5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
103 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
104 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
106 #define HOPc(pos,off) \
107 (char *)(PL_reg_match_utf8 \
108 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
110 #define HOPBACKc(pos, off) \
111 (char*)(PL_reg_match_utf8\
112 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
113 : (pos - off >= PL_bostr) \
117 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
118 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
120 /* these are unrolled below in the CCC_TRY_XXX defined */
121 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
122 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
124 /* Doesn't do an assert to verify that is correct */
125 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
126 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END
128 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
129 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
130 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
132 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
133 LOAD_UTF8_CHARCLASS(X_begin, " "); \
134 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
135 /* These are utf8 constants, and not utf-ebcdic constants, so the \
136 * assert should likely and hopefully fail on an EBCDIC machine */ \
137 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
139 /* No asserts are done for these, in case called on an early \
140 * Unicode version in which they map to nothing */ \
141 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
142 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
143 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
144 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
145 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
146 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
147 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
150 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
151 so that it is possible to override the option here without having to
152 rebuild the entire core. as we are required to do if we change regcomp.h
153 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
155 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
156 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
159 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
160 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS_ALNUM()
161 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS_SPACE()
162 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS_DIGIT()
163 #define RE_utf8_perl_word PL_utf8_alnum
164 #define RE_utf8_perl_space PL_utf8_space
165 #define RE_utf8_posix_digit PL_utf8_digit
166 #define perl_word alnum
167 #define perl_space space
168 #define posix_digit digit
170 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS(perl_word,"a")
171 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS(perl_space," ")
172 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS(posix_digit,"0")
173 #define RE_utf8_perl_word PL_utf8_perl_word
174 #define RE_utf8_perl_space PL_utf8_perl_space
175 #define RE_utf8_posix_digit PL_utf8_posix_digit
179 #define CCC_TRY_AFF(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
181 PL_reg_flags |= RF_tainted; \
186 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
187 if (!CAT2(PL_utf8_,CLASS)) { \
191 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
195 if (!(OP(scan) == NAME \
196 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
197 : LCFUNC_utf8((U8*)locinput))) \
201 locinput += PL_utf8skip[nextchr]; \
202 nextchr = UCHARAT(locinput); \
205 if (!(OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
207 nextchr = UCHARAT(++locinput); \
210 #define CCC_TRY_NEG(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
212 PL_reg_flags |= RF_tainted; \
215 if (!nextchr && locinput >= PL_regeol) \
217 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
218 if (!CAT2(PL_utf8_,CLASS)) { \
222 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
226 if ((OP(scan) == NAME \
227 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
228 : LCFUNC_utf8((U8*)locinput))) \
232 locinput += PL_utf8skip[nextchr]; \
233 nextchr = UCHARAT(locinput); \
236 if ((OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
238 nextchr = UCHARAT(++locinput); \
245 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
247 /* for use after a quantifier and before an EXACT-like node -- japhy */
248 /* it would be nice to rework regcomp.sym to generate this stuff. sigh */
249 #define JUMPABLE(rn) ( \
251 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
253 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
254 OP(rn) == PLUS || OP(rn) == MINMOD || \
255 OP(rn) == KEEPS || (PL_regkind[OP(rn)] == VERB) || \
256 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
258 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
260 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
263 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
264 we don't need this definition. */
265 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
266 #define IS_TEXTF(rn) ( OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
267 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
270 /* ... so we use this as its faster. */
271 #define IS_TEXT(rn) ( OP(rn)==EXACT )
272 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
273 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
278 Search for mandatory following text node; for lookahead, the text must
279 follow but for lookbehind (rn->flags != 0) we skip to the next step.
281 #define FIND_NEXT_IMPT(rn) STMT_START { \
282 while (JUMPABLE(rn)) { \
283 const OPCODE type = OP(rn); \
284 if (type == SUSPEND || PL_regkind[type] == CURLY) \
285 rn = NEXTOPER(NEXTOPER(rn)); \
286 else if (type == PLUS) \
288 else if (type == IFMATCH) \
289 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
290 else rn += NEXT_OFF(rn); \
295 static void restore_pos(pTHX_ void *arg);
297 #define REGCP_PAREN_ELEMS 4
298 #define REGCP_OTHER_ELEMS 5
299 #define REGCP_FRAME_ELEMS 1
300 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
301 * are needed for the regexp context stack bookkeeping. */
304 S_regcppush(pTHX_ I32 parenfloor)
307 const int retval = PL_savestack_ix;
308 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
309 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
310 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
312 GET_RE_DEBUG_FLAGS_DECL;
314 if (paren_elems_to_push < 0)
315 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
317 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
318 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
319 " out of range (%lu-%ld)",
320 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
322 SSGROW(total_elems + REGCP_FRAME_ELEMS);
324 for (p = PL_regsize; p > parenfloor; p--) {
325 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
326 SSPUSHINT(PL_regoffs[p].end);
327 SSPUSHINT(PL_regoffs[p].start);
328 SSPUSHPTR(PL_reg_start_tmp[p]);
330 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
331 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
332 (UV)p, (IV)PL_regoffs[p].start,
333 (IV)(PL_reg_start_tmp[p] - PL_bostr),
334 (IV)PL_regoffs[p].end
337 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
338 SSPUSHPTR(PL_regoffs);
339 SSPUSHINT(PL_regsize);
340 SSPUSHINT(*PL_reglastparen);
341 SSPUSHINT(*PL_reglastcloseparen);
342 SSPUSHPTR(PL_reginput);
343 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
348 /* These are needed since we do not localize EVAL nodes: */
349 #define REGCP_SET(cp) \
351 PerlIO_printf(Perl_debug_log, \
352 " Setting an EVAL scope, savestack=%"IVdf"\n", \
353 (IV)PL_savestack_ix)); \
356 #define REGCP_UNWIND(cp) \
358 if (cp != PL_savestack_ix) \
359 PerlIO_printf(Perl_debug_log, \
360 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
361 (IV)(cp), (IV)PL_savestack_ix)); \
365 S_regcppop(pTHX_ const regexp *rex)
370 GET_RE_DEBUG_FLAGS_DECL;
372 PERL_ARGS_ASSERT_REGCPPOP;
374 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
376 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
377 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
378 input = (char *) SSPOPPTR;
379 *PL_reglastcloseparen = SSPOPINT;
380 *PL_reglastparen = SSPOPINT;
381 PL_regsize = SSPOPINT;
382 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
384 i -= REGCP_OTHER_ELEMS;
385 /* Now restore the parentheses context. */
386 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
388 U32 paren = (U32)SSPOPINT;
389 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
390 PL_regoffs[paren].start = SSPOPINT;
392 if (paren <= *PL_reglastparen)
393 PL_regoffs[paren].end = tmps;
395 PerlIO_printf(Perl_debug_log,
396 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
397 (UV)paren, (IV)PL_regoffs[paren].start,
398 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
399 (IV)PL_regoffs[paren].end,
400 (paren > *PL_reglastparen ? "(no)" : ""));
404 if (*PL_reglastparen + 1 <= rex->nparens) {
405 PerlIO_printf(Perl_debug_log,
406 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
407 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
411 /* It would seem that the similar code in regtry()
412 * already takes care of this, and in fact it is in
413 * a better location to since this code can #if 0-ed out
414 * but the code in regtry() is needed or otherwise tests
415 * requiring null fields (pat.t#187 and split.t#{13,14}
416 * (as of patchlevel 7877) will fail. Then again,
417 * this code seems to be necessary or otherwise
418 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
419 * --jhi updated by dapm */
420 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
422 PL_regoffs[i].start = -1;
423 PL_regoffs[i].end = -1;
429 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
432 * pregexec and friends
435 #ifndef PERL_IN_XSUB_RE
437 - pregexec - match a regexp against a string
440 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
441 char *strbeg, I32 minend, SV *screamer, U32 nosave)
442 /* strend: pointer to null at end of string */
443 /* strbeg: real beginning of string */
444 /* minend: end of match must be >=minend after stringarg. */
445 /* nosave: For optimizations. */
447 PERL_ARGS_ASSERT_PREGEXEC;
450 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
451 nosave ? 0 : REXEC_COPY_STR);
456 * Need to implement the following flags for reg_anch:
458 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
460 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
461 * INTUIT_AUTORITATIVE_ML
462 * INTUIT_ONCE_NOML - Intuit can match in one location only.
465 * Another flag for this function: SECOND_TIME (so that float substrs
466 * with giant delta may be not rechecked).
469 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
471 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
472 Otherwise, only SvCUR(sv) is used to get strbeg. */
474 /* XXXX We assume that strpos is strbeg unless sv. */
476 /* XXXX Some places assume that there is a fixed substring.
477 An update may be needed if optimizer marks as "INTUITable"
478 RExen without fixed substrings. Similarly, it is assumed that
479 lengths of all the strings are no more than minlen, thus they
480 cannot come from lookahead.
481 (Or minlen should take into account lookahead.)
482 NOTE: Some of this comment is not correct. minlen does now take account
483 of lookahead/behind. Further research is required. -- demerphq
487 /* A failure to find a constant substring means that there is no need to make
488 an expensive call to REx engine, thus we celebrate a failure. Similarly,
489 finding a substring too deep into the string means that less calls to
490 regtry() should be needed.
492 REx compiler's optimizer found 4 possible hints:
493 a) Anchored substring;
495 c) Whether we are anchored (beginning-of-line or \G);
496 d) First node (of those at offset 0) which may distingush positions;
497 We use a)b)d) and multiline-part of c), and try to find a position in the
498 string which does not contradict any of them.
501 /* Most of decisions we do here should have been done at compile time.
502 The nodes of the REx which we used for the search should have been
503 deleted from the finite automaton. */
506 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
507 char *strend, const U32 flags, re_scream_pos_data *data)
510 struct regexp *const prog = (struct regexp *)SvANY(rx);
511 register I32 start_shift = 0;
512 /* Should be nonnegative! */
513 register I32 end_shift = 0;
518 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
520 register char *other_last = NULL; /* other substr checked before this */
521 char *check_at = NULL; /* check substr found at this pos */
522 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
523 RXi_GET_DECL(prog,progi);
525 const char * const i_strpos = strpos;
527 GET_RE_DEBUG_FLAGS_DECL;
529 PERL_ARGS_ASSERT_RE_INTUIT_START;
531 RX_MATCH_UTF8_set(rx,utf8_target);
534 PL_reg_flags |= RF_utf8;
537 debug_start_match(rx, utf8_target, strpos, strend,
538 sv ? "Guessing start of match in sv for"
539 : "Guessing start of match in string for");
542 /* CHR_DIST() would be more correct here but it makes things slow. */
543 if (prog->minlen > strend - strpos) {
544 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
545 "String too short... [re_intuit_start]\n"));
549 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
552 if (!prog->check_utf8 && prog->check_substr)
553 to_utf8_substr(prog);
554 check = prog->check_utf8;
556 if (!prog->check_substr && prog->check_utf8)
557 to_byte_substr(prog);
558 check = prog->check_substr;
560 if (check == &PL_sv_undef) {
561 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
562 "Non-utf8 string cannot match utf8 check string\n"));
565 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
566 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
567 || ( (prog->extflags & RXf_ANCH_BOL)
568 && !multiline ) ); /* Check after \n? */
571 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
572 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
573 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
575 && (strpos != strbeg)) {
576 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
579 if (prog->check_offset_min == prog->check_offset_max &&
580 !(prog->extflags & RXf_CANY_SEEN)) {
581 /* Substring at constant offset from beg-of-str... */
584 s = HOP3c(strpos, prog->check_offset_min, strend);
587 slen = SvCUR(check); /* >= 1 */
589 if ( strend - s > slen || strend - s < slen - 1
590 || (strend - s == slen && strend[-1] != '\n')) {
591 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
594 /* Now should match s[0..slen-2] */
596 if (slen && (*SvPVX_const(check) != *s
598 && memNE(SvPVX_const(check), s, slen)))) {
600 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
604 else if (*SvPVX_const(check) != *s
605 || ((slen = SvCUR(check)) > 1
606 && memNE(SvPVX_const(check), s, slen)))
609 goto success_at_start;
612 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
614 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
615 end_shift = prog->check_end_shift;
618 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
619 - (SvTAIL(check) != 0);
620 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
622 if (end_shift < eshift)
626 else { /* Can match at random position */
629 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
630 end_shift = prog->check_end_shift;
632 /* end shift should be non negative here */
635 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
637 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
638 (IV)end_shift, RX_PRECOMP(prog));
642 /* Find a possible match in the region s..strend by looking for
643 the "check" substring in the region corrected by start/end_shift. */
646 I32 srch_start_shift = start_shift;
647 I32 srch_end_shift = end_shift;
648 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
649 srch_end_shift -= ((strbeg - s) - srch_start_shift);
650 srch_start_shift = strbeg - s;
652 DEBUG_OPTIMISE_MORE_r({
653 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
654 (IV)prog->check_offset_min,
655 (IV)srch_start_shift,
657 (IV)prog->check_end_shift);
660 if (flags & REXEC_SCREAM) {
661 I32 p = -1; /* Internal iterator of scream. */
662 I32 * const pp = data ? data->scream_pos : &p;
664 if (PL_screamfirst[BmRARE(check)] >= 0
665 || ( BmRARE(check) == '\n'
666 && (BmPREVIOUS(check) == SvCUR(check) - 1)
668 s = screaminstr(sv, check,
669 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
672 /* we may be pointing at the wrong string */
673 if (s && RXp_MATCH_COPIED(prog))
674 s = strbeg + (s - SvPVX_const(sv));
676 *data->scream_olds = s;
681 if (prog->extflags & RXf_CANY_SEEN) {
682 start_point= (U8*)(s + srch_start_shift);
683 end_point= (U8*)(strend - srch_end_shift);
685 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
686 end_point= HOP3(strend, -srch_end_shift, strbeg);
688 DEBUG_OPTIMISE_MORE_r({
689 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
690 (int)(end_point - start_point),
691 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
695 s = fbm_instr( start_point, end_point,
696 check, multiline ? FBMrf_MULTILINE : 0);
699 /* Update the count-of-usability, remove useless subpatterns,
703 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
704 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
705 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
706 (s ? "Found" : "Did not find"),
707 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
708 ? "anchored" : "floating"),
711 (s ? " at offset " : "...\n") );
716 /* Finish the diagnostic message */
717 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
719 /* XXX dmq: first branch is for positive lookbehind...
720 Our check string is offset from the beginning of the pattern.
721 So we need to do any stclass tests offset forward from that
730 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
731 Start with the other substr.
732 XXXX no SCREAM optimization yet - and a very coarse implementation
733 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
734 *always* match. Probably should be marked during compile...
735 Probably it is right to do no SCREAM here...
738 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
739 : (prog->float_substr && prog->anchored_substr))
741 /* Take into account the "other" substring. */
742 /* XXXX May be hopelessly wrong for UTF... */
745 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
748 char * const last = HOP3c(s, -start_shift, strbeg);
750 char * const saved_s = s;
753 t = s - prog->check_offset_max;
754 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
756 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
761 t = HOP3c(t, prog->anchored_offset, strend);
762 if (t < other_last) /* These positions already checked */
764 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
767 /* XXXX It is not documented what units *_offsets are in.
768 We assume bytes, but this is clearly wrong.
769 Meaning this code needs to be carefully reviewed for errors.
773 /* On end-of-str: see comment below. */
774 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
775 if (must == &PL_sv_undef) {
777 DEBUG_r(must = prog->anchored_utf8); /* for debug */
782 HOP3(HOP3(last1, prog->anchored_offset, strend)
783 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
785 multiline ? FBMrf_MULTILINE : 0
788 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
789 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
790 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
791 (s ? "Found" : "Contradicts"),
792 quoted, RE_SV_TAIL(must));
797 if (last1 >= last2) {
798 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
799 ", giving up...\n"));
802 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
803 ", trying floating at offset %ld...\n",
804 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
805 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
806 s = HOP3c(last, 1, strend);
810 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
811 (long)(s - i_strpos)));
812 t = HOP3c(s, -prog->anchored_offset, strbeg);
813 other_last = HOP3c(s, 1, strend);
821 else { /* Take into account the floating substring. */
823 char * const saved_s = s;
826 t = HOP3c(s, -start_shift, strbeg);
828 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
829 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
830 last = HOP3c(t, prog->float_max_offset, strend);
831 s = HOP3c(t, prog->float_min_offset, strend);
834 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
835 must = utf8_target ? prog->float_utf8 : prog->float_substr;
836 /* fbm_instr() takes into account exact value of end-of-str
837 if the check is SvTAIL(ed). Since false positives are OK,
838 and end-of-str is not later than strend we are OK. */
839 if (must == &PL_sv_undef) {
841 DEBUG_r(must = prog->float_utf8); /* for debug message */
844 s = fbm_instr((unsigned char*)s,
845 (unsigned char*)last + SvCUR(must)
847 must, multiline ? FBMrf_MULTILINE : 0);
849 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
850 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
851 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
852 (s ? "Found" : "Contradicts"),
853 quoted, RE_SV_TAIL(must));
857 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
858 ", giving up...\n"));
861 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
862 ", trying anchored starting at offset %ld...\n",
863 (long)(saved_s + 1 - i_strpos)));
865 s = HOP3c(t, 1, strend);
869 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
870 (long)(s - i_strpos)));
871 other_last = s; /* Fix this later. --Hugo */
881 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
883 DEBUG_OPTIMISE_MORE_r(
884 PerlIO_printf(Perl_debug_log,
885 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
886 (IV)prog->check_offset_min,
887 (IV)prog->check_offset_max,
895 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
897 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
900 /* Fixed substring is found far enough so that the match
901 cannot start at strpos. */
903 if (ml_anch && t[-1] != '\n') {
904 /* Eventually fbm_*() should handle this, but often
905 anchored_offset is not 0, so this check will not be wasted. */
906 /* XXXX In the code below we prefer to look for "^" even in
907 presence of anchored substrings. And we search even
908 beyond the found float position. These pessimizations
909 are historical artefacts only. */
911 while (t < strend - prog->minlen) {
913 if (t < check_at - prog->check_offset_min) {
914 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
915 /* Since we moved from the found position,
916 we definitely contradict the found anchored
917 substr. Due to the above check we do not
918 contradict "check" substr.
919 Thus we can arrive here only if check substr
920 is float. Redo checking for "other"=="fixed".
923 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
924 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
925 goto do_other_anchored;
927 /* We don't contradict the found floating substring. */
928 /* XXXX Why not check for STCLASS? */
930 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
931 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
934 /* Position contradicts check-string */
935 /* XXXX probably better to look for check-string
936 than for "\n", so one should lower the limit for t? */
937 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
938 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
939 other_last = strpos = s = t + 1;
944 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
945 PL_colors[0], PL_colors[1]));
949 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
950 PL_colors[0], PL_colors[1]));
954 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
957 /* The found string does not prohibit matching at strpos,
958 - no optimization of calling REx engine can be performed,
959 unless it was an MBOL and we are not after MBOL,
960 or a future STCLASS check will fail this. */
962 /* Even in this situation we may use MBOL flag if strpos is offset
963 wrt the start of the string. */
964 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
965 && (strpos != strbeg) && strpos[-1] != '\n'
966 /* May be due to an implicit anchor of m{.*foo} */
967 && !(prog->intflags & PREGf_IMPLICIT))
972 DEBUG_EXECUTE_r( if (ml_anch)
973 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
974 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
977 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
979 prog->check_utf8 /* Could be deleted already */
980 && --BmUSEFUL(prog->check_utf8) < 0
981 && (prog->check_utf8 == prog->float_utf8)
983 prog->check_substr /* Could be deleted already */
984 && --BmUSEFUL(prog->check_substr) < 0
985 && (prog->check_substr == prog->float_substr)
988 /* If flags & SOMETHING - do not do it many times on the same match */
989 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
990 /* XXX Does the destruction order has to change with utf8_target? */
991 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
992 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
993 prog->check_substr = prog->check_utf8 = NULL; /* disable */
994 prog->float_substr = prog->float_utf8 = NULL; /* clear */
995 check = NULL; /* abort */
997 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevent flag
998 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
999 if (prog->intflags & PREGf_IMPLICIT)
1000 prog->extflags &= ~RXf_ANCH_MBOL;
1001 /* XXXX This is a remnant of the old implementation. It
1002 looks wasteful, since now INTUIT can use many
1003 other heuristics. */
1004 prog->extflags &= ~RXf_USE_INTUIT;
1005 /* XXXX What other flags might need to be cleared in this branch? */
1011 /* Last resort... */
1012 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1013 /* trie stclasses are too expensive to use here, we are better off to
1014 leave it to regmatch itself */
1015 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1016 /* minlen == 0 is possible if regstclass is \b or \B,
1017 and the fixed substr is ''$.
1018 Since minlen is already taken into account, s+1 is before strend;
1019 accidentally, minlen >= 1 guaranties no false positives at s + 1
1020 even for \b or \B. But (minlen? 1 : 0) below assumes that
1021 regstclass does not come from lookahead... */
1022 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1023 This leaves EXACTF only, which is dealt with in find_byclass(). */
1024 const U8* const str = (U8*)STRING(progi->regstclass);
1025 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1026 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1029 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1030 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1031 else if (prog->float_substr || prog->float_utf8)
1032 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1036 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1037 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1040 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1043 const char *what = NULL;
1045 if (endpos == strend) {
1046 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1047 "Could not match STCLASS...\n") );
1050 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1051 "This position contradicts STCLASS...\n") );
1052 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1054 /* Contradict one of substrings */
1055 if (prog->anchored_substr || prog->anchored_utf8) {
1056 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1057 DEBUG_EXECUTE_r( what = "anchored" );
1059 s = HOP3c(t, 1, strend);
1060 if (s + start_shift + end_shift > strend) {
1061 /* XXXX Should be taken into account earlier? */
1062 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1063 "Could not match STCLASS...\n") );
1068 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1069 "Looking for %s substr starting at offset %ld...\n",
1070 what, (long)(s + start_shift - i_strpos)) );
1073 /* Have both, check_string is floating */
1074 if (t + start_shift >= check_at) /* Contradicts floating=check */
1075 goto retry_floating_check;
1076 /* Recheck anchored substring, but not floating... */
1080 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1081 "Looking for anchored substr starting at offset %ld...\n",
1082 (long)(other_last - i_strpos)) );
1083 goto do_other_anchored;
1085 /* Another way we could have checked stclass at the
1086 current position only: */
1091 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1092 "Looking for /%s^%s/m starting at offset %ld...\n",
1093 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1096 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1098 /* Check is floating subtring. */
1099 retry_floating_check:
1100 t = check_at - start_shift;
1101 DEBUG_EXECUTE_r( what = "floating" );
1102 goto hop_and_restart;
1105 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1106 "By STCLASS: moving %ld --> %ld\n",
1107 (long)(t - i_strpos), (long)(s - i_strpos))
1111 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1112 "Does not contradict STCLASS...\n");
1117 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1118 PL_colors[4], (check ? "Guessed" : "Giving up"),
1119 PL_colors[5], (long)(s - i_strpos)) );
1122 fail_finish: /* Substring not found */
1123 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1124 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1126 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1127 PL_colors[4], PL_colors[5]));
1131 #define DECL_TRIE_TYPE(scan) \
1132 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1133 trie_type = (scan->flags != EXACT) \
1134 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1135 : (utf8_target ? trie_utf8 : trie_plain)
1137 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1138 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1139 switch (trie_type) { \
1140 case trie_utf8_fold: \
1141 if ( foldlen>0 ) { \
1142 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1147 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1148 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1149 foldlen -= UNISKIP( uvc ); \
1150 uscan = foldbuf + UNISKIP( uvc ); \
1153 case trie_latin_utf8_fold: \
1154 if ( foldlen>0 ) { \
1155 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1161 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1162 foldlen -= UNISKIP( uvc ); \
1163 uscan = foldbuf + UNISKIP( uvc ); \
1167 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1174 charid = trie->charmap[ uvc ]; \
1178 if (widecharmap) { \
1179 SV** const svpp = hv_fetch(widecharmap, \
1180 (char*)&uvc, sizeof(UV), 0); \
1182 charid = (U16)SvIV(*svpp); \
1187 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1189 char *my_strend= (char *)strend; \
1192 foldEQ_utf8(s, &my_strend, 0, utf8_target, \
1193 m, NULL, ln, cBOOL(UTF_PATTERN))) \
1194 && (!reginfo || regtry(reginfo, &s)) ) \
1197 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1198 uvchr_to_utf8(tmpbuf, c); \
1199 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1201 && (f == c1 || f == c2) \
1203 foldEQ_utf8(s, &my_strend, 0, utf8_target,\
1204 m, NULL, ln, cBOOL(UTF_PATTERN)))\
1205 && (!reginfo || regtry(reginfo, &s)) ) \
1211 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1215 && (ln == 1 || (OP(c) == EXACTF \
1216 ? foldEQ(s, m, ln) \
1217 : foldEQ_locale(s, m, ln))) \
1218 && (!reginfo || regtry(reginfo, &s)) ) \
1224 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1226 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1232 #define REXEC_FBC_SCAN(CoDe) \
1234 while (s < strend) { \
1240 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1241 REXEC_FBC_UTF8_SCAN( \
1243 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1252 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1255 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1264 #define REXEC_FBC_TRYIT \
1265 if ((!reginfo || regtry(reginfo, &s))) \
1268 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1269 if (utf8_target) { \
1270 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1273 REXEC_FBC_CLASS_SCAN(CoNd); \
1277 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1278 if (utf8_target) { \
1280 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1283 REXEC_FBC_CLASS_SCAN(CoNd); \
1287 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1288 PL_reg_flags |= RF_tainted; \
1289 if (utf8_target) { \
1290 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1293 REXEC_FBC_CLASS_SCAN(CoNd); \
1297 #define DUMP_EXEC_POS(li,s,doutf8) \
1298 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1300 /* We know what class REx starts with. Try to find this position... */
1301 /* if reginfo is NULL, its a dryrun */
1302 /* annoyingly all the vars in this routine have different names from their counterparts
1303 in regmatch. /grrr */
1306 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1307 const char *strend, regmatch_info *reginfo)
1310 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1314 register STRLEN uskip;
1318 register I32 tmp = 1; /* Scratch variable? */
1319 register const bool utf8_target = PL_reg_match_utf8;
1320 RXi_GET_DECL(prog,progi);
1322 PERL_ARGS_ASSERT_FIND_BYCLASS;
1324 /* We know what class it must start with. */
1328 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1329 !UTF8_IS_INVARIANT((U8)s[0]) ?
1330 reginclass(prog, c, (U8*)s, 0, utf8_target) :
1331 REGINCLASS(prog, c, (U8*)s));
1334 while (s < strend) {
1337 if (REGINCLASS(prog, c, (U8*)s) ||
1338 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1339 /* The assignment of 2 is intentional:
1340 * for the folded sharp s, the skip is 2. */
1341 (skip = SHARP_S_SKIP))) {
1342 if (tmp && (!reginfo || regtry(reginfo, &s)))
1355 if (tmp && (!reginfo || regtry(reginfo, &s)))
1363 ln = STR_LEN(c); /* length to match in octets/bytes */
1364 lnc = (I32) ln; /* length to match in characters */
1366 STRLEN ulen1, ulen2;
1368 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1369 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1370 /* used by commented-out code below */
1371 /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
1373 /* XXX: Since the node will be case folded at compile
1374 time this logic is a little odd, although im not
1375 sure that its actually wrong. --dmq */
1377 c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1378 c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1380 /* XXX: This is kinda strange. to_utf8_XYZ returns the
1381 codepoint of the first character in the converted
1382 form, yet originally we did the extra step.
1383 No tests fail by commenting this code out however
1384 so Ive left it out. -- dmq.
1386 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1388 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1393 while (sm < ((U8 *) m + ln)) {
1408 c2 = PL_fold_locale[c1];
1410 e = HOP3c(strend, -((I32)lnc), s);
1412 if (!reginfo && e < s)
1413 e = s; /* Due to minlen logic of intuit() */
1415 /* The idea in the EXACTF* cases is to first find the
1416 * first character of the EXACTF* node and then, if
1417 * necessary, case-insensitively compare the full
1418 * text of the node. The c1 and c2 are the first
1419 * characters (though in Unicode it gets a bit
1420 * more complicated because there are more cases
1421 * than just upper and lower: one needs to use
1422 * the so-called folding case for case-insensitive
1423 * matching (called "loose matching" in Unicode).
1424 * foldEQ_utf8() will do just that. */
1426 if (utf8_target || UTF_PATTERN) {
1428 U8 tmpbuf [UTF8_MAXBYTES+1];
1431 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1433 /* Upper and lower of 1st char are equal -
1434 * probably not a "letter". */
1437 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1442 REXEC_FBC_EXACTISH_CHECK(c == c1);
1448 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1454 /* Handle some of the three Greek sigmas cases.
1455 * Note that not all the possible combinations
1456 * are handled here: some of them are handled
1457 * by the standard folding rules, and some of
1458 * them (the character class or ANYOF cases)
1459 * are handled during compiletime in
1460 * regexec.c:S_regclass(). */
1461 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1462 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1463 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1465 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1470 /* Neither pattern nor string are UTF8 */
1472 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1474 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1478 PL_reg_flags |= RF_tainted;
1485 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1486 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1488 tmp = ((OP(c) == BOUND ?
1489 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1490 LOAD_UTF8_CHARCLASS_ALNUM();
1491 REXEC_FBC_UTF8_SCAN(
1492 if (tmp == !(OP(c) == BOUND ?
1493 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1494 isALNUM_LC_utf8((U8*)s)))
1502 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1503 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1506 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1512 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1516 PL_reg_flags |= RF_tainted;
1523 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1524 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1526 tmp = ((OP(c) == NBOUND ?
1527 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1528 LOAD_UTF8_CHARCLASS_ALNUM();
1529 REXEC_FBC_UTF8_SCAN(
1530 if (tmp == !(OP(c) == NBOUND ?
1531 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1532 isALNUM_LC_utf8((U8*)s)))
1534 else REXEC_FBC_TRYIT;
1538 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1539 tmp = ((OP(c) == NBOUND ?
1540 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1543 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1545 else REXEC_FBC_TRYIT;
1548 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1552 REXEC_FBC_CSCAN_PRELOAD(
1553 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1554 swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1558 REXEC_FBC_CSCAN_TAINT(
1559 isALNUM_LC_utf8((U8*)s),
1563 REXEC_FBC_CSCAN_PRELOAD(
1564 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1565 !swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1569 REXEC_FBC_CSCAN_TAINT(
1570 !isALNUM_LC_utf8((U8*)s),
1574 REXEC_FBC_CSCAN_PRELOAD(
1575 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1576 *s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target),
1580 REXEC_FBC_CSCAN_TAINT(
1581 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1585 REXEC_FBC_CSCAN_PRELOAD(
1586 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1587 !(*s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target)),
1591 REXEC_FBC_CSCAN_TAINT(
1592 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1596 REXEC_FBC_CSCAN_PRELOAD(
1597 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1598 swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1602 REXEC_FBC_CSCAN_TAINT(
1603 isDIGIT_LC_utf8((U8*)s),
1607 REXEC_FBC_CSCAN_PRELOAD(
1608 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1609 !swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1613 REXEC_FBC_CSCAN_TAINT(
1614 !isDIGIT_LC_utf8((U8*)s),
1620 is_LNBREAK_latin1(s)
1630 !is_VERTWS_latin1(s)
1635 is_HORIZWS_latin1(s)
1639 !is_HORIZWS_utf8(s),
1640 !is_HORIZWS_latin1(s)
1646 /* what trie are we using right now */
1648 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1650 = (reg_trie_data*)progi->data->data[ aho->trie ];
1651 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1653 const char *last_start = strend - trie->minlen;
1655 const char *real_start = s;
1657 STRLEN maxlen = trie->maxlen;
1659 U8 **points; /* map of where we were in the input string
1660 when reading a given char. For ASCII this
1661 is unnecessary overhead as the relationship
1662 is always 1:1, but for Unicode, especially
1663 case folded Unicode this is not true. */
1664 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1668 GET_RE_DEBUG_FLAGS_DECL;
1670 /* We can't just allocate points here. We need to wrap it in
1671 * an SV so it gets freed properly if there is a croak while
1672 * running the match */
1675 sv_points=newSV(maxlen * sizeof(U8 *));
1676 SvCUR_set(sv_points,
1677 maxlen * sizeof(U8 *));
1678 SvPOK_on(sv_points);
1679 sv_2mortal(sv_points);
1680 points=(U8**)SvPV_nolen(sv_points );
1681 if ( trie_type != trie_utf8_fold
1682 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1685 bitmap=(U8*)trie->bitmap;
1687 bitmap=(U8*)ANYOF_BITMAP(c);
1689 /* this is the Aho-Corasick algorithm modified a touch
1690 to include special handling for long "unknown char"
1691 sequences. The basic idea being that we use AC as long
1692 as we are dealing with a possible matching char, when
1693 we encounter an unknown char (and we have not encountered
1694 an accepting state) we scan forward until we find a legal
1696 AC matching is basically that of trie matching, except
1697 that when we encounter a failing transition, we fall back
1698 to the current states "fail state", and try the current char
1699 again, a process we repeat until we reach the root state,
1700 state 1, or a legal transition. If we fail on the root state
1701 then we can either terminate if we have reached an accepting
1702 state previously, or restart the entire process from the beginning
1706 while (s <= last_start) {
1707 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1715 U8 *uscan = (U8*)NULL;
1716 U8 *leftmost = NULL;
1718 U32 accepted_word= 0;
1722 while ( state && uc <= (U8*)strend ) {
1724 U32 word = aho->states[ state ].wordnum;
1728 DEBUG_TRIE_EXECUTE_r(
1729 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1730 dump_exec_pos( (char *)uc, c, strend, real_start,
1731 (char *)uc, utf8_target );
1732 PerlIO_printf( Perl_debug_log,
1733 " Scanning for legal start char...\n");
1736 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1741 if (uc >(U8*)last_start) break;
1745 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1746 if (!leftmost || lpos < leftmost) {
1747 DEBUG_r(accepted_word=word);
1753 points[pointpos++ % maxlen]= uc;
1754 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1755 uscan, len, uvc, charid, foldlen,
1757 DEBUG_TRIE_EXECUTE_r({
1758 dump_exec_pos( (char *)uc, c, strend, real_start,
1760 PerlIO_printf(Perl_debug_log,
1761 " Charid:%3u CP:%4"UVxf" ",
1767 word = aho->states[ state ].wordnum;
1769 base = aho->states[ state ].trans.base;
1771 DEBUG_TRIE_EXECUTE_r({
1773 dump_exec_pos( (char *)uc, c, strend, real_start,
1775 PerlIO_printf( Perl_debug_log,
1776 "%sState: %4"UVxf", word=%"UVxf,
1777 failed ? " Fail transition to " : "",
1778 (UV)state, (UV)word);
1784 ( ((offset = base + charid
1785 - 1 - trie->uniquecharcount)) >= 0)
1786 && ((U32)offset < trie->lasttrans)
1787 && trie->trans[offset].check == state
1788 && (tmp=trie->trans[offset].next))
1790 DEBUG_TRIE_EXECUTE_r(
1791 PerlIO_printf( Perl_debug_log," - legal\n"));
1796 DEBUG_TRIE_EXECUTE_r(
1797 PerlIO_printf( Perl_debug_log," - fail\n"));
1799 state = aho->fail[state];
1803 /* we must be accepting here */
1804 DEBUG_TRIE_EXECUTE_r(
1805 PerlIO_printf( Perl_debug_log," - accepting\n"));
1814 if (!state) state = 1;
1817 if ( aho->states[ state ].wordnum ) {
1818 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1819 if (!leftmost || lpos < leftmost) {
1820 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1825 s = (char*)leftmost;
1826 DEBUG_TRIE_EXECUTE_r({
1828 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1829 (UV)accepted_word, (IV)(s - real_start)
1832 if (!reginfo || regtry(reginfo, &s)) {
1838 DEBUG_TRIE_EXECUTE_r({
1839 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1842 DEBUG_TRIE_EXECUTE_r(
1843 PerlIO_printf( Perl_debug_log,"No match.\n"));
1852 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1862 - regexec_flags - match a regexp against a string
1865 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1866 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1867 /* strend: pointer to null at end of string */
1868 /* strbeg: real beginning of string */
1869 /* minend: end of match must be >=minend after stringarg. */
1870 /* data: May be used for some additional optimizations.
1871 Currently its only used, with a U32 cast, for transmitting
1872 the ganch offset when doing a /g match. This will change */
1873 /* nosave: For optimizations. */
1876 struct regexp *const prog = (struct regexp *)SvANY(rx);
1877 /*register*/ char *s;
1878 register regnode *c;
1879 /*register*/ char *startpos = stringarg;
1880 I32 minlen; /* must match at least this many chars */
1881 I32 dontbother = 0; /* how many characters not to try at end */
1882 I32 end_shift = 0; /* Same for the end. */ /* CC */
1883 I32 scream_pos = -1; /* Internal iterator of scream. */
1884 char *scream_olds = NULL;
1885 const bool utf8_target = cBOOL(DO_UTF8(sv));
1887 RXi_GET_DECL(prog,progi);
1888 regmatch_info reginfo; /* create some info to pass to regtry etc */
1889 regexp_paren_pair *swap = NULL;
1890 GET_RE_DEBUG_FLAGS_DECL;
1892 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1893 PERL_UNUSED_ARG(data);
1895 /* Be paranoid... */
1896 if (prog == NULL || startpos == NULL) {
1897 Perl_croak(aTHX_ "NULL regexp parameter");
1901 multiline = prog->extflags & RXf_PMf_MULTILINE;
1902 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
1904 RX_MATCH_UTF8_set(rx, utf8_target);
1906 debug_start_match(rx, utf8_target, startpos, strend,
1910 minlen = prog->minlen;
1912 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1913 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1914 "String too short [regexec_flags]...\n"));
1919 /* Check validity of program. */
1920 if (UCHARAT(progi->program) != REG_MAGIC) {
1921 Perl_croak(aTHX_ "corrupted regexp program");
1925 PL_reg_eval_set = 0;
1929 PL_reg_flags |= RF_utf8;
1931 /* Mark beginning of line for ^ and lookbehind. */
1932 reginfo.bol = startpos; /* XXX not used ??? */
1936 /* Mark end of line for $ (and such) */
1939 /* see how far we have to get to not match where we matched before */
1940 reginfo.till = startpos+minend;
1942 /* If there is a "must appear" string, look for it. */
1945 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1947 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
1948 reginfo.ganch = startpos + prog->gofs;
1949 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1950 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
1951 } else if (sv && SvTYPE(sv) >= SVt_PVMG
1953 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1954 && mg->mg_len >= 0) {
1955 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1956 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1957 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
1959 if (prog->extflags & RXf_ANCH_GPOS) {
1960 if (s > reginfo.ganch)
1962 s = reginfo.ganch - prog->gofs;
1963 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1964 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
1970 reginfo.ganch = strbeg + PTR2UV(data);
1971 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1972 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
1974 } else { /* pos() not defined */
1975 reginfo.ganch = strbeg;
1976 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
1977 "GPOS: reginfo.ganch = strbeg\n"));
1980 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
1981 /* We have to be careful. If the previous successful match
1982 was from this regex we don't want a subsequent partially
1983 successful match to clobber the old results.
1984 So when we detect this possibility we add a swap buffer
1985 to the re, and switch the buffer each match. If we fail
1986 we switch it back, otherwise we leave it swapped.
1989 /* do we need a save destructor here for eval dies? */
1990 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
1992 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1993 re_scream_pos_data d;
1995 d.scream_olds = &scream_olds;
1996 d.scream_pos = &scream_pos;
1997 s = re_intuit_start(rx, sv, s, strend, flags, &d);
1999 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2000 goto phooey; /* not present */
2006 /* Simplest case: anchored match need be tried only once. */
2007 /* [unless only anchor is BOL and multiline is set] */
2008 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2009 if (s == startpos && regtry(®info, &startpos))
2011 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2012 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2017 dontbother = minlen - 1;
2018 end = HOP3c(strend, -dontbother, strbeg) - 1;
2019 /* for multiline we only have to try after newlines */
2020 if (prog->check_substr || prog->check_utf8) {
2024 if (regtry(®info, &s))
2029 if (prog->extflags & RXf_USE_INTUIT) {
2030 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2041 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2042 if (regtry(®info, &s))
2049 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2051 /* the warning about reginfo.ganch being used without intialization
2052 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2053 and we only enter this block when the same bit is set. */
2054 char *tmp_s = reginfo.ganch - prog->gofs;
2056 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2061 /* Messy cases: unanchored match. */
2062 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2063 /* we have /x+whatever/ */
2064 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2069 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2070 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2071 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2076 DEBUG_EXECUTE_r( did_match = 1 );
2077 if (regtry(®info, &s)) goto got_it;
2079 while (s < strend && *s == ch)
2087 DEBUG_EXECUTE_r( did_match = 1 );
2088 if (regtry(®info, &s)) goto got_it;
2090 while (s < strend && *s == ch)
2095 DEBUG_EXECUTE_r(if (!did_match)
2096 PerlIO_printf(Perl_debug_log,
2097 "Did not find anchored character...\n")
2100 else if (prog->anchored_substr != NULL
2101 || prog->anchored_utf8 != NULL
2102 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2103 && prog->float_max_offset < strend - s)) {
2108 char *last1; /* Last position checked before */
2112 if (prog->anchored_substr || prog->anchored_utf8) {
2113 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2114 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2115 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2116 back_max = back_min = prog->anchored_offset;
2118 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2119 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2120 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2121 back_max = prog->float_max_offset;
2122 back_min = prog->float_min_offset;
2126 if (must == &PL_sv_undef)
2127 /* could not downgrade utf8 check substring, so must fail */
2133 last = HOP3c(strend, /* Cannot start after this */
2134 -(I32)(CHR_SVLEN(must)
2135 - (SvTAIL(must) != 0) + back_min), strbeg);
2138 last1 = HOPc(s, -1);
2140 last1 = s - 1; /* bogus */
2142 /* XXXX check_substr already used to find "s", can optimize if
2143 check_substr==must. */
2145 dontbother = end_shift;
2146 strend = HOPc(strend, -dontbother);
2147 while ( (s <= last) &&
2148 ((flags & REXEC_SCREAM)
2149 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2150 end_shift, &scream_pos, 0))
2151 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2152 (unsigned char*)strend, must,
2153 multiline ? FBMrf_MULTILINE : 0))) ) {
2154 /* we may be pointing at the wrong string */
2155 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2156 s = strbeg + (s - SvPVX_const(sv));
2157 DEBUG_EXECUTE_r( did_match = 1 );
2158 if (HOPc(s, -back_max) > last1) {
2159 last1 = HOPc(s, -back_min);
2160 s = HOPc(s, -back_max);
2163 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2165 last1 = HOPc(s, -back_min);
2169 while (s <= last1) {
2170 if (regtry(®info, &s))
2176 while (s <= last1) {
2177 if (regtry(®info, &s))
2183 DEBUG_EXECUTE_r(if (!did_match) {
2184 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2185 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2186 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2187 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2188 ? "anchored" : "floating"),
2189 quoted, RE_SV_TAIL(must));
2193 else if ( (c = progi->regstclass) ) {
2195 const OPCODE op = OP(progi->regstclass);
2196 /* don't bother with what can't match */
2197 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2198 strend = HOPc(strend, -(minlen - 1));
2201 SV * const prop = sv_newmortal();
2202 regprop(prog, prop, c);
2204 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2206 PerlIO_printf(Perl_debug_log,
2207 "Matching stclass %.*s against %s (%d bytes)\n",
2208 (int)SvCUR(prop), SvPVX_const(prop),
2209 quoted, (int)(strend - s));
2212 if (find_byclass(prog, c, s, strend, ®info))
2214 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2218 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2223 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2224 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2225 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2227 if (flags & REXEC_SCREAM) {
2228 last = screaminstr(sv, float_real, s - strbeg,
2229 end_shift, &scream_pos, 1); /* last one */
2231 last = scream_olds; /* Only one occurrence. */
2232 /* we may be pointing at the wrong string */
2233 else if (RXp_MATCH_COPIED(prog))
2234 s = strbeg + (s - SvPVX_const(sv));
2238 const char * const little = SvPV_const(float_real, len);
2240 if (SvTAIL(float_real)) {
2241 if (memEQ(strend - len + 1, little, len - 1))
2242 last = strend - len + 1;
2243 else if (!multiline)
2244 last = memEQ(strend - len, little, len)
2245 ? strend - len : NULL;
2251 last = rninstr(s, strend, little, little + len);
2253 last = strend; /* matching "$" */
2258 PerlIO_printf(Perl_debug_log,
2259 "%sCan't trim the tail, match fails (should not happen)%s\n",
2260 PL_colors[4], PL_colors[5]));
2261 goto phooey; /* Should not happen! */
2263 dontbother = strend - last + prog->float_min_offset;
2265 if (minlen && (dontbother < minlen))
2266 dontbother = minlen - 1;
2267 strend -= dontbother; /* this one's always in bytes! */
2268 /* We don't know much -- general case. */
2271 if (regtry(®info, &s))
2280 if (regtry(®info, &s))
2282 } while (s++ < strend);
2291 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2293 if (PL_reg_eval_set)
2294 restore_pos(aTHX_ prog);
2295 if (RXp_PAREN_NAMES(prog))
2296 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2298 /* make sure $`, $&, $', and $digit will work later */
2299 if ( !(flags & REXEC_NOT_FIRST) ) {
2300 RX_MATCH_COPY_FREE(rx);
2301 if (flags & REXEC_COPY_STR) {
2302 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2303 #ifdef PERL_OLD_COPY_ON_WRITE
2305 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2307 PerlIO_printf(Perl_debug_log,
2308 "Copy on write: regexp capture, type %d\n",
2311 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2312 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2313 assert (SvPOKp(prog->saved_copy));
2317 RX_MATCH_COPIED_on(rx);
2318 s = savepvn(strbeg, i);
2324 prog->subbeg = strbeg;
2325 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2332 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2333 PL_colors[4], PL_colors[5]));
2334 if (PL_reg_eval_set)
2335 restore_pos(aTHX_ prog);
2337 /* we failed :-( roll it back */
2338 Safefree(prog->offs);
2347 - regtry - try match at specific point
2349 STATIC I32 /* 0 failure, 1 success */
2350 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2354 REGEXP *const rx = reginfo->prog;
2355 regexp *const prog = (struct regexp *)SvANY(rx);
2356 RXi_GET_DECL(prog,progi);
2357 GET_RE_DEBUG_FLAGS_DECL;
2359 PERL_ARGS_ASSERT_REGTRY;
2361 reginfo->cutpoint=NULL;
2363 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2366 PL_reg_eval_set = RS_init;
2367 DEBUG_EXECUTE_r(DEBUG_s(
2368 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2369 (IV)(PL_stack_sp - PL_stack_base));
2372 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2373 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2375 /* Apparently this is not needed, judging by wantarray. */
2376 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2377 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2380 /* Make $_ available to executed code. */
2381 if (reginfo->sv != DEFSV) {
2383 DEFSV_set(reginfo->sv);
2386 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2387 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2388 /* prepare for quick setting of pos */
2389 #ifdef PERL_OLD_COPY_ON_WRITE
2390 if (SvIsCOW(reginfo->sv))
2391 sv_force_normal_flags(reginfo->sv, 0);
2393 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2394 &PL_vtbl_mglob, NULL, 0);
2398 PL_reg_oldpos = mg->mg_len;
2399 SAVEDESTRUCTOR_X(restore_pos, prog);
2401 if (!PL_reg_curpm) {
2402 Newxz(PL_reg_curpm, 1, PMOP);
2405 SV* const repointer = &PL_sv_undef;
2406 /* this regexp is also owned by the new PL_reg_curpm, which
2407 will try to free it. */
2408 av_push(PL_regex_padav, repointer);
2409 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2410 PL_regex_pad = AvARRAY(PL_regex_padav);
2415 /* It seems that non-ithreads works both with and without this code.
2416 So for efficiency reasons it seems best not to have the code
2417 compiled when it is not needed. */
2418 /* This is safe against NULLs: */
2419 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2420 /* PM_reg_curpm owns a reference to this regexp. */
2423 PM_SETRE(PL_reg_curpm, rx);
2424 PL_reg_oldcurpm = PL_curpm;
2425 PL_curpm = PL_reg_curpm;
2426 if (RXp_MATCH_COPIED(prog)) {
2427 /* Here is a serious problem: we cannot rewrite subbeg,
2428 since it may be needed if this match fails. Thus
2429 $` inside (?{}) could fail... */
2430 PL_reg_oldsaved = prog->subbeg;
2431 PL_reg_oldsavedlen = prog->sublen;
2432 #ifdef PERL_OLD_COPY_ON_WRITE
2433 PL_nrs = prog->saved_copy;
2435 RXp_MATCH_COPIED_off(prog);
2438 PL_reg_oldsaved = NULL;
2439 prog->subbeg = PL_bostr;
2440 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2442 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2443 prog->offs[0].start = *startpos - PL_bostr;
2444 PL_reginput = *startpos;
2445 PL_reglastparen = &prog->lastparen;
2446 PL_reglastcloseparen = &prog->lastcloseparen;
2447 prog->lastparen = 0;
2448 prog->lastcloseparen = 0;
2450 PL_regoffs = prog->offs;
2451 if (PL_reg_start_tmpl <= prog->nparens) {
2452 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2453 if(PL_reg_start_tmp)
2454 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2456 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2459 /* XXXX What this code is doing here?!!! There should be no need
2460 to do this again and again, PL_reglastparen should take care of
2463 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2464 * Actually, the code in regcppop() (which Ilya may be meaning by
2465 * PL_reglastparen), is not needed at all by the test suite
2466 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2467 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2468 * Meanwhile, this code *is* needed for the
2469 * above-mentioned test suite tests to succeed. The common theme
2470 * on those tests seems to be returning null fields from matches.
2471 * --jhi updated by dapm */
2473 if (prog->nparens) {
2474 regexp_paren_pair *pp = PL_regoffs;
2476 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2484 if (regmatch(reginfo, progi->program + 1)) {
2485 PL_regoffs[0].end = PL_reginput - PL_bostr;
2488 if (reginfo->cutpoint)
2489 *startpos= reginfo->cutpoint;
2490 REGCP_UNWIND(lastcp);
2495 #define sayYES goto yes
2496 #define sayNO goto no
2497 #define sayNO_SILENT goto no_silent
2499 /* we dont use STMT_START/END here because it leads to
2500 "unreachable code" warnings, which are bogus, but distracting. */
2501 #define CACHEsayNO \
2502 if (ST.cache_mask) \
2503 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2506 /* this is used to determine how far from the left messages like
2507 'failed...' are printed. It should be set such that messages
2508 are inline with the regop output that created them.
2510 #define REPORT_CODE_OFF 32
2513 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2514 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2516 #define SLAB_FIRST(s) (&(s)->states[0])
2517 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2519 /* grab a new slab and return the first slot in it */
2521 STATIC regmatch_state *
2524 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2527 regmatch_slab *s = PL_regmatch_slab->next;
2529 Newx(s, 1, regmatch_slab);
2530 s->prev = PL_regmatch_slab;
2532 PL_regmatch_slab->next = s;
2534 PL_regmatch_slab = s;
2535 return SLAB_FIRST(s);
2539 /* push a new state then goto it */
2541 #define PUSH_STATE_GOTO(state, node) \
2543 st->resume_state = state; \
2546 /* push a new state with success backtracking, then goto it */
2548 #define PUSH_YES_STATE_GOTO(state, node) \
2550 st->resume_state = state; \
2551 goto push_yes_state;
2557 regmatch() - main matching routine
2559 This is basically one big switch statement in a loop. We execute an op,
2560 set 'next' to point the next op, and continue. If we come to a point which
2561 we may need to backtrack to on failure such as (A|B|C), we push a
2562 backtrack state onto the backtrack stack. On failure, we pop the top
2563 state, and re-enter the loop at the state indicated. If there are no more
2564 states to pop, we return failure.
2566 Sometimes we also need to backtrack on success; for example /A+/, where
2567 after successfully matching one A, we need to go back and try to
2568 match another one; similarly for lookahead assertions: if the assertion
2569 completes successfully, we backtrack to the state just before the assertion
2570 and then carry on. In these cases, the pushed state is marked as
2571 'backtrack on success too'. This marking is in fact done by a chain of
2572 pointers, each pointing to the previous 'yes' state. On success, we pop to
2573 the nearest yes state, discarding any intermediate failure-only states.
2574 Sometimes a yes state is pushed just to force some cleanup code to be
2575 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2576 it to free the inner regex.
2578 Note that failure backtracking rewinds the cursor position, while
2579 success backtracking leaves it alone.
2581 A pattern is complete when the END op is executed, while a subpattern
2582 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2583 ops trigger the "pop to last yes state if any, otherwise return true"
2586 A common convention in this function is to use A and B to refer to the two
2587 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2588 the subpattern to be matched possibly multiple times, while B is the entire
2589 rest of the pattern. Variable and state names reflect this convention.
2591 The states in the main switch are the union of ops and failure/success of
2592 substates associated with with that op. For example, IFMATCH is the op
2593 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2594 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2595 successfully matched A and IFMATCH_A_fail is a state saying that we have
2596 just failed to match A. Resume states always come in pairs. The backtrack
2597 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2598 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2599 on success or failure.
2601 The struct that holds a backtracking state is actually a big union, with
2602 one variant for each major type of op. The variable st points to the
2603 top-most backtrack struct. To make the code clearer, within each
2604 block of code we #define ST to alias the relevant union.
2606 Here's a concrete example of a (vastly oversimplified) IFMATCH
2612 #define ST st->u.ifmatch
2614 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2615 ST.foo = ...; // some state we wish to save
2617 // push a yes backtrack state with a resume value of
2618 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2620 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2623 case IFMATCH_A: // we have successfully executed A; now continue with B
2625 bar = ST.foo; // do something with the preserved value
2628 case IFMATCH_A_fail: // A failed, so the assertion failed
2629 ...; // do some housekeeping, then ...
2630 sayNO; // propagate the failure
2637 For any old-timers reading this who are familiar with the old recursive
2638 approach, the code above is equivalent to:
2640 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2649 ...; // do some housekeeping, then ...
2650 sayNO; // propagate the failure
2653 The topmost backtrack state, pointed to by st, is usually free. If you
2654 want to claim it, populate any ST.foo fields in it with values you wish to
2655 save, then do one of
2657 PUSH_STATE_GOTO(resume_state, node);
2658 PUSH_YES_STATE_GOTO(resume_state, node);
2660 which sets that backtrack state's resume value to 'resume_state', pushes a
2661 new free entry to the top of the backtrack stack, then goes to 'node'.
2662 On backtracking, the free slot is popped, and the saved state becomes the
2663 new free state. An ST.foo field in this new top state can be temporarily
2664 accessed to retrieve values, but once the main loop is re-entered, it
2665 becomes available for reuse.
2667 Note that the depth of the backtrack stack constantly increases during the
2668 left-to-right execution of the pattern, rather than going up and down with
2669 the pattern nesting. For example the stack is at its maximum at Z at the
2670 end of the pattern, rather than at X in the following:
2672 /(((X)+)+)+....(Y)+....Z/
2674 The only exceptions to this are lookahead/behind assertions and the cut,
2675 (?>A), which pop all the backtrack states associated with A before
2678 Bascktrack state structs are allocated in slabs of about 4K in size.
2679 PL_regmatch_state and st always point to the currently active state,
2680 and PL_regmatch_slab points to the slab currently containing
2681 PL_regmatch_state. The first time regmatch() is called, the first slab is
2682 allocated, and is never freed until interpreter destruction. When the slab
2683 is full, a new one is allocated and chained to the end. At exit from
2684 regmatch(), slabs allocated since entry are freed.
2689 #define DEBUG_STATE_pp(pp) \
2691 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2692 PerlIO_printf(Perl_debug_log, \
2693 " %*s"pp" %s%s%s%s%s\n", \
2695 PL_reg_name[st->resume_state], \
2696 ((st==yes_state||st==mark_state) ? "[" : ""), \
2697 ((st==yes_state) ? "Y" : ""), \
2698 ((st==mark_state) ? "M" : ""), \
2699 ((st==yes_state||st==mark_state) ? "]" : "") \
2704 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2709 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2710 const char *start, const char *end, const char *blurb)
2712 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2714 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2719 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2720 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2722 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2723 start, end - start, 60);
2725 PerlIO_printf(Perl_debug_log,
2726 "%s%s REx%s %s against %s\n",
2727 PL_colors[4], blurb, PL_colors[5], s0, s1);
2729 if (utf8_target||utf8_pat)
2730 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2731 utf8_pat ? "pattern" : "",
2732 utf8_pat && utf8_target ? " and " : "",
2733 utf8_target ? "string" : ""
2739 S_dump_exec_pos(pTHX_ const char *locinput,
2740 const regnode *scan,
2741 const char *loc_regeol,
2742 const char *loc_bostr,
2743 const char *loc_reg_starttry,
2744 const bool utf8_target)
2746 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2747 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2748 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2749 /* The part of the string before starttry has one color
2750 (pref0_len chars), between starttry and current
2751 position another one (pref_len - pref0_len chars),
2752 after the current position the third one.
2753 We assume that pref0_len <= pref_len, otherwise we
2754 decrease pref0_len. */
2755 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2756 ? (5 + taill) - l : locinput - loc_bostr;
2759 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2761 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2763 pref0_len = pref_len - (locinput - loc_reg_starttry);
2764 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2765 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2766 ? (5 + taill) - pref_len : loc_regeol - locinput);
2767 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2771 if (pref0_len > pref_len)
2772 pref0_len = pref_len;
2774 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2776 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2777 (locinput - pref_len),pref0_len, 60, 4, 5);
2779 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2780 (locinput - pref_len + pref0_len),
2781 pref_len - pref0_len, 60, 2, 3);
2783 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2784 locinput, loc_regeol - locinput, 10, 0, 1);
2786 const STRLEN tlen=len0+len1+len2;
2787 PerlIO_printf(Perl_debug_log,
2788 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2789 (IV)(locinput - loc_bostr),
2792 (docolor ? "" : "> <"),
2794 (int)(tlen > 19 ? 0 : 19 - tlen),
2801 /* reg_check_named_buff_matched()
2802 * Checks to see if a named buffer has matched. The data array of
2803 * buffer numbers corresponding to the buffer is expected to reside
2804 * in the regexp->data->data array in the slot stored in the ARG() of
2805 * node involved. Note that this routine doesn't actually care about the
2806 * name, that information is not preserved from compilation to execution.
2807 * Returns the index of the leftmost defined buffer with the given name
2808 * or 0 if non of the buffers matched.
2811 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2814 RXi_GET_DECL(rex,rexi);
2815 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2816 I32 *nums=(I32*)SvPVX(sv_dat);
2818 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2820 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2821 if ((I32)*PL_reglastparen >= nums[n] &&
2822 PL_regoffs[nums[n]].end != -1)
2831 /* free all slabs above current one - called during LEAVE_SCOPE */
2834 S_clear_backtrack_stack(pTHX_ void *p)
2836 regmatch_slab *s = PL_regmatch_slab->next;
2841 PL_regmatch_slab->next = NULL;
2843 regmatch_slab * const osl = s;
2850 #define SETREX(Re1,Re2) \
2851 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2854 STATIC I32 /* 0 failure, 1 success */
2855 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2857 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2861 register const bool utf8_target = PL_reg_match_utf8;
2862 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2863 REGEXP *rex_sv = reginfo->prog;
2864 regexp *rex = (struct regexp *)SvANY(rex_sv);
2865 RXi_GET_DECL(rex,rexi);
2867 /* the current state. This is a cached copy of PL_regmatch_state */
2868 register regmatch_state *st;
2869 /* cache heavy used fields of st in registers */
2870 register regnode *scan;
2871 register regnode *next;
2872 register U32 n = 0; /* general value; init to avoid compiler warning */
2873 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2874 register char *locinput = PL_reginput;
2875 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2877 bool result = 0; /* return value of S_regmatch */
2878 int depth = 0; /* depth of backtrack stack */
2879 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2880 const U32 max_nochange_depth =
2881 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2882 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2883 regmatch_state *yes_state = NULL; /* state to pop to on success of
2885 /* mark_state piggy backs on the yes_state logic so that when we unwind
2886 the stack on success we can update the mark_state as we go */
2887 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2888 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2889 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2891 bool no_final = 0; /* prevent failure from backtracking? */
2892 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2893 char *startpoint = PL_reginput;
2894 SV *popmark = NULL; /* are we looking for a mark? */
2895 SV *sv_commit = NULL; /* last mark name seen in failure */
2896 SV *sv_yes_mark = NULL; /* last mark name we have seen
2897 during a successfull match */
2898 U32 lastopen = 0; /* last open we saw */
2899 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2900 SV* const oreplsv = GvSV(PL_replgv);
2901 /* these three flags are set by various ops to signal information to
2902 * the very next op. They have a useful lifetime of exactly one loop
2903 * iteration, and are not preserved or restored by state pushes/pops
2905 bool sw = 0; /* the condition value in (?(cond)a|b) */
2906 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2907 int logical = 0; /* the following EVAL is:
2911 or the following IFMATCH/UNLESSM is:
2912 false: plain (?=foo)
2913 true: used as a condition: (?(?=foo))
2916 GET_RE_DEBUG_FLAGS_DECL;
2919 PERL_ARGS_ASSERT_REGMATCH;
2921 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
2922 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2924 /* on first ever call to regmatch, allocate first slab */
2925 if (!PL_regmatch_slab) {
2926 Newx(PL_regmatch_slab, 1, regmatch_slab);
2927 PL_regmatch_slab->prev = NULL;
2928 PL_regmatch_slab->next = NULL;
2929 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2932 oldsave = PL_savestack_ix;
2933 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
2934 SAVEVPTR(PL_regmatch_slab);
2935 SAVEVPTR(PL_regmatch_state);
2937 /* grab next free state slot */
2938 st = ++PL_regmatch_state;
2939 if (st > SLAB_LAST(PL_regmatch_slab))
2940 st = PL_regmatch_state = S_push_slab(aTHX);
2942 /* Note that nextchr is a byte even in UTF */
2943 nextchr = UCHARAT(locinput);
2945 while (scan != NULL) {
2948 SV * const prop = sv_newmortal();
2949 regnode *rnext=regnext(scan);
2950 DUMP_EXEC_POS( locinput, scan, utf8_target );
2951 regprop(rex, prop, scan);
2953 PerlIO_printf(Perl_debug_log,
2954 "%3"IVdf":%*s%s(%"IVdf")\n",
2955 (IV)(scan - rexi->program), depth*2, "",
2957 (PL_regkind[OP(scan)] == END || !rnext) ?
2958 0 : (IV)(rnext - rexi->program));
2961 next = scan + NEXT_OFF(scan);
2964 state_num = OP(scan);
2968 assert(PL_reglastparen == &rex->lastparen);
2969 assert(PL_reglastcloseparen == &rex->lastcloseparen);
2970 assert(PL_regoffs == rex->offs);
2972 switch (state_num) {
2974 if (locinput == PL_bostr)
2976 /* reginfo->till = reginfo->bol; */
2981 if (locinput == PL_bostr ||
2982 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2988 if (locinput == PL_bostr)
2992 if (locinput == reginfo->ganch)
2997 /* update the startpoint */
2998 st->u.keeper.val = PL_regoffs[0].start;
2999 PL_reginput = locinput;
3000 PL_regoffs[0].start = locinput - PL_bostr;
3001 PUSH_STATE_GOTO(KEEPS_next, next);
3003 case KEEPS_next_fail:
3004 /* rollback the start point change */
3005 PL_regoffs[0].start = st->u.keeper.val;
3011 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3016 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3018 if (PL_regeol - locinput > 1)
3022 if (PL_regeol != locinput)
3026 if (!nextchr && locinput >= PL_regeol)
3029 locinput += PL_utf8skip[nextchr];
3030 if (locinput > PL_regeol)
3032 nextchr = UCHARAT(locinput);
3035 nextchr = UCHARAT(++locinput);
3038 if (!nextchr && locinput >= PL_regeol)
3040 nextchr = UCHARAT(++locinput);
3043 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3046 locinput += PL_utf8skip[nextchr];
3047 if (locinput > PL_regeol)
3049 nextchr = UCHARAT(locinput);
3052 nextchr = UCHARAT(++locinput);
3056 #define ST st->u.trie
3058 /* In this case the charclass data is available inline so
3059 we can fail fast without a lot of extra overhead.
3061 if (scan->flags == EXACT || !utf8_target) {
3062 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3064 PerlIO_printf(Perl_debug_log,
3065 "%*s %sfailed to match trie start class...%s\n",
3066 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3074 /* the basic plan of execution of the trie is:
3075 * At the beginning, run though all the states, and
3076 * find the longest-matching word. Also remember the position
3077 * of the shortest matching word. For example, this pattern:
3080 * when matched against the string "abcde", will generate
3081 * accept states for all words except 3, with the longest
3082 * matching word being 4, and the shortest being 1 (with
3083 * the position being after char 1 of the string).
3085 * Then for each matching word, in word order (i.e. 1,2,4,5),
3086 * we run the remainder of the pattern; on each try setting
3087 * the current position to the character following the word,
3088 * returning to try the next word on failure.
3090 * We avoid having to build a list of words at runtime by
3091 * using a compile-time structure, wordinfo[].prev, which
3092 * gives, for each word, the previous accepting word (if any).
3093 * In the case above it would contain the mappings 1->2, 2->0,
3094 * 3->0, 4->5, 5->1. We can use this table to generate, from
3095 * the longest word (4 above), a list of all words, by
3096 * following the list of prev pointers; this gives us the
3097 * unordered list 4,5,1,2. Then given the current word we have
3098 * just tried, we can go through the list and find the
3099 * next-biggest word to try (so if we just failed on word 2,
3100 * the next in the list is 4).
3102 * Since at runtime we don't record the matching position in
3103 * the string for each word, we have to work that out for
3104 * each word we're about to process. The wordinfo table holds
3105 * the character length of each word; given that we recorded
3106 * at the start: the position of the shortest word and its
3107 * length in chars, we just need to move the pointer the
3108 * difference between the two char lengths. Depending on
3109 * Unicode status and folding, that's cheap or expensive.
3111 * This algorithm is optimised for the case where are only a
3112 * small number of accept states, i.e. 0,1, or maybe 2.
3113 * With lots of accepts states, and having to try all of them,
3114 * it becomes quadratic on number of accept states to find all
3119 /* what type of TRIE am I? (utf8 makes this contextual) */
3120 DECL_TRIE_TYPE(scan);
3122 /* what trie are we using right now */
3123 reg_trie_data * const trie
3124 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3125 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3126 U32 state = trie->startstate;
3128 if (trie->bitmap && trie_type != trie_utf8_fold &&
3129 !TRIE_BITMAP_TEST(trie,*locinput)
3131 if (trie->states[ state ].wordnum) {
3133 PerlIO_printf(Perl_debug_log,
3134 "%*s %smatched empty string...%s\n",
3135 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3140 PerlIO_printf(Perl_debug_log,
3141 "%*s %sfailed to match trie start class...%s\n",
3142 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3149 U8 *uc = ( U8* )locinput;
3153 U8 *uscan = (U8*)NULL;
3154 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3155 U32 charcount = 0; /* how many input chars we have matched */
3156 U32 accepted = 0; /* have we seen any accepting states? */
3159 ST.jump = trie->jump;
3162 ST.longfold = FALSE; /* char longer if folded => it's harder */
3165 /* fully traverse the TRIE; note the position of the
3166 shortest accept state and the wordnum of the longest
3169 while ( state && uc <= (U8*)PL_regeol ) {
3170 U32 base = trie->states[ state ].trans.base;
3174 wordnum = trie->states[ state ].wordnum;
3176 if (wordnum) { /* it's an accept state */
3179 /* record first match position */
3181 ST.firstpos = (U8*)locinput;
3186 ST.firstchars = charcount;
3189 if (!ST.nextword || wordnum < ST.nextword)
3190 ST.nextword = wordnum;
3191 ST.topword = wordnum;
3194 DEBUG_TRIE_EXECUTE_r({
3195 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3196 PerlIO_printf( Perl_debug_log,
3197 "%*s %sState: %4"UVxf" Accepted: %c ",
3198 2+depth * 2, "", PL_colors[4],
3199 (UV)state, (accepted ? 'Y' : 'N'));
3202 /* read a char and goto next state */
3205 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3206 uscan, len, uvc, charid, foldlen,
3213 base + charid - 1 - trie->uniquecharcount)) >= 0)
3215 && ((U32)offset < trie->lasttrans)
3216 && trie->trans[offset].check == state)
3218 state = trie->trans[offset].next;
3229 DEBUG_TRIE_EXECUTE_r(
3230 PerlIO_printf( Perl_debug_log,
3231 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3232 charid, uvc, (UV)state, PL_colors[5] );
3238 /* calculate total number of accept states */
3243 w = trie->wordinfo[w].prev;
3246 ST.accepted = accepted;
3250 PerlIO_printf( Perl_debug_log,
3251 "%*s %sgot %"IVdf" possible matches%s\n",
3252 REPORT_CODE_OFF + depth * 2, "",
3253 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3255 goto trie_first_try; /* jump into the fail handler */
3259 case TRIE_next_fail: /* we failed - try next alternative */
3261 REGCP_UNWIND(ST.cp);
3262 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3263 PL_regoffs[n].end = -1;
3264 *PL_reglastparen = n;
3266 if (!--ST.accepted) {
3268 PerlIO_printf( Perl_debug_log,
3269 "%*s %sTRIE failed...%s\n",
3270 REPORT_CODE_OFF+depth*2, "",
3277 /* Find next-highest word to process. Note that this code
3278 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3279 register U16 min = 0;
3281 register U16 const nextword = ST.nextword;
3282 register reg_trie_wordinfo * const wordinfo
3283 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3284 for (word=ST.topword; word; word=wordinfo[word].prev) {
3285 if (word > nextword && (!min || word < min))
3298 ST.lastparen = *PL_reglastparen;
3302 /* find start char of end of current word */
3304 U32 chars; /* how many chars to skip */
3305 U8 *uc = ST.firstpos;
3306 reg_trie_data * const trie
3307 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3309 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3311 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3315 /* the hard option - fold each char in turn and find
3316 * its folded length (which may be different */
3317 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3325 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3333 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3338 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3352 PL_reginput = (char *)uc;
3355 scan = (ST.jump && ST.jump[ST.nextword])
3356 ? ST.me + ST.jump[ST.nextword]
3360 PerlIO_printf( Perl_debug_log,
3361 "%*s %sTRIE matched word #%d, continuing%s\n",
3362 REPORT_CODE_OFF+depth*2, "",
3369 if (ST.accepted > 1 || has_cutgroup) {
3370 PUSH_STATE_GOTO(TRIE_next, scan);
3373 /* only one choice left - just continue */
3375 AV *const trie_words
3376 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3377 SV ** const tmp = av_fetch( trie_words,
3379 SV *sv= tmp ? sv_newmortal() : NULL;
3381 PerlIO_printf( Perl_debug_log,
3382 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3383 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3385 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3386 PL_colors[0], PL_colors[1],
3387 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3389 : "not compiled under -Dr",
3393 locinput = PL_reginput;
3394 nextchr = UCHARAT(locinput);
3395 continue; /* execute rest of RE */
3400 char *s = STRING(scan);
3402 if (utf8_target != UTF_PATTERN) {
3403 /* The target and the pattern have differing utf8ness. */
3405 const char * const e = s + ln;
3408 /* The target is utf8, the pattern is not utf8. */
3413 if (NATIVE_TO_UNI(*(U8*)s) !=
3414 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3422 /* The target is not utf8, the pattern is utf8. */
3427 if (NATIVE_TO_UNI(*((U8*)l)) !=
3428 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3436 nextchr = UCHARAT(locinput);
3439 /* The target and the pattern have the same utf8ness. */
3440 /* Inline the first character, for speed. */
3441 if (UCHARAT(s) != nextchr)
3443 if (PL_regeol - locinput < ln)
3445 if (ln > 1 && memNE(s, locinput, ln))
3448 nextchr = UCHARAT(locinput);
3452 PL_reg_flags |= RF_tainted;
3455 char * const s = STRING(scan);
3458 if (utf8_target || UTF_PATTERN) {
3459 /* Either target or the pattern are utf8. */
3460 const char * const l = locinput;
3461 char *e = PL_regeol;
3463 if (! foldEQ_utf8(s, 0, ln, cBOOL(UTF_PATTERN),
3464 l, &e, 0, utf8_target)) {
3465 /* One more case for the sharp s:
3466 * pack("U0U*", 0xDF) =~ /ss/i,
3467 * the 0xC3 0x9F are the UTF-8
3468 * byte sequence for the U+00DF. */
3470 if (!(utf8_target &&
3471 toLOWER(s[0]) == 's' &&
3473 toLOWER(s[1]) == 's' &&
3480 nextchr = UCHARAT(locinput);
3484 /* Neither the target and the pattern are utf8. */
3486 /* Inline the first character, for speed. */
3487 if (UCHARAT(s) != nextchr &&
3488 UCHARAT(s) != ((OP(scan) == EXACTF)
3489 ? PL_fold : PL_fold_locale)[nextchr])
3491 if (PL_regeol - locinput < ln)
3493 if (ln > 1 && (OP(scan) == EXACTF
3494 ? ! foldEQ(s, locinput, ln)
3495 : ! foldEQ_locale(s, locinput, ln)))
3498 nextchr = UCHARAT(locinput);
3503 PL_reg_flags |= RF_tainted;
3507 /* was last char in word? */
3509 if (locinput == PL_bostr)
3512 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3514 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3516 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3517 ln = isALNUM_uni(ln);
3518 LOAD_UTF8_CHARCLASS_ALNUM();
3519 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3522 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3523 n = isALNUM_LC_utf8((U8*)locinput);
3527 ln = (locinput != PL_bostr) ?
3528 UCHARAT(locinput - 1) : '\n';
3529 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3531 n = isALNUM(nextchr);
3534 ln = isALNUM_LC(ln);
3535 n = isALNUM_LC(nextchr);
3538 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3539 OP(scan) == BOUNDL))
3544 STRLEN inclasslen = PL_regeol - locinput;
3546 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3548 if (locinput >= PL_regeol)
3550 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3551 nextchr = UCHARAT(locinput);
3556 nextchr = UCHARAT(locinput);
3557 if (!REGINCLASS(rex, scan, (U8*)locinput))
3559 if (!nextchr && locinput >= PL_regeol)
3561 nextchr = UCHARAT(++locinput);
3565 /* If we might have the case of the German sharp s
3566 * in a casefolding Unicode character class. */
3568 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3569 locinput += SHARP_S_SKIP;
3570 nextchr = UCHARAT(locinput);
3575 /* Special char classes - The defines start on line 129 or so */
3576 CCC_TRY_AFF( ALNUM, ALNUML, perl_word, "a", isALNUM_LC_utf8, isALNUM, isALNUM_LC);
3577 CCC_TRY_NEG(NALNUM, NALNUML, perl_word, "a", isALNUM_LC_utf8, isALNUM, isALNUM_LC);
3579 CCC_TRY_AFF( SPACE, SPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE, isSPACE_LC);
3580 CCC_TRY_NEG(NSPACE, NSPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE, isSPACE_LC);
3582 CCC_TRY_AFF( DIGIT, DIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3583 CCC_TRY_NEG(NDIGIT, NDIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3585 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3586 a Unicode extended Grapheme Cluster */
3587 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3588 extended Grapheme Cluster is:
3591 | Prepend* Begin Extend*
3594 Begin is (Hangul-syllable | ! Control)
3595 Extend is (Grapheme_Extend | Spacing_Mark)
3596 Control is [ GCB_Control CR LF ]
3598 The discussion below shows how the code for CLUMP is derived
3599 from this regex. Note that most of these concepts are from
3600 property values of the Grapheme Cluster Boundary (GCB) property.
3601 No code point can have multiple property values for a given
3602 property. Thus a code point in Prepend can't be in Control, but
3603 it must be in !Control. This is why Control above includes
3604 GCB_Control plus CR plus LF. The latter two are used in the GCB
3605 property separately, and so can't be in GCB_Control, even though
3606 they logically are controls. Control is not the same as gc=cc,
3607 but includes format and other characters as well.
3609 The Unicode definition of Hangul-syllable is:
3611 | (L* ( ( V | LV ) V* | LVT ) T*)
3614 Each of these is a value for the GCB property, and hence must be
3615 disjoint, so the order they are tested is immaterial, so the
3616 above can safely be changed to
3619 | (L* ( LVT | ( V | LV ) V*) T*)
3621 The last two terms can be combined like this:
3623 | (( LVT | ( V | LV ) V*) T*))
3625 And refactored into this:
3626 L* (L | LVT T* | V V* T* | LV V* T*)
3628 That means that if we have seen any L's at all we can quit
3629 there, but if the next character is a LVT, a V or and LV we
3632 There is a subtlety with Prepend* which showed up in testing.
3633 Note that the Begin, and only the Begin is required in:
3634 | Prepend* Begin Extend*
3635 Also, Begin contains '! Control'. A Prepend must be a '!
3636 Control', which means it must be a Begin. What it comes down to
3637 is that if we match Prepend* and then find no suitable Begin
3638 afterwards, that if we backtrack the last Prepend, that one will
3639 be a suitable Begin.
3642 if (locinput >= PL_regeol)
3644 if (! utf8_target) {
3646 /* Match either CR LF or '.', as all the other possibilities
3648 locinput++; /* Match the . or CR */
3650 && locinput < PL_regeol
3651 && UCHARAT(locinput) == '\n') locinput++;
3655 /* Utf8: See if is ( CR LF ); already know that locinput <
3656 * PL_regeol, so locinput+1 is in bounds */
3657 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3661 /* In case have to backtrack to beginning, then match '.' */
3662 char *starting = locinput;
3664 /* In case have to backtrack the last prepend */
3665 char *previous_prepend = 0;
3667 LOAD_UTF8_CHARCLASS_GCB();
3669 /* Match (prepend)* */
3670 while (locinput < PL_regeol
3671 && swash_fetch(PL_utf8_X_prepend,
3672 (U8*)locinput, utf8_target))
3674 previous_prepend = locinput;
3675 locinput += UTF8SKIP(locinput);
3678 /* As noted above, if we matched a prepend character, but
3679 * the next thing won't match, back off the last prepend we
3680 * matched, as it is guaranteed to match the begin */
3681 if (previous_prepend
3682 && (locinput >= PL_regeol
3683 || ! swash_fetch(PL_utf8_X_begin,
3684 (U8*)locinput, utf8_target)))
3686 locinput = previous_prepend;
3689 /* Note that here we know PL_regeol > locinput, as we
3690 * tested that upon input to this switch case, and if we
3691 * moved locinput forward, we tested the result just above
3692 * and it either passed, or we backed off so that it will
3694 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3696 /* Here did not match the required 'Begin' in the
3697 * second term. So just match the very first
3698 * character, the '.' of the final term of the regex */
3699 locinput = starting + UTF8SKIP(starting);
3702 /* Here is the beginning of a character that can have
3703 * an extender. It is either a hangul syllable, or a
3705 if (swash_fetch(PL_utf8_X_non_hangul,
3706 (U8*)locinput, utf8_target))
3709 /* Here not a Hangul syllable, must be a
3710 * ('! * Control') */
3711 locinput += UTF8SKIP(locinput);
3714 /* Here is a Hangul syllable. It can be composed
3715 * of several individual characters. One
3716 * possibility is T+ */
3717 if (swash_fetch(PL_utf8_X_T,
3718 (U8*)locinput, utf8_target))
3720 while (locinput < PL_regeol
3721 && swash_fetch(PL_utf8_X_T,
3722 (U8*)locinput, utf8_target))
3724 locinput += UTF8SKIP(locinput);
3728 /* Here, not T+, but is a Hangul. That means
3729 * it is one of the others: L, LV, LVT or V,
3731 * L* (L | LVT T* | V V* T* | LV V* T*) */
3734 while (locinput < PL_regeol
3735 && swash_fetch(PL_utf8_X_L,
3736 (U8*)locinput, utf8_target))
3738 locinput += UTF8SKIP(locinput);
3741 /* Here, have exhausted L*. If the next
3742 * character is not an LV, LVT nor V, it means
3743 * we had to have at least one L, so matches L+
3744 * in the original equation, we have a complete
3745 * hangul syllable. Are done. */
3747 if (locinput < PL_regeol
3748 && swash_fetch(PL_utf8_X_LV_LVT_V,
3749 (U8*)locinput, utf8_target))
3752 /* Otherwise keep going. Must be LV, LVT
3753 * or V. See if LVT */
3754 if (swash_fetch(PL_utf8_X_LVT,
3755 (U8*)locinput, utf8_target))
3757 locinput += UTF8SKIP(locinput);
3760 /* Must be V or LV. Take it, then
3762 locinput += UTF8SKIP(locinput);
3763 while (locinput < PL_regeol
3764 && swash_fetch(PL_utf8_X_V,
3765 (U8*)locinput, utf8_target))
3767 locinput += UTF8SKIP(locinput);
3771 /* And any of LV, LVT, or V can be followed
3773 while (locinput < PL_regeol
3774 && swash_fetch(PL_utf8_X_T,
3778 locinput += UTF8SKIP(locinput);
3784 /* Match any extender */
3785 while (locinput < PL_regeol
3786 && swash_fetch(PL_utf8_X_extend,
3787 (U8*)locinput, utf8_target))
3789 locinput += UTF8SKIP(locinput);
3793 if (locinput > PL_regeol) sayNO;
3795 nextchr = UCHARAT(locinput);
3802 PL_reg_flags |= RF_tainted;
3807 n = reg_check_named_buff_matched(rex,scan);
3810 type = REF + ( type - NREF );
3817 PL_reg_flags |= RF_tainted;
3821 n = ARG(scan); /* which paren pair */
3824 ln = PL_regoffs[n].start;
3825 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3826 if (*PL_reglastparen < n || ln == -1)
3827 sayNO; /* Do not match unless seen CLOSEn. */
3828 if (ln == PL_regoffs[n].end)
3832 if (utf8_target && type != REF) { /* REF can do byte comparison */
3834 const char *e = PL_bostr + PL_regoffs[n].end;
3836 * Note that we can't do the "other character" lookup trick as
3837 * in the 8-bit case (no pun intended) because in Unicode we
3838 * have to map both upper and title case to lower case.
3842 STRLEN ulen1, ulen2;
3843 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3844 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3848 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3849 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3850 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3857 nextchr = UCHARAT(locinput);
3861 /* Inline the first character, for speed. */
3862 if (UCHARAT(s) != nextchr &&
3864 (UCHARAT(s) != (type == REFF
3865 ? PL_fold : PL_fold_locale)[nextchr])))
3867 ln = PL_regoffs[n].end - ln;
3868 if (locinput + ln > PL_regeol)
3870 if (ln > 1 && (type == REF
3871 ? memNE(s, locinput, ln)
3873 ? ! foldEQ(s, locinput, ln)
3874 : ! foldEQ_locale(s, locinput, ln))))
3877 nextchr = UCHARAT(locinput);
3887 #define ST st->u.eval
3892 regexp_internal *rei;
3893 regnode *startpoint;
3896 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
3897 if (cur_eval && cur_eval->locinput==locinput) {
3898 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3899 Perl_croak(aTHX_ "Infinite recursion in regex");
3900 if ( ++nochange_depth > max_nochange_depth )
3902 "Pattern subroutine nesting without pos change"
3903 " exceeded limit in regex");
3910 (void)ReREFCNT_inc(rex_sv);
3911 if (OP(scan)==GOSUB) {
3912 startpoint = scan + ARG2L(scan);
3913 ST.close_paren = ARG(scan);
3915 startpoint = rei->program+1;
3918 goto eval_recurse_doit;
3920 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3921 if (cur_eval && cur_eval->locinput==locinput) {
3922 if ( ++nochange_depth > max_nochange_depth )
3923 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3928 /* execute the code in the {...} */
3930 SV ** const before = SP;
3931 OP_4tree * const oop = PL_op;
3932 COP * const ocurcop = PL_curcop;
3934 char *saved_regeol = PL_regeol;
3937 PL_op = (OP_4tree*)rexi->data->data[n];
3938 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3939 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3940 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3941 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
3944 SV *sv_mrk = get_sv("REGMARK", 1);
3945 sv_setsv(sv_mrk, sv_yes_mark);
3948 CALLRUNOPS(aTHX); /* Scalar context. */
3951 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3958 PAD_RESTORE_LOCAL(old_comppad);
3959 PL_curcop = ocurcop;
3960 PL_regeol = saved_regeol;
3963 sv_setsv(save_scalar(PL_replgv), ret);
3967 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3970 /* extract RE object from returned value; compiling if
3976 SV *const sv = SvRV(ret);
3978 if (SvTYPE(sv) == SVt_REGEXP) {
3980 } else if (SvSMAGICAL(sv)) {
3981 mg = mg_find(sv, PERL_MAGIC_qr);
3984 } else if (SvTYPE(ret) == SVt_REGEXP) {
3986 } else if (SvSMAGICAL(ret)) {
3987 if (SvGMAGICAL(ret)) {
3988 /* I don't believe that there is ever qr magic
3990 assert(!mg_find(ret, PERL_MAGIC_qr));
3991 sv_unmagic(ret, PERL_MAGIC_qr);
3994 mg = mg_find(ret, PERL_MAGIC_qr);
3995 /* testing suggests mg only ends up non-NULL for
3996 scalars who were upgraded and compiled in the
3997 else block below. In turn, this is only
3998 triggered in the "postponed utf8 string" tests
4004 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4008 rx = reg_temp_copy(NULL, rx);
4012 const I32 osize = PL_regsize;
4015 assert (SvUTF8(ret));
4016 } else if (SvUTF8(ret)) {
4017 /* Not doing UTF-8, despite what the SV says. Is
4018 this only if we're trapped in use 'bytes'? */
4019 /* Make a copy of the octet sequence, but without
4020 the flag on, as the compiler now honours the
4021 SvUTF8 flag on ret. */
4023 const char *const p = SvPV(ret, len);
4024 ret = newSVpvn_flags(p, len, SVs_TEMP);
4026 rx = CALLREGCOMP(ret, pm_flags);
4028 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4030 /* This isn't a first class regexp. Instead, it's
4031 caching a regexp onto an existing, Perl visible
4033 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4038 re = (struct regexp *)SvANY(rx);
4040 RXp_MATCH_COPIED_off(re);
4041 re->subbeg = rex->subbeg;
4042 re->sublen = rex->sublen;
4045 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4046 "Matching embedded");
4048 startpoint = rei->program + 1;
4049 ST.close_paren = 0; /* only used for GOSUB */
4050 /* borrowed from regtry */
4051 if (PL_reg_start_tmpl <= re->nparens) {
4052 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4053 if(PL_reg_start_tmp)
4054 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4056 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4059 eval_recurse_doit: /* Share code with GOSUB below this line */
4060 /* run the pattern returned from (??{...}) */
4061 ST.cp = regcppush(0); /* Save *all* the positions. */
4062 REGCP_SET(ST.lastcp);
4064 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4066 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4067 PL_reglastparen = &re->lastparen;
4068 PL_reglastcloseparen = &re->lastcloseparen;
4070 re->lastcloseparen = 0;
4072 PL_reginput = locinput;
4075 /* XXXX This is too dramatic a measure... */
4078 ST.toggle_reg_flags = PL_reg_flags;
4080 PL_reg_flags |= RF_utf8;
4082 PL_reg_flags &= ~RF_utf8;
4083 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4085 ST.prev_rex = rex_sv;
4086 ST.prev_curlyx = cur_curlyx;
4087 SETREX(rex_sv,re_sv);
4092 ST.prev_eval = cur_eval;
4094 /* now continue from first node in postoned RE */
4095 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4098 /* logical is 1, /(?(?{...})X|Y)/ */
4099 sw = cBOOL(SvTRUE(ret));
4104 case EVAL_AB: /* cleanup after a successful (??{A})B */
4105 /* note: this is called twice; first after popping B, then A */
4106 PL_reg_flags ^= ST.toggle_reg_flags;
4107 ReREFCNT_dec(rex_sv);
4108 SETREX(rex_sv,ST.prev_rex);
4109 rex = (struct regexp *)SvANY(rex_sv);
4110 rexi = RXi_GET(rex);
4112 cur_eval = ST.prev_eval;
4113 cur_curlyx = ST.prev_curlyx;
4115 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4116 PL_reglastparen = &rex->lastparen;
4117 PL_reglastcloseparen = &rex->lastcloseparen;
4118 /* also update PL_regoffs */
4119 PL_regoffs = rex->offs;
4121 /* XXXX This is too dramatic a measure... */
4123 if ( nochange_depth )
4128 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4129 /* note: this is called twice; first after popping B, then A */
4130 PL_reg_flags ^= ST.toggle_reg_flags;
4131 ReREFCNT_dec(rex_sv);
4132 SETREX(rex_sv,ST.prev_rex);
4133 rex = (struct regexp *)SvANY(rex_sv);
4134 rexi = RXi_GET(rex);
4135 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4136 PL_reglastparen = &rex->lastparen;
4137 PL_reglastcloseparen = &rex->lastcloseparen;
4139 PL_reginput = locinput;
4140 REGCP_UNWIND(ST.lastcp);
4142 cur_eval = ST.prev_eval;
4143 cur_curlyx = ST.prev_curlyx;
4144 /* XXXX This is too dramatic a measure... */
4146 if ( nochange_depth )
4152 n = ARG(scan); /* which paren pair */
4153 PL_reg_start_tmp[n] = locinput;
4159 n = ARG(scan); /* which paren pair */
4160 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4161 PL_regoffs[n].end = locinput - PL_bostr;
4162 /*if (n > PL_regsize)
4164 if (n > *PL_reglastparen)
4165 *PL_reglastparen = n;
4166 *PL_reglastcloseparen = n;
4167 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4175 cursor && OP(cursor)!=END;
4176 cursor=regnext(cursor))
4178 if ( OP(cursor)==CLOSE ){
4180 if ( n <= lastopen ) {
4182 = PL_reg_start_tmp[n] - PL_bostr;
4183 PL_regoffs[n].end = locinput - PL_bostr;
4184 /*if (n > PL_regsize)
4186 if (n > *PL_reglastparen)
4187 *PL_reglastparen = n;
4188 *PL_reglastcloseparen = n;
4189 if ( n == ARG(scan) || (cur_eval &&
4190 cur_eval->u.eval.close_paren == n))
4199 n = ARG(scan); /* which paren pair */
4200 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4203 /* reg_check_named_buff_matched returns 0 for no match */
4204 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4208 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4214 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4216 next = NEXTOPER(NEXTOPER(scan));
4218 next = scan + ARG(scan);
4219 if (OP(next) == IFTHEN) /* Fake one. */
4220 next = NEXTOPER(NEXTOPER(next));
4224 logical = scan->flags;
4227 /*******************************************************************
4229 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4230 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4231 STAR/PLUS/CURLY/CURLYN are used instead.)
4233 A*B is compiled as <CURLYX><A><WHILEM><B>
4235 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4236 state, which contains the current count, initialised to -1. It also sets
4237 cur_curlyx to point to this state, with any previous value saved in the
4240 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4241 since the pattern may possibly match zero times (i.e. it's a while {} loop
4242 rather than a do {} while loop).
4244 Each entry to WHILEM represents a successful match of A. The count in the
4245 CURLYX block is incremented, another WHILEM state is pushed, and execution
4246 passes to A or B depending on greediness and the current count.
4248 For example, if matching against the string a1a2a3b (where the aN are
4249 substrings that match /A/), then the match progresses as follows: (the
4250 pushed states are interspersed with the bits of strings matched so far):
4253 <CURLYX cnt=0><WHILEM>
4254 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4255 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4256 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4257 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4259 (Contrast this with something like CURLYM, which maintains only a single
4263 a1 <CURLYM cnt=1> a2
4264 a1 a2 <CURLYM cnt=2> a3
4265 a1 a2 a3 <CURLYM cnt=3> b
4268 Each WHILEM state block marks a point to backtrack to upon partial failure
4269 of A or B, and also contains some minor state data related to that
4270 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4271 overall state, such as the count, and pointers to the A and B ops.
4273 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4274 must always point to the *current* CURLYX block, the rules are:
4276 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4277 and set cur_curlyx to point the new block.
4279 When popping the CURLYX block after a successful or unsuccessful match,
4280 restore the previous cur_curlyx.
4282 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4283 to the outer one saved in the CURLYX block.
4285 When popping the WHILEM block after a successful or unsuccessful B match,
4286 restore the previous cur_curlyx.
4288 Here's an example for the pattern (AI* BI)*BO
4289 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4292 curlyx backtrack stack
4293 ------ ---------------
4295 CO <CO prev=NULL> <WO>
4296 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4297 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4298 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4300 At this point the pattern succeeds, and we work back down the stack to
4301 clean up, restoring as we go:
4303 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4304 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4305 CO <CO prev=NULL> <WO>
4308 *******************************************************************/
4310 #define ST st->u.curlyx
4312 case CURLYX: /* start of /A*B/ (for complex A) */
4314 /* No need to save/restore up to this paren */
4315 I32 parenfloor = scan->flags;
4317 assert(next); /* keep Coverity happy */
4318 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4321 /* XXXX Probably it is better to teach regpush to support
4322 parenfloor > PL_regsize... */
4323 if (parenfloor > (I32)*PL_reglastparen)
4324 parenfloor = *PL_reglastparen; /* Pessimization... */
4326 ST.prev_curlyx= cur_curlyx;
4328 ST.cp = PL_savestack_ix;
4330 /* these fields contain the state of the current curly.
4331 * they are accessed by subsequent WHILEMs */
4332 ST.parenfloor = parenfloor;
4337 ST.count = -1; /* this will be updated by WHILEM */
4338 ST.lastloc = NULL; /* this will be updated by WHILEM */
4340 PL_reginput = locinput;
4341 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4345 case CURLYX_end: /* just finished matching all of A*B */
4346 cur_curlyx = ST.prev_curlyx;
4350 case CURLYX_end_fail: /* just failed to match all of A*B */
4352 cur_curlyx = ST.prev_curlyx;
4358 #define ST st->u.whilem
4360 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4362 /* see the discussion above about CURLYX/WHILEM */
4364 int min = ARG1(cur_curlyx->u.curlyx.me);
4365 int max = ARG2(cur_curlyx->u.curlyx.me);
4366 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4368 assert(cur_curlyx); /* keep Coverity happy */
4369 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4370 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4371 ST.cache_offset = 0;
4374 PL_reginput = locinput;
4376 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4377 "%*s whilem: matched %ld out of %d..%d\n",
4378 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4381 /* First just match a string of min A's. */
4384 cur_curlyx->u.curlyx.lastloc = locinput;
4385 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4389 /* If degenerate A matches "", assume A done. */
4391 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4392 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4393 "%*s whilem: empty match detected, trying continuation...\n",
4394 REPORT_CODE_OFF+depth*2, "")
4396 goto do_whilem_B_max;
4399 /* super-linear cache processing */
4403 if (!PL_reg_maxiter) {
4404 /* start the countdown: Postpone detection until we
4405 * know the match is not *that* much linear. */
4406 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4407 /* possible overflow for long strings and many CURLYX's */
4408 if (PL_reg_maxiter < 0)
4409 PL_reg_maxiter = I32_MAX;
4410 PL_reg_leftiter = PL_reg_maxiter;
4413 if (PL_reg_leftiter-- == 0) {
4414 /* initialise cache */
4415 const I32 size = (PL_reg_maxiter + 7)/8;
4416 if (PL_reg_poscache) {
4417 if ((I32)PL_reg_poscache_size < size) {
4418 Renew(PL_reg_poscache, size, char);
4419 PL_reg_poscache_size = size;
4421 Zero(PL_reg_poscache, size, char);
4424 PL_reg_poscache_size = size;
4425 Newxz(PL_reg_poscache, size, char);
4427 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4428 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4429 PL_colors[4], PL_colors[5])
4433 if (PL_reg_leftiter < 0) {
4434 /* have we already failed at this position? */
4436 offset = (scan->flags & 0xf) - 1
4437 + (locinput - PL_bostr) * (scan->flags>>4);
4438 mask = 1 << (offset % 8);
4440 if (PL_reg_poscache[offset] & mask) {
4441 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4442 "%*s whilem: (cache) already tried at this position...\n",
4443 REPORT_CODE_OFF+depth*2, "")
4445 sayNO; /* cache records failure */
4447 ST.cache_offset = offset;
4448 ST.cache_mask = mask;
4452 /* Prefer B over A for minimal matching. */
4454 if (cur_curlyx->u.curlyx.minmod) {
4455 ST.save_curlyx = cur_curlyx;
4456 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4457 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4458 REGCP_SET(ST.lastcp);
4459 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4463 /* Prefer A over B for maximal matching. */
4465 if (n < max) { /* More greed allowed? */
4466 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4467 cur_curlyx->u.curlyx.lastloc = locinput;
4468 REGCP_SET(ST.lastcp);
4469 PUSH_STATE_GOTO(WHILEM_A_max, A);
4472 goto do_whilem_B_max;
4476 case WHILEM_B_min: /* just matched B in a minimal match */
4477 case WHILEM_B_max: /* just matched B in a maximal match */
4478 cur_curlyx = ST.save_curlyx;
4482 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4483 cur_curlyx = ST.save_curlyx;
4484 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4485 cur_curlyx->u.curlyx.count--;
4489 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4490 REGCP_UNWIND(ST.lastcp);
4493 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4494 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4495 cur_curlyx->u.curlyx.count--;
4499 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4500 REGCP_UNWIND(ST.lastcp);
4501 regcppop(rex); /* Restore some previous $<digit>s? */
4502 PL_reginput = locinput;
4503 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4504 "%*s whilem: failed, trying continuation...\n",
4505 REPORT_CODE_OFF+depth*2, "")
4508 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4509 && ckWARN(WARN_REGEXP)
4510 && !(PL_reg_flags & RF_warned))
4512 PL_reg_flags |= RF_warned;
4513 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4514 "Complex regular subexpression recursion",
4519 ST.save_curlyx = cur_curlyx;
4520 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4521 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4524 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4525 cur_curlyx = ST.save_curlyx;
4526 REGCP_UNWIND(ST.lastcp);
4529 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4530 /* Maximum greed exceeded */
4531 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4532 && ckWARN(WARN_REGEXP)
4533 && !(PL_reg_flags & RF_warned))
4535 PL_reg_flags |= RF_warned;
4536 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4537 "%s limit (%d) exceeded",
4538 "Complex regular subexpression recursion",
4541 cur_curlyx->u.curlyx.count--;
4545 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4546 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4548 /* Try grabbing another A and see if it helps. */
4549 PL_reginput = locinput;
4550 cur_curlyx->u.curlyx.lastloc = locinput;
4551 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4552 REGCP_SET(ST.lastcp);
4553 PUSH_STATE_GOTO(WHILEM_A_min,
4554 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4558 #define ST st->u.branch
4560 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4561 next = scan + ARG(scan);
4564 scan = NEXTOPER(scan);
4567 case BRANCH: /* /(...|A|...)/ */
4568 scan = NEXTOPER(scan); /* scan now points to inner node */
4569 ST.lastparen = *PL_reglastparen;
4570 ST.next_branch = next;
4572 PL_reginput = locinput;
4574 /* Now go into the branch */
4576 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4578 PUSH_STATE_GOTO(BRANCH_next, scan);
4582 PL_reginput = locinput;
4583 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4584 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4585 PUSH_STATE_GOTO(CUTGROUP_next,next);
4587 case CUTGROUP_next_fail:
4590 if (st->u.mark.mark_name)
4591 sv_commit = st->u.mark.mark_name;
4597 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4602 REGCP_UNWIND(ST.cp);
4603 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4604 PL_regoffs[n].end = -1;
4605 *PL_reglastparen = n;
4606 /*dmq: *PL_reglastcloseparen = n; */
4607 scan = ST.next_branch;
4608 /* no more branches? */
4609 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4611 PerlIO_printf( Perl_debug_log,
4612 "%*s %sBRANCH failed...%s\n",
4613 REPORT_CODE_OFF+depth*2, "",
4619 continue; /* execute next BRANCH[J] op */
4627 #define ST st->u.curlym
4629 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4631 /* This is an optimisation of CURLYX that enables us to push
4632 * only a single backtracking state, no matter how many matches
4633 * there are in {m,n}. It relies on the pattern being constant
4634 * length, with no parens to influence future backrefs
4638 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4640 /* if paren positive, emulate an OPEN/CLOSE around A */
4642 U32 paren = ST.me->flags;
4643 if (paren > PL_regsize)
4645 if (paren > *PL_reglastparen)
4646 *PL_reglastparen = paren;
4647 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4655 ST.c1 = CHRTEST_UNINIT;
4658 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4661 curlym_do_A: /* execute the A in /A{m,n}B/ */
4662 PL_reginput = locinput;
4663 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4666 case CURLYM_A: /* we've just matched an A */
4667 locinput = st->locinput;
4668 nextchr = UCHARAT(locinput);
4671 /* after first match, determine A's length: u.curlym.alen */
4672 if (ST.count == 1) {
4673 if (PL_reg_match_utf8) {
4675 while (s < PL_reginput) {
4681 ST.alen = PL_reginput - locinput;
4684 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4687 PerlIO_printf(Perl_debug_log,
4688 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4689 (int)(REPORT_CODE_OFF+(depth*2)), "",
4690 (IV) ST.count, (IV)ST.alen)
4693 locinput = PL_reginput;
4695 if (cur_eval && cur_eval->u.eval.close_paren &&
4696 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4700 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4701 if ( max == REG_INFTY || ST.count < max )
4702 goto curlym_do_A; /* try to match another A */
4704 goto curlym_do_B; /* try to match B */
4706 case CURLYM_A_fail: /* just failed to match an A */
4707 REGCP_UNWIND(ST.cp);
4709 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4710 || (cur_eval && cur_eval->u.eval.close_paren &&
4711 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4714 curlym_do_B: /* execute the B in /A{m,n}B/ */
4715 PL_reginput = locinput;
4716 if (ST.c1 == CHRTEST_UNINIT) {
4717 /* calculate c1 and c2 for possible match of 1st char
4718 * following curly */
4719 ST.c1 = ST.c2 = CHRTEST_VOID;
4720 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4721 regnode *text_node = ST.B;
4722 if (! HAS_TEXT(text_node))
4723 FIND_NEXT_IMPT(text_node);
4726 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4728 But the former is redundant in light of the latter.
4730 if this changes back then the macro for
4731 IS_TEXT and friends need to change.
4733 if (PL_regkind[OP(text_node)] == EXACT)
4736 ST.c1 = (U8)*STRING(text_node);
4738 (IS_TEXTF(text_node))
4740 : (IS_TEXTFL(text_node))
4741 ? PL_fold_locale[ST.c1]
4748 PerlIO_printf(Perl_debug_log,
4749 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4750 (int)(REPORT_CODE_OFF+(depth*2)),
4753 if (ST.c1 != CHRTEST_VOID
4754 && UCHARAT(PL_reginput) != ST.c1
4755 && UCHARAT(PL_reginput) != ST.c2)
4757 /* simulate B failing */
4759 PerlIO_printf(Perl_debug_log,
4760 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
4761 (int)(REPORT_CODE_OFF+(depth*2)),"",
4764 state_num = CURLYM_B_fail;
4765 goto reenter_switch;
4769 /* mark current A as captured */
4770 I32 paren = ST.me->flags;
4772 PL_regoffs[paren].start
4773 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4774 PL_regoffs[paren].end = PL_reginput - PL_bostr;
4775 /*dmq: *PL_reglastcloseparen = paren; */
4778 PL_regoffs[paren].end = -1;
4779 if (cur_eval && cur_eval->u.eval.close_paren &&
4780 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4789 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4792 case CURLYM_B_fail: /* just failed to match a B */
4793 REGCP_UNWIND(ST.cp);
4795 I32 max = ARG2(ST.me);
4796 if (max != REG_INFTY && ST.count == max)
4798 goto curlym_do_A; /* try to match a further A */
4800 /* backtrack one A */
4801 if (ST.count == ARG1(ST.me) /* min */)
4804 locinput = HOPc(locinput, -ST.alen);
4805 goto curlym_do_B; /* try to match B */
4808 #define ST st->u.curly
4810 #define CURLY_SETPAREN(paren, success) \
4813 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
4814 PL_regoffs[paren].end = locinput - PL_bostr; \
4815 *PL_reglastcloseparen = paren; \
4818 PL_regoffs[paren].end = -1; \
4821 case STAR: /* /A*B/ where A is width 1 */
4825 scan = NEXTOPER(scan);
4827 case PLUS: /* /A+B/ where A is width 1 */
4831 scan = NEXTOPER(scan);
4833 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4834 ST.paren = scan->flags; /* Which paren to set */
4835 if (ST.paren > PL_regsize)
4836 PL_regsize = ST.paren;
4837 if (ST.paren > *PL_reglastparen)
4838 *PL_reglastparen = ST.paren;
4839 ST.min = ARG1(scan); /* min to match */
4840 ST.max = ARG2(scan); /* max to match */
4841 if (cur_eval && cur_eval->u.eval.close_paren &&
4842 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4846 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4848 case CURLY: /* /A{m,n}B/ where A is width 1 */
4850 ST.min = ARG1(scan); /* min to match */
4851 ST.max = ARG2(scan); /* max to match */
4852 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4855 * Lookahead to avoid useless match attempts
4856 * when we know what character comes next.
4858 * Used to only do .*x and .*?x, but now it allows
4859 * for )'s, ('s and (?{ ... })'s to be in the way
4860 * of the quantifier and the EXACT-like node. -- japhy
4863 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4865 if (HAS_TEXT(next) || JUMPABLE(next)) {
4867 regnode *text_node = next;
4869 if (! HAS_TEXT(text_node))
4870 FIND_NEXT_IMPT(text_node);
4872 if (! HAS_TEXT(text_node))
4873 ST.c1 = ST.c2 = CHRTEST_VOID;
4875 if ( PL_regkind[OP(text_node)] != EXACT ) {
4876 ST.c1 = ST.c2 = CHRTEST_VOID;
4877 goto assume_ok_easy;
4880 s = (U8*)STRING(text_node);
4882 /* Currently we only get here when
4884 PL_rekind[OP(text_node)] == EXACT
4886 if this changes back then the macro for IS_TEXT and
4887 friends need to change. */
4890 if (IS_TEXTF(text_node))
4891 ST.c2 = PL_fold[ST.c1];
4892 else if (IS_TEXTFL(text_node))
4893 ST.c2 = PL_fold_locale[ST.c1];
4895 else { /* UTF_PATTERN */
4896 if (IS_TEXTF(text_node)) {
4897 STRLEN ulen1, ulen2;
4898 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4899 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4901 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4902 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4904 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4906 0 : UTF8_ALLOW_ANY);
4907 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4909 0 : UTF8_ALLOW_ANY);
4911 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4913 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4918 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4925 ST.c1 = ST.c2 = CHRTEST_VOID;
4930 PL_reginput = locinput;
4933 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4936 locinput = PL_reginput;
4938 if (ST.c1 == CHRTEST_VOID)
4939 goto curly_try_B_min;
4941 ST.oldloc = locinput;
4943 /* set ST.maxpos to the furthest point along the
4944 * string that could possibly match */
4945 if (ST.max == REG_INFTY) {
4946 ST.maxpos = PL_regeol - 1;
4948 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4951 else if (utf8_target) {
4952 int m = ST.max - ST.min;
4953 for (ST.maxpos = locinput;
4954 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4955 ST.maxpos += UTF8SKIP(ST.maxpos);
4958 ST.maxpos = locinput + ST.max - ST.min;
4959 if (ST.maxpos >= PL_regeol)
4960 ST.maxpos = PL_regeol - 1;
4962 goto curly_try_B_min_known;
4966 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4967 locinput = PL_reginput;
4968 if (ST.count < ST.min)
4970 if ((ST.count > ST.min)
4971 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4973 /* A{m,n} must come at the end of the string, there's
4974 * no point in backing off ... */
4976 /* ...except that $ and \Z can match before *and* after
4977 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4978 We may back off by one in this case. */
4979 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4983 goto curly_try_B_max;
4988 case CURLY_B_min_known_fail:
4989 /* failed to find B in a non-greedy match where c1,c2 valid */
4990 if (ST.paren && ST.count)
4991 PL_regoffs[ST.paren].end = -1;
4993 PL_reginput = locinput; /* Could be reset... */
4994 REGCP_UNWIND(ST.cp);
4995 /* Couldn't or didn't -- move forward. */
4996 ST.oldloc = locinput;
4998 locinput += UTF8SKIP(locinput);
5002 curly_try_B_min_known:
5003 /* find the next place where 'B' could work, then call B */
5007 n = (ST.oldloc == locinput) ? 0 : 1;
5008 if (ST.c1 == ST.c2) {
5010 /* set n to utf8_distance(oldloc, locinput) */
5011 while (locinput <= ST.maxpos &&
5012 utf8n_to_uvchr((U8*)locinput,
5013 UTF8_MAXBYTES, &len,
5014 uniflags) != (UV)ST.c1) {
5020 /* set n to utf8_distance(oldloc, locinput) */
5021 while (locinput <= ST.maxpos) {
5023 const UV c = utf8n_to_uvchr((U8*)locinput,
5024 UTF8_MAXBYTES, &len,
5026 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5034 if (ST.c1 == ST.c2) {
5035 while (locinput <= ST.maxpos &&
5036 UCHARAT(locinput) != ST.c1)
5040 while (locinput <= ST.maxpos
5041 && UCHARAT(locinput) != ST.c1
5042 && UCHARAT(locinput) != ST.c2)
5045 n = locinput - ST.oldloc;
5047 if (locinput > ST.maxpos)
5049 /* PL_reginput == oldloc now */
5052 if (regrepeat(rex, ST.A, n, depth) < n)
5055 PL_reginput = locinput;
5056 CURLY_SETPAREN(ST.paren, ST.count);
5057 if (cur_eval && cur_eval->u.eval.close_paren &&
5058 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5061 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5066 case CURLY_B_min_fail:
5067 /* failed to find B in a non-greedy match where c1,c2 invalid */
5068 if (ST.paren && ST.count)
5069 PL_regoffs[ST.paren].end = -1;
5071 REGCP_UNWIND(ST.cp);
5072 /* failed -- move forward one */
5073 PL_reginput = locinput;
5074 if (regrepeat(rex, ST.A, 1, depth)) {
5076 locinput = PL_reginput;
5077 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5078 ST.count > 0)) /* count overflow ? */
5081 CURLY_SETPAREN(ST.paren, ST.count);
5082 if (cur_eval && cur_eval->u.eval.close_paren &&
5083 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5086 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5094 /* a successful greedy match: now try to match B */
5095 if (cur_eval && cur_eval->u.eval.close_paren &&
5096 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5101 if (ST.c1 != CHRTEST_VOID)
5102 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5103 UTF8_MAXBYTES, 0, uniflags)
5104 : (UV) UCHARAT(PL_reginput);
5105 /* If it could work, try it. */
5106 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5107 CURLY_SETPAREN(ST.paren, ST.count);
5108 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5113 case CURLY_B_max_fail:
5114 /* failed to find B in a greedy match */
5115 if (ST.paren && ST.count)
5116 PL_regoffs[ST.paren].end = -1;
5118 REGCP_UNWIND(ST.cp);
5120 if (--ST.count < ST.min)
5122 PL_reginput = locinput = HOPc(locinput, -1);
5123 goto curly_try_B_max;
5130 /* we've just finished A in /(??{A})B/; now continue with B */
5132 st->u.eval.toggle_reg_flags
5133 = cur_eval->u.eval.toggle_reg_flags;
5134 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5136 st->u.eval.prev_rex = rex_sv; /* inner */
5137 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5138 rex = (struct regexp *)SvANY(rex_sv);
5139 rexi = RXi_GET(rex);
5140 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5141 ReREFCNT_inc(rex_sv);
5142 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5144 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5145 PL_reglastparen = &rex->lastparen;
5146 PL_reglastcloseparen = &rex->lastcloseparen;
5148 REGCP_SET(st->u.eval.lastcp);
5149 PL_reginput = locinput;
5151 /* Restore parens of the outer rex without popping the
5153 tmpix = PL_savestack_ix;
5154 PL_savestack_ix = cur_eval->u.eval.lastcp;
5156 PL_savestack_ix = tmpix;
5158 st->u.eval.prev_eval = cur_eval;
5159 cur_eval = cur_eval->u.eval.prev_eval;
5161 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5162 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5163 if ( nochange_depth )
5166 PUSH_YES_STATE_GOTO(EVAL_AB,
5167 st->u.eval.prev_eval->u.eval.B); /* match B */
5170 if (locinput < reginfo->till) {
5171 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5172 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5174 (long)(locinput - PL_reg_starttry),
5175 (long)(reginfo->till - PL_reg_starttry),
5178 sayNO_SILENT; /* Cannot match: too short. */
5180 PL_reginput = locinput; /* put where regtry can find it */
5181 sayYES; /* Success! */
5183 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5185 PerlIO_printf(Perl_debug_log,
5186 "%*s %ssubpattern success...%s\n",
5187 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5188 PL_reginput = locinput; /* put where regtry can find it */
5189 sayYES; /* Success! */
5192 #define ST st->u.ifmatch
5194 case SUSPEND: /* (?>A) */
5196 PL_reginput = locinput;
5199 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5201 goto ifmatch_trivial_fail_test;
5203 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5205 ifmatch_trivial_fail_test:
5207 char * const s = HOPBACKc(locinput, scan->flags);
5212 sw = 1 - cBOOL(ST.wanted);
5216 next = scan + ARG(scan);
5224 PL_reginput = locinput;
5228 ST.logical = logical;
5229 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5231 /* execute body of (?...A) */
5232 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5235 case IFMATCH_A_fail: /* body of (?...A) failed */
5236 ST.wanted = !ST.wanted;
5239 case IFMATCH_A: /* body of (?...A) succeeded */
5241 sw = cBOOL(ST.wanted);
5243 else if (!ST.wanted)
5246 if (OP(ST.me) == SUSPEND)
5247 locinput = PL_reginput;
5249 locinput = PL_reginput = st->locinput;
5250 nextchr = UCHARAT(locinput);
5252 scan = ST.me + ARG(ST.me);
5255 continue; /* execute B */
5260 next = scan + ARG(scan);
5265 reginfo->cutpoint = PL_regeol;
5268 PL_reginput = locinput;
5270 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5271 PUSH_STATE_GOTO(COMMIT_next,next);
5273 case COMMIT_next_fail:
5280 #define ST st->u.mark
5282 ST.prev_mark = mark_state;
5283 ST.mark_name = sv_commit = sv_yes_mark
5284 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5286 ST.mark_loc = PL_reginput = locinput;
5287 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5289 case MARKPOINT_next:
5290 mark_state = ST.prev_mark;
5293 case MARKPOINT_next_fail:
5294 if (popmark && sv_eq(ST.mark_name,popmark))
5296 if (ST.mark_loc > startpoint)
5297 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5298 popmark = NULL; /* we found our mark */
5299 sv_commit = ST.mark_name;
5302 PerlIO_printf(Perl_debug_log,
5303 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5304 REPORT_CODE_OFF+depth*2, "",
5305 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5308 mark_state = ST.prev_mark;
5309 sv_yes_mark = mark_state ?
5310 mark_state->u.mark.mark_name : NULL;
5314 PL_reginput = locinput;
5316 /* (*SKIP) : if we fail we cut here*/
5317 ST.mark_name = NULL;
5318 ST.mark_loc = locinput;
5319 PUSH_STATE_GOTO(SKIP_next,next);
5321 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5322 otherwise do nothing. Meaning we need to scan
5324 regmatch_state *cur = mark_state;
5325 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5328 if ( sv_eq( cur->u.mark.mark_name,
5331 ST.mark_name = find;
5332 PUSH_STATE_GOTO( SKIP_next, next );
5334 cur = cur->u.mark.prev_mark;
5337 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5339 case SKIP_next_fail:
5341 /* (*CUT:NAME) - Set up to search for the name as we
5342 collapse the stack*/
5343 popmark = ST.mark_name;
5345 /* (*CUT) - No name, we cut here.*/
5346 if (ST.mark_loc > startpoint)
5347 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5348 /* but we set sv_commit to latest mark_name if there
5349 is one so they can test to see how things lead to this
5352 sv_commit=mark_state->u.mark.mark_name;
5360 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5362 } else if ( 0xDF == n && !utf8_target && !UTF_PATTERN ) {
5365 U8 folded[UTF8_MAXBYTES_CASE+1];
5367 const char * const l = locinput;
5368 char *e = PL_regeol;
5369 to_uni_fold(n, folded, &foldlen);
5371 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5372 l, &e, 0, utf8_target)) {
5377 nextchr = UCHARAT(locinput);
5380 if ((n=is_LNBREAK(locinput,utf8_target))) {
5382 nextchr = UCHARAT(locinput);
5387 #define CASE_CLASS(nAmE) \
5389 if ((n=is_##nAmE(locinput,utf8_target))) { \
5391 nextchr = UCHARAT(locinput); \
5396 if ((n=is_##nAmE(locinput,utf8_target))) { \
5399 locinput += UTF8SKIP(locinput); \
5400 nextchr = UCHARAT(locinput); \
5405 CASE_CLASS(HORIZWS);
5409 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5410 PTR2UV(scan), OP(scan));
5411 Perl_croak(aTHX_ "regexp memory corruption");
5415 /* switch break jumps here */
5416 scan = next; /* prepare to execute the next op and ... */
5417 continue; /* ... jump back to the top, reusing st */
5421 /* push a state that backtracks on success */
5422 st->u.yes.prev_yes_state = yes_state;
5426 /* push a new regex state, then continue at scan */
5428 regmatch_state *newst;
5431 regmatch_state *cur = st;
5432 regmatch_state *curyes = yes_state;
5434 regmatch_slab *slab = PL_regmatch_slab;
5435 for (;curd > -1;cur--,curd--) {
5436 if (cur < SLAB_FIRST(slab)) {
5438 cur = SLAB_LAST(slab);
5440 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5441 REPORT_CODE_OFF + 2 + depth * 2,"",
5442 curd, PL_reg_name[cur->resume_state],
5443 (curyes == cur) ? "yes" : ""
5446 curyes = cur->u.yes.prev_yes_state;
5449 DEBUG_STATE_pp("push")
5452 st->locinput = locinput;
5454 if (newst > SLAB_LAST(PL_regmatch_slab))
5455 newst = S_push_slab(aTHX);
5456 PL_regmatch_state = newst;
5458 locinput = PL_reginput;
5459 nextchr = UCHARAT(locinput);
5467 * We get here only if there's trouble -- normally "case END" is
5468 * the terminating point.
5470 Perl_croak(aTHX_ "corrupted regexp pointers");
5476 /* we have successfully completed a subexpression, but we must now
5477 * pop to the state marked by yes_state and continue from there */
5478 assert(st != yes_state);
5480 while (st != yes_state) {
5482 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5483 PL_regmatch_slab = PL_regmatch_slab->prev;
5484 st = SLAB_LAST(PL_regmatch_slab);
5488 DEBUG_STATE_pp("pop (no final)");
5490 DEBUG_STATE_pp("pop (yes)");
5496 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5497 || yes_state > SLAB_LAST(PL_regmatch_slab))
5499 /* not in this slab, pop slab */
5500 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5501 PL_regmatch_slab = PL_regmatch_slab->prev;
5502 st = SLAB_LAST(PL_regmatch_slab);
5504 depth -= (st - yes_state);
5507 yes_state = st->u.yes.prev_yes_state;
5508 PL_regmatch_state = st;
5511 locinput= st->locinput;
5512 nextchr = UCHARAT(locinput);
5514 state_num = st->resume_state + no_final;
5515 goto reenter_switch;
5518 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5519 PL_colors[4], PL_colors[5]));
5521 if (PL_reg_eval_set) {
5522 /* each successfully executed (?{...}) block does the equivalent of
5523 * local $^R = do {...}
5524 * When popping the save stack, all these locals would be undone;
5525 * bypass this by setting the outermost saved $^R to the latest
5527 if (oreplsv != GvSV(PL_replgv))
5528 sv_setsv(oreplsv, GvSV(PL_replgv));
5535 PerlIO_printf(Perl_debug_log,
5536 "%*s %sfailed...%s\n",
5537 REPORT_CODE_OFF+depth*2, "",
5538 PL_colors[4], PL_colors[5])
5550 /* there's a previous state to backtrack to */
5552 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5553 PL_regmatch_slab = PL_regmatch_slab->prev;
5554 st = SLAB_LAST(PL_regmatch_slab);
5556 PL_regmatch_state = st;
5557 locinput= st->locinput;
5558 nextchr = UCHARAT(locinput);
5560 DEBUG_STATE_pp("pop");
5562 if (yes_state == st)
5563 yes_state = st->u.yes.prev_yes_state;
5565 state_num = st->resume_state + 1; /* failure = success + 1 */
5566 goto reenter_switch;
5571 if (rex->intflags & PREGf_VERBARG_SEEN) {
5572 SV *sv_err = get_sv("REGERROR", 1);
5573 SV *sv_mrk = get_sv("REGMARK", 1);
5575 sv_commit = &PL_sv_no;
5577 sv_yes_mark = &PL_sv_yes;
5580 sv_commit = &PL_sv_yes;
5581 sv_yes_mark = &PL_sv_no;
5583 sv_setsv(sv_err, sv_commit);
5584 sv_setsv(sv_mrk, sv_yes_mark);
5587 /* clean up; in particular, free all slabs above current one */
5588 LEAVE_SCOPE(oldsave);
5594 - regrepeat - repeatedly match something simple, report how many
5597 * [This routine now assumes that it will only match on things of length 1.
5598 * That was true before, but now we assume scan - reginput is the count,
5599 * rather than incrementing count on every character. [Er, except utf8.]]
5602 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5605 register char *scan;
5607 register char *loceol = PL_regeol;
5608 register I32 hardcount = 0;
5609 register bool utf8_target = PL_reg_match_utf8;
5611 PERL_UNUSED_ARG(depth);
5614 PERL_ARGS_ASSERT_REGREPEAT;
5617 if (max == REG_INFTY)
5619 else if (max < loceol - scan)
5620 loceol = scan + max;
5625 while (scan < loceol && hardcount < max && *scan != '\n') {
5626 scan += UTF8SKIP(scan);
5630 while (scan < loceol && *scan != '\n')
5637 while (scan < loceol && hardcount < max) {
5638 scan += UTF8SKIP(scan);
5648 case EXACT: /* length of string is 1 */
5650 while (scan < loceol && UCHARAT(scan) == c)
5653 case EXACTF: /* length of string is 1 */
5655 while (scan < loceol &&
5656 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5659 case EXACTFL: /* length of string is 1 */
5660 PL_reg_flags |= RF_tainted;
5662 while (scan < loceol &&
5663 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5669 while (hardcount < max && scan < loceol &&
5670 reginclass(prog, p, (U8*)scan, 0, utf8_target)) {
5671 scan += UTF8SKIP(scan);
5675 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5682 LOAD_UTF8_CHARCLASS_ALNUM();
5683 while (hardcount < max && scan < loceol &&
5684 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target)) {
5685 scan += UTF8SKIP(scan);
5689 while (scan < loceol && isALNUM(*scan))
5694 PL_reg_flags |= RF_tainted;
5697 while (hardcount < max && scan < loceol &&
5698 isALNUM_LC_utf8((U8*)scan)) {
5699 scan += UTF8SKIP(scan);
5703 while (scan < loceol && isALNUM_LC(*scan))
5710 LOAD_UTF8_CHARCLASS_ALNUM();
5711 while (hardcount < max && scan < loceol &&
5712 !swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target)) {
5713 scan += UTF8SKIP(scan);
5717 while (scan < loceol && !isALNUM(*scan))
5722 PL_reg_flags |= RF_tainted;
5725 while (hardcount < max && scan < loceol &&
5726 !isALNUM_LC_utf8((U8*)scan)) {
5727 scan += UTF8SKIP(scan);
5731 while (scan < loceol && !isALNUM_LC(*scan))
5738 LOAD_UTF8_CHARCLASS_SPACE();
5739 while (hardcount < max && scan < loceol &&
5741 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target))) {
5742 scan += UTF8SKIP(scan);
5746 while (scan < loceol && isSPACE(*scan))
5751 PL_reg_flags |= RF_tainted;
5754 while (hardcount < max && scan < loceol &&
5755 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5756 scan += UTF8SKIP(scan);
5760 while (scan < loceol && isSPACE_LC(*scan))
5767 LOAD_UTF8_CHARCLASS_SPACE();
5768 while (hardcount < max && scan < loceol &&
5770 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target))) {
5771 scan += UTF8SKIP(scan);
5775 while (scan < loceol && !isSPACE(*scan))
5780 PL_reg_flags |= RF_tainted;
5783 while (hardcount < max && scan < loceol &&
5784 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5785 scan += UTF8SKIP(scan);
5789 while (scan < loceol && !isSPACE_LC(*scan))
5796 LOAD_UTF8_CHARCLASS_DIGIT();
5797 while (hardcount < max && scan < loceol &&
5798 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
5799 scan += UTF8SKIP(scan);
5803 while (scan < loceol && isDIGIT(*scan))
5810 LOAD_UTF8_CHARCLASS_DIGIT();
5811 while (hardcount < max && scan < loceol &&
5812 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
5813 scan += UTF8SKIP(scan);
5817 while (scan < loceol && !isDIGIT(*scan))
5823 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
5829 LNBREAK can match two latin chars, which is ok,
5830 because we have a null terminated string, but we
5831 have to use hardcount in this situation
5833 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
5842 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
5847 while (scan < loceol && is_HORIZWS_latin1(scan))
5854 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
5855 scan += UTF8SKIP(scan);
5859 while (scan < loceol && !is_HORIZWS_latin1(scan))
5867 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
5872 while (scan < loceol && is_VERTWS_latin1(scan))
5880 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
5881 scan += UTF8SKIP(scan);
5885 while (scan < loceol && !is_VERTWS_latin1(scan))
5891 default: /* Called on something of 0 width. */
5892 break; /* So match right here or not at all. */
5898 c = scan - PL_reginput;
5902 GET_RE_DEBUG_FLAGS_DECL;
5904 SV * const prop = sv_newmortal();
5905 regprop(prog, prop, p);
5906 PerlIO_printf(Perl_debug_log,
5907 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5908 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5916 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5918 - regclass_swash - prepare the utf8 swash
5922 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5928 RXi_GET_DECL(prog,progi);
5929 const struct reg_data * const data = prog ? progi->data : NULL;
5931 PERL_ARGS_ASSERT_REGCLASS_SWASH;
5933 if (data && data->count) {
5934 const U32 n = ARG(node);
5936 if (data->what[n] == 's') {
5937 SV * const rv = MUTABLE_SV(data->data[n]);
5938 AV * const av = MUTABLE_AV(SvRV(rv));
5939 SV **const ary = AvARRAY(av);
5942 /* See the end of regcomp.c:S_regclass() for
5943 * documentation of these array elements. */
5946 a = SvROK(ary[1]) ? &ary[1] : NULL;
5947 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
5951 else if (si && doinit) {
5952 sw = swash_init("utf8", "", si, 1, 0);
5953 (void)av_store(av, 1, sw);
5970 - reginclass - determine if a character falls into a character class
5972 The n is the ANYOF regnode, the p is the target string, lenp
5973 is pointer to the maximum length of how far to go in the p
5974 (if the lenp is zero, UTF8SKIP(p) is used),
5975 utf8_target tells whether the target string is in UTF-8.
5980 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool utf8_target)
5983 const char flags = ANYOF_FLAGS(n);
5989 PERL_ARGS_ASSERT_REGINCLASS;
5991 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
5992 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5993 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
5994 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
5995 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
5996 * UTF8_ALLOW_FFFF */
5997 if (len == (STRLEN)-1)
5998 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6001 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
6002 if (utf8_target || (flags & ANYOF_UNICODE)) {
6005 if (utf8_target && !ANYOF_RUNTIME(n)) {
6006 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
6009 if (!match && utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256)
6013 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6021 utf8_p = bytes_to_utf8(p, &len);
6023 if (swash_fetch(sw, utf8_p, 1))
6025 else if (flags & ANYOF_FOLD) {
6026 if (!match && lenp && av) {
6028 for (i = 0; i <= av_len(av); i++) {
6029 SV* const sv = *av_fetch(av, i, FALSE);
6031 const char * const s = SvPV_const(sv, len);
6032 if (len <= plen && memEQ(s, (char*)utf8_p, len)) {
6040 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
6043 to_utf8_fold(utf8_p, tmpbuf, &tmplen);
6044 if (swash_fetch(sw, tmpbuf, 1))
6049 /* If we allocated a string above, free it */
6050 if (! utf8_target) Safefree(utf8_p);
6053 if (match && lenp && *lenp == 0)
6054 *lenp = UNISKIP(NATIVE_TO_UNI(c));
6056 if (!match && c < 256) {
6057 if (ANYOF_BITMAP_TEST(n, c))
6059 else if (flags & ANYOF_FOLD) {
6062 if (flags & ANYOF_LOCALE) {
6063 PL_reg_flags |= RF_tainted;
6064 f = PL_fold_locale[c];
6068 if (f != c && ANYOF_BITMAP_TEST(n, f))
6072 if (!match && (flags & ANYOF_CLASS)) {
6073 PL_reg_flags |= RF_tainted;
6075 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6076 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6077 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6078 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6079 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6080 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6081 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6082 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6083 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6084 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6085 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6086 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6087 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6088 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6089 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6090 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6091 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6092 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6093 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6094 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6095 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6096 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6097 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6098 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6099 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6100 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6101 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6102 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6103 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6104 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6105 ) /* How's that for a conditional? */
6112 return (flags & ANYOF_INVERT) ? !match : match;
6116 S_reghop3(U8 *s, I32 off, const U8* lim)
6120 PERL_ARGS_ASSERT_REGHOP3;
6123 while (off-- && s < lim) {
6124 /* XXX could check well-formedness here */
6129 while (off++ && s > lim) {
6131 if (UTF8_IS_CONTINUED(*s)) {
6132 while (s > lim && UTF8_IS_CONTINUATION(*s))
6135 /* XXX could check well-formedness here */
6142 /* there are a bunch of places where we use two reghop3's that should
6143 be replaced with this routine. but since thats not done yet
6144 we ifdef it out - dmq
6147 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6151 PERL_ARGS_ASSERT_REGHOP4;
6154 while (off-- && s < rlim) {
6155 /* XXX could check well-formedness here */
6160 while (off++ && s > llim) {
6162 if (UTF8_IS_CONTINUED(*s)) {
6163 while (s > llim && UTF8_IS_CONTINUATION(*s))
6166 /* XXX could check well-formedness here */
6174 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6178 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6181 while (off-- && s < lim) {
6182 /* XXX could check well-formedness here */
6189 while (off++ && s > lim) {
6191 if (UTF8_IS_CONTINUED(*s)) {
6192 while (s > lim && UTF8_IS_CONTINUATION(*s))
6195 /* XXX could check well-formedness here */
6204 restore_pos(pTHX_ void *arg)
6207 regexp * const rex = (regexp *)arg;
6208 if (PL_reg_eval_set) {
6209 if (PL_reg_oldsaved) {
6210 rex->subbeg = PL_reg_oldsaved;
6211 rex->sublen = PL_reg_oldsavedlen;
6212 #ifdef PERL_OLD_COPY_ON_WRITE
6213 rex->saved_copy = PL_nrs;
6215 RXp_MATCH_COPIED_on(rex);
6217 PL_reg_magic->mg_len = PL_reg_oldpos;
6218 PL_reg_eval_set = 0;
6219 PL_curpm = PL_reg_oldcurpm;
6224 S_to_utf8_substr(pTHX_ register regexp *prog)
6228 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6231 if (prog->substrs->data[i].substr
6232 && !prog->substrs->data[i].utf8_substr) {
6233 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6234 prog->substrs->data[i].utf8_substr = sv;
6235 sv_utf8_upgrade(sv);
6236 if (SvVALID(prog->substrs->data[i].substr)) {
6237 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6238 if (flags & FBMcf_TAIL) {
6239 /* Trim the trailing \n that fbm_compile added last
6241 SvCUR_set(sv, SvCUR(sv) - 1);
6242 /* Whilst this makes the SV technically "invalid" (as its
6243 buffer is no longer followed by "\0") when fbm_compile()
6244 adds the "\n" back, a "\0" is restored. */
6246 fbm_compile(sv, flags);
6248 if (prog->substrs->data[i].substr == prog->check_substr)
6249 prog->check_utf8 = sv;
6255 S_to_byte_substr(pTHX_ register regexp *prog)
6260 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6263 if (prog->substrs->data[i].utf8_substr
6264 && !prog->substrs->data[i].substr) {
6265 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6266 if (sv_utf8_downgrade(sv, TRUE)) {
6267 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6269 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6270 if (flags & FBMcf_TAIL) {
6271 /* Trim the trailing \n that fbm_compile added last
6273 SvCUR_set(sv, SvCUR(sv) - 1);
6275 fbm_compile(sv, flags);
6281 prog->substrs->data[i].substr = sv;
6282 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6283 prog->check_substr = sv;
6290 * c-indentation-style: bsd
6292 * indent-tabs-mode: t
6295 * ex: set ts=8 sts=4 sw=4 noet: