118 3119 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) { 3120 /* We have to be careful. If the previous successful match 3121 was from this regex we don't want a subsequent partially 3122 successful match to clobber the old results. 3123 So when we detect this possibility we add a swap buffer 3124 to the re, and switch the buffer each match. If we fail, 3125 we switch it back; otherwise we leave it swapped. 3126 */ 3127 swap = prog->offs; 3128 /* do we need a save destructor here for eval dies? */ 3129 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair); 3130 DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ 3131 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n", 3132 PTR2UV(prog), 3133 PTR2UV(swap), 3134 PTR2UV(prog->offs) 3135 )); 3136 } 3137 3138 if (prog->recurse_locinput) 3139 Zero(prog->recurse_locinput,prog->nparens + 1, char *); 3140 3141 /* Simplest case: anchored match need be tried only once, or with 3142 * MBOL, only at the beginning of each line. 3143 * 3144 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets 3145 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't 3146 * match at the start of the string then it won't match anywhere else 3147 * either; while with /.*.../, if it doesn't match at the beginning, 3148 * the earliest it could match is at the start of the next line */ 3149 3150 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) { 3151 char *end; 3152 3153 if (regtry(reginfo, &s)) 3154 goto got_it; 3155 3156 if (!(prog->intflags & PREGf_ANCH_MBOL)) 3157 goto phooey; 3158 3159 /* didn't match at start, try at other newline positions */ 3160 3161 if (minlen) 3162 dontbother = minlen - 1; 3163 end = HOP3c(strend, -dontbother, strbeg) - 1; 3164 3165 /* skip to next newline */ 3166 3167 while (s <= end) { /* note it could be possible to match at the end of the string */ 3168 /* NB: newlines are the same in unicode as they are in latin */ 3169 if (*s++ != '\n') 3170 continue; 3171 if (prog->check_substr || prog->check_utf8) { 3172 /* note that with PREGf_IMPLICIT, intuit can only fail 3173 * or return the start position, so it's of limited utility. 3174 * Nevertheless, I made the decision that the potential for 3175 * quick fail was still worth it - DAPM */ 3176 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL); 3177 if (!s) 3178 goto phooey; 3179 } 3180 if (regtry(reginfo, &s)) 3181 goto got_it; 3182 } 3183 goto phooey; 3184 } /* end anchored search */ 3185 3186 if (prog->intflags & PREGf_ANCH_GPOS) 3187 { 3188 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */ 3189 assert(prog->intflags & PREGf_GPOS_SEEN); 3190 /* For anchored \G, the only position it can match from is 3191 * (ganch-gofs); we already set startpos to this above; if intuit 3192 * moved us on from there, we can't possibly succeed */ 3193 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs)); 3194 if (s == startpos && regtry(reginfo, &s)) 3195 goto got_it; 3196 goto phooey; 3197 } 3198 3199 /* Messy cases: unanchored match. */ 3200 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) { 3201 /* we have /x+whatever/ */ 3202 /* it must be a one character string (XXXX Except is_utf8_pat?) */ 3203 char ch; 3204 #ifdef DEBUGGING 3205 int did_match = 0; 3206 #endif 3207 if (utf8_target) { 3208 if (! prog->anchored_utf8) { 3209 to_utf8_substr(prog); 3210 } 3211 ch = SvPVX_const(prog->anchored_utf8)[0]; 3212 REXEC_FBC_SCAN( 3213 if (*s == ch) { 3214 DEBUG_EXECUTE_r( did_match = 1 ); 3215 if (regtry(reginfo, &s)) goto got_it; 3216 s += UTF8SKIP(s); 3217 while (s < strend && *s == ch) 3218 s += UTF8SKIP(s); 3219 } 3220 ); 3221 3222 } 3223 else { 3224 if (! prog->anchored_substr) { 3225 if (! to_byte_substr(prog)) { 3226 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 3227 } 3228 } 3229 ch = SvPVX_const(prog->anchored_substr)[0]; 3230 REXEC_FBC_SCAN( 3231 if (*s == ch) { 3232 DEBUG_EXECUTE_r( did_match = 1 ); 3233 if (regtry(reginfo, &s)) goto got_it; 3234 s++; 3235 while (s < strend && *s == ch) 3236 s++; 3237 } 3238 ); 3239 } 3240 DEBUG_EXECUTE_r(if (!did_match) 3241 Perl_re_printf( aTHX_ 3242 "Did not find anchored character...\n") 3243 ); 3244 } 3245 else if (prog->anchored_substr != NULL 3246 || prog->anchored_utf8 != NULL 3247 || ((prog->float_substr != NULL || prog->float_utf8 != NULL) 3248 && prog->float_max_offset < strend - s)) { 3249 SV *must; 3250 SSize_t back_max; 3251 SSize_t back_min; 3252 char *last; 3253 char *last1; /* Last position checked before */ 3254 #ifdef DEBUGGING 3255 int did_match = 0; 3256 #endif 3257 if (prog->anchored_substr || prog->anchored_utf8) { 3258 if (utf8_target) { 3259 if (! prog->anchored_utf8) { 3260 to_utf8_substr(prog); 3261 } 3262 must = prog->anchored_utf8; 3263 } 3264 else { 3265 if (! prog->anchored_substr) { 3266 if (! to_byte_substr(prog)) { 3267 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 3268 } 3269 } 3270 must = prog->anchored_substr; 3271 } 3272 back_max = back_min = prog->anchored_offset; 3273 } else { 3274 if (utf8_target) { 3275 if (! prog->float_utf8) { 3276 to_utf8_substr(prog); 3277 } 3278 must = prog->float_utf8; 3279 } 3280 else { 3281 if (! prog->float_substr) { 3282 if (! to_byte_substr(prog)) { 3283 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 3284 } 3285 } 3286 must = prog->float_substr; 3287 } 3288 back_max = prog->float_max_offset; 3289 back_min = prog->float_min_offset; 3290 } 3291 3292 if (back_min<0) { 3293 last = strend; 3294 } else { 3295 last = HOP3c(strend, /* Cannot start after this */ 3296 -(SSize_t)(CHR_SVLEN(must) 3297 - (SvTAIL(must) != 0) + back_min), strbeg); 3298 } 3299 if (s > reginfo->strbeg) 3300 last1 = HOPc(s, -1); 3301 else 3302 last1 = s - 1; /* bogus */ 3303 3304 /* XXXX check_substr already used to find "s", can optimize if 3305 check_substr==must. */ 3306 dontbother = 0; 3307 strend = HOPc(strend, -dontbother); 3308 while ( (s <= last) && 3309 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend), 3310 (unsigned char*)strend, must, 3311 multiline ? FBMrf_MULTILINE : 0)) ) { 3312 DEBUG_EXECUTE_r( did_match = 1 ); 3313 if (HOPc(s, -back_max) > last1) { 3314 last1 = HOPc(s, -back_min); 3315 s = HOPc(s, -back_max); 3316 } 3317 else { 3318 char * const t = (last1 >= reginfo->strbeg) 3319 ? HOPc(last1, 1) : last1 + 1; 3320 3321 last1 = HOPc(s, -back_min); 3322 s = t; 3323 } 3324 if (utf8_target) { 3325 while (s <= last1) { 3326 if (regtry(reginfo, &s)) 3327 goto got_it; 3328 if (s >= last1) { 3329 s++; /* to break out of outer loop */ 3330 break; 3331 } 3332 s += UTF8SKIP(s); 3333 } 3334 } 3335 else { 3336 while (s <= last1) { 3337 if (regtry(reginfo, &s)) 3338 goto got_it; 3339 s++; 3340 } 3341 } 3342 } 3343 DEBUG_EXECUTE_r(if (!did_match) { 3344 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), 3345 SvPVX_const(must), RE_SV_DUMPLEN(must), 30); 3346 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n", 3347 ((must == prog->anchored_substr || must == prog->anchored_utf8) 3348 ? "anchored" : "floating"), 3349 quoted, RE_SV_TAIL(must)); 3350 }); 3351 goto phooey; 3352 } 3353 else if ( (c = progi->regstclass) ) { 3354 if (minlen) { 3355 const OPCODE op = OP(progi->regstclass); 3356 /* don't bother with what can't match */ 3357 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE) 3358 strend = HOPc(strend, -(minlen - 1)); 3359 } 3360 DEBUG_EXECUTE_r({ 3361 SV * const prop = sv_newmortal(); 3362 regprop(prog, prop, c, reginfo, NULL); 3363 { 3364 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1), 3365 s,strend-s,60); 3366 Perl_re_printf( aTHX_ 3367 "Matching stclass %.*s against %s (%d bytes)\n", 3368 (int)SvCUR(prop), SvPVX_const(prop), 3369 quoted, (int)(strend - s)); 3370 } 3371 }); 3372 if (find_byclass(prog, c, s, strend, reginfo)) 3373 goto got_it; 3374 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n")); 3375 } 3376 else { 3377 dontbother = 0; 3378 if (prog->float_substr != NULL || prog->float_utf8 != NULL) { 3379 /* Trim the end. */ 3380 char *last= NULL; 3381 SV* float_real; 3382 STRLEN len; 3383 const char *little; 3384 3385 if (utf8_target) { 3386 if (! prog->float_utf8) { 3387 to_utf8_substr(prog); 3388 } 3389 float_real = prog->float_utf8; 3390 } 3391 else { 3392 if (! prog->float_substr) { 3393 if (! to_byte_substr(prog)) { 3394 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 3395 } 3396 } 3397 float_real = prog->float_substr; 3398 } 3399 3400 little = SvPV_const(float_real, len); 3401 if (SvTAIL(float_real)) { 3402 /* This means that float_real contains an artificial \n on 3403 * the end due to the presence of something like this: 3404 * /foo$/ where we can match both "foo" and "foo\n" at the 3405 * end of the string. So we have to compare the end of the 3406 * string first against the float_real without the \n and 3407 * then against the full float_real with the string. We 3408 * have to watch out for cases where the string might be 3409 * smaller than the float_real or the float_real without 3410 * the \n. */ 3411 char *checkpos= strend - len; 3412 DEBUG_OPTIMISE_r( 3413 Perl_re_printf( aTHX_ 3414 "%sChecking for float_real.%s\n", 3415 PL_colors[4], PL_colors[5])); 3416 if (checkpos + 1 < strbeg) { 3417 /* can't match, even if we remove the trailing \n 3418 * string is too short to match */ 3419 DEBUG_EXECUTE_r( 3420 Perl_re_printf( aTHX_ 3421 "%sString shorter than required trailing substring, cannot match.%s\n", 3422 PL_colors[4], PL_colors[5])); 3423 goto phooey; 3424 } else if (memEQ(checkpos + 1, little, len - 1)) { 3425 /* can match, the end of the string matches without the 3426 * "\n" */ 3427 last = checkpos + 1; 3428 } else if (checkpos < strbeg) { 3429 /* cant match, string is too short when the "\n" is 3430 * included */ 3431 DEBUG_EXECUTE_r( 3432 Perl_re_printf( aTHX_ 3433 "%sString does not contain required trailing substring, cannot match.%s\n", 3434 PL_colors[4], PL_colors[5])); 3435 goto phooey; 3436 } else if (!multiline) { 3437 /* non multiline match, so compare with the "\n" at the 3438 * end of the string */ 3439 if (memEQ(checkpos, little, len)) { 3440 last= checkpos; 3441 } else { 3442 DEBUG_EXECUTE_r( 3443 Perl_re_printf( aTHX_ 3444 "%sString does not contain required trailing substring, cannot match.%s\n", 3445 PL_colors[4], PL_colors[5])); 3446 goto phooey; 3447 } 3448 } else { 3449 /* multiline match, so we have to search for a place 3450 * where the full string is located */ 3451 goto find_last; 3452 } 3453 } else { 3454 find_last: 3455 if (len) 3456 last = rninstr(s, strend, little, little + len); 3457 else 3458 last = strend; /* matching "$" */ 3459 } 3460 if (!last) { 3461 /* at one point this block contained a comment which was 3462 * probably incorrect, which said that this was a "should not 3463 * happen" case. Even if it was true when it was written I am 3464 * pretty sure it is not anymore, so I have removed the comment 3465 * and replaced it with this one. Yves */ 3466 DEBUG_EXECUTE_r( 3467 Perl_re_printf( aTHX_ 3468 "%sString does not contain required substring, cannot match.%s\n", 3469 PL_colors[4], PL_colors[5] 3470 )); 3471 goto phooey; 3472 } 3473 dontbother = strend - last + prog->float_min_offset; 3474 } 3475 if (minlen && (dontbother < minlen)) 3476 dontbother = minlen - 1; 3477 strend -= dontbother; /* this one's always in bytes! */ 3478 /* We don't know much -- general case. */ 3479 if (utf8_target) { 3480 for (;;) { 3481 if (regtry(reginfo, &s)) 3482 goto got_it; 3483 if (s >= strend) 3484 break; 3485 s += UTF8SKIP(s); 3486 }; 3487 } 3488 else { 3489 do { 3490 if (regtry(reginfo, &s)) 3491 goto got_it; 3492 } while (s++ < strend); 3493 } 3494 } 3495 3496 /* Failure. */ 3497 goto phooey; 3498 3499 got_it: 3500 /* s/// doesn't like it if $& is earlier than where we asked it to 3501 * start searching (which can happen on something like /.\G/) */ 3502 if ( (flags & REXEC_FAIL_ON_UNDERFLOW) 3503 && (prog->offs[0].start < stringarg - strbeg)) 3504 { 3505 /* this should only be possible under \G */ 3506 assert(prog->intflags & PREGf_GPOS_SEEN); 3507 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ 3508 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); 3509 goto phooey; 3510 } 3511 3512 DEBUG_BUFFERS_r( 3513 if (swap) 3514 Perl_re_printf( aTHX_ 3515 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n", 3516 PTR2UV(prog), 3517 PTR2UV(swap) 3518 ); 3519 ); 3520 Safefree(swap); 3521 3522 /* clean up; this will trigger destructors that will free all slabs 3523 * above the current one, and cleanup the regmatch_info_aux 3524 * and regmatch_info_aux_eval sructs */ 3525 3526 LEAVE_SCOPE(oldsave); 3527 3528 if (RXp_PAREN_NAMES(prog)) 3529 (void)hv_iterinit(RXp_PAREN_NAMES(prog)); 3530 3531 /* make sure $`, $&, $', and $digit will work later */ 3532 if ( !(flags & REXEC_NOT_FIRST) ) 3533 S_reg_set_capture_string(aTHX_ rx, 3534 strbeg, reginfo->strend, 3535 sv, flags, utf8_target); 3536 3537 return 1; 3538 3539 phooey: 3540 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n", 3541 PL_colors[4], PL_colors[5])); 3542 3543 /* clean up; this will trigger destructors that will free all slabs 3544 * above the current one, and cleanup the regmatch_info_aux 3545 * and regmatch_info_aux_eval sructs */ 3546 3547 LEAVE_SCOPE(oldsave); 3548 3549 if (swap) { 3550 /* we failed :-( roll it back */ 3551 DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ 3552 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n", 3553 PTR2UV(prog), 3554 PTR2UV(prog->offs), 3555 PTR2UV(swap) 3556 )); 3557 Safefree(prog->offs); 3558 prog->offs = swap; 3559 } 3560 return 0; 3561 } 3562 3563 3564 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting. 3565 * Do inc before dec, in case old and new rex are the same */ 3566 #define SET_reg_curpm(Re2) \ 3567 if (reginfo->info_aux_eval) { \ 3568 (void)ReREFCNT_inc(Re2); \ 3569 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \ 3570 PM_SETRE((PL_reg_curpm), (Re2)); \ 3571 } 3572 3573 3574 /* 3575 - regtry - try match at specific point 3576 */ 3577 STATIC bool /* 0 failure, 1 success */ 3578 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp) 3579 { 3580 CHECKPOINT lastcp; 3581 REGEXP *const rx = reginfo->prog; 3582 regexp *const prog = ReANY(rx); 3583 SSize_t result; 3584 #ifdef DEBUGGING 3585 U32 depth = 0; /* used by REGCP_SET */ 3586 #endif 3587 RXi_GET_DECL(prog,progi); 3588 GET_RE_DEBUG_FLAGS_DECL; 3589 3590 PERL_ARGS_ASSERT_REGTRY; 3591 3592 reginfo->cutpoint=NULL; 3593 3594 prog->offs[0].start = *startposp - reginfo->strbeg; 3595 prog->lastparen = 0; 3596 prog->lastcloseparen = 0; 3597 3598 /* XXXX What this code is doing here?!!! There should be no need 3599 to do this again and again, prog->lastparen should take care of 3600 this! --ilya*/ 3601 3602 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code. 3603 * Actually, the code in regcppop() (which Ilya may be meaning by 3604 * prog->lastparen), is not needed at all by the test suite 3605 * (op/regexp, op/pat, op/split), but that code is needed otherwise 3606 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ 3607 * Meanwhile, this code *is* needed for the 3608 * above-mentioned test suite tests to succeed. The common theme 3609 * on those tests seems to be returning null fields from matches. 3610 * --jhi updated by dapm */ 3611 #if 1 3612 if (prog->nparens) { 3613 regexp_paren_pair *pp = prog->offs; 3614 I32 i; 3615 for (i = prog->nparens; i > (I32)prog->lastparen; i--) { 3616 ++pp; 3617 pp->start = -1; 3618 pp->end = -1; 3619 } 3620 } 3621 #endif 3622 REGCP_SET(lastcp); 3623 result = regmatch(reginfo, *startposp, progi->program + 1); 3624 if (result != -1) { 3625 prog->offs[0].end = result; 3626 return 1; 3627 } 3628 if (reginfo->cutpoint) 3629 *startposp= reginfo->cutpoint; 3630 REGCP_UNWIND(lastcp); 3631 return 0; 3632 } 3633 3634 3635 #define sayYES goto yes 3636 #define sayNO goto no 3637 #define sayNO_SILENT goto no_silent 3638 3639 /* we dont use STMT_START/END here because it leads to 3640 "unreachable code" warnings, which are bogus, but distracting. */ 3641 #define CACHEsayNO \ 3642 if (ST.cache_mask) \ 3643 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \ 3644 sayNO 3645 3646 /* this is used to determine how far from the left messages like 3647 'failed...' are printed in regexec.c. It should be set such that 3648 messages are inline with the regop output that created them. 3649 */ 3650 #define REPORT_CODE_OFF 29 3651 #define INDENT_CHARS(depth) ((int)(depth) % 20) 3652 #ifdef DEBUGGING 3653 int 3654 Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...) 3655 { 3656 va_list ap; 3657 int result; 3658 PerlIO *f= Perl_debug_log; 3659 PERL_ARGS_ASSERT_RE_EXEC_INDENTF; 3660 va_start(ap, depth); 3661 PerlIO_printf(f, "%*s|%4"UVuf"| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" ); 3662 result = PerlIO_vprintf(f, fmt, ap); 3663 va_end(ap); 3664 return result; 3665 } 3666 #endif /* DEBUGGING */ 3667 3668 3669 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */ 3670 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */ 3671 #define CHRTEST_NOT_A_CP_1 -999 3672 #define CHRTEST_NOT_A_CP_2 -998 3673 3674 /* grab a new slab and return the first slot in it */ 3675 3676 STATIC regmatch_state * 3677 S_push_slab(pTHX) 3678 { 3679 #if PERL_VERSION < 9 && !defined(PERL_CORE) 3680 dMY_CXT; 3681 #endif 3682 regmatch_slab *s = PL_regmatch_slab->next; 3683 if (!s) { 3684 Newx(s, 1, regmatch_slab); 3685 s->prev = PL_regmatch_slab; 3686 s->next = NULL; 3687 PL_regmatch_slab->next = s; 3688 } 3689 PL_regmatch_slab = s; 3690 return SLAB_FIRST(s); 3691 } 3692 3693 3694 /* push a new state then goto it */ 3695 3696 #define PUSH_STATE_GOTO(state, node, input) \ 3697 pushinput = input; \ 3698 scan = node; \ 3699 st->resume_state = state; \ 3700 goto push_state; 3701 3702 /* push a new state with success backtracking, then goto it */ 3703 3704 #define PUSH_YES_STATE_GOTO(state, node, input) \ 3705 pushinput = input; \ 3706 scan = node; \ 3707 st->resume_state = state; \ 3708 goto push_yes_state; 3709 3710 3711 3712 3713 /* 3714 3715 regmatch() - main matching routine 3716 3717 This is basically one big switch statement in a loop. We execute an op, 3718 set 'next' to point the next op, and continue. If we come to a point which 3719 we may need to backtrack to on failure such as (A|B|C), we push a 3720 backtrack state onto the backtrack stack. On failure, we pop the top 3721 state, and re-enter the loop at the state indicated. If there are no more 3722 states to pop, we return failure. 3723 3724 Sometimes we also need to backtrack on success; for example /A+/, where 3725 after successfully matching one A, we need to go back and try to 3726 match another one; similarly for lookahead assertions: if the assertion 3727 completes successfully, we backtrack to the state just before the assertion 3728 and then carry on. In these cases, the pushed state is marked as 3729 'backtrack on success too'. This marking is in fact done by a chain of 3730 pointers, each pointing to the previous 'yes' state. On success, we pop to 3731 the nearest yes state, discarding any intermediate failure-only states. 3732 Sometimes a yes state is pushed just to force some cleanup code to be 3733 called at the end of a successful match or submatch; e.g. (??{$re}) uses 3734 it to free the inner regex. 3735 3736 Note that failure backtracking rewinds the cursor position, while 3737 success backtracking leaves it alone. 3738 3739 A pattern is complete when the END op is executed, while a subpattern 3740 such as (?=foo) is complete when the SUCCESS op is executed. Both of these 3741 ops trigger the "pop to last yes state if any, otherwise return true" 3742 behaviour. 3743 3744 A common convention in this function is to use A and B to refer to the two 3745 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is 3746 the subpattern to be matched possibly multiple times, while B is the entire 3747 rest of the pattern. Variable and state names reflect this convention. 3748 3749 The states in the main switch are the union of ops and failure/success of 3750 substates associated with with that op. For example, IFMATCH is the op 3751 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means 3752 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just 3753 successfully matched A and IFMATCH_A_fail is a state saying that we have 3754 just failed to match A. Resume states always come in pairs. The backtrack 3755 state we push is marked as 'IFMATCH_A', but when that is popped, we resume 3756 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking 3757 on success or failure. 3758 3759 The struct that holds a backtracking state is actually a big union, with 3760 one variant for each major type of op. The variable st points to the 3761 top-most backtrack struct. To make the code clearer, within each 3762 block of code we #define ST to alias the relevant union. 3763 3764 Here's a concrete example of a (vastly oversimplified) IFMATCH 3765 implementation: 3766 3767 switch (state) { 3768 .... 3769 3770 #define ST st->u.ifmatch 3771 3772 case IFMATCH: // we are executing the IFMATCH op, (?=A)B 3773 ST.foo = ...; // some state we wish to save 3774 ... 3775 // push a yes backtrack state with a resume value of 3776 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the 3777 // first node of A: 3778 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput); 3779 // NOTREACHED 3780 3781 case IFMATCH_A: // we have successfully executed A; now continue with B 3782 next = B; 3783 bar = ST.foo; // do something with the preserved value 3784 break; 3785 3786 case IFMATCH_A_fail: // A failed, so the assertion failed 3787 ...; // do some housekeeping, then ... 3788 sayNO; // propagate the failure 3789 3790 #undef ST 3791 3792 ... 3793 } 3794 3795 For any old-timers reading this who are familiar with the old recursive 3796 approach, the code above is equivalent to: 3797 3798 case IFMATCH: // we are executing the IFMATCH op, (?=A)B 3799 { 3800 int foo = ... 3801 ... 3802 if (regmatch(A)) { 3803 next = B; 3804 bar = foo; 3805 break; 3806 } 3807 ...; // do some housekeeping, then ... 3808 sayNO; // propagate the failure 3809 } 3810 3811 The topmost backtrack state, pointed to by st, is usually free. If you 3812 want to claim it, populate any ST.foo fields in it with values you wish to 3813 save, then do one of 3814 3815 PUSH_STATE_GOTO(resume_state, node, newinput); 3816 PUSH_YES_STATE_GOTO(resume_state, node, newinput); 3817 3818 which sets that backtrack state's resume value to 'resume_state', pushes a 3819 new free entry to the top of the backtrack stack, then goes to 'node'. 3820 On backtracking, the free slot is popped, and the saved state becomes the 3821 new free state. An ST.foo field in this new top state can be temporarily 3822 accessed to retrieve values, but once the main loop is re-entered, it 3823 becomes available for reuse. 3824 3825 Note that the depth of the backtrack stack constantly increases during the 3826 left-to-right execution of the pattern, rather than going up and down with 3827 the pattern nesting. For example the stack is at its maximum at Z at the 3828 end of the pattern, rather than at X in the following: 3829 3830 /(((X)+)+)+....(Y)+....Z/ 3831 3832 The only exceptions to this are lookahead/behind assertions and the cut, 3833 (?>A), which pop all the backtrack states associated with A before 3834 continuing. 3835 3836 Backtrack state structs are allocated in slabs of about 4K in size. 3837 PL_regmatch_state and st always point to the currently active state, 3838 and PL_regmatch_slab points to the slab currently containing 3839 PL_regmatch_state. The first time regmatch() is called, the first slab is 3840 allocated, and is never freed until interpreter destruction. When the slab 3841 is full, a new one is allocated and chained to the end. At exit from 3842 regmatch(), slabs allocated since entry are freed. 3843 3844 */ 3845 3846 3847 #define DEBUG_STATE_pp(pp) \ 3848 DEBUG_STATE_r({ \ 3849 DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \ 3850 Perl_re_printf( aTHX_ \ 3851 "%*s" pp " %s%s%s%s%s\n", \ 3852 INDENT_CHARS(depth), "", \ 3853 PL_reg_name[st->resume_state], \ 3854 ((st==yes_state||st==mark_state) ? "[" : ""), \ 3855 ((st==yes_state) ? "Y" : ""), \ 3856 ((st==mark_state) ? "M" : ""), \ 3857 ((st==yes_state||st==mark_state) ? "]" : "") \ 3858 ); \ 3859 }); 3860 3861 3862 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) 3863 3864 #ifdef DEBUGGING 3865 3866 STATIC void 3867 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target, 3868 const char *start, const char *end, const char *blurb) 3869 { 3870 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0; 3871 3872 PERL_ARGS_ASSERT_DEBUG_START_MATCH; 3873 3874 if (!PL_colorset) 3875 reginitcolors(); 3876 { 3877 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0), 3878 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60); 3879 3880 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1), 3881 start, end - start, 60); 3882 3883 Perl_re_printf( aTHX_ 3884 "%s%s REx%s %s against %s\n", 3885 PL_colors[4], blurb, PL_colors[5], s0, s1); 3886 3887 if (utf8_target||utf8_pat) 3888 Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n", 3889 utf8_pat ? "pattern" : "", 3890 utf8_pat && utf8_target ? " and " : "", 3891 utf8_target ? "string" : "" 3892 ); 3893 } 3894 } 3895 3896 STATIC void 3897 S_dump_exec_pos(pTHX_ const char *locinput, 3898 const regnode *scan, 3899 const char *loc_regeol, 3900 const char *loc_bostr, 3901 const char *loc_reg_starttry, 3902 const bool utf8_target, 3903 const U32 depth 3904 ) 3905 { 3906 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; 3907 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ 3908 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); 3909 /* The part of the string before starttry has one color 3910 (pref0_len chars), between starttry and current 3911 position another one (pref_len - pref0_len chars), 3912 after the current position the third one. 3913 We assume that pref0_len <= pref_len, otherwise we 3914 decrease pref0_len. */ 3915 int pref_len = (locinput - loc_bostr) > (5 + taill) - l 3916 ? (5 + taill) - l : locinput - loc_bostr; 3917 int pref0_len; 3918 3919 PERL_ARGS_ASSERT_DUMP_EXEC_POS; 3920 3921 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) 3922 pref_len++; 3923 pref0_len = pref_len - (locinput - loc_reg_starttry); 3924 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) 3925 l = ( loc_regeol - locinput > (5 + taill) - pref_len 3926 ? (5 + taill) - pref_len : loc_regeol - locinput); 3927 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) 3928 l--; 3929 if (pref0_len < 0) 3930 pref0_len = 0; 3931 if (pref0_len > pref_len) 3932 pref0_len = pref_len; 3933 { 3934 const int is_uni = utf8_target ? 1 : 0; 3935 3936 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0), 3937 (locinput - pref_len),pref0_len, 60, 4, 5); 3938 3939 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1), 3940 (locinput - pref_len + pref0_len), 3941 pref_len - pref0_len, 60, 2, 3); 3942 3943 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), 3944 locinput, loc_regeol - locinput, 10, 0, 1); 3945 3946 const STRLEN tlen=len0+len1+len2; 3947 Perl_re_printf( aTHX_ 3948 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|%4u| ", 3949 (IV)(locinput - loc_bostr), 3950 len0, s0, 3951 len1, s1, 3952 (docolor ? "" : "> <"), 3953 len2, s2, 3954 (int)(tlen > 19 ? 0 : 19 - tlen), 3955 "", 3956 depth); 3957 } 3958 } 3959 3960 #endif 3961 3962 /* reg_check_named_buff_matched() 3963 * Checks to see if a named buffer has matched. The data array of 3964 * buffer numbers corresponding to the buffer is expected to reside 3965 * in the regexp->data->data array in the slot stored in the ARG() of 3966 * node involved. Note that this routine doesn't actually care about the 3967 * name, that information is not preserved from compilation to execution. 3968 * Returns the index of the leftmost defined buffer with the given name 3969 * or 0 if non of the buffers matched. 3970 */ 3971 STATIC I32 3972 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan) 3973 { 3974 I32 n; 3975 RXi_GET_DECL(rex,rexi); 3976 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 3977 I32 *nums=(I32*)SvPVX(sv_dat); 3978 3979 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED; 3980 3981 for ( n=0; n<SvIVX(sv_dat); n++ ) { 3982 if ((I32)rex->lastparen >= nums[n] && 3983 rex->offs[nums[n]].end != -1) 3984 { 3985 return nums[n]; 3986 } 3987 } 3988 return 0; 3989 } 3990 3991 3992 static bool 3993 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, 3994 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo) 3995 { 3996 /* This function determines if there are one or two characters that match 3997 * the first character of the passed-in EXACTish node <text_node>, and if 3998 * so, returns them in the passed-in pointers. 3999 * 4000 * If it determines that no possible character in the target string can 4001 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if 4002 * the first character in <text_node> requires UTF-8 to represent, and the 4003 * target string isn't in UTF-8.) 4004 * 4005 * If there are more than two characters that could match the beginning of 4006 * <text_node>, or if more context is required to determine a match or not, 4007 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID. 4008 * 4009 * The motiviation behind this function is to allow the caller to set up 4010 * tight loops for matching. If <text_node> is of type EXACT, there is 4011 * only one possible character that can match its first character, and so 4012 * the situation is quite simple. But things get much more complicated if 4013 * folding is involved. It may be that the first character of an EXACTFish 4014 * node doesn't participate in any possible fold, e.g., punctuation, so it 4015 * can be matched only by itself. The vast majority of characters that are 4016 * in folds match just two things, their lower and upper-case equivalents. 4017 * But not all are like that; some have multiple possible matches, or match 4018 * sequences of more than one character. This function sorts all that out. 4019 * 4020 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a 4021 * loop of trying to match A*, we know we can't exit where the thing 4022 * following it isn't a B. And something can't be a B unless it is the 4023 * beginning of B. By putting a quick test for that beginning in a tight 4024 * loop, we can rule out things that can't possibly be B without having to 4025 * break out of the loop, thus avoiding work. Similarly, if A is a single 4026 * character, we can make a tight loop matching A*, using the outputs of 4027 * this function. 4028 * 4029 * If the target string to match isn't in UTF-8, and there aren't 4030 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to 4031 * the one or two possible octets (which are characters in this situation) 4032 * that can match. In all cases, if there is only one character that can 4033 * match, *<c1p> and *<c2p> will be identical. 4034 * 4035 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8> 4036 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that 4037 * can match the beginning of <text_node>. They should be declared with at 4038 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is 4039 * undefined what these contain.) If one or both of the buffers are 4040 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the 4041 * corresponding invariant. If variant, the corresponding *<c1p> and/or 4042 * *<c2p> will be set to a negative number(s) that shouldn't match any code 4043 * point (unless inappropriately coerced to unsigned). *<c1p> will equal 4044 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */ 4045 4046 const bool utf8_target = reginfo->is_utf8_target; 4047 4048 UV c1 = (UV)CHRTEST_NOT_A_CP_1; 4049 UV c2 = (UV)CHRTEST_NOT_A_CP_2; 4050 bool use_chrtest_void = FALSE; 4051 const bool is_utf8_pat = reginfo->is_utf8_pat; 4052 4053 /* Used when we have both utf8 input and utf8 output, to avoid converting 4054 * to/from code points */ 4055 bool utf8_has_been_setup = FALSE; 4056 4057 dVAR; 4058 4059 U8 *pat = (U8*)STRING(text_node); 4060 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' }; 4061 4062 if (OP(text_node) == EXACT || OP(text_node) == EXACTL) { 4063 4064 /* In an exact node, only one thing can be matched, that first 4065 * character. If both the pat and the target are UTF-8, we can just 4066 * copy the input to the output, avoiding finding the code point of 4067 * that character */ 4068 if (!is_utf8_pat) { 4069 c2 = c1 = *pat; 4070 } 4071 else if (utf8_target) { 4072 Copy(pat, c1_utf8, UTF8SKIP(pat), U8); 4073 Copy(pat, c2_utf8, UTF8SKIP(pat), U8); 4074 utf8_has_been_setup = TRUE; 4075 } 4076 else { 4077 c2 = c1 = valid_utf8_to_uvchr(pat, NULL); 4078 } 4079 } 4080 else { /* an EXACTFish node */ 4081 U8 *pat_end = pat + STR_LEN(text_node); 4082 4083 /* An EXACTFL node has at least some characters unfolded, because what 4084 * they match is not known until now. So, now is the time to fold 4085 * the first few of them, as many as are needed to determine 'c1' and 4086 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need 4087 * to fold if in a UTF-8 locale, and then only the Sharp S; everything 4088 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we 4089 * need to fold as many characters as a single character can fold to, 4090 * so that later we can check if the first ones are such a multi-char 4091 * fold. But, in such a pattern only locale-problematic characters 4092 * aren't folded, so we can skip this completely if the first character 4093 * in the node isn't one of the tricky ones */ 4094 if (OP(text_node) == EXACTFL) { 4095 4096 if (! is_utf8_pat) { 4097 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S) 4098 { 4099 folded[0] = folded[1] = 's'; 4100 pat = folded; 4101 pat_end = folded + 2; 4102 } 4103 } 4104 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) { 4105 U8 *s = pat; 4106 U8 *d = folded; 4107 int i; 4108 4109 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) { 4110 if (isASCII(*s)) { 4111 *(d++) = (U8) toFOLD_LC(*s); 4112 s++; 4113 } 4114 else { 4115 STRLEN len; 4116 _to_utf8_fold_flags(s, 4117 d, 4118 &len, 4119 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE); 4120 d += len; 4121 s += UTF8SKIP(s); 4122 } 4123 } 4124 4125 pat = folded; 4126 pat_end = d; 4127 } 4128 } 4129 4130 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end)) 4131 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end))) 4132 { 4133 /* Multi-character folds require more context to sort out. Also 4134 * PL_utf8_foldclosures used below doesn't handle them, so have to 4135 * be handled outside this routine */ 4136 use_chrtest_void = TRUE; 4137 } 4138 else { /* an EXACTFish node which doesn't begin with a multi-char fold */ 4139 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat; 4140 if (c1 > 255) { 4141 /* Load the folds hash, if not already done */ 4142 SV** listp; 4143 if (! PL_utf8_foldclosures) { 4144 _load_PL_utf8_foldclosures(); 4145 } 4146 4147 /* The fold closures data structure is a hash with the keys 4148 * being the UTF-8 of every character that is folded to, like 4149 * 'k', and the values each an array of all code points that 4150 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ]. 4151 * Multi-character folds are not included */ 4152 if ((! (listp = hv_fetch(PL_utf8_foldclosures, 4153 (char *) pat, 4154 UTF8SKIP(pat), 4155 FALSE)))) 4156 { 4157 /* Not found in the hash, therefore there are no folds 4158 * containing it, so there is only a single character that 4159 * could match */ 4160 c2 = c1; 4161 } 4162 else { /* Does participate in folds */ 4163 AV* list = (AV*) *listp; 4164 if (av_tindex_nomg(list) != 1) { 4165 4166 /* If there aren't exactly two folds to this, it is 4167 * outside the scope of this function */ 4168 use_chrtest_void = TRUE; 4169 } 4170 else { /* There are two. Get them */ 4171 SV** c_p = av_fetch(list, 0, FALSE); 4172 if (c_p == NULL) { 4173 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); 4174 } 4175 c1 = SvUV(*c_p); 4176 4177 c_p = av_fetch(list, 1, FALSE); 4178 if (c_p == NULL) { 4179 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); 4180 } 4181 c2 = SvUV(*c_p); 4182 4183 /* Folds that cross the 255/256 boundary are forbidden 4184 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and 4185 * one is ASCIII. Since the pattern character is above 4186 * 255, and its only other match is below 256, the only 4187 * legal match will be to itself. We have thrown away 4188 * the original, so have to compute which is the one 4189 * above 255. */ 4190 if ((c1 < 256) != (c2 < 256)) { 4191 if ((OP(text_node) == EXACTFL 4192 && ! IN_UTF8_CTYPE_LOCALE) 4193 || ((OP(text_node) == EXACTFA 4194 || OP(text_node) == EXACTFA_NO_TRIE) 4195 && (isASCII(c1) || isASCII(c2)))) 4196 { 4197 if (c1 < 256) { 4198 c1 = c2; 4199 } 4200 else { 4201 c2 = c1; 4202 } 4203 } 4204 } 4205 } 4206 } 4207 } 4208 else /* Here, c1 is <= 255 */ 4209 if (utf8_target 4210 && HAS_NONLATIN1_FOLD_CLOSURE(c1) 4211 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE)) 4212 && ((OP(text_node) != EXACTFA 4213 && OP(text_node) != EXACTFA_NO_TRIE) 4214 || ! isASCII(c1))) 4215 { 4216 /* Here, there could be something above Latin1 in the target 4217 * which folds to this character in the pattern. All such 4218 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more 4219 * than two characters involved in their folds, so are outside 4220 * the scope of this function */ 4221 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { 4222 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; 4223 } 4224 else { 4225 use_chrtest_void = TRUE; 4226 } 4227 } 4228 else { /* Here nothing above Latin1 can fold to the pattern 4229 character */ 4230 switch (OP(text_node)) { 4231 4232 case EXACTFL: /* /l rules */ 4233 c2 = PL_fold_locale[c1]; 4234 break; 4235 4236 case EXACTF: /* This node only generated for non-utf8 4237 patterns */ 4238 assert(! is_utf8_pat); 4239 if (! utf8_target) { /* /d rules */ 4240 c2 = PL_fold[c1]; 4241 break; 4242 } 4243 /* FALLTHROUGH */ 4244 /* /u rules for all these. This happens to work for 4245 * EXACTFA as nothing in Latin1 folds to ASCII */ 4246 case EXACTFA_NO_TRIE: /* This node only generated for 4247 non-utf8 patterns */ 4248 assert(! is_utf8_pat); 4249 /* FALLTHROUGH */ 4250 case EXACTFA: 4251 case EXACTFU_SS: 4252 case EXACTFU: 4253 c2 = PL_fold_latin1[c1]; 4254 break; 4255 4256 default: 4257 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node)); 4258 NOT_REACHED; /* NOTREACHED */ 4259 } 4260 } 4261 } 4262 } 4263 4264 /* Here have figured things out. Set up the returns */ 4265 if (use_chrtest_void) { 4266 *c2p = *c1p = CHRTEST_VOID; 4267 } 4268 else if (utf8_target) { 4269 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */ 4270 uvchr_to_utf8(c1_utf8, c1); 4271 uvchr_to_utf8(c2_utf8, c2); 4272 } 4273 4274 /* Invariants are stored in both the utf8 and byte outputs; Use 4275 * negative numbers otherwise for the byte ones. Make sure that the 4276 * byte ones are the same iff the utf8 ones are the same */ 4277 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1; 4278 *c2p = (UTF8_IS_INVARIANT(*c2_utf8)) 4279 ? *c2_utf8 4280 : (c1 == c2) 4281 ? CHRTEST_NOT_A_CP_1 4282 : CHRTEST_NOT_A_CP_2; 4283 } 4284 else if (c1 > 255) { 4285 if (c2 > 255) { /* both possibilities are above what a non-utf8 string 4286 can represent */ 4287 return FALSE; 4288 } 4289 4290 *c1p = *c2p = c2; /* c2 is the only representable value */ 4291 } 4292 else { /* c1 is representable; see about c2 */ 4293 *c1p = c1; 4294 *c2p = (c2 < 256) ? c2 : c1; 4295 } 4296 4297 return TRUE; 4298 } 4299 4300 PERL_STATIC_INLINE bool 4301 S_isGCB(const GCB_enum before, const GCB_enum after) 4302 { 4303 /* returns a boolean indicating if there is a Grapheme Cluster Boundary 4304 * between the inputs. See http://www.unicode.org/reports/tr29/ */ 4305 4306 return GCB_table[before][after]; 4307 } 4308 4309 /* Combining marks attach to most classes that precede them, but this defines 4310 * the exceptions (from TR14) */ 4311 #define LB_CM_ATTACHES_TO(prev) ( ! ( prev == LB_EDGE \ 4312 || prev == LB_Mandatory_Break \ 4313 || prev == LB_Carriage_Return \ 4314 || prev == LB_Line_Feed \ 4315 || prev == LB_Next_Line \ 4316 || prev == LB_Space \ 4317 || prev == LB_ZWSpace)) 4318 4319 STATIC bool 4320 S_isLB(pTHX_ LB_enum before, 4321 LB_enum after, 4322 const U8 * const strbeg, 4323 const U8 * const curpos, 4324 const U8 * const strend, 4325 const bool utf8_target) 4326 { 4327 U8 * temp_pos = (U8 *) curpos; 4328 LB_enum prev = before; 4329 4330 /* Is the boundary between 'before' and 'after' line-breakable? 4331 * Most of this is just a table lookup of a generated table from Unicode 4332 * rules. But some rules require context to decide, and so have to be 4333 * implemented in code */ 4334 4335 PERL_ARGS_ASSERT_ISLB; 4336 4337 /* Rule numbers in the comments below are as of Unicode 8.0 */ 4338 4339 redo: 4340 before = prev; 4341 switch (LB_table[before][after]) { 4342 case LB_BREAKABLE: 4343 return TRUE; 4344 4345 case LB_NOBREAK: 4346 case LB_NOBREAK_EVEN_WITH_SP_BETWEEN: 4347 return FALSE; 4348 4349 case LB_SP_foo + LB_BREAKABLE: 4350 case LB_SP_foo + LB_NOBREAK: 4351 case LB_SP_foo + LB_NOBREAK_EVEN_WITH_SP_BETWEEN: 4352 4353 /* When we have something following a SP, we have to look at the 4354 * context in order to know what to do. 4355 * 4356 * SP SP should not reach here because LB7: Do not break before 4357 * spaces. (For two spaces in a row there is nothing that 4358 * overrides that) */ 4359 assert(after != LB_Space); 4360 4361 /* Here we have a space followed by a non-space. Mostly this is a 4362 * case of LB18: "Break after spaces". But there are complications 4363 * as the handling of spaces is somewhat tricky. They are in a 4364 * number of rules, which have to be applied in priority order, but 4365 * something earlier in the string can cause a rule to be skipped 4366 * and a lower priority rule invoked. A prime example is LB7 which 4367 * says don't break before a space. But rule LB8 (lower priority) 4368 * says that the first break opportunity after a ZW is after any 4369 * span of spaces immediately after it. If a ZW comes before a SP 4370 * in the input, rule LB8 applies, and not LB7. Other such rules 4371 * involve combining marks which are rules 9 and 10, but they may 4372 * override higher priority rules if they come earlier in the 4373 * string. Since we're doing random access into the middle of the 4374 * string, we have to look for rules that should get applied based 4375 * on both string position and priority. Combining marks do not 4376 * attach to either ZW nor SP, so we don't have to consider them 4377 * until later. 4378 * 4379 * To check for LB8, we have to find the first non-space character 4380 * before this span of spaces */ 4381 do { 4382 prev = backup_one_LB(strbeg, &temp_pos, utf8_target); 4383 } 4384 while (prev == LB_Space); 4385 4386 /* LB8 Break before any character following a zero-width space, 4387 * even if one or more spaces intervene. 4388 * ZW SP* ÷ 4389 * So if we have a ZW just before this span, and to get here this 4390 * is the final space in the span. */ 4391 if (prev == LB_ZWSpace) { 4392 return TRUE; 4393 } 4394 4395 /* Here, not ZW SP+. There are several rules that have higher 4396 * priority than LB18 and can be resolved now, as they don't depend 4397 * on anything earlier in the string (except ZW, which we have 4398 * already handled). One of these rules is LB11 Do not break 4399 * before Word joiner, but we have specially encoded that in the 4400 * lookup table so it is caught by the single test below which 4401 * catches the other ones. */ 4402 if (LB_table[LB_Space][after] - LB_SP_foo 4403 == LB_NOBREAK_EVEN_WITH_SP_BETWEEN) 4404 { 4405 return FALSE; 4406 } 4407 4408 /* If we get here, we have to XXX consider combining marks. */ 4409 if (prev == LB_Combining_Mark) { 4410 4411 /* What happens with these depends on the character they 4412 * follow. */ 4413 do { 4414 prev = backup_one_LB(strbeg, &temp_pos, utf8_target); 4415 } 4416 while (prev == LB_Combining_Mark); 4417 4418 /* Most times these attach to and inherit the characteristics 4419 * of that character, but not always, and when not, they are to 4420 * be treated as AL by rule LB10. */ 4421 if (! LB_CM_ATTACHES_TO(prev)) { 4422 prev = LB_Alphabetic; 4423 } 4424 } 4425 4426 /* Here, we have the character preceding the span of spaces all set 4427 * up. We follow LB18: "Break after spaces" unless the table shows 4428 * that is overriden */ 4429 return LB_table[prev][after] != LB_NOBREAK_EVEN_WITH_SP_BETWEEN; 4430 4431 case LB_CM_foo: 4432 4433 /* We don't know how to treat the CM except by looking at the first 4434 * non-CM character preceding it */ 4435 do { 4436 prev = backup_one_LB(strbeg, &temp_pos, utf8_target); 4437 } 4438 while (prev == LB_Combining_Mark); 4439 4440 /* Here, 'prev' is that first earlier non-CM character. If the CM 4441 * attatches to it, then it inherits the behavior of 'prev'. If it 4442 * doesn't attach, it is to be treated as an AL */ 4443 if (! LB_CM_ATTACHES_TO(prev)) { 4444 prev = LB_Alphabetic; 4445 } 4446 4447 goto redo; 4448 4449 case LB_HY_or_BA_then_foo + LB_BREAKABLE: 4450 case LB_HY_or_BA_then_foo + LB_NOBREAK: 4451 4452 /* LB21a Don't break after Hebrew + Hyphen. 4453 * HL (HY | BA) × */ 4454 4455 if (backup_one_LB(strbeg, &temp_pos, utf8_target) 4456 == LB_Hebrew_Letter) 4457 { 4458 return FALSE; 4459 } 4460 4461 return LB_table[prev][after] - LB_HY_or_BA_then_foo == LB_BREAKABLE; 4462 4463 case LB_PR_or_PO_then_OP_or_HY + LB_BREAKABLE: 4464 case LB_PR_or_PO_then_OP_or_HY + LB_NOBREAK: 4465 4466 /* LB25a (PR | PO) × ( OP | HY )? NU */ 4467 if (advance_one_LB(&temp_pos, strend, utf8_target) == LB_Numeric) { 4468 return FALSE; 4469 } 4470 4471 return LB_table[prev][after] - LB_PR_or_PO_then_OP_or_HY 4472 == LB_BREAKABLE; 4473 4474 case LB_SY_or_IS_then_various + LB_BREAKABLE: 4475 case LB_SY_or_IS_then_various + LB_NOBREAK: 4476 { 4477 /* LB25d NU (SY | IS)* × (NU | SY | IS | CL | CP ) */ 4478 4479 LB_enum temp = prev; 4480 do { 4481 temp = backup_one_LB(strbeg, &temp_pos, utf8_target); 4482 } 4483 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric); 4484 if (temp == LB_Numeric) { 4485 return FALSE; 4486 } 4487 4488 return LB_table[prev][after] - LB_SY_or_IS_then_various 4489 == LB_BREAKABLE; 4490 } 4491 4492 case LB_various_then_PO_or_PR + LB_BREAKABLE: 4493 case LB_various_then_PO_or_PR + LB_NOBREAK: 4494 { 4495 /* LB25e NU (SY | IS)* (CL | CP)? × (PO | PR) */ 4496 4497 LB_enum temp = prev; 4498 if (temp == LB_Close_Punctuation || temp == LB_Close_Parenthesis) 4499 { 4500 temp = backup_one_LB(strbeg, &temp_pos, utf8_target); 4501 } 4502 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric) { 4503 temp = backup_one_LB(strbeg, &temp_pos, utf8_target); 4504 } 4505 if (temp == LB_Numeric) { 4506 return FALSE; 4507 } 4508 return LB_various_then_PO_or_PR; 4509 } 4510 4511 default: 4512 break; 4513 } 4514 4515 #ifdef DEBUGGING 4516 Perl_re_printf( aTHX_ "Unhandled LB pair: LB_table[%d, %d] = %d\n", 4517 before, after, LB_table[before][after]); 4518 assert(0); 4519 #endif 4520 return TRUE; 4521 } 4522 4523 STATIC LB_enum 4524 S_advance_one_LB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target) 4525 { 4526 LB_enum lb; 4527 4528 PERL_ARGS_ASSERT_ADVANCE_ONE_LB; 4529 4530 if (*curpos >= strend) { 4531 return LB_EDGE; 4532 } 4533 4534 if (utf8_target) { 4535 *curpos += UTF8SKIP(*curpos); 4536 if (*curpos >= strend) { 4537 return LB_EDGE; 4538 } 4539 lb = getLB_VAL_UTF8(*curpos, strend); 4540 } 4541 else { 4542 (*curpos)++; 4543 if (*curpos >= strend) { 4544 return LB_EDGE; 4545 } 4546 lb = getLB_VAL_CP(**curpos); 4547 } 4548 4549 return lb; 4550 } 4551 4552 STATIC LB_enum 4553 S_backup_one_LB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target) 4554 { 4555 LB_enum lb; 4556 4557 PERL_ARGS_ASSERT_BACKUP_ONE_LB; 4558 4559 if (*curpos < strbeg) { 4560 return LB_EDGE; 4561 } 4562 4563 if (utf8_target) { 4564 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg); 4565 U8 * prev_prev_char_pos; 4566 4567 if (! prev_char_pos) { 4568 return LB_EDGE; 4569 } 4570 4571 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) { 4572 lb = getLB_VAL_UTF8(prev_prev_char_pos, prev_char_pos); 4573 *curpos = prev_char_pos; 4574 prev_char_pos = prev_prev_char_pos; 4575 } 4576 else { 4577 *curpos = (U8 *) strbeg; 4578 return LB_EDGE; 4579 } 4580 } 4581 else { 4582 if (*curpos - 2 < strbeg) { 4583 *curpos = (U8 *) strbeg; 4584 return LB_EDGE; 4585 } 4586 (*curpos)--; 4587 lb = getLB_VAL_CP(*(*curpos - 1)); 4588 } 4589 4590 return lb; 4591 } 4592 4593 STATIC bool 4594 S_isSB(pTHX_ SB_enum before, 4595 SB_enum after, 4596 const U8 * const strbeg, 4597 const U8 * const curpos, 4598 const U8 * const strend, 4599 const bool utf8_target) 4600 { 4601 /* returns a boolean indicating if there is a Sentence Boundary Break 4602 * between the inputs. See http://www.unicode.org/reports/tr29/ */ 4603 4604 U8 * lpos = (U8 *) curpos; 4605 bool has_para_sep = FALSE; 4606 bool has_sp = FALSE; 4607 4608 PERL_ARGS_ASSERT_ISSB; 4609 4610 /* Break at the start and end of text. 4611 SB1. sot ÷ 4612 SB2. ÷ eot 4613 But unstated in Unicode is don't break if the text is empty */ 4614 if (before == SB_EDGE || after == SB_EDGE) { 4615 return before != after; 4616 } 4617 4618 /* SB 3: Do not break within CRLF. */ 4619 if (before == SB_CR && after == SB_LF) { 4620 return FALSE; 4621 } 4622 4623 /* Break after paragraph separators. CR and LF are considered 4624 * so because Unicode views text as like word processing text where there 4625 * are no newlines except between paragraphs, and the word processor takes 4626 * care of wrapping without there being hard line-breaks in the text *./ 4627 SB4. Sep | CR | LF ÷ */ 4628 if (before == SB_Sep || before == SB_CR || before == SB_LF) { 4629 return TRUE; 4630 } 4631 4632 /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF. 4633 * (See Section 6.2, Replacing Ignore Rules.) 4634 SB5. X (Extend | Format)* → X */ 4635 if (after == SB_Extend || after == SB_Format) { 4636 4637 /* Implied is that the these characters attach to everything 4638 * immediately prior to them except for those separator-type 4639 * characters. And the rules earlier have already handled the case 4640 * when one of those immediately precedes the extend char */ 4641 return FALSE; 4642 } 4643 4644 if (before == SB_Extend || before == SB_Format) { 4645 U8 * temp_pos = lpos; 4646 const SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target); 4647 if ( backup != SB_EDGE 4648 && backup != SB_Sep 4649 && backup != SB_CR 4650 && backup != SB_LF) 4651 { 4652 before = backup; 4653 lpos = temp_pos; 4654 } 4655 4656 /* Here, both 'before' and 'backup' are these types; implied is that we 4657 * don't break between them */ 4658 if (backup == SB_Extend || backup == SB_Format) { 4659 return FALSE; 4660 } 4661 } 4662 4663 /* Do not break after ambiguous terminators like period, if they are 4664 * immediately followed by a number or lowercase letter, if they are 4665 * between uppercase letters, if the first following letter (optionally 4666 * after certain punctuation) is lowercase, or if they are followed by 4667 * "continuation" punctuation such as comma, colon, or semicolon. For 4668 * example, a period may be an abbreviation or numeric period, and thus may 4669 * not mark the end of a sentence. 4670 4671 * SB6. ATerm × Numeric */ 4672 if (before == SB_ATerm && after == SB_Numeric) { 4673 return FALSE; 4674 } 4675 4676 /* SB7. (Upper | Lower) ATerm × Upper */ 4677 if (before == SB_ATerm && after == SB_Upper) { 4678 U8 * temp_pos = lpos; 4679 SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target); 4680 if (backup == SB_Upper || backup == SB_Lower) { 4681 return FALSE; 4682 } 4683 } 4684 4685 /* The remaining rules that aren't the final one, all require an STerm or 4686 * an ATerm after having backed up over some Close* Sp*, and in one case an 4687 * optional Paragraph separator, although one rule doesn't have any Sp's in it. 4688 * So do that backup now, setting flags if either Sp or a paragraph 4689 * separator are found */ 4690 4691 if (before == SB_Sep || before == SB_CR || before == SB_LF) { 4692 has_para_sep = TRUE; 4693 before = backup_one_SB(strbeg, &lpos, utf8_target); 4694 } 4695 4696 if (before == SB_Sp) { 4697 has_sp = TRUE; 4698 do { 4699 before = backup_one_SB(strbeg, &lpos, utf8_target); 4700 } 4701 while (before == SB_Sp); 4702 } 4703 4704 while (before == SB_Close) { 4705 before = backup_one_SB(strbeg, &lpos, utf8_target); 4706 } 4707 4708 /* The next few rules apply only when the backed-up-to is an ATerm, and in 4709 * most cases an STerm */ 4710 if (before == SB_STerm || before == SB_ATerm) { 4711 4712 /* So, here the lhs matches 4713 * (STerm | ATerm) Close* Sp* (Sep | CR | LF)? 4714 * and we have set flags if we found an Sp, or the optional Sep,CR,LF. 4715 * The rules that apply here are: 4716 * 4717 * SB8 ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR 4718 | LF | STerm | ATerm) )* Lower 4719 SB8a (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm) 4720 SB9 (STerm | ATerm) Close* × (Close | Sp | Sep | CR | LF) 4721 SB10 (STerm | ATerm) Close* Sp* × (Sp | Sep | CR | LF) 4722 SB11 (STerm | ATerm) Close* Sp* (Sep | CR | LF)? ÷ 4723 */ 4724 4725 /* And all but SB11 forbid having seen a paragraph separator */ 4726 if (! has_para_sep) { 4727 if (before == SB_ATerm) { /* SB8 */ 4728 U8 * rpos = (U8 *) curpos; 4729 SB_enum later = after; 4730 4731 while ( later != SB_OLetter 4732 && later != SB_Upper 4733 && later != SB_Lower 4734 && later != SB_Sep 4735 && later != SB_CR 4736 && later != SB_LF 4737 && later != SB_STerm 4738 && later != SB_ATerm 4739 && later != SB_EDGE) 4740 { 4741 later = advance_one_SB(&rpos, strend, utf8_target); 4742 } 4743 if (later == SB_Lower) { 4744 return FALSE; 4745 } 4746 } 4747 4748 if ( after == SB_SContinue /* SB8a */ 4749 || after == SB_STerm 4750 || after == SB_ATerm) 4751 { 4752 return FALSE; 4753 } 4754 4755 if (! has_sp) { /* SB9 applies only if there was no Sp* */ 4756 if ( after == SB_Close 4757 || after == SB_Sp 4758 || after == SB_Sep 4759 || after == SB_CR 4760 || after == SB_LF) 4761 { 4762 return FALSE; 4763 } 4764 } 4765 4766 /* SB10. This and SB9 could probably be combined some way, but khw 4767 * has decided to follow the Unicode rule book precisely for 4768 * simplified maintenance */ 4769 if ( after == SB_Sp 4770 || after == SB_Sep 4771 || after == SB_CR 4772 || after == SB_LF) 4773 { 4774 return FALSE; 4775 } 4776 } 4777 4778 /* SB11. */ 4779 return TRUE; 4780 } 4781 4782 /* Otherwise, do not break. 4783 SB12. Any × Any */ 4784 4785 return FALSE; 4786 } 4787 4788 STATIC SB_enum 4789 S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target) 4790 { 4791 SB_enum sb; 4792 4793 PERL_ARGS_ASSERT_ADVANCE_ONE_SB; 4794 4795 if (*curpos >= strend) { 4796 return SB_EDGE; 4797 } 4798 4799 if (utf8_target) { 4800 do { 4801 *curpos += UTF8SKIP(*curpos); 4802 if (*curpos >= strend) { 4803 return SB_EDGE; 4804 } 4805 sb = getSB_VAL_UTF8(*curpos, strend); 4806 } while (sb == SB_Extend || sb == SB_Format); 4807 } 4808 else { 4809 do { 4810 (*curpos)++; 4811 if (*curpos >= strend) { 4812 return SB_EDGE; 4813 } 4814 sb = getSB_VAL_CP(**curpos); 4815 } while (sb == SB_Extend || sb == SB_Format); 4816 } 4817 4818 return sb; 4819 } 4820 4821 STATIC SB_enum 4822 S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target) 4823 { 4824 SB_enum sb; 4825 4826 PERL_ARGS_ASSERT_BACKUP_ONE_SB; 4827 4828 if (*curpos < strbeg) { 4829 return SB_EDGE; 4830 } 4831 4832 if (utf8_target) { 4833 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg); 4834 if (! prev_char_pos) { 4835 return SB_EDGE; 4836 } 4837 4838 /* Back up over Extend and Format. curpos is always just to the right 4839 * of the characater whose value we are getting */ 4840 do { 4841 U8 * prev_prev_char_pos; 4842 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, 4843 strbeg))) 4844 { 4845 sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos); 4846 *curpos = prev_char_pos; 4847 prev_char_pos = prev_prev_char_pos; 4848 } 4849 else { 4850 *curpos = (U8 *) strbeg; 4851 return SB_EDGE; 4852 } 4853 } while (sb == SB_Extend || sb == SB_Format); 4854 } 4855 else { 4856 do { 4857 if (*curpos - 2 < strbeg) { 4858 *curpos = (U8 *) strbeg; 4859 return SB_EDGE; 4860 } 4861 (*curpos)--; 4862 sb = getSB_VAL_CP(*(*curpos - 1)); 4863 } while (sb == SB_Extend || sb == SB_Format); 4864 } 4865 4866 return sb; 4867 } 4868 4869 STATIC bool 4870 S_isWB(pTHX_ WB_enum previous, 4871 WB_enum before, 4872 WB_enum after, 4873 const U8 * const strbeg, 4874 const U8 * const curpos, 4875 const U8 * const strend, 4876 const bool utf8_target) 4877 { 4878 /* Return a boolean as to if the boundary between 'before' and 'after' is 4879 * a Unicode word break, using their published algorithm, but tailored for 4880 * Perl by treating spans of white space as one unit. Context may be 4881 * needed to make this determination. If the value for the character 4882 * before 'before' is known, it is passed as 'previous'; otherwise that 4883 * should be set to WB_UNKNOWN. The other input parameters give the 4884 * boundaries and current position in the matching of the string. That 4885 * is, 'curpos' marks the position where the character whose wb value is 4886 * 'after' begins. See http://www.unicode.org/reports/tr29/ */ 4887 4888 U8 * before_pos = (U8 *) curpos; 4889 U8 * after_pos = (U8 *) curpos; 4890 WB_enum prev = before; 4891 WB_enum next; 4892 4893 PERL_ARGS_ASSERT_ISWB; 4894 4895 /* Rule numbers in the comments below are as of Unicode 8.0 */ 4896 4897 redo: 4898 before = prev; 4899 switch (WB_table[before][after]) { 4900 case WB_BREAKABLE: 4901 return TRUE; 4902 4903 case WB_NOBREAK: 4904 return FALSE; 4905 4906 case WB_hs_then_hs: /* 2 horizontal spaces in a row */ 4907 next = advance_one_WB(&after_pos, strend, utf8_target, 4908 FALSE /* Don't skip Extend nor Format */ ); 4909 /* A space immediately preceeding an Extend or Format is attached 4910 * to by them, and hence gets separated from previous spaces. 4911 * Otherwise don't break between horizontal white space */ 4912 return next == WB_Extend || next == WB_Format; 4913 4914 /* WB4 Ignore Format and Extend characters, except when they appear at 4915 * the beginning of a region of text. This code currently isn't 4916 * general purpose, but it works as the rules are currently and likely 4917 * to be laid out. The reason it works is that when 'they appear at 4918 * the beginning of a region of text', the rule is to break before 4919 * them, just like any other character. Therefore, the default rule 4920 * applies and we don't have to look in more depth. Should this ever 4921 * change, we would have to have 2 'case' statements, like in the 4922 * rules below, and backup a single character (not spacing over the 4923 * extend ones) and then see if that is one of the region-end 4924 * characters and go from there */ 4925 case WB_Ex_or_FO_then_foo: 4926 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target); 4927 goto redo; 4928 4929 case WB_DQ_then_HL + WB_BREAKABLE: 4930 case WB_DQ_then_HL + WB_NOBREAK: 4931 4932 /* WB7c Hebrew_Letter Double_Quote × Hebrew_Letter */ 4933 4934 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target) 4935 == WB_Hebrew_Letter) 4936 { 4937 return FALSE; 4938 } 4939 4940 return WB_table[before][after] - WB_DQ_then_HL == WB_BREAKABLE; 4941 4942 case WB_HL_then_DQ + WB_BREAKABLE: 4943 case WB_HL_then_DQ + WB_NOBREAK: 4944 4945 /* WB7b Hebrew_Letter × Double_Quote Hebrew_Letter */ 4946 4947 if (advance_one_WB(&after_pos, strend, utf8_target, 4948 TRUE /* Do skip Extend and Format */ ) 4949 == WB_Hebrew_Letter) 4950 { 4951 return FALSE; 4952 } 4953 4954 return WB_table[before][after] - WB_HL_then_DQ == WB_BREAKABLE; 4955 4956 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_NOBREAK: 4957 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_BREAKABLE: 4958 4959 /* WB6 (ALetter | Hebrew_Letter) × (MidLetter | MidNumLet 4960 * | Single_Quote) (ALetter | Hebrew_Letter) */ 4961 4962 next = advance_one_WB(&after_pos, strend, utf8_target, 4963 TRUE /* Do skip Extend and Format */ ); 4964 4965 if (next == WB_ALetter || next == WB_Hebrew_Letter) 4966 { 4967 return FALSE; 4968 } 4969 4970 return WB_table[before][after] 4971 - WB_LE_or_HL_then_MB_or_ML_or_SQ == WB_BREAKABLE; 4972 4973 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_NOBREAK: 4974 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_BREAKABLE: 4975 4976 /* WB7 (ALetter | Hebrew_Letter) (MidLetter | MidNumLet 4977 * | Single_Quote) × (ALetter | Hebrew_Letter) */ 4978 4979 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target); 4980 if (prev == WB_ALetter || prev == WB_Hebrew_Letter) 4981 { 4982 return FALSE; 4983 } 4984 4985 return WB_table[before][after] 4986 - WB_MB_or_ML_or_SQ_then_LE_or_HL == WB_BREAKABLE; 4987 4988 case WB_MB_or_MN_or_SQ_then_NU + WB_NOBREAK: 4989 case WB_MB_or_MN_or_SQ_then_NU + WB_BREAKABLE: 4990 4991 /* WB11 Numeric (MidNum | (MidNumLet | Single_Quote)) × Numeric 4992 * */ 4993 4994 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target) 4995 == WB_Numeric) 4996 { 4997 return FALSE; 4998 } 4999 5000 return WB_table[before][after] 5001 - WB_MB_or_MN_or_SQ_then_NU == WB_BREAKABLE; 5002 5003 case WB_NU_then_MB_or_MN_or_SQ + WB_NOBREAK: 5004 case WB_NU_then_MB_or_MN_or_SQ + WB_BREAKABLE: 5005 5006 /* WB12 Numeric × (MidNum | MidNumLet | Single_Quote) Numeric */ 5007 5008 if (advance_one_WB(&after_pos, strend, utf8_target, 5009 TRUE /* Do skip Extend and Format */ ) 5010 == WB_Numeric) 5011 { 5012 return FALSE; 5013 } 5014 5015 return WB_table[before][after] 5016 - WB_NU_then_MB_or_MN_or_SQ == WB_BREAKABLE; 5017 5018 default: 5019 break; 5020 } 5021 5022 #ifdef DEBUGGING 5023 Perl_re_printf( aTHX_ "Unhandled WB pair: WB_table[%d, %d] = %d\n", 5024 before, after, WB_table[before][after]); 5025 assert(0); 5026 #endif 5027 return TRUE; 5028 } 5029 5030 STATIC WB_enum 5031 S_advance_one_WB(pTHX_ U8 ** curpos, 5032 const U8 * const strend, 5033 const bool utf8_target, 5034 const bool skip_Extend_Format) 5035 { 5036 WB_enum wb; 5037 5038 PERL_ARGS_ASSERT_ADVANCE_ONE_WB; 5039 5040 if (*curpos >= strend) { 5041 return WB_EDGE; 5042 } 5043 5044 if (utf8_target) { 5045 5046 /* Advance over Extend and Format */ 5047 do { 5048 *curpos += UTF8SKIP(*curpos); 5049 if (*curpos >= strend) { 5050 return WB_EDGE; 5051 } 5052 wb = getWB_VAL_UTF8(*curpos, strend); 5053 } while ( skip_Extend_Format 5054 && (wb == WB_Extend || wb == WB_Format)); 5055 } 5056 else { 5057 do { 5058 (*curpos)++; 5059 if (*curpos >= strend) { 5060 return WB_EDGE; 5061 } 5062 wb = getWB_VAL_CP(**curpos); 5063 } while ( skip_Extend_Format 5064 && (wb == WB_Extend || wb == WB_Format)); 5065 } 5066 5067 return wb; 5068 } 5069 5070 STATIC WB_enum 5071 S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target) 5072 { 5073 WB_enum wb; 5074 5075 PERL_ARGS_ASSERT_BACKUP_ONE_WB; 5076 5077 /* If we know what the previous character's break value is, don't have 5078 * to look it up */ 5079 if (*previous != WB_UNKNOWN) { 5080 wb = *previous; 5081 5082 /* But we need to move backwards by one */ 5083 if (utf8_target) { 5084 *curpos = reghopmaybe3(*curpos, -1, strbeg); 5085 if (! *curpos) { 5086 *previous = WB_EDGE; 5087 *curpos = (U8 *) strbeg; 5088 } 5089 else { 5090 *previous = WB_UNKNOWN; 5091 } 5092 } 5093 else { 5094 (*curpos)--; 5095 *previous = (*curpos <= strbeg) ? WB_EDGE : WB_UNKNOWN; 5096 } 5097 5098 /* And we always back up over these two types */ 5099 if (wb != WB_Extend && wb != WB_Format) { 5100 return wb; 5101 } 5102 } 5103 5104 if (*curpos < strbeg) { 5105 return WB_EDGE; 5106 } 5107 5108 if (utf8_target) { 5109 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg); 5110 if (! prev_char_pos) { 5111 return WB_EDGE; 5112 } 5113 5114 /* Back up over Extend and Format. curpos is always just to the right 5115 * of the characater whose value we are getting */ 5116 do { 5117 U8 * prev_prev_char_pos; 5118 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, 5119 -1, 5120 strbeg))) 5121 { 5122 wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos); 5123 *curpos = prev_char_pos; 5124 prev_char_pos = prev_prev_char_pos; 5125 } 5126 else { 5127 *curpos = (U8 *) strbeg; 5128 return WB_EDGE; 5129 } 5130 } while (wb == WB_Extend || wb == WB_Format); 5131 } 5132 else { 5133 do { 5134 if (*curpos - 2 < strbeg) { 5135 *curpos = (U8 *) strbeg; 5136 return WB_EDGE; 5137 } 5138 (*curpos)--; 5139 wb = getWB_VAL_CP(*(*curpos - 1)); 5140 } while (wb == WB_Extend || wb == WB_Format); 5141 } 5142 5143 return wb; 5144 } 5145 5146 #define EVAL_CLOSE_PAREN_IS(st,expr) \ 5147 ( \ 5148 ( ( st ) ) && \ 5149 ( ( st )->u.eval.close_paren ) && \ 5150 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \ 5151 ) 5152 5153 #define EVAL_CLOSE_PAREN_IS_TRUE(st,expr) \ 5154 ( \ 5155 ( ( st ) ) && \ 5156 ( ( st )->u.eval.close_paren ) && \ 5157 ( ( expr ) ) && \ 5158 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \ 5159 ) 5160 5161 5162 #define EVAL_CLOSE_PAREN_SET(st,expr) \ 5163 (st)->u.eval.close_paren = ( (expr) + 1 ) 5164 5165 #define EVAL_CLOSE_PAREN_CLEAR(st) \ 5166 (st)->u.eval.close_paren = 0 5167 5168 /* returns -1 on failure, $+[0] on success */ 5169 STATIC SSize_t 5170 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog) 5171 { 5172 5173 #if PERL_VERSION < 9 && !defined(PERL_CORE) 5174 dMY_CXT; 5175 #endif 5176 dVAR; 5177 const bool utf8_target = reginfo->is_utf8_target; 5178 const U32 uniflags = UTF8_ALLOW_DEFAULT; 5179 REGEXP *rex_sv = reginfo->prog; 5180 regexp *rex = ReANY(rex_sv); 5181 RXi_GET_DECL(rex,rexi); 5182 /* the current state. This is a cached copy of PL_regmatch_state */ 5183 regmatch_state *st; 5184 /* cache heavy used fields of st in registers */ 5185 regnode *scan; 5186 regnode *next; 5187 U32 n = 0; /* general value; init to avoid compiler warning */ 5188 SSize_t ln = 0; /* len or last; init to avoid compiler warning */ 5189 char *locinput = startpos; 5190 char *pushinput; /* where to continue after a PUSH */ 5191 I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */ 5192 5193 bool result = 0; /* return value of S_regmatch */ 5194 int depth = 0; /* depth of backtrack stack */ 5195 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */ 5196 const U32 max_nochange_depth = 5197 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ? 5198 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH; 5199 regmatch_state *yes_state = NULL; /* state to pop to on success of 5200 subpattern */ 5201 /* mark_state piggy backs on the yes_state logic so that when we unwind 5202 the stack on success we can update the mark_state as we go */ 5203 regmatch_state *mark_state = NULL; /* last mark state we have seen */ 5204 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ 5205 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ 5206 U32 state_num; 5207 bool no_final = 0; /* prevent failure from backtracking? */ 5208 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */ 5209 char *startpoint = locinput; 5210 SV *popmark = NULL; /* are we looking for a mark? */ 5211 SV *sv_commit = NULL; /* last mark name seen in failure */ 5212 SV *sv_yes_mark = NULL; /* last mark name we have seen 5213 during a successful match */ 5214 U32 lastopen = 0; /* last open we saw */ 5215 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0; 5216 SV* const oreplsv = GvSVn(PL_replgv); 5217 /* these three flags are set by various ops to signal information to 5218 * the very next op. They have a useful lifetime of exactly one loop 5219 * iteration, and are not preserved or restored by state pushes/pops 5220 */ 5221 bool sw = 0; /* the condition value in (?(cond)a|b) */ 5222 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ 5223 int logical = 0; /* the following EVAL is: 5224 0: (?{...}) 5225 1: (?(?{...})X|Y) 5226 2: (??{...}) 5227 or the following IFMATCH/UNLESSM is: 5228 false: plain (?=foo) 5229 true: used as a condition: (?(?=foo)) 5230 */ 5231 PAD* last_pad = NULL; 5232 dMULTICALL; 5233 U8 gimme = G_SCALAR; 5234 CV *caller_cv = NULL; /* who called us */ 5235 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */ 5236 CHECKPOINT runops_cp; /* savestack position before executing EVAL */ 5237 U32 maxopenparen = 0; /* max '(' index seen so far */ 5238 int to_complement; /* Invert the result? */ 5239 _char_class_number classnum; 5240 bool is_utf8_pat = reginfo->is_utf8_pat; 5241 bool match = FALSE; 5242 5243 /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */ 5244 #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL)) 5245 # define SOLARIS_BAD_OPTIMIZER 5246 const U32 *pl_charclass_dup = PL_charclass; 5247 # define PL_charclass pl_charclass_dup 5248 #endif 5249 5250 #ifdef DEBUGGING 5251 GET_RE_DEBUG_FLAGS_DECL; 5252 #endif 5253 5254 /* protect against undef(*^R) */ 5255 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv)); 5256 5257 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */ 5258 multicall_oldcatch = 0; 5259 PERL_UNUSED_VAR(multicall_cop); 5260 5261 PERL_ARGS_ASSERT_REGMATCH; 5262 5263 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({ 5264 Perl_re_printf( aTHX_ "regmatch start\n"); 5265 })); 5266 5267 st = PL_regmatch_state; 5268 5269 /* Note that nextchr is a byte even in UTF */ 5270 SET_nextchr; 5271 scan = prog; 5272 while (scan != NULL) { 5273 5274 5275 next = scan + NEXT_OFF(scan); 5276 if (next == scan) 5277 next = NULL; 5278 state_num = OP(scan); 5279 5280 reenter_switch: 5281 DEBUG_EXECUTE_r( 5282 if (state_num <= REGNODE_MAX) { 5283 SV * const prop = sv_newmortal(); 5284 regnode *rnext = regnext(scan); 5285 5286 DUMP_EXEC_POS( locinput, scan, utf8_target, depth ); 5287 regprop(rex, prop, scan, reginfo, NULL); 5288 Perl_re_printf( aTHX_ 5289 "%*s%"IVdf":%s(%"IVdf")\n", 5290 INDENT_CHARS(depth), "", 5291 (IV)(scan - rexi->program), 5292 SvPVX_const(prop), 5293 (PL_regkind[OP(scan)] == END || !rnext) ? 5294 0 : (IV)(rnext - rexi->program)); 5295 } 5296 ); 5297 5298 to_complement = 0; 5299 5300 SET_nextchr; 5301 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS)); 5302 5303 switch (state_num) { 5304 case SBOL: /* /^../ and /\A../ */ 5305 if (locinput == reginfo->strbeg) 5306 break; 5307 sayNO; 5308 5309 case MBOL: /* /^../m */ 5310 if (locinput == reginfo->strbeg || 5311 (!NEXTCHR_IS_EOS && locinput[-1] == '\n')) 5312 { 5313 break; 5314 } 5315 sayNO; 5316 5317 case GPOS: /* \G */ 5318 if (locinput == reginfo->ganch) 5319 break; 5320 sayNO; 5321 5322 case KEEPS: /* \K */ 5323 /* update the startpoint */ 5324 st->u.keeper.val = rex->offs[0].start; 5325 rex->offs[0].start = locinput - reginfo->strbeg; 5326 PUSH_STATE_GOTO(KEEPS_next, next, locinput); 5327 NOT_REACHED; /* NOTREACHED */ 5328 5329 case KEEPS_next_fail: 5330 /* rollback the start point change */ 5331 rex->offs[0].start = st->u.keeper.val; 5332 sayNO_SILENT; 5333 NOT_REACHED; /* NOTREACHED */ 5334 5335 case MEOL: /* /..$/m */ 5336 if (!NEXTCHR_IS_EOS && nextchr != '\n') 5337 sayNO; 5338 break; 5339 5340 case SEOL: /* /..$/ */ 5341 if (!NEXTCHR_IS_EOS && nextchr != '\n') 5342 sayNO; 5343 if (reginfo->strend - locinput > 1) 5344 sayNO; 5345 break; 5346 5347 case EOS: /* \z */ 5348 if (!NEXTCHR_IS_EOS) 5349 sayNO; 5350 break; 5351 5352 case SANY: /* /./s */ 5353 if (NEXTCHR_IS_EOS) 5354 sayNO; 5355 goto increment_locinput; 5356 5357 case REG_ANY: /* /./ */ 5358 if ((NEXTCHR_IS_EOS) || nextchr == '\n') 5359 sayNO; 5360 goto increment_locinput; 5361 5362 5363 #undef ST 5364 #define ST st->u.trie 5365 case TRIEC: /* (ab|cd) with known charclass */ 5366 /* In this case the charclass data is available inline so 5367 we can fail fast without a lot of extra overhead. 5368 */ 5369 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) { 5370 DEBUG_EXECUTE_r( 5371 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n", 5372 depth, PL_colors[4], PL_colors[5]) 5373 ); 5374 sayNO_SILENT; 5375 NOT_REACHED; /* NOTREACHED */ 5376 } 5377 /* FALLTHROUGH */ 5378 case TRIE: /* (ab|cd) */ 5379 /* the basic plan of execution of the trie is: 5380 * At the beginning, run though all the states, and 5381 * find the longest-matching word. Also remember the position 5382 * of the shortest matching word. For example, this pattern: 5383 * 1 2 3 4 5 5384 * ab|a|x|abcd|abc 5385 * when matched against the string "abcde", will generate 5386 * accept states for all words except 3, with the longest 5387 * matching word being 4, and the shortest being 2 (with 5388 * the position being after char 1 of the string). 5389 * 5390 * Then for each matching word, in word order (i.e. 1,2,4,5), 5391 * we run the remainder of the pattern; on each try setting 5392 * the current position to the character following the word, 5393 * returning to try the next word on failure. 5394 * 5395 * We avoid having to build a list of words at runtime by 5396 * using a compile-time structure, wordinfo[].prev, which 5397 * gives, for each word, the previous accepting word (if any). 5398 * In the case above it would contain the mappings 1->2, 2->0, 5399 * 3->0, 4->5, 5->1. We can use this table to generate, from 5400 * the longest word (4 above), a list of all words, by 5401 * following the list of prev pointers; this gives us the 5402 * unordered list 4,5,1,2. Then given the current word we have 5403 * just tried, we can go through the list and find the 5404 * next-biggest word to try (so if we just failed on word 2, 5405 * the next in the list is 4). 5406 * 5407 * Since at runtime we don't record the matching position in 5408 * the string for each word, we have to work that out for 5409 * each word we're about to process. The wordinfo table holds 5410 * the character length of each word; given that we recorded 5411 * at the start: the position of the shortest word and its 5412 * length in chars, we just need to move the pointer the 5413 * difference between the two char lengths. Depending on 5414 * Unicode status and folding, that's cheap or expensive. 5415 * 5416 * This algorithm is optimised for the case where are only a 5417 * small number of accept states, i.e. 0,1, or maybe 2. 5418 * With lots of accepts states, and having to try all of them, 5419 * it becomes quadratic on number of accept states to find all 5420 * the next words. 5421 */ 5422 5423 { 5424 /* what type of TRIE am I? (utf8 makes this contextual) */ 5425 DECL_TRIE_TYPE(scan); 5426 5427 /* what trie are we using right now */ 5428 reg_trie_data * const trie 5429 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ]; 5430 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]); 5431 U32 state = trie->startstate; 5432 5433 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) { 5434 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 5435 if (utf8_target 5436 && UTF8_IS_ABOVE_LATIN1(nextchr) 5437 && scan->flags == EXACTL) 5438 { 5439 /* We only output for EXACTL, as we let the folder 5440 * output this message for EXACTFLU8 to avoid 5441 * duplication */ 5442 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, 5443 reginfo->strend); 5444 } 5445 } 5446 if ( trie->bitmap 5447 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr))) 5448 { 5449 if (trie->states[ state ].wordnum) { 5450 DEBUG_EXECUTE_r( 5451 Perl_re_exec_indentf( aTHX_ "%smatched empty string...%s\n", 5452 depth, PL_colors[4], PL_colors[5]) 5453 ); 5454 if (!trie->jump) 5455 break; 5456 } else { 5457 DEBUG_EXECUTE_r( 5458 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n", 5459 depth, PL_colors[4], PL_colors[5]) 5460 ); 5461 sayNO_SILENT; 5462 } 5463 } 5464 5465 { 5466 U8 *uc = ( U8* )locinput; 5467 5468 STRLEN len = 0; 5469 STRLEN foldlen = 0; 5470 U8 *uscan = (U8*)NULL; 5471 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; 5472 U32 charcount = 0; /* how many input chars we have matched */ 5473 U32 accepted = 0; /* have we seen any accepting states? */ 5474 5475 ST.jump = trie->jump; 5476 ST.me = scan; 5477 ST.firstpos = NULL; 5478 ST.longfold = FALSE; /* char longer if folded => it's harder */ 5479 ST.nextword = 0; 5480 5481 /* fully traverse the TRIE; note the position of the 5482 shortest accept state and the wordnum of the longest 5483 accept state */ 5484 5485 while ( state && uc <= (U8*)(reginfo->strend) ) { 5486 U32 base = trie->states[ state ].trans.base; 5487 UV uvc = 0; 5488 U16 charid = 0; 5489 U16 wordnum; 5490 wordnum = trie->states[ state ].wordnum; 5491 5492 if (wordnum) { /* it's an accept state */ 5493 if (!accepted) { 5494 accepted = 1; 5495 /* record first match position */ 5496 if (ST.longfold) { 5497 ST.firstpos = (U8*)locinput; 5498 ST.firstchars = 0; 5499 } 5500 else { 5501 ST.firstpos = uc; 5502 ST.firstchars = charcount; 5503 } 5504 } 5505 if (!ST.nextword || wordnum < ST.nextword) 5506 ST.nextword = wordnum; 5507 ST.topword = wordnum; 5508 } 5509 5510 DEBUG_TRIE_EXECUTE_r({ 5511 DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth ); 5512 Perl_re_exec_indentf( aTHX_ 5513 "%sState: %4"UVxf" Accepted: %c ", 5514 depth, PL_colors[4], 5515 (UV)state, (accepted ? 'Y' : 'N')); 5516 }); 5517 5518 /* read a char and goto next state */ 5519 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) { 5520 I32 offset; 5521 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, 5522 (U8 *) reginfo->strend, uscan, 5523 len, uvc, charid, foldlen, 5524 foldbuf, uniflags); 5525 charcount++; 5526 if (foldlen>0) 5527 ST.longfold = TRUE; 5528 if (charid && 5529 ( ((offset = 5530 base + charid - 1 - trie->uniquecharcount)) >= 0) 5531 5532 && ((U32)offset < trie->lasttrans) 5533 && trie->trans[offset].check == state) 5534 { 5535 state = trie->trans[offset].next; 5536 } 5537 else { 5538 state = 0; 5539 } 5540 uc += len; 5541 5542 } 5543 else { 5544 state = 0; 5545 } 5546 DEBUG_TRIE_EXECUTE_r( 5547 Perl_re_printf( aTHX_ 5548 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", 5549 charid, uvc, (UV)state, PL_colors[5] ); 5550 ); 5551 } 5552 if (!accepted) 5553 sayNO; 5554 5555 /* calculate total number of accept states */ 5556 { 5557 U16 w = ST.topword; 5558 accepted = 0; 5559 while (w) { 5560 w = trie->wordinfo[w].prev; 5561 accepted++; 5562 } 5563 ST.accepted = accepted; 5564 } 5565 5566 DEBUG_EXECUTE_r( 5567 Perl_re_exec_indentf( aTHX_ "%sgot %"IVdf" possible matches%s\n", 5568 depth, 5569 PL_colors[4], (IV)ST.accepted, PL_colors[5] ); 5570 ); 5571 goto trie_first_try; /* jump into the fail handler */ 5572 }} 5573 NOT_REACHED; /* NOTREACHED */ 5574 5575 case TRIE_next_fail: /* we failed - try next alternative */ 5576 { 5577 U8 *uc; 5578 if ( ST.jump) { 5579 REGCP_UNWIND(ST.cp); 5580 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 5581 } 5582 if (!--ST.accepted) { 5583 DEBUG_EXECUTE_r({ 5584 Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n", 5585 depth, 5586 PL_colors[4], 5587 PL_colors[5] ); 5588 }); 5589 sayNO_SILENT; 5590 } 5591 { 5592 /* Find next-highest word to process. Note that this code 5593 * is O(N^2) per trie run (O(N) per branch), so keep tight */ 5594 U16 min = 0; 5595 U16 word; 5596 U16 const nextword = ST.nextword; 5597 reg_trie_wordinfo * const wordinfo 5598 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo; 5599 for (word=ST.topword; word; word=wordinfo[word].prev) { 5600 if (word > nextword && (!min || word < min)) 5601 min = word; 5602 } 5603 ST.nextword = min; 5604 } 5605 5606 trie_first_try: 5607 if (do_cutgroup) { 5608 do_cutgroup = 0; 5609 no_final = 0; 5610 } 5611 5612 if ( ST.jump) { 5613 ST.lastparen = rex->lastparen; 5614 ST.lastcloseparen = rex->lastcloseparen; 5615 REGCP_SET(ST.cp); 5616 } 5617 5618 /* find start char of end of current word */ 5619 { 5620 U32 chars; /* how many chars to skip */ 5621 reg_trie_data * const trie 5622 = (reg_trie_data*)rexi->data->data[ARG(ST.me)]; 5623 5624 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen) 5625 >= ST.firstchars); 5626 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen) 5627 - ST.firstchars; 5628 uc = ST.firstpos; 5629 5630 if (ST.longfold) { 5631 /* the hard option - fold each char in turn and find 5632 * its folded length (which may be different */ 5633 U8 foldbuf[UTF8_MAXBYTES_CASE + 1]; 5634 STRLEN foldlen; 5635 STRLEN len; 5636 UV uvc; 5637 U8 *uscan; 5638 5639 while (chars) { 5640 if (utf8_target) { 5641 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len, 5642 uniflags); 5643 uc += len; 5644 } 5645 else { 5646 uvc = *uc; 5647 uc++; 5648 } 5649 uvc = to_uni_fold(uvc, foldbuf, &foldlen); 5650 uscan = foldbuf; 5651 while (foldlen) { 5652 if (!--chars) 5653 break; 5654 uvc = utf8n_to_uvchr(uscan, foldlen, &len, 5655 uniflags); 5656 uscan += len; 5657 foldlen -= len; 5658 } 5659 } 5660 } 5661 else { 5662 if (utf8_target) 5663 while (chars--) 5664 uc += UTF8SKIP(uc); 5665 else 5666 uc += chars; 5667 } 5668 } 5669 5670 scan = ST.me + ((ST.jump && ST.jump[ST.nextword]) 5671 ? ST.jump[ST.nextword] 5672 : NEXT_OFF(ST.me)); 5673 5674 DEBUG_EXECUTE_r({ 5675 Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n", 5676 depth, 5677 PL_colors[4], 5678 ST.nextword, 5679 PL_colors[5] 5680 ); 5681 }); 5682 5683 if (ST.accepted > 1 || has_cutgroup) { 5684 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc); 5685 NOT_REACHED; /* NOTREACHED */ 5686 } 5687 /* only one choice left - just continue */ 5688 DEBUG_EXECUTE_r({ 5689 AV *const trie_words 5690 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]); 5691 SV ** const tmp = trie_words 5692 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL; 5693 SV *sv= tmp ? sv_newmortal() : NULL; 5694 5695 Perl_re_exec_indentf( aTHX_ "%sonly one match left, short-circuiting: #%d <%s>%s\n", 5696 depth, PL_colors[4], 5697 ST.nextword, 5698 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0, 5699 PL_colors[0], PL_colors[1], 5700 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII 5701 ) 5702 : "not compiled under -Dr", 5703 PL_colors[5] ); 5704 }); 5705 5706 locinput = (char*)uc; 5707 continue; /* execute rest of RE */ 5708 /* NOTREACHED */ 5709 } 5710 #undef ST 5711 5712 case EXACTL: /* /abc/l */ 5713 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 5714 5715 /* Complete checking would involve going through every character 5716 * matched by the string to see if any is above latin1. But the 5717 * comparision otherwise might very well be a fast assembly 5718 * language routine, and I (khw) don't think slowing things down 5719 * just to check for this warning is worth it. So this just checks 5720 * the first character */ 5721 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) { 5722 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend); 5723 } 5724 /* FALLTHROUGH */ 5725 case EXACT: { /* /abc/ */ 5726 char *s = STRING(scan); 5727 ln = STR_LEN(scan); 5728 if (utf8_target != is_utf8_pat) { 5729 /* The target and the pattern have differing utf8ness. */ 5730 char *l = locinput; 5731 const char * const e = s + ln; 5732 5733 if (utf8_target) { 5734 /* The target is utf8, the pattern is not utf8. 5735 * Above-Latin1 code points can't match the pattern; 5736 * invariants match exactly, and the other Latin1 ones need 5737 * to be downgraded to a single byte in order to do the 5738 * comparison. (If we could be confident that the target 5739 * is not malformed, this could be refactored to have fewer 5740 * tests by just assuming that if the first bytes match, it 5741 * is an invariant, but there are tests in the test suite 5742 * dealing with (??{...}) which violate this) */ 5743 while (s < e) { 5744 if (l >= reginfo->strend 5745 || UTF8_IS_ABOVE_LATIN1(* (U8*) l)) 5746 { 5747 sayNO; 5748 } 5749 if (UTF8_IS_INVARIANT(*(U8*)l)) { 5750 if (*l != *s) { 5751 sayNO; 5752 } 5753 l++; 5754 } 5755 else { 5756 if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s) 5757 { 5758 sayNO; 5759 } 5760 l += 2; 5761 } 5762 s++; 5763 } 5764 } 5765 else { 5766 /* The target is not utf8, the pattern is utf8. */ 5767 while (s < e) { 5768 if (l >= reginfo->strend 5769 || UTF8_IS_ABOVE_LATIN1(* (U8*) s)) 5770 { 5771 sayNO; 5772 } 5773 if (UTF8_IS_INVARIANT(*(U8*)s)) { 5774 if (*s != *l) { 5775 sayNO; 5776 } 5777 s++; 5778 } 5779 else { 5780 if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l) 5781 { 5782 sayNO; 5783 } 5784 s += 2; 5785 } 5786 l++; 5787 } 5788 } 5789 locinput = l; 5790 } 5791 else { 5792 /* The target and the pattern have the same utf8ness. */ 5793 /* Inline the first character, for speed. */ 5794 if (reginfo->strend - locinput < ln 5795 || UCHARAT(s) != nextchr 5796 || (ln > 1 && memNE(s, locinput, ln))) 5797 { 5798 sayNO; 5799 } 5800 locinput += ln; 5801 } 5802 break; 5803 } 5804 5805 case EXACTFL: { /* /abc/il */ 5806 re_fold_t folder; 5807 const U8 * fold_array; 5808 const char * s; 5809 U32 fold_utf8_flags; 5810 5811 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 5812 folder = foldEQ_locale; 5813 fold_array = PL_fold_locale; 5814 fold_utf8_flags = FOLDEQ_LOCALE; 5815 goto do_exactf; 5816 5817 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so 5818 is effectively /u; hence to match, target 5819 must be UTF-8. */ 5820 if (! utf8_target) { 5821 sayNO; 5822 } 5823 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S1_ALREADY_FOLDED 5824 | FOLDEQ_S1_FOLDS_SANE; 5825 folder = foldEQ_latin1; 5826 fold_array = PL_fold_latin1; 5827 goto do_exactf; 5828 5829 case EXACTFU_SS: /* /\x{df}/iu */ 5830 case EXACTFU: /* /abc/iu */ 5831 folder = foldEQ_latin1; 5832 fold_array = PL_fold_latin1; 5833 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0; 5834 goto do_exactf; 5835 5836 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 5837 patterns */ 5838 assert(! is_utf8_pat); 5839 /* FALLTHROUGH */ 5840 case EXACTFA: /* /abc/iaa */ 5841 folder = foldEQ_latin1; 5842 fold_array = PL_fold_latin1; 5843 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; 5844 goto do_exactf; 5845 5846 case EXACTF: /* /abc/i This node only generated for 5847 non-utf8 patterns */ 5848 assert(! is_utf8_pat); 5849 folder = foldEQ; 5850 fold_array = PL_fold; 5851 fold_utf8_flags = 0; 5852 5853 do_exactf: 5854 s = STRING(scan); 5855 ln = STR_LEN(scan); 5856 5857 if (utf8_target 5858 || is_utf8_pat 5859 || state_num == EXACTFU_SS 5860 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE)) 5861 { 5862 /* Either target or the pattern are utf8, or has the issue where 5863 * the fold lengths may differ. */ 5864 const char * const l = locinput; 5865 char *e = reginfo->strend; 5866 5867 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat, 5868 l, &e, 0, utf8_target, fold_utf8_flags)) 5869 { 5870 sayNO; 5871 } 5872 locinput = e; 5873 break; 5874 } 5875 5876 /* Neither the target nor the pattern are utf8 */ 5877 if (UCHARAT(s) != nextchr 5878 && !NEXTCHR_IS_EOS 5879 && UCHARAT(s) != fold_array[nextchr]) 5880 { 5881 sayNO; 5882 } 5883 if (reginfo->strend - locinput < ln) 5884 sayNO; 5885 if (ln > 1 && ! folder(s, locinput, ln)) 5886 sayNO; 5887 locinput += ln; 5888 break; 5889 } 5890 5891 case NBOUNDL: /* /\B/l */ 5892 to_complement = 1; 5893 /* FALLTHROUGH */ 5894 5895 case BOUNDL: /* /\b/l */ 5896 { 5897 bool b1, b2; 5898 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 5899 5900 if (FLAGS(scan) != TRADITIONAL_BOUND) { 5901 if (! IN_UTF8_CTYPE_LOCALE) { 5902 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 5903 B_ON_NON_UTF8_LOCALE_IS_WRONG); 5904 } 5905 goto boundu; 5906 } 5907 5908 if (utf8_target) { 5909 if (locinput == reginfo->strbeg) 5910 b1 = isWORDCHAR_LC('\n'); 5911 else { 5912 b1 = isWORDCHAR_LC_utf8(reghop3((U8*)locinput, -1, 5913 (U8*)(reginfo->strbeg))); 5914 } 5915 b2 = (NEXTCHR_IS_EOS) 5916 ? isWORDCHAR_LC('\n') 5917 : isWORDCHAR_LC_utf8((U8*)locinput); 5918 } 5919 else { /* Here the string isn't utf8 */ 5920 b1 = (locinput == reginfo->strbeg) 5921 ? isWORDCHAR_LC('\n') 5922 : isWORDCHAR_LC(UCHARAT(locinput - 1)); 5923 b2 = (NEXTCHR_IS_EOS) 5924 ? isWORDCHAR_LC('\n') 5925 : isWORDCHAR_LC(nextchr); 5926 } 5927 if (to_complement ^ (b1 == b2)) { 5928 sayNO; 5929 } 5930 break; 5931 } 5932 5933 case NBOUND: /* /\B/ */ 5934 to_complement = 1; 5935 /* FALLTHROUGH */ 5936 5937 case BOUND: /* /\b/ */ 5938 if (utf8_target) { 5939 goto bound_utf8; 5940 } 5941 goto bound_ascii_match_only; 5942 5943 case NBOUNDA: /* /\B/a */ 5944 to_complement = 1; 5945 /* FALLTHROUGH */ 5946 5947 case BOUNDA: /* /\b/a */ 5948 { 5949 bool b1, b2; 5950 5951 bound_ascii_match_only: 5952 /* Here the string isn't utf8, or is utf8 and only ascii characters 5953 * are to match \w. In the latter case looking at the byte just 5954 * prior to the current one may be just the final byte of a 5955 * multi-byte character. This is ok. There are two cases: 5956 * 1) it is a single byte character, and then the test is doing 5957 * just what it's supposed to. 5958 * 2) it is a multi-byte character, in which case the final byte is 5959 * never mistakable for ASCII, and so the test will say it is 5960 * not a word character, which is the correct answer. */ 5961 b1 = (locinput == reginfo->strbeg) 5962 ? isWORDCHAR_A('\n') 5963 : isWORDCHAR_A(UCHARAT(locinput - 1)); 5964 b2 = (NEXTCHR_IS_EOS) 5965 ? isWORDCHAR_A('\n') 5966 : isWORDCHAR_A(nextchr); 5967 if (to_complement ^ (b1 == b2)) { 5968 sayNO; 5969 } 5970 break; 5971 } 5972 5973 case NBOUNDU: /* /\B/u */ 5974 to_complement = 1; 5975 /* FALLTHROUGH */ 5976 5977 case BOUNDU: /* /\b/u */ 5978 5979 boundu: 5980 if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) { 5981 match = FALSE; 5982 } 5983 else if (utf8_target) { 5984 bound_utf8: 5985 switch((bound_type) FLAGS(scan)) { 5986 case TRADITIONAL_BOUND: 5987 { 5988 bool b1, b2; 5989 b1 = (locinput == reginfo->strbeg) 5990 ? 0 /* isWORDCHAR_L1('\n') */ 5991 : isWORDCHAR_utf8(reghop3((U8*)locinput, -1, 5992 (U8*)(reginfo->strbeg))); 5993 b2 = (NEXTCHR_IS_EOS) 5994 ? 0 /* isWORDCHAR_L1('\n') */ 5995 : isWORDCHAR_utf8((U8*)locinput); 5996 match = cBOOL(b1 != b2); 5997 break; 5998 } 5999 case GCB_BOUND: 6000 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { 6001 match = TRUE; /* GCB always matches at begin and 6002 end */ 6003 } 6004 else { 6005 /* Find the gcb values of previous and current 6006 * chars, then see if is a break point */ 6007 match = isGCB(getGCB_VAL_UTF8( 6008 reghop3((U8*)locinput, 6009 -1, 6010 (U8*)(reginfo->strbeg)), 6011 (U8*) reginfo->strend), 6012 getGCB_VAL_UTF8((U8*) locinput, 6013 (U8*) reginfo->strend)); 6014 } 6015 break; 6016 6017 case LB_BOUND: 6018 if (locinput == reginfo->strbeg) { 6019 match = FALSE; 6020 } 6021 else if (NEXTCHR_IS_EOS) { 6022 match = TRUE; 6023 } 6024 else { 6025 match = isLB(getLB_VAL_UTF8( 6026 reghop3((U8*)locinput, 6027 -1, 6028 (U8*)(reginfo->strbeg)), 6029 (U8*) reginfo->strend), 6030 getLB_VAL_UTF8((U8*) locinput, 6031 (U8*) reginfo->strend), 6032 (U8*) reginfo->strbeg, 6033 (U8*) locinput, 6034 (U8*) reginfo->strend, 6035 utf8_target); 6036 } 6037 break; 6038 6039 case SB_BOUND: /* Always matches at begin and end */ 6040 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { 6041 match = TRUE; 6042 } 6043 else { 6044 match = isSB(getSB_VAL_UTF8( 6045 reghop3((U8*)locinput, 6046 -1, 6047 (U8*)(reginfo->strbeg)), 6048 (U8*) reginfo->strend), 6049 getSB_VAL_UTF8((U8*) locinput, 6050 (U8*) reginfo->strend), 6051 (U8*) reginfo->strbeg, 6052 (U8*) locinput, 6053 (U8*) reginfo->strend, 6054 utf8_target); 6055 } 6056 break; 6057 6058 case WB_BOUND: 6059 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { 6060 match = TRUE; 6061 } 6062 else { 6063 match = isWB(WB_UNKNOWN, 6064 getWB_VAL_UTF8( 6065 reghop3((U8*)locinput, 6066 -1, 6067 (U8*)(reginfo->strbeg)), 6068 (U8*) reginfo->strend), 6069 getWB_VAL_UTF8((U8*) locinput, 6070 (U8*) reginfo->strend), 6071 (U8*) reginfo->strbeg, 6072 (U8*) locinput, 6073 (U8*) reginfo->strend, 6074 utf8_target); 6075 } 6076 break; 6077 } 6078 } 6079 else { /* Not utf8 target */ 6080 switch((bound_type) FLAGS(scan)) { 6081 case TRADITIONAL_BOUND: 6082 { 6083 bool b1, b2; 6084 b1 = (locinput == reginfo->strbeg) 6085 ? 0 /* isWORDCHAR_L1('\n') */ 6086 : isWORDCHAR_L1(UCHARAT(locinput - 1)); 6087 b2 = (NEXTCHR_IS_EOS) 6088 ? 0 /* isWORDCHAR_L1('\n') */ 6089 : isWORDCHAR_L1(nextchr); 6090 match = cBOOL(b1 != b2); 6091 break; 6092 } 6093 6094 case GCB_BOUND: 6095 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { 6096 match = TRUE; /* GCB always matches at begin and 6097 end */ 6098 } 6099 else { /* Only CR-LF combo isn't a GCB in 0-255 6100 range */ 6101 match = UCHARAT(locinput - 1) != '\r' 6102 || UCHARAT(locinput) != '\n'; 6103 } 6104 break; 6105 6106 case LB_BOUND: 6107 if (locinput == reginfo->strbeg) { 6108 match = FALSE; 6109 } 6110 else if (NEXTCHR_IS_EOS) { 6111 match = TRUE; 6112 } 6113 else { 6114 match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)), 6115 getLB_VAL_CP(UCHARAT(locinput)), 6116 (U8*) reginfo->strbeg, 6117 (U8*) locinput, 6118 (U8*) reginfo->strend, 6119 utf8_target); 6120 } 6121 break; 6122 6123 case SB_BOUND: /* Always matches at begin and end */ 6124 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { 6125 match = TRUE; 6126 } 6127 else { 6128 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)), 6129 getSB_VAL_CP(UCHARAT(locinput)), 6130 (U8*) reginfo->strbeg, 6131 (U8*) locinput, 6132 (U8*) reginfo->strend, 6133 utf8_target); 6134 } 6135 break; 6136 6137 case WB_BOUND: 6138 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) { 6139 match = TRUE; 6140 } 6141 else { 6142 match = isWB(WB_UNKNOWN, 6143 getWB_VAL_CP(UCHARAT(locinput -1)), 6144 getWB_VAL_CP(UCHARAT(locinput)), 6145 (U8*) reginfo->strbeg, 6146 (U8*) locinput, 6147 (U8*) reginfo->strend, 6148 utf8_target); 6149 } 6150 break; 6151 } 6152 } 6153 6154 if (to_complement ^ ! match) { 6155 sayNO; 6156 } 6157 break; 6158 6159 case ANYOFL: /* /[abc]/l */ 6160 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 6161 6162 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE) 6163 { 6164 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required); 6165 } 6166 /* FALLTHROUGH */ 6167 case ANYOFD: /* /[abc]/d */ 6168 case ANYOF: /* /[abc]/ */ 6169 if (NEXTCHR_IS_EOS) 6170 sayNO; 6171 if (utf8_target && ! UTF8_IS_INVARIANT(*locinput)) { 6172 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend, 6173 utf8_target)) 6174 sayNO; 6175 locinput += UTF8SKIP(locinput); 6176 } 6177 else { 6178 if (!REGINCLASS(rex, scan, (U8*)locinput, utf8_target)) 6179 sayNO; 6180 locinput++; 6181 } 6182 break; 6183 6184 /* The argument (FLAGS) to all the POSIX node types is the class number 6185 * */ 6186 6187 case NPOSIXL: /* \W or [:^punct:] etc. under /l */ 6188 to_complement = 1; 6189 /* FALLTHROUGH */ 6190 6191 case POSIXL: /* \w or [:punct:] etc. under /l */ 6192 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 6193 if (NEXTCHR_IS_EOS) 6194 sayNO; 6195 6196 /* Use isFOO_lc() for characters within Latin1. (Note that 6197 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else 6198 * wouldn't be invariant) */ 6199 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { 6200 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) { 6201 sayNO; 6202 } 6203 6204 locinput++; 6205 break; 6206 } 6207 6208 if (! UTF8_IS_DOWNGRADEABLE_START(nextchr)) { /* An above Latin-1 code point */ 6209 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend); 6210 goto utf8_posix_above_latin1; 6211 } 6212 6213 /* Here is a UTF-8 variant code point below 256 and the target is 6214 * UTF-8 */ 6215 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), 6216 EIGHT_BIT_UTF8_TO_NATIVE(nextchr, 6217 *(locinput + 1)))))) 6218 { 6219 sayNO; 6220 } 6221 6222 goto increment_locinput; 6223 6224 case NPOSIXD: /* \W or [:^punct:] etc. under /d */ 6225 to_complement = 1; 6226 /* FALLTHROUGH */ 6227 6228 case POSIXD: /* \w or [:punct:] etc. under /d */ 6229 if (utf8_target) { 6230 goto utf8_posix; 6231 } 6232 goto posixa; 6233 6234 case NPOSIXA: /* \W or [:^punct:] etc. under /a */ 6235 6236 if (NEXTCHR_IS_EOS) { 6237 sayNO; 6238 } 6239 6240 /* All UTF-8 variants match */ 6241 if (! UTF8_IS_INVARIANT(nextchr)) { 6242 goto increment_locinput; 6243 } 6244 6245 to_complement = 1; 6246 goto join_nposixa; 6247 6248 case POSIXA: /* \w or [:punct:] etc. under /a */ 6249 6250 posixa: 6251 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in 6252 * UTF-8, and also from NPOSIXA even in UTF-8 when the current 6253 * character is a single byte */ 6254 6255 if (NEXTCHR_IS_EOS) { 6256 sayNO; 6257 } 6258 6259 join_nposixa: 6260 6261 if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr, 6262 FLAGS(scan))))) 6263 { 6264 sayNO; 6265 } 6266 6267 /* Here we are either not in utf8, or we matched a utf8-invariant, 6268 * so the next char is the next byte */ 6269 locinput++; 6270 break; 6271 6272 case NPOSIXU: /* \W or [:^punct:] etc. under /u */ 6273 to_complement = 1; 6274 /* FALLTHROUGH */ 6275 6276 case POSIXU: /* \w or [:punct:] etc. under /u */ 6277 utf8_posix: 6278 if (NEXTCHR_IS_EOS) { 6279 sayNO; 6280 } 6281 6282 /* Use _generic_isCC() for characters within Latin1. (Note that 6283 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else 6284 * wouldn't be invariant) */ 6285 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { 6286 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr, 6287 FLAGS(scan))))) 6288 { 6289 sayNO; 6290 } 6291 locinput++; 6292 } 6293 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { 6294 if (! (to_complement 6295 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr, 6296 *(locinput + 1)), 6297 FLAGS(scan))))) 6298 { 6299 sayNO; 6300 } 6301 locinput += 2; 6302 } 6303 else { /* Handle above Latin-1 code points */ 6304 utf8_posix_above_latin1: 6305 classnum = (_char_class_number) FLAGS(scan); 6306 if (classnum < _FIRST_NON_SWASH_CC) { 6307 6308 /* Here, uses a swash to find such code points. Load if if 6309 * not done already */ 6310 if (! PL_utf8_swash_ptrs[classnum]) { 6311 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 6312 PL_utf8_swash_ptrs[classnum] 6313 = _core_swash_init("utf8", 6314 "", 6315 &PL_sv_undef, 1, 0, 6316 PL_XPosix_ptrs[classnum], &flags); 6317 } 6318 if (! (to_complement 6319 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], 6320 (U8 *) locinput, TRUE)))) 6321 { 6322 sayNO; 6323 } 6324 } 6325 else { /* Here, uses macros to find above Latin-1 code points */ 6326 switch (classnum) { 6327 case _CC_ENUM_SPACE: 6328 if (! (to_complement 6329 ^ cBOOL(is_XPERLSPACE_high(locinput)))) 6330 { 6331 sayNO; 6332 } 6333 break; 6334 case _CC_ENUM_BLANK: 6335 if (! (to_complement 6336 ^ cBOOL(is_HORIZWS_high(locinput)))) 6337 { 6338 sayNO; 6339 } 6340 break; 6341 case _CC_ENUM_XDIGIT: 6342 if (! (to_complement 6343 ^ cBOOL(is_XDIGIT_high(locinput)))) 6344 { 6345 sayNO; 6346 } 6347 break; 6348 case _CC_ENUM_VERTSPACE: 6349 if (! (to_complement 6350 ^ cBOOL(is_VERTWS_high(locinput)))) 6351 { 6352 sayNO; 6353 } 6354 break; 6355 default: /* The rest, e.g. [:cntrl:], can't match 6356 above Latin1 */ 6357 if (! to_complement) { 6358 sayNO; 6359 } 6360 break; 6361 } 6362 } 6363 locinput += UTF8SKIP(locinput); 6364 } 6365 break; 6366 6367 case CLUMP: /* Match \X: logical Unicode character. This is defined as 6368 a Unicode extended Grapheme Cluster */ 6369 if (NEXTCHR_IS_EOS) 6370 sayNO; 6371 if (! utf8_target) { 6372 6373 /* Match either CR LF or '.', as all the other possibilities 6374 * require utf8 */ 6375 locinput++; /* Match the . or CR */ 6376 if (nextchr == '\r' /* And if it was CR, and the next is LF, 6377 match the LF */ 6378 && locinput < reginfo->strend 6379 && UCHARAT(locinput) == '\n') 6380 { 6381 locinput++; 6382 } 6383 } 6384 else { 6385 6386 /* Get the gcb type for the current character */ 6387 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput, 6388 (U8*) reginfo->strend); 6389 6390 /* Then scan through the input until we get to the first 6391 * character whose type is supposed to be a gcb with the 6392 * current character. (There is always a break at the 6393 * end-of-input) */ 6394 locinput += UTF8SKIP(locinput); 6395 while (locinput < reginfo->strend) { 6396 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput, 6397 (U8*) reginfo->strend); 6398 if (isGCB(prev_gcb, cur_gcb)) { 6399 break; 6400 } 6401 6402 prev_gcb = cur_gcb; 6403 locinput += UTF8SKIP(locinput); 6404 } 6405 6406 6407 } 6408 break; 6409 6410 case NREFFL: /* /\g{name}/il */ 6411 { /* The capture buffer cases. The ones beginning with N for the 6412 named buffers just convert to the equivalent numbered and 6413 pretend they were called as the corresponding numbered buffer 6414 op. */ 6415 /* don't initialize these in the declaration, it makes C++ 6416 unhappy */ 6417 const char *s; 6418 char type; 6419 re_fold_t folder; 6420 const U8 *fold_array; 6421 UV utf8_fold_flags; 6422 6423 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 6424 folder = foldEQ_locale; 6425 fold_array = PL_fold_locale; 6426 type = REFFL; 6427 utf8_fold_flags = FOLDEQ_LOCALE; 6428 goto do_nref; 6429 6430 case NREFFA: /* /\g{name}/iaa */ 6431 folder = foldEQ_latin1; 6432 fold_array = PL_fold_latin1; 6433 type = REFFA; 6434 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; 6435 goto do_nref; 6436 6437 case NREFFU: /* /\g{name}/iu */ 6438 folder = foldEQ_latin1; 6439 fold_array = PL_fold_latin1; 6440 type = REFFU; 6441 utf8_fold_flags = 0; 6442 goto do_nref; 6443 6444 case NREFF: /* /\g{name}/i */ 6445 folder = foldEQ; 6446 fold_array = PL_fold; 6447 type = REFF; 6448 utf8_fold_flags = 0; 6449 goto do_nref; 6450 6451 case NREF: /* /\g{name}/ */ 6452 type = REF; 6453 folder = NULL; 6454 fold_array = NULL; 6455 utf8_fold_flags = 0; 6456 do_nref: 6457 6458 /* For the named back references, find the corresponding buffer 6459 * number */ 6460 n = reg_check_named_buff_matched(rex,scan); 6461 6462 if ( ! n ) { 6463 sayNO; 6464 } 6465 goto do_nref_ref_common; 6466 6467 case REFFL: /* /\1/il */ 6468 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 6469 folder = foldEQ_locale; 6470 fold_array = PL_fold_locale; 6471 utf8_fold_flags = FOLDEQ_LOCALE; 6472 goto do_ref; 6473 6474 case REFFA: /* /\1/iaa */ 6475 folder = foldEQ_latin1; 6476 fold_array = PL_fold_latin1; 6477 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; 6478 goto do_ref; 6479 6480 case REFFU: /* /\1/iu */ 6481 folder = foldEQ_latin1; 6482 fold_array = PL_fold_latin1; 6483 utf8_fold_flags = 0; 6484 goto do_ref; 6485 6486 case REFF: /* /\1/i */ 6487 folder = foldEQ; 6488 fold_array = PL_fold; 6489 utf8_fold_flags = 0; 6490 goto do_ref; 6491 6492 case REF: /* /\1/ */ 6493 folder = NULL; 6494 fold_array = NULL; 6495 utf8_fold_flags = 0; 6496 6497 do_ref: 6498 type = OP(scan); 6499 n = ARG(scan); /* which paren pair */ 6500 6501 do_nref_ref_common: 6502 ln = rex->offs[n].start; 6503 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ 6504 if (rex->lastparen < n || ln == -1) 6505 sayNO; /* Do not match unless seen CLOSEn. */ 6506 if (ln == rex->offs[n].end) 6507 break; 6508 6509 s = reginfo->strbeg + ln; 6510 if (type != REF /* REF can do byte comparison */ 6511 && (utf8_target || type == REFFU || type == REFFL)) 6512 { 6513 char * limit = reginfo->strend; 6514 6515 /* This call case insensitively compares the entire buffer 6516 * at s, with the current input starting at locinput, but 6517 * not going off the end given by reginfo->strend, and 6518 * returns in <limit> upon success, how much of the 6519 * current input was matched */ 6520 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target, 6521 locinput, &limit, 0, utf8_target, utf8_fold_flags)) 6522 { 6523 sayNO; 6524 } 6525 locinput = limit; 6526 break; 6527 } 6528 6529 /* Not utf8: Inline the first character, for speed. */ 6530 if (!NEXTCHR_IS_EOS && 6531 UCHARAT(s) != nextchr && 6532 (type == REF || 6533 UCHARAT(s) != fold_array[nextchr])) 6534 sayNO; 6535 ln = rex->offs[n].end - ln; 6536 if (locinput + ln > reginfo->strend) 6537 sayNO; 6538 if (ln > 1 && (type == REF 6539 ? memNE(s, locinput, ln) 6540 : ! folder(s, locinput, ln))) 6541 sayNO; 6542 locinput += ln; 6543 break; 6544 } 6545 6546 case NOTHING: /* null op; e.g. the 'nothing' following 6547 * the '*' in m{(a+|b)*}' */ 6548 break; 6549 case TAIL: /* placeholder while compiling (A|B|C) */ 6550 break; 6551 6552 #undef ST 6553 #define ST st->u.eval 6554 #define CUR_EVAL cur_eval->u.eval 6555 6556 { 6557 SV *ret; 6558 REGEXP *re_sv; 6559 regexp *re; 6560 regexp_internal *rei; 6561 regnode *startpoint; 6562 U32 arg; 6563 6564 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */ 6565 arg= (U32)ARG(scan); 6566 if (cur_eval && cur_eval->locinput == locinput) { 6567 if ( ++nochange_depth > max_nochange_depth ) 6568 Perl_croak(aTHX_ 6569 "Pattern subroutine nesting without pos change" 6570 " exceeded limit in regex"); 6571 } else { 6572 nochange_depth = 0; 6573 } 6574 re_sv = rex_sv; 6575 re = rex; 6576 rei = rexi; 6577 startpoint = scan + ARG2L(scan); 6578 EVAL_CLOSE_PAREN_SET( st, arg ); 6579 /* Detect infinite recursion 6580 * 6581 * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/ 6582 * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever. 6583 * So we track the position in the string we are at each time 6584 * we recurse and if we try to enter the same routine twice from 6585 * the same position we throw an error. 6586 */ 6587 if ( rex->recurse_locinput[arg] == locinput ) { 6588 /* FIXME: we should show the regop that is failing as part 6589 * of the error message. */ 6590 Perl_croak(aTHX_ "Infinite recursion in regex"); 6591 } else { 6592 ST.prev_recurse_locinput= rex->recurse_locinput[arg]; 6593 rex->recurse_locinput[arg]= locinput; 6594 6595 DEBUG_r({ 6596 GET_RE_DEBUG_FLAGS_DECL; 6597 DEBUG_STACK_r({ 6598 Perl_re_exec_indentf( aTHX_ 6599 "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n", 6600 depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg] 6601 ); 6602 }); 6603 }); 6604 } 6605 6606 /* Save all the positions seen so far. */ 6607 ST.cp = regcppush(rex, 0, maxopenparen); 6608 REGCP_SET(ST.lastcp); 6609 6610 /* and then jump to the code we share with EVAL */ 6611 goto eval_recurse_doit; 6612 /* NOTREACHED */ 6613 6614 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ 6615 if (cur_eval && cur_eval->locinput==locinput) { 6616 if ( ++nochange_depth > max_nochange_depth ) 6617 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex"); 6618 } else { 6619 nochange_depth = 0; 6620 } 6621 { 6622 /* execute the code in the {...} */ 6623 6624 dSP; 6625 IV before; 6626 OP * const oop = PL_op; 6627 COP * const ocurcop = PL_curcop; 6628 OP *nop; 6629 CV *newcv; 6630 6631 /* save *all* paren positions */ 6632 regcppush(rex, 0, maxopenparen); 6633 REGCP_SET(runops_cp); 6634 6635 if (!caller_cv) 6636 caller_cv = find_runcv(NULL); 6637 6638 n = ARG(scan); 6639 6640 if (rexi->data->what[n] == 'r') { /* code from an external qr */ 6641 newcv = (ReANY( 6642 (REGEXP*)(rexi->data->data[n]) 6643 ))->qr_anoncv; 6644 nop = (OP*)rexi->data->data[n+1]; 6645 } 6646 else if (rexi->data->what[n] == 'l') { /* literal code */ 6647 newcv = caller_cv; 6648 nop = (OP*)rexi->data->data[n]; 6649 assert(CvDEPTH(newcv)); 6650 } 6651 else { 6652 /* literal with own CV */ 6653 assert(rexi->data->what[n] == 'L'); 6654 newcv = rex->qr_anoncv; 6655 nop = (OP*)rexi->data->data[n]; 6656 } 6657 6658 /* normally if we're about to execute code from the same 6659 * CV that we used previously, we just use the existing 6660 * CX stack entry. However, its possible that in the 6661 * meantime we may have backtracked, popped from the save 6662 * stack, and undone the SAVECOMPPAD(s) associated with 6663 * PUSH_MULTICALL; in which case PL_comppad no longer 6664 * points to newcv's pad. */ 6665 if (newcv != last_pushed_cv || PL_comppad != last_pad) 6666 { 6667 U8 flags = (CXp_SUB_RE | 6668 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0)); 6669 if (last_pushed_cv) { 6670 /* PUSH/POP_MULTICALL save and restore the 6671 * caller's PL_comppad; if we call multiple subs 6672 * using the same CX block, we have to save and 6673 * unwind the varying PL_comppad's ourselves, 6674 * especially restoring the right PL_comppad on 6675 * backtrack - so save it on the save stack */ 6676 SAVECOMPPAD(); 6677 CHANGE_MULTICALL_FLAGS(newcv, flags); 6678 } 6679 else { 6680 PUSH_MULTICALL_FLAGS(newcv, flags); 6681 } 6682 last_pushed_cv = newcv; 6683 } 6684 else { 6685 /* these assignments are just to silence compiler 6686 * warnings */ 6687 multicall_cop = NULL; 6688 } 6689 last_pad = PL_comppad; 6690 6691 /* the initial nextstate you would normally execute 6692 * at the start of an eval (which would cause error 6693 * messages to come from the eval), may be optimised 6694 * away from the execution path in the regex code blocks; 6695 * so manually set PL_curcop to it initially */ 6696 { 6697 OP *o = cUNOPx(nop)->op_first; 6698 assert(o->op_type == OP_NULL); 6699 if (o->op_targ == OP_SCOPE) { 6700 o = cUNOPo->op_first; 6701 } 6702 else { 6703 assert(o->op_targ == OP_LEAVE); 6704 o = cUNOPo->op_first; 6705 assert(o->op_type == OP_ENTER); 6706 o = OpSIBLING(o); 6707 } 6708 6709 if (o->op_type != OP_STUB) { 6710 assert( o->op_type == OP_NEXTSTATE 6711 || o->op_type == OP_DBSTATE 6712 || (o->op_type == OP_NULL 6713 && ( o->op_targ == OP_NEXTSTATE 6714 || o->op_targ == OP_DBSTATE 6715 ) 6716 ) 6717 ); 6718 PL_curcop = (COP*)o; 6719 } 6720 } 6721 nop = nop->op_next; 6722 6723 DEBUG_STATE_r( Perl_re_printf( aTHX_ 6724 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) ); 6725 6726 rex->offs[0].end = locinput - reginfo->strbeg; 6727 if (reginfo->info_aux_eval->pos_magic) 6728 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic, 6729 reginfo->sv, reginfo->strbeg, 6730 locinput - reginfo->strbeg); 6731 6732 if (sv_yes_mark) { 6733 SV *sv_mrk = get_sv("REGMARK", 1); 6734 sv_setsv(sv_mrk, sv_yes_mark); 6735 } 6736 6737 /* we don't use MULTICALL here as we want to call the 6738 * first op of the block of interest, rather than the 6739 * first op of the sub. Also, we don't want to free 6740 * the savestack frame */ 6741 before = (IV)(SP-PL_stack_base); 6742 PL_op = nop; 6743 CALLRUNOPS(aTHX); /* Scalar context. */ 6744 SPAGAIN; 6745 if ((IV)(SP-PL_stack_base) == before) 6746 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */ 6747 else { 6748 ret = POPs; 6749 PUTBACK; 6750 } 6751 6752 /* before restoring everything, evaluate the returned 6753 * value, so that 'uninit' warnings don't use the wrong 6754 * PL_op or pad. Also need to process any magic vars 6755 * (e.g. $1) *before* parentheses are restored */ 6756 6757 PL_op = NULL; 6758 6759 re_sv = NULL; 6760 if (logical == 0) /* (?{})/ */ 6761 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */ 6762 else if (logical == 1) { /* /(?(?{...})X|Y)/ */ 6763 sw = cBOOL(SvTRUE(ret)); 6764 logical = 0; 6765 } 6766 else { /* /(??{}) */ 6767 /* if its overloaded, let the regex compiler handle 6768 * it; otherwise extract regex, or stringify */ 6769 if (SvGMAGICAL(ret)) 6770 ret = sv_mortalcopy(ret); 6771 if (!SvAMAGIC(ret)) { 6772 SV *sv = ret; 6773 if (SvROK(sv)) 6774 sv = SvRV(sv); 6775 if (SvTYPE(sv) == SVt_REGEXP) 6776 re_sv = (REGEXP*) sv; 6777 else if (SvSMAGICAL(ret)) { 6778 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr); 6779 if (mg) 6780 re_sv = (REGEXP *) mg->mg_obj; 6781 } 6782 6783 /* force any undef warnings here */ 6784 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) { 6785 ret = sv_mortalcopy(ret); 6786 (void) SvPV_force_nolen(ret); 6787 } 6788 } 6789 6790 } 6791 6792 /* *** Note that at this point we don't restore 6793 * PL_comppad, (or pop the CxSUB) on the assumption it may 6794 * be used again soon. This is safe as long as nothing 6795 * in the regexp code uses the pad ! */ 6796 PL_op = oop; 6797 PL_curcop = ocurcop; 6798 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen); 6799 PL_curpm = PL_reg_curpm; 6800 6801 if (logical != 2) 6802 break; 6803 } 6804 6805 /* only /(??{})/ from now on */ 6806 logical = 0; 6807 { 6808 /* extract RE object from returned value; compiling if 6809 * necessary */ 6810 6811 if (re_sv) { 6812 re_sv = reg_temp_copy(NULL, re_sv); 6813 } 6814 else { 6815 U32 pm_flags = 0; 6816 6817 if (SvUTF8(ret) && IN_BYTES) { 6818 /* In use 'bytes': make a copy of the octet 6819 * sequence, but without the flag on */ 6820 STRLEN len; 6821 const char *const p = SvPV(ret, len); 6822 ret = newSVpvn_flags(p, len, SVs_TEMP); 6823 } 6824 if (rex->intflags & PREGf_USE_RE_EVAL) 6825 pm_flags |= PMf_USE_RE_EVAL; 6826 6827 /* if we got here, it should be an engine which 6828 * supports compiling code blocks and stuff */ 6829 assert(rex->engine && rex->engine->op_comp); 6830 assert(!(scan->flags & ~RXf_PMf_COMPILETIME)); 6831 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL, 6832 rex->engine, NULL, NULL, 6833 /* copy /msixn etc to inner pattern */ 6834 ARG2L(scan), 6835 pm_flags); 6836 6837 if (!(SvFLAGS(ret) 6838 & (SVs_TEMP | SVs_GMG | SVf_ROK)) 6839 && (!SvPADTMP(ret) || SvREADONLY(ret))) { 6840 /* This isn't a first class regexp. Instead, it's 6841 caching a regexp onto an existing, Perl visible 6842 scalar. */ 6843 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0); 6844 } 6845 } 6846 SAVEFREESV(re_sv); 6847 re = ReANY(re_sv); 6848 } 6849 RXp_MATCH_COPIED_off(re); 6850 re->subbeg = rex->subbeg; 6851 re->sublen = rex->sublen; 6852 re->suboffset = rex->suboffset; 6853 re->subcoffset = rex->subcoffset; 6854 re->lastparen = 0; 6855 re->lastcloseparen = 0; 6856 rei = RXi_GET(re); 6857 DEBUG_EXECUTE_r( 6858 debug_start_match(re_sv, utf8_target, locinput, 6859 reginfo->strend, "Matching embedded"); 6860 ); 6861 startpoint = rei->program + 1; 6862 EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0; 6863 * close_paren only for GOSUB */ 6864 ST.prev_recurse_locinput= NULL; /* only used for GOSUB */ 6865 /* Save all the seen positions so far. */ 6866 ST.cp = regcppush(rex, 0, maxopenparen); 6867 REGCP_SET(ST.lastcp); 6868 /* and set maxopenparen to 0, since we are starting a "fresh" match */ 6869 maxopenparen = 0; 6870 /* run the pattern returned from (??{...}) */ 6871 6872 eval_recurse_doit: /* Share code with GOSUB below this line 6873 * At this point we expect the stack context to be 6874 * set up correctly */ 6875 6876 /* invalidate the S-L poscache. We're now executing a 6877 * different set of WHILEM ops (and their associated 6878 * indexes) against the same string, so the bits in the 6879 * cache are meaningless. Setting maxiter to zero forces 6880 * the cache to be invalidated and zeroed before reuse. 6881 * XXX This is too dramatic a measure. Ideally we should 6882 * save the old cache and restore when running the outer 6883 * pattern again */ 6884 reginfo->poscache_maxiter = 0; 6885 6886 /* the new regexp might have a different is_utf8_pat than we do */ 6887 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv)); 6888 6889 ST.prev_rex = rex_sv; 6890 ST.prev_curlyx = cur_curlyx; 6891 rex_sv = re_sv; 6892 SET_reg_curpm(rex_sv); 6893 rex = re; 6894 rexi = rei; 6895 cur_curlyx = NULL; 6896 ST.B = next; 6897 ST.prev_eval = cur_eval; 6898 cur_eval = st; 6899 /* now continue from first node in postoned RE */ 6900 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput); 6901 NOT_REACHED; /* NOTREACHED */ 6902 } 6903 6904 case EVAL_AB: /* cleanup after a successful (??{A})B */ 6905 /* note: this is called twice; first after popping B, then A */ 6906 DEBUG_STACK_r({ 6907 Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n", 6908 depth, cur_eval, ST.prev_eval); 6909 }); 6910 6911 #define SET_RECURSE_LOCINPUT(STR,VAL)\ 6912 if ( cur_eval && CUR_EVAL.close_paren ) {\ 6913 DEBUG_STACK_r({ \ 6914 Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\ 6915 depth, \ 6916 CUR_EVAL.close_paren - 1,\ 6917 cur_eval, \ 6918 VAL); \ 6919 }); \ 6920 rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\ 6921 } 6922 6923 SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput); 6924 6925 rex_sv = ST.prev_rex; 6926 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); 6927 SET_reg_curpm(rex_sv); 6928 rex = ReANY(rex_sv); 6929 rexi = RXi_GET(rex); 6930 { 6931 /* preserve $^R across LEAVE's. See Bug 121070. */ 6932 SV *save_sv= GvSV(PL_replgv); 6933 SvREFCNT_inc(save_sv); 6934 regcpblow(ST.cp); /* LEAVE in disguise */ 6935 sv_setsv(GvSV(PL_replgv), save_sv); 6936 SvREFCNT_dec(save_sv); 6937 } 6938 cur_eval = ST.prev_eval; 6939 cur_curlyx = ST.prev_curlyx; 6940 6941 /* Invalidate cache. See "invalidate" comment above. */ 6942 reginfo->poscache_maxiter = 0; 6943 if ( nochange_depth ) 6944 nochange_depth--; 6945 6946 SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput); 6947 sayYES; 6948 6949 6950 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ 6951 /* note: this is called twice; first after popping B, then A */ 6952 DEBUG_STACK_r({ 6953 Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n", 6954 depth, cur_eval, ST.prev_eval); 6955 }); 6956 6957 SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput); 6958 6959 rex_sv = ST.prev_rex; 6960 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); 6961 SET_reg_curpm(rex_sv); 6962 rex = ReANY(rex_sv); 6963 rexi = RXi_GET(rex); 6964 6965 REGCP_UNWIND(ST.lastcp); 6966 regcppop(rex, &maxopenparen); 6967 cur_eval = ST.prev_eval; 6968 cur_curlyx = ST.prev_curlyx; 6969 6970 /* Invalidate cache. See "invalidate" comment above. */ 6971 reginfo->poscache_maxiter = 0; 6972 if ( nochange_depth ) 6973 nochange_depth--; 6974 6975 SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput); 6976 sayNO_SILENT; 6977 #undef ST 6978 6979 case OPEN: /* ( */ 6980 n = ARG(scan); /* which paren pair */ 6981 rex->offs[n].start_tmp = locinput - reginfo->strbeg; 6982 if (n > maxopenparen) 6983 maxopenparen = n; 6984 DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ 6985 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n", 6986 PTR2UV(rex), 6987 PTR2UV(rex->offs), 6988 (UV)n, 6989 (IV)rex->offs[n].start_tmp, 6990 (UV)maxopenparen 6991 )); 6992 lastopen = n; 6993 break; 6994 6995 /* XXX really need to log other places start/end are set too */ 6996 #define CLOSE_CAPTURE \ 6997 rex->offs[n].start = rex->offs[n].start_tmp; \ 6998 rex->offs[n].end = locinput - reginfo->strbeg; \ 6999 DEBUG_BUFFERS_r(Perl_re_printf( aTHX_ \ 7000 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \ 7001 PTR2UV(rex), \ 7002 PTR2UV(rex->offs), \ 7003 (UV)n, \ 7004 (IV)rex->offs[n].start, \ 7005 (IV)rex->offs[n].end \ 7006 )) 7007 7008 case CLOSE: /* ) */ 7009 n = ARG(scan); /* which paren pair */ 7010 CLOSE_CAPTURE; 7011 if (n > rex->lastparen) 7012 rex->lastparen = n; 7013 rex->lastcloseparen = n; 7014 if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) ) 7015 goto fake_end; 7016 7017 break; 7018 7019 case ACCEPT: /* (*ACCEPT) */ 7020 if (scan->flags) 7021 sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 7022 if (ARG2L(scan)){ 7023 regnode *cursor; 7024 for (cursor=scan; 7025 cursor && OP(cursor)!=END; 7026 cursor=regnext(cursor)) 7027 { 7028 if ( OP(cursor)==CLOSE ){ 7029 n = ARG(cursor); 7030 if ( n <= lastopen ) { 7031 CLOSE_CAPTURE; 7032 if (n > rex->lastparen) 7033 rex->lastparen = n; 7034 rex->lastcloseparen = n; 7035 if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) ) 7036 break; 7037 } 7038 } 7039 } 7040 } 7041 goto fake_end; 7042 /* NOTREACHED */ 7043 7044 case GROUPP: /* (?(1)) */ 7045 n = ARG(scan); /* which paren pair */ 7046 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1); 7047 break; 7048 7049 case NGROUPP: /* (?(<name>)) */ 7050 /* reg_check_named_buff_matched returns 0 for no match */ 7051 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan)); 7052 break; 7053 7054 case INSUBP: /* (?(R)) */ 7055 n = ARG(scan); 7056 /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg 7057 * of SCAN is already set up as matches a eval.close_paren */ 7058 sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n); 7059 break; 7060 7061 case DEFINEP: /* (?(DEFINE)) */ 7062 sw = 0; 7063 break; 7064 7065 case IFTHEN: /* (?(cond)A|B) */ 7066 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ 7067 if (sw) 7068 next = NEXTOPER(NEXTOPER(scan)); 7069 else { 7070 next = scan + ARG(scan); 7071 if (OP(next) == IFTHEN) /* Fake one. */ 7072 next = NEXTOPER(NEXTOPER(next)); 7073 } 7074 break; 7075 7076 case LOGICAL: /* modifier for EVAL and IFMATCH */ 7077 logical = scan->flags; 7078 break; 7079 7080 /******************************************************************* 7081 7082 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ 7083 pattern, where A and B are subpatterns. (For simple A, CURLYM or 7084 STAR/PLUS/CURLY/CURLYN are used instead.) 7085 7086 A*B is compiled as <CURLYX><A><WHILEM><B> 7087 7088 On entry to the subpattern, CURLYX is called. This pushes a CURLYX 7089 state, which contains the current count, initialised to -1. It also sets 7090 cur_curlyx to point to this state, with any previous value saved in the 7091 state block. 7092 7093 CURLYX then jumps straight to the WHILEM op, rather than executing A, 7094 since the pattern may possibly match zero times (i.e. it's a while {} loop 7095 rather than a do {} while loop). 7096 7097 Each entry to WHILEM represents a successful match of A. The count in the 7098 CURLYX block is incremented, another WHILEM state is pushed, and execution 7099 passes to A or B depending on greediness and the current count. 7100 7101 For example, if matching against the string a1a2a3b (where the aN are 7102 substrings that match /A/), then the match progresses as follows: (the 7103 pushed states are interspersed with the bits of strings matched so far): 7104 7105 <CURLYX cnt=-1> 7106 <CURLYX cnt=0><WHILEM> 7107 <CURLYX cnt=1><WHILEM> a1 <WHILEM> 7108 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM> 7109 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> 7110 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b 7111 7112 (Contrast this with something like CURLYM, which maintains only a single 7113 backtrack state: 7114 7115 <CURLYM cnt=0> a1 7116 a1 <CURLYM cnt=1> a2 7117 a1 a2 <CURLYM cnt=2> a3 7118 a1 a2 a3 <CURLYM cnt=3> b 7119 ) 7120 7121 Each WHILEM state block marks a point to backtrack to upon partial failure 7122 of A or B, and also contains some minor state data related to that 7123 iteration. The CURLYX block, pointed to by cur_curlyx, contains the 7124 overall state, such as the count, and pointers to the A and B ops. 7125 7126 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx 7127 must always point to the *current* CURLYX block, the rules are: 7128 7129 When executing CURLYX, save the old cur_curlyx in the CURLYX state block, 7130 and set cur_curlyx to point the new block. 7131 7132 When popping the CURLYX block after a successful or unsuccessful match, 7133 restore the previous cur_curlyx. 7134 7135 When WHILEM is about to execute B, save the current cur_curlyx, and set it 7136 to the outer one saved in the CURLYX block. 7137 7138 When popping the WHILEM block after a successful or unsuccessful B match, 7139 restore the previous cur_curlyx. 7140 7141 Here's an example for the pattern (AI* BI)*BO 7142 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: 7143 7144 cur_ 7145 curlyx backtrack stack 7146 ------ --------------- 7147 NULL 7148 CO <CO prev=NULL> <WO> 7149 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai 7150 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi 7151 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo 7152 7153 At this point the pattern succeeds, and we work back down the stack to 7154 clean up, restoring as we go: 7155 7156 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi 7157 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai 7158 CO <CO prev=NULL> <WO> 7159 NULL 7160 7161 *******************************************************************/ 7162 7163 #define ST st->u.curlyx 7164 7165 case CURLYX: /* start of /A*B/ (for complex A) */ 7166 { 7167 /* No need to save/restore up to this paren */ 7168 I32 parenfloor = scan->flags; 7169 7170 assert(next); /* keep Coverity happy */ 7171 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ 7172 next += ARG(next); 7173 7174 /* XXXX Probably it is better to teach regpush to support 7175 parenfloor > maxopenparen ... */ 7176 if (parenfloor > (I32)rex->lastparen) 7177 parenfloor = rex->lastparen; /* Pessimization... */ 7178 7179 ST.prev_curlyx= cur_curlyx; 7180 cur_curlyx = st; 7181 ST.cp = PL_savestack_ix; 7182 7183 /* these fields contain the state of the current curly. 7184 * they are accessed by subsequent WHILEMs */ 7185 ST.parenfloor = parenfloor; 7186 ST.me = scan; 7187 ST.B = next; 7188 ST.minmod = minmod; 7189 minmod = 0; 7190 ST.count = -1; /* this will be updated by WHILEM */ 7191 ST.lastloc = NULL; /* this will be updated by WHILEM */ 7192 7193 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput); 7194 NOT_REACHED; /* NOTREACHED */ 7195 } 7196 7197 case CURLYX_end: /* just finished matching all of A*B */ 7198 cur_curlyx = ST.prev_curlyx; 7199 sayYES; 7200 NOT_REACHED; /* NOTREACHED */ 7201 7202 case CURLYX_end_fail: /* just failed to match all of A*B */ 7203 regcpblow(ST.cp); 7204 cur_curlyx = ST.prev_curlyx; 7205 sayNO; 7206 NOT_REACHED; /* NOTREACHED */ 7207 7208 7209 #undef ST 7210 #define ST st->u.whilem 7211 7212 case WHILEM: /* just matched an A in /A*B/ (for complex A) */ 7213 { 7214 /* see the discussion above about CURLYX/WHILEM */ 7215 I32 n; 7216 int min, max; 7217 regnode *A; 7218 7219 assert(cur_curlyx); /* keep Coverity happy */ 7220 7221 min = ARG1(cur_curlyx->u.curlyx.me); 7222 max = ARG2(cur_curlyx->u.curlyx.me); 7223 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS; 7224 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ 7225 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; 7226 ST.cache_offset = 0; 7227 ST.cache_mask = 0; 7228 7229 7230 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: matched %ld out of %d..%d\n", 7231 depth, (long)n, min, max) 7232 ); 7233 7234 /* First just match a string of min A's. */ 7235 7236 if (n < min) { 7237 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, 7238 maxopenparen); 7239 cur_curlyx->u.curlyx.lastloc = locinput; 7240 REGCP_SET(ST.lastcp); 7241 7242 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput); 7243 NOT_REACHED; /* NOTREACHED */ 7244 } 7245 7246 /* If degenerate A matches "", assume A done. */ 7247 7248 if (locinput == cur_curlyx->u.curlyx.lastloc) { 7249 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: empty match detected, trying continuation...\n", 7250 depth) 7251 ); 7252 goto do_whilem_B_max; 7253 } 7254 7255 /* super-linear cache processing. 7256 * 7257 * The idea here is that for certain types of CURLYX/WHILEM - 7258 * principally those whose upper bound is infinity (and 7259 * excluding regexes that have things like \1 and other very 7260 * non-regular expresssiony things), then if a pattern like 7261 * /....A*.../ fails and we backtrack to the WHILEM, then we 7262 * make a note that this particular WHILEM op was at string 7263 * position 47 (say) when the rest of pattern failed. Then, if 7264 * we ever find ourselves back at that WHILEM, and at string 7265 * position 47 again, we can just fail immediately rather than 7266 * running the rest of the pattern again. 7267 * 7268 * This is very handy when patterns start to go 7269 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up 7270 * with a combinatorial explosion of backtracking. 7271 * 7272 * The cache is implemented as a bit array, with one bit per 7273 * string byte position per WHILEM op (up to 16) - so its 7274 * between 0.25 and 2x the string size. 7275 * 7276 * To avoid allocating a poscache buffer every time, we do an 7277 * initially countdown; only after we have executed a WHILEM 7278 * op (string-length x #WHILEMs) times do we allocate the 7279 * cache. 7280 * 7281 * The top 4 bits of scan->flags byte say how many different 7282 * relevant CURLLYX/WHILEM op pairs there are, while the 7283 * bottom 4-bits is the identifying index number of this 7284 * WHILEM. 7285 */ 7286 7287 if (scan->flags) { 7288 7289 if (!reginfo->poscache_maxiter) { 7290 /* start the countdown: Postpone detection until we 7291 * know the match is not *that* much linear. */ 7292 reginfo->poscache_maxiter 7293 = (reginfo->strend - reginfo->strbeg + 1) 7294 * (scan->flags>>4); 7295 /* possible overflow for long strings and many CURLYX's */ 7296 if (reginfo->poscache_maxiter < 0) 7297 reginfo->poscache_maxiter = I32_MAX; 7298 reginfo->poscache_iter = reginfo->poscache_maxiter; 7299 } 7300 7301 if (reginfo->poscache_iter-- == 0) { 7302 /* initialise cache */ 7303 const SSize_t size = (reginfo->poscache_maxiter + 7)/8; 7304 regmatch_info_aux *const aux = reginfo->info_aux; 7305 if (aux->poscache) { 7306 if ((SSize_t)reginfo->poscache_size < size) { 7307 Renew(aux->poscache, size, char); 7308 reginfo->poscache_size = size; 7309 } 7310 Zero(aux->poscache, size, char); 7311 } 7312 else { 7313 reginfo->poscache_size = size; 7314 Newxz(aux->poscache, size, char); 7315 } 7316 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_ 7317 "%swhilem: Detected a super-linear match, switching on caching%s...\n", 7318 PL_colors[4], PL_colors[5]) 7319 ); 7320 } 7321 7322 if (reginfo->poscache_iter < 0) { 7323 /* have we already failed at this position? */ 7324 SSize_t offset, mask; 7325 7326 reginfo->poscache_iter = -1; /* stop eventual underflow */ 7327 offset = (scan->flags & 0xf) - 1 7328 + (locinput - reginfo->strbeg) 7329 * (scan->flags>>4); 7330 mask = 1 << (offset % 8); 7331 offset /= 8; 7332 if (reginfo->info_aux->poscache[offset] & mask) { 7333 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: (cache) already tried at this position...\n", 7334 depth) 7335 ); 7336 sayNO; /* cache records failure */ 7337 } 7338 ST.cache_offset = offset; 7339 ST.cache_mask = mask; 7340 } 7341 } 7342 7343 /* Prefer B over A for minimal matching. */ 7344 7345 if (cur_curlyx->u.curlyx.minmod) { 7346 ST.save_curlyx = cur_curlyx; 7347 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; 7348 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor, 7349 maxopenparen); 7350 REGCP_SET(ST.lastcp); 7351 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B, 7352 locinput); 7353 NOT_REACHED; /* NOTREACHED */ 7354 } 7355 7356 /* Prefer A over B for maximal matching. */ 7357 7358 if (n < max) { /* More greed allowed? */ 7359 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, 7360 maxopenparen); 7361 cur_curlyx->u.curlyx.lastloc = locinput; 7362 REGCP_SET(ST.lastcp); 7363 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput); 7364 NOT_REACHED; /* NOTREACHED */ 7365 } 7366 goto do_whilem_B_max; 7367 } 7368 NOT_REACHED; /* NOTREACHED */ 7369 7370 case WHILEM_B_min: /* just matched B in a minimal match */ 7371 case WHILEM_B_max: /* just matched B in a maximal match */ 7372 cur_curlyx = ST.save_curlyx; 7373 sayYES; 7374 NOT_REACHED; /* NOTREACHED */ 7375 7376 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ 7377 cur_curlyx = ST.save_curlyx; 7378 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; 7379 cur_curlyx->u.curlyx.count--; 7380 CACHEsayNO; 7381 NOT_REACHED; /* NOTREACHED */ 7382 7383 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ 7384 /* FALLTHROUGH */ 7385 case WHILEM_A_pre_fail: /* just failed to match even minimal A */ 7386 REGCP_UNWIND(ST.lastcp); 7387 regcppop(rex, &maxopenparen); 7388 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; 7389 cur_curlyx->u.curlyx.count--; 7390 CACHEsayNO; 7391 NOT_REACHED; /* NOTREACHED */ 7392 7393 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ 7394 REGCP_UNWIND(ST.lastcp); 7395 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */ 7396 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "whilem: failed, trying continuation...\n", 7397 depth) 7398 ); 7399 do_whilem_B_max: 7400 if (cur_curlyx->u.curlyx.count >= REG_INFTY 7401 && ckWARN(WARN_REGEXP) 7402 && !reginfo->warned) 7403 { 7404 reginfo->warned = TRUE; 7405 Perl_warner(aTHX_ packWARN(WARN_REGEXP), 7406 "Complex regular subexpression recursion limit (%d) " 7407 "exceeded", 7408 REG_INFTY - 1); 7409 } 7410 7411 /* now try B */ 7412 ST.save_curlyx = cur_curlyx; 7413 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; 7414 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B, 7415 locinput); 7416 NOT_REACHED; /* NOTREACHED */ 7417 7418 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ 7419 cur_curlyx = ST.save_curlyx; 7420 REGCP_UNWIND(ST.lastcp); 7421 regcppop(rex, &maxopenparen); 7422 7423 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) { 7424 /* Maximum greed exceeded */ 7425 if (cur_curlyx->u.curlyx.count >= REG_INFTY 7426 && ckWARN(WARN_REGEXP) 7427 && !reginfo->warned) 7428 { 7429 reginfo->warned = TRUE; 7430 Perl_warner(aTHX_ packWARN(WARN_REGEXP), 7431 "Complex regular subexpression recursion " 7432 "limit (%d) exceeded", 7433 REG_INFTY - 1); 7434 } 7435 cur_curlyx->u.curlyx.count--; 7436 CACHEsayNO; 7437 } 7438 7439 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "trying longer...\n", depth) 7440 ); 7441 /* Try grabbing another A and see if it helps. */ 7442 cur_curlyx->u.curlyx.lastloc = locinput; 7443 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, 7444 maxopenparen); 7445 REGCP_SET(ST.lastcp); 7446 PUSH_STATE_GOTO(WHILEM_A_min, 7447 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS, 7448 locinput); 7449 NOT_REACHED; /* NOTREACHED */ 7450 7451 #undef ST 7452 #define ST st->u.branch 7453 7454 case BRANCHJ: /* /(...|A|...)/ with long next pointer */ 7455 next = scan + ARG(scan); 7456 if (next == scan) 7457 next = NULL; 7458 scan = NEXTOPER(scan); 7459 /* FALLTHROUGH */ 7460 7461 case BRANCH: /* /(...|A|...)/ */ 7462 scan = NEXTOPER(scan); /* scan now points to inner node */ 7463 ST.lastparen = rex->lastparen; 7464 ST.lastcloseparen = rex->lastcloseparen; 7465 ST.next_branch = next; 7466 REGCP_SET(ST.cp); 7467 7468 /* Now go into the branch */ 7469 if (has_cutgroup) { 7470 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput); 7471 } else { 7472 PUSH_STATE_GOTO(BRANCH_next, scan, locinput); 7473 } 7474 NOT_REACHED; /* NOTREACHED */ 7475 7476 case CUTGROUP: /* /(*THEN)/ */ 7477 sv_yes_mark = st->u.mark.mark_name = scan->flags 7478 ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ]) 7479 : NULL; 7480 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput); 7481 NOT_REACHED; /* NOTREACHED */ 7482 7483 case CUTGROUP_next_fail: 7484 do_cutgroup = 1; 7485 no_final = 1; 7486 if (st->u.mark.mark_name) 7487 sv_commit = st->u.mark.mark_name; 7488 sayNO; 7489 NOT_REACHED; /* NOTREACHED */ 7490 7491 case BRANCH_next: 7492 sayYES; 7493 NOT_REACHED; /* NOTREACHED */ 7494 7495 case BRANCH_next_fail: /* that branch failed; try the next, if any */ 7496 if (do_cutgroup) { 7497 do_cutgroup = 0; 7498 no_final = 0; 7499 } 7500 REGCP_UNWIND(ST.cp); 7501 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 7502 scan = ST.next_branch; 7503 /* no more branches? */ 7504 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) { 7505 DEBUG_EXECUTE_r({ 7506 Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n", 7507 depth, 7508 PL_colors[4], 7509 PL_colors[5] ); 7510 }); 7511 sayNO_SILENT; 7512 } 7513 continue; /* execute next BRANCH[J] op */ 7514 /* NOTREACHED */ 7515 7516 case MINMOD: /* next op will be non-greedy, e.g. A*? */ 7517 minmod = 1; 7518 break; 7519 7520 #undef ST 7521 #define ST st->u.curlym 7522 7523 case CURLYM: /* /A{m,n}B/ where A is fixed-length */ 7524 7525 /* This is an optimisation of CURLYX that enables us to push 7526 * only a single backtracking state, no matter how many matches 7527 * there are in {m,n}. It relies on the pattern being constant 7528 * length, with no parens to influence future backrefs 7529 */ 7530 7531 ST.me = scan; 7532 scan = NEXTOPER(scan) + NODE_STEP_REGNODE; 7533 7534 ST.lastparen = rex->lastparen; 7535 ST.lastcloseparen = rex->lastcloseparen; 7536 7537 /* if paren positive, emulate an OPEN/CLOSE around A */ 7538 if (ST.me->flags) { 7539 U32 paren = ST.me->flags; 7540 if (paren > maxopenparen) 7541 maxopenparen = paren; 7542 scan += NEXT_OFF(scan); /* Skip former OPEN. */ 7543 } 7544 ST.A = scan; 7545 ST.B = next; 7546 ST.alen = 0; 7547 ST.count = 0; 7548 ST.minmod = minmod; 7549 minmod = 0; 7550 ST.c1 = CHRTEST_UNINIT; 7551 REGCP_SET(ST.cp); 7552 7553 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */ 7554 goto curlym_do_B; 7555 7556 curlym_do_A: /* execute the A in /A{m,n}B/ */ 7557 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */ 7558 NOT_REACHED; /* NOTREACHED */ 7559 7560 case CURLYM_A: /* we've just matched an A */ 7561 ST.count++; 7562 /* after first match, determine A's length: u.curlym.alen */ 7563 if (ST.count == 1) { 7564 if (reginfo->is_utf8_target) { 7565 char *s = st->locinput; 7566 while (s < locinput) { 7567 ST.alen++; 7568 s += UTF8SKIP(s); 7569 } 7570 } 7571 else { 7572 ST.alen = locinput - st->locinput; 7573 } 7574 if (ST.alen == 0) 7575 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me); 7576 } 7577 DEBUG_EXECUTE_r( 7578 Perl_re_exec_indentf( aTHX_ "CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", 7579 depth, (IV) ST.count, (IV)ST.alen) 7580 ); 7581 7582 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags)) 7583 goto fake_end; 7584 7585 { 7586 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)); 7587 if ( max == REG_INFTY || ST.count < max ) 7588 goto curlym_do_A; /* try to match another A */ 7589 } 7590 goto curlym_do_B; /* try to match B */ 7591 7592 case CURLYM_A_fail: /* just failed to match an A */ 7593 REGCP_UNWIND(ST.cp); 7594 7595 7596 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ 7597 || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags)) 7598 sayNO; 7599 7600 curlym_do_B: /* execute the B in /A{m,n}B/ */ 7601 if (ST.c1 == CHRTEST_UNINIT) { 7602 /* calculate c1 and c2 for possible match of 1st char 7603 * following curly */ 7604 ST.c1 = ST.c2 = CHRTEST_VOID; 7605 assert(ST.B); 7606 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) { 7607 regnode *text_node = ST.B; 7608 if (! HAS_TEXT(text_node)) 7609 FIND_NEXT_IMPT(text_node); 7610 /* this used to be 7611 7612 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT) 7613 7614 But the former is redundant in light of the latter. 7615 7616 if this changes back then the macro for 7617 IS_TEXT and friends need to change. 7618 */ 7619 if (PL_regkind[OP(text_node)] == EXACT) { 7620 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ 7621 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, 7622 reginfo)) 7623 { 7624 sayNO; 7625 } 7626 } 7627 } 7628 } 7629 7630 DEBUG_EXECUTE_r( 7631 Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%"IVdf"...\n", 7632 depth, (IV)ST.count) 7633 ); 7634 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) { 7635 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) { 7636 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) 7637 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) 7638 { 7639 /* simulate B failing */ 7640 DEBUG_OPTIMISE_r( 7641 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n", 7642 depth, 7643 valid_utf8_to_uvchr((U8 *) locinput, NULL), 7644 valid_utf8_to_uvchr(ST.c1_utf8, NULL), 7645 valid_utf8_to_uvchr(ST.c2_utf8, NULL)) 7646 ); 7647 state_num = CURLYM_B_fail; 7648 goto reenter_switch; 7649 } 7650 } 7651 else if (nextchr != ST.c1 && nextchr != ST.c2) { 7652 /* simulate B failing */ 7653 DEBUG_OPTIMISE_r( 7654 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n", 7655 depth, 7656 (int) nextchr, ST.c1, ST.c2) 7657 ); 7658 state_num = CURLYM_B_fail; 7659 goto reenter_switch; 7660 } 7661 } 7662 7663 if (ST.me->flags) { 7664 /* emulate CLOSE: mark current A as captured */ 7665 I32 paren = ST.me->flags; 7666 if (ST.count) { 7667 rex->offs[paren].start 7668 = HOPc(locinput, -ST.alen) - reginfo->strbeg; 7669 rex->offs[paren].end = locinput - reginfo->strbeg; 7670 if ((U32)paren > rex->lastparen) 7671 rex->lastparen = paren; 7672 rex->lastcloseparen = paren; 7673 } 7674 else 7675 rex->offs[paren].end = -1; 7676 7677 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags)) 7678 { 7679 if (ST.count) 7680 goto fake_end; 7681 else 7682 sayNO; 7683 } 7684 } 7685 7686 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */ 7687 NOT_REACHED; /* NOTREACHED */ 7688 7689 case CURLYM_B_fail: /* just failed to match a B */ 7690 REGCP_UNWIND(ST.cp); 7691 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 7692 if (ST.minmod) { 7693 I32 max = ARG2(ST.me); 7694 if (max != REG_INFTY && ST.count == max) 7695 sayNO; 7696 goto curlym_do_A; /* try to match a further A */ 7697 } 7698 /* backtrack one A */ 7699 if (ST.count == ARG1(ST.me) /* min */) 7700 sayNO; 7701 ST.count--; 7702 SET_locinput(HOPc(locinput, -ST.alen)); 7703 goto curlym_do_B; /* try to match B */ 7704 7705 #undef ST 7706 #define ST st->u.curly 7707 7708 #define CURLY_SETPAREN(paren, success) \ 7709 if (paren) { \ 7710 if (success) { \ 7711 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \ 7712 rex->offs[paren].end = locinput - reginfo->strbeg; \ 7713 if (paren > rex->lastparen) \ 7714 rex->lastparen = paren; \ 7715 rex->lastcloseparen = paren; \ 7716 } \ 7717 else { \ 7718 rex->offs[paren].end = -1; \ 7719 rex->lastparen = ST.lastparen; \ 7720 rex->lastcloseparen = ST.lastcloseparen; \ 7721 } \ 7722 } 7723 7724 case STAR: /* /A*B/ where A is width 1 char */ 7725 ST.paren = 0; 7726 ST.min = 0; 7727 ST.max = REG_INFTY; 7728 scan = NEXTOPER(scan); 7729 goto repeat; 7730 7731 case PLUS: /* /A+B/ where A is width 1 char */ 7732 ST.paren = 0; 7733 ST.min = 1; 7734 ST.max = REG_INFTY; 7735 scan = NEXTOPER(scan); 7736 goto repeat; 7737 7738 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */ 7739 ST.paren = scan->flags; /* Which paren to set */ 7740 ST.lastparen = rex->lastparen; 7741 ST.lastcloseparen = rex->lastcloseparen; 7742 if (ST.paren > maxopenparen) 7743 maxopenparen = ST.paren; 7744 ST.min = ARG1(scan); /* min to match */ 7745 ST.max = ARG2(scan); /* max to match */ 7746 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) 7747 { 7748 ST.min=1; 7749 ST.max=1; 7750 } 7751 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); 7752 goto repeat; 7753 7754 case CURLY: /* /A{m,n}B/ where A is width 1 char */ 7755 ST.paren = 0; 7756 ST.min = ARG1(scan); /* min to match */ 7757 ST.max = ARG2(scan); /* max to match */ 7758 scan = NEXTOPER(scan) + NODE_STEP_REGNODE; 7759 repeat: 7760 /* 7761 * Lookahead to avoid useless match attempts 7762 * when we know what character comes next. 7763 * 7764 * Used to only do .*x and .*?x, but now it allows 7765 * for )'s, ('s and (?{ ... })'s to be in the way 7766 * of the quantifier and the EXACT-like node. -- japhy 7767 */ 7768 7769 assert(ST.min <= ST.max); 7770 if (! HAS_TEXT(next) && ! JUMPABLE(next)) { 7771 ST.c1 = ST.c2 = CHRTEST_VOID; 7772 } 7773 else { 7774 regnode *text_node = next; 7775 7776 if (! HAS_TEXT(text_node)) 7777 FIND_NEXT_IMPT(text_node); 7778 7779 if (! HAS_TEXT(text_node)) 7780 ST.c1 = ST.c2 = CHRTEST_VOID; 7781 else { 7782 if ( PL_regkind[OP(text_node)] != EXACT ) { 7783 ST.c1 = ST.c2 = CHRTEST_VOID; 7784 } 7785 else { 7786 7787 /* Currently we only get here when 7788 7789 PL_rekind[OP(text_node)] == EXACT 7790 7791 if this changes back then the macro for IS_TEXT and 7792 friends need to change. */ 7793 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ 7794 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, 7795 reginfo)) 7796 { 7797 sayNO; 7798 } 7799 } 7800 } 7801 } 7802 7803 ST.A = scan; 7804 ST.B = next; 7805 if (minmod) { 7806 char *li = locinput; 7807 minmod = 0; 7808 if (ST.min && 7809 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth) 7810 < ST.min) 7811 sayNO; 7812 SET_locinput(li); 7813 ST.count = ST.min; 7814 REGCP_SET(ST.cp); 7815 if (ST.c1 == CHRTEST_VOID) 7816 goto curly_try_B_min; 7817 7818 ST.oldloc = locinput; 7819 7820 /* set ST.maxpos to the furthest point along the 7821 * string that could possibly match */ 7822 if (ST.max == REG_INFTY) { 7823 ST.maxpos = reginfo->strend - 1; 7824 if (utf8_target) 7825 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos)) 7826 ST.maxpos--; 7827 } 7828 else if (utf8_target) { 7829 int m = ST.max - ST.min; 7830 for (ST.maxpos = locinput; 7831 m >0 && ST.maxpos < reginfo->strend; m--) 7832 ST.maxpos += UTF8SKIP(ST.maxpos); 7833 } 7834 else { 7835 ST.maxpos = locinput + ST.max - ST.min; 7836 if (ST.maxpos >= reginfo->strend) 7837 ST.maxpos = reginfo->strend - 1; 7838 } 7839 goto curly_try_B_min_known; 7840 7841 } 7842 else { 7843 /* avoid taking address of locinput, so it can remain 7844 * a register var */ 7845 char *li = locinput; 7846 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth); 7847 if (ST.count < ST.min) 7848 sayNO; 7849 SET_locinput(li); 7850 if ((ST.count > ST.min) 7851 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL)) 7852 { 7853 /* A{m,n} must come at the end of the string, there's 7854 * no point in backing off ... */ 7855 ST.min = ST.count; 7856 /* ...except that $ and \Z can match before *and* after 7857 newline at the end. Consider "\n\n" =~ /\n+\Z\n/. 7858 We may back off by one in this case. */ 7859 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS) 7860 ST.min--; 7861 } 7862 REGCP_SET(ST.cp); 7863 goto curly_try_B_max; 7864 } 7865 NOT_REACHED; /* NOTREACHED */ 7866 7867 case CURLY_B_min_known_fail: 7868 /* failed to find B in a non-greedy match where c1,c2 valid */ 7869 7870 REGCP_UNWIND(ST.cp); 7871 if (ST.paren) { 7872 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 7873 } 7874 /* Couldn't or didn't -- move forward. */ 7875 ST.oldloc = locinput; 7876 if (utf8_target) 7877 locinput += UTF8SKIP(locinput); 7878 else 7879 locinput++; 7880 ST.count++; 7881 curly_try_B_min_known: 7882 /* find the next place where 'B' could work, then call B */ 7883 { 7884 int n; 7885 if (utf8_target) { 7886 n = (ST.oldloc == locinput) ? 0 : 1; 7887 if (ST.c1 == ST.c2) { 7888 /* set n to utf8_distance(oldloc, locinput) */ 7889 while (locinput <= ST.maxpos 7890 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))) 7891 { 7892 locinput += UTF8SKIP(locinput); 7893 n++; 7894 } 7895 } 7896 else { 7897 /* set n to utf8_distance(oldloc, locinput) */ 7898 while (locinput <= ST.maxpos 7899 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) 7900 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) 7901 { 7902 locinput += UTF8SKIP(locinput); 7903 n++; 7904 } 7905 } 7906 } 7907 else { /* Not utf8_target */ 7908 if (ST.c1 == ST.c2) { 7909 while (locinput <= ST.maxpos && 7910 UCHARAT(locinput) != ST.c1) 7911 locinput++; 7912 } 7913 else { 7914 while (locinput <= ST.maxpos 7915 && UCHARAT(locinput) != ST.c1 7916 && UCHARAT(locinput) != ST.c2) 7917 locinput++; 7918 } 7919 n = locinput - ST.oldloc; 7920 } 7921 if (locinput > ST.maxpos) 7922 sayNO; 7923 if (n) { 7924 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is 7925 * at b; check that everything between oldloc and 7926 * locinput matches */ 7927 char *li = ST.oldloc; 7928 ST.count += n; 7929 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n) 7930 sayNO; 7931 assert(n == REG_INFTY || locinput == li); 7932 } 7933 CURLY_SETPAREN(ST.paren, ST.count); 7934 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) 7935 goto fake_end; 7936 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput); 7937 } 7938 NOT_REACHED; /* NOTREACHED */ 7939 7940 case CURLY_B_min_fail: 7941 /* failed to find B in a non-greedy match where c1,c2 invalid */ 7942 7943 REGCP_UNWIND(ST.cp); 7944 if (ST.paren) { 7945 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 7946 } 7947 /* failed -- move forward one */ 7948 { 7949 char *li = locinput; 7950 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) { 7951 sayNO; 7952 } 7953 locinput = li; 7954 } 7955 { 7956 ST.count++; 7957 if (ST.count <= ST.max || (ST.max == REG_INFTY && 7958 ST.count > 0)) /* count overflow ? */ 7959 { 7960 curly_try_B_min: 7961 CURLY_SETPAREN(ST.paren, ST.count); 7962 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) 7963 goto fake_end; 7964 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput); 7965 } 7966 } 7967 sayNO; 7968 NOT_REACHED; /* NOTREACHED */ 7969 7970 curly_try_B_max: 7971 /* a successful greedy match: now try to match B */ 7972 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren)) 7973 goto fake_end; 7974 { 7975 bool could_match = locinput < reginfo->strend; 7976 7977 /* If it could work, try it. */ 7978 if (ST.c1 != CHRTEST_VOID && could_match) { 7979 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target) 7980 { 7981 could_match = memEQ(locinput, 7982 ST.c1_utf8, 7983 UTF8SKIP(locinput)) 7984 || memEQ(locinput, 7985 ST.c2_utf8, 7986 UTF8SKIP(locinput)); 7987 } 7988 else { 7989 could_match = UCHARAT(locinput) == ST.c1 7990 || UCHARAT(locinput) == ST.c2; 7991 } 7992 } 7993 if (ST.c1 == CHRTEST_VOID || could_match) { 7994 CURLY_SETPAREN(ST.paren, ST.count); 7995 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput); 7996 NOT_REACHED; /* NOTREACHED */ 7997 } 7998 } 7999 /* FALLTHROUGH */ 8000 8001 case CURLY_B_max_fail: 8002 /* failed to find B in a greedy match */ 8003 8004 REGCP_UNWIND(ST.cp); 8005 if (ST.paren) { 8006 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 8007 } 8008 /* back up. */ 8009 if (--ST.count < ST.min) 8010 sayNO; 8011 locinput = HOPc(locinput, -1); 8012 goto curly_try_B_max; 8013 8014 #undef ST 8015 8016 case END: /* last op of main pattern */ 8017 fake_end: 8018 if (cur_eval) { 8019 /* we've just finished A in /(??{A})B/; now continue with B */ 8020 SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput); 8021 st->u.eval.prev_rex = rex_sv; /* inner */ 8022 8023 /* Save *all* the positions. */ 8024 st->u.eval.cp = regcppush(rex, 0, maxopenparen); 8025 rex_sv = CUR_EVAL.prev_rex; 8026 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); 8027 SET_reg_curpm(rex_sv); 8028 rex = ReANY(rex_sv); 8029 rexi = RXi_GET(rex); 8030 8031 st->u.eval.prev_curlyx = cur_curlyx; 8032 cur_curlyx = CUR_EVAL.prev_curlyx; 8033 8034 REGCP_SET(st->u.eval.lastcp); 8035 8036 /* Restore parens of the outer rex without popping the 8037 * savestack */ 8038 S_regcp_restore(aTHX_ rex, CUR_EVAL.lastcp, 8039 &maxopenparen); 8040 8041 st->u.eval.prev_eval = cur_eval; 8042 cur_eval = CUR_EVAL.prev_eval; 8043 DEBUG_EXECUTE_r( 8044 Perl_re_exec_indentf( aTHX_ "EVAL trying tail ... (cur_eval=%p)\n", 8045 depth, cur_eval);); 8046 if ( nochange_depth ) 8047 nochange_depth--; 8048 8049 SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput); 8050 8051 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B, 8052 locinput); /* match B */ 8053 } 8054 8055 if (locinput < reginfo->till) { 8056 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ 8057 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", 8058 PL_colors[4], 8059 (long)(locinput - startpos), 8060 (long)(reginfo->till - startpos), 8061 PL_colors[5])); 8062 8063 sayNO_SILENT; /* Cannot match: too short. */ 8064 } 8065 sayYES; /* Success! */ 8066 8067 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ 8068 DEBUG_EXECUTE_r( 8069 Perl_re_exec_indentf( aTHX_ "%ssubpattern success...%s\n", 8070 depth, PL_colors[4], PL_colors[5])); 8071 sayYES; /* Success! */ 8072 8073 #undef ST 8074 #define ST st->u.ifmatch 8075 8076 { 8077 char *newstart; 8078 8079 case SUSPEND: /* (?>A) */ 8080 ST.wanted = 1; 8081 newstart = locinput; 8082 goto do_ifmatch; 8083 8084 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */ 8085 ST.wanted = 0; 8086 goto ifmatch_trivial_fail_test; 8087 8088 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */ 8089 ST.wanted = 1; 8090 ifmatch_trivial_fail_test: 8091 if (scan->flags) { 8092 char * const s = HOPBACKc(locinput, scan->flags); 8093 if (!s) { 8094 /* trivial fail */ 8095 if (logical) { 8096 logical = 0; 8097 sw = 1 - cBOOL(ST.wanted); 8098 } 8099 else if (ST.wanted) 8100 sayNO; 8101 next = scan + ARG(scan); 8102 if (next == scan) 8103 next = NULL; 8104 break; 8105 } 8106 newstart = s; 8107 } 8108 else 8109 newstart = locinput; 8110 8111 do_ifmatch: 8112 ST.me = scan; 8113 ST.logical = logical; 8114 logical = 0; /* XXX: reset state of logical once it has been saved into ST */ 8115 8116 /* execute body of (?...A) */ 8117 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart); 8118 NOT_REACHED; /* NOTREACHED */ 8119 } 8120 8121 case IFMATCH_A_fail: /* body of (?...A) failed */ 8122 ST.wanted = !ST.wanted; 8123 /* FALLTHROUGH */ 8124 8125 case IFMATCH_A: /* body of (?...A) succeeded */ 8126 if (ST.logical) { 8127 sw = cBOOL(ST.wanted); 8128 } 8129 else if (!ST.wanted) 8130 sayNO; 8131 8132 if (OP(ST.me) != SUSPEND) { 8133 /* restore old position except for (?>...) */ 8134 locinput = st->locinput; 8135 } 8136 scan = ST.me + ARG(ST.me); 8137 if (scan == ST.me) 8138 scan = NULL; 8139 continue; /* execute B */ 8140 8141 #undef ST 8142 8143 case LONGJMP: /* alternative with many branches compiles to 8144 * (BRANCHJ; EXACT ...; LONGJMP ) x N */ 8145 next = scan + ARG(scan); 8146 if (next == scan) 8147 next = NULL; 8148 break; 8149 8150 case COMMIT: /* (*COMMIT) */ 8151 reginfo->cutpoint = reginfo->strend; 8152 /* FALLTHROUGH */ 8153 8154 case PRUNE: /* (*PRUNE) */ 8155 if (scan->flags) 8156 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 8157 PUSH_STATE_GOTO(COMMIT_next, next, locinput); 8158 NOT_REACHED; /* NOTREACHED */ 8159 8160 case COMMIT_next_fail: 8161 no_final = 1; 8162 /* FALLTHROUGH */ 8163 sayNO; 8164 NOT_REACHED; /* NOTREACHED */ 8165 8166 case OPFAIL: /* (*FAIL) */ 8167 if (scan->flags) 8168 sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 8169 if (logical) { 8170 /* deal with (?(?!)X|Y) properly, 8171 * make sure we trigger the no branch 8172 * of the trailing IFTHEN structure*/ 8173 sw= 0; 8174 break; 8175 } else { 8176 sayNO; 8177 } 8178 NOT_REACHED; /* NOTREACHED */ 8179 8180 #define ST st->u.mark 8181 case MARKPOINT: /* (*MARK:foo) */ 8182 ST.prev_mark = mark_state; 8183 ST.mark_name = sv_commit = sv_yes_mark 8184 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 8185 mark_state = st; 8186 ST.mark_loc = locinput; 8187 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput); 8188 NOT_REACHED; /* NOTREACHED */ 8189 8190 case MARKPOINT_next: 8191 mark_state = ST.prev_mark; 8192 sayYES; 8193 NOT_REACHED; /* NOTREACHED */ 8194 8195 case MARKPOINT_next_fail: 8196 if (popmark && sv_eq(ST.mark_name,popmark)) 8197 { 8198 if (ST.mark_loc > startpoint) 8199 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); 8200 popmark = NULL; /* we found our mark */ 8201 sv_commit = ST.mark_name; 8202 8203 DEBUG_EXECUTE_r({ 8204 Perl_re_exec_indentf( aTHX_ "%ssetting cutpoint to mark:%"SVf"...%s\n", 8205 depth, 8206 PL_colors[4], SVfARG(sv_commit), PL_colors[5]); 8207 }); 8208 } 8209 mark_state = ST.prev_mark; 8210 sv_yes_mark = mark_state ? 8211 mark_state->u.mark.mark_name : NULL; 8212 sayNO; 8213 NOT_REACHED; /* NOTREACHED */ 8214 8215 case SKIP: /* (*SKIP) */ 8216 if (!scan->flags) { 8217 /* (*SKIP) : if we fail we cut here*/ 8218 ST.mark_name = NULL; 8219 ST.mark_loc = locinput; 8220 PUSH_STATE_GOTO(SKIP_next,next, locinput); 8221 } else { 8222 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was, 8223 otherwise do nothing. Meaning we need to scan 8224 */ 8225 regmatch_state *cur = mark_state; 8226 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 8227 8228 while (cur) { 8229 if ( sv_eq( cur->u.mark.mark_name, 8230 find ) ) 8231 { 8232 ST.mark_name = find; 8233 PUSH_STATE_GOTO( SKIP_next, next, locinput); 8234 } 8235 cur = cur->u.mark.prev_mark; 8236 } 8237 } 8238 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */ 8239 break; 8240 8241 case SKIP_next_fail: 8242 if (ST.mark_name) { 8243 /* (*CUT:NAME) - Set up to search for the name as we 8244 collapse the stack*/ 8245 popmark = ST.mark_name; 8246 } else { 8247 /* (*CUT) - No name, we cut here.*/ 8248 if (ST.mark_loc > startpoint) 8249 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); 8250 /* but we set sv_commit to latest mark_name if there 8251 is one so they can test to see how things lead to this 8252 cut */ 8253 if (mark_state) 8254 sv_commit=mark_state->u.mark.mark_name; 8255 } 8256 no_final = 1; 8257 sayNO; 8258 NOT_REACHED; /* NOTREACHED */ 8259 #undef ST 8260 8261 case LNBREAK: /* \R */ 8262 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) { 8263 locinput += n; 8264 } else 8265 sayNO; 8266 break; 8267 8268 default: 8269 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n", 8270 PTR2UV(scan), OP(scan)); 8271 Perl_croak(aTHX_ "regexp memory corruption"); 8272 8273 /* this is a point to jump to in order to increment 8274 * locinput by one character */ 8275 increment_locinput: 8276 assert(!NEXTCHR_IS_EOS); 8277 if (utf8_target) { 8278 locinput += PL_utf8skip[nextchr]; 8279 /* locinput is allowed to go 1 char off the end, but not 2+ */ 8280 if (locinput > reginfo->strend) 8281 sayNO; 8282 } 8283 else 8284 locinput++; 8285 break; 8286 8287 } /* end switch */ 8288 8289 /* switch break jumps here */ 8290 scan = next; /* prepare to execute the next op and ... */ 8291 continue; /* ... jump back to the top, reusing st */ 8292 /* NOTREACHED */ 8293 8294 push_yes_state: 8295 /* push a state that backtracks on success */ 8296 st->u.yes.prev_yes_state = yes_state; 8297 yes_state = st; 8298 /* FALLTHROUGH */ 8299 push_state: 8300 /* push a new regex state, then continue at scan */ 8301 { 8302 regmatch_state *newst; 8303 8304 DEBUG_STACK_r({ 8305 regmatch_state *cur = st; 8306 regmatch_state *curyes = yes_state; 8307 int curd = depth; 8308 regmatch_slab *slab = PL_regmatch_slab; 8309 for (;curd > -1 && (depth-curd < 3);cur--,curd--) { 8310 if (cur < SLAB_FIRST(slab)) { 8311 slab = slab->prev; 8312 cur = SLAB_LAST(slab); 8313 } 8314 Perl_re_exec_indentf( aTHX_ "#%-3d %-10s %s\n", 8315 depth, 8316 curd, PL_reg_name[cur->resume_state], 8317 (curyes == cur) ? "yes" : "" 8318 ); 8319 if (curyes == cur) 8320 curyes = cur->u.yes.prev_yes_state; 8321 } 8322 } else 8323 DEBUG_STATE_pp("push") 8324 ); 8325 depth++; 8326 st->locinput = locinput; 8327 newst = st+1; 8328 if (newst > SLAB_LAST(PL_regmatch_slab)) 8329 newst = S_push_slab(aTHX); 8330 PL_regmatch_state = newst; 8331 8332 locinput = pushinput; 8333 st = newst; 8334 continue; 8335 /* NOTREACHED */ 8336 } 8337 } 8338 #ifdef SOLARIS_BAD_OPTIMIZER 8339 # undef PL_charclass 8340 #endif 8341 8342 /* 8343 * We get here only if there's trouble -- normally "case END" is 8344 * the terminating point. 8345 */ 8346 Perl_croak(aTHX_ "corrupted regexp pointers"); 8347 NOT_REACHED; /* NOTREACHED */ 8348 8349 yes: 8350 if (yes_state) { 8351 /* we have successfully completed a subexpression, but we must now 8352 * pop to the state marked by yes_state and continue from there */ 8353 assert(st != yes_state); 8354 #ifdef DEBUGGING 8355 while (st != yes_state) { 8356 st--; 8357 if (st < SLAB_FIRST(PL_regmatch_slab)) { 8358 PL_regmatch_slab = PL_regmatch_slab->prev; 8359 st = SLAB_LAST(PL_regmatch_slab); 8360 } 8361 DEBUG_STATE_r({ 8362 if (no_final) { 8363 DEBUG_STATE_pp("pop (no final)"); 8364 } else { 8365 DEBUG_STATE_pp("pop (yes)"); 8366 } 8367 }); 8368 depth--; 8369 } 8370 #else 8371 while (yes_state < SLAB_FIRST(PL_regmatch_slab) 8372 || yes_state > SLAB_LAST(PL_regmatch_slab)) 8373 { 8374 /* not in this slab, pop slab */ 8375 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1); 8376 PL_regmatch_slab = PL_regmatch_slab->prev; 8377 st = SLAB_LAST(PL_regmatch_slab); 8378 } 8379 depth -= (st - yes_state); 8380 #endif 8381 st = yes_state; 8382 yes_state = st->u.yes.prev_yes_state; 8383 PL_regmatch_state = st; 8384 8385 if (no_final) 8386 locinput= st->locinput; 8387 state_num = st->resume_state + no_final; 8388 goto reenter_switch; 8389 } 8390 8391 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n", 8392 PL_colors[4], PL_colors[5])); 8393 8394 if (reginfo->info_aux_eval) { 8395 /* each successfully executed (?{...}) block does the equivalent of 8396 * local $^R = do {...} 8397 * When popping the save stack, all these locals would be undone; 8398 * bypass this by setting the outermost saved $^R to the latest 8399 * value */ 8400 /* I dont know if this is needed or works properly now. 8401 * see code related to PL_replgv elsewhere in this file. 8402 * Yves 8403 */ 8404 if (oreplsv != GvSV(PL_replgv)) 8405 sv_setsv(oreplsv, GvSV(PL_replgv)); 8406 } 8407 result = 1; 8408 goto final_exit; 8409 8410 no: 8411 DEBUG_EXECUTE_r( 8412 Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n", 8413 depth, 8414 PL_colors[4], PL_colors[5]) 8415 ); 8416 8417 no_silent: 8418 if (no_final) { 8419 if (yes_state) { 8420 goto yes; 8421 } else { 8422 goto final_exit; 8423 } 8424 } 8425 if (depth) { 8426 /* there's a previous state to backtrack to */ 8427 st--; 8428 if (st < SLAB_FIRST(PL_regmatch_slab)) { 8429 PL_regmatch_slab = PL_regmatch_slab->prev; 8430 st = SLAB_LAST(PL_regmatch_slab); 8431 } 8432 PL_regmatch_state = st; 8433 locinput= st->locinput; 8434 8435 DEBUG_STATE_pp("pop"); 8436 depth--; 8437 if (yes_state == st) 8438 yes_state = st->u.yes.prev_yes_state; 8439 8440 state_num = st->resume_state + 1; /* failure = success + 1 */ 8441 goto reenter_switch; 8442 } 8443 result = 0; 8444 8445 final_exit: 8446 if (rex->intflags & PREGf_VERBARG_SEEN) { 8447 SV *sv_err = get_sv("REGERROR", 1); 8448 SV *sv_mrk = get_sv("REGMARK", 1); 8449 if (result) { 8450 sv_commit = &PL_sv_no; 8451 if (!sv_yes_mark) 8452 sv_yes_mark = &PL_sv_yes; 8453 } else { 8454 if (!sv_commit) 8455 sv_commit = &PL_sv_yes; 8456 sv_yes_mark = &PL_sv_no; 8457 } 8458 assert(sv_err); 8459 assert(sv_mrk); 8460 sv_setsv(sv_err, sv_commit); 8461 sv_setsv(sv_mrk, sv_yes_mark); 8462 } 8463 8464 8465 if (last_pushed_cv) { 8466 dSP; 8467 POP_MULTICALL; 8468 PERL_UNUSED_VAR(SP); 8469 } 8470 8471 assert(!result || locinput - reginfo->strbeg >= 0); 8472 return result ? locinput - reginfo->strbeg : -1; 8473 } 8474 8475 /* 8476 - regrepeat - repeatedly match something simple, report how many 8477 * 8478 * What 'simple' means is a node which can be the operand of a quantifier like 8479 * '+', or {1,3} 8480 * 8481 * startposp - pointer a pointer to the start position. This is updated 8482 * to point to the byte following the highest successful 8483 * match. 8484 * p - the regnode to be repeatedly matched against. 8485 * reginfo - struct holding match state, such as strend 8486 * max - maximum number of things to match. 8487 * depth - (for debugging) backtracking depth. 8488 */ 8489 STATIC I32 8490 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p, 8491 regmatch_info *const reginfo, I32 max, int depth) 8492 { 8493 char *scan; /* Pointer to current position in target string */ 8494 I32 c; 8495 char *loceol = reginfo->strend; /* local version */ 8496 I32 hardcount = 0; /* How many matches so far */ 8497 bool utf8_target = reginfo->is_utf8_target; 8498 unsigned int to_complement = 0; /* Invert the result? */ 8499 UV utf8_flags; 8500 _char_class_number classnum; 8501 #ifndef DEBUGGING 8502 PERL_UNUSED_ARG(depth); 8503 #endif 8504 8505 PERL_ARGS_ASSERT_REGREPEAT; 8506 8507 scan = *startposp; 8508 if (max == REG_INFTY) 8509 max = I32_MAX; 8510 else if (! utf8_target && loceol - scan > max) 8511 loceol = scan + max; 8512 8513 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down 8514 * to the maximum of how far we should go in it (leaving it set to the real 8515 * end, if the maximum permissible would take us beyond that). This allows 8516 * us to make the loop exit condition that we haven't gone past <loceol> to 8517 * also mean that we haven't exceeded the max permissible count, saving a 8518 * test each time through the loop. But it assumes that the OP matches a 8519 * single byte, which is true for most of the OPs below when applied to a 8520 * non-UTF-8 target. Those relatively few OPs that don't have this 8521 * characteristic will have to compensate. 8522 * 8523 * There is no adjustment for UTF-8 targets, as the number of bytes per 8524 * character varies. OPs will have to test both that the count is less 8525 * than the max permissible (using <hardcount> to keep track), and that we 8526 * are still within the bounds of the string (using <loceol>. A few OPs 8527 * match a single byte no matter what the encoding. They can omit the max 8528 * test if, for the UTF-8 case, they do the adjustment that was skipped 8529 * above. 8530 * 8531 * Thus, the code above sets things up for the common case; and exceptional 8532 * cases need extra work; the common case is to make sure <scan> doesn't 8533 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the 8534 * count doesn't exceed the maximum permissible */ 8535 8536 switch (OP(p)) { 8537 case REG_ANY: 8538 if (utf8_target) { 8539 while (scan < loceol && hardcount < max && *scan != '\n') { 8540 scan += UTF8SKIP(scan); 8541 hardcount++; 8542 } 8543 } else { 8544 while (scan < loceol && *scan != '\n') 8545 scan++; 8546 } 8547 break; 8548 case SANY: 8549 if (utf8_target) { 8550 while (scan < loceol && hardcount < max) { 8551 scan += UTF8SKIP(scan); 8552 hardcount++; 8553 } 8554 } 8555 else 8556 scan = loceol; 8557 break; 8558 case EXACTL: 8559 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 8560 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) { 8561 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol); 8562 } 8563 /* FALLTHROUGH */ 8564 case EXACT: 8565 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); 8566 8567 c = (U8)*STRING(p); 8568 8569 /* Can use a simple loop if the pattern char to match on is invariant 8570 * under UTF-8, or both target and pattern aren't UTF-8. Note that we 8571 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's 8572 * true iff it doesn't matter if the argument is in UTF-8 or not */ 8573 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) { 8574 if (utf8_target && loceol - scan > max) { 8575 /* We didn't adjust <loceol> because is UTF-8, but ok to do so, 8576 * since here, to match at all, 1 char == 1 byte */ 8577 loceol = scan + max; 8578 } 8579 while (scan < loceol && UCHARAT(scan) == c) { 8580 scan++; 8581 } 8582 } 8583 else if (reginfo->is_utf8_pat) { 8584 if (utf8_target) { 8585 STRLEN scan_char_len; 8586 8587 /* When both target and pattern are UTF-8, we have to do 8588 * string EQ */ 8589 while (hardcount < max 8590 && scan < loceol 8591 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p) 8592 && memEQ(scan, STRING(p), scan_char_len)) 8593 { 8594 scan += scan_char_len; 8595 hardcount++; 8596 } 8597 } 8598 else if (! UTF8_IS_ABOVE_LATIN1(c)) { 8599 8600 /* Target isn't utf8; convert the character in the UTF-8 8601 * pattern to non-UTF8, and do a simple loop */ 8602 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1)); 8603 while (scan < loceol && UCHARAT(scan) == c) { 8604 scan++; 8605 } 8606 } /* else pattern char is above Latin1, can't possibly match the 8607 non-UTF-8 target */ 8608 } 8609 else { 8610 8611 /* Here, the string must be utf8; pattern isn't, and <c> is 8612 * different in utf8 than not, so can't compare them directly. 8613 * Outside the loop, find the two utf8 bytes that represent c, and 8614 * then look for those in sequence in the utf8 string */ 8615 U8 high = UTF8_TWO_BYTE_HI(c); 8616 U8 low = UTF8_TWO_BYTE_LO(c); 8617 8618 while (hardcount < max 8619 && scan + 1 < loceol 8620 && UCHARAT(scan) == high 8621 && UCHARAT(scan + 1) == low) 8622 { 8623 scan += 2; 8624 hardcount++; 8625 } 8626 } 8627 break; 8628 8629 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ 8630 assert(! reginfo->is_utf8_pat); 8631 /* FALLTHROUGH */ 8632 case EXACTFA: 8633 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; 8634 goto do_exactf; 8635 8636 case EXACTFL: 8637 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 8638 utf8_flags = FOLDEQ_LOCALE; 8639 goto do_exactf; 8640 8641 case EXACTF: /* This node only generated for non-utf8 patterns */ 8642 assert(! reginfo->is_utf8_pat); 8643 utf8_flags = 0; 8644 goto do_exactf; 8645 8646 case EXACTFLU8: 8647 if (! utf8_target) { 8648 break; 8649 } 8650 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED 8651 | FOLDEQ_S2_FOLDS_SANE; 8652 goto do_exactf; 8653 8654 case EXACTFU_SS: 8655 case EXACTFU: 8656 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; 8657 8658 do_exactf: { 8659 int c1, c2; 8660 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1]; 8661 8662 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); 8663 8664 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8, 8665 reginfo)) 8666 { 8667 if (c1 == CHRTEST_VOID) { 8668 /* Use full Unicode fold matching */ 8669 char *tmpeol = reginfo->strend; 8670 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1; 8671 while (hardcount < max 8672 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target, 8673 STRING(p), NULL, pat_len, 8674 reginfo->is_utf8_pat, utf8_flags)) 8675 { 8676 scan = tmpeol; 8677 tmpeol = reginfo->strend; 8678 hardcount++; 8679 } 8680 } 8681 else if (utf8_target) { 8682 if (c1 == c2) { 8683 while (scan < loceol 8684 && hardcount < max 8685 && memEQ(scan, c1_utf8, UTF8SKIP(scan))) 8686 { 8687 scan += UTF8SKIP(scan); 8688 hardcount++; 8689 } 8690 } 8691 else { 8692 while (scan < loceol 8693 && hardcount < max 8694 && (memEQ(scan, c1_utf8, UTF8SKIP(scan)) 8695 || memEQ(scan, c2_utf8, UTF8SKIP(scan)))) 8696 { 8697 scan += UTF8SKIP(scan); 8698 hardcount++; 8699 } 8700 } 8701 } 8702 else if (c1 == c2) { 8703 while (scan < loceol && UCHARAT(scan) == c1) { 8704 scan++; 8705 } 8706 } 8707 else { 8708 while (scan < loceol && 8709 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2)) 8710 { 8711 scan++; 8712 } 8713 } 8714 } 8715 break; 8716 } 8717 case ANYOFL: 8718 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 8719 8720 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) { 8721 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required); 8722 } 8723 /* FALLTHROUGH */ 8724 case ANYOFD: 8725 case ANYOF: 8726 if (utf8_target) { 8727 while (hardcount < max 8728 && scan < loceol 8729 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target)) 8730 { 8731 scan += UTF8SKIP(scan); 8732 hardcount++; 8733 } 8734 } else { 8735 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan, 0)) 8736 scan++; 8737 } 8738 break; 8739 8740 /* The argument (FLAGS) to all the POSIX node types is the class number */ 8741 8742 case NPOSIXL: 8743 to_complement = 1; 8744 /* FALLTHROUGH */ 8745 8746 case POSIXL: 8747 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 8748 if (! utf8_target) { 8749 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p), 8750 *scan))) 8751 { 8752 scan++; 8753 } 8754 } else { 8755 while (hardcount < max && scan < loceol 8756 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p), 8757 (U8 *) scan))) 8758 { 8759 scan += UTF8SKIP(scan); 8760 hardcount++; 8761 } 8762 } 8763 break; 8764 8765 case POSIXD: 8766 if (utf8_target) { 8767 goto utf8_posix; 8768 } 8769 /* FALLTHROUGH */ 8770 8771 case POSIXA: 8772 if (utf8_target && loceol - scan > max) { 8773 8774 /* We didn't adjust <loceol> at the beginning of this routine 8775 * because is UTF-8, but it is actually ok to do so, since here, to 8776 * match, 1 char == 1 byte. */ 8777 loceol = scan + max; 8778 } 8779 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) { 8780 scan++; 8781 } 8782 break; 8783 8784 case NPOSIXD: 8785 if (utf8_target) { 8786 to_complement = 1; 8787 goto utf8_posix; 8788 } 8789 /* FALLTHROUGH */ 8790 8791 case NPOSIXA: 8792 if (! utf8_target) { 8793 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) { 8794 scan++; 8795 } 8796 } 8797 else { 8798 8799 /* The complement of something that matches only ASCII matches all 8800 * non-ASCII, plus everything in ASCII that isn't in the class. */ 8801 while (hardcount < max && scan < loceol 8802 && (! isASCII_utf8(scan) 8803 || ! _generic_isCC_A((U8) *scan, FLAGS(p)))) 8804 { 8805 scan += UTF8SKIP(scan); 8806 hardcount++; 8807 } 8808 } 8809 break; 8810 8811 case NPOSIXU: 8812 to_complement = 1; 8813 /* FALLTHROUGH */ 8814 8815 case POSIXU: 8816 if (! utf8_target) { 8817 while (scan < loceol && to_complement 8818 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p)))) 8819 { 8820 scan++; 8821 } 8822 } 8823 else { 8824 utf8_posix: 8825 classnum = (_char_class_number) FLAGS(p); 8826 if (classnum < _FIRST_NON_SWASH_CC) { 8827 8828 /* Here, a swash is needed for above-Latin1 code points. 8829 * Process as many Latin1 code points using the built-in rules. 8830 * Go to another loop to finish processing upon encountering 8831 * the first Latin1 code point. We could do that in this loop 8832 * as well, but the other way saves having to test if the swash 8833 * has been loaded every time through the loop: extra space to 8834 * save a test. */ 8835 while (hardcount < max && scan < loceol) { 8836 if (UTF8_IS_INVARIANT(*scan)) { 8837 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan, 8838 classnum)))) 8839 { 8840 break; 8841 } 8842 scan++; 8843 } 8844 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) { 8845 if (! (to_complement 8846 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*scan, 8847 *(scan + 1)), 8848 classnum)))) 8849 { 8850 break; 8851 } 8852 scan += 2; 8853 } 8854 else { 8855 goto found_above_latin1; 8856 } 8857 8858 hardcount++; 8859 } 8860 } 8861 else { 8862 /* For these character classes, the knowledge of how to handle 8863 * every code point is compiled in to Perl via a macro. This 8864 * code is written for making the loops as tight as possible. 8865 * It could be refactored to save space instead */ 8866 switch (classnum) { 8867 case _CC_ENUM_SPACE: 8868 while (hardcount < max 8869 && scan < loceol 8870 && (to_complement ^ cBOOL(isSPACE_utf8(scan)))) 8871 { 8872 scan += UTF8SKIP(scan); 8873 hardcount++; 8874 } 8875 break; 8876 case _CC_ENUM_BLANK: 8877 while (hardcount < max 8878 && scan < loceol 8879 && (to_complement ^ cBOOL(isBLANK_utf8(scan)))) 8880 { 8881 scan += UTF8SKIP(scan); 8882 hardcount++; 8883 } 8884 break; 8885 case _CC_ENUM_XDIGIT: 8886 while (hardcount < max 8887 && scan < loceol 8888 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan)))) 8889 { 8890 scan += UTF8SKIP(scan); 8891 hardcount++; 8892 } 8893 break; 8894 case _CC_ENUM_VERTSPACE: 8895 while (hardcount < max 8896 && scan < loceol 8897 && (to_complement ^ cBOOL(isVERTWS_utf8(scan)))) 8898 { 8899 scan += UTF8SKIP(scan); 8900 hardcount++; 8901 } 8902 break; 8903 case _CC_ENUM_CNTRL: 8904 while (hardcount < max 8905 && scan < loceol 8906 && (to_complement ^ cBOOL(isCNTRL_utf8(scan)))) 8907 { 8908 scan += UTF8SKIP(scan); 8909 hardcount++; 8910 } 8911 break; 8912 default: 8913 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum); 8914 } 8915 } 8916 } 8917 break; 8918 8919 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */ 8920 8921 /* Load the swash if not already present */ 8922 if (! PL_utf8_swash_ptrs[classnum]) { 8923 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 8924 PL_utf8_swash_ptrs[classnum] = _core_swash_init( 8925 "utf8", 8926 "", 8927 &PL_sv_undef, 1, 0, 8928 PL_XPosix_ptrs[classnum], &flags); 8929 } 8930 8931 while (hardcount < max && scan < loceol 8932 && to_complement ^ cBOOL(_generic_utf8( 8933 classnum, 8934 scan, 8935 swash_fetch(PL_utf8_swash_ptrs[classnum], 8936 (U8 *) scan, 8937 TRUE)))) 8938 { 8939 scan += UTF8SKIP(scan); 8940 hardcount++; 8941 } 8942 break; 8943 8944 case LNBREAK: 8945 if (utf8_target) { 8946 while (hardcount < max && scan < loceol && 8947 (c=is_LNBREAK_utf8_safe(scan, loceol))) { 8948 scan += c; 8949 hardcount++; 8950 } 8951 } else { 8952 /* LNBREAK can match one or two latin chars, which is ok, but we 8953 * have to use hardcount in this situation, and throw away the 8954 * adjustment to <loceol> done before the switch statement */ 8955 loceol = reginfo->strend; 8956 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) { 8957 scan+=c; 8958 hardcount++; 8959 } 8960 } 8961 break; 8962 8963 case BOUNDL: 8964 case NBOUNDL: 8965 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; 8966 /* FALLTHROUGH */ 8967 case BOUND: 8968 case BOUNDA: 8969 case BOUNDU: 8970 case EOS: 8971 case GPOS: 8972 case KEEPS: 8973 case NBOUND: 8974 case NBOUNDA: 8975 case NBOUNDU: 8976 case OPFAIL: 8977 case SBOL: 8978 case SEOL: 8979 /* These are all 0 width, so match right here or not at all. */ 8980 break; 8981 8982 default: 8983 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]); 8984 NOT_REACHED; /* NOTREACHED */ 8985 8986 } 8987 8988 if (hardcount) 8989 c = hardcount; 8990 else 8991 c = scan - *startposp; 8992 *startposp = scan; 8993 8994 DEBUG_r({ 8995 GET_RE_DEBUG_FLAGS_DECL; 8996 DEBUG_EXECUTE_r({ 8997 SV * const prop = sv_newmortal(); 8998 regprop(prog, prop, p, reginfo, NULL); 8999 Perl_re_exec_indentf( aTHX_ "%s can match %"IVdf" times out of %"IVdf"...\n", 9000 depth, SvPVX_const(prop),(IV)c,(IV)max); 9001 }); 9002 }); 9003 9004 return(c); 9005 } 9006 9007 9008 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) 9009 /* 9010 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to 9011 create a copy so that changes the caller makes won't change the shared one. 9012 If <altsvp> is non-null, will return NULL in it, for back-compat. 9013 */ 9014 SV * 9015 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp) 9016 { 9017 PERL_ARGS_ASSERT_REGCLASS_SWASH; 9018 9019 if (altsvp) { 9020 *altsvp = NULL; 9021 } 9022 9023 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL)); 9024 } 9025 9026 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */ 9027 9028 /* 9029 - reginclass - determine if a character falls into a character class 9030 9031 n is the ANYOF-type regnode 9032 p is the target string 9033 p_end points to one byte beyond the end of the target string 9034 utf8_target tells whether p is in UTF-8. 9035 9036 Returns true if matched; false otherwise. 9037 9038 Note that this can be a synthetic start class, a combination of various 9039 nodes, so things you think might be mutually exclusive, such as locale, 9040 aren't. It can match both locale and non-locale 9041 9042 */ 9043 9044 STATIC bool 9045 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target) 9046 { 9047 dVAR; 9048 const char flags = ANYOF_FLAGS(n); 9049 bool match = FALSE; 9050 UV c = *p; 9051 9052 PERL_ARGS_ASSERT_REGINCLASS; 9053 9054 /* If c is not already the code point, get it. Note that 9055 * UTF8_IS_INVARIANT() works even if not in UTF-8 */ 9056 if (! UTF8_IS_INVARIANT(c) && utf8_target) { 9057 STRLEN c_len = 0; 9058 c = utf8n_to_uvchr(p, p_end - p, &c_len, 9059 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) 9060 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY); 9061 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for 9062 * UTF8_ALLOW_FFFF */ 9063 if (c_len == (STRLEN)-1) 9064 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); 9065 if (c > 255 && OP(n) == ANYOFL && ! ANYOFL_UTF8_LOCALE_REQD(flags)) { 9066 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c); 9067 } 9068 } 9069 9070 /* If this character is potentially in the bitmap, check it */ 9071 if (c < NUM_ANYOF_CODE_POINTS) { 9072 if (ANYOF_BITMAP_TEST(n, c)) 9073 match = TRUE; 9074 else if ((flags 9075 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER) 9076 && OP(n) == ANYOFD 9077 && ! utf8_target 9078 && ! isASCII(c)) 9079 { 9080 match = TRUE; 9081 } 9082 else if (flags & ANYOF_LOCALE_FLAGS) { 9083 if ((flags & ANYOFL_FOLD) 9084 && c < 256 9085 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) 9086 { 9087 match = TRUE; 9088 } 9089 else if (ANYOF_POSIXL_TEST_ANY_SET(n) 9090 && c < 256 9091 ) { 9092 9093 /* The data structure is arranged so bits 0, 2, 4, ... are set 9094 * if the class includes the Posix character class given by 9095 * bit/2; and 1, 3, 5, ... are set if the class includes the 9096 * complemented Posix class given by int(bit/2). So we loop 9097 * through the bits, each time changing whether we complement 9098 * the result or not. Suppose for the sake of illustration 9099 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0 9100 * is set, it means there is a match for this ANYOF node if the 9101 * character is in the class given by the expression (0 / 2 = 0 9102 * = \w). If it is in that class, isFOO_lc() will return 1, 9103 * and since 'to_complement' is 0, the result will stay TRUE, 9104 * and we exit the loop. Suppose instead that bit 0 is 0, but 9105 * bit 1 is 1. That means there is a match if the character 9106 * matches \W. We won't bother to call isFOO_lc() on bit 0, 9107 * but will on bit 1. On the second iteration 'to_complement' 9108 * will be 1, so the exclusive or will reverse things, so we 9109 * are testing for \W. On the third iteration, 'to_complement' 9110 * will be 0, and we would be testing for \s; the fourth 9111 * iteration would test for \S, etc. 9112 * 9113 * Note that this code assumes that all the classes are closed 9114 * under folding. For example, if a character matches \w, then 9115 * its fold does too; and vice versa. This should be true for 9116 * any well-behaved locale for all the currently defined Posix 9117 * classes, except for :lower: and :upper:, which are handled 9118 * by the pseudo-class :cased: which matches if either of the 9119 * other two does. To get rid of this assumption, an outer 9120 * loop could be used below to iterate over both the source 9121 * character, and its fold (if different) */ 9122 9123 int count = 0; 9124 int to_complement = 0; 9125 9126 while (count < ANYOF_MAX) { 9127 if (ANYOF_POSIXL_TEST(n, count) 9128 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c))) 9129 { 9130 match = TRUE; 9131 break; 9132 } 9133 count++; 9134 to_complement ^= 1; 9135 } 9136 } 9137 } 9138 } 9139 9140 9141 /* If the bitmap didn't (or couldn't) match, and something outside the 9142 * bitmap could match, try that. */ 9143 if (!match) { 9144 if (c >= NUM_ANYOF_CODE_POINTS 9145 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP)) 9146 { 9147 match = TRUE; /* Everything above the bitmap matches */ 9148 } 9149 /* Here doesn't match everything above the bitmap. If there is 9150 * some information available beyond the bitmap, we may find a 9151 * match in it. If so, this is most likely because the code point 9152 * is outside the bitmap range. But rarely, it could be because of 9153 * some other reason. If so, various flags are set to indicate 9154 * this possibility. On ANYOFD nodes, there may be matches that 9155 * happen only when the target string is UTF-8; or for other node 9156 * types, because runtime lookup is needed, regardless of the 9157 * UTF-8ness of the target string. Finally, under /il, there may 9158 * be some matches only possible if the locale is a UTF-8 one. */ 9159 else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP 9160 && ( c >= NUM_ANYOF_CODE_POINTS 9161 || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP) 9162 && ( UNLIKELY(OP(n) != ANYOFD) 9163 || (utf8_target && ! isASCII_uni(c) 9164 # if NUM_ANYOF_CODE_POINTS > 256 9165 && c < 256 9166 # endif 9167 ))) 9168 || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) 9169 && IN_UTF8_CTYPE_LOCALE))) 9170 { 9171 SV* only_utf8_locale = NULL; 9172 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0, 9173 &only_utf8_locale, NULL); 9174 if (sw) { 9175 U8 utf8_buffer[2]; 9176 U8 * utf8_p; 9177 if (utf8_target) { 9178 utf8_p = (U8 *) p; 9179 } else { /* Convert to utf8 */ 9180 utf8_p = utf8_buffer; 9181 append_utf8_from_native_byte(*p, &utf8_p); 9182 utf8_p = utf8_buffer; 9183 } 9184 9185 if (swash_fetch(sw, utf8_p, TRUE)) { 9186 match = TRUE; 9187 } 9188 } 9189 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) { 9190 match = _invlist_contains_cp(only_utf8_locale, c); 9191 } 9192 } 9193 9194 if (UNICODE_IS_SUPER(c) 9195 && (flags 9196 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER) 9197 && OP(n) != ANYOFD 9198 && ckWARN_d(WARN_NON_UNICODE)) 9199 { 9200 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), 9201 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c); 9202 } 9203 } 9204 9205 #if ANYOF_INVERT != 1 9206 /* Depending on compiler optimization cBOOL takes time, so if don't have to 9207 * use it, don't */ 9208 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below, 9209 #endif 9210 9211 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */ 9212 return (flags & ANYOF_INVERT) ^ match; 9213 } 9214 9215 STATIC U8 * 9216 S_reghop3(U8 *s, SSize_t off, const U8* lim) 9217 { 9218 /* return the position 'off' UTF-8 characters away from 's', forward if 9219 * 'off' >= 0, backwards if negative. But don't go outside of position 9220 * 'lim', which better be < s if off < 0 */ 9221 9222 PERL_ARGS_ASSERT_REGHOP3; 9223 9224 if (off >= 0) { 9225 while (off-- && s < lim) { 9226 /* XXX could check well-formedness here */ 9227 s += UTF8SKIP(s); 9228 } 9229 } 9230 else { 9231 while (off++ && s > lim) { 9232 s--; 9233 if (UTF8_IS_CONTINUED(*s)) { 9234 while (s > lim && UTF8_IS_CONTINUATION(*s)) 9235 s--; 9236 if (! UTF8_IS_START(*s)) { 9237 Perl_croak_nocontext("Malformed UTF-8 character (fatal)"); 9238 } 9239 } 9240 /* XXX could check well-formedness here */ 9241 } 9242 } 9243 return s; 9244 } 9245 9246 STATIC U8 * 9247 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim) 9248 { 9249 PERL_ARGS_ASSERT_REGHOP4; 9250 9251 if (off >= 0) { 9252 while (off-- && s < rlim) { 9253 /* XXX could check well-formedness here */ 9254 s += UTF8SKIP(s); 9255 } 9256 } 9257 else { 9258 while (off++ && s > llim) { 9259 s--; 9260 if (UTF8_IS_CONTINUED(*s)) { 9261 while (s > llim && UTF8_IS_CONTINUATION(*s)) 9262 s--; 9263 if (! UTF8_IS_START(*s)) { 9264 Perl_croak_nocontext("Malformed UTF-8 character (fatal)"); 9265 } 9266 } 9267 /* XXX could check well-formedness here */ 9268 } 9269 } 9270 return s; 9271 } 9272 9273 /* like reghop3, but returns NULL on overrun, rather than returning last 9274 * char pos */ 9275 9276 STATIC U8 * 9277 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim) 9278 { 9279 PERL_ARGS_ASSERT_REGHOPMAYBE3; 9280 9281 if (off >= 0) { 9282 while (off-- && s < lim) { 9283 /* XXX could check well-formedness here */ 9284 s += UTF8SKIP(s); 9285 } 9286 if (off >= 0) 9287 return NULL; 9288 } 9289 else { 9290 while (off++ && s > lim) { 9291 s--; 9292 if (UTF8_IS_CONTINUED(*s)) { 9293 while (s > lim && UTF8_IS_CONTINUATION(*s)) 9294 s--; 9295 if (! UTF8_IS_START(*s)) { 9296 Perl_croak_nocontext("Malformed UTF-8 character (fatal)"); 9297 } 9298 } 9299 /* XXX could check well-formedness here */ 9300 } 9301 if (off <= 0) 9302 return NULL; 9303 } 9304 return s; 9305 } 9306 9307 9308 /* when executing a regex that may have (?{}), extra stuff needs setting 9309 up that will be visible to the called code, even before the current 9310 match has finished. In particular: 9311 9312 * $_ is localised to the SV currently being matched; 9313 * pos($_) is created if necessary, ready to be updated on each call-out 9314 to code; 9315 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm 9316 isn't set until the current pattern is successfully finished), so that 9317 $1 etc of the match-so-far can be seen; 9318 * save the old values of subbeg etc of the current regex, and set then 9319 to the current string (again, this is normally only done at the end 9320 of execution) 9321 */ 9322 9323 static void 9324 S_setup_eval_state(pTHX_ regmatch_info *const reginfo) 9325 { 9326 MAGIC *mg; 9327 regexp *const rex = ReANY(reginfo->prog); 9328 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval; 9329 9330 eval_state->rex = rex; 9331 9332 if (reginfo->sv) { 9333 /* Make $_ available to executed code. */ 9334 if (reginfo->sv != DEFSV) { 9335 SAVE_DEFSV; 9336 DEFSV_set(reginfo->sv); 9337 } 9338 9339 if (!(mg = mg_find_mglob(reginfo->sv))) { 9340 /* prepare for quick setting of pos */ 9341 mg = sv_magicext_mglob(reginfo->sv); 9342 mg->mg_len = -1; 9343 } 9344 eval_state->pos_magic = mg; 9345 eval_state->pos = mg->mg_len; 9346 eval_state->pos_flags = mg->mg_flags; 9347 } 9348 else 9349 eval_state->pos_magic = NULL; 9350 9351 if (!PL_reg_curpm) { 9352 /* PL_reg_curpm is a fake PMOP that we can attach the current 9353 * regex to and point PL_curpm at, so that $1 et al are visible 9354 * within a /(?{})/. It's just allocated once per interpreter the 9355 * first time its needed */ 9356 Newxz(PL_reg_curpm, 1, PMOP); 9357 #ifdef USE_ITHREADS 9358 { 9359 SV* const repointer = &PL_sv_undef; 9360 /* this regexp is also owned by the new PL_reg_curpm, which 9361 will try to free it. */ 9362 av_push(PL_regex_padav, repointer); 9363 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav); 9364 PL_regex_pad = AvARRAY(PL_regex_padav); 9365 } 9366 #endif 9367 } 9368 SET_reg_curpm(reginfo->prog); 9369 eval_state->curpm = PL_curpm; 9370 PL_curpm = PL_reg_curpm; 9371 if (RXp_MATCH_COPIED(rex)) { 9372 /* Here is a serious problem: we cannot rewrite subbeg, 9373 since it may be needed if this match fails. Thus 9374 $` inside (?{}) could fail... */ 9375 eval_state->subbeg = rex->subbeg; 9376 eval_state->sublen = rex->sublen; 9377 eval_state->suboffset = rex->suboffset; 9378 eval_state->subcoffset = rex->subcoffset; 9379 #ifdef PERL_ANY_COW 9380 eval_state->saved_copy = rex->saved_copy; 9381 #endif 9382 RXp_MATCH_COPIED_off(rex); 9383 } 9384 else 9385 eval_state->subbeg = NULL; 9386 rex->subbeg = (char *)reginfo->strbeg; 9387 rex->suboffset = 0; 9388 rex->subcoffset = 0; 9389 rex->sublen = reginfo->strend - reginfo->strbeg; 9390 } 9391 9392 9393 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */ 9394 9395 static void 9396 S_cleanup_regmatch_info_aux(pTHX_ void *arg) 9397 { 9398 regmatch_info_aux *aux = (regmatch_info_aux *) arg; 9399 regmatch_info_aux_eval *eval_state = aux->info_aux_eval; 9400 regmatch_slab *s; 9401 9402 Safefree(aux->poscache); 9403 9404 if (eval_state) { 9405 9406 /* undo the effects of S_setup_eval_state() */ 9407 9408 if (eval_state->subbeg) { 9409 regexp * const rex = eval_state->rex; 9410 rex->subbeg = eval_state->subbeg; 9411 rex->sublen = eval_state->sublen; 9412 rex->suboffset = eval_state->suboffset; 9413 rex->subcoffset = eval_state->subcoffset; 9414 #ifdef PERL_ANY_COW 9415 rex->saved_copy = eval_state->saved_copy; 9416 #endif 9417 RXp_MATCH_COPIED_on(rex); 9418 } 9419 if (eval_state->pos_magic) 9420 { 9421 eval_state->pos_magic->mg_len = eval_state->pos; 9422 eval_state->pos_magic->mg_flags = 9423 (eval_state->pos_magic->mg_flags & ~MGf_BYTES) 9424 | (eval_state->pos_flags & MGf_BYTES); 9425 } 9426 9427 PL_curpm = eval_state->curpm; 9428 } 9429 9430 PL_regmatch_state = aux->old_regmatch_state; 9431 PL_regmatch_slab = aux->old_regmatch_slab; 9432 9433 /* free all slabs above current one - this must be the last action 9434 * of this function, as aux and eval_state are allocated within 9435 * slabs and may be freed here */ 9436 9437 s = PL_regmatch_slab->next; 9438 if (s) { 9439 PL_regmatch_slab->next = NULL; 9440 while (s) { 9441 regmatch_slab * const osl = s; 9442 s = s->next; 9443 Safefree(osl); 9444 } 9445 } 9446 } 9447 9448 9449 STATIC void 9450 S_to_utf8_substr(pTHX_ regexp *prog) 9451 { 9452 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile 9453 * on the converted value */ 9454 9455 int i = 1; 9456 9457 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR; 9458 9459 do { 9460 if (prog->substrs->data[i].substr 9461 && !prog->substrs->data[i].utf8_substr) { 9462 SV* const sv = newSVsv(prog->substrs->data[i].substr); 9463 prog->substrs->data[i].utf8_substr = sv; 9464 sv_utf8_upgrade(sv); 9465 if (SvVALID(prog->substrs->data[i].substr)) { 9466 if (SvTAIL(prog->substrs->data[i].substr)) { 9467 /* Trim the trailing \n that fbm_compile added last 9468 time. */ 9469 SvCUR_set(sv, SvCUR(sv) - 1); 9470 /* Whilst this makes the SV technically "invalid" (as its 9471 buffer is no longer followed by "\0") when fbm_compile() 9472 adds the "\n" back, a "\0" is restored. */ 9473 fbm_compile(sv, FBMcf_TAIL); 9474 } else 9475 fbm_compile(sv, 0); 9476 } 9477 if (prog->substrs->data[i].substr == prog->check_substr) 9478 prog->check_utf8 = sv; 9479 } 9480 } while (i--); 9481 } 9482 9483 STATIC bool 9484 S_to_byte_substr(pTHX_ regexp *prog) 9485 { 9486 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile 9487 * on the converted value; returns FALSE if can't be converted. */ 9488 9489 int i = 1; 9490 9491 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR; 9492 9493 do { 9494 if (prog->substrs->data[i].utf8_substr 9495 && !prog->substrs->data[i].substr) { 9496 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr); 9497 if (! sv_utf8_downgrade(sv, TRUE)) { 9498 return FALSE; 9499 } 9500 if (SvVALID(prog->substrs->data[i].utf8_substr)) { 9501 if (SvTAIL(prog->substrs->data[i].utf8_substr)) { 9502 /* Trim the trailing \n that fbm_compile added last 9503 time. */ 9504 SvCUR_set(sv, SvCUR(sv) - 1); 9505 fbm_compile(sv, FBMcf_TAIL); 9506 } else 9507 fbm_compile(sv, 0); 9508 } 9509 prog->substrs->data[i].substr = sv; 9510 if (prog->substrs->data[i].utf8_substr == prog->check_utf8) 9511 prog->check_substr = sv; 9512 } 9513 } while (i--); 9514 9515 return TRUE; 9516 } 9517 9518 /* 9519 * ex: set ts=8 sts=4 sw=4 et: 9520 */ 9521