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 assert(0); /* 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 reginfo->is_utf8_target = cBOOL(utf8_target);
707 reginfo->info_aux = NULL;
708 reginfo->strbeg = strbeg;
709 reginfo->strend = strend;
710 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
712 /* not actually used within intuit, but zero for safety anyway */
713 reginfo->poscache_maxiter = 0;
716 if (!prog->check_utf8 && prog->check_substr)
717 to_utf8_substr(prog);
718 check = prog->check_utf8;
720 if (!prog->check_substr && prog->check_utf8) {
721 if (! to_byte_substr(prog)) {
722 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
725 check = prog->check_substr;
728 /* dump the various substring data */
729 DEBUG_OPTIMISE_MORE_r({
731 for (i=0; i<=2; i++) {
732 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
733 : prog->substrs->data[i].substr);
737 PerlIO_printf(Perl_debug_log,
738 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
739 " useful=%"IVdf" utf8=%d [%s]\n",
741 (IV)prog->substrs->data[i].min_offset,
742 (IV)prog->substrs->data[i].max_offset,
743 (IV)prog->substrs->data[i].end_shift,
750 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
752 /* ml_anch: check after \n?
754 * A note about IMPLICIT: on an un-anchored pattern beginning
755 * with /.*.../, these flags will have been added by the
757 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
758 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
760 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
761 && !(prog->intflags & PREGf_IMPLICIT);
763 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
764 /* we are only allowed to match at BOS or \G */
766 /* trivially reject if there's a BOS anchor and we're not at BOS.
768 * Note that we don't try to do a similar quick reject for
769 * \G, since generally the caller will have calculated strpos
770 * based on pos() and gofs, so the string is already correctly
771 * anchored by definition; and handling the exceptions would
772 * be too fiddly (e.g. REXEC_IGNOREPOS).
774 if ( strpos != strbeg
775 && (prog->intflags & (PREGf_ANCH_BOL|PREGf_ANCH_SBOL)))
777 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
778 " Not at start...\n"));
782 /* in the presence of an anchor, the anchored (relative to the
783 * start of the regex) substr must also be anchored relative
784 * to strpos. So quickly reject if substr isn't found there.
785 * This works for \G too, because the caller will already have
786 * subtracted gofs from pos, and gofs is the offset from the
787 * \G to the start of the regex. For example, in /.abc\Gdef/,
788 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
789 * caller will have set strpos=pos()-4; we look for the substr
790 * at position pos()-4+1, which lines up with the "a" */
792 if (prog->check_offset_min == prog->check_offset_max
793 && !(prog->intflags & PREGf_CANY_SEEN))
795 /* Substring at constant offset from beg-of-str... */
796 SSize_t slen = SvCUR(check);
797 char *s = HOP3c(strpos, prog->check_offset_min, strend);
799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
800 " Looking for check substr at fixed offset %"IVdf"...\n",
801 (IV)prog->check_offset_min));
804 /* In this case, the regex is anchored at the end too.
805 * Unless it's a multiline match, the lengths must match
806 * exactly, give or take a \n. NB: slen >= 1 since
807 * the last char of check is \n */
809 && ( strend - s > slen
810 || strend - s < slen - 1
811 || (strend - s == slen && strend[-1] != '\n')))
813 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
814 " String too long...\n"));
817 /* Now should match s[0..slen-2] */
820 if (slen && (*SvPVX_const(check) != *s
821 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
823 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
824 " String not equal...\n"));
829 goto success_at_start;
834 end_shift = prog->check_end_shift;
836 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
838 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
839 (IV)end_shift, RX_PRECOMP(prog));
844 /* This is the (re)entry point of the main loop in this function.
845 * The goal of this loop is to:
846 * 1) find the "check" substring in the region rx_origin..strend
847 * (adjusted by start_shift / end_shift). If not found, reject
849 * 2) If it exists, look for the "other" substr too if defined; for
850 * example, if the check substr maps to the anchored substr, then
851 * check the floating substr, and vice-versa. If not found, go
852 * back to (1) with rx_origin suitably incremented.
853 * 3) If we find an rx_origin position that doesn't contradict
854 * either of the substrings, then check the possible additional
855 * constraints on rx_origin of /^.../m or a known start class.
856 * If these fail, then depending on which constraints fail, jump
857 * back to here, or to various other re-entry points further along
858 * that skip some of the first steps.
859 * 4) If we pass all those tests, update the BmUSEFUL() count on the
860 * substring. If the start position was determined to be at the
861 * beginning of the string - so, not rejected, but not optimised,
862 * since we have to run regmatch from position 0 - decrement the
863 * BmUSEFUL() count. Otherwise increment it.
867 /* first, look for the 'check' substring */
873 DEBUG_OPTIMISE_MORE_r({
874 PerlIO_printf(Perl_debug_log,
875 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
876 " Start shift: %"IVdf" End shift %"IVdf
877 " Real end Shift: %"IVdf"\n",
878 (IV)(rx_origin - strpos),
879 (IV)prog->check_offset_min,
882 (IV)prog->check_end_shift);
885 if (prog->intflags & PREGf_CANY_SEEN) {
886 start_point= (U8*)(rx_origin + start_shift);
887 end_point= (U8*)(strend - end_shift);
888 if (start_point > end_point)
891 end_point = HOP3(strend, -end_shift, strbeg);
892 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
898 /* If the regex is absolutely anchored to either the start of the
899 * string (BOL,SBOL) or to pos() (ANCH_GPOS), then
900 * check_offset_max represents an upper bound on the string where
901 * the substr could start. For the ANCH_GPOS case, we assume that
902 * the caller of intuit will have already set strpos to
903 * pos()-gofs, so in this case strpos + offset_max will still be
904 * an upper bound on the substr.
907 && prog->intflags & PREGf_ANCH
908 && prog->check_offset_max != SSize_t_MAX)
910 SSize_t len = SvCUR(check) - !!SvTAIL(check);
911 const char * const anchor =
912 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
914 /* do a bytes rather than chars comparison. It's conservative;
915 * so it skips doing the HOP if the result can't possibly end
916 * up earlier than the old value of end_point.
918 if ((char*)end_point - anchor > prog->check_offset_max) {
919 end_point = HOP3lim((U8*)anchor,
920 prog->check_offset_max,
926 DEBUG_OPTIMISE_MORE_r({
927 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
928 (int)(end_point - start_point),
929 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
933 check_at = fbm_instr( start_point, end_point,
934 check, multiline ? FBMrf_MULTILINE : 0);
936 /* Update the count-of-usability, remove useless subpatterns,
940 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
941 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
942 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
943 (check_at ? "Found" : "Did not find"),
944 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
945 ? "anchored" : "floating"),
948 (check_at ? " at offset " : "...\n") );
953 /* Finish the diagnostic message */
954 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
956 /* set rx_origin to the minimum position where the regex could start
957 * matching, given the constraint of the just-matched check substring.
958 * But don't set it lower than previously.
961 if (check_at - rx_origin > prog->check_offset_max)
962 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
966 /* now look for the 'other' substring if defined */
968 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
969 : prog->substrs->data[other_ix].substr)
971 /* Take into account the "other" substring. */
975 struct reg_substr_datum *other;
978 other = &prog->substrs->data[other_ix];
980 /* if "other" is anchored:
981 * we've previously found a floating substr starting at check_at.
982 * This means that the regex origin must lie somewhere
983 * between min (rx_origin): HOP3(check_at, -check_offset_max)
984 * and max: HOP3(check_at, -check_offset_min)
985 * (except that min will be >= strpos)
986 * So the fixed substr must lie somewhere between
987 * HOP3(min, anchored_offset)
988 * HOP3(max, anchored_offset) + SvCUR(substr)
991 /* if "other" is floating
992 * Calculate last1, the absolute latest point where the
993 * floating substr could start in the string, ignoring any
994 * constraints from the earlier fixed match. It is calculated
997 * strend - prog->minlen (in chars) is the absolute latest
998 * position within the string where the origin of the regex
999 * could appear. The latest start point for the floating
1000 * substr is float_min_offset(*) on from the start of the
1001 * regex. last1 simply combines thee two offsets.
1003 * (*) You might think the latest start point should be
1004 * float_max_offset from the regex origin, and technically
1005 * you'd be correct. However, consider
1007 * Here, float min, max are 3,5 and minlen is 7.
1008 * This can match either
1012 * In the first case, the regex matches minlen chars; in the
1013 * second, minlen+1, in the third, minlen+2.
1014 * In the first case, the floating offset is 3 (which equals
1015 * float_min), in the second, 4, and in the third, 5 (which
1016 * equals float_max). In all cases, the floating string bcd
1017 * can never start more than 4 chars from the end of the
1018 * string, which equals minlen - float_min. As the substring
1019 * starts to match more than float_min from the start of the
1020 * regex, it makes the regex match more than minlen chars,
1021 * and the two cancel each other out. So we can always use
1022 * float_min - minlen, rather than float_max - minlen for the
1023 * latest position in the string.
1025 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1026 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1029 assert(prog->minlen >= other->min_offset);
1030 last1 = HOP3c(strend,
1031 other->min_offset - prog->minlen, strbeg);
1033 if (other_ix) {/* i.e. if (other-is-float) */
1034 /* last is the latest point where the floating substr could
1035 * start, *given* any constraints from the earlier fixed
1036 * match. This constraint is that the floating string starts
1037 * <= float_max_offset chars from the regex origin (rx_origin).
1038 * If this value is less than last1, use it instead.
1040 assert(rx_origin <= last1);
1042 /* this condition handles the offset==infinity case, and
1043 * is a short-cut otherwise. Although it's comparing a
1044 * byte offset to a char length, it does so in a safe way,
1045 * since 1 char always occupies 1 or more bytes,
1046 * so if a string range is (last1 - rx_origin) bytes,
1047 * it will be less than or equal to (last1 - rx_origin)
1048 * chars; meaning it errs towards doing the accurate HOP3
1049 * rather than just using last1 as a short-cut */
1050 (last1 - rx_origin) < other->max_offset
1052 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1055 assert(strpos + start_shift <= check_at);
1056 last = HOP4c(check_at, other->min_offset - start_shift,
1060 s = HOP3c(rx_origin, other->min_offset, strend);
1061 if (s < other_last) /* These positions already checked */
1064 must = utf8_target ? other->utf8_substr : other->substr;
1065 assert(SvPOK(must));
1068 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1070 multiline ? FBMrf_MULTILINE : 0
1073 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1074 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1075 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1076 s ? "Found" : "Contradicts",
1077 other_ix ? "floating" : "anchored",
1078 quoted, RE_SV_TAIL(must));
1083 /* last1 is latest possible substr location. If we didn't
1084 * find it before there, we never will */
1085 if (last >= last1) {
1086 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1087 ", giving up...\n"));
1091 /* try to find the check substr again at a later
1092 * position. Maybe next time we'll find the "other" substr
1094 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1095 ", trying %s at offset %ld...\n",
1096 (other_ix ? "floating" : "anchored"),
1097 (long)(HOP3c(check_at, 1, strend) - strpos)));
1099 other_last = HOP3c(last, 1, strend) /* highest failure */;
1101 other_ix /* i.e. if other-is-float */
1102 ? HOP3c(rx_origin, 1, strend)
1103 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1107 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
1108 (long)(s - strpos)));
1110 if (other_ix) { /* if (other-is-float) */
1111 /* other_last is set to s, not s+1, since its possible for
1112 * a floating substr to fail first time, then succeed
1113 * second time at the same floating position; e.g.:
1114 * "-AB--AABZ" =~ /\wAB\d*Z/
1115 * The first time round, anchored and float match at
1116 * "-(AB)--AAB(Z)" then fail on the initial \w character
1117 * class. Second time round, they match at "-AB--A(AB)(Z)".
1122 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1123 other_last = HOP3c(s, 1, strend);
1128 DEBUG_OPTIMISE_MORE_r(
1129 PerlIO_printf(Perl_debug_log,
1130 " Check-only match: offset min:%"IVdf" max:%"IVdf
1131 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
1132 " strend-strpos:%"IVdf"\n",
1133 (IV)prog->check_offset_min,
1134 (IV)prog->check_offset_max,
1135 (IV)(check_at-strpos),
1136 (IV)(rx_origin-strpos),
1137 (IV)(rx_origin-check_at),
1143 postprocess_substr_matches:
1145 /* handle the extra constraint of /^.../m if present */
1147 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1150 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1151 " looking for /^/m anchor"));
1153 /* we have failed the constraint of a \n before rx_origin.
1154 * Find the next \n, if any, even if it's beyond the current
1155 * anchored and/or floating substrings. Whether we should be
1156 * scanning ahead for the next \n or the next substr is debatable.
1157 * On the one hand you'd expect rare substrings to appear less
1158 * often than \n's. On the other hand, searching for \n means
1159 * we're effectively flipping been check_substr and "\n" on each
1160 * iteration as the current "rarest" string candidate, which
1161 * means for example that we'll quickly reject the whole string if
1162 * hasn't got a \n, rather than trying every substr position
1166 s = HOP3c(strend, - prog->minlen, strpos);
1167 if (s <= rx_origin ||
1168 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1170 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1171 " Did not find /%s^%s/m...\n",
1172 PL_colors[0], PL_colors[1]));
1176 /* earliest possible origin is 1 char after the \n.
1177 * (since *rx_origin == '\n', it's safe to ++ here rather than
1178 * HOP(rx_origin, 1)) */
1181 if (prog->substrs->check_ix == 0 /* check is anchored */
1182 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1184 /* Position contradicts check-string; either because
1185 * check was anchored (and thus has no wiggle room),
1186 * or check was float and rx_origin is above the float range */
1187 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1188 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1189 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1193 /* if we get here, the check substr must have been float,
1194 * is in range, and we may or may not have had an anchored
1195 * "other" substr which still contradicts */
1196 assert(prog->substrs->check_ix); /* check is float */
1198 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1199 /* whoops, the anchored "other" substr exists, so we still
1200 * contradict. On the other hand, the float "check" substr
1201 * didn't contradict, so just retry the anchored "other"
1203 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1204 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1205 PL_colors[0], PL_colors[1],
1206 (long)(rx_origin - strpos),
1207 (long)(rx_origin - strpos + prog->anchored_offset)));
1208 goto do_other_substr;
1211 /* success: we don't contradict the found floating substring
1212 * (and there's no anchored substr). */
1213 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1214 " Found /%s^%s/m at offset %ld...\n",
1215 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1218 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1219 " (multiline anchor test skipped)\n"));
1225 /* if we have a starting character class, then test that extra constraint.
1226 * (trie stclasses are too expensive to use here, we are better off to
1227 * leave it to regmatch itself) */
1229 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1230 const U8* const str = (U8*)STRING(progi->regstclass);
1232 /* XXX this value could be pre-computed */
1233 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1234 ? (reginfo->is_utf8_pat
1235 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1236 : STR_LEN(progi->regstclass))
1240 /* latest pos that a matching float substr constrains rx start to */
1241 char *rx_max_float = NULL;
1243 /* if the current rx_origin is anchored, either by satisfying an
1244 * anchored substring constraint, or a /^.../m constraint, then we
1245 * can reject the current origin if the start class isn't found
1246 * at the current position. If we have a float-only match, then
1247 * rx_origin is constrained to a range; so look for the start class
1248 * in that range. if neither, then look for the start class in the
1249 * whole rest of the string */
1251 /* XXX DAPM it's not clear what the minlen test is for, and why
1252 * it's not used in the floating case. Nothing in the test suite
1253 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1254 * Here are some old comments, which may or may not be correct:
1256 * minlen == 0 is possible if regstclass is \b or \B,
1257 * and the fixed substr is ''$.
1258 * Since minlen is already taken into account, rx_origin+1 is
1259 * before strend; accidentally, minlen >= 1 guaranties no false
1260 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1261 * 0) below assumes that regstclass does not come from lookahead...
1262 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1263 * This leaves EXACTF-ish only, which are dealt with in
1267 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1268 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1269 else if (prog->float_substr || prog->float_utf8) {
1270 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1271 endpos= HOP3c(rx_max_float, cl_l, strend);
1276 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1277 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
1278 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1279 (IV)start_shift, (IV)(check_at - strbeg),
1280 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1282 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1285 if (endpos == strend) {
1286 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1287 " Could not match STCLASS...\n") );
1290 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1291 " This position contradicts STCLASS...\n") );
1292 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1293 && !(prog->intflags & PREGf_IMPLICIT))
1296 /* Contradict one of substrings */
1297 if (prog->anchored_substr || prog->anchored_utf8) {
1298 if (prog->substrs->check_ix == 1) { /* check is float */
1299 /* Have both, check_string is floating */
1300 assert(rx_origin + start_shift <= check_at);
1301 if (rx_origin + start_shift != check_at) {
1302 /* not at latest position float substr could match:
1303 * Recheck anchored substring, but not floating.
1304 * The condition above is in bytes rather than
1305 * chars for efficiency. It's conservative, in
1306 * that it errs on the side of doing 'goto
1307 * do_other_substr', where a more accurate
1308 * char-based calculation will be done */
1309 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1310 " Looking for anchored substr starting at offset %ld...\n",
1311 (long)(other_last - strpos)) );
1312 goto do_other_substr;
1320 /* In the presence of ml_anch, we might be able to
1321 * find another \n without breaking the current float
1324 /* strictly speaking this should be HOP3c(..., 1, ...),
1325 * but since we goto a block of code that's going to
1326 * search for the next \n if any, its safe here */
1328 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1329 " Looking for /%s^%s/m starting at offset %ld...\n",
1330 PL_colors[0], PL_colors[1],
1331 (long)(rx_origin - strpos)) );
1332 goto postprocess_substr_matches;
1335 /* strictly speaking this can never be true; but might
1336 * be if we ever allow intuit without substrings */
1337 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1340 rx_origin = rx_max_float;
1343 /* at this point, any matching substrings have been
1344 * contradicted. Start again... */
1346 rx_origin = HOP3c(rx_origin, 1, strend);
1348 /* uses bytes rather than char calculations for efficiency.
1349 * It's conservative: it errs on the side of doing 'goto restart',
1350 * where there is code that does a proper char-based test */
1351 if (rx_origin + start_shift + end_shift > strend) {
1352 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1353 " Could not match STCLASS...\n") );
1356 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1357 " Looking for %s substr starting at offset %ld...\n",
1358 (prog->substrs->check_ix ? "floating" : "anchored"),
1359 (long)(rx_origin + start_shift - strpos)) );
1365 if (rx_origin != s) {
1366 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1367 " By STCLASS: moving %ld --> %ld\n",
1368 (long)(rx_origin - strpos), (long)(s - strpos))
1372 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1373 " Does not contradict STCLASS...\n");
1378 /* Decide whether using the substrings helped */
1380 if (rx_origin != strpos) {
1381 /* Fixed substring is found far enough so that the match
1382 cannot start at strpos. */
1384 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1385 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1388 /* The found rx_origin position does not prohibit matching at
1389 * strpos, so calling intuit didn't gain us anything. Decrement
1390 * the BmUSEFUL() count on the check substring, and if we reach
1392 if (!(prog->intflags & PREGf_NAUGHTY)
1394 prog->check_utf8 /* Could be deleted already */
1395 && --BmUSEFUL(prog->check_utf8) < 0
1396 && (prog->check_utf8 == prog->float_utf8)
1398 prog->check_substr /* Could be deleted already */
1399 && --BmUSEFUL(prog->check_substr) < 0
1400 && (prog->check_substr == prog->float_substr)
1403 /* If flags & SOMETHING - do not do it many times on the same match */
1404 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1405 /* XXX Does the destruction order has to change with utf8_target? */
1406 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1407 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1408 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1409 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1410 check = NULL; /* abort */
1411 /* XXXX This is a remnant of the old implementation. It
1412 looks wasteful, since now INTUIT can use many
1413 other heuristics. */
1414 prog->extflags &= ~RXf_USE_INTUIT;
1418 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1419 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1420 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1424 fail_finish: /* Substring not found */
1425 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1426 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1428 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1429 PL_colors[4], PL_colors[5]));
1434 #define DECL_TRIE_TYPE(scan) \
1435 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1436 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
1437 trie_type = ((scan->flags == EXACT) \
1438 ? (utf8_target ? trie_utf8 : trie_plain) \
1439 : (scan->flags == EXACTFA) \
1440 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1441 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1443 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1446 U8 flags = FOLD_FLAGS_FULL; \
1447 switch (trie_type) { \
1448 case trie_utf8_exactfa_fold: \
1449 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1451 case trie_utf8_fold: \
1452 if ( foldlen>0 ) { \
1453 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1458 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1459 len = UTF8SKIP(uc); \
1460 skiplen = UNISKIP( uvc ); \
1461 foldlen -= skiplen; \
1462 uscan = foldbuf + skiplen; \
1465 case trie_latin_utf8_exactfa_fold: \
1466 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1468 case trie_latin_utf8_fold: \
1469 if ( foldlen>0 ) { \
1470 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1476 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1477 skiplen = UNISKIP( uvc ); \
1478 foldlen -= skiplen; \
1479 uscan = foldbuf + skiplen; \
1483 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1490 charid = trie->charmap[ uvc ]; \
1494 if (widecharmap) { \
1495 SV** const svpp = hv_fetch(widecharmap, \
1496 (char*)&uvc, sizeof(UV), 0); \
1498 charid = (U16)SvIV(*svpp); \
1503 #define DUMP_EXEC_POS(li,s,doutf8) \
1504 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1507 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1511 && (ln == 1 || folder(s, pat_string, ln)) \
1512 && (reginfo->intuit || regtry(reginfo, &s)) )\
1518 #define REXEC_FBC_UTF8_SCAN(CODE) \
1520 while (s < strend) { \
1526 #define REXEC_FBC_SCAN(CODE) \
1528 while (s < strend) { \
1534 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1535 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1537 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1546 #define REXEC_FBC_CLASS_SCAN(COND) \
1547 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1549 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1558 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1559 if (utf8_target) { \
1560 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1563 REXEC_FBC_CLASS_SCAN(COND); \
1566 /* The three macros below are slightly different versions of the same logic.
1568 * The first is for /a and /aa when the target string is UTF-8. This can only
1569 * match ascii, but it must advance based on UTF-8. The other two handle the
1570 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1571 * for the boundary (or non-boundary) between a word and non-word character.
1572 * The utf8 and non-utf8 cases have the same logic, but the details must be
1573 * different. Find the "wordness" of the character just prior to this one, and
1574 * compare it with the wordness of this one. If they differ, we have a
1575 * boundary. At the beginning of the string, pretend that the previous
1576 * character was a new-line.
1578 * All these macros uncleanly have side-effects with each other and outside
1579 * variables. So far it's been too much trouble to clean-up
1581 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1582 * a word character or not.
1583 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1585 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1587 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1588 * are looking for a boundary or for a non-boundary. If we are looking for a
1589 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1590 * see if this tentative match actually works, and if so, to quit the loop
1591 * here. And vice-versa if we are looking for a non-boundary.
1593 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1594 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1595 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1596 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1597 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1598 * complement. But in that branch we complement tmp, meaning that at the
1599 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1600 * which means at the top of the loop in the next iteration, it is
1601 * TEST_NON_UTF8(s-1) */
1602 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1603 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1604 tmp = TEST_NON_UTF8(tmp); \
1605 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1606 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1608 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1615 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1616 * TEST_UTF8 is a macro that for the same input code points returns identically
1617 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1618 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1619 if (s == reginfo->strbeg) { \
1622 else { /* Back-up to the start of the previous character */ \
1623 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1624 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1625 0, UTF8_ALLOW_DEFAULT); \
1627 tmp = TEST_UV(tmp); \
1628 LOAD_UTF8_CHARCLASS_ALNUM(); \
1629 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1630 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1639 /* Like the above two macros. UTF8_CODE is the complete code for handling
1640 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1642 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1643 if (utf8_target) { \
1646 else { /* Not utf8 */ \
1647 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1648 tmp = TEST_NON_UTF8(tmp); \
1649 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1650 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1659 /* Here, things have been set up by the previous code so that tmp is the \
1660 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1661 * utf8ness of the target). We also have to check if this matches against \
1662 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1663 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1665 if (tmp == ! TEST_NON_UTF8('\n')) { \
1672 /* This is the macro to use when we want to see if something that looks like it
1673 * could match, actually does, and if so exits the loop */
1674 #define REXEC_FBC_TRYIT \
1675 if ((reginfo->intuit || regtry(reginfo, &s))) \
1678 /* The only difference between the BOUND and NBOUND cases is that
1679 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1680 * NBOUND. This is accomplished by passing it as either the if or else clause,
1681 * with the other one being empty (PLACEHOLDER is defined as empty).
1683 * The TEST_FOO parameters are for operating on different forms of input, but
1684 * all should be ones that return identically for the same underlying code
1686 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1688 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1689 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1691 #define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1693 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1694 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1696 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1698 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1699 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1701 #define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1703 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1704 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1707 /* We know what class REx starts with. Try to find this position... */
1708 /* if reginfo->intuit, its a dryrun */
1709 /* annoyingly all the vars in this routine have different names from their counterparts
1710 in regmatch. /grrr */
1712 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1713 const char *strend, regmatch_info *reginfo)
1716 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1717 char *pat_string; /* The pattern's exactish string */
1718 char *pat_end; /* ptr to end char of pat_string */
1719 re_fold_t folder; /* Function for computing non-utf8 folds */
1720 const U8 *fold_array; /* array for folding ords < 256 */
1726 I32 tmp = 1; /* Scratch variable? */
1727 const bool utf8_target = reginfo->is_utf8_target;
1728 UV utf8_fold_flags = 0;
1729 const bool is_utf8_pat = reginfo->is_utf8_pat;
1730 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1731 with a result inverts that result, as 0^1 =
1733 _char_class_number classnum;
1735 RXi_GET_DECL(prog,progi);
1737 PERL_ARGS_ASSERT_FIND_BYCLASS;
1739 /* We know what class it must start with. */
1743 REXEC_FBC_UTF8_CLASS_SCAN(
1744 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1747 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1752 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1759 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1760 assert(! is_utf8_pat);
1763 if (is_utf8_pat || utf8_target) {
1764 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1765 goto do_exactf_utf8;
1767 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1768 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1769 goto do_exactf_non_utf8; /* isn't dealt with by these */
1771 case EXACTF: /* This node only generated for non-utf8 patterns */
1772 assert(! is_utf8_pat);
1774 utf8_fold_flags = 0;
1775 goto do_exactf_utf8;
1777 fold_array = PL_fold;
1779 goto do_exactf_non_utf8;
1782 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1783 utf8_fold_flags = FOLDEQ_LOCALE;
1784 goto do_exactf_utf8;
1786 fold_array = PL_fold_locale;
1787 folder = foldEQ_locale;
1788 goto do_exactf_non_utf8;
1792 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1794 goto do_exactf_utf8;
1797 if (is_utf8_pat || utf8_target) {
1798 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1799 goto do_exactf_utf8;
1802 /* Any 'ss' in the pattern should have been replaced by regcomp,
1803 * so we don't have to worry here about this single special case
1804 * in the Latin1 range */
1805 fold_array = PL_fold_latin1;
1806 folder = foldEQ_latin1;
1810 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1811 are no glitches with fold-length differences
1812 between the target string and pattern */
1814 /* The idea in the non-utf8 EXACTF* cases is to first find the
1815 * first character of the EXACTF* node and then, if necessary,
1816 * case-insensitively compare the full text of the node. c1 is the
1817 * first character. c2 is its fold. This logic will not work for
1818 * Unicode semantics and the german sharp ss, which hence should
1819 * not be compiled into a node that gets here. */
1820 pat_string = STRING(c);
1821 ln = STR_LEN(c); /* length to match in octets/bytes */
1823 /* We know that we have to match at least 'ln' bytes (which is the
1824 * same as characters, since not utf8). If we have to match 3
1825 * characters, and there are only 2 availabe, we know without
1826 * trying that it will fail; so don't start a match past the
1827 * required minimum number from the far end */
1828 e = HOP3c(strend, -((SSize_t)ln), s);
1830 if (reginfo->intuit && e < s) {
1831 e = s; /* Due to minlen logic of intuit() */
1835 c2 = fold_array[c1];
1836 if (c1 == c2) { /* If char and fold are the same */
1837 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1840 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1848 /* If one of the operands is in utf8, we can't use the simpler folding
1849 * above, due to the fact that many different characters can have the
1850 * same fold, or portion of a fold, or different- length fold */
1851 pat_string = STRING(c);
1852 ln = STR_LEN(c); /* length to match in octets/bytes */
1853 pat_end = pat_string + ln;
1854 lnc = is_utf8_pat /* length to match in characters */
1855 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1858 /* We have 'lnc' characters to match in the pattern, but because of
1859 * multi-character folding, each character in the target can match
1860 * up to 3 characters (Unicode guarantees it will never exceed
1861 * this) if it is utf8-encoded; and up to 2 if not (based on the
1862 * fact that the Latin 1 folds are already determined, and the
1863 * only multi-char fold in that range is the sharp-s folding to
1864 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1865 * string character. Adjust lnc accordingly, rounding up, so that
1866 * if we need to match at least 4+1/3 chars, that really is 5. */
1867 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1868 lnc = (lnc + expansion - 1) / expansion;
1870 /* As in the non-UTF8 case, if we have to match 3 characters, and
1871 * only 2 are left, it's guaranteed to fail, so don't start a
1872 * match that would require us to go beyond the end of the string
1874 e = HOP3c(strend, -((SSize_t)lnc), s);
1876 if (reginfo->intuit && e < s) {
1877 e = s; /* Due to minlen logic of intuit() */
1880 /* XXX Note that we could recalculate e to stop the loop earlier,
1881 * as the worst case expansion above will rarely be met, and as we
1882 * go along we would usually find that e moves further to the left.
1883 * This would happen only after we reached the point in the loop
1884 * where if there were no expansion we should fail. Unclear if
1885 * worth the expense */
1888 char *my_strend= (char *)strend;
1889 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1890 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1891 && (reginfo->intuit || regtry(reginfo, &s)) )
1895 s += (utf8_target) ? UTF8SKIP(s) : 1;
1901 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1904 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1907 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1910 FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1913 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1916 FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1919 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1922 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1925 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1926 is_LNBREAK_latin1_safe(s, strend)
1930 /* The argument to all the POSIX node types is the class number to pass to
1931 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1938 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1939 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1954 /* The complement of something that matches only ASCII matches all
1955 * non-ASCII, plus everything in ASCII that isn't in the class. */
1956 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
1957 || ! _generic_isCC_A(*s, FLAGS(c)));
1966 /* Don't need to worry about utf8, as it can match only a single
1967 * byte invariant character. */
1968 REXEC_FBC_CLASS_SCAN(
1969 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1977 if (! utf8_target) {
1978 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1984 classnum = (_char_class_number) FLAGS(c);
1985 if (classnum < _FIRST_NON_SWASH_CC) {
1986 while (s < strend) {
1988 /* We avoid loading in the swash as long as possible, but
1989 * should we have to, we jump to a separate loop. This
1990 * extra 'if' statement is what keeps this code from being
1991 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1992 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1993 goto found_above_latin1;
1995 if ((UTF8_IS_INVARIANT(*s)
1996 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1998 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1999 && to_complement ^ cBOOL(
2000 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2004 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2016 else switch (classnum) { /* These classes are implemented as
2018 case _CC_ENUM_SPACE: /* XXX would require separate code if we
2019 revert the change of \v matching this */
2022 case _CC_ENUM_PSXSPC:
2023 REXEC_FBC_UTF8_CLASS_SCAN(
2024 to_complement ^ cBOOL(isSPACE_utf8(s)));
2027 case _CC_ENUM_BLANK:
2028 REXEC_FBC_UTF8_CLASS_SCAN(
2029 to_complement ^ cBOOL(isBLANK_utf8(s)));
2032 case _CC_ENUM_XDIGIT:
2033 REXEC_FBC_UTF8_CLASS_SCAN(
2034 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2037 case _CC_ENUM_VERTSPACE:
2038 REXEC_FBC_UTF8_CLASS_SCAN(
2039 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2042 case _CC_ENUM_CNTRL:
2043 REXEC_FBC_UTF8_CLASS_SCAN(
2044 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2048 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2049 assert(0); /* NOTREACHED */
2054 found_above_latin1: /* Here we have to load a swash to get the result
2055 for the current code point */
2056 if (! PL_utf8_swash_ptrs[classnum]) {
2057 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2058 PL_utf8_swash_ptrs[classnum] =
2059 _core_swash_init("utf8",
2062 PL_XPosix_ptrs[classnum], &flags);
2065 /* This is a copy of the loop above for swash classes, though using the
2066 * FBC macro instead of being expanded out. Since we've loaded the
2067 * swash, we don't have to check for that each time through the loop */
2068 REXEC_FBC_UTF8_CLASS_SCAN(
2069 to_complement ^ cBOOL(_generic_utf8(
2072 swash_fetch(PL_utf8_swash_ptrs[classnum],
2080 /* what trie are we using right now */
2081 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2082 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2083 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2085 const char *last_start = strend - trie->minlen;
2087 const char *real_start = s;
2089 STRLEN maxlen = trie->maxlen;
2091 U8 **points; /* map of where we were in the input string
2092 when reading a given char. For ASCII this
2093 is unnecessary overhead as the relationship
2094 is always 1:1, but for Unicode, especially
2095 case folded Unicode this is not true. */
2096 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2100 GET_RE_DEBUG_FLAGS_DECL;
2102 /* We can't just allocate points here. We need to wrap it in
2103 * an SV so it gets freed properly if there is a croak while
2104 * running the match */
2107 sv_points=newSV(maxlen * sizeof(U8 *));
2108 SvCUR_set(sv_points,
2109 maxlen * sizeof(U8 *));
2110 SvPOK_on(sv_points);
2111 sv_2mortal(sv_points);
2112 points=(U8**)SvPV_nolen(sv_points );
2113 if ( trie_type != trie_utf8_fold
2114 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2117 bitmap=(U8*)trie->bitmap;
2119 bitmap=(U8*)ANYOF_BITMAP(c);
2121 /* this is the Aho-Corasick algorithm modified a touch
2122 to include special handling for long "unknown char" sequences.
2123 The basic idea being that we use AC as long as we are dealing
2124 with a possible matching char, when we encounter an unknown char
2125 (and we have not encountered an accepting state) we scan forward
2126 until we find a legal starting char.
2127 AC matching is basically that of trie matching, except that when
2128 we encounter a failing transition, we fall back to the current
2129 states "fail state", and try the current char again, a process
2130 we repeat until we reach the root state, state 1, or a legal
2131 transition. If we fail on the root state then we can either
2132 terminate if we have reached an accepting state previously, or
2133 restart the entire process from the beginning if we have not.
2136 while (s <= last_start) {
2137 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2145 U8 *uscan = (U8*)NULL;
2146 U8 *leftmost = NULL;
2148 U32 accepted_word= 0;
2152 while ( state && uc <= (U8*)strend ) {
2154 U32 word = aho->states[ state ].wordnum;
2158 DEBUG_TRIE_EXECUTE_r(
2159 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2160 dump_exec_pos( (char *)uc, c, strend, real_start,
2161 (char *)uc, utf8_target );
2162 PerlIO_printf( Perl_debug_log,
2163 " Scanning for legal start char...\n");
2167 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2171 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2177 if (uc >(U8*)last_start) break;
2181 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2182 if (!leftmost || lpos < leftmost) {
2183 DEBUG_r(accepted_word=word);
2189 points[pointpos++ % maxlen]= uc;
2190 if (foldlen || uc < (U8*)strend) {
2191 REXEC_TRIE_READ_CHAR(trie_type, trie,
2193 uscan, len, uvc, charid, foldlen,
2195 DEBUG_TRIE_EXECUTE_r({
2196 dump_exec_pos( (char *)uc, c, strend,
2197 real_start, s, utf8_target);
2198 PerlIO_printf(Perl_debug_log,
2199 " Charid:%3u CP:%4"UVxf" ",
2211 word = aho->states[ state ].wordnum;
2213 base = aho->states[ state ].trans.base;
2215 DEBUG_TRIE_EXECUTE_r({
2217 dump_exec_pos( (char *)uc, c, strend, real_start,
2219 PerlIO_printf( Perl_debug_log,
2220 "%sState: %4"UVxf", word=%"UVxf,
2221 failed ? " Fail transition to " : "",
2222 (UV)state, (UV)word);
2228 ( ((offset = base + charid
2229 - 1 - trie->uniquecharcount)) >= 0)
2230 && ((U32)offset < trie->lasttrans)
2231 && trie->trans[offset].check == state
2232 && (tmp=trie->trans[offset].next))
2234 DEBUG_TRIE_EXECUTE_r(
2235 PerlIO_printf( Perl_debug_log," - legal\n"));
2240 DEBUG_TRIE_EXECUTE_r(
2241 PerlIO_printf( Perl_debug_log," - fail\n"));
2243 state = aho->fail[state];
2247 /* we must be accepting here */
2248 DEBUG_TRIE_EXECUTE_r(
2249 PerlIO_printf( Perl_debug_log," - accepting\n"));
2258 if (!state) state = 1;
2261 if ( aho->states[ state ].wordnum ) {
2262 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2263 if (!leftmost || lpos < leftmost) {
2264 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2269 s = (char*)leftmost;
2270 DEBUG_TRIE_EXECUTE_r({
2272 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2273 (UV)accepted_word, (IV)(s - real_start)
2276 if (reginfo->intuit || regtry(reginfo, &s)) {
2282 DEBUG_TRIE_EXECUTE_r({
2283 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2286 DEBUG_TRIE_EXECUTE_r(
2287 PerlIO_printf( Perl_debug_log,"No match.\n"));
2296 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2303 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2304 * flags have same meanings as with regexec_flags() */
2307 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2314 struct regexp *const prog = ReANY(rx);
2316 if (flags & REXEC_COPY_STR) {
2320 PerlIO_printf(Perl_debug_log,
2321 "Copy on write: regexp capture, type %d\n",
2324 /* Create a new COW SV to share the match string and store
2325 * in saved_copy, unless the current COW SV in saved_copy
2326 * is valid and suitable for our purpose */
2327 if (( prog->saved_copy
2328 && SvIsCOW(prog->saved_copy)
2329 && SvPOKp(prog->saved_copy)
2332 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2334 /* just reuse saved_copy SV */
2335 if (RXp_MATCH_COPIED(prog)) {
2336 Safefree(prog->subbeg);
2337 RXp_MATCH_COPIED_off(prog);
2341 /* create new COW SV to share string */
2342 RX_MATCH_COPY_FREE(rx);
2343 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2345 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2346 assert (SvPOKp(prog->saved_copy));
2347 prog->sublen = strend - strbeg;
2348 prog->suboffset = 0;
2349 prog->subcoffset = 0;
2354 SSize_t max = strend - strbeg;
2357 if ( (flags & REXEC_COPY_SKIP_POST)
2358 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2359 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2360 ) { /* don't copy $' part of string */
2363 /* calculate the right-most part of the string covered
2364 * by a capture. Due to look-ahead, this may be to
2365 * the right of $&, so we have to scan all captures */
2366 while (n <= prog->lastparen) {
2367 if (prog->offs[n].end > max)
2368 max = prog->offs[n].end;
2372 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2373 ? prog->offs[0].start
2375 assert(max >= 0 && max <= strend - strbeg);
2378 if ( (flags & REXEC_COPY_SKIP_PRE)
2379 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2380 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2381 ) { /* don't copy $` part of string */
2384 /* calculate the left-most part of the string covered
2385 * by a capture. Due to look-behind, this may be to
2386 * the left of $&, so we have to scan all captures */
2387 while (min && n <= prog->lastparen) {
2388 if ( prog->offs[n].start != -1
2389 && prog->offs[n].start < min)
2391 min = prog->offs[n].start;
2395 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2396 && min > prog->offs[0].end
2398 min = prog->offs[0].end;
2402 assert(min >= 0 && min <= max && min <= strend - strbeg);
2405 if (RX_MATCH_COPIED(rx)) {
2406 if (sublen > prog->sublen)
2408 (char*)saferealloc(prog->subbeg, sublen+1);
2411 prog->subbeg = (char*)safemalloc(sublen+1);
2412 Copy(strbeg + min, prog->subbeg, sublen, char);
2413 prog->subbeg[sublen] = '\0';
2414 prog->suboffset = min;
2415 prog->sublen = sublen;
2416 RX_MATCH_COPIED_on(rx);
2418 prog->subcoffset = prog->suboffset;
2419 if (prog->suboffset && utf8_target) {
2420 /* Convert byte offset to chars.
2421 * XXX ideally should only compute this if @-/@+
2422 * has been seen, a la PL_sawampersand ??? */
2424 /* If there's a direct correspondence between the
2425 * string which we're matching and the original SV,
2426 * then we can use the utf8 len cache associated with
2427 * the SV. In particular, it means that under //g,
2428 * sv_pos_b2u() will use the previously cached
2429 * position to speed up working out the new length of
2430 * subcoffset, rather than counting from the start of
2431 * the string each time. This stops
2432 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2433 * from going quadratic */
2434 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2435 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2436 SV_GMAGIC|SV_CONST_RETURN);
2438 prog->subcoffset = utf8_length((U8*)strbeg,
2439 (U8*)(strbeg+prog->suboffset));
2443 RX_MATCH_COPY_FREE(rx);
2444 prog->subbeg = strbeg;
2445 prog->suboffset = 0;
2446 prog->subcoffset = 0;
2447 prog->sublen = strend - strbeg;
2455 - regexec_flags - match a regexp against a string
2458 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2459 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2460 /* stringarg: the point in the string at which to begin matching */
2461 /* strend: pointer to null at end of string */
2462 /* strbeg: real beginning of string */
2463 /* minend: end of match must be >= minend bytes after stringarg. */
2464 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2465 * itself is accessed via the pointers above */
2466 /* data: May be used for some additional optimizations.
2467 Currently unused. */
2468 /* flags: For optimizations. See REXEC_* in regexp.h */
2471 struct regexp *const prog = ReANY(rx);
2475 SSize_t minlen; /* must match at least this many chars */
2476 SSize_t dontbother = 0; /* how many characters not to try at end */
2477 const bool utf8_target = cBOOL(DO_UTF8(sv));
2479 RXi_GET_DECL(prog,progi);
2480 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2481 regmatch_info *const reginfo = ®info_buf;
2482 regexp_paren_pair *swap = NULL;
2484 GET_RE_DEBUG_FLAGS_DECL;
2486 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2487 PERL_UNUSED_ARG(data);
2489 /* Be paranoid... */
2490 if (prog == NULL || stringarg == NULL) {
2491 Perl_croak(aTHX_ "NULL regexp parameter");
2495 debug_start_match(rx, utf8_target, stringarg, strend,
2499 startpos = stringarg;
2501 if (prog->intflags & PREGf_GPOS_SEEN) {
2504 /* set reginfo->ganch, the position where \G can match */
2507 (flags & REXEC_IGNOREPOS)
2508 ? stringarg /* use start pos rather than pos() */
2509 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2510 /* Defined pos(): */
2511 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2512 : strbeg; /* pos() not defined; use start of string */
2514 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2515 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
2517 /* in the presence of \G, we may need to start looking earlier in
2518 * the string than the suggested start point of stringarg:
2519 * if prog->gofs is set, then that's a known, fixed minimum
2522 * /ab|c\G/: gofs = 1
2523 * or if the minimum offset isn't known, then we have to go back
2524 * to the start of the string, e.g. /w+\G/
2527 if (prog->intflags & PREGf_ANCH_GPOS) {
2528 startpos = reginfo->ganch - prog->gofs;
2530 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2532 DEBUG_r(PerlIO_printf(Perl_debug_log,
2533 "fail: ganch-gofs before earliest possible start\n"));
2537 else if (prog->gofs) {
2538 if (startpos - prog->gofs < strbeg)
2541 startpos -= prog->gofs;
2543 else if (prog->intflags & PREGf_GPOS_FLOAT)
2547 minlen = prog->minlen;
2548 if ((startpos + minlen) > strend || startpos < strbeg) {
2549 DEBUG_r(PerlIO_printf(Perl_debug_log,
2550 "Regex match can't succeed, so not even tried\n"));
2554 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2555 * which will call destuctors to reset PL_regmatch_state, free higher
2556 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2557 * regmatch_info_aux_eval */
2559 oldsave = PL_savestack_ix;
2563 if ((prog->extflags & RXf_USE_INTUIT)
2564 && !(flags & REXEC_CHECKED))
2566 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2571 if (prog->extflags & RXf_CHECK_ALL) {
2572 /* we can match based purely on the result of INTUIT.
2573 * Set up captures etc just for $& and $-[0]
2574 * (an intuit-only match wont have $1,$2,..) */
2575 assert(!prog->nparens);
2577 /* s/// doesn't like it if $& is earlier than where we asked it to
2578 * start searching (which can happen on something like /.\G/) */
2579 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2582 /* this should only be possible under \G */
2583 assert(prog->intflags & PREGf_GPOS_SEEN);
2584 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2585 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2589 /* match via INTUIT shouldn't have any captures.
2590 * Let @-, @+, $^N know */
2591 prog->lastparen = prog->lastcloseparen = 0;
2592 RX_MATCH_UTF8_set(rx, utf8_target);
2593 prog->offs[0].start = s - strbeg;
2594 prog->offs[0].end = utf8_target
2595 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2596 : s - strbeg + prog->minlenret;
2597 if ( !(flags & REXEC_NOT_FIRST) )
2598 S_reg_set_capture_string(aTHX_ rx,
2600 sv, flags, utf8_target);
2606 multiline = prog->extflags & RXf_PMf_MULTILINE;
2608 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2609 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2610 "String too short [regexec_flags]...\n"));
2614 /* Check validity of program. */
2615 if (UCHARAT(progi->program) != REG_MAGIC) {
2616 Perl_croak(aTHX_ "corrupted regexp program");
2619 RX_MATCH_TAINTED_off(rx);
2621 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2622 reginfo->intuit = 0;
2623 reginfo->is_utf8_target = cBOOL(utf8_target);
2624 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2625 reginfo->warned = FALSE;
2626 reginfo->strbeg = strbeg;
2628 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2629 reginfo->strend = strend;
2630 /* see how far we have to get to not match where we matched before */
2631 reginfo->till = stringarg + minend;
2633 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
2634 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2635 S_cleanup_regmatch_info_aux has executed (registered by
2636 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2637 magic belonging to this SV.
2638 Not newSVsv, either, as it does not COW.
2640 assert(!IS_PADGV(sv));
2641 reginfo->sv = newSV(0);
2642 SvSetSV_nosteal(reginfo->sv, sv);
2643 SAVEFREESV(reginfo->sv);
2646 /* reserve next 2 or 3 slots in PL_regmatch_state:
2647 * slot N+0: may currently be in use: skip it
2648 * slot N+1: use for regmatch_info_aux struct
2649 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2650 * slot N+3: ready for use by regmatch()
2654 regmatch_state *old_regmatch_state;
2655 regmatch_slab *old_regmatch_slab;
2656 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2658 /* on first ever match, allocate first slab */
2659 if (!PL_regmatch_slab) {
2660 Newx(PL_regmatch_slab, 1, regmatch_slab);
2661 PL_regmatch_slab->prev = NULL;
2662 PL_regmatch_slab->next = NULL;
2663 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2666 old_regmatch_state = PL_regmatch_state;
2667 old_regmatch_slab = PL_regmatch_slab;
2669 for (i=0; i <= max; i++) {
2671 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2673 reginfo->info_aux_eval =
2674 reginfo->info_aux->info_aux_eval =
2675 &(PL_regmatch_state->u.info_aux_eval);
2677 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2678 PL_regmatch_state = S_push_slab(aTHX);
2681 /* note initial PL_regmatch_state position; at end of match we'll
2682 * pop back to there and free any higher slabs */
2684 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2685 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2686 reginfo->info_aux->poscache = NULL;
2688 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2690 if ((prog->extflags & RXf_EVAL_SEEN))
2691 S_setup_eval_state(aTHX_ reginfo);
2693 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2696 /* If there is a "must appear" string, look for it. */
2698 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2699 /* We have to be careful. If the previous successful match
2700 was from this regex we don't want a subsequent partially
2701 successful match to clobber the old results.
2702 So when we detect this possibility we add a swap buffer
2703 to the re, and switch the buffer each match. If we fail,
2704 we switch it back; otherwise we leave it swapped.
2707 /* do we need a save destructor here for eval dies? */
2708 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2709 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2710 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2717 /* Simplest case: anchored match need be tried only once. */
2718 /* [unless only anchor is BOL and multiline is set] */
2719 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
2720 if (s == startpos && regtry(reginfo, &s))
2722 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
2727 dontbother = minlen - 1;
2728 end = HOP3c(strend, -dontbother, strbeg) - 1;
2729 /* for multiline we only have to try after newlines */
2730 if (prog->check_substr || prog->check_utf8) {
2731 /* because of the goto we can not easily reuse the macros for bifurcating the
2732 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2735 goto after_try_utf8;
2737 if (regtry(reginfo, &s)) {
2744 if (prog->extflags & RXf_USE_INTUIT) {
2745 s = re_intuit_start(rx, sv, strbeg,
2746 s + UTF8SKIP(s), strend, flags, NULL);
2755 } /* end search for check string in unicode */
2757 if (s == startpos) {
2758 goto after_try_latin;
2761 if (regtry(reginfo, &s)) {
2768 if (prog->extflags & RXf_USE_INTUIT) {
2769 s = re_intuit_start(rx, sv, strbeg,
2770 s + 1, strend, flags, NULL);
2779 } /* end search for check string in latin*/
2780 } /* end search for check string */
2781 else { /* search for newline */
2783 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2786 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2787 while (s <= end) { /* note it could be possible to match at the end of the string */
2788 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2789 if (regtry(reginfo, &s))
2793 } /* end search for newline */
2794 } /* end anchored/multiline check string search */
2796 } else if (prog->intflags & PREGf_ANCH_GPOS)
2798 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2799 assert(prog->intflags & PREGf_GPOS_SEEN);
2800 /* For anchored \G, the only position it can match from is
2801 * (ganch-gofs); we already set startpos to this above; if intuit
2802 * moved us on from there, we can't possibly succeed */
2803 assert(startpos == reginfo->ganch - prog->gofs);
2804 if (s == startpos && regtry(reginfo, &s))
2809 /* Messy cases: unanchored match. */
2810 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2811 /* we have /x+whatever/ */
2812 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2818 if (! prog->anchored_utf8) {
2819 to_utf8_substr(prog);
2821 ch = SvPVX_const(prog->anchored_utf8)[0];
2824 DEBUG_EXECUTE_r( did_match = 1 );
2825 if (regtry(reginfo, &s)) goto got_it;
2827 while (s < strend && *s == ch)
2834 if (! prog->anchored_substr) {
2835 if (! to_byte_substr(prog)) {
2836 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2839 ch = SvPVX_const(prog->anchored_substr)[0];
2842 DEBUG_EXECUTE_r( did_match = 1 );
2843 if (regtry(reginfo, &s)) goto got_it;
2845 while (s < strend && *s == ch)
2850 DEBUG_EXECUTE_r(if (!did_match)
2851 PerlIO_printf(Perl_debug_log,
2852 "Did not find anchored character...\n")
2855 else if (prog->anchored_substr != NULL
2856 || prog->anchored_utf8 != NULL
2857 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2858 && prog->float_max_offset < strend - s)) {
2863 char *last1; /* Last position checked before */
2867 if (prog->anchored_substr || prog->anchored_utf8) {
2869 if (! prog->anchored_utf8) {
2870 to_utf8_substr(prog);
2872 must = prog->anchored_utf8;
2875 if (! prog->anchored_substr) {
2876 if (! to_byte_substr(prog)) {
2877 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2880 must = prog->anchored_substr;
2882 back_max = back_min = prog->anchored_offset;
2885 if (! prog->float_utf8) {
2886 to_utf8_substr(prog);
2888 must = prog->float_utf8;
2891 if (! prog->float_substr) {
2892 if (! to_byte_substr(prog)) {
2893 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2896 must = prog->float_substr;
2898 back_max = prog->float_max_offset;
2899 back_min = prog->float_min_offset;
2905 last = HOP3c(strend, /* Cannot start after this */
2906 -(SSize_t)(CHR_SVLEN(must)
2907 - (SvTAIL(must) != 0) + back_min), strbeg);
2909 if (s > reginfo->strbeg)
2910 last1 = HOPc(s, -1);
2912 last1 = s - 1; /* bogus */
2914 /* XXXX check_substr already used to find "s", can optimize if
2915 check_substr==must. */
2917 strend = HOPc(strend, -dontbother);
2918 while ( (s <= last) &&
2919 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
2920 (unsigned char*)strend, must,
2921 multiline ? FBMrf_MULTILINE : 0)) ) {
2922 DEBUG_EXECUTE_r( did_match = 1 );
2923 if (HOPc(s, -back_max) > last1) {
2924 last1 = HOPc(s, -back_min);
2925 s = HOPc(s, -back_max);
2928 char * const t = (last1 >= reginfo->strbeg)
2929 ? HOPc(last1, 1) : last1 + 1;
2931 last1 = HOPc(s, -back_min);
2935 while (s <= last1) {
2936 if (regtry(reginfo, &s))
2939 s++; /* to break out of outer loop */
2946 while (s <= last1) {
2947 if (regtry(reginfo, &s))
2953 DEBUG_EXECUTE_r(if (!did_match) {
2954 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2955 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2956 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2957 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2958 ? "anchored" : "floating"),
2959 quoted, RE_SV_TAIL(must));
2963 else if ( (c = progi->regstclass) ) {
2965 const OPCODE op = OP(progi->regstclass);
2966 /* don't bother with what can't match */
2967 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2968 strend = HOPc(strend, -(minlen - 1));
2971 SV * const prop = sv_newmortal();
2972 regprop(prog, prop, c, reginfo);
2974 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2976 PerlIO_printf(Perl_debug_log,
2977 "Matching stclass %.*s against %s (%d bytes)\n",
2978 (int)SvCUR(prop), SvPVX_const(prop),
2979 quoted, (int)(strend - s));
2982 if (find_byclass(prog, c, s, strend, reginfo))
2984 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2988 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2996 if (! prog->float_utf8) {
2997 to_utf8_substr(prog);
2999 float_real = prog->float_utf8;
3002 if (! prog->float_substr) {
3003 if (! to_byte_substr(prog)) {
3004 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3007 float_real = prog->float_substr;
3010 little = SvPV_const(float_real, len);
3011 if (SvTAIL(float_real)) {
3012 /* This means that float_real contains an artificial \n on
3013 * the end due to the presence of something like this:
3014 * /foo$/ where we can match both "foo" and "foo\n" at the
3015 * end of the string. So we have to compare the end of the
3016 * string first against the float_real without the \n and
3017 * then against the full float_real with the string. We
3018 * have to watch out for cases where the string might be
3019 * smaller than the float_real or the float_real without
3021 char *checkpos= strend - len;
3023 PerlIO_printf(Perl_debug_log,
3024 "%sChecking for float_real.%s\n",
3025 PL_colors[4], PL_colors[5]));
3026 if (checkpos + 1 < strbeg) {
3027 /* can't match, even if we remove the trailing \n
3028 * string is too short to match */
3030 PerlIO_printf(Perl_debug_log,
3031 "%sString shorter than required trailing substring, cannot match.%s\n",
3032 PL_colors[4], PL_colors[5]));
3034 } else if (memEQ(checkpos + 1, little, len - 1)) {
3035 /* can match, the end of the string matches without the
3037 last = checkpos + 1;
3038 } else if (checkpos < strbeg) {
3039 /* cant match, string is too short when the "\n" is
3042 PerlIO_printf(Perl_debug_log,
3043 "%sString does not contain required trailing substring, cannot match.%s\n",
3044 PL_colors[4], PL_colors[5]));
3046 } else if (!multiline) {
3047 /* non multiline match, so compare with the "\n" at the
3048 * end of the string */
3049 if (memEQ(checkpos, little, len)) {
3053 PerlIO_printf(Perl_debug_log,
3054 "%sString does not contain required trailing substring, cannot match.%s\n",
3055 PL_colors[4], PL_colors[5]));
3059 /* multiline match, so we have to search for a place
3060 * where the full string is located */
3066 last = rninstr(s, strend, little, little + len);
3068 last = strend; /* matching "$" */
3071 /* at one point this block contained a comment which was
3072 * probably incorrect, which said that this was a "should not
3073 * happen" case. Even if it was true when it was written I am
3074 * pretty sure it is not anymore, so I have removed the comment
3075 * and replaced it with this one. Yves */
3077 PerlIO_printf(Perl_debug_log,
3078 "String does not contain required substring, cannot match.\n"
3082 dontbother = strend - last + prog->float_min_offset;
3084 if (minlen && (dontbother < minlen))
3085 dontbother = minlen - 1;
3086 strend -= dontbother; /* this one's always in bytes! */
3087 /* We don't know much -- general case. */
3090 if (regtry(reginfo, &s))
3099 if (regtry(reginfo, &s))
3101 } while (s++ < strend);
3109 /* s/// doesn't like it if $& is earlier than where we asked it to
3110 * start searching (which can happen on something like /.\G/) */
3111 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3112 && (prog->offs[0].start < stringarg - strbeg))
3114 /* this should only be possible under \G */
3115 assert(prog->intflags & PREGf_GPOS_SEEN);
3116 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3117 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3123 PerlIO_printf(Perl_debug_log,
3124 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3131 /* clean up; this will trigger destructors that will free all slabs
3132 * above the current one, and cleanup the regmatch_info_aux
3133 * and regmatch_info_aux_eval sructs */
3135 LEAVE_SCOPE(oldsave);
3137 if (RXp_PAREN_NAMES(prog))
3138 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3140 RX_MATCH_UTF8_set(rx, utf8_target);
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 assert(0); /* 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);
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 BOL: /* /^../ */
4017 case SBOL: /* /^../s */
4018 if (locinput == reginfo->strbeg)
4022 case MBOL: /* /^../m */
4023 if (locinput == reginfo->strbeg ||
4024 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
4031 if (locinput == reginfo->ganch)
4035 case KEEPS: /* \K */
4036 /* update the startpoint */
4037 st->u.keeper.val = rex->offs[0].start;
4038 rex->offs[0].start = locinput - reginfo->strbeg;
4039 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
4043 case KEEPS_next_fail:
4044 /* rollback the start point change */
4045 rex->offs[0].start = st->u.keeper.val;
4050 case MEOL: /* /..$/m */
4051 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4055 case EOL: /* /..$/ */
4057 case SEOL: /* /..$/s */
4058 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4060 if (reginfo->strend - locinput > 1)
4065 if (!NEXTCHR_IS_EOS)
4069 case SANY: /* /./s */
4072 goto increment_locinput;
4080 case REG_ANY: /* /./ */
4081 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
4083 goto increment_locinput;
4087 #define ST st->u.trie
4088 case TRIEC: /* (ab|cd) with known charclass */
4089 /* In this case the charclass data is available inline so
4090 we can fail fast without a lot of extra overhead.
4092 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
4094 PerlIO_printf(Perl_debug_log,
4095 "%*s %sfailed to match trie start class...%s\n",
4096 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4103 case TRIE: /* (ab|cd) */
4104 /* the basic plan of execution of the trie is:
4105 * At the beginning, run though all the states, and
4106 * find the longest-matching word. Also remember the position
4107 * of the shortest matching word. For example, this pattern:
4110 * when matched against the string "abcde", will generate
4111 * accept states for all words except 3, with the longest
4112 * matching word being 4, and the shortest being 2 (with
4113 * the position being after char 1 of the string).
4115 * Then for each matching word, in word order (i.e. 1,2,4,5),
4116 * we run the remainder of the pattern; on each try setting
4117 * the current position to the character following the word,
4118 * returning to try the next word on failure.
4120 * We avoid having to build a list of words at runtime by
4121 * using a compile-time structure, wordinfo[].prev, which
4122 * gives, for each word, the previous accepting word (if any).
4123 * In the case above it would contain the mappings 1->2, 2->0,
4124 * 3->0, 4->5, 5->1. We can use this table to generate, from
4125 * the longest word (4 above), a list of all words, by
4126 * following the list of prev pointers; this gives us the
4127 * unordered list 4,5,1,2. Then given the current word we have
4128 * just tried, we can go through the list and find the
4129 * next-biggest word to try (so if we just failed on word 2,
4130 * the next in the list is 4).
4132 * Since at runtime we don't record the matching position in
4133 * the string for each word, we have to work that out for
4134 * each word we're about to process. The wordinfo table holds
4135 * the character length of each word; given that we recorded
4136 * at the start: the position of the shortest word and its
4137 * length in chars, we just need to move the pointer the
4138 * difference between the two char lengths. Depending on
4139 * Unicode status and folding, that's cheap or expensive.
4141 * This algorithm is optimised for the case where are only a
4142 * small number of accept states, i.e. 0,1, or maybe 2.
4143 * With lots of accepts states, and having to try all of them,
4144 * it becomes quadratic on number of accept states to find all
4149 /* what type of TRIE am I? (utf8 makes this contextual) */
4150 DECL_TRIE_TYPE(scan);
4152 /* what trie are we using right now */
4153 reg_trie_data * const trie
4154 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
4155 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
4156 U32 state = trie->startstate;
4159 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
4161 if (trie->states[ state ].wordnum) {
4163 PerlIO_printf(Perl_debug_log,
4164 "%*s %smatched empty string...%s\n",
4165 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4171 PerlIO_printf(Perl_debug_log,
4172 "%*s %sfailed to match trie start class...%s\n",
4173 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4180 U8 *uc = ( U8* )locinput;
4184 U8 *uscan = (U8*)NULL;
4185 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
4186 U32 charcount = 0; /* how many input chars we have matched */
4187 U32 accepted = 0; /* have we seen any accepting states? */
4189 ST.jump = trie->jump;
4192 ST.longfold = FALSE; /* char longer if folded => it's harder */
4195 /* fully traverse the TRIE; note the position of the
4196 shortest accept state and the wordnum of the longest
4199 while ( state && uc <= (U8*)(reginfo->strend) ) {
4200 U32 base = trie->states[ state ].trans.base;
4204 wordnum = trie->states[ state ].wordnum;
4206 if (wordnum) { /* it's an accept state */
4209 /* record first match position */
4211 ST.firstpos = (U8*)locinput;
4216 ST.firstchars = charcount;
4219 if (!ST.nextword || wordnum < ST.nextword)
4220 ST.nextword = wordnum;
4221 ST.topword = wordnum;
4224 DEBUG_TRIE_EXECUTE_r({
4225 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
4226 PerlIO_printf( Perl_debug_log,
4227 "%*s %sState: %4"UVxf" Accepted: %c ",
4228 2+depth * 2, "", PL_colors[4],
4229 (UV)state, (accepted ? 'Y' : 'N'));
4232 /* read a char and goto next state */
4233 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
4235 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
4236 uscan, len, uvc, charid, foldlen,
4243 base + charid - 1 - trie->uniquecharcount)) >= 0)
4245 && ((U32)offset < trie->lasttrans)
4246 && trie->trans[offset].check == state)
4248 state = trie->trans[offset].next;
4259 DEBUG_TRIE_EXECUTE_r(
4260 PerlIO_printf( Perl_debug_log,
4261 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
4262 charid, uvc, (UV)state, PL_colors[5] );
4268 /* calculate total number of accept states */
4273 w = trie->wordinfo[w].prev;
4276 ST.accepted = accepted;
4280 PerlIO_printf( Perl_debug_log,
4281 "%*s %sgot %"IVdf" possible matches%s\n",
4282 REPORT_CODE_OFF + depth * 2, "",
4283 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
4285 goto trie_first_try; /* jump into the fail handler */
4290 case TRIE_next_fail: /* we failed - try next alternative */
4294 REGCP_UNWIND(ST.cp);
4295 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
4297 if (!--ST.accepted) {
4299 PerlIO_printf( Perl_debug_log,
4300 "%*s %sTRIE failed...%s\n",
4301 REPORT_CODE_OFF+depth*2, "",
4308 /* Find next-highest word to process. Note that this code
4309 * is O(N^2) per trie run (O(N) per branch), so keep tight */
4312 U16 const nextword = ST.nextword;
4313 reg_trie_wordinfo * const wordinfo
4314 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
4315 for (word=ST.topword; word; word=wordinfo[word].prev) {
4316 if (word > nextword && (!min || word < min))
4329 ST.lastparen = rex->lastparen;
4330 ST.lastcloseparen = rex->lastcloseparen;
4334 /* find start char of end of current word */
4336 U32 chars; /* how many chars to skip */
4337 reg_trie_data * const trie
4338 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
4340 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
4342 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
4347 /* the hard option - fold each char in turn and find
4348 * its folded length (which may be different */
4349 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4357 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
4365 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4370 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
4386 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4387 ? ST.jump[ST.nextword]
4391 PerlIO_printf( Perl_debug_log,
4392 "%*s %sTRIE matched word #%d, continuing%s\n",
4393 REPORT_CODE_OFF+depth*2, "",
4400 if (ST.accepted > 1 || has_cutgroup) {
4401 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4405 /* only one choice left - just continue */
4407 AV *const trie_words
4408 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4409 SV ** const tmp = av_fetch( trie_words,
4411 SV *sv= tmp ? sv_newmortal() : NULL;
4413 PerlIO_printf( Perl_debug_log,
4414 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4415 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4417 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4418 PL_colors[0], PL_colors[1],
4419 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4421 : "not compiled under -Dr",
4425 locinput = (char*)uc;
4426 continue; /* execute rest of RE */
4432 case EXACT: { /* /abc/ */
4433 char *s = STRING(scan);
4435 if (utf8_target != is_utf8_pat) {
4436 /* The target and the pattern have differing utf8ness. */
4438 const char * const e = s + ln;
4441 /* The target is utf8, the pattern is not utf8.
4442 * Above-Latin1 code points can't match the pattern;
4443 * invariants match exactly, and the other Latin1 ones need
4444 * to be downgraded to a single byte in order to do the
4445 * comparison. (If we could be confident that the target
4446 * is not malformed, this could be refactored to have fewer
4447 * tests by just assuming that if the first bytes match, it
4448 * is an invariant, but there are tests in the test suite
4449 * dealing with (??{...}) which violate this) */
4451 if (l >= reginfo->strend
4452 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4456 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4463 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
4473 /* The target is not utf8, the pattern is utf8. */
4475 if (l >= reginfo->strend
4476 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4480 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4487 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
4499 /* The target and the pattern have the same utf8ness. */
4500 /* Inline the first character, for speed. */
4501 if (reginfo->strend - locinput < ln
4502 || UCHARAT(s) != nextchr
4503 || (ln > 1 && memNE(s, locinput, ln)))
4512 case EXACTFL: { /* /abc/il */
4514 const U8 * fold_array;
4516 U32 fold_utf8_flags;
4518 folder = foldEQ_locale;
4519 fold_array = PL_fold_locale;
4520 fold_utf8_flags = FOLDEQ_LOCALE;
4523 case EXACTFU_SS: /* /\x{df}/iu */
4524 case EXACTFU: /* /abc/iu */
4525 folder = foldEQ_latin1;
4526 fold_array = PL_fold_latin1;
4527 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4530 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
4532 assert(! is_utf8_pat);
4534 case EXACTFA: /* /abc/iaa */
4535 folder = foldEQ_latin1;
4536 fold_array = PL_fold_latin1;
4537 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4540 case EXACTF: /* /abc/i This node only generated for
4541 non-utf8 patterns */
4542 assert(! is_utf8_pat);
4544 fold_array = PL_fold;
4545 fold_utf8_flags = 0;
4553 || state_num == EXACTFU_SS
4554 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
4556 /* Either target or the pattern are utf8, or has the issue where
4557 * the fold lengths may differ. */
4558 const char * const l = locinput;
4559 char *e = reginfo->strend;
4561 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4562 l, &e, 0, utf8_target, fold_utf8_flags))
4570 /* Neither the target nor the pattern are utf8 */
4571 if (UCHARAT(s) != nextchr
4573 && UCHARAT(s) != fold_array[nextchr])
4577 if (reginfo->strend - locinput < ln)
4579 if (ln > 1 && ! folder(s, locinput, ln))
4585 /* XXX Could improve efficiency by separating these all out using a
4586 * macro or in-line function. At that point regcomp.c would no longer
4587 * have to set the FLAGS fields of these */
4588 case BOUNDL: /* /\b/l */
4589 case NBOUNDL: /* /\B/l */
4590 case BOUND: /* /\b/ */
4591 case BOUNDU: /* /\b/u */
4592 case BOUNDA: /* /\b/a */
4593 case NBOUND: /* /\B/ */
4594 case NBOUNDU: /* /\B/u */
4595 case NBOUNDA: /* /\B/a */
4596 /* was last char in word? */
4598 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4599 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4601 if (locinput == reginfo->strbeg)
4604 const U8 * const r =
4605 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4607 ln = utf8n_to_uvchr(r, (U8*) reginfo->strend - r,
4610 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4611 ln = isWORDCHAR_uni(ln);
4615 LOAD_UTF8_CHARCLASS_ALNUM();
4616 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4621 ln = isWORDCHAR_LC_uvchr(ln);
4622 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4627 /* Here the string isn't utf8, or is utf8 and only ascii
4628 * characters are to match \w. In the latter case looking at
4629 * the byte just prior to the current one may be just the final
4630 * byte of a multi-byte character. This is ok. There are two
4632 * 1) it is a single byte character, and then the test is doing
4633 * just what it's supposed to.
4634 * 2) it is a multi-byte character, in which case the final
4635 * byte is never mistakable for ASCII, and so the test
4636 * will say it is not a word character, which is the
4637 * correct answer. */
4638 ln = (locinput != reginfo->strbeg) ?
4639 UCHARAT(locinput - 1) : '\n';
4640 switch (FLAGS(scan)) {
4641 case REGEX_UNICODE_CHARSET:
4642 ln = isWORDCHAR_L1(ln);
4643 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4645 case REGEX_LOCALE_CHARSET:
4646 ln = isWORDCHAR_LC(ln);
4647 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4649 case REGEX_DEPENDS_CHARSET:
4650 ln = isWORDCHAR(ln);
4651 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4653 case REGEX_ASCII_RESTRICTED_CHARSET:
4654 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4655 ln = isWORDCHAR_A(ln);
4656 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4659 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4662 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4664 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4668 case ANYOF: /* /[abc]/ */
4672 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
4675 locinput += UTF8SKIP(locinput);
4678 if (!REGINCLASS(rex, scan, (U8*)locinput))
4684 /* The argument (FLAGS) to all the POSIX node types is the class number
4687 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4691 case POSIXL: /* \w or [:punct:] etc. under /l */
4695 /* Use isFOO_lc() for characters within Latin1. (Note that
4696 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4697 * wouldn't be invariant) */
4698 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4699 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4703 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4704 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4705 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4706 *(locinput + 1))))))
4711 else { /* Here, must be an above Latin-1 code point */
4712 goto utf8_posix_not_eos;
4715 /* Here, must be utf8 */
4716 locinput += UTF8SKIP(locinput);
4719 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4723 case POSIXD: /* \w or [:punct:] etc. under /d */
4729 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4731 if (NEXTCHR_IS_EOS) {
4735 /* All UTF-8 variants match */
4736 if (! UTF8_IS_INVARIANT(nextchr)) {
4737 goto increment_locinput;
4743 case POSIXA: /* \w or [:punct:] etc. under /a */
4746 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4747 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4748 * character is a single byte */
4751 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4757 /* Here we are either not in utf8, or we matched a utf8-invariant,
4758 * so the next char is the next byte */
4762 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4766 case POSIXU: /* \w or [:punct:] etc. under /u */
4768 if (NEXTCHR_IS_EOS) {
4773 /* Use _generic_isCC() for characters within Latin1. (Note that
4774 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4775 * wouldn't be invariant) */
4776 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4777 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4784 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4785 if (! (to_complement
4786 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4794 else { /* Handle above Latin-1 code points */
4795 classnum = (_char_class_number) FLAGS(scan);
4796 if (classnum < _FIRST_NON_SWASH_CC) {
4798 /* Here, uses a swash to find such code points. Load if if
4799 * not done already */
4800 if (! PL_utf8_swash_ptrs[classnum]) {
4801 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4802 PL_utf8_swash_ptrs[classnum]
4803 = _core_swash_init("utf8",
4806 PL_XPosix_ptrs[classnum], &flags);
4808 if (! (to_complement
4809 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4810 (U8 *) locinput, TRUE))))
4815 else { /* Here, uses macros to find above Latin-1 code points */
4817 case _CC_ENUM_SPACE: /* XXX would require separate
4818 code if we revert the change
4819 of \v matching this */
4820 case _CC_ENUM_PSXSPC:
4821 if (! (to_complement
4822 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4827 case _CC_ENUM_BLANK:
4828 if (! (to_complement
4829 ^ cBOOL(is_HORIZWS_high(locinput))))
4834 case _CC_ENUM_XDIGIT:
4835 if (! (to_complement
4836 ^ cBOOL(is_XDIGIT_high(locinput))))
4841 case _CC_ENUM_VERTSPACE:
4842 if (! (to_complement
4843 ^ cBOOL(is_VERTWS_high(locinput))))
4848 default: /* The rest, e.g. [:cntrl:], can't match
4850 if (! to_complement) {
4856 locinput += UTF8SKIP(locinput);
4860 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4861 a Unicode extended Grapheme Cluster */
4862 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4863 extended Grapheme Cluster is:
4866 | Prepend* Begin Extend*
4869 Begin is: ( Special_Begin | ! Control )
4870 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4871 Extend is: ( Grapheme_Extend | Spacing_Mark )
4872 Control is: [ GCB_Control | CR | LF ]
4873 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4875 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4878 Begin is ( Regular_Begin + Special Begin )
4880 It turns out that 98.4% of all Unicode code points match
4881 Regular_Begin. Doing it this way eliminates a table match in
4882 the previous implementation for almost all Unicode code points.
4884 There is a subtlety with Prepend* which showed up in testing.
4885 Note that the Begin, and only the Begin is required in:
4886 | Prepend* Begin Extend*
4887 Also, Begin contains '! Control'. A Prepend must be a
4888 '! Control', which means it must also be a Begin. What it
4889 comes down to is that if we match Prepend* and then find no
4890 suitable Begin afterwards, that if we backtrack the last
4891 Prepend, that one will be a suitable Begin.
4896 if (! utf8_target) {
4898 /* Match either CR LF or '.', as all the other possibilities
4900 locinput++; /* Match the . or CR */
4901 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4903 && locinput < reginfo->strend
4904 && UCHARAT(locinput) == '\n')
4911 /* Utf8: See if is ( CR LF ); already know that locinput <
4912 * reginfo->strend, so locinput+1 is in bounds */
4913 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4914 && UCHARAT(locinput + 1) == '\n')
4921 /* In case have to backtrack to beginning, then match '.' */
4922 char *starting = locinput;
4924 /* In case have to backtrack the last prepend */
4925 char *previous_prepend = NULL;
4927 LOAD_UTF8_CHARCLASS_GCB();
4929 /* Match (prepend)* */
4930 while (locinput < reginfo->strend
4931 && (len = is_GCB_Prepend_utf8(locinput)))
4933 previous_prepend = locinput;
4937 /* As noted above, if we matched a prepend character, but
4938 * the next thing won't match, back off the last prepend we
4939 * matched, as it is guaranteed to match the begin */
4940 if (previous_prepend
4941 && (locinput >= reginfo->strend
4942 || (! swash_fetch(PL_utf8_X_regular_begin,
4943 (U8*)locinput, utf8_target)
4944 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4947 locinput = previous_prepend;
4950 /* Note that here we know reginfo->strend > locinput, as we
4951 * tested that upon input to this switch case, and if we
4952 * moved locinput forward, we tested the result just above
4953 * and it either passed, or we backed off so that it will
4955 if (swash_fetch(PL_utf8_X_regular_begin,
4956 (U8*)locinput, utf8_target)) {
4957 locinput += UTF8SKIP(locinput);
4959 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4961 /* Here did not match the required 'Begin' in the
4962 * second term. So just match the very first
4963 * character, the '.' of the final term of the regex */
4964 locinput = starting + UTF8SKIP(starting);
4968 /* Here is a special begin. It can be composed of
4969 * several individual characters. One possibility is
4971 if ((len = is_GCB_RI_utf8(locinput))) {
4973 while (locinput < reginfo->strend
4974 && (len = is_GCB_RI_utf8(locinput)))
4978 } else if ((len = is_GCB_T_utf8(locinput))) {
4979 /* Another possibility is T+ */
4981 while (locinput < reginfo->strend
4982 && (len = is_GCB_T_utf8(locinput)))
4988 /* Here, neither RI+ nor T+; must be some other
4989 * Hangul. That means it is one of the others: L,
4990 * LV, LVT or V, and matches:
4991 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4994 while (locinput < reginfo->strend
4995 && (len = is_GCB_L_utf8(locinput)))
5000 /* Here, have exhausted L*. If the next character
5001 * is not an LV, LVT nor V, it means we had to have
5002 * at least one L, so matches L+ in the original
5003 * equation, we have a complete hangul syllable.
5006 if (locinput < reginfo->strend
5007 && is_GCB_LV_LVT_V_utf8(locinput))
5009 /* Otherwise keep going. Must be LV, LVT or V.
5010 * See if LVT, by first ruling out V, then LV */
5011 if (! is_GCB_V_utf8(locinput)
5012 /* All but every TCount one is LV */
5013 && (valid_utf8_to_uvchr((U8 *) locinput,
5018 locinput += UTF8SKIP(locinput);
5021 /* Must be V or LV. Take it, then match
5023 locinput += UTF8SKIP(locinput);
5024 while (locinput < reginfo->strend
5025 && (len = is_GCB_V_utf8(locinput)))
5031 /* And any of LV, LVT, or V can be followed
5033 while (locinput < reginfo->strend
5034 && (len = is_GCB_T_utf8(locinput)))
5042 /* Match any extender */
5043 while (locinput < reginfo->strend
5044 && swash_fetch(PL_utf8_X_extend,
5045 (U8*)locinput, utf8_target))
5047 locinput += UTF8SKIP(locinput);
5051 if (locinput > reginfo->strend) sayNO;
5055 case NREFFL: /* /\g{name}/il */
5056 { /* The capture buffer cases. The ones beginning with N for the
5057 named buffers just convert to the equivalent numbered and
5058 pretend they were called as the corresponding numbered buffer
5060 /* don't initialize these in the declaration, it makes C++
5065 const U8 *fold_array;
5068 folder = foldEQ_locale;
5069 fold_array = PL_fold_locale;
5071 utf8_fold_flags = FOLDEQ_LOCALE;
5074 case NREFFA: /* /\g{name}/iaa */
5075 folder = foldEQ_latin1;
5076 fold_array = PL_fold_latin1;
5078 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5081 case NREFFU: /* /\g{name}/iu */
5082 folder = foldEQ_latin1;
5083 fold_array = PL_fold_latin1;
5085 utf8_fold_flags = 0;
5088 case NREFF: /* /\g{name}/i */
5090 fold_array = PL_fold;
5092 utf8_fold_flags = 0;
5095 case NREF: /* /\g{name}/ */
5099 utf8_fold_flags = 0;
5102 /* For the named back references, find the corresponding buffer
5104 n = reg_check_named_buff_matched(rex,scan);
5109 goto do_nref_ref_common;
5111 case REFFL: /* /\1/il */
5112 folder = foldEQ_locale;
5113 fold_array = PL_fold_locale;
5114 utf8_fold_flags = FOLDEQ_LOCALE;
5117 case REFFA: /* /\1/iaa */
5118 folder = foldEQ_latin1;
5119 fold_array = PL_fold_latin1;
5120 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5123 case REFFU: /* /\1/iu */
5124 folder = foldEQ_latin1;
5125 fold_array = PL_fold_latin1;
5126 utf8_fold_flags = 0;
5129 case REFF: /* /\1/i */
5131 fold_array = PL_fold;
5132 utf8_fold_flags = 0;
5135 case REF: /* /\1/ */
5138 utf8_fold_flags = 0;
5142 n = ARG(scan); /* which paren pair */
5145 ln = rex->offs[n].start;
5146 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5147 if (rex->lastparen < n || ln == -1)
5148 sayNO; /* Do not match unless seen CLOSEn. */
5149 if (ln == rex->offs[n].end)
5152 s = reginfo->strbeg + ln;
5153 if (type != REF /* REF can do byte comparison */
5154 && (utf8_target || type == REFFU || type == REFFL))
5156 char * limit = reginfo->strend;
5158 /* This call case insensitively compares the entire buffer
5159 * at s, with the current input starting at locinput, but
5160 * not going off the end given by reginfo->strend, and
5161 * returns in <limit> upon success, how much of the
5162 * current input was matched */
5163 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
5164 locinput, &limit, 0, utf8_target, utf8_fold_flags))
5172 /* Not utf8: Inline the first character, for speed. */
5173 if (!NEXTCHR_IS_EOS &&
5174 UCHARAT(s) != nextchr &&
5176 UCHARAT(s) != fold_array[nextchr]))
5178 ln = rex->offs[n].end - ln;
5179 if (locinput + ln > reginfo->strend)
5181 if (ln > 1 && (type == REF
5182 ? memNE(s, locinput, ln)
5183 : ! folder(s, locinput, ln)))
5189 case NOTHING: /* null op; e.g. the 'nothing' following
5190 * the '*' in m{(a+|b)*}' */
5192 case TAIL: /* placeholder while compiling (A|B|C) */
5195 case BACK: /* ??? doesn't appear to be used ??? */
5199 #define ST st->u.eval
5204 regexp_internal *rei;
5205 regnode *startpoint;
5207 case GOSTART: /* (?R) */
5208 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
5209 if (cur_eval && cur_eval->locinput==locinput) {
5210 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
5211 Perl_croak(aTHX_ "Infinite recursion in regex");
5212 if ( ++nochange_depth > max_nochange_depth )
5214 "Pattern subroutine nesting without pos change"
5215 " exceeded limit in regex");
5222 if (OP(scan)==GOSUB) {
5223 startpoint = scan + ARG2L(scan);
5224 ST.close_paren = ARG(scan);
5226 startpoint = rei->program+1;
5230 /* Save all the positions seen so far. */
5231 ST.cp = regcppush(rex, 0, maxopenparen);
5232 REGCP_SET(ST.lastcp);
5234 /* and then jump to the code we share with EVAL */
5235 goto eval_recurse_doit;
5240 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
5241 if (cur_eval && cur_eval->locinput==locinput) {
5242 if ( ++nochange_depth > max_nochange_depth )
5243 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
5248 /* execute the code in the {...} */
5252 OP * const oop = PL_op;
5253 COP * const ocurcop = PL_curcop;
5257 /* save *all* paren positions */
5258 regcppush(rex, 0, maxopenparen);
5259 REGCP_SET(runops_cp);
5262 caller_cv = find_runcv(NULL);
5266 if (rexi->data->what[n] == 'r') { /* code from an external qr */
5268 (REGEXP*)(rexi->data->data[n])
5271 nop = (OP*)rexi->data->data[n+1];
5273 else if (rexi->data->what[n] == 'l') { /* literal code */
5275 nop = (OP*)rexi->data->data[n];
5276 assert(CvDEPTH(newcv));
5279 /* literal with own CV */
5280 assert(rexi->data->what[n] == 'L');
5281 newcv = rex->qr_anoncv;
5282 nop = (OP*)rexi->data->data[n];
5285 /* normally if we're about to execute code from the same
5286 * CV that we used previously, we just use the existing
5287 * CX stack entry. However, its possible that in the
5288 * meantime we may have backtracked, popped from the save
5289 * stack, and undone the SAVECOMPPAD(s) associated with
5290 * PUSH_MULTICALL; in which case PL_comppad no longer
5291 * points to newcv's pad. */
5292 if (newcv != last_pushed_cv || PL_comppad != last_pad)
5294 U8 flags = (CXp_SUB_RE |
5295 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
5296 if (last_pushed_cv) {
5297 CHANGE_MULTICALL_FLAGS(newcv, flags);
5300 PUSH_MULTICALL_FLAGS(newcv, flags);
5302 last_pushed_cv = newcv;
5305 /* these assignments are just to silence compiler
5307 multicall_cop = NULL;
5310 last_pad = PL_comppad;
5312 /* the initial nextstate you would normally execute
5313 * at the start of an eval (which would cause error
5314 * messages to come from the eval), may be optimised
5315 * away from the execution path in the regex code blocks;
5316 * so manually set PL_curcop to it initially */
5318 OP *o = cUNOPx(nop)->op_first;
5319 assert(o->op_type == OP_NULL);
5320 if (o->op_targ == OP_SCOPE) {
5321 o = cUNOPo->op_first;
5324 assert(o->op_targ == OP_LEAVE);
5325 o = cUNOPo->op_first;
5326 assert(o->op_type == OP_ENTER);
5330 if (o->op_type != OP_STUB) {
5331 assert( o->op_type == OP_NEXTSTATE
5332 || o->op_type == OP_DBSTATE
5333 || (o->op_type == OP_NULL
5334 && ( o->op_targ == OP_NEXTSTATE
5335 || o->op_targ == OP_DBSTATE
5339 PL_curcop = (COP*)o;
5344 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
5345 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
5347 rex->offs[0].end = locinput - reginfo->strbeg;
5348 if (reginfo->info_aux_eval->pos_magic)
5349 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
5350 reginfo->sv, reginfo->strbeg,
5351 locinput - reginfo->strbeg);
5354 SV *sv_mrk = get_sv("REGMARK", 1);
5355 sv_setsv(sv_mrk, sv_yes_mark);
5358 /* we don't use MULTICALL here as we want to call the
5359 * first op of the block of interest, rather than the
5360 * first op of the sub */
5361 before = (IV)(SP-PL_stack_base);
5363 CALLRUNOPS(aTHX); /* Scalar context. */
5365 if ((IV)(SP-PL_stack_base) == before)
5366 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
5372 /* before restoring everything, evaluate the returned
5373 * value, so that 'uninit' warnings don't use the wrong
5374 * PL_op or pad. Also need to process any magic vars
5375 * (e.g. $1) *before* parentheses are restored */
5380 if (logical == 0) /* (?{})/ */
5381 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5382 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5383 sw = cBOOL(SvTRUE(ret));
5386 else { /* /(??{}) */
5387 /* if its overloaded, let the regex compiler handle
5388 * it; otherwise extract regex, or stringify */
5389 if (SvGMAGICAL(ret))
5390 ret = sv_mortalcopy(ret);
5391 if (!SvAMAGIC(ret)) {
5395 if (SvTYPE(sv) == SVt_REGEXP)
5396 re_sv = (REGEXP*) sv;
5397 else if (SvSMAGICAL(ret)) {
5398 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
5400 re_sv = (REGEXP *) mg->mg_obj;
5403 /* force any undef warnings here */
5404 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
5405 ret = sv_mortalcopy(ret);
5406 (void) SvPV_force_nolen(ret);
5412 /* *** Note that at this point we don't restore
5413 * PL_comppad, (or pop the CxSUB) on the assumption it may
5414 * be used again soon. This is safe as long as nothing
5415 * in the regexp code uses the pad ! */
5417 PL_curcop = ocurcop;
5418 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5419 PL_curpm = PL_reg_curpm;
5425 /* only /(??{})/ from now on */
5428 /* extract RE object from returned value; compiling if
5432 re_sv = reg_temp_copy(NULL, re_sv);
5437 if (SvUTF8(ret) && IN_BYTES) {
5438 /* In use 'bytes': make a copy of the octet
5439 * sequence, but without the flag on */
5441 const char *const p = SvPV(ret, len);
5442 ret = newSVpvn_flags(p, len, SVs_TEMP);
5444 if (rex->intflags & PREGf_USE_RE_EVAL)
5445 pm_flags |= PMf_USE_RE_EVAL;
5447 /* if we got here, it should be an engine which
5448 * supports compiling code blocks and stuff */
5449 assert(rex->engine && rex->engine->op_comp);
5450 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5451 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5452 rex->engine, NULL, NULL,
5453 /* copy /msix etc to inner pattern */
5458 & (SVs_TEMP | SVs_GMG | SVf_ROK))
5459 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
5460 /* This isn't a first class regexp. Instead, it's
5461 caching a regexp onto an existing, Perl visible
5463 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5469 RXp_MATCH_COPIED_off(re);
5470 re->subbeg = rex->subbeg;
5471 re->sublen = rex->sublen;
5472 re->suboffset = rex->suboffset;
5473 re->subcoffset = rex->subcoffset;
5475 re->lastcloseparen = 0;
5478 debug_start_match(re_sv, utf8_target, locinput,
5479 reginfo->strend, "Matching embedded");
5481 startpoint = rei->program + 1;
5482 ST.close_paren = 0; /* only used for GOSUB */
5483 /* Save all the seen positions so far. */
5484 ST.cp = regcppush(rex, 0, maxopenparen);
5485 REGCP_SET(ST.lastcp);
5486 /* and set maxopenparen to 0, since we are starting a "fresh" match */
5488 /* run the pattern returned from (??{...}) */
5490 eval_recurse_doit: /* Share code with GOSUB below this line
5491 * At this point we expect the stack context to be
5492 * set up correctly */
5494 /* invalidate the S-L poscache. We're now executing a
5495 * different set of WHILEM ops (and their associated
5496 * indexes) against the same string, so the bits in the
5497 * cache are meaningless. Setting maxiter to zero forces
5498 * the cache to be invalidated and zeroed before reuse.
5499 * XXX This is too dramatic a measure. Ideally we should
5500 * save the old cache and restore when running the outer
5502 reginfo->poscache_maxiter = 0;
5504 /* the new regexp might have a different is_utf8_pat than we do */
5505 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5507 ST.prev_rex = rex_sv;
5508 ST.prev_curlyx = cur_curlyx;
5510 SET_reg_curpm(rex_sv);
5515 ST.prev_eval = cur_eval;
5517 /* now continue from first node in postoned RE */
5518 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5523 case EVAL_AB: /* cleanup after a successful (??{A})B */
5524 /* note: this is called twice; first after popping B, then A */
5525 rex_sv = ST.prev_rex;
5526 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5527 SET_reg_curpm(rex_sv);
5528 rex = ReANY(rex_sv);
5529 rexi = RXi_GET(rex);
5531 /* preserve $^R across LEAVE's. See Bug 121070. */
5532 SV *save_sv= GvSV(PL_replgv);
5533 SvREFCNT_inc(save_sv);
5534 regcpblow(ST.cp); /* LEAVE in disguise */
5535 sv_setsv(GvSV(PL_replgv), save_sv);
5536 SvREFCNT_dec(save_sv);
5538 cur_eval = ST.prev_eval;
5539 cur_curlyx = ST.prev_curlyx;
5541 /* Invalidate cache. See "invalidate" comment above. */
5542 reginfo->poscache_maxiter = 0;
5543 if ( nochange_depth )
5548 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5549 /* note: this is called twice; first after popping B, then A */
5550 rex_sv = ST.prev_rex;
5551 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5552 SET_reg_curpm(rex_sv);
5553 rex = ReANY(rex_sv);
5554 rexi = RXi_GET(rex);
5556 REGCP_UNWIND(ST.lastcp);
5557 regcppop(rex, &maxopenparen);
5558 cur_eval = ST.prev_eval;
5559 cur_curlyx = ST.prev_curlyx;
5560 /* Invalidate cache. See "invalidate" comment above. */
5561 reginfo->poscache_maxiter = 0;
5562 if ( nochange_depth )
5568 n = ARG(scan); /* which paren pair */
5569 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5570 if (n > maxopenparen)
5572 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5573 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5577 (IV)rex->offs[n].start_tmp,
5583 /* XXX really need to log other places start/end are set too */
5584 #define CLOSE_CAPTURE \
5585 rex->offs[n].start = rex->offs[n].start_tmp; \
5586 rex->offs[n].end = locinput - reginfo->strbeg; \
5587 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5588 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5590 PTR2UV(rex->offs), \
5592 (IV)rex->offs[n].start, \
5593 (IV)rex->offs[n].end \
5597 n = ARG(scan); /* which paren pair */
5599 if (n > rex->lastparen)
5601 rex->lastcloseparen = n;
5602 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5607 case ACCEPT: /* (*ACCEPT) */
5611 cursor && OP(cursor)!=END;
5612 cursor=regnext(cursor))
5614 if ( OP(cursor)==CLOSE ){
5616 if ( n <= lastopen ) {
5618 if (n > rex->lastparen)
5620 rex->lastcloseparen = n;
5621 if ( n == ARG(scan) || (cur_eval &&
5622 cur_eval->u.eval.close_paren == n))
5631 case GROUPP: /* (?(1)) */
5632 n = ARG(scan); /* which paren pair */
5633 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5636 case NGROUPP: /* (?(<name>)) */
5637 /* reg_check_named_buff_matched returns 0 for no match */
5638 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5641 case INSUBP: /* (?(R)) */
5643 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5646 case DEFINEP: /* (?(DEFINE)) */
5650 case IFTHEN: /* (?(cond)A|B) */
5651 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5653 next = NEXTOPER(NEXTOPER(scan));
5655 next = scan + ARG(scan);
5656 if (OP(next) == IFTHEN) /* Fake one. */
5657 next = NEXTOPER(NEXTOPER(next));
5661 case LOGICAL: /* modifier for EVAL and IFMATCH */
5662 logical = scan->flags;
5665 /*******************************************************************
5667 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5668 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5669 STAR/PLUS/CURLY/CURLYN are used instead.)
5671 A*B is compiled as <CURLYX><A><WHILEM><B>
5673 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5674 state, which contains the current count, initialised to -1. It also sets
5675 cur_curlyx to point to this state, with any previous value saved in the
5678 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5679 since the pattern may possibly match zero times (i.e. it's a while {} loop
5680 rather than a do {} while loop).
5682 Each entry to WHILEM represents a successful match of A. The count in the
5683 CURLYX block is incremented, another WHILEM state is pushed, and execution
5684 passes to A or B depending on greediness and the current count.
5686 For example, if matching against the string a1a2a3b (where the aN are
5687 substrings that match /A/), then the match progresses as follows: (the
5688 pushed states are interspersed with the bits of strings matched so far):
5691 <CURLYX cnt=0><WHILEM>
5692 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5693 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5694 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5695 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5697 (Contrast this with something like CURLYM, which maintains only a single
5701 a1 <CURLYM cnt=1> a2
5702 a1 a2 <CURLYM cnt=2> a3
5703 a1 a2 a3 <CURLYM cnt=3> b
5706 Each WHILEM state block marks a point to backtrack to upon partial failure
5707 of A or B, and also contains some minor state data related to that
5708 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5709 overall state, such as the count, and pointers to the A and B ops.
5711 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5712 must always point to the *current* CURLYX block, the rules are:
5714 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5715 and set cur_curlyx to point the new block.
5717 When popping the CURLYX block after a successful or unsuccessful match,
5718 restore the previous cur_curlyx.
5720 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5721 to the outer one saved in the CURLYX block.
5723 When popping the WHILEM block after a successful or unsuccessful B match,
5724 restore the previous cur_curlyx.
5726 Here's an example for the pattern (AI* BI)*BO
5727 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5730 curlyx backtrack stack
5731 ------ ---------------
5733 CO <CO prev=NULL> <WO>
5734 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5735 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5736 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5738 At this point the pattern succeeds, and we work back down the stack to
5739 clean up, restoring as we go:
5741 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5742 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5743 CO <CO prev=NULL> <WO>
5746 *******************************************************************/
5748 #define ST st->u.curlyx
5750 case CURLYX: /* start of /A*B/ (for complex A) */
5752 /* No need to save/restore up to this paren */
5753 I32 parenfloor = scan->flags;
5755 assert(next); /* keep Coverity happy */
5756 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5759 /* XXXX Probably it is better to teach regpush to support
5760 parenfloor > maxopenparen ... */
5761 if (parenfloor > (I32)rex->lastparen)
5762 parenfloor = rex->lastparen; /* Pessimization... */
5764 ST.prev_curlyx= cur_curlyx;
5766 ST.cp = PL_savestack_ix;
5768 /* these fields contain the state of the current curly.
5769 * they are accessed by subsequent WHILEMs */
5770 ST.parenfloor = parenfloor;
5775 ST.count = -1; /* this will be updated by WHILEM */
5776 ST.lastloc = NULL; /* this will be updated by WHILEM */
5778 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5783 case CURLYX_end: /* just finished matching all of A*B */
5784 cur_curlyx = ST.prev_curlyx;
5789 case CURLYX_end_fail: /* just failed to match all of A*B */
5791 cur_curlyx = ST.prev_curlyx;
5798 #define ST st->u.whilem
5800 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5802 /* see the discussion above about CURLYX/WHILEM */
5807 assert(cur_curlyx); /* keep Coverity happy */
5809 min = ARG1(cur_curlyx->u.curlyx.me);
5810 max = ARG2(cur_curlyx->u.curlyx.me);
5811 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5812 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5813 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5814 ST.cache_offset = 0;
5818 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5819 "%*s whilem: matched %ld out of %d..%d\n",
5820 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5823 /* First just match a string of min A's. */
5826 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5828 cur_curlyx->u.curlyx.lastloc = locinput;
5829 REGCP_SET(ST.lastcp);
5831 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5836 /* If degenerate A matches "", assume A done. */
5838 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5839 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5840 "%*s whilem: empty match detected, trying continuation...\n",
5841 REPORT_CODE_OFF+depth*2, "")
5843 goto do_whilem_B_max;
5846 /* super-linear cache processing.
5848 * The idea here is that for certain types of CURLYX/WHILEM -
5849 * principally those whose upper bound is infinity (and
5850 * excluding regexes that have things like \1 and other very
5851 * non-regular expresssiony things), then if a pattern like
5852 * /....A*.../ fails and we backtrack to the WHILEM, then we
5853 * make a note that this particular WHILEM op was at string
5854 * position 47 (say) when the rest of pattern failed. Then, if
5855 * we ever find ourselves back at that WHILEM, and at string
5856 * position 47 again, we can just fail immediately rather than
5857 * running the rest of the pattern again.
5859 * This is very handy when patterns start to go
5860 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5861 * with a combinatorial explosion of backtracking.
5863 * The cache is implemented as a bit array, with one bit per
5864 * string byte position per WHILEM op (up to 16) - so its
5865 * between 0.25 and 2x the string size.
5867 * To avoid allocating a poscache buffer every time, we do an
5868 * initially countdown; only after we have executed a WHILEM
5869 * op (string-length x #WHILEMs) times do we allocate the
5872 * The top 4 bits of scan->flags byte say how many different
5873 * relevant CURLLYX/WHILEM op pairs there are, while the
5874 * bottom 4-bits is the identifying index number of this
5880 if (!reginfo->poscache_maxiter) {
5881 /* start the countdown: Postpone detection until we
5882 * know the match is not *that* much linear. */
5883 reginfo->poscache_maxiter
5884 = (reginfo->strend - reginfo->strbeg + 1)
5886 /* possible overflow for long strings and many CURLYX's */
5887 if (reginfo->poscache_maxiter < 0)
5888 reginfo->poscache_maxiter = I32_MAX;
5889 reginfo->poscache_iter = reginfo->poscache_maxiter;
5892 if (reginfo->poscache_iter-- == 0) {
5893 /* initialise cache */
5894 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
5895 regmatch_info_aux *const aux = reginfo->info_aux;
5896 if (aux->poscache) {
5897 if ((SSize_t)reginfo->poscache_size < size) {
5898 Renew(aux->poscache, size, char);
5899 reginfo->poscache_size = size;
5901 Zero(aux->poscache, size, char);
5904 reginfo->poscache_size = size;
5905 Newxz(aux->poscache, size, char);
5907 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5908 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5909 PL_colors[4], PL_colors[5])
5913 if (reginfo->poscache_iter < 0) {
5914 /* have we already failed at this position? */
5915 SSize_t offset, mask;
5917 reginfo->poscache_iter = -1; /* stop eventual underflow */
5918 offset = (scan->flags & 0xf) - 1
5919 + (locinput - reginfo->strbeg)
5921 mask = 1 << (offset % 8);
5923 if (reginfo->info_aux->poscache[offset] & mask) {
5924 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5925 "%*s whilem: (cache) already tried at this position...\n",
5926 REPORT_CODE_OFF+depth*2, "")
5928 sayNO; /* cache records failure */
5930 ST.cache_offset = offset;
5931 ST.cache_mask = mask;
5935 /* Prefer B over A for minimal matching. */
5937 if (cur_curlyx->u.curlyx.minmod) {
5938 ST.save_curlyx = cur_curlyx;
5939 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5940 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5942 REGCP_SET(ST.lastcp);
5943 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5949 /* Prefer A over B for maximal matching. */
5951 if (n < max) { /* More greed allowed? */
5952 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5954 cur_curlyx->u.curlyx.lastloc = locinput;
5955 REGCP_SET(ST.lastcp);
5956 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5960 goto do_whilem_B_max;
5965 case WHILEM_B_min: /* just matched B in a minimal match */
5966 case WHILEM_B_max: /* just matched B in a maximal match */
5967 cur_curlyx = ST.save_curlyx;
5972 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5973 cur_curlyx = ST.save_curlyx;
5974 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5975 cur_curlyx->u.curlyx.count--;
5980 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5982 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5983 REGCP_UNWIND(ST.lastcp);
5984 regcppop(rex, &maxopenparen);
5985 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5986 cur_curlyx->u.curlyx.count--;
5991 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5992 REGCP_UNWIND(ST.lastcp);
5993 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5994 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5995 "%*s whilem: failed, trying continuation...\n",
5996 REPORT_CODE_OFF+depth*2, "")
5999 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6000 && ckWARN(WARN_REGEXP)
6001 && !reginfo->warned)
6003 reginfo->warned = TRUE;
6004 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6005 "Complex regular subexpression recursion limit (%d) "
6011 ST.save_curlyx = cur_curlyx;
6012 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
6013 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
6018 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
6019 cur_curlyx = ST.save_curlyx;
6020 REGCP_UNWIND(ST.lastcp);
6021 regcppop(rex, &maxopenparen);
6023 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
6024 /* Maximum greed exceeded */
6025 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6026 && ckWARN(WARN_REGEXP)
6027 && !reginfo->warned)
6029 reginfo->warned = TRUE;
6030 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6031 "Complex regular subexpression recursion "
6032 "limit (%d) exceeded",
6035 cur_curlyx->u.curlyx.count--;
6039 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6040 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
6042 /* Try grabbing another A and see if it helps. */
6043 cur_curlyx->u.curlyx.lastloc = locinput;
6044 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6046 REGCP_SET(ST.lastcp);
6047 PUSH_STATE_GOTO(WHILEM_A_min,
6048 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
6054 #define ST st->u.branch
6056 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
6057 next = scan + ARG(scan);
6060 scan = NEXTOPER(scan);
6063 case BRANCH: /* /(...|A|...)/ */
6064 scan = NEXTOPER(scan); /* scan now points to inner node */
6065 ST.lastparen = rex->lastparen;
6066 ST.lastcloseparen = rex->lastcloseparen;
6067 ST.next_branch = next;
6070 /* Now go into the branch */
6072 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
6074 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
6079 case CUTGROUP: /* /(*THEN)/ */
6080 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
6081 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6082 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
6086 case CUTGROUP_next_fail:
6089 if (st->u.mark.mark_name)
6090 sv_commit = st->u.mark.mark_name;
6100 case BRANCH_next_fail: /* that branch failed; try the next, if any */
6105 REGCP_UNWIND(ST.cp);
6106 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6107 scan = ST.next_branch;
6108 /* no more branches? */
6109 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
6111 PerlIO_printf( Perl_debug_log,
6112 "%*s %sBRANCH failed...%s\n",
6113 REPORT_CODE_OFF+depth*2, "",
6119 continue; /* execute next BRANCH[J] op */
6123 case MINMOD: /* next op will be non-greedy, e.g. A*? */
6128 #define ST st->u.curlym
6130 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
6132 /* This is an optimisation of CURLYX that enables us to push
6133 * only a single backtracking state, no matter how many matches
6134 * there are in {m,n}. It relies on the pattern being constant
6135 * length, with no parens to influence future backrefs
6139 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6141 ST.lastparen = rex->lastparen;
6142 ST.lastcloseparen = rex->lastcloseparen;
6144 /* if paren positive, emulate an OPEN/CLOSE around A */
6146 U32 paren = ST.me->flags;
6147 if (paren > maxopenparen)
6148 maxopenparen = paren;
6149 scan += NEXT_OFF(scan); /* Skip former OPEN. */
6157 ST.c1 = CHRTEST_UNINIT;
6160 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
6163 curlym_do_A: /* execute the A in /A{m,n}B/ */
6164 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
6168 case CURLYM_A: /* we've just matched an A */
6170 /* after first match, determine A's length: u.curlym.alen */
6171 if (ST.count == 1) {
6172 if (reginfo->is_utf8_target) {
6173 char *s = st->locinput;
6174 while (s < locinput) {
6180 ST.alen = locinput - st->locinput;
6183 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
6186 PerlIO_printf(Perl_debug_log,
6187 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
6188 (int)(REPORT_CODE_OFF+(depth*2)), "",
6189 (IV) ST.count, (IV)ST.alen)
6192 if (cur_eval && cur_eval->u.eval.close_paren &&
6193 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6197 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
6198 if ( max == REG_INFTY || ST.count < max )
6199 goto curlym_do_A; /* try to match another A */
6201 goto curlym_do_B; /* try to match B */
6203 case CURLYM_A_fail: /* just failed to match an A */
6204 REGCP_UNWIND(ST.cp);
6206 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
6207 || (cur_eval && cur_eval->u.eval.close_paren &&
6208 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
6211 curlym_do_B: /* execute the B in /A{m,n}B/ */
6212 if (ST.c1 == CHRTEST_UNINIT) {
6213 /* calculate c1 and c2 for possible match of 1st char
6214 * following curly */
6215 ST.c1 = ST.c2 = CHRTEST_VOID;
6217 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
6218 regnode *text_node = ST.B;
6219 if (! HAS_TEXT(text_node))
6220 FIND_NEXT_IMPT(text_node);
6223 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
6225 But the former is redundant in light of the latter.
6227 if this changes back then the macro for
6228 IS_TEXT and friends need to change.
6230 if (PL_regkind[OP(text_node)] == EXACT) {
6231 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6232 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6242 PerlIO_printf(Perl_debug_log,
6243 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
6244 (int)(REPORT_CODE_OFF+(depth*2)),
6247 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
6248 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
6249 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6250 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6252 /* simulate B failing */
6254 PerlIO_printf(Perl_debug_log,
6255 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
6256 (int)(REPORT_CODE_OFF+(depth*2)),"",
6257 valid_utf8_to_uvchr((U8 *) locinput, NULL),
6258 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
6259 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
6261 state_num = CURLYM_B_fail;
6262 goto reenter_switch;
6265 else if (nextchr != ST.c1 && nextchr != ST.c2) {
6266 /* simulate B failing */
6268 PerlIO_printf(Perl_debug_log,
6269 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
6270 (int)(REPORT_CODE_OFF+(depth*2)),"",
6271 (int) nextchr, ST.c1, ST.c2)
6273 state_num = CURLYM_B_fail;
6274 goto reenter_switch;
6279 /* emulate CLOSE: mark current A as captured */
6280 I32 paren = ST.me->flags;
6282 rex->offs[paren].start
6283 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
6284 rex->offs[paren].end = locinput - reginfo->strbeg;
6285 if ((U32)paren > rex->lastparen)
6286 rex->lastparen = paren;
6287 rex->lastcloseparen = paren;
6290 rex->offs[paren].end = -1;
6291 if (cur_eval && cur_eval->u.eval.close_paren &&
6292 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6301 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
6305 case CURLYM_B_fail: /* just failed to match a B */
6306 REGCP_UNWIND(ST.cp);
6307 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6309 I32 max = ARG2(ST.me);
6310 if (max != REG_INFTY && ST.count == max)
6312 goto curlym_do_A; /* try to match a further A */
6314 /* backtrack one A */
6315 if (ST.count == ARG1(ST.me) /* min */)
6318 SET_locinput(HOPc(locinput, -ST.alen));
6319 goto curlym_do_B; /* try to match B */
6322 #define ST st->u.curly
6324 #define CURLY_SETPAREN(paren, success) \
6327 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
6328 rex->offs[paren].end = locinput - reginfo->strbeg; \
6329 if (paren > rex->lastparen) \
6330 rex->lastparen = paren; \
6331 rex->lastcloseparen = paren; \
6334 rex->offs[paren].end = -1; \
6335 rex->lastparen = ST.lastparen; \
6336 rex->lastcloseparen = ST.lastcloseparen; \
6340 case STAR: /* /A*B/ where A is width 1 char */
6344 scan = NEXTOPER(scan);
6347 case PLUS: /* /A+B/ where A is width 1 char */
6351 scan = NEXTOPER(scan);
6354 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
6355 ST.paren = scan->flags; /* Which paren to set */
6356 ST.lastparen = rex->lastparen;
6357 ST.lastcloseparen = rex->lastcloseparen;
6358 if (ST.paren > maxopenparen)
6359 maxopenparen = ST.paren;
6360 ST.min = ARG1(scan); /* min to match */
6361 ST.max = ARG2(scan); /* max to match */
6362 if (cur_eval && cur_eval->u.eval.close_paren &&
6363 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6367 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
6370 case CURLY: /* /A{m,n}B/ where A is width 1 char */
6372 ST.min = ARG1(scan); /* min to match */
6373 ST.max = ARG2(scan); /* max to match */
6374 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6377 * Lookahead to avoid useless match attempts
6378 * when we know what character comes next.
6380 * Used to only do .*x and .*?x, but now it allows
6381 * for )'s, ('s and (?{ ... })'s to be in the way
6382 * of the quantifier and the EXACT-like node. -- japhy
6385 assert(ST.min <= ST.max);
6386 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
6387 ST.c1 = ST.c2 = CHRTEST_VOID;
6390 regnode *text_node = next;
6392 if (! HAS_TEXT(text_node))
6393 FIND_NEXT_IMPT(text_node);
6395 if (! HAS_TEXT(text_node))
6396 ST.c1 = ST.c2 = CHRTEST_VOID;
6398 if ( PL_regkind[OP(text_node)] != EXACT ) {
6399 ST.c1 = ST.c2 = CHRTEST_VOID;
6403 /* Currently we only get here when
6405 PL_rekind[OP(text_node)] == EXACT
6407 if this changes back then the macro for IS_TEXT and
6408 friends need to change. */
6409 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6410 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6422 char *li = locinput;
6425 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
6431 if (ST.c1 == CHRTEST_VOID)
6432 goto curly_try_B_min;
6434 ST.oldloc = locinput;
6436 /* set ST.maxpos to the furthest point along the
6437 * string that could possibly match */
6438 if (ST.max == REG_INFTY) {
6439 ST.maxpos = reginfo->strend - 1;
6441 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
6444 else if (utf8_target) {
6445 int m = ST.max - ST.min;
6446 for (ST.maxpos = locinput;
6447 m >0 && ST.maxpos < reginfo->strend; m--)
6448 ST.maxpos += UTF8SKIP(ST.maxpos);
6451 ST.maxpos = locinput + ST.max - ST.min;
6452 if (ST.maxpos >= reginfo->strend)
6453 ST.maxpos = reginfo->strend - 1;
6455 goto curly_try_B_min_known;
6459 /* avoid taking address of locinput, so it can remain
6461 char *li = locinput;
6462 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6463 if (ST.count < ST.min)
6466 if ((ST.count > ST.min)
6467 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6469 /* A{m,n} must come at the end of the string, there's
6470 * no point in backing off ... */
6472 /* ...except that $ and \Z can match before *and* after
6473 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6474 We may back off by one in this case. */
6475 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6479 goto curly_try_B_max;
6484 case CURLY_B_min_known_fail:
6485 /* failed to find B in a non-greedy match where c1,c2 valid */
6487 REGCP_UNWIND(ST.cp);
6489 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6491 /* Couldn't or didn't -- move forward. */
6492 ST.oldloc = locinput;
6494 locinput += UTF8SKIP(locinput);
6498 curly_try_B_min_known:
6499 /* find the next place where 'B' could work, then call B */
6503 n = (ST.oldloc == locinput) ? 0 : 1;
6504 if (ST.c1 == ST.c2) {
6505 /* set n to utf8_distance(oldloc, locinput) */
6506 while (locinput <= ST.maxpos
6507 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6509 locinput += UTF8SKIP(locinput);
6514 /* set n to utf8_distance(oldloc, locinput) */
6515 while (locinput <= ST.maxpos
6516 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6517 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6519 locinput += UTF8SKIP(locinput);
6524 else { /* Not utf8_target */
6525 if (ST.c1 == ST.c2) {
6526 while (locinput <= ST.maxpos &&
6527 UCHARAT(locinput) != ST.c1)
6531 while (locinput <= ST.maxpos
6532 && UCHARAT(locinput) != ST.c1
6533 && UCHARAT(locinput) != ST.c2)
6536 n = locinput - ST.oldloc;
6538 if (locinput > ST.maxpos)
6541 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6542 * at b; check that everything between oldloc and
6543 * locinput matches */
6544 char *li = ST.oldloc;
6546 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6548 assert(n == REG_INFTY || locinput == li);
6550 CURLY_SETPAREN(ST.paren, ST.count);
6551 if (cur_eval && cur_eval->u.eval.close_paren &&
6552 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6555 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6560 case CURLY_B_min_fail:
6561 /* failed to find B in a non-greedy match where c1,c2 invalid */
6563 REGCP_UNWIND(ST.cp);
6565 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6567 /* failed -- move forward one */
6569 char *li = locinput;
6570 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6577 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6578 ST.count > 0)) /* count overflow ? */
6581 CURLY_SETPAREN(ST.paren, ST.count);
6582 if (cur_eval && cur_eval->u.eval.close_paren &&
6583 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6586 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6594 /* a successful greedy match: now try to match B */
6595 if (cur_eval && cur_eval->u.eval.close_paren &&
6596 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6600 bool could_match = locinput < reginfo->strend;
6602 /* If it could work, try it. */
6603 if (ST.c1 != CHRTEST_VOID && could_match) {
6604 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6606 could_match = memEQ(locinput,
6611 UTF8SKIP(locinput));
6614 could_match = UCHARAT(locinput) == ST.c1
6615 || UCHARAT(locinput) == ST.c2;
6618 if (ST.c1 == CHRTEST_VOID || could_match) {
6619 CURLY_SETPAREN(ST.paren, ST.count);
6620 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6627 case CURLY_B_max_fail:
6628 /* failed to find B in a greedy match */
6630 REGCP_UNWIND(ST.cp);
6632 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6635 if (--ST.count < ST.min)
6637 locinput = HOPc(locinput, -1);
6638 goto curly_try_B_max;
6642 case END: /* last op of main pattern */
6645 /* we've just finished A in /(??{A})B/; now continue with B */
6647 st->u.eval.prev_rex = rex_sv; /* inner */
6649 /* Save *all* the positions. */
6650 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6651 rex_sv = cur_eval->u.eval.prev_rex;
6652 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6653 SET_reg_curpm(rex_sv);
6654 rex = ReANY(rex_sv);
6655 rexi = RXi_GET(rex);
6656 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6658 REGCP_SET(st->u.eval.lastcp);
6660 /* Restore parens of the outer rex without popping the
6662 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6665 st->u.eval.prev_eval = cur_eval;
6666 cur_eval = cur_eval->u.eval.prev_eval;
6668 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6669 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6670 if ( nochange_depth )
6673 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6674 locinput); /* match B */
6677 if (locinput < reginfo->till) {
6678 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6679 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6681 (long)(locinput - startpos),
6682 (long)(reginfo->till - startpos),
6685 sayNO_SILENT; /* Cannot match: too short. */
6687 sayYES; /* Success! */
6689 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6691 PerlIO_printf(Perl_debug_log,
6692 "%*s %ssubpattern success...%s\n",
6693 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6694 sayYES; /* Success! */
6697 #define ST st->u.ifmatch
6702 case SUSPEND: /* (?>A) */
6704 newstart = locinput;
6707 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6709 goto ifmatch_trivial_fail_test;
6711 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6713 ifmatch_trivial_fail_test:
6715 char * const s = HOPBACKc(locinput, scan->flags);
6720 sw = 1 - cBOOL(ST.wanted);
6724 next = scan + ARG(scan);
6732 newstart = locinput;
6736 ST.logical = logical;
6737 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6739 /* execute body of (?...A) */
6740 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6745 case IFMATCH_A_fail: /* body of (?...A) failed */
6746 ST.wanted = !ST.wanted;
6749 case IFMATCH_A: /* body of (?...A) succeeded */
6751 sw = cBOOL(ST.wanted);
6753 else if (!ST.wanted)
6756 if (OP(ST.me) != SUSPEND) {
6757 /* restore old position except for (?>...) */
6758 locinput = st->locinput;
6760 scan = ST.me + ARG(ST.me);
6763 continue; /* execute B */
6767 case LONGJMP: /* alternative with many branches compiles to
6768 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6769 next = scan + ARG(scan);
6774 case COMMIT: /* (*COMMIT) */
6775 reginfo->cutpoint = reginfo->strend;
6778 case PRUNE: /* (*PRUNE) */
6780 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6781 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6785 case COMMIT_next_fail:
6789 case OPFAIL: /* (*FAIL) */
6794 #define ST st->u.mark
6795 case MARKPOINT: /* (*MARK:foo) */
6796 ST.prev_mark = mark_state;
6797 ST.mark_name = sv_commit = sv_yes_mark
6798 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6800 ST.mark_loc = locinput;
6801 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6805 case MARKPOINT_next:
6806 mark_state = ST.prev_mark;
6811 case MARKPOINT_next_fail:
6812 if (popmark && sv_eq(ST.mark_name,popmark))
6814 if (ST.mark_loc > startpoint)
6815 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6816 popmark = NULL; /* we found our mark */
6817 sv_commit = ST.mark_name;
6820 PerlIO_printf(Perl_debug_log,
6821 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6822 REPORT_CODE_OFF+depth*2, "",
6823 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6826 mark_state = ST.prev_mark;
6827 sv_yes_mark = mark_state ?
6828 mark_state->u.mark.mark_name : NULL;
6833 case SKIP: /* (*SKIP) */
6835 /* (*SKIP) : if we fail we cut here*/
6836 ST.mark_name = NULL;
6837 ST.mark_loc = locinput;
6838 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6840 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6841 otherwise do nothing. Meaning we need to scan
6843 regmatch_state *cur = mark_state;
6844 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6847 if ( sv_eq( cur->u.mark.mark_name,
6850 ST.mark_name = find;
6851 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6853 cur = cur->u.mark.prev_mark;
6856 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6859 case SKIP_next_fail:
6861 /* (*CUT:NAME) - Set up to search for the name as we
6862 collapse the stack*/
6863 popmark = ST.mark_name;
6865 /* (*CUT) - No name, we cut here.*/
6866 if (ST.mark_loc > startpoint)
6867 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6868 /* but we set sv_commit to latest mark_name if there
6869 is one so they can test to see how things lead to this
6872 sv_commit=mark_state->u.mark.mark_name;
6880 case LNBREAK: /* \R */
6881 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6888 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6889 PTR2UV(scan), OP(scan));
6890 Perl_croak(aTHX_ "regexp memory corruption");
6892 /* this is a point to jump to in order to increment
6893 * locinput by one character */
6895 assert(!NEXTCHR_IS_EOS);
6897 locinput += PL_utf8skip[nextchr];
6898 /* locinput is allowed to go 1 char off the end, but not 2+ */
6899 if (locinput > reginfo->strend)
6908 /* switch break jumps here */
6909 scan = next; /* prepare to execute the next op and ... */
6910 continue; /* ... jump back to the top, reusing st */
6915 /* push a state that backtracks on success */
6916 st->u.yes.prev_yes_state = yes_state;
6920 /* push a new regex state, then continue at scan */
6922 regmatch_state *newst;
6925 regmatch_state *cur = st;
6926 regmatch_state *curyes = yes_state;
6928 regmatch_slab *slab = PL_regmatch_slab;
6929 for (;curd > -1;cur--,curd--) {
6930 if (cur < SLAB_FIRST(slab)) {
6932 cur = SLAB_LAST(slab);
6934 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6935 REPORT_CODE_OFF + 2 + depth * 2,"",
6936 curd, PL_reg_name[cur->resume_state],
6937 (curyes == cur) ? "yes" : ""
6940 curyes = cur->u.yes.prev_yes_state;
6943 DEBUG_STATE_pp("push")
6946 st->locinput = locinput;
6948 if (newst > SLAB_LAST(PL_regmatch_slab))
6949 newst = S_push_slab(aTHX);
6950 PL_regmatch_state = newst;
6952 locinput = pushinput;
6961 * We get here only if there's trouble -- normally "case END" is
6962 * the terminating point.
6964 Perl_croak(aTHX_ "corrupted regexp pointers");
6970 /* we have successfully completed a subexpression, but we must now
6971 * pop to the state marked by yes_state and continue from there */
6972 assert(st != yes_state);
6974 while (st != yes_state) {
6976 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6977 PL_regmatch_slab = PL_regmatch_slab->prev;
6978 st = SLAB_LAST(PL_regmatch_slab);
6982 DEBUG_STATE_pp("pop (no final)");
6984 DEBUG_STATE_pp("pop (yes)");
6990 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6991 || yes_state > SLAB_LAST(PL_regmatch_slab))
6993 /* not in this slab, pop slab */
6994 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6995 PL_regmatch_slab = PL_regmatch_slab->prev;
6996 st = SLAB_LAST(PL_regmatch_slab);
6998 depth -= (st - yes_state);
7001 yes_state = st->u.yes.prev_yes_state;
7002 PL_regmatch_state = st;
7005 locinput= st->locinput;
7006 state_num = st->resume_state + no_final;
7007 goto reenter_switch;
7010 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
7011 PL_colors[4], PL_colors[5]));
7013 if (reginfo->info_aux_eval) {
7014 /* each successfully executed (?{...}) block does the equivalent of
7015 * local $^R = do {...}
7016 * When popping the save stack, all these locals would be undone;
7017 * bypass this by setting the outermost saved $^R to the latest
7019 /* I dont know if this is needed or works properly now.
7020 * see code related to PL_replgv elsewhere in this file.
7023 if (oreplsv != GvSV(PL_replgv))
7024 sv_setsv(oreplsv, GvSV(PL_replgv));
7031 PerlIO_printf(Perl_debug_log,
7032 "%*s %sfailed...%s\n",
7033 REPORT_CODE_OFF+depth*2, "",
7034 PL_colors[4], PL_colors[5])
7046 /* there's a previous state to backtrack to */
7048 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7049 PL_regmatch_slab = PL_regmatch_slab->prev;
7050 st = SLAB_LAST(PL_regmatch_slab);
7052 PL_regmatch_state = st;
7053 locinput= st->locinput;
7055 DEBUG_STATE_pp("pop");
7057 if (yes_state == st)
7058 yes_state = st->u.yes.prev_yes_state;
7060 state_num = st->resume_state + 1; /* failure = success + 1 */
7061 goto reenter_switch;
7066 if (rex->intflags & PREGf_VERBARG_SEEN) {
7067 SV *sv_err = get_sv("REGERROR", 1);
7068 SV *sv_mrk = get_sv("REGMARK", 1);
7070 sv_commit = &PL_sv_no;
7072 sv_yes_mark = &PL_sv_yes;
7075 sv_commit = &PL_sv_yes;
7076 sv_yes_mark = &PL_sv_no;
7080 sv_setsv(sv_err, sv_commit);
7081 sv_setsv(sv_mrk, sv_yes_mark);
7085 if (last_pushed_cv) {
7088 PERL_UNUSED_VAR(SP);
7091 assert(!result || locinput - reginfo->strbeg >= 0);
7092 return result ? locinput - reginfo->strbeg : -1;
7096 - regrepeat - repeatedly match something simple, report how many
7098 * What 'simple' means is a node which can be the operand of a quantifier like
7101 * startposp - pointer a pointer to the start position. This is updated
7102 * to point to the byte following the highest successful
7104 * p - the regnode to be repeatedly matched against.
7105 * reginfo - struct holding match state, such as strend
7106 * max - maximum number of things to match.
7107 * depth - (for debugging) backtracking depth.
7110 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
7111 regmatch_info *const reginfo, I32 max, int depth)
7113 char *scan; /* Pointer to current position in target string */
7115 char *loceol = reginfo->strend; /* local version */
7116 I32 hardcount = 0; /* How many matches so far */
7117 bool utf8_target = reginfo->is_utf8_target;
7118 int to_complement = 0; /* Invert the result? */
7120 _char_class_number classnum;
7122 PERL_UNUSED_ARG(depth);
7125 PERL_ARGS_ASSERT_REGREPEAT;
7128 if (max == REG_INFTY)
7130 else if (! utf8_target && loceol - scan > max)
7131 loceol = scan + max;
7133 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
7134 * to the maximum of how far we should go in it (leaving it set to the real
7135 * end, if the maximum permissible would take us beyond that). This allows
7136 * us to make the loop exit condition that we haven't gone past <loceol> to
7137 * also mean that we haven't exceeded the max permissible count, saving a
7138 * test each time through the loop. But it assumes that the OP matches a
7139 * single byte, which is true for most of the OPs below when applied to a
7140 * non-UTF-8 target. Those relatively few OPs that don't have this
7141 * characteristic will have to compensate.
7143 * There is no adjustment for UTF-8 targets, as the number of bytes per
7144 * character varies. OPs will have to test both that the count is less
7145 * than the max permissible (using <hardcount> to keep track), and that we
7146 * are still within the bounds of the string (using <loceol>. A few OPs
7147 * match a single byte no matter what the encoding. They can omit the max
7148 * test if, for the UTF-8 case, they do the adjustment that was skipped
7151 * Thus, the code above sets things up for the common case; and exceptional
7152 * cases need extra work; the common case is to make sure <scan> doesn't
7153 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
7154 * count doesn't exceed the maximum permissible */
7159 while (scan < loceol && hardcount < max && *scan != '\n') {
7160 scan += UTF8SKIP(scan);
7164 while (scan < loceol && *scan != '\n')
7170 while (scan < loceol && hardcount < max) {
7171 scan += UTF8SKIP(scan);
7178 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
7179 if (utf8_target && loceol - scan > max) {
7181 /* <loceol> hadn't been adjusted in the UTF-8 case */
7189 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7193 /* Can use a simple loop if the pattern char to match on is invariant
7194 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
7195 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
7196 * true iff it doesn't matter if the argument is in UTF-8 or not */
7197 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
7198 if (utf8_target && loceol - scan > max) {
7199 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
7200 * since here, to match at all, 1 char == 1 byte */
7201 loceol = scan + max;
7203 while (scan < loceol && UCHARAT(scan) == c) {
7207 else if (reginfo->is_utf8_pat) {
7209 STRLEN scan_char_len;
7211 /* When both target and pattern are UTF-8, we have to do
7213 while (hardcount < max
7215 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
7216 && memEQ(scan, STRING(p), scan_char_len))
7218 scan += scan_char_len;
7222 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
7224 /* Target isn't utf8; convert the character in the UTF-8
7225 * pattern to non-UTF8, and do a simple loop */
7226 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
7227 while (scan < loceol && UCHARAT(scan) == c) {
7230 } /* else pattern char is above Latin1, can't possibly match the
7235 /* Here, the string must be utf8; pattern isn't, and <c> is
7236 * different in utf8 than not, so can't compare them directly.
7237 * Outside the loop, find the two utf8 bytes that represent c, and
7238 * then look for those in sequence in the utf8 string */
7239 U8 high = UTF8_TWO_BYTE_HI(c);
7240 U8 low = UTF8_TWO_BYTE_LO(c);
7242 while (hardcount < max
7243 && scan + 1 < loceol
7244 && UCHARAT(scan) == high
7245 && UCHARAT(scan + 1) == low)
7253 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
7254 assert(! reginfo->is_utf8_pat);
7257 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7261 utf8_flags = FOLDEQ_LOCALE;
7264 case EXACTF: /* This node only generated for non-utf8 patterns */
7265 assert(! reginfo->is_utf8_pat);
7271 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
7275 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
7277 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7279 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
7282 if (c1 == CHRTEST_VOID) {
7283 /* Use full Unicode fold matching */
7284 char *tmpeol = reginfo->strend;
7285 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
7286 while (hardcount < max
7287 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
7288 STRING(p), NULL, pat_len,
7289 reginfo->is_utf8_pat, utf8_flags))
7292 tmpeol = reginfo->strend;
7296 else if (utf8_target) {
7298 while (scan < loceol
7300 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
7302 scan += UTF8SKIP(scan);
7307 while (scan < loceol
7309 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
7310 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
7312 scan += UTF8SKIP(scan);
7317 else if (c1 == c2) {
7318 while (scan < loceol && UCHARAT(scan) == c1) {
7323 while (scan < loceol &&
7324 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
7334 while (hardcount < max
7336 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
7338 scan += UTF8SKIP(scan);
7342 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
7347 /* The argument (FLAGS) to all the POSIX node types is the class number */
7354 if (! utf8_target) {
7355 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
7361 while (hardcount < max && scan < loceol
7362 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
7365 scan += UTF8SKIP(scan);
7378 if (utf8_target && loceol - scan > max) {
7380 /* We didn't adjust <loceol> at the beginning of this routine
7381 * because is UTF-8, but it is actually ok to do so, since here, to
7382 * match, 1 char == 1 byte. */
7383 loceol = scan + max;
7385 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
7398 if (! utf8_target) {
7399 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
7405 /* The complement of something that matches only ASCII matches all
7406 * non-ASCII, plus everything in ASCII that isn't in the class. */
7407 while (hardcount < max && scan < loceol
7408 && (! isASCII_utf8(scan)
7409 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
7411 scan += UTF8SKIP(scan);
7422 if (! utf8_target) {
7423 while (scan < loceol && to_complement
7424 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
7431 classnum = (_char_class_number) FLAGS(p);
7432 if (classnum < _FIRST_NON_SWASH_CC) {
7434 /* Here, a swash is needed for above-Latin1 code points.
7435 * Process as many Latin1 code points using the built-in rules.
7436 * Go to another loop to finish processing upon encountering
7437 * the first Latin1 code point. We could do that in this loop
7438 * as well, but the other way saves having to test if the swash
7439 * has been loaded every time through the loop: extra space to
7441 while (hardcount < max && scan < loceol) {
7442 if (UTF8_IS_INVARIANT(*scan)) {
7443 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
7450 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
7451 if (! (to_complement
7452 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
7461 goto found_above_latin1;
7468 /* For these character classes, the knowledge of how to handle
7469 * every code point is compiled in to Perl via a macro. This
7470 * code is written for making the loops as tight as possible.
7471 * It could be refactored to save space instead */
7473 case _CC_ENUM_SPACE: /* XXX would require separate code
7474 if we revert the change of \v
7477 case _CC_ENUM_PSXSPC:
7478 while (hardcount < max
7480 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7482 scan += UTF8SKIP(scan);
7486 case _CC_ENUM_BLANK:
7487 while (hardcount < max
7489 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7491 scan += UTF8SKIP(scan);
7495 case _CC_ENUM_XDIGIT:
7496 while (hardcount < max
7498 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7500 scan += UTF8SKIP(scan);
7504 case _CC_ENUM_VERTSPACE:
7505 while (hardcount < max
7507 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7509 scan += UTF8SKIP(scan);
7513 case _CC_ENUM_CNTRL:
7514 while (hardcount < max
7516 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7518 scan += UTF8SKIP(scan);
7523 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7529 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7531 /* Load the swash if not already present */
7532 if (! PL_utf8_swash_ptrs[classnum]) {
7533 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7534 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7538 PL_XPosix_ptrs[classnum], &flags);
7541 while (hardcount < max && scan < loceol
7542 && to_complement ^ cBOOL(_generic_utf8(
7545 swash_fetch(PL_utf8_swash_ptrs[classnum],
7549 scan += UTF8SKIP(scan);
7556 while (hardcount < max && scan < loceol &&
7557 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7562 /* LNBREAK can match one or two latin chars, which is ok, but we
7563 * have to use hardcount in this situation, and throw away the
7564 * adjustment to <loceol> done before the switch statement */
7565 loceol = reginfo->strend;
7566 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7587 /* These are all 0 width, so match right here or not at all. */
7591 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7600 c = scan - *startposp;
7604 GET_RE_DEBUG_FLAGS_DECL;
7606 SV * const prop = sv_newmortal();
7607 regprop(prog, prop, p, reginfo);
7608 PerlIO_printf(Perl_debug_log,
7609 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7610 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7618 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7620 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7621 create a copy so that changes the caller makes won't change the shared one.
7622 If <altsvp> is non-null, will return NULL in it, for back-compat.
7625 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7627 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7633 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL));
7637 Perl__get_regclass_nonbitmap_data(pTHX_ const regexp *prog,
7638 const regnode* node,
7641 SV** only_utf8_locale_ptr)
7643 /* For internal core use only.
7644 * Returns the swash for the input 'node' in the regex 'prog'.
7645 * If <doinit> is 'true', will attempt to create the swash if not already
7647 * If <listsvp> is non-null, will return the printable contents of the
7648 * swash. This can be used to get debugging information even before the
7649 * swash exists, by calling this function with 'doinit' set to false, in
7650 * which case the components that will be used to eventually create the
7651 * swash are returned (in a printable form).
7652 * Tied intimately to how regcomp.c sets up the data structure */
7655 SV *si = NULL; /* Input swash initialization string */
7658 RXi_GET_DECL(prog,progi);
7659 const struct reg_data * const data = prog ? progi->data : NULL;
7661 PERL_ARGS_ASSERT__GET_REGCLASS_NONBITMAP_DATA;
7663 assert(ANYOF_FLAGS(node)
7664 & (ANYOF_UTF8|ANYOF_NONBITMAP_NON_UTF8|ANYOF_LOC_FOLD));
7666 if (data && data->count) {
7667 const U32 n = ARG(node);
7669 if (data->what[n] == 's') {
7670 SV * const rv = MUTABLE_SV(data->data[n]);
7671 AV * const av = MUTABLE_AV(SvRV(rv));
7672 SV **const ary = AvARRAY(av);
7673 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7675 si = *ary; /* ary[0] = the string to initialize the swash with */
7677 /* Elements 3 and 4 are either both present or both absent. [3] is
7678 * any inversion list generated at compile time; [4] indicates if
7679 * that inversion list has any user-defined properties in it. */
7680 if (av_tindex(av) >= 2) {
7681 if (only_utf8_locale_ptr
7683 && ary[2] != &PL_sv_undef)
7685 *only_utf8_locale_ptr = ary[2];
7688 assert(only_utf8_locale_ptr);
7689 *only_utf8_locale_ptr = NULL;
7692 if (av_tindex(av) >= 3) {
7695 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7703 /* Element [1] is reserved for the set-up swash. If already there,
7704 * return it; if not, create it and store it there */
7705 if (ary[1] && SvROK(ary[1])) {
7708 else if (doinit && ((si && si != &PL_sv_undef)
7709 || (invlist && invlist != &PL_sv_undef))) {
7711 sw = _core_swash_init("utf8", /* the utf8 package */
7715 0, /* not from tr/// */
7718 (void)av_store(av, 1, sw);
7723 /* If requested, return a printable version of what this swash matches */
7725 SV* matches_string = newSVpvs("");
7727 /* The swash should be used, if possible, to get the data, as it
7728 * contains the resolved data. But this function can be called at
7729 * compile-time, before everything gets resolved, in which case we
7730 * return the currently best available information, which is the string
7731 * that will eventually be used to do that resolving, 'si' */
7732 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7733 && (si && si != &PL_sv_undef))
7735 sv_catsv(matches_string, si);
7738 /* Add the inversion list to whatever we have. This may have come from
7739 * the swash, or from an input parameter */
7741 sv_catsv(matches_string, _invlist_contents(invlist));
7743 *listsvp = matches_string;
7748 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
7751 - reginclass - determine if a character falls into a character class
7753 n is the ANYOF regnode
7754 p is the target string
7755 p_end points to one byte beyond the end of the target string
7756 utf8_target tells whether p is in UTF-8.
7758 Returns true if matched; false otherwise.
7760 Note that this can be a synthetic start class, a combination of various
7761 nodes, so things you think might be mutually exclusive, such as locale,
7762 aren't. It can match both locale and non-locale
7767 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
7770 const char flags = ANYOF_FLAGS(n);
7774 PERL_ARGS_ASSERT_REGINCLASS;
7776 /* If c is not already the code point, get it. Note that
7777 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7778 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7780 c = utf8n_to_uvchr(p, p_end - p, &c_len,
7781 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7782 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7783 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7784 * UTF8_ALLOW_FFFF */
7785 if (c_len == (STRLEN)-1)
7786 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7789 /* If this character is potentially in the bitmap, check it */
7791 if (ANYOF_BITMAP_TEST(n, c))
7793 else if (flags & ANYOF_NON_UTF8_NON_ASCII_ALL
7799 else if (flags & ANYOF_LOCALE_FLAGS) {
7800 if (flags & ANYOF_LOC_FOLD) {
7801 if (ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) {
7805 if (! match && ANYOF_POSIXL_TEST_ANY_SET(n)) {
7807 /* The data structure is arranged so bits 0, 2, 4, ... are set
7808 * if the class includes the Posix character class given by
7809 * bit/2; and 1, 3, 5, ... are set if the class includes the
7810 * complemented Posix class given by int(bit/2). So we loop
7811 * through the bits, each time changing whether we complement
7812 * the result or not. Suppose for the sake of illustration
7813 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7814 * is set, it means there is a match for this ANYOF node if the
7815 * character is in the class given by the expression (0 / 2 = 0
7816 * = \w). If it is in that class, isFOO_lc() will return 1,
7817 * and since 'to_complement' is 0, the result will stay TRUE,
7818 * and we exit the loop. Suppose instead that bit 0 is 0, but
7819 * bit 1 is 1. That means there is a match if the character
7820 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7821 * but will on bit 1. On the second iteration 'to_complement'
7822 * will be 1, so the exclusive or will reverse things, so we
7823 * are testing for \W. On the third iteration, 'to_complement'
7824 * will be 0, and we would be testing for \s; the fourth
7825 * iteration would test for \S, etc.
7827 * Note that this code assumes that all the classes are closed
7828 * under folding. For example, if a character matches \w, then
7829 * its fold does too; and vice versa. This should be true for
7830 * any well-behaved locale for all the currently defined Posix
7831 * classes, except for :lower: and :upper:, which are handled
7832 * by the pseudo-class :cased: which matches if either of the
7833 * other two does. To get rid of this assumption, an outer
7834 * loop could be used below to iterate over both the source
7835 * character, and its fold (if different) */
7838 int to_complement = 0;
7840 while (count < ANYOF_MAX) {
7841 if (ANYOF_POSIXL_TEST(n, count)
7842 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7855 /* If the bitmap didn't (or couldn't) match, and something outside the
7856 * bitmap could match, try that. */
7858 if (c >= 256 && (flags & ANYOF_ABOVE_LATIN1_ALL)) {
7859 match = TRUE; /* Everything above 255 matches */
7861 else if ((flags & ANYOF_NONBITMAP_NON_UTF8)
7862 || (utf8_target && (flags & ANYOF_UTF8))
7863 || ((flags & ANYOF_LOC_FOLD)
7864 && IN_UTF8_CTYPE_LOCALE
7865 && ARG(n) != ANYOF_NONBITMAP_EMPTY))
7867 SV* only_utf8_locale = NULL;
7868 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
7875 } else { /* Convert to utf8 */
7876 utf8_p = utf8_buffer;
7877 append_utf8_from_native_byte(*p, &utf8_p);
7878 utf8_p = utf8_buffer;
7881 if (swash_fetch(sw, utf8_p, TRUE)) {
7885 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
7886 match = _invlist_contains_cp(only_utf8_locale, c);
7890 if (UNICODE_IS_SUPER(c)
7891 && (flags & ANYOF_WARN_SUPER)
7892 && ckWARN_d(WARN_NON_UNICODE))
7894 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7895 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c);
7899 #if ANYOF_INVERT != 1
7900 /* Depending on compiler optimization cBOOL takes time, so if don't have to
7902 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
7905 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7906 return (flags & ANYOF_INVERT) ^ match;
7910 S_reghop3(U8 *s, SSize_t off, const U8* lim)
7912 /* return the position 'off' UTF-8 characters away from 's', forward if
7913 * 'off' >= 0, backwards if negative. But don't go outside of position
7914 * 'lim', which better be < s if off < 0 */
7916 PERL_ARGS_ASSERT_REGHOP3;
7919 while (off-- && s < lim) {
7920 /* XXX could check well-formedness here */
7925 while (off++ && s > lim) {
7927 if (UTF8_IS_CONTINUED(*s)) {
7928 while (s > lim && UTF8_IS_CONTINUATION(*s))
7931 /* XXX could check well-formedness here */
7938 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
7940 PERL_ARGS_ASSERT_REGHOP4;
7943 while (off-- && s < rlim) {
7944 /* XXX could check well-formedness here */
7949 while (off++ && s > llim) {
7951 if (UTF8_IS_CONTINUED(*s)) {
7952 while (s > llim && UTF8_IS_CONTINUATION(*s))
7955 /* XXX could check well-formedness here */
7961 /* like reghop3, but returns NULL on overrun, rather than returning last
7965 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
7967 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7970 while (off-- && s < lim) {
7971 /* XXX could check well-formedness here */
7978 while (off++ && s > lim) {
7980 if (UTF8_IS_CONTINUED(*s)) {
7981 while (s > lim && UTF8_IS_CONTINUATION(*s))
7984 /* XXX could check well-formedness here */
7993 /* when executing a regex that may have (?{}), extra stuff needs setting
7994 up that will be visible to the called code, even before the current
7995 match has finished. In particular:
7997 * $_ is localised to the SV currently being matched;
7998 * pos($_) is created if necessary, ready to be updated on each call-out
8000 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
8001 isn't set until the current pattern is successfully finished), so that
8002 $1 etc of the match-so-far can be seen;
8003 * save the old values of subbeg etc of the current regex, and set then
8004 to the current string (again, this is normally only done at the end
8009 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
8012 regexp *const rex = ReANY(reginfo->prog);
8013 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
8015 eval_state->rex = rex;
8018 /* Make $_ available to executed code. */
8019 if (reginfo->sv != DEFSV) {
8021 DEFSV_set(reginfo->sv);
8024 if (!(mg = mg_find_mglob(reginfo->sv))) {
8025 /* prepare for quick setting of pos */
8026 mg = sv_magicext_mglob(reginfo->sv);
8029 eval_state->pos_magic = mg;
8030 eval_state->pos = mg->mg_len;
8031 eval_state->pos_flags = mg->mg_flags;
8034 eval_state->pos_magic = NULL;
8036 if (!PL_reg_curpm) {
8037 /* PL_reg_curpm is a fake PMOP that we can attach the current
8038 * regex to and point PL_curpm at, so that $1 et al are visible
8039 * within a /(?{})/. It's just allocated once per interpreter the
8040 * first time its needed */
8041 Newxz(PL_reg_curpm, 1, PMOP);
8044 SV* const repointer = &PL_sv_undef;
8045 /* this regexp is also owned by the new PL_reg_curpm, which
8046 will try to free it. */
8047 av_push(PL_regex_padav, repointer);
8048 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
8049 PL_regex_pad = AvARRAY(PL_regex_padav);
8053 SET_reg_curpm(reginfo->prog);
8054 eval_state->curpm = PL_curpm;
8055 PL_curpm = PL_reg_curpm;
8056 if (RXp_MATCH_COPIED(rex)) {
8057 /* Here is a serious problem: we cannot rewrite subbeg,
8058 since it may be needed if this match fails. Thus
8059 $` inside (?{}) could fail... */
8060 eval_state->subbeg = rex->subbeg;
8061 eval_state->sublen = rex->sublen;
8062 eval_state->suboffset = rex->suboffset;
8063 eval_state->subcoffset = rex->subcoffset;
8065 eval_state->saved_copy = rex->saved_copy;
8067 RXp_MATCH_COPIED_off(rex);
8070 eval_state->subbeg = NULL;
8071 rex->subbeg = (char *)reginfo->strbeg;
8073 rex->subcoffset = 0;
8074 rex->sublen = reginfo->strend - reginfo->strbeg;
8078 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
8081 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
8083 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
8084 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
8087 Safefree(aux->poscache);
8091 /* undo the effects of S_setup_eval_state() */
8093 if (eval_state->subbeg) {
8094 regexp * const rex = eval_state->rex;
8095 rex->subbeg = eval_state->subbeg;
8096 rex->sublen = eval_state->sublen;
8097 rex->suboffset = eval_state->suboffset;
8098 rex->subcoffset = eval_state->subcoffset;
8100 rex->saved_copy = eval_state->saved_copy;
8102 RXp_MATCH_COPIED_on(rex);
8104 if (eval_state->pos_magic)
8106 eval_state->pos_magic->mg_len = eval_state->pos;
8107 eval_state->pos_magic->mg_flags =
8108 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
8109 | (eval_state->pos_flags & MGf_BYTES);
8112 PL_curpm = eval_state->curpm;
8115 PL_regmatch_state = aux->old_regmatch_state;
8116 PL_regmatch_slab = aux->old_regmatch_slab;
8118 /* free all slabs above current one - this must be the last action
8119 * of this function, as aux and eval_state are allocated within
8120 * slabs and may be freed here */
8122 s = PL_regmatch_slab->next;
8124 PL_regmatch_slab->next = NULL;
8126 regmatch_slab * const osl = s;
8135 S_to_utf8_substr(pTHX_ regexp *prog)
8137 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
8138 * on the converted value */
8142 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
8145 if (prog->substrs->data[i].substr
8146 && !prog->substrs->data[i].utf8_substr) {
8147 SV* const sv = newSVsv(prog->substrs->data[i].substr);
8148 prog->substrs->data[i].utf8_substr = sv;
8149 sv_utf8_upgrade(sv);
8150 if (SvVALID(prog->substrs->data[i].substr)) {
8151 if (SvTAIL(prog->substrs->data[i].substr)) {
8152 /* Trim the trailing \n that fbm_compile added last
8154 SvCUR_set(sv, SvCUR(sv) - 1);
8155 /* Whilst this makes the SV technically "invalid" (as its
8156 buffer is no longer followed by "\0") when fbm_compile()
8157 adds the "\n" back, a "\0" is restored. */
8158 fbm_compile(sv, FBMcf_TAIL);
8162 if (prog->substrs->data[i].substr == prog->check_substr)
8163 prog->check_utf8 = sv;
8169 S_to_byte_substr(pTHX_ regexp *prog)
8171 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
8172 * on the converted value; returns FALSE if can't be converted. */
8176 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
8179 if (prog->substrs->data[i].utf8_substr
8180 && !prog->substrs->data[i].substr) {
8181 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
8182 if (! sv_utf8_downgrade(sv, TRUE)) {
8185 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
8186 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
8187 /* Trim the trailing \n that fbm_compile added last
8189 SvCUR_set(sv, SvCUR(sv) - 1);
8190 fbm_compile(sv, FBMcf_TAIL);
8194 prog->substrs->data[i].substr = sv;
8195 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
8196 prog->check_substr = sv;
8205 * c-indentation-style: bsd
8207 * indent-tabs-mode: nil
8210 * ex: set ts=8 sts=4 sw=4 et: