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 "inline_invlist.c" 84 #include "unicode_constants.h" 85 86 #ifdef DEBUGGING 87 /* At least one required character in the target string is expressible only in 88 * UTF-8. */ 89 static const char* const non_utf8_target_but_utf8_required 90 = "Can't match, because target string needs to be in UTF-8\n"; 91 #endif 92 93 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \ 94 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\ 95 goto target; \ 96 } STMT_END 97 98 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i) 99 100 #ifndef STATIC 101 #define STATIC static 102 #endif 103 104 /* Valid only for non-utf8 strings: avoids the reginclass 105 * call if there are no complications: i.e., if everything matchable is 106 * straight forward in the bitmap */ 107 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \ 108 : ANYOF_BITMAP_TEST(p,*(c))) 109 110 /* 111 * Forwards. 112 */ 113 114 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv)) 115 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b) 116 117 #define HOPc(pos,off) \ 118 (char *)(reginfo->is_utf8_target \ 119 ? reghop3((U8*)pos, off, \ 120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \ 121 : (U8*)(pos + off)) 122 123 #define HOPBACKc(pos, off) \ 124 (char*)(reginfo->is_utf8_target \ 125 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \ 126 : (pos - off >= reginfo->strbeg) \ 127 ? (U8*)pos - off \ 128 : NULL) 129 130 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off)) 131 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) 132 133 /* lim must be +ve. Returns NULL on overshoot */ 134 #define HOPMAYBE3(pos,off,lim) \ 135 (reginfo->is_utf8_target \ 136 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \ 137 : ((U8*)pos + off <= lim) \ 138 ? (U8*)pos + off \ 139 : NULL) 140 141 /* like HOP3, but limits the result to <= lim even for the non-utf8 case. 142 * off must be >=0; args should be vars rather than expressions */ 143 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \ 144 ? reghop3((U8*)(pos), off, (U8*)(lim)) \ 145 : (U8*)((pos + off) > lim ? lim : (pos + off))) 146 147 #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \ 148 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \ 149 : (U8*)(pos + off)) 150 #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim)) 151 152 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */ 153 #define NEXTCHR_IS_EOS (nextchr < 0) 154 155 #define SET_nextchr \ 156 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS) 157 158 #define SET_locinput(p) \ 159 locinput = (p); \ 160 SET_nextchr 161 162 163 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \ 164 if (!swash_ptr) { \ 165 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \ 166 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \ 167 1, 0, invlist, &flags); \ 168 assert(swash_ptr); \ 169 } \ 170 } STMT_END 171 172 /* If in debug mode, we test that a known character properly matches */ 173 #ifdef DEBUGGING 174 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \ 175 property_name, \ 176 invlist, \ 177 utf8_char_in_property) \ 178 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \ 179 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE)); 180 #else 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 #endif 187 188 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \ 189 PL_utf8_swash_ptrs[_CC_WORDCHAR], \ 190 "", \ 191 PL_XPosix_ptrs[_CC_WORDCHAR], \ 192 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); 193 194 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \ 195 STMT_START { \ 196 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \ 197 "_X_regular_begin", \ 198 NULL, \ 199 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \ 200 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \ 201 "_X_extend", \ 202 NULL, \ 203 COMBINING_GRAVE_ACCENT_UTF8); \ 204 } STMT_END 205 206 #define PLACEHOLDER /* Something for the preprocessor to grab onto */ 207 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */ 208 209 /* for use after a quantifier and before an EXACT-like node -- japhy */ 210 /* it would be nice to rework regcomp.sym to generate this stuff. sigh 211 * 212 * NOTE that *nothing* that affects backtracking should be in here, specifically 213 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a 214 * node that is in between two EXACT like nodes when ascertaining what the required 215 * "follow" character is. This should probably be moved to regex compile time 216 * although it may be done at run time beause of the REF possibility - more 217 * investigation required. -- demerphq 218 */ 219 #define JUMPABLE(rn) ( \ 220 OP(rn) == OPEN || \ 221 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \ 222 OP(rn) == EVAL || \ 223 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \ 224 OP(rn) == PLUS || OP(rn) == MINMOD || \ 225 OP(rn) == KEEPS || \ 226 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \ 227 ) 228 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT) 229 230 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF ) 231 232 #if 0 233 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so 234 we don't need this definition. */ 235 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF ) 236 #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 ) 237 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL ) 238 239 #else 240 /* ... so we use this as its faster. */ 241 #define IS_TEXT(rn) ( OP(rn)==EXACT ) 242 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE) 243 #define IS_TEXTF(rn) ( OP(rn)==EXACTF ) 244 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL ) 245 246 #endif 247 248 /* 249 Search for mandatory following text node; for lookahead, the text must 250 follow but for lookbehind (rn->flags != 0) we skip to the next step. 251 */ 252 #define FIND_NEXT_IMPT(rn) STMT_START { \ 253 while (JUMPABLE(rn)) { \ 254 const OPCODE type = OP(rn); \ 255 if (type == SUSPEND || PL_regkind[type] == CURLY) \ 256 rn = NEXTOPER(NEXTOPER(rn)); \ 257 else if (type == PLUS) \ 258 rn = NEXTOPER(rn); \ 259 else if (type == IFMATCH) \ 260 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \ 261 else rn += NEXT_OFF(rn); \ 262 } \ 263 } STMT_END 264 265 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode. 266 * These are for the pre-composed Hangul syllables, which are all in a 267 * contiguous block and arranged there in such a way so as to facilitate 268 * alorithmic determination of their characteristics. As such, they don't need 269 * a swash, but can be determined by simple arithmetic. Almost all are 270 * GCB=LVT, but every 28th one is a GCB=LV */ 271 #define SBASE 0xAC00 /* Start of block */ 272 #define SCount 11172 /* Length of block */ 273 #define TCount 28 274 275 #define SLAB_FIRST(s) (&(s)->states[0]) 276 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1]) 277 278 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo); 279 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg); 280 static regmatch_state * S_push_slab(pTHX); 281 282 #define REGCP_PAREN_ELEMS 3 283 #define REGCP_OTHER_ELEMS 3 284 #define REGCP_FRAME_ELEMS 1 285 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and 286 * are needed for the regexp context stack bookkeeping. */ 287 288 STATIC CHECKPOINT 289 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen) 290 { 291 dVAR; 292 const int retval = PL_savestack_ix; 293 const int paren_elems_to_push = 294 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS; 295 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS; 296 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT; 297 I32 p; 298 GET_RE_DEBUG_FLAGS_DECL; 299 300 PERL_ARGS_ASSERT_REGCPPUSH; 301 302 if (paren_elems_to_push < 0) 303 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %i", 304 paren_elems_to_push, maxopenparen, parenfloor, REGCP_PAREN_ELEMS); 305 306 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems) 307 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf 308 " out of range (%lu-%ld)", 309 total_elems, 310 (unsigned long)maxopenparen, 311 (long)parenfloor); 312 313 SSGROW(total_elems + REGCP_FRAME_ELEMS); 314 315 DEBUG_BUFFERS_r( 316 if ((int)maxopenparen > (int)parenfloor) 317 PerlIO_printf(Perl_debug_log, 318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n", 319 PTR2UV(rex), 320 PTR2UV(rex->offs) 321 ); 322 ); 323 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) { 324 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */ 325 SSPUSHIV(rex->offs[p].end); 326 SSPUSHIV(rex->offs[p].start); 327 SSPUSHINT(rex->offs[p].start_tmp); 328 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, 329 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n", 330 (UV)p, 331 (IV)rex->offs[p].start, 332 (IV)rex->offs[p].start_tmp, 333 (IV)rex->offs[p].end 334 )); 335 } 336 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */ 337 SSPUSHINT(maxopenparen); 338 SSPUSHINT(rex->lastparen); 339 SSPUSHINT(rex->lastcloseparen); 340 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */ 341 342 return retval; 343 } 344 345 /* These are needed since we do not localize EVAL nodes: */ 346 #define REGCP_SET(cp) \ 347 DEBUG_STATE_r( \ 348 PerlIO_printf(Perl_debug_log, \ 349 " Setting an EVAL scope, savestack=%"IVdf"\n", \ 350 (IV)PL_savestack_ix)); \ 351 cp = PL_savestack_ix 352 353 #define REGCP_UNWIND(cp) \ 354 DEBUG_STATE_r( \ 355 if (cp != PL_savestack_ix) \ 356 PerlIO_printf(Perl_debug_log, \ 357 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ 358 (IV)(cp), (IV)PL_savestack_ix)); \ 359 regcpblow(cp) 360 361 #define UNWIND_PAREN(lp, lcp) \ 362 for (n = rex->lastparen; n > lp; n--) \ 363 rex->offs[n].end = -1; \ 364 rex->lastparen = n; \ 365 rex->lastcloseparen = lcp; 366 367 368 STATIC void 369 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p) 370 { 371 dVAR; 372 UV i; 373 U32 paren; 374 GET_RE_DEBUG_FLAGS_DECL; 375 376 PERL_ARGS_ASSERT_REGCPPOP; 377 378 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */ 379 i = SSPOPUV; 380 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */ 381 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */ 382 rex->lastcloseparen = SSPOPINT; 383 rex->lastparen = SSPOPINT; 384 *maxopenparen_p = SSPOPINT; 385 386 i -= REGCP_OTHER_ELEMS; 387 /* Now restore the parentheses context. */ 388 DEBUG_BUFFERS_r( 389 if (i || rex->lastparen + 1 <= rex->nparens) 390 PerlIO_printf(Perl_debug_log, 391 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n", 392 PTR2UV(rex), 393 PTR2UV(rex->offs) 394 ); 395 ); 396 paren = *maxopenparen_p; 397 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) { 398 SSize_t tmps; 399 rex->offs[paren].start_tmp = SSPOPINT; 400 rex->offs[paren].start = SSPOPIV; 401 tmps = SSPOPIV; 402 if (paren <= rex->lastparen) 403 rex->offs[paren].end = tmps; 404 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log, 405 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n", 406 (UV)paren, 407 (IV)rex->offs[paren].start, 408 (IV)rex->offs[paren].start_tmp, 409 (IV)rex->offs[paren].end, 410 (paren > rex->lastparen ? "(skipped)" : "")); 411 ); 412 paren--; 413 } 414 #if 1 415 /* It would seem that the similar code in regtry() 416 * already takes care of this, and in fact it is in 417 * a better location to since this code can #if 0-ed out 418 * but the code in regtry() is needed or otherwise tests 419 * requiring null fields (pat.t#187 and split.t#{13,14} 420 * (as of patchlevel 7877) will fail. Then again, 421 * this code seems to be necessary or otherwise 422 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ 423 * --jhi updated by dapm */ 424 for (i = rex->lastparen + 1; i <= rex->nparens; i++) { 425 if (i > *maxopenparen_p) 426 rex->offs[i].start = -1; 427 rex->offs[i].end = -1; 428 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log, 429 " \\%"UVuf": %s ..-1 undeffing\n", 430 (UV)i, 431 (i > *maxopenparen_p) ? "-1" : " " 432 )); 433 } 434 #endif 435 } 436 437 /* restore the parens and associated vars at savestack position ix, 438 * but without popping the stack */ 439 440 STATIC void 441 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p) 442 { 443 I32 tmpix = PL_savestack_ix; 444 PL_savestack_ix = ix; 445 regcppop(rex, maxopenparen_p); 446 PL_savestack_ix = tmpix; 447 } 448 449 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */ 450 451 STATIC bool 452 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character) 453 { 454 /* Returns a boolean as to whether or not 'character' is a member of the 455 * Posix character class given by 'classnum' that should be equivalent to a 456 * value in the typedef '_char_class_number'. 457 * 458 * Ideally this could be replaced by a just an array of function pointers 459 * to the C library functions that implement the macros this calls. 460 * However, to compile, the precise function signatures are required, and 461 * these may vary from platform to to platform. To avoid having to figure 462 * out what those all are on each platform, I (khw) am using this method, 463 * which adds an extra layer of function call overhead (unless the C 464 * optimizer strips it away). But we don't particularly care about 465 * performance with locales anyway. */ 466 467 switch ((_char_class_number) classnum) { 468 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character); 469 case _CC_ENUM_ALPHA: return isALPHA_LC(character); 470 case _CC_ENUM_ASCII: return isASCII_LC(character); 471 case _CC_ENUM_BLANK: return isBLANK_LC(character); 472 case _CC_ENUM_CASED: return isLOWER_LC(character) 473 || isUPPER_LC(character); 474 case _CC_ENUM_CNTRL: return isCNTRL_LC(character); 475 case _CC_ENUM_DIGIT: return isDIGIT_LC(character); 476 case _CC_ENUM_GRAPH: return isGRAPH_LC(character); 477 case _CC_ENUM_LOWER: return isLOWER_LC(character); 478 case _CC_ENUM_PRINT: return isPRINT_LC(character); 479 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character); 480 case _CC_ENUM_PUNCT: return isPUNCT_LC(character); 481 case _CC_ENUM_SPACE: return isSPACE_LC(character); 482 case _CC_ENUM_UPPER: return isUPPER_LC(character); 483 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character); 484 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character); 485 default: /* VERTSPACE should never occur in locales */ 486 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum); 487 } 488 489 assert(0); /* NOTREACHED */ 490 return FALSE; 491 } 492 493 STATIC bool 494 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character) 495 { 496 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded 497 * 'character' is a member of the Posix character class given by 'classnum' 498 * that should be equivalent to a value in the typedef 499 * '_char_class_number'. 500 * 501 * This just calls isFOO_lc on the code point for the character if it is in 502 * the range 0-255. Outside that range, all characters avoid Unicode 503 * rules, ignoring any locale. So use the Unicode function if this class 504 * requires a swash, and use the Unicode macro otherwise. */ 505 506 PERL_ARGS_ASSERT_ISFOO_UTF8_LC; 507 508 if (UTF8_IS_INVARIANT(*character)) { 509 return isFOO_lc(classnum, *character); 510 } 511 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) { 512 return isFOO_lc(classnum, 513 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1))); 514 } 515 516 if (classnum < _FIRST_NON_SWASH_CC) { 517 518 /* Initialize the swash unless done already */ 519 if (! PL_utf8_swash_ptrs[classnum]) { 520 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 521 PL_utf8_swash_ptrs[classnum] = 522 _core_swash_init("utf8", 523 "", 524 &PL_sv_undef, 1, 0, 525 PL_XPosix_ptrs[classnum], &flags); 526 } 527 528 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *) 529 character, 530 TRUE /* is UTF */ )); 531 } 532 533 switch ((_char_class_number) classnum) { 534 case _CC_ENUM_SPACE: 535 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character); 536 537 case _CC_ENUM_BLANK: return is_HORIZWS_high(character); 538 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character); 539 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character); 540 default: return 0; /* Things like CNTRL are always 541 below 256 */ 542 } 543 544 assert(0); /* NOTREACHED */ 545 return FALSE; 546 } 547 548 /* 549 * pregexec and friends 550 */ 551 552 #ifndef PERL_IN_XSUB_RE 553 /* 554 - pregexec - match a regexp against a string 555 */ 556 I32 557 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend, 558 char *strbeg, SSize_t minend, SV *screamer, U32 nosave) 559 /* stringarg: the point in the string at which to begin matching */ 560 /* strend: pointer to null at end of string */ 561 /* strbeg: real beginning of string */ 562 /* minend: end of match must be >= minend bytes after stringarg. */ 563 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string 564 * itself is accessed via the pointers above */ 565 /* nosave: For optimizations. */ 566 { 567 PERL_ARGS_ASSERT_PREGEXEC; 568 569 return 570 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL, 571 nosave ? 0 : REXEC_COPY_STR); 572 } 573 #endif 574 575 576 577 /* re_intuit_start(): 578 * 579 * Based on some optimiser hints, try to find the earliest position in the 580 * string where the regex could match. 581 * 582 * rx: the regex to match against 583 * sv: the SV being matched: only used for utf8 flag; the string 584 * itself is accessed via the pointers below. Note that on 585 * something like an overloaded SV, SvPOK(sv) may be false 586 * and the string pointers may point to something unrelated to 587 * the SV itself. 588 * strbeg: real beginning of string 589 * strpos: the point in the string at which to begin matching 590 * strend: pointer to the byte following the last char of the string 591 * flags currently unused; set to 0 592 * data: currently unused; set to NULL 593 * 594 * The basic idea of re_intuit_start() is to use some known information 595 * about the pattern, namely: 596 * 597 * a) the longest known anchored substring (i.e. one that's at a 598 * constant offset from the beginning of the pattern; but not 599 * necessarily at a fixed offset from the beginning of the 600 * string); 601 * b) the longest floating substring (i.e. one that's not at a constant 602 * offset from the beginning of the pattern); 603 * c) Whether the pattern is anchored to the string; either 604 * an absolute anchor: /^../, or anchored to \n: /^.../m, 605 * or anchored to pos(): /\G/; 606 * d) A start class: a real or synthetic character class which 607 * represents which characters are legal at the start of the pattern; 608 * 609 * to either quickly reject the match, or to find the earliest position 610 * within the string at which the pattern might match, thus avoiding 611 * running the full NFA engine at those earlier locations, only to 612 * eventually fail and retry further along. 613 * 614 * Returns NULL if the pattern can't match, or returns the address within 615 * the string which is the earliest place the match could occur. 616 * 617 * The longest of the anchored and floating substrings is called 'check' 618 * and is checked first. The other is called 'other' and is checked 619 * second. The 'other' substring may not be present. For example, 620 * 621 * /(abc|xyz)ABC\d{0,3}DEFG/ 622 * 623 * will have 624 * 625 * check substr (float) = "DEFG", offset 6..9 chars 626 * other substr (anchored) = "ABC", offset 3..3 chars 627 * stclass = [ax] 628 * 629 * Be aware that during the course of this function, sometimes 'anchored' 630 * refers to a substring being anchored relative to the start of the 631 * pattern, and sometimes to the pattern itself being anchored relative to 632 * the string. For example: 633 * 634 * /\dabc/: "abc" is anchored to the pattern; 635 * /^\dabc/: "abc" is anchored to the pattern and the string; 636 * /\d+abc/: "abc" is anchored to neither the pattern nor the string; 637 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string, 638 * but the pattern is anchored to the string. 639 */ 640 641 char * 642 Perl_re_intuit_start(pTHX_ 643 REGEXP * const rx, 644 SV *sv, 645 const char * const strbeg, 646 char *strpos, 647 char *strend, 648 const U32 flags, 649 re_scream_pos_data *data) 650 { 651 dVAR; 652 struct regexp *const prog = ReANY(rx); 653 SSize_t start_shift = prog->check_offset_min; 654 /* Should be nonnegative! */ 655 SSize_t end_shift = 0; 656 /* current lowest pos in string where the regex can start matching */ 657 char *rx_origin = strpos; 658 SV *check; 659 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */ 660 U8 other_ix = 1 - prog->substrs->check_ix; 661 bool ml_anch = 0; 662 char *other_last = strpos;/* latest pos 'other' substr already checked to */ 663 char *check_at = NULL; /* check substr found at this pos */ 664 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE; 665 RXi_GET_DECL(prog,progi); 666 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */ 667 regmatch_info *const reginfo = ®info_buf; 668 GET_RE_DEBUG_FLAGS_DECL; 669 670 PERL_ARGS_ASSERT_RE_INTUIT_START; 671 PERL_UNUSED_ARG(flags); 672 PERL_UNUSED_ARG(data); 673 674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 675 "Intuit: trying to determine minimum start position...\n")); 676 677 /* for now, assume that all substr offsets are positive. If at some point 678 * in the future someone wants to do clever things with look-behind and 679 * -ve offsets, they'll need to fix up any code in this function 680 * which uses these offsets. See the thread beginning 681 * <20140113145929.GF27210@iabyn.com> 682 */ 683 assert(prog->substrs->data[0].min_offset >= 0); 684 assert(prog->substrs->data[0].max_offset >= 0); 685 assert(prog->substrs->data[1].min_offset >= 0); 686 assert(prog->substrs->data[1].max_offset >= 0); 687 assert(prog->substrs->data[2].min_offset >= 0); 688 assert(prog->substrs->data[2].max_offset >= 0); 689 690 /* for now, assume that if both present, that the floating substring 691 * doesn't start before the anchored substring. 692 * If you break this assumption (e.g. doing better optimisations 693 * with lookahead/behind), then you'll need to audit the code in this 694 * function carefully first 695 */ 696 assert( 697 ! ( (prog->anchored_utf8 || prog->anchored_substr) 698 && (prog->float_utf8 || prog->float_substr)) 699 || (prog->float_min_offset >= prog->anchored_offset)); 700 701 /* byte rather than char calculation for efficiency. It fails 702 * to quickly reject some cases that can't match, but will reject 703 * them later after doing full char arithmetic */ 704 if (prog->minlen > strend - strpos) { 705 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 706 " String too short...\n")); 707 goto fail; 708 } 709 710 reginfo->is_utf8_target = cBOOL(utf8_target); 711 reginfo->info_aux = NULL; 712 reginfo->strbeg = strbeg; 713 reginfo->strend = strend; 714 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx)); 715 reginfo->intuit = 1; 716 /* not actually used within intuit, but zero for safety anyway */ 717 reginfo->poscache_maxiter = 0; 718 719 if (utf8_target) { 720 if (!prog->check_utf8 && prog->check_substr) 721 to_utf8_substr(prog); 722 check = prog->check_utf8; 723 } else { 724 if (!prog->check_substr && prog->check_utf8) { 725 if (! to_byte_substr(prog)) { 726 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail); 727 } 728 } 729 check = prog->check_substr; 730 } 731 732 /* dump the various substring data */ 733 DEBUG_OPTIMISE_MORE_r({ 734 int i; 735 for (i=0; i<=2; i++) { 736 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr 737 : prog->substrs->data[i].substr); 738 if (!sv) 739 continue; 740 741 PerlIO_printf(Perl_debug_log, 742 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf 743 " useful=%"IVdf" utf8=%d [%s]\n", 744 i, 745 (IV)prog->substrs->data[i].min_offset, 746 (IV)prog->substrs->data[i].max_offset, 747 (IV)prog->substrs->data[i].end_shift, 748 BmUSEFUL(sv), 749 utf8_target ? 1 : 0, 750 SvPEEK(sv)); 751 } 752 }); 753 754 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */ 755 756 /* ml_anch: check after \n? 757 * 758 * A note about IMPLICIT: on an un-anchored pattern beginning 759 * with /.*.../, these flags will have been added by the 760 * compiler: 761 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL 762 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL 763 */ 764 ml_anch = (prog->intflags & PREGf_ANCH_MBOL) 765 && !(prog->intflags & PREGf_IMPLICIT); 766 767 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) { 768 /* we are only allowed to match at BOS or \G */ 769 770 /* trivially reject if there's a BOS anchor and we're not at BOS. 771 * 772 * Note that we don't try to do a similar quick reject for 773 * \G, since generally the caller will have calculated strpos 774 * based on pos() and gofs, so the string is already correctly 775 * anchored by definition; and handling the exceptions would 776 * be too fiddly (e.g. REXEC_IGNOREPOS). 777 */ 778 if ( strpos != strbeg 779 && (prog->intflags & (PREGf_ANCH_BOL|PREGf_ANCH_SBOL))) 780 { 781 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 782 " Not at start...\n")); 783 goto fail; 784 } 785 786 /* in the presence of an anchor, the anchored (relative to the 787 * start of the regex) substr must also be anchored relative 788 * to strpos. So quickly reject if substr isn't found there. 789 * This works for \G too, because the caller will already have 790 * subtracted gofs from pos, and gofs is the offset from the 791 * \G to the start of the regex. For example, in /.abc\Gdef/, 792 * where substr="abcdef", pos()=3, gofs=4, offset_min=1: 793 * caller will have set strpos=pos()-4; we look for the substr 794 * at position pos()-4+1, which lines up with the "a" */ 795 796 if (prog->check_offset_min == prog->check_offset_max 797 && !(prog->intflags & PREGf_CANY_SEEN)) 798 { 799 /* Substring at constant offset from beg-of-str... */ 800 SSize_t slen = SvCUR(check); 801 char *s = HOP3c(strpos, prog->check_offset_min, strend); 802 803 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 804 " Looking for check substr at fixed offset %"IVdf"...\n", 805 (IV)prog->check_offset_min)); 806 807 if (SvTAIL(check)) { 808 /* In this case, the regex is anchored at the end too. 809 * Unless it's a multiline match, the lengths must match 810 * exactly, give or take a \n. NB: slen >= 1 since 811 * the last char of check is \n */ 812 if (!multiline 813 && ( strend - s > slen 814 || strend - s < slen - 1 815 || (strend - s == slen && strend[-1] != '\n'))) 816 { 817 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 818 " String too long...\n")); 819 goto fail_finish; 820 } 821 /* Now should match s[0..slen-2] */ 822 slen--; 823 } 824 if (slen && (*SvPVX_const(check) != *s 825 || (slen > 1 && memNE(SvPVX_const(check), s, slen)))) 826 { 827 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 828 " String not equal...\n")); 829 goto fail_finish; 830 } 831 832 check_at = s; 833 goto success_at_start; 834 } 835 } 836 } 837 838 end_shift = prog->check_end_shift; 839 840 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */ 841 if (end_shift < 0) 842 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ", 843 (IV)end_shift, RX_PRECOMP(prog)); 844 #endif 845 846 restart: 847 848 /* This is the (re)entry point of the main loop in this function. 849 * The goal of this loop is to: 850 * 1) find the "check" substring in the region rx_origin..strend 851 * (adjusted by start_shift / end_shift). If not found, reject 852 * immediately. 853 * 2) If it exists, look for the "other" substr too if defined; for 854 * example, if the check substr maps to the anchored substr, then 855 * check the floating substr, and vice-versa. If not found, go 856 * back to (1) with rx_origin suitably incremented. 857 * 3) If we find an rx_origin position that doesn't contradict 858 * either of the substrings, then check the possible additional 859 * constraints on rx_origin of /^.../m or a known start class. 860 * If these fail, then depending on which constraints fail, jump 861 * back to here, or to various other re-entry points further along 862 * that skip some of the first steps. 863 * 4) If we pass all those tests, update the BmUSEFUL() count on the 864 * substring. If the start position was determined to be at the 865 * beginning of the string - so, not rejected, but not optimised, 866 * since we have to run regmatch from position 0 - decrement the 867 * BmUSEFUL() count. Otherwise increment it. 868 */ 869 870 871 /* first, look for the 'check' substring */ 872 873 { 874 U8* start_point; 875 U8* end_point; 876 877 DEBUG_OPTIMISE_MORE_r({ 878 PerlIO_printf(Perl_debug_log, 879 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf 880 " Start shift: %"IVdf" End shift %"IVdf 881 " Real end Shift: %"IVdf"\n", 882 (IV)(rx_origin - strpos), 883 (IV)prog->check_offset_min, 884 (IV)start_shift, 885 (IV)end_shift, 886 (IV)prog->check_end_shift); 887 }); 888 889 if (prog->intflags & PREGf_CANY_SEEN) { 890 start_point= (U8*)(rx_origin + start_shift); 891 end_point= (U8*)(strend - end_shift); 892 if (start_point > end_point) 893 goto fail_finish; 894 } else { 895 end_point = HOP3(strend, -end_shift, strbeg); 896 start_point = HOPMAYBE3(rx_origin, start_shift, end_point); 897 if (!start_point) 898 goto fail_finish; 899 } 900 901 902 /* If the regex is absolutely anchored to either the start of the 903 * string (BOL,SBOL) or to pos() (ANCH_GPOS), then 904 * check_offset_max represents an upper bound on the string where 905 * the substr could start. For the ANCH_GPOS case, we assume that 906 * the caller of intuit will have already set strpos to 907 * pos()-gofs, so in this case strpos + offset_max will still be 908 * an upper bound on the substr. 909 */ 910 if (!ml_anch 911 && prog->intflags & PREGf_ANCH 912 && prog->check_offset_max != SSize_t_MAX) 913 { 914 SSize_t len = SvCUR(check) - !!SvTAIL(check); 915 const char * const anchor = 916 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg); 917 918 /* do a bytes rather than chars comparison. It's conservative; 919 * so it skips doing the HOP if the result can't possibly end 920 * up earlier than the old value of end_point. 921 */ 922 if ((char*)end_point - anchor > prog->check_offset_max) { 923 end_point = HOP3lim((U8*)anchor, 924 prog->check_offset_max, 925 end_point -len) 926 + len; 927 } 928 } 929 930 DEBUG_OPTIMISE_MORE_r({ 931 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n", 932 (int)(end_point - start_point), 933 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point), 934 start_point); 935 }); 936 937 check_at = fbm_instr( start_point, end_point, 938 check, multiline ? FBMrf_MULTILINE : 0); 939 940 /* Update the count-of-usability, remove useless subpatterns, 941 unshift s. */ 942 943 DEBUG_EXECUTE_r({ 944 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), 945 SvPVX_const(check), RE_SV_DUMPLEN(check), 30); 946 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s", 947 (check_at ? "Found" : "Did not find"), 948 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) 949 ? "anchored" : "floating"), 950 quoted, 951 RE_SV_TAIL(check), 952 (check_at ? " at offset " : "...\n") ); 953 }); 954 955 if (!check_at) 956 goto fail_finish; 957 /* Finish the diagnostic message */ 958 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) ); 959 960 /* set rx_origin to the minimum position where the regex could start 961 * matching, given the constraint of the just-matched check substring. 962 * But don't set it lower than previously. 963 */ 964 965 if (check_at - rx_origin > prog->check_offset_max) 966 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin); 967 } 968 969 970 /* now look for the 'other' substring if defined */ 971 972 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr 973 : prog->substrs->data[other_ix].substr) 974 { 975 /* Take into account the "other" substring. */ 976 char *last, *last1; 977 char *s; 978 SV* must; 979 struct reg_substr_datum *other; 980 981 do_other_substr: 982 other = &prog->substrs->data[other_ix]; 983 984 /* if "other" is anchored: 985 * we've previously found a floating substr starting at check_at. 986 * This means that the regex origin must lie somewhere 987 * between min (rx_origin): HOP3(check_at, -check_offset_max) 988 * and max: HOP3(check_at, -check_offset_min) 989 * (except that min will be >= strpos) 990 * So the fixed substr must lie somewhere between 991 * HOP3(min, anchored_offset) 992 * HOP3(max, anchored_offset) + SvCUR(substr) 993 */ 994 995 /* if "other" is floating 996 * Calculate last1, the absolute latest point where the 997 * floating substr could start in the string, ignoring any 998 * constraints from the earlier fixed match. It is calculated 999 * as follows: 1000 * 1001 * strend - prog->minlen (in chars) is the absolute latest 1002 * position within the string where the origin of the regex 1003 * could appear. The latest start point for the floating 1004 * substr is float_min_offset(*) on from the start of the 1005 * regex. last1 simply combines thee two offsets. 1006 * 1007 * (*) You might think the latest start point should be 1008 * float_max_offset from the regex origin, and technically 1009 * you'd be correct. However, consider 1010 * /a\d{2,4}bcd\w/ 1011 * Here, float min, max are 3,5 and minlen is 7. 1012 * This can match either 1013 * /a\d\dbcd\w/ 1014 * /a\d\d\dbcd\w/ 1015 * /a\d\d\d\dbcd\w/ 1016 * In the first case, the regex matches minlen chars; in the 1017 * second, minlen+1, in the third, minlen+2. 1018 * In the first case, the floating offset is 3 (which equals 1019 * float_min), in the second, 4, and in the third, 5 (which 1020 * equals float_max). In all cases, the floating string bcd 1021 * can never start more than 4 chars from the end of the 1022 * string, which equals minlen - float_min. As the substring 1023 * starts to match more than float_min from the start of the 1024 * regex, it makes the regex match more than minlen chars, 1025 * and the two cancel each other out. So we can always use 1026 * float_min - minlen, rather than float_max - minlen for the 1027 * latest position in the string. 1028 * 1029 * Note that -minlen + float_min_offset is equivalent (AFAIKT) 1030 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift 1031 */ 1032 1033 assert(prog->minlen >= other->min_offset); 1034 last1 = HOP3c(strend, 1035 other->min_offset - prog->minlen, strbeg); 1036 1037 if (other_ix) {/* i.e. if (other-is-float) */ 1038 /* last is the latest point where the floating substr could 1039 * start, *given* any constraints from the earlier fixed 1040 * match. This constraint is that the floating string starts 1041 * <= float_max_offset chars from the regex origin (rx_origin). 1042 * If this value is less than last1, use it instead. 1043 */ 1044 assert(rx_origin <= last1); 1045 last = 1046 /* this condition handles the offset==infinity case, and 1047 * is a short-cut otherwise. Although it's comparing a 1048 * byte offset to a char length, it does so in a safe way, 1049 * since 1 char always occupies 1 or more bytes, 1050 * so if a string range is (last1 - rx_origin) bytes, 1051 * it will be less than or equal to (last1 - rx_origin) 1052 * chars; meaning it errs towards doing the accurate HOP3 1053 * rather than just using last1 as a short-cut */ 1054 (last1 - rx_origin) < other->max_offset 1055 ? last1 1056 : (char*)HOP3lim(rx_origin, other->max_offset, last1); 1057 } 1058 else { 1059 assert(strpos + start_shift <= check_at); 1060 last = HOP4c(check_at, other->min_offset - start_shift, 1061 strbeg, strend); 1062 } 1063 1064 s = HOP3c(rx_origin, other->min_offset, strend); 1065 if (s < other_last) /* These positions already checked */ 1066 s = other_last; 1067 1068 must = utf8_target ? other->utf8_substr : other->substr; 1069 assert(SvPOK(must)); 1070 s = fbm_instr( 1071 (unsigned char*)s, 1072 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0), 1073 must, 1074 multiline ? FBMrf_MULTILINE : 0 1075 ); 1076 DEBUG_EXECUTE_r({ 1077 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), 1078 SvPVX_const(must), RE_SV_DUMPLEN(must), 30); 1079 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s", 1080 s ? "Found" : "Contradicts", 1081 other_ix ? "floating" : "anchored", 1082 quoted, RE_SV_TAIL(must)); 1083 }); 1084 1085 1086 if (!s) { 1087 /* last1 is latest possible substr location. If we didn't 1088 * find it before there, we never will */ 1089 if (last >= last1) { 1090 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1091 ", giving up...\n")); 1092 goto fail_finish; 1093 } 1094 1095 /* try to find the check substr again at a later 1096 * position. Maybe next time we'll find the "other" substr 1097 * in range too */ 1098 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1099 ", trying %s at offset %ld...\n", 1100 (other_ix ? "floating" : "anchored"), 1101 (long)(HOP3c(check_at, 1, strend) - strpos))); 1102 1103 other_last = HOP3c(last, 1, strend) /* highest failure */; 1104 rx_origin = 1105 other_ix /* i.e. if other-is-float */ 1106 ? HOP3c(rx_origin, 1, strend) 1107 : HOP4c(last, 1 - other->min_offset, strbeg, strend); 1108 goto restart; 1109 } 1110 else { 1111 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n", 1112 (long)(s - strpos))); 1113 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 } 1130 } 1131 else { 1132 DEBUG_OPTIMISE_MORE_r( 1133 PerlIO_printf(Perl_debug_log, 1134 " Check-only match: offset min:%"IVdf" max:%"IVdf 1135 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf 1136 " strend-strpos:%"IVdf"\n", 1137 (IV)prog->check_offset_min, 1138 (IV)prog->check_offset_max, 1139 (IV)(check_at-strpos), 1140 (IV)(rx_origin-strpos), 1141 (IV)(rx_origin-check_at), 1142 (IV)(strend-strpos) 1143 ) 1144 ); 1145 } 1146 1147 postprocess_substr_matches: 1148 1149 /* handle the extra constraint of /^.../m if present */ 1150 1151 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') { 1152 char *s; 1153 1154 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1155 " looking for /^/m anchor")); 1156 1157 /* we have failed the constraint of a \n before rx_origin. 1158 * Find the next \n, if any, even if it's beyond the current 1159 * anchored and/or floating substrings. Whether we should be 1160 * scanning ahead for the next \n or the next substr is debatable. 1161 * On the one hand you'd expect rare substrings to appear less 1162 * often than \n's. On the other hand, searching for \n means 1163 * we're effectively flipping been check_substr and "\n" on each 1164 * iteration as the current "rarest" string candidate, which 1165 * means for example that we'll quickly reject the whole string if 1166 * hasn't got a \n, rather than trying every substr position 1167 * first 1168 */ 1169 1170 s = HOP3c(strend, - prog->minlen, strpos); 1171 if (s <= rx_origin || 1172 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin))) 1173 { 1174 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1175 " Did not find /%s^%s/m...\n", 1176 PL_colors[0], PL_colors[1])); 1177 goto fail_finish; 1178 } 1179 1180 /* earliest possible origin is 1 char after the \n. 1181 * (since *rx_origin == '\n', it's safe to ++ here rather than 1182 * HOP(rx_origin, 1)) */ 1183 rx_origin++; 1184 1185 if (prog->substrs->check_ix == 0 /* check is anchored */ 1186 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos)) 1187 { 1188 /* Position contradicts check-string; either because 1189 * check was anchored (and thus has no wiggle room), 1190 * or check was float and rx_origin is above the float range */ 1191 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1192 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n", 1193 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos))); 1194 goto restart; 1195 } 1196 1197 /* if we get here, the check substr must have been float, 1198 * is in range, and we may or may not have had an anchored 1199 * "other" substr which still contradicts */ 1200 assert(prog->substrs->check_ix); /* check is float */ 1201 1202 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) { 1203 /* whoops, the anchored "other" substr exists, so we still 1204 * contradict. On the other hand, the float "check" substr 1205 * didn't contradict, so just retry the anchored "other" 1206 * substr */ 1207 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1208 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n", 1209 PL_colors[0], PL_colors[1], 1210 (long)(rx_origin - strpos), 1211 (long)(rx_origin - strpos + prog->anchored_offset))); 1212 goto do_other_substr; 1213 } 1214 1215 /* success: we don't contradict the found floating substring 1216 * (and there's no anchored substr). */ 1217 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1218 " Found /%s^%s/m at offset %ld...\n", 1219 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos))); 1220 } 1221 else { 1222 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1223 " (multiline anchor test skipped)\n")); 1224 } 1225 1226 success_at_start: 1227 1228 1229 /* if we have a starting character class, then test that extra constraint. 1230 * (trie stclasses are too expensive to use here, we are better off to 1231 * leave it to regmatch itself) */ 1232 1233 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) { 1234 const U8* const str = (U8*)STRING(progi->regstclass); 1235 1236 /* XXX this value could be pre-computed */ 1237 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT 1238 ? (reginfo->is_utf8_pat 1239 ? utf8_distance(str + STR_LEN(progi->regstclass), str) 1240 : STR_LEN(progi->regstclass)) 1241 : 1); 1242 char * endpos; 1243 char *s; 1244 /* latest pos that a matching float substr constrains rx start to */ 1245 char *rx_max_float = NULL; 1246 1247 /* if the current rx_origin is anchored, either by satisfying an 1248 * anchored substring constraint, or a /^.../m constraint, then we 1249 * can reject the current origin if the start class isn't found 1250 * at the current position. If we have a float-only match, then 1251 * rx_origin is constrained to a range; so look for the start class 1252 * in that range. if neither, then look for the start class in the 1253 * whole rest of the string */ 1254 1255 /* XXX DAPM it's not clear what the minlen test is for, and why 1256 * it's not used in the floating case. Nothing in the test suite 1257 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>. 1258 * Here are some old comments, which may or may not be correct: 1259 * 1260 * minlen == 0 is possible if regstclass is \b or \B, 1261 * and the fixed substr is ''$. 1262 * Since minlen is already taken into account, rx_origin+1 is 1263 * before strend; accidentally, minlen >= 1 guaranties no false 1264 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 : 1265 * 0) below assumes that regstclass does not come from lookahead... 1266 * If regstclass takes bytelength more than 1: If charlength==1, OK. 1267 * This leaves EXACTF-ish only, which are dealt with in 1268 * find_byclass(). 1269 */ 1270 1271 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch) 1272 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend); 1273 else if (prog->float_substr || prog->float_utf8) { 1274 rx_max_float = HOP3c(check_at, -start_shift, strbeg); 1275 endpos= HOP3c(rx_max_float, cl_l, strend); 1276 } 1277 else 1278 endpos= strend; 1279 1280 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1281 " looking for class: start_shift: %"IVdf" check_at: %"IVdf 1282 " rx_origin: %"IVdf" endpos: %"IVdf"\n", 1283 (IV)start_shift, (IV)(check_at - strbeg), 1284 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg))); 1285 1286 s = find_byclass(prog, progi->regstclass, rx_origin, endpos, 1287 reginfo); 1288 if (!s) { 1289 if (endpos == strend) { 1290 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 1291 " Could not match STCLASS...\n") ); 1292 goto fail; 1293 } 1294 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 1295 " This position contradicts STCLASS...\n") ); 1296 if ((prog->intflags & PREGf_ANCH) && !ml_anch 1297 && !(prog->intflags & PREGf_IMPLICIT)) 1298 goto fail; 1299 1300 /* Contradict one of substrings */ 1301 if (prog->anchored_substr || prog->anchored_utf8) { 1302 if (prog->substrs->check_ix == 1) { /* check is float */ 1303 /* Have both, check_string is floating */ 1304 assert(rx_origin + start_shift <= check_at); 1305 if (rx_origin + start_shift != check_at) { 1306 /* not at latest position float substr could match: 1307 * Recheck anchored substring, but not floating. 1308 * The condition above is in bytes rather than 1309 * chars for efficiency. It's conservative, in 1310 * that it errs on the side of doing 'goto 1311 * do_other_substr', where a more accurate 1312 * char-based calculation will be done */ 1313 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 1314 " Looking for anchored substr starting at offset %ld...\n", 1315 (long)(other_last - strpos)) ); 1316 goto do_other_substr; 1317 } 1318 } 1319 } 1320 else { 1321 /* float-only */ 1322 1323 if (ml_anch) { 1324 /* In the presence of ml_anch, we might be able to 1325 * find another \n without breaking the current float 1326 * constraint. */ 1327 1328 /* strictly speaking this should be HOP3c(..., 1, ...), 1329 * but since we goto a block of code that's going to 1330 * search for the next \n if any, its safe here */ 1331 rx_origin++; 1332 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 1333 " Looking for /%s^%s/m starting at offset %ld...\n", 1334 PL_colors[0], PL_colors[1], 1335 (long)(rx_origin - strpos)) ); 1336 goto postprocess_substr_matches; 1337 } 1338 1339 /* strictly speaking this can never be true; but might 1340 * be if we ever allow intuit without substrings */ 1341 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) 1342 goto fail; 1343 1344 rx_origin = rx_max_float; 1345 } 1346 1347 /* at this point, any matching substrings have been 1348 * contradicted. Start again... */ 1349 1350 rx_origin = HOP3c(rx_origin, 1, strend); 1351 1352 /* uses bytes rather than char calculations for efficiency. 1353 * It's conservative: it errs on the side of doing 'goto restart', 1354 * where there is code that does a proper char-based test */ 1355 if (rx_origin + start_shift + end_shift > strend) { 1356 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 1357 " Could not match STCLASS...\n") ); 1358 goto fail; 1359 } 1360 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 1361 " Looking for %s substr starting at offset %ld...\n", 1362 (prog->substrs->check_ix ? "floating" : "anchored"), 1363 (long)(rx_origin + start_shift - strpos)) ); 1364 goto restart; 1365 } 1366 1367 /* Success !!! */ 1368 1369 if (rx_origin != s) { 1370 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1371 " By STCLASS: moving %ld --> %ld\n", 1372 (long)(rx_origin - strpos), (long)(s - strpos)) 1373 ); 1374 } 1375 else { 1376 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1377 " Does not contradict STCLASS...\n"); 1378 ); 1379 } 1380 } 1381 1382 /* Decide whether using the substrings helped */ 1383 1384 if (rx_origin != strpos) { 1385 /* Fixed substring is found far enough so that the match 1386 cannot start at strpos. */ 1387 1388 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n")); 1389 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */ 1390 } 1391 else { 1392 /* The found rx_origin position does not prohibit matching at 1393 * strpos, so calling intuit didn't gain us anything. Decrement 1394 * the BmUSEFUL() count on the check substring, and if we reach 1395 * zero, free it. */ 1396 if (!(prog->intflags & PREGf_NAUGHTY) 1397 && (utf8_target ? ( 1398 prog->check_utf8 /* Could be deleted already */ 1399 && --BmUSEFUL(prog->check_utf8) < 0 1400 && (prog->check_utf8 == prog->float_utf8) 1401 ) : ( 1402 prog->check_substr /* Could be deleted already */ 1403 && --BmUSEFUL(prog->check_substr) < 0 1404 && (prog->check_substr == prog->float_substr) 1405 ))) 1406 { 1407 /* If flags & SOMETHING - do not do it many times on the same match */ 1408 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n")); 1409 /* XXX Does the destruction order has to change with utf8_target? */ 1410 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr); 1411 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8); 1412 prog->check_substr = prog->check_utf8 = NULL; /* disable */ 1413 prog->float_substr = prog->float_utf8 = NULL; /* clear */ 1414 check = NULL; /* abort */ 1415 /* XXXX This is a remnant of the old implementation. It 1416 looks wasteful, since now INTUIT can use many 1417 other heuristics. */ 1418 prog->extflags &= ~RXf_USE_INTUIT; 1419 } 1420 } 1421 1422 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 1423 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n", 1424 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) ); 1425 1426 return rx_origin; 1427 1428 fail_finish: /* Substring not found */ 1429 if (prog->check_substr || prog->check_utf8) /* could be removed already */ 1430 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */ 1431 fail: 1432 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n", 1433 PL_colors[4], PL_colors[5])); 1434 return NULL; 1435 } 1436 1437 1438 #define DECL_TRIE_TYPE(scan) \ 1439 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \ 1440 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \ 1441 trie_type = ((scan->flags == EXACT) \ 1442 ? (utf8_target ? trie_utf8 : trie_plain) \ 1443 : (scan->flags == EXACTFA) \ 1444 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \ 1445 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold)) 1446 1447 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \ 1448 STMT_START { \ 1449 STRLEN skiplen; \ 1450 U8 flags = FOLD_FLAGS_FULL; \ 1451 switch (trie_type) { \ 1452 case trie_utf8_exactfa_fold: \ 1453 flags |= FOLD_FLAGS_NOMIX_ASCII; \ 1454 /* FALL THROUGH */ \ 1455 case trie_utf8_fold: \ 1456 if ( foldlen>0 ) { \ 1457 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ 1458 foldlen -= len; \ 1459 uscan += len; \ 1460 len=0; \ 1461 } else { \ 1462 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \ 1463 len = UTF8SKIP(uc); \ 1464 skiplen = UNISKIP( uvc ); \ 1465 foldlen -= skiplen; \ 1466 uscan = foldbuf + skiplen; \ 1467 } \ 1468 break; \ 1469 case trie_latin_utf8_exactfa_fold: \ 1470 flags |= FOLD_FLAGS_NOMIX_ASCII; \ 1471 /* FALL THROUGH */ \ 1472 case trie_latin_utf8_fold: \ 1473 if ( foldlen>0 ) { \ 1474 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ 1475 foldlen -= len; \ 1476 uscan += len; \ 1477 len=0; \ 1478 } else { \ 1479 len = 1; \ 1480 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \ 1481 skiplen = UNISKIP( uvc ); \ 1482 foldlen -= skiplen; \ 1483 uscan = foldbuf + skiplen; \ 1484 } \ 1485 break; \ 1486 case trie_utf8: \ 1487 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \ 1488 break; \ 1489 case trie_plain: \ 1490 uvc = (UV)*uc; \ 1491 len = 1; \ 1492 } \ 1493 if (uvc < 256) { \ 1494 charid = trie->charmap[ uvc ]; \ 1495 } \ 1496 else { \ 1497 charid = 0; \ 1498 if (widecharmap) { \ 1499 SV** const svpp = hv_fetch(widecharmap, \ 1500 (char*)&uvc, sizeof(UV), 0); \ 1501 if (svpp) \ 1502 charid = (U16)SvIV(*svpp); \ 1503 } \ 1504 } \ 1505 } STMT_END 1506 1507 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \ 1508 STMT_START { \ 1509 while (s <= e) { \ 1510 if ( (CoNd) \ 1511 && (ln == 1 || folder(s, pat_string, ln)) \ 1512 && (reginfo->intuit || regtry(reginfo, &s)) )\ 1513 goto got_it; \ 1514 s++; \ 1515 } \ 1516 } STMT_END 1517 1518 #define REXEC_FBC_UTF8_SCAN(CoDe) \ 1519 STMT_START { \ 1520 while (s < strend) { \ 1521 CoDe \ 1522 s += UTF8SKIP(s); \ 1523 } \ 1524 } STMT_END 1525 1526 #define REXEC_FBC_SCAN(CoDe) \ 1527 STMT_START { \ 1528 while (s < strend) { \ 1529 CoDe \ 1530 s++; \ 1531 } \ 1532 } STMT_END 1533 1534 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \ 1535 REXEC_FBC_UTF8_SCAN( \ 1536 if (CoNd) { \ 1537 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ 1538 goto got_it; \ 1539 else \ 1540 tmp = doevery; \ 1541 } \ 1542 else \ 1543 tmp = 1; \ 1544 ) 1545 1546 #define REXEC_FBC_CLASS_SCAN(CoNd) \ 1547 REXEC_FBC_SCAN( \ 1548 if (CoNd) { \ 1549 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ 1550 goto got_it; \ 1551 else \ 1552 tmp = doevery; \ 1553 } \ 1554 else \ 1555 tmp = 1; \ 1556 ) 1557 1558 #define REXEC_FBC_TRYIT \ 1559 if ((reginfo->intuit || regtry(reginfo, &s))) \ 1560 goto got_it 1561 1562 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \ 1563 if (utf8_target) { \ 1564 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \ 1565 } \ 1566 else { \ 1567 REXEC_FBC_CLASS_SCAN(CoNd); \ 1568 } 1569 1570 #define DUMP_EXEC_POS(li,s,doutf8) \ 1571 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \ 1572 startpos, doutf8) 1573 1574 1575 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ 1576 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ 1577 tmp = TEST_NON_UTF8(tmp); \ 1578 REXEC_FBC_UTF8_SCAN( \ 1579 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ 1580 tmp = !tmp; \ 1581 IF_SUCCESS; \ 1582 } \ 1583 else { \ 1584 IF_FAIL; \ 1585 } \ 1586 ); \ 1587 1588 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \ 1589 if (s == reginfo->strbeg) { \ 1590 tmp = '\n'; \ 1591 } \ 1592 else { \ 1593 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \ 1594 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \ 1595 0, UTF8_ALLOW_DEFAULT); \ 1596 } \ 1597 tmp = TeSt1_UtF8; \ 1598 LOAD_UTF8_CHARCLASS_ALNUM(); \ 1599 REXEC_FBC_UTF8_SCAN( \ 1600 if (tmp == ! (TeSt2_UtF8)) { \ 1601 tmp = !tmp; \ 1602 IF_SUCCESS; \ 1603 } \ 1604 else { \ 1605 IF_FAIL; \ 1606 } \ 1607 ); \ 1608 1609 /* The only difference between the BOUND and NBOUND cases is that 1610 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in 1611 * NBOUND. This is accomplished by passing it in either the if or else clause, 1612 * with the other one being empty */ 1613 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ 1614 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) 1615 1616 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ 1617 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) 1618 1619 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ 1620 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) 1621 1622 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ 1623 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) 1624 1625 1626 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to 1627 * be passed in completely with the variable name being tested, which isn't 1628 * such a clean interface, but this is easier to read than it was before. We 1629 * are looking for the boundary (or non-boundary between a word and non-word 1630 * character. The utf8 and non-utf8 cases have the same logic, but the details 1631 * must be different. Find the "wordness" of the character just prior to this 1632 * one, and compare it with the wordness of this one. If they differ, we have 1633 * a boundary. At the beginning of the string, pretend that the previous 1634 * character was a new-line */ 1635 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ 1636 if (utf8_target) { \ 1637 UTF8_CODE \ 1638 } \ 1639 else { /* Not utf8 */ \ 1640 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ 1641 tmp = TEST_NON_UTF8(tmp); \ 1642 REXEC_FBC_SCAN( \ 1643 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ 1644 tmp = !tmp; \ 1645 IF_SUCCESS; \ 1646 } \ 1647 else { \ 1648 IF_FAIL; \ 1649 } \ 1650 ); \ 1651 } \ 1652 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \ 1653 goto got_it; 1654 1655 /* We know what class REx starts with. Try to find this position... */ 1656 /* if reginfo->intuit, its a dryrun */ 1657 /* annoyingly all the vars in this routine have different names from their counterparts 1658 in regmatch. /grrr */ 1659 1660 STATIC char * 1661 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, 1662 const char *strend, regmatch_info *reginfo) 1663 { 1664 dVAR; 1665 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0; 1666 char *pat_string; /* The pattern's exactish string */ 1667 char *pat_end; /* ptr to end char of pat_string */ 1668 re_fold_t folder; /* Function for computing non-utf8 folds */ 1669 const U8 *fold_array; /* array for folding ords < 256 */ 1670 STRLEN ln; 1671 STRLEN lnc; 1672 U8 c1; 1673 U8 c2; 1674 char *e; 1675 I32 tmp = 1; /* Scratch variable? */ 1676 const bool utf8_target = reginfo->is_utf8_target; 1677 UV utf8_fold_flags = 0; 1678 const bool is_utf8_pat = reginfo->is_utf8_pat; 1679 bool to_complement = FALSE; /* Invert the result? Taking the xor of this 1680 with a result inverts that result, as 0^1 = 1681 1 and 1^1 = 0 */ 1682 _char_class_number classnum; 1683 1684 RXi_GET_DECL(prog,progi); 1685 1686 PERL_ARGS_ASSERT_FIND_BYCLASS; 1687 1688 /* We know what class it must start with. */ 1689 switch (OP(c)) { 1690 case ANYOF: 1691 if (utf8_target) { 1692 REXEC_FBC_UTF8_CLASS_SCAN( 1693 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target)); 1694 } 1695 else { 1696 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s)); 1697 } 1698 break; 1699 case CANY: 1700 REXEC_FBC_SCAN( 1701 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) 1702 goto got_it; 1703 else 1704 tmp = doevery; 1705 ); 1706 break; 1707 1708 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ 1709 assert(! is_utf8_pat); 1710 /* FALL THROUGH */ 1711 case EXACTFA: 1712 if (is_utf8_pat || utf8_target) { 1713 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; 1714 goto do_exactf_utf8; 1715 } 1716 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */ 1717 folder = foldEQ_latin1; /* /a, except the sharp s one which */ 1718 goto do_exactf_non_utf8; /* isn't dealt with by these */ 1719 1720 case EXACTF: /* This node only generated for non-utf8 patterns */ 1721 assert(! is_utf8_pat); 1722 if (utf8_target) { 1723 utf8_fold_flags = 0; 1724 goto do_exactf_utf8; 1725 } 1726 fold_array = PL_fold; 1727 folder = foldEQ; 1728 goto do_exactf_non_utf8; 1729 1730 case EXACTFL: 1731 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) { 1732 utf8_fold_flags = FOLDEQ_LOCALE; 1733 goto do_exactf_utf8; 1734 } 1735 fold_array = PL_fold_locale; 1736 folder = foldEQ_locale; 1737 goto do_exactf_non_utf8; 1738 1739 case EXACTFU_SS: 1740 if (is_utf8_pat) { 1741 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED; 1742 } 1743 goto do_exactf_utf8; 1744 1745 case EXACTFU: 1746 if (is_utf8_pat || utf8_target) { 1747 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; 1748 goto do_exactf_utf8; 1749 } 1750 1751 /* Any 'ss' in the pattern should have been replaced by regcomp, 1752 * so we don't have to worry here about this single special case 1753 * in the Latin1 range */ 1754 fold_array = PL_fold_latin1; 1755 folder = foldEQ_latin1; 1756 1757 /* FALL THROUGH */ 1758 1759 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there 1760 are no glitches with fold-length differences 1761 between the target string and pattern */ 1762 1763 /* The idea in the non-utf8 EXACTF* cases is to first find the 1764 * first character of the EXACTF* node and then, if necessary, 1765 * case-insensitively compare the full text of the node. c1 is the 1766 * first character. c2 is its fold. This logic will not work for 1767 * Unicode semantics and the german sharp ss, which hence should 1768 * not be compiled into a node that gets here. */ 1769 pat_string = STRING(c); 1770 ln = STR_LEN(c); /* length to match in octets/bytes */ 1771 1772 /* We know that we have to match at least 'ln' bytes (which is the 1773 * same as characters, since not utf8). If we have to match 3 1774 * characters, and there are only 2 availabe, we know without 1775 * trying that it will fail; so don't start a match past the 1776 * required minimum number from the far end */ 1777 e = HOP3c(strend, -((SSize_t)ln), s); 1778 1779 if (reginfo->intuit && e < s) { 1780 e = s; /* Due to minlen logic of intuit() */ 1781 } 1782 1783 c1 = *pat_string; 1784 c2 = fold_array[c1]; 1785 if (c1 == c2) { /* If char and fold are the same */ 1786 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1); 1787 } 1788 else { 1789 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2); 1790 } 1791 break; 1792 1793 do_exactf_utf8: 1794 { 1795 unsigned expansion; 1796 1797 /* If one of the operands is in utf8, we can't use the simpler folding 1798 * above, due to the fact that many different characters can have the 1799 * same fold, or portion of a fold, or different- length fold */ 1800 pat_string = STRING(c); 1801 ln = STR_LEN(c); /* length to match in octets/bytes */ 1802 pat_end = pat_string + ln; 1803 lnc = is_utf8_pat /* length to match in characters */ 1804 ? utf8_length((U8 *) pat_string, (U8 *) pat_end) 1805 : ln; 1806 1807 /* We have 'lnc' characters to match in the pattern, but because of 1808 * multi-character folding, each character in the target can match 1809 * up to 3 characters (Unicode guarantees it will never exceed 1810 * this) if it is utf8-encoded; and up to 2 if not (based on the 1811 * fact that the Latin 1 folds are already determined, and the 1812 * only multi-char fold in that range is the sharp-s folding to 1813 * 'ss'. Thus, a pattern character can match as little as 1/3 of a 1814 * string character. Adjust lnc accordingly, rounding up, so that 1815 * if we need to match at least 4+1/3 chars, that really is 5. */ 1816 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2; 1817 lnc = (lnc + expansion - 1) / expansion; 1818 1819 /* As in the non-UTF8 case, if we have to match 3 characters, and 1820 * only 2 are left, it's guaranteed to fail, so don't start a 1821 * match that would require us to go beyond the end of the string 1822 */ 1823 e = HOP3c(strend, -((SSize_t)lnc), s); 1824 1825 if (reginfo->intuit && e < s) { 1826 e = s; /* Due to minlen logic of intuit() */ 1827 } 1828 1829 /* XXX Note that we could recalculate e to stop the loop earlier, 1830 * as the worst case expansion above will rarely be met, and as we 1831 * go along we would usually find that e moves further to the left. 1832 * This would happen only after we reached the point in the loop 1833 * where if there were no expansion we should fail. Unclear if 1834 * worth the expense */ 1835 1836 while (s <= e) { 1837 char *my_strend= (char *)strend; 1838 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target, 1839 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags) 1840 && (reginfo->intuit || regtry(reginfo, &s)) ) 1841 { 1842 goto got_it; 1843 } 1844 s += (utf8_target) ? UTF8SKIP(s) : 1; 1845 } 1846 break; 1847 } 1848 case BOUNDL: 1849 FBC_BOUND(isWORDCHAR_LC, 1850 isWORDCHAR_LC_uvchr(tmp), 1851 isWORDCHAR_LC_utf8((U8*)s)); 1852 break; 1853 case NBOUNDL: 1854 FBC_NBOUND(isWORDCHAR_LC, 1855 isWORDCHAR_LC_uvchr(tmp), 1856 isWORDCHAR_LC_utf8((U8*)s)); 1857 break; 1858 case BOUND: 1859 FBC_BOUND(isWORDCHAR, 1860 isWORDCHAR_uni(tmp), 1861 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); 1862 break; 1863 case BOUNDA: 1864 FBC_BOUND_NOLOAD(isWORDCHAR_A, 1865 isWORDCHAR_A(tmp), 1866 isWORDCHAR_A((U8*)s)); 1867 break; 1868 case NBOUND: 1869 FBC_NBOUND(isWORDCHAR, 1870 isWORDCHAR_uni(tmp), 1871 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); 1872 break; 1873 case NBOUNDA: 1874 FBC_NBOUND_NOLOAD(isWORDCHAR_A, 1875 isWORDCHAR_A(tmp), 1876 isWORDCHAR_A((U8*)s)); 1877 break; 1878 case BOUNDU: 1879 FBC_BOUND(isWORDCHAR_L1, 1880 isWORDCHAR_uni(tmp), 1881 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); 1882 break; 1883 case NBOUNDU: 1884 FBC_NBOUND(isWORDCHAR_L1, 1885 isWORDCHAR_uni(tmp), 1886 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); 1887 break; 1888 case LNBREAK: 1889 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend), 1890 is_LNBREAK_latin1_safe(s, strend) 1891 ); 1892 break; 1893 1894 /* The argument to all the POSIX node types is the class number to pass to 1895 * _generic_isCC() to build a mask for searching in PL_charclass[] */ 1896 1897 case NPOSIXL: 1898 to_complement = 1; 1899 /* FALLTHROUGH */ 1900 1901 case POSIXL: 1902 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)), 1903 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s))); 1904 break; 1905 1906 case NPOSIXD: 1907 to_complement = 1; 1908 /* FALLTHROUGH */ 1909 1910 case POSIXD: 1911 if (utf8_target) { 1912 goto posix_utf8; 1913 } 1914 goto posixa; 1915 1916 case NPOSIXA: 1917 if (utf8_target) { 1918 /* The complement of something that matches only ASCII matches all 1919 * UTF-8 variant code points, plus everything in ASCII that isn't 1920 * in the class */ 1921 REXEC_FBC_UTF8_CLASS_SCAN(! UTF8_IS_INVARIANT(*s) 1922 || ! _generic_isCC_A(*s, FLAGS(c))); 1923 break; 1924 } 1925 1926 to_complement = 1; 1927 /* FALLTHROUGH */ 1928 1929 case POSIXA: 1930 posixa: 1931 /* Don't need to worry about utf8, as it can match only a single 1932 * byte invariant character. */ 1933 REXEC_FBC_CLASS_SCAN( 1934 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c)))); 1935 break; 1936 1937 case NPOSIXU: 1938 to_complement = 1; 1939 /* FALLTHROUGH */ 1940 1941 case POSIXU: 1942 if (! utf8_target) { 1943 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s, 1944 FLAGS(c)))); 1945 } 1946 else { 1947 1948 posix_utf8: 1949 classnum = (_char_class_number) FLAGS(c); 1950 if (classnum < _FIRST_NON_SWASH_CC) { 1951 while (s < strend) { 1952 1953 /* We avoid loading in the swash as long as possible, but 1954 * should we have to, we jump to a separate loop. This 1955 * extra 'if' statement is what keeps this code from being 1956 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */ 1957 if (UTF8_IS_ABOVE_LATIN1(*s)) { 1958 goto found_above_latin1; 1959 } 1960 if ((UTF8_IS_INVARIANT(*s) 1961 && to_complement ^ cBOOL(_generic_isCC((U8) *s, 1962 classnum))) 1963 || (UTF8_IS_DOWNGRADEABLE_START(*s) 1964 && to_complement ^ cBOOL( 1965 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s, 1966 *(s + 1)), 1967 classnum)))) 1968 { 1969 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) 1970 goto got_it; 1971 else { 1972 tmp = doevery; 1973 } 1974 } 1975 else { 1976 tmp = 1; 1977 } 1978 s += UTF8SKIP(s); 1979 } 1980 } 1981 else switch (classnum) { /* These classes are implemented as 1982 macros */ 1983 case _CC_ENUM_SPACE: /* XXX would require separate code if we 1984 revert the change of \v matching this */ 1985 /* FALL THROUGH */ 1986 1987 case _CC_ENUM_PSXSPC: 1988 REXEC_FBC_UTF8_CLASS_SCAN( 1989 to_complement ^ cBOOL(isSPACE_utf8(s))); 1990 break; 1991 1992 case _CC_ENUM_BLANK: 1993 REXEC_FBC_UTF8_CLASS_SCAN( 1994 to_complement ^ cBOOL(isBLANK_utf8(s))); 1995 break; 1996 1997 case _CC_ENUM_XDIGIT: 1998 REXEC_FBC_UTF8_CLASS_SCAN( 1999 to_complement ^ cBOOL(isXDIGIT_utf8(s))); 2000 break; 2001 2002 case _CC_ENUM_VERTSPACE: 2003 REXEC_FBC_UTF8_CLASS_SCAN( 2004 to_complement ^ cBOOL(isVERTWS_utf8(s))); 2005 break; 2006 2007 case _CC_ENUM_CNTRL: 2008 REXEC_FBC_UTF8_CLASS_SCAN( 2009 to_complement ^ cBOOL(isCNTRL_utf8(s))); 2010 break; 2011 2012 default: 2013 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum); 2014 assert(0); /* NOTREACHED */ 2015 } 2016 } 2017 break; 2018 2019 found_above_latin1: /* Here we have to load a swash to get the result 2020 for the current code point */ 2021 if (! PL_utf8_swash_ptrs[classnum]) { 2022 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 2023 PL_utf8_swash_ptrs[classnum] = 2024 _core_swash_init("utf8", 2025 "", 2026 &PL_sv_undef, 1, 0, 2027 PL_XPosix_ptrs[classnum], &flags); 2028 } 2029 2030 /* This is a copy of the loop above for swash classes, though using the 2031 * FBC macro instead of being expanded out. Since we've loaded the 2032 * swash, we don't have to check for that each time through the loop */ 2033 REXEC_FBC_UTF8_CLASS_SCAN( 2034 to_complement ^ cBOOL(_generic_utf8( 2035 classnum, 2036 s, 2037 swash_fetch(PL_utf8_swash_ptrs[classnum], 2038 (U8 *) s, TRUE)))); 2039 break; 2040 2041 case AHOCORASICKC: 2042 case AHOCORASICK: 2043 { 2044 DECL_TRIE_TYPE(c); 2045 /* what trie are we using right now */ 2046 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ]; 2047 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ]; 2048 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]); 2049 2050 const char *last_start = strend - trie->minlen; 2051 #ifdef DEBUGGING 2052 const char *real_start = s; 2053 #endif 2054 STRLEN maxlen = trie->maxlen; 2055 SV *sv_points; 2056 U8 **points; /* map of where we were in the input string 2057 when reading a given char. For ASCII this 2058 is unnecessary overhead as the relationship 2059 is always 1:1, but for Unicode, especially 2060 case folded Unicode this is not true. */ 2061 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; 2062 U8 *bitmap=NULL; 2063 2064 2065 GET_RE_DEBUG_FLAGS_DECL; 2066 2067 /* We can't just allocate points here. We need to wrap it in 2068 * an SV so it gets freed properly if there is a croak while 2069 * running the match */ 2070 ENTER; 2071 SAVETMPS; 2072 sv_points=newSV(maxlen * sizeof(U8 *)); 2073 SvCUR_set(sv_points, 2074 maxlen * sizeof(U8 *)); 2075 SvPOK_on(sv_points); 2076 sv_2mortal(sv_points); 2077 points=(U8**)SvPV_nolen(sv_points ); 2078 if ( trie_type != trie_utf8_fold 2079 && (trie->bitmap || OP(c)==AHOCORASICKC) ) 2080 { 2081 if (trie->bitmap) 2082 bitmap=(U8*)trie->bitmap; 2083 else 2084 bitmap=(U8*)ANYOF_BITMAP(c); 2085 } 2086 /* this is the Aho-Corasick algorithm modified a touch 2087 to include special handling for long "unknown char" sequences. 2088 The basic idea being that we use AC as long as we are dealing 2089 with a possible matching char, when we encounter an unknown char 2090 (and we have not encountered an accepting state) we scan forward 2091 until we find a legal starting char. 2092 AC matching is basically that of trie matching, except that when 2093 we encounter a failing transition, we fall back to the current 2094 states "fail state", and try the current char again, a process 2095 we repeat until we reach the root state, state 1, or a legal 2096 transition. If we fail on the root state then we can either 2097 terminate if we have reached an accepting state previously, or 2098 restart the entire process from the beginning if we have not. 2099 2100 */ 2101 while (s <= last_start) { 2102 const U32 uniflags = UTF8_ALLOW_DEFAULT; 2103 U8 *uc = (U8*)s; 2104 U16 charid = 0; 2105 U32 base = 1; 2106 U32 state = 1; 2107 UV uvc = 0; 2108 STRLEN len = 0; 2109 STRLEN foldlen = 0; 2110 U8 *uscan = (U8*)NULL; 2111 U8 *leftmost = NULL; 2112 #ifdef DEBUGGING 2113 U32 accepted_word= 0; 2114 #endif 2115 U32 pointpos = 0; 2116 2117 while ( state && uc <= (U8*)strend ) { 2118 int failed=0; 2119 U32 word = aho->states[ state ].wordnum; 2120 2121 if( state==1 ) { 2122 if ( bitmap ) { 2123 DEBUG_TRIE_EXECUTE_r( 2124 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { 2125 dump_exec_pos( (char *)uc, c, strend, real_start, 2126 (char *)uc, utf8_target ); 2127 PerlIO_printf( Perl_debug_log, 2128 " Scanning for legal start char...\n"); 2129 } 2130 ); 2131 if (utf8_target) { 2132 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { 2133 uc += UTF8SKIP(uc); 2134 } 2135 } else { 2136 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { 2137 uc++; 2138 } 2139 } 2140 s= (char *)uc; 2141 } 2142 if (uc >(U8*)last_start) break; 2143 } 2144 2145 if ( word ) { 2146 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ]; 2147 if (!leftmost || lpos < leftmost) { 2148 DEBUG_r(accepted_word=word); 2149 leftmost= lpos; 2150 } 2151 if (base==0) break; 2152 2153 } 2154 points[pointpos++ % maxlen]= uc; 2155 if (foldlen || uc < (U8*)strend) { 2156 REXEC_TRIE_READ_CHAR(trie_type, trie, 2157 widecharmap, uc, 2158 uscan, len, uvc, charid, foldlen, 2159 foldbuf, uniflags); 2160 DEBUG_TRIE_EXECUTE_r({ 2161 dump_exec_pos( (char *)uc, c, strend, 2162 real_start, s, utf8_target); 2163 PerlIO_printf(Perl_debug_log, 2164 " Charid:%3u CP:%4"UVxf" ", 2165 charid, uvc); 2166 }); 2167 } 2168 else { 2169 len = 0; 2170 charid = 0; 2171 } 2172 2173 2174 do { 2175 #ifdef DEBUGGING 2176 word = aho->states[ state ].wordnum; 2177 #endif 2178 base = aho->states[ state ].trans.base; 2179 2180 DEBUG_TRIE_EXECUTE_r({ 2181 if (failed) 2182 dump_exec_pos( (char *)uc, c, strend, real_start, 2183 s, utf8_target ); 2184 PerlIO_printf( Perl_debug_log, 2185 "%sState: %4"UVxf", word=%"UVxf, 2186 failed ? " Fail transition to " : "", 2187 (UV)state, (UV)word); 2188 }); 2189 if ( base ) { 2190 U32 tmp; 2191 I32 offset; 2192 if (charid && 2193 ( ((offset = base + charid 2194 - 1 - trie->uniquecharcount)) >= 0) 2195 && ((U32)offset < trie->lasttrans) 2196 && trie->trans[offset].check == state 2197 && (tmp=trie->trans[offset].next)) 2198 { 2199 DEBUG_TRIE_EXECUTE_r( 2200 PerlIO_printf( Perl_debug_log," - legal\n")); 2201 state = tmp; 2202 break; 2203 } 2204 else { 2205 DEBUG_TRIE_EXECUTE_r( 2206 PerlIO_printf( Perl_debug_log," - fail\n")); 2207 failed = 1; 2208 state = aho->fail[state]; 2209 } 2210 } 2211 else { 2212 /* we must be accepting here */ 2213 DEBUG_TRIE_EXECUTE_r( 2214 PerlIO_printf( Perl_debug_log," - accepting\n")); 2215 failed = 1; 2216 break; 2217 } 2218 } while(state); 2219 uc += len; 2220 if (failed) { 2221 if (leftmost) 2222 break; 2223 if (!state) state = 1; 2224 } 2225 } 2226 if ( aho->states[ state ].wordnum ) { 2227 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ]; 2228 if (!leftmost || lpos < leftmost) { 2229 DEBUG_r(accepted_word=aho->states[ state ].wordnum); 2230 leftmost = lpos; 2231 } 2232 } 2233 if (leftmost) { 2234 s = (char*)leftmost; 2235 DEBUG_TRIE_EXECUTE_r({ 2236 PerlIO_printf( 2237 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n", 2238 (UV)accepted_word, (IV)(s - real_start) 2239 ); 2240 }); 2241 if (reginfo->intuit || regtry(reginfo, &s)) { 2242 FREETMPS; 2243 LEAVE; 2244 goto got_it; 2245 } 2246 s = HOPc(s,1); 2247 DEBUG_TRIE_EXECUTE_r({ 2248 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n"); 2249 }); 2250 } else { 2251 DEBUG_TRIE_EXECUTE_r( 2252 PerlIO_printf( Perl_debug_log,"No match.\n")); 2253 break; 2254 } 2255 } 2256 FREETMPS; 2257 LEAVE; 2258 } 2259 break; 2260 default: 2261 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c)); 2262 break; 2263 } 2264 return 0; 2265 got_it: 2266 return s; 2267 } 2268 2269 /* set RX_SAVED_COPY, RX_SUBBEG etc. 2270 * flags have same meanings as with regexec_flags() */ 2271 2272 static void 2273 S_reg_set_capture_string(pTHX_ REGEXP * const rx, 2274 char *strbeg, 2275 char *strend, 2276 SV *sv, 2277 U32 flags, 2278 bool utf8_target) 2279 { 2280 struct regexp *const prog = ReANY(rx); 2281 2282 if (flags & REXEC_COPY_STR) { 2283 #ifdef PERL_ANY_COW 2284 if (SvCANCOW(sv)) { 2285 if (DEBUG_C_TEST) { 2286 PerlIO_printf(Perl_debug_log, 2287 "Copy on write: regexp capture, type %d\n", 2288 (int) SvTYPE(sv)); 2289 } 2290 /* Create a new COW SV to share the match string and store 2291 * in saved_copy, unless the current COW SV in saved_copy 2292 * is valid and suitable for our purpose */ 2293 if (( prog->saved_copy 2294 && SvIsCOW(prog->saved_copy) 2295 && SvPOKp(prog->saved_copy) 2296 && SvIsCOW(sv) 2297 && SvPOKp(sv) 2298 && SvPVX(sv) == SvPVX(prog->saved_copy))) 2299 { 2300 /* just reuse saved_copy SV */ 2301 if (RXp_MATCH_COPIED(prog)) { 2302 Safefree(prog->subbeg); 2303 RXp_MATCH_COPIED_off(prog); 2304 } 2305 } 2306 else { 2307 /* create new COW SV to share string */ 2308 RX_MATCH_COPY_FREE(rx); 2309 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv); 2310 } 2311 prog->subbeg = (char *)SvPVX_const(prog->saved_copy); 2312 assert (SvPOKp(prog->saved_copy)); 2313 prog->sublen = strend - strbeg; 2314 prog->suboffset = 0; 2315 prog->subcoffset = 0; 2316 } else 2317 #endif 2318 { 2319 SSize_t min = 0; 2320 SSize_t max = strend - strbeg; 2321 SSize_t sublen; 2322 2323 if ( (flags & REXEC_COPY_SKIP_POST) 2324 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ 2325 && !(PL_sawampersand & SAWAMPERSAND_RIGHT) 2326 ) { /* don't copy $' part of string */ 2327 U32 n = 0; 2328 max = -1; 2329 /* calculate the right-most part of the string covered 2330 * by a capture. Due to look-ahead, this may be to 2331 * the right of $&, so we have to scan all captures */ 2332 while (n <= prog->lastparen) { 2333 if (prog->offs[n].end > max) 2334 max = prog->offs[n].end; 2335 n++; 2336 } 2337 if (max == -1) 2338 max = (PL_sawampersand & SAWAMPERSAND_LEFT) 2339 ? prog->offs[0].start 2340 : 0; 2341 assert(max >= 0 && max <= strend - strbeg); 2342 } 2343 2344 if ( (flags & REXEC_COPY_SKIP_PRE) 2345 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ 2346 && !(PL_sawampersand & SAWAMPERSAND_LEFT) 2347 ) { /* don't copy $` part of string */ 2348 U32 n = 0; 2349 min = max; 2350 /* calculate the left-most part of the string covered 2351 * by a capture. Due to look-behind, this may be to 2352 * the left of $&, so we have to scan all captures */ 2353 while (min && n <= prog->lastparen) { 2354 if ( prog->offs[n].start != -1 2355 && prog->offs[n].start < min) 2356 { 2357 min = prog->offs[n].start; 2358 } 2359 n++; 2360 } 2361 if ((PL_sawampersand & SAWAMPERSAND_RIGHT) 2362 && min > prog->offs[0].end 2363 ) 2364 min = prog->offs[0].end; 2365 2366 } 2367 2368 assert(min >= 0 && min <= max && min <= strend - strbeg); 2369 sublen = max - min; 2370 2371 if (RX_MATCH_COPIED(rx)) { 2372 if (sublen > prog->sublen) 2373 prog->subbeg = 2374 (char*)saferealloc(prog->subbeg, sublen+1); 2375 } 2376 else 2377 prog->subbeg = (char*)safemalloc(sublen+1); 2378 Copy(strbeg + min, prog->subbeg, sublen, char); 2379 prog->subbeg[sublen] = '\0'; 2380 prog->suboffset = min; 2381 prog->sublen = sublen; 2382 RX_MATCH_COPIED_on(rx); 2383 } 2384 prog->subcoffset = prog->suboffset; 2385 if (prog->suboffset && utf8_target) { 2386 /* Convert byte offset to chars. 2387 * XXX ideally should only compute this if @-/@+ 2388 * has been seen, a la PL_sawampersand ??? */ 2389 2390 /* If there's a direct correspondence between the 2391 * string which we're matching and the original SV, 2392 * then we can use the utf8 len cache associated with 2393 * the SV. In particular, it means that under //g, 2394 * sv_pos_b2u() will use the previously cached 2395 * position to speed up working out the new length of 2396 * subcoffset, rather than counting from the start of 2397 * the string each time. This stops 2398 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g; 2399 * from going quadratic */ 2400 if (SvPOKp(sv) && SvPVX(sv) == strbeg) 2401 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset, 2402 SV_GMAGIC|SV_CONST_RETURN); 2403 else 2404 prog->subcoffset = utf8_length((U8*)strbeg, 2405 (U8*)(strbeg+prog->suboffset)); 2406 } 2407 } 2408 else { 2409 RX_MATCH_COPY_FREE(rx); 2410 prog->subbeg = strbeg; 2411 prog->suboffset = 0; 2412 prog->subcoffset = 0; 2413 prog->sublen = strend - strbeg; 2414 } 2415 } 2416 2417 2418 2419 2420 /* 2421 - regexec_flags - match a regexp against a string 2422 */ 2423 I32 2424 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend, 2425 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags) 2426 /* stringarg: the point in the string at which to begin matching */ 2427 /* strend: pointer to null at end of string */ 2428 /* strbeg: real beginning of string */ 2429 /* minend: end of match must be >= minend bytes after stringarg. */ 2430 /* sv: SV being matched: only used for utf8 flag, pos() etc; string 2431 * itself is accessed via the pointers above */ 2432 /* data: May be used for some additional optimizations. 2433 Currently unused. */ 2434 /* flags: For optimizations. See REXEC_* in regexp.h */ 2435 2436 { 2437 dVAR; 2438 struct regexp *const prog = ReANY(rx); 2439 char *s; 2440 regnode *c; 2441 char *startpos; 2442 SSize_t minlen; /* must match at least this many chars */ 2443 SSize_t dontbother = 0; /* how many characters not to try at end */ 2444 const bool utf8_target = cBOOL(DO_UTF8(sv)); 2445 I32 multiline; 2446 RXi_GET_DECL(prog,progi); 2447 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */ 2448 regmatch_info *const reginfo = ®info_buf; 2449 regexp_paren_pair *swap = NULL; 2450 I32 oldsave; 2451 GET_RE_DEBUG_FLAGS_DECL; 2452 2453 PERL_ARGS_ASSERT_REGEXEC_FLAGS; 2454 PERL_UNUSED_ARG(data); 2455 2456 /* Be paranoid... */ 2457 if (prog == NULL || stringarg == NULL) { 2458 Perl_croak(aTHX_ "NULL regexp parameter"); 2459 return 0; 2460 } 2461 2462 DEBUG_EXECUTE_r( 2463 debug_start_match(rx, utf8_target, stringarg, strend, 2464 "Matching"); 2465 ); 2466 2467 startpos = stringarg; 2468 2469 if (prog->intflags & PREGf_GPOS_SEEN) { 2470 MAGIC *mg; 2471 2472 /* set reginfo->ganch, the position where \G can match */ 2473 2474 reginfo->ganch = 2475 (flags & REXEC_IGNOREPOS) 2476 ? stringarg /* use start pos rather than pos() */ 2477 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0) 2478 /* Defined pos(): */ 2479 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg) 2480 : strbeg; /* pos() not defined; use start of string */ 2481 2482 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, 2483 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg))); 2484 2485 /* in the presence of \G, we may need to start looking earlier in 2486 * the string than the suggested start point of stringarg: 2487 * if prog->gofs is set, then that's a known, fixed minimum 2488 * offset, such as 2489 * /..\G/: gofs = 2 2490 * /ab|c\G/: gofs = 1 2491 * or if the minimum offset isn't known, then we have to go back 2492 * to the start of the string, e.g. /w+\G/ 2493 */ 2494 2495 if (prog->intflags & PREGf_ANCH_GPOS) { 2496 startpos = reginfo->ganch - prog->gofs; 2497 if (startpos < 2498 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg)) 2499 { 2500 DEBUG_r(PerlIO_printf(Perl_debug_log, 2501 "fail: ganch-gofs before earliest possible start\n")); 2502 return 0; 2503 } 2504 } 2505 else if (prog->gofs) { 2506 if (startpos - prog->gofs < strbeg) 2507 startpos = strbeg; 2508 else 2509 startpos -= prog->gofs; 2510 } 2511 else if (prog->intflags & PREGf_GPOS_FLOAT) 2512 startpos = strbeg; 2513 } 2514 2515 minlen = prog->minlen; 2516 if ((startpos + minlen) > strend || startpos < strbeg) { 2517 DEBUG_r(PerlIO_printf(Perl_debug_log, 2518 "Regex match can't succeed, so not even tried\n")); 2519 return 0; 2520 } 2521 2522 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave), 2523 * which will call destuctors to reset PL_regmatch_state, free higher 2524 * PL_regmatch_slabs, and clean up regmatch_info_aux and 2525 * regmatch_info_aux_eval */ 2526 2527 oldsave = PL_savestack_ix; 2528 2529 s = startpos; 2530 2531 if ((prog->extflags & RXf_USE_INTUIT) 2532 && !(flags & REXEC_CHECKED)) 2533 { 2534 s = re_intuit_start(rx, sv, strbeg, startpos, strend, 2535 flags, NULL); 2536 if (!s) 2537 return 0; 2538 2539 if (prog->extflags & RXf_CHECK_ALL) { 2540 /* we can match based purely on the result of INTUIT. 2541 * Set up captures etc just for $& and $-[0] 2542 * (an intuit-only match wont have $1,$2,..) */ 2543 assert(!prog->nparens); 2544 2545 /* s/// doesn't like it if $& is earlier than where we asked it to 2546 * start searching (which can happen on something like /.\G/) */ 2547 if ( (flags & REXEC_FAIL_ON_UNDERFLOW) 2548 && (s < stringarg)) 2549 { 2550 /* this should only be possible under \G */ 2551 assert(prog->intflags & PREGf_GPOS_SEEN); 2552 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 2553 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); 2554 goto phooey; 2555 } 2556 2557 /* match via INTUIT shouldn't have any captures. 2558 * Let @-, @+, $^N know */ 2559 prog->lastparen = prog->lastcloseparen = 0; 2560 RX_MATCH_UTF8_set(rx, utf8_target); 2561 prog->offs[0].start = s - strbeg; 2562 prog->offs[0].end = utf8_target 2563 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg 2564 : s - strbeg + prog->minlenret; 2565 if ( !(flags & REXEC_NOT_FIRST) ) 2566 S_reg_set_capture_string(aTHX_ rx, 2567 strbeg, strend, 2568 sv, flags, utf8_target); 2569 2570 return 1; 2571 } 2572 } 2573 2574 multiline = prog->extflags & RXf_PMf_MULTILINE; 2575 2576 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { 2577 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 2578 "String too short [regexec_flags]...\n")); 2579 goto phooey; 2580 } 2581 2582 /* Check validity of program. */ 2583 if (UCHARAT(progi->program) != REG_MAGIC) { 2584 Perl_croak(aTHX_ "corrupted regexp program"); 2585 } 2586 2587 RX_MATCH_TAINTED_off(rx); 2588 2589 reginfo->prog = rx; /* Yes, sorry that this is confusing. */ 2590 reginfo->intuit = 0; 2591 reginfo->is_utf8_target = cBOOL(utf8_target); 2592 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx)); 2593 reginfo->warned = FALSE; 2594 reginfo->strbeg = strbeg; 2595 reginfo->sv = sv; 2596 reginfo->poscache_maxiter = 0; /* not yet started a countdown */ 2597 reginfo->strend = strend; 2598 /* see how far we have to get to not match where we matched before */ 2599 reginfo->till = stringarg + minend; 2600 2601 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) { 2602 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after 2603 S_cleanup_regmatch_info_aux has executed (registered by 2604 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies 2605 magic belonging to this SV. 2606 Not newSVsv, either, as it does not COW. 2607 */ 2608 assert(!IS_PADGV(sv)); 2609 reginfo->sv = newSV(0); 2610 SvSetSV_nosteal(reginfo->sv, sv); 2611 SAVEFREESV(reginfo->sv); 2612 } 2613 2614 /* reserve next 2 or 3 slots in PL_regmatch_state: 2615 * slot N+0: may currently be in use: skip it 2616 * slot N+1: use for regmatch_info_aux struct 2617 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s 2618 * slot N+3: ready for use by regmatch() 2619 */ 2620 2621 { 2622 regmatch_state *old_regmatch_state; 2623 regmatch_slab *old_regmatch_slab; 2624 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1; 2625 2626 /* on first ever match, allocate first slab */ 2627 if (!PL_regmatch_slab) { 2628 Newx(PL_regmatch_slab, 1, regmatch_slab); 2629 PL_regmatch_slab->prev = NULL; 2630 PL_regmatch_slab->next = NULL; 2631 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab); 2632 } 2633 2634 old_regmatch_state = PL_regmatch_state; 2635 old_regmatch_slab = PL_regmatch_slab; 2636 2637 for (i=0; i <= max; i++) { 2638 if (i == 1) 2639 reginfo->info_aux = &(PL_regmatch_state->u.info_aux); 2640 else if (i ==2) 2641 reginfo->info_aux_eval = 2642 reginfo->info_aux->info_aux_eval = 2643 &(PL_regmatch_state->u.info_aux_eval); 2644 2645 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab)) 2646 PL_regmatch_state = S_push_slab(aTHX); 2647 } 2648 2649 /* note initial PL_regmatch_state position; at end of match we'll 2650 * pop back to there and free any higher slabs */ 2651 2652 reginfo->info_aux->old_regmatch_state = old_regmatch_state; 2653 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab; 2654 reginfo->info_aux->poscache = NULL; 2655 2656 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux); 2657 2658 if ((prog->extflags & RXf_EVAL_SEEN)) 2659 S_setup_eval_state(aTHX_ reginfo); 2660 else 2661 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL; 2662 } 2663 2664 /* If there is a "must appear" string, look for it. */ 2665 2666 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) { 2667 /* We have to be careful. If the previous successful match 2668 was from this regex we don't want a subsequent partially 2669 successful match to clobber the old results. 2670 So when we detect this possibility we add a swap buffer 2671 to the re, and switch the buffer each match. If we fail, 2672 we switch it back; otherwise we leave it swapped. 2673 */ 2674 swap = prog->offs; 2675 /* do we need a save destructor here for eval dies? */ 2676 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair); 2677 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, 2678 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n", 2679 PTR2UV(prog), 2680 PTR2UV(swap), 2681 PTR2UV(prog->offs) 2682 )); 2683 } 2684 2685 /* Simplest case: anchored match need be tried only once. */ 2686 /* [unless only anchor is BOL and multiline is set] */ 2687 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) { 2688 if (s == startpos && regtry(reginfo, &s)) 2689 goto got_it; 2690 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */ 2691 { 2692 char *end; 2693 2694 if (minlen) 2695 dontbother = minlen - 1; 2696 end = HOP3c(strend, -dontbother, strbeg) - 1; 2697 /* for multiline we only have to try after newlines */ 2698 if (prog->check_substr || prog->check_utf8) { 2699 /* because of the goto we can not easily reuse the macros for bifurcating the 2700 unicode/non-unicode match modes here like we do elsewhere - demerphq */ 2701 if (utf8_target) { 2702 if (s == startpos) 2703 goto after_try_utf8; 2704 while (1) { 2705 if (regtry(reginfo, &s)) { 2706 goto got_it; 2707 } 2708 after_try_utf8: 2709 if (s > end) { 2710 goto phooey; 2711 } 2712 if (prog->extflags & RXf_USE_INTUIT) { 2713 s = re_intuit_start(rx, sv, strbeg, 2714 s + UTF8SKIP(s), strend, flags, NULL); 2715 if (!s) { 2716 goto phooey; 2717 } 2718 } 2719 else { 2720 s += UTF8SKIP(s); 2721 } 2722 } 2723 } /* end search for check string in unicode */ 2724 else { 2725 if (s == startpos) { 2726 goto after_try_latin; 2727 } 2728 while (1) { 2729 if (regtry(reginfo, &s)) { 2730 goto got_it; 2731 } 2732 after_try_latin: 2733 if (s > end) { 2734 goto phooey; 2735 } 2736 if (prog->extflags & RXf_USE_INTUIT) { 2737 s = re_intuit_start(rx, sv, strbeg, 2738 s + 1, strend, flags, NULL); 2739 if (!s) { 2740 goto phooey; 2741 } 2742 } 2743 else { 2744 s++; 2745 } 2746 } 2747 } /* end search for check string in latin*/ 2748 } /* end search for check string */ 2749 else { /* search for newline */ 2750 if (s > startpos) { 2751 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/ 2752 s--; 2753 } 2754 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */ 2755 while (s <= end) { /* note it could be possible to match at the end of the string */ 2756 if (*s++ == '\n') { /* don't need PL_utf8skip here */ 2757 if (regtry(reginfo, &s)) 2758 goto got_it; 2759 } 2760 } 2761 } /* end search for newline */ 2762 } /* end anchored/multiline check string search */ 2763 goto phooey; 2764 } else if (prog->intflags & PREGf_ANCH_GPOS) 2765 { 2766 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */ 2767 assert(prog->intflags & PREGf_GPOS_SEEN); 2768 /* For anchored \G, the only position it can match from is 2769 * (ganch-gofs); we already set startpos to this above; if intuit 2770 * moved us on from there, we can't possibly succeed */ 2771 assert(startpos == reginfo->ganch - prog->gofs); 2772 if (s == startpos && regtry(reginfo, &s)) 2773 goto got_it; 2774 goto phooey; 2775 } 2776 2777 /* Messy cases: unanchored match. */ 2778 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) { 2779 /* we have /x+whatever/ */ 2780 /* it must be a one character string (XXXX Except is_utf8_pat?) */ 2781 char ch; 2782 #ifdef DEBUGGING 2783 int did_match = 0; 2784 #endif 2785 if (utf8_target) { 2786 if (! prog->anchored_utf8) { 2787 to_utf8_substr(prog); 2788 } 2789 ch = SvPVX_const(prog->anchored_utf8)[0]; 2790 REXEC_FBC_SCAN( 2791 if (*s == ch) { 2792 DEBUG_EXECUTE_r( did_match = 1 ); 2793 if (regtry(reginfo, &s)) goto got_it; 2794 s += UTF8SKIP(s); 2795 while (s < strend && *s == ch) 2796 s += UTF8SKIP(s); 2797 } 2798 ); 2799 2800 } 2801 else { 2802 if (! prog->anchored_substr) { 2803 if (! to_byte_substr(prog)) { 2804 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 2805 } 2806 } 2807 ch = SvPVX_const(prog->anchored_substr)[0]; 2808 REXEC_FBC_SCAN( 2809 if (*s == ch) { 2810 DEBUG_EXECUTE_r( did_match = 1 ); 2811 if (regtry(reginfo, &s)) goto got_it; 2812 s++; 2813 while (s < strend && *s == ch) 2814 s++; 2815 } 2816 ); 2817 } 2818 DEBUG_EXECUTE_r(if (!did_match) 2819 PerlIO_printf(Perl_debug_log, 2820 "Did not find anchored character...\n") 2821 ); 2822 } 2823 else if (prog->anchored_substr != NULL 2824 || prog->anchored_utf8 != NULL 2825 || ((prog->float_substr != NULL || prog->float_utf8 != NULL) 2826 && prog->float_max_offset < strend - s)) { 2827 SV *must; 2828 SSize_t back_max; 2829 SSize_t back_min; 2830 char *last; 2831 char *last1; /* Last position checked before */ 2832 #ifdef DEBUGGING 2833 int did_match = 0; 2834 #endif 2835 if (prog->anchored_substr || prog->anchored_utf8) { 2836 if (utf8_target) { 2837 if (! prog->anchored_utf8) { 2838 to_utf8_substr(prog); 2839 } 2840 must = prog->anchored_utf8; 2841 } 2842 else { 2843 if (! prog->anchored_substr) { 2844 if (! to_byte_substr(prog)) { 2845 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 2846 } 2847 } 2848 must = prog->anchored_substr; 2849 } 2850 back_max = back_min = prog->anchored_offset; 2851 } else { 2852 if (utf8_target) { 2853 if (! prog->float_utf8) { 2854 to_utf8_substr(prog); 2855 } 2856 must = prog->float_utf8; 2857 } 2858 else { 2859 if (! prog->float_substr) { 2860 if (! to_byte_substr(prog)) { 2861 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 2862 } 2863 } 2864 must = prog->float_substr; 2865 } 2866 back_max = prog->float_max_offset; 2867 back_min = prog->float_min_offset; 2868 } 2869 2870 if (back_min<0) { 2871 last = strend; 2872 } else { 2873 last = HOP3c(strend, /* Cannot start after this */ 2874 -(SSize_t)(CHR_SVLEN(must) 2875 - (SvTAIL(must) != 0) + back_min), strbeg); 2876 } 2877 if (s > reginfo->strbeg) 2878 last1 = HOPc(s, -1); 2879 else 2880 last1 = s - 1; /* bogus */ 2881 2882 /* XXXX check_substr already used to find "s", can optimize if 2883 check_substr==must. */ 2884 dontbother = 0; 2885 strend = HOPc(strend, -dontbother); 2886 while ( (s <= last) && 2887 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend), 2888 (unsigned char*)strend, must, 2889 multiline ? FBMrf_MULTILINE : 0)) ) { 2890 DEBUG_EXECUTE_r( did_match = 1 ); 2891 if (HOPc(s, -back_max) > last1) { 2892 last1 = HOPc(s, -back_min); 2893 s = HOPc(s, -back_max); 2894 } 2895 else { 2896 char * const t = (last1 >= reginfo->strbeg) 2897 ? HOPc(last1, 1) : last1 + 1; 2898 2899 last1 = HOPc(s, -back_min); 2900 s = t; 2901 } 2902 if (utf8_target) { 2903 while (s <= last1) { 2904 if (regtry(reginfo, &s)) 2905 goto got_it; 2906 if (s >= last1) { 2907 s++; /* to break out of outer loop */ 2908 break; 2909 } 2910 s += UTF8SKIP(s); 2911 } 2912 } 2913 else { 2914 while (s <= last1) { 2915 if (regtry(reginfo, &s)) 2916 goto got_it; 2917 s++; 2918 } 2919 } 2920 } 2921 DEBUG_EXECUTE_r(if (!did_match) { 2922 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), 2923 SvPVX_const(must), RE_SV_DUMPLEN(must), 30); 2924 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n", 2925 ((must == prog->anchored_substr || must == prog->anchored_utf8) 2926 ? "anchored" : "floating"), 2927 quoted, RE_SV_TAIL(must)); 2928 }); 2929 goto phooey; 2930 } 2931 else if ( (c = progi->regstclass) ) { 2932 if (minlen) { 2933 const OPCODE op = OP(progi->regstclass); 2934 /* don't bother with what can't match */ 2935 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE) 2936 strend = HOPc(strend, -(minlen - 1)); 2937 } 2938 DEBUG_EXECUTE_r({ 2939 SV * const prop = sv_newmortal(); 2940 regprop(prog, prop, c, reginfo); 2941 { 2942 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1), 2943 s,strend-s,60); 2944 PerlIO_printf(Perl_debug_log, 2945 "Matching stclass %.*s against %s (%d bytes)\n", 2946 (int)SvCUR(prop), SvPVX_const(prop), 2947 quoted, (int)(strend - s)); 2948 } 2949 }); 2950 if (find_byclass(prog, c, s, strend, reginfo)) 2951 goto got_it; 2952 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n")); 2953 } 2954 else { 2955 dontbother = 0; 2956 if (prog->float_substr != NULL || prog->float_utf8 != NULL) { 2957 /* Trim the end. */ 2958 char *last= NULL; 2959 SV* float_real; 2960 STRLEN len; 2961 const char *little; 2962 2963 if (utf8_target) { 2964 if (! prog->float_utf8) { 2965 to_utf8_substr(prog); 2966 } 2967 float_real = prog->float_utf8; 2968 } 2969 else { 2970 if (! prog->float_substr) { 2971 if (! to_byte_substr(prog)) { 2972 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); 2973 } 2974 } 2975 float_real = prog->float_substr; 2976 } 2977 2978 little = SvPV_const(float_real, len); 2979 if (SvTAIL(float_real)) { 2980 /* This means that float_real contains an artificial \n on 2981 * the end due to the presence of something like this: 2982 * /foo$/ where we can match both "foo" and "foo\n" at the 2983 * end of the string. So we have to compare the end of the 2984 * string first against the float_real without the \n and 2985 * then against the full float_real with the string. We 2986 * have to watch out for cases where the string might be 2987 * smaller than the float_real or the float_real without 2988 * the \n. */ 2989 char *checkpos= strend - len; 2990 DEBUG_OPTIMISE_r( 2991 PerlIO_printf(Perl_debug_log, 2992 "%sChecking for float_real.%s\n", 2993 PL_colors[4], PL_colors[5])); 2994 if (checkpos + 1 < strbeg) { 2995 /* can't match, even if we remove the trailing \n 2996 * string is too short to match */ 2997 DEBUG_EXECUTE_r( 2998 PerlIO_printf(Perl_debug_log, 2999 "%sString shorter than required trailing substring, cannot match.%s\n", 3000 PL_colors[4], PL_colors[5])); 3001 goto phooey; 3002 } else if (memEQ(checkpos + 1, little, len - 1)) { 3003 /* can match, the end of the string matches without the 3004 * "\n" */ 3005 last = checkpos + 1; 3006 } else if (checkpos < strbeg) { 3007 /* cant match, string is too short when the "\n" is 3008 * included */ 3009 DEBUG_EXECUTE_r( 3010 PerlIO_printf(Perl_debug_log, 3011 "%sString does not contain required trailing substring, cannot match.%s\n", 3012 PL_colors[4], PL_colors[5])); 3013 goto phooey; 3014 } else if (!multiline) { 3015 /* non multiline match, so compare with the "\n" at the 3016 * end of the string */ 3017 if (memEQ(checkpos, little, len)) { 3018 last= checkpos; 3019 } else { 3020 DEBUG_EXECUTE_r( 3021 PerlIO_printf(Perl_debug_log, 3022 "%sString does not contain required trailing substring, cannot match.%s\n", 3023 PL_colors[4], PL_colors[5])); 3024 goto phooey; 3025 } 3026 } else { 3027 /* multiline match, so we have to search for a place 3028 * where the full string is located */ 3029 goto find_last; 3030 } 3031 } else { 3032 find_last: 3033 if (len) 3034 last = rninstr(s, strend, little, little + len); 3035 else 3036 last = strend; /* matching "$" */ 3037 } 3038 if (!last) { 3039 /* at one point this block contained a comment which was 3040 * probably incorrect, which said that this was a "should not 3041 * happen" case. Even if it was true when it was written I am 3042 * pretty sure it is not anymore, so I have removed the comment 3043 * and replaced it with this one. Yves */ 3044 DEBUG_EXECUTE_r( 3045 PerlIO_printf(Perl_debug_log, 3046 "String does not contain required substring, cannot match.\n" 3047 )); 3048 goto phooey; 3049 } 3050 dontbother = strend - last + prog->float_min_offset; 3051 } 3052 if (minlen && (dontbother < minlen)) 3053 dontbother = minlen - 1; 3054 strend -= dontbother; /* this one's always in bytes! */ 3055 /* We don't know much -- general case. */ 3056 if (utf8_target) { 3057 for (;;) { 3058 if (regtry(reginfo, &s)) 3059 goto got_it; 3060 if (s >= strend) 3061 break; 3062 s += UTF8SKIP(s); 3063 }; 3064 } 3065 else { 3066 do { 3067 if (regtry(reginfo, &s)) 3068 goto got_it; 3069 } while (s++ < strend); 3070 } 3071 } 3072 3073 /* Failure. */ 3074 goto phooey; 3075 3076 got_it: 3077 /* s/// doesn't like it if $& is earlier than where we asked it to 3078 * start searching (which can happen on something like /.\G/) */ 3079 if ( (flags & REXEC_FAIL_ON_UNDERFLOW) 3080 && (prog->offs[0].start < stringarg - strbeg)) 3081 { 3082 /* this should only be possible under \G */ 3083 assert(prog->intflags & PREGf_GPOS_SEEN); 3084 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 3085 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); 3086 goto phooey; 3087 } 3088 3089 DEBUG_BUFFERS_r( 3090 if (swap) 3091 PerlIO_printf(Perl_debug_log, 3092 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n", 3093 PTR2UV(prog), 3094 PTR2UV(swap) 3095 ); 3096 ); 3097 Safefree(swap); 3098 3099 /* clean up; this will trigger destructors that will free all slabs 3100 * above the current one, and cleanup the regmatch_info_aux 3101 * and regmatch_info_aux_eval sructs */ 3102 3103 LEAVE_SCOPE(oldsave); 3104 3105 if (RXp_PAREN_NAMES(prog)) 3106 (void)hv_iterinit(RXp_PAREN_NAMES(prog)); 3107 3108 RX_MATCH_UTF8_set(rx, utf8_target); 3109 3110 /* make sure $`, $&, $', and $digit will work later */ 3111 if ( !(flags & REXEC_NOT_FIRST) ) 3112 S_reg_set_capture_string(aTHX_ rx, 3113 strbeg, reginfo->strend, 3114 sv, flags, utf8_target); 3115 3116 return 1; 3117 3118 phooey: 3119 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n", 3120 PL_colors[4], PL_colors[5])); 3121 3122 /* clean up; this will trigger destructors that will free all slabs 3123 * above the current one, and cleanup the regmatch_info_aux 3124 * and regmatch_info_aux_eval sructs */ 3125 3126 LEAVE_SCOPE(oldsave); 3127 3128 if (swap) { 3129 /* we failed :-( roll it back */ 3130 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, 3131 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n", 3132 PTR2UV(prog), 3133 PTR2UV(prog->offs), 3134 PTR2UV(swap) 3135 )); 3136 Safefree(prog->offs); 3137 prog->offs = swap; 3138 } 3139 return 0; 3140 } 3141 3142 3143 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting. 3144 * Do inc before dec, in case old and new rex are the same */ 3145 #define SET_reg_curpm(Re2) \ 3146 if (reginfo->info_aux_eval) { \ 3147 (void)ReREFCNT_inc(Re2); \ 3148 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \ 3149 PM_SETRE((PL_reg_curpm), (Re2)); \ 3150 } 3151 3152 3153 /* 3154 - regtry - try match at specific point 3155 */ 3156 STATIC I32 /* 0 failure, 1 success */ 3157 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp) 3158 { 3159 dVAR; 3160 CHECKPOINT lastcp; 3161 REGEXP *const rx = reginfo->prog; 3162 regexp *const prog = ReANY(rx); 3163 SSize_t result; 3164 RXi_GET_DECL(prog,progi); 3165 GET_RE_DEBUG_FLAGS_DECL; 3166 3167 PERL_ARGS_ASSERT_REGTRY; 3168 3169 reginfo->cutpoint=NULL; 3170 3171 prog->offs[0].start = *startposp - reginfo->strbeg; 3172 prog->lastparen = 0; 3173 prog->lastcloseparen = 0; 3174 3175 /* XXXX What this code is doing here?!!! There should be no need 3176 to do this again and again, prog->lastparen should take care of 3177 this! --ilya*/ 3178 3179 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code. 3180 * Actually, the code in regcppop() (which Ilya may be meaning by 3181 * prog->lastparen), is not needed at all by the test suite 3182 * (op/regexp, op/pat, op/split), but that code is needed otherwise 3183 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ 3184 * Meanwhile, this code *is* needed for the 3185 * above-mentioned test suite tests to succeed. The common theme 3186 * on those tests seems to be returning null fields from matches. 3187 * --jhi updated by dapm */ 3188 #if 1 3189 if (prog->nparens) { 3190 regexp_paren_pair *pp = prog->offs; 3191 I32 i; 3192 for (i = prog->nparens; i > (I32)prog->lastparen; i--) { 3193 ++pp; 3194 pp->start = -1; 3195 pp->end = -1; 3196 } 3197 } 3198 #endif 3199 REGCP_SET(lastcp); 3200 result = regmatch(reginfo, *startposp, progi->program + 1); 3201 if (result != -1) { 3202 prog->offs[0].end = result; 3203 return 1; 3204 } 3205 if (reginfo->cutpoint) 3206 *startposp= reginfo->cutpoint; 3207 REGCP_UNWIND(lastcp); 3208 return 0; 3209 } 3210 3211 3212 #define sayYES goto yes 3213 #define sayNO goto no 3214 #define sayNO_SILENT goto no_silent 3215 3216 /* we dont use STMT_START/END here because it leads to 3217 "unreachable code" warnings, which are bogus, but distracting. */ 3218 #define CACHEsayNO \ 3219 if (ST.cache_mask) \ 3220 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \ 3221 sayNO 3222 3223 /* this is used to determine how far from the left messages like 3224 'failed...' are printed. It should be set such that messages 3225 are inline with the regop output that created them. 3226 */ 3227 #define REPORT_CODE_OFF 32 3228 3229 3230 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */ 3231 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */ 3232 #define CHRTEST_NOT_A_CP_1 -999 3233 #define CHRTEST_NOT_A_CP_2 -998 3234 3235 /* grab a new slab and return the first slot in it */ 3236 3237 STATIC regmatch_state * 3238 S_push_slab(pTHX) 3239 { 3240 #if PERL_VERSION < 9 && !defined(PERL_CORE) 3241 dMY_CXT; 3242 #endif 3243 regmatch_slab *s = PL_regmatch_slab->next; 3244 if (!s) { 3245 Newx(s, 1, regmatch_slab); 3246 s->prev = PL_regmatch_slab; 3247 s->next = NULL; 3248 PL_regmatch_slab->next = s; 3249 } 3250 PL_regmatch_slab = s; 3251 return SLAB_FIRST(s); 3252 } 3253 3254 3255 /* push a new state then goto it */ 3256 3257 #define PUSH_STATE_GOTO(state, node, input) \ 3258 pushinput = input; \ 3259 scan = node; \ 3260 st->resume_state = state; \ 3261 goto push_state; 3262 3263 /* push a new state with success backtracking, then goto it */ 3264 3265 #define PUSH_YES_STATE_GOTO(state, node, input) \ 3266 pushinput = input; \ 3267 scan = node; \ 3268 st->resume_state = state; \ 3269 goto push_yes_state; 3270 3271 3272 3273 3274 /* 3275 3276 regmatch() - main matching routine 3277 3278 This is basically one big switch statement in a loop. We execute an op, 3279 set 'next' to point the next op, and continue. If we come to a point which 3280 we may need to backtrack to on failure such as (A|B|C), we push a 3281 backtrack state onto the backtrack stack. On failure, we pop the top 3282 state, and re-enter the loop at the state indicated. If there are no more 3283 states to pop, we return failure. 3284 3285 Sometimes we also need to backtrack on success; for example /A+/, where 3286 after successfully matching one A, we need to go back and try to 3287 match another one; similarly for lookahead assertions: if the assertion 3288 completes successfully, we backtrack to the state just before the assertion 3289 and then carry on. In these cases, the pushed state is marked as 3290 'backtrack on success too'. This marking is in fact done by a chain of 3291 pointers, each pointing to the previous 'yes' state. On success, we pop to 3292 the nearest yes state, discarding any intermediate failure-only states. 3293 Sometimes a yes state is pushed just to force some cleanup code to be 3294 called at the end of a successful match or submatch; e.g. (??{$re}) uses 3295 it to free the inner regex. 3296 3297 Note that failure backtracking rewinds the cursor position, while 3298 success backtracking leaves it alone. 3299 3300 A pattern is complete when the END op is executed, while a subpattern 3301 such as (?=foo) is complete when the SUCCESS op is executed. Both of these 3302 ops trigger the "pop to last yes state if any, otherwise return true" 3303 behaviour. 3304 3305 A common convention in this function is to use A and B to refer to the two 3306 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is 3307 the subpattern to be matched possibly multiple times, while B is the entire 3308 rest of the pattern. Variable and state names reflect this convention. 3309 3310 The states in the main switch are the union of ops and failure/success of 3311 substates associated with with that op. For example, IFMATCH is the op 3312 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means 3313 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just 3314 successfully matched A and IFMATCH_A_fail is a state saying that we have 3315 just failed to match A. Resume states always come in pairs. The backtrack 3316 state we push is marked as 'IFMATCH_A', but when that is popped, we resume 3317 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking 3318 on success or failure. 3319 3320 The struct that holds a backtracking state is actually a big union, with 3321 one variant for each major type of op. The variable st points to the 3322 top-most backtrack struct. To make the code clearer, within each 3323 block of code we #define ST to alias the relevant union. 3324 3325 Here's a concrete example of a (vastly oversimplified) IFMATCH 3326 implementation: 3327 3328 switch (state) { 3329 .... 3330 3331 #define ST st->u.ifmatch 3332 3333 case IFMATCH: // we are executing the IFMATCH op, (?=A)B 3334 ST.foo = ...; // some state we wish to save 3335 ... 3336 // push a yes backtrack state with a resume value of 3337 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the 3338 // first node of A: 3339 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput); 3340 // NOTREACHED 3341 3342 case IFMATCH_A: // we have successfully executed A; now continue with B 3343 next = B; 3344 bar = ST.foo; // do something with the preserved value 3345 break; 3346 3347 case IFMATCH_A_fail: // A failed, so the assertion failed 3348 ...; // do some housekeeping, then ... 3349 sayNO; // propagate the failure 3350 3351 #undef ST 3352 3353 ... 3354 } 3355 3356 For any old-timers reading this who are familiar with the old recursive 3357 approach, the code above is equivalent to: 3358 3359 case IFMATCH: // we are executing the IFMATCH op, (?=A)B 3360 { 3361 int foo = ... 3362 ... 3363 if (regmatch(A)) { 3364 next = B; 3365 bar = foo; 3366 break; 3367 } 3368 ...; // do some housekeeping, then ... 3369 sayNO; // propagate the failure 3370 } 3371 3372 The topmost backtrack state, pointed to by st, is usually free. If you 3373 want to claim it, populate any ST.foo fields in it with values you wish to 3374 save, then do one of 3375 3376 PUSH_STATE_GOTO(resume_state, node, newinput); 3377 PUSH_YES_STATE_GOTO(resume_state, node, newinput); 3378 3379 which sets that backtrack state's resume value to 'resume_state', pushes a 3380 new free entry to the top of the backtrack stack, then goes to 'node'. 3381 On backtracking, the free slot is popped, and the saved state becomes the 3382 new free state. An ST.foo field in this new top state can be temporarily 3383 accessed to retrieve values, but once the main loop is re-entered, it 3384 becomes available for reuse. 3385 3386 Note that the depth of the backtrack stack constantly increases during the 3387 left-to-right execution of the pattern, rather than going up and down with 3388 the pattern nesting. For example the stack is at its maximum at Z at the 3389 end of the pattern, rather than at X in the following: 3390 3391 /(((X)+)+)+....(Y)+....Z/ 3392 3393 The only exceptions to this are lookahead/behind assertions and the cut, 3394 (?>A), which pop all the backtrack states associated with A before 3395 continuing. 3396 3397 Backtrack state structs are allocated in slabs of about 4K in size. 3398 PL_regmatch_state and st always point to the currently active state, 3399 and PL_regmatch_slab points to the slab currently containing 3400 PL_regmatch_state. The first time regmatch() is called, the first slab is 3401 allocated, and is never freed until interpreter destruction. When the slab 3402 is full, a new one is allocated and chained to the end. At exit from 3403 regmatch(), slabs allocated since entry are freed. 3404 3405 */ 3406 3407 3408 #define DEBUG_STATE_pp(pp) \ 3409 DEBUG_STATE_r({ \ 3410 DUMP_EXEC_POS(locinput, scan, utf8_target); \ 3411 PerlIO_printf(Perl_debug_log, \ 3412 " %*s"pp" %s%s%s%s%s\n", \ 3413 depth*2, "", \ 3414 PL_reg_name[st->resume_state], \ 3415 ((st==yes_state||st==mark_state) ? "[" : ""), \ 3416 ((st==yes_state) ? "Y" : ""), \ 3417 ((st==mark_state) ? "M" : ""), \ 3418 ((st==yes_state||st==mark_state) ? "]" : "") \ 3419 ); \ 3420 }); 3421 3422 3423 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) 3424 3425 #ifdef DEBUGGING 3426 3427 STATIC void 3428 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target, 3429 const char *start, const char *end, const char *blurb) 3430 { 3431 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0; 3432 3433 PERL_ARGS_ASSERT_DEBUG_START_MATCH; 3434 3435 if (!PL_colorset) 3436 reginitcolors(); 3437 { 3438 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0), 3439 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60); 3440 3441 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1), 3442 start, end - start, 60); 3443 3444 PerlIO_printf(Perl_debug_log, 3445 "%s%s REx%s %s against %s\n", 3446 PL_colors[4], blurb, PL_colors[5], s0, s1); 3447 3448 if (utf8_target||utf8_pat) 3449 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n", 3450 utf8_pat ? "pattern" : "", 3451 utf8_pat && utf8_target ? " and " : "", 3452 utf8_target ? "string" : "" 3453 ); 3454 } 3455 } 3456 3457 STATIC void 3458 S_dump_exec_pos(pTHX_ const char *locinput, 3459 const regnode *scan, 3460 const char *loc_regeol, 3461 const char *loc_bostr, 3462 const char *loc_reg_starttry, 3463 const bool utf8_target) 3464 { 3465 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; 3466 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ 3467 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); 3468 /* The part of the string before starttry has one color 3469 (pref0_len chars), between starttry and current 3470 position another one (pref_len - pref0_len chars), 3471 after the current position the third one. 3472 We assume that pref0_len <= pref_len, otherwise we 3473 decrease pref0_len. */ 3474 int pref_len = (locinput - loc_bostr) > (5 + taill) - l 3475 ? (5 + taill) - l : locinput - loc_bostr; 3476 int pref0_len; 3477 3478 PERL_ARGS_ASSERT_DUMP_EXEC_POS; 3479 3480 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) 3481 pref_len++; 3482 pref0_len = pref_len - (locinput - loc_reg_starttry); 3483 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) 3484 l = ( loc_regeol - locinput > (5 + taill) - pref_len 3485 ? (5 + taill) - pref_len : loc_regeol - locinput); 3486 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) 3487 l--; 3488 if (pref0_len < 0) 3489 pref0_len = 0; 3490 if (pref0_len > pref_len) 3491 pref0_len = pref_len; 3492 { 3493 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0; 3494 3495 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0), 3496 (locinput - pref_len),pref0_len, 60, 4, 5); 3497 3498 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1), 3499 (locinput - pref_len + pref0_len), 3500 pref_len - pref0_len, 60, 2, 3); 3501 3502 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), 3503 locinput, loc_regeol - locinput, 10, 0, 1); 3504 3505 const STRLEN tlen=len0+len1+len2; 3506 PerlIO_printf(Perl_debug_log, 3507 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|", 3508 (IV)(locinput - loc_bostr), 3509 len0, s0, 3510 len1, s1, 3511 (docolor ? "" : "> <"), 3512 len2, s2, 3513 (int)(tlen > 19 ? 0 : 19 - tlen), 3514 ""); 3515 } 3516 } 3517 3518 #endif 3519 3520 /* reg_check_named_buff_matched() 3521 * Checks to see if a named buffer has matched. The data array of 3522 * buffer numbers corresponding to the buffer is expected to reside 3523 * in the regexp->data->data array in the slot stored in the ARG() of 3524 * node involved. Note that this routine doesn't actually care about the 3525 * name, that information is not preserved from compilation to execution. 3526 * Returns the index of the leftmost defined buffer with the given name 3527 * or 0 if non of the buffers matched. 3528 */ 3529 STATIC I32 3530 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) 3531 { 3532 I32 n; 3533 RXi_GET_DECL(rex,rexi); 3534 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 3535 I32 *nums=(I32*)SvPVX(sv_dat); 3536 3537 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED; 3538 3539 for ( n=0; n<SvIVX(sv_dat); n++ ) { 3540 if ((I32)rex->lastparen >= nums[n] && 3541 rex->offs[nums[n]].end != -1) 3542 { 3543 return nums[n]; 3544 } 3545 } 3546 return 0; 3547 } 3548 3549 3550 static bool 3551 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, 3552 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo) 3553 { 3554 /* This function determines if there are one or two characters that match 3555 * the first character of the passed-in EXACTish node <text_node>, and if 3556 * so, returns them in the passed-in pointers. 3557 * 3558 * If it determines that no possible character in the target string can 3559 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if 3560 * the first character in <text_node> requires UTF-8 to represent, and the 3561 * target string isn't in UTF-8.) 3562 * 3563 * If there are more than two characters that could match the beginning of 3564 * <text_node>, or if more context is required to determine a match or not, 3565 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID. 3566 * 3567 * The motiviation behind this function is to allow the caller to set up 3568 * tight loops for matching. If <text_node> is of type EXACT, there is 3569 * only one possible character that can match its first character, and so 3570 * the situation is quite simple. But things get much more complicated if 3571 * folding is involved. It may be that the first character of an EXACTFish 3572 * node doesn't participate in any possible fold, e.g., punctuation, so it 3573 * can be matched only by itself. The vast majority of characters that are 3574 * in folds match just two things, their lower and upper-case equivalents. 3575 * But not all are like that; some have multiple possible matches, or match 3576 * sequences of more than one character. This function sorts all that out. 3577 * 3578 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a 3579 * loop of trying to match A*, we know we can't exit where the thing 3580 * following it isn't a B. And something can't be a B unless it is the 3581 * beginning of B. By putting a quick test for that beginning in a tight 3582 * loop, we can rule out things that can't possibly be B without having to 3583 * break out of the loop, thus avoiding work. Similarly, if A is a single 3584 * character, we can make a tight loop matching A*, using the outputs of 3585 * this function. 3586 * 3587 * If the target string to match isn't in UTF-8, and there aren't 3588 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to 3589 * the one or two possible octets (which are characters in this situation) 3590 * that can match. In all cases, if there is only one character that can 3591 * match, *<c1p> and *<c2p> will be identical. 3592 * 3593 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8> 3594 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that 3595 * can match the beginning of <text_node>. They should be declared with at 3596 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is 3597 * undefined what these contain.) If one or both of the buffers are 3598 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the 3599 * corresponding invariant. If variant, the corresponding *<c1p> and/or 3600 * *<c2p> will be set to a negative number(s) that shouldn't match any code 3601 * point (unless inappropriately coerced to unsigned). *<c1p> will equal 3602 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */ 3603 3604 const bool utf8_target = reginfo->is_utf8_target; 3605 3606 UV c1 = CHRTEST_NOT_A_CP_1; 3607 UV c2 = CHRTEST_NOT_A_CP_2; 3608 bool use_chrtest_void = FALSE; 3609 const bool is_utf8_pat = reginfo->is_utf8_pat; 3610 3611 /* Used when we have both utf8 input and utf8 output, to avoid converting 3612 * to/from code points */ 3613 bool utf8_has_been_setup = FALSE; 3614 3615 dVAR; 3616 3617 U8 *pat = (U8*)STRING(text_node); 3618 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' }; 3619 3620 if (OP(text_node) == EXACT) { 3621 3622 /* In an exact node, only one thing can be matched, that first 3623 * character. If both the pat and the target are UTF-8, we can just 3624 * copy the input to the output, avoiding finding the code point of 3625 * that character */ 3626 if (!is_utf8_pat) { 3627 c2 = c1 = *pat; 3628 } 3629 else if (utf8_target) { 3630 Copy(pat, c1_utf8, UTF8SKIP(pat), U8); 3631 Copy(pat, c2_utf8, UTF8SKIP(pat), U8); 3632 utf8_has_been_setup = TRUE; 3633 } 3634 else { 3635 c2 = c1 = valid_utf8_to_uvchr(pat, NULL); 3636 } 3637 } 3638 else { /* an EXACTFish node */ 3639 U8 *pat_end = pat + STR_LEN(text_node); 3640 3641 /* An EXACTFL node has at least some characters unfolded, because what 3642 * they match is not known until now. So, now is the time to fold 3643 * the first few of them, as many as are needed to determine 'c1' and 3644 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need 3645 * to fold if in a UTF-8 locale, and then only the Sharp S; everything 3646 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we 3647 * need to fold as many characters as a single character can fold to, 3648 * so that later we can check if the first ones are such a multi-char 3649 * fold. But, in such a pattern only locale-problematic characters 3650 * aren't folded, so we can skip this completely if the first character 3651 * in the node isn't one of the tricky ones */ 3652 if (OP(text_node) == EXACTFL) { 3653 3654 if (! is_utf8_pat) { 3655 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S) 3656 { 3657 folded[0] = folded[1] = 's'; 3658 pat = folded; 3659 pat_end = folded + 2; 3660 } 3661 } 3662 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) { 3663 U8 *s = pat; 3664 U8 *d = folded; 3665 int i; 3666 3667 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) { 3668 if (isASCII(*s)) { 3669 *(d++) = (U8) toFOLD_LC(*s); 3670 s++; 3671 } 3672 else { 3673 STRLEN len; 3674 _to_utf8_fold_flags(s, 3675 d, 3676 &len, 3677 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE); 3678 d += len; 3679 s += UTF8SKIP(s); 3680 } 3681 } 3682 3683 pat = folded; 3684 pat_end = d; 3685 } 3686 } 3687 3688 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end)) 3689 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end))) 3690 { 3691 /* Multi-character folds require more context to sort out. Also 3692 * PL_utf8_foldclosures used below doesn't handle them, so have to 3693 * be handled outside this routine */ 3694 use_chrtest_void = TRUE; 3695 } 3696 else { /* an EXACTFish node which doesn't begin with a multi-char fold */ 3697 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat; 3698 if (c1 > 256) { 3699 /* Load the folds hash, if not already done */ 3700 SV** listp; 3701 if (! PL_utf8_foldclosures) { 3702 if (! PL_utf8_tofold) { 3703 U8 dummy[UTF8_MAXBYTES_CASE+1]; 3704 3705 /* Force loading this by folding an above-Latin1 char */ 3706 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL); 3707 assert(PL_utf8_tofold); /* Verify that worked */ 3708 } 3709 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold); 3710 } 3711 3712 /* The fold closures data structure is a hash with the keys 3713 * being the UTF-8 of every character that is folded to, like 3714 * 'k', and the values each an array of all code points that 3715 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ]. 3716 * Multi-character folds are not included */ 3717 if ((! (listp = hv_fetch(PL_utf8_foldclosures, 3718 (char *) pat, 3719 UTF8SKIP(pat), 3720 FALSE)))) 3721 { 3722 /* Not found in the hash, therefore there are no folds 3723 * containing it, so there is only a single character that 3724 * could match */ 3725 c2 = c1; 3726 } 3727 else { /* Does participate in folds */ 3728 AV* list = (AV*) *listp; 3729 if (av_tindex(list) != 1) { 3730 3731 /* If there aren't exactly two folds to this, it is 3732 * outside the scope of this function */ 3733 use_chrtest_void = TRUE; 3734 } 3735 else { /* There are two. Get them */ 3736 SV** c_p = av_fetch(list, 0, FALSE); 3737 if (c_p == NULL) { 3738 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); 3739 } 3740 c1 = SvUV(*c_p); 3741 3742 c_p = av_fetch(list, 1, FALSE); 3743 if (c_p == NULL) { 3744 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); 3745 } 3746 c2 = SvUV(*c_p); 3747 3748 /* Folds that cross the 255/256 boundary are forbidden 3749 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and 3750 * one is ASCIII. Since the pattern character is above 3751 * 256, and its only other match is below 256, the only 3752 * legal match will be to itself. We have thrown away 3753 * the original, so have to compute which is the one 3754 * above 255 */ 3755 if ((c1 < 256) != (c2 < 256)) { 3756 if ((OP(text_node) == EXACTFL 3757 && ! IN_UTF8_CTYPE_LOCALE) 3758 || ((OP(text_node) == EXACTFA 3759 || OP(text_node) == EXACTFA_NO_TRIE) 3760 && (isASCII(c1) || isASCII(c2)))) 3761 { 3762 if (c1 < 256) { 3763 c1 = c2; 3764 } 3765 else { 3766 c2 = c1; 3767 } 3768 } 3769 } 3770 } 3771 } 3772 } 3773 else /* Here, c1 is < 255 */ 3774 if (utf8_target 3775 && HAS_NONLATIN1_FOLD_CLOSURE(c1) 3776 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE)) 3777 && ((OP(text_node) != EXACTFA 3778 && OP(text_node) != EXACTFA_NO_TRIE) 3779 || ! isASCII(c1))) 3780 { 3781 /* Here, there could be something above Latin1 in the target 3782 * which folds to this character in the pattern. All such 3783 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more 3784 * than two characters involved in their folds, so are outside 3785 * the scope of this function */ 3786 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { 3787 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; 3788 } 3789 else { 3790 use_chrtest_void = TRUE; 3791 } 3792 } 3793 else { /* Here nothing above Latin1 can fold to the pattern 3794 character */ 3795 switch (OP(text_node)) { 3796 3797 case EXACTFL: /* /l rules */ 3798 c2 = PL_fold_locale[c1]; 3799 break; 3800 3801 case EXACTF: /* This node only generated for non-utf8 3802 patterns */ 3803 assert(! is_utf8_pat); 3804 if (! utf8_target) { /* /d rules */ 3805 c2 = PL_fold[c1]; 3806 break; 3807 } 3808 /* FALLTHROUGH */ 3809 /* /u rules for all these. This happens to work for 3810 * EXACTFA as nothing in Latin1 folds to ASCII */ 3811 case EXACTFA_NO_TRIE: /* This node only generated for 3812 non-utf8 patterns */ 3813 assert(! is_utf8_pat); 3814 /* FALL THROUGH */ 3815 case EXACTFA: 3816 case EXACTFU_SS: 3817 case EXACTFU: 3818 c2 = PL_fold_latin1[c1]; 3819 break; 3820 3821 default: 3822 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node)); 3823 assert(0); /* NOTREACHED */ 3824 } 3825 } 3826 } 3827 } 3828 3829 /* Here have figured things out. Set up the returns */ 3830 if (use_chrtest_void) { 3831 *c2p = *c1p = CHRTEST_VOID; 3832 } 3833 else if (utf8_target) { 3834 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */ 3835 uvchr_to_utf8(c1_utf8, c1); 3836 uvchr_to_utf8(c2_utf8, c2); 3837 } 3838 3839 /* Invariants are stored in both the utf8 and byte outputs; Use 3840 * negative numbers otherwise for the byte ones. Make sure that the 3841 * byte ones are the same iff the utf8 ones are the same */ 3842 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1; 3843 *c2p = (UTF8_IS_INVARIANT(*c2_utf8)) 3844 ? *c2_utf8 3845 : (c1 == c2) 3846 ? CHRTEST_NOT_A_CP_1 3847 : CHRTEST_NOT_A_CP_2; 3848 } 3849 else if (c1 > 255) { 3850 if (c2 > 255) { /* both possibilities are above what a non-utf8 string 3851 can represent */ 3852 return FALSE; 3853 } 3854 3855 *c1p = *c2p = c2; /* c2 is the only representable value */ 3856 } 3857 else { /* c1 is representable; see about c2 */ 3858 *c1p = c1; 3859 *c2p = (c2 < 256) ? c2 : c1; 3860 } 3861 3862 return TRUE; 3863 } 3864 3865 /* returns -1 on failure, $+[0] on success */ 3866 STATIC SSize_t 3867 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog) 3868 { 3869 #if PERL_VERSION < 9 && !defined(PERL_CORE) 3870 dMY_CXT; 3871 #endif 3872 dVAR; 3873 const bool utf8_target = reginfo->is_utf8_target; 3874 const U32 uniflags = UTF8_ALLOW_DEFAULT; 3875 REGEXP *rex_sv = reginfo->prog; 3876 regexp *rex = ReANY(rex_sv); 3877 RXi_GET_DECL(rex,rexi); 3878 /* the current state. This is a cached copy of PL_regmatch_state */ 3879 regmatch_state *st; 3880 /* cache heavy used fields of st in registers */ 3881 regnode *scan; 3882 regnode *next; 3883 U32 n = 0; /* general value; init to avoid compiler warning */ 3884 SSize_t ln = 0; /* len or last; init to avoid compiler warning */ 3885 char *locinput = startpos; 3886 char *pushinput; /* where to continue after a PUSH */ 3887 I32 nextchr; /* is always set to UCHARAT(locinput) */ 3888 3889 bool result = 0; /* return value of S_regmatch */ 3890 int depth = 0; /* depth of backtrack stack */ 3891 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */ 3892 const U32 max_nochange_depth = 3893 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ? 3894 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH; 3895 regmatch_state *yes_state = NULL; /* state to pop to on success of 3896 subpattern */ 3897 /* mark_state piggy backs on the yes_state logic so that when we unwind 3898 the stack on success we can update the mark_state as we go */ 3899 regmatch_state *mark_state = NULL; /* last mark state we have seen */ 3900 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ 3901 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ 3902 U32 state_num; 3903 bool no_final = 0; /* prevent failure from backtracking? */ 3904 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */ 3905 char *startpoint = locinput; 3906 SV *popmark = NULL; /* are we looking for a mark? */ 3907 SV *sv_commit = NULL; /* last mark name seen in failure */ 3908 SV *sv_yes_mark = NULL; /* last mark name we have seen 3909 during a successful match */ 3910 U32 lastopen = 0; /* last open we saw */ 3911 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0; 3912 SV* const oreplsv = GvSVn(PL_replgv); 3913 /* these three flags are set by various ops to signal information to 3914 * the very next op. They have a useful lifetime of exactly one loop 3915 * iteration, and are not preserved or restored by state pushes/pops 3916 */ 3917 bool sw = 0; /* the condition value in (?(cond)a|b) */ 3918 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ 3919 int logical = 0; /* the following EVAL is: 3920 0: (?{...}) 3921 1: (?(?{...})X|Y) 3922 2: (??{...}) 3923 or the following IFMATCH/UNLESSM is: 3924 false: plain (?=foo) 3925 true: used as a condition: (?(?=foo)) 3926 */ 3927 PAD* last_pad = NULL; 3928 dMULTICALL; 3929 I32 gimme = G_SCALAR; 3930 CV *caller_cv = NULL; /* who called us */ 3931 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */ 3932 CHECKPOINT runops_cp; /* savestack position before executing EVAL */ 3933 U32 maxopenparen = 0; /* max '(' index seen so far */ 3934 int to_complement; /* Invert the result? */ 3935 _char_class_number classnum; 3936 bool is_utf8_pat = reginfo->is_utf8_pat; 3937 3938 #ifdef DEBUGGING 3939 GET_RE_DEBUG_FLAGS_DECL; 3940 #endif 3941 3942 /* protect against undef(*^R) */ 3943 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv)); 3944 3945 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */ 3946 multicall_oldcatch = 0; 3947 multicall_cv = NULL; 3948 cx = NULL; 3949 PERL_UNUSED_VAR(multicall_cop); 3950 PERL_UNUSED_VAR(newsp); 3951 3952 3953 PERL_ARGS_ASSERT_REGMATCH; 3954 3955 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({ 3956 PerlIO_printf(Perl_debug_log,"regmatch start\n"); 3957 })); 3958 3959 st = PL_regmatch_state; 3960 3961 /* Note that nextchr is a byte even in UTF */ 3962 SET_nextchr; 3963 scan = prog; 3964 while (scan != NULL) { 3965 3966 DEBUG_EXECUTE_r( { 3967 SV * const prop = sv_newmortal(); 3968 regnode *rnext=regnext(scan); 3969 DUMP_EXEC_POS( locinput, scan, utf8_target ); 3970 regprop(rex, prop, scan, reginfo); 3971 3972 PerlIO_printf(Perl_debug_log, 3973 "%3"IVdf":%*s%s(%"IVdf")\n", 3974 (IV)(scan - rexi->program), depth*2, "", 3975 SvPVX_const(prop), 3976 (PL_regkind[OP(scan)] == END || !rnext) ? 3977 0 : (IV)(rnext - rexi->program)); 3978 }); 3979 3980 next = scan + NEXT_OFF(scan); 3981 if (next == scan) 3982 next = NULL; 3983 state_num = OP(scan); 3984 3985 reenter_switch: 3986 to_complement = 0; 3987 3988 SET_nextchr; 3989 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS)); 3990 3991 switch (state_num) { 3992 case BOL: /* /^../ */ 3993 case SBOL: /* /^../s */ 3994 if (locinput == reginfo->strbeg) 3995 break; 3996 sayNO; 3997 3998 case MBOL: /* /^../m */ 3999 if (locinput == reginfo->strbeg || 4000 (!NEXTCHR_IS_EOS && locinput[-1] == '\n')) 4001 { 4002 break; 4003 } 4004 sayNO; 4005 4006 case GPOS: /* \G */ 4007 if (locinput == reginfo->ganch) 4008 break; 4009 sayNO; 4010 4011 case KEEPS: /* \K */ 4012 /* update the startpoint */ 4013 st->u.keeper.val = rex->offs[0].start; 4014 rex->offs[0].start = locinput - reginfo->strbeg; 4015 PUSH_STATE_GOTO(KEEPS_next, next, locinput); 4016 assert(0); /*NOTREACHED*/ 4017 case KEEPS_next_fail: 4018 /* rollback the start point change */ 4019 rex->offs[0].start = st->u.keeper.val; 4020 sayNO_SILENT; 4021 assert(0); /*NOTREACHED*/ 4022 4023 case MEOL: /* /..$/m */ 4024 if (!NEXTCHR_IS_EOS && nextchr != '\n') 4025 sayNO; 4026 break; 4027 4028 case EOL: /* /..$/ */ 4029 /* FALL THROUGH */ 4030 case SEOL: /* /..$/s */ 4031 if (!NEXTCHR_IS_EOS && nextchr != '\n') 4032 sayNO; 4033 if (reginfo->strend - locinput > 1) 4034 sayNO; 4035 break; 4036 4037 case EOS: /* \z */ 4038 if (!NEXTCHR_IS_EOS) 4039 sayNO; 4040 break; 4041 4042 case SANY: /* /./s */ 4043 if (NEXTCHR_IS_EOS) 4044 sayNO; 4045 goto increment_locinput; 4046 4047 case CANY: /* \C */ 4048 if (NEXTCHR_IS_EOS) 4049 sayNO; 4050 locinput++; 4051 break; 4052 4053 case REG_ANY: /* /./ */ 4054 if ((NEXTCHR_IS_EOS) || nextchr == '\n') 4055 sayNO; 4056 goto increment_locinput; 4057 4058 4059 #undef ST 4060 #define ST st->u.trie 4061 case TRIEC: /* (ab|cd) with known charclass */ 4062 /* In this case the charclass data is available inline so 4063 we can fail fast without a lot of extra overhead. 4064 */ 4065 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) { 4066 DEBUG_EXECUTE_r( 4067 PerlIO_printf(Perl_debug_log, 4068 "%*s %sfailed to match trie start class...%s\n", 4069 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) 4070 ); 4071 sayNO_SILENT; 4072 assert(0); /* NOTREACHED */ 4073 } 4074 /* FALL THROUGH */ 4075 case TRIE: /* (ab|cd) */ 4076 /* the basic plan of execution of the trie is: 4077 * At the beginning, run though all the states, and 4078 * find the longest-matching word. Also remember the position 4079 * of the shortest matching word. For example, this pattern: 4080 * 1 2 3 4 5 4081 * ab|a|x|abcd|abc 4082 * when matched against the string "abcde", will generate 4083 * accept states for all words except 3, with the longest 4084 * matching word being 4, and the shortest being 2 (with 4085 * the position being after char 1 of the string). 4086 * 4087 * Then for each matching word, in word order (i.e. 1,2,4,5), 4088 * we run the remainder of the pattern; on each try setting 4089 * the current position to the character following the word, 4090 * returning to try the next word on failure. 4091 * 4092 * We avoid having to build a list of words at runtime by 4093 * using a compile-time structure, wordinfo[].prev, which 4094 * gives, for each word, the previous accepting word (if any). 4095 * In the case above it would contain the mappings 1->2, 2->0, 4096 * 3->0, 4->5, 5->1. We can use this table to generate, from 4097 * the longest word (4 above), a list of all words, by 4098 * following the list of prev pointers; this gives us the 4099 * unordered list 4,5,1,2. Then given the current word we have 4100 * just tried, we can go through the list and find the 4101 * next-biggest word to try (so if we just failed on word 2, 4102 * the next in the list is 4). 4103 * 4104 * Since at runtime we don't record the matching position in 4105 * the string for each word, we have to work that out for 4106 * each word we're about to process. The wordinfo table holds 4107 * the character length of each word; given that we recorded 4108 * at the start: the position of the shortest word and its 4109 * length in chars, we just need to move the pointer the 4110 * difference between the two char lengths. Depending on 4111 * Unicode status and folding, that's cheap or expensive. 4112 * 4113 * This algorithm is optimised for the case where are only a 4114 * small number of accept states, i.e. 0,1, or maybe 2. 4115 * With lots of accepts states, and having to try all of them, 4116 * it becomes quadratic on number of accept states to find all 4117 * the next words. 4118 */ 4119 4120 { 4121 /* what type of TRIE am I? (utf8 makes this contextual) */ 4122 DECL_TRIE_TYPE(scan); 4123 4124 /* what trie are we using right now */ 4125 reg_trie_data * const trie 4126 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ]; 4127 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]); 4128 U32 state = trie->startstate; 4129 4130 if ( trie->bitmap 4131 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr))) 4132 { 4133 if (trie->states[ state ].wordnum) { 4134 DEBUG_EXECUTE_r( 4135 PerlIO_printf(Perl_debug_log, 4136 "%*s %smatched empty string...%s\n", 4137 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) 4138 ); 4139 if (!trie->jump) 4140 break; 4141 } else { 4142 DEBUG_EXECUTE_r( 4143 PerlIO_printf(Perl_debug_log, 4144 "%*s %sfailed to match trie start class...%s\n", 4145 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) 4146 ); 4147 sayNO_SILENT; 4148 } 4149 } 4150 4151 { 4152 U8 *uc = ( U8* )locinput; 4153 4154 STRLEN len = 0; 4155 STRLEN foldlen = 0; 4156 U8 *uscan = (U8*)NULL; 4157 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; 4158 U32 charcount = 0; /* how many input chars we have matched */ 4159 U32 accepted = 0; /* have we seen any accepting states? */ 4160 4161 ST.jump = trie->jump; 4162 ST.me = scan; 4163 ST.firstpos = NULL; 4164 ST.longfold = FALSE; /* char longer if folded => it's harder */ 4165 ST.nextword = 0; 4166 4167 /* fully traverse the TRIE; note the position of the 4168 shortest accept state and the wordnum of the longest 4169 accept state */ 4170 4171 while ( state && uc <= (U8*)(reginfo->strend) ) { 4172 U32 base = trie->states[ state ].trans.base; 4173 UV uvc = 0; 4174 U16 charid = 0; 4175 U16 wordnum; 4176 wordnum = trie->states[ state ].wordnum; 4177 4178 if (wordnum) { /* it's an accept state */ 4179 if (!accepted) { 4180 accepted = 1; 4181 /* record first match position */ 4182 if (ST.longfold) { 4183 ST.firstpos = (U8*)locinput; 4184 ST.firstchars = 0; 4185 } 4186 else { 4187 ST.firstpos = uc; 4188 ST.firstchars = charcount; 4189 } 4190 } 4191 if (!ST.nextword || wordnum < ST.nextword) 4192 ST.nextword = wordnum; 4193 ST.topword = wordnum; 4194 } 4195 4196 DEBUG_TRIE_EXECUTE_r({ 4197 DUMP_EXEC_POS( (char *)uc, scan, utf8_target ); 4198 PerlIO_printf( Perl_debug_log, 4199 "%*s %sState: %4"UVxf" Accepted: %c ", 4200 2+depth * 2, "", PL_colors[4], 4201 (UV)state, (accepted ? 'Y' : 'N')); 4202 }); 4203 4204 /* read a char and goto next state */ 4205 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) { 4206 I32 offset; 4207 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, 4208 uscan, len, uvc, charid, foldlen, 4209 foldbuf, uniflags); 4210 charcount++; 4211 if (foldlen>0) 4212 ST.longfold = TRUE; 4213 if (charid && 4214 ( ((offset = 4215 base + charid - 1 - trie->uniquecharcount)) >= 0) 4216 4217 && ((U32)offset < trie->lasttrans) 4218 && trie->trans[offset].check == state) 4219 { 4220 state = trie->trans[offset].next; 4221 } 4222 else { 4223 state = 0; 4224 } 4225 uc += len; 4226 4227 } 4228 else { 4229 state = 0; 4230 } 4231 DEBUG_TRIE_EXECUTE_r( 4232 PerlIO_printf( Perl_debug_log, 4233 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", 4234 charid, uvc, (UV)state, PL_colors[5] ); 4235 ); 4236 } 4237 if (!accepted) 4238 sayNO; 4239 4240 /* calculate total number of accept states */ 4241 { 4242 U16 w = ST.topword; 4243 accepted = 0; 4244 while (w) { 4245 w = trie->wordinfo[w].prev; 4246 accepted++; 4247 } 4248 ST.accepted = accepted; 4249 } 4250 4251 DEBUG_EXECUTE_r( 4252 PerlIO_printf( Perl_debug_log, 4253 "%*s %sgot %"IVdf" possible matches%s\n", 4254 REPORT_CODE_OFF + depth * 2, "", 4255 PL_colors[4], (IV)ST.accepted, PL_colors[5] ); 4256 ); 4257 goto trie_first_try; /* jump into the fail handler */ 4258 }} 4259 assert(0); /* NOTREACHED */ 4260 4261 case TRIE_next_fail: /* we failed - try next alternative */ 4262 { 4263 U8 *uc; 4264 if ( ST.jump) { 4265 REGCP_UNWIND(ST.cp); 4266 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 4267 } 4268 if (!--ST.accepted) { 4269 DEBUG_EXECUTE_r({ 4270 PerlIO_printf( Perl_debug_log, 4271 "%*s %sTRIE failed...%s\n", 4272 REPORT_CODE_OFF+depth*2, "", 4273 PL_colors[4], 4274 PL_colors[5] ); 4275 }); 4276 sayNO_SILENT; 4277 } 4278 { 4279 /* Find next-highest word to process. Note that this code 4280 * is O(N^2) per trie run (O(N) per branch), so keep tight */ 4281 U16 min = 0; 4282 U16 word; 4283 U16 const nextword = ST.nextword; 4284 reg_trie_wordinfo * const wordinfo 4285 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo; 4286 for (word=ST.topword; word; word=wordinfo[word].prev) { 4287 if (word > nextword && (!min || word < min)) 4288 min = word; 4289 } 4290 ST.nextword = min; 4291 } 4292 4293 trie_first_try: 4294 if (do_cutgroup) { 4295 do_cutgroup = 0; 4296 no_final = 0; 4297 } 4298 4299 if ( ST.jump) { 4300 ST.lastparen = rex->lastparen; 4301 ST.lastcloseparen = rex->lastcloseparen; 4302 REGCP_SET(ST.cp); 4303 } 4304 4305 /* find start char of end of current word */ 4306 { 4307 U32 chars; /* how many chars to skip */ 4308 reg_trie_data * const trie 4309 = (reg_trie_data*)rexi->data->data[ARG(ST.me)]; 4310 4311 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen) 4312 >= ST.firstchars); 4313 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen) 4314 - ST.firstchars; 4315 uc = ST.firstpos; 4316 4317 if (ST.longfold) { 4318 /* the hard option - fold each char in turn and find 4319 * its folded length (which may be different */ 4320 U8 foldbuf[UTF8_MAXBYTES_CASE + 1]; 4321 STRLEN foldlen; 4322 STRLEN len; 4323 UV uvc; 4324 U8 *uscan; 4325 4326 while (chars) { 4327 if (utf8_target) { 4328 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len, 4329 uniflags); 4330 uc += len; 4331 } 4332 else { 4333 uvc = *uc; 4334 uc++; 4335 } 4336 uvc = to_uni_fold(uvc, foldbuf, &foldlen); 4337 uscan = foldbuf; 4338 while (foldlen) { 4339 if (!--chars) 4340 break; 4341 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len, 4342 uniflags); 4343 uscan += len; 4344 foldlen -= len; 4345 } 4346 } 4347 } 4348 else { 4349 if (utf8_target) 4350 while (chars--) 4351 uc += UTF8SKIP(uc); 4352 else 4353 uc += chars; 4354 } 4355 } 4356 4357 scan = ST.me + ((ST.jump && ST.jump[ST.nextword]) 4358 ? ST.jump[ST.nextword] 4359 : NEXT_OFF(ST.me)); 4360 4361 DEBUG_EXECUTE_r({ 4362 PerlIO_printf( Perl_debug_log, 4363 "%*s %sTRIE matched word #%d, continuing%s\n", 4364 REPORT_CODE_OFF+depth*2, "", 4365 PL_colors[4], 4366 ST.nextword, 4367 PL_colors[5] 4368 ); 4369 }); 4370 4371 if (ST.accepted > 1 || has_cutgroup) { 4372 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc); 4373 assert(0); /* NOTREACHED */ 4374 } 4375 /* only one choice left - just continue */ 4376 DEBUG_EXECUTE_r({ 4377 AV *const trie_words 4378 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]); 4379 SV ** const tmp = av_fetch( trie_words, 4380 ST.nextword-1, 0 ); 4381 SV *sv= tmp ? sv_newmortal() : NULL; 4382 4383 PerlIO_printf( Perl_debug_log, 4384 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n", 4385 REPORT_CODE_OFF+depth*2, "", PL_colors[4], 4386 ST.nextword, 4387 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0, 4388 PL_colors[0], PL_colors[1], 4389 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII 4390 ) 4391 : "not compiled under -Dr", 4392 PL_colors[5] ); 4393 }); 4394 4395 locinput = (char*)uc; 4396 continue; /* execute rest of RE */ 4397 assert(0); /* NOTREACHED */ 4398 } 4399 #undef ST 4400 4401 case EXACT: { /* /abc/ */ 4402 char *s = STRING(scan); 4403 ln = STR_LEN(scan); 4404 if (utf8_target != is_utf8_pat) { 4405 /* The target and the pattern have differing utf8ness. */ 4406 char *l = locinput; 4407 const char * const e = s + ln; 4408 4409 if (utf8_target) { 4410 /* The target is utf8, the pattern is not utf8. 4411 * Above-Latin1 code points can't match the pattern; 4412 * invariants match exactly, and the other Latin1 ones need 4413 * to be downgraded to a single byte in order to do the 4414 * comparison. (If we could be confident that the target 4415 * is not malformed, this could be refactored to have fewer 4416 * tests by just assuming that if the first bytes match, it 4417 * is an invariant, but there are tests in the test suite 4418 * dealing with (??{...}) which violate this) */ 4419 while (s < e) { 4420 if (l >= reginfo->strend 4421 || UTF8_IS_ABOVE_LATIN1(* (U8*) l)) 4422 { 4423 sayNO; 4424 } 4425 if (UTF8_IS_INVARIANT(*(U8*)l)) { 4426 if (*l != *s) { 4427 sayNO; 4428 } 4429 l++; 4430 } 4431 else { 4432 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s) 4433 { 4434 sayNO; 4435 } 4436 l += 2; 4437 } 4438 s++; 4439 } 4440 } 4441 else { 4442 /* The target is not utf8, the pattern is utf8. */ 4443 while (s < e) { 4444 if (l >= reginfo->strend 4445 || UTF8_IS_ABOVE_LATIN1(* (U8*) s)) 4446 { 4447 sayNO; 4448 } 4449 if (UTF8_IS_INVARIANT(*(U8*)s)) { 4450 if (*s != *l) { 4451 sayNO; 4452 } 4453 s++; 4454 } 4455 else { 4456 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l) 4457 { 4458 sayNO; 4459 } 4460 s += 2; 4461 } 4462 l++; 4463 } 4464 } 4465 locinput = l; 4466 } 4467 else { 4468 /* The target and the pattern have the same utf8ness. */ 4469 /* Inline the first character, for speed. */ 4470 if (reginfo->strend - locinput < ln 4471 || UCHARAT(s) != nextchr 4472 || (ln > 1 && memNE(s, locinput, ln))) 4473 { 4474 sayNO; 4475 } 4476 locinput += ln; 4477 } 4478 break; 4479 } 4480 4481 case EXACTFL: { /* /abc/il */ 4482 re_fold_t folder; 4483 const U8 * fold_array; 4484 const char * s; 4485 U32 fold_utf8_flags; 4486 4487 folder = foldEQ_locale; 4488 fold_array = PL_fold_locale; 4489 fold_utf8_flags = FOLDEQ_LOCALE; 4490 goto do_exactf; 4491 4492 case EXACTFU_SS: /* /\x{df}/iu */ 4493 case EXACTFU: /* /abc/iu */ 4494 folder = foldEQ_latin1; 4495 fold_array = PL_fold_latin1; 4496 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0; 4497 goto do_exactf; 4498 4499 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 4500 patterns */ 4501 assert(! is_utf8_pat); 4502 /* FALL THROUGH */ 4503 case EXACTFA: /* /abc/iaa */ 4504 folder = foldEQ_latin1; 4505 fold_array = PL_fold_latin1; 4506 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; 4507 goto do_exactf; 4508 4509 case EXACTF: /* /abc/i This node only generated for 4510 non-utf8 patterns */ 4511 assert(! is_utf8_pat); 4512 folder = foldEQ; 4513 fold_array = PL_fold; 4514 fold_utf8_flags = 0; 4515 4516 do_exactf: 4517 s = STRING(scan); 4518 ln = STR_LEN(scan); 4519 4520 if (utf8_target 4521 || is_utf8_pat 4522 || state_num == EXACTFU_SS 4523 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE)) 4524 { 4525 /* Either target or the pattern are utf8, or has the issue where 4526 * the fold lengths may differ. */ 4527 const char * const l = locinput; 4528 char *e = reginfo->strend; 4529 4530 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat, 4531 l, &e, 0, utf8_target, fold_utf8_flags)) 4532 { 4533 sayNO; 4534 } 4535 locinput = e; 4536 break; 4537 } 4538 4539 /* Neither the target nor the pattern are utf8 */ 4540 if (UCHARAT(s) != nextchr 4541 && !NEXTCHR_IS_EOS 4542 && UCHARAT(s) != fold_array[nextchr]) 4543 { 4544 sayNO; 4545 } 4546 if (reginfo->strend - locinput < ln) 4547 sayNO; 4548 if (ln > 1 && ! folder(s, locinput, ln)) 4549 sayNO; 4550 locinput += ln; 4551 break; 4552 } 4553 4554 /* XXX Could improve efficiency by separating these all out using a 4555 * macro or in-line function. At that point regcomp.c would no longer 4556 * have to set the FLAGS fields of these */ 4557 case BOUNDL: /* /\b/l */ 4558 case NBOUNDL: /* /\B/l */ 4559 case BOUND: /* /\b/ */ 4560 case BOUNDU: /* /\b/u */ 4561 case BOUNDA: /* /\b/a */ 4562 case NBOUND: /* /\B/ */ 4563 case NBOUNDU: /* /\B/u */ 4564 case NBOUNDA: /* /\B/a */ 4565 /* was last char in word? */ 4566 if (utf8_target 4567 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET 4568 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET) 4569 { 4570 if (locinput == reginfo->strbeg) 4571 ln = '\n'; 4572 else { 4573 const U8 * const r = 4574 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg)); 4575 4576 ln = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, 4577 0, uniflags); 4578 } 4579 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) { 4580 ln = isWORDCHAR_uni(ln); 4581 if (NEXTCHR_IS_EOS) 4582 n = 0; 4583 else { 4584 LOAD_UTF8_CHARCLASS_ALNUM(); 4585 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput, 4586 utf8_target); 4587 } 4588 } 4589 else { 4590 ln = isWORDCHAR_LC_uvchr(ln); 4591 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput); 4592 } 4593 } 4594 else { 4595 4596 /* Here the string isn't utf8, or is utf8 and only ascii 4597 * characters are to match \w. In the latter case looking at 4598 * the byte just prior to the current one may be just the final 4599 * byte of a multi-byte character. This is ok. There are two 4600 * cases: 4601 * 1) it is a single byte character, and then the test is doing 4602 * just what it's supposed to. 4603 * 2) it is a multi-byte character, in which case the final 4604 * byte is never mistakable for ASCII, and so the test 4605 * will say it is not a word character, which is the 4606 * correct answer. */ 4607 ln = (locinput != reginfo->strbeg) ? 4608 UCHARAT(locinput - 1) : '\n'; 4609 switch (FLAGS(scan)) { 4610 case REGEX_UNICODE_CHARSET: 4611 ln = isWORDCHAR_L1(ln); 4612 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr); 4613 break; 4614 case REGEX_LOCALE_CHARSET: 4615 ln = isWORDCHAR_LC(ln); 4616 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr); 4617 break; 4618 case REGEX_DEPENDS_CHARSET: 4619 ln = isWORDCHAR(ln); 4620 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr); 4621 break; 4622 case REGEX_ASCII_RESTRICTED_CHARSET: 4623 case REGEX_ASCII_MORE_RESTRICTED_CHARSET: 4624 ln = isWORDCHAR_A(ln); 4625 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr); 4626 break; 4627 default: 4628 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan)); 4629 break; 4630 } 4631 } 4632 /* Note requires that all BOUNDs be lower than all NBOUNDs in 4633 * regcomp.sym */ 4634 if (((!ln) == (!n)) == (OP(scan) < NBOUND)) 4635 sayNO; 4636 break; 4637 4638 case ANYOF: /* /[abc]/ */ 4639 if (NEXTCHR_IS_EOS) 4640 sayNO; 4641 if (utf8_target) { 4642 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend, 4643 utf8_target)) 4644 sayNO; 4645 locinput += UTF8SKIP(locinput); 4646 } 4647 else { 4648 if (!REGINCLASS(rex, scan, (U8*)locinput)) 4649 sayNO; 4650 locinput++; 4651 } 4652 break; 4653 4654 /* The argument (FLAGS) to all the POSIX node types is the class number 4655 * */ 4656 4657 case NPOSIXL: /* \W or [:^punct:] etc. under /l */ 4658 to_complement = 1; 4659 /* FALLTHROUGH */ 4660 4661 case POSIXL: /* \w or [:punct:] etc. under /l */ 4662 if (NEXTCHR_IS_EOS) 4663 sayNO; 4664 4665 /* Use isFOO_lc() for characters within Latin1. (Note that 4666 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else 4667 * wouldn't be invariant) */ 4668 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { 4669 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) { 4670 sayNO; 4671 } 4672 } 4673 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { 4674 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), 4675 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr, 4676 *(locinput + 1)))))) 4677 { 4678 sayNO; 4679 } 4680 } 4681 else { /* Here, must be an above Latin-1 code point */ 4682 goto utf8_posix_not_eos; 4683 } 4684 4685 /* Here, must be utf8 */ 4686 locinput += UTF8SKIP(locinput); 4687 break; 4688 4689 case NPOSIXD: /* \W or [:^punct:] etc. under /d */ 4690 to_complement = 1; 4691 /* FALLTHROUGH */ 4692 4693 case POSIXD: /* \w or [:punct:] etc. under /d */ 4694 if (utf8_target) { 4695 goto utf8_posix; 4696 } 4697 goto posixa; 4698 4699 case NPOSIXA: /* \W or [:^punct:] etc. under /a */ 4700 4701 if (NEXTCHR_IS_EOS) { 4702 sayNO; 4703 } 4704 4705 /* All UTF-8 variants match */ 4706 if (! UTF8_IS_INVARIANT(nextchr)) { 4707 goto increment_locinput; 4708 } 4709 4710 to_complement = 1; 4711 /* FALLTHROUGH */ 4712 4713 case POSIXA: /* \w or [:punct:] etc. under /a */ 4714 4715 posixa: 4716 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in 4717 * UTF-8, and also from NPOSIXA even in UTF-8 when the current 4718 * character is a single byte */ 4719 4720 if (NEXTCHR_IS_EOS 4721 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr, 4722 FLAGS(scan))))) 4723 { 4724 sayNO; 4725 } 4726 4727 /* Here we are either not in utf8, or we matched a utf8-invariant, 4728 * so the next char is the next byte */ 4729 locinput++; 4730 break; 4731 4732 case NPOSIXU: /* \W or [:^punct:] etc. under /u */ 4733 to_complement = 1; 4734 /* FALLTHROUGH */ 4735 4736 case POSIXU: /* \w or [:punct:] etc. under /u */ 4737 utf8_posix: 4738 if (NEXTCHR_IS_EOS) { 4739 sayNO; 4740 } 4741 utf8_posix_not_eos: 4742 4743 /* Use _generic_isCC() for characters within Latin1. (Note that 4744 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else 4745 * wouldn't be invariant) */ 4746 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { 4747 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr, 4748 FLAGS(scan))))) 4749 { 4750 sayNO; 4751 } 4752 locinput++; 4753 } 4754 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { 4755 if (! (to_complement 4756 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr, 4757 *(locinput + 1)), 4758 FLAGS(scan))))) 4759 { 4760 sayNO; 4761 } 4762 locinput += 2; 4763 } 4764 else { /* Handle above Latin-1 code points */ 4765 classnum = (_char_class_number) FLAGS(scan); 4766 if (classnum < _FIRST_NON_SWASH_CC) { 4767 4768 /* Here, uses a swash to find such code points. Load if if 4769 * not done already */ 4770 if (! PL_utf8_swash_ptrs[classnum]) { 4771 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 4772 PL_utf8_swash_ptrs[classnum] 4773 = _core_swash_init("utf8", 4774 "", 4775 &PL_sv_undef, 1, 0, 4776 PL_XPosix_ptrs[classnum], &flags); 4777 } 4778 if (! (to_complement 4779 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], 4780 (U8 *) locinput, TRUE)))) 4781 { 4782 sayNO; 4783 } 4784 } 4785 else { /* Here, uses macros to find above Latin-1 code points */ 4786 switch (classnum) { 4787 case _CC_ENUM_SPACE: /* XXX would require separate 4788 code if we revert the change 4789 of \v matching this */ 4790 case _CC_ENUM_PSXSPC: 4791 if (! (to_complement 4792 ^ cBOOL(is_XPERLSPACE_high(locinput)))) 4793 { 4794 sayNO; 4795 } 4796 break; 4797 case _CC_ENUM_BLANK: 4798 if (! (to_complement 4799 ^ cBOOL(is_HORIZWS_high(locinput)))) 4800 { 4801 sayNO; 4802 } 4803 break; 4804 case _CC_ENUM_XDIGIT: 4805 if (! (to_complement 4806 ^ cBOOL(is_XDIGIT_high(locinput)))) 4807 { 4808 sayNO; 4809 } 4810 break; 4811 case _CC_ENUM_VERTSPACE: 4812 if (! (to_complement 4813 ^ cBOOL(is_VERTWS_high(locinput)))) 4814 { 4815 sayNO; 4816 } 4817 break; 4818 default: /* The rest, e.g. [:cntrl:], can't match 4819 above Latin1 */ 4820 if (! to_complement) { 4821 sayNO; 4822 } 4823 break; 4824 } 4825 } 4826 locinput += UTF8SKIP(locinput); 4827 } 4828 break; 4829 4830 case CLUMP: /* Match \X: logical Unicode character. This is defined as 4831 a Unicode extended Grapheme Cluster */ 4832 /* From http://www.unicode.org/reports/tr29 (5.2 version). An 4833 extended Grapheme Cluster is: 4834 4835 CR LF 4836 | Prepend* Begin Extend* 4837 | . 4838 4839 Begin is: ( Special_Begin | ! Control ) 4840 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable ) 4841 Extend is: ( Grapheme_Extend | Spacing_Mark ) 4842 Control is: [ GCB_Control | CR | LF ] 4843 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) )) 4844 4845 If we create a 'Regular_Begin' = Begin - Special_Begin, then 4846 we can rewrite 4847 4848 Begin is ( Regular_Begin + Special Begin ) 4849 4850 It turns out that 98.4% of all Unicode code points match 4851 Regular_Begin. Doing it this way eliminates a table match in 4852 the previous implementation for almost all Unicode code points. 4853 4854 There is a subtlety with Prepend* which showed up in testing. 4855 Note that the Begin, and only the Begin is required in: 4856 | Prepend* Begin Extend* 4857 Also, Begin contains '! Control'. A Prepend must be a 4858 '! Control', which means it must also be a Begin. What it 4859 comes down to is that if we match Prepend* and then find no 4860 suitable Begin afterwards, that if we backtrack the last 4861 Prepend, that one will be a suitable Begin. 4862 */ 4863 4864 if (NEXTCHR_IS_EOS) 4865 sayNO; 4866 if (! utf8_target) { 4867 4868 /* Match either CR LF or '.', as all the other possibilities 4869 * require utf8 */ 4870 locinput++; /* Match the . or CR */ 4871 if (nextchr == '\r' /* And if it was CR, and the next is LF, 4872 match the LF */ 4873 && locinput < reginfo->strend 4874 && UCHARAT(locinput) == '\n') 4875 { 4876 locinput++; 4877 } 4878 } 4879 else { 4880 4881 /* Utf8: See if is ( CR LF ); already know that locinput < 4882 * reginfo->strend, so locinput+1 is in bounds */ 4883 if ( nextchr == '\r' && locinput+1 < reginfo->strend 4884 && UCHARAT(locinput + 1) == '\n') 4885 { 4886 locinput += 2; 4887 } 4888 else { 4889 STRLEN len; 4890 4891 /* In case have to backtrack to beginning, then match '.' */ 4892 char *starting = locinput; 4893 4894 /* In case have to backtrack the last prepend */ 4895 char *previous_prepend = NULL; 4896 4897 LOAD_UTF8_CHARCLASS_GCB(); 4898 4899 /* Match (prepend)* */ 4900 while (locinput < reginfo->strend 4901 && (len = is_GCB_Prepend_utf8(locinput))) 4902 { 4903 previous_prepend = locinput; 4904 locinput += len; 4905 } 4906 4907 /* As noted above, if we matched a prepend character, but 4908 * the next thing won't match, back off the last prepend we 4909 * matched, as it is guaranteed to match the begin */ 4910 if (previous_prepend 4911 && (locinput >= reginfo->strend 4912 || (! swash_fetch(PL_utf8_X_regular_begin, 4913 (U8*)locinput, utf8_target) 4914 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput))) 4915 ) 4916 { 4917 locinput = previous_prepend; 4918 } 4919 4920 /* Note that here we know reginfo->strend > locinput, as we 4921 * tested that upon input to this switch case, and if we 4922 * moved locinput forward, we tested the result just above 4923 * and it either passed, or we backed off so that it will 4924 * now pass */ 4925 if (swash_fetch(PL_utf8_X_regular_begin, 4926 (U8*)locinput, utf8_target)) { 4927 locinput += UTF8SKIP(locinput); 4928 } 4929 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) { 4930 4931 /* Here did not match the required 'Begin' in the 4932 * second term. So just match the very first 4933 * character, the '.' of the final term of the regex */ 4934 locinput = starting + UTF8SKIP(starting); 4935 goto exit_utf8; 4936 } else { 4937 4938 /* Here is a special begin. It can be composed of 4939 * several individual characters. One possibility is 4940 * RI+ */ 4941 if ((len = is_GCB_RI_utf8(locinput))) { 4942 locinput += len; 4943 while (locinput < reginfo->strend 4944 && (len = is_GCB_RI_utf8(locinput))) 4945 { 4946 locinput += len; 4947 } 4948 } else if ((len = is_GCB_T_utf8(locinput))) { 4949 /* Another possibility is T+ */ 4950 locinput += len; 4951 while (locinput < reginfo->strend 4952 && (len = is_GCB_T_utf8(locinput))) 4953 { 4954 locinput += len; 4955 } 4956 } else { 4957 4958 /* Here, neither RI+ nor T+; must be some other 4959 * Hangul. That means it is one of the others: L, 4960 * LV, LVT or V, and matches: 4961 * L* (L | LVT T* | V * V* T* | LV V* T*) */ 4962 4963 /* Match L* */ 4964 while (locinput < reginfo->strend 4965 && (len = is_GCB_L_utf8(locinput))) 4966 { 4967 locinput += len; 4968 } 4969 4970 /* Here, have exhausted L*. If the next character 4971 * is not an LV, LVT nor V, it means we had to have 4972 * at least one L, so matches L+ in the original 4973 * equation, we have a complete hangul syllable. 4974 * Are done. */ 4975 4976 if (locinput < reginfo->strend 4977 && is_GCB_LV_LVT_V_utf8(locinput)) 4978 { 4979 /* Otherwise keep going. Must be LV, LVT or V. 4980 * See if LVT, by first ruling out V, then LV */ 4981 if (! is_GCB_V_utf8(locinput) 4982 /* All but every TCount one is LV */ 4983 && (valid_utf8_to_uvchr((U8 *) locinput, 4984 NULL) 4985 - SBASE) 4986 % TCount != 0) 4987 { 4988 locinput += UTF8SKIP(locinput); 4989 } else { 4990 4991 /* Must be V or LV. Take it, then match 4992 * V* */ 4993 locinput += UTF8SKIP(locinput); 4994 while (locinput < reginfo->strend 4995 && (len = is_GCB_V_utf8(locinput))) 4996 { 4997 locinput += len; 4998 } 4999 } 5000 5001 /* And any of LV, LVT, or V can be followed 5002 * by T* */ 5003 while (locinput < reginfo->strend 5004 && (len = is_GCB_T_utf8(locinput))) 5005 { 5006 locinput += len; 5007 } 5008 } 5009 } 5010 } 5011 5012 /* Match any extender */ 5013 while (locinput < reginfo->strend 5014 && swash_fetch(PL_utf8_X_extend, 5015 (U8*)locinput, utf8_target)) 5016 { 5017 locinput += UTF8SKIP(locinput); 5018 } 5019 } 5020 exit_utf8: 5021 if (locinput > reginfo->strend) sayNO; 5022 } 5023 break; 5024 5025 case NREFFL: /* /\g{name}/il */ 5026 { /* The capture buffer cases. The ones beginning with N for the 5027 named buffers just convert to the equivalent numbered and 5028 pretend they were called as the corresponding numbered buffer 5029 op. */ 5030 /* don't initialize these in the declaration, it makes C++ 5031 unhappy */ 5032 const char *s; 5033 char type; 5034 re_fold_t folder; 5035 const U8 *fold_array; 5036 UV utf8_fold_flags; 5037 5038 folder = foldEQ_locale; 5039 fold_array = PL_fold_locale; 5040 type = REFFL; 5041 utf8_fold_flags = FOLDEQ_LOCALE; 5042 goto do_nref; 5043 5044 case NREFFA: /* /\g{name}/iaa */ 5045 folder = foldEQ_latin1; 5046 fold_array = PL_fold_latin1; 5047 type = REFFA; 5048 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; 5049 goto do_nref; 5050 5051 case NREFFU: /* /\g{name}/iu */ 5052 folder = foldEQ_latin1; 5053 fold_array = PL_fold_latin1; 5054 type = REFFU; 5055 utf8_fold_flags = 0; 5056 goto do_nref; 5057 5058 case NREFF: /* /\g{name}/i */ 5059 folder = foldEQ; 5060 fold_array = PL_fold; 5061 type = REFF; 5062 utf8_fold_flags = 0; 5063 goto do_nref; 5064 5065 case NREF: /* /\g{name}/ */ 5066 type = REF; 5067 folder = NULL; 5068 fold_array = NULL; 5069 utf8_fold_flags = 0; 5070 do_nref: 5071 5072 /* For the named back references, find the corresponding buffer 5073 * number */ 5074 n = reg_check_named_buff_matched(rex,scan); 5075 5076 if ( ! n ) { 5077 sayNO; 5078 } 5079 goto do_nref_ref_common; 5080 5081 case REFFL: /* /\1/il */ 5082 folder = foldEQ_locale; 5083 fold_array = PL_fold_locale; 5084 utf8_fold_flags = FOLDEQ_LOCALE; 5085 goto do_ref; 5086 5087 case REFFA: /* /\1/iaa */ 5088 folder = foldEQ_latin1; 5089 fold_array = PL_fold_latin1; 5090 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; 5091 goto do_ref; 5092 5093 case REFFU: /* /\1/iu */ 5094 folder = foldEQ_latin1; 5095 fold_array = PL_fold_latin1; 5096 utf8_fold_flags = 0; 5097 goto do_ref; 5098 5099 case REFF: /* /\1/i */ 5100 folder = foldEQ; 5101 fold_array = PL_fold; 5102 utf8_fold_flags = 0; 5103 goto do_ref; 5104 5105 case REF: /* /\1/ */ 5106 folder = NULL; 5107 fold_array = NULL; 5108 utf8_fold_flags = 0; 5109 5110 do_ref: 5111 type = OP(scan); 5112 n = ARG(scan); /* which paren pair */ 5113 5114 do_nref_ref_common: 5115 ln = rex->offs[n].start; 5116 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ 5117 if (rex->lastparen < n || ln == -1) 5118 sayNO; /* Do not match unless seen CLOSEn. */ 5119 if (ln == rex->offs[n].end) 5120 break; 5121 5122 s = reginfo->strbeg + ln; 5123 if (type != REF /* REF can do byte comparison */ 5124 && (utf8_target || type == REFFU || type == REFFL)) 5125 { 5126 char * limit = reginfo->strend; 5127 5128 /* This call case insensitively compares the entire buffer 5129 * at s, with the current input starting at locinput, but 5130 * not going off the end given by reginfo->strend, and 5131 * returns in <limit> upon success, how much of the 5132 * current input was matched */ 5133 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target, 5134 locinput, &limit, 0, utf8_target, utf8_fold_flags)) 5135 { 5136 sayNO; 5137 } 5138 locinput = limit; 5139 break; 5140 } 5141 5142 /* Not utf8: Inline the first character, for speed. */ 5143 if (!NEXTCHR_IS_EOS && 5144 UCHARAT(s) != nextchr && 5145 (type == REF || 5146 UCHARAT(s) != fold_array[nextchr])) 5147 sayNO; 5148 ln = rex->offs[n].end - ln; 5149 if (locinput + ln > reginfo->strend) 5150 sayNO; 5151 if (ln > 1 && (type == REF 5152 ? memNE(s, locinput, ln) 5153 : ! folder(s, locinput, ln))) 5154 sayNO; 5155 locinput += ln; 5156 break; 5157 } 5158 5159 case NOTHING: /* null op; e.g. the 'nothing' following 5160 * the '*' in m{(a+|b)*}' */ 5161 break; 5162 case TAIL: /* placeholder while compiling (A|B|C) */ 5163 break; 5164 5165 case BACK: /* ??? doesn't appear to be used ??? */ 5166 break; 5167 5168 #undef ST 5169 #define ST st->u.eval 5170 { 5171 SV *ret; 5172 REGEXP *re_sv; 5173 regexp *re; 5174 regexp_internal *rei; 5175 regnode *startpoint; 5176 5177 case GOSTART: /* (?R) */ 5178 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */ 5179 if (cur_eval && cur_eval->locinput==locinput) { 5180 if (cur_eval->u.eval.close_paren == (U32)ARG(scan)) 5181 Perl_croak(aTHX_ "Infinite recursion in regex"); 5182 if ( ++nochange_depth > max_nochange_depth ) 5183 Perl_croak(aTHX_ 5184 "Pattern subroutine nesting without pos change" 5185 " exceeded limit in regex"); 5186 } else { 5187 nochange_depth = 0; 5188 } 5189 re_sv = rex_sv; 5190 re = rex; 5191 rei = rexi; 5192 if (OP(scan)==GOSUB) { 5193 startpoint = scan + ARG2L(scan); 5194 ST.close_paren = ARG(scan); 5195 } else { 5196 startpoint = rei->program+1; 5197 ST.close_paren = 0; 5198 } 5199 5200 /* Save all the positions seen so far. */ 5201 ST.cp = regcppush(rex, 0, maxopenparen); 5202 REGCP_SET(ST.lastcp); 5203 5204 /* and then jump to the code we share with EVAL */ 5205 goto eval_recurse_doit; 5206 5207 assert(0); /* NOTREACHED */ 5208 5209 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ 5210 if (cur_eval && cur_eval->locinput==locinput) { 5211 if ( ++nochange_depth > max_nochange_depth ) 5212 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex"); 5213 } else { 5214 nochange_depth = 0; 5215 } 5216 { 5217 /* execute the code in the {...} */ 5218 5219 dSP; 5220 IV before; 5221 OP * const oop = PL_op; 5222 COP * const ocurcop = PL_curcop; 5223 OP *nop; 5224 CV *newcv; 5225 5226 /* save *all* paren positions */ 5227 regcppush(rex, 0, maxopenparen); 5228 REGCP_SET(runops_cp); 5229 5230 if (!caller_cv) 5231 caller_cv = find_runcv(NULL); 5232 5233 n = ARG(scan); 5234 5235 if (rexi->data->what[n] == 'r') { /* code from an external qr */ 5236 newcv = (ReANY( 5237 (REGEXP*)(rexi->data->data[n]) 5238 ))->qr_anoncv 5239 ; 5240 nop = (OP*)rexi->data->data[n+1]; 5241 } 5242 else if (rexi->data->what[n] == 'l') { /* literal code */ 5243 newcv = caller_cv; 5244 nop = (OP*)rexi->data->data[n]; 5245 assert(CvDEPTH(newcv)); 5246 } 5247 else { 5248 /* literal with own CV */ 5249 assert(rexi->data->what[n] == 'L'); 5250 newcv = rex->qr_anoncv; 5251 nop = (OP*)rexi->data->data[n]; 5252 } 5253 5254 /* normally if we're about to execute code from the same 5255 * CV that we used previously, we just use the existing 5256 * CX stack entry. However, its possible that in the 5257 * meantime we may have backtracked, popped from the save 5258 * stack, and undone the SAVECOMPPAD(s) associated with 5259 * PUSH_MULTICALL; in which case PL_comppad no longer 5260 * points to newcv's pad. */ 5261 if (newcv != last_pushed_cv || PL_comppad != last_pad) 5262 { 5263 U8 flags = (CXp_SUB_RE | 5264 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0)); 5265 if (last_pushed_cv) { 5266 CHANGE_MULTICALL_FLAGS(newcv, flags); 5267 } 5268 else { 5269 PUSH_MULTICALL_FLAGS(newcv, flags); 5270 } 5271 last_pushed_cv = newcv; 5272 } 5273 else { 5274 /* these assignments are just to silence compiler 5275 * warnings */ 5276 multicall_cop = NULL; 5277 newsp = NULL; 5278 } 5279 last_pad = PL_comppad; 5280 5281 /* the initial nextstate you would normally execute 5282 * at the start of an eval (which would cause error 5283 * messages to come from the eval), may be optimised 5284 * away from the execution path in the regex code blocks; 5285 * so manually set PL_curcop to it initially */ 5286 { 5287 OP *o = cUNOPx(nop)->op_first; 5288 assert(o->op_type == OP_NULL); 5289 if (o->op_targ == OP_SCOPE) { 5290 o = cUNOPo->op_first; 5291 } 5292 else { 5293 assert(o->op_targ == OP_LEAVE); 5294 o = cUNOPo->op_first; 5295 assert(o->op_type == OP_ENTER); 5296 o = o->op_sibling; 5297 } 5298 5299 if (o->op_type != OP_STUB) { 5300 assert( o->op_type == OP_NEXTSTATE 5301 || o->op_type == OP_DBSTATE 5302 || (o->op_type == OP_NULL 5303 && ( o->op_targ == OP_NEXTSTATE 5304 || o->op_targ == OP_DBSTATE 5305 ) 5306 ) 5307 ); 5308 PL_curcop = (COP*)o; 5309 } 5310 } 5311 nop = nop->op_next; 5312 5313 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log, 5314 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) ); 5315 5316 rex->offs[0].end = locinput - reginfo->strbeg; 5317 if (reginfo->info_aux_eval->pos_magic) 5318 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic, 5319 reginfo->sv, reginfo->strbeg, 5320 locinput - reginfo->strbeg); 5321 5322 if (sv_yes_mark) { 5323 SV *sv_mrk = get_sv("REGMARK", 1); 5324 sv_setsv(sv_mrk, sv_yes_mark); 5325 } 5326 5327 /* we don't use MULTICALL here as we want to call the 5328 * first op of the block of interest, rather than the 5329 * first op of the sub */ 5330 before = (IV)(SP-PL_stack_base); 5331 PL_op = nop; 5332 CALLRUNOPS(aTHX); /* Scalar context. */ 5333 SPAGAIN; 5334 if ((IV)(SP-PL_stack_base) == before) 5335 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */ 5336 else { 5337 ret = POPs; 5338 PUTBACK; 5339 } 5340 5341 /* before restoring everything, evaluate the returned 5342 * value, so that 'uninit' warnings don't use the wrong 5343 * PL_op or pad. Also need to process any magic vars 5344 * (e.g. $1) *before* parentheses are restored */ 5345 5346 PL_op = NULL; 5347 5348 re_sv = NULL; 5349 if (logical == 0) /* (?{})/ */ 5350 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */ 5351 else if (logical == 1) { /* /(?(?{...})X|Y)/ */ 5352 sw = cBOOL(SvTRUE(ret)); 5353 logical = 0; 5354 } 5355 else { /* /(??{}) */ 5356 /* if its overloaded, let the regex compiler handle 5357 * it; otherwise extract regex, or stringify */ 5358 if (SvGMAGICAL(ret)) 5359 ret = sv_mortalcopy(ret); 5360 if (!SvAMAGIC(ret)) { 5361 SV *sv = ret; 5362 if (SvROK(sv)) 5363 sv = SvRV(sv); 5364 if (SvTYPE(sv) == SVt_REGEXP) 5365 re_sv = (REGEXP*) sv; 5366 else if (SvSMAGICAL(ret)) { 5367 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr); 5368 if (mg) 5369 re_sv = (REGEXP *) mg->mg_obj; 5370 } 5371 5372 /* force any undef warnings here */ 5373 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) { 5374 ret = sv_mortalcopy(ret); 5375 (void) SvPV_force_nolen(ret); 5376 } 5377 } 5378 5379 } 5380 5381 /* *** Note that at this point we don't restore 5382 * PL_comppad, (or pop the CxSUB) on the assumption it may 5383 * be used again soon. This is safe as long as nothing 5384 * in the regexp code uses the pad ! */ 5385 PL_op = oop; 5386 PL_curcop = ocurcop; 5387 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen); 5388 PL_curpm = PL_reg_curpm; 5389 5390 if (logical != 2) 5391 break; 5392 } 5393 5394 /* only /(??{})/ from now on */ 5395 logical = 0; 5396 { 5397 /* extract RE object from returned value; compiling if 5398 * necessary */ 5399 5400 if (re_sv) { 5401 re_sv = reg_temp_copy(NULL, re_sv); 5402 } 5403 else { 5404 U32 pm_flags = 0; 5405 5406 if (SvUTF8(ret) && IN_BYTES) { 5407 /* In use 'bytes': make a copy of the octet 5408 * sequence, but without the flag on */ 5409 STRLEN len; 5410 const char *const p = SvPV(ret, len); 5411 ret = newSVpvn_flags(p, len, SVs_TEMP); 5412 } 5413 if (rex->intflags & PREGf_USE_RE_EVAL) 5414 pm_flags |= PMf_USE_RE_EVAL; 5415 5416 /* if we got here, it should be an engine which 5417 * supports compiling code blocks and stuff */ 5418 assert(rex->engine && rex->engine->op_comp); 5419 assert(!(scan->flags & ~RXf_PMf_COMPILETIME)); 5420 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL, 5421 rex->engine, NULL, NULL, 5422 /* copy /msix etc to inner pattern */ 5423 scan->flags, 5424 pm_flags); 5425 5426 if (!(SvFLAGS(ret) 5427 & (SVs_TEMP | SVs_GMG | SVf_ROK)) 5428 && (!SvPADTMP(ret) || SvREADONLY(ret))) { 5429 /* This isn't a first class regexp. Instead, it's 5430 caching a regexp onto an existing, Perl visible 5431 scalar. */ 5432 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0); 5433 } 5434 } 5435 SAVEFREESV(re_sv); 5436 re = ReANY(re_sv); 5437 } 5438 RXp_MATCH_COPIED_off(re); 5439 re->subbeg = rex->subbeg; 5440 re->sublen = rex->sublen; 5441 re->suboffset = rex->suboffset; 5442 re->subcoffset = rex->subcoffset; 5443 re->lastparen = 0; 5444 re->lastcloseparen = 0; 5445 rei = RXi_GET(re); 5446 DEBUG_EXECUTE_r( 5447 debug_start_match(re_sv, utf8_target, locinput, 5448 reginfo->strend, "Matching embedded"); 5449 ); 5450 startpoint = rei->program + 1; 5451 ST.close_paren = 0; /* only used for GOSUB */ 5452 /* Save all the seen positions so far. */ 5453 ST.cp = regcppush(rex, 0, maxopenparen); 5454 REGCP_SET(ST.lastcp); 5455 /* and set maxopenparen to 0, since we are starting a "fresh" match */ 5456 maxopenparen = 0; 5457 /* run the pattern returned from (??{...}) */ 5458 5459 eval_recurse_doit: /* Share code with GOSUB below this line 5460 * At this point we expect the stack context to be 5461 * set up correctly */ 5462 5463 /* invalidate the S-L poscache. We're now executing a 5464 * different set of WHILEM ops (and their associated 5465 * indexes) against the same string, so the bits in the 5466 * cache are meaningless. Setting maxiter to zero forces 5467 * the cache to be invalidated and zeroed before reuse. 5468 * XXX This is too dramatic a measure. Ideally we should 5469 * save the old cache and restore when running the outer 5470 * pattern again */ 5471 reginfo->poscache_maxiter = 0; 5472 5473 /* the new regexp might have a different is_utf8_pat than we do */ 5474 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv)); 5475 5476 ST.prev_rex = rex_sv; 5477 ST.prev_curlyx = cur_curlyx; 5478 rex_sv = re_sv; 5479 SET_reg_curpm(rex_sv); 5480 rex = re; 5481 rexi = rei; 5482 cur_curlyx = NULL; 5483 ST.B = next; 5484 ST.prev_eval = cur_eval; 5485 cur_eval = st; 5486 /* now continue from first node in postoned RE */ 5487 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput); 5488 assert(0); /* NOTREACHED */ 5489 } 5490 5491 case EVAL_AB: /* cleanup after a successful (??{A})B */ 5492 /* note: this is called twice; first after popping B, then A */ 5493 rex_sv = ST.prev_rex; 5494 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); 5495 SET_reg_curpm(rex_sv); 5496 rex = ReANY(rex_sv); 5497 rexi = RXi_GET(rex); 5498 { 5499 /* preserve $^R across LEAVE's. See Bug 121070. */ 5500 SV *save_sv= GvSV(PL_replgv); 5501 SvREFCNT_inc(save_sv); 5502 regcpblow(ST.cp); /* LEAVE in disguise */ 5503 sv_setsv(GvSV(PL_replgv), save_sv); 5504 SvREFCNT_dec(save_sv); 5505 } 5506 cur_eval = ST.prev_eval; 5507 cur_curlyx = ST.prev_curlyx; 5508 5509 /* Invalidate cache. See "invalidate" comment above. */ 5510 reginfo->poscache_maxiter = 0; 5511 if ( nochange_depth ) 5512 nochange_depth--; 5513 sayYES; 5514 5515 5516 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ 5517 /* note: this is called twice; first after popping B, then A */ 5518 rex_sv = ST.prev_rex; 5519 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); 5520 SET_reg_curpm(rex_sv); 5521 rex = ReANY(rex_sv); 5522 rexi = RXi_GET(rex); 5523 5524 REGCP_UNWIND(ST.lastcp); 5525 regcppop(rex, &maxopenparen); 5526 cur_eval = ST.prev_eval; 5527 cur_curlyx = ST.prev_curlyx; 5528 /* Invalidate cache. See "invalidate" comment above. */ 5529 reginfo->poscache_maxiter = 0; 5530 if ( nochange_depth ) 5531 nochange_depth--; 5532 sayNO_SILENT; 5533 #undef ST 5534 5535 case OPEN: /* ( */ 5536 n = ARG(scan); /* which paren pair */ 5537 rex->offs[n].start_tmp = locinput - reginfo->strbeg; 5538 if (n > maxopenparen) 5539 maxopenparen = n; 5540 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, 5541 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n", 5542 PTR2UV(rex), 5543 PTR2UV(rex->offs), 5544 (UV)n, 5545 (IV)rex->offs[n].start_tmp, 5546 (UV)maxopenparen 5547 )); 5548 lastopen = n; 5549 break; 5550 5551 /* XXX really need to log other places start/end are set too */ 5552 #define CLOSE_CAPTURE \ 5553 rex->offs[n].start = rex->offs[n].start_tmp; \ 5554 rex->offs[n].end = locinput - reginfo->strbeg; \ 5555 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \ 5556 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \ 5557 PTR2UV(rex), \ 5558 PTR2UV(rex->offs), \ 5559 (UV)n, \ 5560 (IV)rex->offs[n].start, \ 5561 (IV)rex->offs[n].end \ 5562 )) 5563 5564 case CLOSE: /* ) */ 5565 n = ARG(scan); /* which paren pair */ 5566 CLOSE_CAPTURE; 5567 if (n > rex->lastparen) 5568 rex->lastparen = n; 5569 rex->lastcloseparen = n; 5570 if (cur_eval && cur_eval->u.eval.close_paren == n) { 5571 goto fake_end; 5572 } 5573 break; 5574 5575 case ACCEPT: /* (*ACCEPT) */ 5576 if (ARG(scan)){ 5577 regnode *cursor; 5578 for (cursor=scan; 5579 cursor && OP(cursor)!=END; 5580 cursor=regnext(cursor)) 5581 { 5582 if ( OP(cursor)==CLOSE ){ 5583 n = ARG(cursor); 5584 if ( n <= lastopen ) { 5585 CLOSE_CAPTURE; 5586 if (n > rex->lastparen) 5587 rex->lastparen = n; 5588 rex->lastcloseparen = n; 5589 if ( n == ARG(scan) || (cur_eval && 5590 cur_eval->u.eval.close_paren == n)) 5591 break; 5592 } 5593 } 5594 } 5595 } 5596 goto fake_end; 5597 /*NOTREACHED*/ 5598 5599 case GROUPP: /* (?(1)) */ 5600 n = ARG(scan); /* which paren pair */ 5601 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1); 5602 break; 5603 5604 case NGROUPP: /* (?(<name>)) */ 5605 /* reg_check_named_buff_matched returns 0 for no match */ 5606 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan)); 5607 break; 5608 5609 case INSUBP: /* (?(R)) */ 5610 n = ARG(scan); 5611 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n)); 5612 break; 5613 5614 case DEFINEP: /* (?(DEFINE)) */ 5615 sw = 0; 5616 break; 5617 5618 case IFTHEN: /* (?(cond)A|B) */ 5619 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ 5620 if (sw) 5621 next = NEXTOPER(NEXTOPER(scan)); 5622 else { 5623 next = scan + ARG(scan); 5624 if (OP(next) == IFTHEN) /* Fake one. */ 5625 next = NEXTOPER(NEXTOPER(next)); 5626 } 5627 break; 5628 5629 case LOGICAL: /* modifier for EVAL and IFMATCH */ 5630 logical = scan->flags; 5631 break; 5632 5633 /******************************************************************* 5634 5635 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ 5636 pattern, where A and B are subpatterns. (For simple A, CURLYM or 5637 STAR/PLUS/CURLY/CURLYN are used instead.) 5638 5639 A*B is compiled as <CURLYX><A><WHILEM><B> 5640 5641 On entry to the subpattern, CURLYX is called. This pushes a CURLYX 5642 state, which contains the current count, initialised to -1. It also sets 5643 cur_curlyx to point to this state, with any previous value saved in the 5644 state block. 5645 5646 CURLYX then jumps straight to the WHILEM op, rather than executing A, 5647 since the pattern may possibly match zero times (i.e. it's a while {} loop 5648 rather than a do {} while loop). 5649 5650 Each entry to WHILEM represents a successful match of A. The count in the 5651 CURLYX block is incremented, another WHILEM state is pushed, and execution 5652 passes to A or B depending on greediness and the current count. 5653 5654 For example, if matching against the string a1a2a3b (where the aN are 5655 substrings that match /A/), then the match progresses as follows: (the 5656 pushed states are interspersed with the bits of strings matched so far): 5657 5658 <CURLYX cnt=-1> 5659 <CURLYX cnt=0><WHILEM> 5660 <CURLYX cnt=1><WHILEM> a1 <WHILEM> 5661 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM> 5662 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> 5663 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b 5664 5665 (Contrast this with something like CURLYM, which maintains only a single 5666 backtrack state: 5667 5668 <CURLYM cnt=0> a1 5669 a1 <CURLYM cnt=1> a2 5670 a1 a2 <CURLYM cnt=2> a3 5671 a1 a2 a3 <CURLYM cnt=3> b 5672 ) 5673 5674 Each WHILEM state block marks a point to backtrack to upon partial failure 5675 of A or B, and also contains some minor state data related to that 5676 iteration. The CURLYX block, pointed to by cur_curlyx, contains the 5677 overall state, such as the count, and pointers to the A and B ops. 5678 5679 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx 5680 must always point to the *current* CURLYX block, the rules are: 5681 5682 When executing CURLYX, save the old cur_curlyx in the CURLYX state block, 5683 and set cur_curlyx to point the new block. 5684 5685 When popping the CURLYX block after a successful or unsuccessful match, 5686 restore the previous cur_curlyx. 5687 5688 When WHILEM is about to execute B, save the current cur_curlyx, and set it 5689 to the outer one saved in the CURLYX block. 5690 5691 When popping the WHILEM block after a successful or unsuccessful B match, 5692 restore the previous cur_curlyx. 5693 5694 Here's an example for the pattern (AI* BI)*BO 5695 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: 5696 5697 cur_ 5698 curlyx backtrack stack 5699 ------ --------------- 5700 NULL 5701 CO <CO prev=NULL> <WO> 5702 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai 5703 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi 5704 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo 5705 5706 At this point the pattern succeeds, and we work back down the stack to 5707 clean up, restoring as we go: 5708 5709 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi 5710 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai 5711 CO <CO prev=NULL> <WO> 5712 NULL 5713 5714 *******************************************************************/ 5715 5716 #define ST st->u.curlyx 5717 5718 case CURLYX: /* start of /A*B/ (for complex A) */ 5719 { 5720 /* No need to save/restore up to this paren */ 5721 I32 parenfloor = scan->flags; 5722 5723 assert(next); /* keep Coverity happy */ 5724 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ 5725 next += ARG(next); 5726 5727 /* XXXX Probably it is better to teach regpush to support 5728 parenfloor > maxopenparen ... */ 5729 if (parenfloor > (I32)rex->lastparen) 5730 parenfloor = rex->lastparen; /* Pessimization... */ 5731 5732 ST.prev_curlyx= cur_curlyx; 5733 cur_curlyx = st; 5734 ST.cp = PL_savestack_ix; 5735 5736 /* these fields contain the state of the current curly. 5737 * they are accessed by subsequent WHILEMs */ 5738 ST.parenfloor = parenfloor; 5739 ST.me = scan; 5740 ST.B = next; 5741 ST.minmod = minmod; 5742 minmod = 0; 5743 ST.count = -1; /* this will be updated by WHILEM */ 5744 ST.lastloc = NULL; /* this will be updated by WHILEM */ 5745 5746 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput); 5747 assert(0); /* NOTREACHED */ 5748 } 5749 5750 case CURLYX_end: /* just finished matching all of A*B */ 5751 cur_curlyx = ST.prev_curlyx; 5752 sayYES; 5753 assert(0); /* NOTREACHED */ 5754 5755 case CURLYX_end_fail: /* just failed to match all of A*B */ 5756 regcpblow(ST.cp); 5757 cur_curlyx = ST.prev_curlyx; 5758 sayNO; 5759 assert(0); /* NOTREACHED */ 5760 5761 5762 #undef ST 5763 #define ST st->u.whilem 5764 5765 case WHILEM: /* just matched an A in /A*B/ (for complex A) */ 5766 { 5767 /* see the discussion above about CURLYX/WHILEM */ 5768 I32 n; 5769 int min = ARG1(cur_curlyx->u.curlyx.me); 5770 int max = ARG2(cur_curlyx->u.curlyx.me); 5771 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS; 5772 5773 assert(cur_curlyx); /* keep Coverity happy */ 5774 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ 5775 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; 5776 ST.cache_offset = 0; 5777 ST.cache_mask = 0; 5778 5779 5780 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 5781 "%*s whilem: matched %ld out of %d..%d\n", 5782 REPORT_CODE_OFF+depth*2, "", (long)n, min, max) 5783 ); 5784 5785 /* First just match a string of min A's. */ 5786 5787 if (n < min) { 5788 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, 5789 maxopenparen); 5790 cur_curlyx->u.curlyx.lastloc = locinput; 5791 REGCP_SET(ST.lastcp); 5792 5793 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput); 5794 assert(0); /* NOTREACHED */ 5795 } 5796 5797 /* If degenerate A matches "", assume A done. */ 5798 5799 if (locinput == cur_curlyx->u.curlyx.lastloc) { 5800 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 5801 "%*s whilem: empty match detected, trying continuation...\n", 5802 REPORT_CODE_OFF+depth*2, "") 5803 ); 5804 goto do_whilem_B_max; 5805 } 5806 5807 /* super-linear cache processing. 5808 * 5809 * The idea here is that for certain types of CURLYX/WHILEM - 5810 * principally those whose upper bound is infinity (and 5811 * excluding regexes that have things like \1 and other very 5812 * non-regular expresssiony things), then if a pattern like 5813 * /....A*.../ fails and we backtrack to the WHILEM, then we 5814 * make a note that this particular WHILEM op was at string 5815 * position 47 (say) when the rest of pattern failed. Then, if 5816 * we ever find ourselves back at that WHILEM, and at string 5817 * position 47 again, we can just fail immediately rather than 5818 * running the rest of the pattern again. 5819 * 5820 * This is very handy when patterns start to go 5821 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up 5822 * with a combinatorial explosion of backtracking. 5823 * 5824 * The cache is implemented as a bit array, with one bit per 5825 * string byte position per WHILEM op (up to 16) - so its 5826 * between 0.25 and 2x the string size. 5827 * 5828 * To avoid allocating a poscache buffer every time, we do an 5829 * initially countdown; only after we have executed a WHILEM 5830 * op (string-length x #WHILEMs) times do we allocate the 5831 * cache. 5832 * 5833 * The top 4 bits of scan->flags byte say how many different 5834 * relevant CURLLYX/WHILEM op pairs there are, while the 5835 * bottom 4-bits is the identifying index number of this 5836 * WHILEM. 5837 */ 5838 5839 if (scan->flags) { 5840 5841 if (!reginfo->poscache_maxiter) { 5842 /* start the countdown: Postpone detection until we 5843 * know the match is not *that* much linear. */ 5844 reginfo->poscache_maxiter 5845 = (reginfo->strend - reginfo->strbeg + 1) 5846 * (scan->flags>>4); 5847 /* possible overflow for long strings and many CURLYX's */ 5848 if (reginfo->poscache_maxiter < 0) 5849 reginfo->poscache_maxiter = I32_MAX; 5850 reginfo->poscache_iter = reginfo->poscache_maxiter; 5851 } 5852 5853 if (reginfo->poscache_iter-- == 0) { 5854 /* initialise cache */ 5855 const SSize_t size = (reginfo->poscache_maxiter + 7)/8; 5856 regmatch_info_aux *const aux = reginfo->info_aux; 5857 if (aux->poscache) { 5858 if ((SSize_t)reginfo->poscache_size < size) { 5859 Renew(aux->poscache, size, char); 5860 reginfo->poscache_size = size; 5861 } 5862 Zero(aux->poscache, size, char); 5863 } 5864 else { 5865 reginfo->poscache_size = size; 5866 Newxz(aux->poscache, size, char); 5867 } 5868 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 5869 "%swhilem: Detected a super-linear match, switching on caching%s...\n", 5870 PL_colors[4], PL_colors[5]) 5871 ); 5872 } 5873 5874 if (reginfo->poscache_iter < 0) { 5875 /* have we already failed at this position? */ 5876 SSize_t offset, mask; 5877 5878 reginfo->poscache_iter = -1; /* stop eventual underflow */ 5879 offset = (scan->flags & 0xf) - 1 5880 + (locinput - reginfo->strbeg) 5881 * (scan->flags>>4); 5882 mask = 1 << (offset % 8); 5883 offset /= 8; 5884 if (reginfo->info_aux->poscache[offset] & mask) { 5885 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, 5886 "%*s whilem: (cache) already tried at this position...\n", 5887 REPORT_CODE_OFF+depth*2, "") 5888 ); 5889 sayNO; /* cache records failure */ 5890 } 5891 ST.cache_offset = offset; 5892 ST.cache_mask = mask; 5893 } 5894 } 5895 5896 /* Prefer B over A for minimal matching. */ 5897 5898 if (cur_curlyx->u.curlyx.minmod) { 5899 ST.save_curlyx = cur_curlyx; 5900 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; 5901 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor, 5902 maxopenparen); 5903 REGCP_SET(ST.lastcp); 5904 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B, 5905 locinput); 5906 assert(0); /* NOTREACHED */ 5907 } 5908 5909 /* Prefer A over B for maximal matching. */ 5910 5911 if (n < max) { /* More greed allowed? */ 5912 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, 5913 maxopenparen); 5914 cur_curlyx->u.curlyx.lastloc = locinput; 5915 REGCP_SET(ST.lastcp); 5916 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput); 5917 assert(0); /* NOTREACHED */ 5918 } 5919 goto do_whilem_B_max; 5920 } 5921 assert(0); /* NOTREACHED */ 5922 5923 case WHILEM_B_min: /* just matched B in a minimal match */ 5924 case WHILEM_B_max: /* just matched B in a maximal match */ 5925 cur_curlyx = ST.save_curlyx; 5926 sayYES; 5927 assert(0); /* NOTREACHED */ 5928 5929 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ 5930 cur_curlyx = ST.save_curlyx; 5931 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; 5932 cur_curlyx->u.curlyx.count--; 5933 CACHEsayNO; 5934 assert(0); /* NOTREACHED */ 5935 5936 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ 5937 /* FALL THROUGH */ 5938 case WHILEM_A_pre_fail: /* just failed to match even minimal A */ 5939 REGCP_UNWIND(ST.lastcp); 5940 regcppop(rex, &maxopenparen); 5941 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; 5942 cur_curlyx->u.curlyx.count--; 5943 CACHEsayNO; 5944 assert(0); /* NOTREACHED */ 5945 5946 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ 5947 REGCP_UNWIND(ST.lastcp); 5948 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */ 5949 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 5950 "%*s whilem: failed, trying continuation...\n", 5951 REPORT_CODE_OFF+depth*2, "") 5952 ); 5953 do_whilem_B_max: 5954 if (cur_curlyx->u.curlyx.count >= REG_INFTY 5955 && ckWARN(WARN_REGEXP) 5956 && !reginfo->warned) 5957 { 5958 reginfo->warned = TRUE; 5959 Perl_warner(aTHX_ packWARN(WARN_REGEXP), 5960 "Complex regular subexpression recursion limit (%d) " 5961 "exceeded", 5962 REG_INFTY - 1); 5963 } 5964 5965 /* now try B */ 5966 ST.save_curlyx = cur_curlyx; 5967 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; 5968 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B, 5969 locinput); 5970 assert(0); /* NOTREACHED */ 5971 5972 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ 5973 cur_curlyx = ST.save_curlyx; 5974 REGCP_UNWIND(ST.lastcp); 5975 regcppop(rex, &maxopenparen); 5976 5977 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) { 5978 /* Maximum greed exceeded */ 5979 if (cur_curlyx->u.curlyx.count >= REG_INFTY 5980 && ckWARN(WARN_REGEXP) 5981 && !reginfo->warned) 5982 { 5983 reginfo->warned = TRUE; 5984 Perl_warner(aTHX_ packWARN(WARN_REGEXP), 5985 "Complex regular subexpression recursion " 5986 "limit (%d) exceeded", 5987 REG_INFTY - 1); 5988 } 5989 cur_curlyx->u.curlyx.count--; 5990 CACHEsayNO; 5991 } 5992 5993 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 5994 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "") 5995 ); 5996 /* Try grabbing another A and see if it helps. */ 5997 cur_curlyx->u.curlyx.lastloc = locinput; 5998 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, 5999 maxopenparen); 6000 REGCP_SET(ST.lastcp); 6001 PUSH_STATE_GOTO(WHILEM_A_min, 6002 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS, 6003 locinput); 6004 assert(0); /* NOTREACHED */ 6005 6006 #undef ST 6007 #define ST st->u.branch 6008 6009 case BRANCHJ: /* /(...|A|...)/ with long next pointer */ 6010 next = scan + ARG(scan); 6011 if (next == scan) 6012 next = NULL; 6013 scan = NEXTOPER(scan); 6014 /* FALL THROUGH */ 6015 6016 case BRANCH: /* /(...|A|...)/ */ 6017 scan = NEXTOPER(scan); /* scan now points to inner node */ 6018 ST.lastparen = rex->lastparen; 6019 ST.lastcloseparen = rex->lastcloseparen; 6020 ST.next_branch = next; 6021 REGCP_SET(ST.cp); 6022 6023 /* Now go into the branch */ 6024 if (has_cutgroup) { 6025 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput); 6026 } else { 6027 PUSH_STATE_GOTO(BRANCH_next, scan, locinput); 6028 } 6029 assert(0); /* NOTREACHED */ 6030 6031 case CUTGROUP: /* /(*THEN)/ */ 6032 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL : 6033 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 6034 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput); 6035 assert(0); /* NOTREACHED */ 6036 6037 case CUTGROUP_next_fail: 6038 do_cutgroup = 1; 6039 no_final = 1; 6040 if (st->u.mark.mark_name) 6041 sv_commit = st->u.mark.mark_name; 6042 sayNO; 6043 assert(0); /* NOTREACHED */ 6044 6045 case BRANCH_next: 6046 sayYES; 6047 assert(0); /* NOTREACHED */ 6048 6049 case BRANCH_next_fail: /* that branch failed; try the next, if any */ 6050 if (do_cutgroup) { 6051 do_cutgroup = 0; 6052 no_final = 0; 6053 } 6054 REGCP_UNWIND(ST.cp); 6055 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 6056 scan = ST.next_branch; 6057 /* no more branches? */ 6058 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) { 6059 DEBUG_EXECUTE_r({ 6060 PerlIO_printf( Perl_debug_log, 6061 "%*s %sBRANCH failed...%s\n", 6062 REPORT_CODE_OFF+depth*2, "", 6063 PL_colors[4], 6064 PL_colors[5] ); 6065 }); 6066 sayNO_SILENT; 6067 } 6068 continue; /* execute next BRANCH[J] op */ 6069 assert(0); /* NOTREACHED */ 6070 6071 case MINMOD: /* next op will be non-greedy, e.g. A*? */ 6072 minmod = 1; 6073 break; 6074 6075 #undef ST 6076 #define ST st->u.curlym 6077 6078 case CURLYM: /* /A{m,n}B/ where A is fixed-length */ 6079 6080 /* This is an optimisation of CURLYX that enables us to push 6081 * only a single backtracking state, no matter how many matches 6082 * there are in {m,n}. It relies on the pattern being constant 6083 * length, with no parens to influence future backrefs 6084 */ 6085 6086 ST.me = scan; 6087 scan = NEXTOPER(scan) + NODE_STEP_REGNODE; 6088 6089 ST.lastparen = rex->lastparen; 6090 ST.lastcloseparen = rex->lastcloseparen; 6091 6092 /* if paren positive, emulate an OPEN/CLOSE around A */ 6093 if (ST.me->flags) { 6094 U32 paren = ST.me->flags; 6095 if (paren > maxopenparen) 6096 maxopenparen = paren; 6097 scan += NEXT_OFF(scan); /* Skip former OPEN. */ 6098 } 6099 ST.A = scan; 6100 ST.B = next; 6101 ST.alen = 0; 6102 ST.count = 0; 6103 ST.minmod = minmod; 6104 minmod = 0; 6105 ST.c1 = CHRTEST_UNINIT; 6106 REGCP_SET(ST.cp); 6107 6108 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */ 6109 goto curlym_do_B; 6110 6111 curlym_do_A: /* execute the A in /A{m,n}B/ */ 6112 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */ 6113 assert(0); /* NOTREACHED */ 6114 6115 case CURLYM_A: /* we've just matched an A */ 6116 ST.count++; 6117 /* after first match, determine A's length: u.curlym.alen */ 6118 if (ST.count == 1) { 6119 if (reginfo->is_utf8_target) { 6120 char *s = st->locinput; 6121 while (s < locinput) { 6122 ST.alen++; 6123 s += UTF8SKIP(s); 6124 } 6125 } 6126 else { 6127 ST.alen = locinput - st->locinput; 6128 } 6129 if (ST.alen == 0) 6130 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me); 6131 } 6132 DEBUG_EXECUTE_r( 6133 PerlIO_printf(Perl_debug_log, 6134 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", 6135 (int)(REPORT_CODE_OFF+(depth*2)), "", 6136 (IV) ST.count, (IV)ST.alen) 6137 ); 6138 6139 if (cur_eval && cur_eval->u.eval.close_paren && 6140 cur_eval->u.eval.close_paren == (U32)ST.me->flags) 6141 goto fake_end; 6142 6143 { 6144 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)); 6145 if ( max == REG_INFTY || ST.count < max ) 6146 goto curlym_do_A; /* try to match another A */ 6147 } 6148 goto curlym_do_B; /* try to match B */ 6149 6150 case CURLYM_A_fail: /* just failed to match an A */ 6151 REGCP_UNWIND(ST.cp); 6152 6153 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ 6154 || (cur_eval && cur_eval->u.eval.close_paren && 6155 cur_eval->u.eval.close_paren == (U32)ST.me->flags)) 6156 sayNO; 6157 6158 curlym_do_B: /* execute the B in /A{m,n}B/ */ 6159 if (ST.c1 == CHRTEST_UNINIT) { 6160 /* calculate c1 and c2 for possible match of 1st char 6161 * following curly */ 6162 ST.c1 = ST.c2 = CHRTEST_VOID; 6163 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) { 6164 regnode *text_node = ST.B; 6165 if (! HAS_TEXT(text_node)) 6166 FIND_NEXT_IMPT(text_node); 6167 /* this used to be 6168 6169 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT) 6170 6171 But the former is redundant in light of the latter. 6172 6173 if this changes back then the macro for 6174 IS_TEXT and friends need to change. 6175 */ 6176 if (PL_regkind[OP(text_node)] == EXACT) { 6177 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ 6178 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, 6179 reginfo)) 6180 { 6181 sayNO; 6182 } 6183 } 6184 } 6185 } 6186 6187 DEBUG_EXECUTE_r( 6188 PerlIO_printf(Perl_debug_log, 6189 "%*s CURLYM trying tail with matches=%"IVdf"...\n", 6190 (int)(REPORT_CODE_OFF+(depth*2)), 6191 "", (IV)ST.count) 6192 ); 6193 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) { 6194 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) { 6195 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) 6196 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) 6197 { 6198 /* simulate B failing */ 6199 DEBUG_OPTIMISE_r( 6200 PerlIO_printf(Perl_debug_log, 6201 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n", 6202 (int)(REPORT_CODE_OFF+(depth*2)),"", 6203 valid_utf8_to_uvchr((U8 *) locinput, NULL), 6204 valid_utf8_to_uvchr(ST.c1_utf8, NULL), 6205 valid_utf8_to_uvchr(ST.c2_utf8, NULL)) 6206 ); 6207 state_num = CURLYM_B_fail; 6208 goto reenter_switch; 6209 } 6210 } 6211 else if (nextchr != ST.c1 && nextchr != ST.c2) { 6212 /* simulate B failing */ 6213 DEBUG_OPTIMISE_r( 6214 PerlIO_printf(Perl_debug_log, 6215 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n", 6216 (int)(REPORT_CODE_OFF+(depth*2)),"", 6217 (int) nextchr, ST.c1, ST.c2) 6218 ); 6219 state_num = CURLYM_B_fail; 6220 goto reenter_switch; 6221 } 6222 } 6223 6224 if (ST.me->flags) { 6225 /* emulate CLOSE: mark current A as captured */ 6226 I32 paren = ST.me->flags; 6227 if (ST.count) { 6228 rex->offs[paren].start 6229 = HOPc(locinput, -ST.alen) - reginfo->strbeg; 6230 rex->offs[paren].end = locinput - reginfo->strbeg; 6231 if ((U32)paren > rex->lastparen) 6232 rex->lastparen = paren; 6233 rex->lastcloseparen = paren; 6234 } 6235 else 6236 rex->offs[paren].end = -1; 6237 if (cur_eval && cur_eval->u.eval.close_paren && 6238 cur_eval->u.eval.close_paren == (U32)ST.me->flags) 6239 { 6240 if (ST.count) 6241 goto fake_end; 6242 else 6243 sayNO; 6244 } 6245 } 6246 6247 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */ 6248 assert(0); /* NOTREACHED */ 6249 6250 case CURLYM_B_fail: /* just failed to match a B */ 6251 REGCP_UNWIND(ST.cp); 6252 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 6253 if (ST.minmod) { 6254 I32 max = ARG2(ST.me); 6255 if (max != REG_INFTY && ST.count == max) 6256 sayNO; 6257 goto curlym_do_A; /* try to match a further A */ 6258 } 6259 /* backtrack one A */ 6260 if (ST.count == ARG1(ST.me) /* min */) 6261 sayNO; 6262 ST.count--; 6263 SET_locinput(HOPc(locinput, -ST.alen)); 6264 goto curlym_do_B; /* try to match B */ 6265 6266 #undef ST 6267 #define ST st->u.curly 6268 6269 #define CURLY_SETPAREN(paren, success) \ 6270 if (paren) { \ 6271 if (success) { \ 6272 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \ 6273 rex->offs[paren].end = locinput - reginfo->strbeg; \ 6274 if (paren > rex->lastparen) \ 6275 rex->lastparen = paren; \ 6276 rex->lastcloseparen = paren; \ 6277 } \ 6278 else { \ 6279 rex->offs[paren].end = -1; \ 6280 rex->lastparen = ST.lastparen; \ 6281 rex->lastcloseparen = ST.lastcloseparen; \ 6282 } \ 6283 } 6284 6285 case STAR: /* /A*B/ where A is width 1 char */ 6286 ST.paren = 0; 6287 ST.min = 0; 6288 ST.max = REG_INFTY; 6289 scan = NEXTOPER(scan); 6290 goto repeat; 6291 6292 case PLUS: /* /A+B/ where A is width 1 char */ 6293 ST.paren = 0; 6294 ST.min = 1; 6295 ST.max = REG_INFTY; 6296 scan = NEXTOPER(scan); 6297 goto repeat; 6298 6299 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */ 6300 ST.paren = scan->flags; /* Which paren to set */ 6301 ST.lastparen = rex->lastparen; 6302 ST.lastcloseparen = rex->lastcloseparen; 6303 if (ST.paren > maxopenparen) 6304 maxopenparen = ST.paren; 6305 ST.min = ARG1(scan); /* min to match */ 6306 ST.max = ARG2(scan); /* max to match */ 6307 if (cur_eval && cur_eval->u.eval.close_paren && 6308 cur_eval->u.eval.close_paren == (U32)ST.paren) { 6309 ST.min=1; 6310 ST.max=1; 6311 } 6312 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); 6313 goto repeat; 6314 6315 case CURLY: /* /A{m,n}B/ where A is width 1 char */ 6316 ST.paren = 0; 6317 ST.min = ARG1(scan); /* min to match */ 6318 ST.max = ARG2(scan); /* max to match */ 6319 scan = NEXTOPER(scan) + NODE_STEP_REGNODE; 6320 repeat: 6321 /* 6322 * Lookahead to avoid useless match attempts 6323 * when we know what character comes next. 6324 * 6325 * Used to only do .*x and .*?x, but now it allows 6326 * for )'s, ('s and (?{ ... })'s to be in the way 6327 * of the quantifier and the EXACT-like node. -- japhy 6328 */ 6329 6330 assert(ST.min <= ST.max); 6331 if (! HAS_TEXT(next) && ! JUMPABLE(next)) { 6332 ST.c1 = ST.c2 = CHRTEST_VOID; 6333 } 6334 else { 6335 regnode *text_node = next; 6336 6337 if (! HAS_TEXT(text_node)) 6338 FIND_NEXT_IMPT(text_node); 6339 6340 if (! HAS_TEXT(text_node)) 6341 ST.c1 = ST.c2 = CHRTEST_VOID; 6342 else { 6343 if ( PL_regkind[OP(text_node)] != EXACT ) { 6344 ST.c1 = ST.c2 = CHRTEST_VOID; 6345 } 6346 else { 6347 6348 /* Currently we only get here when 6349 6350 PL_rekind[OP(text_node)] == EXACT 6351 6352 if this changes back then the macro for IS_TEXT and 6353 friends need to change. */ 6354 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ 6355 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, 6356 reginfo)) 6357 { 6358 sayNO; 6359 } 6360 } 6361 } 6362 } 6363 6364 ST.A = scan; 6365 ST.B = next; 6366 if (minmod) { 6367 char *li = locinput; 6368 minmod = 0; 6369 if (ST.min && 6370 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth) 6371 < ST.min) 6372 sayNO; 6373 SET_locinput(li); 6374 ST.count = ST.min; 6375 REGCP_SET(ST.cp); 6376 if (ST.c1 == CHRTEST_VOID) 6377 goto curly_try_B_min; 6378 6379 ST.oldloc = locinput; 6380 6381 /* set ST.maxpos to the furthest point along the 6382 * string that could possibly match */ 6383 if (ST.max == REG_INFTY) { 6384 ST.maxpos = reginfo->strend - 1; 6385 if (utf8_target) 6386 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos)) 6387 ST.maxpos--; 6388 } 6389 else if (utf8_target) { 6390 int m = ST.max - ST.min; 6391 for (ST.maxpos = locinput; 6392 m >0 && ST.maxpos < reginfo->strend; m--) 6393 ST.maxpos += UTF8SKIP(ST.maxpos); 6394 } 6395 else { 6396 ST.maxpos = locinput + ST.max - ST.min; 6397 if (ST.maxpos >= reginfo->strend) 6398 ST.maxpos = reginfo->strend - 1; 6399 } 6400 goto curly_try_B_min_known; 6401 6402 } 6403 else { 6404 /* avoid taking address of locinput, so it can remain 6405 * a register var */ 6406 char *li = locinput; 6407 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth); 6408 if (ST.count < ST.min) 6409 sayNO; 6410 SET_locinput(li); 6411 if ((ST.count > ST.min) 6412 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL)) 6413 { 6414 /* A{m,n} must come at the end of the string, there's 6415 * no point in backing off ... */ 6416 ST.min = ST.count; 6417 /* ...except that $ and \Z can match before *and* after 6418 newline at the end. Consider "\n\n" =~ /\n+\Z\n/. 6419 We may back off by one in this case. */ 6420 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS) 6421 ST.min--; 6422 } 6423 REGCP_SET(ST.cp); 6424 goto curly_try_B_max; 6425 } 6426 assert(0); /* NOTREACHED */ 6427 6428 6429 case CURLY_B_min_known_fail: 6430 /* failed to find B in a non-greedy match where c1,c2 valid */ 6431 6432 REGCP_UNWIND(ST.cp); 6433 if (ST.paren) { 6434 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 6435 } 6436 /* Couldn't or didn't -- move forward. */ 6437 ST.oldloc = locinput; 6438 if (utf8_target) 6439 locinput += UTF8SKIP(locinput); 6440 else 6441 locinput++; 6442 ST.count++; 6443 curly_try_B_min_known: 6444 /* find the next place where 'B' could work, then call B */ 6445 { 6446 int n; 6447 if (utf8_target) { 6448 n = (ST.oldloc == locinput) ? 0 : 1; 6449 if (ST.c1 == ST.c2) { 6450 /* set n to utf8_distance(oldloc, locinput) */ 6451 while (locinput <= ST.maxpos 6452 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))) 6453 { 6454 locinput += UTF8SKIP(locinput); 6455 n++; 6456 } 6457 } 6458 else { 6459 /* set n to utf8_distance(oldloc, locinput) */ 6460 while (locinput <= ST.maxpos 6461 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) 6462 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) 6463 { 6464 locinput += UTF8SKIP(locinput); 6465 n++; 6466 } 6467 } 6468 } 6469 else { /* Not utf8_target */ 6470 if (ST.c1 == ST.c2) { 6471 while (locinput <= ST.maxpos && 6472 UCHARAT(locinput) != ST.c1) 6473 locinput++; 6474 } 6475 else { 6476 while (locinput <= ST.maxpos 6477 && UCHARAT(locinput) != ST.c1 6478 && UCHARAT(locinput) != ST.c2) 6479 locinput++; 6480 } 6481 n = locinput - ST.oldloc; 6482 } 6483 if (locinput > ST.maxpos) 6484 sayNO; 6485 if (n) { 6486 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is 6487 * at b; check that everything between oldloc and 6488 * locinput matches */ 6489 char *li = ST.oldloc; 6490 ST.count += n; 6491 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n) 6492 sayNO; 6493 assert(n == REG_INFTY || locinput == li); 6494 } 6495 CURLY_SETPAREN(ST.paren, ST.count); 6496 if (cur_eval && cur_eval->u.eval.close_paren && 6497 cur_eval->u.eval.close_paren == (U32)ST.paren) { 6498 goto fake_end; 6499 } 6500 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput); 6501 } 6502 assert(0); /* NOTREACHED */ 6503 6504 6505 case CURLY_B_min_fail: 6506 /* failed to find B in a non-greedy match where c1,c2 invalid */ 6507 6508 REGCP_UNWIND(ST.cp); 6509 if (ST.paren) { 6510 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 6511 } 6512 /* failed -- move forward one */ 6513 { 6514 char *li = locinput; 6515 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) { 6516 sayNO; 6517 } 6518 locinput = li; 6519 } 6520 { 6521 ST.count++; 6522 if (ST.count <= ST.max || (ST.max == REG_INFTY && 6523 ST.count > 0)) /* count overflow ? */ 6524 { 6525 curly_try_B_min: 6526 CURLY_SETPAREN(ST.paren, ST.count); 6527 if (cur_eval && cur_eval->u.eval.close_paren && 6528 cur_eval->u.eval.close_paren == (U32)ST.paren) { 6529 goto fake_end; 6530 } 6531 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput); 6532 } 6533 } 6534 sayNO; 6535 assert(0); /* NOTREACHED */ 6536 6537 6538 curly_try_B_max: 6539 /* a successful greedy match: now try to match B */ 6540 if (cur_eval && cur_eval->u.eval.close_paren && 6541 cur_eval->u.eval.close_paren == (U32)ST.paren) { 6542 goto fake_end; 6543 } 6544 { 6545 bool could_match = locinput < reginfo->strend; 6546 6547 /* If it could work, try it. */ 6548 if (ST.c1 != CHRTEST_VOID && could_match) { 6549 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target) 6550 { 6551 could_match = memEQ(locinput, 6552 ST.c1_utf8, 6553 UTF8SKIP(locinput)) 6554 || memEQ(locinput, 6555 ST.c2_utf8, 6556 UTF8SKIP(locinput)); 6557 } 6558 else { 6559 could_match = UCHARAT(locinput) == ST.c1 6560 || UCHARAT(locinput) == ST.c2; 6561 } 6562 } 6563 if (ST.c1 == CHRTEST_VOID || could_match) { 6564 CURLY_SETPAREN(ST.paren, ST.count); 6565 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput); 6566 assert(0); /* NOTREACHED */ 6567 } 6568 } 6569 /* FALL THROUGH */ 6570 6571 case CURLY_B_max_fail: 6572 /* failed to find B in a greedy match */ 6573 6574 REGCP_UNWIND(ST.cp); 6575 if (ST.paren) { 6576 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); 6577 } 6578 /* back up. */ 6579 if (--ST.count < ST.min) 6580 sayNO; 6581 locinput = HOPc(locinput, -1); 6582 goto curly_try_B_max; 6583 6584 #undef ST 6585 6586 case END: /* last op of main pattern */ 6587 fake_end: 6588 if (cur_eval) { 6589 /* we've just finished A in /(??{A})B/; now continue with B */ 6590 6591 st->u.eval.prev_rex = rex_sv; /* inner */ 6592 6593 /* Save *all* the positions. */ 6594 st->u.eval.cp = regcppush(rex, 0, maxopenparen); 6595 rex_sv = cur_eval->u.eval.prev_rex; 6596 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); 6597 SET_reg_curpm(rex_sv); 6598 rex = ReANY(rex_sv); 6599 rexi = RXi_GET(rex); 6600 cur_curlyx = cur_eval->u.eval.prev_curlyx; 6601 6602 REGCP_SET(st->u.eval.lastcp); 6603 6604 /* Restore parens of the outer rex without popping the 6605 * savestack */ 6606 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp, 6607 &maxopenparen); 6608 6609 st->u.eval.prev_eval = cur_eval; 6610 cur_eval = cur_eval->u.eval.prev_eval; 6611 DEBUG_EXECUTE_r( 6612 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n", 6613 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval));); 6614 if ( nochange_depth ) 6615 nochange_depth--; 6616 6617 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B, 6618 locinput); /* match B */ 6619 } 6620 6621 if (locinput < reginfo->till) { 6622 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, 6623 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", 6624 PL_colors[4], 6625 (long)(locinput - startpos), 6626 (long)(reginfo->till - startpos), 6627 PL_colors[5])); 6628 6629 sayNO_SILENT; /* Cannot match: too short. */ 6630 } 6631 sayYES; /* Success! */ 6632 6633 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ 6634 DEBUG_EXECUTE_r( 6635 PerlIO_printf(Perl_debug_log, 6636 "%*s %ssubpattern success...%s\n", 6637 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])); 6638 sayYES; /* Success! */ 6639 6640 #undef ST 6641 #define ST st->u.ifmatch 6642 6643 { 6644 char *newstart; 6645 6646 case SUSPEND: /* (?>A) */ 6647 ST.wanted = 1; 6648 newstart = locinput; 6649 goto do_ifmatch; 6650 6651 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */ 6652 ST.wanted = 0; 6653 goto ifmatch_trivial_fail_test; 6654 6655 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */ 6656 ST.wanted = 1; 6657 ifmatch_trivial_fail_test: 6658 if (scan->flags) { 6659 char * const s = HOPBACKc(locinput, scan->flags); 6660 if (!s) { 6661 /* trivial fail */ 6662 if (logical) { 6663 logical = 0; 6664 sw = 1 - cBOOL(ST.wanted); 6665 } 6666 else if (ST.wanted) 6667 sayNO; 6668 next = scan + ARG(scan); 6669 if (next == scan) 6670 next = NULL; 6671 break; 6672 } 6673 newstart = s; 6674 } 6675 else 6676 newstart = locinput; 6677 6678 do_ifmatch: 6679 ST.me = scan; 6680 ST.logical = logical; 6681 logical = 0; /* XXX: reset state of logical once it has been saved into ST */ 6682 6683 /* execute body of (?...A) */ 6684 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart); 6685 assert(0); /* NOTREACHED */ 6686 } 6687 6688 case IFMATCH_A_fail: /* body of (?...A) failed */ 6689 ST.wanted = !ST.wanted; 6690 /* FALL THROUGH */ 6691 6692 case IFMATCH_A: /* body of (?...A) succeeded */ 6693 if (ST.logical) { 6694 sw = cBOOL(ST.wanted); 6695 } 6696 else if (!ST.wanted) 6697 sayNO; 6698 6699 if (OP(ST.me) != SUSPEND) { 6700 /* restore old position except for (?>...) */ 6701 locinput = st->locinput; 6702 } 6703 scan = ST.me + ARG(ST.me); 6704 if (scan == ST.me) 6705 scan = NULL; 6706 continue; /* execute B */ 6707 6708 #undef ST 6709 6710 case LONGJMP: /* alternative with many branches compiles to 6711 * (BRANCHJ; EXACT ...; LONGJMP ) x N */ 6712 next = scan + ARG(scan); 6713 if (next == scan) 6714 next = NULL; 6715 break; 6716 6717 case COMMIT: /* (*COMMIT) */ 6718 reginfo->cutpoint = reginfo->strend; 6719 /* FALLTHROUGH */ 6720 6721 case PRUNE: /* (*PRUNE) */ 6722 if (!scan->flags) 6723 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 6724 PUSH_STATE_GOTO(COMMIT_next, next, locinput); 6725 assert(0); /* NOTREACHED */ 6726 6727 case COMMIT_next_fail: 6728 no_final = 1; 6729 /* FALLTHROUGH */ 6730 6731 case OPFAIL: /* (*FAIL) */ 6732 sayNO; 6733 assert(0); /* NOTREACHED */ 6734 6735 #define ST st->u.mark 6736 case MARKPOINT: /* (*MARK:foo) */ 6737 ST.prev_mark = mark_state; 6738 ST.mark_name = sv_commit = sv_yes_mark 6739 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 6740 mark_state = st; 6741 ST.mark_loc = locinput; 6742 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput); 6743 assert(0); /* NOTREACHED */ 6744 6745 case MARKPOINT_next: 6746 mark_state = ST.prev_mark; 6747 sayYES; 6748 assert(0); /* NOTREACHED */ 6749 6750 case MARKPOINT_next_fail: 6751 if (popmark && sv_eq(ST.mark_name,popmark)) 6752 { 6753 if (ST.mark_loc > startpoint) 6754 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); 6755 popmark = NULL; /* we found our mark */ 6756 sv_commit = ST.mark_name; 6757 6758 DEBUG_EXECUTE_r({ 6759 PerlIO_printf(Perl_debug_log, 6760 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n", 6761 REPORT_CODE_OFF+depth*2, "", 6762 PL_colors[4], SVfARG(sv_commit), PL_colors[5]); 6763 }); 6764 } 6765 mark_state = ST.prev_mark; 6766 sv_yes_mark = mark_state ? 6767 mark_state->u.mark.mark_name : NULL; 6768 sayNO; 6769 assert(0); /* NOTREACHED */ 6770 6771 case SKIP: /* (*SKIP) */ 6772 if (scan->flags) { 6773 /* (*SKIP) : if we fail we cut here*/ 6774 ST.mark_name = NULL; 6775 ST.mark_loc = locinput; 6776 PUSH_STATE_GOTO(SKIP_next,next, locinput); 6777 } else { 6778 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was, 6779 otherwise do nothing. Meaning we need to scan 6780 */ 6781 regmatch_state *cur = mark_state; 6782 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); 6783 6784 while (cur) { 6785 if ( sv_eq( cur->u.mark.mark_name, 6786 find ) ) 6787 { 6788 ST.mark_name = find; 6789 PUSH_STATE_GOTO( SKIP_next, next, locinput); 6790 } 6791 cur = cur->u.mark.prev_mark; 6792 } 6793 } 6794 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */ 6795 break; 6796 6797 case SKIP_next_fail: 6798 if (ST.mark_name) { 6799 /* (*CUT:NAME) - Set up to search for the name as we 6800 collapse the stack*/ 6801 popmark = ST.mark_name; 6802 } else { 6803 /* (*CUT) - No name, we cut here.*/ 6804 if (ST.mark_loc > startpoint) 6805 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); 6806 /* but we set sv_commit to latest mark_name if there 6807 is one so they can test to see how things lead to this 6808 cut */ 6809 if (mark_state) 6810 sv_commit=mark_state->u.mark.mark_name; 6811 } 6812 no_final = 1; 6813 sayNO; 6814 assert(0); /* NOTREACHED */ 6815 #undef ST 6816 6817 case LNBREAK: /* \R */ 6818 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) { 6819 locinput += n; 6820 } else 6821 sayNO; 6822 break; 6823 6824 default: 6825 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n", 6826 PTR2UV(scan), OP(scan)); 6827 Perl_croak(aTHX_ "regexp memory corruption"); 6828 6829 /* this is a point to jump to in order to increment 6830 * locinput by one character */ 6831 increment_locinput: 6832 assert(!NEXTCHR_IS_EOS); 6833 if (utf8_target) { 6834 locinput += PL_utf8skip[nextchr]; 6835 /* locinput is allowed to go 1 char off the end, but not 2+ */ 6836 if (locinput > reginfo->strend) 6837 sayNO; 6838 } 6839 else 6840 locinput++; 6841 break; 6842 6843 } /* end switch */ 6844 6845 /* switch break jumps here */ 6846 scan = next; /* prepare to execute the next op and ... */ 6847 continue; /* ... jump back to the top, reusing st */ 6848 assert(0); /* NOTREACHED */ 6849 6850 push_yes_state: 6851 /* push a state that backtracks on success */ 6852 st->u.yes.prev_yes_state = yes_state; 6853 yes_state = st; 6854 /* FALL THROUGH */ 6855 push_state: 6856 /* push a new regex state, then continue at scan */ 6857 { 6858 regmatch_state *newst; 6859 6860 DEBUG_STACK_r({ 6861 regmatch_state *cur = st; 6862 regmatch_state *curyes = yes_state; 6863 int curd = depth; 6864 regmatch_slab *slab = PL_regmatch_slab; 6865 for (;curd > -1;cur--,curd--) { 6866 if (cur < SLAB_FIRST(slab)) { 6867 slab = slab->prev; 6868 cur = SLAB_LAST(slab); 6869 } 6870 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n", 6871 REPORT_CODE_OFF + 2 + depth * 2,"", 6872 curd, PL_reg_name[cur->resume_state], 6873 (curyes == cur) ? "yes" : "" 6874 ); 6875 if (curyes == cur) 6876 curyes = cur->u.yes.prev_yes_state; 6877 } 6878 } else 6879 DEBUG_STATE_pp("push") 6880 ); 6881 depth++; 6882 st->locinput = locinput; 6883 newst = st+1; 6884 if (newst > SLAB_LAST(PL_regmatch_slab)) 6885 newst = S_push_slab(aTHX); 6886 PL_regmatch_state = newst; 6887 6888 locinput = pushinput; 6889 st = newst; 6890 continue; 6891 assert(0); /* NOTREACHED */ 6892 } 6893 } 6894 6895 /* 6896 * We get here only if there's trouble -- normally "case END" is 6897 * the terminating point. 6898 */ 6899 Perl_croak(aTHX_ "corrupted regexp pointers"); 6900 /*NOTREACHED*/ 6901 sayNO; 6902 6903 yes: 6904 if (yes_state) { 6905 /* we have successfully completed a subexpression, but we must now 6906 * pop to the state marked by yes_state and continue from there */ 6907 assert(st != yes_state); 6908 #ifdef DEBUGGING 6909 while (st != yes_state) { 6910 st--; 6911 if (st < SLAB_FIRST(PL_regmatch_slab)) { 6912 PL_regmatch_slab = PL_regmatch_slab->prev; 6913 st = SLAB_LAST(PL_regmatch_slab); 6914 } 6915 DEBUG_STATE_r({ 6916 if (no_final) { 6917 DEBUG_STATE_pp("pop (no final)"); 6918 } else { 6919 DEBUG_STATE_pp("pop (yes)"); 6920 } 6921 }); 6922 depth--; 6923 } 6924 #else 6925 while (yes_state < SLAB_FIRST(PL_regmatch_slab) 6926 || yes_state > SLAB_LAST(PL_regmatch_slab)) 6927 { 6928 /* not in this slab, pop slab */ 6929 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1); 6930 PL_regmatch_slab = PL_regmatch_slab->prev; 6931 st = SLAB_LAST(PL_regmatch_slab); 6932 } 6933 depth -= (st - yes_state); 6934 #endif 6935 st = yes_state; 6936 yes_state = st->u.yes.prev_yes_state; 6937 PL_regmatch_state = st; 6938 6939 if (no_final) 6940 locinput= st->locinput; 6941 state_num = st->resume_state + no_final; 6942 goto reenter_switch; 6943 } 6944 6945 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n", 6946 PL_colors[4], PL_colors[5])); 6947 6948 if (reginfo->info_aux_eval) { 6949 /* each successfully executed (?{...}) block does the equivalent of 6950 * local $^R = do {...} 6951 * When popping the save stack, all these locals would be undone; 6952 * bypass this by setting the outermost saved $^R to the latest 6953 * value */ 6954 /* I dont know if this is needed or works properly now. 6955 * see code related to PL_replgv elsewhere in this file. 6956 * Yves 6957 */ 6958 if (oreplsv != GvSV(PL_replgv)) 6959 sv_setsv(oreplsv, GvSV(PL_replgv)); 6960 } 6961 result = 1; 6962 goto final_exit; 6963 6964 no: 6965 DEBUG_EXECUTE_r( 6966 PerlIO_printf(Perl_debug_log, 6967 "%*s %sfailed...%s\n", 6968 REPORT_CODE_OFF+depth*2, "", 6969 PL_colors[4], PL_colors[5]) 6970 ); 6971 6972 no_silent: 6973 if (no_final) { 6974 if (yes_state) { 6975 goto yes; 6976 } else { 6977 goto final_exit; 6978 } 6979 } 6980 if (depth) { 6981 /* there's a previous state to backtrack to */ 6982 st--; 6983 if (st < SLAB_FIRST(PL_regmatch_slab)) { 6984 PL_regmatch_slab = PL_regmatch_slab->prev; 6985 st = SLAB_LAST(PL_regmatch_slab); 6986 } 6987 PL_regmatch_state = st; 6988 locinput= st->locinput; 6989 6990 DEBUG_STATE_pp("pop"); 6991 depth--; 6992 if (yes_state == st) 6993 yes_state = st->u.yes.prev_yes_state; 6994 6995 state_num = st->resume_state + 1; /* failure = success + 1 */ 6996 goto reenter_switch; 6997 } 6998 result = 0; 6999 7000 final_exit: 7001 if (rex->intflags & PREGf_VERBARG_SEEN) { 7002 SV *sv_err = get_sv("REGERROR", 1); 7003 SV *sv_mrk = get_sv("REGMARK", 1); 7004 if (result) { 7005 sv_commit = &PL_sv_no; 7006 if (!sv_yes_mark) 7007 sv_yes_mark = &PL_sv_yes; 7008 } else { 7009 if (!sv_commit) 7010 sv_commit = &PL_sv_yes; 7011 sv_yes_mark = &PL_sv_no; 7012 } 7013 sv_setsv(sv_err, sv_commit); 7014 sv_setsv(sv_mrk, sv_yes_mark); 7015 } 7016 7017 7018 if (last_pushed_cv) { 7019 dSP; 7020 POP_MULTICALL; 7021 PERL_UNUSED_VAR(SP); 7022 } 7023 7024 assert(!result || locinput - reginfo->strbeg >= 0); 7025 return result ? locinput - reginfo->strbeg : -1; 7026 } 7027 7028 /* 7029 - regrepeat - repeatedly match something simple, report how many 7030 * 7031 * What 'simple' means is a node which can be the operand of a quantifier like 7032 * '+', or {1,3} 7033 * 7034 * startposp - pointer a pointer to the start position. This is updated 7035 * to point to the byte following the highest successful 7036 * match. 7037 * p - the regnode to be repeatedly matched against. 7038 * reginfo - struct holding match state, such as strend 7039 * max - maximum number of things to match. 7040 * depth - (for debugging) backtracking depth. 7041 */ 7042 STATIC I32 7043 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p, 7044 regmatch_info *const reginfo, I32 max, int depth) 7045 { 7046 dVAR; 7047 char *scan; /* Pointer to current position in target string */ 7048 I32 c; 7049 char *loceol = reginfo->strend; /* local version */ 7050 I32 hardcount = 0; /* How many matches so far */ 7051 bool utf8_target = reginfo->is_utf8_target; 7052 int to_complement = 0; /* Invert the result? */ 7053 UV utf8_flags; 7054 _char_class_number classnum; 7055 #ifndef DEBUGGING 7056 PERL_UNUSED_ARG(depth); 7057 #endif 7058 7059 PERL_ARGS_ASSERT_REGREPEAT; 7060 7061 scan = *startposp; 7062 if (max == REG_INFTY) 7063 max = I32_MAX; 7064 else if (! utf8_target && loceol - scan > max) 7065 loceol = scan + max; 7066 7067 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down 7068 * to the maximum of how far we should go in it (leaving it set to the real 7069 * end, if the maximum permissible would take us beyond that). This allows 7070 * us to make the loop exit condition that we haven't gone past <loceol> to 7071 * also mean that we haven't exceeded the max permissible count, saving a 7072 * test each time through the loop. But it assumes that the OP matches a 7073 * single byte, which is true for most of the OPs below when applied to a 7074 * non-UTF-8 target. Those relatively few OPs that don't have this 7075 * characteristic will have to compensate. 7076 * 7077 * There is no adjustment for UTF-8 targets, as the number of bytes per 7078 * character varies. OPs will have to test both that the count is less 7079 * than the max permissible (using <hardcount> to keep track), and that we 7080 * are still within the bounds of the string (using <loceol>. A few OPs 7081 * match a single byte no matter what the encoding. They can omit the max 7082 * test if, for the UTF-8 case, they do the adjustment that was skipped 7083 * above. 7084 * 7085 * Thus, the code above sets things up for the common case; and exceptional 7086 * cases need extra work; the common case is to make sure <scan> doesn't 7087 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the 7088 * count doesn't exceed the maximum permissible */ 7089 7090 switch (OP(p)) { 7091 case REG_ANY: 7092 if (utf8_target) { 7093 while (scan < loceol && hardcount < max && *scan != '\n') { 7094 scan += UTF8SKIP(scan); 7095 hardcount++; 7096 } 7097 } else { 7098 while (scan < loceol && *scan != '\n') 7099 scan++; 7100 } 7101 break; 7102 case SANY: 7103 if (utf8_target) { 7104 while (scan < loceol && hardcount < max) { 7105 scan += UTF8SKIP(scan); 7106 hardcount++; 7107 } 7108 } 7109 else 7110 scan = loceol; 7111 break; 7112 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */ 7113 if (utf8_target && loceol - scan > max) { 7114 7115 /* <loceol> hadn't been adjusted in the UTF-8 case */ 7116 scan += max; 7117 } 7118 else { 7119 scan = loceol; 7120 } 7121 break; 7122 case EXACT: 7123 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); 7124 7125 c = (U8)*STRING(p); 7126 7127 /* Can use a simple loop if the pattern char to match on is invariant 7128 * under UTF-8, or both target and pattern aren't UTF-8. Note that we 7129 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's 7130 * true iff it doesn't matter if the argument is in UTF-8 or not */ 7131 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) { 7132 if (utf8_target && loceol - scan > max) { 7133 /* We didn't adjust <loceol> because is UTF-8, but ok to do so, 7134 * since here, to match at all, 1 char == 1 byte */ 7135 loceol = scan + max; 7136 } 7137 while (scan < loceol && UCHARAT(scan) == c) { 7138 scan++; 7139 } 7140 } 7141 else if (reginfo->is_utf8_pat) { 7142 if (utf8_target) { 7143 STRLEN scan_char_len; 7144 7145 /* When both target and pattern are UTF-8, we have to do 7146 * string EQ */ 7147 while (hardcount < max 7148 && scan < loceol 7149 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p) 7150 && memEQ(scan, STRING(p), scan_char_len)) 7151 { 7152 scan += scan_char_len; 7153 hardcount++; 7154 } 7155 } 7156 else if (! UTF8_IS_ABOVE_LATIN1(c)) { 7157 7158 /* Target isn't utf8; convert the character in the UTF-8 7159 * pattern to non-UTF8, and do a simple loop */ 7160 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1)); 7161 while (scan < loceol && UCHARAT(scan) == c) { 7162 scan++; 7163 } 7164 } /* else pattern char is above Latin1, can't possibly match the 7165 non-UTF-8 target */ 7166 } 7167 else { 7168 7169 /* Here, the string must be utf8; pattern isn't, and <c> is 7170 * different in utf8 than not, so can't compare them directly. 7171 * Outside the loop, find the two utf8 bytes that represent c, and 7172 * then look for those in sequence in the utf8 string */ 7173 U8 high = UTF8_TWO_BYTE_HI(c); 7174 U8 low = UTF8_TWO_BYTE_LO(c); 7175 7176 while (hardcount < max 7177 && scan + 1 < loceol 7178 && UCHARAT(scan) == high 7179 && UCHARAT(scan + 1) == low) 7180 { 7181 scan += 2; 7182 hardcount++; 7183 } 7184 } 7185 break; 7186 7187 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ 7188 assert(! reginfo->is_utf8_pat); 7189 /* FALL THROUGH */ 7190 case EXACTFA: 7191 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; 7192 goto do_exactf; 7193 7194 case EXACTFL: 7195 utf8_flags = FOLDEQ_LOCALE; 7196 goto do_exactf; 7197 7198 case EXACTF: /* This node only generated for non-utf8 patterns */ 7199 assert(! reginfo->is_utf8_pat); 7200 utf8_flags = 0; 7201 goto do_exactf; 7202 7203 case EXACTFU_SS: 7204 case EXACTFU: 7205 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; 7206 7207 do_exactf: { 7208 int c1, c2; 7209 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1]; 7210 7211 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); 7212 7213 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8, 7214 reginfo)) 7215 { 7216 if (c1 == CHRTEST_VOID) { 7217 /* Use full Unicode fold matching */ 7218 char *tmpeol = reginfo->strend; 7219 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1; 7220 while (hardcount < max 7221 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target, 7222 STRING(p), NULL, pat_len, 7223 reginfo->is_utf8_pat, utf8_flags)) 7224 { 7225 scan = tmpeol; 7226 tmpeol = reginfo->strend; 7227 hardcount++; 7228 } 7229 } 7230 else if (utf8_target) { 7231 if (c1 == c2) { 7232 while (scan < loceol 7233 && hardcount < max 7234 && memEQ(scan, c1_utf8, UTF8SKIP(scan))) 7235 { 7236 scan += UTF8SKIP(scan); 7237 hardcount++; 7238 } 7239 } 7240 else { 7241 while (scan < loceol 7242 && hardcount < max 7243 && (memEQ(scan, c1_utf8, UTF8SKIP(scan)) 7244 || memEQ(scan, c2_utf8, UTF8SKIP(scan)))) 7245 { 7246 scan += UTF8SKIP(scan); 7247 hardcount++; 7248 } 7249 } 7250 } 7251 else if (c1 == c2) { 7252 while (scan < loceol && UCHARAT(scan) == c1) { 7253 scan++; 7254 } 7255 } 7256 else { 7257 while (scan < loceol && 7258 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2)) 7259 { 7260 scan++; 7261 } 7262 } 7263 } 7264 break; 7265 } 7266 case ANYOF: 7267 if (utf8_target) { 7268 while (hardcount < max 7269 && scan < loceol 7270 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target)) 7271 { 7272 scan += UTF8SKIP(scan); 7273 hardcount++; 7274 } 7275 } else { 7276 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan)) 7277 scan++; 7278 } 7279 break; 7280 7281 /* The argument (FLAGS) to all the POSIX node types is the class number */ 7282 7283 case NPOSIXL: 7284 to_complement = 1; 7285 /* FALLTHROUGH */ 7286 7287 case POSIXL: 7288 if (! utf8_target) { 7289 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p), 7290 *scan))) 7291 { 7292 scan++; 7293 } 7294 } else { 7295 while (hardcount < max && scan < loceol 7296 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p), 7297 (U8 *) scan))) 7298 { 7299 scan += UTF8SKIP(scan); 7300 hardcount++; 7301 } 7302 } 7303 break; 7304 7305 case POSIXD: 7306 if (utf8_target) { 7307 goto utf8_posix; 7308 } 7309 /* FALLTHROUGH */ 7310 7311 case POSIXA: 7312 if (utf8_target && loceol - scan > max) { 7313 7314 /* We didn't adjust <loceol> at the beginning of this routine 7315 * because is UTF-8, but it is actually ok to do so, since here, to 7316 * match, 1 char == 1 byte. */ 7317 loceol = scan + max; 7318 } 7319 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) { 7320 scan++; 7321 } 7322 break; 7323 7324 case NPOSIXD: 7325 if (utf8_target) { 7326 to_complement = 1; 7327 goto utf8_posix; 7328 } 7329 /* FALL THROUGH */ 7330 7331 case NPOSIXA: 7332 if (! utf8_target) { 7333 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) { 7334 scan++; 7335 } 7336 } 7337 else { 7338 7339 /* The complement of something that matches only ASCII matches all 7340 * UTF-8 variant code points, plus everything in ASCII that isn't 7341 * in the class. */ 7342 while (hardcount < max && scan < loceol 7343 && (! UTF8_IS_INVARIANT(*scan) 7344 || ! _generic_isCC_A((U8) *scan, FLAGS(p)))) 7345 { 7346 scan += UTF8SKIP(scan); 7347 hardcount++; 7348 } 7349 } 7350 break; 7351 7352 case NPOSIXU: 7353 to_complement = 1; 7354 /* FALLTHROUGH */ 7355 7356 case POSIXU: 7357 if (! utf8_target) { 7358 while (scan < loceol && to_complement 7359 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p)))) 7360 { 7361 scan++; 7362 } 7363 } 7364 else { 7365 utf8_posix: 7366 classnum = (_char_class_number) FLAGS(p); 7367 if (classnum < _FIRST_NON_SWASH_CC) { 7368 7369 /* Here, a swash is needed for above-Latin1 code points. 7370 * Process as many Latin1 code points using the built-in rules. 7371 * Go to another loop to finish processing upon encountering 7372 * the first Latin1 code point. We could do that in this loop 7373 * as well, but the other way saves having to test if the swash 7374 * has been loaded every time through the loop: extra space to 7375 * save a test. */ 7376 while (hardcount < max && scan < loceol) { 7377 if (UTF8_IS_INVARIANT(*scan)) { 7378 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan, 7379 classnum)))) 7380 { 7381 break; 7382 } 7383 scan++; 7384 } 7385 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) { 7386 if (! (to_complement 7387 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan, 7388 *(scan + 1)), 7389 classnum)))) 7390 { 7391 break; 7392 } 7393 scan += 2; 7394 } 7395 else { 7396 goto found_above_latin1; 7397 } 7398 7399 hardcount++; 7400 } 7401 } 7402 else { 7403 /* For these character classes, the knowledge of how to handle 7404 * every code point is compiled in to Perl via a macro. This 7405 * code is written for making the loops as tight as possible. 7406 * It could be refactored to save space instead */ 7407 switch (classnum) { 7408 case _CC_ENUM_SPACE: /* XXX would require separate code 7409 if we revert the change of \v 7410 matching this */ 7411 /* FALL THROUGH */ 7412 case _CC_ENUM_PSXSPC: 7413 while (hardcount < max 7414 && scan < loceol 7415 && (to_complement ^ cBOOL(isSPACE_utf8(scan)))) 7416 { 7417 scan += UTF8SKIP(scan); 7418 hardcount++; 7419 } 7420 break; 7421 case _CC_ENUM_BLANK: 7422 while (hardcount < max 7423 && scan < loceol 7424 && (to_complement ^ cBOOL(isBLANK_utf8(scan)))) 7425 { 7426 scan += UTF8SKIP(scan); 7427 hardcount++; 7428 } 7429 break; 7430 case _CC_ENUM_XDIGIT: 7431 while (hardcount < max 7432 && scan < loceol 7433 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan)))) 7434 { 7435 scan += UTF8SKIP(scan); 7436 hardcount++; 7437 } 7438 break; 7439 case _CC_ENUM_VERTSPACE: 7440 while (hardcount < max 7441 && scan < loceol 7442 && (to_complement ^ cBOOL(isVERTWS_utf8(scan)))) 7443 { 7444 scan += UTF8SKIP(scan); 7445 hardcount++; 7446 } 7447 break; 7448 case _CC_ENUM_CNTRL: 7449 while (hardcount < max 7450 && scan < loceol 7451 && (to_complement ^ cBOOL(isCNTRL_utf8(scan)))) 7452 { 7453 scan += UTF8SKIP(scan); 7454 hardcount++; 7455 } 7456 break; 7457 default: 7458 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum); 7459 } 7460 } 7461 } 7462 break; 7463 7464 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */ 7465 7466 /* Load the swash if not already present */ 7467 if (! PL_utf8_swash_ptrs[classnum]) { 7468 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 7469 PL_utf8_swash_ptrs[classnum] = _core_swash_init( 7470 "utf8", 7471 "", 7472 &PL_sv_undef, 1, 0, 7473 PL_XPosix_ptrs[classnum], &flags); 7474 } 7475 7476 while (hardcount < max && scan < loceol 7477 && to_complement ^ cBOOL(_generic_utf8( 7478 classnum, 7479 scan, 7480 swash_fetch(PL_utf8_swash_ptrs[classnum], 7481 (U8 *) scan, 7482 TRUE)))) 7483 { 7484 scan += UTF8SKIP(scan); 7485 hardcount++; 7486 } 7487 break; 7488 7489 case LNBREAK: 7490 if (utf8_target) { 7491 while (hardcount < max && scan < loceol && 7492 (c=is_LNBREAK_utf8_safe(scan, loceol))) { 7493 scan += c; 7494 hardcount++; 7495 } 7496 } else { 7497 /* LNBREAK can match one or two latin chars, which is ok, but we 7498 * have to use hardcount in this situation, and throw away the 7499 * adjustment to <loceol> done before the switch statement */ 7500 loceol = reginfo->strend; 7501 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) { 7502 scan+=c; 7503 hardcount++; 7504 } 7505 } 7506 break; 7507 7508 case BOUND: 7509 case BOUNDA: 7510 case BOUNDL: 7511 case BOUNDU: 7512 case EOS: 7513 case GPOS: 7514 case KEEPS: 7515 case NBOUND: 7516 case NBOUNDA: 7517 case NBOUNDL: 7518 case NBOUNDU: 7519 case OPFAIL: 7520 case SBOL: 7521 case SEOL: 7522 /* These are all 0 width, so match right here or not at all. */ 7523 break; 7524 7525 default: 7526 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]); 7527 assert(0); /* NOTREACHED */ 7528 7529 } 7530 7531 if (hardcount) 7532 c = hardcount; 7533 else 7534 c = scan - *startposp; 7535 *startposp = scan; 7536 7537 DEBUG_r({ 7538 GET_RE_DEBUG_FLAGS_DECL; 7539 DEBUG_EXECUTE_r({ 7540 SV * const prop = sv_newmortal(); 7541 regprop(prog, prop, p, reginfo); 7542 PerlIO_printf(Perl_debug_log, 7543 "%*s %s can match %"IVdf" times out of %"IVdf"...\n", 7544 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max); 7545 }); 7546 }); 7547 7548 return(c); 7549 } 7550 7551 7552 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) 7553 /* 7554 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to 7555 create a copy so that changes the caller makes won't change the shared one. 7556 If <altsvp> is non-null, will return NULL in it, for back-compat. 7557 */ 7558 SV * 7559 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp) 7560 { 7561 PERL_ARGS_ASSERT_REGCLASS_SWASH; 7562 7563 if (altsvp) { 7564 *altsvp = NULL; 7565 } 7566 7567 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL)); 7568 } 7569 7570 SV * 7571 Perl__get_regclass_nonbitmap_data(pTHX_ const regexp *prog, 7572 const regnode* node, 7573 bool doinit, 7574 SV** listsvp, 7575 SV** only_utf8_locale_ptr) 7576 { 7577 /* For internal core use only. 7578 * Returns the swash for the input 'node' in the regex 'prog'. 7579 * If <doinit> is 'true', will attempt to create the swash if not already 7580 * done. 7581 * If <listsvp> is non-null, will return the printable contents of the 7582 * swash. This can be used to get debugging information even before the 7583 * swash exists, by calling this function with 'doinit' set to false, in 7584 * which case the components that will be used to eventually create the 7585 * swash are returned (in a printable form). 7586 * Tied intimately to how regcomp.c sets up the data structure */ 7587 7588 dVAR; 7589 SV *sw = NULL; 7590 SV *si = NULL; /* Input swash initialization string */ 7591 SV* invlist = NULL; 7592 7593 RXi_GET_DECL(prog,progi); 7594 const struct reg_data * const data = prog ? progi->data : NULL; 7595 7596 PERL_ARGS_ASSERT__GET_REGCLASS_NONBITMAP_DATA; 7597 7598 assert(ANYOF_FLAGS(node) 7599 & (ANYOF_UTF8|ANYOF_NONBITMAP_NON_UTF8|ANYOF_LOC_FOLD)); 7600 7601 if (data && data->count) { 7602 const U32 n = ARG(node); 7603 7604 if (data->what[n] == 's') { 7605 SV * const rv = MUTABLE_SV(data->data[n]); 7606 AV * const av = MUTABLE_AV(SvRV(rv)); 7607 SV **const ary = AvARRAY(av); 7608 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; 7609 7610 si = *ary; /* ary[0] = the string to initialize the swash with */ 7611 7612 /* Elements 3 and 4 are either both present or both absent. [3] is 7613 * any inversion list generated at compile time; [4] indicates if 7614 * that inversion list has any user-defined properties in it. */ 7615 if (av_tindex(av) >= 2) { 7616 if (only_utf8_locale_ptr 7617 && ary[2] 7618 && ary[2] != &PL_sv_undef) 7619 { 7620 *only_utf8_locale_ptr = ary[2]; 7621 } 7622 else { 7623 *only_utf8_locale_ptr = NULL; 7624 } 7625 7626 if (av_tindex(av) >= 3) { 7627 invlist = ary[3]; 7628 if (SvUV(ary[4])) { 7629 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY; 7630 } 7631 } 7632 else { 7633 invlist = NULL; 7634 } 7635 } 7636 7637 /* Element [1] is reserved for the set-up swash. If already there, 7638 * return it; if not, create it and store it there */ 7639 if (ary[1] && SvROK(ary[1])) { 7640 sw = ary[1]; 7641 } 7642 else if (doinit && ((si && si != &PL_sv_undef) 7643 || (invlist && invlist != &PL_sv_undef))) { 7644 7645 sw = _core_swash_init("utf8", /* the utf8 package */ 7646 "", /* nameless */ 7647 si, 7648 1, /* binary */ 7649 0, /* not from tr/// */ 7650 invlist, 7651 &swash_init_flags); 7652 (void)av_store(av, 1, sw); 7653 } 7654 } 7655 } 7656 7657 /* If requested, return a printable version of what this swash matches */ 7658 if (listsvp) { 7659 SV* matches_string = newSVpvn("", 0); 7660 7661 /* The swash should be used, if possible, to get the data, as it 7662 * contains the resolved data. But this function can be called at 7663 * compile-time, before everything gets resolved, in which case we 7664 * return the currently best available information, which is the string 7665 * that will eventually be used to do that resolving, 'si' */ 7666 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL) 7667 && (si && si != &PL_sv_undef)) 7668 { 7669 sv_catsv(matches_string, si); 7670 } 7671 7672 /* Add the inversion list to whatever we have. This may have come from 7673 * the swash, or from an input parameter */ 7674 if (invlist) { 7675 sv_catsv(matches_string, _invlist_contents(invlist)); 7676 } 7677 *listsvp = matches_string; 7678 } 7679 7680 return sw; 7681 } 7682 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */ 7683 7684 /* 7685 - reginclass - determine if a character falls into a character class 7686 7687 n is the ANYOF regnode 7688 p is the target string 7689 p_end points to one byte beyond the end of the target string 7690 utf8_target tells whether p is in UTF-8. 7691 7692 Returns true if matched; false otherwise. 7693 7694 Note that this can be a synthetic start class, a combination of various 7695 nodes, so things you think might be mutually exclusive, such as locale, 7696 aren't. It can match both locale and non-locale 7697 7698 */ 7699 7700 STATIC bool 7701 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target) 7702 { 7703 dVAR; 7704 const char flags = ANYOF_FLAGS(n); 7705 bool match = FALSE; 7706 UV c = *p; 7707 7708 PERL_ARGS_ASSERT_REGINCLASS; 7709 7710 /* If c is not already the code point, get it. Note that 7711 * UTF8_IS_INVARIANT() works even if not in UTF-8 */ 7712 if (! UTF8_IS_INVARIANT(c) && utf8_target) { 7713 STRLEN c_len = 0; 7714 c = utf8n_to_uvchr(p, p_end - p, &c_len, 7715 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) 7716 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY); 7717 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for 7718 * UTF8_ALLOW_FFFF */ 7719 if (c_len == (STRLEN)-1) 7720 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); 7721 } 7722 7723 /* If this character is potentially in the bitmap, check it */ 7724 if (c < 256) { 7725 if (ANYOF_BITMAP_TEST(n, c)) 7726 match = TRUE; 7727 else if (flags & ANYOF_NON_UTF8_NON_ASCII_ALL 7728 && ! utf8_target 7729 && ! isASCII(c)) 7730 { 7731 match = TRUE; 7732 } 7733 else if (flags & ANYOF_LOCALE_FLAGS) { 7734 if (flags & ANYOF_LOC_FOLD) { 7735 if (ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) { 7736 match = TRUE; 7737 } 7738 } 7739 if (! match && ANYOF_POSIXL_TEST_ANY_SET(n)) { 7740 7741 /* The data structure is arranged so bits 0, 2, 4, ... are set 7742 * if the class includes the Posix character class given by 7743 * bit/2; and 1, 3, 5, ... are set if the class includes the 7744 * complemented Posix class given by int(bit/2). So we loop 7745 * through the bits, each time changing whether we complement 7746 * the result or not. Suppose for the sake of illustration 7747 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0 7748 * is set, it means there is a match for this ANYOF node if the 7749 * character is in the class given by the expression (0 / 2 = 0 7750 * = \w). If it is in that class, isFOO_lc() will return 1, 7751 * and since 'to_complement' is 0, the result will stay TRUE, 7752 * and we exit the loop. Suppose instead that bit 0 is 0, but 7753 * bit 1 is 1. That means there is a match if the character 7754 * matches \W. We won't bother to call isFOO_lc() on bit 0, 7755 * but will on bit 1. On the second iteration 'to_complement' 7756 * will be 1, so the exclusive or will reverse things, so we 7757 * are testing for \W. On the third iteration, 'to_complement' 7758 * will be 0, and we would be testing for \s; the fourth 7759 * iteration would test for \S, etc. 7760 * 7761 * Note that this code assumes that all the classes are closed 7762 * under folding. For example, if a character matches \w, then 7763 * its fold does too; and vice versa. This should be true for 7764 * any well-behaved locale for all the currently defined Posix 7765 * classes, except for :lower: and :upper:, which are handled 7766 * by the pseudo-class :cased: which matches if either of the 7767 * other two does. To get rid of this assumption, an outer 7768 * loop could be used below to iterate over both the source 7769 * character, and its fold (if different) */ 7770 7771 int count = 0; 7772 int to_complement = 0; 7773 7774 while (count < ANYOF_MAX) { 7775 if (ANYOF_POSIXL_TEST(n, count) 7776 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c))) 7777 { 7778 match = TRUE; 7779 break; 7780 } 7781 count++; 7782 to_complement ^= 1; 7783 } 7784 } 7785 } 7786 } 7787 7788 7789 /* If the bitmap didn't (or couldn't) match, and something outside the 7790 * bitmap could match, try that. */ 7791 if (!match) { 7792 if (c >= 256 && (flags & ANYOF_ABOVE_LATIN1_ALL)) { 7793 match = TRUE; /* Everything above 255 matches */ 7794 } 7795 else if ((flags & ANYOF_NONBITMAP_NON_UTF8) 7796 || (utf8_target && (flags & ANYOF_UTF8)) 7797 || ((flags & ANYOF_LOC_FOLD) 7798 && IN_UTF8_CTYPE_LOCALE 7799 && ARG(n) != ANYOF_NONBITMAP_EMPTY)) 7800 { 7801 SV* only_utf8_locale = NULL; 7802 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0, 7803 &only_utf8_locale); 7804 if (sw) { 7805 U8 * utf8_p; 7806 if (utf8_target) { 7807 utf8_p = (U8 *) p; 7808 } else { /* Convert to utf8 */ 7809 STRLEN len = 1; 7810 utf8_p = bytes_to_utf8(p, &len); 7811 } 7812 7813 if (swash_fetch(sw, utf8_p, TRUE)) { 7814 match = TRUE; 7815 } 7816 7817 /* If we allocated a string above, free it */ 7818 if (! utf8_target) Safefree(utf8_p); 7819 } 7820 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) { 7821 match = _invlist_contains_cp(only_utf8_locale, c); 7822 } 7823 } 7824 7825 if (UNICODE_IS_SUPER(c) 7826 && (flags & ANYOF_WARN_SUPER) 7827 && ckWARN_d(WARN_NON_UNICODE)) 7828 { 7829 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), 7830 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c); 7831 } 7832 } 7833 7834 #if ANYOF_INVERT != 1 7835 /* Depending on compiler optimization cBOOL takes time, so if don't have to 7836 * use it, don't */ 7837 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below, 7838 #endif 7839 7840 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */ 7841 return (flags & ANYOF_INVERT) ^ match; 7842 } 7843 7844 STATIC U8 * 7845 S_reghop3(U8 *s, SSize_t off, const U8* lim) 7846 { 7847 /* return the position 'off' UTF-8 characters away from 's', forward if 7848 * 'off' >= 0, backwards if negative. But don't go outside of position 7849 * 'lim', which better be < s if off < 0 */ 7850 7851 dVAR; 7852 7853 PERL_ARGS_ASSERT_REGHOP3; 7854 7855 if (off >= 0) { 7856 while (off-- && s < lim) { 7857 /* XXX could check well-formedness here */ 7858 s += UTF8SKIP(s); 7859 } 7860 } 7861 else { 7862 while (off++ && s > lim) { 7863 s--; 7864 if (UTF8_IS_CONTINUED(*s)) { 7865 while (s > lim && UTF8_IS_CONTINUATION(*s)) 7866 s--; 7867 } 7868 /* XXX could check well-formedness here */ 7869 } 7870 } 7871 return s; 7872 } 7873 7874 STATIC U8 * 7875 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim) 7876 { 7877 dVAR; 7878 7879 PERL_ARGS_ASSERT_REGHOP4; 7880 7881 if (off >= 0) { 7882 while (off-- && s < rlim) { 7883 /* XXX could check well-formedness here */ 7884 s += UTF8SKIP(s); 7885 } 7886 } 7887 else { 7888 while (off++ && s > llim) { 7889 s--; 7890 if (UTF8_IS_CONTINUED(*s)) { 7891 while (s > llim && UTF8_IS_CONTINUATION(*s)) 7892 s--; 7893 } 7894 /* XXX could check well-formedness here */ 7895 } 7896 } 7897 return s; 7898 } 7899 7900 /* like reghop3, but returns NULL on overrun, rather than returning last 7901 * char pos */ 7902 7903 STATIC U8 * 7904 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim) 7905 { 7906 dVAR; 7907 7908 PERL_ARGS_ASSERT_REGHOPMAYBE3; 7909 7910 if (off >= 0) { 7911 while (off-- && s < lim) { 7912 /* XXX could check well-formedness here */ 7913 s += UTF8SKIP(s); 7914 } 7915 if (off >= 0) 7916 return NULL; 7917 } 7918 else { 7919 while (off++ && s > lim) { 7920 s--; 7921 if (UTF8_IS_CONTINUED(*s)) { 7922 while (s > lim && UTF8_IS_CONTINUATION(*s)) 7923 s--; 7924 } 7925 /* XXX could check well-formedness here */ 7926 } 7927 if (off <= 0) 7928 return NULL; 7929 } 7930 return s; 7931 } 7932 7933 7934 /* when executing a regex that may have (?{}), extra stuff needs setting 7935 up that will be visible to the called code, even before the current 7936 match has finished. In particular: 7937 7938 * $_ is localised to the SV currently being matched; 7939 * pos($_) is created if necessary, ready to be updated on each call-out 7940 to code; 7941 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm 7942 isn't set until the current pattern is successfully finished), so that 7943 $1 etc of the match-so-far can be seen; 7944 * save the old values of subbeg etc of the current regex, and set then 7945 to the current string (again, this is normally only done at the end 7946 of execution) 7947 */ 7948 7949 static void 7950 S_setup_eval_state(pTHX_ regmatch_info *const reginfo) 7951 { 7952 MAGIC *mg; 7953 regexp *const rex = ReANY(reginfo->prog); 7954 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval; 7955 7956 eval_state->rex = rex; 7957 7958 if (reginfo->sv) { 7959 /* Make $_ available to executed code. */ 7960 if (reginfo->sv != DEFSV) { 7961 SAVE_DEFSV; 7962 DEFSV_set(reginfo->sv); 7963 } 7964 7965 if (!(mg = mg_find_mglob(reginfo->sv))) { 7966 /* prepare for quick setting of pos */ 7967 mg = sv_magicext_mglob(reginfo->sv); 7968 mg->mg_len = -1; 7969 } 7970 eval_state->pos_magic = mg; 7971 eval_state->pos = mg->mg_len; 7972 eval_state->pos_flags = mg->mg_flags; 7973 } 7974 else 7975 eval_state->pos_magic = NULL; 7976 7977 if (!PL_reg_curpm) { 7978 /* PL_reg_curpm is a fake PMOP that we can attach the current 7979 * regex to and point PL_curpm at, so that $1 et al are visible 7980 * within a /(?{})/. It's just allocated once per interpreter the 7981 * first time its needed */ 7982 Newxz(PL_reg_curpm, 1, PMOP); 7983 #ifdef USE_ITHREADS 7984 { 7985 SV* const repointer = &PL_sv_undef; 7986 /* this regexp is also owned by the new PL_reg_curpm, which 7987 will try to free it. */ 7988 av_push(PL_regex_padav, repointer); 7989 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav); 7990 PL_regex_pad = AvARRAY(PL_regex_padav); 7991 } 7992 #endif 7993 } 7994 SET_reg_curpm(reginfo->prog); 7995 eval_state->curpm = PL_curpm; 7996 PL_curpm = PL_reg_curpm; 7997 if (RXp_MATCH_COPIED(rex)) { 7998 /* Here is a serious problem: we cannot rewrite subbeg, 7999 since it may be needed if this match fails. Thus 8000 $` inside (?{}) could fail... */ 8001 eval_state->subbeg = rex->subbeg; 8002 eval_state->sublen = rex->sublen; 8003 eval_state->suboffset = rex->suboffset; 8004 eval_state->subcoffset = rex->subcoffset; 8005 #ifdef PERL_ANY_COW 8006 eval_state->saved_copy = rex->saved_copy; 8007 #endif 8008 RXp_MATCH_COPIED_off(rex); 8009 } 8010 else 8011 eval_state->subbeg = NULL; 8012 rex->subbeg = (char *)reginfo->strbeg; 8013 rex->suboffset = 0; 8014 rex->subcoffset = 0; 8015 rex->sublen = reginfo->strend - reginfo->strbeg; 8016 } 8017 8018 8019 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */ 8020 8021 static void 8022 S_cleanup_regmatch_info_aux(pTHX_ void *arg) 8023 { 8024 dVAR; 8025 regmatch_info_aux *aux = (regmatch_info_aux *) arg; 8026 regmatch_info_aux_eval *eval_state = aux->info_aux_eval; 8027 regmatch_slab *s; 8028 8029 Safefree(aux->poscache); 8030 8031 if (eval_state) { 8032 8033 /* undo the effects of S_setup_eval_state() */ 8034 8035 if (eval_state->subbeg) { 8036 regexp * const rex = eval_state->rex; 8037 rex->subbeg = eval_state->subbeg; 8038 rex->sublen = eval_state->sublen; 8039 rex->suboffset = eval_state->suboffset; 8040 rex->subcoffset = eval_state->subcoffset; 8041 #ifdef PERL_ANY_COW 8042 rex->saved_copy = eval_state->saved_copy; 8043 #endif 8044 RXp_MATCH_COPIED_on(rex); 8045 } 8046 if (eval_state->pos_magic) 8047 { 8048 eval_state->pos_magic->mg_len = eval_state->pos; 8049 eval_state->pos_magic->mg_flags = 8050 (eval_state->pos_magic->mg_flags & ~MGf_BYTES) 8051 | (eval_state->pos_flags & MGf_BYTES); 8052 } 8053 8054 PL_curpm = eval_state->curpm; 8055 } 8056 8057 PL_regmatch_state = aux->old_regmatch_state; 8058 PL_regmatch_slab = aux->old_regmatch_slab; 8059 8060 /* free all slabs above current one - this must be the last action 8061 * of this function, as aux and eval_state are allocated within 8062 * slabs and may be freed here */ 8063 8064 s = PL_regmatch_slab->next; 8065 if (s) { 8066 PL_regmatch_slab->next = NULL; 8067 while (s) { 8068 regmatch_slab * const osl = s; 8069 s = s->next; 8070 Safefree(osl); 8071 } 8072 } 8073 } 8074 8075 8076 STATIC void 8077 S_to_utf8_substr(pTHX_ regexp *prog) 8078 { 8079 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile 8080 * on the converted value */ 8081 8082 int i = 1; 8083 8084 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR; 8085 8086 do { 8087 if (prog->substrs->data[i].substr 8088 && !prog->substrs->data[i].utf8_substr) { 8089 SV* const sv = newSVsv(prog->substrs->data[i].substr); 8090 prog->substrs->data[i].utf8_substr = sv; 8091 sv_utf8_upgrade(sv); 8092 if (SvVALID(prog->substrs->data[i].substr)) { 8093 if (SvTAIL(prog->substrs->data[i].substr)) { 8094 /* Trim the trailing \n that fbm_compile added last 8095 time. */ 8096 SvCUR_set(sv, SvCUR(sv) - 1); 8097 /* Whilst this makes the SV technically "invalid" (as its 8098 buffer is no longer followed by "\0") when fbm_compile() 8099 adds the "\n" back, a "\0" is restored. */ 8100 fbm_compile(sv, FBMcf_TAIL); 8101 } else 8102 fbm_compile(sv, 0); 8103 } 8104 if (prog->substrs->data[i].substr == prog->check_substr) 8105 prog->check_utf8 = sv; 8106 } 8107 } while (i--); 8108 } 8109 8110 STATIC bool 8111 S_to_byte_substr(pTHX_ regexp *prog) 8112 { 8113 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile 8114 * on the converted value; returns FALSE if can't be converted. */ 8115 8116 dVAR; 8117 int i = 1; 8118 8119 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR; 8120 8121 do { 8122 if (prog->substrs->data[i].utf8_substr 8123 && !prog->substrs->data[i].substr) { 8124 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr); 8125 if (! sv_utf8_downgrade(sv, TRUE)) { 8126 return FALSE; 8127 } 8128 if (SvVALID(prog->substrs->data[i].utf8_substr)) { 8129 if (SvTAIL(prog->substrs->data[i].utf8_substr)) { 8130 /* Trim the trailing \n that fbm_compile added last 8131 time. */ 8132 SvCUR_set(sv, SvCUR(sv) - 1); 8133 fbm_compile(sv, FBMcf_TAIL); 8134 } else 8135 fbm_compile(sv, 0); 8136 } 8137 prog->substrs->data[i].substr = sv; 8138 if (prog->substrs->data[i].utf8_substr == prog->check_utf8) 8139 prog->check_substr = sv; 8140 } 8141 } while (i--); 8142 8143 return TRUE; 8144 } 8145 8146 /* 8147 * Local variables: 8148 * c-indentation-style: bsd 8149 * c-basic-offset: 4 8150 * indent-tabs-mode: nil 8151 * End: 8152 * 8153 * ex: set ts=8 sts=4 sw=4 et: 8154 */ 8155