5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "inline_invlist.c"
84 #include "unicode_constants.h"
87 /* At least one required character in the target string is expressible only in
89 static const char* const non_utf8_target_but_utf8_required
90 = "Can't match, because target string needs to be in UTF-8\n";
93 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
94 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
98 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
101 #define STATIC static
104 /* Valid only for non-utf8 strings: avoids the reginclass
105 * call if there are no complications: i.e., if everything matchable is
106 * straight forward in the bitmap */
107 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \
108 : ANYOF_BITMAP_TEST(p,*(c)))
114 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
115 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
117 #define HOPc(pos,off) \
118 (char *)(reginfo->is_utf8_target \
119 ? reghop3((U8*)pos, off, \
120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
123 #define HOPBACKc(pos, off) \
124 (char*)(reginfo->is_utf8_target \
125 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
126 : (pos - off >= reginfo->strbeg) \
130 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
131 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
133 /* lim must be +ve. Returns NULL on overshoot */
134 #define HOPMAYBE3(pos,off,lim) \
135 (reginfo->is_utf8_target \
136 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
137 : ((U8*)pos + off <= lim) \
141 /* like HOP3, but limits the result to <= lim even for the non-utf8 case.
142 * off must be >=0; args should be vars rather than expressions */
143 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
144 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
145 : (U8*)((pos + off) > lim ? lim : (pos + off)))
147 #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
148 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
150 #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
152 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
153 #define NEXTCHR_IS_EOS (nextchr < 0)
155 #define SET_nextchr \
156 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
158 #define SET_locinput(p) \
163 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
165 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
166 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
167 1, 0, invlist, &flags); \
172 /* If in debug mode, we test that a known character properly matches */
174 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
177 utf8_char_in_property) \
178 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \
179 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
181 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
184 utf8_char_in_property) \
185 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
188 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
189 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
191 PL_XPosix_ptrs[_CC_WORDCHAR], \
192 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
194 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
196 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
197 "_X_regular_begin", \
199 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \
200 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
203 COMBINING_GRAVE_ACCENT_UTF8); \
206 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
207 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
209 /* for use after a quantifier and before an EXACT-like node -- japhy */
210 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
212 * NOTE that *nothing* that affects backtracking should be in here, specifically
213 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
214 * node that is in between two EXACT like nodes when ascertaining what the required
215 * "follow" character is. This should probably be moved to regex compile time
216 * although it may be done at run time beause of the REF possibility - more
217 * investigation required. -- demerphq
219 #define JUMPABLE(rn) ( \
221 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
223 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
224 OP(rn) == PLUS || OP(rn) == MINMOD || \
226 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
228 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
230 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
233 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
234 we don't need this definition. */
235 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
236 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
237 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
240 /* ... so we use this as its faster. */
241 #define IS_TEXT(rn) ( OP(rn)==EXACT )
242 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
243 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
244 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
249 Search for mandatory following text node; for lookahead, the text must
250 follow but for lookbehind (rn->flags != 0) we skip to the next step.
252 #define FIND_NEXT_IMPT(rn) STMT_START { \
253 while (JUMPABLE(rn)) { \
254 const OPCODE type = OP(rn); \
255 if (type == SUSPEND || PL_regkind[type] == CURLY) \
256 rn = NEXTOPER(NEXTOPER(rn)); \
257 else if (type == PLUS) \
259 else if (type == IFMATCH) \
260 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
261 else rn += NEXT_OFF(rn); \
265 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
266 * These are for the pre-composed Hangul syllables, which are all in a
267 * contiguous block and arranged there in such a way so as to facilitate
268 * alorithmic determination of their characteristics. As such, they don't need
269 * a swash, but can be determined by simple arithmetic. Almost all are
270 * GCB=LVT, but every 28th one is a GCB=LV */
271 #define SBASE 0xAC00 /* Start of block */
272 #define SCount 11172 /* Length of block */
275 #define SLAB_FIRST(s) (&(s)->states[0])
276 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
278 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
279 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
280 static regmatch_state * S_push_slab(pTHX);
282 #define REGCP_PAREN_ELEMS 3
283 #define REGCP_OTHER_ELEMS 3
284 #define REGCP_FRAME_ELEMS 1
285 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
286 * are needed for the regexp context stack bookkeeping. */
289 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
291 const int retval = PL_savestack_ix;
292 const int paren_elems_to_push =
293 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
294 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
295 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
297 GET_RE_DEBUG_FLAGS_DECL;
299 PERL_ARGS_ASSERT_REGCPPUSH;
301 if (paren_elems_to_push < 0)
302 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
303 (int)paren_elems_to_push, (int)maxopenparen,
304 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
306 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
307 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
308 " out of range (%lu-%ld)",
310 (unsigned long)maxopenparen,
313 SSGROW(total_elems + REGCP_FRAME_ELEMS);
316 if ((int)maxopenparen > (int)parenfloor)
317 PerlIO_printf(Perl_debug_log,
318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
323 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
324 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
325 SSPUSHIV(rex->offs[p].end);
326 SSPUSHIV(rex->offs[p].start);
327 SSPUSHINT(rex->offs[p].start_tmp);
328 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
329 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
331 (IV)rex->offs[p].start,
332 (IV)rex->offs[p].start_tmp,
336 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
337 SSPUSHINT(maxopenparen);
338 SSPUSHINT(rex->lastparen);
339 SSPUSHINT(rex->lastcloseparen);
340 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
345 /* These are needed since we do not localize EVAL nodes: */
346 #define REGCP_SET(cp) \
348 PerlIO_printf(Perl_debug_log, \
349 " Setting an EVAL scope, savestack=%"IVdf"\n", \
350 (IV)PL_savestack_ix)); \
353 #define REGCP_UNWIND(cp) \
355 if (cp != PL_savestack_ix) \
356 PerlIO_printf(Perl_debug_log, \
357 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
358 (IV)(cp), (IV)PL_savestack_ix)); \
361 #define UNWIND_PAREN(lp, lcp) \
362 for (n = rex->lastparen; n > lp; n--) \
363 rex->offs[n].end = -1; \
364 rex->lastparen = n; \
365 rex->lastcloseparen = lcp;
369 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
373 GET_RE_DEBUG_FLAGS_DECL;
375 PERL_ARGS_ASSERT_REGCPPOP;
377 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
379 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
380 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
381 rex->lastcloseparen = SSPOPINT;
382 rex->lastparen = SSPOPINT;
383 *maxopenparen_p = SSPOPINT;
385 i -= REGCP_OTHER_ELEMS;
386 /* Now restore the parentheses context. */
388 if (i || rex->lastparen + 1 <= rex->nparens)
389 PerlIO_printf(Perl_debug_log,
390 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
395 paren = *maxopenparen_p;
396 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
398 rex->offs[paren].start_tmp = SSPOPINT;
399 rex->offs[paren].start = SSPOPIV;
401 if (paren <= rex->lastparen)
402 rex->offs[paren].end = tmps;
403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
406 (IV)rex->offs[paren].start,
407 (IV)rex->offs[paren].start_tmp,
408 (IV)rex->offs[paren].end,
409 (paren > rex->lastparen ? "(skipped)" : ""));
414 /* It would seem that the similar code in regtry()
415 * already takes care of this, and in fact it is in
416 * a better location to since this code can #if 0-ed out
417 * but the code in regtry() is needed or otherwise tests
418 * requiring null fields (pat.t#187 and split.t#{13,14}
419 * (as of patchlevel 7877) will fail. Then again,
420 * this code seems to be necessary or otherwise
421 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
422 * --jhi updated by dapm */
423 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
424 if (i > *maxopenparen_p)
425 rex->offs[i].start = -1;
426 rex->offs[i].end = -1;
427 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
428 " \\%"UVuf": %s ..-1 undeffing\n",
430 (i > *maxopenparen_p) ? "-1" : " "
436 /* restore the parens and associated vars at savestack position ix,
437 * but without popping the stack */
440 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
442 I32 tmpix = PL_savestack_ix;
443 PL_savestack_ix = ix;
444 regcppop(rex, maxopenparen_p);
445 PL_savestack_ix = tmpix;
448 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
451 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
453 /* Returns a boolean as to whether or not 'character' is a member of the
454 * Posix character class given by 'classnum' that should be equivalent to a
455 * value in the typedef '_char_class_number'.
457 * Ideally this could be replaced by a just an array of function pointers
458 * to the C library functions that implement the macros this calls.
459 * However, to compile, the precise function signatures are required, and
460 * these may vary from platform to to platform. To avoid having to figure
461 * out what those all are on each platform, I (khw) am using this method,
462 * which adds an extra layer of function call overhead (unless the C
463 * optimizer strips it away). But we don't particularly care about
464 * performance with locales anyway. */
466 switch ((_char_class_number) classnum) {
467 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
468 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
469 case _CC_ENUM_ASCII: return isASCII_LC(character);
470 case _CC_ENUM_BLANK: return isBLANK_LC(character);
471 case _CC_ENUM_CASED: return isLOWER_LC(character)
472 || isUPPER_LC(character);
473 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
474 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
475 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
476 case _CC_ENUM_LOWER: return isLOWER_LC(character);
477 case _CC_ENUM_PRINT: return isPRINT_LC(character);
478 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
479 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
480 case _CC_ENUM_SPACE: return isSPACE_LC(character);
481 case _CC_ENUM_UPPER: return isUPPER_LC(character);
482 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
483 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
484 default: /* VERTSPACE should never occur in locales */
485 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
488 assert(0); /* NOTREACHED */
493 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
495 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
496 * 'character' is a member of the Posix character class given by 'classnum'
497 * that should be equivalent to a value in the typedef
498 * '_char_class_number'.
500 * This just calls isFOO_lc on the code point for the character if it is in
501 * the range 0-255. Outside that range, all characters avoid Unicode
502 * rules, ignoring any locale. So use the Unicode function if this class
503 * requires a swash, and use the Unicode macro otherwise. */
505 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
507 if (UTF8_IS_INVARIANT(*character)) {
508 return isFOO_lc(classnum, *character);
510 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
511 return isFOO_lc(classnum,
512 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
515 if (classnum < _FIRST_NON_SWASH_CC) {
517 /* Initialize the swash unless done already */
518 if (! PL_utf8_swash_ptrs[classnum]) {
519 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
520 PL_utf8_swash_ptrs[classnum] =
521 _core_swash_init("utf8",
524 PL_XPosix_ptrs[classnum], &flags);
527 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
529 TRUE /* is UTF */ ));
532 switch ((_char_class_number) classnum) {
534 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
536 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
537 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
538 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
542 return FALSE; /* Things like CNTRL are always below 256 */
546 * pregexec and friends
549 #ifndef PERL_IN_XSUB_RE
551 - pregexec - match a regexp against a string
554 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
555 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
556 /* stringarg: the point in the string at which to begin matching */
557 /* strend: pointer to null at end of string */
558 /* strbeg: real beginning of string */
559 /* minend: end of match must be >= minend bytes after stringarg. */
560 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
561 * itself is accessed via the pointers above */
562 /* nosave: For optimizations. */
564 PERL_ARGS_ASSERT_PREGEXEC;
567 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
568 nosave ? 0 : REXEC_COPY_STR);
574 /* re_intuit_start():
576 * Based on some optimiser hints, try to find the earliest position in the
577 * string where the regex could match.
579 * rx: the regex to match against
580 * sv: the SV being matched: only used for utf8 flag; the string
581 * itself is accessed via the pointers below. Note that on
582 * something like an overloaded SV, SvPOK(sv) may be false
583 * and the string pointers may point to something unrelated to
585 * strbeg: real beginning of string
586 * strpos: the point in the string at which to begin matching
587 * strend: pointer to the byte following the last char of the string
588 * flags currently unused; set to 0
589 * data: currently unused; set to NULL
591 * The basic idea of re_intuit_start() is to use some known information
592 * about the pattern, namely:
594 * a) the longest known anchored substring (i.e. one that's at a
595 * constant offset from the beginning of the pattern; but not
596 * necessarily at a fixed offset from the beginning of the
598 * b) the longest floating substring (i.e. one that's not at a constant
599 * offset from the beginning of the pattern);
600 * c) Whether the pattern is anchored to the string; either
601 * an absolute anchor: /^../, or anchored to \n: /^.../m,
602 * or anchored to pos(): /\G/;
603 * d) A start class: a real or synthetic character class which
604 * represents which characters are legal at the start of the pattern;
606 * to either quickly reject the match, or to find the earliest position
607 * within the string at which the pattern might match, thus avoiding
608 * running the full NFA engine at those earlier locations, only to
609 * eventually fail and retry further along.
611 * Returns NULL if the pattern can't match, or returns the address within
612 * the string which is the earliest place the match could occur.
614 * The longest of the anchored and floating substrings is called 'check'
615 * and is checked first. The other is called 'other' and is checked
616 * second. The 'other' substring may not be present. For example,
618 * /(abc|xyz)ABC\d{0,3}DEFG/
622 * check substr (float) = "DEFG", offset 6..9 chars
623 * other substr (anchored) = "ABC", offset 3..3 chars
626 * Be aware that during the course of this function, sometimes 'anchored'
627 * refers to a substring being anchored relative to the start of the
628 * pattern, and sometimes to the pattern itself being anchored relative to
629 * the string. For example:
631 * /\dabc/: "abc" is anchored to the pattern;
632 * /^\dabc/: "abc" is anchored to the pattern and the string;
633 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
634 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
635 * but the pattern is anchored to the string.
639 Perl_re_intuit_start(pTHX_
642 const char * const strbeg,
646 re_scream_pos_data *data)
648 struct regexp *const prog = ReANY(rx);
649 SSize_t start_shift = prog->check_offset_min;
650 /* Should be nonnegative! */
651 SSize_t end_shift = 0;
652 /* current lowest pos in string where the regex can start matching */
653 char *rx_origin = strpos;
655 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
656 U8 other_ix = 1 - prog->substrs->check_ix;
658 char *other_last = strpos;/* latest pos 'other' substr already checked to */
659 char *check_at = NULL; /* check substr found at this pos */
660 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
661 RXi_GET_DECL(prog,progi);
662 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
663 regmatch_info *const reginfo = ®info_buf;
664 GET_RE_DEBUG_FLAGS_DECL;
666 PERL_ARGS_ASSERT_RE_INTUIT_START;
667 PERL_UNUSED_ARG(flags);
668 PERL_UNUSED_ARG(data);
670 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
671 "Intuit: trying to determine minimum start position...\n"));
673 /* for now, assume that all substr offsets are positive. If at some point
674 * in the future someone wants to do clever things with look-behind and
675 * -ve offsets, they'll need to fix up any code in this function
676 * which uses these offsets. See the thread beginning
677 * <20140113145929.GF27210@iabyn.com>
679 assert(prog->substrs->data[0].min_offset >= 0);
680 assert(prog->substrs->data[0].max_offset >= 0);
681 assert(prog->substrs->data[1].min_offset >= 0);
682 assert(prog->substrs->data[1].max_offset >= 0);
683 assert(prog->substrs->data[2].min_offset >= 0);
684 assert(prog->substrs->data[2].max_offset >= 0);
686 /* for now, assume that if both present, that the floating substring
687 * doesn't start before the anchored substring.
688 * If you break this assumption (e.g. doing better optimisations
689 * with lookahead/behind), then you'll need to audit the code in this
690 * function carefully first
693 ! ( (prog->anchored_utf8 || prog->anchored_substr)
694 && (prog->float_utf8 || prog->float_substr))
695 || (prog->float_min_offset >= prog->anchored_offset));
697 /* byte rather than char calculation for efficiency. It fails
698 * to quickly reject some cases that can't match, but will reject
699 * them later after doing full char arithmetic */
700 if (prog->minlen > strend - strpos) {
701 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
702 " String too short...\n"));
706 reginfo->is_utf8_target = cBOOL(utf8_target);
707 reginfo->info_aux = NULL;
708 reginfo->strbeg = strbeg;
709 reginfo->strend = strend;
710 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
712 /* not actually used within intuit, but zero for safety anyway */
713 reginfo->poscache_maxiter = 0;
716 if (!prog->check_utf8 && prog->check_substr)
717 to_utf8_substr(prog);
718 check = prog->check_utf8;
720 if (!prog->check_substr && prog->check_utf8) {
721 if (! to_byte_substr(prog)) {
722 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
725 check = prog->check_substr;
728 /* dump the various substring data */
729 DEBUG_OPTIMISE_MORE_r({
731 for (i=0; i<=2; i++) {
732 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
733 : prog->substrs->data[i].substr);
737 PerlIO_printf(Perl_debug_log,
738 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
739 " useful=%"IVdf" utf8=%d [%s]\n",
741 (IV)prog->substrs->data[i].min_offset,
742 (IV)prog->substrs->data[i].max_offset,
743 (IV)prog->substrs->data[i].end_shift,
750 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
752 /* ml_anch: check after \n?
754 * A note about IMPLICIT: on an un-anchored pattern beginning
755 * with /.*.../, these flags will have been added by the
757 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
758 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
760 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
761 && !(prog->intflags & PREGf_IMPLICIT);
763 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
764 /* we are only allowed to match at BOS or \G */
766 /* trivially reject if there's a BOS anchor and we're not at BOS.
768 * Note that we don't try to do a similar quick reject for
769 * \G, since generally the caller will have calculated strpos
770 * based on pos() and gofs, so the string is already correctly
771 * anchored by definition; and handling the exceptions would
772 * be too fiddly (e.g. REXEC_IGNOREPOS).
774 if ( strpos != strbeg
775 && (prog->intflags & PREGf_ANCH_SBOL))
777 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
778 " Not at start...\n"));
782 /* in the presence of an anchor, the anchored (relative to the
783 * start of the regex) substr must also be anchored relative
784 * to strpos. So quickly reject if substr isn't found there.
785 * This works for \G too, because the caller will already have
786 * subtracted gofs from pos, and gofs is the offset from the
787 * \G to the start of the regex. For example, in /.abc\Gdef/,
788 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
789 * caller will have set strpos=pos()-4; we look for the substr
790 * at position pos()-4+1, which lines up with the "a" */
792 if (prog->check_offset_min == prog->check_offset_max
793 && !(prog->intflags & PREGf_CANY_SEEN))
795 /* Substring at constant offset from beg-of-str... */
796 SSize_t slen = SvCUR(check);
797 char *s = HOP3c(strpos, prog->check_offset_min, strend);
799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
800 " Looking for check substr at fixed offset %"IVdf"...\n",
801 (IV)prog->check_offset_min));
804 /* In this case, the regex is anchored at the end too.
805 * Unless it's a multiline match, the lengths must match
806 * exactly, give or take a \n. NB: slen >= 1 since
807 * the last char of check is \n */
809 && ( strend - s > slen
810 || strend - s < slen - 1
811 || (strend - s == slen && strend[-1] != '\n')))
813 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
814 " String too long...\n"));
817 /* Now should match s[0..slen-2] */
820 if (slen && (*SvPVX_const(check) != *s
821 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
823 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
824 " String not equal...\n"));
829 goto success_at_start;
834 end_shift = prog->check_end_shift;
836 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
838 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
839 (IV)end_shift, RX_PRECOMP(prog));
844 /* This is the (re)entry point of the main loop in this function.
845 * The goal of this loop is to:
846 * 1) find the "check" substring in the region rx_origin..strend
847 * (adjusted by start_shift / end_shift). If not found, reject
849 * 2) If it exists, look for the "other" substr too if defined; for
850 * example, if the check substr maps to the anchored substr, then
851 * check the floating substr, and vice-versa. If not found, go
852 * back to (1) with rx_origin suitably incremented.
853 * 3) If we find an rx_origin position that doesn't contradict
854 * either of the substrings, then check the possible additional
855 * constraints on rx_origin of /^.../m or a known start class.
856 * If these fail, then depending on which constraints fail, jump
857 * back to here, or to various other re-entry points further along
858 * that skip some of the first steps.
859 * 4) If we pass all those tests, update the BmUSEFUL() count on the
860 * substring. If the start position was determined to be at the
861 * beginning of the string - so, not rejected, but not optimised,
862 * since we have to run regmatch from position 0 - decrement the
863 * BmUSEFUL() count. Otherwise increment it.
867 /* first, look for the 'check' substring */
873 DEBUG_OPTIMISE_MORE_r({
874 PerlIO_printf(Perl_debug_log,
875 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
876 " Start shift: %"IVdf" End shift %"IVdf
877 " Real end Shift: %"IVdf"\n",
878 (IV)(rx_origin - strpos),
879 (IV)prog->check_offset_min,
882 (IV)prog->check_end_shift);
885 if (prog->intflags & PREGf_CANY_SEEN) {
886 start_point= (U8*)(rx_origin + start_shift);
887 end_point= (U8*)(strend - end_shift);
888 if (start_point > end_point)
891 end_point = HOP3(strend, -end_shift, strbeg);
892 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
898 /* If the regex is absolutely anchored to either the start of the
899 * string (SBOL) or to pos() (ANCH_GPOS), then
900 * check_offset_max represents an upper bound on the string where
901 * the substr could start. For the ANCH_GPOS case, we assume that
902 * the caller of intuit will have already set strpos to
903 * pos()-gofs, so in this case strpos + offset_max will still be
904 * an upper bound on the substr.
907 && prog->intflags & PREGf_ANCH
908 && prog->check_offset_max != SSize_t_MAX)
910 SSize_t len = SvCUR(check) - !!SvTAIL(check);
911 const char * const anchor =
912 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
914 /* do a bytes rather than chars comparison. It's conservative;
915 * so it skips doing the HOP if the result can't possibly end
916 * up earlier than the old value of end_point.
918 if ((char*)end_point - anchor > prog->check_offset_max) {
919 end_point = HOP3lim((U8*)anchor,
920 prog->check_offset_max,
926 DEBUG_OPTIMISE_MORE_r({
927 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
928 (int)(end_point - start_point),
929 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
933 check_at = fbm_instr( start_point, end_point,
934 check, multiline ? FBMrf_MULTILINE : 0);
936 /* Update the count-of-usability, remove useless subpatterns,
940 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
941 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
942 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
943 (check_at ? "Found" : "Did not find"),
944 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
945 ? "anchored" : "floating"),
948 (check_at ? " at offset " : "...\n") );
953 /* Finish the diagnostic message */
954 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
956 /* set rx_origin to the minimum position where the regex could start
957 * matching, given the constraint of the just-matched check substring.
958 * But don't set it lower than previously.
961 if (check_at - rx_origin > prog->check_offset_max)
962 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
966 /* now look for the 'other' substring if defined */
968 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
969 : prog->substrs->data[other_ix].substr)
971 /* Take into account the "other" substring. */
975 struct reg_substr_datum *other;
978 other = &prog->substrs->data[other_ix];
980 /* if "other" is anchored:
981 * we've previously found a floating substr starting at check_at.
982 * This means that the regex origin must lie somewhere
983 * between min (rx_origin): HOP3(check_at, -check_offset_max)
984 * and max: HOP3(check_at, -check_offset_min)
985 * (except that min will be >= strpos)
986 * So the fixed substr must lie somewhere between
987 * HOP3(min, anchored_offset)
988 * HOP3(max, anchored_offset) + SvCUR(substr)
991 /* if "other" is floating
992 * Calculate last1, the absolute latest point where the
993 * floating substr could start in the string, ignoring any
994 * constraints from the earlier fixed match. It is calculated
997 * strend - prog->minlen (in chars) is the absolute latest
998 * position within the string where the origin of the regex
999 * could appear. The latest start point for the floating
1000 * substr is float_min_offset(*) on from the start of the
1001 * regex. last1 simply combines thee two offsets.
1003 * (*) You might think the latest start point should be
1004 * float_max_offset from the regex origin, and technically
1005 * you'd be correct. However, consider
1007 * Here, float min, max are 3,5 and minlen is 7.
1008 * This can match either
1012 * In the first case, the regex matches minlen chars; in the
1013 * second, minlen+1, in the third, minlen+2.
1014 * In the first case, the floating offset is 3 (which equals
1015 * float_min), in the second, 4, and in the third, 5 (which
1016 * equals float_max). In all cases, the floating string bcd
1017 * can never start more than 4 chars from the end of the
1018 * string, which equals minlen - float_min. As the substring
1019 * starts to match more than float_min from the start of the
1020 * regex, it makes the regex match more than minlen chars,
1021 * and the two cancel each other out. So we can always use
1022 * float_min - minlen, rather than float_max - minlen for the
1023 * latest position in the string.
1025 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1026 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1029 assert(prog->minlen >= other->min_offset);
1030 last1 = HOP3c(strend,
1031 other->min_offset - prog->minlen, strbeg);
1033 if (other_ix) {/* i.e. if (other-is-float) */
1034 /* last is the latest point where the floating substr could
1035 * start, *given* any constraints from the earlier fixed
1036 * match. This constraint is that the floating string starts
1037 * <= float_max_offset chars from the regex origin (rx_origin).
1038 * If this value is less than last1, use it instead.
1040 assert(rx_origin <= last1);
1042 /* this condition handles the offset==infinity case, and
1043 * is a short-cut otherwise. Although it's comparing a
1044 * byte offset to a char length, it does so in a safe way,
1045 * since 1 char always occupies 1 or more bytes,
1046 * so if a string range is (last1 - rx_origin) bytes,
1047 * it will be less than or equal to (last1 - rx_origin)
1048 * chars; meaning it errs towards doing the accurate HOP3
1049 * rather than just using last1 as a short-cut */
1050 (last1 - rx_origin) < other->max_offset
1052 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1055 assert(strpos + start_shift <= check_at);
1056 last = HOP4c(check_at, other->min_offset - start_shift,
1060 s = HOP3c(rx_origin, other->min_offset, strend);
1061 if (s < other_last) /* These positions already checked */
1064 must = utf8_target ? other->utf8_substr : other->substr;
1065 assert(SvPOK(must));
1068 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1070 multiline ? FBMrf_MULTILINE : 0
1073 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1074 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1075 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1076 s ? "Found" : "Contradicts",
1077 other_ix ? "floating" : "anchored",
1078 quoted, RE_SV_TAIL(must));
1083 /* last1 is latest possible substr location. If we didn't
1084 * find it before there, we never will */
1085 if (last >= last1) {
1086 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1087 ", giving up...\n"));
1091 /* try to find the check substr again at a later
1092 * position. Maybe next time we'll find the "other" substr
1094 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1095 ", trying %s at offset %ld...\n",
1096 (other_ix ? "floating" : "anchored"),
1097 (long)(HOP3c(check_at, 1, strend) - strpos)));
1099 other_last = HOP3c(last, 1, strend) /* highest failure */;
1101 other_ix /* i.e. if other-is-float */
1102 ? HOP3c(rx_origin, 1, strend)
1103 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1107 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
1108 (long)(s - strpos)));
1110 if (other_ix) { /* if (other-is-float) */
1111 /* other_last is set to s, not s+1, since its possible for
1112 * a floating substr to fail first time, then succeed
1113 * second time at the same floating position; e.g.:
1114 * "-AB--AABZ" =~ /\wAB\d*Z/
1115 * The first time round, anchored and float match at
1116 * "-(AB)--AAB(Z)" then fail on the initial \w character
1117 * class. Second time round, they match at "-AB--A(AB)(Z)".
1122 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1123 other_last = HOP3c(s, 1, strend);
1128 DEBUG_OPTIMISE_MORE_r(
1129 PerlIO_printf(Perl_debug_log,
1130 " Check-only match: offset min:%"IVdf" max:%"IVdf
1131 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
1132 " strend-strpos:%"IVdf"\n",
1133 (IV)prog->check_offset_min,
1134 (IV)prog->check_offset_max,
1135 (IV)(check_at-strpos),
1136 (IV)(rx_origin-strpos),
1137 (IV)(rx_origin-check_at),
1143 postprocess_substr_matches:
1145 /* handle the extra constraint of /^.../m if present */
1147 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1150 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1151 " looking for /^/m anchor"));
1153 /* we have failed the constraint of a \n before rx_origin.
1154 * Find the next \n, if any, even if it's beyond the current
1155 * anchored and/or floating substrings. Whether we should be
1156 * scanning ahead for the next \n or the next substr is debatable.
1157 * On the one hand you'd expect rare substrings to appear less
1158 * often than \n's. On the other hand, searching for \n means
1159 * we're effectively flipping been check_substr and "\n" on each
1160 * iteration as the current "rarest" string candidate, which
1161 * means for example that we'll quickly reject the whole string if
1162 * hasn't got a \n, rather than trying every substr position
1166 s = HOP3c(strend, - prog->minlen, strpos);
1167 if (s <= rx_origin ||
1168 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1170 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1171 " Did not find /%s^%s/m...\n",
1172 PL_colors[0], PL_colors[1]));
1176 /* earliest possible origin is 1 char after the \n.
1177 * (since *rx_origin == '\n', it's safe to ++ here rather than
1178 * HOP(rx_origin, 1)) */
1181 if (prog->substrs->check_ix == 0 /* check is anchored */
1182 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1184 /* Position contradicts check-string; either because
1185 * check was anchored (and thus has no wiggle room),
1186 * or check was float and rx_origin is above the float range */
1187 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1188 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1189 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1193 /* if we get here, the check substr must have been float,
1194 * is in range, and we may or may not have had an anchored
1195 * "other" substr which still contradicts */
1196 assert(prog->substrs->check_ix); /* check is float */
1198 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1199 /* whoops, the anchored "other" substr exists, so we still
1200 * contradict. On the other hand, the float "check" substr
1201 * didn't contradict, so just retry the anchored "other"
1203 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1204 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1205 PL_colors[0], PL_colors[1],
1206 (long)(rx_origin - strpos),
1207 (long)(rx_origin - strpos + prog->anchored_offset)));
1208 goto do_other_substr;
1211 /* success: we don't contradict the found floating substring
1212 * (and there's no anchored substr). */
1213 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1214 " Found /%s^%s/m at offset %ld...\n",
1215 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1218 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1219 " (multiline anchor test skipped)\n"));
1225 /* if we have a starting character class, then test that extra constraint.
1226 * (trie stclasses are too expensive to use here, we are better off to
1227 * leave it to regmatch itself) */
1229 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1230 const U8* const str = (U8*)STRING(progi->regstclass);
1232 /* XXX this value could be pre-computed */
1233 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1234 ? (reginfo->is_utf8_pat
1235 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1236 : STR_LEN(progi->regstclass))
1240 /* latest pos that a matching float substr constrains rx start to */
1241 char *rx_max_float = NULL;
1243 /* if the current rx_origin is anchored, either by satisfying an
1244 * anchored substring constraint, or a /^.../m constraint, then we
1245 * can reject the current origin if the start class isn't found
1246 * at the current position. If we have a float-only match, then
1247 * rx_origin is constrained to a range; so look for the start class
1248 * in that range. if neither, then look for the start class in the
1249 * whole rest of the string */
1251 /* XXX DAPM it's not clear what the minlen test is for, and why
1252 * it's not used in the floating case. Nothing in the test suite
1253 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1254 * Here are some old comments, which may or may not be correct:
1256 * minlen == 0 is possible if regstclass is \b or \B,
1257 * and the fixed substr is ''$.
1258 * Since minlen is already taken into account, rx_origin+1 is
1259 * before strend; accidentally, minlen >= 1 guaranties no false
1260 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1261 * 0) below assumes that regstclass does not come from lookahead...
1262 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1263 * This leaves EXACTF-ish only, which are dealt with in
1267 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1268 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1269 else if (prog->float_substr || prog->float_utf8) {
1270 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1271 endpos= HOP3c(rx_max_float, cl_l, strend);
1276 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1277 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
1278 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1279 (IV)start_shift, (IV)(check_at - strbeg),
1280 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1282 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1285 if (endpos == strend) {
1286 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1287 " Could not match STCLASS...\n") );
1290 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1291 " This position contradicts STCLASS...\n") );
1292 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1293 && !(prog->intflags & PREGf_IMPLICIT))
1296 /* Contradict one of substrings */
1297 if (prog->anchored_substr || prog->anchored_utf8) {
1298 if (prog->substrs->check_ix == 1) { /* check is float */
1299 /* Have both, check_string is floating */
1300 assert(rx_origin + start_shift <= check_at);
1301 if (rx_origin + start_shift != check_at) {
1302 /* not at latest position float substr could match:
1303 * Recheck anchored substring, but not floating.
1304 * The condition above is in bytes rather than
1305 * chars for efficiency. It's conservative, in
1306 * that it errs on the side of doing 'goto
1307 * do_other_substr', where a more accurate
1308 * char-based calculation will be done */
1309 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1310 " Looking for anchored substr starting at offset %ld...\n",
1311 (long)(other_last - strpos)) );
1312 goto do_other_substr;
1320 /* In the presence of ml_anch, we might be able to
1321 * find another \n without breaking the current float
1324 /* strictly speaking this should be HOP3c(..., 1, ...),
1325 * but since we goto a block of code that's going to
1326 * search for the next \n if any, its safe here */
1328 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1329 " Looking for /%s^%s/m starting at offset %ld...\n",
1330 PL_colors[0], PL_colors[1],
1331 (long)(rx_origin - strpos)) );
1332 goto postprocess_substr_matches;
1335 /* strictly speaking this can never be true; but might
1336 * be if we ever allow intuit without substrings */
1337 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1340 rx_origin = rx_max_float;
1343 /* at this point, any matching substrings have been
1344 * contradicted. Start again... */
1346 rx_origin = HOP3c(rx_origin, 1, strend);
1348 /* uses bytes rather than char calculations for efficiency.
1349 * It's conservative: it errs on the side of doing 'goto restart',
1350 * where there is code that does a proper char-based test */
1351 if (rx_origin + start_shift + end_shift > strend) {
1352 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1353 " Could not match STCLASS...\n") );
1356 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1357 " Looking for %s substr starting at offset %ld...\n",
1358 (prog->substrs->check_ix ? "floating" : "anchored"),
1359 (long)(rx_origin + start_shift - strpos)) );
1365 if (rx_origin != s) {
1366 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1367 " By STCLASS: moving %ld --> %ld\n",
1368 (long)(rx_origin - strpos), (long)(s - strpos))
1372 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1373 " Does not contradict STCLASS...\n");
1378 /* Decide whether using the substrings helped */
1380 if (rx_origin != strpos) {
1381 /* Fixed substring is found far enough so that the match
1382 cannot start at strpos. */
1384 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1385 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1388 /* The found rx_origin position does not prohibit matching at
1389 * strpos, so calling intuit didn't gain us anything. Decrement
1390 * the BmUSEFUL() count on the check substring, and if we reach
1392 if (!(prog->intflags & PREGf_NAUGHTY)
1394 prog->check_utf8 /* Could be deleted already */
1395 && --BmUSEFUL(prog->check_utf8) < 0
1396 && (prog->check_utf8 == prog->float_utf8)
1398 prog->check_substr /* Could be deleted already */
1399 && --BmUSEFUL(prog->check_substr) < 0
1400 && (prog->check_substr == prog->float_substr)
1403 /* If flags & SOMETHING - do not do it many times on the same match */
1404 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1405 /* XXX Does the destruction order has to change with utf8_target? */
1406 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1407 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1408 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1409 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1410 check = NULL; /* abort */
1411 /* XXXX This is a remnant of the old implementation. It
1412 looks wasteful, since now INTUIT can use many
1413 other heuristics. */
1414 prog->extflags &= ~RXf_USE_INTUIT;
1418 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1419 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1420 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1424 fail_finish: /* Substring not found */
1425 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1426 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1428 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1429 PL_colors[4], PL_colors[5]));
1434 #define DECL_TRIE_TYPE(scan) \
1435 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1436 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
1437 trie_type = ((scan->flags == EXACT) \
1438 ? (utf8_target ? trie_utf8 : trie_plain) \
1439 : (scan->flags == EXACTFA) \
1440 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1441 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1443 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1446 U8 flags = FOLD_FLAGS_FULL; \
1447 switch (trie_type) { \
1448 case trie_utf8_exactfa_fold: \
1449 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1451 case trie_utf8_fold: \
1452 if ( foldlen>0 ) { \
1453 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1458 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1459 len = UTF8SKIP(uc); \
1460 skiplen = UNISKIP( uvc ); \
1461 foldlen -= skiplen; \
1462 uscan = foldbuf + skiplen; \
1465 case trie_latin_utf8_exactfa_fold: \
1466 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1468 case trie_latin_utf8_fold: \
1469 if ( foldlen>0 ) { \
1470 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1476 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1477 skiplen = UNISKIP( uvc ); \
1478 foldlen -= skiplen; \
1479 uscan = foldbuf + skiplen; \
1483 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1490 charid = trie->charmap[ uvc ]; \
1494 if (widecharmap) { \
1495 SV** const svpp = hv_fetch(widecharmap, \
1496 (char*)&uvc, sizeof(UV), 0); \
1498 charid = (U16)SvIV(*svpp); \
1503 #define DUMP_EXEC_POS(li,s,doutf8) \
1504 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1507 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1511 && (ln == 1 || folder(s, pat_string, ln)) \
1512 && (reginfo->intuit || regtry(reginfo, &s)) )\
1518 #define REXEC_FBC_UTF8_SCAN(CODE) \
1520 while (s < strend) { \
1526 #define REXEC_FBC_SCAN(CODE) \
1528 while (s < strend) { \
1534 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1535 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1537 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1546 #define REXEC_FBC_CLASS_SCAN(COND) \
1547 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1549 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1558 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1559 if (utf8_target) { \
1560 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1563 REXEC_FBC_CLASS_SCAN(COND); \
1566 /* The three macros below are slightly different versions of the same logic.
1568 * The first is for /a and /aa when the target string is UTF-8. This can only
1569 * match ascii, but it must advance based on UTF-8. The other two handle the
1570 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1571 * for the boundary (or non-boundary) between a word and non-word character.
1572 * The utf8 and non-utf8 cases have the same logic, but the details must be
1573 * different. Find the "wordness" of the character just prior to this one, and
1574 * compare it with the wordness of this one. If they differ, we have a
1575 * boundary. At the beginning of the string, pretend that the previous
1576 * character was a new-line.
1578 * All these macros uncleanly have side-effects with each other and outside
1579 * variables. So far it's been too much trouble to clean-up
1581 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1582 * a word character or not.
1583 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1585 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1587 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1588 * are looking for a boundary or for a non-boundary. If we are looking for a
1589 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1590 * see if this tentative match actually works, and if so, to quit the loop
1591 * here. And vice-versa if we are looking for a non-boundary.
1593 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1594 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1595 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1596 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1597 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1598 * complement. But in that branch we complement tmp, meaning that at the
1599 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1600 * which means at the top of the loop in the next iteration, it is
1601 * TEST_NON_UTF8(s-1) */
1602 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1603 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1604 tmp = TEST_NON_UTF8(tmp); \
1605 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1606 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1608 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1615 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1616 * TEST_UTF8 is a macro that for the same input code points returns identically
1617 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1618 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1619 if (s == reginfo->strbeg) { \
1622 else { /* Back-up to the start of the previous character */ \
1623 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1624 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1625 0, UTF8_ALLOW_DEFAULT); \
1627 tmp = TEST_UV(tmp); \
1628 LOAD_UTF8_CHARCLASS_ALNUM(); \
1629 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1630 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1639 /* Like the above two macros. UTF8_CODE is the complete code for handling
1640 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1642 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1643 if (utf8_target) { \
1646 else { /* Not utf8 */ \
1647 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1648 tmp = TEST_NON_UTF8(tmp); \
1649 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1650 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1659 /* Here, things have been set up by the previous code so that tmp is the \
1660 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1661 * utf8ness of the target). We also have to check if this matches against \
1662 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1663 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1665 if (tmp == ! TEST_NON_UTF8('\n')) { \
1672 /* This is the macro to use when we want to see if something that looks like it
1673 * could match, actually does, and if so exits the loop */
1674 #define REXEC_FBC_TRYIT \
1675 if ((reginfo->intuit || regtry(reginfo, &s))) \
1678 /* The only difference between the BOUND and NBOUND cases is that
1679 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1680 * NBOUND. This is accomplished by passing it as either the if or else clause,
1681 * with the other one being empty (PLACEHOLDER is defined as empty).
1683 * The TEST_FOO parameters are for operating on different forms of input, but
1684 * all should be ones that return identically for the same underlying code
1686 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1688 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1689 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1691 #define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1693 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1694 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1696 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1698 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1699 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1701 #define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1703 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1704 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1707 /* We know what class REx starts with. Try to find this position... */
1708 /* if reginfo->intuit, its a dryrun */
1709 /* annoyingly all the vars in this routine have different names from their counterparts
1710 in regmatch. /grrr */
1712 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1713 const char *strend, regmatch_info *reginfo)
1716 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1717 char *pat_string; /* The pattern's exactish string */
1718 char *pat_end; /* ptr to end char of pat_string */
1719 re_fold_t folder; /* Function for computing non-utf8 folds */
1720 const U8 *fold_array; /* array for folding ords < 256 */
1726 I32 tmp = 1; /* Scratch variable? */
1727 const bool utf8_target = reginfo->is_utf8_target;
1728 UV utf8_fold_flags = 0;
1729 const bool is_utf8_pat = reginfo->is_utf8_pat;
1730 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1731 with a result inverts that result, as 0^1 =
1733 _char_class_number classnum;
1735 RXi_GET_DECL(prog,progi);
1737 PERL_ARGS_ASSERT_FIND_BYCLASS;
1739 /* We know what class it must start with. */
1743 REXEC_FBC_UTF8_CLASS_SCAN(
1744 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1747 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1752 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1759 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1760 assert(! is_utf8_pat);
1763 if (is_utf8_pat || utf8_target) {
1764 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1765 goto do_exactf_utf8;
1767 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1768 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1769 goto do_exactf_non_utf8; /* isn't dealt with by these */
1771 case EXACTF: /* This node only generated for non-utf8 patterns */
1772 assert(! is_utf8_pat);
1774 utf8_fold_flags = 0;
1775 goto do_exactf_utf8;
1777 fold_array = PL_fold;
1779 goto do_exactf_non_utf8;
1782 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1783 utf8_fold_flags = FOLDEQ_LOCALE;
1784 goto do_exactf_utf8;
1786 fold_array = PL_fold_locale;
1787 folder = foldEQ_locale;
1788 goto do_exactf_non_utf8;
1792 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1794 goto do_exactf_utf8;
1797 if (is_utf8_pat || utf8_target) {
1798 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1799 goto do_exactf_utf8;
1802 /* Any 'ss' in the pattern should have been replaced by regcomp,
1803 * so we don't have to worry here about this single special case
1804 * in the Latin1 range */
1805 fold_array = PL_fold_latin1;
1806 folder = foldEQ_latin1;
1810 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1811 are no glitches with fold-length differences
1812 between the target string and pattern */
1814 /* The idea in the non-utf8 EXACTF* cases is to first find the
1815 * first character of the EXACTF* node and then, if necessary,
1816 * case-insensitively compare the full text of the node. c1 is the
1817 * first character. c2 is its fold. This logic will not work for
1818 * Unicode semantics and the german sharp ss, which hence should
1819 * not be compiled into a node that gets here. */
1820 pat_string = STRING(c);
1821 ln = STR_LEN(c); /* length to match in octets/bytes */
1823 /* We know that we have to match at least 'ln' bytes (which is the
1824 * same as characters, since not utf8). If we have to match 3
1825 * characters, and there are only 2 availabe, we know without
1826 * trying that it will fail; so don't start a match past the
1827 * required minimum number from the far end */
1828 e = HOP3c(strend, -((SSize_t)ln), s);
1830 if (reginfo->intuit && e < s) {
1831 e = s; /* Due to minlen logic of intuit() */
1835 c2 = fold_array[c1];
1836 if (c1 == c2) { /* If char and fold are the same */
1837 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1840 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1848 /* If one of the operands is in utf8, we can't use the simpler folding
1849 * above, due to the fact that many different characters can have the
1850 * same fold, or portion of a fold, or different- length fold */
1851 pat_string = STRING(c);
1852 ln = STR_LEN(c); /* length to match in octets/bytes */
1853 pat_end = pat_string + ln;
1854 lnc = is_utf8_pat /* length to match in characters */
1855 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1858 /* We have 'lnc' characters to match in the pattern, but because of
1859 * multi-character folding, each character in the target can match
1860 * up to 3 characters (Unicode guarantees it will never exceed
1861 * this) if it is utf8-encoded; and up to 2 if not (based on the
1862 * fact that the Latin 1 folds are already determined, and the
1863 * only multi-char fold in that range is the sharp-s folding to
1864 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1865 * string character. Adjust lnc accordingly, rounding up, so that
1866 * if we need to match at least 4+1/3 chars, that really is 5. */
1867 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1868 lnc = (lnc + expansion - 1) / expansion;
1870 /* As in the non-UTF8 case, if we have to match 3 characters, and
1871 * only 2 are left, it's guaranteed to fail, so don't start a
1872 * match that would require us to go beyond the end of the string
1874 e = HOP3c(strend, -((SSize_t)lnc), s);
1876 if (reginfo->intuit && e < s) {
1877 e = s; /* Due to minlen logic of intuit() */
1880 /* XXX Note that we could recalculate e to stop the loop earlier,
1881 * as the worst case expansion above will rarely be met, and as we
1882 * go along we would usually find that e moves further to the left.
1883 * This would happen only after we reached the point in the loop
1884 * where if there were no expansion we should fail. Unclear if
1885 * worth the expense */
1888 char *my_strend= (char *)strend;
1889 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1890 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1891 && (reginfo->intuit || regtry(reginfo, &s)) )
1895 s += (utf8_target) ? UTF8SKIP(s) : 1;
1901 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1904 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1907 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1910 FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1913 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1916 FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1919 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1922 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1925 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1926 is_LNBREAK_latin1_safe(s, strend)
1930 /* The argument to all the POSIX node types is the class number to pass to
1931 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1938 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1939 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1954 /* The complement of something that matches only ASCII matches all
1955 * non-ASCII, plus everything in ASCII that isn't in the class. */
1956 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
1957 || ! _generic_isCC_A(*s, FLAGS(c)));
1966 /* Don't need to worry about utf8, as it can match only a single
1967 * byte invariant character. */
1968 REXEC_FBC_CLASS_SCAN(
1969 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1977 if (! utf8_target) {
1978 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1984 classnum = (_char_class_number) FLAGS(c);
1985 if (classnum < _FIRST_NON_SWASH_CC) {
1986 while (s < strend) {
1988 /* We avoid loading in the swash as long as possible, but
1989 * should we have to, we jump to a separate loop. This
1990 * extra 'if' statement is what keeps this code from being
1991 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1992 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1993 goto found_above_latin1;
1995 if ((UTF8_IS_INVARIANT(*s)
1996 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1998 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1999 && to_complement ^ cBOOL(
2000 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2004 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2016 else switch (classnum) { /* These classes are implemented as
2018 case _CC_ENUM_SPACE: /* XXX would require separate code if we
2019 revert the change of \v matching this */
2022 case _CC_ENUM_PSXSPC:
2023 REXEC_FBC_UTF8_CLASS_SCAN(
2024 to_complement ^ cBOOL(isSPACE_utf8(s)));
2027 case _CC_ENUM_BLANK:
2028 REXEC_FBC_UTF8_CLASS_SCAN(
2029 to_complement ^ cBOOL(isBLANK_utf8(s)));
2032 case _CC_ENUM_XDIGIT:
2033 REXEC_FBC_UTF8_CLASS_SCAN(
2034 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2037 case _CC_ENUM_VERTSPACE:
2038 REXEC_FBC_UTF8_CLASS_SCAN(
2039 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2042 case _CC_ENUM_CNTRL:
2043 REXEC_FBC_UTF8_CLASS_SCAN(
2044 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2048 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2049 assert(0); /* NOTREACHED */
2054 found_above_latin1: /* Here we have to load a swash to get the result
2055 for the current code point */
2056 if (! PL_utf8_swash_ptrs[classnum]) {
2057 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2058 PL_utf8_swash_ptrs[classnum] =
2059 _core_swash_init("utf8",
2062 PL_XPosix_ptrs[classnum], &flags);
2065 /* This is a copy of the loop above for swash classes, though using the
2066 * FBC macro instead of being expanded out. Since we've loaded the
2067 * swash, we don't have to check for that each time through the loop */
2068 REXEC_FBC_UTF8_CLASS_SCAN(
2069 to_complement ^ cBOOL(_generic_utf8(
2072 swash_fetch(PL_utf8_swash_ptrs[classnum],
2080 /* what trie are we using right now */
2081 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2082 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2083 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2085 const char *last_start = strend - trie->minlen;
2087 const char *real_start = s;
2089 STRLEN maxlen = trie->maxlen;
2091 U8 **points; /* map of where we were in the input string
2092 when reading a given char. For ASCII this
2093 is unnecessary overhead as the relationship
2094 is always 1:1, but for Unicode, especially
2095 case folded Unicode this is not true. */
2096 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2100 GET_RE_DEBUG_FLAGS_DECL;
2102 /* We can't just allocate points here. We need to wrap it in
2103 * an SV so it gets freed properly if there is a croak while
2104 * running the match */
2107 sv_points=newSV(maxlen * sizeof(U8 *));
2108 SvCUR_set(sv_points,
2109 maxlen * sizeof(U8 *));
2110 SvPOK_on(sv_points);
2111 sv_2mortal(sv_points);
2112 points=(U8**)SvPV_nolen(sv_points );
2113 if ( trie_type != trie_utf8_fold
2114 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2117 bitmap=(U8*)trie->bitmap;
2119 bitmap=(U8*)ANYOF_BITMAP(c);
2121 /* this is the Aho-Corasick algorithm modified a touch
2122 to include special handling for long "unknown char" sequences.
2123 The basic idea being that we use AC as long as we are dealing
2124 with a possible matching char, when we encounter an unknown char
2125 (and we have not encountered an accepting state) we scan forward
2126 until we find a legal starting char.
2127 AC matching is basically that of trie matching, except that when
2128 we encounter a failing transition, we fall back to the current
2129 states "fail state", and try the current char again, a process
2130 we repeat until we reach the root state, state 1, or a legal
2131 transition. If we fail on the root state then we can either
2132 terminate if we have reached an accepting state previously, or
2133 restart the entire process from the beginning if we have not.
2136 while (s <= last_start) {
2137 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2145 U8 *uscan = (U8*)NULL;
2146 U8 *leftmost = NULL;
2148 U32 accepted_word= 0;
2152 while ( state && uc <= (U8*)strend ) {
2154 U32 word = aho->states[ state ].wordnum;
2158 DEBUG_TRIE_EXECUTE_r(
2159 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2160 dump_exec_pos( (char *)uc, c, strend, real_start,
2161 (char *)uc, utf8_target );
2162 PerlIO_printf( Perl_debug_log,
2163 " Scanning for legal start char...\n");
2167 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2171 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2177 if (uc >(U8*)last_start) break;
2181 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2182 if (!leftmost || lpos < leftmost) {
2183 DEBUG_r(accepted_word=word);
2189 points[pointpos++ % maxlen]= uc;
2190 if (foldlen || uc < (U8*)strend) {
2191 REXEC_TRIE_READ_CHAR(trie_type, trie,
2193 uscan, len, uvc, charid, foldlen,
2195 DEBUG_TRIE_EXECUTE_r({
2196 dump_exec_pos( (char *)uc, c, strend,
2197 real_start, s, utf8_target);
2198 PerlIO_printf(Perl_debug_log,
2199 " Charid:%3u CP:%4"UVxf" ",
2211 word = aho->states[ state ].wordnum;
2213 base = aho->states[ state ].trans.base;
2215 DEBUG_TRIE_EXECUTE_r({
2217 dump_exec_pos( (char *)uc, c, strend, real_start,
2219 PerlIO_printf( Perl_debug_log,
2220 "%sState: %4"UVxf", word=%"UVxf,
2221 failed ? " Fail transition to " : "",
2222 (UV)state, (UV)word);
2228 ( ((offset = base + charid
2229 - 1 - trie->uniquecharcount)) >= 0)
2230 && ((U32)offset < trie->lasttrans)
2231 && trie->trans[offset].check == state
2232 && (tmp=trie->trans[offset].next))
2234 DEBUG_TRIE_EXECUTE_r(
2235 PerlIO_printf( Perl_debug_log," - legal\n"));
2240 DEBUG_TRIE_EXECUTE_r(
2241 PerlIO_printf( Perl_debug_log," - fail\n"));
2243 state = aho->fail[state];
2247 /* we must be accepting here */
2248 DEBUG_TRIE_EXECUTE_r(
2249 PerlIO_printf( Perl_debug_log," - accepting\n"));
2258 if (!state) state = 1;
2261 if ( aho->states[ state ].wordnum ) {
2262 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2263 if (!leftmost || lpos < leftmost) {
2264 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2269 s = (char*)leftmost;
2270 DEBUG_TRIE_EXECUTE_r({
2272 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2273 (UV)accepted_word, (IV)(s - real_start)
2276 if (reginfo->intuit || regtry(reginfo, &s)) {
2282 DEBUG_TRIE_EXECUTE_r({
2283 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2286 DEBUG_TRIE_EXECUTE_r(
2287 PerlIO_printf( Perl_debug_log,"No match.\n"));
2296 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2303 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2304 * flags have same meanings as with regexec_flags() */
2307 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2314 struct regexp *const prog = ReANY(rx);
2316 if (flags & REXEC_COPY_STR) {
2320 PerlIO_printf(Perl_debug_log,
2321 "Copy on write: regexp capture, type %d\n",
2324 /* Create a new COW SV to share the match string and store
2325 * in saved_copy, unless the current COW SV in saved_copy
2326 * is valid and suitable for our purpose */
2327 if (( prog->saved_copy
2328 && SvIsCOW(prog->saved_copy)
2329 && SvPOKp(prog->saved_copy)
2332 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2334 /* just reuse saved_copy SV */
2335 if (RXp_MATCH_COPIED(prog)) {
2336 Safefree(prog->subbeg);
2337 RXp_MATCH_COPIED_off(prog);
2341 /* create new COW SV to share string */
2342 RX_MATCH_COPY_FREE(rx);
2343 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2345 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2346 assert (SvPOKp(prog->saved_copy));
2347 prog->sublen = strend - strbeg;
2348 prog->suboffset = 0;
2349 prog->subcoffset = 0;
2354 SSize_t max = strend - strbeg;
2357 if ( (flags & REXEC_COPY_SKIP_POST)
2358 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2359 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2360 ) { /* don't copy $' part of string */
2363 /* calculate the right-most part of the string covered
2364 * by a capture. Due to look-ahead, this may be to
2365 * the right of $&, so we have to scan all captures */
2366 while (n <= prog->lastparen) {
2367 if (prog->offs[n].end > max)
2368 max = prog->offs[n].end;
2372 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2373 ? prog->offs[0].start
2375 assert(max >= 0 && max <= strend - strbeg);
2378 if ( (flags & REXEC_COPY_SKIP_PRE)
2379 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2380 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2381 ) { /* don't copy $` part of string */
2384 /* calculate the left-most part of the string covered
2385 * by a capture. Due to look-behind, this may be to
2386 * the left of $&, so we have to scan all captures */
2387 while (min && n <= prog->lastparen) {
2388 if ( prog->offs[n].start != -1
2389 && prog->offs[n].start < min)
2391 min = prog->offs[n].start;
2395 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2396 && min > prog->offs[0].end
2398 min = prog->offs[0].end;
2402 assert(min >= 0 && min <= max && min <= strend - strbeg);
2405 if (RX_MATCH_COPIED(rx)) {
2406 if (sublen > prog->sublen)
2408 (char*)saferealloc(prog->subbeg, sublen+1);
2411 prog->subbeg = (char*)safemalloc(sublen+1);
2412 Copy(strbeg + min, prog->subbeg, sublen, char);
2413 prog->subbeg[sublen] = '\0';
2414 prog->suboffset = min;
2415 prog->sublen = sublen;
2416 RX_MATCH_COPIED_on(rx);
2418 prog->subcoffset = prog->suboffset;
2419 if (prog->suboffset && utf8_target) {
2420 /* Convert byte offset to chars.
2421 * XXX ideally should only compute this if @-/@+
2422 * has been seen, a la PL_sawampersand ??? */
2424 /* If there's a direct correspondence between the
2425 * string which we're matching and the original SV,
2426 * then we can use the utf8 len cache associated with
2427 * the SV. In particular, it means that under //g,
2428 * sv_pos_b2u() will use the previously cached
2429 * position to speed up working out the new length of
2430 * subcoffset, rather than counting from the start of
2431 * the string each time. This stops
2432 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2433 * from going quadratic */
2434 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2435 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2436 SV_GMAGIC|SV_CONST_RETURN);
2438 prog->subcoffset = utf8_length((U8*)strbeg,
2439 (U8*)(strbeg+prog->suboffset));
2443 RX_MATCH_COPY_FREE(rx);
2444 prog->subbeg = strbeg;
2445 prog->suboffset = 0;
2446 prog->subcoffset = 0;
2447 prog->sublen = strend - strbeg;
2455 - regexec_flags - match a regexp against a string
2458 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2459 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2460 /* stringarg: the point in the string at which to begin matching */
2461 /* strend: pointer to null at end of string */
2462 /* strbeg: real beginning of string */
2463 /* minend: end of match must be >= minend bytes after stringarg. */
2464 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2465 * itself is accessed via the pointers above */
2466 /* data: May be used for some additional optimizations.
2467 Currently unused. */
2468 /* flags: For optimizations. See REXEC_* in regexp.h */
2471 struct regexp *const prog = ReANY(rx);
2475 SSize_t minlen; /* must match at least this many chars */
2476 SSize_t dontbother = 0; /* how many characters not to try at end */
2477 const bool utf8_target = cBOOL(DO_UTF8(sv));
2479 RXi_GET_DECL(prog,progi);
2480 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2481 regmatch_info *const reginfo = ®info_buf;
2482 regexp_paren_pair *swap = NULL;
2484 GET_RE_DEBUG_FLAGS_DECL;
2486 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2487 PERL_UNUSED_ARG(data);
2489 /* Be paranoid... */
2490 if (prog == NULL || stringarg == NULL) {
2491 Perl_croak(aTHX_ "NULL regexp parameter");
2495 debug_start_match(rx, utf8_target, stringarg, strend,
2499 startpos = stringarg;
2501 if (prog->intflags & PREGf_GPOS_SEEN) {
2504 /* set reginfo->ganch, the position where \G can match */
2507 (flags & REXEC_IGNOREPOS)
2508 ? stringarg /* use start pos rather than pos() */
2509 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2510 /* Defined pos(): */
2511 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2512 : strbeg; /* pos() not defined; use start of string */
2514 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2515 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
2517 /* in the presence of \G, we may need to start looking earlier in
2518 * the string than the suggested start point of stringarg:
2519 * if prog->gofs is set, then that's a known, fixed minimum
2522 * /ab|c\G/: gofs = 1
2523 * or if the minimum offset isn't known, then we have to go back
2524 * to the start of the string, e.g. /w+\G/
2527 if (prog->intflags & PREGf_ANCH_GPOS) {
2528 startpos = reginfo->ganch - prog->gofs;
2530 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2532 DEBUG_r(PerlIO_printf(Perl_debug_log,
2533 "fail: ganch-gofs before earliest possible start\n"));
2537 else if (prog->gofs) {
2538 if (startpos - prog->gofs < strbeg)
2541 startpos -= prog->gofs;
2543 else if (prog->intflags & PREGf_GPOS_FLOAT)
2547 minlen = prog->minlen;
2548 if ((startpos + minlen) > strend || startpos < strbeg) {
2549 DEBUG_r(PerlIO_printf(Perl_debug_log,
2550 "Regex match can't succeed, so not even tried\n"));
2554 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2555 * which will call destuctors to reset PL_regmatch_state, free higher
2556 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2557 * regmatch_info_aux_eval */
2559 oldsave = PL_savestack_ix;
2563 if ((prog->extflags & RXf_USE_INTUIT)
2564 && !(flags & REXEC_CHECKED))
2566 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2571 if (prog->extflags & RXf_CHECK_ALL) {
2572 /* we can match based purely on the result of INTUIT.
2573 * Set up captures etc just for $& and $-[0]
2574 * (an intuit-only match wont have $1,$2,..) */
2575 assert(!prog->nparens);
2577 /* s/// doesn't like it if $& is earlier than where we asked it to
2578 * start searching (which can happen on something like /.\G/) */
2579 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2582 /* this should only be possible under \G */
2583 assert(prog->intflags & PREGf_GPOS_SEEN);
2584 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2585 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2589 /* match via INTUIT shouldn't have any captures.
2590 * Let @-, @+, $^N know */
2591 prog->lastparen = prog->lastcloseparen = 0;
2592 RX_MATCH_UTF8_set(rx, utf8_target);
2593 prog->offs[0].start = s - strbeg;
2594 prog->offs[0].end = utf8_target
2595 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2596 : s - strbeg + prog->minlenret;
2597 if ( !(flags & REXEC_NOT_FIRST) )
2598 S_reg_set_capture_string(aTHX_ rx,
2600 sv, flags, utf8_target);
2606 multiline = prog->extflags & RXf_PMf_MULTILINE;
2608 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2609 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2610 "String too short [regexec_flags]...\n"));
2614 /* Check validity of program. */
2615 if (UCHARAT(progi->program) != REG_MAGIC) {
2616 Perl_croak(aTHX_ "corrupted regexp program");
2619 RX_MATCH_TAINTED_off(rx);
2621 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2622 reginfo->intuit = 0;
2623 reginfo->is_utf8_target = cBOOL(utf8_target);
2624 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2625 reginfo->warned = FALSE;
2626 reginfo->strbeg = strbeg;
2628 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2629 reginfo->strend = strend;
2630 /* see how far we have to get to not match where we matched before */
2631 reginfo->till = stringarg + minend;
2633 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
2634 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2635 S_cleanup_regmatch_info_aux has executed (registered by
2636 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2637 magic belonging to this SV.
2638 Not newSVsv, either, as it does not COW.
2640 reginfo->sv = newSV(0);
2641 SvSetSV_nosteal(reginfo->sv, sv);
2642 SAVEFREESV(reginfo->sv);
2645 /* reserve next 2 or 3 slots in PL_regmatch_state:
2646 * slot N+0: may currently be in use: skip it
2647 * slot N+1: use for regmatch_info_aux struct
2648 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2649 * slot N+3: ready for use by regmatch()
2653 regmatch_state *old_regmatch_state;
2654 regmatch_slab *old_regmatch_slab;
2655 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2657 /* on first ever match, allocate first slab */
2658 if (!PL_regmatch_slab) {
2659 Newx(PL_regmatch_slab, 1, regmatch_slab);
2660 PL_regmatch_slab->prev = NULL;
2661 PL_regmatch_slab->next = NULL;
2662 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2665 old_regmatch_state = PL_regmatch_state;
2666 old_regmatch_slab = PL_regmatch_slab;
2668 for (i=0; i <= max; i++) {
2670 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2672 reginfo->info_aux_eval =
2673 reginfo->info_aux->info_aux_eval =
2674 &(PL_regmatch_state->u.info_aux_eval);
2676 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2677 PL_regmatch_state = S_push_slab(aTHX);
2680 /* note initial PL_regmatch_state position; at end of match we'll
2681 * pop back to there and free any higher slabs */
2683 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2684 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2685 reginfo->info_aux->poscache = NULL;
2687 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2689 if ((prog->extflags & RXf_EVAL_SEEN))
2690 S_setup_eval_state(aTHX_ reginfo);
2692 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2695 /* If there is a "must appear" string, look for it. */
2697 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2698 /* We have to be careful. If the previous successful match
2699 was from this regex we don't want a subsequent partially
2700 successful match to clobber the old results.
2701 So when we detect this possibility we add a swap buffer
2702 to the re, and switch the buffer each match. If we fail,
2703 we switch it back; otherwise we leave it swapped.
2706 /* do we need a save destructor here for eval dies? */
2707 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2708 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2709 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2716 /* Simplest case: anchored match need be tried only once. */
2717 /* [unless only anchor is MBOL - implying multiline is set] */
2718 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
2719 if (s == startpos && regtry(reginfo, &s))
2721 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
2726 dontbother = minlen - 1;
2727 end = HOP3c(strend, -dontbother, strbeg) - 1;
2728 /* for multiline we only have to try after newlines */
2729 if (prog->check_substr || prog->check_utf8) {
2730 /* because of the goto we can not easily reuse the macros for bifurcating the
2731 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2734 goto after_try_utf8;
2736 if (regtry(reginfo, &s)) {
2743 if (prog->extflags & RXf_USE_INTUIT) {
2744 s = re_intuit_start(rx, sv, strbeg,
2745 s + UTF8SKIP(s), strend, flags, NULL);
2754 } /* end search for check string in unicode */
2756 if (s == startpos) {
2757 goto after_try_latin;
2760 if (regtry(reginfo, &s)) {
2767 if (prog->extflags & RXf_USE_INTUIT) {
2768 s = re_intuit_start(rx, sv, strbeg,
2769 s + 1, strend, flags, NULL);
2778 } /* end search for check string in latin*/
2779 } /* end search for check string */
2780 else { /* search for newline */
2782 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2785 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2786 while (s <= end) { /* note it could be possible to match at the end of the string */
2787 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2788 if (regtry(reginfo, &s))
2792 } /* end search for newline */
2793 } /* end anchored/multiline check string search */
2795 } else if (prog->intflags & PREGf_ANCH_GPOS)
2797 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2798 assert(prog->intflags & PREGf_GPOS_SEEN);
2799 /* For anchored \G, the only position it can match from is
2800 * (ganch-gofs); we already set startpos to this above; if intuit
2801 * moved us on from there, we can't possibly succeed */
2802 assert(startpos == reginfo->ganch - prog->gofs);
2803 if (s == startpos && regtry(reginfo, &s))
2808 /* Messy cases: unanchored match. */
2809 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2810 /* we have /x+whatever/ */
2811 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2817 if (! prog->anchored_utf8) {
2818 to_utf8_substr(prog);
2820 ch = SvPVX_const(prog->anchored_utf8)[0];
2823 DEBUG_EXECUTE_r( did_match = 1 );
2824 if (regtry(reginfo, &s)) goto got_it;
2826 while (s < strend && *s == ch)
2833 if (! prog->anchored_substr) {
2834 if (! to_byte_substr(prog)) {
2835 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2838 ch = SvPVX_const(prog->anchored_substr)[0];
2841 DEBUG_EXECUTE_r( did_match = 1 );
2842 if (regtry(reginfo, &s)) goto got_it;
2844 while (s < strend && *s == ch)
2849 DEBUG_EXECUTE_r(if (!did_match)
2850 PerlIO_printf(Perl_debug_log,
2851 "Did not find anchored character...\n")
2854 else if (prog->anchored_substr != NULL
2855 || prog->anchored_utf8 != NULL
2856 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2857 && prog->float_max_offset < strend - s)) {
2862 char *last1; /* Last position checked before */
2866 if (prog->anchored_substr || prog->anchored_utf8) {
2868 if (! prog->anchored_utf8) {
2869 to_utf8_substr(prog);
2871 must = prog->anchored_utf8;
2874 if (! prog->anchored_substr) {
2875 if (! to_byte_substr(prog)) {
2876 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2879 must = prog->anchored_substr;
2881 back_max = back_min = prog->anchored_offset;
2884 if (! prog->float_utf8) {
2885 to_utf8_substr(prog);
2887 must = prog->float_utf8;
2890 if (! prog->float_substr) {
2891 if (! to_byte_substr(prog)) {
2892 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2895 must = prog->float_substr;
2897 back_max = prog->float_max_offset;
2898 back_min = prog->float_min_offset;
2904 last = HOP3c(strend, /* Cannot start after this */
2905 -(SSize_t)(CHR_SVLEN(must)
2906 - (SvTAIL(must) != 0) + back_min), strbeg);
2908 if (s > reginfo->strbeg)
2909 last1 = HOPc(s, -1);
2911 last1 = s - 1; /* bogus */
2913 /* XXXX check_substr already used to find "s", can optimize if
2914 check_substr==must. */
2916 strend = HOPc(strend, -dontbother);
2917 while ( (s <= last) &&
2918 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
2919 (unsigned char*)strend, must,
2920 multiline ? FBMrf_MULTILINE : 0)) ) {
2921 DEBUG_EXECUTE_r( did_match = 1 );
2922 if (HOPc(s, -back_max) > last1) {
2923 last1 = HOPc(s, -back_min);
2924 s = HOPc(s, -back_max);
2927 char * const t = (last1 >= reginfo->strbeg)
2928 ? HOPc(last1, 1) : last1 + 1;
2930 last1 = HOPc(s, -back_min);
2934 while (s <= last1) {
2935 if (regtry(reginfo, &s))
2938 s++; /* to break out of outer loop */
2945 while (s <= last1) {
2946 if (regtry(reginfo, &s))
2952 DEBUG_EXECUTE_r(if (!did_match) {
2953 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2954 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2955 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2956 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2957 ? "anchored" : "floating"),
2958 quoted, RE_SV_TAIL(must));
2962 else if ( (c = progi->regstclass) ) {
2964 const OPCODE op = OP(progi->regstclass);
2965 /* don't bother with what can't match */
2966 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2967 strend = HOPc(strend, -(minlen - 1));
2970 SV * const prop = sv_newmortal();
2971 regprop(prog, prop, c, reginfo);
2973 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2975 PerlIO_printf(Perl_debug_log,
2976 "Matching stclass %.*s against %s (%d bytes)\n",
2977 (int)SvCUR(prop), SvPVX_const(prop),
2978 quoted, (int)(strend - s));
2981 if (find_byclass(prog, c, s, strend, reginfo))
2983 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2987 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2995 if (! prog->float_utf8) {
2996 to_utf8_substr(prog);
2998 float_real = prog->float_utf8;
3001 if (! prog->float_substr) {
3002 if (! to_byte_substr(prog)) {
3003 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3006 float_real = prog->float_substr;
3009 little = SvPV_const(float_real, len);
3010 if (SvTAIL(float_real)) {
3011 /* This means that float_real contains an artificial \n on
3012 * the end due to the presence of something like this:
3013 * /foo$/ where we can match both "foo" and "foo\n" at the
3014 * end of the string. So we have to compare the end of the
3015 * string first against the float_real without the \n and
3016 * then against the full float_real with the string. We
3017 * have to watch out for cases where the string might be
3018 * smaller than the float_real or the float_real without
3020 char *checkpos= strend - len;
3022 PerlIO_printf(Perl_debug_log,
3023 "%sChecking for float_real.%s\n",
3024 PL_colors[4], PL_colors[5]));
3025 if (checkpos + 1 < strbeg) {
3026 /* can't match, even if we remove the trailing \n
3027 * string is too short to match */
3029 PerlIO_printf(Perl_debug_log,
3030 "%sString shorter than required trailing substring, cannot match.%s\n",
3031 PL_colors[4], PL_colors[5]));
3033 } else if (memEQ(checkpos + 1, little, len - 1)) {
3034 /* can match, the end of the string matches without the
3036 last = checkpos + 1;
3037 } else if (checkpos < strbeg) {
3038 /* cant match, string is too short when the "\n" is
3041 PerlIO_printf(Perl_debug_log,
3042 "%sString does not contain required trailing substring, cannot match.%s\n",
3043 PL_colors[4], PL_colors[5]));
3045 } else if (!multiline) {
3046 /* non multiline match, so compare with the "\n" at the
3047 * end of the string */
3048 if (memEQ(checkpos, little, len)) {
3052 PerlIO_printf(Perl_debug_log,
3053 "%sString does not contain required trailing substring, cannot match.%s\n",
3054 PL_colors[4], PL_colors[5]));
3058 /* multiline match, so we have to search for a place
3059 * where the full string is located */
3065 last = rninstr(s, strend, little, little + len);
3067 last = strend; /* matching "$" */
3070 /* at one point this block contained a comment which was
3071 * probably incorrect, which said that this was a "should not
3072 * happen" case. Even if it was true when it was written I am
3073 * pretty sure it is not anymore, so I have removed the comment
3074 * and replaced it with this one. Yves */
3076 PerlIO_printf(Perl_debug_log,
3077 "String does not contain required substring, cannot match.\n"
3081 dontbother = strend - last + prog->float_min_offset;
3083 if (minlen && (dontbother < minlen))
3084 dontbother = minlen - 1;
3085 strend -= dontbother; /* this one's always in bytes! */
3086 /* We don't know much -- general case. */
3089 if (regtry(reginfo, &s))
3098 if (regtry(reginfo, &s))
3100 } while (s++ < strend);
3108 /* s/// doesn't like it if $& is earlier than where we asked it to
3109 * start searching (which can happen on something like /.\G/) */
3110 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3111 && (prog->offs[0].start < stringarg - strbeg))
3113 /* this should only be possible under \G */
3114 assert(prog->intflags & PREGf_GPOS_SEEN);
3115 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3116 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3122 PerlIO_printf(Perl_debug_log,
3123 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3130 /* clean up; this will trigger destructors that will free all slabs
3131 * above the current one, and cleanup the regmatch_info_aux
3132 * and regmatch_info_aux_eval sructs */
3134 LEAVE_SCOPE(oldsave);
3136 if (RXp_PAREN_NAMES(prog))
3137 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3139 RX_MATCH_UTF8_set(rx, utf8_target);
3141 /* make sure $`, $&, $', and $digit will work later */
3142 if ( !(flags & REXEC_NOT_FIRST) )
3143 S_reg_set_capture_string(aTHX_ rx,
3144 strbeg, reginfo->strend,
3145 sv, flags, utf8_target);
3150 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
3151 PL_colors[4], PL_colors[5]));
3153 /* clean up; this will trigger destructors that will free all slabs
3154 * above the current one, and cleanup the regmatch_info_aux
3155 * and regmatch_info_aux_eval sructs */
3157 LEAVE_SCOPE(oldsave);
3160 /* we failed :-( roll it back */
3161 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3162 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3167 Safefree(prog->offs);
3174 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3175 * Do inc before dec, in case old and new rex are the same */
3176 #define SET_reg_curpm(Re2) \
3177 if (reginfo->info_aux_eval) { \
3178 (void)ReREFCNT_inc(Re2); \
3179 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3180 PM_SETRE((PL_reg_curpm), (Re2)); \
3185 - regtry - try match at specific point
3187 STATIC I32 /* 0 failure, 1 success */
3188 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3191 REGEXP *const rx = reginfo->prog;
3192 regexp *const prog = ReANY(rx);
3194 RXi_GET_DECL(prog,progi);
3195 GET_RE_DEBUG_FLAGS_DECL;
3197 PERL_ARGS_ASSERT_REGTRY;
3199 reginfo->cutpoint=NULL;
3201 prog->offs[0].start = *startposp - reginfo->strbeg;
3202 prog->lastparen = 0;
3203 prog->lastcloseparen = 0;
3205 /* XXXX What this code is doing here?!!! There should be no need
3206 to do this again and again, prog->lastparen should take care of
3209 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3210 * Actually, the code in regcppop() (which Ilya may be meaning by
3211 * prog->lastparen), is not needed at all by the test suite
3212 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3213 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3214 * Meanwhile, this code *is* needed for the
3215 * above-mentioned test suite tests to succeed. The common theme
3216 * on those tests seems to be returning null fields from matches.
3217 * --jhi updated by dapm */
3219 if (prog->nparens) {
3220 regexp_paren_pair *pp = prog->offs;
3222 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3230 result = regmatch(reginfo, *startposp, progi->program + 1);
3232 prog->offs[0].end = result;
3235 if (reginfo->cutpoint)
3236 *startposp= reginfo->cutpoint;
3237 REGCP_UNWIND(lastcp);
3242 #define sayYES goto yes
3243 #define sayNO goto no
3244 #define sayNO_SILENT goto no_silent
3246 /* we dont use STMT_START/END here because it leads to
3247 "unreachable code" warnings, which are bogus, but distracting. */
3248 #define CACHEsayNO \
3249 if (ST.cache_mask) \
3250 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3253 /* this is used to determine how far from the left messages like
3254 'failed...' are printed. It should be set such that messages
3255 are inline with the regop output that created them.
3257 #define REPORT_CODE_OFF 32
3260 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3261 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3262 #define CHRTEST_NOT_A_CP_1 -999
3263 #define CHRTEST_NOT_A_CP_2 -998
3265 /* grab a new slab and return the first slot in it */
3267 STATIC regmatch_state *
3270 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3273 regmatch_slab *s = PL_regmatch_slab->next;
3275 Newx(s, 1, regmatch_slab);
3276 s->prev = PL_regmatch_slab;
3278 PL_regmatch_slab->next = s;
3280 PL_regmatch_slab = s;
3281 return SLAB_FIRST(s);
3285 /* push a new state then goto it */
3287 #define PUSH_STATE_GOTO(state, node, input) \
3288 pushinput = input; \
3290 st->resume_state = state; \
3293 /* push a new state with success backtracking, then goto it */
3295 #define PUSH_YES_STATE_GOTO(state, node, input) \
3296 pushinput = input; \
3298 st->resume_state = state; \
3299 goto push_yes_state;
3306 regmatch() - main matching routine
3308 This is basically one big switch statement in a loop. We execute an op,
3309 set 'next' to point the next op, and continue. If we come to a point which
3310 we may need to backtrack to on failure such as (A|B|C), we push a
3311 backtrack state onto the backtrack stack. On failure, we pop the top
3312 state, and re-enter the loop at the state indicated. If there are no more
3313 states to pop, we return failure.
3315 Sometimes we also need to backtrack on success; for example /A+/, where
3316 after successfully matching one A, we need to go back and try to
3317 match another one; similarly for lookahead assertions: if the assertion
3318 completes successfully, we backtrack to the state just before the assertion
3319 and then carry on. In these cases, the pushed state is marked as
3320 'backtrack on success too'. This marking is in fact done by a chain of
3321 pointers, each pointing to the previous 'yes' state. On success, we pop to
3322 the nearest yes state, discarding any intermediate failure-only states.
3323 Sometimes a yes state is pushed just to force some cleanup code to be
3324 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3325 it to free the inner regex.
3327 Note that failure backtracking rewinds the cursor position, while
3328 success backtracking leaves it alone.
3330 A pattern is complete when the END op is executed, while a subpattern
3331 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3332 ops trigger the "pop to last yes state if any, otherwise return true"
3335 A common convention in this function is to use A and B to refer to the two
3336 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3337 the subpattern to be matched possibly multiple times, while B is the entire
3338 rest of the pattern. Variable and state names reflect this convention.
3340 The states in the main switch are the union of ops and failure/success of
3341 substates associated with with that op. For example, IFMATCH is the op
3342 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3343 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3344 successfully matched A and IFMATCH_A_fail is a state saying that we have
3345 just failed to match A. Resume states always come in pairs. The backtrack
3346 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3347 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3348 on success or failure.
3350 The struct that holds a backtracking state is actually a big union, with
3351 one variant for each major type of op. The variable st points to the
3352 top-most backtrack struct. To make the code clearer, within each
3353 block of code we #define ST to alias the relevant union.
3355 Here's a concrete example of a (vastly oversimplified) IFMATCH
3361 #define ST st->u.ifmatch
3363 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3364 ST.foo = ...; // some state we wish to save
3366 // push a yes backtrack state with a resume value of
3367 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3369 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3372 case IFMATCH_A: // we have successfully executed A; now continue with B
3374 bar = ST.foo; // do something with the preserved value
3377 case IFMATCH_A_fail: // A failed, so the assertion failed
3378 ...; // do some housekeeping, then ...
3379 sayNO; // propagate the failure
3386 For any old-timers reading this who are familiar with the old recursive
3387 approach, the code above is equivalent to:
3389 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3398 ...; // do some housekeeping, then ...
3399 sayNO; // propagate the failure
3402 The topmost backtrack state, pointed to by st, is usually free. If you
3403 want to claim it, populate any ST.foo fields in it with values you wish to
3404 save, then do one of
3406 PUSH_STATE_GOTO(resume_state, node, newinput);
3407 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3409 which sets that backtrack state's resume value to 'resume_state', pushes a
3410 new free entry to the top of the backtrack stack, then goes to 'node'.
3411 On backtracking, the free slot is popped, and the saved state becomes the
3412 new free state. An ST.foo field in this new top state can be temporarily
3413 accessed to retrieve values, but once the main loop is re-entered, it
3414 becomes available for reuse.
3416 Note that the depth of the backtrack stack constantly increases during the
3417 left-to-right execution of the pattern, rather than going up and down with
3418 the pattern nesting. For example the stack is at its maximum at Z at the
3419 end of the pattern, rather than at X in the following:
3421 /(((X)+)+)+....(Y)+....Z/
3423 The only exceptions to this are lookahead/behind assertions and the cut,
3424 (?>A), which pop all the backtrack states associated with A before
3427 Backtrack state structs are allocated in slabs of about 4K in size.
3428 PL_regmatch_state and st always point to the currently active state,
3429 and PL_regmatch_slab points to the slab currently containing
3430 PL_regmatch_state. The first time regmatch() is called, the first slab is
3431 allocated, and is never freed until interpreter destruction. When the slab
3432 is full, a new one is allocated and chained to the end. At exit from
3433 regmatch(), slabs allocated since entry are freed.
3438 #define DEBUG_STATE_pp(pp) \
3440 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3441 PerlIO_printf(Perl_debug_log, \
3442 " %*s"pp" %s%s%s%s%s\n", \
3444 PL_reg_name[st->resume_state], \
3445 ((st==yes_state||st==mark_state) ? "[" : ""), \
3446 ((st==yes_state) ? "Y" : ""), \
3447 ((st==mark_state) ? "M" : ""), \
3448 ((st==yes_state||st==mark_state) ? "]" : "") \
3453 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3458 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3459 const char *start, const char *end, const char *blurb)
3461 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3463 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3468 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3469 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3471 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3472 start, end - start, 60);
3474 PerlIO_printf(Perl_debug_log,
3475 "%s%s REx%s %s against %s\n",
3476 PL_colors[4], blurb, PL_colors[5], s0, s1);
3478 if (utf8_target||utf8_pat)
3479 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3480 utf8_pat ? "pattern" : "",
3481 utf8_pat && utf8_target ? " and " : "",
3482 utf8_target ? "string" : ""
3488 S_dump_exec_pos(pTHX_ const char *locinput,
3489 const regnode *scan,
3490 const char *loc_regeol,
3491 const char *loc_bostr,
3492 const char *loc_reg_starttry,
3493 const bool utf8_target)
3495 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3496 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3497 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3498 /* The part of the string before starttry has one color
3499 (pref0_len chars), between starttry and current
3500 position another one (pref_len - pref0_len chars),
3501 after the current position the third one.
3502 We assume that pref0_len <= pref_len, otherwise we
3503 decrease pref0_len. */
3504 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3505 ? (5 + taill) - l : locinput - loc_bostr;
3508 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3510 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3512 pref0_len = pref_len - (locinput - loc_reg_starttry);
3513 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3514 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3515 ? (5 + taill) - pref_len : loc_regeol - locinput);
3516 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3520 if (pref0_len > pref_len)
3521 pref0_len = pref_len;
3523 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3525 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3526 (locinput - pref_len),pref0_len, 60, 4, 5);
3528 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3529 (locinput - pref_len + pref0_len),
3530 pref_len - pref0_len, 60, 2, 3);
3532 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3533 locinput, loc_regeol - locinput, 10, 0, 1);
3535 const STRLEN tlen=len0+len1+len2;
3536 PerlIO_printf(Perl_debug_log,
3537 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3538 (IV)(locinput - loc_bostr),
3541 (docolor ? "" : "> <"),
3543 (int)(tlen > 19 ? 0 : 19 - tlen),
3550 /* reg_check_named_buff_matched()
3551 * Checks to see if a named buffer has matched. The data array of
3552 * buffer numbers corresponding to the buffer is expected to reside
3553 * in the regexp->data->data array in the slot stored in the ARG() of
3554 * node involved. Note that this routine doesn't actually care about the
3555 * name, that information is not preserved from compilation to execution.
3556 * Returns the index of the leftmost defined buffer with the given name
3557 * or 0 if non of the buffers matched.
3560 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3563 RXi_GET_DECL(rex,rexi);
3564 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3565 I32 *nums=(I32*)SvPVX(sv_dat);
3567 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3569 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3570 if ((I32)rex->lastparen >= nums[n] &&
3571 rex->offs[nums[n]].end != -1)
3581 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3582 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3584 /* This function determines if there are one or two characters that match
3585 * the first character of the passed-in EXACTish node <text_node>, and if
3586 * so, returns them in the passed-in pointers.
3588 * If it determines that no possible character in the target string can
3589 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3590 * the first character in <text_node> requires UTF-8 to represent, and the
3591 * target string isn't in UTF-8.)
3593 * If there are more than two characters that could match the beginning of
3594 * <text_node>, or if more context is required to determine a match or not,
3595 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3597 * The motiviation behind this function is to allow the caller to set up
3598 * tight loops for matching. If <text_node> is of type EXACT, there is
3599 * only one possible character that can match its first character, and so
3600 * the situation is quite simple. But things get much more complicated if
3601 * folding is involved. It may be that the first character of an EXACTFish
3602 * node doesn't participate in any possible fold, e.g., punctuation, so it
3603 * can be matched only by itself. The vast majority of characters that are
3604 * in folds match just two things, their lower and upper-case equivalents.
3605 * But not all are like that; some have multiple possible matches, or match
3606 * sequences of more than one character. This function sorts all that out.
3608 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3609 * loop of trying to match A*, we know we can't exit where the thing
3610 * following it isn't a B. And something can't be a B unless it is the
3611 * beginning of B. By putting a quick test for that beginning in a tight
3612 * loop, we can rule out things that can't possibly be B without having to
3613 * break out of the loop, thus avoiding work. Similarly, if A is a single
3614 * character, we can make a tight loop matching A*, using the outputs of
3617 * If the target string to match isn't in UTF-8, and there aren't
3618 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3619 * the one or two possible octets (which are characters in this situation)
3620 * that can match. In all cases, if there is only one character that can
3621 * match, *<c1p> and *<c2p> will be identical.
3623 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3624 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3625 * can match the beginning of <text_node>. They should be declared with at
3626 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3627 * undefined what these contain.) If one or both of the buffers are
3628 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3629 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3630 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3631 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3632 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3634 const bool utf8_target = reginfo->is_utf8_target;
3636 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
3637 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
3638 bool use_chrtest_void = FALSE;
3639 const bool is_utf8_pat = reginfo->is_utf8_pat;
3641 /* Used when we have both utf8 input and utf8 output, to avoid converting
3642 * to/from code points */
3643 bool utf8_has_been_setup = FALSE;
3647 U8 *pat = (U8*)STRING(text_node);
3648 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
3650 if (OP(text_node) == EXACT) {
3652 /* In an exact node, only one thing can be matched, that first
3653 * character. If both the pat and the target are UTF-8, we can just
3654 * copy the input to the output, avoiding finding the code point of
3659 else if (utf8_target) {
3660 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3661 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3662 utf8_has_been_setup = TRUE;
3665 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3668 else { /* an EXACTFish node */
3669 U8 *pat_end = pat + STR_LEN(text_node);
3671 /* An EXACTFL node has at least some characters unfolded, because what
3672 * they match is not known until now. So, now is the time to fold
3673 * the first few of them, as many as are needed to determine 'c1' and
3674 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
3675 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
3676 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
3677 * need to fold as many characters as a single character can fold to,
3678 * so that later we can check if the first ones are such a multi-char
3679 * fold. But, in such a pattern only locale-problematic characters
3680 * aren't folded, so we can skip this completely if the first character
3681 * in the node isn't one of the tricky ones */
3682 if (OP(text_node) == EXACTFL) {
3684 if (! is_utf8_pat) {
3685 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
3687 folded[0] = folded[1] = 's';
3689 pat_end = folded + 2;
3692 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
3697 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
3699 *(d++) = (U8) toFOLD_LC(*s);
3704 _to_utf8_fold_flags(s,
3707 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
3718 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
3719 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
3721 /* Multi-character folds require more context to sort out. Also
3722 * PL_utf8_foldclosures used below doesn't handle them, so have to
3723 * be handled outside this routine */
3724 use_chrtest_void = TRUE;
3726 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3727 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3729 /* Load the folds hash, if not already done */
3731 if (! PL_utf8_foldclosures) {
3732 _load_PL_utf8_foldclosures();
3735 /* The fold closures data structure is a hash with the keys
3736 * being the UTF-8 of every character that is folded to, like
3737 * 'k', and the values each an array of all code points that
3738 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
3739 * Multi-character folds are not included */
3740 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3745 /* Not found in the hash, therefore there are no folds
3746 * containing it, so there is only a single character that
3750 else { /* Does participate in folds */
3751 AV* list = (AV*) *listp;
3752 if (av_tindex(list) != 1) {
3754 /* If there aren't exactly two folds to this, it is
3755 * outside the scope of this function */
3756 use_chrtest_void = TRUE;
3758 else { /* There are two. Get them */
3759 SV** c_p = av_fetch(list, 0, FALSE);
3761 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3765 c_p = av_fetch(list, 1, FALSE);
3767 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3771 /* Folds that cross the 255/256 boundary are forbidden
3772 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
3773 * one is ASCIII. Since the pattern character is above
3774 * 255, and its only other match is below 256, the only
3775 * legal match will be to itself. We have thrown away
3776 * the original, so have to compute which is the one
3778 if ((c1 < 256) != (c2 < 256)) {
3779 if ((OP(text_node) == EXACTFL
3780 && ! IN_UTF8_CTYPE_LOCALE)
3781 || ((OP(text_node) == EXACTFA
3782 || OP(text_node) == EXACTFA_NO_TRIE)
3783 && (isASCII(c1) || isASCII(c2))))
3796 else /* Here, c1 is <= 255 */
3798 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3799 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
3800 && ((OP(text_node) != EXACTFA
3801 && OP(text_node) != EXACTFA_NO_TRIE)
3804 /* Here, there could be something above Latin1 in the target
3805 * which folds to this character in the pattern. All such
3806 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
3807 * than two characters involved in their folds, so are outside
3808 * the scope of this function */
3809 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3810 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3813 use_chrtest_void = TRUE;
3816 else { /* Here nothing above Latin1 can fold to the pattern
3818 switch (OP(text_node)) {
3820 case EXACTFL: /* /l rules */
3821 c2 = PL_fold_locale[c1];
3824 case EXACTF: /* This node only generated for non-utf8
3826 assert(! is_utf8_pat);
3827 if (! utf8_target) { /* /d rules */
3832 /* /u rules for all these. This happens to work for
3833 * EXACTFA as nothing in Latin1 folds to ASCII */
3834 case EXACTFA_NO_TRIE: /* This node only generated for
3835 non-utf8 patterns */
3836 assert(! is_utf8_pat);
3841 c2 = PL_fold_latin1[c1];
3845 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3846 assert(0); /* NOTREACHED */
3852 /* Here have figured things out. Set up the returns */
3853 if (use_chrtest_void) {
3854 *c2p = *c1p = CHRTEST_VOID;
3856 else if (utf8_target) {
3857 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3858 uvchr_to_utf8(c1_utf8, c1);
3859 uvchr_to_utf8(c2_utf8, c2);
3862 /* Invariants are stored in both the utf8 and byte outputs; Use
3863 * negative numbers otherwise for the byte ones. Make sure that the
3864 * byte ones are the same iff the utf8 ones are the same */
3865 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3866 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3869 ? CHRTEST_NOT_A_CP_1
3870 : CHRTEST_NOT_A_CP_2;
3872 else if (c1 > 255) {
3873 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3878 *c1p = *c2p = c2; /* c2 is the only representable value */
3880 else { /* c1 is representable; see about c2 */
3882 *c2p = (c2 < 256) ? c2 : c1;
3888 /* returns -1 on failure, $+[0] on success */
3890 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3892 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3896 const bool utf8_target = reginfo->is_utf8_target;
3897 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3898 REGEXP *rex_sv = reginfo->prog;
3899 regexp *rex = ReANY(rex_sv);
3900 RXi_GET_DECL(rex,rexi);
3901 /* the current state. This is a cached copy of PL_regmatch_state */
3903 /* cache heavy used fields of st in registers */
3906 U32 n = 0; /* general value; init to avoid compiler warning */
3907 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
3908 char *locinput = startpos;
3909 char *pushinput; /* where to continue after a PUSH */
3910 I32 nextchr; /* is always set to UCHARAT(locinput) */
3912 bool result = 0; /* return value of S_regmatch */
3913 int depth = 0; /* depth of backtrack stack */
3914 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3915 const U32 max_nochange_depth =
3916 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3917 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3918 regmatch_state *yes_state = NULL; /* state to pop to on success of
3920 /* mark_state piggy backs on the yes_state logic so that when we unwind
3921 the stack on success we can update the mark_state as we go */
3922 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3923 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3924 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3926 bool no_final = 0; /* prevent failure from backtracking? */
3927 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3928 char *startpoint = locinput;
3929 SV *popmark = NULL; /* are we looking for a mark? */
3930 SV *sv_commit = NULL; /* last mark name seen in failure */
3931 SV *sv_yes_mark = NULL; /* last mark name we have seen
3932 during a successful match */
3933 U32 lastopen = 0; /* last open we saw */
3934 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3935 SV* const oreplsv = GvSVn(PL_replgv);
3936 /* these three flags are set by various ops to signal information to
3937 * the very next op. They have a useful lifetime of exactly one loop
3938 * iteration, and are not preserved or restored by state pushes/pops
3940 bool sw = 0; /* the condition value in (?(cond)a|b) */
3941 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3942 int logical = 0; /* the following EVAL is:
3946 or the following IFMATCH/UNLESSM is:
3947 false: plain (?=foo)
3948 true: used as a condition: (?(?=foo))
3950 PAD* last_pad = NULL;
3952 I32 gimme = G_SCALAR;
3953 CV *caller_cv = NULL; /* who called us */
3954 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3955 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3956 U32 maxopenparen = 0; /* max '(' index seen so far */
3957 int to_complement; /* Invert the result? */
3958 _char_class_number classnum;
3959 bool is_utf8_pat = reginfo->is_utf8_pat;
3962 GET_RE_DEBUG_FLAGS_DECL;
3965 /* protect against undef(*^R) */
3966 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
3968 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3969 multicall_oldcatch = 0;
3970 multicall_cv = NULL;
3972 PERL_UNUSED_VAR(multicall_cop);
3973 PERL_UNUSED_VAR(newsp);
3976 PERL_ARGS_ASSERT_REGMATCH;
3978 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3979 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3982 st = PL_regmatch_state;
3984 /* Note that nextchr is a byte even in UTF */
3987 while (scan != NULL) {
3990 SV * const prop = sv_newmortal();
3991 regnode *rnext=regnext(scan);
3992 DUMP_EXEC_POS( locinput, scan, utf8_target );
3993 regprop(rex, prop, scan, reginfo);
3995 PerlIO_printf(Perl_debug_log,
3996 "%3"IVdf":%*s%s(%"IVdf")\n",
3997 (IV)(scan - rexi->program), depth*2, "",
3999 (PL_regkind[OP(scan)] == END || !rnext) ?
4000 0 : (IV)(rnext - rexi->program));
4003 next = scan + NEXT_OFF(scan);
4006 state_num = OP(scan);
4012 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
4014 switch (state_num) {
4015 case SBOL: /* /^../ and /\A../ */
4016 if (locinput == reginfo->strbeg)
4020 case MBOL: /* /^../m */
4021 if (locinput == reginfo->strbeg ||
4022 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
4029 if (locinput == reginfo->ganch)
4033 case KEEPS: /* \K */
4034 /* update the startpoint */
4035 st->u.keeper.val = rex->offs[0].start;
4036 rex->offs[0].start = locinput - reginfo->strbeg;
4037 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
4041 case KEEPS_next_fail:
4042 /* rollback the start point change */
4043 rex->offs[0].start = st->u.keeper.val;
4048 case MEOL: /* /..$/m */
4049 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4053 case SEOL: /* /..$/ */
4054 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4056 if (reginfo->strend - locinput > 1)
4061 if (!NEXTCHR_IS_EOS)
4065 case SANY: /* /./s */
4068 goto increment_locinput;
4076 case REG_ANY: /* /./ */
4077 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
4079 goto increment_locinput;
4083 #define ST st->u.trie
4084 case TRIEC: /* (ab|cd) with known charclass */
4085 /* In this case the charclass data is available inline so
4086 we can fail fast without a lot of extra overhead.
4088 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
4090 PerlIO_printf(Perl_debug_log,
4091 "%*s %sfailed to match trie start class...%s\n",
4092 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4099 case TRIE: /* (ab|cd) */
4100 /* the basic plan of execution of the trie is:
4101 * At the beginning, run though all the states, and
4102 * find the longest-matching word. Also remember the position
4103 * of the shortest matching word. For example, this pattern:
4106 * when matched against the string "abcde", will generate
4107 * accept states for all words except 3, with the longest
4108 * matching word being 4, and the shortest being 2 (with
4109 * the position being after char 1 of the string).
4111 * Then for each matching word, in word order (i.e. 1,2,4,5),
4112 * we run the remainder of the pattern; on each try setting
4113 * the current position to the character following the word,
4114 * returning to try the next word on failure.
4116 * We avoid having to build a list of words at runtime by
4117 * using a compile-time structure, wordinfo[].prev, which
4118 * gives, for each word, the previous accepting word (if any).
4119 * In the case above it would contain the mappings 1->2, 2->0,
4120 * 3->0, 4->5, 5->1. We can use this table to generate, from
4121 * the longest word (4 above), a list of all words, by
4122 * following the list of prev pointers; this gives us the
4123 * unordered list 4,5,1,2. Then given the current word we have
4124 * just tried, we can go through the list and find the
4125 * next-biggest word to try (so if we just failed on word 2,
4126 * the next in the list is 4).
4128 * Since at runtime we don't record the matching position in
4129 * the string for each word, we have to work that out for
4130 * each word we're about to process. The wordinfo table holds
4131 * the character length of each word; given that we recorded
4132 * at the start: the position of the shortest word and its
4133 * length in chars, we just need to move the pointer the
4134 * difference between the two char lengths. Depending on
4135 * Unicode status and folding, that's cheap or expensive.
4137 * This algorithm is optimised for the case where are only a
4138 * small number of accept states, i.e. 0,1, or maybe 2.
4139 * With lots of accepts states, and having to try all of them,
4140 * it becomes quadratic on number of accept states to find all
4145 /* what type of TRIE am I? (utf8 makes this contextual) */
4146 DECL_TRIE_TYPE(scan);
4148 /* what trie are we using right now */
4149 reg_trie_data * const trie
4150 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
4151 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
4152 U32 state = trie->startstate;
4155 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
4157 if (trie->states[ state ].wordnum) {
4159 PerlIO_printf(Perl_debug_log,
4160 "%*s %smatched empty string...%s\n",
4161 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4167 PerlIO_printf(Perl_debug_log,
4168 "%*s %sfailed to match trie start class...%s\n",
4169 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4176 U8 *uc = ( U8* )locinput;
4180 U8 *uscan = (U8*)NULL;
4181 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
4182 U32 charcount = 0; /* how many input chars we have matched */
4183 U32 accepted = 0; /* have we seen any accepting states? */
4185 ST.jump = trie->jump;
4188 ST.longfold = FALSE; /* char longer if folded => it's harder */
4191 /* fully traverse the TRIE; note the position of the
4192 shortest accept state and the wordnum of the longest
4195 while ( state && uc <= (U8*)(reginfo->strend) ) {
4196 U32 base = trie->states[ state ].trans.base;
4200 wordnum = trie->states[ state ].wordnum;
4202 if (wordnum) { /* it's an accept state */
4205 /* record first match position */
4207 ST.firstpos = (U8*)locinput;
4212 ST.firstchars = charcount;
4215 if (!ST.nextword || wordnum < ST.nextword)
4216 ST.nextword = wordnum;
4217 ST.topword = wordnum;
4220 DEBUG_TRIE_EXECUTE_r({
4221 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
4222 PerlIO_printf( Perl_debug_log,
4223 "%*s %sState: %4"UVxf" Accepted: %c ",
4224 2+depth * 2, "", PL_colors[4],
4225 (UV)state, (accepted ? 'Y' : 'N'));
4228 /* read a char and goto next state */
4229 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
4231 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
4232 uscan, len, uvc, charid, foldlen,
4239 base + charid - 1 - trie->uniquecharcount)) >= 0)
4241 && ((U32)offset < trie->lasttrans)
4242 && trie->trans[offset].check == state)
4244 state = trie->trans[offset].next;
4255 DEBUG_TRIE_EXECUTE_r(
4256 PerlIO_printf( Perl_debug_log,
4257 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
4258 charid, uvc, (UV)state, PL_colors[5] );
4264 /* calculate total number of accept states */
4269 w = trie->wordinfo[w].prev;
4272 ST.accepted = accepted;
4276 PerlIO_printf( Perl_debug_log,
4277 "%*s %sgot %"IVdf" possible matches%s\n",
4278 REPORT_CODE_OFF + depth * 2, "",
4279 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
4281 goto trie_first_try; /* jump into the fail handler */
4286 case TRIE_next_fail: /* we failed - try next alternative */
4290 REGCP_UNWIND(ST.cp);
4291 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
4293 if (!--ST.accepted) {
4295 PerlIO_printf( Perl_debug_log,
4296 "%*s %sTRIE failed...%s\n",
4297 REPORT_CODE_OFF+depth*2, "",
4304 /* Find next-highest word to process. Note that this code
4305 * is O(N^2) per trie run (O(N) per branch), so keep tight */
4308 U16 const nextword = ST.nextword;
4309 reg_trie_wordinfo * const wordinfo
4310 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
4311 for (word=ST.topword; word; word=wordinfo[word].prev) {
4312 if (word > nextword && (!min || word < min))
4325 ST.lastparen = rex->lastparen;
4326 ST.lastcloseparen = rex->lastcloseparen;
4330 /* find start char of end of current word */
4332 U32 chars; /* how many chars to skip */
4333 reg_trie_data * const trie
4334 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
4336 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
4338 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
4343 /* the hard option - fold each char in turn and find
4344 * its folded length (which may be different */
4345 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4353 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
4361 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4366 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
4382 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4383 ? ST.jump[ST.nextword]
4387 PerlIO_printf( Perl_debug_log,
4388 "%*s %sTRIE matched word #%d, continuing%s\n",
4389 REPORT_CODE_OFF+depth*2, "",
4396 if (ST.accepted > 1 || has_cutgroup) {
4397 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4401 /* only one choice left - just continue */
4403 AV *const trie_words
4404 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4405 SV ** const tmp = av_fetch( trie_words,
4407 SV *sv= tmp ? sv_newmortal() : NULL;
4409 PerlIO_printf( Perl_debug_log,
4410 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4411 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4413 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4414 PL_colors[0], PL_colors[1],
4415 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4417 : "not compiled under -Dr",
4421 locinput = (char*)uc;
4422 continue; /* execute rest of RE */
4428 case EXACT: { /* /abc/ */
4429 char *s = STRING(scan);
4431 if (utf8_target != is_utf8_pat) {
4432 /* The target and the pattern have differing utf8ness. */
4434 const char * const e = s + ln;
4437 /* The target is utf8, the pattern is not utf8.
4438 * Above-Latin1 code points can't match the pattern;
4439 * invariants match exactly, and the other Latin1 ones need
4440 * to be downgraded to a single byte in order to do the
4441 * comparison. (If we could be confident that the target
4442 * is not malformed, this could be refactored to have fewer
4443 * tests by just assuming that if the first bytes match, it
4444 * is an invariant, but there are tests in the test suite
4445 * dealing with (??{...}) which violate this) */
4447 if (l >= reginfo->strend
4448 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4452 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4459 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
4469 /* The target is not utf8, the pattern is utf8. */
4471 if (l >= reginfo->strend
4472 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4476 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4483 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
4495 /* The target and the pattern have the same utf8ness. */
4496 /* Inline the first character, for speed. */
4497 if (reginfo->strend - locinput < ln
4498 || UCHARAT(s) != nextchr
4499 || (ln > 1 && memNE(s, locinput, ln)))
4508 case EXACTFL: { /* /abc/il */
4510 const U8 * fold_array;
4512 U32 fold_utf8_flags;
4514 folder = foldEQ_locale;
4515 fold_array = PL_fold_locale;
4516 fold_utf8_flags = FOLDEQ_LOCALE;
4519 case EXACTFU_SS: /* /\x{df}/iu */
4520 case EXACTFU: /* /abc/iu */
4521 folder = foldEQ_latin1;
4522 fold_array = PL_fold_latin1;
4523 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4526 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
4528 assert(! is_utf8_pat);
4530 case EXACTFA: /* /abc/iaa */
4531 folder = foldEQ_latin1;
4532 fold_array = PL_fold_latin1;
4533 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4536 case EXACTF: /* /abc/i This node only generated for
4537 non-utf8 patterns */
4538 assert(! is_utf8_pat);
4540 fold_array = PL_fold;
4541 fold_utf8_flags = 0;
4549 || state_num == EXACTFU_SS
4550 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
4552 /* Either target or the pattern are utf8, or has the issue where
4553 * the fold lengths may differ. */
4554 const char * const l = locinput;
4555 char *e = reginfo->strend;
4557 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4558 l, &e, 0, utf8_target, fold_utf8_flags))
4566 /* Neither the target nor the pattern are utf8 */
4567 if (UCHARAT(s) != nextchr
4569 && UCHARAT(s) != fold_array[nextchr])
4573 if (reginfo->strend - locinput < ln)
4575 if (ln > 1 && ! folder(s, locinput, ln))
4581 /* XXX Could improve efficiency by separating these all out using a
4582 * macro or in-line function. At that point regcomp.c would no longer
4583 * have to set the FLAGS fields of these */
4584 case BOUNDL: /* /\b/l */
4585 case NBOUNDL: /* /\B/l */
4586 case BOUND: /* /\b/ */
4587 case BOUNDU: /* /\b/u */
4588 case BOUNDA: /* /\b/a */
4589 case NBOUND: /* /\B/ */
4590 case NBOUNDU: /* /\B/u */
4591 case NBOUNDA: /* /\B/a */
4592 /* was last char in word? */
4594 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4595 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4597 if (locinput == reginfo->strbeg)
4600 const U8 * const r =
4601 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4603 ln = utf8n_to_uvchr(r, (U8*) reginfo->strend - r,
4606 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4607 ln = isWORDCHAR_uni(ln);
4611 LOAD_UTF8_CHARCLASS_ALNUM();
4612 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4617 ln = isWORDCHAR_LC_uvchr(ln);
4618 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4623 /* Here the string isn't utf8, or is utf8 and only ascii
4624 * characters are to match \w. In the latter case looking at
4625 * the byte just prior to the current one may be just the final
4626 * byte of a multi-byte character. This is ok. There are two
4628 * 1) it is a single byte character, and then the test is doing
4629 * just what it's supposed to.
4630 * 2) it is a multi-byte character, in which case the final
4631 * byte is never mistakable for ASCII, and so the test
4632 * will say it is not a word character, which is the
4633 * correct answer. */
4634 ln = (locinput != reginfo->strbeg) ?
4635 UCHARAT(locinput - 1) : '\n';
4636 switch (FLAGS(scan)) {
4637 case REGEX_UNICODE_CHARSET:
4638 ln = isWORDCHAR_L1(ln);
4639 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4641 case REGEX_LOCALE_CHARSET:
4642 ln = isWORDCHAR_LC(ln);
4643 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4645 case REGEX_DEPENDS_CHARSET:
4646 ln = isWORDCHAR(ln);
4647 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4649 case REGEX_ASCII_RESTRICTED_CHARSET:
4650 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4651 ln = isWORDCHAR_A(ln);
4652 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4655 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4658 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4660 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4664 case ANYOF: /* /[abc]/ */
4668 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
4671 locinput += UTF8SKIP(locinput);
4674 if (!REGINCLASS(rex, scan, (U8*)locinput))
4680 /* The argument (FLAGS) to all the POSIX node types is the class number
4683 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4687 case POSIXL: /* \w or [:punct:] etc. under /l */
4691 /* Use isFOO_lc() for characters within Latin1. (Note that
4692 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4693 * wouldn't be invariant) */
4694 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4695 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4699 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4700 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4701 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4702 *(locinput + 1))))))
4707 else { /* Here, must be an above Latin-1 code point */
4708 goto utf8_posix_not_eos;
4711 /* Here, must be utf8 */
4712 locinput += UTF8SKIP(locinput);
4715 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4719 case POSIXD: /* \w or [:punct:] etc. under /d */
4725 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4727 if (NEXTCHR_IS_EOS) {
4731 /* All UTF-8 variants match */
4732 if (! UTF8_IS_INVARIANT(nextchr)) {
4733 goto increment_locinput;
4739 case POSIXA: /* \w or [:punct:] etc. under /a */
4742 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4743 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4744 * character is a single byte */
4747 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4753 /* Here we are either not in utf8, or we matched a utf8-invariant,
4754 * so the next char is the next byte */
4758 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4762 case POSIXU: /* \w or [:punct:] etc. under /u */
4764 if (NEXTCHR_IS_EOS) {
4769 /* Use _generic_isCC() for characters within Latin1. (Note that
4770 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4771 * wouldn't be invariant) */
4772 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4773 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4780 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4781 if (! (to_complement
4782 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4790 else { /* Handle above Latin-1 code points */
4791 classnum = (_char_class_number) FLAGS(scan);
4792 if (classnum < _FIRST_NON_SWASH_CC) {
4794 /* Here, uses a swash to find such code points. Load if if
4795 * not done already */
4796 if (! PL_utf8_swash_ptrs[classnum]) {
4797 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4798 PL_utf8_swash_ptrs[classnum]
4799 = _core_swash_init("utf8",
4802 PL_XPosix_ptrs[classnum], &flags);
4804 if (! (to_complement
4805 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4806 (U8 *) locinput, TRUE))))
4811 else { /* Here, uses macros to find above Latin-1 code points */
4813 case _CC_ENUM_SPACE: /* XXX would require separate
4814 code if we revert the change
4815 of \v matching this */
4816 case _CC_ENUM_PSXSPC:
4817 if (! (to_complement
4818 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4823 case _CC_ENUM_BLANK:
4824 if (! (to_complement
4825 ^ cBOOL(is_HORIZWS_high(locinput))))
4830 case _CC_ENUM_XDIGIT:
4831 if (! (to_complement
4832 ^ cBOOL(is_XDIGIT_high(locinput))))
4837 case _CC_ENUM_VERTSPACE:
4838 if (! (to_complement
4839 ^ cBOOL(is_VERTWS_high(locinput))))
4844 default: /* The rest, e.g. [:cntrl:], can't match
4846 if (! to_complement) {
4852 locinput += UTF8SKIP(locinput);
4856 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4857 a Unicode extended Grapheme Cluster */
4858 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4859 extended Grapheme Cluster is:
4862 | Prepend* Begin Extend*
4865 Begin is: ( Special_Begin | ! Control )
4866 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4867 Extend is: ( Grapheme_Extend | Spacing_Mark )
4868 Control is: [ GCB_Control | CR | LF ]
4869 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4871 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4874 Begin is ( Regular_Begin + Special Begin )
4876 It turns out that 98.4% of all Unicode code points match
4877 Regular_Begin. Doing it this way eliminates a table match in
4878 the previous implementation for almost all Unicode code points.
4880 There is a subtlety with Prepend* which showed up in testing.
4881 Note that the Begin, and only the Begin is required in:
4882 | Prepend* Begin Extend*
4883 Also, Begin contains '! Control'. A Prepend must be a
4884 '! Control', which means it must also be a Begin. What it
4885 comes down to is that if we match Prepend* and then find no
4886 suitable Begin afterwards, that if we backtrack the last
4887 Prepend, that one will be a suitable Begin.
4892 if (! utf8_target) {
4894 /* Match either CR LF or '.', as all the other possibilities
4896 locinput++; /* Match the . or CR */
4897 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4899 && locinput < reginfo->strend
4900 && UCHARAT(locinput) == '\n')
4907 /* Utf8: See if is ( CR LF ); already know that locinput <
4908 * reginfo->strend, so locinput+1 is in bounds */
4909 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4910 && UCHARAT(locinput + 1) == '\n')
4917 /* In case have to backtrack to beginning, then match '.' */
4918 char *starting = locinput;
4920 /* In case have to backtrack the last prepend */
4921 char *previous_prepend = NULL;
4923 LOAD_UTF8_CHARCLASS_GCB();
4925 /* Match (prepend)* */
4926 while (locinput < reginfo->strend
4927 && (len = is_GCB_Prepend_utf8(locinput)))
4929 previous_prepend = locinput;
4933 /* As noted above, if we matched a prepend character, but
4934 * the next thing won't match, back off the last prepend we
4935 * matched, as it is guaranteed to match the begin */
4936 if (previous_prepend
4937 && (locinput >= reginfo->strend
4938 || (! swash_fetch(PL_utf8_X_regular_begin,
4939 (U8*)locinput, utf8_target)
4940 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4943 locinput = previous_prepend;
4946 /* Note that here we know reginfo->strend > locinput, as we
4947 * tested that upon input to this switch case, and if we
4948 * moved locinput forward, we tested the result just above
4949 * and it either passed, or we backed off so that it will
4951 if (swash_fetch(PL_utf8_X_regular_begin,
4952 (U8*)locinput, utf8_target)) {
4953 locinput += UTF8SKIP(locinput);
4955 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4957 /* Here did not match the required 'Begin' in the
4958 * second term. So just match the very first
4959 * character, the '.' of the final term of the regex */
4960 locinput = starting + UTF8SKIP(starting);
4964 /* Here is a special begin. It can be composed of
4965 * several individual characters. One possibility is
4967 if ((len = is_GCB_RI_utf8(locinput))) {
4969 while (locinput < reginfo->strend
4970 && (len = is_GCB_RI_utf8(locinput)))
4974 } else if ((len = is_GCB_T_utf8(locinput))) {
4975 /* Another possibility is T+ */
4977 while (locinput < reginfo->strend
4978 && (len = is_GCB_T_utf8(locinput)))
4984 /* Here, neither RI+ nor T+; must be some other
4985 * Hangul. That means it is one of the others: L,
4986 * LV, LVT or V, and matches:
4987 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4990 while (locinput < reginfo->strend
4991 && (len = is_GCB_L_utf8(locinput)))
4996 /* Here, have exhausted L*. If the next character
4997 * is not an LV, LVT nor V, it means we had to have
4998 * at least one L, so matches L+ in the original
4999 * equation, we have a complete hangul syllable.
5002 if (locinput < reginfo->strend
5003 && is_GCB_LV_LVT_V_utf8(locinput))
5005 /* Otherwise keep going. Must be LV, LVT or V.
5006 * See if LVT, by first ruling out V, then LV */
5007 if (! is_GCB_V_utf8(locinput)
5008 /* All but every TCount one is LV */
5009 && (valid_utf8_to_uvchr((U8 *) locinput,
5014 locinput += UTF8SKIP(locinput);
5017 /* Must be V or LV. Take it, then match
5019 locinput += UTF8SKIP(locinput);
5020 while (locinput < reginfo->strend
5021 && (len = is_GCB_V_utf8(locinput)))
5027 /* And any of LV, LVT, or V can be followed
5029 while (locinput < reginfo->strend
5030 && (len = is_GCB_T_utf8(locinput)))
5038 /* Match any extender */
5039 while (locinput < reginfo->strend
5040 && swash_fetch(PL_utf8_X_extend,
5041 (U8*)locinput, utf8_target))
5043 locinput += UTF8SKIP(locinput);
5047 if (locinput > reginfo->strend) sayNO;
5051 case NREFFL: /* /\g{name}/il */
5052 { /* The capture buffer cases. The ones beginning with N for the
5053 named buffers just convert to the equivalent numbered and
5054 pretend they were called as the corresponding numbered buffer
5056 /* don't initialize these in the declaration, it makes C++
5061 const U8 *fold_array;
5064 folder = foldEQ_locale;
5065 fold_array = PL_fold_locale;
5067 utf8_fold_flags = FOLDEQ_LOCALE;
5070 case NREFFA: /* /\g{name}/iaa */
5071 folder = foldEQ_latin1;
5072 fold_array = PL_fold_latin1;
5074 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5077 case NREFFU: /* /\g{name}/iu */
5078 folder = foldEQ_latin1;
5079 fold_array = PL_fold_latin1;
5081 utf8_fold_flags = 0;
5084 case NREFF: /* /\g{name}/i */
5086 fold_array = PL_fold;
5088 utf8_fold_flags = 0;
5091 case NREF: /* /\g{name}/ */
5095 utf8_fold_flags = 0;
5098 /* For the named back references, find the corresponding buffer
5100 n = reg_check_named_buff_matched(rex,scan);
5105 goto do_nref_ref_common;
5107 case REFFL: /* /\1/il */
5108 folder = foldEQ_locale;
5109 fold_array = PL_fold_locale;
5110 utf8_fold_flags = FOLDEQ_LOCALE;
5113 case REFFA: /* /\1/iaa */
5114 folder = foldEQ_latin1;
5115 fold_array = PL_fold_latin1;
5116 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5119 case REFFU: /* /\1/iu */
5120 folder = foldEQ_latin1;
5121 fold_array = PL_fold_latin1;
5122 utf8_fold_flags = 0;
5125 case REFF: /* /\1/i */
5127 fold_array = PL_fold;
5128 utf8_fold_flags = 0;
5131 case REF: /* /\1/ */
5134 utf8_fold_flags = 0;
5138 n = ARG(scan); /* which paren pair */
5141 ln = rex->offs[n].start;
5142 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5143 if (rex->lastparen < n || ln == -1)
5144 sayNO; /* Do not match unless seen CLOSEn. */
5145 if (ln == rex->offs[n].end)
5148 s = reginfo->strbeg + ln;
5149 if (type != REF /* REF can do byte comparison */
5150 && (utf8_target || type == REFFU || type == REFFL))
5152 char * limit = reginfo->strend;
5154 /* This call case insensitively compares the entire buffer
5155 * at s, with the current input starting at locinput, but
5156 * not going off the end given by reginfo->strend, and
5157 * returns in <limit> upon success, how much of the
5158 * current input was matched */
5159 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
5160 locinput, &limit, 0, utf8_target, utf8_fold_flags))
5168 /* Not utf8: Inline the first character, for speed. */
5169 if (!NEXTCHR_IS_EOS &&
5170 UCHARAT(s) != nextchr &&
5172 UCHARAT(s) != fold_array[nextchr]))
5174 ln = rex->offs[n].end - ln;
5175 if (locinput + ln > reginfo->strend)
5177 if (ln > 1 && (type == REF
5178 ? memNE(s, locinput, ln)
5179 : ! folder(s, locinput, ln)))
5185 case NOTHING: /* null op; e.g. the 'nothing' following
5186 * the '*' in m{(a+|b)*}' */
5188 case TAIL: /* placeholder while compiling (A|B|C) */
5191 case BACK: /* ??? doesn't appear to be used ??? */
5195 #define ST st->u.eval
5200 regexp_internal *rei;
5201 regnode *startpoint;
5203 case GOSTART: /* (?R) */
5204 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
5205 if (cur_eval && cur_eval->locinput==locinput) {
5206 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
5207 Perl_croak(aTHX_ "Infinite recursion in regex");
5208 if ( ++nochange_depth > max_nochange_depth )
5210 "Pattern subroutine nesting without pos change"
5211 " exceeded limit in regex");
5218 if (OP(scan)==GOSUB) {
5219 startpoint = scan + ARG2L(scan);
5220 ST.close_paren = ARG(scan);
5222 startpoint = rei->program+1;
5226 /* Save all the positions seen so far. */
5227 ST.cp = regcppush(rex, 0, maxopenparen);
5228 REGCP_SET(ST.lastcp);
5230 /* and then jump to the code we share with EVAL */
5231 goto eval_recurse_doit;
5236 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
5237 if (cur_eval && cur_eval->locinput==locinput) {
5238 if ( ++nochange_depth > max_nochange_depth )
5239 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
5244 /* execute the code in the {...} */
5248 OP * const oop = PL_op;
5249 COP * const ocurcop = PL_curcop;
5253 /* save *all* paren positions */
5254 regcppush(rex, 0, maxopenparen);
5255 REGCP_SET(runops_cp);
5258 caller_cv = find_runcv(NULL);
5262 if (rexi->data->what[n] == 'r') { /* code from an external qr */
5264 (REGEXP*)(rexi->data->data[n])
5267 nop = (OP*)rexi->data->data[n+1];
5269 else if (rexi->data->what[n] == 'l') { /* literal code */
5271 nop = (OP*)rexi->data->data[n];
5272 assert(CvDEPTH(newcv));
5275 /* literal with own CV */
5276 assert(rexi->data->what[n] == 'L');
5277 newcv = rex->qr_anoncv;
5278 nop = (OP*)rexi->data->data[n];
5281 /* normally if we're about to execute code from the same
5282 * CV that we used previously, we just use the existing
5283 * CX stack entry. However, its possible that in the
5284 * meantime we may have backtracked, popped from the save
5285 * stack, and undone the SAVECOMPPAD(s) associated with
5286 * PUSH_MULTICALL; in which case PL_comppad no longer
5287 * points to newcv's pad. */
5288 if (newcv != last_pushed_cv || PL_comppad != last_pad)
5290 U8 flags = (CXp_SUB_RE |
5291 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
5292 if (last_pushed_cv) {
5293 CHANGE_MULTICALL_FLAGS(newcv, flags);
5296 PUSH_MULTICALL_FLAGS(newcv, flags);
5298 last_pushed_cv = newcv;
5301 /* these assignments are just to silence compiler
5303 multicall_cop = NULL;
5306 last_pad = PL_comppad;
5308 /* the initial nextstate you would normally execute
5309 * at the start of an eval (which would cause error
5310 * messages to come from the eval), may be optimised
5311 * away from the execution path in the regex code blocks;
5312 * so manually set PL_curcop to it initially */
5314 OP *o = cUNOPx(nop)->op_first;
5315 assert(o->op_type == OP_NULL);
5316 if (o->op_targ == OP_SCOPE) {
5317 o = cUNOPo->op_first;
5320 assert(o->op_targ == OP_LEAVE);
5321 o = cUNOPo->op_first;
5322 assert(o->op_type == OP_ENTER);
5326 if (o->op_type != OP_STUB) {
5327 assert( o->op_type == OP_NEXTSTATE
5328 || o->op_type == OP_DBSTATE
5329 || (o->op_type == OP_NULL
5330 && ( o->op_targ == OP_NEXTSTATE
5331 || o->op_targ == OP_DBSTATE
5335 PL_curcop = (COP*)o;
5340 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
5341 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
5343 rex->offs[0].end = locinput - reginfo->strbeg;
5344 if (reginfo->info_aux_eval->pos_magic)
5345 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
5346 reginfo->sv, reginfo->strbeg,
5347 locinput - reginfo->strbeg);
5350 SV *sv_mrk = get_sv("REGMARK", 1);
5351 sv_setsv(sv_mrk, sv_yes_mark);
5354 /* we don't use MULTICALL here as we want to call the
5355 * first op of the block of interest, rather than the
5356 * first op of the sub */
5357 before = (IV)(SP-PL_stack_base);
5359 CALLRUNOPS(aTHX); /* Scalar context. */
5361 if ((IV)(SP-PL_stack_base) == before)
5362 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
5368 /* before restoring everything, evaluate the returned
5369 * value, so that 'uninit' warnings don't use the wrong
5370 * PL_op or pad. Also need to process any magic vars
5371 * (e.g. $1) *before* parentheses are restored */
5376 if (logical == 0) /* (?{})/ */
5377 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5378 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5379 sw = cBOOL(SvTRUE(ret));
5382 else { /* /(??{}) */
5383 /* if its overloaded, let the regex compiler handle
5384 * it; otherwise extract regex, or stringify */
5385 if (SvGMAGICAL(ret))
5386 ret = sv_mortalcopy(ret);
5387 if (!SvAMAGIC(ret)) {
5391 if (SvTYPE(sv) == SVt_REGEXP)
5392 re_sv = (REGEXP*) sv;
5393 else if (SvSMAGICAL(ret)) {
5394 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
5396 re_sv = (REGEXP *) mg->mg_obj;
5399 /* force any undef warnings here */
5400 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
5401 ret = sv_mortalcopy(ret);
5402 (void) SvPV_force_nolen(ret);
5408 /* *** Note that at this point we don't restore
5409 * PL_comppad, (or pop the CxSUB) on the assumption it may
5410 * be used again soon. This is safe as long as nothing
5411 * in the regexp code uses the pad ! */
5413 PL_curcop = ocurcop;
5414 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5415 PL_curpm = PL_reg_curpm;
5421 /* only /(??{})/ from now on */
5424 /* extract RE object from returned value; compiling if
5428 re_sv = reg_temp_copy(NULL, re_sv);
5433 if (SvUTF8(ret) && IN_BYTES) {
5434 /* In use 'bytes': make a copy of the octet
5435 * sequence, but without the flag on */
5437 const char *const p = SvPV(ret, len);
5438 ret = newSVpvn_flags(p, len, SVs_TEMP);
5440 if (rex->intflags & PREGf_USE_RE_EVAL)
5441 pm_flags |= PMf_USE_RE_EVAL;
5443 /* if we got here, it should be an engine which
5444 * supports compiling code blocks and stuff */
5445 assert(rex->engine && rex->engine->op_comp);
5446 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5447 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5448 rex->engine, NULL, NULL,
5449 /* copy /msix etc to inner pattern */
5454 & (SVs_TEMP | SVs_GMG | SVf_ROK))
5455 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
5456 /* This isn't a first class regexp. Instead, it's
5457 caching a regexp onto an existing, Perl visible
5459 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5465 RXp_MATCH_COPIED_off(re);
5466 re->subbeg = rex->subbeg;
5467 re->sublen = rex->sublen;
5468 re->suboffset = rex->suboffset;
5469 re->subcoffset = rex->subcoffset;
5471 re->lastcloseparen = 0;
5474 debug_start_match(re_sv, utf8_target, locinput,
5475 reginfo->strend, "Matching embedded");
5477 startpoint = rei->program + 1;
5478 ST.close_paren = 0; /* only used for GOSUB */
5479 /* Save all the seen positions so far. */
5480 ST.cp = regcppush(rex, 0, maxopenparen);
5481 REGCP_SET(ST.lastcp);
5482 /* and set maxopenparen to 0, since we are starting a "fresh" match */
5484 /* run the pattern returned from (??{...}) */
5486 eval_recurse_doit: /* Share code with GOSUB below this line
5487 * At this point we expect the stack context to be
5488 * set up correctly */
5490 /* invalidate the S-L poscache. We're now executing a
5491 * different set of WHILEM ops (and their associated
5492 * indexes) against the same string, so the bits in the
5493 * cache are meaningless. Setting maxiter to zero forces
5494 * the cache to be invalidated and zeroed before reuse.
5495 * XXX This is too dramatic a measure. Ideally we should
5496 * save the old cache and restore when running the outer
5498 reginfo->poscache_maxiter = 0;
5500 /* the new regexp might have a different is_utf8_pat than we do */
5501 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5503 ST.prev_rex = rex_sv;
5504 ST.prev_curlyx = cur_curlyx;
5506 SET_reg_curpm(rex_sv);
5511 ST.prev_eval = cur_eval;
5513 /* now continue from first node in postoned RE */
5514 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5519 case EVAL_AB: /* cleanup after a successful (??{A})B */
5520 /* note: this is called twice; first after popping B, then A */
5521 rex_sv = ST.prev_rex;
5522 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5523 SET_reg_curpm(rex_sv);
5524 rex = ReANY(rex_sv);
5525 rexi = RXi_GET(rex);
5527 /* preserve $^R across LEAVE's. See Bug 121070. */
5528 SV *save_sv= GvSV(PL_replgv);
5529 SvREFCNT_inc(save_sv);
5530 regcpblow(ST.cp); /* LEAVE in disguise */
5531 sv_setsv(GvSV(PL_replgv), save_sv);
5532 SvREFCNT_dec(save_sv);
5534 cur_eval = ST.prev_eval;
5535 cur_curlyx = ST.prev_curlyx;
5537 /* Invalidate cache. See "invalidate" comment above. */
5538 reginfo->poscache_maxiter = 0;
5539 if ( nochange_depth )
5544 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5545 /* note: this is called twice; first after popping B, then A */
5546 rex_sv = ST.prev_rex;
5547 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5548 SET_reg_curpm(rex_sv);
5549 rex = ReANY(rex_sv);
5550 rexi = RXi_GET(rex);
5552 REGCP_UNWIND(ST.lastcp);
5553 regcppop(rex, &maxopenparen);
5554 cur_eval = ST.prev_eval;
5555 cur_curlyx = ST.prev_curlyx;
5556 /* Invalidate cache. See "invalidate" comment above. */
5557 reginfo->poscache_maxiter = 0;
5558 if ( nochange_depth )
5564 n = ARG(scan); /* which paren pair */
5565 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5566 if (n > maxopenparen)
5568 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5569 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5573 (IV)rex->offs[n].start_tmp,
5579 /* XXX really need to log other places start/end are set too */
5580 #define CLOSE_CAPTURE \
5581 rex->offs[n].start = rex->offs[n].start_tmp; \
5582 rex->offs[n].end = locinput - reginfo->strbeg; \
5583 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5584 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5586 PTR2UV(rex->offs), \
5588 (IV)rex->offs[n].start, \
5589 (IV)rex->offs[n].end \
5593 n = ARG(scan); /* which paren pair */
5595 if (n > rex->lastparen)
5597 rex->lastcloseparen = n;
5598 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5603 case ACCEPT: /* (*ACCEPT) */
5607 cursor && OP(cursor)!=END;
5608 cursor=regnext(cursor))
5610 if ( OP(cursor)==CLOSE ){
5612 if ( n <= lastopen ) {
5614 if (n > rex->lastparen)
5616 rex->lastcloseparen = n;
5617 if ( n == ARG(scan) || (cur_eval &&
5618 cur_eval->u.eval.close_paren == n))
5627 case GROUPP: /* (?(1)) */
5628 n = ARG(scan); /* which paren pair */
5629 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5632 case NGROUPP: /* (?(<name>)) */
5633 /* reg_check_named_buff_matched returns 0 for no match */
5634 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5637 case INSUBP: /* (?(R)) */
5639 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5642 case DEFINEP: /* (?(DEFINE)) */
5646 case IFTHEN: /* (?(cond)A|B) */
5647 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5649 next = NEXTOPER(NEXTOPER(scan));
5651 next = scan + ARG(scan);
5652 if (OP(next) == IFTHEN) /* Fake one. */
5653 next = NEXTOPER(NEXTOPER(next));
5657 case LOGICAL: /* modifier for EVAL and IFMATCH */
5658 logical = scan->flags;
5661 /*******************************************************************
5663 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5664 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5665 STAR/PLUS/CURLY/CURLYN are used instead.)
5667 A*B is compiled as <CURLYX><A><WHILEM><B>
5669 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5670 state, which contains the current count, initialised to -1. It also sets
5671 cur_curlyx to point to this state, with any previous value saved in the
5674 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5675 since the pattern may possibly match zero times (i.e. it's a while {} loop
5676 rather than a do {} while loop).
5678 Each entry to WHILEM represents a successful match of A. The count in the
5679 CURLYX block is incremented, another WHILEM state is pushed, and execution
5680 passes to A or B depending on greediness and the current count.
5682 For example, if matching against the string a1a2a3b (where the aN are
5683 substrings that match /A/), then the match progresses as follows: (the
5684 pushed states are interspersed with the bits of strings matched so far):
5687 <CURLYX cnt=0><WHILEM>
5688 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5689 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5690 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5691 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5693 (Contrast this with something like CURLYM, which maintains only a single
5697 a1 <CURLYM cnt=1> a2
5698 a1 a2 <CURLYM cnt=2> a3
5699 a1 a2 a3 <CURLYM cnt=3> b
5702 Each WHILEM state block marks a point to backtrack to upon partial failure
5703 of A or B, and also contains some minor state data related to that
5704 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5705 overall state, such as the count, and pointers to the A and B ops.
5707 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5708 must always point to the *current* CURLYX block, the rules are:
5710 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5711 and set cur_curlyx to point the new block.
5713 When popping the CURLYX block after a successful or unsuccessful match,
5714 restore the previous cur_curlyx.
5716 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5717 to the outer one saved in the CURLYX block.
5719 When popping the WHILEM block after a successful or unsuccessful B match,
5720 restore the previous cur_curlyx.
5722 Here's an example for the pattern (AI* BI)*BO
5723 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5726 curlyx backtrack stack
5727 ------ ---------------
5729 CO <CO prev=NULL> <WO>
5730 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5731 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5732 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5734 At this point the pattern succeeds, and we work back down the stack to
5735 clean up, restoring as we go:
5737 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5738 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5739 CO <CO prev=NULL> <WO>
5742 *******************************************************************/
5744 #define ST st->u.curlyx
5746 case CURLYX: /* start of /A*B/ (for complex A) */
5748 /* No need to save/restore up to this paren */
5749 I32 parenfloor = scan->flags;
5751 assert(next); /* keep Coverity happy */
5752 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5755 /* XXXX Probably it is better to teach regpush to support
5756 parenfloor > maxopenparen ... */
5757 if (parenfloor > (I32)rex->lastparen)
5758 parenfloor = rex->lastparen; /* Pessimization... */
5760 ST.prev_curlyx= cur_curlyx;
5762 ST.cp = PL_savestack_ix;
5764 /* these fields contain the state of the current curly.
5765 * they are accessed by subsequent WHILEMs */
5766 ST.parenfloor = parenfloor;
5771 ST.count = -1; /* this will be updated by WHILEM */
5772 ST.lastloc = NULL; /* this will be updated by WHILEM */
5774 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5779 case CURLYX_end: /* just finished matching all of A*B */
5780 cur_curlyx = ST.prev_curlyx;
5785 case CURLYX_end_fail: /* just failed to match all of A*B */
5787 cur_curlyx = ST.prev_curlyx;
5794 #define ST st->u.whilem
5796 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5798 /* see the discussion above about CURLYX/WHILEM */
5803 assert(cur_curlyx); /* keep Coverity happy */
5805 min = ARG1(cur_curlyx->u.curlyx.me);
5806 max = ARG2(cur_curlyx->u.curlyx.me);
5807 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5808 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5809 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5810 ST.cache_offset = 0;
5814 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5815 "%*s whilem: matched %ld out of %d..%d\n",
5816 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5819 /* First just match a string of min A's. */
5822 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5824 cur_curlyx->u.curlyx.lastloc = locinput;
5825 REGCP_SET(ST.lastcp);
5827 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5832 /* If degenerate A matches "", assume A done. */
5834 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5835 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5836 "%*s whilem: empty match detected, trying continuation...\n",
5837 REPORT_CODE_OFF+depth*2, "")
5839 goto do_whilem_B_max;
5842 /* super-linear cache processing.
5844 * The idea here is that for certain types of CURLYX/WHILEM -
5845 * principally those whose upper bound is infinity (and
5846 * excluding regexes that have things like \1 and other very
5847 * non-regular expresssiony things), then if a pattern like
5848 * /....A*.../ fails and we backtrack to the WHILEM, then we
5849 * make a note that this particular WHILEM op was at string
5850 * position 47 (say) when the rest of pattern failed. Then, if
5851 * we ever find ourselves back at that WHILEM, and at string
5852 * position 47 again, we can just fail immediately rather than
5853 * running the rest of the pattern again.
5855 * This is very handy when patterns start to go
5856 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5857 * with a combinatorial explosion of backtracking.
5859 * The cache is implemented as a bit array, with one bit per
5860 * string byte position per WHILEM op (up to 16) - so its
5861 * between 0.25 and 2x the string size.
5863 * To avoid allocating a poscache buffer every time, we do an
5864 * initially countdown; only after we have executed a WHILEM
5865 * op (string-length x #WHILEMs) times do we allocate the
5868 * The top 4 bits of scan->flags byte say how many different
5869 * relevant CURLLYX/WHILEM op pairs there are, while the
5870 * bottom 4-bits is the identifying index number of this
5876 if (!reginfo->poscache_maxiter) {
5877 /* start the countdown: Postpone detection until we
5878 * know the match is not *that* much linear. */
5879 reginfo->poscache_maxiter
5880 = (reginfo->strend - reginfo->strbeg + 1)
5882 /* possible overflow for long strings and many CURLYX's */
5883 if (reginfo->poscache_maxiter < 0)
5884 reginfo->poscache_maxiter = I32_MAX;
5885 reginfo->poscache_iter = reginfo->poscache_maxiter;
5888 if (reginfo->poscache_iter-- == 0) {
5889 /* initialise cache */
5890 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
5891 regmatch_info_aux *const aux = reginfo->info_aux;
5892 if (aux->poscache) {
5893 if ((SSize_t)reginfo->poscache_size < size) {
5894 Renew(aux->poscache, size, char);
5895 reginfo->poscache_size = size;
5897 Zero(aux->poscache, size, char);
5900 reginfo->poscache_size = size;
5901 Newxz(aux->poscache, size, char);
5903 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5904 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5905 PL_colors[4], PL_colors[5])
5909 if (reginfo->poscache_iter < 0) {
5910 /* have we already failed at this position? */
5911 SSize_t offset, mask;
5913 reginfo->poscache_iter = -1; /* stop eventual underflow */
5914 offset = (scan->flags & 0xf) - 1
5915 + (locinput - reginfo->strbeg)
5917 mask = 1 << (offset % 8);
5919 if (reginfo->info_aux->poscache[offset] & mask) {
5920 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5921 "%*s whilem: (cache) already tried at this position...\n",
5922 REPORT_CODE_OFF+depth*2, "")
5924 sayNO; /* cache records failure */
5926 ST.cache_offset = offset;
5927 ST.cache_mask = mask;
5931 /* Prefer B over A for minimal matching. */
5933 if (cur_curlyx->u.curlyx.minmod) {
5934 ST.save_curlyx = cur_curlyx;
5935 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5936 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5938 REGCP_SET(ST.lastcp);
5939 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5945 /* Prefer A over B for maximal matching. */
5947 if (n < max) { /* More greed allowed? */
5948 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5950 cur_curlyx->u.curlyx.lastloc = locinput;
5951 REGCP_SET(ST.lastcp);
5952 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5956 goto do_whilem_B_max;
5961 case WHILEM_B_min: /* just matched B in a minimal match */
5962 case WHILEM_B_max: /* just matched B in a maximal match */
5963 cur_curlyx = ST.save_curlyx;
5968 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5969 cur_curlyx = ST.save_curlyx;
5970 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5971 cur_curlyx->u.curlyx.count--;
5976 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5978 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5979 REGCP_UNWIND(ST.lastcp);
5980 regcppop(rex, &maxopenparen);
5981 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5982 cur_curlyx->u.curlyx.count--;
5987 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5988 REGCP_UNWIND(ST.lastcp);
5989 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5990 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5991 "%*s whilem: failed, trying continuation...\n",
5992 REPORT_CODE_OFF+depth*2, "")
5995 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5996 && ckWARN(WARN_REGEXP)
5997 && !reginfo->warned)
5999 reginfo->warned = TRUE;
6000 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6001 "Complex regular subexpression recursion limit (%d) "
6007 ST.save_curlyx = cur_curlyx;
6008 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
6009 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
6014 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
6015 cur_curlyx = ST.save_curlyx;
6016 REGCP_UNWIND(ST.lastcp);
6017 regcppop(rex, &maxopenparen);
6019 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
6020 /* Maximum greed exceeded */
6021 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6022 && ckWARN(WARN_REGEXP)
6023 && !reginfo->warned)
6025 reginfo->warned = TRUE;
6026 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6027 "Complex regular subexpression recursion "
6028 "limit (%d) exceeded",
6031 cur_curlyx->u.curlyx.count--;
6035 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6036 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
6038 /* Try grabbing another A and see if it helps. */
6039 cur_curlyx->u.curlyx.lastloc = locinput;
6040 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6042 REGCP_SET(ST.lastcp);
6043 PUSH_STATE_GOTO(WHILEM_A_min,
6044 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
6050 #define ST st->u.branch
6052 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
6053 next = scan + ARG(scan);
6056 scan = NEXTOPER(scan);
6059 case BRANCH: /* /(...|A|...)/ */
6060 scan = NEXTOPER(scan); /* scan now points to inner node */
6061 ST.lastparen = rex->lastparen;
6062 ST.lastcloseparen = rex->lastcloseparen;
6063 ST.next_branch = next;
6066 /* Now go into the branch */
6068 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
6070 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
6075 case CUTGROUP: /* /(*THEN)/ */
6076 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
6077 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6078 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
6082 case CUTGROUP_next_fail:
6085 if (st->u.mark.mark_name)
6086 sv_commit = st->u.mark.mark_name;
6096 case BRANCH_next_fail: /* that branch failed; try the next, if any */
6101 REGCP_UNWIND(ST.cp);
6102 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6103 scan = ST.next_branch;
6104 /* no more branches? */
6105 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
6107 PerlIO_printf( Perl_debug_log,
6108 "%*s %sBRANCH failed...%s\n",
6109 REPORT_CODE_OFF+depth*2, "",
6115 continue; /* execute next BRANCH[J] op */
6119 case MINMOD: /* next op will be non-greedy, e.g. A*? */
6124 #define ST st->u.curlym
6126 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
6128 /* This is an optimisation of CURLYX that enables us to push
6129 * only a single backtracking state, no matter how many matches
6130 * there are in {m,n}. It relies on the pattern being constant
6131 * length, with no parens to influence future backrefs
6135 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6137 ST.lastparen = rex->lastparen;
6138 ST.lastcloseparen = rex->lastcloseparen;
6140 /* if paren positive, emulate an OPEN/CLOSE around A */
6142 U32 paren = ST.me->flags;
6143 if (paren > maxopenparen)
6144 maxopenparen = paren;
6145 scan += NEXT_OFF(scan); /* Skip former OPEN. */
6153 ST.c1 = CHRTEST_UNINIT;
6156 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
6159 curlym_do_A: /* execute the A in /A{m,n}B/ */
6160 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
6164 case CURLYM_A: /* we've just matched an A */
6166 /* after first match, determine A's length: u.curlym.alen */
6167 if (ST.count == 1) {
6168 if (reginfo->is_utf8_target) {
6169 char *s = st->locinput;
6170 while (s < locinput) {
6176 ST.alen = locinput - st->locinput;
6179 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
6182 PerlIO_printf(Perl_debug_log,
6183 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
6184 (int)(REPORT_CODE_OFF+(depth*2)), "",
6185 (IV) ST.count, (IV)ST.alen)
6188 if (cur_eval && cur_eval->u.eval.close_paren &&
6189 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6193 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
6194 if ( max == REG_INFTY || ST.count < max )
6195 goto curlym_do_A; /* try to match another A */
6197 goto curlym_do_B; /* try to match B */
6199 case CURLYM_A_fail: /* just failed to match an A */
6200 REGCP_UNWIND(ST.cp);
6202 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
6203 || (cur_eval && cur_eval->u.eval.close_paren &&
6204 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
6207 curlym_do_B: /* execute the B in /A{m,n}B/ */
6208 if (ST.c1 == CHRTEST_UNINIT) {
6209 /* calculate c1 and c2 for possible match of 1st char
6210 * following curly */
6211 ST.c1 = ST.c2 = CHRTEST_VOID;
6213 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
6214 regnode *text_node = ST.B;
6215 if (! HAS_TEXT(text_node))
6216 FIND_NEXT_IMPT(text_node);
6219 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
6221 But the former is redundant in light of the latter.
6223 if this changes back then the macro for
6224 IS_TEXT and friends need to change.
6226 if (PL_regkind[OP(text_node)] == EXACT) {
6227 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6228 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6238 PerlIO_printf(Perl_debug_log,
6239 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
6240 (int)(REPORT_CODE_OFF+(depth*2)),
6243 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
6244 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
6245 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6246 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6248 /* simulate B failing */
6250 PerlIO_printf(Perl_debug_log,
6251 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
6252 (int)(REPORT_CODE_OFF+(depth*2)),"",
6253 valid_utf8_to_uvchr((U8 *) locinput, NULL),
6254 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
6255 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
6257 state_num = CURLYM_B_fail;
6258 goto reenter_switch;
6261 else if (nextchr != ST.c1 && nextchr != ST.c2) {
6262 /* simulate B failing */
6264 PerlIO_printf(Perl_debug_log,
6265 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
6266 (int)(REPORT_CODE_OFF+(depth*2)),"",
6267 (int) nextchr, ST.c1, ST.c2)
6269 state_num = CURLYM_B_fail;
6270 goto reenter_switch;
6275 /* emulate CLOSE: mark current A as captured */
6276 I32 paren = ST.me->flags;
6278 rex->offs[paren].start
6279 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
6280 rex->offs[paren].end = locinput - reginfo->strbeg;
6281 if ((U32)paren > rex->lastparen)
6282 rex->lastparen = paren;
6283 rex->lastcloseparen = paren;
6286 rex->offs[paren].end = -1;
6287 if (cur_eval && cur_eval->u.eval.close_paren &&
6288 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6297 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
6301 case CURLYM_B_fail: /* just failed to match a B */
6302 REGCP_UNWIND(ST.cp);
6303 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6305 I32 max = ARG2(ST.me);
6306 if (max != REG_INFTY && ST.count == max)
6308 goto curlym_do_A; /* try to match a further A */
6310 /* backtrack one A */
6311 if (ST.count == ARG1(ST.me) /* min */)
6314 SET_locinput(HOPc(locinput, -ST.alen));
6315 goto curlym_do_B; /* try to match B */
6318 #define ST st->u.curly
6320 #define CURLY_SETPAREN(paren, success) \
6323 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
6324 rex->offs[paren].end = locinput - reginfo->strbeg; \
6325 if (paren > rex->lastparen) \
6326 rex->lastparen = paren; \
6327 rex->lastcloseparen = paren; \
6330 rex->offs[paren].end = -1; \
6331 rex->lastparen = ST.lastparen; \
6332 rex->lastcloseparen = ST.lastcloseparen; \
6336 case STAR: /* /A*B/ where A is width 1 char */
6340 scan = NEXTOPER(scan);
6343 case PLUS: /* /A+B/ where A is width 1 char */
6347 scan = NEXTOPER(scan);
6350 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
6351 ST.paren = scan->flags; /* Which paren to set */
6352 ST.lastparen = rex->lastparen;
6353 ST.lastcloseparen = rex->lastcloseparen;
6354 if (ST.paren > maxopenparen)
6355 maxopenparen = ST.paren;
6356 ST.min = ARG1(scan); /* min to match */
6357 ST.max = ARG2(scan); /* max to match */
6358 if (cur_eval && cur_eval->u.eval.close_paren &&
6359 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6363 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
6366 case CURLY: /* /A{m,n}B/ where A is width 1 char */
6368 ST.min = ARG1(scan); /* min to match */
6369 ST.max = ARG2(scan); /* max to match */
6370 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6373 * Lookahead to avoid useless match attempts
6374 * when we know what character comes next.
6376 * Used to only do .*x and .*?x, but now it allows
6377 * for )'s, ('s and (?{ ... })'s to be in the way
6378 * of the quantifier and the EXACT-like node. -- japhy
6381 assert(ST.min <= ST.max);
6382 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
6383 ST.c1 = ST.c2 = CHRTEST_VOID;
6386 regnode *text_node = next;
6388 if (! HAS_TEXT(text_node))
6389 FIND_NEXT_IMPT(text_node);
6391 if (! HAS_TEXT(text_node))
6392 ST.c1 = ST.c2 = CHRTEST_VOID;
6394 if ( PL_regkind[OP(text_node)] != EXACT ) {
6395 ST.c1 = ST.c2 = CHRTEST_VOID;
6399 /* Currently we only get here when
6401 PL_rekind[OP(text_node)] == EXACT
6403 if this changes back then the macro for IS_TEXT and
6404 friends need to change. */
6405 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6406 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6418 char *li = locinput;
6421 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
6427 if (ST.c1 == CHRTEST_VOID)
6428 goto curly_try_B_min;
6430 ST.oldloc = locinput;
6432 /* set ST.maxpos to the furthest point along the
6433 * string that could possibly match */
6434 if (ST.max == REG_INFTY) {
6435 ST.maxpos = reginfo->strend - 1;
6437 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
6440 else if (utf8_target) {
6441 int m = ST.max - ST.min;
6442 for (ST.maxpos = locinput;
6443 m >0 && ST.maxpos < reginfo->strend; m--)
6444 ST.maxpos += UTF8SKIP(ST.maxpos);
6447 ST.maxpos = locinput + ST.max - ST.min;
6448 if (ST.maxpos >= reginfo->strend)
6449 ST.maxpos = reginfo->strend - 1;
6451 goto curly_try_B_min_known;
6455 /* avoid taking address of locinput, so it can remain
6457 char *li = locinput;
6458 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6459 if (ST.count < ST.min)
6462 if ((ST.count > ST.min)
6463 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6465 /* A{m,n} must come at the end of the string, there's
6466 * no point in backing off ... */
6468 /* ...except that $ and \Z can match before *and* after
6469 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6470 We may back off by one in this case. */
6471 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6475 goto curly_try_B_max;
6480 case CURLY_B_min_known_fail:
6481 /* failed to find B in a non-greedy match where c1,c2 valid */
6483 REGCP_UNWIND(ST.cp);
6485 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6487 /* Couldn't or didn't -- move forward. */
6488 ST.oldloc = locinput;
6490 locinput += UTF8SKIP(locinput);
6494 curly_try_B_min_known:
6495 /* find the next place where 'B' could work, then call B */
6499 n = (ST.oldloc == locinput) ? 0 : 1;
6500 if (ST.c1 == ST.c2) {
6501 /* set n to utf8_distance(oldloc, locinput) */
6502 while (locinput <= ST.maxpos
6503 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6505 locinput += UTF8SKIP(locinput);
6510 /* set n to utf8_distance(oldloc, locinput) */
6511 while (locinput <= ST.maxpos
6512 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6513 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6515 locinput += UTF8SKIP(locinput);
6520 else { /* Not utf8_target */
6521 if (ST.c1 == ST.c2) {
6522 while (locinput <= ST.maxpos &&
6523 UCHARAT(locinput) != ST.c1)
6527 while (locinput <= ST.maxpos
6528 && UCHARAT(locinput) != ST.c1
6529 && UCHARAT(locinput) != ST.c2)
6532 n = locinput - ST.oldloc;
6534 if (locinput > ST.maxpos)
6537 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6538 * at b; check that everything between oldloc and
6539 * locinput matches */
6540 char *li = ST.oldloc;
6542 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6544 assert(n == REG_INFTY || locinput == li);
6546 CURLY_SETPAREN(ST.paren, ST.count);
6547 if (cur_eval && cur_eval->u.eval.close_paren &&
6548 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6551 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6556 case CURLY_B_min_fail:
6557 /* failed to find B in a non-greedy match where c1,c2 invalid */
6559 REGCP_UNWIND(ST.cp);
6561 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6563 /* failed -- move forward one */
6565 char *li = locinput;
6566 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6573 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6574 ST.count > 0)) /* count overflow ? */
6577 CURLY_SETPAREN(ST.paren, ST.count);
6578 if (cur_eval && cur_eval->u.eval.close_paren &&
6579 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6582 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6590 /* a successful greedy match: now try to match B */
6591 if (cur_eval && cur_eval->u.eval.close_paren &&
6592 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6596 bool could_match = locinput < reginfo->strend;
6598 /* If it could work, try it. */
6599 if (ST.c1 != CHRTEST_VOID && could_match) {
6600 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6602 could_match = memEQ(locinput,
6607 UTF8SKIP(locinput));
6610 could_match = UCHARAT(locinput) == ST.c1
6611 || UCHARAT(locinput) == ST.c2;
6614 if (ST.c1 == CHRTEST_VOID || could_match) {
6615 CURLY_SETPAREN(ST.paren, ST.count);
6616 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6623 case CURLY_B_max_fail:
6624 /* failed to find B in a greedy match */
6626 REGCP_UNWIND(ST.cp);
6628 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6631 if (--ST.count < ST.min)
6633 locinput = HOPc(locinput, -1);
6634 goto curly_try_B_max;
6638 case END: /* last op of main pattern */
6641 /* we've just finished A in /(??{A})B/; now continue with B */
6643 st->u.eval.prev_rex = rex_sv; /* inner */
6645 /* Save *all* the positions. */
6646 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6647 rex_sv = cur_eval->u.eval.prev_rex;
6648 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6649 SET_reg_curpm(rex_sv);
6650 rex = ReANY(rex_sv);
6651 rexi = RXi_GET(rex);
6652 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6654 REGCP_SET(st->u.eval.lastcp);
6656 /* Restore parens of the outer rex without popping the
6658 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6661 st->u.eval.prev_eval = cur_eval;
6662 cur_eval = cur_eval->u.eval.prev_eval;
6664 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6665 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6666 if ( nochange_depth )
6669 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6670 locinput); /* match B */
6673 if (locinput < reginfo->till) {
6674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6675 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6677 (long)(locinput - startpos),
6678 (long)(reginfo->till - startpos),
6681 sayNO_SILENT; /* Cannot match: too short. */
6683 sayYES; /* Success! */
6685 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6687 PerlIO_printf(Perl_debug_log,
6688 "%*s %ssubpattern success...%s\n",
6689 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6690 sayYES; /* Success! */
6693 #define ST st->u.ifmatch
6698 case SUSPEND: /* (?>A) */
6700 newstart = locinput;
6703 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6705 goto ifmatch_trivial_fail_test;
6707 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6709 ifmatch_trivial_fail_test:
6711 char * const s = HOPBACKc(locinput, scan->flags);
6716 sw = 1 - cBOOL(ST.wanted);
6720 next = scan + ARG(scan);
6728 newstart = locinput;
6732 ST.logical = logical;
6733 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6735 /* execute body of (?...A) */
6736 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6741 case IFMATCH_A_fail: /* body of (?...A) failed */
6742 ST.wanted = !ST.wanted;
6745 case IFMATCH_A: /* body of (?...A) succeeded */
6747 sw = cBOOL(ST.wanted);
6749 else if (!ST.wanted)
6752 if (OP(ST.me) != SUSPEND) {
6753 /* restore old position except for (?>...) */
6754 locinput = st->locinput;
6756 scan = ST.me + ARG(ST.me);
6759 continue; /* execute B */
6763 case LONGJMP: /* alternative with many branches compiles to
6764 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6765 next = scan + ARG(scan);
6770 case COMMIT: /* (*COMMIT) */
6771 reginfo->cutpoint = reginfo->strend;
6774 case PRUNE: /* (*PRUNE) */
6776 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6777 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6781 case COMMIT_next_fail:
6785 case OPFAIL: /* (*FAIL) */
6790 #define ST st->u.mark
6791 case MARKPOINT: /* (*MARK:foo) */
6792 ST.prev_mark = mark_state;
6793 ST.mark_name = sv_commit = sv_yes_mark
6794 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6796 ST.mark_loc = locinput;
6797 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6801 case MARKPOINT_next:
6802 mark_state = ST.prev_mark;
6807 case MARKPOINT_next_fail:
6808 if (popmark && sv_eq(ST.mark_name,popmark))
6810 if (ST.mark_loc > startpoint)
6811 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6812 popmark = NULL; /* we found our mark */
6813 sv_commit = ST.mark_name;
6816 PerlIO_printf(Perl_debug_log,
6817 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6818 REPORT_CODE_OFF+depth*2, "",
6819 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6822 mark_state = ST.prev_mark;
6823 sv_yes_mark = mark_state ?
6824 mark_state->u.mark.mark_name : NULL;
6829 case SKIP: /* (*SKIP) */
6831 /* (*SKIP) : if we fail we cut here*/
6832 ST.mark_name = NULL;
6833 ST.mark_loc = locinput;
6834 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6836 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6837 otherwise do nothing. Meaning we need to scan
6839 regmatch_state *cur = mark_state;
6840 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6843 if ( sv_eq( cur->u.mark.mark_name,
6846 ST.mark_name = find;
6847 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6849 cur = cur->u.mark.prev_mark;
6852 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6855 case SKIP_next_fail:
6857 /* (*CUT:NAME) - Set up to search for the name as we
6858 collapse the stack*/
6859 popmark = ST.mark_name;
6861 /* (*CUT) - No name, we cut here.*/
6862 if (ST.mark_loc > startpoint)
6863 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6864 /* but we set sv_commit to latest mark_name if there
6865 is one so they can test to see how things lead to this
6868 sv_commit=mark_state->u.mark.mark_name;
6876 case LNBREAK: /* \R */
6877 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6884 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6885 PTR2UV(scan), OP(scan));
6886 Perl_croak(aTHX_ "regexp memory corruption");
6888 /* this is a point to jump to in order to increment
6889 * locinput by one character */
6891 assert(!NEXTCHR_IS_EOS);
6893 locinput += PL_utf8skip[nextchr];
6894 /* locinput is allowed to go 1 char off the end, but not 2+ */
6895 if (locinput > reginfo->strend)
6904 /* switch break jumps here */
6905 scan = next; /* prepare to execute the next op and ... */
6906 continue; /* ... jump back to the top, reusing st */
6911 /* push a state that backtracks on success */
6912 st->u.yes.prev_yes_state = yes_state;
6916 /* push a new regex state, then continue at scan */
6918 regmatch_state *newst;
6921 regmatch_state *cur = st;
6922 regmatch_state *curyes = yes_state;
6924 regmatch_slab *slab = PL_regmatch_slab;
6925 for (;curd > -1;cur--,curd--) {
6926 if (cur < SLAB_FIRST(slab)) {
6928 cur = SLAB_LAST(slab);
6930 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6931 REPORT_CODE_OFF + 2 + depth * 2,"",
6932 curd, PL_reg_name[cur->resume_state],
6933 (curyes == cur) ? "yes" : ""
6936 curyes = cur->u.yes.prev_yes_state;
6939 DEBUG_STATE_pp("push")
6942 st->locinput = locinput;
6944 if (newst > SLAB_LAST(PL_regmatch_slab))
6945 newst = S_push_slab(aTHX);
6946 PL_regmatch_state = newst;
6948 locinput = pushinput;
6957 * We get here only if there's trouble -- normally "case END" is
6958 * the terminating point.
6960 Perl_croak(aTHX_ "corrupted regexp pointers");
6966 /* we have successfully completed a subexpression, but we must now
6967 * pop to the state marked by yes_state and continue from there */
6968 assert(st != yes_state);
6970 while (st != yes_state) {
6972 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6973 PL_regmatch_slab = PL_regmatch_slab->prev;
6974 st = SLAB_LAST(PL_regmatch_slab);
6978 DEBUG_STATE_pp("pop (no final)");
6980 DEBUG_STATE_pp("pop (yes)");
6986 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6987 || yes_state > SLAB_LAST(PL_regmatch_slab))
6989 /* not in this slab, pop slab */
6990 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6991 PL_regmatch_slab = PL_regmatch_slab->prev;
6992 st = SLAB_LAST(PL_regmatch_slab);
6994 depth -= (st - yes_state);
6997 yes_state = st->u.yes.prev_yes_state;
6998 PL_regmatch_state = st;
7001 locinput= st->locinput;
7002 state_num = st->resume_state + no_final;
7003 goto reenter_switch;
7006 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
7007 PL_colors[4], PL_colors[5]));
7009 if (reginfo->info_aux_eval) {
7010 /* each successfully executed (?{...}) block does the equivalent of
7011 * local $^R = do {...}
7012 * When popping the save stack, all these locals would be undone;
7013 * bypass this by setting the outermost saved $^R to the latest
7015 /* I dont know if this is needed or works properly now.
7016 * see code related to PL_replgv elsewhere in this file.
7019 if (oreplsv != GvSV(PL_replgv))
7020 sv_setsv(oreplsv, GvSV(PL_replgv));
7027 PerlIO_printf(Perl_debug_log,
7028 "%*s %sfailed...%s\n",
7029 REPORT_CODE_OFF+depth*2, "",
7030 PL_colors[4], PL_colors[5])
7042 /* there's a previous state to backtrack to */
7044 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7045 PL_regmatch_slab = PL_regmatch_slab->prev;
7046 st = SLAB_LAST(PL_regmatch_slab);
7048 PL_regmatch_state = st;
7049 locinput= st->locinput;
7051 DEBUG_STATE_pp("pop");
7053 if (yes_state == st)
7054 yes_state = st->u.yes.prev_yes_state;
7056 state_num = st->resume_state + 1; /* failure = success + 1 */
7057 goto reenter_switch;
7062 if (rex->intflags & PREGf_VERBARG_SEEN) {
7063 SV *sv_err = get_sv("REGERROR", 1);
7064 SV *sv_mrk = get_sv("REGMARK", 1);
7066 sv_commit = &PL_sv_no;
7068 sv_yes_mark = &PL_sv_yes;
7071 sv_commit = &PL_sv_yes;
7072 sv_yes_mark = &PL_sv_no;
7076 sv_setsv(sv_err, sv_commit);
7077 sv_setsv(sv_mrk, sv_yes_mark);
7081 if (last_pushed_cv) {
7084 PERL_UNUSED_VAR(SP);
7087 assert(!result || locinput - reginfo->strbeg >= 0);
7088 return result ? locinput - reginfo->strbeg : -1;
7092 - regrepeat - repeatedly match something simple, report how many
7094 * What 'simple' means is a node which can be the operand of a quantifier like
7097 * startposp - pointer a pointer to the start position. This is updated
7098 * to point to the byte following the highest successful
7100 * p - the regnode to be repeatedly matched against.
7101 * reginfo - struct holding match state, such as strend
7102 * max - maximum number of things to match.
7103 * depth - (for debugging) backtracking depth.
7106 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
7107 regmatch_info *const reginfo, I32 max, int depth)
7109 char *scan; /* Pointer to current position in target string */
7111 char *loceol = reginfo->strend; /* local version */
7112 I32 hardcount = 0; /* How many matches so far */
7113 bool utf8_target = reginfo->is_utf8_target;
7114 int to_complement = 0; /* Invert the result? */
7116 _char_class_number classnum;
7118 PERL_UNUSED_ARG(depth);
7121 PERL_ARGS_ASSERT_REGREPEAT;
7124 if (max == REG_INFTY)
7126 else if (! utf8_target && loceol - scan > max)
7127 loceol = scan + max;
7129 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
7130 * to the maximum of how far we should go in it (leaving it set to the real
7131 * end, if the maximum permissible would take us beyond that). This allows
7132 * us to make the loop exit condition that we haven't gone past <loceol> to
7133 * also mean that we haven't exceeded the max permissible count, saving a
7134 * test each time through the loop. But it assumes that the OP matches a
7135 * single byte, which is true for most of the OPs below when applied to a
7136 * non-UTF-8 target. Those relatively few OPs that don't have this
7137 * characteristic will have to compensate.
7139 * There is no adjustment for UTF-8 targets, as the number of bytes per
7140 * character varies. OPs will have to test both that the count is less
7141 * than the max permissible (using <hardcount> to keep track), and that we
7142 * are still within the bounds of the string (using <loceol>. A few OPs
7143 * match a single byte no matter what the encoding. They can omit the max
7144 * test if, for the UTF-8 case, they do the adjustment that was skipped
7147 * Thus, the code above sets things up for the common case; and exceptional
7148 * cases need extra work; the common case is to make sure <scan> doesn't
7149 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
7150 * count doesn't exceed the maximum permissible */
7155 while (scan < loceol && hardcount < max && *scan != '\n') {
7156 scan += UTF8SKIP(scan);
7160 while (scan < loceol && *scan != '\n')
7166 while (scan < loceol && hardcount < max) {
7167 scan += UTF8SKIP(scan);
7174 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
7175 if (utf8_target && loceol - scan > max) {
7177 /* <loceol> hadn't been adjusted in the UTF-8 case */
7185 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7189 /* Can use a simple loop if the pattern char to match on is invariant
7190 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
7191 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
7192 * true iff it doesn't matter if the argument is in UTF-8 or not */
7193 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
7194 if (utf8_target && loceol - scan > max) {
7195 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
7196 * since here, to match at all, 1 char == 1 byte */
7197 loceol = scan + max;
7199 while (scan < loceol && UCHARAT(scan) == c) {
7203 else if (reginfo->is_utf8_pat) {
7205 STRLEN scan_char_len;
7207 /* When both target and pattern are UTF-8, we have to do
7209 while (hardcount < max
7211 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
7212 && memEQ(scan, STRING(p), scan_char_len))
7214 scan += scan_char_len;
7218 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
7220 /* Target isn't utf8; convert the character in the UTF-8
7221 * pattern to non-UTF8, and do a simple loop */
7222 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
7223 while (scan < loceol && UCHARAT(scan) == c) {
7226 } /* else pattern char is above Latin1, can't possibly match the
7231 /* Here, the string must be utf8; pattern isn't, and <c> is
7232 * different in utf8 than not, so can't compare them directly.
7233 * Outside the loop, find the two utf8 bytes that represent c, and
7234 * then look for those in sequence in the utf8 string */
7235 U8 high = UTF8_TWO_BYTE_HI(c);
7236 U8 low = UTF8_TWO_BYTE_LO(c);
7238 while (hardcount < max
7239 && scan + 1 < loceol
7240 && UCHARAT(scan) == high
7241 && UCHARAT(scan + 1) == low)
7249 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
7250 assert(! reginfo->is_utf8_pat);
7253 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7257 utf8_flags = FOLDEQ_LOCALE;
7260 case EXACTF: /* This node only generated for non-utf8 patterns */
7261 assert(! reginfo->is_utf8_pat);
7267 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
7271 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
7273 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7275 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
7278 if (c1 == CHRTEST_VOID) {
7279 /* Use full Unicode fold matching */
7280 char *tmpeol = reginfo->strend;
7281 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
7282 while (hardcount < max
7283 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
7284 STRING(p), NULL, pat_len,
7285 reginfo->is_utf8_pat, utf8_flags))
7288 tmpeol = reginfo->strend;
7292 else if (utf8_target) {
7294 while (scan < loceol
7296 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
7298 scan += UTF8SKIP(scan);
7303 while (scan < loceol
7305 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
7306 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
7308 scan += UTF8SKIP(scan);
7313 else if (c1 == c2) {
7314 while (scan < loceol && UCHARAT(scan) == c1) {
7319 while (scan < loceol &&
7320 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
7330 while (hardcount < max
7332 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
7334 scan += UTF8SKIP(scan);
7338 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
7343 /* The argument (FLAGS) to all the POSIX node types is the class number */
7350 if (! utf8_target) {
7351 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
7357 while (hardcount < max && scan < loceol
7358 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
7361 scan += UTF8SKIP(scan);
7374 if (utf8_target && loceol - scan > max) {
7376 /* We didn't adjust <loceol> at the beginning of this routine
7377 * because is UTF-8, but it is actually ok to do so, since here, to
7378 * match, 1 char == 1 byte. */
7379 loceol = scan + max;
7381 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
7394 if (! utf8_target) {
7395 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
7401 /* The complement of something that matches only ASCII matches all
7402 * non-ASCII, plus everything in ASCII that isn't in the class. */
7403 while (hardcount < max && scan < loceol
7404 && (! isASCII_utf8(scan)
7405 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
7407 scan += UTF8SKIP(scan);
7418 if (! utf8_target) {
7419 while (scan < loceol && to_complement
7420 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
7427 classnum = (_char_class_number) FLAGS(p);
7428 if (classnum < _FIRST_NON_SWASH_CC) {
7430 /* Here, a swash is needed for above-Latin1 code points.
7431 * Process as many Latin1 code points using the built-in rules.
7432 * Go to another loop to finish processing upon encountering
7433 * the first Latin1 code point. We could do that in this loop
7434 * as well, but the other way saves having to test if the swash
7435 * has been loaded every time through the loop: extra space to
7437 while (hardcount < max && scan < loceol) {
7438 if (UTF8_IS_INVARIANT(*scan)) {
7439 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
7446 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
7447 if (! (to_complement
7448 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
7457 goto found_above_latin1;
7464 /* For these character classes, the knowledge of how to handle
7465 * every code point is compiled in to Perl via a macro. This
7466 * code is written for making the loops as tight as possible.
7467 * It could be refactored to save space instead */
7469 case _CC_ENUM_SPACE: /* XXX would require separate code
7470 if we revert the change of \v
7473 case _CC_ENUM_PSXSPC:
7474 while (hardcount < max
7476 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7478 scan += UTF8SKIP(scan);
7482 case _CC_ENUM_BLANK:
7483 while (hardcount < max
7485 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7487 scan += UTF8SKIP(scan);
7491 case _CC_ENUM_XDIGIT:
7492 while (hardcount < max
7494 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7496 scan += UTF8SKIP(scan);
7500 case _CC_ENUM_VERTSPACE:
7501 while (hardcount < max
7503 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7505 scan += UTF8SKIP(scan);
7509 case _CC_ENUM_CNTRL:
7510 while (hardcount < max
7512 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7514 scan += UTF8SKIP(scan);
7519 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7525 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7527 /* Load the swash if not already present */
7528 if (! PL_utf8_swash_ptrs[classnum]) {
7529 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7530 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7534 PL_XPosix_ptrs[classnum], &flags);
7537 while (hardcount < max && scan < loceol
7538 && to_complement ^ cBOOL(_generic_utf8(
7541 swash_fetch(PL_utf8_swash_ptrs[classnum],
7545 scan += UTF8SKIP(scan);
7552 while (hardcount < max && scan < loceol &&
7553 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7558 /* LNBREAK can match one or two latin chars, which is ok, but we
7559 * have to use hardcount in this situation, and throw away the
7560 * adjustment to <loceol> done before the switch statement */
7561 loceol = reginfo->strend;
7562 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7583 /* These are all 0 width, so match right here or not at all. */
7587 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7596 c = scan - *startposp;
7600 GET_RE_DEBUG_FLAGS_DECL;
7602 SV * const prop = sv_newmortal();
7603 regprop(prog, prop, p, reginfo);
7604 PerlIO_printf(Perl_debug_log,
7605 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7606 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7614 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7616 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7617 create a copy so that changes the caller makes won't change the shared one.
7618 If <altsvp> is non-null, will return NULL in it, for back-compat.
7621 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7623 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7629 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL));
7632 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
7635 - reginclass - determine if a character falls into a character class
7637 n is the ANYOF regnode
7638 p is the target string
7639 p_end points to one byte beyond the end of the target string
7640 utf8_target tells whether p is in UTF-8.
7642 Returns true if matched; false otherwise.
7644 Note that this can be a synthetic start class, a combination of various
7645 nodes, so things you think might be mutually exclusive, such as locale,
7646 aren't. It can match both locale and non-locale
7651 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
7654 const char flags = ANYOF_FLAGS(n);
7658 PERL_ARGS_ASSERT_REGINCLASS;
7660 /* If c is not already the code point, get it. Note that
7661 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7662 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7664 c = utf8n_to_uvchr(p, p_end - p, &c_len,
7665 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7666 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7667 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7668 * UTF8_ALLOW_FFFF */
7669 if (c_len == (STRLEN)-1)
7670 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7673 /* If this character is potentially in the bitmap, check it */
7674 if (c < NUM_ANYOF_CODE_POINTS) {
7675 if (ANYOF_BITMAP_TEST(n, c))
7677 else if ((flags & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII)
7683 else if (flags & ANYOF_LOCALE_FLAGS) {
7684 if ((flags & ANYOF_LOC_FOLD)
7686 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7690 else if (ANYOF_POSIXL_TEST_ANY_SET(n)
7694 /* The data structure is arranged so bits 0, 2, 4, ... are set
7695 * if the class includes the Posix character class given by
7696 * bit/2; and 1, 3, 5, ... are set if the class includes the
7697 * complemented Posix class given by int(bit/2). So we loop
7698 * through the bits, each time changing whether we complement
7699 * the result or not. Suppose for the sake of illustration
7700 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7701 * is set, it means there is a match for this ANYOF node if the
7702 * character is in the class given by the expression (0 / 2 = 0
7703 * = \w). If it is in that class, isFOO_lc() will return 1,
7704 * and since 'to_complement' is 0, the result will stay TRUE,
7705 * and we exit the loop. Suppose instead that bit 0 is 0, but
7706 * bit 1 is 1. That means there is a match if the character
7707 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7708 * but will on bit 1. On the second iteration 'to_complement'
7709 * will be 1, so the exclusive or will reverse things, so we
7710 * are testing for \W. On the third iteration, 'to_complement'
7711 * will be 0, and we would be testing for \s; the fourth
7712 * iteration would test for \S, etc.
7714 * Note that this code assumes that all the classes are closed
7715 * under folding. For example, if a character matches \w, then
7716 * its fold does too; and vice versa. This should be true for
7717 * any well-behaved locale for all the currently defined Posix
7718 * classes, except for :lower: and :upper:, which are handled
7719 * by the pseudo-class :cased: which matches if either of the
7720 * other two does. To get rid of this assumption, an outer
7721 * loop could be used below to iterate over both the source
7722 * character, and its fold (if different) */
7725 int to_complement = 0;
7727 while (count < ANYOF_MAX) {
7728 if (ANYOF_POSIXL_TEST(n, count)
7729 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7742 /* If the bitmap didn't (or couldn't) match, and something outside the
7743 * bitmap could match, try that. */
7745 if (c >= NUM_ANYOF_CODE_POINTS
7746 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
7748 match = TRUE; /* Everything above the bitmap matches */
7750 else if ((flags & ANYOF_HAS_NONBITMAP_NON_UTF8_MATCHES)
7751 || (utf8_target && (flags & ANYOF_HAS_UTF8_NONBITMAP_MATCHES))
7752 || ((flags & ANYOF_LOC_FOLD)
7753 && IN_UTF8_CTYPE_LOCALE
7754 && ARG(n) != ANYOF_ONLY_HAS_BITMAP))
7756 SV* only_utf8_locale = NULL;
7757 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
7758 &only_utf8_locale, NULL);
7764 } else { /* Convert to utf8 */
7765 utf8_p = utf8_buffer;
7766 append_utf8_from_native_byte(*p, &utf8_p);
7767 utf8_p = utf8_buffer;
7770 if (swash_fetch(sw, utf8_p, TRUE)) {
7774 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
7775 match = _invlist_contains_cp(only_utf8_locale, c);
7779 if (UNICODE_IS_SUPER(c)
7780 && (flags & ANYOF_WARN_SUPER)
7781 && ckWARN_d(WARN_NON_UNICODE))
7783 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7784 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c);
7788 #if ANYOF_INVERT != 1
7789 /* Depending on compiler optimization cBOOL takes time, so if don't have to
7791 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
7794 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7795 return (flags & ANYOF_INVERT) ^ match;
7799 S_reghop3(U8 *s, SSize_t off, const U8* lim)
7801 /* return the position 'off' UTF-8 characters away from 's', forward if
7802 * 'off' >= 0, backwards if negative. But don't go outside of position
7803 * 'lim', which better be < s if off < 0 */
7805 PERL_ARGS_ASSERT_REGHOP3;
7808 while (off-- && s < lim) {
7809 /* XXX could check well-formedness here */
7814 while (off++ && s > lim) {
7816 if (UTF8_IS_CONTINUED(*s)) {
7817 while (s > lim && UTF8_IS_CONTINUATION(*s))
7820 /* XXX could check well-formedness here */
7827 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
7829 PERL_ARGS_ASSERT_REGHOP4;
7832 while (off-- && s < rlim) {
7833 /* XXX could check well-formedness here */
7838 while (off++ && s > llim) {
7840 if (UTF8_IS_CONTINUED(*s)) {
7841 while (s > llim && UTF8_IS_CONTINUATION(*s))
7844 /* XXX could check well-formedness here */
7850 /* like reghop3, but returns NULL on overrun, rather than returning last
7854 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
7856 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7859 while (off-- && s < lim) {
7860 /* XXX could check well-formedness here */
7867 while (off++ && s > lim) {
7869 if (UTF8_IS_CONTINUED(*s)) {
7870 while (s > lim && UTF8_IS_CONTINUATION(*s))
7873 /* XXX could check well-formedness here */
7882 /* when executing a regex that may have (?{}), extra stuff needs setting
7883 up that will be visible to the called code, even before the current
7884 match has finished. In particular:
7886 * $_ is localised to the SV currently being matched;
7887 * pos($_) is created if necessary, ready to be updated on each call-out
7889 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
7890 isn't set until the current pattern is successfully finished), so that
7891 $1 etc of the match-so-far can be seen;
7892 * save the old values of subbeg etc of the current regex, and set then
7893 to the current string (again, this is normally only done at the end
7898 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
7901 regexp *const rex = ReANY(reginfo->prog);
7902 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
7904 eval_state->rex = rex;
7907 /* Make $_ available to executed code. */
7908 if (reginfo->sv != DEFSV) {
7910 DEFSV_set(reginfo->sv);
7913 if (!(mg = mg_find_mglob(reginfo->sv))) {
7914 /* prepare for quick setting of pos */
7915 mg = sv_magicext_mglob(reginfo->sv);
7918 eval_state->pos_magic = mg;
7919 eval_state->pos = mg->mg_len;
7920 eval_state->pos_flags = mg->mg_flags;
7923 eval_state->pos_magic = NULL;
7925 if (!PL_reg_curpm) {
7926 /* PL_reg_curpm is a fake PMOP that we can attach the current
7927 * regex to and point PL_curpm at, so that $1 et al are visible
7928 * within a /(?{})/. It's just allocated once per interpreter the
7929 * first time its needed */
7930 Newxz(PL_reg_curpm, 1, PMOP);
7933 SV* const repointer = &PL_sv_undef;
7934 /* this regexp is also owned by the new PL_reg_curpm, which
7935 will try to free it. */
7936 av_push(PL_regex_padav, repointer);
7937 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
7938 PL_regex_pad = AvARRAY(PL_regex_padav);
7942 SET_reg_curpm(reginfo->prog);
7943 eval_state->curpm = PL_curpm;
7944 PL_curpm = PL_reg_curpm;
7945 if (RXp_MATCH_COPIED(rex)) {
7946 /* Here is a serious problem: we cannot rewrite subbeg,
7947 since it may be needed if this match fails. Thus
7948 $` inside (?{}) could fail... */
7949 eval_state->subbeg = rex->subbeg;
7950 eval_state->sublen = rex->sublen;
7951 eval_state->suboffset = rex->suboffset;
7952 eval_state->subcoffset = rex->subcoffset;
7954 eval_state->saved_copy = rex->saved_copy;
7956 RXp_MATCH_COPIED_off(rex);
7959 eval_state->subbeg = NULL;
7960 rex->subbeg = (char *)reginfo->strbeg;
7962 rex->subcoffset = 0;
7963 rex->sublen = reginfo->strend - reginfo->strbeg;
7967 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
7970 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
7972 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
7973 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
7976 Safefree(aux->poscache);
7980 /* undo the effects of S_setup_eval_state() */
7982 if (eval_state->subbeg) {
7983 regexp * const rex = eval_state->rex;
7984 rex->subbeg = eval_state->subbeg;
7985 rex->sublen = eval_state->sublen;
7986 rex->suboffset = eval_state->suboffset;
7987 rex->subcoffset = eval_state->subcoffset;
7989 rex->saved_copy = eval_state->saved_copy;
7991 RXp_MATCH_COPIED_on(rex);
7993 if (eval_state->pos_magic)
7995 eval_state->pos_magic->mg_len = eval_state->pos;
7996 eval_state->pos_magic->mg_flags =
7997 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
7998 | (eval_state->pos_flags & MGf_BYTES);
8001 PL_curpm = eval_state->curpm;
8004 PL_regmatch_state = aux->old_regmatch_state;
8005 PL_regmatch_slab = aux->old_regmatch_slab;
8007 /* free all slabs above current one - this must be the last action
8008 * of this function, as aux and eval_state are allocated within
8009 * slabs and may be freed here */
8011 s = PL_regmatch_slab->next;
8013 PL_regmatch_slab->next = NULL;
8015 regmatch_slab * const osl = s;
8024 S_to_utf8_substr(pTHX_ regexp *prog)
8026 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
8027 * on the converted value */
8031 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
8034 if (prog->substrs->data[i].substr
8035 && !prog->substrs->data[i].utf8_substr) {
8036 SV* const sv = newSVsv(prog->substrs->data[i].substr);
8037 prog->substrs->data[i].utf8_substr = sv;
8038 sv_utf8_upgrade(sv);
8039 if (SvVALID(prog->substrs->data[i].substr)) {
8040 if (SvTAIL(prog->substrs->data[i].substr)) {
8041 /* Trim the trailing \n that fbm_compile added last
8043 SvCUR_set(sv, SvCUR(sv) - 1);
8044 /* Whilst this makes the SV technically "invalid" (as its
8045 buffer is no longer followed by "\0") when fbm_compile()
8046 adds the "\n" back, a "\0" is restored. */
8047 fbm_compile(sv, FBMcf_TAIL);
8051 if (prog->substrs->data[i].substr == prog->check_substr)
8052 prog->check_utf8 = sv;
8058 S_to_byte_substr(pTHX_ regexp *prog)
8060 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
8061 * on the converted value; returns FALSE if can't be converted. */
8065 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
8068 if (prog->substrs->data[i].utf8_substr
8069 && !prog->substrs->data[i].substr) {
8070 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
8071 if (! sv_utf8_downgrade(sv, TRUE)) {
8074 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
8075 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
8076 /* Trim the trailing \n that fbm_compile added last
8078 SvCUR_set(sv, SvCUR(sv) - 1);
8079 fbm_compile(sv, FBMcf_TAIL);
8083 prog->substrs->data[i].substr = sv;
8084 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
8085 prog->check_substr = sv;
8094 * c-indentation-style: bsd
8096 * indent-tabs-mode: nil
8099 * ex: set ts=8 sts=4 sw=4 et: