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 && (mg = mg_find_mglob(sv))
2199 && mg->mg_len >= 0) {
2200 reginfo->ganch = strbeg + mg->mg_len; /* Defined pos() */
2201 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2202 "GPOS MAGIC: reginfo->ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2204 if (prog->extflags & RXf_ANCH_GPOS) {
2205 if (s > reginfo->ganch)
2207 s = reginfo->ganch - prog->gofs;
2208 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2209 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2215 reginfo->ganch = strbeg + PTR2UV(data);
2216 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2217 "GPOS DATA: reginfo->ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2219 } else { /* pos() not defined */
2220 reginfo->ganch = strbeg;
2221 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2222 "GPOS: reginfo->ganch = strbeg\n"));
2225 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2226 /* We have to be careful. If the previous successful match
2227 was from this regex we don't want a subsequent partially
2228 successful match to clobber the old results.
2229 So when we detect this possibility we add a swap buffer
2230 to the re, and switch the buffer each match. If we fail,
2231 we switch it back; otherwise we leave it swapped.
2234 /* do we need a save destructor here for eval dies? */
2235 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2236 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2237 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2243 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2244 re_scream_pos_data d;
2246 d.scream_olds = &scream_olds;
2247 d.scream_pos = &scream_pos;
2248 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, &d);
2250 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2251 goto phooey; /* not present */
2257 /* Simplest case: anchored match need be tried only once. */
2258 /* [unless only anchor is BOL and multiline is set] */
2259 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2260 if (s == startpos && regtry(reginfo, &startpos))
2262 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2263 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2268 dontbother = minlen - 1;
2269 end = HOP3c(strend, -dontbother, strbeg) - 1;
2270 /* for multiline we only have to try after newlines */
2271 if (prog->check_substr || prog->check_utf8) {
2272 /* because of the goto we can not easily reuse the macros for bifurcating the
2273 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2276 goto after_try_utf8;
2278 if (regtry(reginfo, &s)) {
2285 if (prog->extflags & RXf_USE_INTUIT) {
2286 s = re_intuit_start(rx, sv, strbeg,
2287 s + UTF8SKIP(s), strend, flags, NULL);
2296 } /* end search for check string in unicode */
2298 if (s == startpos) {
2299 goto after_try_latin;
2302 if (regtry(reginfo, &s)) {
2309 if (prog->extflags & RXf_USE_INTUIT) {
2310 s = re_intuit_start(rx, sv, strbeg,
2311 s + 1, strend, flags, NULL);
2320 } /* end search for check string in latin*/
2321 } /* end search for check string */
2322 else { /* search for newline */
2324 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2327 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2328 while (s <= end) { /* note it could be possible to match at the end of the string */
2329 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2330 if (regtry(reginfo, &s))
2334 } /* end search for newline */
2335 } /* end anchored/multiline check string search */
2337 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2339 /* the warning about reginfo->ganch being used without initialization
2340 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2341 and we only enter this block when the same bit is set. */
2342 char *tmp_s = reginfo->ganch - prog->gofs;
2344 if (tmp_s >= strbeg && regtry(reginfo, &tmp_s))
2349 /* Messy cases: unanchored match. */
2350 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2351 /* we have /x+whatever/ */
2352 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2358 if (! prog->anchored_utf8) {
2359 to_utf8_substr(prog);
2361 ch = SvPVX_const(prog->anchored_utf8)[0];
2364 DEBUG_EXECUTE_r( did_match = 1 );
2365 if (regtry(reginfo, &s)) goto got_it;
2367 while (s < strend && *s == ch)
2374 if (! prog->anchored_substr) {
2375 if (! to_byte_substr(prog)) {
2376 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2379 ch = SvPVX_const(prog->anchored_substr)[0];
2382 DEBUG_EXECUTE_r( did_match = 1 );
2383 if (regtry(reginfo, &s)) goto got_it;
2385 while (s < strend && *s == ch)
2390 DEBUG_EXECUTE_r(if (!did_match)
2391 PerlIO_printf(Perl_debug_log,
2392 "Did not find anchored character...\n")
2395 else if (prog->anchored_substr != NULL
2396 || prog->anchored_utf8 != NULL
2397 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2398 && prog->float_max_offset < strend - s)) {
2403 char *last1; /* Last position checked before */
2407 if (prog->anchored_substr || prog->anchored_utf8) {
2409 if (! prog->anchored_utf8) {
2410 to_utf8_substr(prog);
2412 must = prog->anchored_utf8;
2415 if (! prog->anchored_substr) {
2416 if (! to_byte_substr(prog)) {
2417 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2420 must = prog->anchored_substr;
2422 back_max = back_min = prog->anchored_offset;
2425 if (! prog->float_utf8) {
2426 to_utf8_substr(prog);
2428 must = prog->float_utf8;
2431 if (! prog->float_substr) {
2432 if (! to_byte_substr(prog)) {
2433 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2436 must = prog->float_substr;
2438 back_max = prog->float_max_offset;
2439 back_min = prog->float_min_offset;
2445 last = HOP3c(strend, /* Cannot start after this */
2446 -(I32)(CHR_SVLEN(must)
2447 - (SvTAIL(must) != 0) + back_min), strbeg);
2449 if (s > reginfo->strbeg)
2450 last1 = HOPc(s, -1);
2452 last1 = s - 1; /* bogus */
2454 /* XXXX check_substr already used to find "s", can optimize if
2455 check_substr==must. */
2457 dontbother = end_shift;
2458 strend = HOPc(strend, -dontbother);
2459 while ( (s <= last) &&
2460 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2461 (unsigned char*)strend, must,
2462 multiline ? FBMrf_MULTILINE : 0)) ) {
2463 DEBUG_EXECUTE_r( did_match = 1 );
2464 if (HOPc(s, -back_max) > last1) {
2465 last1 = HOPc(s, -back_min);
2466 s = HOPc(s, -back_max);
2469 char * const t = (last1 >= reginfo->strbeg)
2470 ? HOPc(last1, 1) : last1 + 1;
2472 last1 = HOPc(s, -back_min);
2476 while (s <= last1) {
2477 if (regtry(reginfo, &s))
2480 s++; /* to break out of outer loop */
2487 while (s <= last1) {
2488 if (regtry(reginfo, &s))
2494 DEBUG_EXECUTE_r(if (!did_match) {
2495 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2496 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2497 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2498 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2499 ? "anchored" : "floating"),
2500 quoted, RE_SV_TAIL(must));
2504 else if ( (c = progi->regstclass) ) {
2506 const OPCODE op = OP(progi->regstclass);
2507 /* don't bother with what can't match */
2508 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2509 strend = HOPc(strend, -(minlen - 1));
2512 SV * const prop = sv_newmortal();
2513 regprop(prog, prop, c);
2515 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2517 PerlIO_printf(Perl_debug_log,
2518 "Matching stclass %.*s against %s (%d bytes)\n",
2519 (int)SvCUR(prop), SvPVX_const(prop),
2520 quoted, (int)(strend - s));
2523 if (find_byclass(prog, c, s, strend, reginfo))
2525 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2529 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2537 if (! prog->float_utf8) {
2538 to_utf8_substr(prog);
2540 float_real = prog->float_utf8;
2543 if (! prog->float_substr) {
2544 if (! to_byte_substr(prog)) {
2545 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2548 float_real = prog->float_substr;
2551 little = SvPV_const(float_real, len);
2552 if (SvTAIL(float_real)) {
2553 /* This means that float_real contains an artificial \n on
2554 * the end due to the presence of something like this:
2555 * /foo$/ where we can match both "foo" and "foo\n" at the
2556 * end of the string. So we have to compare the end of the
2557 * string first against the float_real without the \n and
2558 * then against the full float_real with the string. We
2559 * have to watch out for cases where the string might be
2560 * smaller than the float_real or the float_real without
2562 char *checkpos= strend - len;
2564 PerlIO_printf(Perl_debug_log,
2565 "%sChecking for float_real.%s\n",
2566 PL_colors[4], PL_colors[5]));
2567 if (checkpos + 1 < strbeg) {
2568 /* can't match, even if we remove the trailing \n
2569 * string is too short to match */
2571 PerlIO_printf(Perl_debug_log,
2572 "%sString shorter than required trailing substring, cannot match.%s\n",
2573 PL_colors[4], PL_colors[5]));
2575 } else if (memEQ(checkpos + 1, little, len - 1)) {
2576 /* can match, the end of the string matches without the
2578 last = checkpos + 1;
2579 } else if (checkpos < strbeg) {
2580 /* cant match, string is too short when the "\n" is
2583 PerlIO_printf(Perl_debug_log,
2584 "%sString does not contain required trailing substring, cannot match.%s\n",
2585 PL_colors[4], PL_colors[5]));
2587 } else if (!multiline) {
2588 /* non multiline match, so compare with the "\n" at the
2589 * end of the string */
2590 if (memEQ(checkpos, little, len)) {
2594 PerlIO_printf(Perl_debug_log,
2595 "%sString does not contain required trailing substring, cannot match.%s\n",
2596 PL_colors[4], PL_colors[5]));
2600 /* multiline match, so we have to search for a place
2601 * where the full string is located */
2607 last = rninstr(s, strend, little, little + len);
2609 last = strend; /* matching "$" */
2612 /* at one point this block contained a comment which was
2613 * probably incorrect, which said that this was a "should not
2614 * happen" case. Even if it was true when it was written I am
2615 * pretty sure it is not anymore, so I have removed the comment
2616 * and replaced it with this one. Yves */
2618 PerlIO_printf(Perl_debug_log,
2619 "String does not contain required substring, cannot match.\n"
2623 dontbother = strend - last + prog->float_min_offset;
2625 if (minlen && (dontbother < minlen))
2626 dontbother = minlen - 1;
2627 strend -= dontbother; /* this one's always in bytes! */
2628 /* We don't know much -- general case. */
2631 if (regtry(reginfo, &s))
2640 if (regtry(reginfo, &s))
2642 } while (s++ < strend);
2652 PerlIO_printf(Perl_debug_log,
2653 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2660 /* clean up; this will trigger destructors that will free all slabs
2661 * above the current one, and cleanup the regmatch_info_aux
2662 * and regmatch_info_aux_eval sructs */
2664 LEAVE_SCOPE(oldsave);
2666 if (RXp_PAREN_NAMES(prog))
2667 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2669 RX_MATCH_UTF8_set(rx, utf8_target);
2671 /* make sure $`, $&, $', and $digit will work later */
2672 if ( !(flags & REXEC_NOT_FIRST) ) {
2673 if (flags & REXEC_COPY_STR) {
2677 PerlIO_printf(Perl_debug_log,
2678 "Copy on write: regexp capture, type %d\n",
2681 RX_MATCH_COPY_FREE(rx);
2682 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2683 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2684 assert (SvPOKp(prog->saved_copy));
2685 prog->sublen = reginfo->strend - strbeg;
2686 prog->suboffset = 0;
2687 prog->subcoffset = 0;
2692 I32 max = reginfo->strend - strbeg;
2695 if ( (flags & REXEC_COPY_SKIP_POST)
2696 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2697 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2698 ) { /* don't copy $' part of string */
2701 /* calculate the right-most part of the string covered
2702 * by a capture. Due to look-ahead, this may be to
2703 * the right of $&, so we have to scan all captures */
2704 while (n <= prog->lastparen) {
2705 if (prog->offs[n].end > max)
2706 max = prog->offs[n].end;
2710 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2711 ? prog->offs[0].start
2713 assert(max >= 0 && max <= reginfo->strend - strbeg);
2716 if ( (flags & REXEC_COPY_SKIP_PRE)
2717 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2718 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2719 ) { /* don't copy $` part of string */
2722 /* calculate the left-most part of the string covered
2723 * by a capture. Due to look-behind, this may be to
2724 * the left of $&, so we have to scan all captures */
2725 while (min && n <= prog->lastparen) {
2726 if ( prog->offs[n].start != -1
2727 && prog->offs[n].start < min)
2729 min = prog->offs[n].start;
2733 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2734 && min > prog->offs[0].end
2736 min = prog->offs[0].end;
2740 assert(min >= 0 && min <= max
2741 && min <= reginfo->strend - strbeg);
2744 if (RX_MATCH_COPIED(rx)) {
2745 if (sublen > prog->sublen)
2747 (char*)saferealloc(prog->subbeg, sublen+1);
2750 prog->subbeg = (char*)safemalloc(sublen+1);
2751 Copy(strbeg + min, prog->subbeg, sublen, char);
2752 prog->subbeg[sublen] = '\0';
2753 prog->suboffset = min;
2754 prog->sublen = sublen;
2755 RX_MATCH_COPIED_on(rx);
2757 prog->subcoffset = prog->suboffset;
2758 if (prog->suboffset && utf8_target) {
2759 /* Convert byte offset to chars.
2760 * XXX ideally should only compute this if @-/@+
2761 * has been seen, a la PL_sawampersand ??? */
2763 /* If there's a direct correspondence between the
2764 * string which we're matching and the original SV,
2765 * then we can use the utf8 len cache associated with
2766 * the SV. In particular, it means that under //g,
2767 * sv_pos_b2u() will use the previously cached
2768 * position to speed up working out the new length of
2769 * subcoffset, rather than counting from the start of
2770 * the string each time. This stops
2771 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2772 * from going quadratic */
2773 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2774 sv_pos_b2u(sv, &(prog->subcoffset));
2776 prog->subcoffset = utf8_length((U8*)strbeg,
2777 (U8*)(strbeg+prog->suboffset));
2781 RX_MATCH_COPY_FREE(rx);
2782 prog->subbeg = strbeg;
2783 prog->suboffset = 0;
2784 prog->subcoffset = 0;
2785 /* use reginfo->strend, as strend may have been modified */
2786 prog->sublen = reginfo->strend - strbeg;
2793 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2794 PL_colors[4], PL_colors[5]));
2796 /* clean up; this will trigger destructors that will free all slabs
2797 * above the current one, and cleanup the regmatch_info_aux
2798 * and regmatch_info_aux_eval sructs */
2800 LEAVE_SCOPE(oldsave);
2803 /* we failed :-( roll it back */
2804 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2805 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2810 Safefree(prog->offs);
2817 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
2818 * Do inc before dec, in case old and new rex are the same */
2819 #define SET_reg_curpm(Re2) \
2820 if (reginfo->info_aux_eval) { \
2821 (void)ReREFCNT_inc(Re2); \
2822 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2823 PM_SETRE((PL_reg_curpm), (Re2)); \
2828 - regtry - try match at specific point
2830 STATIC I32 /* 0 failure, 1 success */
2831 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2835 REGEXP *const rx = reginfo->prog;
2836 regexp *const prog = ReANY(rx);
2838 RXi_GET_DECL(prog,progi);
2839 GET_RE_DEBUG_FLAGS_DECL;
2841 PERL_ARGS_ASSERT_REGTRY;
2843 reginfo->cutpoint=NULL;
2845 prog->offs[0].start = *startposp - reginfo->strbeg;
2846 prog->lastparen = 0;
2847 prog->lastcloseparen = 0;
2849 /* XXXX What this code is doing here?!!! There should be no need
2850 to do this again and again, prog->lastparen should take care of
2853 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2854 * Actually, the code in regcppop() (which Ilya may be meaning by
2855 * prog->lastparen), is not needed at all by the test suite
2856 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2857 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2858 * Meanwhile, this code *is* needed for the
2859 * above-mentioned test suite tests to succeed. The common theme
2860 * on those tests seems to be returning null fields from matches.
2861 * --jhi updated by dapm */
2863 if (prog->nparens) {
2864 regexp_paren_pair *pp = prog->offs;
2866 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2874 result = regmatch(reginfo, *startposp, progi->program + 1);
2876 prog->offs[0].end = result;
2879 if (reginfo->cutpoint)
2880 *startposp= reginfo->cutpoint;
2881 REGCP_UNWIND(lastcp);
2886 #define sayYES goto yes
2887 #define sayNO goto no
2888 #define sayNO_SILENT goto no_silent
2890 /* we dont use STMT_START/END here because it leads to
2891 "unreachable code" warnings, which are bogus, but distracting. */
2892 #define CACHEsayNO \
2893 if (ST.cache_mask) \
2894 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
2897 /* this is used to determine how far from the left messages like
2898 'failed...' are printed. It should be set such that messages
2899 are inline with the regop output that created them.
2901 #define REPORT_CODE_OFF 32
2904 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2905 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2906 #define CHRTEST_NOT_A_CP_1 -999
2907 #define CHRTEST_NOT_A_CP_2 -998
2909 /* grab a new slab and return the first slot in it */
2911 STATIC regmatch_state *
2914 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2917 regmatch_slab *s = PL_regmatch_slab->next;
2919 Newx(s, 1, regmatch_slab);
2920 s->prev = PL_regmatch_slab;
2922 PL_regmatch_slab->next = s;
2924 PL_regmatch_slab = s;
2925 return SLAB_FIRST(s);
2929 /* push a new state then goto it */
2931 #define PUSH_STATE_GOTO(state, node, input) \
2932 pushinput = input; \
2934 st->resume_state = state; \
2937 /* push a new state with success backtracking, then goto it */
2939 #define PUSH_YES_STATE_GOTO(state, node, input) \
2940 pushinput = input; \
2942 st->resume_state = state; \
2943 goto push_yes_state;
2950 regmatch() - main matching routine
2952 This is basically one big switch statement in a loop. We execute an op,
2953 set 'next' to point the next op, and continue. If we come to a point which
2954 we may need to backtrack to on failure such as (A|B|C), we push a
2955 backtrack state onto the backtrack stack. On failure, we pop the top
2956 state, and re-enter the loop at the state indicated. If there are no more
2957 states to pop, we return failure.
2959 Sometimes we also need to backtrack on success; for example /A+/, where
2960 after successfully matching one A, we need to go back and try to
2961 match another one; similarly for lookahead assertions: if the assertion
2962 completes successfully, we backtrack to the state just before the assertion
2963 and then carry on. In these cases, the pushed state is marked as
2964 'backtrack on success too'. This marking is in fact done by a chain of
2965 pointers, each pointing to the previous 'yes' state. On success, we pop to
2966 the nearest yes state, discarding any intermediate failure-only states.
2967 Sometimes a yes state is pushed just to force some cleanup code to be
2968 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2969 it to free the inner regex.
2971 Note that failure backtracking rewinds the cursor position, while
2972 success backtracking leaves it alone.
2974 A pattern is complete when the END op is executed, while a subpattern
2975 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2976 ops trigger the "pop to last yes state if any, otherwise return true"
2979 A common convention in this function is to use A and B to refer to the two
2980 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2981 the subpattern to be matched possibly multiple times, while B is the entire
2982 rest of the pattern. Variable and state names reflect this convention.
2984 The states in the main switch are the union of ops and failure/success of
2985 substates associated with with that op. For example, IFMATCH is the op
2986 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2987 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2988 successfully matched A and IFMATCH_A_fail is a state saying that we have
2989 just failed to match A. Resume states always come in pairs. The backtrack
2990 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2991 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2992 on success or failure.
2994 The struct that holds a backtracking state is actually a big union, with
2995 one variant for each major type of op. The variable st points to the
2996 top-most backtrack struct. To make the code clearer, within each
2997 block of code we #define ST to alias the relevant union.
2999 Here's a concrete example of a (vastly oversimplified) IFMATCH
3005 #define ST st->u.ifmatch
3007 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3008 ST.foo = ...; // some state we wish to save
3010 // push a yes backtrack state with a resume value of
3011 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3013 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3016 case IFMATCH_A: // we have successfully executed A; now continue with B
3018 bar = ST.foo; // do something with the preserved value
3021 case IFMATCH_A_fail: // A failed, so the assertion failed
3022 ...; // do some housekeeping, then ...
3023 sayNO; // propagate the failure
3030 For any old-timers reading this who are familiar with the old recursive
3031 approach, the code above is equivalent to:
3033 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3042 ...; // do some housekeeping, then ...
3043 sayNO; // propagate the failure
3046 The topmost backtrack state, pointed to by st, is usually free. If you
3047 want to claim it, populate any ST.foo fields in it with values you wish to
3048 save, then do one of
3050 PUSH_STATE_GOTO(resume_state, node, newinput);
3051 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3053 which sets that backtrack state's resume value to 'resume_state', pushes a
3054 new free entry to the top of the backtrack stack, then goes to 'node'.
3055 On backtracking, the free slot is popped, and the saved state becomes the
3056 new free state. An ST.foo field in this new top state can be temporarily
3057 accessed to retrieve values, but once the main loop is re-entered, it
3058 becomes available for reuse.
3060 Note that the depth of the backtrack stack constantly increases during the
3061 left-to-right execution of the pattern, rather than going up and down with
3062 the pattern nesting. For example the stack is at its maximum at Z at the
3063 end of the pattern, rather than at X in the following:
3065 /(((X)+)+)+....(Y)+....Z/
3067 The only exceptions to this are lookahead/behind assertions and the cut,
3068 (?>A), which pop all the backtrack states associated with A before
3071 Backtrack state structs are allocated in slabs of about 4K in size.
3072 PL_regmatch_state and st always point to the currently active state,
3073 and PL_regmatch_slab points to the slab currently containing
3074 PL_regmatch_state. The first time regmatch() is called, the first slab is
3075 allocated, and is never freed until interpreter destruction. When the slab
3076 is full, a new one is allocated and chained to the end. At exit from
3077 regmatch(), slabs allocated since entry are freed.
3082 #define DEBUG_STATE_pp(pp) \
3084 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3085 PerlIO_printf(Perl_debug_log, \
3086 " %*s"pp" %s%s%s%s%s\n", \
3088 PL_reg_name[st->resume_state], \
3089 ((st==yes_state||st==mark_state) ? "[" : ""), \
3090 ((st==yes_state) ? "Y" : ""), \
3091 ((st==mark_state) ? "M" : ""), \
3092 ((st==yes_state||st==mark_state) ? "]" : "") \
3097 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3102 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3103 const char *start, const char *end, const char *blurb)
3105 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3107 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3112 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3113 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3115 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3116 start, end - start, 60);
3118 PerlIO_printf(Perl_debug_log,
3119 "%s%s REx%s %s against %s\n",
3120 PL_colors[4], blurb, PL_colors[5], s0, s1);
3122 if (utf8_target||utf8_pat)
3123 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3124 utf8_pat ? "pattern" : "",
3125 utf8_pat && utf8_target ? " and " : "",
3126 utf8_target ? "string" : ""
3132 S_dump_exec_pos(pTHX_ const char *locinput,
3133 const regnode *scan,
3134 const char *loc_regeol,
3135 const char *loc_bostr,
3136 const char *loc_reg_starttry,
3137 const bool utf8_target)
3139 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3140 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3141 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3142 /* The part of the string before starttry has one color
3143 (pref0_len chars), between starttry and current
3144 position another one (pref_len - pref0_len chars),
3145 after the current position the third one.
3146 We assume that pref0_len <= pref_len, otherwise we
3147 decrease pref0_len. */
3148 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3149 ? (5 + taill) - l : locinput - loc_bostr;
3152 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3154 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3156 pref0_len = pref_len - (locinput - loc_reg_starttry);
3157 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3158 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3159 ? (5 + taill) - pref_len : loc_regeol - locinput);
3160 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3164 if (pref0_len > pref_len)
3165 pref0_len = pref_len;
3167 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3169 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3170 (locinput - pref_len),pref0_len, 60, 4, 5);
3172 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3173 (locinput - pref_len + pref0_len),
3174 pref_len - pref0_len, 60, 2, 3);
3176 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3177 locinput, loc_regeol - locinput, 10, 0, 1);
3179 const STRLEN tlen=len0+len1+len2;
3180 PerlIO_printf(Perl_debug_log,
3181 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3182 (IV)(locinput - loc_bostr),
3185 (docolor ? "" : "> <"),
3187 (int)(tlen > 19 ? 0 : 19 - tlen),
3194 /* reg_check_named_buff_matched()
3195 * Checks to see if a named buffer has matched. The data array of
3196 * buffer numbers corresponding to the buffer is expected to reside
3197 * in the regexp->data->data array in the slot stored in the ARG() of
3198 * node involved. Note that this routine doesn't actually care about the
3199 * name, that information is not preserved from compilation to execution.
3200 * Returns the index of the leftmost defined buffer with the given name
3201 * or 0 if non of the buffers matched.
3204 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3207 RXi_GET_DECL(rex,rexi);
3208 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3209 I32 *nums=(I32*)SvPVX(sv_dat);
3211 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3213 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3214 if ((I32)rex->lastparen >= nums[n] &&
3215 rex->offs[nums[n]].end != -1)
3225 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3226 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3228 /* This function determines if there are one or two characters that match
3229 * the first character of the passed-in EXACTish node <text_node>, and if
3230 * so, returns them in the passed-in pointers.
3232 * If it determines that no possible character in the target string can
3233 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3234 * the first character in <text_node> requires UTF-8 to represent, and the
3235 * target string isn't in UTF-8.)
3237 * If there are more than two characters that could match the beginning of
3238 * <text_node>, or if more context is required to determine a match or not,
3239 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3241 * The motiviation behind this function is to allow the caller to set up
3242 * tight loops for matching. If <text_node> is of type EXACT, there is
3243 * only one possible character that can match its first character, and so
3244 * the situation is quite simple. But things get much more complicated if
3245 * folding is involved. It may be that the first character of an EXACTFish
3246 * node doesn't participate in any possible fold, e.g., punctuation, so it
3247 * can be matched only by itself. The vast majority of characters that are
3248 * in folds match just two things, their lower and upper-case equivalents.
3249 * But not all are like that; some have multiple possible matches, or match
3250 * sequences of more than one character. This function sorts all that out.
3252 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3253 * loop of trying to match A*, we know we can't exit where the thing
3254 * following it isn't a B. And something can't be a B unless it is the
3255 * beginning of B. By putting a quick test for that beginning in a tight
3256 * loop, we can rule out things that can't possibly be B without having to
3257 * break out of the loop, thus avoiding work. Similarly, if A is a single
3258 * character, we can make a tight loop matching A*, using the outputs of
3261 * If the target string to match isn't in UTF-8, and there aren't
3262 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3263 * the one or two possible octets (which are characters in this situation)
3264 * that can match. In all cases, if there is only one character that can
3265 * match, *<c1p> and *<c2p> will be identical.
3267 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3268 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3269 * can match the beginning of <text_node>. They should be declared with at
3270 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3271 * undefined what these contain.) If one or both of the buffers are
3272 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3273 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3274 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3275 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3276 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3278 const bool utf8_target = reginfo->is_utf8_target;
3280 UV c1 = CHRTEST_NOT_A_CP_1;
3281 UV c2 = CHRTEST_NOT_A_CP_2;
3282 bool use_chrtest_void = FALSE;
3283 const bool is_utf8_pat = reginfo->is_utf8_pat;
3285 /* Used when we have both utf8 input and utf8 output, to avoid converting
3286 * to/from code points */
3287 bool utf8_has_been_setup = FALSE;
3291 U8 *pat = (U8*)STRING(text_node);
3293 if (OP(text_node) == EXACT) {
3295 /* In an exact node, only one thing can be matched, that first
3296 * character. If both the pat and the target are UTF-8, we can just
3297 * copy the input to the output, avoiding finding the code point of
3302 else if (utf8_target) {
3303 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3304 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3305 utf8_has_been_setup = TRUE;
3308 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3311 else /* an EXACTFish node */
3313 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3314 pat + STR_LEN(text_node)))
3316 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3317 pat + STR_LEN(text_node))))
3319 /* Multi-character folds require more context to sort out. Also
3320 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3321 * handled outside this routine */
3322 use_chrtest_void = TRUE;
3324 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3325 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3327 /* Load the folds hash, if not already done */
3329 if (! PL_utf8_foldclosures) {
3330 if (! PL_utf8_tofold) {
3331 U8 dummy[UTF8_MAXBYTES+1];
3333 /* Force loading this by folding an above-Latin1 char */
3334 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3335 assert(PL_utf8_tofold); /* Verify that worked */
3337 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3340 /* The fold closures data structure is a hash with the keys being
3341 * the UTF-8 of every character that is folded to, like 'k', and
3342 * the values each an array of all code points that fold to its
3343 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3345 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3350 /* Not found in the hash, therefore there are no folds
3351 * containing it, so there is only a single character that
3355 else { /* Does participate in folds */
3356 AV* list = (AV*) *listp;
3357 if (av_len(list) != 1) {
3359 /* If there aren't exactly two folds to this, it is outside
3360 * the scope of this function */
3361 use_chrtest_void = TRUE;
3363 else { /* There are two. Get them */
3364 SV** c_p = av_fetch(list, 0, FALSE);
3366 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3370 c_p = av_fetch(list, 1, FALSE);
3372 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3376 /* Folds that cross the 255/256 boundary are forbidden if
3377 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3378 * pattern character is above 256, and its only other match
3379 * is below 256, the only legal match will be to itself.
3380 * We have thrown away the original, so have to compute
3381 * which is the one above 255 */
3382 if ((c1 < 256) != (c2 < 256)) {
3383 if (OP(text_node) == EXACTFL
3384 || (OP(text_node) == EXACTFA
3385 && (isASCII(c1) || isASCII(c2))))
3398 else /* Here, c1 is < 255 */
3400 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3401 && OP(text_node) != EXACTFL
3402 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3404 /* Here, there could be something above Latin1 in the target which
3405 * folds to this character in the pattern. All such cases except
3406 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3407 * involved in their folds, so are outside the scope of this
3409 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3410 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3413 use_chrtest_void = TRUE;
3416 else { /* Here nothing above Latin1 can fold to the pattern character */
3417 switch (OP(text_node)) {
3419 case EXACTFL: /* /l rules */
3420 c2 = PL_fold_locale[c1];
3424 if (! utf8_target) { /* /d rules */
3429 /* /u rules for all these. This happens to work for
3430 * EXACTFA as nothing in Latin1 folds to ASCII */
3432 case EXACTFU_TRICKYFOLD:
3435 c2 = PL_fold_latin1[c1];
3439 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3440 assert(0); /* NOTREACHED */
3445 /* Here have figured things out. Set up the returns */
3446 if (use_chrtest_void) {
3447 *c2p = *c1p = CHRTEST_VOID;
3449 else if (utf8_target) {
3450 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3451 uvchr_to_utf8(c1_utf8, c1);
3452 uvchr_to_utf8(c2_utf8, c2);
3455 /* Invariants are stored in both the utf8 and byte outputs; Use
3456 * negative numbers otherwise for the byte ones. Make sure that the
3457 * byte ones are the same iff the utf8 ones are the same */
3458 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3459 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3462 ? CHRTEST_NOT_A_CP_1
3463 : CHRTEST_NOT_A_CP_2;
3465 else if (c1 > 255) {
3466 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3471 *c1p = *c2p = c2; /* c2 is the only representable value */
3473 else { /* c1 is representable; see about c2 */
3475 *c2p = (c2 < 256) ? c2 : c1;
3481 /* returns -1 on failure, $+[0] on success */
3483 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3485 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3489 const bool utf8_target = reginfo->is_utf8_target;
3490 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3491 REGEXP *rex_sv = reginfo->prog;
3492 regexp *rex = ReANY(rex_sv);
3493 RXi_GET_DECL(rex,rexi);
3494 /* the current state. This is a cached copy of PL_regmatch_state */
3496 /* cache heavy used fields of st in registers */
3499 U32 n = 0; /* general value; init to avoid compiler warning */
3500 I32 ln = 0; /* len or last; init to avoid compiler warning */
3501 char *locinput = startpos;
3502 char *pushinput; /* where to continue after a PUSH */
3503 I32 nextchr; /* is always set to UCHARAT(locinput) */
3505 bool result = 0; /* return value of S_regmatch */
3506 int depth = 0; /* depth of backtrack stack */
3507 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3508 const U32 max_nochange_depth =
3509 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3510 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3511 regmatch_state *yes_state = NULL; /* state to pop to on success of
3513 /* mark_state piggy backs on the yes_state logic so that when we unwind
3514 the stack on success we can update the mark_state as we go */
3515 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3516 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3517 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3519 bool no_final = 0; /* prevent failure from backtracking? */
3520 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3521 char *startpoint = locinput;
3522 SV *popmark = NULL; /* are we looking for a mark? */
3523 SV *sv_commit = NULL; /* last mark name seen in failure */
3524 SV *sv_yes_mark = NULL; /* last mark name we have seen
3525 during a successful match */
3526 U32 lastopen = 0; /* last open we saw */
3527 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3528 SV* const oreplsv = GvSV(PL_replgv);
3529 /* these three flags are set by various ops to signal information to
3530 * the very next op. They have a useful lifetime of exactly one loop
3531 * iteration, and are not preserved or restored by state pushes/pops
3533 bool sw = 0; /* the condition value in (?(cond)a|b) */
3534 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3535 int logical = 0; /* the following EVAL is:
3539 or the following IFMATCH/UNLESSM is:
3540 false: plain (?=foo)
3541 true: used as a condition: (?(?=foo))
3543 PAD* last_pad = NULL;
3545 I32 gimme = G_SCALAR;
3546 CV *caller_cv = NULL; /* who called us */
3547 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3548 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3549 U32 maxopenparen = 0; /* max '(' index seen so far */
3550 int to_complement; /* Invert the result? */
3551 _char_class_number classnum;
3552 bool is_utf8_pat = reginfo->is_utf8_pat;
3555 GET_RE_DEBUG_FLAGS_DECL;
3558 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3559 multicall_oldcatch = 0;
3560 multicall_cv = NULL;
3562 PERL_UNUSED_VAR(multicall_cop);
3563 PERL_UNUSED_VAR(newsp);
3566 PERL_ARGS_ASSERT_REGMATCH;
3568 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3569 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3572 st = PL_regmatch_state;
3574 /* Note that nextchr is a byte even in UTF */
3577 while (scan != NULL) {
3580 SV * const prop = sv_newmortal();
3581 regnode *rnext=regnext(scan);
3582 DUMP_EXEC_POS( locinput, scan, utf8_target );
3583 regprop(rex, prop, scan);
3585 PerlIO_printf(Perl_debug_log,
3586 "%3"IVdf":%*s%s(%"IVdf")\n",
3587 (IV)(scan - rexi->program), depth*2, "",
3589 (PL_regkind[OP(scan)] == END || !rnext) ?
3590 0 : (IV)(rnext - rexi->program));
3593 next = scan + NEXT_OFF(scan);
3596 state_num = OP(scan);
3598 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3603 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3605 switch (state_num) {
3606 case BOL: /* /^../ */
3607 if (locinput == reginfo->strbeg)
3611 case MBOL: /* /^../m */
3612 if (locinput == reginfo->strbeg ||
3613 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3619 case SBOL: /* /^../s */
3620 if (locinput == reginfo->strbeg)
3625 if (locinput == reginfo->ganch)
3629 case KEEPS: /* \K */
3630 /* update the startpoint */
3631 st->u.keeper.val = rex->offs[0].start;
3632 rex->offs[0].start = locinput - reginfo->strbeg;
3633 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3634 assert(0); /*NOTREACHED*/
3635 case KEEPS_next_fail:
3636 /* rollback the start point change */
3637 rex->offs[0].start = st->u.keeper.val;
3639 assert(0); /*NOTREACHED*/
3641 case EOL: /* /..$/ */
3644 case MEOL: /* /..$/m */
3645 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3649 case SEOL: /* /..$/s */
3651 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3653 if (reginfo->strend - locinput > 1)
3658 if (!NEXTCHR_IS_EOS)
3662 case SANY: /* /./s */
3665 goto increment_locinput;
3673 case REG_ANY: /* /./ */
3674 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3676 goto increment_locinput;
3680 #define ST st->u.trie
3681 case TRIEC: /* (ab|cd) with known charclass */
3682 /* In this case the charclass data is available inline so
3683 we can fail fast without a lot of extra overhead.
3685 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3687 PerlIO_printf(Perl_debug_log,
3688 "%*s %sfailed to match trie start class...%s\n",
3689 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3692 assert(0); /* NOTREACHED */
3695 case TRIE: /* (ab|cd) */
3696 /* the basic plan of execution of the trie is:
3697 * At the beginning, run though all the states, and
3698 * find the longest-matching word. Also remember the position
3699 * of the shortest matching word. For example, this pattern:
3702 * when matched against the string "abcde", will generate
3703 * accept states for all words except 3, with the longest
3704 * matching word being 4, and the shortest being 2 (with
3705 * the position being after char 1 of the string).
3707 * Then for each matching word, in word order (i.e. 1,2,4,5),
3708 * we run the remainder of the pattern; on each try setting
3709 * the current position to the character following the word,
3710 * returning to try the next word on failure.
3712 * We avoid having to build a list of words at runtime by
3713 * using a compile-time structure, wordinfo[].prev, which
3714 * gives, for each word, the previous accepting word (if any).
3715 * In the case above it would contain the mappings 1->2, 2->0,
3716 * 3->0, 4->5, 5->1. We can use this table to generate, from
3717 * the longest word (4 above), a list of all words, by
3718 * following the list of prev pointers; this gives us the
3719 * unordered list 4,5,1,2. Then given the current word we have
3720 * just tried, we can go through the list and find the
3721 * next-biggest word to try (so if we just failed on word 2,
3722 * the next in the list is 4).
3724 * Since at runtime we don't record the matching position in
3725 * the string for each word, we have to work that out for
3726 * each word we're about to process. The wordinfo table holds
3727 * the character length of each word; given that we recorded
3728 * at the start: the position of the shortest word and its
3729 * length in chars, we just need to move the pointer the
3730 * difference between the two char lengths. Depending on
3731 * Unicode status and folding, that's cheap or expensive.
3733 * This algorithm is optimised for the case where are only a
3734 * small number of accept states, i.e. 0,1, or maybe 2.
3735 * With lots of accepts states, and having to try all of them,
3736 * it becomes quadratic on number of accept states to find all
3741 /* what type of TRIE am I? (utf8 makes this contextual) */
3742 DECL_TRIE_TYPE(scan);
3744 /* what trie are we using right now */
3745 reg_trie_data * const trie
3746 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3747 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3748 U32 state = trie->startstate;
3751 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3753 if (trie->states[ state ].wordnum) {
3755 PerlIO_printf(Perl_debug_log,
3756 "%*s %smatched empty string...%s\n",
3757 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3763 PerlIO_printf(Perl_debug_log,
3764 "%*s %sfailed to match trie start class...%s\n",
3765 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3772 U8 *uc = ( U8* )locinput;
3776 U8 *uscan = (U8*)NULL;
3777 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3778 U32 charcount = 0; /* how many input chars we have matched */
3779 U32 accepted = 0; /* have we seen any accepting states? */
3781 ST.jump = trie->jump;
3784 ST.longfold = FALSE; /* char longer if folded => it's harder */
3787 /* fully traverse the TRIE; note the position of the
3788 shortest accept state and the wordnum of the longest
3791 while ( state && uc <= (U8*)(reginfo->strend) ) {
3792 U32 base = trie->states[ state ].trans.base;
3796 wordnum = trie->states[ state ].wordnum;
3798 if (wordnum) { /* it's an accept state */
3801 /* record first match position */
3803 ST.firstpos = (U8*)locinput;
3808 ST.firstchars = charcount;
3811 if (!ST.nextword || wordnum < ST.nextword)
3812 ST.nextword = wordnum;
3813 ST.topword = wordnum;
3816 DEBUG_TRIE_EXECUTE_r({
3817 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3818 PerlIO_printf( Perl_debug_log,
3819 "%*s %sState: %4"UVxf" Accepted: %c ",
3820 2+depth * 2, "", PL_colors[4],
3821 (UV)state, (accepted ? 'Y' : 'N'));
3824 /* read a char and goto next state */
3825 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
3827 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3828 uscan, len, uvc, charid, foldlen,
3835 base + charid - 1 - trie->uniquecharcount)) >= 0)
3837 && ((U32)offset < trie->lasttrans)
3838 && trie->trans[offset].check == state)
3840 state = trie->trans[offset].next;
3851 DEBUG_TRIE_EXECUTE_r(
3852 PerlIO_printf( Perl_debug_log,
3853 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3854 charid, uvc, (UV)state, PL_colors[5] );
3860 /* calculate total number of accept states */
3865 w = trie->wordinfo[w].prev;
3868 ST.accepted = accepted;
3872 PerlIO_printf( Perl_debug_log,
3873 "%*s %sgot %"IVdf" possible matches%s\n",
3874 REPORT_CODE_OFF + depth * 2, "",
3875 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3877 goto trie_first_try; /* jump into the fail handler */
3879 assert(0); /* NOTREACHED */
3881 case TRIE_next_fail: /* we failed - try next alternative */
3885 REGCP_UNWIND(ST.cp);
3886 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3888 if (!--ST.accepted) {
3890 PerlIO_printf( Perl_debug_log,
3891 "%*s %sTRIE failed...%s\n",
3892 REPORT_CODE_OFF+depth*2, "",
3899 /* Find next-highest word to process. Note that this code
3900 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3903 U16 const nextword = ST.nextword;
3904 reg_trie_wordinfo * const wordinfo
3905 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3906 for (word=ST.topword; word; word=wordinfo[word].prev) {
3907 if (word > nextword && (!min || word < min))
3920 ST.lastparen = rex->lastparen;
3921 ST.lastcloseparen = rex->lastcloseparen;
3925 /* find start char of end of current word */
3927 U32 chars; /* how many chars to skip */
3928 reg_trie_data * const trie
3929 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3931 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3933 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3938 /* the hard option - fold each char in turn and find
3939 * its folded length (which may be different */
3940 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3948 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3956 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3961 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3977 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
3978 ? ST.jump[ST.nextword]
3982 PerlIO_printf( Perl_debug_log,
3983 "%*s %sTRIE matched word #%d, continuing%s\n",
3984 REPORT_CODE_OFF+depth*2, "",
3991 if (ST.accepted > 1 || has_cutgroup) {
3992 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
3993 assert(0); /* NOTREACHED */
3995 /* only one choice left - just continue */
3997 AV *const trie_words
3998 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3999 SV ** const tmp = av_fetch( trie_words,
4001 SV *sv= tmp ? sv_newmortal() : NULL;
4003 PerlIO_printf( Perl_debug_log,
4004 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4005 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4007 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4008 PL_colors[0], PL_colors[1],
4009 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4011 : "not compiled under -Dr",
4015 locinput = (char*)uc;
4016 continue; /* execute rest of RE */
4017 assert(0); /* NOTREACHED */
4021 case EXACT: { /* /abc/ */
4022 char *s = STRING(scan);
4024 if (utf8_target != is_utf8_pat) {
4025 /* The target and the pattern have differing utf8ness. */
4027 const char * const e = s + ln;
4030 /* The target is utf8, the pattern is not utf8.
4031 * Above-Latin1 code points can't match the pattern;
4032 * invariants match exactly, and the other Latin1 ones need
4033 * to be downgraded to a single byte in order to do the
4034 * comparison. (If we could be confident that the target
4035 * is not malformed, this could be refactored to have fewer
4036 * tests by just assuming that if the first bytes match, it
4037 * is an invariant, but there are tests in the test suite
4038 * dealing with (??{...}) which violate this) */
4040 if (l >= reginfo->strend
4041 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4045 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4052 if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
4061 /* The target is not utf8, the pattern is utf8. */
4063 if (l >= reginfo->strend
4064 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4068 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4075 if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
4086 /* The target and the pattern have the same utf8ness. */
4087 /* Inline the first character, for speed. */
4088 if (reginfo->strend - locinput < ln
4089 || UCHARAT(s) != nextchr
4090 || (ln > 1 && memNE(s, locinput, ln)))
4099 case EXACTFL: { /* /abc/il */
4101 const U8 * fold_array;
4103 U32 fold_utf8_flags;
4105 RX_MATCH_TAINTED_on(reginfo->prog);
4106 folder = foldEQ_locale;
4107 fold_array = PL_fold_locale;
4108 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
4111 case EXACTFU_SS: /* /\x{df}/iu */
4112 case EXACTFU_TRICKYFOLD: /* /\x{390}/iu */
4113 case EXACTFU: /* /abc/iu */
4114 folder = foldEQ_latin1;
4115 fold_array = PL_fold_latin1;
4116 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4119 case EXACTFA: /* /abc/iaa */
4120 folder = foldEQ_latin1;
4121 fold_array = PL_fold_latin1;
4122 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4125 case EXACTF: /* /abc/i */
4127 fold_array = PL_fold;
4128 fold_utf8_flags = 0;
4134 if (utf8_target || is_utf8_pat || state_num == EXACTFU_SS) {
4135 /* Either target or the pattern are utf8, or has the issue where
4136 * the fold lengths may differ. */
4137 const char * const l = locinput;
4138 char *e = reginfo->strend;
4140 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4141 l, &e, 0, utf8_target, fold_utf8_flags))
4149 /* Neither the target nor the pattern are utf8 */
4150 if (UCHARAT(s) != nextchr
4152 && UCHARAT(s) != fold_array[nextchr])
4156 if (reginfo->strend - locinput < ln)
4158 if (ln > 1 && ! folder(s, locinput, ln))
4164 /* XXX Could improve efficiency by separating these all out using a
4165 * macro or in-line function. At that point regcomp.c would no longer
4166 * have to set the FLAGS fields of these */
4167 case BOUNDL: /* /\b/l */
4168 case NBOUNDL: /* /\B/l */
4169 RX_MATCH_TAINTED_on(reginfo->prog);
4171 case BOUND: /* /\b/ */
4172 case BOUNDU: /* /\b/u */
4173 case BOUNDA: /* /\b/a */
4174 case NBOUND: /* /\B/ */
4175 case NBOUNDU: /* /\B/u */
4176 case NBOUNDA: /* /\B/a */
4177 /* was last char in word? */
4179 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4180 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4182 if (locinput == reginfo->strbeg)
4185 const U8 * const r =
4186 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4188 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
4190 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4191 ln = isWORDCHAR_uni(ln);
4195 LOAD_UTF8_CHARCLASS_ALNUM();
4196 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4201 ln = isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(ln));
4202 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4207 /* Here the string isn't utf8, or is utf8 and only ascii
4208 * characters are to match \w. In the latter case looking at
4209 * the byte just prior to the current one may be just the final
4210 * byte of a multi-byte character. This is ok. There are two
4212 * 1) it is a single byte character, and then the test is doing
4213 * just what it's supposed to.
4214 * 2) it is a multi-byte character, in which case the final
4215 * byte is never mistakable for ASCII, and so the test
4216 * will say it is not a word character, which is the
4217 * correct answer. */
4218 ln = (locinput != reginfo->strbeg) ?
4219 UCHARAT(locinput - 1) : '\n';
4220 switch (FLAGS(scan)) {
4221 case REGEX_UNICODE_CHARSET:
4222 ln = isWORDCHAR_L1(ln);
4223 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4225 case REGEX_LOCALE_CHARSET:
4226 ln = isWORDCHAR_LC(ln);
4227 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4229 case REGEX_DEPENDS_CHARSET:
4230 ln = isWORDCHAR(ln);
4231 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4233 case REGEX_ASCII_RESTRICTED_CHARSET:
4234 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4235 ln = isWORDCHAR_A(ln);
4236 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4239 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4243 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4245 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4249 case ANYOF: /* /[abc]/ */
4250 case ANYOF_WARN_SUPER:
4254 if (!reginclass(rex, scan, (U8*)locinput, utf8_target))
4256 locinput += UTF8SKIP(locinput);
4259 if (!REGINCLASS(rex, scan, (U8*)locinput))
4265 /* The argument (FLAGS) to all the POSIX node types is the class number
4268 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4272 case POSIXL: /* \w or [:punct:] etc. under /l */
4276 /* The locale hasn't influenced the outcome before this, so defer
4277 * tainting until now */
4278 RX_MATCH_TAINTED_on(reginfo->prog);
4280 /* Use isFOO_lc() for characters within Latin1. (Note that
4281 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4282 * wouldn't be invariant) */
4283 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4284 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4288 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4289 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4290 (U8) TWO_BYTE_UTF8_TO_UNI(nextchr,
4291 *(locinput + 1))))))
4296 else { /* Here, must be an above Latin-1 code point */
4297 goto utf8_posix_not_eos;
4300 /* Here, must be utf8 */
4301 locinput += UTF8SKIP(locinput);
4304 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4308 case POSIXD: /* \w or [:punct:] etc. under /d */
4314 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4316 if (NEXTCHR_IS_EOS) {
4320 /* All UTF-8 variants match */
4321 if (! UTF8_IS_INVARIANT(nextchr)) {
4322 goto increment_locinput;
4328 case POSIXA: /* \w or [:punct:] etc. under /a */
4331 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4332 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4333 * character is a single byte */
4336 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4342 /* Here we are either not in utf8, or we matched a utf8-invariant,
4343 * so the next char is the next byte */
4347 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4351 case POSIXU: /* \w or [:punct:] etc. under /u */
4353 if (NEXTCHR_IS_EOS) {
4358 /* Use _generic_isCC() for characters within Latin1. (Note that
4359 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4360 * wouldn't be invariant) */
4361 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4362 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4369 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4370 if (! (to_complement
4371 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(nextchr,
4379 else { /* Handle above Latin-1 code points */
4380 classnum = (_char_class_number) FLAGS(scan);
4381 if (classnum < _FIRST_NON_SWASH_CC) {
4383 /* Here, uses a swash to find such code points. Load if if
4384 * not done already */
4385 if (! PL_utf8_swash_ptrs[classnum]) {
4386 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4387 PL_utf8_swash_ptrs[classnum]
4388 = _core_swash_init("utf8",
4389 swash_property_names[classnum],
4390 &PL_sv_undef, 1, 0, NULL, &flags);
4392 if (! (to_complement
4393 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4394 (U8 *) locinput, TRUE))))
4399 else { /* Here, uses macros to find above Latin-1 code points */
4401 case _CC_ENUM_SPACE: /* XXX would require separate
4402 code if we revert the change
4403 of \v matching this */
4404 case _CC_ENUM_PSXSPC:
4405 if (! (to_complement
4406 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4411 case _CC_ENUM_BLANK:
4412 if (! (to_complement
4413 ^ cBOOL(is_HORIZWS_high(locinput))))
4418 case _CC_ENUM_XDIGIT:
4419 if (! (to_complement
4420 ^ cBOOL(is_XDIGIT_high(locinput))))
4425 case _CC_ENUM_VERTSPACE:
4426 if (! (to_complement
4427 ^ cBOOL(is_VERTWS_high(locinput))))
4432 default: /* The rest, e.g. [:cntrl:], can't match
4434 if (! to_complement) {
4440 locinput += UTF8SKIP(locinput);
4444 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4445 a Unicode extended Grapheme Cluster */
4446 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4447 extended Grapheme Cluster is:
4450 | Prepend* Begin Extend*
4453 Begin is: ( Special_Begin | ! Control )
4454 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4455 Extend is: ( Grapheme_Extend | Spacing_Mark )
4456 Control is: [ GCB_Control | CR | LF ]
4457 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4459 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4462 Begin is ( Regular_Begin + Special Begin )
4464 It turns out that 98.4% of all Unicode code points match
4465 Regular_Begin. Doing it this way eliminates a table match in
4466 the previous implementation for almost all Unicode code points.
4468 There is a subtlety with Prepend* which showed up in testing.
4469 Note that the Begin, and only the Begin is required in:
4470 | Prepend* Begin Extend*
4471 Also, Begin contains '! Control'. A Prepend must be a
4472 '! Control', which means it must also be a Begin. What it
4473 comes down to is that if we match Prepend* and then find no
4474 suitable Begin afterwards, that if we backtrack the last
4475 Prepend, that one will be a suitable Begin.
4480 if (! utf8_target) {
4482 /* Match either CR LF or '.', as all the other possibilities
4484 locinput++; /* Match the . or CR */
4485 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4487 && locinput < reginfo->strend
4488 && UCHARAT(locinput) == '\n')
4495 /* Utf8: See if is ( CR LF ); already know that locinput <
4496 * reginfo->strend, so locinput+1 is in bounds */
4497 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4498 && UCHARAT(locinput + 1) == '\n')
4505 /* In case have to backtrack to beginning, then match '.' */
4506 char *starting = locinput;
4508 /* In case have to backtrack the last prepend */
4509 char *previous_prepend = NULL;
4511 LOAD_UTF8_CHARCLASS_GCB();
4513 /* Match (prepend)* */
4514 while (locinput < reginfo->strend
4515 && (len = is_GCB_Prepend_utf8(locinput)))
4517 previous_prepend = locinput;
4521 /* As noted above, if we matched a prepend character, but
4522 * the next thing won't match, back off the last prepend we
4523 * matched, as it is guaranteed to match the begin */
4524 if (previous_prepend
4525 && (locinput >= reginfo->strend
4526 || (! swash_fetch(PL_utf8_X_regular_begin,
4527 (U8*)locinput, utf8_target)
4528 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4531 locinput = previous_prepend;
4534 /* Note that here we know reginfo->strend > locinput, as we
4535 * tested that upon input to this switch case, and if we
4536 * moved locinput forward, we tested the result just above
4537 * and it either passed, or we backed off so that it will
4539 if (swash_fetch(PL_utf8_X_regular_begin,
4540 (U8*)locinput, utf8_target)) {
4541 locinput += UTF8SKIP(locinput);
4543 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4545 /* Here did not match the required 'Begin' in the
4546 * second term. So just match the very first
4547 * character, the '.' of the final term of the regex */
4548 locinput = starting + UTF8SKIP(starting);
4552 /* Here is a special begin. It can be composed of
4553 * several individual characters. One possibility is
4555 if ((len = is_GCB_RI_utf8(locinput))) {
4557 while (locinput < reginfo->strend
4558 && (len = is_GCB_RI_utf8(locinput)))
4562 } else if ((len = is_GCB_T_utf8(locinput))) {
4563 /* Another possibility is T+ */
4565 while (locinput < reginfo->strend
4566 && (len = is_GCB_T_utf8(locinput)))
4572 /* Here, neither RI+ nor T+; must be some other
4573 * Hangul. That means it is one of the others: L,
4574 * LV, LVT or V, and matches:
4575 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4578 while (locinput < reginfo->strend
4579 && (len = is_GCB_L_utf8(locinput)))
4584 /* Here, have exhausted L*. If the next character
4585 * is not an LV, LVT nor V, it means we had to have
4586 * at least one L, so matches L+ in the original
4587 * equation, we have a complete hangul syllable.
4590 if (locinput < reginfo->strend
4591 && is_GCB_LV_LVT_V_utf8(locinput))
4593 /* Otherwise keep going. Must be LV, LVT or V.
4594 * See if LVT, by first ruling out V, then LV */
4595 if (! is_GCB_V_utf8(locinput)
4596 /* All but every TCount one is LV */
4597 && (valid_utf8_to_uvchr((U8 *) locinput,
4602 locinput += UTF8SKIP(locinput);
4605 /* Must be V or LV. Take it, then match
4607 locinput += UTF8SKIP(locinput);
4608 while (locinput < reginfo->strend
4609 && (len = is_GCB_V_utf8(locinput)))
4615 /* And any of LV, LVT, or V can be followed
4617 while (locinput < reginfo->strend
4618 && (len = is_GCB_T_utf8(locinput)))
4626 /* Match any extender */
4627 while (locinput < reginfo->strend
4628 && swash_fetch(PL_utf8_X_extend,
4629 (U8*)locinput, utf8_target))
4631 locinput += UTF8SKIP(locinput);
4635 if (locinput > reginfo->strend) sayNO;
4639 case NREFFL: /* /\g{name}/il */
4640 { /* The capture buffer cases. The ones beginning with N for the
4641 named buffers just convert to the equivalent numbered and
4642 pretend they were called as the corresponding numbered buffer
4644 /* don't initialize these in the declaration, it makes C++
4649 const U8 *fold_array;
4652 RX_MATCH_TAINTED_on(reginfo->prog);
4653 folder = foldEQ_locale;
4654 fold_array = PL_fold_locale;
4656 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4659 case NREFFA: /* /\g{name}/iaa */
4660 folder = foldEQ_latin1;
4661 fold_array = PL_fold_latin1;
4663 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4666 case NREFFU: /* /\g{name}/iu */
4667 folder = foldEQ_latin1;
4668 fold_array = PL_fold_latin1;
4670 utf8_fold_flags = 0;
4673 case NREFF: /* /\g{name}/i */
4675 fold_array = PL_fold;
4677 utf8_fold_flags = 0;
4680 case NREF: /* /\g{name}/ */
4684 utf8_fold_flags = 0;
4687 /* For the named back references, find the corresponding buffer
4689 n = reg_check_named_buff_matched(rex,scan);
4694 goto do_nref_ref_common;
4696 case REFFL: /* /\1/il */
4697 RX_MATCH_TAINTED_on(reginfo->prog);
4698 folder = foldEQ_locale;
4699 fold_array = PL_fold_locale;
4700 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4703 case REFFA: /* /\1/iaa */
4704 folder = foldEQ_latin1;
4705 fold_array = PL_fold_latin1;
4706 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4709 case REFFU: /* /\1/iu */
4710 folder = foldEQ_latin1;
4711 fold_array = PL_fold_latin1;
4712 utf8_fold_flags = 0;
4715 case REFF: /* /\1/i */
4717 fold_array = PL_fold;
4718 utf8_fold_flags = 0;
4721 case REF: /* /\1/ */
4724 utf8_fold_flags = 0;
4728 n = ARG(scan); /* which paren pair */
4731 ln = rex->offs[n].start;
4732 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
4733 if (rex->lastparen < n || ln == -1)
4734 sayNO; /* Do not match unless seen CLOSEn. */
4735 if (ln == rex->offs[n].end)
4738 s = reginfo->strbeg + ln;
4739 if (type != REF /* REF can do byte comparison */
4740 && (utf8_target || type == REFFU))
4741 { /* XXX handle REFFL better */
4742 char * limit = reginfo->strend;
4744 /* This call case insensitively compares the entire buffer
4745 * at s, with the current input starting at locinput, but
4746 * not going off the end given by reginfo->strend, and
4747 * returns in <limit> upon success, how much of the
4748 * current input was matched */
4749 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4750 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4758 /* Not utf8: Inline the first character, for speed. */
4759 if (!NEXTCHR_IS_EOS &&
4760 UCHARAT(s) != nextchr &&
4762 UCHARAT(s) != fold_array[nextchr]))
4764 ln = rex->offs[n].end - ln;
4765 if (locinput + ln > reginfo->strend)
4767 if (ln > 1 && (type == REF
4768 ? memNE(s, locinput, ln)
4769 : ! folder(s, locinput, ln)))
4775 case NOTHING: /* null op; e.g. the 'nothing' following
4776 * the '*' in m{(a+|b)*}' */
4778 case TAIL: /* placeholder while compiling (A|B|C) */
4781 case BACK: /* ??? doesn't appear to be used ??? */
4785 #define ST st->u.eval
4790 regexp_internal *rei;
4791 regnode *startpoint;
4793 case GOSTART: /* (?R) */
4794 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4795 if (cur_eval && cur_eval->locinput==locinput) {
4796 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4797 Perl_croak(aTHX_ "Infinite recursion in regex");
4798 if ( ++nochange_depth > max_nochange_depth )
4800 "Pattern subroutine nesting without pos change"
4801 " exceeded limit in regex");
4808 if (OP(scan)==GOSUB) {
4809 startpoint = scan + ARG2L(scan);
4810 ST.close_paren = ARG(scan);
4812 startpoint = rei->program+1;
4815 goto eval_recurse_doit;
4816 assert(0); /* NOTREACHED */
4818 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4819 if (cur_eval && cur_eval->locinput==locinput) {
4820 if ( ++nochange_depth > max_nochange_depth )
4821 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4826 /* execute the code in the {...} */
4830 OP * const oop = PL_op;
4831 COP * const ocurcop = PL_curcop;
4835 /* save *all* paren positions */
4836 regcppush(rex, 0, maxopenparen);
4837 REGCP_SET(runops_cp);
4840 caller_cv = find_runcv(NULL);
4844 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4846 (REGEXP*)(rexi->data->data[n])
4849 nop = (OP*)rexi->data->data[n+1];
4851 else if (rexi->data->what[n] == 'l') { /* literal code */
4853 nop = (OP*)rexi->data->data[n];
4854 assert(CvDEPTH(newcv));
4857 /* literal with own CV */
4858 assert(rexi->data->what[n] == 'L');
4859 newcv = rex->qr_anoncv;
4860 nop = (OP*)rexi->data->data[n];
4863 /* normally if we're about to execute code from the same
4864 * CV that we used previously, we just use the existing
4865 * CX stack entry. However, its possible that in the
4866 * meantime we may have backtracked, popped from the save
4867 * stack, and undone the SAVECOMPPAD(s) associated with
4868 * PUSH_MULTICALL; in which case PL_comppad no longer
4869 * points to newcv's pad. */
4870 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4872 U8 flags = (CXp_SUB_RE |
4873 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
4874 if (last_pushed_cv) {
4875 CHANGE_MULTICALL_FLAGS(newcv, flags);
4878 PUSH_MULTICALL_FLAGS(newcv, flags);
4880 last_pushed_cv = newcv;
4883 /* these assignments are just to silence compiler
4885 multicall_cop = NULL;
4888 last_pad = PL_comppad;
4890 /* the initial nextstate you would normally execute
4891 * at the start of an eval (which would cause error
4892 * messages to come from the eval), may be optimised
4893 * away from the execution path in the regex code blocks;
4894 * so manually set PL_curcop to it initially */
4896 OP *o = cUNOPx(nop)->op_first;
4897 assert(o->op_type == OP_NULL);
4898 if (o->op_targ == OP_SCOPE) {
4899 o = cUNOPo->op_first;
4902 assert(o->op_targ == OP_LEAVE);
4903 o = cUNOPo->op_first;
4904 assert(o->op_type == OP_ENTER);
4908 if (o->op_type != OP_STUB) {
4909 assert( o->op_type == OP_NEXTSTATE
4910 || o->op_type == OP_DBSTATE
4911 || (o->op_type == OP_NULL
4912 && ( o->op_targ == OP_NEXTSTATE
4913 || o->op_targ == OP_DBSTATE
4917 PL_curcop = (COP*)o;
4922 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4923 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4925 rex->offs[0].end = locinput - reginfo->strbeg;
4926 if (reginfo->info_aux_eval->pos_magic)
4927 reginfo->info_aux_eval->pos_magic->mg_len
4928 = locinput - reginfo->strbeg;
4931 SV *sv_mrk = get_sv("REGMARK", 1);
4932 sv_setsv(sv_mrk, sv_yes_mark);
4935 /* we don't use MULTICALL here as we want to call the
4936 * first op of the block of interest, rather than the
4937 * first op of the sub */
4938 before = (IV)(SP-PL_stack_base);
4940 CALLRUNOPS(aTHX); /* Scalar context. */
4942 if ((IV)(SP-PL_stack_base) == before)
4943 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4949 /* before restoring everything, evaluate the returned
4950 * value, so that 'uninit' warnings don't use the wrong
4951 * PL_op or pad. Also need to process any magic vars
4952 * (e.g. $1) *before* parentheses are restored */
4957 if (logical == 0) /* (?{})/ */
4958 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4959 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4960 sw = cBOOL(SvTRUE(ret));
4963 else { /* /(??{}) */
4964 /* if its overloaded, let the regex compiler handle
4965 * it; otherwise extract regex, or stringify */
4966 if (!SvAMAGIC(ret)) {
4970 if (SvTYPE(sv) == SVt_REGEXP)
4971 re_sv = (REGEXP*) sv;
4972 else if (SvSMAGICAL(sv)) {
4973 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
4975 re_sv = (REGEXP *) mg->mg_obj;
4978 /* force any magic, undef warnings here */
4980 ret = sv_mortalcopy(ret);
4981 (void) SvPV_force_nolen(ret);
4987 /* *** Note that at this point we don't restore
4988 * PL_comppad, (or pop the CxSUB) on the assumption it may
4989 * be used again soon. This is safe as long as nothing
4990 * in the regexp code uses the pad ! */
4992 PL_curcop = ocurcop;
4993 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
4994 PL_curpm = PL_reg_curpm;
5000 /* only /(??{})/ from now on */
5003 /* extract RE object from returned value; compiling if
5007 re_sv = reg_temp_copy(NULL, re_sv);
5012 if (SvUTF8(ret) && IN_BYTES) {
5013 /* In use 'bytes': make a copy of the octet
5014 * sequence, but without the flag on */
5016 const char *const p = SvPV(ret, len);
5017 ret = newSVpvn_flags(p, len, SVs_TEMP);
5019 if (rex->intflags & PREGf_USE_RE_EVAL)
5020 pm_flags |= PMf_USE_RE_EVAL;
5022 /* if we got here, it should be an engine which
5023 * supports compiling code blocks and stuff */
5024 assert(rex->engine && rex->engine->op_comp);
5025 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5026 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5027 rex->engine, NULL, NULL,
5028 /* copy /msix etc to inner pattern */
5033 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
5035 /* This isn't a first class regexp. Instead, it's
5036 caching a regexp onto an existing, Perl visible
5038 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5040 /* safe to do now that any $1 etc has been
5041 * interpolated into the new pattern string and
5043 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5048 RXp_MATCH_COPIED_off(re);
5049 re->subbeg = rex->subbeg;
5050 re->sublen = rex->sublen;
5051 re->suboffset = rex->suboffset;
5052 re->subcoffset = rex->subcoffset;
5055 debug_start_match(re_sv, utf8_target, locinput,
5056 reginfo->strend, "Matching embedded");
5058 startpoint = rei->program + 1;
5059 ST.close_paren = 0; /* only used for GOSUB */
5061 eval_recurse_doit: /* Share code with GOSUB below this line */
5062 /* run the pattern returned from (??{...}) */
5064 /* Save *all* the positions. */
5065 ST.cp = regcppush(rex, 0, maxopenparen);
5066 REGCP_SET(ST.lastcp);
5069 re->lastcloseparen = 0;
5073 /* invalidate the S-L poscache. We're now executing a
5074 * different set of WHILEM ops (and their associated
5075 * indexes) against the same string, so the bits in the
5076 * cache are meaningless. Setting maxiter to zero forces
5077 * the cache to be invalidated and zeroed before reuse.
5078 * XXX This is too dramatic a measure. Ideally we should
5079 * save the old cache and restore when running the outer
5081 reginfo->poscache_maxiter = 0;
5083 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5085 ST.prev_rex = rex_sv;
5086 ST.prev_curlyx = cur_curlyx;
5088 SET_reg_curpm(rex_sv);
5093 ST.prev_eval = cur_eval;
5095 /* now continue from first node in postoned RE */
5096 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5097 assert(0); /* NOTREACHED */
5100 case EVAL_AB: /* cleanup after a successful (??{A})B */
5101 /* note: this is called twice; first after popping B, then A */
5102 rex_sv = ST.prev_rex;
5103 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5104 SET_reg_curpm(rex_sv);
5105 rex = ReANY(rex_sv);
5106 rexi = RXi_GET(rex);
5108 cur_eval = ST.prev_eval;
5109 cur_curlyx = ST.prev_curlyx;
5111 /* Invalidate cache. See "invalidate" comment above. */
5112 reginfo->poscache_maxiter = 0;
5113 if ( nochange_depth )
5118 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5119 /* note: this is called twice; first after popping B, then A */
5120 rex_sv = ST.prev_rex;
5121 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5122 SET_reg_curpm(rex_sv);
5123 rex = ReANY(rex_sv);
5124 rexi = RXi_GET(rex);
5126 REGCP_UNWIND(ST.lastcp);
5127 regcppop(rex, &maxopenparen);
5128 cur_eval = ST.prev_eval;
5129 cur_curlyx = ST.prev_curlyx;
5130 /* Invalidate cache. See "invalidate" comment above. */
5131 reginfo->poscache_maxiter = 0;
5132 if ( nochange_depth )
5138 n = ARG(scan); /* which paren pair */
5139 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5140 if (n > maxopenparen)
5142 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5143 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5147 (IV)rex->offs[n].start_tmp,
5153 /* XXX really need to log other places start/end are set too */
5154 #define CLOSE_CAPTURE \
5155 rex->offs[n].start = rex->offs[n].start_tmp; \
5156 rex->offs[n].end = locinput - reginfo->strbeg; \
5157 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5158 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5160 PTR2UV(rex->offs), \
5162 (IV)rex->offs[n].start, \
5163 (IV)rex->offs[n].end \
5167 n = ARG(scan); /* which paren pair */
5169 if (n > rex->lastparen)
5171 rex->lastcloseparen = n;
5172 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5177 case ACCEPT: /* (*ACCEPT) */
5181 cursor && OP(cursor)!=END;
5182 cursor=regnext(cursor))
5184 if ( OP(cursor)==CLOSE ){
5186 if ( n <= lastopen ) {
5188 if (n > rex->lastparen)
5190 rex->lastcloseparen = n;
5191 if ( n == ARG(scan) || (cur_eval &&
5192 cur_eval->u.eval.close_paren == n))
5201 case GROUPP: /* (?(1)) */
5202 n = ARG(scan); /* which paren pair */
5203 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5206 case NGROUPP: /* (?(<name>)) */
5207 /* reg_check_named_buff_matched returns 0 for no match */
5208 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5211 case INSUBP: /* (?(R)) */
5213 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5216 case DEFINEP: /* (?(DEFINE)) */
5220 case IFTHEN: /* (?(cond)A|B) */
5221 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5223 next = NEXTOPER(NEXTOPER(scan));
5225 next = scan + ARG(scan);
5226 if (OP(next) == IFTHEN) /* Fake one. */
5227 next = NEXTOPER(NEXTOPER(next));
5231 case LOGICAL: /* modifier for EVAL and IFMATCH */
5232 logical = scan->flags;
5235 /*******************************************************************
5237 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5238 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5239 STAR/PLUS/CURLY/CURLYN are used instead.)
5241 A*B is compiled as <CURLYX><A><WHILEM><B>
5243 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5244 state, which contains the current count, initialised to -1. It also sets
5245 cur_curlyx to point to this state, with any previous value saved in the
5248 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5249 since the pattern may possibly match zero times (i.e. it's a while {} loop
5250 rather than a do {} while loop).
5252 Each entry to WHILEM represents a successful match of A. The count in the
5253 CURLYX block is incremented, another WHILEM state is pushed, and execution
5254 passes to A or B depending on greediness and the current count.
5256 For example, if matching against the string a1a2a3b (where the aN are
5257 substrings that match /A/), then the match progresses as follows: (the
5258 pushed states are interspersed with the bits of strings matched so far):
5261 <CURLYX cnt=0><WHILEM>
5262 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5263 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5264 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5265 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5267 (Contrast this with something like CURLYM, which maintains only a single
5271 a1 <CURLYM cnt=1> a2
5272 a1 a2 <CURLYM cnt=2> a3
5273 a1 a2 a3 <CURLYM cnt=3> b
5276 Each WHILEM state block marks a point to backtrack to upon partial failure
5277 of A or B, and also contains some minor state data related to that
5278 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5279 overall state, such as the count, and pointers to the A and B ops.
5281 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5282 must always point to the *current* CURLYX block, the rules are:
5284 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5285 and set cur_curlyx to point the new block.
5287 When popping the CURLYX block after a successful or unsuccessful match,
5288 restore the previous cur_curlyx.
5290 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5291 to the outer one saved in the CURLYX block.
5293 When popping the WHILEM block after a successful or unsuccessful B match,
5294 restore the previous cur_curlyx.
5296 Here's an example for the pattern (AI* BI)*BO
5297 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5300 curlyx backtrack stack
5301 ------ ---------------
5303 CO <CO prev=NULL> <WO>
5304 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5305 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5306 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5308 At this point the pattern succeeds, and we work back down the stack to
5309 clean up, restoring as we go:
5311 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5312 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5313 CO <CO prev=NULL> <WO>
5316 *******************************************************************/
5318 #define ST st->u.curlyx
5320 case CURLYX: /* start of /A*B/ (for complex A) */
5322 /* No need to save/restore up to this paren */
5323 I32 parenfloor = scan->flags;
5325 assert(next); /* keep Coverity happy */
5326 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5329 /* XXXX Probably it is better to teach regpush to support
5330 parenfloor > maxopenparen ... */
5331 if (parenfloor > (I32)rex->lastparen)
5332 parenfloor = rex->lastparen; /* Pessimization... */
5334 ST.prev_curlyx= cur_curlyx;
5336 ST.cp = PL_savestack_ix;
5338 /* these fields contain the state of the current curly.
5339 * they are accessed by subsequent WHILEMs */
5340 ST.parenfloor = parenfloor;
5345 ST.count = -1; /* this will be updated by WHILEM */
5346 ST.lastloc = NULL; /* this will be updated by WHILEM */
5348 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5349 assert(0); /* NOTREACHED */
5352 case CURLYX_end: /* just finished matching all of A*B */
5353 cur_curlyx = ST.prev_curlyx;
5355 assert(0); /* NOTREACHED */
5357 case CURLYX_end_fail: /* just failed to match all of A*B */
5359 cur_curlyx = ST.prev_curlyx;
5361 assert(0); /* NOTREACHED */
5365 #define ST st->u.whilem
5367 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5369 /* see the discussion above about CURLYX/WHILEM */
5371 int min = ARG1(cur_curlyx->u.curlyx.me);
5372 int max = ARG2(cur_curlyx->u.curlyx.me);
5373 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5375 assert(cur_curlyx); /* keep Coverity happy */
5376 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5377 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5378 ST.cache_offset = 0;
5382 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5383 "%*s whilem: matched %ld out of %d..%d\n",
5384 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5387 /* First just match a string of min A's. */
5390 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5392 cur_curlyx->u.curlyx.lastloc = locinput;
5393 REGCP_SET(ST.lastcp);
5395 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5396 assert(0); /* NOTREACHED */
5399 /* If degenerate A matches "", assume A done. */
5401 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5402 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5403 "%*s whilem: empty match detected, trying continuation...\n",
5404 REPORT_CODE_OFF+depth*2, "")
5406 goto do_whilem_B_max;
5409 /* super-linear cache processing.
5411 * The idea here is that for certain types of CURLYX/WHILEM -
5412 * principally those whose upper bound is infinity (and
5413 * excluding regexes that have things like \1 and other very
5414 * non-regular expresssiony things), then if a pattern like
5415 * /....A*.../ fails and we backtrack to the WHILEM, then we
5416 * make a note that this particular WHILEM op was at string
5417 * position 47 (say) when the rest of pattern failed. Then, if
5418 * we ever find ourselves back at that WHILEM, and at string
5419 * position 47 again, we can just fail immediately rather than
5420 * running the rest of the pattern again.
5422 * This is very handy when patterns start to go
5423 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5424 * with a combinatorial explosion of backtracking.
5426 * The cache is implemented as a bit array, with one bit per
5427 * string byte position per WHILEM op (up to 16) - so its
5428 * between 0.25 and 2x the string size.
5430 * To avoid allocating a poscache buffer every time, we do an
5431 * initially countdown; only after we have executed a WHILEM
5432 * op (string-length x #WHILEMs) times do we allocate the
5435 * The top 4 bits of scan->flags byte say how many different
5436 * relevant CURLLYX/WHILEM op pairs there are, while the
5437 * bottom 4-bits is the identifying index number of this
5443 if (!reginfo->poscache_maxiter) {
5444 /* start the countdown: Postpone detection until we
5445 * know the match is not *that* much linear. */
5446 reginfo->poscache_maxiter
5447 = (reginfo->strend - reginfo->strbeg + 1)
5449 /* possible overflow for long strings and many CURLYX's */
5450 if (reginfo->poscache_maxiter < 0)
5451 reginfo->poscache_maxiter = I32_MAX;
5452 reginfo->poscache_iter = reginfo->poscache_maxiter;
5455 if (reginfo->poscache_iter-- == 0) {
5456 /* initialise cache */
5457 const I32 size = (reginfo->poscache_maxiter + 7)/8;
5458 regmatch_info_aux *const aux = reginfo->info_aux;
5459 if (aux->poscache) {
5460 if ((I32)reginfo->poscache_size < size) {
5461 Renew(aux->poscache, size, char);
5462 reginfo->poscache_size = size;
5464 Zero(aux->poscache, size, char);
5467 reginfo->poscache_size = size;
5468 Newxz(aux->poscache, size, char);
5470 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5471 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5472 PL_colors[4], PL_colors[5])
5476 if (reginfo->poscache_iter < 0) {
5477 /* have we already failed at this position? */
5480 reginfo->poscache_iter = -1; /* stop eventual underflow */
5481 offset = (scan->flags & 0xf) - 1
5482 + (locinput - reginfo->strbeg)
5484 mask = 1 << (offset % 8);
5486 if (reginfo->info_aux->poscache[offset] & mask) {
5487 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5488 "%*s whilem: (cache) already tried at this position...\n",
5489 REPORT_CODE_OFF+depth*2, "")
5491 sayNO; /* cache records failure */
5493 ST.cache_offset = offset;
5494 ST.cache_mask = mask;
5498 /* Prefer B over A for minimal matching. */
5500 if (cur_curlyx->u.curlyx.minmod) {
5501 ST.save_curlyx = cur_curlyx;
5502 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5503 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5505 REGCP_SET(ST.lastcp);
5506 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5508 assert(0); /* NOTREACHED */
5511 /* Prefer A over B for maximal matching. */
5513 if (n < max) { /* More greed allowed? */
5514 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5516 cur_curlyx->u.curlyx.lastloc = locinput;
5517 REGCP_SET(ST.lastcp);
5518 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5519 assert(0); /* NOTREACHED */
5521 goto do_whilem_B_max;
5523 assert(0); /* NOTREACHED */
5525 case WHILEM_B_min: /* just matched B in a minimal match */
5526 case WHILEM_B_max: /* just matched B in a maximal match */
5527 cur_curlyx = ST.save_curlyx;
5529 assert(0); /* NOTREACHED */
5531 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5532 cur_curlyx = ST.save_curlyx;
5533 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5534 cur_curlyx->u.curlyx.count--;
5536 assert(0); /* NOTREACHED */
5538 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5540 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5541 REGCP_UNWIND(ST.lastcp);
5542 regcppop(rex, &maxopenparen);
5543 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5544 cur_curlyx->u.curlyx.count--;
5546 assert(0); /* NOTREACHED */
5548 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5549 REGCP_UNWIND(ST.lastcp);
5550 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5551 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5552 "%*s whilem: failed, trying continuation...\n",
5553 REPORT_CODE_OFF+depth*2, "")
5556 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5557 && ckWARN(WARN_REGEXP)
5558 && !reginfo->warned)
5560 reginfo->warned = TRUE;
5561 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5562 "Complex regular subexpression recursion limit (%d) "
5568 ST.save_curlyx = cur_curlyx;
5569 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5570 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5572 assert(0); /* NOTREACHED */
5574 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5575 cur_curlyx = ST.save_curlyx;
5576 REGCP_UNWIND(ST.lastcp);
5577 regcppop(rex, &maxopenparen);
5579 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5580 /* Maximum greed exceeded */
5581 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5582 && ckWARN(WARN_REGEXP)
5583 && !reginfo->warned)
5585 reginfo->warned = TRUE;
5586 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5587 "Complex regular subexpression recursion "
5588 "limit (%d) exceeded",
5591 cur_curlyx->u.curlyx.count--;
5595 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5596 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5598 /* Try grabbing another A and see if it helps. */
5599 cur_curlyx->u.curlyx.lastloc = locinput;
5600 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5602 REGCP_SET(ST.lastcp);
5603 PUSH_STATE_GOTO(WHILEM_A_min,
5604 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5606 assert(0); /* NOTREACHED */
5609 #define ST st->u.branch
5611 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5612 next = scan + ARG(scan);
5615 scan = NEXTOPER(scan);
5618 case BRANCH: /* /(...|A|...)/ */
5619 scan = NEXTOPER(scan); /* scan now points to inner node */
5620 ST.lastparen = rex->lastparen;
5621 ST.lastcloseparen = rex->lastcloseparen;
5622 ST.next_branch = next;
5625 /* Now go into the branch */
5627 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5629 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5631 assert(0); /* NOTREACHED */
5633 case CUTGROUP: /* /(*THEN)/ */
5634 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5635 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5636 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5637 assert(0); /* NOTREACHED */
5639 case CUTGROUP_next_fail:
5642 if (st->u.mark.mark_name)
5643 sv_commit = st->u.mark.mark_name;
5645 assert(0); /* NOTREACHED */
5649 assert(0); /* NOTREACHED */
5651 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5656 REGCP_UNWIND(ST.cp);
5657 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5658 scan = ST.next_branch;
5659 /* no more branches? */
5660 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5662 PerlIO_printf( Perl_debug_log,
5663 "%*s %sBRANCH failed...%s\n",
5664 REPORT_CODE_OFF+depth*2, "",
5670 continue; /* execute next BRANCH[J] op */
5671 assert(0); /* NOTREACHED */
5673 case MINMOD: /* next op will be non-greedy, e.g. A*? */
5678 #define ST st->u.curlym
5680 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5682 /* This is an optimisation of CURLYX that enables us to push
5683 * only a single backtracking state, no matter how many matches
5684 * there are in {m,n}. It relies on the pattern being constant
5685 * length, with no parens to influence future backrefs
5689 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5691 ST.lastparen = rex->lastparen;
5692 ST.lastcloseparen = rex->lastcloseparen;
5694 /* if paren positive, emulate an OPEN/CLOSE around A */
5696 U32 paren = ST.me->flags;
5697 if (paren > maxopenparen)
5698 maxopenparen = paren;
5699 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5707 ST.c1 = CHRTEST_UNINIT;
5710 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5713 curlym_do_A: /* execute the A in /A{m,n}B/ */
5714 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5715 assert(0); /* NOTREACHED */
5717 case CURLYM_A: /* we've just matched an A */
5719 /* after first match, determine A's length: u.curlym.alen */
5720 if (ST.count == 1) {
5721 if (reginfo->is_utf8_target) {
5722 char *s = st->locinput;
5723 while (s < locinput) {
5729 ST.alen = locinput - st->locinput;
5732 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5735 PerlIO_printf(Perl_debug_log,
5736 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5737 (int)(REPORT_CODE_OFF+(depth*2)), "",
5738 (IV) ST.count, (IV)ST.alen)
5741 if (cur_eval && cur_eval->u.eval.close_paren &&
5742 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5746 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5747 if ( max == REG_INFTY || ST.count < max )
5748 goto curlym_do_A; /* try to match another A */
5750 goto curlym_do_B; /* try to match B */
5752 case CURLYM_A_fail: /* just failed to match an A */
5753 REGCP_UNWIND(ST.cp);
5755 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5756 || (cur_eval && cur_eval->u.eval.close_paren &&
5757 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5760 curlym_do_B: /* execute the B in /A{m,n}B/ */
5761 if (ST.c1 == CHRTEST_UNINIT) {
5762 /* calculate c1 and c2 for possible match of 1st char
5763 * following curly */
5764 ST.c1 = ST.c2 = CHRTEST_VOID;
5765 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5766 regnode *text_node = ST.B;
5767 if (! HAS_TEXT(text_node))
5768 FIND_NEXT_IMPT(text_node);
5771 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5773 But the former is redundant in light of the latter.
5775 if this changes back then the macro for
5776 IS_TEXT and friends need to change.
5778 if (PL_regkind[OP(text_node)] == EXACT) {
5779 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5780 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5790 PerlIO_printf(Perl_debug_log,
5791 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5792 (int)(REPORT_CODE_OFF+(depth*2)),
5795 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
5796 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
5797 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
5798 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
5800 /* simulate B failing */
5802 PerlIO_printf(Perl_debug_log,
5803 "%*s CURLYM Fast bail next target=U+%"UVXf" c1=U+%"UVXf" c2=U+%"UVXf"\n",
5804 (int)(REPORT_CODE_OFF+(depth*2)),"",
5805 valid_utf8_to_uvchr((U8 *) locinput, NULL),
5806 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
5807 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
5809 state_num = CURLYM_B_fail;
5810 goto reenter_switch;
5813 else if (nextchr != ST.c1 && nextchr != ST.c2) {
5814 /* simulate B failing */
5816 PerlIO_printf(Perl_debug_log,
5817 "%*s CURLYM Fast bail next target=U+%X c1=U+%X c2=U+%X\n",
5818 (int)(REPORT_CODE_OFF+(depth*2)),"",
5819 (int) nextchr, ST.c1, ST.c2)
5821 state_num = CURLYM_B_fail;
5822 goto reenter_switch;
5827 /* emulate CLOSE: mark current A as captured */
5828 I32 paren = ST.me->flags;
5830 rex->offs[paren].start
5831 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
5832 rex->offs[paren].end = locinput - reginfo->strbeg;
5833 if ((U32)paren > rex->lastparen)
5834 rex->lastparen = paren;
5835 rex->lastcloseparen = paren;
5838 rex->offs[paren].end = -1;
5839 if (cur_eval && cur_eval->u.eval.close_paren &&
5840 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5849 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5850 assert(0); /* NOTREACHED */
5852 case CURLYM_B_fail: /* just failed to match a B */
5853 REGCP_UNWIND(ST.cp);
5854 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5856 I32 max = ARG2(ST.me);
5857 if (max != REG_INFTY && ST.count == max)
5859 goto curlym_do_A; /* try to match a further A */
5861 /* backtrack one A */
5862 if (ST.count == ARG1(ST.me) /* min */)
5865 SET_locinput(HOPc(locinput, -ST.alen));
5866 goto curlym_do_B; /* try to match B */
5869 #define ST st->u.curly
5871 #define CURLY_SETPAREN(paren, success) \
5874 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
5875 rex->offs[paren].end = locinput - reginfo->strbeg; \
5876 if (paren > rex->lastparen) \
5877 rex->lastparen = paren; \
5878 rex->lastcloseparen = paren; \
5881 rex->offs[paren].end = -1; \
5882 rex->lastparen = ST.lastparen; \
5883 rex->lastcloseparen = ST.lastcloseparen; \
5887 case STAR: /* /A*B/ where A is width 1 char */
5891 scan = NEXTOPER(scan);
5894 case PLUS: /* /A+B/ where A is width 1 char */
5898 scan = NEXTOPER(scan);
5901 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
5902 ST.paren = scan->flags; /* Which paren to set */
5903 ST.lastparen = rex->lastparen;
5904 ST.lastcloseparen = rex->lastcloseparen;
5905 if (ST.paren > maxopenparen)
5906 maxopenparen = ST.paren;
5907 ST.min = ARG1(scan); /* min to match */
5908 ST.max = ARG2(scan); /* max to match */
5909 if (cur_eval && cur_eval->u.eval.close_paren &&
5910 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5914 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5917 case CURLY: /* /A{m,n}B/ where A is width 1 char */
5919 ST.min = ARG1(scan); /* min to match */
5920 ST.max = ARG2(scan); /* max to match */
5921 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5924 * Lookahead to avoid useless match attempts
5925 * when we know what character comes next.
5927 * Used to only do .*x and .*?x, but now it allows
5928 * for )'s, ('s and (?{ ... })'s to be in the way
5929 * of the quantifier and the EXACT-like node. -- japhy
5932 assert(ST.min <= ST.max);
5933 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
5934 ST.c1 = ST.c2 = CHRTEST_VOID;
5937 regnode *text_node = next;
5939 if (! HAS_TEXT(text_node))
5940 FIND_NEXT_IMPT(text_node);
5942 if (! HAS_TEXT(text_node))
5943 ST.c1 = ST.c2 = CHRTEST_VOID;
5945 if ( PL_regkind[OP(text_node)] != EXACT ) {
5946 ST.c1 = ST.c2 = CHRTEST_VOID;
5950 /* Currently we only get here when
5952 PL_rekind[OP(text_node)] == EXACT
5954 if this changes back then the macro for IS_TEXT and
5955 friends need to change. */
5956 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5957 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5969 char *li = locinput;
5972 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
5978 if (ST.c1 == CHRTEST_VOID)
5979 goto curly_try_B_min;
5981 ST.oldloc = locinput;
5983 /* set ST.maxpos to the furthest point along the
5984 * string that could possibly match */
5985 if (ST.max == REG_INFTY) {
5986 ST.maxpos = reginfo->strend - 1;
5988 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5991 else if (utf8_target) {
5992 int m = ST.max - ST.min;
5993 for (ST.maxpos = locinput;
5994 m >0 && ST.maxpos < reginfo->strend; m--)
5995 ST.maxpos += UTF8SKIP(ST.maxpos);
5998 ST.maxpos = locinput + ST.max - ST.min;
5999 if (ST.maxpos >= reginfo->strend)
6000 ST.maxpos = reginfo->strend - 1;
6002 goto curly_try_B_min_known;
6006 /* avoid taking address of locinput, so it can remain
6008 char *li = locinput;
6009 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6010 if (ST.count < ST.min)
6013 if ((ST.count > ST.min)
6014 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6016 /* A{m,n} must come at the end of the string, there's
6017 * no point in backing off ... */
6019 /* ...except that $ and \Z can match before *and* after
6020 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6021 We may back off by one in this case. */
6022 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6026 goto curly_try_B_max;
6028 assert(0); /* NOTREACHED */
6031 case CURLY_B_min_known_fail:
6032 /* failed to find B in a non-greedy match where c1,c2 valid */
6034 REGCP_UNWIND(ST.cp);
6036 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6038 /* Couldn't or didn't -- move forward. */
6039 ST.oldloc = locinput;
6041 locinput += UTF8SKIP(locinput);
6045 curly_try_B_min_known:
6046 /* find the next place where 'B' could work, then call B */
6050 n = (ST.oldloc == locinput) ? 0 : 1;
6051 if (ST.c1 == ST.c2) {
6052 /* set n to utf8_distance(oldloc, locinput) */
6053 while (locinput <= ST.maxpos
6054 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6056 locinput += UTF8SKIP(locinput);
6061 /* set n to utf8_distance(oldloc, locinput) */
6062 while (locinput <= ST.maxpos
6063 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6064 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6066 locinput += UTF8SKIP(locinput);
6071 else { /* Not utf8_target */
6072 if (ST.c1 == ST.c2) {
6073 while (locinput <= ST.maxpos &&
6074 UCHARAT(locinput) != ST.c1)
6078 while (locinput <= ST.maxpos
6079 && UCHARAT(locinput) != ST.c1
6080 && UCHARAT(locinput) != ST.c2)
6083 n = locinput - ST.oldloc;
6085 if (locinput > ST.maxpos)
6088 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6089 * at b; check that everything between oldloc and
6090 * locinput matches */
6091 char *li = ST.oldloc;
6093 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6095 assert(n == REG_INFTY || locinput == li);
6097 CURLY_SETPAREN(ST.paren, ST.count);
6098 if (cur_eval && cur_eval->u.eval.close_paren &&
6099 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6102 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6104 assert(0); /* NOTREACHED */
6107 case CURLY_B_min_fail:
6108 /* failed to find B in a non-greedy match where c1,c2 invalid */
6110 REGCP_UNWIND(ST.cp);
6112 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6114 /* failed -- move forward one */
6116 char *li = locinput;
6117 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6124 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6125 ST.count > 0)) /* count overflow ? */
6128 CURLY_SETPAREN(ST.paren, ST.count);
6129 if (cur_eval && cur_eval->u.eval.close_paren &&
6130 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6133 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6137 assert(0); /* NOTREACHED */
6141 /* a successful greedy match: now try to match B */
6142 if (cur_eval && cur_eval->u.eval.close_paren &&
6143 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6147 bool could_match = locinput < reginfo->strend;
6149 /* If it could work, try it. */
6150 if (ST.c1 != CHRTEST_VOID && could_match) {
6151 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6153 could_match = memEQ(locinput,
6158 UTF8SKIP(locinput));
6161 could_match = UCHARAT(locinput) == ST.c1
6162 || UCHARAT(locinput) == ST.c2;
6165 if (ST.c1 == CHRTEST_VOID || could_match) {
6166 CURLY_SETPAREN(ST.paren, ST.count);
6167 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6168 assert(0); /* NOTREACHED */
6173 case CURLY_B_max_fail:
6174 /* failed to find B in a greedy match */
6176 REGCP_UNWIND(ST.cp);
6178 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6181 if (--ST.count < ST.min)
6183 locinput = HOPc(locinput, -1);
6184 goto curly_try_B_max;
6188 case END: /* last op of main pattern */
6191 /* we've just finished A in /(??{A})B/; now continue with B */
6193 st->u.eval.prev_rex = rex_sv; /* inner */
6195 /* Save *all* the positions. */
6196 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6197 rex_sv = cur_eval->u.eval.prev_rex;
6198 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6199 SET_reg_curpm(rex_sv);
6200 rex = ReANY(rex_sv);
6201 rexi = RXi_GET(rex);
6202 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6204 REGCP_SET(st->u.eval.lastcp);
6206 /* Restore parens of the outer rex without popping the
6208 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6211 st->u.eval.prev_eval = cur_eval;
6212 cur_eval = cur_eval->u.eval.prev_eval;
6214 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6215 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6216 if ( nochange_depth )
6219 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6220 locinput); /* match B */
6223 if (locinput < reginfo->till) {
6224 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6225 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6227 (long)(locinput - startpos),
6228 (long)(reginfo->till - startpos),
6231 sayNO_SILENT; /* Cannot match: too short. */
6233 sayYES; /* Success! */
6235 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6237 PerlIO_printf(Perl_debug_log,
6238 "%*s %ssubpattern success...%s\n",
6239 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6240 sayYES; /* Success! */
6243 #define ST st->u.ifmatch
6248 case SUSPEND: /* (?>A) */
6250 newstart = locinput;
6253 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6255 goto ifmatch_trivial_fail_test;
6257 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6259 ifmatch_trivial_fail_test:
6261 char * const s = HOPBACKc(locinput, scan->flags);
6266 sw = 1 - cBOOL(ST.wanted);
6270 next = scan + ARG(scan);
6278 newstart = locinput;
6282 ST.logical = logical;
6283 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6285 /* execute body of (?...A) */
6286 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6287 assert(0); /* NOTREACHED */
6290 case IFMATCH_A_fail: /* body of (?...A) failed */
6291 ST.wanted = !ST.wanted;
6294 case IFMATCH_A: /* body of (?...A) succeeded */
6296 sw = cBOOL(ST.wanted);
6298 else if (!ST.wanted)
6301 if (OP(ST.me) != SUSPEND) {
6302 /* restore old position except for (?>...) */
6303 locinput = st->locinput;
6305 scan = ST.me + ARG(ST.me);
6308 continue; /* execute B */
6312 case LONGJMP: /* alternative with many branches compiles to
6313 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6314 next = scan + ARG(scan);
6319 case COMMIT: /* (*COMMIT) */
6320 reginfo->cutpoint = reginfo->strend;
6323 case PRUNE: /* (*PRUNE) */
6325 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6326 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6327 assert(0); /* NOTREACHED */
6329 case COMMIT_next_fail:
6333 case OPFAIL: /* (*FAIL) */
6335 assert(0); /* NOTREACHED */
6337 #define ST st->u.mark
6338 case MARKPOINT: /* (*MARK:foo) */
6339 ST.prev_mark = mark_state;
6340 ST.mark_name = sv_commit = sv_yes_mark
6341 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6343 ST.mark_loc = locinput;
6344 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6345 assert(0); /* NOTREACHED */
6347 case MARKPOINT_next:
6348 mark_state = ST.prev_mark;
6350 assert(0); /* NOTREACHED */
6352 case MARKPOINT_next_fail:
6353 if (popmark && sv_eq(ST.mark_name,popmark))
6355 if (ST.mark_loc > startpoint)
6356 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6357 popmark = NULL; /* we found our mark */
6358 sv_commit = ST.mark_name;
6361 PerlIO_printf(Perl_debug_log,
6362 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6363 REPORT_CODE_OFF+depth*2, "",
6364 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6367 mark_state = ST.prev_mark;
6368 sv_yes_mark = mark_state ?
6369 mark_state->u.mark.mark_name : NULL;
6371 assert(0); /* NOTREACHED */
6373 case SKIP: /* (*SKIP) */
6375 /* (*SKIP) : if we fail we cut here*/
6376 ST.mark_name = NULL;
6377 ST.mark_loc = locinput;
6378 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6380 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6381 otherwise do nothing. Meaning we need to scan
6383 regmatch_state *cur = mark_state;
6384 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6387 if ( sv_eq( cur->u.mark.mark_name,
6390 ST.mark_name = find;
6391 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6393 cur = cur->u.mark.prev_mark;
6396 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6399 case SKIP_next_fail:
6401 /* (*CUT:NAME) - Set up to search for the name as we
6402 collapse the stack*/
6403 popmark = ST.mark_name;
6405 /* (*CUT) - No name, we cut here.*/
6406 if (ST.mark_loc > startpoint)
6407 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6408 /* but we set sv_commit to latest mark_name if there
6409 is one so they can test to see how things lead to this
6412 sv_commit=mark_state->u.mark.mark_name;
6416 assert(0); /* NOTREACHED */
6419 case LNBREAK: /* \R */
6420 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6427 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6428 PTR2UV(scan), OP(scan));
6429 Perl_croak(aTHX_ "regexp memory corruption");
6431 /* this is a point to jump to in order to increment
6432 * locinput by one character */
6434 assert(!NEXTCHR_IS_EOS);
6436 locinput += PL_utf8skip[nextchr];
6437 /* locinput is allowed to go 1 char off the end, but not 2+ */
6438 if (locinput > reginfo->strend)
6447 /* switch break jumps here */
6448 scan = next; /* prepare to execute the next op and ... */
6449 continue; /* ... jump back to the top, reusing st */
6450 assert(0); /* NOTREACHED */
6453 /* push a state that backtracks on success */
6454 st->u.yes.prev_yes_state = yes_state;
6458 /* push a new regex state, then continue at scan */
6460 regmatch_state *newst;
6463 regmatch_state *cur = st;
6464 regmatch_state *curyes = yes_state;
6466 regmatch_slab *slab = PL_regmatch_slab;
6467 for (;curd > -1;cur--,curd--) {
6468 if (cur < SLAB_FIRST(slab)) {
6470 cur = SLAB_LAST(slab);
6472 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6473 REPORT_CODE_OFF + 2 + depth * 2,"",
6474 curd, PL_reg_name[cur->resume_state],
6475 (curyes == cur) ? "yes" : ""
6478 curyes = cur->u.yes.prev_yes_state;
6481 DEBUG_STATE_pp("push")
6484 st->locinput = locinput;
6486 if (newst > SLAB_LAST(PL_regmatch_slab))
6487 newst = S_push_slab(aTHX);
6488 PL_regmatch_state = newst;
6490 locinput = pushinput;
6493 assert(0); /* NOTREACHED */
6498 * We get here only if there's trouble -- normally "case END" is
6499 * the terminating point.
6501 Perl_croak(aTHX_ "corrupted regexp pointers");
6507 /* we have successfully completed a subexpression, but we must now
6508 * pop to the state marked by yes_state and continue from there */
6509 assert(st != yes_state);
6511 while (st != yes_state) {
6513 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6514 PL_regmatch_slab = PL_regmatch_slab->prev;
6515 st = SLAB_LAST(PL_regmatch_slab);
6519 DEBUG_STATE_pp("pop (no final)");
6521 DEBUG_STATE_pp("pop (yes)");
6527 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6528 || yes_state > SLAB_LAST(PL_regmatch_slab))
6530 /* not in this slab, pop slab */
6531 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6532 PL_regmatch_slab = PL_regmatch_slab->prev;
6533 st = SLAB_LAST(PL_regmatch_slab);
6535 depth -= (st - yes_state);
6538 yes_state = st->u.yes.prev_yes_state;
6539 PL_regmatch_state = st;
6542 locinput= st->locinput;
6543 state_num = st->resume_state + no_final;
6544 goto reenter_switch;
6547 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6548 PL_colors[4], PL_colors[5]));
6550 if (reginfo->info_aux_eval) {
6551 /* each successfully executed (?{...}) block does the equivalent of
6552 * local $^R = do {...}
6553 * When popping the save stack, all these locals would be undone;
6554 * bypass this by setting the outermost saved $^R to the latest
6556 if (oreplsv != GvSV(PL_replgv))
6557 sv_setsv(oreplsv, GvSV(PL_replgv));
6564 PerlIO_printf(Perl_debug_log,
6565 "%*s %sfailed...%s\n",
6566 REPORT_CODE_OFF+depth*2, "",
6567 PL_colors[4], PL_colors[5])
6579 /* there's a previous state to backtrack to */
6581 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6582 PL_regmatch_slab = PL_regmatch_slab->prev;
6583 st = SLAB_LAST(PL_regmatch_slab);
6585 PL_regmatch_state = st;
6586 locinput= st->locinput;
6588 DEBUG_STATE_pp("pop");
6590 if (yes_state == st)
6591 yes_state = st->u.yes.prev_yes_state;
6593 state_num = st->resume_state + 1; /* failure = success + 1 */
6594 goto reenter_switch;
6599 if (rex->intflags & PREGf_VERBARG_SEEN) {
6600 SV *sv_err = get_sv("REGERROR", 1);
6601 SV *sv_mrk = get_sv("REGMARK", 1);
6603 sv_commit = &PL_sv_no;
6605 sv_yes_mark = &PL_sv_yes;
6608 sv_commit = &PL_sv_yes;
6609 sv_yes_mark = &PL_sv_no;
6611 sv_setsv(sv_err, sv_commit);
6612 sv_setsv(sv_mrk, sv_yes_mark);
6616 if (last_pushed_cv) {
6619 PERL_UNUSED_VAR(SP);
6622 assert(!result || locinput - reginfo->strbeg >= 0);
6623 return result ? locinput - reginfo->strbeg : -1;
6627 - regrepeat - repeatedly match something simple, report how many
6629 * What 'simple' means is a node which can be the operand of a quantifier like
6632 * startposp - pointer a pointer to the start position. This is updated
6633 * to point to the byte following the highest successful
6635 * p - the regnode to be repeatedly matched against.
6636 * reginfo - struct holding match state, such as strend
6637 * max - maximum number of things to match.
6638 * depth - (for debugging) backtracking depth.
6641 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
6642 regmatch_info *const reginfo, I32 max, int depth)
6645 char *scan; /* Pointer to current position in target string */
6647 char *loceol = reginfo->strend; /* local version */
6648 I32 hardcount = 0; /* How many matches so far */
6649 bool utf8_target = reginfo->is_utf8_target;
6650 int to_complement = 0; /* Invert the result? */
6652 _char_class_number classnum;
6654 PERL_UNUSED_ARG(depth);
6657 PERL_ARGS_ASSERT_REGREPEAT;
6660 if (max == REG_INFTY)
6662 else if (! utf8_target && loceol - scan > max)
6663 loceol = scan + max;
6665 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
6666 * to the maximum of how far we should go in it (leaving it set to the real
6667 * end, if the maximum permissible would take us beyond that). This allows
6668 * us to make the loop exit condition that we haven't gone past <loceol> to
6669 * also mean that we haven't exceeded the max permissible count, saving a
6670 * test each time through the loop. But it assumes that the OP matches a
6671 * single byte, which is true for most of the OPs below when applied to a
6672 * non-UTF-8 target. Those relatively few OPs that don't have this
6673 * characteristic will have to compensate.
6675 * There is no adjustment for UTF-8 targets, as the number of bytes per
6676 * character varies. OPs will have to test both that the count is less
6677 * than the max permissible (using <hardcount> to keep track), and that we
6678 * are still within the bounds of the string (using <loceol>. A few OPs
6679 * match a single byte no matter what the encoding. They can omit the max
6680 * test if, for the UTF-8 case, they do the adjustment that was skipped
6683 * Thus, the code above sets things up for the common case; and exceptional
6684 * cases need extra work; the common case is to make sure <scan> doesn't
6685 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
6686 * count doesn't exceed the maximum permissible */
6691 while (scan < loceol && hardcount < max && *scan != '\n') {
6692 scan += UTF8SKIP(scan);
6696 while (scan < loceol && *scan != '\n')
6702 while (scan < loceol && hardcount < max) {
6703 scan += UTF8SKIP(scan);
6710 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
6711 if (utf8_target && loceol - scan > max) {
6713 /* <loceol> hadn't been adjusted in the UTF-8 case */
6721 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6725 /* Can use a simple loop if the pattern char to match on is invariant
6726 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
6727 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
6728 * true iff it doesn't matter if the argument is in UTF-8 or not */
6729 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
6730 if (utf8_target && loceol - scan > max) {
6731 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6732 * since here, to match at all, 1 char == 1 byte */
6733 loceol = scan + max;
6735 while (scan < loceol && UCHARAT(scan) == c) {
6739 else if (reginfo->is_utf8_pat) {
6741 STRLEN scan_char_len;
6743 /* When both target and pattern are UTF-8, we have to do
6745 while (hardcount < max
6747 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
6748 && memEQ(scan, STRING(p), scan_char_len))
6750 scan += scan_char_len;
6754 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
6756 /* Target isn't utf8; convert the character in the UTF-8
6757 * pattern to non-UTF8, and do a simple loop */
6758 c = TWO_BYTE_UTF8_TO_UNI(c, *(STRING(p) + 1));
6759 while (scan < loceol && UCHARAT(scan) == c) {
6762 } /* else pattern char is above Latin1, can't possibly match the
6767 /* Here, the string must be utf8; pattern isn't, and <c> is
6768 * different in utf8 than not, so can't compare them directly.
6769 * Outside the loop, find the two utf8 bytes that represent c, and
6770 * then look for those in sequence in the utf8 string */
6771 U8 high = UTF8_TWO_BYTE_HI(c);
6772 U8 low = UTF8_TWO_BYTE_LO(c);
6774 while (hardcount < max
6775 && scan + 1 < loceol
6776 && UCHARAT(scan) == high
6777 && UCHARAT(scan + 1) == low)
6786 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6790 RXp_MATCH_TAINTED_on(prog);
6791 utf8_flags = FOLDEQ_UTF8_LOCALE;
6799 case EXACTFU_TRICKYFOLD:
6801 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6805 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
6807 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6809 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
6812 if (c1 == CHRTEST_VOID) {
6813 /* Use full Unicode fold matching */
6814 char *tmpeol = reginfo->strend;
6815 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
6816 while (hardcount < max
6817 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6818 STRING(p), NULL, pat_len,
6819 reginfo->is_utf8_pat, utf8_flags))
6822 tmpeol = reginfo->strend;
6826 else if (utf8_target) {
6828 while (scan < loceol
6830 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
6832 scan += UTF8SKIP(scan);
6837 while (scan < loceol
6839 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
6840 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
6842 scan += UTF8SKIP(scan);
6847 else if (c1 == c2) {
6848 while (scan < loceol && UCHARAT(scan) == c1) {
6853 while (scan < loceol &&
6854 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
6863 case ANYOF_WARN_SUPER:
6865 while (hardcount < max
6867 && reginclass(prog, p, (U8*)scan, utf8_target))
6869 scan += UTF8SKIP(scan);
6873 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6878 /* The argument (FLAGS) to all the POSIX node types is the class number */
6885 RXp_MATCH_TAINTED_on(prog);
6886 if (! utf8_target) {
6887 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
6893 while (hardcount < max && scan < loceol
6894 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
6897 scan += UTF8SKIP(scan);
6910 if (utf8_target && loceol - scan > max) {
6912 /* We didn't adjust <loceol> at the beginning of this routine
6913 * because is UTF-8, but it is actually ok to do so, since here, to
6914 * match, 1 char == 1 byte. */
6915 loceol = scan + max;
6917 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6930 if (! utf8_target) {
6931 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6937 /* The complement of something that matches only ASCII matches all
6938 * UTF-8 variant code points, plus everything in ASCII that isn't
6940 while (hardcount < max && scan < loceol
6941 && (! UTF8_IS_INVARIANT(*scan)
6942 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
6944 scan += UTF8SKIP(scan);
6955 if (! utf8_target) {
6956 while (scan < loceol && to_complement
6957 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
6964 classnum = (_char_class_number) FLAGS(p);
6965 if (classnum < _FIRST_NON_SWASH_CC) {
6967 /* Here, a swash is needed for above-Latin1 code points.
6968 * Process as many Latin1 code points using the built-in rules.
6969 * Go to another loop to finish processing upon encountering
6970 * the first Latin1 code point. We could do that in this loop
6971 * as well, but the other way saves having to test if the swash
6972 * has been loaded every time through the loop: extra space to
6974 while (hardcount < max && scan < loceol) {
6975 if (UTF8_IS_INVARIANT(*scan)) {
6976 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
6983 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
6984 if (! (to_complement
6985 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(*scan,
6994 goto found_above_latin1;
7001 /* For these character classes, the knowledge of how to handle
7002 * every code point is compiled in to Perl via a macro. This
7003 * code is written for making the loops as tight as possible.
7004 * It could be refactored to save space instead */
7006 case _CC_ENUM_SPACE: /* XXX would require separate code
7007 if we revert the change of \v
7010 case _CC_ENUM_PSXSPC:
7011 while (hardcount < max
7013 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7015 scan += UTF8SKIP(scan);
7019 case _CC_ENUM_BLANK:
7020 while (hardcount < max
7022 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7024 scan += UTF8SKIP(scan);
7028 case _CC_ENUM_XDIGIT:
7029 while (hardcount < max
7031 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7033 scan += UTF8SKIP(scan);
7037 case _CC_ENUM_VERTSPACE:
7038 while (hardcount < max
7040 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7042 scan += UTF8SKIP(scan);
7046 case _CC_ENUM_CNTRL:
7047 while (hardcount < max
7049 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7051 scan += UTF8SKIP(scan);
7056 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7062 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7064 /* Load the swash if not already present */
7065 if (! PL_utf8_swash_ptrs[classnum]) {
7066 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7067 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7068 "utf8", swash_property_names[classnum],
7069 &PL_sv_undef, 1, 0, NULL, &flags);
7072 while (hardcount < max && scan < loceol
7073 && to_complement ^ cBOOL(_generic_utf8(
7076 swash_fetch(PL_utf8_swash_ptrs[classnum],
7080 scan += UTF8SKIP(scan);
7087 while (hardcount < max && scan < loceol &&
7088 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7093 /* LNBREAK can match one or two latin chars, which is ok, but we
7094 * have to use hardcount in this situation, and throw away the
7095 * adjustment to <loceol> done before the switch statement */
7096 loceol = reginfo->strend;
7097 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7118 /* These are all 0 width, so match right here or not at all. */
7122 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7123 assert(0); /* NOTREACHED */
7130 c = scan - *startposp;
7134 GET_RE_DEBUG_FLAGS_DECL;
7136 SV * const prop = sv_newmortal();
7137 regprop(prog, prop, p);
7138 PerlIO_printf(Perl_debug_log,
7139 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7140 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7148 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7150 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7151 create a copy so that changes the caller makes won't change the shared one.
7152 If <altsvp> is non-null, will return NULL in it, for back-compat.
7155 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7157 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7163 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp));
7168 S_core_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp)
7170 /* Returns the swash for the input 'node' in the regex 'prog'.
7171 * If <doinit> is true, will attempt to create the swash if not already
7173 * If <listsvp> is non-null, will return the swash initialization string in
7175 * Tied intimately to how regcomp.c sets up the data structure */
7182 RXi_GET_DECL(prog,progi);
7183 const struct reg_data * const data = prog ? progi->data : NULL;
7185 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
7187 assert(ANYOF_NONBITMAP(node));
7189 if (data && data->count) {
7190 const U32 n = ARG(node);
7192 if (data->what[n] == 's') {
7193 SV * const rv = MUTABLE_SV(data->data[n]);
7194 AV * const av = MUTABLE_AV(SvRV(rv));
7195 SV **const ary = AvARRAY(av);
7196 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7198 si = *ary; /* ary[0] = the string to initialize the swash with */
7200 /* Elements 2 and 3 are either both present or both absent. [2] is
7201 * any inversion list generated at compile time; [3] indicates if
7202 * that inversion list has any user-defined properties in it. */
7203 if (av_len(av) >= 2) {
7206 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7213 /* Element [1] is reserved for the set-up swash. If already there,
7214 * return it; if not, create it and store it there */
7215 if (SvROK(ary[1])) {
7218 else if (si && doinit) {
7220 sw = _core_swash_init("utf8", /* the utf8 package */
7224 0, /* not from tr/// */
7227 (void)av_store(av, 1, sw);
7233 SV* matches_string = newSVpvn("", 0);
7235 /* Use the swash, if any, which has to have incorporated into it all
7237 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7238 && (si && si != &PL_sv_undef))
7241 /* If no swash, use the input initialization string, if available */
7242 sv_catsv(matches_string, si);
7245 /* Add the inversion list to whatever we have. This may have come from
7246 * the swash, or from an input parameter */
7248 sv_catsv(matches_string, _invlist_contents(invlist));
7250 *listsvp = matches_string;
7257 - reginclass - determine if a character falls into a character class
7259 n is the ANYOF regnode
7260 p is the target string
7261 utf8_target tells whether p is in UTF-8.
7263 Returns true if matched; false otherwise.
7265 Note that this can be a synthetic start class, a combination of various
7266 nodes, so things you think might be mutually exclusive, such as locale,
7267 aren't. It can match both locale and non-locale
7272 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const bool utf8_target)
7275 const char flags = ANYOF_FLAGS(n);
7279 PERL_ARGS_ASSERT_REGINCLASS;
7281 /* If c is not already the code point, get it. Note that
7282 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7283 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7285 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
7286 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7287 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7288 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7289 * UTF8_ALLOW_FFFF */
7290 if (c_len == (STRLEN)-1)
7291 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7294 /* If this character is potentially in the bitmap, check it */
7296 if (ANYOF_BITMAP_TEST(n, c))
7298 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
7304 else if (flags & ANYOF_LOCALE) {
7305 RXp_MATCH_TAINTED_on(prog);
7307 if ((flags & ANYOF_LOC_FOLD)
7308 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7312 else if (ANYOF_CLASS_TEST_ANY_SET(n)) {
7314 /* The data structure is arranged so bits 0, 2, 4, ... are set
7315 * if the class includes the Posix character class given by
7316 * bit/2; and 1, 3, 5, ... are set if the class includes the
7317 * complemented Posix class given by int(bit/2). So we loop
7318 * through the bits, each time changing whether we complement
7319 * the result or not. Suppose for the sake of illustration
7320 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7321 * is set, it means there is a match for this ANYOF node if the
7322 * character is in the class given by the expression (0 / 2 = 0
7323 * = \w). If it is in that class, isFOO_lc() will return 1,
7324 * and since 'to_complement' is 0, the result will stay TRUE,
7325 * and we exit the loop. Suppose instead that bit 0 is 0, but
7326 * bit 1 is 1. That means there is a match if the character
7327 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7328 * but will on bit 1. On the second iteration 'to_complement'
7329 * will be 1, so the exclusive or will reverse things, so we
7330 * are testing for \W. On the third iteration, 'to_complement'
7331 * will be 0, and we would be testing for \s; the fourth
7332 * iteration would test for \S, etc.
7334 * Note that this code assumes that all the classes are closed
7335 * under folding. For example, if a character matches \w, then
7336 * its fold does too; and vice versa. This should be true for
7337 * any well-behaved locale for all the currently defined Posix
7338 * classes, except for :lower: and :upper:, which are handled
7339 * by the pseudo-class :cased: which matches if either of the
7340 * other two does. To get rid of this assumption, an outer
7341 * loop could be used below to iterate over both the source
7342 * character, and its fold (if different) */
7345 int to_complement = 0;
7346 while (count < ANYOF_MAX) {
7347 if (ANYOF_CLASS_TEST(n, count)
7348 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7360 /* If the bitmap didn't (or couldn't) match, and something outside the
7361 * bitmap could match, try that. Locale nodes specify completely the
7362 * behavior of code points in the bit map (otherwise, a utf8 target would
7363 * cause them to be treated as Unicode and not locale), except in
7364 * the very unlikely event when this node is a synthetic start class, which
7365 * could be a combination of locale and non-locale nodes. So allow locale
7366 * to match for the synthetic start class, which will give a false
7367 * positive that will be resolved when the match is done again as not part
7368 * of the synthetic start class */
7370 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
7371 match = TRUE; /* Everything above 255 matches */
7373 else if (ANYOF_NONBITMAP(n)
7374 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
7377 || (! (flags & ANYOF_LOCALE))
7378 || OP(n) == ANYOF_SYNTHETIC))))
7380 SV * const sw = core_regclass_swash(prog, n, TRUE, 0);
7385 } else { /* Convert to utf8 */
7387 utf8_p = bytes_to_utf8(p, &len);
7390 if (swash_fetch(sw, utf8_p, TRUE)) {
7394 /* If we allocated a string above, free it */
7395 if (! utf8_target) Safefree(utf8_p);
7399 if (UNICODE_IS_SUPER(c)
7400 && OP(n) == ANYOF_WARN_SUPER
7401 && ckWARN_d(WARN_NON_UNICODE))
7403 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7404 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7408 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7409 return cBOOL(flags & ANYOF_INVERT) ^ match;
7413 S_reghop3(U8 *s, I32 off, const U8* lim)
7415 /* return the position 'off' UTF-8 characters away from 's', forward if
7416 * 'off' >= 0, backwards if negative. But don't go outside of position
7417 * 'lim', which better be < s if off < 0 */
7421 PERL_ARGS_ASSERT_REGHOP3;
7424 while (off-- && s < lim) {
7425 /* XXX could check well-formedness here */
7430 while (off++ && s > lim) {
7432 if (UTF8_IS_CONTINUED(*s)) {
7433 while (s > lim && UTF8_IS_CONTINUATION(*s))
7436 /* XXX could check well-formedness here */
7443 /* there are a bunch of places where we use two reghop3's that should
7444 be replaced with this routine. but since thats not done yet
7445 we ifdef it out - dmq
7448 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7452 PERL_ARGS_ASSERT_REGHOP4;
7455 while (off-- && s < rlim) {
7456 /* XXX could check well-formedness here */
7461 while (off++ && s > llim) {
7463 if (UTF8_IS_CONTINUED(*s)) {
7464 while (s > llim && UTF8_IS_CONTINUATION(*s))
7467 /* XXX could check well-formedness here */
7475 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7479 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7482 while (off-- && s < lim) {
7483 /* XXX could check well-formedness here */
7490 while (off++ && s > lim) {
7492 if (UTF8_IS_CONTINUED(*s)) {
7493 while (s > lim && UTF8_IS_CONTINUATION(*s))
7496 /* XXX could check well-formedness here */
7505 /* when executing a regex that may have (?{}), extra stuff needs setting
7506 up that will be visible to the called code, even before the current
7507 match has finished. In particular:
7509 * $_ is localised to the SV currently being matched;
7510 * pos($_) is created if necessary, ready to be updated on each call-out
7512 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
7513 isn't set until the current pattern is successfully finished), so that
7514 $1 etc of the match-so-far can be seen;
7515 * save the old values of subbeg etc of the current regex, and set then
7516 to the current string (again, this is normally only done at the end
7521 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
7524 regexp *const rex = ReANY(reginfo->prog);
7525 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
7527 eval_state->rex = rex;
7530 /* Make $_ available to executed code. */
7531 if (reginfo->sv != DEFSV) {
7533 DEFSV_set(reginfo->sv);
7536 if (!(mg = mg_find_mglob(reginfo->sv))) {
7537 /* prepare for quick setting of pos */
7538 mg = sv_magicext_mglob(reginfo->sv);
7541 eval_state->pos_magic = mg;
7542 eval_state->pos = mg->mg_len;
7545 eval_state->pos_magic = NULL;
7547 if (!PL_reg_curpm) {
7548 /* PL_reg_curpm is a fake PMOP that we can attach the current
7549 * regex to and point PL_curpm at, so that $1 et al are visible
7550 * within a /(?{})/. It's just allocated once per interpreter the
7551 * first time its needed */
7552 Newxz(PL_reg_curpm, 1, PMOP);
7555 SV* const repointer = &PL_sv_undef;
7556 /* this regexp is also owned by the new PL_reg_curpm, which
7557 will try to free it. */
7558 av_push(PL_regex_padav, repointer);
7559 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
7560 PL_regex_pad = AvARRAY(PL_regex_padav);
7564 SET_reg_curpm(reginfo->prog);
7565 eval_state->curpm = PL_curpm;
7566 PL_curpm = PL_reg_curpm;
7567 if (RXp_MATCH_COPIED(rex)) {
7568 /* Here is a serious problem: we cannot rewrite subbeg,
7569 since it may be needed if this match fails. Thus
7570 $` inside (?{}) could fail... */
7571 eval_state->subbeg = rex->subbeg;
7572 eval_state->sublen = rex->sublen;
7573 eval_state->suboffset = rex->suboffset;
7574 eval_state->subcoffset = rex->subcoffset;
7576 eval_state->saved_copy = rex->saved_copy;
7578 RXp_MATCH_COPIED_off(rex);
7581 eval_state->subbeg = NULL;
7582 rex->subbeg = (char *)reginfo->strbeg;
7584 rex->subcoffset = 0;
7585 rex->sublen = reginfo->strend - reginfo->strbeg;
7589 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
7592 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
7595 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
7596 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
7599 Safefree(aux->poscache);
7603 /* undo the effects of S_setup_eval_state() */
7605 if (eval_state->subbeg) {
7606 regexp * const rex = eval_state->rex;
7607 rex->subbeg = eval_state->subbeg;
7608 rex->sublen = eval_state->sublen;
7609 rex->suboffset = eval_state->suboffset;
7610 rex->subcoffset = eval_state->subcoffset;
7612 rex->saved_copy = eval_state->saved_copy;
7614 RXp_MATCH_COPIED_on(rex);
7616 if (eval_state->pos_magic)
7617 eval_state->pos_magic->mg_len = eval_state->pos;
7619 PL_curpm = eval_state->curpm;
7622 PL_regmatch_state = aux->old_regmatch_state;
7623 PL_regmatch_slab = aux->old_regmatch_slab;
7625 /* free all slabs above current one - this must be the last action
7626 * of this function, as aux and eval_state are allocated within
7627 * slabs and may be freed here */
7629 s = PL_regmatch_slab->next;
7631 PL_regmatch_slab->next = NULL;
7633 regmatch_slab * const osl = s;
7642 S_to_utf8_substr(pTHX_ regexp *prog)
7644 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
7645 * on the converted value */
7649 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7652 if (prog->substrs->data[i].substr
7653 && !prog->substrs->data[i].utf8_substr) {
7654 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7655 prog->substrs->data[i].utf8_substr = sv;
7656 sv_utf8_upgrade(sv);
7657 if (SvVALID(prog->substrs->data[i].substr)) {
7658 if (SvTAIL(prog->substrs->data[i].substr)) {
7659 /* Trim the trailing \n that fbm_compile added last
7661 SvCUR_set(sv, SvCUR(sv) - 1);
7662 /* Whilst this makes the SV technically "invalid" (as its
7663 buffer is no longer followed by "\0") when fbm_compile()
7664 adds the "\n" back, a "\0" is restored. */
7665 fbm_compile(sv, FBMcf_TAIL);
7669 if (prog->substrs->data[i].substr == prog->check_substr)
7670 prog->check_utf8 = sv;
7676 S_to_byte_substr(pTHX_ regexp *prog)
7678 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
7679 * on the converted value; returns FALSE if can't be converted. */
7684 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7687 if (prog->substrs->data[i].utf8_substr
7688 && !prog->substrs->data[i].substr) {
7689 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7690 if (! sv_utf8_downgrade(sv, TRUE)) {
7693 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7694 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7695 /* Trim the trailing \n that fbm_compile added last
7697 SvCUR_set(sv, SvCUR(sv) - 1);
7698 fbm_compile(sv, FBMcf_TAIL);
7702 prog->substrs->data[i].substr = sv;
7703 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7704 prog->check_substr = sv;
7713 * c-indentation-style: bsd
7715 * indent-tabs-mode: nil
7718 * ex: set ts=8 sts=4 sw=4 et: