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)
292 const int retval = PL_savestack_ix;
293 const int paren_elems_to_push =
294 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
295 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
296 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
298 GET_RE_DEBUG_FLAGS_DECL;
300 PERL_ARGS_ASSERT_REGCPPUSH;
302 if (paren_elems_to_push < 0)
303 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %i",
304 paren_elems_to_push, maxopenparen, parenfloor, REGCP_PAREN_ELEMS);
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)
374 GET_RE_DEBUG_FLAGS_DECL;
376 PERL_ARGS_ASSERT_REGCPPOP;
378 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
380 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
381 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
382 rex->lastcloseparen = SSPOPINT;
383 rex->lastparen = SSPOPINT;
384 *maxopenparen_p = SSPOPINT;
386 i -= REGCP_OTHER_ELEMS;
387 /* Now restore the parentheses context. */
389 if (i || rex->lastparen + 1 <= rex->nparens)
390 PerlIO_printf(Perl_debug_log,
391 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
396 paren = *maxopenparen_p;
397 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
399 rex->offs[paren].start_tmp = SSPOPINT;
400 rex->offs[paren].start = SSPOPIV;
402 if (paren <= rex->lastparen)
403 rex->offs[paren].end = tmps;
404 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
405 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
407 (IV)rex->offs[paren].start,
408 (IV)rex->offs[paren].start_tmp,
409 (IV)rex->offs[paren].end,
410 (paren > rex->lastparen ? "(skipped)" : ""));
415 /* It would seem that the similar code in regtry()
416 * already takes care of this, and in fact it is in
417 * a better location to since this code can #if 0-ed out
418 * but the code in regtry() is needed or otherwise tests
419 * requiring null fields (pat.t#187 and split.t#{13,14}
420 * (as of patchlevel 7877) will fail. Then again,
421 * this code seems to be necessary or otherwise
422 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
423 * --jhi updated by dapm */
424 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
425 if (i > *maxopenparen_p)
426 rex->offs[i].start = -1;
427 rex->offs[i].end = -1;
428 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
429 " \\%"UVuf": %s ..-1 undeffing\n",
431 (i > *maxopenparen_p) ? "-1" : " "
437 /* restore the parens and associated vars at savestack position ix,
438 * but without popping the stack */
441 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
443 I32 tmpix = PL_savestack_ix;
444 PL_savestack_ix = ix;
445 regcppop(rex, maxopenparen_p);
446 PL_savestack_ix = tmpix;
449 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
452 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
454 /* Returns a boolean as to whether or not 'character' is a member of the
455 * Posix character class given by 'classnum' that should be equivalent to a
456 * value in the typedef '_char_class_number'.
458 * Ideally this could be replaced by a just an array of function pointers
459 * to the C library functions that implement the macros this calls.
460 * However, to compile, the precise function signatures are required, and
461 * these may vary from platform to to platform. To avoid having to figure
462 * out what those all are on each platform, I (khw) am using this method,
463 * which adds an extra layer of function call overhead (unless the C
464 * optimizer strips it away). But we don't particularly care about
465 * performance with locales anyway. */
467 switch ((_char_class_number) classnum) {
468 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
469 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
470 case _CC_ENUM_ASCII: return isASCII_LC(character);
471 case _CC_ENUM_BLANK: return isBLANK_LC(character);
472 case _CC_ENUM_CASED: return isLOWER_LC(character)
473 || isUPPER_LC(character);
474 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
475 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
476 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
477 case _CC_ENUM_LOWER: return isLOWER_LC(character);
478 case _CC_ENUM_PRINT: return isPRINT_LC(character);
479 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
480 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
481 case _CC_ENUM_SPACE: return isSPACE_LC(character);
482 case _CC_ENUM_UPPER: return isUPPER_LC(character);
483 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
484 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
485 default: /* VERTSPACE should never occur in locales */
486 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
489 assert(0); /* NOTREACHED */
494 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
496 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
497 * 'character' is a member of the Posix character class given by 'classnum'
498 * that should be equivalent to a value in the typedef
499 * '_char_class_number'.
501 * This just calls isFOO_lc on the code point for the character if it is in
502 * the range 0-255. Outside that range, all characters avoid Unicode
503 * rules, ignoring any locale. So use the Unicode function if this class
504 * requires a swash, and use the Unicode macro otherwise. */
506 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
508 if (UTF8_IS_INVARIANT(*character)) {
509 return isFOO_lc(classnum, *character);
511 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
512 return isFOO_lc(classnum,
513 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
516 if (classnum < _FIRST_NON_SWASH_CC) {
518 /* Initialize the swash unless done already */
519 if (! PL_utf8_swash_ptrs[classnum]) {
520 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
521 PL_utf8_swash_ptrs[classnum] =
522 _core_swash_init("utf8",
525 PL_XPosix_ptrs[classnum], &flags);
528 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
530 TRUE /* is UTF */ ));
533 switch ((_char_class_number) classnum) {
535 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
537 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
538 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
539 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
540 default: return 0; /* Things like CNTRL are always
544 assert(0); /* NOTREACHED */
549 * pregexec and friends
552 #ifndef PERL_IN_XSUB_RE
554 - pregexec - match a regexp against a string
557 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
558 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
559 /* stringarg: the point in the string at which to begin matching */
560 /* strend: pointer to null at end of string */
561 /* strbeg: real beginning of string */
562 /* minend: end of match must be >= minend bytes after stringarg. */
563 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
564 * itself is accessed via the pointers above */
565 /* nosave: For optimizations. */
567 PERL_ARGS_ASSERT_PREGEXEC;
570 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
571 nosave ? 0 : REXEC_COPY_STR);
577 /* re_intuit_start():
579 * Based on some optimiser hints, try to find the earliest position in the
580 * string where the regex could match.
582 * rx: the regex to match against
583 * sv: the SV being matched: only used for utf8 flag; the string
584 * itself is accessed via the pointers below. Note that on
585 * something like an overloaded SV, SvPOK(sv) may be false
586 * and the string pointers may point to something unrelated to
588 * strbeg: real beginning of string
589 * strpos: the point in the string at which to begin matching
590 * strend: pointer to the byte following the last char of the string
591 * flags currently unused; set to 0
592 * data: currently unused; set to NULL
594 * The basic idea of re_intuit_start() is to use some known information
595 * about the pattern, namely:
597 * a) the longest known anchored substring (i.e. one that's at a
598 * constant offset from the beginning of the pattern; but not
599 * necessarily at a fixed offset from the beginning of the
601 * b) the longest floating substring (i.e. one that's not at a constant
602 * offset from the beginning of the pattern);
603 * c) Whether the pattern is anchored to the string; either
604 * an absolute anchor: /^../, or anchored to \n: /^.../m,
605 * or anchored to pos(): /\G/;
606 * d) A start class: a real or synthetic character class which
607 * represents which characters are legal at the start of the pattern;
609 * to either quickly reject the match, or to find the earliest position
610 * within the string at which the pattern might match, thus avoiding
611 * running the full NFA engine at those earlier locations, only to
612 * eventually fail and retry further along.
614 * Returns NULL if the pattern can't match, or returns the address within
615 * the string which is the earliest place the match could occur.
617 * The longest of the anchored and floating substrings is called 'check'
618 * and is checked first. The other is called 'other' and is checked
619 * second. The 'other' substring may not be present. For example,
621 * /(abc|xyz)ABC\d{0,3}DEFG/
625 * check substr (float) = "DEFG", offset 6..9 chars
626 * other substr (anchored) = "ABC", offset 3..3 chars
629 * Be aware that during the course of this function, sometimes 'anchored'
630 * refers to a substring being anchored relative to the start of the
631 * pattern, and sometimes to the pattern itself being anchored relative to
632 * the string. For example:
634 * /\dabc/: "abc" is anchored to the pattern;
635 * /^\dabc/: "abc" is anchored to the pattern and the string;
636 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
637 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
638 * but the pattern is anchored to the string.
642 Perl_re_intuit_start(pTHX_
645 const char * const strbeg,
649 re_scream_pos_data *data)
652 struct regexp *const prog = ReANY(rx);
653 SSize_t start_shift = prog->check_offset_min;
654 /* Should be nonnegative! */
655 SSize_t end_shift = 0;
656 /* current lowest pos in string where the regex can start matching */
657 char *rx_origin = strpos;
659 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
660 U8 other_ix = 1 - prog->substrs->check_ix;
662 char *other_last = strpos;/* latest pos 'other' substr already checked to */
663 char *check_at = NULL; /* check substr found at this pos */
664 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
665 RXi_GET_DECL(prog,progi);
666 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
667 regmatch_info *const reginfo = ®info_buf;
668 GET_RE_DEBUG_FLAGS_DECL;
670 PERL_ARGS_ASSERT_RE_INTUIT_START;
671 PERL_UNUSED_ARG(flags);
672 PERL_UNUSED_ARG(data);
674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
675 "Intuit: trying to determine minimum start position...\n"));
677 /* for now, assume that all substr offsets are positive. If at some point
678 * in the future someone wants to do clever things with look-behind and
679 * -ve offsets, they'll need to fix up any code in this function
680 * which uses these offsets. See the thread beginning
681 * <20140113145929.GF27210@iabyn.com>
683 assert(prog->substrs->data[0].min_offset >= 0);
684 assert(prog->substrs->data[0].max_offset >= 0);
685 assert(prog->substrs->data[1].min_offset >= 0);
686 assert(prog->substrs->data[1].max_offset >= 0);
687 assert(prog->substrs->data[2].min_offset >= 0);
688 assert(prog->substrs->data[2].max_offset >= 0);
690 /* for now, assume that if both present, that the floating substring
691 * doesn't start before the anchored substring.
692 * If you break this assumption (e.g. doing better optimisations
693 * with lookahead/behind), then you'll need to audit the code in this
694 * function carefully first
697 ! ( (prog->anchored_utf8 || prog->anchored_substr)
698 && (prog->float_utf8 || prog->float_substr))
699 || (prog->float_min_offset >= prog->anchored_offset));
701 /* byte rather than char calculation for efficiency. It fails
702 * to quickly reject some cases that can't match, but will reject
703 * them later after doing full char arithmetic */
704 if (prog->minlen > strend - strpos) {
705 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
706 " String too short...\n"));
710 reginfo->is_utf8_target = cBOOL(utf8_target);
711 reginfo->info_aux = NULL;
712 reginfo->strbeg = strbeg;
713 reginfo->strend = strend;
714 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
716 /* not actually used within intuit, but zero for safety anyway */
717 reginfo->poscache_maxiter = 0;
720 if (!prog->check_utf8 && prog->check_substr)
721 to_utf8_substr(prog);
722 check = prog->check_utf8;
724 if (!prog->check_substr && prog->check_utf8) {
725 if (! to_byte_substr(prog)) {
726 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
729 check = prog->check_substr;
732 /* dump the various substring data */
733 DEBUG_OPTIMISE_MORE_r({
735 for (i=0; i<=2; i++) {
736 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
737 : prog->substrs->data[i].substr);
741 PerlIO_printf(Perl_debug_log,
742 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
743 " useful=%"IVdf" utf8=%d [%s]\n",
745 (IV)prog->substrs->data[i].min_offset,
746 (IV)prog->substrs->data[i].max_offset,
747 (IV)prog->substrs->data[i].end_shift,
754 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
756 /* ml_anch: check after \n?
758 * A note about IMPLICIT: on an un-anchored pattern beginning
759 * with /.*.../, these flags will have been added by the
761 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
762 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
764 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
765 && !(prog->intflags & PREGf_IMPLICIT);
767 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
768 /* we are only allowed to match at BOS or \G */
770 /* trivially reject if there's a BOS anchor and we're not at BOS.
772 * Note that we don't try to do a similar quick reject for
773 * \G, since generally the caller will have calculated strpos
774 * based on pos() and gofs, so the string is already correctly
775 * anchored by definition; and handling the exceptions would
776 * be too fiddly (e.g. REXEC_IGNOREPOS).
778 if ( strpos != strbeg
779 && (prog->intflags & (PREGf_ANCH_BOL|PREGf_ANCH_SBOL)))
781 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
782 " Not at start...\n"));
786 /* in the presence of an anchor, the anchored (relative to the
787 * start of the regex) substr must also be anchored relative
788 * to strpos. So quickly reject if substr isn't found there.
789 * This works for \G too, because the caller will already have
790 * subtracted gofs from pos, and gofs is the offset from the
791 * \G to the start of the regex. For example, in /.abc\Gdef/,
792 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
793 * caller will have set strpos=pos()-4; we look for the substr
794 * at position pos()-4+1, which lines up with the "a" */
796 if (prog->check_offset_min == prog->check_offset_max
797 && !(prog->intflags & PREGf_CANY_SEEN))
799 /* Substring at constant offset from beg-of-str... */
800 SSize_t slen = SvCUR(check);
801 char *s = HOP3c(strpos, prog->check_offset_min, strend);
803 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
804 " Looking for check substr at fixed offset %"IVdf"...\n",
805 (IV)prog->check_offset_min));
808 /* In this case, the regex is anchored at the end too.
809 * Unless it's a multiline match, the lengths must match
810 * exactly, give or take a \n. NB: slen >= 1 since
811 * the last char of check is \n */
813 && ( strend - s > slen
814 || strend - s < slen - 1
815 || (strend - s == slen && strend[-1] != '\n')))
817 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
818 " String too long...\n"));
821 /* Now should match s[0..slen-2] */
824 if (slen && (*SvPVX_const(check) != *s
825 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
827 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
828 " String not equal...\n"));
833 goto success_at_start;
838 end_shift = prog->check_end_shift;
840 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
842 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
843 (IV)end_shift, RX_PRECOMP(prog));
848 /* This is the (re)entry point of the main loop in this function.
849 * The goal of this loop is to:
850 * 1) find the "check" substring in the region rx_origin..strend
851 * (adjusted by start_shift / end_shift). If not found, reject
853 * 2) If it exists, look for the "other" substr too if defined; for
854 * example, if the check substr maps to the anchored substr, then
855 * check the floating substr, and vice-versa. If not found, go
856 * back to (1) with rx_origin suitably incremented.
857 * 3) If we find an rx_origin position that doesn't contradict
858 * either of the substrings, then check the possible additional
859 * constraints on rx_origin of /^.../m or a known start class.
860 * If these fail, then depending on which constraints fail, jump
861 * back to here, or to various other re-entry points further along
862 * that skip some of the first steps.
863 * 4) If we pass all those tests, update the BmUSEFUL() count on the
864 * substring. If the start position was determined to be at the
865 * beginning of the string - so, not rejected, but not optimised,
866 * since we have to run regmatch from position 0 - decrement the
867 * BmUSEFUL() count. Otherwise increment it.
871 /* first, look for the 'check' substring */
877 DEBUG_OPTIMISE_MORE_r({
878 PerlIO_printf(Perl_debug_log,
879 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
880 " Start shift: %"IVdf" End shift %"IVdf
881 " Real end Shift: %"IVdf"\n",
882 (IV)(rx_origin - strpos),
883 (IV)prog->check_offset_min,
886 (IV)prog->check_end_shift);
889 if (prog->intflags & PREGf_CANY_SEEN) {
890 start_point= (U8*)(rx_origin + start_shift);
891 end_point= (U8*)(strend - end_shift);
892 if (start_point > end_point)
895 end_point = HOP3(strend, -end_shift, strbeg);
896 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
902 /* If the regex is absolutely anchored to either the start of the
903 * string (BOL,SBOL) or to pos() (ANCH_GPOS), then
904 * check_offset_max represents an upper bound on the string where
905 * the substr could start. For the ANCH_GPOS case, we assume that
906 * the caller of intuit will have already set strpos to
907 * pos()-gofs, so in this case strpos + offset_max will still be
908 * an upper bound on the substr.
911 && prog->intflags & PREGf_ANCH
912 && prog->check_offset_max != SSize_t_MAX)
914 SSize_t len = SvCUR(check) - !!SvTAIL(check);
915 const char * const anchor =
916 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
918 /* do a bytes rather than chars comparison. It's conservative;
919 * so it skips doing the HOP if the result can't possibly end
920 * up earlier than the old value of end_point.
922 if ((char*)end_point - anchor > prog->check_offset_max) {
923 end_point = HOP3lim((U8*)anchor,
924 prog->check_offset_max,
930 DEBUG_OPTIMISE_MORE_r({
931 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
932 (int)(end_point - start_point),
933 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
937 check_at = fbm_instr( start_point, end_point,
938 check, multiline ? FBMrf_MULTILINE : 0);
940 /* Update the count-of-usability, remove useless subpatterns,
944 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
945 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
946 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
947 (check_at ? "Found" : "Did not find"),
948 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
949 ? "anchored" : "floating"),
952 (check_at ? " at offset " : "...\n") );
957 /* Finish the diagnostic message */
958 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
960 /* set rx_origin to the minimum position where the regex could start
961 * matching, given the constraint of the just-matched check substring.
962 * But don't set it lower than previously.
965 if (check_at - rx_origin > prog->check_offset_max)
966 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
970 /* now look for the 'other' substring if defined */
972 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
973 : prog->substrs->data[other_ix].substr)
975 /* Take into account the "other" substring. */
979 struct reg_substr_datum *other;
982 other = &prog->substrs->data[other_ix];
984 /* if "other" is anchored:
985 * we've previously found a floating substr starting at check_at.
986 * This means that the regex origin must lie somewhere
987 * between min (rx_origin): HOP3(check_at, -check_offset_max)
988 * and max: HOP3(check_at, -check_offset_min)
989 * (except that min will be >= strpos)
990 * So the fixed substr must lie somewhere between
991 * HOP3(min, anchored_offset)
992 * HOP3(max, anchored_offset) + SvCUR(substr)
995 /* if "other" is floating
996 * Calculate last1, the absolute latest point where the
997 * floating substr could start in the string, ignoring any
998 * constraints from the earlier fixed match. It is calculated
1001 * strend - prog->minlen (in chars) is the absolute latest
1002 * position within the string where the origin of the regex
1003 * could appear. The latest start point for the floating
1004 * substr is float_min_offset(*) on from the start of the
1005 * regex. last1 simply combines thee two offsets.
1007 * (*) You might think the latest start point should be
1008 * float_max_offset from the regex origin, and technically
1009 * you'd be correct. However, consider
1011 * Here, float min, max are 3,5 and minlen is 7.
1012 * This can match either
1016 * In the first case, the regex matches minlen chars; in the
1017 * second, minlen+1, in the third, minlen+2.
1018 * In the first case, the floating offset is 3 (which equals
1019 * float_min), in the second, 4, and in the third, 5 (which
1020 * equals float_max). In all cases, the floating string bcd
1021 * can never start more than 4 chars from the end of the
1022 * string, which equals minlen - float_min. As the substring
1023 * starts to match more than float_min from the start of the
1024 * regex, it makes the regex match more than minlen chars,
1025 * and the two cancel each other out. So we can always use
1026 * float_min - minlen, rather than float_max - minlen for the
1027 * latest position in the string.
1029 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1030 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1033 assert(prog->minlen >= other->min_offset);
1034 last1 = HOP3c(strend,
1035 other->min_offset - prog->minlen, strbeg);
1037 if (other_ix) {/* i.e. if (other-is-float) */
1038 /* last is the latest point where the floating substr could
1039 * start, *given* any constraints from the earlier fixed
1040 * match. This constraint is that the floating string starts
1041 * <= float_max_offset chars from the regex origin (rx_origin).
1042 * If this value is less than last1, use it instead.
1044 assert(rx_origin <= last1);
1046 /* this condition handles the offset==infinity case, and
1047 * is a short-cut otherwise. Although it's comparing a
1048 * byte offset to a char length, it does so in a safe way,
1049 * since 1 char always occupies 1 or more bytes,
1050 * so if a string range is (last1 - rx_origin) bytes,
1051 * it will be less than or equal to (last1 - rx_origin)
1052 * chars; meaning it errs towards doing the accurate HOP3
1053 * rather than just using last1 as a short-cut */
1054 (last1 - rx_origin) < other->max_offset
1056 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1059 assert(strpos + start_shift <= check_at);
1060 last = HOP4c(check_at, other->min_offset - start_shift,
1064 s = HOP3c(rx_origin, other->min_offset, strend);
1065 if (s < other_last) /* These positions already checked */
1068 must = utf8_target ? other->utf8_substr : other->substr;
1069 assert(SvPOK(must));
1072 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1074 multiline ? FBMrf_MULTILINE : 0
1077 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1078 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1079 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1080 s ? "Found" : "Contradicts",
1081 other_ix ? "floating" : "anchored",
1082 quoted, RE_SV_TAIL(must));
1087 /* last1 is latest possible substr location. If we didn't
1088 * find it before there, we never will */
1089 if (last >= last1) {
1090 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1091 ", giving up...\n"));
1095 /* try to find the check substr again at a later
1096 * position. Maybe next time we'll find the "other" substr
1098 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1099 ", trying %s at offset %ld...\n",
1100 (other_ix ? "floating" : "anchored"),
1101 (long)(HOP3c(check_at, 1, strend) - strpos)));
1103 other_last = HOP3c(last, 1, strend) /* highest failure */;
1105 other_ix /* i.e. if other-is-float */
1106 ? HOP3c(rx_origin, 1, strend)
1107 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1111 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
1112 (long)(s - strpos)));
1114 if (other_ix) { /* if (other-is-float) */
1115 /* other_last is set to s, not s+1, since its possible for
1116 * a floating substr to fail first time, then succeed
1117 * second time at the same floating position; e.g.:
1118 * "-AB--AABZ" =~ /\wAB\d*Z/
1119 * The first time round, anchored and float match at
1120 * "-(AB)--AAB(Z)" then fail on the initial \w character
1121 * class. Second time round, they match at "-AB--A(AB)(Z)".
1126 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1127 other_last = HOP3c(s, 1, strend);
1132 DEBUG_OPTIMISE_MORE_r(
1133 PerlIO_printf(Perl_debug_log,
1134 " Check-only match: offset min:%"IVdf" max:%"IVdf
1135 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
1136 " strend-strpos:%"IVdf"\n",
1137 (IV)prog->check_offset_min,
1138 (IV)prog->check_offset_max,
1139 (IV)(check_at-strpos),
1140 (IV)(rx_origin-strpos),
1141 (IV)(rx_origin-check_at),
1147 postprocess_substr_matches:
1149 /* handle the extra constraint of /^.../m if present */
1151 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1154 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1155 " looking for /^/m anchor"));
1157 /* we have failed the constraint of a \n before rx_origin.
1158 * Find the next \n, if any, even if it's beyond the current
1159 * anchored and/or floating substrings. Whether we should be
1160 * scanning ahead for the next \n or the next substr is debatable.
1161 * On the one hand you'd expect rare substrings to appear less
1162 * often than \n's. On the other hand, searching for \n means
1163 * we're effectively flipping been check_substr and "\n" on each
1164 * iteration as the current "rarest" string candidate, which
1165 * means for example that we'll quickly reject the whole string if
1166 * hasn't got a \n, rather than trying every substr position
1170 s = HOP3c(strend, - prog->minlen, strpos);
1171 if (s <= rx_origin ||
1172 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1174 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1175 " Did not find /%s^%s/m...\n",
1176 PL_colors[0], PL_colors[1]));
1180 /* earliest possible origin is 1 char after the \n.
1181 * (since *rx_origin == '\n', it's safe to ++ here rather than
1182 * HOP(rx_origin, 1)) */
1185 if (prog->substrs->check_ix == 0 /* check is anchored */
1186 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1188 /* Position contradicts check-string; either because
1189 * check was anchored (and thus has no wiggle room),
1190 * or check was float and rx_origin is above the float range */
1191 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1192 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1193 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1197 /* if we get here, the check substr must have been float,
1198 * is in range, and we may or may not have had an anchored
1199 * "other" substr which still contradicts */
1200 assert(prog->substrs->check_ix); /* check is float */
1202 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1203 /* whoops, the anchored "other" substr exists, so we still
1204 * contradict. On the other hand, the float "check" substr
1205 * didn't contradict, so just retry the anchored "other"
1207 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1208 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1209 PL_colors[0], PL_colors[1],
1210 (long)(rx_origin - strpos),
1211 (long)(rx_origin - strpos + prog->anchored_offset)));
1212 goto do_other_substr;
1215 /* success: we don't contradict the found floating substring
1216 * (and there's no anchored substr). */
1217 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1218 " Found /%s^%s/m at offset %ld...\n",
1219 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1222 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1223 " (multiline anchor test skipped)\n"));
1229 /* if we have a starting character class, then test that extra constraint.
1230 * (trie stclasses are too expensive to use here, we are better off to
1231 * leave it to regmatch itself) */
1233 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1234 const U8* const str = (U8*)STRING(progi->regstclass);
1236 /* XXX this value could be pre-computed */
1237 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1238 ? (reginfo->is_utf8_pat
1239 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1240 : STR_LEN(progi->regstclass))
1244 /* latest pos that a matching float substr constrains rx start to */
1245 char *rx_max_float = NULL;
1247 /* if the current rx_origin is anchored, either by satisfying an
1248 * anchored substring constraint, or a /^.../m constraint, then we
1249 * can reject the current origin if the start class isn't found
1250 * at the current position. If we have a float-only match, then
1251 * rx_origin is constrained to a range; so look for the start class
1252 * in that range. if neither, then look for the start class in the
1253 * whole rest of the string */
1255 /* XXX DAPM it's not clear what the minlen test is for, and why
1256 * it's not used in the floating case. Nothing in the test suite
1257 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1258 * Here are some old comments, which may or may not be correct:
1260 * minlen == 0 is possible if regstclass is \b or \B,
1261 * and the fixed substr is ''$.
1262 * Since minlen is already taken into account, rx_origin+1 is
1263 * before strend; accidentally, minlen >= 1 guaranties no false
1264 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1265 * 0) below assumes that regstclass does not come from lookahead...
1266 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1267 * This leaves EXACTF-ish only, which are dealt with in
1271 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1272 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1273 else if (prog->float_substr || prog->float_utf8) {
1274 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1275 endpos= HOP3c(rx_max_float, cl_l, strend);
1280 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1281 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
1282 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1283 (IV)start_shift, (IV)(check_at - strbeg),
1284 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1286 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1289 if (endpos == strend) {
1290 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1291 " Could not match STCLASS...\n") );
1294 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1295 " This position contradicts STCLASS...\n") );
1296 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1297 && !(prog->intflags & PREGf_IMPLICIT))
1300 /* Contradict one of substrings */
1301 if (prog->anchored_substr || prog->anchored_utf8) {
1302 if (prog->substrs->check_ix == 1) { /* check is float */
1303 /* Have both, check_string is floating */
1304 assert(rx_origin + start_shift <= check_at);
1305 if (rx_origin + start_shift != check_at) {
1306 /* not at latest position float substr could match:
1307 * Recheck anchored substring, but not floating.
1308 * The condition above is in bytes rather than
1309 * chars for efficiency. It's conservative, in
1310 * that it errs on the side of doing 'goto
1311 * do_other_substr', where a more accurate
1312 * char-based calculation will be done */
1313 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1314 " Looking for anchored substr starting at offset %ld...\n",
1315 (long)(other_last - strpos)) );
1316 goto do_other_substr;
1324 /* In the presence of ml_anch, we might be able to
1325 * find another \n without breaking the current float
1328 /* strictly speaking this should be HOP3c(..., 1, ...),
1329 * but since we goto a block of code that's going to
1330 * search for the next \n if any, its safe here */
1332 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1333 " Looking for /%s^%s/m starting at offset %ld...\n",
1334 PL_colors[0], PL_colors[1],
1335 (long)(rx_origin - strpos)) );
1336 goto postprocess_substr_matches;
1339 /* strictly speaking this can never be true; but might
1340 * be if we ever allow intuit without substrings */
1341 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1344 rx_origin = rx_max_float;
1347 /* at this point, any matching substrings have been
1348 * contradicted. Start again... */
1350 rx_origin = HOP3c(rx_origin, 1, strend);
1352 /* uses bytes rather than char calculations for efficiency.
1353 * It's conservative: it errs on the side of doing 'goto restart',
1354 * where there is code that does a proper char-based test */
1355 if (rx_origin + start_shift + end_shift > strend) {
1356 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1357 " Could not match STCLASS...\n") );
1360 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1361 " Looking for %s substr starting at offset %ld...\n",
1362 (prog->substrs->check_ix ? "floating" : "anchored"),
1363 (long)(rx_origin + start_shift - strpos)) );
1369 if (rx_origin != s) {
1370 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1371 " By STCLASS: moving %ld --> %ld\n",
1372 (long)(rx_origin - strpos), (long)(s - strpos))
1376 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1377 " Does not contradict STCLASS...\n");
1382 /* Decide whether using the substrings helped */
1384 if (rx_origin != strpos) {
1385 /* Fixed substring is found far enough so that the match
1386 cannot start at strpos. */
1388 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1389 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1392 /* The found rx_origin position does not prohibit matching at
1393 * strpos, so calling intuit didn't gain us anything. Decrement
1394 * the BmUSEFUL() count on the check substring, and if we reach
1396 if (!(prog->intflags & PREGf_NAUGHTY)
1398 prog->check_utf8 /* Could be deleted already */
1399 && --BmUSEFUL(prog->check_utf8) < 0
1400 && (prog->check_utf8 == prog->float_utf8)
1402 prog->check_substr /* Could be deleted already */
1403 && --BmUSEFUL(prog->check_substr) < 0
1404 && (prog->check_substr == prog->float_substr)
1407 /* If flags & SOMETHING - do not do it many times on the same match */
1408 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1409 /* XXX Does the destruction order has to change with utf8_target? */
1410 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1411 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1412 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1413 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1414 check = NULL; /* abort */
1415 /* XXXX This is a remnant of the old implementation. It
1416 looks wasteful, since now INTUIT can use many
1417 other heuristics. */
1418 prog->extflags &= ~RXf_USE_INTUIT;
1422 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1423 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1424 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1428 fail_finish: /* Substring not found */
1429 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1430 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1432 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1433 PL_colors[4], PL_colors[5]));
1438 #define DECL_TRIE_TYPE(scan) \
1439 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1440 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
1441 trie_type = ((scan->flags == EXACT) \
1442 ? (utf8_target ? trie_utf8 : trie_plain) \
1443 : (scan->flags == EXACTFA) \
1444 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1445 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1447 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1450 U8 flags = FOLD_FLAGS_FULL; \
1451 switch (trie_type) { \
1452 case trie_utf8_exactfa_fold: \
1453 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1455 case trie_utf8_fold: \
1456 if ( foldlen>0 ) { \
1457 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1462 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1463 len = UTF8SKIP(uc); \
1464 skiplen = UNISKIP( uvc ); \
1465 foldlen -= skiplen; \
1466 uscan = foldbuf + skiplen; \
1469 case trie_latin_utf8_exactfa_fold: \
1470 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1472 case trie_latin_utf8_fold: \
1473 if ( foldlen>0 ) { \
1474 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1480 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1481 skiplen = UNISKIP( uvc ); \
1482 foldlen -= skiplen; \
1483 uscan = foldbuf + skiplen; \
1487 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1494 charid = trie->charmap[ uvc ]; \
1498 if (widecharmap) { \
1499 SV** const svpp = hv_fetch(widecharmap, \
1500 (char*)&uvc, sizeof(UV), 0); \
1502 charid = (U16)SvIV(*svpp); \
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( \
1537 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1546 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1549 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1558 #define REXEC_FBC_TRYIT \
1559 if ((reginfo->intuit || regtry(reginfo, &s))) \
1562 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1563 if (utf8_target) { \
1564 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1567 REXEC_FBC_CLASS_SCAN(CoNd); \
1570 #define DUMP_EXEC_POS(li,s,doutf8) \
1571 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1575 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1576 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1577 tmp = TEST_NON_UTF8(tmp); \
1578 REXEC_FBC_UTF8_SCAN( \
1579 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1588 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1589 if (s == reginfo->strbeg) { \
1593 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1594 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1595 0, UTF8_ALLOW_DEFAULT); \
1598 LOAD_UTF8_CHARCLASS_ALNUM(); \
1599 REXEC_FBC_UTF8_SCAN( \
1600 if (tmp == ! (TeSt2_UtF8)) { \
1609 /* The only difference between the BOUND and NBOUND cases is that
1610 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1611 * NBOUND. This is accomplished by passing it in either the if or else clause,
1612 * with the other one being empty */
1613 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1614 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1616 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1617 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1619 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1620 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1622 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1623 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1626 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1627 * be passed in completely with the variable name being tested, which isn't
1628 * such a clean interface, but this is easier to read than it was before. We
1629 * are looking for the boundary (or non-boundary between a word and non-word
1630 * character. The utf8 and non-utf8 cases have the same logic, but the details
1631 * must be different. Find the "wordness" of the character just prior to this
1632 * one, and compare it with the wordness of this one. If they differ, we have
1633 * a boundary. At the beginning of the string, pretend that the previous
1634 * character was a new-line */
1635 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1636 if (utf8_target) { \
1639 else { /* Not utf8 */ \
1640 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1641 tmp = TEST_NON_UTF8(tmp); \
1643 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1652 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \
1655 /* We know what class REx starts with. Try to find this position... */
1656 /* if reginfo->intuit, its a dryrun */
1657 /* annoyingly all the vars in this routine have different names from their counterparts
1658 in regmatch. /grrr */
1661 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1662 const char *strend, regmatch_info *reginfo)
1665 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1666 char *pat_string; /* The pattern's exactish string */
1667 char *pat_end; /* ptr to end char of pat_string */
1668 re_fold_t folder; /* Function for computing non-utf8 folds */
1669 const U8 *fold_array; /* array for folding ords < 256 */
1675 I32 tmp = 1; /* Scratch variable? */
1676 const bool utf8_target = reginfo->is_utf8_target;
1677 UV utf8_fold_flags = 0;
1678 const bool is_utf8_pat = reginfo->is_utf8_pat;
1679 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1680 with a result inverts that result, as 0^1 =
1682 _char_class_number classnum;
1684 RXi_GET_DECL(prog,progi);
1686 PERL_ARGS_ASSERT_FIND_BYCLASS;
1688 /* We know what class it must start with. */
1692 REXEC_FBC_UTF8_CLASS_SCAN(
1693 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1696 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1701 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1708 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1709 assert(! is_utf8_pat);
1712 if (is_utf8_pat || utf8_target) {
1713 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1714 goto do_exactf_utf8;
1716 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1717 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1718 goto do_exactf_non_utf8; /* isn't dealt with by these */
1720 case EXACTF: /* This node only generated for non-utf8 patterns */
1721 assert(! is_utf8_pat);
1723 utf8_fold_flags = 0;
1724 goto do_exactf_utf8;
1726 fold_array = PL_fold;
1728 goto do_exactf_non_utf8;
1731 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1732 utf8_fold_flags = FOLDEQ_LOCALE;
1733 goto do_exactf_utf8;
1735 fold_array = PL_fold_locale;
1736 folder = foldEQ_locale;
1737 goto do_exactf_non_utf8;
1741 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1743 goto do_exactf_utf8;
1746 if (is_utf8_pat || utf8_target) {
1747 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1748 goto do_exactf_utf8;
1751 /* Any 'ss' in the pattern should have been replaced by regcomp,
1752 * so we don't have to worry here about this single special case
1753 * in the Latin1 range */
1754 fold_array = PL_fold_latin1;
1755 folder = foldEQ_latin1;
1759 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1760 are no glitches with fold-length differences
1761 between the target string and pattern */
1763 /* The idea in the non-utf8 EXACTF* cases is to first find the
1764 * first character of the EXACTF* node and then, if necessary,
1765 * case-insensitively compare the full text of the node. c1 is the
1766 * first character. c2 is its fold. This logic will not work for
1767 * Unicode semantics and the german sharp ss, which hence should
1768 * not be compiled into a node that gets here. */
1769 pat_string = STRING(c);
1770 ln = STR_LEN(c); /* length to match in octets/bytes */
1772 /* We know that we have to match at least 'ln' bytes (which is the
1773 * same as characters, since not utf8). If we have to match 3
1774 * characters, and there are only 2 availabe, we know without
1775 * trying that it will fail; so don't start a match past the
1776 * required minimum number from the far end */
1777 e = HOP3c(strend, -((SSize_t)ln), s);
1779 if (reginfo->intuit && e < s) {
1780 e = s; /* Due to minlen logic of intuit() */
1784 c2 = fold_array[c1];
1785 if (c1 == c2) { /* If char and fold are the same */
1786 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1789 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1797 /* If one of the operands is in utf8, we can't use the simpler folding
1798 * above, due to the fact that many different characters can have the
1799 * same fold, or portion of a fold, or different- length fold */
1800 pat_string = STRING(c);
1801 ln = STR_LEN(c); /* length to match in octets/bytes */
1802 pat_end = pat_string + ln;
1803 lnc = is_utf8_pat /* length to match in characters */
1804 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1807 /* We have 'lnc' characters to match in the pattern, but because of
1808 * multi-character folding, each character in the target can match
1809 * up to 3 characters (Unicode guarantees it will never exceed
1810 * this) if it is utf8-encoded; and up to 2 if not (based on the
1811 * fact that the Latin 1 folds are already determined, and the
1812 * only multi-char fold in that range is the sharp-s folding to
1813 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1814 * string character. Adjust lnc accordingly, rounding up, so that
1815 * if we need to match at least 4+1/3 chars, that really is 5. */
1816 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1817 lnc = (lnc + expansion - 1) / expansion;
1819 /* As in the non-UTF8 case, if we have to match 3 characters, and
1820 * only 2 are left, it's guaranteed to fail, so don't start a
1821 * match that would require us to go beyond the end of the string
1823 e = HOP3c(strend, -((SSize_t)lnc), s);
1825 if (reginfo->intuit && e < s) {
1826 e = s; /* Due to minlen logic of intuit() */
1829 /* XXX Note that we could recalculate e to stop the loop earlier,
1830 * as the worst case expansion above will rarely be met, and as we
1831 * go along we would usually find that e moves further to the left.
1832 * This would happen only after we reached the point in the loop
1833 * where if there were no expansion we should fail. Unclear if
1834 * worth the expense */
1837 char *my_strend= (char *)strend;
1838 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1839 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1840 && (reginfo->intuit || regtry(reginfo, &s)) )
1844 s += (utf8_target) ? UTF8SKIP(s) : 1;
1849 FBC_BOUND(isWORDCHAR_LC,
1850 isWORDCHAR_LC_uvchr(tmp),
1851 isWORDCHAR_LC_utf8((U8*)s));
1854 FBC_NBOUND(isWORDCHAR_LC,
1855 isWORDCHAR_LC_uvchr(tmp),
1856 isWORDCHAR_LC_utf8((U8*)s));
1859 FBC_BOUND(isWORDCHAR,
1860 isWORDCHAR_uni(tmp),
1861 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1864 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1866 isWORDCHAR_A((U8*)s));
1869 FBC_NBOUND(isWORDCHAR,
1870 isWORDCHAR_uni(tmp),
1871 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1874 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1876 isWORDCHAR_A((U8*)s));
1879 FBC_BOUND(isWORDCHAR_L1,
1880 isWORDCHAR_uni(tmp),
1881 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1884 FBC_NBOUND(isWORDCHAR_L1,
1885 isWORDCHAR_uni(tmp),
1886 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1889 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1890 is_LNBREAK_latin1_safe(s, strend)
1894 /* The argument to all the POSIX node types is the class number to pass to
1895 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1902 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1903 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1918 /* The complement of something that matches only ASCII matches all
1919 * non-ASCII, plus everything in ASCII that isn't in the class. */
1920 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
1921 || ! _generic_isCC_A(*s, FLAGS(c)));
1930 /* Don't need to worry about utf8, as it can match only a single
1931 * byte invariant character. */
1932 REXEC_FBC_CLASS_SCAN(
1933 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1941 if (! utf8_target) {
1942 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1948 classnum = (_char_class_number) FLAGS(c);
1949 if (classnum < _FIRST_NON_SWASH_CC) {
1950 while (s < strend) {
1952 /* We avoid loading in the swash as long as possible, but
1953 * should we have to, we jump to a separate loop. This
1954 * extra 'if' statement is what keeps this code from being
1955 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1956 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1957 goto found_above_latin1;
1959 if ((UTF8_IS_INVARIANT(*s)
1960 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1962 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1963 && to_complement ^ cBOOL(
1964 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
1968 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1980 else switch (classnum) { /* These classes are implemented as
1982 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1983 revert the change of \v matching this */
1986 case _CC_ENUM_PSXSPC:
1987 REXEC_FBC_UTF8_CLASS_SCAN(
1988 to_complement ^ cBOOL(isSPACE_utf8(s)));
1991 case _CC_ENUM_BLANK:
1992 REXEC_FBC_UTF8_CLASS_SCAN(
1993 to_complement ^ cBOOL(isBLANK_utf8(s)));
1996 case _CC_ENUM_XDIGIT:
1997 REXEC_FBC_UTF8_CLASS_SCAN(
1998 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2001 case _CC_ENUM_VERTSPACE:
2002 REXEC_FBC_UTF8_CLASS_SCAN(
2003 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2006 case _CC_ENUM_CNTRL:
2007 REXEC_FBC_UTF8_CLASS_SCAN(
2008 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2012 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2013 assert(0); /* NOTREACHED */
2018 found_above_latin1: /* Here we have to load a swash to get the result
2019 for the current code point */
2020 if (! PL_utf8_swash_ptrs[classnum]) {
2021 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2022 PL_utf8_swash_ptrs[classnum] =
2023 _core_swash_init("utf8",
2026 PL_XPosix_ptrs[classnum], &flags);
2029 /* This is a copy of the loop above for swash classes, though using the
2030 * FBC macro instead of being expanded out. Since we've loaded the
2031 * swash, we don't have to check for that each time through the loop */
2032 REXEC_FBC_UTF8_CLASS_SCAN(
2033 to_complement ^ cBOOL(_generic_utf8(
2036 swash_fetch(PL_utf8_swash_ptrs[classnum],
2044 /* what trie are we using right now */
2045 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2046 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2047 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2049 const char *last_start = strend - trie->minlen;
2051 const char *real_start = s;
2053 STRLEN maxlen = trie->maxlen;
2055 U8 **points; /* map of where we were in the input string
2056 when reading a given char. For ASCII this
2057 is unnecessary overhead as the relationship
2058 is always 1:1, but for Unicode, especially
2059 case folded Unicode this is not true. */
2060 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2064 GET_RE_DEBUG_FLAGS_DECL;
2066 /* We can't just allocate points here. We need to wrap it in
2067 * an SV so it gets freed properly if there is a croak while
2068 * running the match */
2071 sv_points=newSV(maxlen * sizeof(U8 *));
2072 SvCUR_set(sv_points,
2073 maxlen * sizeof(U8 *));
2074 SvPOK_on(sv_points);
2075 sv_2mortal(sv_points);
2076 points=(U8**)SvPV_nolen(sv_points );
2077 if ( trie_type != trie_utf8_fold
2078 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2081 bitmap=(U8*)trie->bitmap;
2083 bitmap=(U8*)ANYOF_BITMAP(c);
2085 /* this is the Aho-Corasick algorithm modified a touch
2086 to include special handling for long "unknown char" sequences.
2087 The basic idea being that we use AC as long as we are dealing
2088 with a possible matching char, when we encounter an unknown char
2089 (and we have not encountered an accepting state) we scan forward
2090 until we find a legal starting char.
2091 AC matching is basically that of trie matching, except that when
2092 we encounter a failing transition, we fall back to the current
2093 states "fail state", and try the current char again, a process
2094 we repeat until we reach the root state, state 1, or a legal
2095 transition. If we fail on the root state then we can either
2096 terminate if we have reached an accepting state previously, or
2097 restart the entire process from the beginning if we have not.
2100 while (s <= last_start) {
2101 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2109 U8 *uscan = (U8*)NULL;
2110 U8 *leftmost = NULL;
2112 U32 accepted_word= 0;
2116 while ( state && uc <= (U8*)strend ) {
2118 U32 word = aho->states[ state ].wordnum;
2122 DEBUG_TRIE_EXECUTE_r(
2123 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2124 dump_exec_pos( (char *)uc, c, strend, real_start,
2125 (char *)uc, utf8_target );
2126 PerlIO_printf( Perl_debug_log,
2127 " Scanning for legal start char...\n");
2131 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2135 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2141 if (uc >(U8*)last_start) break;
2145 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2146 if (!leftmost || lpos < leftmost) {
2147 DEBUG_r(accepted_word=word);
2153 points[pointpos++ % maxlen]= uc;
2154 if (foldlen || uc < (U8*)strend) {
2155 REXEC_TRIE_READ_CHAR(trie_type, trie,
2157 uscan, len, uvc, charid, foldlen,
2159 DEBUG_TRIE_EXECUTE_r({
2160 dump_exec_pos( (char *)uc, c, strend,
2161 real_start, s, utf8_target);
2162 PerlIO_printf(Perl_debug_log,
2163 " Charid:%3u CP:%4"UVxf" ",
2175 word = aho->states[ state ].wordnum;
2177 base = aho->states[ state ].trans.base;
2179 DEBUG_TRIE_EXECUTE_r({
2181 dump_exec_pos( (char *)uc, c, strend, real_start,
2183 PerlIO_printf( Perl_debug_log,
2184 "%sState: %4"UVxf", word=%"UVxf,
2185 failed ? " Fail transition to " : "",
2186 (UV)state, (UV)word);
2192 ( ((offset = base + charid
2193 - 1 - trie->uniquecharcount)) >= 0)
2194 && ((U32)offset < trie->lasttrans)
2195 && trie->trans[offset].check == state
2196 && (tmp=trie->trans[offset].next))
2198 DEBUG_TRIE_EXECUTE_r(
2199 PerlIO_printf( Perl_debug_log," - legal\n"));
2204 DEBUG_TRIE_EXECUTE_r(
2205 PerlIO_printf( Perl_debug_log," - fail\n"));
2207 state = aho->fail[state];
2211 /* we must be accepting here */
2212 DEBUG_TRIE_EXECUTE_r(
2213 PerlIO_printf( Perl_debug_log," - accepting\n"));
2222 if (!state) state = 1;
2225 if ( aho->states[ state ].wordnum ) {
2226 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2227 if (!leftmost || lpos < leftmost) {
2228 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2233 s = (char*)leftmost;
2234 DEBUG_TRIE_EXECUTE_r({
2236 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2237 (UV)accepted_word, (IV)(s - real_start)
2240 if (reginfo->intuit || regtry(reginfo, &s)) {
2246 DEBUG_TRIE_EXECUTE_r({
2247 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2250 DEBUG_TRIE_EXECUTE_r(
2251 PerlIO_printf( Perl_debug_log,"No match.\n"));
2260 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2268 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2269 * flags have same meanings as with regexec_flags() */
2272 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2279 struct regexp *const prog = ReANY(rx);
2281 if (flags & REXEC_COPY_STR) {
2285 PerlIO_printf(Perl_debug_log,
2286 "Copy on write: regexp capture, type %d\n",
2289 /* Create a new COW SV to share the match string and store
2290 * in saved_copy, unless the current COW SV in saved_copy
2291 * is valid and suitable for our purpose */
2292 if (( prog->saved_copy
2293 && SvIsCOW(prog->saved_copy)
2294 && SvPOKp(prog->saved_copy)
2297 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2299 /* just reuse saved_copy SV */
2300 if (RXp_MATCH_COPIED(prog)) {
2301 Safefree(prog->subbeg);
2302 RXp_MATCH_COPIED_off(prog);
2306 /* create new COW SV to share string */
2307 RX_MATCH_COPY_FREE(rx);
2308 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2310 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2311 assert (SvPOKp(prog->saved_copy));
2312 prog->sublen = strend - strbeg;
2313 prog->suboffset = 0;
2314 prog->subcoffset = 0;
2319 SSize_t max = strend - strbeg;
2322 if ( (flags & REXEC_COPY_SKIP_POST)
2323 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2324 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2325 ) { /* don't copy $' part of string */
2328 /* calculate the right-most part of the string covered
2329 * by a capture. Due to look-ahead, this may be to
2330 * the right of $&, so we have to scan all captures */
2331 while (n <= prog->lastparen) {
2332 if (prog->offs[n].end > max)
2333 max = prog->offs[n].end;
2337 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2338 ? prog->offs[0].start
2340 assert(max >= 0 && max <= strend - strbeg);
2343 if ( (flags & REXEC_COPY_SKIP_PRE)
2344 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2345 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2346 ) { /* don't copy $` part of string */
2349 /* calculate the left-most part of the string covered
2350 * by a capture. Due to look-behind, this may be to
2351 * the left of $&, so we have to scan all captures */
2352 while (min && n <= prog->lastparen) {
2353 if ( prog->offs[n].start != -1
2354 && prog->offs[n].start < min)
2356 min = prog->offs[n].start;
2360 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2361 && min > prog->offs[0].end
2363 min = prog->offs[0].end;
2367 assert(min >= 0 && min <= max && min <= strend - strbeg);
2370 if (RX_MATCH_COPIED(rx)) {
2371 if (sublen > prog->sublen)
2373 (char*)saferealloc(prog->subbeg, sublen+1);
2376 prog->subbeg = (char*)safemalloc(sublen+1);
2377 Copy(strbeg + min, prog->subbeg, sublen, char);
2378 prog->subbeg[sublen] = '\0';
2379 prog->suboffset = min;
2380 prog->sublen = sublen;
2381 RX_MATCH_COPIED_on(rx);
2383 prog->subcoffset = prog->suboffset;
2384 if (prog->suboffset && utf8_target) {
2385 /* Convert byte offset to chars.
2386 * XXX ideally should only compute this if @-/@+
2387 * has been seen, a la PL_sawampersand ??? */
2389 /* If there's a direct correspondence between the
2390 * string which we're matching and the original SV,
2391 * then we can use the utf8 len cache associated with
2392 * the SV. In particular, it means that under //g,
2393 * sv_pos_b2u() will use the previously cached
2394 * position to speed up working out the new length of
2395 * subcoffset, rather than counting from the start of
2396 * the string each time. This stops
2397 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2398 * from going quadratic */
2399 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2400 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2401 SV_GMAGIC|SV_CONST_RETURN);
2403 prog->subcoffset = utf8_length((U8*)strbeg,
2404 (U8*)(strbeg+prog->suboffset));
2408 RX_MATCH_COPY_FREE(rx);
2409 prog->subbeg = strbeg;
2410 prog->suboffset = 0;
2411 prog->subcoffset = 0;
2412 prog->sublen = strend - strbeg;
2420 - regexec_flags - match a regexp against a string
2423 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2424 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2425 /* stringarg: the point in the string at which to begin matching */
2426 /* strend: pointer to null at end of string */
2427 /* strbeg: real beginning of string */
2428 /* minend: end of match must be >= minend bytes after stringarg. */
2429 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2430 * itself is accessed via the pointers above */
2431 /* data: May be used for some additional optimizations.
2432 Currently unused. */
2433 /* flags: For optimizations. See REXEC_* in regexp.h */
2437 struct regexp *const prog = ReANY(rx);
2441 SSize_t minlen; /* must match at least this many chars */
2442 SSize_t dontbother = 0; /* how many characters not to try at end */
2443 const bool utf8_target = cBOOL(DO_UTF8(sv));
2445 RXi_GET_DECL(prog,progi);
2446 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2447 regmatch_info *const reginfo = ®info_buf;
2448 regexp_paren_pair *swap = NULL;
2450 GET_RE_DEBUG_FLAGS_DECL;
2452 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2453 PERL_UNUSED_ARG(data);
2455 /* Be paranoid... */
2456 if (prog == NULL || stringarg == NULL) {
2457 Perl_croak(aTHX_ "NULL regexp parameter");
2462 debug_start_match(rx, utf8_target, stringarg, strend,
2466 startpos = stringarg;
2468 if (prog->intflags & PREGf_GPOS_SEEN) {
2471 /* set reginfo->ganch, the position where \G can match */
2474 (flags & REXEC_IGNOREPOS)
2475 ? stringarg /* use start pos rather than pos() */
2476 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2477 /* Defined pos(): */
2478 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2479 : strbeg; /* pos() not defined; use start of string */
2481 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2482 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
2484 /* in the presence of \G, we may need to start looking earlier in
2485 * the string than the suggested start point of stringarg:
2486 * if prog->gofs is set, then that's a known, fixed minimum
2489 * /ab|c\G/: gofs = 1
2490 * or if the minimum offset isn't known, then we have to go back
2491 * to the start of the string, e.g. /w+\G/
2494 if (prog->intflags & PREGf_ANCH_GPOS) {
2495 startpos = reginfo->ganch - prog->gofs;
2497 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2499 DEBUG_r(PerlIO_printf(Perl_debug_log,
2500 "fail: ganch-gofs before earliest possible start\n"));
2504 else if (prog->gofs) {
2505 if (startpos - prog->gofs < strbeg)
2508 startpos -= prog->gofs;
2510 else if (prog->intflags & PREGf_GPOS_FLOAT)
2514 minlen = prog->minlen;
2515 if ((startpos + minlen) > strend || startpos < strbeg) {
2516 DEBUG_r(PerlIO_printf(Perl_debug_log,
2517 "Regex match can't succeed, so not even tried\n"));
2521 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2522 * which will call destuctors to reset PL_regmatch_state, free higher
2523 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2524 * regmatch_info_aux_eval */
2526 oldsave = PL_savestack_ix;
2530 if ((prog->extflags & RXf_USE_INTUIT)
2531 && !(flags & REXEC_CHECKED))
2533 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2538 if (prog->extflags & RXf_CHECK_ALL) {
2539 /* we can match based purely on the result of INTUIT.
2540 * Set up captures etc just for $& and $-[0]
2541 * (an intuit-only match wont have $1,$2,..) */
2542 assert(!prog->nparens);
2544 /* s/// doesn't like it if $& is earlier than where we asked it to
2545 * start searching (which can happen on something like /.\G/) */
2546 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2549 /* this should only be possible under \G */
2550 assert(prog->intflags & PREGf_GPOS_SEEN);
2551 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2552 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2556 /* match via INTUIT shouldn't have any captures.
2557 * Let @-, @+, $^N know */
2558 prog->lastparen = prog->lastcloseparen = 0;
2559 RX_MATCH_UTF8_set(rx, utf8_target);
2560 prog->offs[0].start = s - strbeg;
2561 prog->offs[0].end = utf8_target
2562 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2563 : s - strbeg + prog->minlenret;
2564 if ( !(flags & REXEC_NOT_FIRST) )
2565 S_reg_set_capture_string(aTHX_ rx,
2567 sv, flags, utf8_target);
2573 multiline = prog->extflags & RXf_PMf_MULTILINE;
2575 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2576 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2577 "String too short [regexec_flags]...\n"));
2581 /* Check validity of program. */
2582 if (UCHARAT(progi->program) != REG_MAGIC) {
2583 Perl_croak(aTHX_ "corrupted regexp program");
2586 RX_MATCH_TAINTED_off(rx);
2588 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2589 reginfo->intuit = 0;
2590 reginfo->is_utf8_target = cBOOL(utf8_target);
2591 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2592 reginfo->warned = FALSE;
2593 reginfo->strbeg = strbeg;
2595 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2596 reginfo->strend = strend;
2597 /* see how far we have to get to not match where we matched before */
2598 reginfo->till = stringarg + minend;
2600 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
2601 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2602 S_cleanup_regmatch_info_aux has executed (registered by
2603 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2604 magic belonging to this SV.
2605 Not newSVsv, either, as it does not COW.
2607 assert(!IS_PADGV(sv));
2608 reginfo->sv = newSV(0);
2609 SvSetSV_nosteal(reginfo->sv, sv);
2610 SAVEFREESV(reginfo->sv);
2613 /* reserve next 2 or 3 slots in PL_regmatch_state:
2614 * slot N+0: may currently be in use: skip it
2615 * slot N+1: use for regmatch_info_aux struct
2616 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2617 * slot N+3: ready for use by regmatch()
2621 regmatch_state *old_regmatch_state;
2622 regmatch_slab *old_regmatch_slab;
2623 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2625 /* on first ever match, allocate first slab */
2626 if (!PL_regmatch_slab) {
2627 Newx(PL_regmatch_slab, 1, regmatch_slab);
2628 PL_regmatch_slab->prev = NULL;
2629 PL_regmatch_slab->next = NULL;
2630 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2633 old_regmatch_state = PL_regmatch_state;
2634 old_regmatch_slab = PL_regmatch_slab;
2636 for (i=0; i <= max; i++) {
2638 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2640 reginfo->info_aux_eval =
2641 reginfo->info_aux->info_aux_eval =
2642 &(PL_regmatch_state->u.info_aux_eval);
2644 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2645 PL_regmatch_state = S_push_slab(aTHX);
2648 /* note initial PL_regmatch_state position; at end of match we'll
2649 * pop back to there and free any higher slabs */
2651 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2652 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2653 reginfo->info_aux->poscache = NULL;
2655 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2657 if ((prog->extflags & RXf_EVAL_SEEN))
2658 S_setup_eval_state(aTHX_ reginfo);
2660 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2663 /* If there is a "must appear" string, look for it. */
2665 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2666 /* We have to be careful. If the previous successful match
2667 was from this regex we don't want a subsequent partially
2668 successful match to clobber the old results.
2669 So when we detect this possibility we add a swap buffer
2670 to the re, and switch the buffer each match. If we fail,
2671 we switch it back; otherwise we leave it swapped.
2674 /* do we need a save destructor here for eval dies? */
2675 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2676 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2677 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2684 /* Simplest case: anchored match need be tried only once. */
2685 /* [unless only anchor is BOL and multiline is set] */
2686 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
2687 if (s == startpos && regtry(reginfo, &s))
2689 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
2694 dontbother = minlen - 1;
2695 end = HOP3c(strend, -dontbother, strbeg) - 1;
2696 /* for multiline we only have to try after newlines */
2697 if (prog->check_substr || prog->check_utf8) {
2698 /* because of the goto we can not easily reuse the macros for bifurcating the
2699 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2702 goto after_try_utf8;
2704 if (regtry(reginfo, &s)) {
2711 if (prog->extflags & RXf_USE_INTUIT) {
2712 s = re_intuit_start(rx, sv, strbeg,
2713 s + UTF8SKIP(s), strend, flags, NULL);
2722 } /* end search for check string in unicode */
2724 if (s == startpos) {
2725 goto after_try_latin;
2728 if (regtry(reginfo, &s)) {
2735 if (prog->extflags & RXf_USE_INTUIT) {
2736 s = re_intuit_start(rx, sv, strbeg,
2737 s + 1, strend, flags, NULL);
2746 } /* end search for check string in latin*/
2747 } /* end search for check string */
2748 else { /* search for newline */
2750 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2753 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2754 while (s <= end) { /* note it could be possible to match at the end of the string */
2755 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2756 if (regtry(reginfo, &s))
2760 } /* end search for newline */
2761 } /* end anchored/multiline check string search */
2763 } else if (prog->intflags & PREGf_ANCH_GPOS)
2765 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2766 assert(prog->intflags & PREGf_GPOS_SEEN);
2767 /* For anchored \G, the only position it can match from is
2768 * (ganch-gofs); we already set startpos to this above; if intuit
2769 * moved us on from there, we can't possibly succeed */
2770 assert(startpos == reginfo->ganch - prog->gofs);
2771 if (s == startpos && regtry(reginfo, &s))
2776 /* Messy cases: unanchored match. */
2777 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2778 /* we have /x+whatever/ */
2779 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2785 if (! prog->anchored_utf8) {
2786 to_utf8_substr(prog);
2788 ch = SvPVX_const(prog->anchored_utf8)[0];
2791 DEBUG_EXECUTE_r( did_match = 1 );
2792 if (regtry(reginfo, &s)) goto got_it;
2794 while (s < strend && *s == ch)
2801 if (! prog->anchored_substr) {
2802 if (! to_byte_substr(prog)) {
2803 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2806 ch = SvPVX_const(prog->anchored_substr)[0];
2809 DEBUG_EXECUTE_r( did_match = 1 );
2810 if (regtry(reginfo, &s)) goto got_it;
2812 while (s < strend && *s == ch)
2817 DEBUG_EXECUTE_r(if (!did_match)
2818 PerlIO_printf(Perl_debug_log,
2819 "Did not find anchored character...\n")
2822 else if (prog->anchored_substr != NULL
2823 || prog->anchored_utf8 != NULL
2824 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2825 && prog->float_max_offset < strend - s)) {
2830 char *last1; /* Last position checked before */
2834 if (prog->anchored_substr || prog->anchored_utf8) {
2836 if (! prog->anchored_utf8) {
2837 to_utf8_substr(prog);
2839 must = prog->anchored_utf8;
2842 if (! prog->anchored_substr) {
2843 if (! to_byte_substr(prog)) {
2844 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2847 must = prog->anchored_substr;
2849 back_max = back_min = prog->anchored_offset;
2852 if (! prog->float_utf8) {
2853 to_utf8_substr(prog);
2855 must = prog->float_utf8;
2858 if (! prog->float_substr) {
2859 if (! to_byte_substr(prog)) {
2860 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2863 must = prog->float_substr;
2865 back_max = prog->float_max_offset;
2866 back_min = prog->float_min_offset;
2872 last = HOP3c(strend, /* Cannot start after this */
2873 -(SSize_t)(CHR_SVLEN(must)
2874 - (SvTAIL(must) != 0) + back_min), strbeg);
2876 if (s > reginfo->strbeg)
2877 last1 = HOPc(s, -1);
2879 last1 = s - 1; /* bogus */
2881 /* XXXX check_substr already used to find "s", can optimize if
2882 check_substr==must. */
2884 strend = HOPc(strend, -dontbother);
2885 while ( (s <= last) &&
2886 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
2887 (unsigned char*)strend, must,
2888 multiline ? FBMrf_MULTILINE : 0)) ) {
2889 DEBUG_EXECUTE_r( did_match = 1 );
2890 if (HOPc(s, -back_max) > last1) {
2891 last1 = HOPc(s, -back_min);
2892 s = HOPc(s, -back_max);
2895 char * const t = (last1 >= reginfo->strbeg)
2896 ? HOPc(last1, 1) : last1 + 1;
2898 last1 = HOPc(s, -back_min);
2902 while (s <= last1) {
2903 if (regtry(reginfo, &s))
2906 s++; /* to break out of outer loop */
2913 while (s <= last1) {
2914 if (regtry(reginfo, &s))
2920 DEBUG_EXECUTE_r(if (!did_match) {
2921 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2922 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2923 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2924 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2925 ? "anchored" : "floating"),
2926 quoted, RE_SV_TAIL(must));
2930 else if ( (c = progi->regstclass) ) {
2932 const OPCODE op = OP(progi->regstclass);
2933 /* don't bother with what can't match */
2934 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2935 strend = HOPc(strend, -(minlen - 1));
2938 SV * const prop = sv_newmortal();
2939 regprop(prog, prop, c, reginfo);
2941 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2943 PerlIO_printf(Perl_debug_log,
2944 "Matching stclass %.*s against %s (%d bytes)\n",
2945 (int)SvCUR(prop), SvPVX_const(prop),
2946 quoted, (int)(strend - s));
2949 if (find_byclass(prog, c, s, strend, reginfo))
2951 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2955 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2963 if (! prog->float_utf8) {
2964 to_utf8_substr(prog);
2966 float_real = prog->float_utf8;
2969 if (! prog->float_substr) {
2970 if (! to_byte_substr(prog)) {
2971 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2974 float_real = prog->float_substr;
2977 little = SvPV_const(float_real, len);
2978 if (SvTAIL(float_real)) {
2979 /* This means that float_real contains an artificial \n on
2980 * the end due to the presence of something like this:
2981 * /foo$/ where we can match both "foo" and "foo\n" at the
2982 * end of the string. So we have to compare the end of the
2983 * string first against the float_real without the \n and
2984 * then against the full float_real with the string. We
2985 * have to watch out for cases where the string might be
2986 * smaller than the float_real or the float_real without
2988 char *checkpos= strend - len;
2990 PerlIO_printf(Perl_debug_log,
2991 "%sChecking for float_real.%s\n",
2992 PL_colors[4], PL_colors[5]));
2993 if (checkpos + 1 < strbeg) {
2994 /* can't match, even if we remove the trailing \n
2995 * string is too short to match */
2997 PerlIO_printf(Perl_debug_log,
2998 "%sString shorter than required trailing substring, cannot match.%s\n",
2999 PL_colors[4], PL_colors[5]));
3001 } else if (memEQ(checkpos + 1, little, len - 1)) {
3002 /* can match, the end of the string matches without the
3004 last = checkpos + 1;
3005 } else if (checkpos < strbeg) {
3006 /* cant match, string is too short when the "\n" is
3009 PerlIO_printf(Perl_debug_log,
3010 "%sString does not contain required trailing substring, cannot match.%s\n",
3011 PL_colors[4], PL_colors[5]));
3013 } else if (!multiline) {
3014 /* non multiline match, so compare with the "\n" at the
3015 * end of the string */
3016 if (memEQ(checkpos, little, len)) {
3020 PerlIO_printf(Perl_debug_log,
3021 "%sString does not contain required trailing substring, cannot match.%s\n",
3022 PL_colors[4], PL_colors[5]));
3026 /* multiline match, so we have to search for a place
3027 * where the full string is located */
3033 last = rninstr(s, strend, little, little + len);
3035 last = strend; /* matching "$" */
3038 /* at one point this block contained a comment which was
3039 * probably incorrect, which said that this was a "should not
3040 * happen" case. Even if it was true when it was written I am
3041 * pretty sure it is not anymore, so I have removed the comment
3042 * and replaced it with this one. Yves */
3044 PerlIO_printf(Perl_debug_log,
3045 "String does not contain required substring, cannot match.\n"
3049 dontbother = strend - last + prog->float_min_offset;
3051 if (minlen && (dontbother < minlen))
3052 dontbother = minlen - 1;
3053 strend -= dontbother; /* this one's always in bytes! */
3054 /* We don't know much -- general case. */
3057 if (regtry(reginfo, &s))
3066 if (regtry(reginfo, &s))
3068 } while (s++ < strend);
3076 /* s/// doesn't like it if $& is earlier than where we asked it to
3077 * start searching (which can happen on something like /.\G/) */
3078 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3079 && (prog->offs[0].start < stringarg - strbeg))
3081 /* this should only be possible under \G */
3082 assert(prog->intflags & PREGf_GPOS_SEEN);
3083 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3084 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3090 PerlIO_printf(Perl_debug_log,
3091 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3098 /* clean up; this will trigger destructors that will free all slabs
3099 * above the current one, and cleanup the regmatch_info_aux
3100 * and regmatch_info_aux_eval sructs */
3102 LEAVE_SCOPE(oldsave);
3104 if (RXp_PAREN_NAMES(prog))
3105 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3107 RX_MATCH_UTF8_set(rx, utf8_target);
3109 /* make sure $`, $&, $', and $digit will work later */
3110 if ( !(flags & REXEC_NOT_FIRST) )
3111 S_reg_set_capture_string(aTHX_ rx,
3112 strbeg, reginfo->strend,
3113 sv, flags, utf8_target);
3118 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
3119 PL_colors[4], PL_colors[5]));
3121 /* clean up; this will trigger destructors that will free all slabs
3122 * above the current one, and cleanup the regmatch_info_aux
3123 * and regmatch_info_aux_eval sructs */
3125 LEAVE_SCOPE(oldsave);
3128 /* we failed :-( roll it back */
3129 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3130 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3135 Safefree(prog->offs);
3142 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3143 * Do inc before dec, in case old and new rex are the same */
3144 #define SET_reg_curpm(Re2) \
3145 if (reginfo->info_aux_eval) { \
3146 (void)ReREFCNT_inc(Re2); \
3147 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3148 PM_SETRE((PL_reg_curpm), (Re2)); \
3153 - regtry - try match at specific point
3155 STATIC I32 /* 0 failure, 1 success */
3156 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3160 REGEXP *const rx = reginfo->prog;
3161 regexp *const prog = ReANY(rx);
3163 RXi_GET_DECL(prog,progi);
3164 GET_RE_DEBUG_FLAGS_DECL;
3166 PERL_ARGS_ASSERT_REGTRY;
3168 reginfo->cutpoint=NULL;
3170 prog->offs[0].start = *startposp - reginfo->strbeg;
3171 prog->lastparen = 0;
3172 prog->lastcloseparen = 0;
3174 /* XXXX What this code is doing here?!!! There should be no need
3175 to do this again and again, prog->lastparen should take care of
3178 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3179 * Actually, the code in regcppop() (which Ilya may be meaning by
3180 * prog->lastparen), is not needed at all by the test suite
3181 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3182 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3183 * Meanwhile, this code *is* needed for the
3184 * above-mentioned test suite tests to succeed. The common theme
3185 * on those tests seems to be returning null fields from matches.
3186 * --jhi updated by dapm */
3188 if (prog->nparens) {
3189 regexp_paren_pair *pp = prog->offs;
3191 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3199 result = regmatch(reginfo, *startposp, progi->program + 1);
3201 prog->offs[0].end = result;
3204 if (reginfo->cutpoint)
3205 *startposp= reginfo->cutpoint;
3206 REGCP_UNWIND(lastcp);
3211 #define sayYES goto yes
3212 #define sayNO goto no
3213 #define sayNO_SILENT goto no_silent
3215 /* we dont use STMT_START/END here because it leads to
3216 "unreachable code" warnings, which are bogus, but distracting. */
3217 #define CACHEsayNO \
3218 if (ST.cache_mask) \
3219 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3222 /* this is used to determine how far from the left messages like
3223 'failed...' are printed. It should be set such that messages
3224 are inline with the regop output that created them.
3226 #define REPORT_CODE_OFF 32
3229 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3230 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3231 #define CHRTEST_NOT_A_CP_1 -999
3232 #define CHRTEST_NOT_A_CP_2 -998
3234 /* grab a new slab and return the first slot in it */
3236 STATIC regmatch_state *
3239 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3242 regmatch_slab *s = PL_regmatch_slab->next;
3244 Newx(s, 1, regmatch_slab);
3245 s->prev = PL_regmatch_slab;
3247 PL_regmatch_slab->next = s;
3249 PL_regmatch_slab = s;
3250 return SLAB_FIRST(s);
3254 /* push a new state then goto it */
3256 #define PUSH_STATE_GOTO(state, node, input) \
3257 pushinput = input; \
3259 st->resume_state = state; \
3262 /* push a new state with success backtracking, then goto it */
3264 #define PUSH_YES_STATE_GOTO(state, node, input) \
3265 pushinput = input; \
3267 st->resume_state = state; \
3268 goto push_yes_state;
3275 regmatch() - main matching routine
3277 This is basically one big switch statement in a loop. We execute an op,
3278 set 'next' to point the next op, and continue. If we come to a point which
3279 we may need to backtrack to on failure such as (A|B|C), we push a
3280 backtrack state onto the backtrack stack. On failure, we pop the top
3281 state, and re-enter the loop at the state indicated. If there are no more
3282 states to pop, we return failure.
3284 Sometimes we also need to backtrack on success; for example /A+/, where
3285 after successfully matching one A, we need to go back and try to
3286 match another one; similarly for lookahead assertions: if the assertion
3287 completes successfully, we backtrack to the state just before the assertion
3288 and then carry on. In these cases, the pushed state is marked as
3289 'backtrack on success too'. This marking is in fact done by a chain of
3290 pointers, each pointing to the previous 'yes' state. On success, we pop to
3291 the nearest yes state, discarding any intermediate failure-only states.
3292 Sometimes a yes state is pushed just to force some cleanup code to be
3293 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3294 it to free the inner regex.
3296 Note that failure backtracking rewinds the cursor position, while
3297 success backtracking leaves it alone.
3299 A pattern is complete when the END op is executed, while a subpattern
3300 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3301 ops trigger the "pop to last yes state if any, otherwise return true"
3304 A common convention in this function is to use A and B to refer to the two
3305 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3306 the subpattern to be matched possibly multiple times, while B is the entire
3307 rest of the pattern. Variable and state names reflect this convention.
3309 The states in the main switch are the union of ops and failure/success of
3310 substates associated with with that op. For example, IFMATCH is the op
3311 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3312 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3313 successfully matched A and IFMATCH_A_fail is a state saying that we have
3314 just failed to match A. Resume states always come in pairs. The backtrack
3315 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3316 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3317 on success or failure.
3319 The struct that holds a backtracking state is actually a big union, with
3320 one variant for each major type of op. The variable st points to the
3321 top-most backtrack struct. To make the code clearer, within each
3322 block of code we #define ST to alias the relevant union.
3324 Here's a concrete example of a (vastly oversimplified) IFMATCH
3330 #define ST st->u.ifmatch
3332 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3333 ST.foo = ...; // some state we wish to save
3335 // push a yes backtrack state with a resume value of
3336 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3338 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3341 case IFMATCH_A: // we have successfully executed A; now continue with B
3343 bar = ST.foo; // do something with the preserved value
3346 case IFMATCH_A_fail: // A failed, so the assertion failed
3347 ...; // do some housekeeping, then ...
3348 sayNO; // propagate the failure
3355 For any old-timers reading this who are familiar with the old recursive
3356 approach, the code above is equivalent to:
3358 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3367 ...; // do some housekeeping, then ...
3368 sayNO; // propagate the failure
3371 The topmost backtrack state, pointed to by st, is usually free. If you
3372 want to claim it, populate any ST.foo fields in it with values you wish to
3373 save, then do one of
3375 PUSH_STATE_GOTO(resume_state, node, newinput);
3376 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3378 which sets that backtrack state's resume value to 'resume_state', pushes a
3379 new free entry to the top of the backtrack stack, then goes to 'node'.
3380 On backtracking, the free slot is popped, and the saved state becomes the
3381 new free state. An ST.foo field in this new top state can be temporarily
3382 accessed to retrieve values, but once the main loop is re-entered, it
3383 becomes available for reuse.
3385 Note that the depth of the backtrack stack constantly increases during the
3386 left-to-right execution of the pattern, rather than going up and down with
3387 the pattern nesting. For example the stack is at its maximum at Z at the
3388 end of the pattern, rather than at X in the following:
3390 /(((X)+)+)+....(Y)+....Z/
3392 The only exceptions to this are lookahead/behind assertions and the cut,
3393 (?>A), which pop all the backtrack states associated with A before
3396 Backtrack state structs are allocated in slabs of about 4K in size.
3397 PL_regmatch_state and st always point to the currently active state,
3398 and PL_regmatch_slab points to the slab currently containing
3399 PL_regmatch_state. The first time regmatch() is called, the first slab is
3400 allocated, and is never freed until interpreter destruction. When the slab
3401 is full, a new one is allocated and chained to the end. At exit from
3402 regmatch(), slabs allocated since entry are freed.
3407 #define DEBUG_STATE_pp(pp) \
3409 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3410 PerlIO_printf(Perl_debug_log, \
3411 " %*s"pp" %s%s%s%s%s\n", \
3413 PL_reg_name[st->resume_state], \
3414 ((st==yes_state||st==mark_state) ? "[" : ""), \
3415 ((st==yes_state) ? "Y" : ""), \
3416 ((st==mark_state) ? "M" : ""), \
3417 ((st==yes_state||st==mark_state) ? "]" : "") \
3422 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3427 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3428 const char *start, const char *end, const char *blurb)
3430 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3432 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3437 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3438 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3440 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3441 start, end - start, 60);
3443 PerlIO_printf(Perl_debug_log,
3444 "%s%s REx%s %s against %s\n",
3445 PL_colors[4], blurb, PL_colors[5], s0, s1);
3447 if (utf8_target||utf8_pat)
3448 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3449 utf8_pat ? "pattern" : "",
3450 utf8_pat && utf8_target ? " and " : "",
3451 utf8_target ? "string" : ""
3457 S_dump_exec_pos(pTHX_ const char *locinput,
3458 const regnode *scan,
3459 const char *loc_regeol,
3460 const char *loc_bostr,
3461 const char *loc_reg_starttry,
3462 const bool utf8_target)
3464 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3465 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3466 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3467 /* The part of the string before starttry has one color
3468 (pref0_len chars), between starttry and current
3469 position another one (pref_len - pref0_len chars),
3470 after the current position the third one.
3471 We assume that pref0_len <= pref_len, otherwise we
3472 decrease pref0_len. */
3473 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3474 ? (5 + taill) - l : locinput - loc_bostr;
3477 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3479 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3481 pref0_len = pref_len - (locinput - loc_reg_starttry);
3482 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3483 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3484 ? (5 + taill) - pref_len : loc_regeol - locinput);
3485 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3489 if (pref0_len > pref_len)
3490 pref0_len = pref_len;
3492 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3494 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3495 (locinput - pref_len),pref0_len, 60, 4, 5);
3497 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3498 (locinput - pref_len + pref0_len),
3499 pref_len - pref0_len, 60, 2, 3);
3501 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3502 locinput, loc_regeol - locinput, 10, 0, 1);
3504 const STRLEN tlen=len0+len1+len2;
3505 PerlIO_printf(Perl_debug_log,
3506 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3507 (IV)(locinput - loc_bostr),
3510 (docolor ? "" : "> <"),
3512 (int)(tlen > 19 ? 0 : 19 - tlen),
3519 /* reg_check_named_buff_matched()
3520 * Checks to see if a named buffer has matched. The data array of
3521 * buffer numbers corresponding to the buffer is expected to reside
3522 * in the regexp->data->data array in the slot stored in the ARG() of
3523 * node involved. Note that this routine doesn't actually care about the
3524 * name, that information is not preserved from compilation to execution.
3525 * Returns the index of the leftmost defined buffer with the given name
3526 * or 0 if non of the buffers matched.
3529 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3532 RXi_GET_DECL(rex,rexi);
3533 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3534 I32 *nums=(I32*)SvPVX(sv_dat);
3536 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3538 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3539 if ((I32)rex->lastparen >= nums[n] &&
3540 rex->offs[nums[n]].end != -1)
3550 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3551 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3553 /* This function determines if there are one or two characters that match
3554 * the first character of the passed-in EXACTish node <text_node>, and if
3555 * so, returns them in the passed-in pointers.
3557 * If it determines that no possible character in the target string can
3558 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3559 * the first character in <text_node> requires UTF-8 to represent, and the
3560 * target string isn't in UTF-8.)
3562 * If there are more than two characters that could match the beginning of
3563 * <text_node>, or if more context is required to determine a match or not,
3564 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3566 * The motiviation behind this function is to allow the caller to set up
3567 * tight loops for matching. If <text_node> is of type EXACT, there is
3568 * only one possible character that can match its first character, and so
3569 * the situation is quite simple. But things get much more complicated if
3570 * folding is involved. It may be that the first character of an EXACTFish
3571 * node doesn't participate in any possible fold, e.g., punctuation, so it
3572 * can be matched only by itself. The vast majority of characters that are
3573 * in folds match just two things, their lower and upper-case equivalents.
3574 * But not all are like that; some have multiple possible matches, or match
3575 * sequences of more than one character. This function sorts all that out.
3577 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3578 * loop of trying to match A*, we know we can't exit where the thing
3579 * following it isn't a B. And something can't be a B unless it is the
3580 * beginning of B. By putting a quick test for that beginning in a tight
3581 * loop, we can rule out things that can't possibly be B without having to
3582 * break out of the loop, thus avoiding work. Similarly, if A is a single
3583 * character, we can make a tight loop matching A*, using the outputs of
3586 * If the target string to match isn't in UTF-8, and there aren't
3587 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3588 * the one or two possible octets (which are characters in this situation)
3589 * that can match. In all cases, if there is only one character that can
3590 * match, *<c1p> and *<c2p> will be identical.
3592 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3593 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3594 * can match the beginning of <text_node>. They should be declared with at
3595 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3596 * undefined what these contain.) If one or both of the buffers are
3597 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3598 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3599 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3600 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3601 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3603 const bool utf8_target = reginfo->is_utf8_target;
3605 UV c1 = CHRTEST_NOT_A_CP_1;
3606 UV c2 = CHRTEST_NOT_A_CP_2;
3607 bool use_chrtest_void = FALSE;
3608 const bool is_utf8_pat = reginfo->is_utf8_pat;
3610 /* Used when we have both utf8 input and utf8 output, to avoid converting
3611 * to/from code points */
3612 bool utf8_has_been_setup = FALSE;
3616 U8 *pat = (U8*)STRING(text_node);
3617 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
3619 if (OP(text_node) == EXACT) {
3621 /* In an exact node, only one thing can be matched, that first
3622 * character. If both the pat and the target are UTF-8, we can just
3623 * copy the input to the output, avoiding finding the code point of
3628 else if (utf8_target) {
3629 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3630 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3631 utf8_has_been_setup = TRUE;
3634 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3637 else { /* an EXACTFish node */
3638 U8 *pat_end = pat + STR_LEN(text_node);
3640 /* An EXACTFL node has at least some characters unfolded, because what
3641 * they match is not known until now. So, now is the time to fold
3642 * the first few of them, as many as are needed to determine 'c1' and
3643 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
3644 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
3645 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
3646 * need to fold as many characters as a single character can fold to,
3647 * so that later we can check if the first ones are such a multi-char
3648 * fold. But, in such a pattern only locale-problematic characters
3649 * aren't folded, so we can skip this completely if the first character
3650 * in the node isn't one of the tricky ones */
3651 if (OP(text_node) == EXACTFL) {
3653 if (! is_utf8_pat) {
3654 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
3656 folded[0] = folded[1] = 's';
3658 pat_end = folded + 2;
3661 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
3666 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
3668 *(d++) = (U8) toFOLD_LC(*s);
3673 _to_utf8_fold_flags(s,
3676 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
3687 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
3688 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
3690 /* Multi-character folds require more context to sort out. Also
3691 * PL_utf8_foldclosures used below doesn't handle them, so have to
3692 * be handled outside this routine */
3693 use_chrtest_void = TRUE;
3695 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3696 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3698 /* Load the folds hash, if not already done */
3700 if (! PL_utf8_foldclosures) {
3701 _load_PL_utf8_foldclosures();
3704 /* The fold closures data structure is a hash with the keys
3705 * being the UTF-8 of every character that is folded to, like
3706 * 'k', and the values each an array of all code points that
3707 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
3708 * Multi-character folds are not included */
3709 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3714 /* Not found in the hash, therefore there are no folds
3715 * containing it, so there is only a single character that
3719 else { /* Does participate in folds */
3720 AV* list = (AV*) *listp;
3721 if (av_tindex(list) != 1) {
3723 /* If there aren't exactly two folds to this, it is
3724 * outside the scope of this function */
3725 use_chrtest_void = TRUE;
3727 else { /* There are two. Get them */
3728 SV** c_p = av_fetch(list, 0, FALSE);
3730 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3734 c_p = av_fetch(list, 1, FALSE);
3736 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3740 /* Folds that cross the 255/256 boundary are forbidden
3741 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
3742 * one is ASCIII. Since the pattern character is above
3743 * 255, and its only other match is below 256, the only
3744 * legal match will be to itself. We have thrown away
3745 * the original, so have to compute which is the one
3747 if ((c1 < 256) != (c2 < 256)) {
3748 if ((OP(text_node) == EXACTFL
3749 && ! IN_UTF8_CTYPE_LOCALE)
3750 || ((OP(text_node) == EXACTFA
3751 || OP(text_node) == EXACTFA_NO_TRIE)
3752 && (isASCII(c1) || isASCII(c2))))
3765 else /* Here, c1 is <= 255 */
3767 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3768 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
3769 && ((OP(text_node) != EXACTFA
3770 && OP(text_node) != EXACTFA_NO_TRIE)
3773 /* Here, there could be something above Latin1 in the target
3774 * which folds to this character in the pattern. All such
3775 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
3776 * than two characters involved in their folds, so are outside
3777 * the scope of this function */
3778 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3779 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3782 use_chrtest_void = TRUE;
3785 else { /* Here nothing above Latin1 can fold to the pattern
3787 switch (OP(text_node)) {
3789 case EXACTFL: /* /l rules */
3790 c2 = PL_fold_locale[c1];
3793 case EXACTF: /* This node only generated for non-utf8
3795 assert(! is_utf8_pat);
3796 if (! utf8_target) { /* /d rules */
3801 /* /u rules for all these. This happens to work for
3802 * EXACTFA as nothing in Latin1 folds to ASCII */
3803 case EXACTFA_NO_TRIE: /* This node only generated for
3804 non-utf8 patterns */
3805 assert(! is_utf8_pat);
3810 c2 = PL_fold_latin1[c1];
3814 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3815 assert(0); /* NOTREACHED */
3821 /* Here have figured things out. Set up the returns */
3822 if (use_chrtest_void) {
3823 *c2p = *c1p = CHRTEST_VOID;
3825 else if (utf8_target) {
3826 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3827 uvchr_to_utf8(c1_utf8, c1);
3828 uvchr_to_utf8(c2_utf8, c2);
3831 /* Invariants are stored in both the utf8 and byte outputs; Use
3832 * negative numbers otherwise for the byte ones. Make sure that the
3833 * byte ones are the same iff the utf8 ones are the same */
3834 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3835 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3838 ? CHRTEST_NOT_A_CP_1
3839 : CHRTEST_NOT_A_CP_2;
3841 else if (c1 > 255) {
3842 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3847 *c1p = *c2p = c2; /* c2 is the only representable value */
3849 else { /* c1 is representable; see about c2 */
3851 *c2p = (c2 < 256) ? c2 : c1;
3857 /* returns -1 on failure, $+[0] on success */
3859 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3861 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3865 const bool utf8_target = reginfo->is_utf8_target;
3866 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3867 REGEXP *rex_sv = reginfo->prog;
3868 regexp *rex = ReANY(rex_sv);
3869 RXi_GET_DECL(rex,rexi);
3870 /* the current state. This is a cached copy of PL_regmatch_state */
3872 /* cache heavy used fields of st in registers */
3875 U32 n = 0; /* general value; init to avoid compiler warning */
3876 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
3877 char *locinput = startpos;
3878 char *pushinput; /* where to continue after a PUSH */
3879 I32 nextchr; /* is always set to UCHARAT(locinput) */
3881 bool result = 0; /* return value of S_regmatch */
3882 int depth = 0; /* depth of backtrack stack */
3883 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3884 const U32 max_nochange_depth =
3885 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3886 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3887 regmatch_state *yes_state = NULL; /* state to pop to on success of
3889 /* mark_state piggy backs on the yes_state logic so that when we unwind
3890 the stack on success we can update the mark_state as we go */
3891 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3892 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3893 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3895 bool no_final = 0; /* prevent failure from backtracking? */
3896 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3897 char *startpoint = locinput;
3898 SV *popmark = NULL; /* are we looking for a mark? */
3899 SV *sv_commit = NULL; /* last mark name seen in failure */
3900 SV *sv_yes_mark = NULL; /* last mark name we have seen
3901 during a successful match */
3902 U32 lastopen = 0; /* last open we saw */
3903 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3904 SV* const oreplsv = GvSVn(PL_replgv);
3905 /* these three flags are set by various ops to signal information to
3906 * the very next op. They have a useful lifetime of exactly one loop
3907 * iteration, and are not preserved or restored by state pushes/pops
3909 bool sw = 0; /* the condition value in (?(cond)a|b) */
3910 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3911 int logical = 0; /* the following EVAL is:
3915 or the following IFMATCH/UNLESSM is:
3916 false: plain (?=foo)
3917 true: used as a condition: (?(?=foo))
3919 PAD* last_pad = NULL;
3921 I32 gimme = G_SCALAR;
3922 CV *caller_cv = NULL; /* who called us */
3923 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3924 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3925 U32 maxopenparen = 0; /* max '(' index seen so far */
3926 int to_complement; /* Invert the result? */
3927 _char_class_number classnum;
3928 bool is_utf8_pat = reginfo->is_utf8_pat;
3931 GET_RE_DEBUG_FLAGS_DECL;
3934 /* protect against undef(*^R) */
3935 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
3937 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3938 multicall_oldcatch = 0;
3939 multicall_cv = NULL;
3941 PERL_UNUSED_VAR(multicall_cop);
3942 PERL_UNUSED_VAR(newsp);
3945 PERL_ARGS_ASSERT_REGMATCH;
3947 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3948 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3951 st = PL_regmatch_state;
3953 /* Note that nextchr is a byte even in UTF */
3956 while (scan != NULL) {
3959 SV * const prop = sv_newmortal();
3960 regnode *rnext=regnext(scan);
3961 DUMP_EXEC_POS( locinput, scan, utf8_target );
3962 regprop(rex, prop, scan, reginfo);
3964 PerlIO_printf(Perl_debug_log,
3965 "%3"IVdf":%*s%s(%"IVdf")\n",
3966 (IV)(scan - rexi->program), depth*2, "",
3968 (PL_regkind[OP(scan)] == END || !rnext) ?
3969 0 : (IV)(rnext - rexi->program));
3972 next = scan + NEXT_OFF(scan);
3975 state_num = OP(scan);
3981 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3983 switch (state_num) {
3984 case BOL: /* /^../ */
3985 case SBOL: /* /^../s */
3986 if (locinput == reginfo->strbeg)
3990 case MBOL: /* /^../m */
3991 if (locinput == reginfo->strbeg ||
3992 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3999 if (locinput == reginfo->ganch)
4003 case KEEPS: /* \K */
4004 /* update the startpoint */
4005 st->u.keeper.val = rex->offs[0].start;
4006 rex->offs[0].start = locinput - reginfo->strbeg;
4007 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
4008 assert(0); /*NOTREACHED*/
4009 case KEEPS_next_fail:
4010 /* rollback the start point change */
4011 rex->offs[0].start = st->u.keeper.val;
4013 assert(0); /*NOTREACHED*/
4015 case MEOL: /* /..$/m */
4016 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4020 case EOL: /* /..$/ */
4022 case SEOL: /* /..$/s */
4023 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4025 if (reginfo->strend - locinput > 1)
4030 if (!NEXTCHR_IS_EOS)
4034 case SANY: /* /./s */
4037 goto increment_locinput;
4045 case REG_ANY: /* /./ */
4046 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
4048 goto increment_locinput;
4052 #define ST st->u.trie
4053 case TRIEC: /* (ab|cd) with known charclass */
4054 /* In this case the charclass data is available inline so
4055 we can fail fast without a lot of extra overhead.
4057 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
4059 PerlIO_printf(Perl_debug_log,
4060 "%*s %sfailed to match trie start class...%s\n",
4061 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4064 assert(0); /* NOTREACHED */
4067 case TRIE: /* (ab|cd) */
4068 /* the basic plan of execution of the trie is:
4069 * At the beginning, run though all the states, and
4070 * find the longest-matching word. Also remember the position
4071 * of the shortest matching word. For example, this pattern:
4074 * when matched against the string "abcde", will generate
4075 * accept states for all words except 3, with the longest
4076 * matching word being 4, and the shortest being 2 (with
4077 * the position being after char 1 of the string).
4079 * Then for each matching word, in word order (i.e. 1,2,4,5),
4080 * we run the remainder of the pattern; on each try setting
4081 * the current position to the character following the word,
4082 * returning to try the next word on failure.
4084 * We avoid having to build a list of words at runtime by
4085 * using a compile-time structure, wordinfo[].prev, which
4086 * gives, for each word, the previous accepting word (if any).
4087 * In the case above it would contain the mappings 1->2, 2->0,
4088 * 3->0, 4->5, 5->1. We can use this table to generate, from
4089 * the longest word (4 above), a list of all words, by
4090 * following the list of prev pointers; this gives us the
4091 * unordered list 4,5,1,2. Then given the current word we have
4092 * just tried, we can go through the list and find the
4093 * next-biggest word to try (so if we just failed on word 2,
4094 * the next in the list is 4).
4096 * Since at runtime we don't record the matching position in
4097 * the string for each word, we have to work that out for
4098 * each word we're about to process. The wordinfo table holds
4099 * the character length of each word; given that we recorded
4100 * at the start: the position of the shortest word and its
4101 * length in chars, we just need to move the pointer the
4102 * difference between the two char lengths. Depending on
4103 * Unicode status and folding, that's cheap or expensive.
4105 * This algorithm is optimised for the case where are only a
4106 * small number of accept states, i.e. 0,1, or maybe 2.
4107 * With lots of accepts states, and having to try all of them,
4108 * it becomes quadratic on number of accept states to find all
4113 /* what type of TRIE am I? (utf8 makes this contextual) */
4114 DECL_TRIE_TYPE(scan);
4116 /* what trie are we using right now */
4117 reg_trie_data * const trie
4118 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
4119 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
4120 U32 state = trie->startstate;
4123 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
4125 if (trie->states[ state ].wordnum) {
4127 PerlIO_printf(Perl_debug_log,
4128 "%*s %smatched empty string...%s\n",
4129 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4135 PerlIO_printf(Perl_debug_log,
4136 "%*s %sfailed to match trie start class...%s\n",
4137 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4144 U8 *uc = ( U8* )locinput;
4148 U8 *uscan = (U8*)NULL;
4149 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
4150 U32 charcount = 0; /* how many input chars we have matched */
4151 U32 accepted = 0; /* have we seen any accepting states? */
4153 ST.jump = trie->jump;
4156 ST.longfold = FALSE; /* char longer if folded => it's harder */
4159 /* fully traverse the TRIE; note the position of the
4160 shortest accept state and the wordnum of the longest
4163 while ( state && uc <= (U8*)(reginfo->strend) ) {
4164 U32 base = trie->states[ state ].trans.base;
4168 wordnum = trie->states[ state ].wordnum;
4170 if (wordnum) { /* it's an accept state */
4173 /* record first match position */
4175 ST.firstpos = (U8*)locinput;
4180 ST.firstchars = charcount;
4183 if (!ST.nextword || wordnum < ST.nextword)
4184 ST.nextword = wordnum;
4185 ST.topword = wordnum;
4188 DEBUG_TRIE_EXECUTE_r({
4189 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
4190 PerlIO_printf( Perl_debug_log,
4191 "%*s %sState: %4"UVxf" Accepted: %c ",
4192 2+depth * 2, "", PL_colors[4],
4193 (UV)state, (accepted ? 'Y' : 'N'));
4196 /* read a char and goto next state */
4197 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
4199 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
4200 uscan, len, uvc, charid, foldlen,
4207 base + charid - 1 - trie->uniquecharcount)) >= 0)
4209 && ((U32)offset < trie->lasttrans)
4210 && trie->trans[offset].check == state)
4212 state = trie->trans[offset].next;
4223 DEBUG_TRIE_EXECUTE_r(
4224 PerlIO_printf( Perl_debug_log,
4225 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
4226 charid, uvc, (UV)state, PL_colors[5] );
4232 /* calculate total number of accept states */
4237 w = trie->wordinfo[w].prev;
4240 ST.accepted = accepted;
4244 PerlIO_printf( Perl_debug_log,
4245 "%*s %sgot %"IVdf" possible matches%s\n",
4246 REPORT_CODE_OFF + depth * 2, "",
4247 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
4249 goto trie_first_try; /* jump into the fail handler */
4251 assert(0); /* NOTREACHED */
4253 case TRIE_next_fail: /* we failed - try next alternative */
4257 REGCP_UNWIND(ST.cp);
4258 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
4260 if (!--ST.accepted) {
4262 PerlIO_printf( Perl_debug_log,
4263 "%*s %sTRIE failed...%s\n",
4264 REPORT_CODE_OFF+depth*2, "",
4271 /* Find next-highest word to process. Note that this code
4272 * is O(N^2) per trie run (O(N) per branch), so keep tight */
4275 U16 const nextword = ST.nextword;
4276 reg_trie_wordinfo * const wordinfo
4277 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
4278 for (word=ST.topword; word; word=wordinfo[word].prev) {
4279 if (word > nextword && (!min || word < min))
4292 ST.lastparen = rex->lastparen;
4293 ST.lastcloseparen = rex->lastcloseparen;
4297 /* find start char of end of current word */
4299 U32 chars; /* how many chars to skip */
4300 reg_trie_data * const trie
4301 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
4303 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
4305 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
4310 /* the hard option - fold each char in turn and find
4311 * its folded length (which may be different */
4312 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4320 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
4328 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4333 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
4349 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4350 ? ST.jump[ST.nextword]
4354 PerlIO_printf( Perl_debug_log,
4355 "%*s %sTRIE matched word #%d, continuing%s\n",
4356 REPORT_CODE_OFF+depth*2, "",
4363 if (ST.accepted > 1 || has_cutgroup) {
4364 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4365 assert(0); /* NOTREACHED */
4367 /* only one choice left - just continue */
4369 AV *const trie_words
4370 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4371 SV ** const tmp = av_fetch( trie_words,
4373 SV *sv= tmp ? sv_newmortal() : NULL;
4375 PerlIO_printf( Perl_debug_log,
4376 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4377 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4379 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4380 PL_colors[0], PL_colors[1],
4381 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4383 : "not compiled under -Dr",
4387 locinput = (char*)uc;
4388 continue; /* execute rest of RE */
4389 assert(0); /* NOTREACHED */
4393 case EXACT: { /* /abc/ */
4394 char *s = STRING(scan);
4396 if (utf8_target != is_utf8_pat) {
4397 /* The target and the pattern have differing utf8ness. */
4399 const char * const e = s + ln;
4402 /* The target is utf8, the pattern is not utf8.
4403 * Above-Latin1 code points can't match the pattern;
4404 * invariants match exactly, and the other Latin1 ones need
4405 * to be downgraded to a single byte in order to do the
4406 * comparison. (If we could be confident that the target
4407 * is not malformed, this could be refactored to have fewer
4408 * tests by just assuming that if the first bytes match, it
4409 * is an invariant, but there are tests in the test suite
4410 * dealing with (??{...}) which violate this) */
4412 if (l >= reginfo->strend
4413 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4417 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4424 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
4434 /* The target is not utf8, the pattern is utf8. */
4436 if (l >= reginfo->strend
4437 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4441 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4448 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
4460 /* The target and the pattern have the same utf8ness. */
4461 /* Inline the first character, for speed. */
4462 if (reginfo->strend - locinput < ln
4463 || UCHARAT(s) != nextchr
4464 || (ln > 1 && memNE(s, locinput, ln)))
4473 case EXACTFL: { /* /abc/il */
4475 const U8 * fold_array;
4477 U32 fold_utf8_flags;
4479 folder = foldEQ_locale;
4480 fold_array = PL_fold_locale;
4481 fold_utf8_flags = FOLDEQ_LOCALE;
4484 case EXACTFU_SS: /* /\x{df}/iu */
4485 case EXACTFU: /* /abc/iu */
4486 folder = foldEQ_latin1;
4487 fold_array = PL_fold_latin1;
4488 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4491 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
4493 assert(! is_utf8_pat);
4495 case EXACTFA: /* /abc/iaa */
4496 folder = foldEQ_latin1;
4497 fold_array = PL_fold_latin1;
4498 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4501 case EXACTF: /* /abc/i This node only generated for
4502 non-utf8 patterns */
4503 assert(! is_utf8_pat);
4505 fold_array = PL_fold;
4506 fold_utf8_flags = 0;
4514 || state_num == EXACTFU_SS
4515 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
4517 /* Either target or the pattern are utf8, or has the issue where
4518 * the fold lengths may differ. */
4519 const char * const l = locinput;
4520 char *e = reginfo->strend;
4522 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4523 l, &e, 0, utf8_target, fold_utf8_flags))
4531 /* Neither the target nor the pattern are utf8 */
4532 if (UCHARAT(s) != nextchr
4534 && UCHARAT(s) != fold_array[nextchr])
4538 if (reginfo->strend - locinput < ln)
4540 if (ln > 1 && ! folder(s, locinput, ln))
4546 /* XXX Could improve efficiency by separating these all out using a
4547 * macro or in-line function. At that point regcomp.c would no longer
4548 * have to set the FLAGS fields of these */
4549 case BOUNDL: /* /\b/l */
4550 case NBOUNDL: /* /\B/l */
4551 case BOUND: /* /\b/ */
4552 case BOUNDU: /* /\b/u */
4553 case BOUNDA: /* /\b/a */
4554 case NBOUND: /* /\B/ */
4555 case NBOUNDU: /* /\B/u */
4556 case NBOUNDA: /* /\B/a */
4557 /* was last char in word? */
4559 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4560 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4562 if (locinput == reginfo->strbeg)
4565 const U8 * const r =
4566 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4568 ln = utf8n_to_uvchr(r, (U8*) reginfo->strend - r,
4571 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4572 ln = isWORDCHAR_uni(ln);
4576 LOAD_UTF8_CHARCLASS_ALNUM();
4577 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4582 ln = isWORDCHAR_LC_uvchr(ln);
4583 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4588 /* Here the string isn't utf8, or is utf8 and only ascii
4589 * characters are to match \w. In the latter case looking at
4590 * the byte just prior to the current one may be just the final
4591 * byte of a multi-byte character. This is ok. There are two
4593 * 1) it is a single byte character, and then the test is doing
4594 * just what it's supposed to.
4595 * 2) it is a multi-byte character, in which case the final
4596 * byte is never mistakable for ASCII, and so the test
4597 * will say it is not a word character, which is the
4598 * correct answer. */
4599 ln = (locinput != reginfo->strbeg) ?
4600 UCHARAT(locinput - 1) : '\n';
4601 switch (FLAGS(scan)) {
4602 case REGEX_UNICODE_CHARSET:
4603 ln = isWORDCHAR_L1(ln);
4604 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4606 case REGEX_LOCALE_CHARSET:
4607 ln = isWORDCHAR_LC(ln);
4608 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4610 case REGEX_DEPENDS_CHARSET:
4611 ln = isWORDCHAR(ln);
4612 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4614 case REGEX_ASCII_RESTRICTED_CHARSET:
4615 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4616 ln = isWORDCHAR_A(ln);
4617 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4620 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4624 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4626 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4630 case ANYOF: /* /[abc]/ */
4634 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
4637 locinput += UTF8SKIP(locinput);
4640 if (!REGINCLASS(rex, scan, (U8*)locinput))
4646 /* The argument (FLAGS) to all the POSIX node types is the class number
4649 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4653 case POSIXL: /* \w or [:punct:] etc. under /l */
4657 /* Use isFOO_lc() for characters within Latin1. (Note that
4658 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4659 * wouldn't be invariant) */
4660 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4661 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4665 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4666 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4667 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4668 *(locinput + 1))))))
4673 else { /* Here, must be an above Latin-1 code point */
4674 goto utf8_posix_not_eos;
4677 /* Here, must be utf8 */
4678 locinput += UTF8SKIP(locinput);
4681 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4685 case POSIXD: /* \w or [:punct:] etc. under /d */
4691 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4693 if (NEXTCHR_IS_EOS) {
4697 /* All UTF-8 variants match */
4698 if (! UTF8_IS_INVARIANT(nextchr)) {
4699 goto increment_locinput;
4705 case POSIXA: /* \w or [:punct:] etc. under /a */
4708 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4709 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4710 * character is a single byte */
4713 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4719 /* Here we are either not in utf8, or we matched a utf8-invariant,
4720 * so the next char is the next byte */
4724 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4728 case POSIXU: /* \w or [:punct:] etc. under /u */
4730 if (NEXTCHR_IS_EOS) {
4735 /* Use _generic_isCC() for characters within Latin1. (Note that
4736 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4737 * wouldn't be invariant) */
4738 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4739 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4746 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4747 if (! (to_complement
4748 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4756 else { /* Handle above Latin-1 code points */
4757 classnum = (_char_class_number) FLAGS(scan);
4758 if (classnum < _FIRST_NON_SWASH_CC) {
4760 /* Here, uses a swash to find such code points. Load if if
4761 * not done already */
4762 if (! PL_utf8_swash_ptrs[classnum]) {
4763 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4764 PL_utf8_swash_ptrs[classnum]
4765 = _core_swash_init("utf8",
4768 PL_XPosix_ptrs[classnum], &flags);
4770 if (! (to_complement
4771 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4772 (U8 *) locinput, TRUE))))
4777 else { /* Here, uses macros to find above Latin-1 code points */
4779 case _CC_ENUM_SPACE: /* XXX would require separate
4780 code if we revert the change
4781 of \v matching this */
4782 case _CC_ENUM_PSXSPC:
4783 if (! (to_complement
4784 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4789 case _CC_ENUM_BLANK:
4790 if (! (to_complement
4791 ^ cBOOL(is_HORIZWS_high(locinput))))
4796 case _CC_ENUM_XDIGIT:
4797 if (! (to_complement
4798 ^ cBOOL(is_XDIGIT_high(locinput))))
4803 case _CC_ENUM_VERTSPACE:
4804 if (! (to_complement
4805 ^ cBOOL(is_VERTWS_high(locinput))))
4810 default: /* The rest, e.g. [:cntrl:], can't match
4812 if (! to_complement) {
4818 locinput += UTF8SKIP(locinput);
4822 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4823 a Unicode extended Grapheme Cluster */
4824 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4825 extended Grapheme Cluster is:
4828 | Prepend* Begin Extend*
4831 Begin is: ( Special_Begin | ! Control )
4832 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4833 Extend is: ( Grapheme_Extend | Spacing_Mark )
4834 Control is: [ GCB_Control | CR | LF ]
4835 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4837 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4840 Begin is ( Regular_Begin + Special Begin )
4842 It turns out that 98.4% of all Unicode code points match
4843 Regular_Begin. Doing it this way eliminates a table match in
4844 the previous implementation for almost all Unicode code points.
4846 There is a subtlety with Prepend* which showed up in testing.
4847 Note that the Begin, and only the Begin is required in:
4848 | Prepend* Begin Extend*
4849 Also, Begin contains '! Control'. A Prepend must be a
4850 '! Control', which means it must also be a Begin. What it
4851 comes down to is that if we match Prepend* and then find no
4852 suitable Begin afterwards, that if we backtrack the last
4853 Prepend, that one will be a suitable Begin.
4858 if (! utf8_target) {
4860 /* Match either CR LF or '.', as all the other possibilities
4862 locinput++; /* Match the . or CR */
4863 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4865 && locinput < reginfo->strend
4866 && UCHARAT(locinput) == '\n')
4873 /* Utf8: See if is ( CR LF ); already know that locinput <
4874 * reginfo->strend, so locinput+1 is in bounds */
4875 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4876 && UCHARAT(locinput + 1) == '\n')
4883 /* In case have to backtrack to beginning, then match '.' */
4884 char *starting = locinput;
4886 /* In case have to backtrack the last prepend */
4887 char *previous_prepend = NULL;
4889 LOAD_UTF8_CHARCLASS_GCB();
4891 /* Match (prepend)* */
4892 while (locinput < reginfo->strend
4893 && (len = is_GCB_Prepend_utf8(locinput)))
4895 previous_prepend = locinput;
4899 /* As noted above, if we matched a prepend character, but
4900 * the next thing won't match, back off the last prepend we
4901 * matched, as it is guaranteed to match the begin */
4902 if (previous_prepend
4903 && (locinput >= reginfo->strend
4904 || (! swash_fetch(PL_utf8_X_regular_begin,
4905 (U8*)locinput, utf8_target)
4906 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4909 locinput = previous_prepend;
4912 /* Note that here we know reginfo->strend > locinput, as we
4913 * tested that upon input to this switch case, and if we
4914 * moved locinput forward, we tested the result just above
4915 * and it either passed, or we backed off so that it will
4917 if (swash_fetch(PL_utf8_X_regular_begin,
4918 (U8*)locinput, utf8_target)) {
4919 locinput += UTF8SKIP(locinput);
4921 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4923 /* Here did not match the required 'Begin' in the
4924 * second term. So just match the very first
4925 * character, the '.' of the final term of the regex */
4926 locinput = starting + UTF8SKIP(starting);
4930 /* Here is a special begin. It can be composed of
4931 * several individual characters. One possibility is
4933 if ((len = is_GCB_RI_utf8(locinput))) {
4935 while (locinput < reginfo->strend
4936 && (len = is_GCB_RI_utf8(locinput)))
4940 } else if ((len = is_GCB_T_utf8(locinput))) {
4941 /* Another possibility is T+ */
4943 while (locinput < reginfo->strend
4944 && (len = is_GCB_T_utf8(locinput)))
4950 /* Here, neither RI+ nor T+; must be some other
4951 * Hangul. That means it is one of the others: L,
4952 * LV, LVT or V, and matches:
4953 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4956 while (locinput < reginfo->strend
4957 && (len = is_GCB_L_utf8(locinput)))
4962 /* Here, have exhausted L*. If the next character
4963 * is not an LV, LVT nor V, it means we had to have
4964 * at least one L, so matches L+ in the original
4965 * equation, we have a complete hangul syllable.
4968 if (locinput < reginfo->strend
4969 && is_GCB_LV_LVT_V_utf8(locinput))
4971 /* Otherwise keep going. Must be LV, LVT or V.
4972 * See if LVT, by first ruling out V, then LV */
4973 if (! is_GCB_V_utf8(locinput)
4974 /* All but every TCount one is LV */
4975 && (valid_utf8_to_uvchr((U8 *) locinput,
4980 locinput += UTF8SKIP(locinput);
4983 /* Must be V or LV. Take it, then match
4985 locinput += UTF8SKIP(locinput);
4986 while (locinput < reginfo->strend
4987 && (len = is_GCB_V_utf8(locinput)))
4993 /* And any of LV, LVT, or V can be followed
4995 while (locinput < reginfo->strend
4996 && (len = is_GCB_T_utf8(locinput)))
5004 /* Match any extender */
5005 while (locinput < reginfo->strend
5006 && swash_fetch(PL_utf8_X_extend,
5007 (U8*)locinput, utf8_target))
5009 locinput += UTF8SKIP(locinput);
5013 if (locinput > reginfo->strend) sayNO;
5017 case NREFFL: /* /\g{name}/il */
5018 { /* The capture buffer cases. The ones beginning with N for the
5019 named buffers just convert to the equivalent numbered and
5020 pretend they were called as the corresponding numbered buffer
5022 /* don't initialize these in the declaration, it makes C++
5027 const U8 *fold_array;
5030 folder = foldEQ_locale;
5031 fold_array = PL_fold_locale;
5033 utf8_fold_flags = FOLDEQ_LOCALE;
5036 case NREFFA: /* /\g{name}/iaa */
5037 folder = foldEQ_latin1;
5038 fold_array = PL_fold_latin1;
5040 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5043 case NREFFU: /* /\g{name}/iu */
5044 folder = foldEQ_latin1;
5045 fold_array = PL_fold_latin1;
5047 utf8_fold_flags = 0;
5050 case NREFF: /* /\g{name}/i */
5052 fold_array = PL_fold;
5054 utf8_fold_flags = 0;
5057 case NREF: /* /\g{name}/ */
5061 utf8_fold_flags = 0;
5064 /* For the named back references, find the corresponding buffer
5066 n = reg_check_named_buff_matched(rex,scan);
5071 goto do_nref_ref_common;
5073 case REFFL: /* /\1/il */
5074 folder = foldEQ_locale;
5075 fold_array = PL_fold_locale;
5076 utf8_fold_flags = FOLDEQ_LOCALE;
5079 case REFFA: /* /\1/iaa */
5080 folder = foldEQ_latin1;
5081 fold_array = PL_fold_latin1;
5082 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5085 case REFFU: /* /\1/iu */
5086 folder = foldEQ_latin1;
5087 fold_array = PL_fold_latin1;
5088 utf8_fold_flags = 0;
5091 case REFF: /* /\1/i */
5093 fold_array = PL_fold;
5094 utf8_fold_flags = 0;
5097 case REF: /* /\1/ */
5100 utf8_fold_flags = 0;
5104 n = ARG(scan); /* which paren pair */
5107 ln = rex->offs[n].start;
5108 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5109 if (rex->lastparen < n || ln == -1)
5110 sayNO; /* Do not match unless seen CLOSEn. */
5111 if (ln == rex->offs[n].end)
5114 s = reginfo->strbeg + ln;
5115 if (type != REF /* REF can do byte comparison */
5116 && (utf8_target || type == REFFU || type == REFFL))
5118 char * limit = reginfo->strend;
5120 /* This call case insensitively compares the entire buffer
5121 * at s, with the current input starting at locinput, but
5122 * not going off the end given by reginfo->strend, and
5123 * returns in <limit> upon success, how much of the
5124 * current input was matched */
5125 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
5126 locinput, &limit, 0, utf8_target, utf8_fold_flags))
5134 /* Not utf8: Inline the first character, for speed. */
5135 if (!NEXTCHR_IS_EOS &&
5136 UCHARAT(s) != nextchr &&
5138 UCHARAT(s) != fold_array[nextchr]))
5140 ln = rex->offs[n].end - ln;
5141 if (locinput + ln > reginfo->strend)
5143 if (ln > 1 && (type == REF
5144 ? memNE(s, locinput, ln)
5145 : ! folder(s, locinput, ln)))
5151 case NOTHING: /* null op; e.g. the 'nothing' following
5152 * the '*' in m{(a+|b)*}' */
5154 case TAIL: /* placeholder while compiling (A|B|C) */
5157 case BACK: /* ??? doesn't appear to be used ??? */
5161 #define ST st->u.eval
5166 regexp_internal *rei;
5167 regnode *startpoint;
5169 case GOSTART: /* (?R) */
5170 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
5171 if (cur_eval && cur_eval->locinput==locinput) {
5172 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
5173 Perl_croak(aTHX_ "Infinite recursion in regex");
5174 if ( ++nochange_depth > max_nochange_depth )
5176 "Pattern subroutine nesting without pos change"
5177 " exceeded limit in regex");
5184 if (OP(scan)==GOSUB) {
5185 startpoint = scan + ARG2L(scan);
5186 ST.close_paren = ARG(scan);
5188 startpoint = rei->program+1;
5192 /* Save all the positions seen so far. */
5193 ST.cp = regcppush(rex, 0, maxopenparen);
5194 REGCP_SET(ST.lastcp);
5196 /* and then jump to the code we share with EVAL */
5197 goto eval_recurse_doit;
5199 assert(0); /* NOTREACHED */
5201 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
5202 if (cur_eval && cur_eval->locinput==locinput) {
5203 if ( ++nochange_depth > max_nochange_depth )
5204 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
5209 /* execute the code in the {...} */
5213 OP * const oop = PL_op;
5214 COP * const ocurcop = PL_curcop;
5218 /* save *all* paren positions */
5219 regcppush(rex, 0, maxopenparen);
5220 REGCP_SET(runops_cp);
5223 caller_cv = find_runcv(NULL);
5227 if (rexi->data->what[n] == 'r') { /* code from an external qr */
5229 (REGEXP*)(rexi->data->data[n])
5232 nop = (OP*)rexi->data->data[n+1];
5234 else if (rexi->data->what[n] == 'l') { /* literal code */
5236 nop = (OP*)rexi->data->data[n];
5237 assert(CvDEPTH(newcv));
5240 /* literal with own CV */
5241 assert(rexi->data->what[n] == 'L');
5242 newcv = rex->qr_anoncv;
5243 nop = (OP*)rexi->data->data[n];
5246 /* normally if we're about to execute code from the same
5247 * CV that we used previously, we just use the existing
5248 * CX stack entry. However, its possible that in the
5249 * meantime we may have backtracked, popped from the save
5250 * stack, and undone the SAVECOMPPAD(s) associated with
5251 * PUSH_MULTICALL; in which case PL_comppad no longer
5252 * points to newcv's pad. */
5253 if (newcv != last_pushed_cv || PL_comppad != last_pad)
5255 U8 flags = (CXp_SUB_RE |
5256 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
5257 if (last_pushed_cv) {
5258 CHANGE_MULTICALL_FLAGS(newcv, flags);
5261 PUSH_MULTICALL_FLAGS(newcv, flags);
5263 last_pushed_cv = newcv;
5266 /* these assignments are just to silence compiler
5268 multicall_cop = NULL;
5271 last_pad = PL_comppad;
5273 /* the initial nextstate you would normally execute
5274 * at the start of an eval (which would cause error
5275 * messages to come from the eval), may be optimised
5276 * away from the execution path in the regex code blocks;
5277 * so manually set PL_curcop to it initially */
5279 OP *o = cUNOPx(nop)->op_first;
5280 assert(o->op_type == OP_NULL);
5281 if (o->op_targ == OP_SCOPE) {
5282 o = cUNOPo->op_first;
5285 assert(o->op_targ == OP_LEAVE);
5286 o = cUNOPo->op_first;
5287 assert(o->op_type == OP_ENTER);
5291 if (o->op_type != OP_STUB) {
5292 assert( o->op_type == OP_NEXTSTATE
5293 || o->op_type == OP_DBSTATE
5294 || (o->op_type == OP_NULL
5295 && ( o->op_targ == OP_NEXTSTATE
5296 || o->op_targ == OP_DBSTATE
5300 PL_curcop = (COP*)o;
5305 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
5306 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
5308 rex->offs[0].end = locinput - reginfo->strbeg;
5309 if (reginfo->info_aux_eval->pos_magic)
5310 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
5311 reginfo->sv, reginfo->strbeg,
5312 locinput - reginfo->strbeg);
5315 SV *sv_mrk = get_sv("REGMARK", 1);
5316 sv_setsv(sv_mrk, sv_yes_mark);
5319 /* we don't use MULTICALL here as we want to call the
5320 * first op of the block of interest, rather than the
5321 * first op of the sub */
5322 before = (IV)(SP-PL_stack_base);
5324 CALLRUNOPS(aTHX); /* Scalar context. */
5326 if ((IV)(SP-PL_stack_base) == before)
5327 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
5333 /* before restoring everything, evaluate the returned
5334 * value, so that 'uninit' warnings don't use the wrong
5335 * PL_op or pad. Also need to process any magic vars
5336 * (e.g. $1) *before* parentheses are restored */
5341 if (logical == 0) /* (?{})/ */
5342 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5343 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5344 sw = cBOOL(SvTRUE(ret));
5347 else { /* /(??{}) */
5348 /* if its overloaded, let the regex compiler handle
5349 * it; otherwise extract regex, or stringify */
5350 if (SvGMAGICAL(ret))
5351 ret = sv_mortalcopy(ret);
5352 if (!SvAMAGIC(ret)) {
5356 if (SvTYPE(sv) == SVt_REGEXP)
5357 re_sv = (REGEXP*) sv;
5358 else if (SvSMAGICAL(ret)) {
5359 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
5361 re_sv = (REGEXP *) mg->mg_obj;
5364 /* force any undef warnings here */
5365 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
5366 ret = sv_mortalcopy(ret);
5367 (void) SvPV_force_nolen(ret);
5373 /* *** Note that at this point we don't restore
5374 * PL_comppad, (or pop the CxSUB) on the assumption it may
5375 * be used again soon. This is safe as long as nothing
5376 * in the regexp code uses the pad ! */
5378 PL_curcop = ocurcop;
5379 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5380 PL_curpm = PL_reg_curpm;
5386 /* only /(??{})/ from now on */
5389 /* extract RE object from returned value; compiling if
5393 re_sv = reg_temp_copy(NULL, re_sv);
5398 if (SvUTF8(ret) && IN_BYTES) {
5399 /* In use 'bytes': make a copy of the octet
5400 * sequence, but without the flag on */
5402 const char *const p = SvPV(ret, len);
5403 ret = newSVpvn_flags(p, len, SVs_TEMP);
5405 if (rex->intflags & PREGf_USE_RE_EVAL)
5406 pm_flags |= PMf_USE_RE_EVAL;
5408 /* if we got here, it should be an engine which
5409 * supports compiling code blocks and stuff */
5410 assert(rex->engine && rex->engine->op_comp);
5411 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5412 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5413 rex->engine, NULL, NULL,
5414 /* copy /msix etc to inner pattern */
5419 & (SVs_TEMP | SVs_GMG | SVf_ROK))
5420 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
5421 /* This isn't a first class regexp. Instead, it's
5422 caching a regexp onto an existing, Perl visible
5424 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5430 RXp_MATCH_COPIED_off(re);
5431 re->subbeg = rex->subbeg;
5432 re->sublen = rex->sublen;
5433 re->suboffset = rex->suboffset;
5434 re->subcoffset = rex->subcoffset;
5436 re->lastcloseparen = 0;
5439 debug_start_match(re_sv, utf8_target, locinput,
5440 reginfo->strend, "Matching embedded");
5442 startpoint = rei->program + 1;
5443 ST.close_paren = 0; /* only used for GOSUB */
5444 /* Save all the seen positions so far. */
5445 ST.cp = regcppush(rex, 0, maxopenparen);
5446 REGCP_SET(ST.lastcp);
5447 /* and set maxopenparen to 0, since we are starting a "fresh" match */
5449 /* run the pattern returned from (??{...}) */
5451 eval_recurse_doit: /* Share code with GOSUB below this line
5452 * At this point we expect the stack context to be
5453 * set up correctly */
5455 /* invalidate the S-L poscache. We're now executing a
5456 * different set of WHILEM ops (and their associated
5457 * indexes) against the same string, so the bits in the
5458 * cache are meaningless. Setting maxiter to zero forces
5459 * the cache to be invalidated and zeroed before reuse.
5460 * XXX This is too dramatic a measure. Ideally we should
5461 * save the old cache and restore when running the outer
5463 reginfo->poscache_maxiter = 0;
5465 /* the new regexp might have a different is_utf8_pat than we do */
5466 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5468 ST.prev_rex = rex_sv;
5469 ST.prev_curlyx = cur_curlyx;
5471 SET_reg_curpm(rex_sv);
5476 ST.prev_eval = cur_eval;
5478 /* now continue from first node in postoned RE */
5479 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5480 assert(0); /* NOTREACHED */
5483 case EVAL_AB: /* cleanup after a successful (??{A})B */
5484 /* note: this is called twice; first after popping B, then A */
5485 rex_sv = ST.prev_rex;
5486 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5487 SET_reg_curpm(rex_sv);
5488 rex = ReANY(rex_sv);
5489 rexi = RXi_GET(rex);
5491 /* preserve $^R across LEAVE's. See Bug 121070. */
5492 SV *save_sv= GvSV(PL_replgv);
5493 SvREFCNT_inc(save_sv);
5494 regcpblow(ST.cp); /* LEAVE in disguise */
5495 sv_setsv(GvSV(PL_replgv), save_sv);
5496 SvREFCNT_dec(save_sv);
5498 cur_eval = ST.prev_eval;
5499 cur_curlyx = ST.prev_curlyx;
5501 /* Invalidate cache. See "invalidate" comment above. */
5502 reginfo->poscache_maxiter = 0;
5503 if ( nochange_depth )
5508 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5509 /* note: this is called twice; first after popping B, then A */
5510 rex_sv = ST.prev_rex;
5511 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5512 SET_reg_curpm(rex_sv);
5513 rex = ReANY(rex_sv);
5514 rexi = RXi_GET(rex);
5516 REGCP_UNWIND(ST.lastcp);
5517 regcppop(rex, &maxopenparen);
5518 cur_eval = ST.prev_eval;
5519 cur_curlyx = ST.prev_curlyx;
5520 /* Invalidate cache. See "invalidate" comment above. */
5521 reginfo->poscache_maxiter = 0;
5522 if ( nochange_depth )
5528 n = ARG(scan); /* which paren pair */
5529 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5530 if (n > maxopenparen)
5532 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5533 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5537 (IV)rex->offs[n].start_tmp,
5543 /* XXX really need to log other places start/end are set too */
5544 #define CLOSE_CAPTURE \
5545 rex->offs[n].start = rex->offs[n].start_tmp; \
5546 rex->offs[n].end = locinput - reginfo->strbeg; \
5547 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5548 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5550 PTR2UV(rex->offs), \
5552 (IV)rex->offs[n].start, \
5553 (IV)rex->offs[n].end \
5557 n = ARG(scan); /* which paren pair */
5559 if (n > rex->lastparen)
5561 rex->lastcloseparen = n;
5562 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5567 case ACCEPT: /* (*ACCEPT) */
5571 cursor && OP(cursor)!=END;
5572 cursor=regnext(cursor))
5574 if ( OP(cursor)==CLOSE ){
5576 if ( n <= lastopen ) {
5578 if (n > rex->lastparen)
5580 rex->lastcloseparen = n;
5581 if ( n == ARG(scan) || (cur_eval &&
5582 cur_eval->u.eval.close_paren == n))
5591 case GROUPP: /* (?(1)) */
5592 n = ARG(scan); /* which paren pair */
5593 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5596 case NGROUPP: /* (?(<name>)) */
5597 /* reg_check_named_buff_matched returns 0 for no match */
5598 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5601 case INSUBP: /* (?(R)) */
5603 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5606 case DEFINEP: /* (?(DEFINE)) */
5610 case IFTHEN: /* (?(cond)A|B) */
5611 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5613 next = NEXTOPER(NEXTOPER(scan));
5615 next = scan + ARG(scan);
5616 if (OP(next) == IFTHEN) /* Fake one. */
5617 next = NEXTOPER(NEXTOPER(next));
5621 case LOGICAL: /* modifier for EVAL and IFMATCH */
5622 logical = scan->flags;
5625 /*******************************************************************
5627 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5628 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5629 STAR/PLUS/CURLY/CURLYN are used instead.)
5631 A*B is compiled as <CURLYX><A><WHILEM><B>
5633 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5634 state, which contains the current count, initialised to -1. It also sets
5635 cur_curlyx to point to this state, with any previous value saved in the
5638 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5639 since the pattern may possibly match zero times (i.e. it's a while {} loop
5640 rather than a do {} while loop).
5642 Each entry to WHILEM represents a successful match of A. The count in the
5643 CURLYX block is incremented, another WHILEM state is pushed, and execution
5644 passes to A or B depending on greediness and the current count.
5646 For example, if matching against the string a1a2a3b (where the aN are
5647 substrings that match /A/), then the match progresses as follows: (the
5648 pushed states are interspersed with the bits of strings matched so far):
5651 <CURLYX cnt=0><WHILEM>
5652 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5653 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5654 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5655 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5657 (Contrast this with something like CURLYM, which maintains only a single
5661 a1 <CURLYM cnt=1> a2
5662 a1 a2 <CURLYM cnt=2> a3
5663 a1 a2 a3 <CURLYM cnt=3> b
5666 Each WHILEM state block marks a point to backtrack to upon partial failure
5667 of A or B, and also contains some minor state data related to that
5668 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5669 overall state, such as the count, and pointers to the A and B ops.
5671 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5672 must always point to the *current* CURLYX block, the rules are:
5674 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5675 and set cur_curlyx to point the new block.
5677 When popping the CURLYX block after a successful or unsuccessful match,
5678 restore the previous cur_curlyx.
5680 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5681 to the outer one saved in the CURLYX block.
5683 When popping the WHILEM block after a successful or unsuccessful B match,
5684 restore the previous cur_curlyx.
5686 Here's an example for the pattern (AI* BI)*BO
5687 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5690 curlyx backtrack stack
5691 ------ ---------------
5693 CO <CO prev=NULL> <WO>
5694 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5695 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5696 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5698 At this point the pattern succeeds, and we work back down the stack to
5699 clean up, restoring as we go:
5701 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5702 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5703 CO <CO prev=NULL> <WO>
5706 *******************************************************************/
5708 #define ST st->u.curlyx
5710 case CURLYX: /* start of /A*B/ (for complex A) */
5712 /* No need to save/restore up to this paren */
5713 I32 parenfloor = scan->flags;
5715 assert(next); /* keep Coverity happy */
5716 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5719 /* XXXX Probably it is better to teach regpush to support
5720 parenfloor > maxopenparen ... */
5721 if (parenfloor > (I32)rex->lastparen)
5722 parenfloor = rex->lastparen; /* Pessimization... */
5724 ST.prev_curlyx= cur_curlyx;
5726 ST.cp = PL_savestack_ix;
5728 /* these fields contain the state of the current curly.
5729 * they are accessed by subsequent WHILEMs */
5730 ST.parenfloor = parenfloor;
5735 ST.count = -1; /* this will be updated by WHILEM */
5736 ST.lastloc = NULL; /* this will be updated by WHILEM */
5738 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5739 assert(0); /* NOTREACHED */
5742 case CURLYX_end: /* just finished matching all of A*B */
5743 cur_curlyx = ST.prev_curlyx;
5745 assert(0); /* NOTREACHED */
5747 case CURLYX_end_fail: /* just failed to match all of A*B */
5749 cur_curlyx = ST.prev_curlyx;
5751 assert(0); /* NOTREACHED */
5755 #define ST st->u.whilem
5757 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5759 /* see the discussion above about CURLYX/WHILEM */
5761 int min = ARG1(cur_curlyx->u.curlyx.me);
5762 int max = ARG2(cur_curlyx->u.curlyx.me);
5763 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5765 assert(cur_curlyx); /* keep Coverity happy */
5766 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5767 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5768 ST.cache_offset = 0;
5772 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5773 "%*s whilem: matched %ld out of %d..%d\n",
5774 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5777 /* First just match a string of min A's. */
5780 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5782 cur_curlyx->u.curlyx.lastloc = locinput;
5783 REGCP_SET(ST.lastcp);
5785 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5786 assert(0); /* NOTREACHED */
5789 /* If degenerate A matches "", assume A done. */
5791 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5792 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5793 "%*s whilem: empty match detected, trying continuation...\n",
5794 REPORT_CODE_OFF+depth*2, "")
5796 goto do_whilem_B_max;
5799 /* super-linear cache processing.
5801 * The idea here is that for certain types of CURLYX/WHILEM -
5802 * principally those whose upper bound is infinity (and
5803 * excluding regexes that have things like \1 and other very
5804 * non-regular expresssiony things), then if a pattern like
5805 * /....A*.../ fails and we backtrack to the WHILEM, then we
5806 * make a note that this particular WHILEM op was at string
5807 * position 47 (say) when the rest of pattern failed. Then, if
5808 * we ever find ourselves back at that WHILEM, and at string
5809 * position 47 again, we can just fail immediately rather than
5810 * running the rest of the pattern again.
5812 * This is very handy when patterns start to go
5813 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5814 * with a combinatorial explosion of backtracking.
5816 * The cache is implemented as a bit array, with one bit per
5817 * string byte position per WHILEM op (up to 16) - so its
5818 * between 0.25 and 2x the string size.
5820 * To avoid allocating a poscache buffer every time, we do an
5821 * initially countdown; only after we have executed a WHILEM
5822 * op (string-length x #WHILEMs) times do we allocate the
5825 * The top 4 bits of scan->flags byte say how many different
5826 * relevant CURLLYX/WHILEM op pairs there are, while the
5827 * bottom 4-bits is the identifying index number of this
5833 if (!reginfo->poscache_maxiter) {
5834 /* start the countdown: Postpone detection until we
5835 * know the match is not *that* much linear. */
5836 reginfo->poscache_maxiter
5837 = (reginfo->strend - reginfo->strbeg + 1)
5839 /* possible overflow for long strings and many CURLYX's */
5840 if (reginfo->poscache_maxiter < 0)
5841 reginfo->poscache_maxiter = I32_MAX;
5842 reginfo->poscache_iter = reginfo->poscache_maxiter;
5845 if (reginfo->poscache_iter-- == 0) {
5846 /* initialise cache */
5847 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
5848 regmatch_info_aux *const aux = reginfo->info_aux;
5849 if (aux->poscache) {
5850 if ((SSize_t)reginfo->poscache_size < size) {
5851 Renew(aux->poscache, size, char);
5852 reginfo->poscache_size = size;
5854 Zero(aux->poscache, size, char);
5857 reginfo->poscache_size = size;
5858 Newxz(aux->poscache, size, char);
5860 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5861 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5862 PL_colors[4], PL_colors[5])
5866 if (reginfo->poscache_iter < 0) {
5867 /* have we already failed at this position? */
5868 SSize_t offset, mask;
5870 reginfo->poscache_iter = -1; /* stop eventual underflow */
5871 offset = (scan->flags & 0xf) - 1
5872 + (locinput - reginfo->strbeg)
5874 mask = 1 << (offset % 8);
5876 if (reginfo->info_aux->poscache[offset] & mask) {
5877 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5878 "%*s whilem: (cache) already tried at this position...\n",
5879 REPORT_CODE_OFF+depth*2, "")
5881 sayNO; /* cache records failure */
5883 ST.cache_offset = offset;
5884 ST.cache_mask = mask;
5888 /* Prefer B over A for minimal matching. */
5890 if (cur_curlyx->u.curlyx.minmod) {
5891 ST.save_curlyx = cur_curlyx;
5892 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5893 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5895 REGCP_SET(ST.lastcp);
5896 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5898 assert(0); /* NOTREACHED */
5901 /* Prefer A over B for maximal matching. */
5903 if (n < max) { /* More greed allowed? */
5904 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5906 cur_curlyx->u.curlyx.lastloc = locinput;
5907 REGCP_SET(ST.lastcp);
5908 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5909 assert(0); /* NOTREACHED */
5911 goto do_whilem_B_max;
5913 assert(0); /* NOTREACHED */
5915 case WHILEM_B_min: /* just matched B in a minimal match */
5916 case WHILEM_B_max: /* just matched B in a maximal match */
5917 cur_curlyx = ST.save_curlyx;
5919 assert(0); /* NOTREACHED */
5921 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5922 cur_curlyx = ST.save_curlyx;
5923 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5924 cur_curlyx->u.curlyx.count--;
5926 assert(0); /* NOTREACHED */
5928 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5930 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5931 REGCP_UNWIND(ST.lastcp);
5932 regcppop(rex, &maxopenparen);
5933 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5934 cur_curlyx->u.curlyx.count--;
5936 assert(0); /* NOTREACHED */
5938 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5939 REGCP_UNWIND(ST.lastcp);
5940 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5941 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5942 "%*s whilem: failed, trying continuation...\n",
5943 REPORT_CODE_OFF+depth*2, "")
5946 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5947 && ckWARN(WARN_REGEXP)
5948 && !reginfo->warned)
5950 reginfo->warned = TRUE;
5951 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5952 "Complex regular subexpression recursion limit (%d) "
5958 ST.save_curlyx = cur_curlyx;
5959 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5960 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5962 assert(0); /* NOTREACHED */
5964 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5965 cur_curlyx = ST.save_curlyx;
5966 REGCP_UNWIND(ST.lastcp);
5967 regcppop(rex, &maxopenparen);
5969 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5970 /* Maximum greed exceeded */
5971 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5972 && ckWARN(WARN_REGEXP)
5973 && !reginfo->warned)
5975 reginfo->warned = TRUE;
5976 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5977 "Complex regular subexpression recursion "
5978 "limit (%d) exceeded",
5981 cur_curlyx->u.curlyx.count--;
5985 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5986 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5988 /* Try grabbing another A and see if it helps. */
5989 cur_curlyx->u.curlyx.lastloc = locinput;
5990 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5992 REGCP_SET(ST.lastcp);
5993 PUSH_STATE_GOTO(WHILEM_A_min,
5994 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5996 assert(0); /* NOTREACHED */
5999 #define ST st->u.branch
6001 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
6002 next = scan + ARG(scan);
6005 scan = NEXTOPER(scan);
6008 case BRANCH: /* /(...|A|...)/ */
6009 scan = NEXTOPER(scan); /* scan now points to inner node */
6010 ST.lastparen = rex->lastparen;
6011 ST.lastcloseparen = rex->lastcloseparen;
6012 ST.next_branch = next;
6015 /* Now go into the branch */
6017 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
6019 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
6021 assert(0); /* NOTREACHED */
6023 case CUTGROUP: /* /(*THEN)/ */
6024 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
6025 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6026 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
6027 assert(0); /* NOTREACHED */
6029 case CUTGROUP_next_fail:
6032 if (st->u.mark.mark_name)
6033 sv_commit = st->u.mark.mark_name;
6035 assert(0); /* NOTREACHED */
6039 assert(0); /* NOTREACHED */
6041 case BRANCH_next_fail: /* that branch failed; try the next, if any */
6046 REGCP_UNWIND(ST.cp);
6047 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6048 scan = ST.next_branch;
6049 /* no more branches? */
6050 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
6052 PerlIO_printf( Perl_debug_log,
6053 "%*s %sBRANCH failed...%s\n",
6054 REPORT_CODE_OFF+depth*2, "",
6060 continue; /* execute next BRANCH[J] op */
6061 assert(0); /* NOTREACHED */
6063 case MINMOD: /* next op will be non-greedy, e.g. A*? */
6068 #define ST st->u.curlym
6070 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
6072 /* This is an optimisation of CURLYX that enables us to push
6073 * only a single backtracking state, no matter how many matches
6074 * there are in {m,n}. It relies on the pattern being constant
6075 * length, with no parens to influence future backrefs
6079 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6081 ST.lastparen = rex->lastparen;
6082 ST.lastcloseparen = rex->lastcloseparen;
6084 /* if paren positive, emulate an OPEN/CLOSE around A */
6086 U32 paren = ST.me->flags;
6087 if (paren > maxopenparen)
6088 maxopenparen = paren;
6089 scan += NEXT_OFF(scan); /* Skip former OPEN. */
6097 ST.c1 = CHRTEST_UNINIT;
6100 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
6103 curlym_do_A: /* execute the A in /A{m,n}B/ */
6104 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
6105 assert(0); /* NOTREACHED */
6107 case CURLYM_A: /* we've just matched an A */
6109 /* after first match, determine A's length: u.curlym.alen */
6110 if (ST.count == 1) {
6111 if (reginfo->is_utf8_target) {
6112 char *s = st->locinput;
6113 while (s < locinput) {
6119 ST.alen = locinput - st->locinput;
6122 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
6125 PerlIO_printf(Perl_debug_log,
6126 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
6127 (int)(REPORT_CODE_OFF+(depth*2)), "",
6128 (IV) ST.count, (IV)ST.alen)
6131 if (cur_eval && cur_eval->u.eval.close_paren &&
6132 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6136 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
6137 if ( max == REG_INFTY || ST.count < max )
6138 goto curlym_do_A; /* try to match another A */
6140 goto curlym_do_B; /* try to match B */
6142 case CURLYM_A_fail: /* just failed to match an A */
6143 REGCP_UNWIND(ST.cp);
6145 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
6146 || (cur_eval && cur_eval->u.eval.close_paren &&
6147 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
6150 curlym_do_B: /* execute the B in /A{m,n}B/ */
6151 if (ST.c1 == CHRTEST_UNINIT) {
6152 /* calculate c1 and c2 for possible match of 1st char
6153 * following curly */
6154 ST.c1 = ST.c2 = CHRTEST_VOID;
6155 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
6156 regnode *text_node = ST.B;
6157 if (! HAS_TEXT(text_node))
6158 FIND_NEXT_IMPT(text_node);
6161 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
6163 But the former is redundant in light of the latter.
6165 if this changes back then the macro for
6166 IS_TEXT and friends need to change.
6168 if (PL_regkind[OP(text_node)] == EXACT) {
6169 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6170 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6180 PerlIO_printf(Perl_debug_log,
6181 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
6182 (int)(REPORT_CODE_OFF+(depth*2)),
6185 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
6186 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
6187 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6188 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6190 /* simulate B failing */
6192 PerlIO_printf(Perl_debug_log,
6193 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
6194 (int)(REPORT_CODE_OFF+(depth*2)),"",
6195 valid_utf8_to_uvchr((U8 *) locinput, NULL),
6196 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
6197 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
6199 state_num = CURLYM_B_fail;
6200 goto reenter_switch;
6203 else if (nextchr != ST.c1 && nextchr != ST.c2) {
6204 /* simulate B failing */
6206 PerlIO_printf(Perl_debug_log,
6207 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
6208 (int)(REPORT_CODE_OFF+(depth*2)),"",
6209 (int) nextchr, ST.c1, ST.c2)
6211 state_num = CURLYM_B_fail;
6212 goto reenter_switch;
6217 /* emulate CLOSE: mark current A as captured */
6218 I32 paren = ST.me->flags;
6220 rex->offs[paren].start
6221 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
6222 rex->offs[paren].end = locinput - reginfo->strbeg;
6223 if ((U32)paren > rex->lastparen)
6224 rex->lastparen = paren;
6225 rex->lastcloseparen = paren;
6228 rex->offs[paren].end = -1;
6229 if (cur_eval && cur_eval->u.eval.close_paren &&
6230 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6239 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
6240 assert(0); /* NOTREACHED */
6242 case CURLYM_B_fail: /* just failed to match a B */
6243 REGCP_UNWIND(ST.cp);
6244 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6246 I32 max = ARG2(ST.me);
6247 if (max != REG_INFTY && ST.count == max)
6249 goto curlym_do_A; /* try to match a further A */
6251 /* backtrack one A */
6252 if (ST.count == ARG1(ST.me) /* min */)
6255 SET_locinput(HOPc(locinput, -ST.alen));
6256 goto curlym_do_B; /* try to match B */
6259 #define ST st->u.curly
6261 #define CURLY_SETPAREN(paren, success) \
6264 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
6265 rex->offs[paren].end = locinput - reginfo->strbeg; \
6266 if (paren > rex->lastparen) \
6267 rex->lastparen = paren; \
6268 rex->lastcloseparen = paren; \
6271 rex->offs[paren].end = -1; \
6272 rex->lastparen = ST.lastparen; \
6273 rex->lastcloseparen = ST.lastcloseparen; \
6277 case STAR: /* /A*B/ where A is width 1 char */
6281 scan = NEXTOPER(scan);
6284 case PLUS: /* /A+B/ where A is width 1 char */
6288 scan = NEXTOPER(scan);
6291 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
6292 ST.paren = scan->flags; /* Which paren to set */
6293 ST.lastparen = rex->lastparen;
6294 ST.lastcloseparen = rex->lastcloseparen;
6295 if (ST.paren > maxopenparen)
6296 maxopenparen = ST.paren;
6297 ST.min = ARG1(scan); /* min to match */
6298 ST.max = ARG2(scan); /* max to match */
6299 if (cur_eval && cur_eval->u.eval.close_paren &&
6300 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6304 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
6307 case CURLY: /* /A{m,n}B/ where A is width 1 char */
6309 ST.min = ARG1(scan); /* min to match */
6310 ST.max = ARG2(scan); /* max to match */
6311 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6314 * Lookahead to avoid useless match attempts
6315 * when we know what character comes next.
6317 * Used to only do .*x and .*?x, but now it allows
6318 * for )'s, ('s and (?{ ... })'s to be in the way
6319 * of the quantifier and the EXACT-like node. -- japhy
6322 assert(ST.min <= ST.max);
6323 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
6324 ST.c1 = ST.c2 = CHRTEST_VOID;
6327 regnode *text_node = next;
6329 if (! HAS_TEXT(text_node))
6330 FIND_NEXT_IMPT(text_node);
6332 if (! HAS_TEXT(text_node))
6333 ST.c1 = ST.c2 = CHRTEST_VOID;
6335 if ( PL_regkind[OP(text_node)] != EXACT ) {
6336 ST.c1 = ST.c2 = CHRTEST_VOID;
6340 /* Currently we only get here when
6342 PL_rekind[OP(text_node)] == EXACT
6344 if this changes back then the macro for IS_TEXT and
6345 friends need to change. */
6346 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6347 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6359 char *li = locinput;
6362 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
6368 if (ST.c1 == CHRTEST_VOID)
6369 goto curly_try_B_min;
6371 ST.oldloc = locinput;
6373 /* set ST.maxpos to the furthest point along the
6374 * string that could possibly match */
6375 if (ST.max == REG_INFTY) {
6376 ST.maxpos = reginfo->strend - 1;
6378 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
6381 else if (utf8_target) {
6382 int m = ST.max - ST.min;
6383 for (ST.maxpos = locinput;
6384 m >0 && ST.maxpos < reginfo->strend; m--)
6385 ST.maxpos += UTF8SKIP(ST.maxpos);
6388 ST.maxpos = locinput + ST.max - ST.min;
6389 if (ST.maxpos >= reginfo->strend)
6390 ST.maxpos = reginfo->strend - 1;
6392 goto curly_try_B_min_known;
6396 /* avoid taking address of locinput, so it can remain
6398 char *li = locinput;
6399 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6400 if (ST.count < ST.min)
6403 if ((ST.count > ST.min)
6404 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6406 /* A{m,n} must come at the end of the string, there's
6407 * no point in backing off ... */
6409 /* ...except that $ and \Z can match before *and* after
6410 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6411 We may back off by one in this case. */
6412 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6416 goto curly_try_B_max;
6418 assert(0); /* NOTREACHED */
6421 case CURLY_B_min_known_fail:
6422 /* failed to find B in a non-greedy match where c1,c2 valid */
6424 REGCP_UNWIND(ST.cp);
6426 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6428 /* Couldn't or didn't -- move forward. */
6429 ST.oldloc = locinput;
6431 locinput += UTF8SKIP(locinput);
6435 curly_try_B_min_known:
6436 /* find the next place where 'B' could work, then call B */
6440 n = (ST.oldloc == locinput) ? 0 : 1;
6441 if (ST.c1 == ST.c2) {
6442 /* set n to utf8_distance(oldloc, locinput) */
6443 while (locinput <= ST.maxpos
6444 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6446 locinput += UTF8SKIP(locinput);
6451 /* set n to utf8_distance(oldloc, locinput) */
6452 while (locinput <= ST.maxpos
6453 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6454 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6456 locinput += UTF8SKIP(locinput);
6461 else { /* Not utf8_target */
6462 if (ST.c1 == ST.c2) {
6463 while (locinput <= ST.maxpos &&
6464 UCHARAT(locinput) != ST.c1)
6468 while (locinput <= ST.maxpos
6469 && UCHARAT(locinput) != ST.c1
6470 && UCHARAT(locinput) != ST.c2)
6473 n = locinput - ST.oldloc;
6475 if (locinput > ST.maxpos)
6478 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6479 * at b; check that everything between oldloc and
6480 * locinput matches */
6481 char *li = ST.oldloc;
6483 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6485 assert(n == REG_INFTY || locinput == li);
6487 CURLY_SETPAREN(ST.paren, ST.count);
6488 if (cur_eval && cur_eval->u.eval.close_paren &&
6489 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6492 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6494 assert(0); /* NOTREACHED */
6497 case CURLY_B_min_fail:
6498 /* failed to find B in a non-greedy match where c1,c2 invalid */
6500 REGCP_UNWIND(ST.cp);
6502 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6504 /* failed -- move forward one */
6506 char *li = locinput;
6507 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6514 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6515 ST.count > 0)) /* count overflow ? */
6518 CURLY_SETPAREN(ST.paren, ST.count);
6519 if (cur_eval && cur_eval->u.eval.close_paren &&
6520 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6523 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6527 assert(0); /* NOTREACHED */
6531 /* a successful greedy match: now try to match B */
6532 if (cur_eval && cur_eval->u.eval.close_paren &&
6533 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6537 bool could_match = locinput < reginfo->strend;
6539 /* If it could work, try it. */
6540 if (ST.c1 != CHRTEST_VOID && could_match) {
6541 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6543 could_match = memEQ(locinput,
6548 UTF8SKIP(locinput));
6551 could_match = UCHARAT(locinput) == ST.c1
6552 || UCHARAT(locinput) == ST.c2;
6555 if (ST.c1 == CHRTEST_VOID || could_match) {
6556 CURLY_SETPAREN(ST.paren, ST.count);
6557 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6558 assert(0); /* NOTREACHED */
6563 case CURLY_B_max_fail:
6564 /* failed to find B in a greedy match */
6566 REGCP_UNWIND(ST.cp);
6568 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6571 if (--ST.count < ST.min)
6573 locinput = HOPc(locinput, -1);
6574 goto curly_try_B_max;
6578 case END: /* last op of main pattern */
6581 /* we've just finished A in /(??{A})B/; now continue with B */
6583 st->u.eval.prev_rex = rex_sv; /* inner */
6585 /* Save *all* the positions. */
6586 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6587 rex_sv = cur_eval->u.eval.prev_rex;
6588 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6589 SET_reg_curpm(rex_sv);
6590 rex = ReANY(rex_sv);
6591 rexi = RXi_GET(rex);
6592 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6594 REGCP_SET(st->u.eval.lastcp);
6596 /* Restore parens of the outer rex without popping the
6598 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6601 st->u.eval.prev_eval = cur_eval;
6602 cur_eval = cur_eval->u.eval.prev_eval;
6604 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6605 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6606 if ( nochange_depth )
6609 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6610 locinput); /* match B */
6613 if (locinput < reginfo->till) {
6614 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6615 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6617 (long)(locinput - startpos),
6618 (long)(reginfo->till - startpos),
6621 sayNO_SILENT; /* Cannot match: too short. */
6623 sayYES; /* Success! */
6625 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6627 PerlIO_printf(Perl_debug_log,
6628 "%*s %ssubpattern success...%s\n",
6629 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6630 sayYES; /* Success! */
6633 #define ST st->u.ifmatch
6638 case SUSPEND: /* (?>A) */
6640 newstart = locinput;
6643 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6645 goto ifmatch_trivial_fail_test;
6647 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6649 ifmatch_trivial_fail_test:
6651 char * const s = HOPBACKc(locinput, scan->flags);
6656 sw = 1 - cBOOL(ST.wanted);
6660 next = scan + ARG(scan);
6668 newstart = locinput;
6672 ST.logical = logical;
6673 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6675 /* execute body of (?...A) */
6676 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6677 assert(0); /* NOTREACHED */
6680 case IFMATCH_A_fail: /* body of (?...A) failed */
6681 ST.wanted = !ST.wanted;
6684 case IFMATCH_A: /* body of (?...A) succeeded */
6686 sw = cBOOL(ST.wanted);
6688 else if (!ST.wanted)
6691 if (OP(ST.me) != SUSPEND) {
6692 /* restore old position except for (?>...) */
6693 locinput = st->locinput;
6695 scan = ST.me + ARG(ST.me);
6698 continue; /* execute B */
6702 case LONGJMP: /* alternative with many branches compiles to
6703 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6704 next = scan + ARG(scan);
6709 case COMMIT: /* (*COMMIT) */
6710 reginfo->cutpoint = reginfo->strend;
6713 case PRUNE: /* (*PRUNE) */
6715 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6716 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6717 assert(0); /* NOTREACHED */
6719 case COMMIT_next_fail:
6723 case OPFAIL: /* (*FAIL) */
6725 assert(0); /* NOTREACHED */
6727 #define ST st->u.mark
6728 case MARKPOINT: /* (*MARK:foo) */
6729 ST.prev_mark = mark_state;
6730 ST.mark_name = sv_commit = sv_yes_mark
6731 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6733 ST.mark_loc = locinput;
6734 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6735 assert(0); /* NOTREACHED */
6737 case MARKPOINT_next:
6738 mark_state = ST.prev_mark;
6740 assert(0); /* NOTREACHED */
6742 case MARKPOINT_next_fail:
6743 if (popmark && sv_eq(ST.mark_name,popmark))
6745 if (ST.mark_loc > startpoint)
6746 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6747 popmark = NULL; /* we found our mark */
6748 sv_commit = ST.mark_name;
6751 PerlIO_printf(Perl_debug_log,
6752 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6753 REPORT_CODE_OFF+depth*2, "",
6754 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6757 mark_state = ST.prev_mark;
6758 sv_yes_mark = mark_state ?
6759 mark_state->u.mark.mark_name : NULL;
6761 assert(0); /* NOTREACHED */
6763 case SKIP: /* (*SKIP) */
6765 /* (*SKIP) : if we fail we cut here*/
6766 ST.mark_name = NULL;
6767 ST.mark_loc = locinput;
6768 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6770 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6771 otherwise do nothing. Meaning we need to scan
6773 regmatch_state *cur = mark_state;
6774 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6777 if ( sv_eq( cur->u.mark.mark_name,
6780 ST.mark_name = find;
6781 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6783 cur = cur->u.mark.prev_mark;
6786 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6789 case SKIP_next_fail:
6791 /* (*CUT:NAME) - Set up to search for the name as we
6792 collapse the stack*/
6793 popmark = ST.mark_name;
6795 /* (*CUT) - No name, we cut here.*/
6796 if (ST.mark_loc > startpoint)
6797 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6798 /* but we set sv_commit to latest mark_name if there
6799 is one so they can test to see how things lead to this
6802 sv_commit=mark_state->u.mark.mark_name;
6806 assert(0); /* NOTREACHED */
6809 case LNBREAK: /* \R */
6810 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6817 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6818 PTR2UV(scan), OP(scan));
6819 Perl_croak(aTHX_ "regexp memory corruption");
6821 /* this is a point to jump to in order to increment
6822 * locinput by one character */
6824 assert(!NEXTCHR_IS_EOS);
6826 locinput += PL_utf8skip[nextchr];
6827 /* locinput is allowed to go 1 char off the end, but not 2+ */
6828 if (locinput > reginfo->strend)
6837 /* switch break jumps here */
6838 scan = next; /* prepare to execute the next op and ... */
6839 continue; /* ... jump back to the top, reusing st */
6840 assert(0); /* NOTREACHED */
6843 /* push a state that backtracks on success */
6844 st->u.yes.prev_yes_state = yes_state;
6848 /* push a new regex state, then continue at scan */
6850 regmatch_state *newst;
6853 regmatch_state *cur = st;
6854 regmatch_state *curyes = yes_state;
6856 regmatch_slab *slab = PL_regmatch_slab;
6857 for (;curd > -1;cur--,curd--) {
6858 if (cur < SLAB_FIRST(slab)) {
6860 cur = SLAB_LAST(slab);
6862 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6863 REPORT_CODE_OFF + 2 + depth * 2,"",
6864 curd, PL_reg_name[cur->resume_state],
6865 (curyes == cur) ? "yes" : ""
6868 curyes = cur->u.yes.prev_yes_state;
6871 DEBUG_STATE_pp("push")
6874 st->locinput = locinput;
6876 if (newst > SLAB_LAST(PL_regmatch_slab))
6877 newst = S_push_slab(aTHX);
6878 PL_regmatch_state = newst;
6880 locinput = pushinput;
6883 assert(0); /* NOTREACHED */
6888 * We get here only if there's trouble -- normally "case END" is
6889 * the terminating point.
6891 Perl_croak(aTHX_ "corrupted regexp pointers");
6897 /* we have successfully completed a subexpression, but we must now
6898 * pop to the state marked by yes_state and continue from there */
6899 assert(st != yes_state);
6901 while (st != yes_state) {
6903 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6904 PL_regmatch_slab = PL_regmatch_slab->prev;
6905 st = SLAB_LAST(PL_regmatch_slab);
6909 DEBUG_STATE_pp("pop (no final)");
6911 DEBUG_STATE_pp("pop (yes)");
6917 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6918 || yes_state > SLAB_LAST(PL_regmatch_slab))
6920 /* not in this slab, pop slab */
6921 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6922 PL_regmatch_slab = PL_regmatch_slab->prev;
6923 st = SLAB_LAST(PL_regmatch_slab);
6925 depth -= (st - yes_state);
6928 yes_state = st->u.yes.prev_yes_state;
6929 PL_regmatch_state = st;
6932 locinput= st->locinput;
6933 state_num = st->resume_state + no_final;
6934 goto reenter_switch;
6937 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6938 PL_colors[4], PL_colors[5]));
6940 if (reginfo->info_aux_eval) {
6941 /* each successfully executed (?{...}) block does the equivalent of
6942 * local $^R = do {...}
6943 * When popping the save stack, all these locals would be undone;
6944 * bypass this by setting the outermost saved $^R to the latest
6946 /* I dont know if this is needed or works properly now.
6947 * see code related to PL_replgv elsewhere in this file.
6950 if (oreplsv != GvSV(PL_replgv))
6951 sv_setsv(oreplsv, GvSV(PL_replgv));
6958 PerlIO_printf(Perl_debug_log,
6959 "%*s %sfailed...%s\n",
6960 REPORT_CODE_OFF+depth*2, "",
6961 PL_colors[4], PL_colors[5])
6973 /* there's a previous state to backtrack to */
6975 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6976 PL_regmatch_slab = PL_regmatch_slab->prev;
6977 st = SLAB_LAST(PL_regmatch_slab);
6979 PL_regmatch_state = st;
6980 locinput= st->locinput;
6982 DEBUG_STATE_pp("pop");
6984 if (yes_state == st)
6985 yes_state = st->u.yes.prev_yes_state;
6987 state_num = st->resume_state + 1; /* failure = success + 1 */
6988 goto reenter_switch;
6993 if (rex->intflags & PREGf_VERBARG_SEEN) {
6994 SV *sv_err = get_sv("REGERROR", 1);
6995 SV *sv_mrk = get_sv("REGMARK", 1);
6997 sv_commit = &PL_sv_no;
6999 sv_yes_mark = &PL_sv_yes;
7002 sv_commit = &PL_sv_yes;
7003 sv_yes_mark = &PL_sv_no;
7007 sv_setsv(sv_err, sv_commit);
7008 sv_setsv(sv_mrk, sv_yes_mark);
7012 if (last_pushed_cv) {
7015 PERL_UNUSED_VAR(SP);
7018 assert(!result || locinput - reginfo->strbeg >= 0);
7019 return result ? locinput - reginfo->strbeg : -1;
7023 - regrepeat - repeatedly match something simple, report how many
7025 * What 'simple' means is a node which can be the operand of a quantifier like
7028 * startposp - pointer a pointer to the start position. This is updated
7029 * to point to the byte following the highest successful
7031 * p - the regnode to be repeatedly matched against.
7032 * reginfo - struct holding match state, such as strend
7033 * max - maximum number of things to match.
7034 * depth - (for debugging) backtracking depth.
7037 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
7038 regmatch_info *const reginfo, I32 max, int depth)
7041 char *scan; /* Pointer to current position in target string */
7043 char *loceol = reginfo->strend; /* local version */
7044 I32 hardcount = 0; /* How many matches so far */
7045 bool utf8_target = reginfo->is_utf8_target;
7046 int to_complement = 0; /* Invert the result? */
7048 _char_class_number classnum;
7050 PERL_UNUSED_ARG(depth);
7053 PERL_ARGS_ASSERT_REGREPEAT;
7056 if (max == REG_INFTY)
7058 else if (! utf8_target && loceol - scan > max)
7059 loceol = scan + max;
7061 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
7062 * to the maximum of how far we should go in it (leaving it set to the real
7063 * end, if the maximum permissible would take us beyond that). This allows
7064 * us to make the loop exit condition that we haven't gone past <loceol> to
7065 * also mean that we haven't exceeded the max permissible count, saving a
7066 * test each time through the loop. But it assumes that the OP matches a
7067 * single byte, which is true for most of the OPs below when applied to a
7068 * non-UTF-8 target. Those relatively few OPs that don't have this
7069 * characteristic will have to compensate.
7071 * There is no adjustment for UTF-8 targets, as the number of bytes per
7072 * character varies. OPs will have to test both that the count is less
7073 * than the max permissible (using <hardcount> to keep track), and that we
7074 * are still within the bounds of the string (using <loceol>. A few OPs
7075 * match a single byte no matter what the encoding. They can omit the max
7076 * test if, for the UTF-8 case, they do the adjustment that was skipped
7079 * Thus, the code above sets things up for the common case; and exceptional
7080 * cases need extra work; the common case is to make sure <scan> doesn't
7081 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
7082 * count doesn't exceed the maximum permissible */
7087 while (scan < loceol && hardcount < max && *scan != '\n') {
7088 scan += UTF8SKIP(scan);
7092 while (scan < loceol && *scan != '\n')
7098 while (scan < loceol && hardcount < max) {
7099 scan += UTF8SKIP(scan);
7106 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
7107 if (utf8_target && loceol - scan > max) {
7109 /* <loceol> hadn't been adjusted in the UTF-8 case */
7117 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7121 /* Can use a simple loop if the pattern char to match on is invariant
7122 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
7123 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
7124 * true iff it doesn't matter if the argument is in UTF-8 or not */
7125 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
7126 if (utf8_target && loceol - scan > max) {
7127 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
7128 * since here, to match at all, 1 char == 1 byte */
7129 loceol = scan + max;
7131 while (scan < loceol && UCHARAT(scan) == c) {
7135 else if (reginfo->is_utf8_pat) {
7137 STRLEN scan_char_len;
7139 /* When both target and pattern are UTF-8, we have to do
7141 while (hardcount < max
7143 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
7144 && memEQ(scan, STRING(p), scan_char_len))
7146 scan += scan_char_len;
7150 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
7152 /* Target isn't utf8; convert the character in the UTF-8
7153 * pattern to non-UTF8, and do a simple loop */
7154 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
7155 while (scan < loceol && UCHARAT(scan) == c) {
7158 } /* else pattern char is above Latin1, can't possibly match the
7163 /* Here, the string must be utf8; pattern isn't, and <c> is
7164 * different in utf8 than not, so can't compare them directly.
7165 * Outside the loop, find the two utf8 bytes that represent c, and
7166 * then look for those in sequence in the utf8 string */
7167 U8 high = UTF8_TWO_BYTE_HI(c);
7168 U8 low = UTF8_TWO_BYTE_LO(c);
7170 while (hardcount < max
7171 && scan + 1 < loceol
7172 && UCHARAT(scan) == high
7173 && UCHARAT(scan + 1) == low)
7181 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
7182 assert(! reginfo->is_utf8_pat);
7185 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7189 utf8_flags = FOLDEQ_LOCALE;
7192 case EXACTF: /* This node only generated for non-utf8 patterns */
7193 assert(! reginfo->is_utf8_pat);
7199 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
7203 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
7205 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7207 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
7210 if (c1 == CHRTEST_VOID) {
7211 /* Use full Unicode fold matching */
7212 char *tmpeol = reginfo->strend;
7213 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
7214 while (hardcount < max
7215 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
7216 STRING(p), NULL, pat_len,
7217 reginfo->is_utf8_pat, utf8_flags))
7220 tmpeol = reginfo->strend;
7224 else if (utf8_target) {
7226 while (scan < loceol
7228 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
7230 scan += UTF8SKIP(scan);
7235 while (scan < loceol
7237 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
7238 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
7240 scan += UTF8SKIP(scan);
7245 else if (c1 == c2) {
7246 while (scan < loceol && UCHARAT(scan) == c1) {
7251 while (scan < loceol &&
7252 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
7262 while (hardcount < max
7264 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
7266 scan += UTF8SKIP(scan);
7270 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
7275 /* The argument (FLAGS) to all the POSIX node types is the class number */
7282 if (! utf8_target) {
7283 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
7289 while (hardcount < max && scan < loceol
7290 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
7293 scan += UTF8SKIP(scan);
7306 if (utf8_target && loceol - scan > max) {
7308 /* We didn't adjust <loceol> at the beginning of this routine
7309 * because is UTF-8, but it is actually ok to do so, since here, to
7310 * match, 1 char == 1 byte. */
7311 loceol = scan + max;
7313 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
7326 if (! utf8_target) {
7327 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
7333 /* The complement of something that matches only ASCII matches all
7334 * non-ASCII, plus everything in ASCII that isn't in the class. */
7335 while (hardcount < max && scan < loceol
7336 && (! isASCII_utf8(scan)
7337 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
7339 scan += UTF8SKIP(scan);
7350 if (! utf8_target) {
7351 while (scan < loceol && to_complement
7352 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
7359 classnum = (_char_class_number) FLAGS(p);
7360 if (classnum < _FIRST_NON_SWASH_CC) {
7362 /* Here, a swash is needed for above-Latin1 code points.
7363 * Process as many Latin1 code points using the built-in rules.
7364 * Go to another loop to finish processing upon encountering
7365 * the first Latin1 code point. We could do that in this loop
7366 * as well, but the other way saves having to test if the swash
7367 * has been loaded every time through the loop: extra space to
7369 while (hardcount < max && scan < loceol) {
7370 if (UTF8_IS_INVARIANT(*scan)) {
7371 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
7378 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
7379 if (! (to_complement
7380 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
7389 goto found_above_latin1;
7396 /* For these character classes, the knowledge of how to handle
7397 * every code point is compiled in to Perl via a macro. This
7398 * code is written for making the loops as tight as possible.
7399 * It could be refactored to save space instead */
7401 case _CC_ENUM_SPACE: /* XXX would require separate code
7402 if we revert the change of \v
7405 case _CC_ENUM_PSXSPC:
7406 while (hardcount < max
7408 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7410 scan += UTF8SKIP(scan);
7414 case _CC_ENUM_BLANK:
7415 while (hardcount < max
7417 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7419 scan += UTF8SKIP(scan);
7423 case _CC_ENUM_XDIGIT:
7424 while (hardcount < max
7426 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7428 scan += UTF8SKIP(scan);
7432 case _CC_ENUM_VERTSPACE:
7433 while (hardcount < max
7435 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7437 scan += UTF8SKIP(scan);
7441 case _CC_ENUM_CNTRL:
7442 while (hardcount < max
7444 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7446 scan += UTF8SKIP(scan);
7451 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7457 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7459 /* Load the swash if not already present */
7460 if (! PL_utf8_swash_ptrs[classnum]) {
7461 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7462 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7466 PL_XPosix_ptrs[classnum], &flags);
7469 while (hardcount < max && scan < loceol
7470 && to_complement ^ cBOOL(_generic_utf8(
7473 swash_fetch(PL_utf8_swash_ptrs[classnum],
7477 scan += UTF8SKIP(scan);
7484 while (hardcount < max && scan < loceol &&
7485 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7490 /* LNBREAK can match one or two latin chars, which is ok, but we
7491 * have to use hardcount in this situation, and throw away the
7492 * adjustment to <loceol> done before the switch statement */
7493 loceol = reginfo->strend;
7494 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7515 /* These are all 0 width, so match right here or not at all. */
7519 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7520 assert(0); /* NOTREACHED */
7527 c = scan - *startposp;
7531 GET_RE_DEBUG_FLAGS_DECL;
7533 SV * const prop = sv_newmortal();
7534 regprop(prog, prop, p, reginfo);
7535 PerlIO_printf(Perl_debug_log,
7536 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7537 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7545 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7547 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7548 create a copy so that changes the caller makes won't change the shared one.
7549 If <altsvp> is non-null, will return NULL in it, for back-compat.
7552 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7554 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7560 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL));
7564 Perl__get_regclass_nonbitmap_data(pTHX_ const regexp *prog,
7565 const regnode* node,
7568 SV** only_utf8_locale_ptr)
7570 /* For internal core use only.
7571 * Returns the swash for the input 'node' in the regex 'prog'.
7572 * If <doinit> is 'true', will attempt to create the swash if not already
7574 * If <listsvp> is non-null, will return the printable contents of the
7575 * swash. This can be used to get debugging information even before the
7576 * swash exists, by calling this function with 'doinit' set to false, in
7577 * which case the components that will be used to eventually create the
7578 * swash are returned (in a printable form).
7579 * Tied intimately to how regcomp.c sets up the data structure */
7583 SV *si = NULL; /* Input swash initialization string */
7586 RXi_GET_DECL(prog,progi);
7587 const struct reg_data * const data = prog ? progi->data : NULL;
7589 PERL_ARGS_ASSERT__GET_REGCLASS_NONBITMAP_DATA;
7591 assert(ANYOF_FLAGS(node)
7592 & (ANYOF_UTF8|ANYOF_NONBITMAP_NON_UTF8|ANYOF_LOC_FOLD));
7594 if (data && data->count) {
7595 const U32 n = ARG(node);
7597 if (data->what[n] == 's') {
7598 SV * const rv = MUTABLE_SV(data->data[n]);
7599 AV * const av = MUTABLE_AV(SvRV(rv));
7600 SV **const ary = AvARRAY(av);
7601 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7603 si = *ary; /* ary[0] = the string to initialize the swash with */
7605 /* Elements 3 and 4 are either both present or both absent. [3] is
7606 * any inversion list generated at compile time; [4] indicates if
7607 * that inversion list has any user-defined properties in it. */
7608 if (av_tindex(av) >= 2) {
7609 if (only_utf8_locale_ptr
7611 && ary[2] != &PL_sv_undef)
7613 *only_utf8_locale_ptr = ary[2];
7616 assert(only_utf8_locale_ptr);
7617 *only_utf8_locale_ptr = NULL;
7620 if (av_tindex(av) >= 3) {
7623 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7631 /* Element [1] is reserved for the set-up swash. If already there,
7632 * return it; if not, create it and store it there */
7633 if (ary[1] && SvROK(ary[1])) {
7636 else if (doinit && ((si && si != &PL_sv_undef)
7637 || (invlist && invlist != &PL_sv_undef))) {
7639 sw = _core_swash_init("utf8", /* the utf8 package */
7643 0, /* not from tr/// */
7646 (void)av_store(av, 1, sw);
7651 /* If requested, return a printable version of what this swash matches */
7653 SV* matches_string = newSVpvn("", 0);
7655 /* The swash should be used, if possible, to get the data, as it
7656 * contains the resolved data. But this function can be called at
7657 * compile-time, before everything gets resolved, in which case we
7658 * return the currently best available information, which is the string
7659 * that will eventually be used to do that resolving, 'si' */
7660 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7661 && (si && si != &PL_sv_undef))
7663 sv_catsv(matches_string, si);
7666 /* Add the inversion list to whatever we have. This may have come from
7667 * the swash, or from an input parameter */
7669 sv_catsv(matches_string, _invlist_contents(invlist));
7671 *listsvp = matches_string;
7676 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
7679 - reginclass - determine if a character falls into a character class
7681 n is the ANYOF regnode
7682 p is the target string
7683 p_end points to one byte beyond the end of the target string
7684 utf8_target tells whether p is in UTF-8.
7686 Returns true if matched; false otherwise.
7688 Note that this can be a synthetic start class, a combination of various
7689 nodes, so things you think might be mutually exclusive, such as locale,
7690 aren't. It can match both locale and non-locale
7695 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
7698 const char flags = ANYOF_FLAGS(n);
7702 PERL_ARGS_ASSERT_REGINCLASS;
7704 /* If c is not already the code point, get it. Note that
7705 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7706 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7708 c = utf8n_to_uvchr(p, p_end - p, &c_len,
7709 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7710 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7711 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7712 * UTF8_ALLOW_FFFF */
7713 if (c_len == (STRLEN)-1)
7714 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7717 /* If this character is potentially in the bitmap, check it */
7719 if (ANYOF_BITMAP_TEST(n, c))
7721 else if (flags & ANYOF_NON_UTF8_NON_ASCII_ALL
7727 else if (flags & ANYOF_LOCALE_FLAGS) {
7728 if (flags & ANYOF_LOC_FOLD) {
7729 if (ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) {
7733 if (! match && ANYOF_POSIXL_TEST_ANY_SET(n)) {
7735 /* The data structure is arranged so bits 0, 2, 4, ... are set
7736 * if the class includes the Posix character class given by
7737 * bit/2; and 1, 3, 5, ... are set if the class includes the
7738 * complemented Posix class given by int(bit/2). So we loop
7739 * through the bits, each time changing whether we complement
7740 * the result or not. Suppose for the sake of illustration
7741 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7742 * is set, it means there is a match for this ANYOF node if the
7743 * character is in the class given by the expression (0 / 2 = 0
7744 * = \w). If it is in that class, isFOO_lc() will return 1,
7745 * and since 'to_complement' is 0, the result will stay TRUE,
7746 * and we exit the loop. Suppose instead that bit 0 is 0, but
7747 * bit 1 is 1. That means there is a match if the character
7748 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7749 * but will on bit 1. On the second iteration 'to_complement'
7750 * will be 1, so the exclusive or will reverse things, so we
7751 * are testing for \W. On the third iteration, 'to_complement'
7752 * will be 0, and we would be testing for \s; the fourth
7753 * iteration would test for \S, etc.
7755 * Note that this code assumes that all the classes are closed
7756 * under folding. For example, if a character matches \w, then
7757 * its fold does too; and vice versa. This should be true for
7758 * any well-behaved locale for all the currently defined Posix
7759 * classes, except for :lower: and :upper:, which are handled
7760 * by the pseudo-class :cased: which matches if either of the
7761 * other two does. To get rid of this assumption, an outer
7762 * loop could be used below to iterate over both the source
7763 * character, and its fold (if different) */
7766 int to_complement = 0;
7768 while (count < ANYOF_MAX) {
7769 if (ANYOF_POSIXL_TEST(n, count)
7770 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7783 /* If the bitmap didn't (or couldn't) match, and something outside the
7784 * bitmap could match, try that. */
7786 if (c >= 256 && (flags & ANYOF_ABOVE_LATIN1_ALL)) {
7787 match = TRUE; /* Everything above 255 matches */
7789 else if ((flags & ANYOF_NONBITMAP_NON_UTF8)
7790 || (utf8_target && (flags & ANYOF_UTF8))
7791 || ((flags & ANYOF_LOC_FOLD)
7792 && IN_UTF8_CTYPE_LOCALE
7793 && ARG(n) != ANYOF_NONBITMAP_EMPTY))
7795 SV* only_utf8_locale = NULL;
7796 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
7802 } else { /* Convert to utf8 */
7804 utf8_p = bytes_to_utf8(p, &len);
7807 if (swash_fetch(sw, utf8_p, TRUE)) {
7811 /* If we allocated a string above, free it */
7812 if (! utf8_target) Safefree(utf8_p);
7814 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
7815 match = _invlist_contains_cp(only_utf8_locale, c);
7819 if (UNICODE_IS_SUPER(c)
7820 && (flags & ANYOF_WARN_SUPER)
7821 && ckWARN_d(WARN_NON_UNICODE))
7823 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7824 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c);
7828 #if ANYOF_INVERT != 1
7829 /* Depending on compiler optimization cBOOL takes time, so if don't have to
7831 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
7834 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7835 return (flags & ANYOF_INVERT) ^ match;
7839 S_reghop3(U8 *s, SSize_t off, const U8* lim)
7841 /* return the position 'off' UTF-8 characters away from 's', forward if
7842 * 'off' >= 0, backwards if negative. But don't go outside of position
7843 * 'lim', which better be < s if off < 0 */
7847 PERL_ARGS_ASSERT_REGHOP3;
7850 while (off-- && s < lim) {
7851 /* XXX could check well-formedness here */
7856 while (off++ && s > lim) {
7858 if (UTF8_IS_CONTINUED(*s)) {
7859 while (s > lim && UTF8_IS_CONTINUATION(*s))
7862 /* XXX could check well-formedness here */
7869 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
7873 PERL_ARGS_ASSERT_REGHOP4;
7876 while (off-- && s < rlim) {
7877 /* XXX could check well-formedness here */
7882 while (off++ && s > llim) {
7884 if (UTF8_IS_CONTINUED(*s)) {
7885 while (s > llim && UTF8_IS_CONTINUATION(*s))
7888 /* XXX could check well-formedness here */
7894 /* like reghop3, but returns NULL on overrun, rather than returning last
7898 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
7902 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7905 while (off-- && s < lim) {
7906 /* XXX could check well-formedness here */
7913 while (off++ && s > lim) {
7915 if (UTF8_IS_CONTINUED(*s)) {
7916 while (s > lim && UTF8_IS_CONTINUATION(*s))
7919 /* XXX could check well-formedness here */
7928 /* when executing a regex that may have (?{}), extra stuff needs setting
7929 up that will be visible to the called code, even before the current
7930 match has finished. In particular:
7932 * $_ is localised to the SV currently being matched;
7933 * pos($_) is created if necessary, ready to be updated on each call-out
7935 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
7936 isn't set until the current pattern is successfully finished), so that
7937 $1 etc of the match-so-far can be seen;
7938 * save the old values of subbeg etc of the current regex, and set then
7939 to the current string (again, this is normally only done at the end
7944 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
7947 regexp *const rex = ReANY(reginfo->prog);
7948 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
7950 eval_state->rex = rex;
7953 /* Make $_ available to executed code. */
7954 if (reginfo->sv != DEFSV) {
7956 DEFSV_set(reginfo->sv);
7959 if (!(mg = mg_find_mglob(reginfo->sv))) {
7960 /* prepare for quick setting of pos */
7961 mg = sv_magicext_mglob(reginfo->sv);
7964 eval_state->pos_magic = mg;
7965 eval_state->pos = mg->mg_len;
7966 eval_state->pos_flags = mg->mg_flags;
7969 eval_state->pos_magic = NULL;
7971 if (!PL_reg_curpm) {
7972 /* PL_reg_curpm is a fake PMOP that we can attach the current
7973 * regex to and point PL_curpm at, so that $1 et al are visible
7974 * within a /(?{})/. It's just allocated once per interpreter the
7975 * first time its needed */
7976 Newxz(PL_reg_curpm, 1, PMOP);
7979 SV* const repointer = &PL_sv_undef;
7980 /* this regexp is also owned by the new PL_reg_curpm, which
7981 will try to free it. */
7982 av_push(PL_regex_padav, repointer);
7983 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
7984 PL_regex_pad = AvARRAY(PL_regex_padav);
7988 SET_reg_curpm(reginfo->prog);
7989 eval_state->curpm = PL_curpm;
7990 PL_curpm = PL_reg_curpm;
7991 if (RXp_MATCH_COPIED(rex)) {
7992 /* Here is a serious problem: we cannot rewrite subbeg,
7993 since it may be needed if this match fails. Thus
7994 $` inside (?{}) could fail... */
7995 eval_state->subbeg = rex->subbeg;
7996 eval_state->sublen = rex->sublen;
7997 eval_state->suboffset = rex->suboffset;
7998 eval_state->subcoffset = rex->subcoffset;
8000 eval_state->saved_copy = rex->saved_copy;
8002 RXp_MATCH_COPIED_off(rex);
8005 eval_state->subbeg = NULL;
8006 rex->subbeg = (char *)reginfo->strbeg;
8008 rex->subcoffset = 0;
8009 rex->sublen = reginfo->strend - reginfo->strbeg;
8013 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
8016 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
8019 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
8020 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
8023 Safefree(aux->poscache);
8027 /* undo the effects of S_setup_eval_state() */
8029 if (eval_state->subbeg) {
8030 regexp * const rex = eval_state->rex;
8031 rex->subbeg = eval_state->subbeg;
8032 rex->sublen = eval_state->sublen;
8033 rex->suboffset = eval_state->suboffset;
8034 rex->subcoffset = eval_state->subcoffset;
8036 rex->saved_copy = eval_state->saved_copy;
8038 RXp_MATCH_COPIED_on(rex);
8040 if (eval_state->pos_magic)
8042 eval_state->pos_magic->mg_len = eval_state->pos;
8043 eval_state->pos_magic->mg_flags =
8044 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
8045 | (eval_state->pos_flags & MGf_BYTES);
8048 PL_curpm = eval_state->curpm;
8051 PL_regmatch_state = aux->old_regmatch_state;
8052 PL_regmatch_slab = aux->old_regmatch_slab;
8054 /* free all slabs above current one - this must be the last action
8055 * of this function, as aux and eval_state are allocated within
8056 * slabs and may be freed here */
8058 s = PL_regmatch_slab->next;
8060 PL_regmatch_slab->next = NULL;
8062 regmatch_slab * const osl = s;
8071 S_to_utf8_substr(pTHX_ regexp *prog)
8073 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
8074 * on the converted value */
8078 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
8081 if (prog->substrs->data[i].substr
8082 && !prog->substrs->data[i].utf8_substr) {
8083 SV* const sv = newSVsv(prog->substrs->data[i].substr);
8084 prog->substrs->data[i].utf8_substr = sv;
8085 sv_utf8_upgrade(sv);
8086 if (SvVALID(prog->substrs->data[i].substr)) {
8087 if (SvTAIL(prog->substrs->data[i].substr)) {
8088 /* Trim the trailing \n that fbm_compile added last
8090 SvCUR_set(sv, SvCUR(sv) - 1);
8091 /* Whilst this makes the SV technically "invalid" (as its
8092 buffer is no longer followed by "\0") when fbm_compile()
8093 adds the "\n" back, a "\0" is restored. */
8094 fbm_compile(sv, FBMcf_TAIL);
8098 if (prog->substrs->data[i].substr == prog->check_substr)
8099 prog->check_utf8 = sv;
8105 S_to_byte_substr(pTHX_ regexp *prog)
8107 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
8108 * on the converted value; returns FALSE if can't be converted. */
8113 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
8116 if (prog->substrs->data[i].utf8_substr
8117 && !prog->substrs->data[i].substr) {
8118 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
8119 if (! sv_utf8_downgrade(sv, TRUE)) {
8122 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
8123 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
8124 /* Trim the trailing \n that fbm_compile added last
8126 SvCUR_set(sv, SvCUR(sv) - 1);
8127 fbm_compile(sv, FBMcf_TAIL);
8131 prog->substrs->data[i].substr = sv;
8132 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
8133 prog->check_substr = sv;
8142 * c-indentation-style: bsd
8144 * indent-tabs-mode: nil
8147 * ex: set ts=8 sts=4 sw=4 et: