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
40 /* At least one required character in the target string is expressible only in
42 static const char* const non_utf8_target_but_utf8_required
43 = "Can't match, because target string needs to be in UTF-8\n";
45 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
46 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
51 * pregcomp and pregexec -- regsub and regerror are not used in perl
53 * Copyright (c) 1986 by University of Toronto.
54 * Written by Henry Spencer. Not derived from licensed software.
56 * Permission is granted to anyone to use this software for any
57 * purpose on any computer system, and to redistribute it freely,
58 * subject to the following restrictions:
60 * 1. The author is not responsible for the consequences of use of
61 * this software, no matter how awful, even if they arise
64 * 2. The origin of this software must not be misrepresented, either
65 * by explicit claim or by omission.
67 * 3. Altered versions must be plainly marked as such, and must not
68 * be misrepresented as being the original software.
70 **** Alterations to Henry's code are...
72 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
73 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
74 **** by Larry Wall and others
76 **** You may distribute under the terms of either the GNU General Public
77 **** License or the Artistic License, as specified in the README file.
79 * Beware that some of this code is subtly aware of the way operator
80 * precedence is structured in regular expressions. Serious changes in
81 * regular-expression syntax might require a total rethink.
84 #define PERL_IN_REGEXEC_C
88 #ifdef PERL_IN_XSUB_RE
94 #include "inline_invlist.c"
95 #include "unicode_constants.h"
97 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
100 #define STATIC static
103 /* Valid for non-utf8 strings: avoids the reginclass
104 * call if there are no complications: i.e., if everything matchable is
105 * straight forward in the bitmap */
106 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \
107 : ANYOF_BITMAP_TEST(p,*(c)))
113 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
114 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
116 #define HOPc(pos,off) \
117 (char *)(reginfo->is_utf8_target \
118 ? reghop3((U8*)pos, off, \
119 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
121 #define HOPBACKc(pos, off) \
122 (char*)(reginfo->is_utf8_target \
123 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
124 : (pos - off >= reginfo->strbeg) \
128 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
129 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
132 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
133 #define NEXTCHR_IS_EOS (nextchr < 0)
135 #define SET_nextchr \
136 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
138 #define SET_locinput(p) \
143 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name) STMT_START { \
145 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
146 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
147 1, 0, NULL, &flags); \
152 /* If in debug mode, we test that a known character properly matches */
154 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
156 utf8_char_in_property) \
157 LOAD_UTF8_CHARCLASS(swash_ptr, property_name); \
158 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
160 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
162 utf8_char_in_property) \
163 LOAD_UTF8_CHARCLASS(swash_ptr, property_name)
166 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
167 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
168 swash_property_names[_CC_WORDCHAR], \
169 GREEK_SMALL_LETTER_IOTA_UTF8)
171 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
173 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
174 "_X_regular_begin", \
175 GREEK_SMALL_LETTER_IOTA_UTF8); \
176 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
178 COMBINING_GRAVE_ACCENT_UTF8); \
181 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
182 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
184 /* for use after a quantifier and before an EXACT-like node -- japhy */
185 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
187 * NOTE that *nothing* that affects backtracking should be in here, specifically
188 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
189 * node that is in between two EXACT like nodes when ascertaining what the required
190 * "follow" character is. This should probably be moved to regex compile time
191 * although it may be done at run time beause of the REF possibility - more
192 * investigation required. -- demerphq
194 #define JUMPABLE(rn) ( \
196 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
198 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
199 OP(rn) == PLUS || OP(rn) == MINMOD || \
201 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
203 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
205 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
208 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
209 we don't need this definition. */
210 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
211 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
212 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
215 /* ... so we use this as its faster. */
216 #define IS_TEXT(rn) ( OP(rn)==EXACT )
217 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
218 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
219 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
224 Search for mandatory following text node; for lookahead, the text must
225 follow but for lookbehind (rn->flags != 0) we skip to the next step.
227 #define FIND_NEXT_IMPT(rn) STMT_START { \
228 while (JUMPABLE(rn)) { \
229 const OPCODE type = OP(rn); \
230 if (type == SUSPEND || PL_regkind[type] == CURLY) \
231 rn = NEXTOPER(NEXTOPER(rn)); \
232 else if (type == PLUS) \
234 else if (type == IFMATCH) \
235 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
236 else rn += NEXT_OFF(rn); \
240 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
241 * These are for the pre-composed Hangul syllables, which are all in a
242 * contiguous block and arranged there in such a way so as to facilitate
243 * alorithmic determination of their characteristics. As such, they don't need
244 * a swash, but can be determined by simple arithmetic. Almost all are
245 * GCB=LVT, but every 28th one is a GCB=LV */
246 #define SBASE 0xAC00 /* Start of block */
247 #define SCount 11172 /* Length of block */
250 #define SLAB_FIRST(s) (&(s)->states[0])
251 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
253 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
254 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
255 static regmatch_state * S_push_slab(pTHX);
257 #define REGCP_PAREN_ELEMS 3
258 #define REGCP_OTHER_ELEMS 3
259 #define REGCP_FRAME_ELEMS 1
260 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
261 * are needed for the regexp context stack bookkeeping. */
264 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
267 const int retval = PL_savestack_ix;
268 const int paren_elems_to_push =
269 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
270 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
271 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
273 GET_RE_DEBUG_FLAGS_DECL;
275 PERL_ARGS_ASSERT_REGCPPUSH;
277 if (paren_elems_to_push < 0)
278 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
279 paren_elems_to_push);
281 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
282 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
283 " out of range (%lu-%ld)",
285 (unsigned long)maxopenparen,
288 SSGROW(total_elems + REGCP_FRAME_ELEMS);
291 if ((int)maxopenparen > (int)parenfloor)
292 PerlIO_printf(Perl_debug_log,
293 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
298 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
299 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
300 SSPUSHINT(rex->offs[p].end);
301 SSPUSHINT(rex->offs[p].start);
302 SSPUSHINT(rex->offs[p].start_tmp);
303 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
304 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
306 (IV)rex->offs[p].start,
307 (IV)rex->offs[p].start_tmp,
311 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
312 SSPUSHINT(maxopenparen);
313 SSPUSHINT(rex->lastparen);
314 SSPUSHINT(rex->lastcloseparen);
315 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
320 /* These are needed since we do not localize EVAL nodes: */
321 #define REGCP_SET(cp) \
323 PerlIO_printf(Perl_debug_log, \
324 " Setting an EVAL scope, savestack=%"IVdf"\n", \
325 (IV)PL_savestack_ix)); \
328 #define REGCP_UNWIND(cp) \
330 if (cp != PL_savestack_ix) \
331 PerlIO_printf(Perl_debug_log, \
332 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
333 (IV)(cp), (IV)PL_savestack_ix)); \
336 #define UNWIND_PAREN(lp, lcp) \
337 for (n = rex->lastparen; n > lp; n--) \
338 rex->offs[n].end = -1; \
339 rex->lastparen = n; \
340 rex->lastcloseparen = lcp;
344 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
349 GET_RE_DEBUG_FLAGS_DECL;
351 PERL_ARGS_ASSERT_REGCPPOP;
353 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
355 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
356 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
357 rex->lastcloseparen = SSPOPINT;
358 rex->lastparen = SSPOPINT;
359 *maxopenparen_p = SSPOPINT;
361 i -= REGCP_OTHER_ELEMS;
362 /* Now restore the parentheses context. */
364 if (i || rex->lastparen + 1 <= rex->nparens)
365 PerlIO_printf(Perl_debug_log,
366 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
371 paren = *maxopenparen_p;
372 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
374 rex->offs[paren].start_tmp = SSPOPINT;
375 rex->offs[paren].start = SSPOPINT;
377 if (paren <= rex->lastparen)
378 rex->offs[paren].end = tmps;
379 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
380 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
382 (IV)rex->offs[paren].start,
383 (IV)rex->offs[paren].start_tmp,
384 (IV)rex->offs[paren].end,
385 (paren > rex->lastparen ? "(skipped)" : ""));
390 /* It would seem that the similar code in regtry()
391 * already takes care of this, and in fact it is in
392 * a better location to since this code can #if 0-ed out
393 * but the code in regtry() is needed or otherwise tests
394 * requiring null fields (pat.t#187 and split.t#{13,14}
395 * (as of patchlevel 7877) will fail. Then again,
396 * this code seems to be necessary or otherwise
397 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
398 * --jhi updated by dapm */
399 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
400 if (i > *maxopenparen_p)
401 rex->offs[i].start = -1;
402 rex->offs[i].end = -1;
403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %s ..-1 undeffing\n",
406 (i > *maxopenparen_p) ? "-1" : " "
412 /* restore the parens and associated vars at savestack position ix,
413 * but without popping the stack */
416 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
418 I32 tmpix = PL_savestack_ix;
419 PL_savestack_ix = ix;
420 regcppop(rex, maxopenparen_p);
421 PL_savestack_ix = tmpix;
424 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
427 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
429 /* Returns a boolean as to whether or not 'character' is a member of the
430 * Posix character class given by 'classnum' that should be equivalent to a
431 * value in the typedef '_char_class_number'.
433 * Ideally this could be replaced by a just an array of function pointers
434 * to the C library functions that implement the macros this calls.
435 * However, to compile, the precise function signatures are required, and
436 * these may vary from platform to to platform. To avoid having to figure
437 * out what those all are on each platform, I (khw) am using this method,
438 * which adds an extra layer of function call overhead (unless the C
439 * optimizer strips it away). But we don't particularly care about
440 * performance with locales anyway. */
442 switch ((_char_class_number) classnum) {
443 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
444 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
445 case _CC_ENUM_ASCII: return isASCII_LC(character);
446 case _CC_ENUM_BLANK: return isBLANK_LC(character);
447 case _CC_ENUM_CASED: return isLOWER_LC(character)
448 || isUPPER_LC(character);
449 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
450 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
451 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
452 case _CC_ENUM_LOWER: return isLOWER_LC(character);
453 case _CC_ENUM_PRINT: return isPRINT_LC(character);
454 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
455 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
456 case _CC_ENUM_SPACE: return isSPACE_LC(character);
457 case _CC_ENUM_UPPER: return isUPPER_LC(character);
458 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
459 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
460 default: /* VERTSPACE should never occur in locales */
461 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
464 assert(0); /* NOTREACHED */
469 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
471 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
472 * 'character' is a member of the Posix character class given by 'classnum'
473 * that should be equivalent to a value in the typedef
474 * '_char_class_number'.
476 * This just calls isFOO_lc on the code point for the character if it is in
477 * the range 0-255. Outside that range, all characters avoid Unicode
478 * rules, ignoring any locale. So use the Unicode function if this class
479 * requires a swash, and use the Unicode macro otherwise. */
481 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
483 if (UTF8_IS_INVARIANT(*character)) {
484 return isFOO_lc(classnum, *character);
486 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
487 return isFOO_lc(classnum,
488 TWO_BYTE_UTF8_TO_UNI(*character, *(character + 1)));
491 if (classnum < _FIRST_NON_SWASH_CC) {
493 /* Initialize the swash unless done already */
494 if (! PL_utf8_swash_ptrs[classnum]) {
495 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
496 PL_utf8_swash_ptrs[classnum] = _core_swash_init("utf8",
497 swash_property_names[classnum], &PL_sv_undef, 1, 0, NULL, &flags);
500 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
502 TRUE /* is UTF */ ));
505 switch ((_char_class_number) classnum) {
507 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
509 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
510 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
511 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
512 default: return 0; /* Things like CNTRL are always
516 assert(0); /* NOTREACHED */
521 * pregexec and friends
524 #ifndef PERL_IN_XSUB_RE
526 - pregexec - match a regexp against a string
529 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
530 char *strbeg, I32 minend, SV *screamer, U32 nosave)
531 /* stringarg: the point in the string at which to begin matching */
532 /* strend: pointer to null at end of string */
533 /* strbeg: real beginning of string */
534 /* minend: end of match must be >= minend bytes after stringarg. */
535 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
536 * itself is accessed via the pointers above */
537 /* nosave: For optimizations. */
539 PERL_ARGS_ASSERT_PREGEXEC;
542 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
543 nosave ? 0 : REXEC_COPY_STR);
548 * Need to implement the following flags for reg_anch:
550 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
552 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
553 * INTUIT_AUTORITATIVE_ML
554 * INTUIT_ONCE_NOML - Intuit can match in one location only.
557 * Another flag for this function: SECOND_TIME (so that float substrs
558 * with giant delta may be not rechecked).
561 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
563 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
564 Otherwise, only SvCUR(sv) is used to get strbeg. */
566 /* XXXX We assume that strpos is strbeg unless sv. */
568 /* XXXX Some places assume that there is a fixed substring.
569 An update may be needed if optimizer marks as "INTUITable"
570 RExen without fixed substrings. Similarly, it is assumed that
571 lengths of all the strings are no more than minlen, thus they
572 cannot come from lookahead.
573 (Or minlen should take into account lookahead.)
574 NOTE: Some of this comment is not correct. minlen does now take account
575 of lookahead/behind. Further research is required. -- demerphq
579 /* A failure to find a constant substring means that there is no need to make
580 an expensive call to REx engine, thus we celebrate a failure. Similarly,
581 finding a substring too deep into the string means that fewer calls to
582 regtry() should be needed.
584 REx compiler's optimizer found 4 possible hints:
585 a) Anchored substring;
587 c) Whether we are anchored (beginning-of-line or \G);
588 d) First node (of those at offset 0) which may distinguish positions;
589 We use a)b)d) and multiline-part of c), and try to find a position in the
590 string which does not contradict any of them.
593 /* Most of decisions we do here should have been done at compile time.
594 The nodes of the REx which we used for the search should have been
595 deleted from the finite automaton. */
598 * rx: the regex to match against
599 * sv: the SV being matched: only used for utf8 flag; the string
600 * itself is accessed via the pointers below. Note that on
601 * something like an overloaded SV, SvPOK(sv) may be false
602 * and the string pointers may point to something unrelated to
604 * strbeg: real beginning of string
605 * strpos: the point in the string at which to begin matching
606 * strend: pointer to the byte following the last char of the string
607 * flags currently unused; set to 0
608 * data: currently unused; set to NULL
612 Perl_re_intuit_start(pTHX_
615 const char * const strbeg,
619 re_scream_pos_data *data)
622 struct regexp *const prog = ReANY(rx);
624 /* Should be nonnegative! */
629 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
631 char *other_last = NULL; /* other substr checked before this */
632 char *check_at = NULL; /* check substr found at this pos */
633 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
634 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
635 RXi_GET_DECL(prog,progi);
636 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
637 regmatch_info *const reginfo = ®info_buf;
639 const char * const i_strpos = strpos;
641 GET_RE_DEBUG_FLAGS_DECL;
643 PERL_ARGS_ASSERT_RE_INTUIT_START;
644 PERL_UNUSED_ARG(flags);
645 PERL_UNUSED_ARG(data);
647 /* CHR_DIST() would be more correct here but it makes things slow. */
648 if (prog->minlen > strend - strpos) {
649 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
650 "String too short... [re_intuit_start]\n"));
654 reginfo->is_utf8_target = cBOOL(utf8_target);
655 reginfo->info_aux = NULL;
656 reginfo->strbeg = strbeg;
657 reginfo->strend = strend;
658 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
660 /* not actually used within intuit, but zero for safety anyway */
661 reginfo->poscache_maxiter = 0;
664 if (!prog->check_utf8 && prog->check_substr)
665 to_utf8_substr(prog);
666 check = prog->check_utf8;
668 if (!prog->check_substr && prog->check_utf8) {
669 if (! to_byte_substr(prog)) {
670 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
673 check = prog->check_substr;
675 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
676 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
677 || ( (prog->extflags & RXf_ANCH_BOL)
678 && !multiline ) ); /* Check after \n? */
681 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
682 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
683 && (strpos != strbeg)) {
684 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
687 if (prog->check_offset_min == prog->check_offset_max
688 && !(prog->extflags & RXf_CANY_SEEN)
689 && ! multiline) /* /m can cause \n's to match that aren't
690 accounted for in the string max length.
691 See [perl #115242] */
693 /* Substring at constant offset from beg-of-str... */
696 s = HOP3c(strpos, prog->check_offset_min, strend);
699 slen = SvCUR(check); /* >= 1 */
701 if ( strend - s > slen || strend - s < slen - 1
702 || (strend - s == slen && strend[-1] != '\n')) {
703 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
706 /* Now should match s[0..slen-2] */
708 if (slen && (*SvPVX_const(check) != *s
710 && memNE(SvPVX_const(check), s, slen)))) {
712 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
716 else if (*SvPVX_const(check) != *s
717 || ((slen = SvCUR(check)) > 1
718 && memNE(SvPVX_const(check), s, slen)))
721 goto success_at_start;
724 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
726 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
727 end_shift = prog->check_end_shift;
730 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
731 - (SvTAIL(check) != 0);
732 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
734 if (end_shift < eshift)
738 else { /* Can match at random position */
741 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
742 end_shift = prog->check_end_shift;
744 /* end shift should be non negative here */
747 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
749 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
750 (IV)end_shift, RX_PRECOMP(prog));
754 /* Find a possible match in the region s..strend by looking for
755 the "check" substring in the region corrected by start/end_shift. */
758 I32 srch_start_shift = start_shift;
759 I32 srch_end_shift = end_shift;
762 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
763 srch_end_shift -= ((strbeg - s) - srch_start_shift);
764 srch_start_shift = strbeg - s;
766 DEBUG_OPTIMISE_MORE_r({
767 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
768 (IV)prog->check_offset_min,
769 (IV)srch_start_shift,
771 (IV)prog->check_end_shift);
774 if (prog->extflags & RXf_CANY_SEEN) {
775 start_point= (U8*)(s + srch_start_shift);
776 end_point= (U8*)(strend - srch_end_shift);
778 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
779 end_point= HOP3(strend, -srch_end_shift, strbeg);
781 DEBUG_OPTIMISE_MORE_r({
782 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
783 (int)(end_point - start_point),
784 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
788 s = fbm_instr( start_point, end_point,
789 check, multiline ? FBMrf_MULTILINE : 0);
791 /* Update the count-of-usability, remove useless subpatterns,
795 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
796 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
797 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
798 (s ? "Found" : "Did not find"),
799 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
800 ? "anchored" : "floating"),
803 (s ? " at offset " : "...\n") );
808 /* Finish the diagnostic message */
809 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
811 /* XXX dmq: first branch is for positive lookbehind...
812 Our check string is offset from the beginning of the pattern.
813 So we need to do any stclass tests offset forward from that
822 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
823 Start with the other substr.
824 XXXX no SCREAM optimization yet - and a very coarse implementation
825 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
826 *always* match. Probably should be marked during compile...
827 Probably it is right to do no SCREAM here...
830 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
831 : (prog->float_substr && prog->anchored_substr))
833 /* Take into account the "other" substring. */
834 /* XXXX May be hopelessly wrong for UTF... */
837 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
840 char * const last = HOP3c(s, -start_shift, strbeg);
842 char * const saved_s = s;
845 t = s - prog->check_offset_max;
846 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
848 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
853 t = HOP3c(t, prog->anchored_offset, strend);
854 if (t < other_last) /* These positions already checked */
856 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
859 /* XXXX It is not documented what units *_offsets are in.
860 We assume bytes, but this is clearly wrong.
861 Meaning this code needs to be carefully reviewed for errors.
865 /* On end-of-str: see comment below. */
866 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
867 if (must == &PL_sv_undef) {
869 DEBUG_r(must = prog->anchored_utf8); /* for debug */
874 HOP3(HOP3(last1, prog->anchored_offset, strend)
875 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
877 multiline ? FBMrf_MULTILINE : 0
880 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
881 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
882 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
883 (s ? "Found" : "Contradicts"),
884 quoted, RE_SV_TAIL(must));
889 if (last1 >= last2) {
890 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
891 ", giving up...\n"));
894 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
895 ", trying floating at offset %ld...\n",
896 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
897 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
898 s = HOP3c(last, 1, strend);
902 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
903 (long)(s - i_strpos)));
904 t = HOP3c(s, -prog->anchored_offset, strbeg);
905 other_last = HOP3c(s, 1, strend);
913 else { /* Take into account the floating substring. */
915 char * const saved_s = s;
918 t = HOP3c(s, -start_shift, strbeg);
920 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
921 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
922 last = HOP3c(t, prog->float_max_offset, strend);
923 s = HOP3c(t, prog->float_min_offset, strend);
926 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
927 must = utf8_target ? prog->float_utf8 : prog->float_substr;
928 /* fbm_instr() takes into account exact value of end-of-str
929 if the check is SvTAIL(ed). Since false positives are OK,
930 and end-of-str is not later than strend we are OK. */
931 if (must == &PL_sv_undef) {
933 DEBUG_r(must = prog->float_utf8); /* for debug message */
936 s = fbm_instr((unsigned char*)s,
937 (unsigned char*)last + SvCUR(must)
939 must, multiline ? FBMrf_MULTILINE : 0);
941 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
942 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
943 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
944 (s ? "Found" : "Contradicts"),
945 quoted, RE_SV_TAIL(must));
949 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
950 ", giving up...\n"));
953 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
954 ", trying anchored starting at offset %ld...\n",
955 (long)(saved_s + 1 - i_strpos)));
957 s = HOP3c(t, 1, strend);
961 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
962 (long)(s - i_strpos)));
963 other_last = s; /* Fix this later. --Hugo */
973 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
975 DEBUG_OPTIMISE_MORE_r(
976 PerlIO_printf(Perl_debug_log,
977 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
978 (IV)prog->check_offset_min,
979 (IV)prog->check_offset_max,
987 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
989 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
992 /* Fixed substring is found far enough so that the match
993 cannot start at strpos. */
995 if (ml_anch && t[-1] != '\n') {
996 /* Eventually fbm_*() should handle this, but often
997 anchored_offset is not 0, so this check will not be wasted. */
998 /* XXXX In the code below we prefer to look for "^" even in
999 presence of anchored substrings. And we search even
1000 beyond the found float position. These pessimizations
1001 are historical artefacts only. */
1003 while (t < strend - prog->minlen) {
1005 if (t < check_at - prog->check_offset_min) {
1006 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1007 /* Since we moved from the found position,
1008 we definitely contradict the found anchored
1009 substr. Due to the above check we do not
1010 contradict "check" substr.
1011 Thus we can arrive here only if check substr
1012 is float. Redo checking for "other"=="fixed".
1015 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1016 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
1017 goto do_other_anchored;
1019 /* We don't contradict the found floating substring. */
1020 /* XXXX Why not check for STCLASS? */
1022 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1023 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1026 /* Position contradicts check-string */
1027 /* XXXX probably better to look for check-string
1028 than for "\n", so one should lower the limit for t? */
1029 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1030 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1031 other_last = strpos = s = t + 1;
1036 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1037 PL_colors[0], PL_colors[1]));
1041 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1042 PL_colors[0], PL_colors[1]));
1046 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1049 /* The found string does not prohibit matching at strpos,
1050 - no optimization of calling REx engine can be performed,
1051 unless it was an MBOL and we are not after MBOL,
1052 or a future STCLASS check will fail this. */
1054 /* Even in this situation we may use MBOL flag if strpos is offset
1055 wrt the start of the string. */
1056 if (ml_anch && (strpos != strbeg) && strpos[-1] != '\n'
1057 /* May be due to an implicit anchor of m{.*foo} */
1058 && !(prog->intflags & PREGf_IMPLICIT))
1063 DEBUG_EXECUTE_r( if (ml_anch)
1064 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1065 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1068 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1070 prog->check_utf8 /* Could be deleted already */
1071 && --BmUSEFUL(prog->check_utf8) < 0
1072 && (prog->check_utf8 == prog->float_utf8)
1074 prog->check_substr /* Could be deleted already */
1075 && --BmUSEFUL(prog->check_substr) < 0
1076 && (prog->check_substr == prog->float_substr)
1079 /* If flags & SOMETHING - do not do it many times on the same match */
1080 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1081 /* XXX Does the destruction order has to change with utf8_target? */
1082 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1083 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1084 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1085 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1086 check = NULL; /* abort */
1088 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1089 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1090 if (prog->intflags & PREGf_IMPLICIT)
1091 prog->extflags &= ~RXf_ANCH_MBOL;
1092 /* XXXX This is a remnant of the old implementation. It
1093 looks wasteful, since now INTUIT can use many
1094 other heuristics. */
1095 prog->extflags &= ~RXf_USE_INTUIT;
1096 /* XXXX What other flags might need to be cleared in this branch? */
1102 /* Last resort... */
1103 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1104 /* trie stclasses are too expensive to use here, we are better off to
1105 leave it to regmatch itself */
1106 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1107 /* minlen == 0 is possible if regstclass is \b or \B,
1108 and the fixed substr is ''$.
1109 Since minlen is already taken into account, s+1 is before strend;
1110 accidentally, minlen >= 1 guaranties no false positives at s + 1
1111 even for \b or \B. But (minlen? 1 : 0) below assumes that
1112 regstclass does not come from lookahead... */
1113 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1114 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1115 const U8* const str = (U8*)STRING(progi->regstclass);
1116 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1117 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1120 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1121 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1122 else if (prog->float_substr || prog->float_utf8)
1123 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1127 if (checked_upto < s)
1129 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1130 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1133 s = find_byclass(prog, progi->regstclass, checked_upto, endpos,
1139 const char *what = NULL;
1141 if (endpos == strend) {
1142 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1143 "Could not match STCLASS...\n") );
1146 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1147 "This position contradicts STCLASS...\n") );
1148 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1150 checked_upto = HOPBACKc(endpos, start_shift);
1151 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1152 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1153 /* Contradict one of substrings */
1154 if (prog->anchored_substr || prog->anchored_utf8) {
1155 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1156 DEBUG_EXECUTE_r( what = "anchored" );
1158 s = HOP3c(t, 1, strend);
1159 if (s + start_shift + end_shift > strend) {
1160 /* XXXX Should be taken into account earlier? */
1161 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1162 "Could not match STCLASS...\n") );
1167 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1168 "Looking for %s substr starting at offset %ld...\n",
1169 what, (long)(s + start_shift - i_strpos)) );
1172 /* Have both, check_string is floating */
1173 if (t + start_shift >= check_at) /* Contradicts floating=check */
1174 goto retry_floating_check;
1175 /* Recheck anchored substring, but not floating... */
1179 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1180 "Looking for anchored substr starting at offset %ld...\n",
1181 (long)(other_last - i_strpos)) );
1182 goto do_other_anchored;
1184 /* Another way we could have checked stclass at the
1185 current position only: */
1190 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1191 "Looking for /%s^%s/m starting at offset %ld...\n",
1192 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1195 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1197 /* Check is floating substring. */
1198 retry_floating_check:
1199 t = check_at - start_shift;
1200 DEBUG_EXECUTE_r( what = "floating" );
1201 goto hop_and_restart;
1204 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1205 "By STCLASS: moving %ld --> %ld\n",
1206 (long)(t - i_strpos), (long)(s - i_strpos))
1210 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1211 "Does not contradict STCLASS...\n");
1216 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1217 PL_colors[4], (check ? "Guessed" : "Giving up"),
1218 PL_colors[5], (long)(s - i_strpos)) );
1221 fail_finish: /* Substring not found */
1222 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1223 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1225 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1226 PL_colors[4], PL_colors[5]));
1230 #define DECL_TRIE_TYPE(scan) \
1231 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1232 trie_type = ((scan->flags == EXACT) \
1233 ? (utf8_target ? trie_utf8 : trie_plain) \
1234 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1236 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1239 switch (trie_type) { \
1240 case trie_utf8_fold: \
1241 if ( foldlen>0 ) { \
1242 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1247 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1248 len = UTF8SKIP(uc); \
1249 skiplen = UNISKIP( uvc ); \
1250 foldlen -= skiplen; \
1251 uscan = foldbuf + skiplen; \
1254 case trie_latin_utf8_fold: \
1255 if ( foldlen>0 ) { \
1256 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1262 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, FOLD_FLAGS_FULL); \
1263 skiplen = UNISKIP( uvc ); \
1264 foldlen -= skiplen; \
1265 uscan = foldbuf + skiplen; \
1269 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1276 charid = trie->charmap[ uvc ]; \
1280 if (widecharmap) { \
1281 SV** const svpp = hv_fetch(widecharmap, \
1282 (char*)&uvc, sizeof(UV), 0); \
1284 charid = (U16)SvIV(*svpp); \
1289 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1293 && (ln == 1 || folder(s, pat_string, ln)) \
1294 && (reginfo->intuit || regtry(reginfo, &s)) )\
1300 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1302 while (s < strend) { \
1308 #define REXEC_FBC_SCAN(CoDe) \
1310 while (s < strend) { \
1316 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1317 REXEC_FBC_UTF8_SCAN( \
1319 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1328 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1331 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1340 #define REXEC_FBC_TRYIT \
1341 if ((reginfo->intuit || regtry(reginfo, &s))) \
1344 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1345 if (utf8_target) { \
1346 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1349 REXEC_FBC_CLASS_SCAN(CoNd); \
1352 #define DUMP_EXEC_POS(li,s,doutf8) \
1353 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1357 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1358 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1359 tmp = TEST_NON_UTF8(tmp); \
1360 REXEC_FBC_UTF8_SCAN( \
1361 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1370 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1371 if (s == reginfo->strbeg) { \
1375 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1376 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1379 LOAD_UTF8_CHARCLASS_ALNUM(); \
1380 REXEC_FBC_UTF8_SCAN( \
1381 if (tmp == ! (TeSt2_UtF8)) { \
1390 /* The only difference between the BOUND and NBOUND cases is that
1391 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1392 * NBOUND. This is accomplished by passing it in either the if or else clause,
1393 * with the other one being empty */
1394 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1395 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1397 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1398 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1400 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1401 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1403 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1404 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1407 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1408 * be passed in completely with the variable name being tested, which isn't
1409 * such a clean interface, but this is easier to read than it was before. We
1410 * are looking for the boundary (or non-boundary between a word and non-word
1411 * character. The utf8 and non-utf8 cases have the same logic, but the details
1412 * must be different. Find the "wordness" of the character just prior to this
1413 * one, and compare it with the wordness of this one. If they differ, we have
1414 * a boundary. At the beginning of the string, pretend that the previous
1415 * character was a new-line */
1416 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1417 if (utf8_target) { \
1420 else { /* Not utf8 */ \
1421 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1422 tmp = TEST_NON_UTF8(tmp); \
1424 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1433 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \
1436 /* We know what class REx starts with. Try to find this position... */
1437 /* if reginfo->intuit, its a dryrun */
1438 /* annoyingly all the vars in this routine have different names from their counterparts
1439 in regmatch. /grrr */
1442 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1443 const char *strend, regmatch_info *reginfo)
1446 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1447 char *pat_string; /* The pattern's exactish string */
1448 char *pat_end; /* ptr to end char of pat_string */
1449 re_fold_t folder; /* Function for computing non-utf8 folds */
1450 const U8 *fold_array; /* array for folding ords < 256 */
1456 I32 tmp = 1; /* Scratch variable? */
1457 const bool utf8_target = reginfo->is_utf8_target;
1458 UV utf8_fold_flags = 0;
1459 const bool is_utf8_pat = reginfo->is_utf8_pat;
1460 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1461 with a result inverts that result, as 0^1 =
1463 _char_class_number classnum;
1465 RXi_GET_DECL(prog,progi);
1467 PERL_ARGS_ASSERT_FIND_BYCLASS;
1469 /* We know what class it must start with. */
1472 case ANYOF_SYNTHETIC:
1473 case ANYOF_WARN_SUPER:
1475 REXEC_FBC_UTF8_CLASS_SCAN(
1476 reginclass(prog, c, (U8*)s, utf8_target));
1479 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1484 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1492 if (is_utf8_pat || utf8_target) {
1493 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1494 goto do_exactf_utf8;
1496 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1497 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1498 goto do_exactf_non_utf8; /* isn't dealt with by these */
1503 /* regcomp.c already folded this if pattern is in UTF-8 */
1504 utf8_fold_flags = 0;
1505 goto do_exactf_utf8;
1507 fold_array = PL_fold;
1509 goto do_exactf_non_utf8;
1512 if (is_utf8_pat || utf8_target) {
1513 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1514 goto do_exactf_utf8;
1516 fold_array = PL_fold_locale;
1517 folder = foldEQ_locale;
1518 goto do_exactf_non_utf8;
1522 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1524 goto do_exactf_utf8;
1526 case EXACTFU_TRICKYFOLD:
1528 if (is_utf8_pat || utf8_target) {
1529 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1530 goto do_exactf_utf8;
1533 /* Any 'ss' in the pattern should have been replaced by regcomp,
1534 * so we don't have to worry here about this single special case
1535 * in the Latin1 range */
1536 fold_array = PL_fold_latin1;
1537 folder = foldEQ_latin1;
1541 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1542 are no glitches with fold-length differences
1543 between the target string and pattern */
1545 /* The idea in the non-utf8 EXACTF* cases is to first find the
1546 * first character of the EXACTF* node and then, if necessary,
1547 * case-insensitively compare the full text of the node. c1 is the
1548 * first character. c2 is its fold. This logic will not work for
1549 * Unicode semantics and the german sharp ss, which hence should
1550 * not be compiled into a node that gets here. */
1551 pat_string = STRING(c);
1552 ln = STR_LEN(c); /* length to match in octets/bytes */
1554 /* We know that we have to match at least 'ln' bytes (which is the
1555 * same as characters, since not utf8). If we have to match 3
1556 * characters, and there are only 2 availabe, we know without
1557 * trying that it will fail; so don't start a match past the
1558 * required minimum number from the far end */
1559 e = HOP3c(strend, -((I32)ln), s);
1561 if (reginfo->intuit && e < s) {
1562 e = s; /* Due to minlen logic of intuit() */
1566 c2 = fold_array[c1];
1567 if (c1 == c2) { /* If char and fold are the same */
1568 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1571 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1579 /* If one of the operands is in utf8, we can't use the simpler folding
1580 * above, due to the fact that many different characters can have the
1581 * same fold, or portion of a fold, or different- length fold */
1582 pat_string = STRING(c);
1583 ln = STR_LEN(c); /* length to match in octets/bytes */
1584 pat_end = pat_string + ln;
1585 lnc = is_utf8_pat /* length to match in characters */
1586 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1589 /* We have 'lnc' characters to match in the pattern, but because of
1590 * multi-character folding, each character in the target can match
1591 * up to 3 characters (Unicode guarantees it will never exceed
1592 * this) if it is utf8-encoded; and up to 2 if not (based on the
1593 * fact that the Latin 1 folds are already determined, and the
1594 * only multi-char fold in that range is the sharp-s folding to
1595 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1596 * string character. Adjust lnc accordingly, rounding up, so that
1597 * if we need to match at least 4+1/3 chars, that really is 5. */
1598 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1599 lnc = (lnc + expansion - 1) / expansion;
1601 /* As in the non-UTF8 case, if we have to match 3 characters, and
1602 * only 2 are left, it's guaranteed to fail, so don't start a
1603 * match that would require us to go beyond the end of the string
1605 e = HOP3c(strend, -((I32)lnc), s);
1607 if (reginfo->intuit && e < s) {
1608 e = s; /* Due to minlen logic of intuit() */
1611 /* XXX Note that we could recalculate e to stop the loop earlier,
1612 * as the worst case expansion above will rarely be met, and as we
1613 * go along we would usually find that e moves further to the left.
1614 * This would happen only after we reached the point in the loop
1615 * where if there were no expansion we should fail. Unclear if
1616 * worth the expense */
1619 char *my_strend= (char *)strend;
1620 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1621 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1622 && (reginfo->intuit || regtry(reginfo, &s)) )
1626 s += (utf8_target) ? UTF8SKIP(s) : 1;
1631 RXp_MATCH_TAINTED_on(prog);
1632 FBC_BOUND(isWORDCHAR_LC,
1633 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1634 isWORDCHAR_LC_utf8((U8*)s));
1637 RXp_MATCH_TAINTED_on(prog);
1638 FBC_NBOUND(isWORDCHAR_LC,
1639 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1640 isWORDCHAR_LC_utf8((U8*)s));
1643 FBC_BOUND(isWORDCHAR,
1644 isWORDCHAR_uni(tmp),
1645 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1648 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1650 isWORDCHAR_A((U8*)s));
1653 FBC_NBOUND(isWORDCHAR,
1654 isWORDCHAR_uni(tmp),
1655 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1658 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1660 isWORDCHAR_A((U8*)s));
1663 FBC_BOUND(isWORDCHAR_L1,
1664 isWORDCHAR_uni(tmp),
1665 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1668 FBC_NBOUND(isWORDCHAR_L1,
1669 isWORDCHAR_uni(tmp),
1670 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1673 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1674 is_LNBREAK_latin1_safe(s, strend)
1678 /* The argument to all the POSIX node types is the class number to pass to
1679 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1686 RXp_MATCH_TAINTED_on(prog);
1687 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1688 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1703 /* The complement of something that matches only ASCII matches all
1704 * UTF-8 variant code points, plus everything in ASCII that isn't
1706 REXEC_FBC_UTF8_CLASS_SCAN(! UTF8_IS_INVARIANT(*s)
1707 || ! _generic_isCC_A(*s, FLAGS(c)));
1716 /* Don't need to worry about utf8, as it can match only a single
1717 * byte invariant character. */
1718 REXEC_FBC_CLASS_SCAN(
1719 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1727 if (! utf8_target) {
1728 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1734 classnum = (_char_class_number) FLAGS(c);
1735 if (classnum < _FIRST_NON_SWASH_CC) {
1736 while (s < strend) {
1738 /* We avoid loading in the swash as long as possible, but
1739 * should we have to, we jump to a separate loop. This
1740 * extra 'if' statement is what keeps this code from being
1741 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1742 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1743 goto found_above_latin1;
1745 if ((UTF8_IS_INVARIANT(*s)
1746 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1748 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1749 && to_complement ^ cBOOL(
1750 _generic_isCC(TWO_BYTE_UTF8_TO_UNI(*s, *(s + 1)),
1753 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1765 else switch (classnum) { /* These classes are implemented as
1767 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1768 revert the change of \v matching this */
1771 case _CC_ENUM_PSXSPC:
1772 REXEC_FBC_UTF8_CLASS_SCAN(
1773 to_complement ^ cBOOL(isSPACE_utf8(s)));
1776 case _CC_ENUM_BLANK:
1777 REXEC_FBC_UTF8_CLASS_SCAN(
1778 to_complement ^ cBOOL(isBLANK_utf8(s)));
1781 case _CC_ENUM_XDIGIT:
1782 REXEC_FBC_UTF8_CLASS_SCAN(
1783 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1786 case _CC_ENUM_VERTSPACE:
1787 REXEC_FBC_UTF8_CLASS_SCAN(
1788 to_complement ^ cBOOL(isVERTWS_utf8(s)));
1791 case _CC_ENUM_CNTRL:
1792 REXEC_FBC_UTF8_CLASS_SCAN(
1793 to_complement ^ cBOOL(isCNTRL_utf8(s)));
1797 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
1798 assert(0); /* NOTREACHED */
1803 found_above_latin1: /* Here we have to load a swash to get the result
1804 for the current code point */
1805 if (! PL_utf8_swash_ptrs[classnum]) {
1806 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1807 PL_utf8_swash_ptrs[classnum] =
1808 _core_swash_init("utf8", swash_property_names[classnum],
1809 &PL_sv_undef, 1, 0, NULL, &flags);
1812 /* This is a copy of the loop above for swash classes, though using the
1813 * FBC macro instead of being expanded out. Since we've loaded the
1814 * swash, we don't have to check for that each time through the loop */
1815 REXEC_FBC_UTF8_CLASS_SCAN(
1816 to_complement ^ cBOOL(_generic_utf8(
1819 swash_fetch(PL_utf8_swash_ptrs[classnum],
1827 /* what trie are we using right now */
1828 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1829 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
1830 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1832 const char *last_start = strend - trie->minlen;
1834 const char *real_start = s;
1836 STRLEN maxlen = trie->maxlen;
1838 U8 **points; /* map of where we were in the input string
1839 when reading a given char. For ASCII this
1840 is unnecessary overhead as the relationship
1841 is always 1:1, but for Unicode, especially
1842 case folded Unicode this is not true. */
1843 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1847 GET_RE_DEBUG_FLAGS_DECL;
1849 /* We can't just allocate points here. We need to wrap it in
1850 * an SV so it gets freed properly if there is a croak while
1851 * running the match */
1854 sv_points=newSV(maxlen * sizeof(U8 *));
1855 SvCUR_set(sv_points,
1856 maxlen * sizeof(U8 *));
1857 SvPOK_on(sv_points);
1858 sv_2mortal(sv_points);
1859 points=(U8**)SvPV_nolen(sv_points );
1860 if ( trie_type != trie_utf8_fold
1861 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1864 bitmap=(U8*)trie->bitmap;
1866 bitmap=(U8*)ANYOF_BITMAP(c);
1868 /* this is the Aho-Corasick algorithm modified a touch
1869 to include special handling for long "unknown char" sequences.
1870 The basic idea being that we use AC as long as we are dealing
1871 with a possible matching char, when we encounter an unknown char
1872 (and we have not encountered an accepting state) we scan forward
1873 until we find a legal starting char.
1874 AC matching is basically that of trie matching, except that when
1875 we encounter a failing transition, we fall back to the current
1876 states "fail state", and try the current char again, a process
1877 we repeat until we reach the root state, state 1, or a legal
1878 transition. If we fail on the root state then we can either
1879 terminate if we have reached an accepting state previously, or
1880 restart the entire process from the beginning if we have not.
1883 while (s <= last_start) {
1884 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1892 U8 *uscan = (U8*)NULL;
1893 U8 *leftmost = NULL;
1895 U32 accepted_word= 0;
1899 while ( state && uc <= (U8*)strend ) {
1901 U32 word = aho->states[ state ].wordnum;
1905 DEBUG_TRIE_EXECUTE_r(
1906 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1907 dump_exec_pos( (char *)uc, c, strend, real_start,
1908 (char *)uc, utf8_target );
1909 PerlIO_printf( Perl_debug_log,
1910 " Scanning for legal start char...\n");
1914 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1918 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1924 if (uc >(U8*)last_start) break;
1928 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1929 if (!leftmost || lpos < leftmost) {
1930 DEBUG_r(accepted_word=word);
1936 points[pointpos++ % maxlen]= uc;
1937 if (foldlen || uc < (U8*)strend) {
1938 REXEC_TRIE_READ_CHAR(trie_type, trie,
1940 uscan, len, uvc, charid, foldlen,
1942 DEBUG_TRIE_EXECUTE_r({
1943 dump_exec_pos( (char *)uc, c, strend,
1944 real_start, s, utf8_target);
1945 PerlIO_printf(Perl_debug_log,
1946 " Charid:%3u CP:%4"UVxf" ",
1958 word = aho->states[ state ].wordnum;
1960 base = aho->states[ state ].trans.base;
1962 DEBUG_TRIE_EXECUTE_r({
1964 dump_exec_pos( (char *)uc, c, strend, real_start,
1966 PerlIO_printf( Perl_debug_log,
1967 "%sState: %4"UVxf", word=%"UVxf,
1968 failed ? " Fail transition to " : "",
1969 (UV)state, (UV)word);
1975 ( ((offset = base + charid
1976 - 1 - trie->uniquecharcount)) >= 0)
1977 && ((U32)offset < trie->lasttrans)
1978 && trie->trans[offset].check == state
1979 && (tmp=trie->trans[offset].next))
1981 DEBUG_TRIE_EXECUTE_r(
1982 PerlIO_printf( Perl_debug_log," - legal\n"));
1987 DEBUG_TRIE_EXECUTE_r(
1988 PerlIO_printf( Perl_debug_log," - fail\n"));
1990 state = aho->fail[state];
1994 /* we must be accepting here */
1995 DEBUG_TRIE_EXECUTE_r(
1996 PerlIO_printf( Perl_debug_log," - accepting\n"));
2005 if (!state) state = 1;
2008 if ( aho->states[ state ].wordnum ) {
2009 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2010 if (!leftmost || lpos < leftmost) {
2011 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2016 s = (char*)leftmost;
2017 DEBUG_TRIE_EXECUTE_r({
2019 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2020 (UV)accepted_word, (IV)(s - real_start)
2023 if (reginfo->intuit || regtry(reginfo, &s)) {
2029 DEBUG_TRIE_EXECUTE_r({
2030 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2033 DEBUG_TRIE_EXECUTE_r(
2034 PerlIO_printf( Perl_debug_log,"No match.\n"));
2043 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2053 - regexec_flags - match a regexp against a string
2056 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2057 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2058 /* stringarg: the point in the string at which to begin matching */
2059 /* strend: pointer to null at end of string */
2060 /* strbeg: real beginning of string */
2061 /* minend: end of match must be >= minend bytes after stringarg. */
2062 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2063 * itself is accessed via the pointers above */
2064 /* data: May be used for some additional optimizations.
2065 Currently its only used, with a U32 cast, for transmitting
2066 the ganch offset when doing a /g match. This will change */
2067 /* nosave: For optimizations. */
2071 struct regexp *const prog = ReANY(rx);
2074 char *startpos = stringarg;
2075 I32 minlen; /* must match at least this many chars */
2076 I32 dontbother = 0; /* how many characters not to try at end */
2077 I32 end_shift = 0; /* Same for the end. */ /* CC */
2078 I32 scream_pos = -1; /* Internal iterator of scream. */
2079 char *scream_olds = NULL;
2080 const bool utf8_target = cBOOL(DO_UTF8(sv));
2082 RXi_GET_DECL(prog,progi);
2083 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2084 regmatch_info *const reginfo = ®info_buf;
2085 regexp_paren_pair *swap = NULL;
2087 GET_RE_DEBUG_FLAGS_DECL;
2089 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2090 PERL_UNUSED_ARG(data);
2092 /* Be paranoid... */
2093 if (prog == NULL || startpos == NULL) {
2094 Perl_croak(aTHX_ "NULL regexp parameter");
2099 debug_start_match(rx, utf8_target, startpos, strend,
2104 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2105 * which will call destuctors to reset PL_regmatch_state, free higher
2106 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2107 * regmatch_info_aux_eval */
2109 oldsave = PL_savestack_ix;
2111 multiline = prog->extflags & RXf_PMf_MULTILINE;
2112 minlen = prog->minlen;
2114 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2115 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2116 "String too short [regexec_flags]...\n"));
2120 /* Check validity of program. */
2121 if (UCHARAT(progi->program) != REG_MAGIC) {
2122 Perl_croak(aTHX_ "corrupted regexp program");
2125 RX_MATCH_TAINTED_off(rx);
2127 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2128 reginfo->intuit = 0;
2129 reginfo->is_utf8_target = cBOOL(utf8_target);
2130 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2131 reginfo->warned = FALSE;
2132 reginfo->strbeg = strbeg;
2134 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2135 reginfo->strend = strend;
2136 /* see how far we have to get to not match where we matched before */
2137 reginfo->till = startpos+minend;
2139 /* reserve next 2 or 3 slots in PL_regmatch_state:
2140 * slot N+0: may currently be in use: skip it
2141 * slot N+1: use for regmatch_info_aux struct
2142 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2143 * slot N+3: ready for use by regmatch()
2147 regmatch_state *old_regmatch_state;
2148 regmatch_slab *old_regmatch_slab;
2149 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2151 /* on first ever match, allocate first slab */
2152 if (!PL_regmatch_slab) {
2153 Newx(PL_regmatch_slab, 1, regmatch_slab);
2154 PL_regmatch_slab->prev = NULL;
2155 PL_regmatch_slab->next = NULL;
2156 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2159 old_regmatch_state = PL_regmatch_state;
2160 old_regmatch_slab = PL_regmatch_slab;
2162 for (i=0; i <= max; i++) {
2164 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2166 reginfo->info_aux_eval =
2167 reginfo->info_aux->info_aux_eval =
2168 &(PL_regmatch_state->u.info_aux_eval);
2170 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2171 PL_regmatch_state = S_push_slab(aTHX);
2174 /* note initial PL_regmatch_state position; at end of match we'll
2175 * pop back to there and free any higher slabs */
2177 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2178 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2179 reginfo->info_aux->poscache = NULL;
2181 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2183 if ((prog->extflags & RXf_EVAL_SEEN))
2184 S_setup_eval_state(aTHX_ reginfo);
2186 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2189 /* If there is a "must appear" string, look for it. */
2192 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2194 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2195 reginfo->ganch = startpos + prog->gofs;
2196 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2197 "GPOS IGNOREPOS: reginfo->ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2198 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2200 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2201 && mg->mg_len >= 0) {
2202 reginfo->ganch = strbeg + mg->mg_len; /* Defined pos() */
2203 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2204 "GPOS MAGIC: reginfo->ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2206 if (prog->extflags & RXf_ANCH_GPOS) {
2207 if (s > reginfo->ganch)
2209 s = reginfo->ganch - prog->gofs;
2210 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2211 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2217 reginfo->ganch = strbeg + PTR2UV(data);
2218 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2219 "GPOS DATA: reginfo->ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2221 } else { /* pos() not defined */
2222 reginfo->ganch = strbeg;
2223 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2224 "GPOS: reginfo->ganch = strbeg\n"));
2227 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2228 /* We have to be careful. If the previous successful match
2229 was from this regex we don't want a subsequent partially
2230 successful match to clobber the old results.
2231 So when we detect this possibility we add a swap buffer
2232 to the re, and switch the buffer each match. If we fail,
2233 we switch it back; otherwise we leave it swapped.
2236 /* do we need a save destructor here for eval dies? */
2237 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2238 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2239 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2245 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2246 re_scream_pos_data d;
2248 d.scream_olds = &scream_olds;
2249 d.scream_pos = &scream_pos;
2250 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, &d);
2252 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2253 goto phooey; /* not present */
2259 /* Simplest case: anchored match need be tried only once. */
2260 /* [unless only anchor is BOL and multiline is set] */
2261 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2262 if (s == startpos && regtry(reginfo, &startpos))
2264 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2265 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2270 dontbother = minlen - 1;
2271 end = HOP3c(strend, -dontbother, strbeg) - 1;
2272 /* for multiline we only have to try after newlines */
2273 if (prog->check_substr || prog->check_utf8) {
2274 /* because of the goto we can not easily reuse the macros for bifurcating the
2275 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2278 goto after_try_utf8;
2280 if (regtry(reginfo, &s)) {
2287 if (prog->extflags & RXf_USE_INTUIT) {
2288 s = re_intuit_start(rx, sv, strbeg,
2289 s + UTF8SKIP(s), strend, flags, NULL);
2298 } /* end search for check string in unicode */
2300 if (s == startpos) {
2301 goto after_try_latin;
2304 if (regtry(reginfo, &s)) {
2311 if (prog->extflags & RXf_USE_INTUIT) {
2312 s = re_intuit_start(rx, sv, strbeg,
2313 s + 1, strend, flags, NULL);
2322 } /* end search for check string in latin*/
2323 } /* end search for check string */
2324 else { /* search for newline */
2326 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2329 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2330 while (s <= end) { /* note it could be possible to match at the end of the string */
2331 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2332 if (regtry(reginfo, &s))
2336 } /* end search for newline */
2337 } /* end anchored/multiline check string search */
2339 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2341 /* the warning about reginfo->ganch being used without initialization
2342 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2343 and we only enter this block when the same bit is set. */
2344 char *tmp_s = reginfo->ganch - prog->gofs;
2346 if (tmp_s >= strbeg && regtry(reginfo, &tmp_s))
2351 /* Messy cases: unanchored match. */
2352 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2353 /* we have /x+whatever/ */
2354 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2360 if (! prog->anchored_utf8) {
2361 to_utf8_substr(prog);
2363 ch = SvPVX_const(prog->anchored_utf8)[0];
2366 DEBUG_EXECUTE_r( did_match = 1 );
2367 if (regtry(reginfo, &s)) goto got_it;
2369 while (s < strend && *s == ch)
2376 if (! prog->anchored_substr) {
2377 if (! to_byte_substr(prog)) {
2378 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2381 ch = SvPVX_const(prog->anchored_substr)[0];
2384 DEBUG_EXECUTE_r( did_match = 1 );
2385 if (regtry(reginfo, &s)) goto got_it;
2387 while (s < strend && *s == ch)
2392 DEBUG_EXECUTE_r(if (!did_match)
2393 PerlIO_printf(Perl_debug_log,
2394 "Did not find anchored character...\n")
2397 else if (prog->anchored_substr != NULL
2398 || prog->anchored_utf8 != NULL
2399 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2400 && prog->float_max_offset < strend - s)) {
2405 char *last1; /* Last position checked before */
2409 if (prog->anchored_substr || prog->anchored_utf8) {
2411 if (! prog->anchored_utf8) {
2412 to_utf8_substr(prog);
2414 must = prog->anchored_utf8;
2417 if (! prog->anchored_substr) {
2418 if (! to_byte_substr(prog)) {
2419 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2422 must = prog->anchored_substr;
2424 back_max = back_min = prog->anchored_offset;
2427 if (! prog->float_utf8) {
2428 to_utf8_substr(prog);
2430 must = prog->float_utf8;
2433 if (! prog->float_substr) {
2434 if (! to_byte_substr(prog)) {
2435 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2438 must = prog->float_substr;
2440 back_max = prog->float_max_offset;
2441 back_min = prog->float_min_offset;
2447 last = HOP3c(strend, /* Cannot start after this */
2448 -(I32)(CHR_SVLEN(must)
2449 - (SvTAIL(must) != 0) + back_min), strbeg);
2451 if (s > reginfo->strbeg)
2452 last1 = HOPc(s, -1);
2454 last1 = s - 1; /* bogus */
2456 /* XXXX check_substr already used to find "s", can optimize if
2457 check_substr==must. */
2459 dontbother = end_shift;
2460 strend = HOPc(strend, -dontbother);
2461 while ( (s <= last) &&
2462 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2463 (unsigned char*)strend, must,
2464 multiline ? FBMrf_MULTILINE : 0)) ) {
2465 DEBUG_EXECUTE_r( did_match = 1 );
2466 if (HOPc(s, -back_max) > last1) {
2467 last1 = HOPc(s, -back_min);
2468 s = HOPc(s, -back_max);
2471 char * const t = (last1 >= reginfo->strbeg)
2472 ? HOPc(last1, 1) : last1 + 1;
2474 last1 = HOPc(s, -back_min);
2478 while (s <= last1) {
2479 if (regtry(reginfo, &s))
2482 s++; /* to break out of outer loop */
2489 while (s <= last1) {
2490 if (regtry(reginfo, &s))
2496 DEBUG_EXECUTE_r(if (!did_match) {
2497 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2498 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2499 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2500 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2501 ? "anchored" : "floating"),
2502 quoted, RE_SV_TAIL(must));
2506 else if ( (c = progi->regstclass) ) {
2508 const OPCODE op = OP(progi->regstclass);
2509 /* don't bother with what can't match */
2510 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2511 strend = HOPc(strend, -(minlen - 1));
2514 SV * const prop = sv_newmortal();
2515 regprop(prog, prop, c);
2517 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2519 PerlIO_printf(Perl_debug_log,
2520 "Matching stclass %.*s against %s (%d bytes)\n",
2521 (int)SvCUR(prop), SvPVX_const(prop),
2522 quoted, (int)(strend - s));
2525 if (find_byclass(prog, c, s, strend, reginfo))
2527 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2531 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2539 if (! prog->float_utf8) {
2540 to_utf8_substr(prog);
2542 float_real = prog->float_utf8;
2545 if (! prog->float_substr) {
2546 if (! to_byte_substr(prog)) {
2547 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2550 float_real = prog->float_substr;
2553 little = SvPV_const(float_real, len);
2554 if (SvTAIL(float_real)) {
2555 /* This means that float_real contains an artificial \n on
2556 * the end due to the presence of something like this:
2557 * /foo$/ where we can match both "foo" and "foo\n" at the
2558 * end of the string. So we have to compare the end of the
2559 * string first against the float_real without the \n and
2560 * then against the full float_real with the string. We
2561 * have to watch out for cases where the string might be
2562 * smaller than the float_real or the float_real without
2564 char *checkpos= strend - len;
2566 PerlIO_printf(Perl_debug_log,
2567 "%sChecking for float_real.%s\n",
2568 PL_colors[4], PL_colors[5]));
2569 if (checkpos + 1 < strbeg) {
2570 /* can't match, even if we remove the trailing \n
2571 * string is too short to match */
2573 PerlIO_printf(Perl_debug_log,
2574 "%sString shorter than required trailing substring, cannot match.%s\n",
2575 PL_colors[4], PL_colors[5]));
2577 } else if (memEQ(checkpos + 1, little, len - 1)) {
2578 /* can match, the end of the string matches without the
2580 last = checkpos + 1;
2581 } else if (checkpos < strbeg) {
2582 /* cant match, string is too short when the "\n" is
2585 PerlIO_printf(Perl_debug_log,
2586 "%sString does not contain required trailing substring, cannot match.%s\n",
2587 PL_colors[4], PL_colors[5]));
2589 } else if (!multiline) {
2590 /* non multiline match, so compare with the "\n" at the
2591 * end of the string */
2592 if (memEQ(checkpos, little, len)) {
2596 PerlIO_printf(Perl_debug_log,
2597 "%sString does not contain required trailing substring, cannot match.%s\n",
2598 PL_colors[4], PL_colors[5]));
2602 /* multiline match, so we have to search for a place
2603 * where the full string is located */
2609 last = rninstr(s, strend, little, little + len);
2611 last = strend; /* matching "$" */
2614 /* at one point this block contained a comment which was
2615 * probably incorrect, which said that this was a "should not
2616 * happen" case. Even if it was true when it was written I am
2617 * pretty sure it is not anymore, so I have removed the comment
2618 * and replaced it with this one. Yves */
2620 PerlIO_printf(Perl_debug_log,
2621 "String does not contain required substring, cannot match.\n"
2625 dontbother = strend - last + prog->float_min_offset;
2627 if (minlen && (dontbother < minlen))
2628 dontbother = minlen - 1;
2629 strend -= dontbother; /* this one's always in bytes! */
2630 /* We don't know much -- general case. */
2633 if (regtry(reginfo, &s))
2642 if (regtry(reginfo, &s))
2644 } while (s++ < strend);
2654 PerlIO_printf(Perl_debug_log,
2655 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2662 /* clean up; this will trigger destructors that will free all slabs
2663 * above the current one, and cleanup the regmatch_info_aux
2664 * and regmatch_info_aux_eval sructs */
2666 LEAVE_SCOPE(oldsave);
2668 if (RXp_PAREN_NAMES(prog))
2669 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2671 RX_MATCH_UTF8_set(rx, utf8_target);
2673 /* make sure $`, $&, $', and $digit will work later */
2674 if ( !(flags & REXEC_NOT_FIRST) ) {
2675 if (flags & REXEC_COPY_STR) {
2679 PerlIO_printf(Perl_debug_log,
2680 "Copy on write: regexp capture, type %d\n",
2683 RX_MATCH_COPY_FREE(rx);
2684 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2685 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2686 assert (SvPOKp(prog->saved_copy));
2687 prog->sublen = reginfo->strend - strbeg;
2688 prog->suboffset = 0;
2689 prog->subcoffset = 0;
2694 I32 max = reginfo->strend - strbeg;
2697 if ( (flags & REXEC_COPY_SKIP_POST)
2698 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2699 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2700 ) { /* don't copy $' part of string */
2703 /* calculate the right-most part of the string covered
2704 * by a capture. Due to look-ahead, this may be to
2705 * the right of $&, so we have to scan all captures */
2706 while (n <= prog->lastparen) {
2707 if (prog->offs[n].end > max)
2708 max = prog->offs[n].end;
2712 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2713 ? prog->offs[0].start
2715 assert(max >= 0 && max <= reginfo->strend - strbeg);
2718 if ( (flags & REXEC_COPY_SKIP_PRE)
2719 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2720 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2721 ) { /* don't copy $` part of string */
2724 /* calculate the left-most part of the string covered
2725 * by a capture. Due to look-behind, this may be to
2726 * the left of $&, so we have to scan all captures */
2727 while (min && n <= prog->lastparen) {
2728 if ( prog->offs[n].start != -1
2729 && prog->offs[n].start < min)
2731 min = prog->offs[n].start;
2735 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2736 && min > prog->offs[0].end
2738 min = prog->offs[0].end;
2742 assert(min >= 0 && min <= max
2743 && min <= reginfo->strend - strbeg);
2746 if (RX_MATCH_COPIED(rx)) {
2747 if (sublen > prog->sublen)
2749 (char*)saferealloc(prog->subbeg, sublen+1);
2752 prog->subbeg = (char*)safemalloc(sublen+1);
2753 Copy(strbeg + min, prog->subbeg, sublen, char);
2754 prog->subbeg[sublen] = '\0';
2755 prog->suboffset = min;
2756 prog->sublen = sublen;
2757 RX_MATCH_COPIED_on(rx);
2759 prog->subcoffset = prog->suboffset;
2760 if (prog->suboffset && utf8_target) {
2761 /* Convert byte offset to chars.
2762 * XXX ideally should only compute this if @-/@+
2763 * has been seen, a la PL_sawampersand ??? */
2765 /* If there's a direct correspondence between the
2766 * string which we're matching and the original SV,
2767 * then we can use the utf8 len cache associated with
2768 * the SV. In particular, it means that under //g,
2769 * sv_pos_b2u() will use the previously cached
2770 * position to speed up working out the new length of
2771 * subcoffset, rather than counting from the start of
2772 * the string each time. This stops
2773 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2774 * from going quadratic */
2775 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2776 sv_pos_b2u(sv, &(prog->subcoffset));
2778 prog->subcoffset = utf8_length((U8*)strbeg,
2779 (U8*)(strbeg+prog->suboffset));
2783 RX_MATCH_COPY_FREE(rx);
2784 prog->subbeg = strbeg;
2785 prog->suboffset = 0;
2786 prog->subcoffset = 0;
2787 /* use reginfo->strend, as strend may have been modified */
2788 prog->sublen = reginfo->strend - strbeg;
2795 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2796 PL_colors[4], PL_colors[5]));
2798 /* clean up; this will trigger destructors that will free all slabs
2799 * above the current one, and cleanup the regmatch_info_aux
2800 * and regmatch_info_aux_eval sructs */
2802 LEAVE_SCOPE(oldsave);
2805 /* we failed :-( roll it back */
2806 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2807 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2812 Safefree(prog->offs);
2819 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
2820 * Do inc before dec, in case old and new rex are the same */
2821 #define SET_reg_curpm(Re2) \
2822 if (reginfo->info_aux_eval) { \
2823 (void)ReREFCNT_inc(Re2); \
2824 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2825 PM_SETRE((PL_reg_curpm), (Re2)); \
2830 - regtry - try match at specific point
2832 STATIC I32 /* 0 failure, 1 success */
2833 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2837 REGEXP *const rx = reginfo->prog;
2838 regexp *const prog = ReANY(rx);
2840 RXi_GET_DECL(prog,progi);
2841 GET_RE_DEBUG_FLAGS_DECL;
2843 PERL_ARGS_ASSERT_REGTRY;
2845 reginfo->cutpoint=NULL;
2847 prog->offs[0].start = *startposp - reginfo->strbeg;
2848 prog->lastparen = 0;
2849 prog->lastcloseparen = 0;
2851 /* XXXX What this code is doing here?!!! There should be no need
2852 to do this again and again, prog->lastparen should take care of
2855 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2856 * Actually, the code in regcppop() (which Ilya may be meaning by
2857 * prog->lastparen), is not needed at all by the test suite
2858 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2859 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2860 * Meanwhile, this code *is* needed for the
2861 * above-mentioned test suite tests to succeed. The common theme
2862 * on those tests seems to be returning null fields from matches.
2863 * --jhi updated by dapm */
2865 if (prog->nparens) {
2866 regexp_paren_pair *pp = prog->offs;
2868 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2876 result = regmatch(reginfo, *startposp, progi->program + 1);
2878 prog->offs[0].end = result;
2881 if (reginfo->cutpoint)
2882 *startposp= reginfo->cutpoint;
2883 REGCP_UNWIND(lastcp);
2888 #define sayYES goto yes
2889 #define sayNO goto no
2890 #define sayNO_SILENT goto no_silent
2892 /* we dont use STMT_START/END here because it leads to
2893 "unreachable code" warnings, which are bogus, but distracting. */
2894 #define CACHEsayNO \
2895 if (ST.cache_mask) \
2896 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
2899 /* this is used to determine how far from the left messages like
2900 'failed...' are printed. It should be set such that messages
2901 are inline with the regop output that created them.
2903 #define REPORT_CODE_OFF 32
2906 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2907 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2908 #define CHRTEST_NOT_A_CP_1 -999
2909 #define CHRTEST_NOT_A_CP_2 -998
2911 /* grab a new slab and return the first slot in it */
2913 STATIC regmatch_state *
2916 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2919 regmatch_slab *s = PL_regmatch_slab->next;
2921 Newx(s, 1, regmatch_slab);
2922 s->prev = PL_regmatch_slab;
2924 PL_regmatch_slab->next = s;
2926 PL_regmatch_slab = s;
2927 return SLAB_FIRST(s);
2931 /* push a new state then goto it */
2933 #define PUSH_STATE_GOTO(state, node, input) \
2934 pushinput = input; \
2936 st->resume_state = state; \
2939 /* push a new state with success backtracking, then goto it */
2941 #define PUSH_YES_STATE_GOTO(state, node, input) \
2942 pushinput = input; \
2944 st->resume_state = state; \
2945 goto push_yes_state;
2952 regmatch() - main matching routine
2954 This is basically one big switch statement in a loop. We execute an op,
2955 set 'next' to point the next op, and continue. If we come to a point which
2956 we may need to backtrack to on failure such as (A|B|C), we push a
2957 backtrack state onto the backtrack stack. On failure, we pop the top
2958 state, and re-enter the loop at the state indicated. If there are no more
2959 states to pop, we return failure.
2961 Sometimes we also need to backtrack on success; for example /A+/, where
2962 after successfully matching one A, we need to go back and try to
2963 match another one; similarly for lookahead assertions: if the assertion
2964 completes successfully, we backtrack to the state just before the assertion
2965 and then carry on. In these cases, the pushed state is marked as
2966 'backtrack on success too'. This marking is in fact done by a chain of
2967 pointers, each pointing to the previous 'yes' state. On success, we pop to
2968 the nearest yes state, discarding any intermediate failure-only states.
2969 Sometimes a yes state is pushed just to force some cleanup code to be
2970 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2971 it to free the inner regex.
2973 Note that failure backtracking rewinds the cursor position, while
2974 success backtracking leaves it alone.
2976 A pattern is complete when the END op is executed, while a subpattern
2977 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2978 ops trigger the "pop to last yes state if any, otherwise return true"
2981 A common convention in this function is to use A and B to refer to the two
2982 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2983 the subpattern to be matched possibly multiple times, while B is the entire
2984 rest of the pattern. Variable and state names reflect this convention.
2986 The states in the main switch are the union of ops and failure/success of
2987 substates associated with with that op. For example, IFMATCH is the op
2988 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2989 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2990 successfully matched A and IFMATCH_A_fail is a state saying that we have
2991 just failed to match A. Resume states always come in pairs. The backtrack
2992 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2993 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2994 on success or failure.
2996 The struct that holds a backtracking state is actually a big union, with
2997 one variant for each major type of op. The variable st points to the
2998 top-most backtrack struct. To make the code clearer, within each
2999 block of code we #define ST to alias the relevant union.
3001 Here's a concrete example of a (vastly oversimplified) IFMATCH
3007 #define ST st->u.ifmatch
3009 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3010 ST.foo = ...; // some state we wish to save
3012 // push a yes backtrack state with a resume value of
3013 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3015 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3018 case IFMATCH_A: // we have successfully executed A; now continue with B
3020 bar = ST.foo; // do something with the preserved value
3023 case IFMATCH_A_fail: // A failed, so the assertion failed
3024 ...; // do some housekeeping, then ...
3025 sayNO; // propagate the failure
3032 For any old-timers reading this who are familiar with the old recursive
3033 approach, the code above is equivalent to:
3035 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3044 ...; // do some housekeeping, then ...
3045 sayNO; // propagate the failure
3048 The topmost backtrack state, pointed to by st, is usually free. If you
3049 want to claim it, populate any ST.foo fields in it with values you wish to
3050 save, then do one of
3052 PUSH_STATE_GOTO(resume_state, node, newinput);
3053 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3055 which sets that backtrack state's resume value to 'resume_state', pushes a
3056 new free entry to the top of the backtrack stack, then goes to 'node'.
3057 On backtracking, the free slot is popped, and the saved state becomes the
3058 new free state. An ST.foo field in this new top state can be temporarily
3059 accessed to retrieve values, but once the main loop is re-entered, it
3060 becomes available for reuse.
3062 Note that the depth of the backtrack stack constantly increases during the
3063 left-to-right execution of the pattern, rather than going up and down with
3064 the pattern nesting. For example the stack is at its maximum at Z at the
3065 end of the pattern, rather than at X in the following:
3067 /(((X)+)+)+....(Y)+....Z/
3069 The only exceptions to this are lookahead/behind assertions and the cut,
3070 (?>A), which pop all the backtrack states associated with A before
3073 Backtrack state structs are allocated in slabs of about 4K in size.
3074 PL_regmatch_state and st always point to the currently active state,
3075 and PL_regmatch_slab points to the slab currently containing
3076 PL_regmatch_state. The first time regmatch() is called, the first slab is
3077 allocated, and is never freed until interpreter destruction. When the slab
3078 is full, a new one is allocated and chained to the end. At exit from
3079 regmatch(), slabs allocated since entry are freed.
3084 #define DEBUG_STATE_pp(pp) \
3086 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3087 PerlIO_printf(Perl_debug_log, \
3088 " %*s"pp" %s%s%s%s%s\n", \
3090 PL_reg_name[st->resume_state], \
3091 ((st==yes_state||st==mark_state) ? "[" : ""), \
3092 ((st==yes_state) ? "Y" : ""), \
3093 ((st==mark_state) ? "M" : ""), \
3094 ((st==yes_state||st==mark_state) ? "]" : "") \
3099 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3104 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3105 const char *start, const char *end, const char *blurb)
3107 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3109 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3114 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3115 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3117 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3118 start, end - start, 60);
3120 PerlIO_printf(Perl_debug_log,
3121 "%s%s REx%s %s against %s\n",
3122 PL_colors[4], blurb, PL_colors[5], s0, s1);
3124 if (utf8_target||utf8_pat)
3125 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3126 utf8_pat ? "pattern" : "",
3127 utf8_pat && utf8_target ? " and " : "",
3128 utf8_target ? "string" : ""
3134 S_dump_exec_pos(pTHX_ const char *locinput,
3135 const regnode *scan,
3136 const char *loc_regeol,
3137 const char *loc_bostr,
3138 const char *loc_reg_starttry,
3139 const bool utf8_target)
3141 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3142 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3143 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3144 /* The part of the string before starttry has one color
3145 (pref0_len chars), between starttry and current
3146 position another one (pref_len - pref0_len chars),
3147 after the current position the third one.
3148 We assume that pref0_len <= pref_len, otherwise we
3149 decrease pref0_len. */
3150 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3151 ? (5 + taill) - l : locinput - loc_bostr;
3154 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3156 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3158 pref0_len = pref_len - (locinput - loc_reg_starttry);
3159 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3160 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3161 ? (5 + taill) - pref_len : loc_regeol - locinput);
3162 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3166 if (pref0_len > pref_len)
3167 pref0_len = pref_len;
3169 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3171 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3172 (locinput - pref_len),pref0_len, 60, 4, 5);
3174 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3175 (locinput - pref_len + pref0_len),
3176 pref_len - pref0_len, 60, 2, 3);
3178 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3179 locinput, loc_regeol - locinput, 10, 0, 1);
3181 const STRLEN tlen=len0+len1+len2;
3182 PerlIO_printf(Perl_debug_log,
3183 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3184 (IV)(locinput - loc_bostr),
3187 (docolor ? "" : "> <"),
3189 (int)(tlen > 19 ? 0 : 19 - tlen),
3196 /* reg_check_named_buff_matched()
3197 * Checks to see if a named buffer has matched. The data array of
3198 * buffer numbers corresponding to the buffer is expected to reside
3199 * in the regexp->data->data array in the slot stored in the ARG() of
3200 * node involved. Note that this routine doesn't actually care about the
3201 * name, that information is not preserved from compilation to execution.
3202 * Returns the index of the leftmost defined buffer with the given name
3203 * or 0 if non of the buffers matched.
3206 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3209 RXi_GET_DECL(rex,rexi);
3210 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3211 I32 *nums=(I32*)SvPVX(sv_dat);
3213 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3215 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3216 if ((I32)rex->lastparen >= nums[n] &&
3217 rex->offs[nums[n]].end != -1)
3227 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3228 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3230 /* This function determines if there are one or two characters that match
3231 * the first character of the passed-in EXACTish node <text_node>, and if
3232 * so, returns them in the passed-in pointers.
3234 * If it determines that no possible character in the target string can
3235 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3236 * the first character in <text_node> requires UTF-8 to represent, and the
3237 * target string isn't in UTF-8.)
3239 * If there are more than two characters that could match the beginning of
3240 * <text_node>, or if more context is required to determine a match or not,
3241 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3243 * The motiviation behind this function is to allow the caller to set up
3244 * tight loops for matching. If <text_node> is of type EXACT, there is
3245 * only one possible character that can match its first character, and so
3246 * the situation is quite simple. But things get much more complicated if
3247 * folding is involved. It may be that the first character of an EXACTFish
3248 * node doesn't participate in any possible fold, e.g., punctuation, so it
3249 * can be matched only by itself. The vast majority of characters that are
3250 * in folds match just two things, their lower and upper-case equivalents.
3251 * But not all are like that; some have multiple possible matches, or match
3252 * sequences of more than one character. This function sorts all that out.
3254 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3255 * loop of trying to match A*, we know we can't exit where the thing
3256 * following it isn't a B. And something can't be a B unless it is the
3257 * beginning of B. By putting a quick test for that beginning in a tight
3258 * loop, we can rule out things that can't possibly be B without having to
3259 * break out of the loop, thus avoiding work. Similarly, if A is a single
3260 * character, we can make a tight loop matching A*, using the outputs of
3263 * If the target string to match isn't in UTF-8, and there aren't
3264 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3265 * the one or two possible octets (which are characters in this situation)
3266 * that can match. In all cases, if there is only one character that can
3267 * match, *<c1p> and *<c2p> will be identical.
3269 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3270 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3271 * can match the beginning of <text_node>. They should be declared with at
3272 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3273 * undefined what these contain.) If one or both of the buffers are
3274 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3275 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3276 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3277 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3278 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3280 const bool utf8_target = reginfo->is_utf8_target;
3282 UV c1 = CHRTEST_NOT_A_CP_1;
3283 UV c2 = CHRTEST_NOT_A_CP_2;
3284 bool use_chrtest_void = FALSE;
3285 const bool is_utf8_pat = reginfo->is_utf8_pat;
3287 /* Used when we have both utf8 input and utf8 output, to avoid converting
3288 * to/from code points */
3289 bool utf8_has_been_setup = FALSE;
3293 U8 *pat = (U8*)STRING(text_node);
3295 if (OP(text_node) == EXACT) {
3297 /* In an exact node, only one thing can be matched, that first
3298 * character. If both the pat and the target are UTF-8, we can just
3299 * copy the input to the output, avoiding finding the code point of
3304 else if (utf8_target) {
3305 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3306 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3307 utf8_has_been_setup = TRUE;
3310 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3313 else /* an EXACTFish node */
3315 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3316 pat + STR_LEN(text_node)))
3318 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3319 pat + STR_LEN(text_node))))
3321 /* Multi-character folds require more context to sort out. Also
3322 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3323 * handled outside this routine */
3324 use_chrtest_void = TRUE;
3326 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3327 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3329 /* Load the folds hash, if not already done */
3331 if (! PL_utf8_foldclosures) {
3332 if (! PL_utf8_tofold) {
3333 U8 dummy[UTF8_MAXBYTES+1];
3335 /* Force loading this by folding an above-Latin1 char */
3336 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3337 assert(PL_utf8_tofold); /* Verify that worked */
3339 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3342 /* The fold closures data structure is a hash with the keys being
3343 * the UTF-8 of every character that is folded to, like 'k', and
3344 * the values each an array of all code points that fold to its
3345 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3347 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3352 /* Not found in the hash, therefore there are no folds
3353 * containing it, so there is only a single character that
3357 else { /* Does participate in folds */
3358 AV* list = (AV*) *listp;
3359 if (av_len(list) != 1) {
3361 /* If there aren't exactly two folds to this, it is outside
3362 * the scope of this function */
3363 use_chrtest_void = TRUE;
3365 else { /* There are two. Get them */
3366 SV** c_p = av_fetch(list, 0, FALSE);
3368 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3372 c_p = av_fetch(list, 1, FALSE);
3374 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3378 /* Folds that cross the 255/256 boundary are forbidden if
3379 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3380 * pattern character is above 256, and its only other match
3381 * is below 256, the only legal match will be to itself.
3382 * We have thrown away the original, so have to compute
3383 * which is the one above 255 */
3384 if ((c1 < 256) != (c2 < 256)) {
3385 if (OP(text_node) == EXACTFL
3386 || (OP(text_node) == EXACTFA
3387 && (isASCII(c1) || isASCII(c2))))
3400 else /* Here, c1 is < 255 */
3402 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3403 && OP(text_node) != EXACTFL
3404 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3406 /* Here, there could be something above Latin1 in the target which
3407 * folds to this character in the pattern. All such cases except
3408 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3409 * involved in their folds, so are outside the scope of this
3411 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3412 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3415 use_chrtest_void = TRUE;
3418 else { /* Here nothing above Latin1 can fold to the pattern character */
3419 switch (OP(text_node)) {
3421 case EXACTFL: /* /l rules */
3422 c2 = PL_fold_locale[c1];
3426 if (! utf8_target) { /* /d rules */
3431 /* /u rules for all these. This happens to work for
3432 * EXACTFA as nothing in Latin1 folds to ASCII */
3434 case EXACTFU_TRICKYFOLD:
3437 c2 = PL_fold_latin1[c1];
3441 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3442 assert(0); /* NOTREACHED */
3447 /* Here have figured things out. Set up the returns */
3448 if (use_chrtest_void) {
3449 *c2p = *c1p = CHRTEST_VOID;
3451 else if (utf8_target) {
3452 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3453 uvchr_to_utf8(c1_utf8, c1);
3454 uvchr_to_utf8(c2_utf8, c2);
3457 /* Invariants are stored in both the utf8 and byte outputs; Use
3458 * negative numbers otherwise for the byte ones. Make sure that the
3459 * byte ones are the same iff the utf8 ones are the same */
3460 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3461 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3464 ? CHRTEST_NOT_A_CP_1
3465 : CHRTEST_NOT_A_CP_2;
3467 else if (c1 > 255) {
3468 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3473 *c1p = *c2p = c2; /* c2 is the only representable value */
3475 else { /* c1 is representable; see about c2 */
3477 *c2p = (c2 < 256) ? c2 : c1;
3483 /* returns -1 on failure, $+[0] on success */
3485 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3487 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3491 const bool utf8_target = reginfo->is_utf8_target;
3492 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3493 REGEXP *rex_sv = reginfo->prog;
3494 regexp *rex = ReANY(rex_sv);
3495 RXi_GET_DECL(rex,rexi);
3496 /* the current state. This is a cached copy of PL_regmatch_state */
3498 /* cache heavy used fields of st in registers */
3501 U32 n = 0; /* general value; init to avoid compiler warning */
3502 I32 ln = 0; /* len or last; init to avoid compiler warning */
3503 char *locinput = startpos;
3504 char *pushinput; /* where to continue after a PUSH */
3505 I32 nextchr; /* is always set to UCHARAT(locinput) */
3507 bool result = 0; /* return value of S_regmatch */
3508 int depth = 0; /* depth of backtrack stack */
3509 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3510 const U32 max_nochange_depth =
3511 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3512 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3513 regmatch_state *yes_state = NULL; /* state to pop to on success of
3515 /* mark_state piggy backs on the yes_state logic so that when we unwind
3516 the stack on success we can update the mark_state as we go */
3517 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3518 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3519 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3521 bool no_final = 0; /* prevent failure from backtracking? */
3522 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3523 char *startpoint = locinput;
3524 SV *popmark = NULL; /* are we looking for a mark? */
3525 SV *sv_commit = NULL; /* last mark name seen in failure */
3526 SV *sv_yes_mark = NULL; /* last mark name we have seen
3527 during a successful match */
3528 U32 lastopen = 0; /* last open we saw */
3529 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3530 SV* const oreplsv = GvSV(PL_replgv);
3531 /* these three flags are set by various ops to signal information to
3532 * the very next op. They have a useful lifetime of exactly one loop
3533 * iteration, and are not preserved or restored by state pushes/pops
3535 bool sw = 0; /* the condition value in (?(cond)a|b) */
3536 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3537 int logical = 0; /* the following EVAL is:
3541 or the following IFMATCH/UNLESSM is:
3542 false: plain (?=foo)
3543 true: used as a condition: (?(?=foo))
3545 PAD* last_pad = NULL;
3547 I32 gimme = G_SCALAR;
3548 CV *caller_cv = NULL; /* who called us */
3549 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3550 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3551 U32 maxopenparen = 0; /* max '(' index seen so far */
3552 int to_complement; /* Invert the result? */
3553 _char_class_number classnum;
3554 bool is_utf8_pat = reginfo->is_utf8_pat;
3557 GET_RE_DEBUG_FLAGS_DECL;
3560 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3561 multicall_oldcatch = 0;
3562 multicall_cv = NULL;
3564 PERL_UNUSED_VAR(multicall_cop);
3565 PERL_UNUSED_VAR(newsp);
3568 PERL_ARGS_ASSERT_REGMATCH;
3570 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3571 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3574 st = PL_regmatch_state;
3576 /* Note that nextchr is a byte even in UTF */
3579 while (scan != NULL) {
3582 SV * const prop = sv_newmortal();
3583 regnode *rnext=regnext(scan);
3584 DUMP_EXEC_POS( locinput, scan, utf8_target );
3585 regprop(rex, prop, scan);
3587 PerlIO_printf(Perl_debug_log,
3588 "%3"IVdf":%*s%s(%"IVdf")\n",
3589 (IV)(scan - rexi->program), depth*2, "",
3591 (PL_regkind[OP(scan)] == END || !rnext) ?
3592 0 : (IV)(rnext - rexi->program));
3595 next = scan + NEXT_OFF(scan);
3598 state_num = OP(scan);
3600 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3605 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3607 switch (state_num) {
3608 case BOL: /* /^../ */
3609 if (locinput == reginfo->strbeg)
3613 case MBOL: /* /^../m */
3614 if (locinput == reginfo->strbeg ||
3615 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3621 case SBOL: /* /^../s */
3622 if (locinput == reginfo->strbeg)
3627 if (locinput == reginfo->ganch)
3631 case KEEPS: /* \K */
3632 /* update the startpoint */
3633 st->u.keeper.val = rex->offs[0].start;
3634 rex->offs[0].start = locinput - reginfo->strbeg;
3635 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3636 assert(0); /*NOTREACHED*/
3637 case KEEPS_next_fail:
3638 /* rollback the start point change */
3639 rex->offs[0].start = st->u.keeper.val;
3641 assert(0); /*NOTREACHED*/
3643 case EOL: /* /..$/ */
3646 case MEOL: /* /..$/m */
3647 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3651 case SEOL: /* /..$/s */
3653 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3655 if (reginfo->strend - locinput > 1)
3660 if (!NEXTCHR_IS_EOS)
3664 case SANY: /* /./s */
3667 goto increment_locinput;
3675 case REG_ANY: /* /./ */
3676 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3678 goto increment_locinput;
3682 #define ST st->u.trie
3683 case TRIEC: /* (ab|cd) with known charclass */
3684 /* In this case the charclass data is available inline so
3685 we can fail fast without a lot of extra overhead.
3687 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3689 PerlIO_printf(Perl_debug_log,
3690 "%*s %sfailed to match trie start class...%s\n",
3691 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3694 assert(0); /* NOTREACHED */
3697 case TRIE: /* (ab|cd) */
3698 /* the basic plan of execution of the trie is:
3699 * At the beginning, run though all the states, and
3700 * find the longest-matching word. Also remember the position
3701 * of the shortest matching word. For example, this pattern:
3704 * when matched against the string "abcde", will generate
3705 * accept states for all words except 3, with the longest
3706 * matching word being 4, and the shortest being 2 (with
3707 * the position being after char 1 of the string).
3709 * Then for each matching word, in word order (i.e. 1,2,4,5),
3710 * we run the remainder of the pattern; on each try setting
3711 * the current position to the character following the word,
3712 * returning to try the next word on failure.
3714 * We avoid having to build a list of words at runtime by
3715 * using a compile-time structure, wordinfo[].prev, which
3716 * gives, for each word, the previous accepting word (if any).
3717 * In the case above it would contain the mappings 1->2, 2->0,
3718 * 3->0, 4->5, 5->1. We can use this table to generate, from
3719 * the longest word (4 above), a list of all words, by
3720 * following the list of prev pointers; this gives us the
3721 * unordered list 4,5,1,2. Then given the current word we have
3722 * just tried, we can go through the list and find the
3723 * next-biggest word to try (so if we just failed on word 2,
3724 * the next in the list is 4).
3726 * Since at runtime we don't record the matching position in
3727 * the string for each word, we have to work that out for
3728 * each word we're about to process. The wordinfo table holds
3729 * the character length of each word; given that we recorded
3730 * at the start: the position of the shortest word and its
3731 * length in chars, we just need to move the pointer the
3732 * difference between the two char lengths. Depending on
3733 * Unicode status and folding, that's cheap or expensive.
3735 * This algorithm is optimised for the case where are only a
3736 * small number of accept states, i.e. 0,1, or maybe 2.
3737 * With lots of accepts states, and having to try all of them,
3738 * it becomes quadratic on number of accept states to find all
3743 /* what type of TRIE am I? (utf8 makes this contextual) */
3744 DECL_TRIE_TYPE(scan);
3746 /* what trie are we using right now */
3747 reg_trie_data * const trie
3748 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3749 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3750 U32 state = trie->startstate;
3753 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3755 if (trie->states[ state ].wordnum) {
3757 PerlIO_printf(Perl_debug_log,
3758 "%*s %smatched empty string...%s\n",
3759 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3765 PerlIO_printf(Perl_debug_log,
3766 "%*s %sfailed to match trie start class...%s\n",
3767 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3774 U8 *uc = ( U8* )locinput;
3778 U8 *uscan = (U8*)NULL;
3779 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3780 U32 charcount = 0; /* how many input chars we have matched */
3781 U32 accepted = 0; /* have we seen any accepting states? */
3783 ST.jump = trie->jump;
3786 ST.longfold = FALSE; /* char longer if folded => it's harder */
3789 /* fully traverse the TRIE; note the position of the
3790 shortest accept state and the wordnum of the longest
3793 while ( state && uc <= (U8*)(reginfo->strend) ) {
3794 U32 base = trie->states[ state ].trans.base;
3798 wordnum = trie->states[ state ].wordnum;
3800 if (wordnum) { /* it's an accept state */
3803 /* record first match position */
3805 ST.firstpos = (U8*)locinput;
3810 ST.firstchars = charcount;
3813 if (!ST.nextword || wordnum < ST.nextword)
3814 ST.nextword = wordnum;
3815 ST.topword = wordnum;
3818 DEBUG_TRIE_EXECUTE_r({
3819 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3820 PerlIO_printf( Perl_debug_log,
3821 "%*s %sState: %4"UVxf" Accepted: %c ",
3822 2+depth * 2, "", PL_colors[4],
3823 (UV)state, (accepted ? 'Y' : 'N'));
3826 /* read a char and goto next state */
3827 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
3829 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3830 uscan, len, uvc, charid, foldlen,
3837 base + charid - 1 - trie->uniquecharcount)) >= 0)
3839 && ((U32)offset < trie->lasttrans)
3840 && trie->trans[offset].check == state)
3842 state = trie->trans[offset].next;
3853 DEBUG_TRIE_EXECUTE_r(
3854 PerlIO_printf( Perl_debug_log,
3855 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3856 charid, uvc, (UV)state, PL_colors[5] );
3862 /* calculate total number of accept states */
3867 w = trie->wordinfo[w].prev;
3870 ST.accepted = accepted;
3874 PerlIO_printf( Perl_debug_log,
3875 "%*s %sgot %"IVdf" possible matches%s\n",
3876 REPORT_CODE_OFF + depth * 2, "",
3877 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3879 goto trie_first_try; /* jump into the fail handler */
3881 assert(0); /* NOTREACHED */
3883 case TRIE_next_fail: /* we failed - try next alternative */
3887 REGCP_UNWIND(ST.cp);
3888 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3890 if (!--ST.accepted) {
3892 PerlIO_printf( Perl_debug_log,
3893 "%*s %sTRIE failed...%s\n",
3894 REPORT_CODE_OFF+depth*2, "",
3901 /* Find next-highest word to process. Note that this code
3902 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3905 U16 const nextword = ST.nextword;
3906 reg_trie_wordinfo * const wordinfo
3907 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3908 for (word=ST.topword; word; word=wordinfo[word].prev) {
3909 if (word > nextword && (!min || word < min))
3922 ST.lastparen = rex->lastparen;
3923 ST.lastcloseparen = rex->lastcloseparen;
3927 /* find start char of end of current word */
3929 U32 chars; /* how many chars to skip */
3930 reg_trie_data * const trie
3931 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3933 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3935 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3940 /* the hard option - fold each char in turn and find
3941 * its folded length (which may be different */
3942 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3950 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3958 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3963 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3979 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
3980 ? ST.jump[ST.nextword]
3984 PerlIO_printf( Perl_debug_log,
3985 "%*s %sTRIE matched word #%d, continuing%s\n",
3986 REPORT_CODE_OFF+depth*2, "",
3993 if (ST.accepted > 1 || has_cutgroup) {
3994 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
3995 assert(0); /* NOTREACHED */
3997 /* only one choice left - just continue */
3999 AV *const trie_words
4000 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4001 SV ** const tmp = av_fetch( trie_words,
4003 SV *sv= tmp ? sv_newmortal() : NULL;
4005 PerlIO_printf( Perl_debug_log,
4006 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4007 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4009 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4010 PL_colors[0], PL_colors[1],
4011 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4013 : "not compiled under -Dr",
4017 locinput = (char*)uc;
4018 continue; /* execute rest of RE */
4019 assert(0); /* NOTREACHED */
4023 case EXACT: { /* /abc/ */
4024 char *s = STRING(scan);
4026 if (utf8_target != is_utf8_pat) {
4027 /* The target and the pattern have differing utf8ness. */
4029 const char * const e = s + ln;
4032 /* The target is utf8, the pattern is not utf8.
4033 * Above-Latin1 code points can't match the pattern;
4034 * invariants match exactly, and the other Latin1 ones need
4035 * to be downgraded to a single byte in order to do the
4036 * comparison. (If we could be confident that the target
4037 * is not malformed, this could be refactored to have fewer
4038 * tests by just assuming that if the first bytes match, it
4039 * is an invariant, but there are tests in the test suite
4040 * dealing with (??{...}) which violate this) */
4042 if (l >= reginfo->strend
4043 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4047 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4054 if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
4063 /* The target is not utf8, the pattern is utf8. */
4065 if (l >= reginfo->strend
4066 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4070 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4077 if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
4088 /* The target and the pattern have the same utf8ness. */
4089 /* Inline the first character, for speed. */
4090 if (reginfo->strend - locinput < ln
4091 || UCHARAT(s) != nextchr
4092 || (ln > 1 && memNE(s, locinput, ln)))
4101 case EXACTFL: { /* /abc/il */
4103 const U8 * fold_array;
4105 U32 fold_utf8_flags;
4107 RX_MATCH_TAINTED_on(reginfo->prog);
4108 folder = foldEQ_locale;
4109 fold_array = PL_fold_locale;
4110 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
4113 case EXACTFU_SS: /* /\x{df}/iu */
4114 case EXACTFU_TRICKYFOLD: /* /\x{390}/iu */
4115 case EXACTFU: /* /abc/iu */
4116 folder = foldEQ_latin1;
4117 fold_array = PL_fold_latin1;
4118 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4121 case EXACTFA: /* /abc/iaa */
4122 folder = foldEQ_latin1;
4123 fold_array = PL_fold_latin1;
4124 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4127 case EXACTF: /* /abc/i */
4129 fold_array = PL_fold;
4130 fold_utf8_flags = 0;
4136 if (utf8_target || is_utf8_pat || state_num == EXACTFU_SS) {
4137 /* Either target or the pattern are utf8, or has the issue where
4138 * the fold lengths may differ. */
4139 const char * const l = locinput;
4140 char *e = reginfo->strend;
4142 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4143 l, &e, 0, utf8_target, fold_utf8_flags))
4151 /* Neither the target nor the pattern are utf8 */
4152 if (UCHARAT(s) != nextchr
4154 && UCHARAT(s) != fold_array[nextchr])
4158 if (reginfo->strend - locinput < ln)
4160 if (ln > 1 && ! folder(s, locinput, ln))
4166 /* XXX Could improve efficiency by separating these all out using a
4167 * macro or in-line function. At that point regcomp.c would no longer
4168 * have to set the FLAGS fields of these */
4169 case BOUNDL: /* /\b/l */
4170 case NBOUNDL: /* /\B/l */
4171 RX_MATCH_TAINTED_on(reginfo->prog);
4173 case BOUND: /* /\b/ */
4174 case BOUNDU: /* /\b/u */
4175 case BOUNDA: /* /\b/a */
4176 case NBOUND: /* /\B/ */
4177 case NBOUNDU: /* /\B/u */
4178 case NBOUNDA: /* /\B/a */
4179 /* was last char in word? */
4181 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4182 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4184 if (locinput == reginfo->strbeg)
4187 const U8 * const r =
4188 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4190 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
4192 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4193 ln = isWORDCHAR_uni(ln);
4197 LOAD_UTF8_CHARCLASS_ALNUM();
4198 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4203 ln = isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(ln));
4204 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4209 /* Here the string isn't utf8, or is utf8 and only ascii
4210 * characters are to match \w. In the latter case looking at
4211 * the byte just prior to the current one may be just the final
4212 * byte of a multi-byte character. This is ok. There are two
4214 * 1) it is a single byte character, and then the test is doing
4215 * just what it's supposed to.
4216 * 2) it is a multi-byte character, in which case the final
4217 * byte is never mistakable for ASCII, and so the test
4218 * will say it is not a word character, which is the
4219 * correct answer. */
4220 ln = (locinput != reginfo->strbeg) ?
4221 UCHARAT(locinput - 1) : '\n';
4222 switch (FLAGS(scan)) {
4223 case REGEX_UNICODE_CHARSET:
4224 ln = isWORDCHAR_L1(ln);
4225 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4227 case REGEX_LOCALE_CHARSET:
4228 ln = isWORDCHAR_LC(ln);
4229 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4231 case REGEX_DEPENDS_CHARSET:
4232 ln = isWORDCHAR(ln);
4233 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4235 case REGEX_ASCII_RESTRICTED_CHARSET:
4236 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4237 ln = isWORDCHAR_A(ln);
4238 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4241 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4245 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4247 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4251 case ANYOF: /* /[abc]/ */
4252 case ANYOF_WARN_SUPER:
4256 if (!reginclass(rex, scan, (U8*)locinput, utf8_target))
4258 locinput += UTF8SKIP(locinput);
4261 if (!REGINCLASS(rex, scan, (U8*)locinput))
4267 /* The argument (FLAGS) to all the POSIX node types is the class number
4270 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4274 case POSIXL: /* \w or [:punct:] etc. under /l */
4278 /* The locale hasn't influenced the outcome before this, so defer
4279 * tainting until now */
4280 RX_MATCH_TAINTED_on(reginfo->prog);
4282 /* Use isFOO_lc() for characters within Latin1. (Note that
4283 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4284 * wouldn't be invariant) */
4285 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4286 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4290 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4291 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4292 (U8) TWO_BYTE_UTF8_TO_UNI(nextchr,
4293 *(locinput + 1))))))
4298 else { /* Here, must be an above Latin-1 code point */
4299 goto utf8_posix_not_eos;
4302 /* Here, must be utf8 */
4303 locinput += UTF8SKIP(locinput);
4306 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4310 case POSIXD: /* \w or [:punct:] etc. under /d */
4316 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4318 if (NEXTCHR_IS_EOS) {
4322 /* All UTF-8 variants match */
4323 if (! UTF8_IS_INVARIANT(nextchr)) {
4324 goto increment_locinput;
4330 case POSIXA: /* \w or [:punct:] etc. under /a */
4333 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4334 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4335 * character is a single byte */
4338 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4344 /* Here we are either not in utf8, or we matched a utf8-invariant,
4345 * so the next char is the next byte */
4349 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4353 case POSIXU: /* \w or [:punct:] etc. under /u */
4355 if (NEXTCHR_IS_EOS) {
4360 /* Use _generic_isCC() for characters within Latin1. (Note that
4361 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4362 * wouldn't be invariant) */
4363 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4364 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4371 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4372 if (! (to_complement
4373 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(nextchr,
4381 else { /* Handle above Latin-1 code points */
4382 classnum = (_char_class_number) FLAGS(scan);
4383 if (classnum < _FIRST_NON_SWASH_CC) {
4385 /* Here, uses a swash to find such code points. Load if if
4386 * not done already */
4387 if (! PL_utf8_swash_ptrs[classnum]) {
4388 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4389 PL_utf8_swash_ptrs[classnum]
4390 = _core_swash_init("utf8",
4391 swash_property_names[classnum],
4392 &PL_sv_undef, 1, 0, NULL, &flags);
4394 if (! (to_complement
4395 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4396 (U8 *) locinput, TRUE))))
4401 else { /* Here, uses macros to find above Latin-1 code points */
4403 case _CC_ENUM_SPACE: /* XXX would require separate
4404 code if we revert the change
4405 of \v matching this */
4406 case _CC_ENUM_PSXSPC:
4407 if (! (to_complement
4408 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4413 case _CC_ENUM_BLANK:
4414 if (! (to_complement
4415 ^ cBOOL(is_HORIZWS_high(locinput))))
4420 case _CC_ENUM_XDIGIT:
4421 if (! (to_complement
4422 ^ cBOOL(is_XDIGIT_high(locinput))))
4427 case _CC_ENUM_VERTSPACE:
4428 if (! (to_complement
4429 ^ cBOOL(is_VERTWS_high(locinput))))
4434 default: /* The rest, e.g. [:cntrl:], can't match
4436 if (! to_complement) {
4442 locinput += UTF8SKIP(locinput);
4446 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4447 a Unicode extended Grapheme Cluster */
4448 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4449 extended Grapheme Cluster is:
4452 | Prepend* Begin Extend*
4455 Begin is: ( Special_Begin | ! Control )
4456 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4457 Extend is: ( Grapheme_Extend | Spacing_Mark )
4458 Control is: [ GCB_Control | CR | LF ]
4459 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4461 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4464 Begin is ( Regular_Begin + Special Begin )
4466 It turns out that 98.4% of all Unicode code points match
4467 Regular_Begin. Doing it this way eliminates a table match in
4468 the previous implementation for almost all Unicode code points.
4470 There is a subtlety with Prepend* which showed up in testing.
4471 Note that the Begin, and only the Begin is required in:
4472 | Prepend* Begin Extend*
4473 Also, Begin contains '! Control'. A Prepend must be a
4474 '! Control', which means it must also be a Begin. What it
4475 comes down to is that if we match Prepend* and then find no
4476 suitable Begin afterwards, that if we backtrack the last
4477 Prepend, that one will be a suitable Begin.
4482 if (! utf8_target) {
4484 /* Match either CR LF or '.', as all the other possibilities
4486 locinput++; /* Match the . or CR */
4487 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4489 && locinput < reginfo->strend
4490 && UCHARAT(locinput) == '\n')
4497 /* Utf8: See if is ( CR LF ); already know that locinput <
4498 * reginfo->strend, so locinput+1 is in bounds */
4499 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4500 && UCHARAT(locinput + 1) == '\n')
4507 /* In case have to backtrack to beginning, then match '.' */
4508 char *starting = locinput;
4510 /* In case have to backtrack the last prepend */
4511 char *previous_prepend = NULL;
4513 LOAD_UTF8_CHARCLASS_GCB();
4515 /* Match (prepend)* */
4516 while (locinput < reginfo->strend
4517 && (len = is_GCB_Prepend_utf8(locinput)))
4519 previous_prepend = locinput;
4523 /* As noted above, if we matched a prepend character, but
4524 * the next thing won't match, back off the last prepend we
4525 * matched, as it is guaranteed to match the begin */
4526 if (previous_prepend
4527 && (locinput >= reginfo->strend
4528 || (! swash_fetch(PL_utf8_X_regular_begin,
4529 (U8*)locinput, utf8_target)
4530 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4533 locinput = previous_prepend;
4536 /* Note that here we know reginfo->strend > locinput, as we
4537 * tested that upon input to this switch case, and if we
4538 * moved locinput forward, we tested the result just above
4539 * and it either passed, or we backed off so that it will
4541 if (swash_fetch(PL_utf8_X_regular_begin,
4542 (U8*)locinput, utf8_target)) {
4543 locinput += UTF8SKIP(locinput);
4545 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4547 /* Here did not match the required 'Begin' in the
4548 * second term. So just match the very first
4549 * character, the '.' of the final term of the regex */
4550 locinput = starting + UTF8SKIP(starting);
4554 /* Here is a special begin. It can be composed of
4555 * several individual characters. One possibility is
4557 if ((len = is_GCB_RI_utf8(locinput))) {
4559 while (locinput < reginfo->strend
4560 && (len = is_GCB_RI_utf8(locinput)))
4564 } else if ((len = is_GCB_T_utf8(locinput))) {
4565 /* Another possibility is T+ */
4567 while (locinput < reginfo->strend
4568 && (len = is_GCB_T_utf8(locinput)))
4574 /* Here, neither RI+ nor T+; must be some other
4575 * Hangul. That means it is one of the others: L,
4576 * LV, LVT or V, and matches:
4577 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4580 while (locinput < reginfo->strend
4581 && (len = is_GCB_L_utf8(locinput)))
4586 /* Here, have exhausted L*. If the next character
4587 * is not an LV, LVT nor V, it means we had to have
4588 * at least one L, so matches L+ in the original
4589 * equation, we have a complete hangul syllable.
4592 if (locinput < reginfo->strend
4593 && is_GCB_LV_LVT_V_utf8(locinput))
4595 /* Otherwise keep going. Must be LV, LVT or V.
4596 * See if LVT, by first ruling out V, then LV */
4597 if (! is_GCB_V_utf8(locinput)
4598 /* All but every TCount one is LV */
4599 && (valid_utf8_to_uvchr((U8 *) locinput,
4604 locinput += UTF8SKIP(locinput);
4607 /* Must be V or LV. Take it, then match
4609 locinput += UTF8SKIP(locinput);
4610 while (locinput < reginfo->strend
4611 && (len = is_GCB_V_utf8(locinput)))
4617 /* And any of LV, LVT, or V can be followed
4619 while (locinput < reginfo->strend
4620 && (len = is_GCB_T_utf8(locinput)))
4628 /* Match any extender */
4629 while (locinput < reginfo->strend
4630 && swash_fetch(PL_utf8_X_extend,
4631 (U8*)locinput, utf8_target))
4633 locinput += UTF8SKIP(locinput);
4637 if (locinput > reginfo->strend) sayNO;
4641 case NREFFL: /* /\g{name}/il */
4642 { /* The capture buffer cases. The ones beginning with N for the
4643 named buffers just convert to the equivalent numbered and
4644 pretend they were called as the corresponding numbered buffer
4646 /* don't initialize these in the declaration, it makes C++
4651 const U8 *fold_array;
4654 RX_MATCH_TAINTED_on(reginfo->prog);
4655 folder = foldEQ_locale;
4656 fold_array = PL_fold_locale;
4658 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4661 case NREFFA: /* /\g{name}/iaa */
4662 folder = foldEQ_latin1;
4663 fold_array = PL_fold_latin1;
4665 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4668 case NREFFU: /* /\g{name}/iu */
4669 folder = foldEQ_latin1;
4670 fold_array = PL_fold_latin1;
4672 utf8_fold_flags = 0;
4675 case NREFF: /* /\g{name}/i */
4677 fold_array = PL_fold;
4679 utf8_fold_flags = 0;
4682 case NREF: /* /\g{name}/ */
4686 utf8_fold_flags = 0;
4689 /* For the named back references, find the corresponding buffer
4691 n = reg_check_named_buff_matched(rex,scan);
4696 goto do_nref_ref_common;
4698 case REFFL: /* /\1/il */
4699 RX_MATCH_TAINTED_on(reginfo->prog);
4700 folder = foldEQ_locale;
4701 fold_array = PL_fold_locale;
4702 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4705 case REFFA: /* /\1/iaa */
4706 folder = foldEQ_latin1;
4707 fold_array = PL_fold_latin1;
4708 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4711 case REFFU: /* /\1/iu */
4712 folder = foldEQ_latin1;
4713 fold_array = PL_fold_latin1;
4714 utf8_fold_flags = 0;
4717 case REFF: /* /\1/i */
4719 fold_array = PL_fold;
4720 utf8_fold_flags = 0;
4723 case REF: /* /\1/ */
4726 utf8_fold_flags = 0;
4730 n = ARG(scan); /* which paren pair */
4733 ln = rex->offs[n].start;
4734 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
4735 if (rex->lastparen < n || ln == -1)
4736 sayNO; /* Do not match unless seen CLOSEn. */
4737 if (ln == rex->offs[n].end)
4740 s = reginfo->strbeg + ln;
4741 if (type != REF /* REF can do byte comparison */
4742 && (utf8_target || type == REFFU))
4743 { /* XXX handle REFFL better */
4744 char * limit = reginfo->strend;
4746 /* This call case insensitively compares the entire buffer
4747 * at s, with the current input starting at locinput, but
4748 * not going off the end given by reginfo->strend, and
4749 * returns in <limit> upon success, how much of the
4750 * current input was matched */
4751 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4752 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4760 /* Not utf8: Inline the first character, for speed. */
4761 if (!NEXTCHR_IS_EOS &&
4762 UCHARAT(s) != nextchr &&
4764 UCHARAT(s) != fold_array[nextchr]))
4766 ln = rex->offs[n].end - ln;
4767 if (locinput + ln > reginfo->strend)
4769 if (ln > 1 && (type == REF
4770 ? memNE(s, locinput, ln)
4771 : ! folder(s, locinput, ln)))
4777 case NOTHING: /* null op; e.g. the 'nothing' following
4778 * the '*' in m{(a+|b)*}' */
4780 case TAIL: /* placeholder while compiling (A|B|C) */
4783 case BACK: /* ??? doesn't appear to be used ??? */
4787 #define ST st->u.eval
4792 regexp_internal *rei;
4793 regnode *startpoint;
4795 case GOSTART: /* (?R) */
4796 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4797 if (cur_eval && cur_eval->locinput==locinput) {
4798 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4799 Perl_croak(aTHX_ "Infinite recursion in regex");
4800 if ( ++nochange_depth > max_nochange_depth )
4802 "Pattern subroutine nesting without pos change"
4803 " exceeded limit in regex");
4810 if (OP(scan)==GOSUB) {
4811 startpoint = scan + ARG2L(scan);
4812 ST.close_paren = ARG(scan);
4814 startpoint = rei->program+1;
4817 goto eval_recurse_doit;
4818 assert(0); /* NOTREACHED */
4820 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4821 if (cur_eval && cur_eval->locinput==locinput) {
4822 if ( ++nochange_depth > max_nochange_depth )
4823 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4828 /* execute the code in the {...} */
4832 OP * const oop = PL_op;
4833 COP * const ocurcop = PL_curcop;
4837 /* save *all* paren positions */
4838 regcppush(rex, 0, maxopenparen);
4839 REGCP_SET(runops_cp);
4842 caller_cv = find_runcv(NULL);
4846 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4848 (REGEXP*)(rexi->data->data[n])
4851 nop = (OP*)rexi->data->data[n+1];
4853 else if (rexi->data->what[n] == 'l') { /* literal code */
4855 nop = (OP*)rexi->data->data[n];
4856 assert(CvDEPTH(newcv));
4859 /* literal with own CV */
4860 assert(rexi->data->what[n] == 'L');
4861 newcv = rex->qr_anoncv;
4862 nop = (OP*)rexi->data->data[n];
4865 /* normally if we're about to execute code from the same
4866 * CV that we used previously, we just use the existing
4867 * CX stack entry. However, its possible that in the
4868 * meantime we may have backtracked, popped from the save
4869 * stack, and undone the SAVECOMPPAD(s) associated with
4870 * PUSH_MULTICALL; in which case PL_comppad no longer
4871 * points to newcv's pad. */
4872 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4874 U8 flags = (CXp_SUB_RE |
4875 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
4876 if (last_pushed_cv) {
4877 CHANGE_MULTICALL_FLAGS(newcv, flags);
4880 PUSH_MULTICALL_FLAGS(newcv, flags);
4882 last_pushed_cv = newcv;
4885 /* these assignments are just to silence compiler
4887 multicall_cop = NULL;
4890 last_pad = PL_comppad;
4892 /* the initial nextstate you would normally execute
4893 * at the start of an eval (which would cause error
4894 * messages to come from the eval), may be optimised
4895 * away from the execution path in the regex code blocks;
4896 * so manually set PL_curcop to it initially */
4898 OP *o = cUNOPx(nop)->op_first;
4899 assert(o->op_type == OP_NULL);
4900 if (o->op_targ == OP_SCOPE) {
4901 o = cUNOPo->op_first;
4904 assert(o->op_targ == OP_LEAVE);
4905 o = cUNOPo->op_first;
4906 assert(o->op_type == OP_ENTER);
4910 if (o->op_type != OP_STUB) {
4911 assert( o->op_type == OP_NEXTSTATE
4912 || o->op_type == OP_DBSTATE
4913 || (o->op_type == OP_NULL
4914 && ( o->op_targ == OP_NEXTSTATE
4915 || o->op_targ == OP_DBSTATE
4919 PL_curcop = (COP*)o;
4924 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4925 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4927 rex->offs[0].end = locinput - reginfo->strbeg;
4928 if (reginfo->info_aux_eval->pos_magic)
4929 reginfo->info_aux_eval->pos_magic->mg_len
4930 = locinput - reginfo->strbeg;
4933 SV *sv_mrk = get_sv("REGMARK", 1);
4934 sv_setsv(sv_mrk, sv_yes_mark);
4937 /* we don't use MULTICALL here as we want to call the
4938 * first op of the block of interest, rather than the
4939 * first op of the sub */
4940 before = (IV)(SP-PL_stack_base);
4942 CALLRUNOPS(aTHX); /* Scalar context. */
4944 if ((IV)(SP-PL_stack_base) == before)
4945 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4951 /* before restoring everything, evaluate the returned
4952 * value, so that 'uninit' warnings don't use the wrong
4953 * PL_op or pad. Also need to process any magic vars
4954 * (e.g. $1) *before* parentheses are restored */
4959 if (logical == 0) /* (?{})/ */
4960 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4961 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4962 sw = cBOOL(SvTRUE(ret));
4965 else { /* /(??{}) */
4966 /* if its overloaded, let the regex compiler handle
4967 * it; otherwise extract regex, or stringify */
4968 if (!SvAMAGIC(ret)) {
4972 if (SvTYPE(sv) == SVt_REGEXP)
4973 re_sv = (REGEXP*) sv;
4974 else if (SvSMAGICAL(sv)) {
4975 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
4977 re_sv = (REGEXP *) mg->mg_obj;
4980 /* force any magic, undef warnings here */
4982 ret = sv_mortalcopy(ret);
4983 (void) SvPV_force_nolen(ret);
4989 /* *** Note that at this point we don't restore
4990 * PL_comppad, (or pop the CxSUB) on the assumption it may
4991 * be used again soon. This is safe as long as nothing
4992 * in the regexp code uses the pad ! */
4994 PL_curcop = ocurcop;
4995 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5001 /* only /(??{})/ from now on */
5004 /* extract RE object from returned value; compiling if
5008 re_sv = reg_temp_copy(NULL, re_sv);
5013 if (SvUTF8(ret) && IN_BYTES) {
5014 /* In use 'bytes': make a copy of the octet
5015 * sequence, but without the flag on */
5017 const char *const p = SvPV(ret, len);
5018 ret = newSVpvn_flags(p, len, SVs_TEMP);
5020 if (rex->intflags & PREGf_USE_RE_EVAL)
5021 pm_flags |= PMf_USE_RE_EVAL;
5023 /* if we got here, it should be an engine which
5024 * supports compiling code blocks and stuff */
5025 assert(rex->engine && rex->engine->op_comp);
5026 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5027 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5028 rex->engine, NULL, NULL,
5029 /* copy /msix etc to inner pattern */
5034 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
5036 /* This isn't a first class regexp. Instead, it's
5037 caching a regexp onto an existing, Perl visible
5039 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5041 /* safe to do now that any $1 etc has been
5042 * interpolated into the new pattern string and
5044 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5049 RXp_MATCH_COPIED_off(re);
5050 re->subbeg = rex->subbeg;
5051 re->sublen = rex->sublen;
5052 re->suboffset = rex->suboffset;
5053 re->subcoffset = rex->subcoffset;
5056 debug_start_match(re_sv, utf8_target, locinput,
5057 reginfo->strend, "Matching embedded");
5059 startpoint = rei->program + 1;
5060 ST.close_paren = 0; /* only used for GOSUB */
5062 eval_recurse_doit: /* Share code with GOSUB below this line */
5063 /* run the pattern returned from (??{...}) */
5065 /* Save *all* the positions. */
5066 ST.cp = regcppush(rex, 0, maxopenparen);
5067 REGCP_SET(ST.lastcp);
5070 re->lastcloseparen = 0;
5074 /* invalidate the S-L poscache. We're now executing a
5075 * different set of WHILEM ops (and their associated
5076 * indexes) against the same string, so the bits in the
5077 * cache are meaningless. Setting maxiter to zero forces
5078 * the cache to be invalidated and zeroed before reuse.
5079 * XXX This is too dramatic a measure. Ideally we should
5080 * save the old cache and restore when running the outer
5082 reginfo->poscache_maxiter = 0;
5084 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5086 ST.prev_rex = rex_sv;
5087 ST.prev_curlyx = cur_curlyx;
5089 SET_reg_curpm(rex_sv);
5094 ST.prev_eval = cur_eval;
5096 /* now continue from first node in postoned RE */
5097 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5098 assert(0); /* NOTREACHED */
5101 case EVAL_AB: /* cleanup after a successful (??{A})B */
5102 /* note: this is called twice; first after popping B, then A */
5103 rex_sv = ST.prev_rex;
5104 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5105 SET_reg_curpm(rex_sv);
5106 rex = ReANY(rex_sv);
5107 rexi = RXi_GET(rex);
5109 cur_eval = ST.prev_eval;
5110 cur_curlyx = ST.prev_curlyx;
5112 /* Invalidate cache. See "invalidate" comment above. */
5113 reginfo->poscache_maxiter = 0;
5114 if ( nochange_depth )
5119 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5120 /* note: this is called twice; first after popping B, then A */
5121 rex_sv = ST.prev_rex;
5122 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5123 SET_reg_curpm(rex_sv);
5124 rex = ReANY(rex_sv);
5125 rexi = RXi_GET(rex);
5127 REGCP_UNWIND(ST.lastcp);
5128 regcppop(rex, &maxopenparen);
5129 cur_eval = ST.prev_eval;
5130 cur_curlyx = ST.prev_curlyx;
5131 /* Invalidate cache. See "invalidate" comment above. */
5132 reginfo->poscache_maxiter = 0;
5133 if ( nochange_depth )
5139 n = ARG(scan); /* which paren pair */
5140 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5141 if (n > maxopenparen)
5143 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5144 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5148 (IV)rex->offs[n].start_tmp,
5154 /* XXX really need to log other places start/end are set too */
5155 #define CLOSE_CAPTURE \
5156 rex->offs[n].start = rex->offs[n].start_tmp; \
5157 rex->offs[n].end = locinput - reginfo->strbeg; \
5158 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5159 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5161 PTR2UV(rex->offs), \
5163 (IV)rex->offs[n].start, \
5164 (IV)rex->offs[n].end \
5168 n = ARG(scan); /* which paren pair */
5170 if (n > rex->lastparen)
5172 rex->lastcloseparen = n;
5173 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5178 case ACCEPT: /* (*ACCEPT) */
5182 cursor && OP(cursor)!=END;
5183 cursor=regnext(cursor))
5185 if ( OP(cursor)==CLOSE ){
5187 if ( n <= lastopen ) {
5189 if (n > rex->lastparen)
5191 rex->lastcloseparen = n;
5192 if ( n == ARG(scan) || (cur_eval &&
5193 cur_eval->u.eval.close_paren == n))
5202 case GROUPP: /* (?(1)) */
5203 n = ARG(scan); /* which paren pair */
5204 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5207 case NGROUPP: /* (?(<name>)) */
5208 /* reg_check_named_buff_matched returns 0 for no match */
5209 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5212 case INSUBP: /* (?(R)) */
5214 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5217 case DEFINEP: /* (?(DEFINE)) */
5221 case IFTHEN: /* (?(cond)A|B) */
5222 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5224 next = NEXTOPER(NEXTOPER(scan));
5226 next = scan + ARG(scan);
5227 if (OP(next) == IFTHEN) /* Fake one. */
5228 next = NEXTOPER(NEXTOPER(next));
5232 case LOGICAL: /* modifier for EVAL and IFMATCH */
5233 logical = scan->flags;
5236 /*******************************************************************
5238 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5239 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5240 STAR/PLUS/CURLY/CURLYN are used instead.)
5242 A*B is compiled as <CURLYX><A><WHILEM><B>
5244 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5245 state, which contains the current count, initialised to -1. It also sets
5246 cur_curlyx to point to this state, with any previous value saved in the
5249 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5250 since the pattern may possibly match zero times (i.e. it's a while {} loop
5251 rather than a do {} while loop).
5253 Each entry to WHILEM represents a successful match of A. The count in the
5254 CURLYX block is incremented, another WHILEM state is pushed, and execution
5255 passes to A or B depending on greediness and the current count.
5257 For example, if matching against the string a1a2a3b (where the aN are
5258 substrings that match /A/), then the match progresses as follows: (the
5259 pushed states are interspersed with the bits of strings matched so far):
5262 <CURLYX cnt=0><WHILEM>
5263 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5264 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5265 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5266 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5268 (Contrast this with something like CURLYM, which maintains only a single
5272 a1 <CURLYM cnt=1> a2
5273 a1 a2 <CURLYM cnt=2> a3
5274 a1 a2 a3 <CURLYM cnt=3> b
5277 Each WHILEM state block marks a point to backtrack to upon partial failure
5278 of A or B, and also contains some minor state data related to that
5279 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5280 overall state, such as the count, and pointers to the A and B ops.
5282 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5283 must always point to the *current* CURLYX block, the rules are:
5285 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5286 and set cur_curlyx to point the new block.
5288 When popping the CURLYX block after a successful or unsuccessful match,
5289 restore the previous cur_curlyx.
5291 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5292 to the outer one saved in the CURLYX block.
5294 When popping the WHILEM block after a successful or unsuccessful B match,
5295 restore the previous cur_curlyx.
5297 Here's an example for the pattern (AI* BI)*BO
5298 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5301 curlyx backtrack stack
5302 ------ ---------------
5304 CO <CO prev=NULL> <WO>
5305 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5306 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5307 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5309 At this point the pattern succeeds, and we work back down the stack to
5310 clean up, restoring as we go:
5312 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5313 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5314 CO <CO prev=NULL> <WO>
5317 *******************************************************************/
5319 #define ST st->u.curlyx
5321 case CURLYX: /* start of /A*B/ (for complex A) */
5323 /* No need to save/restore up to this paren */
5324 I32 parenfloor = scan->flags;
5326 assert(next); /* keep Coverity happy */
5327 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5330 /* XXXX Probably it is better to teach regpush to support
5331 parenfloor > maxopenparen ... */
5332 if (parenfloor > (I32)rex->lastparen)
5333 parenfloor = rex->lastparen; /* Pessimization... */
5335 ST.prev_curlyx= cur_curlyx;
5337 ST.cp = PL_savestack_ix;
5339 /* these fields contain the state of the current curly.
5340 * they are accessed by subsequent WHILEMs */
5341 ST.parenfloor = parenfloor;
5346 ST.count = -1; /* this will be updated by WHILEM */
5347 ST.lastloc = NULL; /* this will be updated by WHILEM */
5349 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5350 assert(0); /* NOTREACHED */
5353 case CURLYX_end: /* just finished matching all of A*B */
5354 cur_curlyx = ST.prev_curlyx;
5356 assert(0); /* NOTREACHED */
5358 case CURLYX_end_fail: /* just failed to match all of A*B */
5360 cur_curlyx = ST.prev_curlyx;
5362 assert(0); /* NOTREACHED */
5366 #define ST st->u.whilem
5368 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5370 /* see the discussion above about CURLYX/WHILEM */
5372 int min = ARG1(cur_curlyx->u.curlyx.me);
5373 int max = ARG2(cur_curlyx->u.curlyx.me);
5374 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5376 assert(cur_curlyx); /* keep Coverity happy */
5377 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5378 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5379 ST.cache_offset = 0;
5383 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5384 "%*s whilem: matched %ld out of %d..%d\n",
5385 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5388 /* First just match a string of min A's. */
5391 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5393 cur_curlyx->u.curlyx.lastloc = locinput;
5394 REGCP_SET(ST.lastcp);
5396 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5397 assert(0); /* NOTREACHED */
5400 /* If degenerate A matches "", assume A done. */
5402 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5403 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5404 "%*s whilem: empty match detected, trying continuation...\n",
5405 REPORT_CODE_OFF+depth*2, "")
5407 goto do_whilem_B_max;
5410 /* super-linear cache processing.
5412 * The idea here is that for certain types of CURLYX/WHILEM -
5413 * principally those whose upper bound is infinity (and
5414 * excluding regexes that have things like \1 and other very
5415 * non-regular expresssiony things), then if a pattern like
5416 * /....A*.../ fails and we backtrack to the WHILEM, then we
5417 * make a note that this particular WHILEM op was at string
5418 * position 47 (say) when the rest of pattern failed. Then, if
5419 * we ever find ourselves back at that WHILEM, and at string
5420 * position 47 again, we can just fail immediately rather than
5421 * running the rest of the pattern again.
5423 * This is very handy when patterns start to go
5424 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5425 * with a combinatorial explosion of backtracking.
5427 * The cache is implemented as a bit array, with one bit per
5428 * string byte position per WHILEM op (up to 16) - so its
5429 * between 0.25 and 2x the string size.
5431 * To avoid allocating a poscache buffer every time, we do an
5432 * initially countdown; only after we have executed a WHILEM
5433 * op (string-length x #WHILEMs) times do we allocate the
5436 * The top 4 bits of scan->flags byte say how many different
5437 * relevant CURLLYX/WHILEM op pairs there are, while the
5438 * bottom 4-bits is the identifying index number of this
5444 if (!reginfo->poscache_maxiter) {
5445 /* start the countdown: Postpone detection until we
5446 * know the match is not *that* much linear. */
5447 reginfo->poscache_maxiter
5448 = (reginfo->strend - reginfo->strbeg + 1)
5450 /* possible overflow for long strings and many CURLYX's */
5451 if (reginfo->poscache_maxiter < 0)
5452 reginfo->poscache_maxiter = I32_MAX;
5453 reginfo->poscache_iter = reginfo->poscache_maxiter;
5456 if (reginfo->poscache_iter-- == 0) {
5457 /* initialise cache */
5458 const I32 size = (reginfo->poscache_maxiter + 7)/8;
5459 regmatch_info_aux *const aux = reginfo->info_aux;
5460 if (aux->poscache) {
5461 if ((I32)reginfo->poscache_size < size) {
5462 Renew(aux->poscache, size, char);
5463 reginfo->poscache_size = size;
5465 Zero(aux->poscache, size, char);
5468 reginfo->poscache_size = size;
5469 Newxz(aux->poscache, size, char);
5471 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5472 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5473 PL_colors[4], PL_colors[5])
5477 if (reginfo->poscache_iter < 0) {
5478 /* have we already failed at this position? */
5481 reginfo->poscache_iter = -1; /* stop eventual underflow */
5482 offset = (scan->flags & 0xf) - 1
5483 + (locinput - reginfo->strbeg)
5485 mask = 1 << (offset % 8);
5487 if (reginfo->info_aux->poscache[offset] & mask) {
5488 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5489 "%*s whilem: (cache) already tried at this position...\n",
5490 REPORT_CODE_OFF+depth*2, "")
5492 sayNO; /* cache records failure */
5494 ST.cache_offset = offset;
5495 ST.cache_mask = mask;
5499 /* Prefer B over A for minimal matching. */
5501 if (cur_curlyx->u.curlyx.minmod) {
5502 ST.save_curlyx = cur_curlyx;
5503 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5504 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5506 REGCP_SET(ST.lastcp);
5507 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5509 assert(0); /* NOTREACHED */
5512 /* Prefer A over B for maximal matching. */
5514 if (n < max) { /* More greed allowed? */
5515 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5517 cur_curlyx->u.curlyx.lastloc = locinput;
5518 REGCP_SET(ST.lastcp);
5519 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5520 assert(0); /* NOTREACHED */
5522 goto do_whilem_B_max;
5524 assert(0); /* NOTREACHED */
5526 case WHILEM_B_min: /* just matched B in a minimal match */
5527 case WHILEM_B_max: /* just matched B in a maximal match */
5528 cur_curlyx = ST.save_curlyx;
5530 assert(0); /* NOTREACHED */
5532 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5533 cur_curlyx = ST.save_curlyx;
5534 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5535 cur_curlyx->u.curlyx.count--;
5537 assert(0); /* NOTREACHED */
5539 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5541 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5542 REGCP_UNWIND(ST.lastcp);
5543 regcppop(rex, &maxopenparen);
5544 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5545 cur_curlyx->u.curlyx.count--;
5547 assert(0); /* NOTREACHED */
5549 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5550 REGCP_UNWIND(ST.lastcp);
5551 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5552 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5553 "%*s whilem: failed, trying continuation...\n",
5554 REPORT_CODE_OFF+depth*2, "")
5557 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5558 && ckWARN(WARN_REGEXP)
5559 && !reginfo->warned)
5561 reginfo->warned = TRUE;
5562 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5563 "Complex regular subexpression recursion limit (%d) "
5569 ST.save_curlyx = cur_curlyx;
5570 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5571 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5573 assert(0); /* NOTREACHED */
5575 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5576 cur_curlyx = ST.save_curlyx;
5577 REGCP_UNWIND(ST.lastcp);
5578 regcppop(rex, &maxopenparen);
5580 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5581 /* Maximum greed exceeded */
5582 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5583 && ckWARN(WARN_REGEXP)
5584 && !reginfo->warned)
5586 reginfo->warned = TRUE;
5587 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5588 "Complex regular subexpression recursion "
5589 "limit (%d) exceeded",
5592 cur_curlyx->u.curlyx.count--;
5596 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5597 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5599 /* Try grabbing another A and see if it helps. */
5600 cur_curlyx->u.curlyx.lastloc = locinput;
5601 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5603 REGCP_SET(ST.lastcp);
5604 PUSH_STATE_GOTO(WHILEM_A_min,
5605 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5607 assert(0); /* NOTREACHED */
5610 #define ST st->u.branch
5612 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5613 next = scan + ARG(scan);
5616 scan = NEXTOPER(scan);
5619 case BRANCH: /* /(...|A|...)/ */
5620 scan = NEXTOPER(scan); /* scan now points to inner node */
5621 ST.lastparen = rex->lastparen;
5622 ST.lastcloseparen = rex->lastcloseparen;
5623 ST.next_branch = next;
5626 /* Now go into the branch */
5628 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5630 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5632 assert(0); /* NOTREACHED */
5634 case CUTGROUP: /* /(*THEN)/ */
5635 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5636 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5637 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5638 assert(0); /* NOTREACHED */
5640 case CUTGROUP_next_fail:
5643 if (st->u.mark.mark_name)
5644 sv_commit = st->u.mark.mark_name;
5646 assert(0); /* NOTREACHED */
5650 assert(0); /* NOTREACHED */
5652 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5657 REGCP_UNWIND(ST.cp);
5658 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5659 scan = ST.next_branch;
5660 /* no more branches? */
5661 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5663 PerlIO_printf( Perl_debug_log,
5664 "%*s %sBRANCH failed...%s\n",
5665 REPORT_CODE_OFF+depth*2, "",
5671 continue; /* execute next BRANCH[J] op */
5672 assert(0); /* NOTREACHED */
5674 case MINMOD: /* next op will be non-greedy, e.g. A*? */
5679 #define ST st->u.curlym
5681 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5683 /* This is an optimisation of CURLYX that enables us to push
5684 * only a single backtracking state, no matter how many matches
5685 * there are in {m,n}. It relies on the pattern being constant
5686 * length, with no parens to influence future backrefs
5690 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5692 ST.lastparen = rex->lastparen;
5693 ST.lastcloseparen = rex->lastcloseparen;
5695 /* if paren positive, emulate an OPEN/CLOSE around A */
5697 U32 paren = ST.me->flags;
5698 if (paren > maxopenparen)
5699 maxopenparen = paren;
5700 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5708 ST.c1 = CHRTEST_UNINIT;
5711 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5714 curlym_do_A: /* execute the A in /A{m,n}B/ */
5715 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5716 assert(0); /* NOTREACHED */
5718 case CURLYM_A: /* we've just matched an A */
5720 /* after first match, determine A's length: u.curlym.alen */
5721 if (ST.count == 1) {
5722 if (reginfo->is_utf8_target) {
5723 char *s = st->locinput;
5724 while (s < locinput) {
5730 ST.alen = locinput - st->locinput;
5733 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5736 PerlIO_printf(Perl_debug_log,
5737 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5738 (int)(REPORT_CODE_OFF+(depth*2)), "",
5739 (IV) ST.count, (IV)ST.alen)
5742 if (cur_eval && cur_eval->u.eval.close_paren &&
5743 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5747 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5748 if ( max == REG_INFTY || ST.count < max )
5749 goto curlym_do_A; /* try to match another A */
5751 goto curlym_do_B; /* try to match B */
5753 case CURLYM_A_fail: /* just failed to match an A */
5754 REGCP_UNWIND(ST.cp);
5756 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5757 || (cur_eval && cur_eval->u.eval.close_paren &&
5758 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5761 curlym_do_B: /* execute the B in /A{m,n}B/ */
5762 if (ST.c1 == CHRTEST_UNINIT) {
5763 /* calculate c1 and c2 for possible match of 1st char
5764 * following curly */
5765 ST.c1 = ST.c2 = CHRTEST_VOID;
5766 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5767 regnode *text_node = ST.B;
5768 if (! HAS_TEXT(text_node))
5769 FIND_NEXT_IMPT(text_node);
5772 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5774 But the former is redundant in light of the latter.
5776 if this changes back then the macro for
5777 IS_TEXT and friends need to change.
5779 if (PL_regkind[OP(text_node)] == EXACT) {
5780 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5781 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5791 PerlIO_printf(Perl_debug_log,
5792 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5793 (int)(REPORT_CODE_OFF+(depth*2)),
5796 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
5797 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
5798 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
5799 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
5801 /* simulate B failing */
5803 PerlIO_printf(Perl_debug_log,
5804 "%*s CURLYM Fast bail next target=U+%"UVXf" c1=U+%"UVXf" c2=U+%"UVXf"\n",
5805 (int)(REPORT_CODE_OFF+(depth*2)),"",
5806 valid_utf8_to_uvchr((U8 *) locinput, NULL),
5807 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
5808 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
5810 state_num = CURLYM_B_fail;
5811 goto reenter_switch;
5814 else if (nextchr != ST.c1 && nextchr != ST.c2) {
5815 /* simulate B failing */
5817 PerlIO_printf(Perl_debug_log,
5818 "%*s CURLYM Fast bail next target=U+%X c1=U+%X c2=U+%X\n",
5819 (int)(REPORT_CODE_OFF+(depth*2)),"",
5820 (int) nextchr, ST.c1, ST.c2)
5822 state_num = CURLYM_B_fail;
5823 goto reenter_switch;
5828 /* emulate CLOSE: mark current A as captured */
5829 I32 paren = ST.me->flags;
5831 rex->offs[paren].start
5832 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
5833 rex->offs[paren].end = locinput - reginfo->strbeg;
5834 if ((U32)paren > rex->lastparen)
5835 rex->lastparen = paren;
5836 rex->lastcloseparen = paren;
5839 rex->offs[paren].end = -1;
5840 if (cur_eval && cur_eval->u.eval.close_paren &&
5841 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5850 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5851 assert(0); /* NOTREACHED */
5853 case CURLYM_B_fail: /* just failed to match a B */
5854 REGCP_UNWIND(ST.cp);
5855 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5857 I32 max = ARG2(ST.me);
5858 if (max != REG_INFTY && ST.count == max)
5860 goto curlym_do_A; /* try to match a further A */
5862 /* backtrack one A */
5863 if (ST.count == ARG1(ST.me) /* min */)
5866 SET_locinput(HOPc(locinput, -ST.alen));
5867 goto curlym_do_B; /* try to match B */
5870 #define ST st->u.curly
5872 #define CURLY_SETPAREN(paren, success) \
5875 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
5876 rex->offs[paren].end = locinput - reginfo->strbeg; \
5877 if (paren > rex->lastparen) \
5878 rex->lastparen = paren; \
5879 rex->lastcloseparen = paren; \
5882 rex->offs[paren].end = -1; \
5883 rex->lastparen = ST.lastparen; \
5884 rex->lastcloseparen = ST.lastcloseparen; \
5888 case STAR: /* /A*B/ where A is width 1 char */
5892 scan = NEXTOPER(scan);
5895 case PLUS: /* /A+B/ where A is width 1 char */
5899 scan = NEXTOPER(scan);
5902 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
5903 ST.paren = scan->flags; /* Which paren to set */
5904 ST.lastparen = rex->lastparen;
5905 ST.lastcloseparen = rex->lastcloseparen;
5906 if (ST.paren > maxopenparen)
5907 maxopenparen = ST.paren;
5908 ST.min = ARG1(scan); /* min to match */
5909 ST.max = ARG2(scan); /* max to match */
5910 if (cur_eval && cur_eval->u.eval.close_paren &&
5911 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5915 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5918 case CURLY: /* /A{m,n}B/ where A is width 1 char */
5920 ST.min = ARG1(scan); /* min to match */
5921 ST.max = ARG2(scan); /* max to match */
5922 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5925 * Lookahead to avoid useless match attempts
5926 * when we know what character comes next.
5928 * Used to only do .*x and .*?x, but now it allows
5929 * for )'s, ('s and (?{ ... })'s to be in the way
5930 * of the quantifier and the EXACT-like node. -- japhy
5933 assert(ST.min <= ST.max);
5934 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
5935 ST.c1 = ST.c2 = CHRTEST_VOID;
5938 regnode *text_node = next;
5940 if (! HAS_TEXT(text_node))
5941 FIND_NEXT_IMPT(text_node);
5943 if (! HAS_TEXT(text_node))
5944 ST.c1 = ST.c2 = CHRTEST_VOID;
5946 if ( PL_regkind[OP(text_node)] != EXACT ) {
5947 ST.c1 = ST.c2 = CHRTEST_VOID;
5951 /* Currently we only get here when
5953 PL_rekind[OP(text_node)] == EXACT
5955 if this changes back then the macro for IS_TEXT and
5956 friends need to change. */
5957 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5958 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5970 char *li = locinput;
5973 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
5979 if (ST.c1 == CHRTEST_VOID)
5980 goto curly_try_B_min;
5982 ST.oldloc = locinput;
5984 /* set ST.maxpos to the furthest point along the
5985 * string that could possibly match */
5986 if (ST.max == REG_INFTY) {
5987 ST.maxpos = reginfo->strend - 1;
5989 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5992 else if (utf8_target) {
5993 int m = ST.max - ST.min;
5994 for (ST.maxpos = locinput;
5995 m >0 && ST.maxpos < reginfo->strend; m--)
5996 ST.maxpos += UTF8SKIP(ST.maxpos);
5999 ST.maxpos = locinput + ST.max - ST.min;
6000 if (ST.maxpos >= reginfo->strend)
6001 ST.maxpos = reginfo->strend - 1;
6003 goto curly_try_B_min_known;
6007 /* avoid taking address of locinput, so it can remain
6009 char *li = locinput;
6010 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6011 if (ST.count < ST.min)
6014 if ((ST.count > ST.min)
6015 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6017 /* A{m,n} must come at the end of the string, there's
6018 * no point in backing off ... */
6020 /* ...except that $ and \Z can match before *and* after
6021 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6022 We may back off by one in this case. */
6023 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6027 goto curly_try_B_max;
6029 assert(0); /* NOTREACHED */
6032 case CURLY_B_min_known_fail:
6033 /* failed to find B in a non-greedy match where c1,c2 valid */
6035 REGCP_UNWIND(ST.cp);
6037 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6039 /* Couldn't or didn't -- move forward. */
6040 ST.oldloc = locinput;
6042 locinput += UTF8SKIP(locinput);
6046 curly_try_B_min_known:
6047 /* find the next place where 'B' could work, then call B */
6051 n = (ST.oldloc == locinput) ? 0 : 1;
6052 if (ST.c1 == ST.c2) {
6053 /* set n to utf8_distance(oldloc, locinput) */
6054 while (locinput <= ST.maxpos
6055 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6057 locinput += UTF8SKIP(locinput);
6062 /* set n to utf8_distance(oldloc, locinput) */
6063 while (locinput <= ST.maxpos
6064 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6065 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6067 locinput += UTF8SKIP(locinput);
6072 else { /* Not utf8_target */
6073 if (ST.c1 == ST.c2) {
6074 while (locinput <= ST.maxpos &&
6075 UCHARAT(locinput) != ST.c1)
6079 while (locinput <= ST.maxpos
6080 && UCHARAT(locinput) != ST.c1
6081 && UCHARAT(locinput) != ST.c2)
6084 n = locinput - ST.oldloc;
6086 if (locinput > ST.maxpos)
6089 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6090 * at b; check that everything between oldloc and
6091 * locinput matches */
6092 char *li = ST.oldloc;
6094 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6096 assert(n == REG_INFTY || locinput == li);
6098 CURLY_SETPAREN(ST.paren, ST.count);
6099 if (cur_eval && cur_eval->u.eval.close_paren &&
6100 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6103 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6105 assert(0); /* NOTREACHED */
6108 case CURLY_B_min_fail:
6109 /* failed to find B in a non-greedy match where c1,c2 invalid */
6111 REGCP_UNWIND(ST.cp);
6113 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6115 /* failed -- move forward one */
6117 char *li = locinput;
6118 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6125 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6126 ST.count > 0)) /* count overflow ? */
6129 CURLY_SETPAREN(ST.paren, ST.count);
6130 if (cur_eval && cur_eval->u.eval.close_paren &&
6131 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6134 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6138 assert(0); /* NOTREACHED */
6142 /* a successful greedy match: now try to match B */
6143 if (cur_eval && cur_eval->u.eval.close_paren &&
6144 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6148 bool could_match = locinput < reginfo->strend;
6150 /* If it could work, try it. */
6151 if (ST.c1 != CHRTEST_VOID && could_match) {
6152 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6154 could_match = memEQ(locinput,
6159 UTF8SKIP(locinput));
6162 could_match = UCHARAT(locinput) == ST.c1
6163 || UCHARAT(locinput) == ST.c2;
6166 if (ST.c1 == CHRTEST_VOID || could_match) {
6167 CURLY_SETPAREN(ST.paren, ST.count);
6168 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6169 assert(0); /* NOTREACHED */
6174 case CURLY_B_max_fail:
6175 /* failed to find B in a greedy match */
6177 REGCP_UNWIND(ST.cp);
6179 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6182 if (--ST.count < ST.min)
6184 locinput = HOPc(locinput, -1);
6185 goto curly_try_B_max;
6189 case END: /* last op of main pattern */
6192 /* we've just finished A in /(??{A})B/; now continue with B */
6194 st->u.eval.prev_rex = rex_sv; /* inner */
6196 /* Save *all* the positions. */
6197 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6198 rex_sv = cur_eval->u.eval.prev_rex;
6199 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6200 SET_reg_curpm(rex_sv);
6201 rex = ReANY(rex_sv);
6202 rexi = RXi_GET(rex);
6203 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6205 REGCP_SET(st->u.eval.lastcp);
6207 /* Restore parens of the outer rex without popping the
6209 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6212 st->u.eval.prev_eval = cur_eval;
6213 cur_eval = cur_eval->u.eval.prev_eval;
6215 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6216 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6217 if ( nochange_depth )
6220 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6221 locinput); /* match B */
6224 if (locinput < reginfo->till) {
6225 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6226 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6228 (long)(locinput - startpos),
6229 (long)(reginfo->till - startpos),
6232 sayNO_SILENT; /* Cannot match: too short. */
6234 sayYES; /* Success! */
6236 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6238 PerlIO_printf(Perl_debug_log,
6239 "%*s %ssubpattern success...%s\n",
6240 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6241 sayYES; /* Success! */
6244 #define ST st->u.ifmatch
6249 case SUSPEND: /* (?>A) */
6251 newstart = locinput;
6254 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6256 goto ifmatch_trivial_fail_test;
6258 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6260 ifmatch_trivial_fail_test:
6262 char * const s = HOPBACKc(locinput, scan->flags);
6267 sw = 1 - cBOOL(ST.wanted);
6271 next = scan + ARG(scan);
6279 newstart = locinput;
6283 ST.logical = logical;
6284 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6286 /* execute body of (?...A) */
6287 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6288 assert(0); /* NOTREACHED */
6291 case IFMATCH_A_fail: /* body of (?...A) failed */
6292 ST.wanted = !ST.wanted;
6295 case IFMATCH_A: /* body of (?...A) succeeded */
6297 sw = cBOOL(ST.wanted);
6299 else if (!ST.wanted)
6302 if (OP(ST.me) != SUSPEND) {
6303 /* restore old position except for (?>...) */
6304 locinput = st->locinput;
6306 scan = ST.me + ARG(ST.me);
6309 continue; /* execute B */
6313 case LONGJMP: /* alternative with many branches compiles to
6314 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6315 next = scan + ARG(scan);
6320 case COMMIT: /* (*COMMIT) */
6321 reginfo->cutpoint = reginfo->strend;
6324 case PRUNE: /* (*PRUNE) */
6326 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6327 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6328 assert(0); /* NOTREACHED */
6330 case COMMIT_next_fail:
6334 case OPFAIL: /* (*FAIL) */
6336 assert(0); /* NOTREACHED */
6338 #define ST st->u.mark
6339 case MARKPOINT: /* (*MARK:foo) */
6340 ST.prev_mark = mark_state;
6341 ST.mark_name = sv_commit = sv_yes_mark
6342 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6344 ST.mark_loc = locinput;
6345 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6346 assert(0); /* NOTREACHED */
6348 case MARKPOINT_next:
6349 mark_state = ST.prev_mark;
6351 assert(0); /* NOTREACHED */
6353 case MARKPOINT_next_fail:
6354 if (popmark && sv_eq(ST.mark_name,popmark))
6356 if (ST.mark_loc > startpoint)
6357 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6358 popmark = NULL; /* we found our mark */
6359 sv_commit = ST.mark_name;
6362 PerlIO_printf(Perl_debug_log,
6363 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6364 REPORT_CODE_OFF+depth*2, "",
6365 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6368 mark_state = ST.prev_mark;
6369 sv_yes_mark = mark_state ?
6370 mark_state->u.mark.mark_name : NULL;
6372 assert(0); /* NOTREACHED */
6374 case SKIP: /* (*SKIP) */
6376 /* (*SKIP) : if we fail we cut here*/
6377 ST.mark_name = NULL;
6378 ST.mark_loc = locinput;
6379 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6381 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6382 otherwise do nothing. Meaning we need to scan
6384 regmatch_state *cur = mark_state;
6385 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6388 if ( sv_eq( cur->u.mark.mark_name,
6391 ST.mark_name = find;
6392 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6394 cur = cur->u.mark.prev_mark;
6397 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6400 case SKIP_next_fail:
6402 /* (*CUT:NAME) - Set up to search for the name as we
6403 collapse the stack*/
6404 popmark = ST.mark_name;
6406 /* (*CUT) - No name, we cut here.*/
6407 if (ST.mark_loc > startpoint)
6408 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6409 /* but we set sv_commit to latest mark_name if there
6410 is one so they can test to see how things lead to this
6413 sv_commit=mark_state->u.mark.mark_name;
6417 assert(0); /* NOTREACHED */
6420 case LNBREAK: /* \R */
6421 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6428 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6429 PTR2UV(scan), OP(scan));
6430 Perl_croak(aTHX_ "regexp memory corruption");
6432 /* this is a point to jump to in order to increment
6433 * locinput by one character */
6435 assert(!NEXTCHR_IS_EOS);
6437 locinput += PL_utf8skip[nextchr];
6438 /* locinput is allowed to go 1 char off the end, but not 2+ */
6439 if (locinput > reginfo->strend)
6448 /* switch break jumps here */
6449 scan = next; /* prepare to execute the next op and ... */
6450 continue; /* ... jump back to the top, reusing st */
6451 assert(0); /* NOTREACHED */
6454 /* push a state that backtracks on success */
6455 st->u.yes.prev_yes_state = yes_state;
6459 /* push a new regex state, then continue at scan */
6461 regmatch_state *newst;
6464 regmatch_state *cur = st;
6465 regmatch_state *curyes = yes_state;
6467 regmatch_slab *slab = PL_regmatch_slab;
6468 for (;curd > -1;cur--,curd--) {
6469 if (cur < SLAB_FIRST(slab)) {
6471 cur = SLAB_LAST(slab);
6473 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6474 REPORT_CODE_OFF + 2 + depth * 2,"",
6475 curd, PL_reg_name[cur->resume_state],
6476 (curyes == cur) ? "yes" : ""
6479 curyes = cur->u.yes.prev_yes_state;
6482 DEBUG_STATE_pp("push")
6485 st->locinput = locinput;
6487 if (newst > SLAB_LAST(PL_regmatch_slab))
6488 newst = S_push_slab(aTHX);
6489 PL_regmatch_state = newst;
6491 locinput = pushinput;
6494 assert(0); /* NOTREACHED */
6499 * We get here only if there's trouble -- normally "case END" is
6500 * the terminating point.
6502 Perl_croak(aTHX_ "corrupted regexp pointers");
6508 /* we have successfully completed a subexpression, but we must now
6509 * pop to the state marked by yes_state and continue from there */
6510 assert(st != yes_state);
6512 while (st != yes_state) {
6514 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6515 PL_regmatch_slab = PL_regmatch_slab->prev;
6516 st = SLAB_LAST(PL_regmatch_slab);
6520 DEBUG_STATE_pp("pop (no final)");
6522 DEBUG_STATE_pp("pop (yes)");
6528 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6529 || yes_state > SLAB_LAST(PL_regmatch_slab))
6531 /* not in this slab, pop slab */
6532 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6533 PL_regmatch_slab = PL_regmatch_slab->prev;
6534 st = SLAB_LAST(PL_regmatch_slab);
6536 depth -= (st - yes_state);
6539 yes_state = st->u.yes.prev_yes_state;
6540 PL_regmatch_state = st;
6543 locinput= st->locinput;
6544 state_num = st->resume_state + no_final;
6545 goto reenter_switch;
6548 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6549 PL_colors[4], PL_colors[5]));
6551 if (reginfo->info_aux_eval) {
6552 /* each successfully executed (?{...}) block does the equivalent of
6553 * local $^R = do {...}
6554 * When popping the save stack, all these locals would be undone;
6555 * bypass this by setting the outermost saved $^R to the latest
6557 if (oreplsv != GvSV(PL_replgv))
6558 sv_setsv(oreplsv, GvSV(PL_replgv));
6565 PerlIO_printf(Perl_debug_log,
6566 "%*s %sfailed...%s\n",
6567 REPORT_CODE_OFF+depth*2, "",
6568 PL_colors[4], PL_colors[5])
6580 /* there's a previous state to backtrack to */
6582 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6583 PL_regmatch_slab = PL_regmatch_slab->prev;
6584 st = SLAB_LAST(PL_regmatch_slab);
6586 PL_regmatch_state = st;
6587 locinput= st->locinput;
6589 DEBUG_STATE_pp("pop");
6591 if (yes_state == st)
6592 yes_state = st->u.yes.prev_yes_state;
6594 state_num = st->resume_state + 1; /* failure = success + 1 */
6595 goto reenter_switch;
6600 if (rex->intflags & PREGf_VERBARG_SEEN) {
6601 SV *sv_err = get_sv("REGERROR", 1);
6602 SV *sv_mrk = get_sv("REGMARK", 1);
6604 sv_commit = &PL_sv_no;
6606 sv_yes_mark = &PL_sv_yes;
6609 sv_commit = &PL_sv_yes;
6610 sv_yes_mark = &PL_sv_no;
6612 sv_setsv(sv_err, sv_commit);
6613 sv_setsv(sv_mrk, sv_yes_mark);
6617 if (last_pushed_cv) {
6620 PERL_UNUSED_VAR(SP);
6623 assert(!result || locinput - reginfo->strbeg >= 0);
6624 return result ? locinput - reginfo->strbeg : -1;
6628 - regrepeat - repeatedly match something simple, report how many
6630 * What 'simple' means is a node which can be the operand of a quantifier like
6633 * startposp - pointer a pointer to the start position. This is updated
6634 * to point to the byte following the highest successful
6636 * p - the regnode to be repeatedly matched against.
6637 * reginfo - struct holding match state, such as strend
6638 * max - maximum number of things to match.
6639 * depth - (for debugging) backtracking depth.
6642 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
6643 regmatch_info *const reginfo, I32 max, int depth)
6646 char *scan; /* Pointer to current position in target string */
6648 char *loceol = reginfo->strend; /* local version */
6649 I32 hardcount = 0; /* How many matches so far */
6650 bool utf8_target = reginfo->is_utf8_target;
6651 int to_complement = 0; /* Invert the result? */
6653 _char_class_number classnum;
6655 PERL_UNUSED_ARG(depth);
6658 PERL_ARGS_ASSERT_REGREPEAT;
6661 if (max == REG_INFTY)
6663 else if (! utf8_target && loceol - scan > max)
6664 loceol = scan + max;
6666 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
6667 * to the maximum of how far we should go in it (leaving it set to the real
6668 * end, if the maximum permissible would take us beyond that). This allows
6669 * us to make the loop exit condition that we haven't gone past <loceol> to
6670 * also mean that we haven't exceeded the max permissible count, saving a
6671 * test each time through the loop. But it assumes that the OP matches a
6672 * single byte, which is true for most of the OPs below when applied to a
6673 * non-UTF-8 target. Those relatively few OPs that don't have this
6674 * characteristic will have to compensate.
6676 * There is no adjustment for UTF-8 targets, as the number of bytes per
6677 * character varies. OPs will have to test both that the count is less
6678 * than the max permissible (using <hardcount> to keep track), and that we
6679 * are still within the bounds of the string (using <loceol>. A few OPs
6680 * match a single byte no matter what the encoding. They can omit the max
6681 * test if, for the UTF-8 case, they do the adjustment that was skipped
6684 * Thus, the code above sets things up for the common case; and exceptional
6685 * cases need extra work; the common case is to make sure <scan> doesn't
6686 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
6687 * count doesn't exceed the maximum permissible */
6692 while (scan < loceol && hardcount < max && *scan != '\n') {
6693 scan += UTF8SKIP(scan);
6697 while (scan < loceol && *scan != '\n')
6703 while (scan < loceol && hardcount < max) {
6704 scan += UTF8SKIP(scan);
6711 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
6712 if (utf8_target && loceol - scan > max) {
6714 /* <loceol> hadn't been adjusted in the UTF-8 case */
6722 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6726 /* Can use a simple loop if the pattern char to match on is invariant
6727 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
6728 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
6729 * true iff it doesn't matter if the argument is in UTF-8 or not */
6730 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
6731 if (utf8_target && loceol - scan > max) {
6732 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6733 * since here, to match at all, 1 char == 1 byte */
6734 loceol = scan + max;
6736 while (scan < loceol && UCHARAT(scan) == c) {
6740 else if (reginfo->is_utf8_pat) {
6742 STRLEN scan_char_len;
6744 /* When both target and pattern are UTF-8, we have to do
6746 while (hardcount < max
6748 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
6749 && memEQ(scan, STRING(p), scan_char_len))
6751 scan += scan_char_len;
6755 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
6757 /* Target isn't utf8; convert the character in the UTF-8
6758 * pattern to non-UTF8, and do a simple loop */
6759 c = TWO_BYTE_UTF8_TO_UNI(c, *(STRING(p) + 1));
6760 while (scan < loceol && UCHARAT(scan) == c) {
6763 } /* else pattern char is above Latin1, can't possibly match the
6768 /* Here, the string must be utf8; pattern isn't, and <c> is
6769 * different in utf8 than not, so can't compare them directly.
6770 * Outside the loop, find the two utf8 bytes that represent c, and
6771 * then look for those in sequence in the utf8 string */
6772 U8 high = UTF8_TWO_BYTE_HI(c);
6773 U8 low = UTF8_TWO_BYTE_LO(c);
6775 while (hardcount < max
6776 && scan + 1 < loceol
6777 && UCHARAT(scan) == high
6778 && UCHARAT(scan + 1) == low)
6787 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6791 RXp_MATCH_TAINTED_on(prog);
6792 utf8_flags = FOLDEQ_UTF8_LOCALE;
6800 case EXACTFU_TRICKYFOLD:
6802 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6806 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
6808 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6810 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
6813 if (c1 == CHRTEST_VOID) {
6814 /* Use full Unicode fold matching */
6815 char *tmpeol = reginfo->strend;
6816 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
6817 while (hardcount < max
6818 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6819 STRING(p), NULL, pat_len,
6820 reginfo->is_utf8_pat, utf8_flags))
6823 tmpeol = reginfo->strend;
6827 else if (utf8_target) {
6829 while (scan < loceol
6831 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
6833 scan += UTF8SKIP(scan);
6838 while (scan < loceol
6840 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
6841 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
6843 scan += UTF8SKIP(scan);
6848 else if (c1 == c2) {
6849 while (scan < loceol && UCHARAT(scan) == c1) {
6854 while (scan < loceol &&
6855 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
6864 case ANYOF_WARN_SUPER:
6866 while (hardcount < max
6868 && reginclass(prog, p, (U8*)scan, utf8_target))
6870 scan += UTF8SKIP(scan);
6874 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6879 /* The argument (FLAGS) to all the POSIX node types is the class number */
6886 RXp_MATCH_TAINTED_on(prog);
6887 if (! utf8_target) {
6888 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
6894 while (hardcount < max && scan < loceol
6895 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
6898 scan += UTF8SKIP(scan);
6911 if (utf8_target && loceol - scan > max) {
6913 /* We didn't adjust <loceol> at the beginning of this routine
6914 * because is UTF-8, but it is actually ok to do so, since here, to
6915 * match, 1 char == 1 byte. */
6916 loceol = scan + max;
6918 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6931 if (! utf8_target) {
6932 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6938 /* The complement of something that matches only ASCII matches all
6939 * UTF-8 variant code points, plus everything in ASCII that isn't
6941 while (hardcount < max && scan < loceol
6942 && (! UTF8_IS_INVARIANT(*scan)
6943 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
6945 scan += UTF8SKIP(scan);
6956 if (! utf8_target) {
6957 while (scan < loceol && to_complement
6958 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
6965 classnum = (_char_class_number) FLAGS(p);
6966 if (classnum < _FIRST_NON_SWASH_CC) {
6968 /* Here, a swash is needed for above-Latin1 code points.
6969 * Process as many Latin1 code points using the built-in rules.
6970 * Go to another loop to finish processing upon encountering
6971 * the first Latin1 code point. We could do that in this loop
6972 * as well, but the other way saves having to test if the swash
6973 * has been loaded every time through the loop: extra space to
6975 while (hardcount < max && scan < loceol) {
6976 if (UTF8_IS_INVARIANT(*scan)) {
6977 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
6984 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
6985 if (! (to_complement
6986 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(*scan,
6995 goto found_above_latin1;
7002 /* For these character classes, the knowledge of how to handle
7003 * every code point is compiled in to Perl via a macro. This
7004 * code is written for making the loops as tight as possible.
7005 * It could be refactored to save space instead */
7007 case _CC_ENUM_SPACE: /* XXX would require separate code
7008 if we revert the change of \v
7011 case _CC_ENUM_PSXSPC:
7012 while (hardcount < max
7014 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7016 scan += UTF8SKIP(scan);
7020 case _CC_ENUM_BLANK:
7021 while (hardcount < max
7023 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7025 scan += UTF8SKIP(scan);
7029 case _CC_ENUM_XDIGIT:
7030 while (hardcount < max
7032 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7034 scan += UTF8SKIP(scan);
7038 case _CC_ENUM_VERTSPACE:
7039 while (hardcount < max
7041 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7043 scan += UTF8SKIP(scan);
7047 case _CC_ENUM_CNTRL:
7048 while (hardcount < max
7050 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7052 scan += UTF8SKIP(scan);
7057 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7063 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7065 /* Load the swash if not already present */
7066 if (! PL_utf8_swash_ptrs[classnum]) {
7067 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7068 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7069 "utf8", swash_property_names[classnum],
7070 &PL_sv_undef, 1, 0, NULL, &flags);
7073 while (hardcount < max && scan < loceol
7074 && to_complement ^ cBOOL(_generic_utf8(
7077 swash_fetch(PL_utf8_swash_ptrs[classnum],
7081 scan += UTF8SKIP(scan);
7088 while (hardcount < max && scan < loceol &&
7089 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7094 /* LNBREAK can match one or two latin chars, which is ok, but we
7095 * have to use hardcount in this situation, and throw away the
7096 * adjustment to <loceol> done before the switch statement */
7097 loceol = reginfo->strend;
7098 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7119 /* These are all 0 width, so match right here or not at all. */
7123 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7124 assert(0); /* NOTREACHED */
7131 c = scan - *startposp;
7135 GET_RE_DEBUG_FLAGS_DECL;
7137 SV * const prop = sv_newmortal();
7138 regprop(prog, prop, p);
7139 PerlIO_printf(Perl_debug_log,
7140 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7141 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7149 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7151 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7152 create a copy so that changes the caller makes won't change the shared one.
7153 If <altsvp> is non-null, will return NULL in it, for back-compat.
7156 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7158 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7164 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp));
7169 S_core_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp)
7171 /* Returns the swash for the input 'node' in the regex 'prog'.
7172 * If <doinit> is true, will attempt to create the swash if not already
7174 * If <listsvp> is non-null, will return the swash initialization string in
7176 * Tied intimately to how regcomp.c sets up the data structure */
7183 RXi_GET_DECL(prog,progi);
7184 const struct reg_data * const data = prog ? progi->data : NULL;
7186 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
7188 assert(ANYOF_NONBITMAP(node));
7190 if (data && data->count) {
7191 const U32 n = ARG(node);
7193 if (data->what[n] == 's') {
7194 SV * const rv = MUTABLE_SV(data->data[n]);
7195 AV * const av = MUTABLE_AV(SvRV(rv));
7196 SV **const ary = AvARRAY(av);
7197 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7199 si = *ary; /* ary[0] = the string to initialize the swash with */
7201 /* Elements 2 and 3 are either both present or both absent. [2] is
7202 * any inversion list generated at compile time; [3] indicates if
7203 * that inversion list has any user-defined properties in it. */
7204 if (av_len(av) >= 2) {
7207 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7214 /* Element [1] is reserved for the set-up swash. If already there,
7215 * return it; if not, create it and store it there */
7216 if (SvROK(ary[1])) {
7219 else if (si && doinit) {
7221 sw = _core_swash_init("utf8", /* the utf8 package */
7225 0, /* not from tr/// */
7228 (void)av_store(av, 1, sw);
7234 SV* matches_string = newSVpvn("", 0);
7236 /* Use the swash, if any, which has to have incorporated into it all
7238 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7239 && (si && si != &PL_sv_undef))
7242 /* If no swash, use the input initialization string, if available */
7243 sv_catsv(matches_string, si);
7246 /* Add the inversion list to whatever we have. This may have come from
7247 * the swash, or from an input parameter */
7249 sv_catsv(matches_string, _invlist_contents(invlist));
7251 *listsvp = matches_string;
7258 - reginclass - determine if a character falls into a character class
7260 n is the ANYOF regnode
7261 p is the target string
7262 utf8_target tells whether p is in UTF-8.
7264 Returns true if matched; false otherwise.
7266 Note that this can be a synthetic start class, a combination of various
7267 nodes, so things you think might be mutually exclusive, such as locale,
7268 aren't. It can match both locale and non-locale
7273 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const bool utf8_target)
7276 const char flags = ANYOF_FLAGS(n);
7280 PERL_ARGS_ASSERT_REGINCLASS;
7282 /* If c is not already the code point, get it. Note that
7283 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7284 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7286 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
7287 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7288 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7289 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7290 * UTF8_ALLOW_FFFF */
7291 if (c_len == (STRLEN)-1)
7292 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7295 /* If this character is potentially in the bitmap, check it */
7297 if (ANYOF_BITMAP_TEST(n, c))
7299 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
7305 else if (flags & ANYOF_LOCALE) {
7306 RXp_MATCH_TAINTED_on(prog);
7308 if ((flags & ANYOF_LOC_FOLD)
7309 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7313 else if (ANYOF_CLASS_TEST_ANY_SET(n)) {
7315 /* The data structure is arranged so bits 0, 2, 4, ... are set
7316 * if the class includes the Posix character class given by
7317 * bit/2; and 1, 3, 5, ... are set if the class includes the
7318 * complemented Posix class given by int(bit/2). So we loop
7319 * through the bits, each time changing whether we complement
7320 * the result or not. Suppose for the sake of illustration
7321 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7322 * is set, it means there is a match for this ANYOF node if the
7323 * character is in the class given by the expression (0 / 2 = 0
7324 * = \w). If it is in that class, isFOO_lc() will return 1,
7325 * and since 'to_complement' is 0, the result will stay TRUE,
7326 * and we exit the loop. Suppose instead that bit 0 is 0, but
7327 * bit 1 is 1. That means there is a match if the character
7328 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7329 * but will on bit 1. On the second iteration 'to_complement'
7330 * will be 1, so the exclusive or will reverse things, so we
7331 * are testing for \W. On the third iteration, 'to_complement'
7332 * will be 0, and we would be testing for \s; the fourth
7333 * iteration would test for \S, etc.
7335 * Note that this code assumes that all the classes are closed
7336 * under folding. For example, if a character matches \w, then
7337 * its fold does too; and vice versa. This should be true for
7338 * any well-behaved locale for all the currently defined Posix
7339 * classes, except for :lower: and :upper:, which are handled
7340 * by the pseudo-class :cased: which matches if either of the
7341 * other two does. To get rid of this assumption, an outer
7342 * loop could be used below to iterate over both the source
7343 * character, and its fold (if different) */
7346 int to_complement = 0;
7347 while (count < ANYOF_MAX) {
7348 if (ANYOF_CLASS_TEST(n, count)
7349 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7361 /* If the bitmap didn't (or couldn't) match, and something outside the
7362 * bitmap could match, try that. Locale nodes specify completely the
7363 * behavior of code points in the bit map (otherwise, a utf8 target would
7364 * cause them to be treated as Unicode and not locale), except in
7365 * the very unlikely event when this node is a synthetic start class, which
7366 * could be a combination of locale and non-locale nodes. So allow locale
7367 * to match for the synthetic start class, which will give a false
7368 * positive that will be resolved when the match is done again as not part
7369 * of the synthetic start class */
7371 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
7372 match = TRUE; /* Everything above 255 matches */
7374 else if (ANYOF_NONBITMAP(n)
7375 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
7378 || (! (flags & ANYOF_LOCALE))
7379 || OP(n) == ANYOF_SYNTHETIC))))
7381 SV * const sw = core_regclass_swash(prog, n, TRUE, 0);
7386 } else { /* Convert to utf8 */
7388 utf8_p = bytes_to_utf8(p, &len);
7391 if (swash_fetch(sw, utf8_p, TRUE)) {
7395 /* If we allocated a string above, free it */
7396 if (! utf8_target) Safefree(utf8_p);
7400 if (UNICODE_IS_SUPER(c)
7401 && OP(n) == ANYOF_WARN_SUPER
7402 && ckWARN_d(WARN_NON_UNICODE))
7404 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7405 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7409 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7410 return cBOOL(flags & ANYOF_INVERT) ^ match;
7414 S_reghop3(U8 *s, I32 off, const U8* lim)
7416 /* return the position 'off' UTF-8 characters away from 's', forward if
7417 * 'off' >= 0, backwards if negative. But don't go outside of position
7418 * 'lim', which better be < s if off < 0 */
7422 PERL_ARGS_ASSERT_REGHOP3;
7425 while (off-- && s < lim) {
7426 /* XXX could check well-formedness here */
7431 while (off++ && s > lim) {
7433 if (UTF8_IS_CONTINUED(*s)) {
7434 while (s > lim && UTF8_IS_CONTINUATION(*s))
7437 /* XXX could check well-formedness here */
7444 /* there are a bunch of places where we use two reghop3's that should
7445 be replaced with this routine. but since thats not done yet
7446 we ifdef it out - dmq
7449 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7453 PERL_ARGS_ASSERT_REGHOP4;
7456 while (off-- && s < rlim) {
7457 /* XXX could check well-formedness here */
7462 while (off++ && s > llim) {
7464 if (UTF8_IS_CONTINUED(*s)) {
7465 while (s > llim && UTF8_IS_CONTINUATION(*s))
7468 /* XXX could check well-formedness here */
7476 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7480 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7483 while (off-- && s < lim) {
7484 /* XXX could check well-formedness here */
7491 while (off++ && s > lim) {
7493 if (UTF8_IS_CONTINUED(*s)) {
7494 while (s > lim && UTF8_IS_CONTINUATION(*s))
7497 /* XXX could check well-formedness here */
7506 /* when executing a regex that may have (?{}), extra stuff needs setting
7507 up that will be visible to the called code, even before the current
7508 match has finished. In particular:
7510 * $_ is localised to the SV currently being matched;
7511 * pos($_) is created if necessary, ready to be updated on each call-out
7513 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
7514 isn't set until the current pattern is successfully finished), so that
7515 $1 etc of the match-so-far can be seen;
7516 * save the old values of subbeg etc of the current regex, and set then
7517 to the current string (again, this is normally only done at the end
7522 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
7525 regexp *const rex = ReANY(reginfo->prog);
7526 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
7528 eval_state->rex = rex;
7531 /* Make $_ available to executed code. */
7532 if (reginfo->sv != DEFSV) {
7534 DEFSV_set(reginfo->sv);
7537 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
7538 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
7539 /* prepare for quick setting of pos */
7540 #ifdef PERL_OLD_COPY_ON_WRITE
7541 if (SvIsCOW(reginfo->sv))
7542 sv_force_normal_flags(reginfo->sv, 0);
7544 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
7545 &PL_vtbl_mglob, NULL, 0);
7548 eval_state->pos_magic = mg;
7549 eval_state->pos = mg->mg_len;
7552 eval_state->pos_magic = NULL;
7554 if (!PL_reg_curpm) {
7555 /* PL_reg_curpm is a fake PMOP that we can attach the current
7556 * regex to and point PL_curpm at, so that $1 et al are visible
7557 * within a /(?{})/. It's just allocated once per interpreter the
7558 * first time its needed */
7559 Newxz(PL_reg_curpm, 1, PMOP);
7562 SV* const repointer = &PL_sv_undef;
7563 /* this regexp is also owned by the new PL_reg_curpm, which
7564 will try to free it. */
7565 av_push(PL_regex_padav, repointer);
7566 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
7567 PL_regex_pad = AvARRAY(PL_regex_padav);
7571 SET_reg_curpm(reginfo->prog);
7572 eval_state->curpm = PL_curpm;
7573 PL_curpm = PL_reg_curpm;
7574 if (RXp_MATCH_COPIED(rex)) {
7575 /* Here is a serious problem: we cannot rewrite subbeg,
7576 since it may be needed if this match fails. Thus
7577 $` inside (?{}) could fail... */
7578 eval_state->subbeg = rex->subbeg;
7579 eval_state->sublen = rex->sublen;
7580 eval_state->suboffset = rex->suboffset;
7581 eval_state->subcoffset = rex->subcoffset;
7583 eval_state->saved_copy = rex->saved_copy;
7585 RXp_MATCH_COPIED_off(rex);
7588 eval_state->subbeg = NULL;
7589 rex->subbeg = (char *)reginfo->strbeg;
7591 rex->subcoffset = 0;
7592 rex->sublen = reginfo->strend - reginfo->strbeg;
7596 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
7599 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
7602 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
7603 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
7606 Safefree(aux->poscache);
7610 /* undo the effects of S_setup_eval_state() */
7612 if (eval_state->subbeg) {
7613 regexp * const rex = eval_state->rex;
7614 rex->subbeg = eval_state->subbeg;
7615 rex->sublen = eval_state->sublen;
7616 rex->suboffset = eval_state->suboffset;
7617 rex->subcoffset = eval_state->subcoffset;
7619 rex->saved_copy = eval_state->saved_copy;
7621 RXp_MATCH_COPIED_on(rex);
7623 if (eval_state->pos_magic)
7624 eval_state->pos_magic->mg_len = eval_state->pos;
7626 PL_curpm = eval_state->curpm;
7629 PL_regmatch_state = aux->old_regmatch_state;
7630 PL_regmatch_slab = aux->old_regmatch_slab;
7632 /* free all slabs above current one - this must be the last action
7633 * of this function, as aux and eval_state are allocated within
7634 * slabs and may be freed here */
7636 s = PL_regmatch_slab->next;
7638 PL_regmatch_slab->next = NULL;
7640 regmatch_slab * const osl = s;
7649 S_to_utf8_substr(pTHX_ regexp *prog)
7651 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
7652 * on the converted value */
7656 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7659 if (prog->substrs->data[i].substr
7660 && !prog->substrs->data[i].utf8_substr) {
7661 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7662 prog->substrs->data[i].utf8_substr = sv;
7663 sv_utf8_upgrade(sv);
7664 if (SvVALID(prog->substrs->data[i].substr)) {
7665 if (SvTAIL(prog->substrs->data[i].substr)) {
7666 /* Trim the trailing \n that fbm_compile added last
7668 SvCUR_set(sv, SvCUR(sv) - 1);
7669 /* Whilst this makes the SV technically "invalid" (as its
7670 buffer is no longer followed by "\0") when fbm_compile()
7671 adds the "\n" back, a "\0" is restored. */
7672 fbm_compile(sv, FBMcf_TAIL);
7676 if (prog->substrs->data[i].substr == prog->check_substr)
7677 prog->check_utf8 = sv;
7683 S_to_byte_substr(pTHX_ regexp *prog)
7685 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
7686 * on the converted value; returns FALSE if can't be converted. */
7691 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7694 if (prog->substrs->data[i].utf8_substr
7695 && !prog->substrs->data[i].substr) {
7696 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7697 if (! sv_utf8_downgrade(sv, TRUE)) {
7700 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7701 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7702 /* Trim the trailing \n that fbm_compile added last
7704 SvCUR_set(sv, SvCUR(sv) - 1);
7705 fbm_compile(sv, FBMcf_TAIL);
7709 prog->substrs->data[i].substr = sv;
7710 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7711 prog->check_substr = sv;
7720 * c-indentation-style: bsd
7722 * indent-tabs-mode: nil
7725 * ex: set ts=8 sts=4 sw=4 et: