5 * One Ring to rule them all, One Ring to find them
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"]
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.
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.
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!
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.
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.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
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:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
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
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.
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.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "inline_invlist.c"
84 #include "unicode_constants.h"
87 /* At least one required character in the target string is expressible only in
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";
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));\
98 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
101 #define STATIC static
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)))
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)
117 #define HOPc(pos,off) \
118 (char *)(reginfo->is_utf8_target \
119 ? reghop3((U8*)pos, off, \
120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
123 #define HOPBACKc(pos, off) \
124 (char*)(reginfo->is_utf8_target \
125 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
126 : (pos - off >= reginfo->strbeg) \
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))
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) \
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)))
147 #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
148 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
150 #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
152 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
153 #define NEXTCHR_IS_EOS (nextchr < 0)
155 #define SET_nextchr \
156 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
158 #define SET_locinput(p) \
163 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
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); \
172 /* If in debug mode, we test that a known character properly matches */
174 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
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));
181 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
184 utf8_char_in_property) \
185 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
188 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
189 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
191 PL_XPosix_ptrs[_CC_WORDCHAR], \
192 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
194 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
196 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
197 "_X_regular_begin", \
199 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \
200 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
203 COMBINING_GRAVE_ACCENT_UTF8); \
206 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
207 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
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
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
219 #define JUMPABLE(rn) ( \
221 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
223 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
224 OP(rn) == PLUS || OP(rn) == MINMOD || \
226 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
228 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
230 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
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 )
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 )
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.
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) \
259 else if (type == IFMATCH) \
260 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
261 else rn += NEXT_OFF(rn); \
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 */
275 #define SLAB_FIRST(s) (&(s)->states[0])
276 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
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);
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. */
289 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
291 const int retval = PL_savestack_ix;
292 const int paren_elems_to_push =
293 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
294 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
295 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
297 GET_RE_DEBUG_FLAGS_DECL;
299 PERL_ARGS_ASSERT_REGCPPUSH;
301 if (paren_elems_to_push < 0)
302 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
303 (int)paren_elems_to_push, (int)maxopenparen,
304 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
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)",
310 (unsigned long)maxopenparen,
313 SSGROW(total_elems + REGCP_FRAME_ELEMS);
316 if ((int)maxopenparen > (int)parenfloor)
317 PerlIO_printf(Perl_debug_log,
318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
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",
331 (IV)rex->offs[p].start,
332 (IV)rex->offs[p].start_tmp,
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. */
345 /* These are needed since we do not localize EVAL nodes: */
346 #define REGCP_SET(cp) \
348 PerlIO_printf(Perl_debug_log, \
349 " Setting an EVAL scope, savestack=%"IVdf"\n", \
350 (IV)PL_savestack_ix)); \
353 #define REGCP_UNWIND(cp) \
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)); \
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;
369 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
373 GET_RE_DEBUG_FLAGS_DECL;
375 PERL_ARGS_ASSERT_REGCPPOP;
377 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
379 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
380 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
381 rex->lastcloseparen = SSPOPINT;
382 rex->lastparen = SSPOPINT;
383 *maxopenparen_p = SSPOPINT;
385 i -= REGCP_OTHER_ELEMS;
386 /* Now restore the parentheses context. */
388 if (i || rex->lastparen + 1 <= rex->nparens)
389 PerlIO_printf(Perl_debug_log,
390 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
395 paren = *maxopenparen_p;
396 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
398 rex->offs[paren].start_tmp = SSPOPINT;
399 rex->offs[paren].start = SSPOPIV;
401 if (paren <= rex->lastparen)
402 rex->offs[paren].end = tmps;
403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
406 (IV)rex->offs[paren].start,
407 (IV)rex->offs[paren].start_tmp,
408 (IV)rex->offs[paren].end,
409 (paren > rex->lastparen ? "(skipped)" : ""));
414 /* It would seem that the similar code in regtry()
415 * already takes care of this, and in fact it is in
416 * a better location to since this code can #if 0-ed out
417 * but the code in regtry() is needed or otherwise tests
418 * requiring null fields (pat.t#187 and split.t#{13,14}
419 * (as of patchlevel 7877) will fail. Then again,
420 * this code seems to be necessary or otherwise
421 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
422 * --jhi updated by dapm */
423 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
424 if (i > *maxopenparen_p)
425 rex->offs[i].start = -1;
426 rex->offs[i].end = -1;
427 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
428 " \\%"UVuf": %s ..-1 undeffing\n",
430 (i > *maxopenparen_p) ? "-1" : " "
436 /* restore the parens and associated vars at savestack position ix,
437 * but without popping the stack */
440 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
442 I32 tmpix = PL_savestack_ix;
443 PL_savestack_ix = ix;
444 regcppop(rex, maxopenparen_p);
445 PL_savestack_ix = tmpix;
448 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
451 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
453 /* Returns a boolean as to whether or not 'character' is a member of the
454 * Posix character class given by 'classnum' that should be equivalent to a
455 * value in the typedef '_char_class_number'.
457 * Ideally this could be replaced by a just an array of function pointers
458 * to the C library functions that implement the macros this calls.
459 * However, to compile, the precise function signatures are required, and
460 * these may vary from platform to to platform. To avoid having to figure
461 * out what those all are on each platform, I (khw) am using this method,
462 * which adds an extra layer of function call overhead (unless the C
463 * optimizer strips it away). But we don't particularly care about
464 * performance with locales anyway. */
466 switch ((_char_class_number) classnum) {
467 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
468 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
469 case _CC_ENUM_ASCII: return isASCII_LC(character);
470 case _CC_ENUM_BLANK: return isBLANK_LC(character);
471 case _CC_ENUM_CASED: return isLOWER_LC(character)
472 || isUPPER_LC(character);
473 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
474 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
475 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
476 case _CC_ENUM_LOWER: return isLOWER_LC(character);
477 case _CC_ENUM_PRINT: return isPRINT_LC(character);
478 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
479 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
480 case _CC_ENUM_SPACE: return isSPACE_LC(character);
481 case _CC_ENUM_UPPER: return isUPPER_LC(character);
482 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
483 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
484 default: /* VERTSPACE should never occur in locales */
485 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
488 NOT_REACHED; /* NOTREACHED */
493 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
495 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
496 * 'character' is a member of the Posix character class given by 'classnum'
497 * that should be equivalent to a value in the typedef
498 * '_char_class_number'.
500 * This just calls isFOO_lc on the code point for the character if it is in
501 * the range 0-255. Outside that range, all characters avoid Unicode
502 * rules, ignoring any locale. So use the Unicode function if this class
503 * requires a swash, and use the Unicode macro otherwise. */
505 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
507 if (UTF8_IS_INVARIANT(*character)) {
508 return isFOO_lc(classnum, *character);
510 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
511 return isFOO_lc(classnum,
512 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
515 if (classnum < _FIRST_NON_SWASH_CC) {
517 /* Initialize the swash unless done already */
518 if (! PL_utf8_swash_ptrs[classnum]) {
519 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
520 PL_utf8_swash_ptrs[classnum] =
521 _core_swash_init("utf8",
524 PL_XPosix_ptrs[classnum], &flags);
527 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
529 TRUE /* is UTF */ ));
532 switch ((_char_class_number) classnum) {
534 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
536 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
537 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
538 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
542 return FALSE; /* Things like CNTRL are always below 256 */
546 * pregexec and friends
549 #ifndef PERL_IN_XSUB_RE
551 - pregexec - match a regexp against a string
554 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
555 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
556 /* stringarg: the point in the string at which to begin matching */
557 /* strend: pointer to null at end of string */
558 /* strbeg: real beginning of string */
559 /* minend: end of match must be >= minend bytes after stringarg. */
560 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
561 * itself is accessed via the pointers above */
562 /* nosave: For optimizations. */
564 PERL_ARGS_ASSERT_PREGEXEC;
567 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
568 nosave ? 0 : REXEC_COPY_STR);
574 /* re_intuit_start():
576 * Based on some optimiser hints, try to find the earliest position in the
577 * string where the regex could match.
579 * rx: the regex to match against
580 * sv: the SV being matched: only used for utf8 flag; the string
581 * itself is accessed via the pointers below. Note that on
582 * something like an overloaded SV, SvPOK(sv) may be false
583 * and the string pointers may point to something unrelated to
585 * strbeg: real beginning of string
586 * strpos: the point in the string at which to begin matching
587 * strend: pointer to the byte following the last char of the string
588 * flags currently unused; set to 0
589 * data: currently unused; set to NULL
591 * The basic idea of re_intuit_start() is to use some known information
592 * about the pattern, namely:
594 * a) the longest known anchored substring (i.e. one that's at a
595 * constant offset from the beginning of the pattern; but not
596 * necessarily at a fixed offset from the beginning of the
598 * b) the longest floating substring (i.e. one that's not at a constant
599 * offset from the beginning of the pattern);
600 * c) Whether the pattern is anchored to the string; either
601 * an absolute anchor: /^../, or anchored to \n: /^.../m,
602 * or anchored to pos(): /\G/;
603 * d) A start class: a real or synthetic character class which
604 * represents which characters are legal at the start of the pattern;
606 * to either quickly reject the match, or to find the earliest position
607 * within the string at which the pattern might match, thus avoiding
608 * running the full NFA engine at those earlier locations, only to
609 * eventually fail and retry further along.
611 * Returns NULL if the pattern can't match, or returns the address within
612 * the string which is the earliest place the match could occur.
614 * The longest of the anchored and floating substrings is called 'check'
615 * and is checked first. The other is called 'other' and is checked
616 * second. The 'other' substring may not be present. For example,
618 * /(abc|xyz)ABC\d{0,3}DEFG/
622 * check substr (float) = "DEFG", offset 6..9 chars
623 * other substr (anchored) = "ABC", offset 3..3 chars
626 * Be aware that during the course of this function, sometimes 'anchored'
627 * refers to a substring being anchored relative to the start of the
628 * pattern, and sometimes to the pattern itself being anchored relative to
629 * the string. For example:
631 * /\dabc/: "abc" is anchored to the pattern;
632 * /^\dabc/: "abc" is anchored to the pattern and the string;
633 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
634 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
635 * but the pattern is anchored to the string.
639 Perl_re_intuit_start(pTHX_
642 const char * const strbeg,
646 re_scream_pos_data *data)
648 struct regexp *const prog = ReANY(rx);
649 SSize_t start_shift = prog->check_offset_min;
650 /* Should be nonnegative! */
651 SSize_t end_shift = 0;
652 /* current lowest pos in string where the regex can start matching */
653 char *rx_origin = strpos;
655 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
656 U8 other_ix = 1 - prog->substrs->check_ix;
658 char *other_last = strpos;/* latest pos 'other' substr already checked to */
659 char *check_at = NULL; /* check substr found at this pos */
660 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
661 RXi_GET_DECL(prog,progi);
662 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
663 regmatch_info *const reginfo = ®info_buf;
664 GET_RE_DEBUG_FLAGS_DECL;
666 PERL_ARGS_ASSERT_RE_INTUIT_START;
667 PERL_UNUSED_ARG(flags);
668 PERL_UNUSED_ARG(data);
670 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
671 "Intuit: trying to determine minimum start position...\n"));
673 /* for now, assume that all substr offsets are positive. If at some point
674 * in the future someone wants to do clever things with look-behind and
675 * -ve offsets, they'll need to fix up any code in this function
676 * which uses these offsets. See the thread beginning
677 * <20140113145929.GF27210@iabyn.com>
679 assert(prog->substrs->data[0].min_offset >= 0);
680 assert(prog->substrs->data[0].max_offset >= 0);
681 assert(prog->substrs->data[1].min_offset >= 0);
682 assert(prog->substrs->data[1].max_offset >= 0);
683 assert(prog->substrs->data[2].min_offset >= 0);
684 assert(prog->substrs->data[2].max_offset >= 0);
686 /* for now, assume that if both present, that the floating substring
687 * doesn't start before the anchored substring.
688 * If you break this assumption (e.g. doing better optimisations
689 * with lookahead/behind), then you'll need to audit the code in this
690 * function carefully first
693 ! ( (prog->anchored_utf8 || prog->anchored_substr)
694 && (prog->float_utf8 || prog->float_substr))
695 || (prog->float_min_offset >= prog->anchored_offset));
697 /* byte rather than char calculation for efficiency. It fails
698 * to quickly reject some cases that can't match, but will reject
699 * them later after doing full char arithmetic */
700 if (prog->minlen > strend - strpos) {
701 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
702 " String too short...\n"));
706 RX_MATCH_UTF8_set(rx,utf8_target);
707 reginfo->is_utf8_target = cBOOL(utf8_target);
708 reginfo->info_aux = NULL;
709 reginfo->strbeg = strbeg;
710 reginfo->strend = strend;
711 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
713 /* not actually used within intuit, but zero for safety anyway */
714 reginfo->poscache_maxiter = 0;
717 if (!prog->check_utf8 && prog->check_substr)
718 to_utf8_substr(prog);
719 check = prog->check_utf8;
721 if (!prog->check_substr && prog->check_utf8) {
722 if (! to_byte_substr(prog)) {
723 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
726 check = prog->check_substr;
729 /* dump the various substring data */
730 DEBUG_OPTIMISE_MORE_r({
732 for (i=0; i<=2; i++) {
733 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
734 : prog->substrs->data[i].substr);
738 PerlIO_printf(Perl_debug_log,
739 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
740 " useful=%"IVdf" utf8=%d [%s]\n",
742 (IV)prog->substrs->data[i].min_offset,
743 (IV)prog->substrs->data[i].max_offset,
744 (IV)prog->substrs->data[i].end_shift,
751 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
753 /* ml_anch: check after \n?
755 * A note about IMPLICIT: on an un-anchored pattern beginning
756 * with /.*.../, these flags will have been added by the
758 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
759 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
761 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
762 && !(prog->intflags & PREGf_IMPLICIT);
764 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
765 /* we are only allowed to match at BOS or \G */
767 /* trivially reject if there's a BOS anchor and we're not at BOS.
769 * Note that we don't try to do a similar quick reject for
770 * \G, since generally the caller will have calculated strpos
771 * based on pos() and gofs, so the string is already correctly
772 * anchored by definition; and handling the exceptions would
773 * be too fiddly (e.g. REXEC_IGNOREPOS).
775 if ( strpos != strbeg
776 && (prog->intflags & PREGf_ANCH_SBOL))
778 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
779 " Not at start...\n"));
783 /* in the presence of an anchor, the anchored (relative to the
784 * start of the regex) substr must also be anchored relative
785 * to strpos. So quickly reject if substr isn't found there.
786 * This works for \G too, because the caller will already have
787 * subtracted gofs from pos, and gofs is the offset from the
788 * \G to the start of the regex. For example, in /.abc\Gdef/,
789 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
790 * caller will have set strpos=pos()-4; we look for the substr
791 * at position pos()-4+1, which lines up with the "a" */
793 if (prog->check_offset_min == prog->check_offset_max
794 && !(prog->intflags & PREGf_CANY_SEEN))
796 /* Substring at constant offset from beg-of-str... */
797 SSize_t slen = SvCUR(check);
798 char *s = HOP3c(strpos, prog->check_offset_min, strend);
800 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
801 " Looking for check substr at fixed offset %"IVdf"...\n",
802 (IV)prog->check_offset_min));
805 /* In this case, the regex is anchored at the end too.
806 * Unless it's a multiline match, the lengths must match
807 * exactly, give or take a \n. NB: slen >= 1 since
808 * the last char of check is \n */
810 && ( strend - s > slen
811 || strend - s < slen - 1
812 || (strend - s == slen && strend[-1] != '\n')))
814 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
815 " String too long...\n"));
818 /* Now should match s[0..slen-2] */
821 if (slen && (*SvPVX_const(check) != *s
822 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
824 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
825 " String not equal...\n"));
830 goto success_at_start;
835 end_shift = prog->check_end_shift;
837 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
839 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
840 (IV)end_shift, RX_PRECOMP(prog));
845 /* This is the (re)entry point of the main loop in this function.
846 * The goal of this loop is to:
847 * 1) find the "check" substring in the region rx_origin..strend
848 * (adjusted by start_shift / end_shift). If not found, reject
850 * 2) If it exists, look for the "other" substr too if defined; for
851 * example, if the check substr maps to the anchored substr, then
852 * check the floating substr, and vice-versa. If not found, go
853 * back to (1) with rx_origin suitably incremented.
854 * 3) If we find an rx_origin position that doesn't contradict
855 * either of the substrings, then check the possible additional
856 * constraints on rx_origin of /^.../m or a known start class.
857 * If these fail, then depending on which constraints fail, jump
858 * back to here, or to various other re-entry points further along
859 * that skip some of the first steps.
860 * 4) If we pass all those tests, update the BmUSEFUL() count on the
861 * substring. If the start position was determined to be at the
862 * beginning of the string - so, not rejected, but not optimised,
863 * since we have to run regmatch from position 0 - decrement the
864 * BmUSEFUL() count. Otherwise increment it.
868 /* first, look for the 'check' substring */
874 DEBUG_OPTIMISE_MORE_r({
875 PerlIO_printf(Perl_debug_log,
876 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
877 " Start shift: %"IVdf" End shift %"IVdf
878 " Real end Shift: %"IVdf"\n",
879 (IV)(rx_origin - strpos),
880 (IV)prog->check_offset_min,
883 (IV)prog->check_end_shift);
886 if (prog->intflags & PREGf_CANY_SEEN) {
887 start_point= (U8*)(rx_origin + start_shift);
888 end_point= (U8*)(strend - end_shift);
889 if (start_point > end_point)
892 end_point = HOP3(strend, -end_shift, strbeg);
893 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
899 /* If the regex is absolutely anchored to either the start of the
900 * string (SBOL) or to pos() (ANCH_GPOS), then
901 * check_offset_max represents an upper bound on the string where
902 * the substr could start. For the ANCH_GPOS case, we assume that
903 * the caller of intuit will have already set strpos to
904 * pos()-gofs, so in this case strpos + offset_max will still be
905 * an upper bound on the substr.
908 && prog->intflags & PREGf_ANCH
909 && prog->check_offset_max != SSize_t_MAX)
911 SSize_t len = SvCUR(check) - !!SvTAIL(check);
912 const char * const anchor =
913 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
915 /* do a bytes rather than chars comparison. It's conservative;
916 * so it skips doing the HOP if the result can't possibly end
917 * up earlier than the old value of end_point.
919 if ((char*)end_point - anchor > prog->check_offset_max) {
920 end_point = HOP3lim((U8*)anchor,
921 prog->check_offset_max,
927 DEBUG_OPTIMISE_MORE_r({
928 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
929 (int)(end_point - start_point),
930 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
934 check_at = fbm_instr( start_point, end_point,
935 check, multiline ? FBMrf_MULTILINE : 0);
937 /* Update the count-of-usability, remove useless subpatterns,
941 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
942 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
943 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
944 (check_at ? "Found" : "Did not find"),
945 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
946 ? "anchored" : "floating"),
949 (check_at ? " at offset " : "...\n") );
954 /* Finish the diagnostic message */
955 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
957 /* set rx_origin to the minimum position where the regex could start
958 * matching, given the constraint of the just-matched check substring.
959 * But don't set it lower than previously.
962 if (check_at - rx_origin > prog->check_offset_max)
963 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
967 /* now look for the 'other' substring if defined */
969 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
970 : prog->substrs->data[other_ix].substr)
972 /* Take into account the "other" substring. */
976 struct reg_substr_datum *other;
979 other = &prog->substrs->data[other_ix];
981 /* if "other" is anchored:
982 * we've previously found a floating substr starting at check_at.
983 * This means that the regex origin must lie somewhere
984 * between min (rx_origin): HOP3(check_at, -check_offset_max)
985 * and max: HOP3(check_at, -check_offset_min)
986 * (except that min will be >= strpos)
987 * So the fixed substr must lie somewhere between
988 * HOP3(min, anchored_offset)
989 * HOP3(max, anchored_offset) + SvCUR(substr)
992 /* if "other" is floating
993 * Calculate last1, the absolute latest point where the
994 * floating substr could start in the string, ignoring any
995 * constraints from the earlier fixed match. It is calculated
998 * strend - prog->minlen (in chars) is the absolute latest
999 * position within the string where the origin of the regex
1000 * could appear. The latest start point for the floating
1001 * substr is float_min_offset(*) on from the start of the
1002 * regex. last1 simply combines thee two offsets.
1004 * (*) You might think the latest start point should be
1005 * float_max_offset from the regex origin, and technically
1006 * you'd be correct. However, consider
1008 * Here, float min, max are 3,5 and minlen is 7.
1009 * This can match either
1013 * In the first case, the regex matches minlen chars; in the
1014 * second, minlen+1, in the third, minlen+2.
1015 * In the first case, the floating offset is 3 (which equals
1016 * float_min), in the second, 4, and in the third, 5 (which
1017 * equals float_max). In all cases, the floating string bcd
1018 * can never start more than 4 chars from the end of the
1019 * string, which equals minlen - float_min. As the substring
1020 * starts to match more than float_min from the start of the
1021 * regex, it makes the regex match more than minlen chars,
1022 * and the two cancel each other out. So we can always use
1023 * float_min - minlen, rather than float_max - minlen for the
1024 * latest position in the string.
1026 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1027 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1030 assert(prog->minlen >= other->min_offset);
1031 last1 = HOP3c(strend,
1032 other->min_offset - prog->minlen, strbeg);
1034 if (other_ix) {/* i.e. if (other-is-float) */
1035 /* last is the latest point where the floating substr could
1036 * start, *given* any constraints from the earlier fixed
1037 * match. This constraint is that the floating string starts
1038 * <= float_max_offset chars from the regex origin (rx_origin).
1039 * If this value is less than last1, use it instead.
1041 assert(rx_origin <= last1);
1043 /* this condition handles the offset==infinity case, and
1044 * is a short-cut otherwise. Although it's comparing a
1045 * byte offset to a char length, it does so in a safe way,
1046 * since 1 char always occupies 1 or more bytes,
1047 * so if a string range is (last1 - rx_origin) bytes,
1048 * it will be less than or equal to (last1 - rx_origin)
1049 * chars; meaning it errs towards doing the accurate HOP3
1050 * rather than just using last1 as a short-cut */
1051 (last1 - rx_origin) < other->max_offset
1053 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1056 assert(strpos + start_shift <= check_at);
1057 last = HOP4c(check_at, other->min_offset - start_shift,
1061 s = HOP3c(rx_origin, other->min_offset, strend);
1062 if (s < other_last) /* These positions already checked */
1065 must = utf8_target ? other->utf8_substr : other->substr;
1066 assert(SvPOK(must));
1069 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1071 multiline ? FBMrf_MULTILINE : 0
1074 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1075 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1076 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1077 s ? "Found" : "Contradicts",
1078 other_ix ? "floating" : "anchored",
1079 quoted, RE_SV_TAIL(must));
1084 /* last1 is latest possible substr location. If we didn't
1085 * find it before there, we never will */
1086 if (last >= last1) {
1087 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1088 ", giving up...\n"));
1092 /* try to find the check substr again at a later
1093 * position. Maybe next time we'll find the "other" substr
1095 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1096 ", trying %s at offset %ld...\n",
1097 (other_ix ? "floating" : "anchored"),
1098 (long)(HOP3c(check_at, 1, strend) - strpos)));
1100 other_last = HOP3c(last, 1, strend) /* highest failure */;
1102 other_ix /* i.e. if other-is-float */
1103 ? HOP3c(rx_origin, 1, strend)
1104 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1108 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
1109 (long)(s - strpos)));
1111 if (other_ix) { /* if (other-is-float) */
1112 /* other_last is set to s, not s+1, since its possible for
1113 * a floating substr to fail first time, then succeed
1114 * second time at the same floating position; e.g.:
1115 * "-AB--AABZ" =~ /\wAB\d*Z/
1116 * The first time round, anchored and float match at
1117 * "-(AB)--AAB(Z)" then fail on the initial \w character
1118 * class. Second time round, they match at "-AB--A(AB)(Z)".
1123 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1124 other_last = HOP3c(s, 1, strend);
1129 DEBUG_OPTIMISE_MORE_r(
1130 PerlIO_printf(Perl_debug_log,
1131 " Check-only match: offset min:%"IVdf" max:%"IVdf
1132 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
1133 " strend-strpos:%"IVdf"\n",
1134 (IV)prog->check_offset_min,
1135 (IV)prog->check_offset_max,
1136 (IV)(check_at-strpos),
1137 (IV)(rx_origin-strpos),
1138 (IV)(rx_origin-check_at),
1144 postprocess_substr_matches:
1146 /* handle the extra constraint of /^.../m if present */
1148 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1151 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1152 " looking for /^/m anchor"));
1154 /* we have failed the constraint of a \n before rx_origin.
1155 * Find the next \n, if any, even if it's beyond the current
1156 * anchored and/or floating substrings. Whether we should be
1157 * scanning ahead for the next \n or the next substr is debatable.
1158 * On the one hand you'd expect rare substrings to appear less
1159 * often than \n's. On the other hand, searching for \n means
1160 * we're effectively flipping been check_substr and "\n" on each
1161 * iteration as the current "rarest" string candidate, which
1162 * means for example that we'll quickly reject the whole string if
1163 * hasn't got a \n, rather than trying every substr position
1167 s = HOP3c(strend, - prog->minlen, strpos);
1168 if (s <= rx_origin ||
1169 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1171 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1172 " Did not find /%s^%s/m...\n",
1173 PL_colors[0], PL_colors[1]));
1177 /* earliest possible origin is 1 char after the \n.
1178 * (since *rx_origin == '\n', it's safe to ++ here rather than
1179 * HOP(rx_origin, 1)) */
1182 if (prog->substrs->check_ix == 0 /* check is anchored */
1183 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1185 /* Position contradicts check-string; either because
1186 * check was anchored (and thus has no wiggle room),
1187 * or check was float and rx_origin is above the float range */
1188 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1189 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1190 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1194 /* if we get here, the check substr must have been float,
1195 * is in range, and we may or may not have had an anchored
1196 * "other" substr which still contradicts */
1197 assert(prog->substrs->check_ix); /* check is float */
1199 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1200 /* whoops, the anchored "other" substr exists, so we still
1201 * contradict. On the other hand, the float "check" substr
1202 * didn't contradict, so just retry the anchored "other"
1204 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1205 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1206 PL_colors[0], PL_colors[1],
1207 (long)(rx_origin - strpos),
1208 (long)(rx_origin - strpos + prog->anchored_offset)));
1209 goto do_other_substr;
1212 /* success: we don't contradict the found floating substring
1213 * (and there's no anchored substr). */
1214 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1215 " Found /%s^%s/m at offset %ld...\n",
1216 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1219 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1220 " (multiline anchor test skipped)\n"));
1226 /* if we have a starting character class, then test that extra constraint.
1227 * (trie stclasses are too expensive to use here, we are better off to
1228 * leave it to regmatch itself) */
1230 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1231 const U8* const str = (U8*)STRING(progi->regstclass);
1233 /* XXX this value could be pre-computed */
1234 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1235 ? (reginfo->is_utf8_pat
1236 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1237 : STR_LEN(progi->regstclass))
1241 /* latest pos that a matching float substr constrains rx start to */
1242 char *rx_max_float = NULL;
1244 /* if the current rx_origin is anchored, either by satisfying an
1245 * anchored substring constraint, or a /^.../m constraint, then we
1246 * can reject the current origin if the start class isn't found
1247 * at the current position. If we have a float-only match, then
1248 * rx_origin is constrained to a range; so look for the start class
1249 * in that range. if neither, then look for the start class in the
1250 * whole rest of the string */
1252 /* XXX DAPM it's not clear what the minlen test is for, and why
1253 * it's not used in the floating case. Nothing in the test suite
1254 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1255 * Here are some old comments, which may or may not be correct:
1257 * minlen == 0 is possible if regstclass is \b or \B,
1258 * and the fixed substr is ''$.
1259 * Since minlen is already taken into account, rx_origin+1 is
1260 * before strend; accidentally, minlen >= 1 guaranties no false
1261 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1262 * 0) below assumes that regstclass does not come from lookahead...
1263 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1264 * This leaves EXACTF-ish only, which are dealt with in
1268 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1269 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1270 else if (prog->float_substr || prog->float_utf8) {
1271 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1272 endpos= HOP3c(rx_max_float, cl_l, strend);
1277 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1278 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
1279 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1280 (IV)start_shift, (IV)(check_at - strbeg),
1281 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1283 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1286 if (endpos == strend) {
1287 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1288 " Could not match STCLASS...\n") );
1291 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1292 " This position contradicts STCLASS...\n") );
1293 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1294 && !(prog->intflags & PREGf_IMPLICIT))
1297 /* Contradict one of substrings */
1298 if (prog->anchored_substr || prog->anchored_utf8) {
1299 if (prog->substrs->check_ix == 1) { /* check is float */
1300 /* Have both, check_string is floating */
1301 assert(rx_origin + start_shift <= check_at);
1302 if (rx_origin + start_shift != check_at) {
1303 /* not at latest position float substr could match:
1304 * Recheck anchored substring, but not floating.
1305 * The condition above is in bytes rather than
1306 * chars for efficiency. It's conservative, in
1307 * that it errs on the side of doing 'goto
1308 * do_other_substr', where a more accurate
1309 * char-based calculation will be done */
1310 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1311 " Looking for anchored substr starting at offset %ld...\n",
1312 (long)(other_last - strpos)) );
1313 goto do_other_substr;
1321 /* In the presence of ml_anch, we might be able to
1322 * find another \n without breaking the current float
1325 /* strictly speaking this should be HOP3c(..., 1, ...),
1326 * but since we goto a block of code that's going to
1327 * search for the next \n if any, its safe here */
1329 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1330 " Looking for /%s^%s/m starting at offset %ld...\n",
1331 PL_colors[0], PL_colors[1],
1332 (long)(rx_origin - strpos)) );
1333 goto postprocess_substr_matches;
1336 /* strictly speaking this can never be true; but might
1337 * be if we ever allow intuit without substrings */
1338 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1341 rx_origin = rx_max_float;
1344 /* at this point, any matching substrings have been
1345 * contradicted. Start again... */
1347 rx_origin = HOP3c(rx_origin, 1, strend);
1349 /* uses bytes rather than char calculations for efficiency.
1350 * It's conservative: it errs on the side of doing 'goto restart',
1351 * where there is code that does a proper char-based test */
1352 if (rx_origin + start_shift + end_shift > strend) {
1353 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1354 " Could not match STCLASS...\n") );
1357 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1358 " Looking for %s substr starting at offset %ld...\n",
1359 (prog->substrs->check_ix ? "floating" : "anchored"),
1360 (long)(rx_origin + start_shift - strpos)) );
1366 if (rx_origin != s) {
1367 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1368 " By STCLASS: moving %ld --> %ld\n",
1369 (long)(rx_origin - strpos), (long)(s - strpos))
1373 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1374 " Does not contradict STCLASS...\n");
1379 /* Decide whether using the substrings helped */
1381 if (rx_origin != strpos) {
1382 /* Fixed substring is found far enough so that the match
1383 cannot start at strpos. */
1385 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1386 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1389 /* The found rx_origin position does not prohibit matching at
1390 * strpos, so calling intuit didn't gain us anything. Decrement
1391 * the BmUSEFUL() count on the check substring, and if we reach
1393 if (!(prog->intflags & PREGf_NAUGHTY)
1395 prog->check_utf8 /* Could be deleted already */
1396 && --BmUSEFUL(prog->check_utf8) < 0
1397 && (prog->check_utf8 == prog->float_utf8)
1399 prog->check_substr /* Could be deleted already */
1400 && --BmUSEFUL(prog->check_substr) < 0
1401 && (prog->check_substr == prog->float_substr)
1404 /* If flags & SOMETHING - do not do it many times on the same match */
1405 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1406 /* XXX Does the destruction order has to change with utf8_target? */
1407 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1408 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1409 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1410 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1411 check = NULL; /* abort */
1412 /* XXXX This is a remnant of the old implementation. It
1413 looks wasteful, since now INTUIT can use many
1414 other heuristics. */
1415 prog->extflags &= ~RXf_USE_INTUIT;
1419 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1420 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1421 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1425 fail_finish: /* Substring not found */
1426 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1427 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1429 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1430 PL_colors[4], PL_colors[5]));
1435 #define DECL_TRIE_TYPE(scan) \
1436 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1437 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
1438 trie_type = ((scan->flags == EXACT) \
1439 ? (utf8_target ? trie_utf8 : trie_plain) \
1440 : (scan->flags == EXACTFA) \
1441 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1442 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1444 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1447 U8 flags = FOLD_FLAGS_FULL; \
1448 switch (trie_type) { \
1449 case trie_utf8_exactfa_fold: \
1450 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1452 case trie_utf8_fold: \
1453 if ( foldlen>0 ) { \
1454 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1459 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1460 len = UTF8SKIP(uc); \
1461 skiplen = UNISKIP( uvc ); \
1462 foldlen -= skiplen; \
1463 uscan = foldbuf + skiplen; \
1466 case trie_latin_utf8_exactfa_fold: \
1467 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1469 case trie_latin_utf8_fold: \
1470 if ( foldlen>0 ) { \
1471 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1477 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1478 skiplen = UNISKIP( uvc ); \
1479 foldlen -= skiplen; \
1480 uscan = foldbuf + skiplen; \
1484 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1491 charid = trie->charmap[ uvc ]; \
1495 if (widecharmap) { \
1496 SV** const svpp = hv_fetch(widecharmap, \
1497 (char*)&uvc, sizeof(UV), 0); \
1499 charid = (U16)SvIV(*svpp); \
1504 #define DUMP_EXEC_POS(li,s,doutf8) \
1505 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1508 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1512 && (ln == 1 || folder(s, pat_string, ln)) \
1513 && (reginfo->intuit || regtry(reginfo, &s)) )\
1519 #define REXEC_FBC_UTF8_SCAN(CODE) \
1521 while (s < strend) { \
1527 #define REXEC_FBC_SCAN(CODE) \
1529 while (s < strend) { \
1535 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1536 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1538 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1547 #define REXEC_FBC_CLASS_SCAN(COND) \
1548 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1550 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1559 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1560 if (utf8_target) { \
1561 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1564 REXEC_FBC_CLASS_SCAN(COND); \
1567 /* The three macros below are slightly different versions of the same logic.
1569 * The first is for /a and /aa when the target string is UTF-8. This can only
1570 * match ascii, but it must advance based on UTF-8. The other two handle the
1571 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1572 * for the boundary (or non-boundary) between a word and non-word character.
1573 * The utf8 and non-utf8 cases have the same logic, but the details must be
1574 * different. Find the "wordness" of the character just prior to this one, and
1575 * compare it with the wordness of this one. If they differ, we have a
1576 * boundary. At the beginning of the string, pretend that the previous
1577 * character was a new-line.
1579 * All these macros uncleanly have side-effects with each other and outside
1580 * variables. So far it's been too much trouble to clean-up
1582 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1583 * a word character or not.
1584 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1586 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1588 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1589 * are looking for a boundary or for a non-boundary. If we are looking for a
1590 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1591 * see if this tentative match actually works, and if so, to quit the loop
1592 * here. And vice-versa if we are looking for a non-boundary.
1594 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1595 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1596 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1597 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1598 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1599 * complement. But in that branch we complement tmp, meaning that at the
1600 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1601 * which means at the top of the loop in the next iteration, it is
1602 * TEST_NON_UTF8(s-1) */
1603 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1604 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1605 tmp = TEST_NON_UTF8(tmp); \
1606 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1607 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1609 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1616 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1617 * TEST_UTF8 is a macro that for the same input code points returns identically
1618 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1619 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1620 if (s == reginfo->strbeg) { \
1623 else { /* Back-up to the start of the previous character */ \
1624 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1625 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1626 0, UTF8_ALLOW_DEFAULT); \
1628 tmp = TEST_UV(tmp); \
1629 LOAD_UTF8_CHARCLASS_ALNUM(); \
1630 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1631 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1640 /* Like the above two macros. UTF8_CODE is the complete code for handling
1641 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1643 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1644 if (utf8_target) { \
1647 else { /* Not utf8 */ \
1648 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1649 tmp = TEST_NON_UTF8(tmp); \
1650 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1651 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1660 /* Here, things have been set up by the previous code so that tmp is the \
1661 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1662 * utf8ness of the target). We also have to check if this matches against \
1663 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1664 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1666 if (tmp == ! TEST_NON_UTF8('\n')) { \
1673 /* This is the macro to use when we want to see if something that looks like it
1674 * could match, actually does, and if so exits the loop */
1675 #define REXEC_FBC_TRYIT \
1676 if ((reginfo->intuit || regtry(reginfo, &s))) \
1679 /* The only difference between the BOUND and NBOUND cases is that
1680 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1681 * NBOUND. This is accomplished by passing it as either the if or else clause,
1682 * with the other one being empty (PLACEHOLDER is defined as empty).
1684 * The TEST_FOO parameters are for operating on different forms of input, but
1685 * all should be ones that return identically for the same underlying code
1687 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1689 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1690 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1692 #define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1694 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1695 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1697 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1699 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1700 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1702 #define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1704 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1705 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1708 /* We know what class REx starts with. Try to find this position... */
1709 /* if reginfo->intuit, its a dryrun */
1710 /* annoyingly all the vars in this routine have different names from their counterparts
1711 in regmatch. /grrr */
1713 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1714 const char *strend, regmatch_info *reginfo)
1717 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1718 char *pat_string; /* The pattern's exactish string */
1719 char *pat_end; /* ptr to end char of pat_string */
1720 re_fold_t folder; /* Function for computing non-utf8 folds */
1721 const U8 *fold_array; /* array for folding ords < 256 */
1727 I32 tmp = 1; /* Scratch variable? */
1728 const bool utf8_target = reginfo->is_utf8_target;
1729 UV utf8_fold_flags = 0;
1730 const bool is_utf8_pat = reginfo->is_utf8_pat;
1731 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1732 with a result inverts that result, as 0^1 =
1734 _char_class_number classnum;
1736 RXi_GET_DECL(prog,progi);
1738 PERL_ARGS_ASSERT_FIND_BYCLASS;
1740 /* We know what class it must start with. */
1744 REXEC_FBC_UTF8_CLASS_SCAN(
1745 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1748 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1753 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1760 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1761 assert(! is_utf8_pat);
1764 if (is_utf8_pat || utf8_target) {
1765 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1766 goto do_exactf_utf8;
1768 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1769 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1770 goto do_exactf_non_utf8; /* isn't dealt with by these */
1772 case EXACTF: /* This node only generated for non-utf8 patterns */
1773 assert(! is_utf8_pat);
1775 utf8_fold_flags = 0;
1776 goto do_exactf_utf8;
1778 fold_array = PL_fold;
1780 goto do_exactf_non_utf8;
1783 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1784 utf8_fold_flags = FOLDEQ_LOCALE;
1785 goto do_exactf_utf8;
1787 fold_array = PL_fold_locale;
1788 folder = foldEQ_locale;
1789 goto do_exactf_non_utf8;
1793 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1795 goto do_exactf_utf8;
1798 if (is_utf8_pat || utf8_target) {
1799 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1800 goto do_exactf_utf8;
1803 /* Any 'ss' in the pattern should have been replaced by regcomp,
1804 * so we don't have to worry here about this single special case
1805 * in the Latin1 range */
1806 fold_array = PL_fold_latin1;
1807 folder = foldEQ_latin1;
1811 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1812 are no glitches with fold-length differences
1813 between the target string and pattern */
1815 /* The idea in the non-utf8 EXACTF* cases is to first find the
1816 * first character of the EXACTF* node and then, if necessary,
1817 * case-insensitively compare the full text of the node. c1 is the
1818 * first character. c2 is its fold. This logic will not work for
1819 * Unicode semantics and the german sharp ss, which hence should
1820 * not be compiled into a node that gets here. */
1821 pat_string = STRING(c);
1822 ln = STR_LEN(c); /* length to match in octets/bytes */
1824 /* We know that we have to match at least 'ln' bytes (which is the
1825 * same as characters, since not utf8). If we have to match 3
1826 * characters, and there are only 2 availabe, we know without
1827 * trying that it will fail; so don't start a match past the
1828 * required minimum number from the far end */
1829 e = HOP3c(strend, -((SSize_t)ln), s);
1831 if (reginfo->intuit && e < s) {
1832 e = s; /* Due to minlen logic of intuit() */
1836 c2 = fold_array[c1];
1837 if (c1 == c2) { /* If char and fold are the same */
1838 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1841 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1849 /* If one of the operands is in utf8, we can't use the simpler folding
1850 * above, due to the fact that many different characters can have the
1851 * same fold, or portion of a fold, or different- length fold */
1852 pat_string = STRING(c);
1853 ln = STR_LEN(c); /* length to match in octets/bytes */
1854 pat_end = pat_string + ln;
1855 lnc = is_utf8_pat /* length to match in characters */
1856 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1859 /* We have 'lnc' characters to match in the pattern, but because of
1860 * multi-character folding, each character in the target can match
1861 * up to 3 characters (Unicode guarantees it will never exceed
1862 * this) if it is utf8-encoded; and up to 2 if not (based on the
1863 * fact that the Latin 1 folds are already determined, and the
1864 * only multi-char fold in that range is the sharp-s folding to
1865 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1866 * string character. Adjust lnc accordingly, rounding up, so that
1867 * if we need to match at least 4+1/3 chars, that really is 5. */
1868 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1869 lnc = (lnc + expansion - 1) / expansion;
1871 /* As in the non-UTF8 case, if we have to match 3 characters, and
1872 * only 2 are left, it's guaranteed to fail, so don't start a
1873 * match that would require us to go beyond the end of the string
1875 e = HOP3c(strend, -((SSize_t)lnc), s);
1877 if (reginfo->intuit && e < s) {
1878 e = s; /* Due to minlen logic of intuit() */
1881 /* XXX Note that we could recalculate e to stop the loop earlier,
1882 * as the worst case expansion above will rarely be met, and as we
1883 * go along we would usually find that e moves further to the left.
1884 * This would happen only after we reached the point in the loop
1885 * where if there were no expansion we should fail. Unclear if
1886 * worth the expense */
1889 char *my_strend= (char *)strend;
1890 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1891 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1892 && (reginfo->intuit || regtry(reginfo, &s)) )
1896 s += (utf8_target) ? UTF8SKIP(s) : 1;
1902 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1905 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1908 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1911 FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1914 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1917 FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1920 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1923 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1926 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1927 is_LNBREAK_latin1_safe(s, strend)
1931 /* The argument to all the POSIX node types is the class number to pass to
1932 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1939 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1940 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1955 /* The complement of something that matches only ASCII matches all
1956 * non-ASCII, plus everything in ASCII that isn't in the class. */
1957 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
1958 || ! _generic_isCC_A(*s, FLAGS(c)));
1967 /* Don't need to worry about utf8, as it can match only a single
1968 * byte invariant character. */
1969 REXEC_FBC_CLASS_SCAN(
1970 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1978 if (! utf8_target) {
1979 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1985 classnum = (_char_class_number) FLAGS(c);
1986 if (classnum < _FIRST_NON_SWASH_CC) {
1987 while (s < strend) {
1989 /* We avoid loading in the swash as long as possible, but
1990 * should we have to, we jump to a separate loop. This
1991 * extra 'if' statement is what keeps this code from being
1992 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1993 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1994 goto found_above_latin1;
1996 if ((UTF8_IS_INVARIANT(*s)
1997 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1999 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2000 && to_complement ^ cBOOL(
2001 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2005 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2017 else switch (classnum) { /* These classes are implemented as
2019 case _CC_ENUM_SPACE: /* XXX would require separate code if we
2020 revert the change of \v matching this */
2023 case _CC_ENUM_PSXSPC:
2024 REXEC_FBC_UTF8_CLASS_SCAN(
2025 to_complement ^ cBOOL(isSPACE_utf8(s)));
2028 case _CC_ENUM_BLANK:
2029 REXEC_FBC_UTF8_CLASS_SCAN(
2030 to_complement ^ cBOOL(isBLANK_utf8(s)));
2033 case _CC_ENUM_XDIGIT:
2034 REXEC_FBC_UTF8_CLASS_SCAN(
2035 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2038 case _CC_ENUM_VERTSPACE:
2039 REXEC_FBC_UTF8_CLASS_SCAN(
2040 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2043 case _CC_ENUM_CNTRL:
2044 REXEC_FBC_UTF8_CLASS_SCAN(
2045 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2049 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2050 NOT_REACHED; /* NOTREACHED */
2055 found_above_latin1: /* Here we have to load a swash to get the result
2056 for the current code point */
2057 if (! PL_utf8_swash_ptrs[classnum]) {
2058 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2059 PL_utf8_swash_ptrs[classnum] =
2060 _core_swash_init("utf8",
2063 PL_XPosix_ptrs[classnum], &flags);
2066 /* This is a copy of the loop above for swash classes, though using the
2067 * FBC macro instead of being expanded out. Since we've loaded the
2068 * swash, we don't have to check for that each time through the loop */
2069 REXEC_FBC_UTF8_CLASS_SCAN(
2070 to_complement ^ cBOOL(_generic_utf8(
2073 swash_fetch(PL_utf8_swash_ptrs[classnum],
2081 /* what trie are we using right now */
2082 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2083 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2084 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2086 const char *last_start = strend - trie->minlen;
2088 const char *real_start = s;
2090 STRLEN maxlen = trie->maxlen;
2092 U8 **points; /* map of where we were in the input string
2093 when reading a given char. For ASCII this
2094 is unnecessary overhead as the relationship
2095 is always 1:1, but for Unicode, especially
2096 case folded Unicode this is not true. */
2097 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2101 GET_RE_DEBUG_FLAGS_DECL;
2103 /* We can't just allocate points here. We need to wrap it in
2104 * an SV so it gets freed properly if there is a croak while
2105 * running the match */
2108 sv_points=newSV(maxlen * sizeof(U8 *));
2109 SvCUR_set(sv_points,
2110 maxlen * sizeof(U8 *));
2111 SvPOK_on(sv_points);
2112 sv_2mortal(sv_points);
2113 points=(U8**)SvPV_nolen(sv_points );
2114 if ( trie_type != trie_utf8_fold
2115 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2118 bitmap=(U8*)trie->bitmap;
2120 bitmap=(U8*)ANYOF_BITMAP(c);
2122 /* this is the Aho-Corasick algorithm modified a touch
2123 to include special handling for long "unknown char" sequences.
2124 The basic idea being that we use AC as long as we are dealing
2125 with a possible matching char, when we encounter an unknown char
2126 (and we have not encountered an accepting state) we scan forward
2127 until we find a legal starting char.
2128 AC matching is basically that of trie matching, except that when
2129 we encounter a failing transition, we fall back to the current
2130 states "fail state", and try the current char again, a process
2131 we repeat until we reach the root state, state 1, or a legal
2132 transition. If we fail on the root state then we can either
2133 terminate if we have reached an accepting state previously, or
2134 restart the entire process from the beginning if we have not.
2137 while (s <= last_start) {
2138 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2146 U8 *uscan = (U8*)NULL;
2147 U8 *leftmost = NULL;
2149 U32 accepted_word= 0;
2153 while ( state && uc <= (U8*)strend ) {
2155 U32 word = aho->states[ state ].wordnum;
2159 DEBUG_TRIE_EXECUTE_r(
2160 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2161 dump_exec_pos( (char *)uc, c, strend, real_start,
2162 (char *)uc, utf8_target );
2163 PerlIO_printf( Perl_debug_log,
2164 " Scanning for legal start char...\n");
2168 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2172 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2178 if (uc >(U8*)last_start) break;
2182 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2183 if (!leftmost || lpos < leftmost) {
2184 DEBUG_r(accepted_word=word);
2190 points[pointpos++ % maxlen]= uc;
2191 if (foldlen || uc < (U8*)strend) {
2192 REXEC_TRIE_READ_CHAR(trie_type, trie,
2194 uscan, len, uvc, charid, foldlen,
2196 DEBUG_TRIE_EXECUTE_r({
2197 dump_exec_pos( (char *)uc, c, strend,
2198 real_start, s, utf8_target);
2199 PerlIO_printf(Perl_debug_log,
2200 " Charid:%3u CP:%4"UVxf" ",
2212 word = aho->states[ state ].wordnum;
2214 base = aho->states[ state ].trans.base;
2216 DEBUG_TRIE_EXECUTE_r({
2218 dump_exec_pos( (char *)uc, c, strend, real_start,
2220 PerlIO_printf( Perl_debug_log,
2221 "%sState: %4"UVxf", word=%"UVxf,
2222 failed ? " Fail transition to " : "",
2223 (UV)state, (UV)word);
2229 ( ((offset = base + charid
2230 - 1 - trie->uniquecharcount)) >= 0)
2231 && ((U32)offset < trie->lasttrans)
2232 && trie->trans[offset].check == state
2233 && (tmp=trie->trans[offset].next))
2235 DEBUG_TRIE_EXECUTE_r(
2236 PerlIO_printf( Perl_debug_log," - legal\n"));
2241 DEBUG_TRIE_EXECUTE_r(
2242 PerlIO_printf( Perl_debug_log," - fail\n"));
2244 state = aho->fail[state];
2248 /* we must be accepting here */
2249 DEBUG_TRIE_EXECUTE_r(
2250 PerlIO_printf( Perl_debug_log," - accepting\n"));
2259 if (!state) state = 1;
2262 if ( aho->states[ state ].wordnum ) {
2263 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2264 if (!leftmost || lpos < leftmost) {
2265 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2270 s = (char*)leftmost;
2271 DEBUG_TRIE_EXECUTE_r({
2273 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2274 (UV)accepted_word, (IV)(s - real_start)
2277 if (reginfo->intuit || regtry(reginfo, &s)) {
2283 DEBUG_TRIE_EXECUTE_r({
2284 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2287 DEBUG_TRIE_EXECUTE_r(
2288 PerlIO_printf( Perl_debug_log,"No match.\n"));
2297 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2304 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2305 * flags have same meanings as with regexec_flags() */
2308 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2315 struct regexp *const prog = ReANY(rx);
2317 if (flags & REXEC_COPY_STR) {
2321 PerlIO_printf(Perl_debug_log,
2322 "Copy on write: regexp capture, type %d\n",
2325 /* Create a new COW SV to share the match string and store
2326 * in saved_copy, unless the current COW SV in saved_copy
2327 * is valid and suitable for our purpose */
2328 if (( prog->saved_copy
2329 && SvIsCOW(prog->saved_copy)
2330 && SvPOKp(prog->saved_copy)
2333 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2335 /* just reuse saved_copy SV */
2336 if (RXp_MATCH_COPIED(prog)) {
2337 Safefree(prog->subbeg);
2338 RXp_MATCH_COPIED_off(prog);
2342 /* create new COW SV to share string */
2343 RX_MATCH_COPY_FREE(rx);
2344 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2346 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2347 assert (SvPOKp(prog->saved_copy));
2348 prog->sublen = strend - strbeg;
2349 prog->suboffset = 0;
2350 prog->subcoffset = 0;
2355 SSize_t max = strend - strbeg;
2358 if ( (flags & REXEC_COPY_SKIP_POST)
2359 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2360 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2361 ) { /* don't copy $' part of string */
2364 /* calculate the right-most part of the string covered
2365 * by a capture. Due to look-ahead, this may be to
2366 * the right of $&, so we have to scan all captures */
2367 while (n <= prog->lastparen) {
2368 if (prog->offs[n].end > max)
2369 max = prog->offs[n].end;
2373 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2374 ? prog->offs[0].start
2376 assert(max >= 0 && max <= strend - strbeg);
2379 if ( (flags & REXEC_COPY_SKIP_PRE)
2380 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2381 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2382 ) { /* don't copy $` part of string */
2385 /* calculate the left-most part of the string covered
2386 * by a capture. Due to look-behind, this may be to
2387 * the left of $&, so we have to scan all captures */
2388 while (min && n <= prog->lastparen) {
2389 if ( prog->offs[n].start != -1
2390 && prog->offs[n].start < min)
2392 min = prog->offs[n].start;
2396 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2397 && min > prog->offs[0].end
2399 min = prog->offs[0].end;
2403 assert(min >= 0 && min <= max && min <= strend - strbeg);
2406 if (RX_MATCH_COPIED(rx)) {
2407 if (sublen > prog->sublen)
2409 (char*)saferealloc(prog->subbeg, sublen+1);
2412 prog->subbeg = (char*)safemalloc(sublen+1);
2413 Copy(strbeg + min, prog->subbeg, sublen, char);
2414 prog->subbeg[sublen] = '\0';
2415 prog->suboffset = min;
2416 prog->sublen = sublen;
2417 RX_MATCH_COPIED_on(rx);
2419 prog->subcoffset = prog->suboffset;
2420 if (prog->suboffset && utf8_target) {
2421 /* Convert byte offset to chars.
2422 * XXX ideally should only compute this if @-/@+
2423 * has been seen, a la PL_sawampersand ??? */
2425 /* If there's a direct correspondence between the
2426 * string which we're matching and the original SV,
2427 * then we can use the utf8 len cache associated with
2428 * the SV. In particular, it means that under //g,
2429 * sv_pos_b2u() will use the previously cached
2430 * position to speed up working out the new length of
2431 * subcoffset, rather than counting from the start of
2432 * the string each time. This stops
2433 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2434 * from going quadratic */
2435 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2436 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2437 SV_GMAGIC|SV_CONST_RETURN);
2439 prog->subcoffset = utf8_length((U8*)strbeg,
2440 (U8*)(strbeg+prog->suboffset));
2444 RX_MATCH_COPY_FREE(rx);
2445 prog->subbeg = strbeg;
2446 prog->suboffset = 0;
2447 prog->subcoffset = 0;
2448 prog->sublen = strend - strbeg;
2456 - regexec_flags - match a regexp against a string
2459 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2460 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2461 /* stringarg: the point in the string at which to begin matching */
2462 /* strend: pointer to null at end of string */
2463 /* strbeg: real beginning of string */
2464 /* minend: end of match must be >= minend bytes after stringarg. */
2465 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2466 * itself is accessed via the pointers above */
2467 /* data: May be used for some additional optimizations.
2468 Currently unused. */
2469 /* flags: For optimizations. See REXEC_* in regexp.h */
2472 struct regexp *const prog = ReANY(rx);
2476 SSize_t minlen; /* must match at least this many chars */
2477 SSize_t dontbother = 0; /* how many characters not to try at end */
2478 const bool utf8_target = cBOOL(DO_UTF8(sv));
2480 RXi_GET_DECL(prog,progi);
2481 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2482 regmatch_info *const reginfo = ®info_buf;
2483 regexp_paren_pair *swap = NULL;
2485 GET_RE_DEBUG_FLAGS_DECL;
2487 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2488 PERL_UNUSED_ARG(data);
2490 /* Be paranoid... */
2491 if (prog == NULL || stringarg == NULL) {
2492 Perl_croak(aTHX_ "NULL regexp parameter");
2496 debug_start_match(rx, utf8_target, stringarg, strend,
2500 startpos = stringarg;
2502 if (prog->intflags & PREGf_GPOS_SEEN) {
2505 /* set reginfo->ganch, the position where \G can match */
2508 (flags & REXEC_IGNOREPOS)
2509 ? stringarg /* use start pos rather than pos() */
2510 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2511 /* Defined pos(): */
2512 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2513 : strbeg; /* pos() not defined; use start of string */
2515 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2516 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
2518 /* in the presence of \G, we may need to start looking earlier in
2519 * the string than the suggested start point of stringarg:
2520 * if prog->gofs is set, then that's a known, fixed minimum
2523 * /ab|c\G/: gofs = 1
2524 * or if the minimum offset isn't known, then we have to go back
2525 * to the start of the string, e.g. /w+\G/
2528 if (prog->intflags & PREGf_ANCH_GPOS) {
2529 startpos = reginfo->ganch - prog->gofs;
2531 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2533 DEBUG_r(PerlIO_printf(Perl_debug_log,
2534 "fail: ganch-gofs before earliest possible start\n"));
2538 else if (prog->gofs) {
2539 if (startpos - prog->gofs < strbeg)
2542 startpos -= prog->gofs;
2544 else if (prog->intflags & PREGf_GPOS_FLOAT)
2548 minlen = prog->minlen;
2549 if ((startpos + minlen) > strend || startpos < strbeg) {
2550 DEBUG_r(PerlIO_printf(Perl_debug_log,
2551 "Regex match can't succeed, so not even tried\n"));
2555 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2556 * which will call destuctors to reset PL_regmatch_state, free higher
2557 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2558 * regmatch_info_aux_eval */
2560 oldsave = PL_savestack_ix;
2564 if ((prog->extflags & RXf_USE_INTUIT)
2565 && !(flags & REXEC_CHECKED))
2567 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2572 if (prog->extflags & RXf_CHECK_ALL) {
2573 /* we can match based purely on the result of INTUIT.
2574 * Set up captures etc just for $& and $-[0]
2575 * (an intuit-only match wont have $1,$2,..) */
2576 assert(!prog->nparens);
2578 /* s/// doesn't like it if $& is earlier than where we asked it to
2579 * start searching (which can happen on something like /.\G/) */
2580 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2583 /* this should only be possible under \G */
2584 assert(prog->intflags & PREGf_GPOS_SEEN);
2585 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2586 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2590 /* match via INTUIT shouldn't have any captures.
2591 * Let @-, @+, $^N know */
2592 prog->lastparen = prog->lastcloseparen = 0;
2593 RX_MATCH_UTF8_set(rx, utf8_target);
2594 prog->offs[0].start = s - strbeg;
2595 prog->offs[0].end = utf8_target
2596 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2597 : s - strbeg + prog->minlenret;
2598 if ( !(flags & REXEC_NOT_FIRST) )
2599 S_reg_set_capture_string(aTHX_ rx,
2601 sv, flags, utf8_target);
2607 multiline = prog->extflags & RXf_PMf_MULTILINE;
2609 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2610 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2611 "String too short [regexec_flags]...\n"));
2615 /* Check validity of program. */
2616 if (UCHARAT(progi->program) != REG_MAGIC) {
2617 Perl_croak(aTHX_ "corrupted regexp program");
2620 RX_MATCH_TAINTED_off(rx);
2621 RX_MATCH_UTF8_set(rx, utf8_target);
2623 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2624 reginfo->intuit = 0;
2625 reginfo->is_utf8_target = cBOOL(utf8_target);
2626 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2627 reginfo->warned = FALSE;
2628 reginfo->strbeg = strbeg;
2630 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2631 reginfo->strend = strend;
2632 /* see how far we have to get to not match where we matched before */
2633 reginfo->till = stringarg + minend;
2635 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
2636 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2637 S_cleanup_regmatch_info_aux has executed (registered by
2638 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2639 magic belonging to this SV.
2640 Not newSVsv, either, as it does not COW.
2642 reginfo->sv = newSV(0);
2643 SvSetSV_nosteal(reginfo->sv, sv);
2644 SAVEFREESV(reginfo->sv);
2647 /* reserve next 2 or 3 slots in PL_regmatch_state:
2648 * slot N+0: may currently be in use: skip it
2649 * slot N+1: use for regmatch_info_aux struct
2650 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2651 * slot N+3: ready for use by regmatch()
2655 regmatch_state *old_regmatch_state;
2656 regmatch_slab *old_regmatch_slab;
2657 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2659 /* on first ever match, allocate first slab */
2660 if (!PL_regmatch_slab) {
2661 Newx(PL_regmatch_slab, 1, regmatch_slab);
2662 PL_regmatch_slab->prev = NULL;
2663 PL_regmatch_slab->next = NULL;
2664 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2667 old_regmatch_state = PL_regmatch_state;
2668 old_regmatch_slab = PL_regmatch_slab;
2670 for (i=0; i <= max; i++) {
2672 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2674 reginfo->info_aux_eval =
2675 reginfo->info_aux->info_aux_eval =
2676 &(PL_regmatch_state->u.info_aux_eval);
2678 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2679 PL_regmatch_state = S_push_slab(aTHX);
2682 /* note initial PL_regmatch_state position; at end of match we'll
2683 * pop back to there and free any higher slabs */
2685 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2686 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2687 reginfo->info_aux->poscache = NULL;
2689 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2691 if ((prog->extflags & RXf_EVAL_SEEN))
2692 S_setup_eval_state(aTHX_ reginfo);
2694 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2697 /* If there is a "must appear" string, look for it. */
2699 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2700 /* We have to be careful. If the previous successful match
2701 was from this regex we don't want a subsequent partially
2702 successful match to clobber the old results.
2703 So when we detect this possibility we add a swap buffer
2704 to the re, and switch the buffer each match. If we fail,
2705 we switch it back; otherwise we leave it swapped.
2708 /* do we need a save destructor here for eval dies? */
2709 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2710 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2711 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2718 /* Simplest case: anchored match need be tried only once. */
2719 /* [unless only anchor is MBOL - implying multiline is set] */
2720 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
2721 if (s == startpos && regtry(reginfo, &s))
2723 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
2728 dontbother = minlen - 1;
2729 end = HOP3c(strend, -dontbother, strbeg) - 1;
2730 /* for multiline we only have to try after newlines */
2731 if (prog->check_substr || prog->check_utf8) {
2732 /* because of the goto we can not easily reuse the macros for bifurcating the
2733 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2736 goto after_try_utf8;
2738 if (regtry(reginfo, &s)) {
2745 if (prog->extflags & RXf_USE_INTUIT) {
2746 s = re_intuit_start(rx, sv, strbeg,
2747 s + UTF8SKIP(s), strend, flags, NULL);
2756 } /* end search for check string in unicode */
2758 if (s == startpos) {
2759 goto after_try_latin;
2762 if (regtry(reginfo, &s)) {
2769 if (prog->extflags & RXf_USE_INTUIT) {
2770 s = re_intuit_start(rx, sv, strbeg,
2771 s + 1, strend, flags, NULL);
2780 } /* end search for check string in latin*/
2781 } /* end search for check string */
2782 else { /* search for newline */
2784 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2787 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2788 while (s <= end) { /* note it could be possible to match at the end of the string */
2789 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2790 if (regtry(reginfo, &s))
2794 } /* end search for newline */
2795 } /* end anchored/multiline check string search */
2797 } else if (prog->intflags & PREGf_ANCH_GPOS)
2799 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2800 assert(prog->intflags & PREGf_GPOS_SEEN);
2801 /* For anchored \G, the only position it can match from is
2802 * (ganch-gofs); we already set startpos to this above; if intuit
2803 * moved us on from there, we can't possibly succeed */
2804 assert(startpos == reginfo->ganch - prog->gofs);
2805 if (s == startpos && regtry(reginfo, &s))
2810 /* Messy cases: unanchored match. */
2811 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2812 /* we have /x+whatever/ */
2813 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2819 if (! prog->anchored_utf8) {
2820 to_utf8_substr(prog);
2822 ch = SvPVX_const(prog->anchored_utf8)[0];
2825 DEBUG_EXECUTE_r( did_match = 1 );
2826 if (regtry(reginfo, &s)) goto got_it;
2828 while (s < strend && *s == ch)
2835 if (! prog->anchored_substr) {
2836 if (! to_byte_substr(prog)) {
2837 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2840 ch = SvPVX_const(prog->anchored_substr)[0];
2843 DEBUG_EXECUTE_r( did_match = 1 );
2844 if (regtry(reginfo, &s)) goto got_it;
2846 while (s < strend && *s == ch)
2851 DEBUG_EXECUTE_r(if (!did_match)
2852 PerlIO_printf(Perl_debug_log,
2853 "Did not find anchored character...\n")
2856 else if (prog->anchored_substr != NULL
2857 || prog->anchored_utf8 != NULL
2858 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2859 && prog->float_max_offset < strend - s)) {
2864 char *last1; /* Last position checked before */
2868 if (prog->anchored_substr || prog->anchored_utf8) {
2870 if (! prog->anchored_utf8) {
2871 to_utf8_substr(prog);
2873 must = prog->anchored_utf8;
2876 if (! prog->anchored_substr) {
2877 if (! to_byte_substr(prog)) {
2878 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2881 must = prog->anchored_substr;
2883 back_max = back_min = prog->anchored_offset;
2886 if (! prog->float_utf8) {
2887 to_utf8_substr(prog);
2889 must = prog->float_utf8;
2892 if (! prog->float_substr) {
2893 if (! to_byte_substr(prog)) {
2894 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2897 must = prog->float_substr;
2899 back_max = prog->float_max_offset;
2900 back_min = prog->float_min_offset;
2906 last = HOP3c(strend, /* Cannot start after this */
2907 -(SSize_t)(CHR_SVLEN(must)
2908 - (SvTAIL(must) != 0) + back_min), strbeg);
2910 if (s > reginfo->strbeg)
2911 last1 = HOPc(s, -1);
2913 last1 = s - 1; /* bogus */
2915 /* XXXX check_substr already used to find "s", can optimize if
2916 check_substr==must. */
2918 strend = HOPc(strend, -dontbother);
2919 while ( (s <= last) &&
2920 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
2921 (unsigned char*)strend, must,
2922 multiline ? FBMrf_MULTILINE : 0)) ) {
2923 DEBUG_EXECUTE_r( did_match = 1 );
2924 if (HOPc(s, -back_max) > last1) {
2925 last1 = HOPc(s, -back_min);
2926 s = HOPc(s, -back_max);
2929 char * const t = (last1 >= reginfo->strbeg)
2930 ? HOPc(last1, 1) : last1 + 1;
2932 last1 = HOPc(s, -back_min);
2936 while (s <= last1) {
2937 if (regtry(reginfo, &s))
2940 s++; /* to break out of outer loop */
2947 while (s <= last1) {
2948 if (regtry(reginfo, &s))
2954 DEBUG_EXECUTE_r(if (!did_match) {
2955 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2956 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2957 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2958 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2959 ? "anchored" : "floating"),
2960 quoted, RE_SV_TAIL(must));
2964 else if ( (c = progi->regstclass) ) {
2966 const OPCODE op = OP(progi->regstclass);
2967 /* don't bother with what can't match */
2968 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2969 strend = HOPc(strend, -(minlen - 1));
2972 SV * const prop = sv_newmortal();
2973 regprop(prog, prop, c, reginfo, NULL);
2975 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2977 PerlIO_printf(Perl_debug_log,
2978 "Matching stclass %.*s against %s (%d bytes)\n",
2979 (int)SvCUR(prop), SvPVX_const(prop),
2980 quoted, (int)(strend - s));
2983 if (find_byclass(prog, c, s, strend, reginfo))
2985 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2989 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2997 if (! prog->float_utf8) {
2998 to_utf8_substr(prog);
3000 float_real = prog->float_utf8;
3003 if (! prog->float_substr) {
3004 if (! to_byte_substr(prog)) {
3005 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3008 float_real = prog->float_substr;
3011 little = SvPV_const(float_real, len);
3012 if (SvTAIL(float_real)) {
3013 /* This means that float_real contains an artificial \n on
3014 * the end due to the presence of something like this:
3015 * /foo$/ where we can match both "foo" and "foo\n" at the
3016 * end of the string. So we have to compare the end of the
3017 * string first against the float_real without the \n and
3018 * then against the full float_real with the string. We
3019 * have to watch out for cases where the string might be
3020 * smaller than the float_real or the float_real without
3022 char *checkpos= strend - len;
3024 PerlIO_printf(Perl_debug_log,
3025 "%sChecking for float_real.%s\n",
3026 PL_colors[4], PL_colors[5]));
3027 if (checkpos + 1 < strbeg) {
3028 /* can't match, even if we remove the trailing \n
3029 * string is too short to match */
3031 PerlIO_printf(Perl_debug_log,
3032 "%sString shorter than required trailing substring, cannot match.%s\n",
3033 PL_colors[4], PL_colors[5]));
3035 } else if (memEQ(checkpos + 1, little, len - 1)) {
3036 /* can match, the end of the string matches without the
3038 last = checkpos + 1;
3039 } else if (checkpos < strbeg) {
3040 /* cant match, string is too short when the "\n" is
3043 PerlIO_printf(Perl_debug_log,
3044 "%sString does not contain required trailing substring, cannot match.%s\n",
3045 PL_colors[4], PL_colors[5]));
3047 } else if (!multiline) {
3048 /* non multiline match, so compare with the "\n" at the
3049 * end of the string */
3050 if (memEQ(checkpos, little, len)) {
3054 PerlIO_printf(Perl_debug_log,
3055 "%sString does not contain required trailing substring, cannot match.%s\n",
3056 PL_colors[4], PL_colors[5]));
3060 /* multiline match, so we have to search for a place
3061 * where the full string is located */
3067 last = rninstr(s, strend, little, little + len);
3069 last = strend; /* matching "$" */
3072 /* at one point this block contained a comment which was
3073 * probably incorrect, which said that this was a "should not
3074 * happen" case. Even if it was true when it was written I am
3075 * pretty sure it is not anymore, so I have removed the comment
3076 * and replaced it with this one. Yves */
3078 PerlIO_printf(Perl_debug_log,
3079 "%sString does not contain required substring, cannot match.%s\n",
3080 PL_colors[4], PL_colors[5]
3084 dontbother = strend - last + prog->float_min_offset;
3086 if (minlen && (dontbother < minlen))
3087 dontbother = minlen - 1;
3088 strend -= dontbother; /* this one's always in bytes! */
3089 /* We don't know much -- general case. */
3092 if (regtry(reginfo, &s))
3101 if (regtry(reginfo, &s))
3103 } while (s++ < strend);
3111 /* s/// doesn't like it if $& is earlier than where we asked it to
3112 * start searching (which can happen on something like /.\G/) */
3113 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3114 && (prog->offs[0].start < stringarg - strbeg))
3116 /* this should only be possible under \G */
3117 assert(prog->intflags & PREGf_GPOS_SEEN);
3118 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3119 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3125 PerlIO_printf(Perl_debug_log,
3126 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3133 /* clean up; this will trigger destructors that will free all slabs
3134 * above the current one, and cleanup the regmatch_info_aux
3135 * and regmatch_info_aux_eval sructs */
3137 LEAVE_SCOPE(oldsave);
3139 if (RXp_PAREN_NAMES(prog))
3140 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3142 /* make sure $`, $&, $', and $digit will work later */
3143 if ( !(flags & REXEC_NOT_FIRST) )
3144 S_reg_set_capture_string(aTHX_ rx,
3145 strbeg, reginfo->strend,
3146 sv, flags, utf8_target);
3151 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
3152 PL_colors[4], PL_colors[5]));
3154 /* clean up; this will trigger destructors that will free all slabs
3155 * above the current one, and cleanup the regmatch_info_aux
3156 * and regmatch_info_aux_eval sructs */
3158 LEAVE_SCOPE(oldsave);
3161 /* we failed :-( roll it back */
3162 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3163 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3168 Safefree(prog->offs);
3175 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3176 * Do inc before dec, in case old and new rex are the same */
3177 #define SET_reg_curpm(Re2) \
3178 if (reginfo->info_aux_eval) { \
3179 (void)ReREFCNT_inc(Re2); \
3180 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3181 PM_SETRE((PL_reg_curpm), (Re2)); \
3186 - regtry - try match at specific point
3188 STATIC I32 /* 0 failure, 1 success */
3189 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3192 REGEXP *const rx = reginfo->prog;
3193 regexp *const prog = ReANY(rx);
3195 RXi_GET_DECL(prog,progi);
3196 GET_RE_DEBUG_FLAGS_DECL;
3198 PERL_ARGS_ASSERT_REGTRY;
3200 reginfo->cutpoint=NULL;
3202 prog->offs[0].start = *startposp - reginfo->strbeg;
3203 prog->lastparen = 0;
3204 prog->lastcloseparen = 0;
3206 /* XXXX What this code is doing here?!!! There should be no need
3207 to do this again and again, prog->lastparen should take care of
3210 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3211 * Actually, the code in regcppop() (which Ilya may be meaning by
3212 * prog->lastparen), is not needed at all by the test suite
3213 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3214 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3215 * Meanwhile, this code *is* needed for the
3216 * above-mentioned test suite tests to succeed. The common theme
3217 * on those tests seems to be returning null fields from matches.
3218 * --jhi updated by dapm */
3220 if (prog->nparens) {
3221 regexp_paren_pair *pp = prog->offs;
3223 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3231 result = regmatch(reginfo, *startposp, progi->program + 1);
3233 prog->offs[0].end = result;
3236 if (reginfo->cutpoint)
3237 *startposp= reginfo->cutpoint;
3238 REGCP_UNWIND(lastcp);
3243 #define sayYES goto yes
3244 #define sayNO goto no
3245 #define sayNO_SILENT goto no_silent
3247 /* we dont use STMT_START/END here because it leads to
3248 "unreachable code" warnings, which are bogus, but distracting. */
3249 #define CACHEsayNO \
3250 if (ST.cache_mask) \
3251 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3254 /* this is used to determine how far from the left messages like
3255 'failed...' are printed. It should be set such that messages
3256 are inline with the regop output that created them.
3258 #define REPORT_CODE_OFF 32
3261 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3262 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3263 #define CHRTEST_NOT_A_CP_1 -999
3264 #define CHRTEST_NOT_A_CP_2 -998
3266 /* grab a new slab and return the first slot in it */
3268 STATIC regmatch_state *
3271 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3274 regmatch_slab *s = PL_regmatch_slab->next;
3276 Newx(s, 1, regmatch_slab);
3277 s->prev = PL_regmatch_slab;
3279 PL_regmatch_slab->next = s;
3281 PL_regmatch_slab = s;
3282 return SLAB_FIRST(s);
3286 /* push a new state then goto it */
3288 #define PUSH_STATE_GOTO(state, node, input) \
3289 pushinput = input; \
3291 st->resume_state = state; \
3294 /* push a new state with success backtracking, then goto it */
3296 #define PUSH_YES_STATE_GOTO(state, node, input) \
3297 pushinput = input; \
3299 st->resume_state = state; \
3300 goto push_yes_state;
3307 regmatch() - main matching routine
3309 This is basically one big switch statement in a loop. We execute an op,
3310 set 'next' to point the next op, and continue. If we come to a point which
3311 we may need to backtrack to on failure such as (A|B|C), we push a
3312 backtrack state onto the backtrack stack. On failure, we pop the top
3313 state, and re-enter the loop at the state indicated. If there are no more
3314 states to pop, we return failure.
3316 Sometimes we also need to backtrack on success; for example /A+/, where
3317 after successfully matching one A, we need to go back and try to
3318 match another one; similarly for lookahead assertions: if the assertion
3319 completes successfully, we backtrack to the state just before the assertion
3320 and then carry on. In these cases, the pushed state is marked as
3321 'backtrack on success too'. This marking is in fact done by a chain of
3322 pointers, each pointing to the previous 'yes' state. On success, we pop to
3323 the nearest yes state, discarding any intermediate failure-only states.
3324 Sometimes a yes state is pushed just to force some cleanup code to be
3325 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3326 it to free the inner regex.
3328 Note that failure backtracking rewinds the cursor position, while
3329 success backtracking leaves it alone.
3331 A pattern is complete when the END op is executed, while a subpattern
3332 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3333 ops trigger the "pop to last yes state if any, otherwise return true"
3336 A common convention in this function is to use A and B to refer to the two
3337 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3338 the subpattern to be matched possibly multiple times, while B is the entire
3339 rest of the pattern. Variable and state names reflect this convention.
3341 The states in the main switch are the union of ops and failure/success of
3342 substates associated with with that op. For example, IFMATCH is the op
3343 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3344 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3345 successfully matched A and IFMATCH_A_fail is a state saying that we have
3346 just failed to match A. Resume states always come in pairs. The backtrack
3347 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3348 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3349 on success or failure.
3351 The struct that holds a backtracking state is actually a big union, with
3352 one variant for each major type of op. The variable st points to the
3353 top-most backtrack struct. To make the code clearer, within each
3354 block of code we #define ST to alias the relevant union.
3356 Here's a concrete example of a (vastly oversimplified) IFMATCH
3362 #define ST st->u.ifmatch
3364 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3365 ST.foo = ...; // some state we wish to save
3367 // push a yes backtrack state with a resume value of
3368 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3370 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3373 case IFMATCH_A: // we have successfully executed A; now continue with B
3375 bar = ST.foo; // do something with the preserved value
3378 case IFMATCH_A_fail: // A failed, so the assertion failed
3379 ...; // do some housekeeping, then ...
3380 sayNO; // propagate the failure
3387 For any old-timers reading this who are familiar with the old recursive
3388 approach, the code above is equivalent to:
3390 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3399 ...; // do some housekeeping, then ...
3400 sayNO; // propagate the failure
3403 The topmost backtrack state, pointed to by st, is usually free. If you
3404 want to claim it, populate any ST.foo fields in it with values you wish to
3405 save, then do one of
3407 PUSH_STATE_GOTO(resume_state, node, newinput);
3408 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3410 which sets that backtrack state's resume value to 'resume_state', pushes a
3411 new free entry to the top of the backtrack stack, then goes to 'node'.
3412 On backtracking, the free slot is popped, and the saved state becomes the
3413 new free state. An ST.foo field in this new top state can be temporarily
3414 accessed to retrieve values, but once the main loop is re-entered, it
3415 becomes available for reuse.
3417 Note that the depth of the backtrack stack constantly increases during the
3418 left-to-right execution of the pattern, rather than going up and down with
3419 the pattern nesting. For example the stack is at its maximum at Z at the
3420 end of the pattern, rather than at X in the following:
3422 /(((X)+)+)+....(Y)+....Z/
3424 The only exceptions to this are lookahead/behind assertions and the cut,
3425 (?>A), which pop all the backtrack states associated with A before
3428 Backtrack state structs are allocated in slabs of about 4K in size.
3429 PL_regmatch_state and st always point to the currently active state,
3430 and PL_regmatch_slab points to the slab currently containing
3431 PL_regmatch_state. The first time regmatch() is called, the first slab is
3432 allocated, and is never freed until interpreter destruction. When the slab
3433 is full, a new one is allocated and chained to the end. At exit from
3434 regmatch(), slabs allocated since entry are freed.
3439 #define DEBUG_STATE_pp(pp) \
3441 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3442 PerlIO_printf(Perl_debug_log, \
3443 " %*s"pp" %s%s%s%s%s\n", \
3445 PL_reg_name[st->resume_state], \
3446 ((st==yes_state||st==mark_state) ? "[" : ""), \
3447 ((st==yes_state) ? "Y" : ""), \
3448 ((st==mark_state) ? "M" : ""), \
3449 ((st==yes_state||st==mark_state) ? "]" : "") \
3454 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3459 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3460 const char *start, const char *end, const char *blurb)
3462 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3464 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3469 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3470 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3472 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3473 start, end - start, 60);
3475 PerlIO_printf(Perl_debug_log,
3476 "%s%s REx%s %s against %s\n",
3477 PL_colors[4], blurb, PL_colors[5], s0, s1);
3479 if (utf8_target||utf8_pat)
3480 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3481 utf8_pat ? "pattern" : "",
3482 utf8_pat && utf8_target ? " and " : "",
3483 utf8_target ? "string" : ""
3489 S_dump_exec_pos(pTHX_ const char *locinput,
3490 const regnode *scan,
3491 const char *loc_regeol,
3492 const char *loc_bostr,
3493 const char *loc_reg_starttry,
3494 const bool utf8_target)
3496 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3497 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3498 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3499 /* The part of the string before starttry has one color
3500 (pref0_len chars), between starttry and current
3501 position another one (pref_len - pref0_len chars),
3502 after the current position the third one.
3503 We assume that pref0_len <= pref_len, otherwise we
3504 decrease pref0_len. */
3505 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3506 ? (5 + taill) - l : locinput - loc_bostr;
3509 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3511 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3513 pref0_len = pref_len - (locinput - loc_reg_starttry);
3514 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3515 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3516 ? (5 + taill) - pref_len : loc_regeol - locinput);
3517 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3521 if (pref0_len > pref_len)
3522 pref0_len = pref_len;
3524 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3526 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3527 (locinput - pref_len),pref0_len, 60, 4, 5);
3529 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3530 (locinput - pref_len + pref0_len),
3531 pref_len - pref0_len, 60, 2, 3);
3533 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3534 locinput, loc_regeol - locinput, 10, 0, 1);
3536 const STRLEN tlen=len0+len1+len2;
3537 PerlIO_printf(Perl_debug_log,
3538 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3539 (IV)(locinput - loc_bostr),
3542 (docolor ? "" : "> <"),
3544 (int)(tlen > 19 ? 0 : 19 - tlen),
3551 /* reg_check_named_buff_matched()
3552 * Checks to see if a named buffer has matched. The data array of
3553 * buffer numbers corresponding to the buffer is expected to reside
3554 * in the regexp->data->data array in the slot stored in the ARG() of
3555 * node involved. Note that this routine doesn't actually care about the
3556 * name, that information is not preserved from compilation to execution.
3557 * Returns the index of the leftmost defined buffer with the given name
3558 * or 0 if non of the buffers matched.
3561 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3564 RXi_GET_DECL(rex,rexi);
3565 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3566 I32 *nums=(I32*)SvPVX(sv_dat);
3568 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3570 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3571 if ((I32)rex->lastparen >= nums[n] &&
3572 rex->offs[nums[n]].end != -1)
3582 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3583 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3585 /* This function determines if there are one or two characters that match
3586 * the first character of the passed-in EXACTish node <text_node>, and if
3587 * so, returns them in the passed-in pointers.
3589 * If it determines that no possible character in the target string can
3590 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3591 * the first character in <text_node> requires UTF-8 to represent, and the
3592 * target string isn't in UTF-8.)
3594 * If there are more than two characters that could match the beginning of
3595 * <text_node>, or if more context is required to determine a match or not,
3596 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3598 * The motiviation behind this function is to allow the caller to set up
3599 * tight loops for matching. If <text_node> is of type EXACT, there is
3600 * only one possible character that can match its first character, and so
3601 * the situation is quite simple. But things get much more complicated if
3602 * folding is involved. It may be that the first character of an EXACTFish
3603 * node doesn't participate in any possible fold, e.g., punctuation, so it
3604 * can be matched only by itself. The vast majority of characters that are
3605 * in folds match just two things, their lower and upper-case equivalents.
3606 * But not all are like that; some have multiple possible matches, or match
3607 * sequences of more than one character. This function sorts all that out.
3609 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3610 * loop of trying to match A*, we know we can't exit where the thing
3611 * following it isn't a B. And something can't be a B unless it is the
3612 * beginning of B. By putting a quick test for that beginning in a tight
3613 * loop, we can rule out things that can't possibly be B without having to
3614 * break out of the loop, thus avoiding work. Similarly, if A is a single
3615 * character, we can make a tight loop matching A*, using the outputs of
3618 * If the target string to match isn't in UTF-8, and there aren't
3619 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3620 * the one or two possible octets (which are characters in this situation)
3621 * that can match. In all cases, if there is only one character that can
3622 * match, *<c1p> and *<c2p> will be identical.
3624 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3625 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3626 * can match the beginning of <text_node>. They should be declared with at
3627 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3628 * undefined what these contain.) If one or both of the buffers are
3629 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3630 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3631 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3632 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3633 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3635 const bool utf8_target = reginfo->is_utf8_target;
3637 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
3638 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
3639 bool use_chrtest_void = FALSE;
3640 const bool is_utf8_pat = reginfo->is_utf8_pat;
3642 /* Used when we have both utf8 input and utf8 output, to avoid converting
3643 * to/from code points */
3644 bool utf8_has_been_setup = FALSE;
3648 U8 *pat = (U8*)STRING(text_node);
3649 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
3651 if (OP(text_node) == EXACT) {
3653 /* In an exact node, only one thing can be matched, that first
3654 * character. If both the pat and the target are UTF-8, we can just
3655 * copy the input to the output, avoiding finding the code point of
3660 else if (utf8_target) {
3661 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3662 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3663 utf8_has_been_setup = TRUE;
3666 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3669 else { /* an EXACTFish node */
3670 U8 *pat_end = pat + STR_LEN(text_node);
3672 /* An EXACTFL node has at least some characters unfolded, because what
3673 * they match is not known until now. So, now is the time to fold
3674 * the first few of them, as many as are needed to determine 'c1' and
3675 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
3676 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
3677 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
3678 * need to fold as many characters as a single character can fold to,
3679 * so that later we can check if the first ones are such a multi-char
3680 * fold. But, in such a pattern only locale-problematic characters
3681 * aren't folded, so we can skip this completely if the first character
3682 * in the node isn't one of the tricky ones */
3683 if (OP(text_node) == EXACTFL) {
3685 if (! is_utf8_pat) {
3686 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
3688 folded[0] = folded[1] = 's';
3690 pat_end = folded + 2;
3693 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
3698 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
3700 *(d++) = (U8) toFOLD_LC(*s);
3705 _to_utf8_fold_flags(s,
3708 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
3719 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
3720 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
3722 /* Multi-character folds require more context to sort out. Also
3723 * PL_utf8_foldclosures used below doesn't handle them, so have to
3724 * be handled outside this routine */
3725 use_chrtest_void = TRUE;
3727 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3728 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3730 /* Load the folds hash, if not already done */
3732 if (! PL_utf8_foldclosures) {
3733 _load_PL_utf8_foldclosures();
3736 /* The fold closures data structure is a hash with the keys
3737 * being the UTF-8 of every character that is folded to, like
3738 * 'k', and the values each an array of all code points that
3739 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
3740 * Multi-character folds are not included */
3741 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3746 /* Not found in the hash, therefore there are no folds
3747 * containing it, so there is only a single character that
3751 else { /* Does participate in folds */
3752 AV* list = (AV*) *listp;
3753 if (av_tindex(list) != 1) {
3755 /* If there aren't exactly two folds to this, it is
3756 * outside the scope of this function */
3757 use_chrtest_void = TRUE;
3759 else { /* There are two. Get them */
3760 SV** c_p = av_fetch(list, 0, FALSE);
3762 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3766 c_p = av_fetch(list, 1, FALSE);
3768 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3772 /* Folds that cross the 255/256 boundary are forbidden
3773 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
3774 * one is ASCIII. Since the pattern character is above
3775 * 255, and its only other match is below 256, the only
3776 * legal match will be to itself. We have thrown away
3777 * the original, so have to compute which is the one
3779 if ((c1 < 256) != (c2 < 256)) {
3780 if ((OP(text_node) == EXACTFL
3781 && ! IN_UTF8_CTYPE_LOCALE)
3782 || ((OP(text_node) == EXACTFA
3783 || OP(text_node) == EXACTFA_NO_TRIE)
3784 && (isASCII(c1) || isASCII(c2))))
3797 else /* Here, c1 is <= 255 */
3799 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3800 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
3801 && ((OP(text_node) != EXACTFA
3802 && OP(text_node) != EXACTFA_NO_TRIE)
3805 /* Here, there could be something above Latin1 in the target
3806 * which folds to this character in the pattern. All such
3807 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
3808 * than two characters involved in their folds, so are outside
3809 * the scope of this function */
3810 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3811 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3814 use_chrtest_void = TRUE;
3817 else { /* Here nothing above Latin1 can fold to the pattern
3819 switch (OP(text_node)) {
3821 case EXACTFL: /* /l rules */
3822 c2 = PL_fold_locale[c1];
3825 case EXACTF: /* This node only generated for non-utf8
3827 assert(! is_utf8_pat);
3828 if (! utf8_target) { /* /d rules */
3833 /* /u rules for all these. This happens to work for
3834 * EXACTFA as nothing in Latin1 folds to ASCII */
3835 case EXACTFA_NO_TRIE: /* This node only generated for
3836 non-utf8 patterns */
3837 assert(! is_utf8_pat);
3842 c2 = PL_fold_latin1[c1];
3846 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3847 NOT_REACHED; /* NOTREACHED */
3853 /* Here have figured things out. Set up the returns */
3854 if (use_chrtest_void) {
3855 *c2p = *c1p = CHRTEST_VOID;
3857 else if (utf8_target) {
3858 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3859 uvchr_to_utf8(c1_utf8, c1);
3860 uvchr_to_utf8(c2_utf8, c2);
3863 /* Invariants are stored in both the utf8 and byte outputs; Use
3864 * negative numbers otherwise for the byte ones. Make sure that the
3865 * byte ones are the same iff the utf8 ones are the same */
3866 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3867 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3870 ? CHRTEST_NOT_A_CP_1
3871 : CHRTEST_NOT_A_CP_2;
3873 else if (c1 > 255) {
3874 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3879 *c1p = *c2p = c2; /* c2 is the only representable value */
3881 else { /* c1 is representable; see about c2 */
3883 *c2p = (c2 < 256) ? c2 : c1;
3889 /* returns -1 on failure, $+[0] on success */
3891 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3893 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3897 const bool utf8_target = reginfo->is_utf8_target;
3898 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3899 REGEXP *rex_sv = reginfo->prog;
3900 regexp *rex = ReANY(rex_sv);
3901 RXi_GET_DECL(rex,rexi);
3902 /* the current state. This is a cached copy of PL_regmatch_state */
3904 /* cache heavy used fields of st in registers */
3907 U32 n = 0; /* general value; init to avoid compiler warning */
3908 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
3909 char *locinput = startpos;
3910 char *pushinput; /* where to continue after a PUSH */
3911 I32 nextchr; /* is always set to UCHARAT(locinput) */
3913 bool result = 0; /* return value of S_regmatch */
3914 int depth = 0; /* depth of backtrack stack */
3915 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3916 const U32 max_nochange_depth =
3917 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3918 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3919 regmatch_state *yes_state = NULL; /* state to pop to on success of
3921 /* mark_state piggy backs on the yes_state logic so that when we unwind
3922 the stack on success we can update the mark_state as we go */
3923 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3924 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3925 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3927 bool no_final = 0; /* prevent failure from backtracking? */
3928 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3929 char *startpoint = locinput;
3930 SV *popmark = NULL; /* are we looking for a mark? */
3931 SV *sv_commit = NULL; /* last mark name seen in failure */
3932 SV *sv_yes_mark = NULL; /* last mark name we have seen
3933 during a successful match */
3934 U32 lastopen = 0; /* last open we saw */
3935 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3936 SV* const oreplsv = GvSVn(PL_replgv);
3937 /* these three flags are set by various ops to signal information to
3938 * the very next op. They have a useful lifetime of exactly one loop
3939 * iteration, and are not preserved or restored by state pushes/pops
3941 bool sw = 0; /* the condition value in (?(cond)a|b) */
3942 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3943 int logical = 0; /* the following EVAL is:
3947 or the following IFMATCH/UNLESSM is:
3948 false: plain (?=foo)
3949 true: used as a condition: (?(?=foo))
3951 PAD* last_pad = NULL;
3953 I32 gimme = G_SCALAR;
3954 CV *caller_cv = NULL; /* who called us */
3955 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3956 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3957 U32 maxopenparen = 0; /* max '(' index seen so far */
3958 int to_complement; /* Invert the result? */
3959 _char_class_number classnum;
3960 bool is_utf8_pat = reginfo->is_utf8_pat;
3963 GET_RE_DEBUG_FLAGS_DECL;
3966 /* protect against undef(*^R) */
3967 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
3969 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3970 multicall_oldcatch = 0;
3971 multicall_cv = NULL;
3973 PERL_UNUSED_VAR(multicall_cop);
3974 PERL_UNUSED_VAR(newsp);
3977 PERL_ARGS_ASSERT_REGMATCH;
3979 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3980 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3983 st = PL_regmatch_state;
3985 /* Note that nextchr is a byte even in UTF */
3988 while (scan != NULL) {
3991 SV * const prop = sv_newmortal();
3992 regnode *rnext=regnext(scan);
3993 DUMP_EXEC_POS( locinput, scan, utf8_target );
3994 regprop(rex, prop, scan, reginfo, NULL);
3996 PerlIO_printf(Perl_debug_log,
3997 "%3"IVdf":%*s%s(%"IVdf")\n",
3998 (IV)(scan - rexi->program), depth*2, "",
4000 (PL_regkind[OP(scan)] == END || !rnext) ?
4001 0 : (IV)(rnext - rexi->program));
4004 next = scan + NEXT_OFF(scan);
4007 state_num = OP(scan);
4013 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
4015 switch (state_num) {
4016 case SBOL: /* /^../ and /\A../ */
4017 if (locinput == reginfo->strbeg)
4021 case MBOL: /* /^../m */
4022 if (locinput == reginfo->strbeg ||
4023 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
4030 if (locinput == reginfo->ganch)
4034 case KEEPS: /* \K */
4035 /* update the startpoint */
4036 st->u.keeper.val = rex->offs[0].start;
4037 rex->offs[0].start = locinput - reginfo->strbeg;
4038 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
4042 case KEEPS_next_fail:
4043 /* rollback the start point change */
4044 rex->offs[0].start = st->u.keeper.val;
4049 case MEOL: /* /..$/m */
4050 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4054 case SEOL: /* /..$/ */
4055 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4057 if (reginfo->strend - locinput > 1)
4062 if (!NEXTCHR_IS_EOS)
4066 case SANY: /* /./s */
4069 goto increment_locinput;
4077 case REG_ANY: /* /./ */
4078 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
4080 goto increment_locinput;
4084 #define ST st->u.trie
4085 case TRIEC: /* (ab|cd) with known charclass */
4086 /* In this case the charclass data is available inline so
4087 we can fail fast without a lot of extra overhead.
4089 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
4091 PerlIO_printf(Perl_debug_log,
4092 "%*s %sfailed to match trie start class...%s\n",
4093 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4100 case TRIE: /* (ab|cd) */
4101 /* the basic plan of execution of the trie is:
4102 * At the beginning, run though all the states, and
4103 * find the longest-matching word. Also remember the position
4104 * of the shortest matching word. For example, this pattern:
4107 * when matched against the string "abcde", will generate
4108 * accept states for all words except 3, with the longest
4109 * matching word being 4, and the shortest being 2 (with
4110 * the position being after char 1 of the string).
4112 * Then for each matching word, in word order (i.e. 1,2,4,5),
4113 * we run the remainder of the pattern; on each try setting
4114 * the current position to the character following the word,
4115 * returning to try the next word on failure.
4117 * We avoid having to build a list of words at runtime by
4118 * using a compile-time structure, wordinfo[].prev, which
4119 * gives, for each word, the previous accepting word (if any).
4120 * In the case above it would contain the mappings 1->2, 2->0,
4121 * 3->0, 4->5, 5->1. We can use this table to generate, from
4122 * the longest word (4 above), a list of all words, by
4123 * following the list of prev pointers; this gives us the
4124 * unordered list 4,5,1,2. Then given the current word we have
4125 * just tried, we can go through the list and find the
4126 * next-biggest word to try (so if we just failed on word 2,
4127 * the next in the list is 4).
4129 * Since at runtime we don't record the matching position in
4130 * the string for each word, we have to work that out for
4131 * each word we're about to process. The wordinfo table holds
4132 * the character length of each word; given that we recorded
4133 * at the start: the position of the shortest word and its
4134 * length in chars, we just need to move the pointer the
4135 * difference between the two char lengths. Depending on
4136 * Unicode status and folding, that's cheap or expensive.
4138 * This algorithm is optimised for the case where are only a
4139 * small number of accept states, i.e. 0,1, or maybe 2.
4140 * With lots of accepts states, and having to try all of them,
4141 * it becomes quadratic on number of accept states to find all
4146 /* what type of TRIE am I? (utf8 makes this contextual) */
4147 DECL_TRIE_TYPE(scan);
4149 /* what trie are we using right now */
4150 reg_trie_data * const trie
4151 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
4152 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
4153 U32 state = trie->startstate;
4156 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
4158 if (trie->states[ state ].wordnum) {
4160 PerlIO_printf(Perl_debug_log,
4161 "%*s %smatched empty string...%s\n",
4162 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4168 PerlIO_printf(Perl_debug_log,
4169 "%*s %sfailed to match trie start class...%s\n",
4170 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4177 U8 *uc = ( U8* )locinput;
4181 U8 *uscan = (U8*)NULL;
4182 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
4183 U32 charcount = 0; /* how many input chars we have matched */
4184 U32 accepted = 0; /* have we seen any accepting states? */
4186 ST.jump = trie->jump;
4189 ST.longfold = FALSE; /* char longer if folded => it's harder */
4192 /* fully traverse the TRIE; note the position of the
4193 shortest accept state and the wordnum of the longest
4196 while ( state && uc <= (U8*)(reginfo->strend) ) {
4197 U32 base = trie->states[ state ].trans.base;
4201 wordnum = trie->states[ state ].wordnum;
4203 if (wordnum) { /* it's an accept state */
4206 /* record first match position */
4208 ST.firstpos = (U8*)locinput;
4213 ST.firstchars = charcount;
4216 if (!ST.nextword || wordnum < ST.nextword)
4217 ST.nextword = wordnum;
4218 ST.topword = wordnum;
4221 DEBUG_TRIE_EXECUTE_r({
4222 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
4223 PerlIO_printf( Perl_debug_log,
4224 "%*s %sState: %4"UVxf" Accepted: %c ",
4225 2+depth * 2, "", PL_colors[4],
4226 (UV)state, (accepted ? 'Y' : 'N'));
4229 /* read a char and goto next state */
4230 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
4232 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
4233 uscan, len, uvc, charid, foldlen,
4240 base + charid - 1 - trie->uniquecharcount)) >= 0)
4242 && ((U32)offset < trie->lasttrans)
4243 && trie->trans[offset].check == state)
4245 state = trie->trans[offset].next;
4256 DEBUG_TRIE_EXECUTE_r(
4257 PerlIO_printf( Perl_debug_log,
4258 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
4259 charid, uvc, (UV)state, PL_colors[5] );
4265 /* calculate total number of accept states */
4270 w = trie->wordinfo[w].prev;
4273 ST.accepted = accepted;
4277 PerlIO_printf( Perl_debug_log,
4278 "%*s %sgot %"IVdf" possible matches%s\n",
4279 REPORT_CODE_OFF + depth * 2, "",
4280 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
4282 goto trie_first_try; /* jump into the fail handler */
4287 case TRIE_next_fail: /* we failed - try next alternative */
4291 REGCP_UNWIND(ST.cp);
4292 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
4294 if (!--ST.accepted) {
4296 PerlIO_printf( Perl_debug_log,
4297 "%*s %sTRIE failed...%s\n",
4298 REPORT_CODE_OFF+depth*2, "",
4305 /* Find next-highest word to process. Note that this code
4306 * is O(N^2) per trie run (O(N) per branch), so keep tight */
4309 U16 const nextword = ST.nextword;
4310 reg_trie_wordinfo * const wordinfo
4311 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
4312 for (word=ST.topword; word; word=wordinfo[word].prev) {
4313 if (word > nextword && (!min || word < min))
4326 ST.lastparen = rex->lastparen;
4327 ST.lastcloseparen = rex->lastcloseparen;
4331 /* find start char of end of current word */
4333 U32 chars; /* how many chars to skip */
4334 reg_trie_data * const trie
4335 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
4337 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
4339 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
4344 /* the hard option - fold each char in turn and find
4345 * its folded length (which may be different */
4346 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4354 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
4362 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4367 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
4383 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4384 ? ST.jump[ST.nextword]
4388 PerlIO_printf( Perl_debug_log,
4389 "%*s %sTRIE matched word #%d, continuing%s\n",
4390 REPORT_CODE_OFF+depth*2, "",
4397 if (ST.accepted > 1 || has_cutgroup) {
4398 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4402 /* only one choice left - just continue */
4404 AV *const trie_words
4405 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4406 SV ** const tmp = av_fetch( trie_words,
4408 SV *sv= tmp ? sv_newmortal() : NULL;
4410 PerlIO_printf( Perl_debug_log,
4411 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4412 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4414 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4415 PL_colors[0], PL_colors[1],
4416 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4418 : "not compiled under -Dr",
4422 locinput = (char*)uc;
4423 continue; /* execute rest of RE */
4428 case EXACT: { /* /abc/ */
4429 char *s = STRING(scan);
4431 if (utf8_target != is_utf8_pat) {
4432 /* The target and the pattern have differing utf8ness. */
4434 const char * const e = s + ln;
4437 /* The target is utf8, the pattern is not utf8.
4438 * Above-Latin1 code points can't match the pattern;
4439 * invariants match exactly, and the other Latin1 ones need
4440 * to be downgraded to a single byte in order to do the
4441 * comparison. (If we could be confident that the target
4442 * is not malformed, this could be refactored to have fewer
4443 * tests by just assuming that if the first bytes match, it
4444 * is an invariant, but there are tests in the test suite
4445 * dealing with (??{...}) which violate this) */
4447 if (l >= reginfo->strend
4448 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4452 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4459 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
4469 /* The target is not utf8, the pattern is utf8. */
4471 if (l >= reginfo->strend
4472 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4476 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4483 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
4495 /* The target and the pattern have the same utf8ness. */
4496 /* Inline the first character, for speed. */
4497 if (reginfo->strend - locinput < ln
4498 || UCHARAT(s) != nextchr
4499 || (ln > 1 && memNE(s, locinput, ln)))
4508 case EXACTFL: { /* /abc/il */
4510 const U8 * fold_array;
4512 U32 fold_utf8_flags;
4514 folder = foldEQ_locale;
4515 fold_array = PL_fold_locale;
4516 fold_utf8_flags = FOLDEQ_LOCALE;
4519 case EXACTFU_SS: /* /\x{df}/iu */
4520 case EXACTFU: /* /abc/iu */
4521 folder = foldEQ_latin1;
4522 fold_array = PL_fold_latin1;
4523 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4526 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
4528 assert(! is_utf8_pat);
4530 case EXACTFA: /* /abc/iaa */
4531 folder = foldEQ_latin1;
4532 fold_array = PL_fold_latin1;
4533 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4536 case EXACTF: /* /abc/i This node only generated for
4537 non-utf8 patterns */
4538 assert(! is_utf8_pat);
4540 fold_array = PL_fold;
4541 fold_utf8_flags = 0;
4549 || state_num == EXACTFU_SS
4550 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
4552 /* Either target or the pattern are utf8, or has the issue where
4553 * the fold lengths may differ. */
4554 const char * const l = locinput;
4555 char *e = reginfo->strend;
4557 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4558 l, &e, 0, utf8_target, fold_utf8_flags))
4566 /* Neither the target nor the pattern are utf8 */
4567 if (UCHARAT(s) != nextchr
4569 && UCHARAT(s) != fold_array[nextchr])
4573 if (reginfo->strend - locinput < ln)
4575 if (ln > 1 && ! folder(s, locinput, ln))
4581 /* XXX Could improve efficiency by separating these all out using a
4582 * macro or in-line function. At that point regcomp.c would no longer
4583 * have to set the FLAGS fields of these */
4584 case BOUNDL: /* /\b/l */
4585 case NBOUNDL: /* /\B/l */
4586 case BOUND: /* /\b/ */
4587 case BOUNDU: /* /\b/u */
4588 case BOUNDA: /* /\b/a */
4589 case NBOUND: /* /\B/ */
4590 case NBOUNDU: /* /\B/u */
4591 case NBOUNDA: /* /\B/a */
4592 /* was last char in word? */
4594 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4595 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4597 if (locinput == reginfo->strbeg)
4600 const U8 * const r =
4601 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4603 ln = utf8n_to_uvchr(r, (U8*) reginfo->strend - r,
4606 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4607 ln = isWORDCHAR_uni(ln);
4611 LOAD_UTF8_CHARCLASS_ALNUM();
4612 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4617 ln = isWORDCHAR_LC_uvchr(ln);
4618 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4623 /* Here the string isn't utf8, or is utf8 and only ascii
4624 * characters are to match \w. In the latter case looking at
4625 * the byte just prior to the current one may be just the final
4626 * byte of a multi-byte character. This is ok. There are two
4628 * 1) it is a single byte character, and then the test is doing
4629 * just what it's supposed to.
4630 * 2) it is a multi-byte character, in which case the final
4631 * byte is never mistakable for ASCII, and so the test
4632 * will say it is not a word character, which is the
4633 * correct answer. */
4634 ln = (locinput != reginfo->strbeg) ?
4635 UCHARAT(locinput - 1) : '\n';
4636 switch (FLAGS(scan)) {
4637 case REGEX_UNICODE_CHARSET:
4638 ln = isWORDCHAR_L1(ln);
4639 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4641 case REGEX_LOCALE_CHARSET:
4642 ln = isWORDCHAR_LC(ln);
4643 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4645 case REGEX_DEPENDS_CHARSET:
4646 ln = isWORDCHAR(ln);
4647 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4649 case REGEX_ASCII_RESTRICTED_CHARSET:
4650 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4651 ln = isWORDCHAR_A(ln);
4652 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4655 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4658 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4660 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4664 case ANYOF: /* /[abc]/ */
4668 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
4671 locinput += UTF8SKIP(locinput);
4674 if (!REGINCLASS(rex, scan, (U8*)locinput))
4680 /* The argument (FLAGS) to all the POSIX node types is the class number
4683 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4687 case POSIXL: /* \w or [:punct:] etc. under /l */
4691 /* Use isFOO_lc() for characters within Latin1. (Note that
4692 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4693 * wouldn't be invariant) */
4694 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4695 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4699 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4700 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4701 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4702 *(locinput + 1))))))
4707 else { /* Here, must be an above Latin-1 code point */
4708 goto utf8_posix_not_eos;
4711 /* Here, must be utf8 */
4712 locinput += UTF8SKIP(locinput);
4715 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4719 case POSIXD: /* \w or [:punct:] etc. under /d */
4725 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4727 if (NEXTCHR_IS_EOS) {
4731 /* All UTF-8 variants match */
4732 if (! UTF8_IS_INVARIANT(nextchr)) {
4733 goto increment_locinput;
4739 case POSIXA: /* \w or [:punct:] etc. under /a */
4742 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4743 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4744 * character is a single byte */
4747 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4753 /* Here we are either not in utf8, or we matched a utf8-invariant,
4754 * so the next char is the next byte */
4758 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4762 case POSIXU: /* \w or [:punct:] etc. under /u */
4764 if (NEXTCHR_IS_EOS) {
4769 /* Use _generic_isCC() for characters within Latin1. (Note that
4770 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4771 * wouldn't be invariant) */
4772 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4773 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4780 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4781 if (! (to_complement
4782 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4790 else { /* Handle above Latin-1 code points */
4791 classnum = (_char_class_number) FLAGS(scan);
4792 if (classnum < _FIRST_NON_SWASH_CC) {
4794 /* Here, uses a swash to find such code points. Load if if
4795 * not done already */
4796 if (! PL_utf8_swash_ptrs[classnum]) {
4797 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4798 PL_utf8_swash_ptrs[classnum]
4799 = _core_swash_init("utf8",
4802 PL_XPosix_ptrs[classnum], &flags);
4804 if (! (to_complement
4805 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4806 (U8 *) locinput, TRUE))))
4811 else { /* Here, uses macros to find above Latin-1 code points */
4813 case _CC_ENUM_SPACE: /* XXX would require separate
4814 code if we revert the change
4815 of \v matching this */
4816 case _CC_ENUM_PSXSPC:
4817 if (! (to_complement
4818 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4823 case _CC_ENUM_BLANK:
4824 if (! (to_complement
4825 ^ cBOOL(is_HORIZWS_high(locinput))))
4830 case _CC_ENUM_XDIGIT:
4831 if (! (to_complement
4832 ^ cBOOL(is_XDIGIT_high(locinput))))
4837 case _CC_ENUM_VERTSPACE:
4838 if (! (to_complement
4839 ^ cBOOL(is_VERTWS_high(locinput))))
4844 default: /* The rest, e.g. [:cntrl:], can't match
4846 if (! to_complement) {
4852 locinput += UTF8SKIP(locinput);
4856 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4857 a Unicode extended Grapheme Cluster */
4858 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4859 extended Grapheme Cluster is:
4862 | Prepend* Begin Extend*
4865 Begin is: ( Special_Begin | ! Control )
4866 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4867 Extend is: ( Grapheme_Extend | Spacing_Mark )
4868 Control is: [ GCB_Control | CR | LF ]
4869 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4871 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4874 Begin is ( Regular_Begin + Special Begin )
4876 It turns out that 98.4% of all Unicode code points match
4877 Regular_Begin. Doing it this way eliminates a table match in
4878 the previous implementation for almost all Unicode code points.
4880 There is a subtlety with Prepend* which showed up in testing.
4881 Note that the Begin, and only the Begin is required in:
4882 | Prepend* Begin Extend*
4883 Also, Begin contains '! Control'. A Prepend must be a
4884 '! Control', which means it must also be a Begin. What it
4885 comes down to is that if we match Prepend* and then find no
4886 suitable Begin afterwards, that if we backtrack the last
4887 Prepend, that one will be a suitable Begin.
4892 if (! utf8_target) {
4894 /* Match either CR LF or '.', as all the other possibilities
4896 locinput++; /* Match the . or CR */
4897 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4899 && locinput < reginfo->strend
4900 && UCHARAT(locinput) == '\n')
4907 /* Utf8: See if is ( CR LF ); already know that locinput <
4908 * reginfo->strend, so locinput+1 is in bounds */
4909 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4910 && UCHARAT(locinput + 1) == '\n')
4917 /* In case have to backtrack to beginning, then match '.' */
4918 char *starting = locinput;
4920 /* In case have to backtrack the last prepend */
4921 char *previous_prepend = NULL;
4923 LOAD_UTF8_CHARCLASS_GCB();
4925 /* Match (prepend)* */
4926 while (locinput < reginfo->strend
4927 && (len = is_GCB_Prepend_utf8(locinput)))
4929 previous_prepend = locinput;
4933 /* As noted above, if we matched a prepend character, but
4934 * the next thing won't match, back off the last prepend we
4935 * matched, as it is guaranteed to match the begin */
4936 if (previous_prepend
4937 && (locinput >= reginfo->strend
4938 || (! swash_fetch(PL_utf8_X_regular_begin,
4939 (U8*)locinput, utf8_target)
4940 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4943 locinput = previous_prepend;
4946 /* Note that here we know reginfo->strend > locinput, as we
4947 * tested that upon input to this switch case, and if we
4948 * moved locinput forward, we tested the result just above
4949 * and it either passed, or we backed off so that it will
4951 if (swash_fetch(PL_utf8_X_regular_begin,
4952 (U8*)locinput, utf8_target)) {
4953 locinput += UTF8SKIP(locinput);
4955 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4957 /* Here did not match the required 'Begin' in the
4958 * second term. So just match the very first
4959 * character, the '.' of the final term of the regex */
4960 locinput = starting + UTF8SKIP(starting);
4964 /* Here is a special begin. It can be composed of
4965 * several individual characters. One possibility is
4967 if ((len = is_GCB_RI_utf8(locinput))) {
4969 while (locinput < reginfo->strend
4970 && (len = is_GCB_RI_utf8(locinput)))
4974 } else if ((len = is_GCB_T_utf8(locinput))) {
4975 /* Another possibility is T+ */
4977 while (locinput < reginfo->strend
4978 && (len = is_GCB_T_utf8(locinput)))
4984 /* Here, neither RI+ nor T+; must be some other
4985 * Hangul. That means it is one of the others: L,
4986 * LV, LVT or V, and matches:
4987 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4990 while (locinput < reginfo->strend
4991 && (len = is_GCB_L_utf8(locinput)))
4996 /* Here, have exhausted L*. If the next character
4997 * is not an LV, LVT nor V, it means we had to have
4998 * at least one L, so matches L+ in the original
4999 * equation, we have a complete hangul syllable.
5002 if (locinput < reginfo->strend
5003 && is_GCB_LV_LVT_V_utf8(locinput))
5005 /* Otherwise keep going. Must be LV, LVT or V.
5006 * See if LVT, by first ruling out V, then LV */
5007 if (! is_GCB_V_utf8(locinput)
5008 /* All but every TCount one is LV */
5009 && (valid_utf8_to_uvchr((U8 *) locinput,
5014 locinput += UTF8SKIP(locinput);
5017 /* Must be V or LV. Take it, then match
5019 locinput += UTF8SKIP(locinput);
5020 while (locinput < reginfo->strend
5021 && (len = is_GCB_V_utf8(locinput)))
5027 /* And any of LV, LVT, or V can be followed
5029 while (locinput < reginfo->strend
5030 && (len = is_GCB_T_utf8(locinput)))
5038 /* Match any extender */
5039 while (locinput < reginfo->strend
5040 && swash_fetch(PL_utf8_X_extend,
5041 (U8*)locinput, utf8_target))
5043 locinput += UTF8SKIP(locinput);
5047 if (locinput > reginfo->strend) sayNO;
5051 case NREFFL: /* /\g{name}/il */
5052 { /* The capture buffer cases. The ones beginning with N for the
5053 named buffers just convert to the equivalent numbered and
5054 pretend they were called as the corresponding numbered buffer
5056 /* don't initialize these in the declaration, it makes C++
5061 const U8 *fold_array;
5064 folder = foldEQ_locale;
5065 fold_array = PL_fold_locale;
5067 utf8_fold_flags = FOLDEQ_LOCALE;
5070 case NREFFA: /* /\g{name}/iaa */
5071 folder = foldEQ_latin1;
5072 fold_array = PL_fold_latin1;
5074 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5077 case NREFFU: /* /\g{name}/iu */
5078 folder = foldEQ_latin1;
5079 fold_array = PL_fold_latin1;
5081 utf8_fold_flags = 0;
5084 case NREFF: /* /\g{name}/i */
5086 fold_array = PL_fold;
5088 utf8_fold_flags = 0;
5091 case NREF: /* /\g{name}/ */
5095 utf8_fold_flags = 0;
5098 /* For the named back references, find the corresponding buffer
5100 n = reg_check_named_buff_matched(rex,scan);
5105 goto do_nref_ref_common;
5107 case REFFL: /* /\1/il */
5108 folder = foldEQ_locale;
5109 fold_array = PL_fold_locale;
5110 utf8_fold_flags = FOLDEQ_LOCALE;
5113 case REFFA: /* /\1/iaa */
5114 folder = foldEQ_latin1;
5115 fold_array = PL_fold_latin1;
5116 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5119 case REFFU: /* /\1/iu */
5120 folder = foldEQ_latin1;
5121 fold_array = PL_fold_latin1;
5122 utf8_fold_flags = 0;
5125 case REFF: /* /\1/i */
5127 fold_array = PL_fold;
5128 utf8_fold_flags = 0;
5131 case REF: /* /\1/ */
5134 utf8_fold_flags = 0;
5138 n = ARG(scan); /* which paren pair */
5141 ln = rex->offs[n].start;
5142 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5143 if (rex->lastparen < n || ln == -1)
5144 sayNO; /* Do not match unless seen CLOSEn. */
5145 if (ln == rex->offs[n].end)
5148 s = reginfo->strbeg + ln;
5149 if (type != REF /* REF can do byte comparison */
5150 && (utf8_target || type == REFFU || type == REFFL))
5152 char * limit = reginfo->strend;
5154 /* This call case insensitively compares the entire buffer
5155 * at s, with the current input starting at locinput, but
5156 * not going off the end given by reginfo->strend, and
5157 * returns in <limit> upon success, how much of the
5158 * current input was matched */
5159 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
5160 locinput, &limit, 0, utf8_target, utf8_fold_flags))
5168 /* Not utf8: Inline the first character, for speed. */
5169 if (!NEXTCHR_IS_EOS &&
5170 UCHARAT(s) != nextchr &&
5172 UCHARAT(s) != fold_array[nextchr]))
5174 ln = rex->offs[n].end - ln;
5175 if (locinput + ln > reginfo->strend)
5177 if (ln > 1 && (type == REF
5178 ? memNE(s, locinput, ln)
5179 : ! folder(s, locinput, ln)))
5185 case NOTHING: /* null op; e.g. the 'nothing' following
5186 * the '*' in m{(a+|b)*}' */
5188 case TAIL: /* placeholder while compiling (A|B|C) */
5192 #define ST st->u.eval
5197 regexp_internal *rei;
5198 regnode *startpoint;
5200 case GOSTART: /* (?R) */
5201 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
5202 if (cur_eval && cur_eval->locinput==locinput) {
5203 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
5204 Perl_croak(aTHX_ "Infinite recursion in regex");
5205 if ( ++nochange_depth > max_nochange_depth )
5207 "Pattern subroutine nesting without pos change"
5208 " exceeded limit in regex");
5215 if (OP(scan)==GOSUB) {
5216 startpoint = scan + ARG2L(scan);
5217 ST.close_paren = ARG(scan);
5219 startpoint = rei->program+1;
5223 /* Save all the positions seen so far. */
5224 ST.cp = regcppush(rex, 0, maxopenparen);
5225 REGCP_SET(ST.lastcp);
5227 /* and then jump to the code we share with EVAL */
5228 goto eval_recurse_doit;
5231 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
5232 if (cur_eval && cur_eval->locinput==locinput) {
5233 if ( ++nochange_depth > max_nochange_depth )
5234 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
5239 /* execute the code in the {...} */
5243 OP * const oop = PL_op;
5244 COP * const ocurcop = PL_curcop;
5248 /* save *all* paren positions */
5249 regcppush(rex, 0, maxopenparen);
5250 REGCP_SET(runops_cp);
5253 caller_cv = find_runcv(NULL);
5257 if (rexi->data->what[n] == 'r') { /* code from an external qr */
5259 (REGEXP*)(rexi->data->data[n])
5262 nop = (OP*)rexi->data->data[n+1];
5264 else if (rexi->data->what[n] == 'l') { /* literal code */
5266 nop = (OP*)rexi->data->data[n];
5267 assert(CvDEPTH(newcv));
5270 /* literal with own CV */
5271 assert(rexi->data->what[n] == 'L');
5272 newcv = rex->qr_anoncv;
5273 nop = (OP*)rexi->data->data[n];
5276 /* normally if we're about to execute code from the same
5277 * CV that we used previously, we just use the existing
5278 * CX stack entry. However, its possible that in the
5279 * meantime we may have backtracked, popped from the save
5280 * stack, and undone the SAVECOMPPAD(s) associated with
5281 * PUSH_MULTICALL; in which case PL_comppad no longer
5282 * points to newcv's pad. */
5283 if (newcv != last_pushed_cv || PL_comppad != last_pad)
5285 U8 flags = (CXp_SUB_RE |
5286 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
5287 if (last_pushed_cv) {
5288 CHANGE_MULTICALL_FLAGS(newcv, flags);
5291 PUSH_MULTICALL_FLAGS(newcv, flags);
5293 last_pushed_cv = newcv;
5296 /* these assignments are just to silence compiler
5298 multicall_cop = NULL;
5301 last_pad = PL_comppad;
5303 /* the initial nextstate you would normally execute
5304 * at the start of an eval (which would cause error
5305 * messages to come from the eval), may be optimised
5306 * away from the execution path in the regex code blocks;
5307 * so manually set PL_curcop to it initially */
5309 OP *o = cUNOPx(nop)->op_first;
5310 assert(o->op_type == OP_NULL);
5311 if (o->op_targ == OP_SCOPE) {
5312 o = cUNOPo->op_first;
5315 assert(o->op_targ == OP_LEAVE);
5316 o = cUNOPo->op_first;
5317 assert(o->op_type == OP_ENTER);
5321 if (o->op_type != OP_STUB) {
5322 assert( o->op_type == OP_NEXTSTATE
5323 || o->op_type == OP_DBSTATE
5324 || (o->op_type == OP_NULL
5325 && ( o->op_targ == OP_NEXTSTATE
5326 || o->op_targ == OP_DBSTATE
5330 PL_curcop = (COP*)o;
5335 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
5336 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
5338 rex->offs[0].end = locinput - reginfo->strbeg;
5339 if (reginfo->info_aux_eval->pos_magic)
5340 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
5341 reginfo->sv, reginfo->strbeg,
5342 locinput - reginfo->strbeg);
5345 SV *sv_mrk = get_sv("REGMARK", 1);
5346 sv_setsv(sv_mrk, sv_yes_mark);
5349 /* we don't use MULTICALL here as we want to call the
5350 * first op of the block of interest, rather than the
5351 * first op of the sub */
5352 before = (IV)(SP-PL_stack_base);
5354 CALLRUNOPS(aTHX); /* Scalar context. */
5356 if ((IV)(SP-PL_stack_base) == before)
5357 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
5363 /* before restoring everything, evaluate the returned
5364 * value, so that 'uninit' warnings don't use the wrong
5365 * PL_op or pad. Also need to process any magic vars
5366 * (e.g. $1) *before* parentheses are restored */
5371 if (logical == 0) /* (?{})/ */
5372 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5373 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5374 sw = cBOOL(SvTRUE(ret));
5377 else { /* /(??{}) */
5378 /* if its overloaded, let the regex compiler handle
5379 * it; otherwise extract regex, or stringify */
5380 if (SvGMAGICAL(ret))
5381 ret = sv_mortalcopy(ret);
5382 if (!SvAMAGIC(ret)) {
5386 if (SvTYPE(sv) == SVt_REGEXP)
5387 re_sv = (REGEXP*) sv;
5388 else if (SvSMAGICAL(ret)) {
5389 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
5391 re_sv = (REGEXP *) mg->mg_obj;
5394 /* force any undef warnings here */
5395 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
5396 ret = sv_mortalcopy(ret);
5397 (void) SvPV_force_nolen(ret);
5403 /* *** Note that at this point we don't restore
5404 * PL_comppad, (or pop the CxSUB) on the assumption it may
5405 * be used again soon. This is safe as long as nothing
5406 * in the regexp code uses the pad ! */
5408 PL_curcop = ocurcop;
5409 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5410 PL_curpm = PL_reg_curpm;
5416 /* only /(??{})/ from now on */
5419 /* extract RE object from returned value; compiling if
5423 re_sv = reg_temp_copy(NULL, re_sv);
5428 if (SvUTF8(ret) && IN_BYTES) {
5429 /* In use 'bytes': make a copy of the octet
5430 * sequence, but without the flag on */
5432 const char *const p = SvPV(ret, len);
5433 ret = newSVpvn_flags(p, len, SVs_TEMP);
5435 if (rex->intflags & PREGf_USE_RE_EVAL)
5436 pm_flags |= PMf_USE_RE_EVAL;
5438 /* if we got here, it should be an engine which
5439 * supports compiling code blocks and stuff */
5440 assert(rex->engine && rex->engine->op_comp);
5441 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5442 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5443 rex->engine, NULL, NULL,
5444 /* copy /msix etc to inner pattern */
5449 & (SVs_TEMP | SVs_GMG | SVf_ROK))
5450 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
5451 /* This isn't a first class regexp. Instead, it's
5452 caching a regexp onto an existing, Perl visible
5454 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5460 RXp_MATCH_COPIED_off(re);
5461 re->subbeg = rex->subbeg;
5462 re->sublen = rex->sublen;
5463 re->suboffset = rex->suboffset;
5464 re->subcoffset = rex->subcoffset;
5466 re->lastcloseparen = 0;
5469 debug_start_match(re_sv, utf8_target, locinput,
5470 reginfo->strend, "Matching embedded");
5472 startpoint = rei->program + 1;
5473 ST.close_paren = 0; /* only used for GOSUB */
5474 /* Save all the seen positions so far. */
5475 ST.cp = regcppush(rex, 0, maxopenparen);
5476 REGCP_SET(ST.lastcp);
5477 /* and set maxopenparen to 0, since we are starting a "fresh" match */
5479 /* run the pattern returned from (??{...}) */
5481 eval_recurse_doit: /* Share code with GOSUB below this line
5482 * At this point we expect the stack context to be
5483 * set up correctly */
5485 /* invalidate the S-L poscache. We're now executing a
5486 * different set of WHILEM ops (and their associated
5487 * indexes) against the same string, so the bits in the
5488 * cache are meaningless. Setting maxiter to zero forces
5489 * the cache to be invalidated and zeroed before reuse.
5490 * XXX This is too dramatic a measure. Ideally we should
5491 * save the old cache and restore when running the outer
5493 reginfo->poscache_maxiter = 0;
5495 /* the new regexp might have a different is_utf8_pat than we do */
5496 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5498 ST.prev_rex = rex_sv;
5499 ST.prev_curlyx = cur_curlyx;
5501 SET_reg_curpm(rex_sv);
5506 ST.prev_eval = cur_eval;
5508 /* now continue from first node in postoned RE */
5509 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5514 case EVAL_AB: /* cleanup after a successful (??{A})B */
5515 /* note: this is called twice; first after popping B, then A */
5516 rex_sv = ST.prev_rex;
5517 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5518 SET_reg_curpm(rex_sv);
5519 rex = ReANY(rex_sv);
5520 rexi = RXi_GET(rex);
5522 /* preserve $^R across LEAVE's. See Bug 121070. */
5523 SV *save_sv= GvSV(PL_replgv);
5524 SvREFCNT_inc(save_sv);
5525 regcpblow(ST.cp); /* LEAVE in disguise */
5526 sv_setsv(GvSV(PL_replgv), save_sv);
5527 SvREFCNT_dec(save_sv);
5529 cur_eval = ST.prev_eval;
5530 cur_curlyx = ST.prev_curlyx;
5532 /* Invalidate cache. See "invalidate" comment above. */
5533 reginfo->poscache_maxiter = 0;
5534 if ( nochange_depth )
5539 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5540 /* note: this is called twice; first after popping B, then A */
5541 rex_sv = ST.prev_rex;
5542 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5543 SET_reg_curpm(rex_sv);
5544 rex = ReANY(rex_sv);
5545 rexi = RXi_GET(rex);
5547 REGCP_UNWIND(ST.lastcp);
5548 regcppop(rex, &maxopenparen);
5549 cur_eval = ST.prev_eval;
5550 cur_curlyx = ST.prev_curlyx;
5551 /* Invalidate cache. See "invalidate" comment above. */
5552 reginfo->poscache_maxiter = 0;
5553 if ( nochange_depth )
5559 n = ARG(scan); /* which paren pair */
5560 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5561 if (n > maxopenparen)
5563 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5564 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5568 (IV)rex->offs[n].start_tmp,
5574 /* XXX really need to log other places start/end are set too */
5575 #define CLOSE_CAPTURE \
5576 rex->offs[n].start = rex->offs[n].start_tmp; \
5577 rex->offs[n].end = locinput - reginfo->strbeg; \
5578 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5579 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5581 PTR2UV(rex->offs), \
5583 (IV)rex->offs[n].start, \
5584 (IV)rex->offs[n].end \
5588 n = ARG(scan); /* which paren pair */
5590 if (n > rex->lastparen)
5592 rex->lastcloseparen = n;
5593 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5598 case ACCEPT: /* (*ACCEPT) */
5602 cursor && OP(cursor)!=END;
5603 cursor=regnext(cursor))
5605 if ( OP(cursor)==CLOSE ){
5607 if ( n <= lastopen ) {
5609 if (n > rex->lastparen)
5611 rex->lastcloseparen = n;
5612 if ( n == ARG(scan) || (cur_eval &&
5613 cur_eval->u.eval.close_paren == n))
5622 case GROUPP: /* (?(1)) */
5623 n = ARG(scan); /* which paren pair */
5624 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5627 case NGROUPP: /* (?(<name>)) */
5628 /* reg_check_named_buff_matched returns 0 for no match */
5629 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5632 case INSUBP: /* (?(R)) */
5634 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5637 case DEFINEP: /* (?(DEFINE)) */
5641 case IFTHEN: /* (?(cond)A|B) */
5642 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5644 next = NEXTOPER(NEXTOPER(scan));
5646 next = scan + ARG(scan);
5647 if (OP(next) == IFTHEN) /* Fake one. */
5648 next = NEXTOPER(NEXTOPER(next));
5652 case LOGICAL: /* modifier for EVAL and IFMATCH */
5653 logical = scan->flags;
5656 /*******************************************************************
5658 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5659 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5660 STAR/PLUS/CURLY/CURLYN are used instead.)
5662 A*B is compiled as <CURLYX><A><WHILEM><B>
5664 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5665 state, which contains the current count, initialised to -1. It also sets
5666 cur_curlyx to point to this state, with any previous value saved in the
5669 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5670 since the pattern may possibly match zero times (i.e. it's a while {} loop
5671 rather than a do {} while loop).
5673 Each entry to WHILEM represents a successful match of A. The count in the
5674 CURLYX block is incremented, another WHILEM state is pushed, and execution
5675 passes to A or B depending on greediness and the current count.
5677 For example, if matching against the string a1a2a3b (where the aN are
5678 substrings that match /A/), then the match progresses as follows: (the
5679 pushed states are interspersed with the bits of strings matched so far):
5682 <CURLYX cnt=0><WHILEM>
5683 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5684 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5685 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5686 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5688 (Contrast this with something like CURLYM, which maintains only a single
5692 a1 <CURLYM cnt=1> a2
5693 a1 a2 <CURLYM cnt=2> a3
5694 a1 a2 a3 <CURLYM cnt=3> b
5697 Each WHILEM state block marks a point to backtrack to upon partial failure
5698 of A or B, and also contains some minor state data related to that
5699 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5700 overall state, such as the count, and pointers to the A and B ops.
5702 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5703 must always point to the *current* CURLYX block, the rules are:
5705 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5706 and set cur_curlyx to point the new block.
5708 When popping the CURLYX block after a successful or unsuccessful match,
5709 restore the previous cur_curlyx.
5711 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5712 to the outer one saved in the CURLYX block.
5714 When popping the WHILEM block after a successful or unsuccessful B match,
5715 restore the previous cur_curlyx.
5717 Here's an example for the pattern (AI* BI)*BO
5718 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5721 curlyx backtrack stack
5722 ------ ---------------
5724 CO <CO prev=NULL> <WO>
5725 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5726 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5727 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5729 At this point the pattern succeeds, and we work back down the stack to
5730 clean up, restoring as we go:
5732 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5733 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5734 CO <CO prev=NULL> <WO>
5737 *******************************************************************/
5739 #define ST st->u.curlyx
5741 case CURLYX: /* start of /A*B/ (for complex A) */
5743 /* No need to save/restore up to this paren */
5744 I32 parenfloor = scan->flags;
5746 assert(next); /* keep Coverity happy */
5747 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5750 /* XXXX Probably it is better to teach regpush to support
5751 parenfloor > maxopenparen ... */
5752 if (parenfloor > (I32)rex->lastparen)
5753 parenfloor = rex->lastparen; /* Pessimization... */
5755 ST.prev_curlyx= cur_curlyx;
5757 ST.cp = PL_savestack_ix;
5759 /* these fields contain the state of the current curly.
5760 * they are accessed by subsequent WHILEMs */
5761 ST.parenfloor = parenfloor;
5766 ST.count = -1; /* this will be updated by WHILEM */
5767 ST.lastloc = NULL; /* this will be updated by WHILEM */
5769 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5774 case CURLYX_end: /* just finished matching all of A*B */
5775 cur_curlyx = ST.prev_curlyx;
5780 case CURLYX_end_fail: /* just failed to match all of A*B */
5782 cur_curlyx = ST.prev_curlyx;
5789 #define ST st->u.whilem
5791 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5793 /* see the discussion above about CURLYX/WHILEM */
5798 assert(cur_curlyx); /* keep Coverity happy */
5800 min = ARG1(cur_curlyx->u.curlyx.me);
5801 max = ARG2(cur_curlyx->u.curlyx.me);
5802 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5803 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5804 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5805 ST.cache_offset = 0;
5809 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5810 "%*s whilem: matched %ld out of %d..%d\n",
5811 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5814 /* First just match a string of min A's. */
5817 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5819 cur_curlyx->u.curlyx.lastloc = locinput;
5820 REGCP_SET(ST.lastcp);
5822 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5827 /* If degenerate A matches "", assume A done. */
5829 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5830 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5831 "%*s whilem: empty match detected, trying continuation...\n",
5832 REPORT_CODE_OFF+depth*2, "")
5834 goto do_whilem_B_max;
5837 /* super-linear cache processing.
5839 * The idea here is that for certain types of CURLYX/WHILEM -
5840 * principally those whose upper bound is infinity (and
5841 * excluding regexes that have things like \1 and other very
5842 * non-regular expresssiony things), then if a pattern like
5843 * /....A*.../ fails and we backtrack to the WHILEM, then we
5844 * make a note that this particular WHILEM op was at string
5845 * position 47 (say) when the rest of pattern failed. Then, if
5846 * we ever find ourselves back at that WHILEM, and at string
5847 * position 47 again, we can just fail immediately rather than
5848 * running the rest of the pattern again.
5850 * This is very handy when patterns start to go
5851 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5852 * with a combinatorial explosion of backtracking.
5854 * The cache is implemented as a bit array, with one bit per
5855 * string byte position per WHILEM op (up to 16) - so its
5856 * between 0.25 and 2x the string size.
5858 * To avoid allocating a poscache buffer every time, we do an
5859 * initially countdown; only after we have executed a WHILEM
5860 * op (string-length x #WHILEMs) times do we allocate the
5863 * The top 4 bits of scan->flags byte say how many different
5864 * relevant CURLLYX/WHILEM op pairs there are, while the
5865 * bottom 4-bits is the identifying index number of this
5871 if (!reginfo->poscache_maxiter) {
5872 /* start the countdown: Postpone detection until we
5873 * know the match is not *that* much linear. */
5874 reginfo->poscache_maxiter
5875 = (reginfo->strend - reginfo->strbeg + 1)
5877 /* possible overflow for long strings and many CURLYX's */
5878 if (reginfo->poscache_maxiter < 0)
5879 reginfo->poscache_maxiter = I32_MAX;
5880 reginfo->poscache_iter = reginfo->poscache_maxiter;
5883 if (reginfo->poscache_iter-- == 0) {
5884 /* initialise cache */
5885 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
5886 regmatch_info_aux *const aux = reginfo->info_aux;
5887 if (aux->poscache) {
5888 if ((SSize_t)reginfo->poscache_size < size) {
5889 Renew(aux->poscache, size, char);
5890 reginfo->poscache_size = size;
5892 Zero(aux->poscache, size, char);
5895 reginfo->poscache_size = size;
5896 Newxz(aux->poscache, size, char);
5898 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5899 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5900 PL_colors[4], PL_colors[5])
5904 if (reginfo->poscache_iter < 0) {
5905 /* have we already failed at this position? */
5906 SSize_t offset, mask;
5908 reginfo->poscache_iter = -1; /* stop eventual underflow */
5909 offset = (scan->flags & 0xf) - 1
5910 + (locinput - reginfo->strbeg)
5912 mask = 1 << (offset % 8);
5914 if (reginfo->info_aux->poscache[offset] & mask) {
5915 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5916 "%*s whilem: (cache) already tried at this position...\n",
5917 REPORT_CODE_OFF+depth*2, "")
5919 sayNO; /* cache records failure */
5921 ST.cache_offset = offset;
5922 ST.cache_mask = mask;
5926 /* Prefer B over A for minimal matching. */
5928 if (cur_curlyx->u.curlyx.minmod) {
5929 ST.save_curlyx = cur_curlyx;
5930 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5931 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5933 REGCP_SET(ST.lastcp);
5934 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5940 /* Prefer A over B for maximal matching. */
5942 if (n < max) { /* More greed allowed? */
5943 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5945 cur_curlyx->u.curlyx.lastloc = locinput;
5946 REGCP_SET(ST.lastcp);
5947 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5951 goto do_whilem_B_max;
5956 case WHILEM_B_min: /* just matched B in a minimal match */
5957 case WHILEM_B_max: /* just matched B in a maximal match */
5958 cur_curlyx = ST.save_curlyx;
5963 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5964 cur_curlyx = ST.save_curlyx;
5965 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5966 cur_curlyx->u.curlyx.count--;
5971 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5973 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5974 REGCP_UNWIND(ST.lastcp);
5975 regcppop(rex, &maxopenparen);
5976 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5977 cur_curlyx->u.curlyx.count--;
5982 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5983 REGCP_UNWIND(ST.lastcp);
5984 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5985 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5986 "%*s whilem: failed, trying continuation...\n",
5987 REPORT_CODE_OFF+depth*2, "")
5990 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5991 && ckWARN(WARN_REGEXP)
5992 && !reginfo->warned)
5994 reginfo->warned = TRUE;
5995 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5996 "Complex regular subexpression recursion limit (%d) "
6002 ST.save_curlyx = cur_curlyx;
6003 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
6004 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
6009 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
6010 cur_curlyx = ST.save_curlyx;
6011 REGCP_UNWIND(ST.lastcp);
6012 regcppop(rex, &maxopenparen);
6014 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
6015 /* Maximum greed exceeded */
6016 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6017 && ckWARN(WARN_REGEXP)
6018 && !reginfo->warned)
6020 reginfo->warned = TRUE;
6021 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6022 "Complex regular subexpression recursion "
6023 "limit (%d) exceeded",
6026 cur_curlyx->u.curlyx.count--;
6030 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6031 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
6033 /* Try grabbing another A and see if it helps. */
6034 cur_curlyx->u.curlyx.lastloc = locinput;
6035 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6037 REGCP_SET(ST.lastcp);
6038 PUSH_STATE_GOTO(WHILEM_A_min,
6039 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
6045 #define ST st->u.branch
6047 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
6048 next = scan + ARG(scan);
6051 scan = NEXTOPER(scan);
6054 case BRANCH: /* /(...|A|...)/ */
6055 scan = NEXTOPER(scan); /* scan now points to inner node */
6056 ST.lastparen = rex->lastparen;
6057 ST.lastcloseparen = rex->lastcloseparen;
6058 ST.next_branch = next;
6061 /* Now go into the branch */
6063 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
6065 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
6070 case CUTGROUP: /* /(*THEN)/ */
6071 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
6072 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6073 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
6077 case CUTGROUP_next_fail:
6080 if (st->u.mark.mark_name)
6081 sv_commit = st->u.mark.mark_name;
6091 case BRANCH_next_fail: /* that branch failed; try the next, if any */
6096 REGCP_UNWIND(ST.cp);
6097 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6098 scan = ST.next_branch;
6099 /* no more branches? */
6100 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
6102 PerlIO_printf( Perl_debug_log,
6103 "%*s %sBRANCH failed...%s\n",
6104 REPORT_CODE_OFF+depth*2, "",
6110 continue; /* execute next BRANCH[J] op */
6113 case MINMOD: /* next op will be non-greedy, e.g. A*? */
6118 #define ST st->u.curlym
6120 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
6122 /* This is an optimisation of CURLYX that enables us to push
6123 * only a single backtracking state, no matter how many matches
6124 * there are in {m,n}. It relies on the pattern being constant
6125 * length, with no parens to influence future backrefs
6129 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6131 ST.lastparen = rex->lastparen;
6132 ST.lastcloseparen = rex->lastcloseparen;
6134 /* if paren positive, emulate an OPEN/CLOSE around A */
6136 U32 paren = ST.me->flags;
6137 if (paren > maxopenparen)
6138 maxopenparen = paren;
6139 scan += NEXT_OFF(scan); /* Skip former OPEN. */
6147 ST.c1 = CHRTEST_UNINIT;
6150 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
6153 curlym_do_A: /* execute the A in /A{m,n}B/ */
6154 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
6158 case CURLYM_A: /* we've just matched an A */
6160 /* after first match, determine A's length: u.curlym.alen */
6161 if (ST.count == 1) {
6162 if (reginfo->is_utf8_target) {
6163 char *s = st->locinput;
6164 while (s < locinput) {
6170 ST.alen = locinput - st->locinput;
6173 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
6176 PerlIO_printf(Perl_debug_log,
6177 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
6178 (int)(REPORT_CODE_OFF+(depth*2)), "",
6179 (IV) ST.count, (IV)ST.alen)
6182 if (cur_eval && cur_eval->u.eval.close_paren &&
6183 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6187 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
6188 if ( max == REG_INFTY || ST.count < max )
6189 goto curlym_do_A; /* try to match another A */
6191 goto curlym_do_B; /* try to match B */
6193 case CURLYM_A_fail: /* just failed to match an A */
6194 REGCP_UNWIND(ST.cp);
6196 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
6197 || (cur_eval && cur_eval->u.eval.close_paren &&
6198 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
6201 curlym_do_B: /* execute the B in /A{m,n}B/ */
6202 if (ST.c1 == CHRTEST_UNINIT) {
6203 /* calculate c1 and c2 for possible match of 1st char
6204 * following curly */
6205 ST.c1 = ST.c2 = CHRTEST_VOID;
6207 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
6208 regnode *text_node = ST.B;
6209 if (! HAS_TEXT(text_node))
6210 FIND_NEXT_IMPT(text_node);
6213 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
6215 But the former is redundant in light of the latter.
6217 if this changes back then the macro for
6218 IS_TEXT and friends need to change.
6220 if (PL_regkind[OP(text_node)] == EXACT) {
6221 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6222 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6232 PerlIO_printf(Perl_debug_log,
6233 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
6234 (int)(REPORT_CODE_OFF+(depth*2)),
6237 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
6238 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
6239 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6240 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6242 /* simulate B failing */
6244 PerlIO_printf(Perl_debug_log,
6245 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
6246 (int)(REPORT_CODE_OFF+(depth*2)),"",
6247 valid_utf8_to_uvchr((U8 *) locinput, NULL),
6248 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
6249 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
6251 state_num = CURLYM_B_fail;
6252 goto reenter_switch;
6255 else if (nextchr != ST.c1 && nextchr != ST.c2) {
6256 /* simulate B failing */
6258 PerlIO_printf(Perl_debug_log,
6259 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
6260 (int)(REPORT_CODE_OFF+(depth*2)),"",
6261 (int) nextchr, ST.c1, ST.c2)
6263 state_num = CURLYM_B_fail;
6264 goto reenter_switch;
6269 /* emulate CLOSE: mark current A as captured */
6270 I32 paren = ST.me->flags;
6272 rex->offs[paren].start
6273 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
6274 rex->offs[paren].end = locinput - reginfo->strbeg;
6275 if ((U32)paren > rex->lastparen)
6276 rex->lastparen = paren;
6277 rex->lastcloseparen = paren;
6280 rex->offs[paren].end = -1;
6281 if (cur_eval && cur_eval->u.eval.close_paren &&
6282 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6291 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
6295 case CURLYM_B_fail: /* just failed to match a B */
6296 REGCP_UNWIND(ST.cp);
6297 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6299 I32 max = ARG2(ST.me);
6300 if (max != REG_INFTY && ST.count == max)
6302 goto curlym_do_A; /* try to match a further A */
6304 /* backtrack one A */
6305 if (ST.count == ARG1(ST.me) /* min */)
6308 SET_locinput(HOPc(locinput, -ST.alen));
6309 goto curlym_do_B; /* try to match B */
6312 #define ST st->u.curly
6314 #define CURLY_SETPAREN(paren, success) \
6317 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
6318 rex->offs[paren].end = locinput - reginfo->strbeg; \
6319 if (paren > rex->lastparen) \
6320 rex->lastparen = paren; \
6321 rex->lastcloseparen = paren; \
6324 rex->offs[paren].end = -1; \
6325 rex->lastparen = ST.lastparen; \
6326 rex->lastcloseparen = ST.lastcloseparen; \
6330 case STAR: /* /A*B/ where A is width 1 char */
6334 scan = NEXTOPER(scan);
6337 case PLUS: /* /A+B/ where A is width 1 char */
6341 scan = NEXTOPER(scan);
6344 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
6345 ST.paren = scan->flags; /* Which paren to set */
6346 ST.lastparen = rex->lastparen;
6347 ST.lastcloseparen = rex->lastcloseparen;
6348 if (ST.paren > maxopenparen)
6349 maxopenparen = ST.paren;
6350 ST.min = ARG1(scan); /* min to match */
6351 ST.max = ARG2(scan); /* max to match */
6352 if (cur_eval && cur_eval->u.eval.close_paren &&
6353 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6357 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
6360 case CURLY: /* /A{m,n}B/ where A is width 1 char */
6362 ST.min = ARG1(scan); /* min to match */
6363 ST.max = ARG2(scan); /* max to match */
6364 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6367 * Lookahead to avoid useless match attempts
6368 * when we know what character comes next.
6370 * Used to only do .*x and .*?x, but now it allows
6371 * for )'s, ('s and (?{ ... })'s to be in the way
6372 * of the quantifier and the EXACT-like node. -- japhy
6375 assert(ST.min <= ST.max);
6376 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
6377 ST.c1 = ST.c2 = CHRTEST_VOID;
6380 regnode *text_node = next;
6382 if (! HAS_TEXT(text_node))
6383 FIND_NEXT_IMPT(text_node);
6385 if (! HAS_TEXT(text_node))
6386 ST.c1 = ST.c2 = CHRTEST_VOID;
6388 if ( PL_regkind[OP(text_node)] != EXACT ) {
6389 ST.c1 = ST.c2 = CHRTEST_VOID;
6393 /* Currently we only get here when
6395 PL_rekind[OP(text_node)] == EXACT
6397 if this changes back then the macro for IS_TEXT and
6398 friends need to change. */
6399 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6400 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6412 char *li = locinput;
6415 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
6421 if (ST.c1 == CHRTEST_VOID)
6422 goto curly_try_B_min;
6424 ST.oldloc = locinput;
6426 /* set ST.maxpos to the furthest point along the
6427 * string that could possibly match */
6428 if (ST.max == REG_INFTY) {
6429 ST.maxpos = reginfo->strend - 1;
6431 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
6434 else if (utf8_target) {
6435 int m = ST.max - ST.min;
6436 for (ST.maxpos = locinput;
6437 m >0 && ST.maxpos < reginfo->strend; m--)
6438 ST.maxpos += UTF8SKIP(ST.maxpos);
6441 ST.maxpos = locinput + ST.max - ST.min;
6442 if (ST.maxpos >= reginfo->strend)
6443 ST.maxpos = reginfo->strend - 1;
6445 goto curly_try_B_min_known;
6449 /* avoid taking address of locinput, so it can remain
6451 char *li = locinput;
6452 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6453 if (ST.count < ST.min)
6456 if ((ST.count > ST.min)
6457 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6459 /* A{m,n} must come at the end of the string, there's
6460 * no point in backing off ... */
6462 /* ...except that $ and \Z can match before *and* after
6463 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6464 We may back off by one in this case. */
6465 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6469 goto curly_try_B_max;
6474 case CURLY_B_min_known_fail:
6475 /* failed to find B in a non-greedy match where c1,c2 valid */
6477 REGCP_UNWIND(ST.cp);
6479 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6481 /* Couldn't or didn't -- move forward. */
6482 ST.oldloc = locinput;
6484 locinput += UTF8SKIP(locinput);
6488 curly_try_B_min_known:
6489 /* find the next place where 'B' could work, then call B */
6493 n = (ST.oldloc == locinput) ? 0 : 1;
6494 if (ST.c1 == ST.c2) {
6495 /* set n to utf8_distance(oldloc, locinput) */
6496 while (locinput <= ST.maxpos
6497 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6499 locinput += UTF8SKIP(locinput);
6504 /* set n to utf8_distance(oldloc, locinput) */
6505 while (locinput <= ST.maxpos
6506 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6507 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6509 locinput += UTF8SKIP(locinput);
6514 else { /* Not utf8_target */
6515 if (ST.c1 == ST.c2) {
6516 while (locinput <= ST.maxpos &&
6517 UCHARAT(locinput) != ST.c1)
6521 while (locinput <= ST.maxpos
6522 && UCHARAT(locinput) != ST.c1
6523 && UCHARAT(locinput) != ST.c2)
6526 n = locinput - ST.oldloc;
6528 if (locinput > ST.maxpos)
6531 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6532 * at b; check that everything between oldloc and
6533 * locinput matches */
6534 char *li = ST.oldloc;
6536 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6538 assert(n == REG_INFTY || locinput == li);
6540 CURLY_SETPAREN(ST.paren, ST.count);
6541 if (cur_eval && cur_eval->u.eval.close_paren &&
6542 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6545 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6550 case CURLY_B_min_fail:
6551 /* failed to find B in a non-greedy match where c1,c2 invalid */
6553 REGCP_UNWIND(ST.cp);
6555 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6557 /* failed -- move forward one */
6559 char *li = locinput;
6560 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6567 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6568 ST.count > 0)) /* count overflow ? */
6571 CURLY_SETPAREN(ST.paren, ST.count);
6572 if (cur_eval && cur_eval->u.eval.close_paren &&
6573 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6576 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6584 /* a successful greedy match: now try to match B */
6585 if (cur_eval && cur_eval->u.eval.close_paren &&
6586 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6590 bool could_match = locinput < reginfo->strend;
6592 /* If it could work, try it. */
6593 if (ST.c1 != CHRTEST_VOID && could_match) {
6594 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6596 could_match = memEQ(locinput,
6601 UTF8SKIP(locinput));
6604 could_match = UCHARAT(locinput) == ST.c1
6605 || UCHARAT(locinput) == ST.c2;
6608 if (ST.c1 == CHRTEST_VOID || could_match) {
6609 CURLY_SETPAREN(ST.paren, ST.count);
6610 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6617 case CURLY_B_max_fail:
6618 /* failed to find B in a greedy match */
6620 REGCP_UNWIND(ST.cp);
6622 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6625 if (--ST.count < ST.min)
6627 locinput = HOPc(locinput, -1);
6628 goto curly_try_B_max;
6632 case END: /* last op of main pattern */
6635 /* we've just finished A in /(??{A})B/; now continue with B */
6637 st->u.eval.prev_rex = rex_sv; /* inner */
6639 /* Save *all* the positions. */
6640 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6641 rex_sv = cur_eval->u.eval.prev_rex;
6642 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6643 SET_reg_curpm(rex_sv);
6644 rex = ReANY(rex_sv);
6645 rexi = RXi_GET(rex);
6646 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6648 REGCP_SET(st->u.eval.lastcp);
6650 /* Restore parens of the outer rex without popping the
6652 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6655 st->u.eval.prev_eval = cur_eval;
6656 cur_eval = cur_eval->u.eval.prev_eval;
6658 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6659 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6660 if ( nochange_depth )
6663 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6664 locinput); /* match B */
6667 if (locinput < reginfo->till) {
6668 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6669 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6671 (long)(locinput - startpos),
6672 (long)(reginfo->till - startpos),
6675 sayNO_SILENT; /* Cannot match: too short. */
6677 sayYES; /* Success! */
6679 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6681 PerlIO_printf(Perl_debug_log,
6682 "%*s %ssubpattern success...%s\n",
6683 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6684 sayYES; /* Success! */
6687 #define ST st->u.ifmatch
6692 case SUSPEND: /* (?>A) */
6694 newstart = locinput;
6697 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6699 goto ifmatch_trivial_fail_test;
6701 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6703 ifmatch_trivial_fail_test:
6705 char * const s = HOPBACKc(locinput, scan->flags);
6710 sw = 1 - cBOOL(ST.wanted);
6714 next = scan + ARG(scan);
6722 newstart = locinput;
6726 ST.logical = logical;
6727 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6729 /* execute body of (?...A) */
6730 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6735 case IFMATCH_A_fail: /* body of (?...A) failed */
6736 ST.wanted = !ST.wanted;
6739 case IFMATCH_A: /* body of (?...A) succeeded */
6741 sw = cBOOL(ST.wanted);
6743 else if (!ST.wanted)
6746 if (OP(ST.me) != SUSPEND) {
6747 /* restore old position except for (?>...) */
6748 locinput = st->locinput;
6750 scan = ST.me + ARG(ST.me);
6753 continue; /* execute B */
6757 case LONGJMP: /* alternative with many branches compiles to
6758 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6759 next = scan + ARG(scan);
6764 case COMMIT: /* (*COMMIT) */
6765 reginfo->cutpoint = reginfo->strend;
6768 case PRUNE: /* (*PRUNE) */
6770 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6771 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6775 case COMMIT_next_fail:
6779 case OPFAIL: /* (*FAIL) */
6784 #define ST st->u.mark
6785 case MARKPOINT: /* (*MARK:foo) */
6786 ST.prev_mark = mark_state;
6787 ST.mark_name = sv_commit = sv_yes_mark
6788 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6790 ST.mark_loc = locinput;
6791 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6795 case MARKPOINT_next:
6796 mark_state = ST.prev_mark;
6801 case MARKPOINT_next_fail:
6802 if (popmark && sv_eq(ST.mark_name,popmark))
6804 if (ST.mark_loc > startpoint)
6805 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6806 popmark = NULL; /* we found our mark */
6807 sv_commit = ST.mark_name;
6810 PerlIO_printf(Perl_debug_log,
6811 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6812 REPORT_CODE_OFF+depth*2, "",
6813 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6816 mark_state = ST.prev_mark;
6817 sv_yes_mark = mark_state ?
6818 mark_state->u.mark.mark_name : NULL;
6823 case SKIP: /* (*SKIP) */
6825 /* (*SKIP) : if we fail we cut here*/
6826 ST.mark_name = NULL;
6827 ST.mark_loc = locinput;
6828 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6830 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6831 otherwise do nothing. Meaning we need to scan
6833 regmatch_state *cur = mark_state;
6834 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6837 if ( sv_eq( cur->u.mark.mark_name,
6840 ST.mark_name = find;
6841 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6843 cur = cur->u.mark.prev_mark;
6846 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6849 case SKIP_next_fail:
6851 /* (*CUT:NAME) - Set up to search for the name as we
6852 collapse the stack*/
6853 popmark = ST.mark_name;
6855 /* (*CUT) - No name, we cut here.*/
6856 if (ST.mark_loc > startpoint)
6857 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6858 /* but we set sv_commit to latest mark_name if there
6859 is one so they can test to see how things lead to this
6862 sv_commit=mark_state->u.mark.mark_name;
6870 case LNBREAK: /* \R */
6871 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6878 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6879 PTR2UV(scan), OP(scan));
6880 Perl_croak(aTHX_ "regexp memory corruption");
6882 /* this is a point to jump to in order to increment
6883 * locinput by one character */
6885 assert(!NEXTCHR_IS_EOS);
6887 locinput += PL_utf8skip[nextchr];
6888 /* locinput is allowed to go 1 char off the end, but not 2+ */
6889 if (locinput > reginfo->strend)
6898 /* switch break jumps here */
6899 scan = next; /* prepare to execute the next op and ... */
6900 continue; /* ... jump back to the top, reusing st */
6904 /* push a state that backtracks on success */
6905 st->u.yes.prev_yes_state = yes_state;
6909 /* push a new regex state, then continue at scan */
6911 regmatch_state *newst;
6914 regmatch_state *cur = st;
6915 regmatch_state *curyes = yes_state;
6917 regmatch_slab *slab = PL_regmatch_slab;
6918 for (;curd > -1;cur--,curd--) {
6919 if (cur < SLAB_FIRST(slab)) {
6921 cur = SLAB_LAST(slab);
6923 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6924 REPORT_CODE_OFF + 2 + depth * 2,"",
6925 curd, PL_reg_name[cur->resume_state],
6926 (curyes == cur) ? "yes" : ""
6929 curyes = cur->u.yes.prev_yes_state;
6932 DEBUG_STATE_pp("push")
6935 st->locinput = locinput;
6937 if (newst > SLAB_LAST(PL_regmatch_slab))
6938 newst = S_push_slab(aTHX);
6939 PL_regmatch_state = newst;
6941 locinput = pushinput;
6949 * We get here only if there's trouble -- normally "case END" is
6950 * the terminating point.
6952 Perl_croak(aTHX_ "corrupted regexp pointers");
6959 /* we have successfully completed a subexpression, but we must now
6960 * pop to the state marked by yes_state and continue from there */
6961 assert(st != yes_state);
6963 while (st != yes_state) {
6965 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6966 PL_regmatch_slab = PL_regmatch_slab->prev;
6967 st = SLAB_LAST(PL_regmatch_slab);
6971 DEBUG_STATE_pp("pop (no final)");
6973 DEBUG_STATE_pp("pop (yes)");
6979 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6980 || yes_state > SLAB_LAST(PL_regmatch_slab))
6982 /* not in this slab, pop slab */
6983 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6984 PL_regmatch_slab = PL_regmatch_slab->prev;
6985 st = SLAB_LAST(PL_regmatch_slab);
6987 depth -= (st - yes_state);
6990 yes_state = st->u.yes.prev_yes_state;
6991 PL_regmatch_state = st;
6994 locinput= st->locinput;
6995 state_num = st->resume_state + no_final;
6996 goto reenter_switch;
6999 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
7000 PL_colors[4], PL_colors[5]));
7002 if (reginfo->info_aux_eval) {
7003 /* each successfully executed (?{...}) block does the equivalent of
7004 * local $^R = do {...}
7005 * When popping the save stack, all these locals would be undone;
7006 * bypass this by setting the outermost saved $^R to the latest
7008 /* I dont know if this is needed or works properly now.
7009 * see code related to PL_replgv elsewhere in this file.
7012 if (oreplsv != GvSV(PL_replgv))
7013 sv_setsv(oreplsv, GvSV(PL_replgv));
7020 PerlIO_printf(Perl_debug_log,
7021 "%*s %sfailed...%s\n",
7022 REPORT_CODE_OFF+depth*2, "",
7023 PL_colors[4], PL_colors[5])
7035 /* there's a previous state to backtrack to */
7037 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7038 PL_regmatch_slab = PL_regmatch_slab->prev;
7039 st = SLAB_LAST(PL_regmatch_slab);
7041 PL_regmatch_state = st;
7042 locinput= st->locinput;
7044 DEBUG_STATE_pp("pop");
7046 if (yes_state == st)
7047 yes_state = st->u.yes.prev_yes_state;
7049 state_num = st->resume_state + 1; /* failure = success + 1 */
7050 goto reenter_switch;
7055 if (rex->intflags & PREGf_VERBARG_SEEN) {
7056 SV *sv_err = get_sv("REGERROR", 1);
7057 SV *sv_mrk = get_sv("REGMARK", 1);
7059 sv_commit = &PL_sv_no;
7061 sv_yes_mark = &PL_sv_yes;
7064 sv_commit = &PL_sv_yes;
7065 sv_yes_mark = &PL_sv_no;
7069 sv_setsv(sv_err, sv_commit);
7070 sv_setsv(sv_mrk, sv_yes_mark);
7074 if (last_pushed_cv) {
7077 PERL_UNUSED_VAR(SP);
7080 assert(!result || locinput - reginfo->strbeg >= 0);
7081 return result ? locinput - reginfo->strbeg : -1;
7085 - regrepeat - repeatedly match something simple, report how many
7087 * What 'simple' means is a node which can be the operand of a quantifier like
7090 * startposp - pointer a pointer to the start position. This is updated
7091 * to point to the byte following the highest successful
7093 * p - the regnode to be repeatedly matched against.
7094 * reginfo - struct holding match state, such as strend
7095 * max - maximum number of things to match.
7096 * depth - (for debugging) backtracking depth.
7099 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
7100 regmatch_info *const reginfo, I32 max, int depth)
7102 char *scan; /* Pointer to current position in target string */
7104 char *loceol = reginfo->strend; /* local version */
7105 I32 hardcount = 0; /* How many matches so far */
7106 bool utf8_target = reginfo->is_utf8_target;
7107 int to_complement = 0; /* Invert the result? */
7109 _char_class_number classnum;
7111 PERL_UNUSED_ARG(depth);
7114 PERL_ARGS_ASSERT_REGREPEAT;
7117 if (max == REG_INFTY)
7119 else if (! utf8_target && loceol - scan > max)
7120 loceol = scan + max;
7122 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
7123 * to the maximum of how far we should go in it (leaving it set to the real
7124 * end, if the maximum permissible would take us beyond that). This allows
7125 * us to make the loop exit condition that we haven't gone past <loceol> to
7126 * also mean that we haven't exceeded the max permissible count, saving a
7127 * test each time through the loop. But it assumes that the OP matches a
7128 * single byte, which is true for most of the OPs below when applied to a
7129 * non-UTF-8 target. Those relatively few OPs that don't have this
7130 * characteristic will have to compensate.
7132 * There is no adjustment for UTF-8 targets, as the number of bytes per
7133 * character varies. OPs will have to test both that the count is less
7134 * than the max permissible (using <hardcount> to keep track), and that we
7135 * are still within the bounds of the string (using <loceol>. A few OPs
7136 * match a single byte no matter what the encoding. They can omit the max
7137 * test if, for the UTF-8 case, they do the adjustment that was skipped
7140 * Thus, the code above sets things up for the common case; and exceptional
7141 * cases need extra work; the common case is to make sure <scan> doesn't
7142 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
7143 * count doesn't exceed the maximum permissible */
7148 while (scan < loceol && hardcount < max && *scan != '\n') {
7149 scan += UTF8SKIP(scan);
7153 while (scan < loceol && *scan != '\n')
7159 while (scan < loceol && hardcount < max) {
7160 scan += UTF8SKIP(scan);
7167 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
7168 if (utf8_target && loceol - scan > max) {
7170 /* <loceol> hadn't been adjusted in the UTF-8 case */
7178 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7182 /* Can use a simple loop if the pattern char to match on is invariant
7183 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
7184 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
7185 * true iff it doesn't matter if the argument is in UTF-8 or not */
7186 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
7187 if (utf8_target && loceol - scan > max) {
7188 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
7189 * since here, to match at all, 1 char == 1 byte */
7190 loceol = scan + max;
7192 while (scan < loceol && UCHARAT(scan) == c) {
7196 else if (reginfo->is_utf8_pat) {
7198 STRLEN scan_char_len;
7200 /* When both target and pattern are UTF-8, we have to do
7202 while (hardcount < max
7204 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
7205 && memEQ(scan, STRING(p), scan_char_len))
7207 scan += scan_char_len;
7211 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
7213 /* Target isn't utf8; convert the character in the UTF-8
7214 * pattern to non-UTF8, and do a simple loop */
7215 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
7216 while (scan < loceol && UCHARAT(scan) == c) {
7219 } /* else pattern char is above Latin1, can't possibly match the
7224 /* Here, the string must be utf8; pattern isn't, and <c> is
7225 * different in utf8 than not, so can't compare them directly.
7226 * Outside the loop, find the two utf8 bytes that represent c, and
7227 * then look for those in sequence in the utf8 string */
7228 U8 high = UTF8_TWO_BYTE_HI(c);
7229 U8 low = UTF8_TWO_BYTE_LO(c);
7231 while (hardcount < max
7232 && scan + 1 < loceol
7233 && UCHARAT(scan) == high
7234 && UCHARAT(scan + 1) == low)
7242 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
7243 assert(! reginfo->is_utf8_pat);
7246 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7250 utf8_flags = FOLDEQ_LOCALE;
7253 case EXACTF: /* This node only generated for non-utf8 patterns */
7254 assert(! reginfo->is_utf8_pat);
7260 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
7264 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
7266 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7268 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
7271 if (c1 == CHRTEST_VOID) {
7272 /* Use full Unicode fold matching */
7273 char *tmpeol = reginfo->strend;
7274 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
7275 while (hardcount < max
7276 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
7277 STRING(p), NULL, pat_len,
7278 reginfo->is_utf8_pat, utf8_flags))
7281 tmpeol = reginfo->strend;
7285 else if (utf8_target) {
7287 while (scan < loceol
7289 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
7291 scan += UTF8SKIP(scan);
7296 while (scan < loceol
7298 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
7299 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
7301 scan += UTF8SKIP(scan);
7306 else if (c1 == c2) {
7307 while (scan < loceol && UCHARAT(scan) == c1) {
7312 while (scan < loceol &&
7313 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
7323 while (hardcount < max
7325 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
7327 scan += UTF8SKIP(scan);
7331 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
7336 /* The argument (FLAGS) to all the POSIX node types is the class number */
7343 if (! utf8_target) {
7344 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
7350 while (hardcount < max && scan < loceol
7351 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
7354 scan += UTF8SKIP(scan);
7367 if (utf8_target && loceol - scan > max) {
7369 /* We didn't adjust <loceol> at the beginning of this routine
7370 * because is UTF-8, but it is actually ok to do so, since here, to
7371 * match, 1 char == 1 byte. */
7372 loceol = scan + max;
7374 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
7387 if (! utf8_target) {
7388 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
7394 /* The complement of something that matches only ASCII matches all
7395 * non-ASCII, plus everything in ASCII that isn't in the class. */
7396 while (hardcount < max && scan < loceol
7397 && (! isASCII_utf8(scan)
7398 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
7400 scan += UTF8SKIP(scan);
7411 if (! utf8_target) {
7412 while (scan < loceol && to_complement
7413 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
7420 classnum = (_char_class_number) FLAGS(p);
7421 if (classnum < _FIRST_NON_SWASH_CC) {
7423 /* Here, a swash is needed for above-Latin1 code points.
7424 * Process as many Latin1 code points using the built-in rules.
7425 * Go to another loop to finish processing upon encountering
7426 * the first Latin1 code point. We could do that in this loop
7427 * as well, but the other way saves having to test if the swash
7428 * has been loaded every time through the loop: extra space to
7430 while (hardcount < max && scan < loceol) {
7431 if (UTF8_IS_INVARIANT(*scan)) {
7432 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
7439 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
7440 if (! (to_complement
7441 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
7450 goto found_above_latin1;
7457 /* For these character classes, the knowledge of how to handle
7458 * every code point is compiled in to Perl via a macro. This
7459 * code is written for making the loops as tight as possible.
7460 * It could be refactored to save space instead */
7462 case _CC_ENUM_SPACE: /* XXX would require separate code
7463 if we revert the change of \v
7466 case _CC_ENUM_PSXSPC:
7467 while (hardcount < max
7469 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7471 scan += UTF8SKIP(scan);
7475 case _CC_ENUM_BLANK:
7476 while (hardcount < max
7478 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7480 scan += UTF8SKIP(scan);
7484 case _CC_ENUM_XDIGIT:
7485 while (hardcount < max
7487 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7489 scan += UTF8SKIP(scan);
7493 case _CC_ENUM_VERTSPACE:
7494 while (hardcount < max
7496 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7498 scan += UTF8SKIP(scan);
7502 case _CC_ENUM_CNTRL:
7503 while (hardcount < max
7505 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7507 scan += UTF8SKIP(scan);
7512 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7518 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7520 /* Load the swash if not already present */
7521 if (! PL_utf8_swash_ptrs[classnum]) {
7522 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7523 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7527 PL_XPosix_ptrs[classnum], &flags);
7530 while (hardcount < max && scan < loceol
7531 && to_complement ^ cBOOL(_generic_utf8(
7534 swash_fetch(PL_utf8_swash_ptrs[classnum],
7538 scan += UTF8SKIP(scan);
7545 while (hardcount < max && scan < loceol &&
7546 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7551 /* LNBREAK can match one or two latin chars, which is ok, but we
7552 * have to use hardcount in this situation, and throw away the
7553 * adjustment to <loceol> done before the switch statement */
7554 loceol = reginfo->strend;
7555 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7576 /* These are all 0 width, so match right here or not at all. */
7580 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7589 c = scan - *startposp;
7593 GET_RE_DEBUG_FLAGS_DECL;
7595 SV * const prop = sv_newmortal();
7596 regprop(prog, prop, p, reginfo, NULL);
7597 PerlIO_printf(Perl_debug_log,
7598 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7599 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7607 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7609 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7610 create a copy so that changes the caller makes won't change the shared one.
7611 If <altsvp> is non-null, will return NULL in it, for back-compat.
7614 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7616 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7622 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL));
7625 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
7628 - reginclass - determine if a character falls into a character class
7630 n is the ANYOF regnode
7631 p is the target string
7632 p_end points to one byte beyond the end of the target string
7633 utf8_target tells whether p is in UTF-8.
7635 Returns true if matched; false otherwise.
7637 Note that this can be a synthetic start class, a combination of various
7638 nodes, so things you think might be mutually exclusive, such as locale,
7639 aren't. It can match both locale and non-locale
7644 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
7647 const char flags = ANYOF_FLAGS(n);
7651 PERL_ARGS_ASSERT_REGINCLASS;
7653 /* If c is not already the code point, get it. Note that
7654 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7655 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7657 c = utf8n_to_uvchr(p, p_end - p, &c_len,
7658 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7659 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7660 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7661 * UTF8_ALLOW_FFFF */
7662 if (c_len == (STRLEN)-1)
7663 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7666 /* If this character is potentially in the bitmap, check it */
7667 if (c < NUM_ANYOF_CODE_POINTS) {
7668 if (ANYOF_BITMAP_TEST(n, c))
7670 else if ((flags & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII)
7676 else if (flags & ANYOF_LOCALE_FLAGS) {
7677 if ((flags & ANYOF_LOC_FOLD)
7679 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7683 else if (ANYOF_POSIXL_TEST_ANY_SET(n)
7687 /* The data structure is arranged so bits 0, 2, 4, ... are set
7688 * if the class includes the Posix character class given by
7689 * bit/2; and 1, 3, 5, ... are set if the class includes the
7690 * complemented Posix class given by int(bit/2). So we loop
7691 * through the bits, each time changing whether we complement
7692 * the result or not. Suppose for the sake of illustration
7693 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7694 * is set, it means there is a match for this ANYOF node if the
7695 * character is in the class given by the expression (0 / 2 = 0
7696 * = \w). If it is in that class, isFOO_lc() will return 1,
7697 * and since 'to_complement' is 0, the result will stay TRUE,
7698 * and we exit the loop. Suppose instead that bit 0 is 0, but
7699 * bit 1 is 1. That means there is a match if the character
7700 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7701 * but will on bit 1. On the second iteration 'to_complement'
7702 * will be 1, so the exclusive or will reverse things, so we
7703 * are testing for \W. On the third iteration, 'to_complement'
7704 * will be 0, and we would be testing for \s; the fourth
7705 * iteration would test for \S, etc.
7707 * Note that this code assumes that all the classes are closed
7708 * under folding. For example, if a character matches \w, then
7709 * its fold does too; and vice versa. This should be true for
7710 * any well-behaved locale for all the currently defined Posix
7711 * classes, except for :lower: and :upper:, which are handled
7712 * by the pseudo-class :cased: which matches if either of the
7713 * other two does. To get rid of this assumption, an outer
7714 * loop could be used below to iterate over both the source
7715 * character, and its fold (if different) */
7718 int to_complement = 0;
7720 while (count < ANYOF_MAX) {
7721 if (ANYOF_POSIXL_TEST(n, count)
7722 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7735 /* If the bitmap didn't (or couldn't) match, and something outside the
7736 * bitmap could match, try that. */
7738 if (c >= NUM_ANYOF_CODE_POINTS
7739 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
7741 match = TRUE; /* Everything above the bitmap matches */
7743 else if ((flags & ANYOF_HAS_NONBITMAP_NON_UTF8_MATCHES)
7744 || (utf8_target && (flags & ANYOF_HAS_UTF8_NONBITMAP_MATCHES))
7745 || ((flags & ANYOF_LOC_FOLD)
7746 && IN_UTF8_CTYPE_LOCALE
7747 && ARG(n) != ANYOF_ONLY_HAS_BITMAP))
7749 SV* only_utf8_locale = NULL;
7750 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
7751 &only_utf8_locale, NULL);
7757 } else { /* Convert to utf8 */
7758 utf8_p = utf8_buffer;
7759 append_utf8_from_native_byte(*p, &utf8_p);
7760 utf8_p = utf8_buffer;
7763 if (swash_fetch(sw, utf8_p, TRUE)) {
7767 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
7768 match = _invlist_contains_cp(only_utf8_locale, c);
7772 if (UNICODE_IS_SUPER(c)
7773 && (flags & ANYOF_WARN_SUPER)
7774 && ckWARN_d(WARN_NON_UNICODE))
7776 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7777 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c);
7781 #if ANYOF_INVERT != 1
7782 /* Depending on compiler optimization cBOOL takes time, so if don't have to
7784 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
7787 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7788 return (flags & ANYOF_INVERT) ^ match;
7792 S_reghop3(U8 *s, SSize_t off, const U8* lim)
7794 /* return the position 'off' UTF-8 characters away from 's', forward if
7795 * 'off' >= 0, backwards if negative. But don't go outside of position
7796 * 'lim', which better be < s if off < 0 */
7798 PERL_ARGS_ASSERT_REGHOP3;
7801 while (off-- && s < lim) {
7802 /* XXX could check well-formedness here */
7807 while (off++ && s > lim) {
7809 if (UTF8_IS_CONTINUED(*s)) {
7810 while (s > lim && UTF8_IS_CONTINUATION(*s))
7813 /* XXX could check well-formedness here */
7820 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
7822 PERL_ARGS_ASSERT_REGHOP4;
7825 while (off-- && s < rlim) {
7826 /* XXX could check well-formedness here */
7831 while (off++ && s > llim) {
7833 if (UTF8_IS_CONTINUED(*s)) {
7834 while (s > llim && UTF8_IS_CONTINUATION(*s))
7837 /* XXX could check well-formedness here */
7843 /* like reghop3, but returns NULL on overrun, rather than returning last
7847 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
7849 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7852 while (off-- && s < lim) {
7853 /* XXX could check well-formedness here */
7860 while (off++ && s > lim) {
7862 if (UTF8_IS_CONTINUED(*s)) {
7863 while (s > lim && UTF8_IS_CONTINUATION(*s))
7866 /* XXX could check well-formedness here */
7875 /* when executing a regex that may have (?{}), extra stuff needs setting
7876 up that will be visible to the called code, even before the current
7877 match has finished. In particular:
7879 * $_ is localised to the SV currently being matched;
7880 * pos($_) is created if necessary, ready to be updated on each call-out
7882 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
7883 isn't set until the current pattern is successfully finished), so that
7884 $1 etc of the match-so-far can be seen;
7885 * save the old values of subbeg etc of the current regex, and set then
7886 to the current string (again, this is normally only done at the end
7891 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
7894 regexp *const rex = ReANY(reginfo->prog);
7895 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
7897 eval_state->rex = rex;
7900 /* Make $_ available to executed code. */
7901 if (reginfo->sv != DEFSV) {
7903 DEFSV_set(reginfo->sv);
7906 if (!(mg = mg_find_mglob(reginfo->sv))) {
7907 /* prepare for quick setting of pos */
7908 mg = sv_magicext_mglob(reginfo->sv);
7911 eval_state->pos_magic = mg;
7912 eval_state->pos = mg->mg_len;
7913 eval_state->pos_flags = mg->mg_flags;
7916 eval_state->pos_magic = NULL;
7918 if (!PL_reg_curpm) {
7919 /* PL_reg_curpm is a fake PMOP that we can attach the current
7920 * regex to and point PL_curpm at, so that $1 et al are visible
7921 * within a /(?{})/. It's just allocated once per interpreter the
7922 * first time its needed */
7923 Newxz(PL_reg_curpm, 1, PMOP);
7926 SV* const repointer = &PL_sv_undef;
7927 /* this regexp is also owned by the new PL_reg_curpm, which
7928 will try to free it. */
7929 av_push(PL_regex_padav, repointer);
7930 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
7931 PL_regex_pad = AvARRAY(PL_regex_padav);
7935 SET_reg_curpm(reginfo->prog);
7936 eval_state->curpm = PL_curpm;
7937 PL_curpm = PL_reg_curpm;
7938 if (RXp_MATCH_COPIED(rex)) {
7939 /* Here is a serious problem: we cannot rewrite subbeg,
7940 since it may be needed if this match fails. Thus
7941 $` inside (?{}) could fail... */
7942 eval_state->subbeg = rex->subbeg;
7943 eval_state->sublen = rex->sublen;
7944 eval_state->suboffset = rex->suboffset;
7945 eval_state->subcoffset = rex->subcoffset;
7947 eval_state->saved_copy = rex->saved_copy;
7949 RXp_MATCH_COPIED_off(rex);
7952 eval_state->subbeg = NULL;
7953 rex->subbeg = (char *)reginfo->strbeg;
7955 rex->subcoffset = 0;
7956 rex->sublen = reginfo->strend - reginfo->strbeg;
7960 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
7963 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
7965 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
7966 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
7969 Safefree(aux->poscache);
7973 /* undo the effects of S_setup_eval_state() */
7975 if (eval_state->subbeg) {
7976 regexp * const rex = eval_state->rex;
7977 rex->subbeg = eval_state->subbeg;
7978 rex->sublen = eval_state->sublen;
7979 rex->suboffset = eval_state->suboffset;
7980 rex->subcoffset = eval_state->subcoffset;
7982 rex->saved_copy = eval_state->saved_copy;
7984 RXp_MATCH_COPIED_on(rex);
7986 if (eval_state->pos_magic)
7988 eval_state->pos_magic->mg_len = eval_state->pos;
7989 eval_state->pos_magic->mg_flags =
7990 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
7991 | (eval_state->pos_flags & MGf_BYTES);
7994 PL_curpm = eval_state->curpm;
7997 PL_regmatch_state = aux->old_regmatch_state;
7998 PL_regmatch_slab = aux->old_regmatch_slab;
8000 /* free all slabs above current one - this must be the last action
8001 * of this function, as aux and eval_state are allocated within
8002 * slabs and may be freed here */
8004 s = PL_regmatch_slab->next;
8006 PL_regmatch_slab->next = NULL;
8008 regmatch_slab * const osl = s;
8017 S_to_utf8_substr(pTHX_ regexp *prog)
8019 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
8020 * on the converted value */
8024 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
8027 if (prog->substrs->data[i].substr
8028 && !prog->substrs->data[i].utf8_substr) {
8029 SV* const sv = newSVsv(prog->substrs->data[i].substr);
8030 prog->substrs->data[i].utf8_substr = sv;
8031 sv_utf8_upgrade(sv);
8032 if (SvVALID(prog->substrs->data[i].substr)) {
8033 if (SvTAIL(prog->substrs->data[i].substr)) {
8034 /* Trim the trailing \n that fbm_compile added last
8036 SvCUR_set(sv, SvCUR(sv) - 1);
8037 /* Whilst this makes the SV technically "invalid" (as its
8038 buffer is no longer followed by "\0") when fbm_compile()
8039 adds the "\n" back, a "\0" is restored. */
8040 fbm_compile(sv, FBMcf_TAIL);
8044 if (prog->substrs->data[i].substr == prog->check_substr)
8045 prog->check_utf8 = sv;
8051 S_to_byte_substr(pTHX_ regexp *prog)
8053 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
8054 * on the converted value; returns FALSE if can't be converted. */
8058 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
8061 if (prog->substrs->data[i].utf8_substr
8062 && !prog->substrs->data[i].substr) {
8063 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
8064 if (! sv_utf8_downgrade(sv, TRUE)) {
8067 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
8068 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
8069 /* Trim the trailing \n that fbm_compile added last
8071 SvCUR_set(sv, SvCUR(sv) - 1);
8072 fbm_compile(sv, FBMcf_TAIL);
8076 prog->substrs->data[i].substr = sv;
8077 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
8078 prog->check_substr = sv;
8087 * c-indentation-style: bsd
8089 * indent-tabs-mode: nil
8092 * ex: set ts=8 sts=4 sw=4 et: