5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
56 **** Alterations to Henry's code are...
58 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
59 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
61 **** You may distribute under the terms of either the GNU General Public
62 **** License or the Artistic License, as specified in the README file.
64 * Beware that some of this code is subtly aware of the way operator
65 * precedence is structured in regular expressions. Serious changes in
66 * regular-expression syntax might require a total rethink.
69 #define PERL_IN_REGEXEC_C
72 #ifdef PERL_IN_XSUB_RE
78 #define RF_tainted 1 /* tainted information used? */
79 #define RF_warned 2 /* warned about big count? */
81 #define RF_utf8 8 /* Pattern contains multibyte chars? */
83 #define UTF ((PL_reg_flags & RF_utf8) != 0)
85 #define RS_init 1 /* eval environment created */
86 #define RS_set 2 /* replsv value is set */
92 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
98 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
99 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
101 #define HOPc(pos,off) \
102 (char *)(PL_reg_match_utf8 \
103 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
105 #define HOPBACKc(pos, off) \
106 (char*)(PL_reg_match_utf8\
107 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
108 : (pos - off >= PL_bostr) \
112 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
113 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
115 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
116 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
117 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
118 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
119 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
120 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
122 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
124 /* for use after a quantifier and before an EXACT-like node -- japhy */
125 #define JUMPABLE(rn) ( \
126 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
127 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
128 OP(rn) == PLUS || OP(rn) == MINMOD || \
129 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
132 #define HAS_TEXT(rn) ( \
133 PL_regkind[OP(rn)] == EXACT || PL_regkind[OP(rn)] == REF \
137 Search for mandatory following text node; for lookahead, the text must
138 follow but for lookbehind (rn->flags != 0) we skip to the next step.
140 #define FIND_NEXT_IMPT(rn) STMT_START { \
141 while (JUMPABLE(rn)) { \
142 const OPCODE type = OP(rn); \
143 if (type == SUSPEND || PL_regkind[type] == CURLY) \
144 rn = NEXTOPER(NEXTOPER(rn)); \
145 else if (type == PLUS) \
147 else if (type == IFMATCH) \
148 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
149 else rn += NEXT_OFF(rn); \
153 static void restore_pos(pTHX_ void *arg);
156 S_regcppush(pTHX_ I32 parenfloor)
159 const int retval = PL_savestack_ix;
160 #define REGCP_PAREN_ELEMS 4
161 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
163 GET_RE_DEBUG_FLAGS_DECL;
165 if (paren_elems_to_push < 0)
166 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
168 #define REGCP_OTHER_ELEMS 6
169 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
170 for (p = PL_regsize; p > parenfloor; p--) {
171 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
172 SSPUSHINT(PL_regendp[p]);
173 SSPUSHINT(PL_regstartp[p]);
174 SSPUSHPTR(PL_reg_start_tmp[p]);
176 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
177 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
178 (UV)p, (IV)PL_regstartp[p],
179 (IV)(PL_reg_start_tmp[p] - PL_bostr),
183 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
184 SSPUSHINT(PL_regsize);
185 SSPUSHINT(*PL_reglastparen);
186 SSPUSHINT(*PL_reglastcloseparen);
187 SSPUSHPTR(PL_reginput);
188 #define REGCP_FRAME_ELEMS 2
189 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
190 * are needed for the regexp context stack bookkeeping. */
191 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
192 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
197 /* These are needed since we do not localize EVAL nodes: */
198 #define REGCP_SET(cp) \
200 PerlIO_printf(Perl_debug_log, \
201 " Setting an EVAL scope, savestack=%"IVdf"\n", \
202 (IV)PL_savestack_ix)); \
205 #define REGCP_UNWIND(cp) \
207 if (cp != PL_savestack_ix) \
208 PerlIO_printf(Perl_debug_log, \
209 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
210 (IV)(cp), (IV)PL_savestack_ix)); \
214 S_regcppop(pTHX_ const regexp *rex)
220 GET_RE_DEBUG_FLAGS_DECL;
222 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
224 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
225 i = SSPOPINT; /* Parentheses elements to pop. */
226 input = (char *) SSPOPPTR;
227 *PL_reglastcloseparen = SSPOPINT;
228 *PL_reglastparen = SSPOPINT;
229 PL_regsize = SSPOPINT;
231 /* Now restore the parentheses context. */
232 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
233 i > 0; i -= REGCP_PAREN_ELEMS) {
235 U32 paren = (U32)SSPOPINT;
236 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
237 PL_regstartp[paren] = SSPOPINT;
239 if (paren <= *PL_reglastparen)
240 PL_regendp[paren] = tmps;
242 PerlIO_printf(Perl_debug_log,
243 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
244 (UV)paren, (IV)PL_regstartp[paren],
245 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
246 (IV)PL_regendp[paren],
247 (paren > *PL_reglastparen ? "(no)" : ""));
251 if (*PL_reglastparen + 1 <= rex->nparens) {
252 PerlIO_printf(Perl_debug_log,
253 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
254 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
258 /* It would seem that the similar code in regtry()
259 * already takes care of this, and in fact it is in
260 * a better location to since this code can #if 0-ed out
261 * but the code in regtry() is needed or otherwise tests
262 * requiring null fields (pat.t#187 and split.t#{13,14}
263 * (as of patchlevel 7877) will fail. Then again,
264 * this code seems to be necessary or otherwise
265 * building DynaLoader will fail:
266 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
268 for (i = *PL_reglastparen + 1; (U32)i <= rex->nparens; i++) {
270 PL_regstartp[i] = -1;
277 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
280 * pregexec and friends
283 #ifndef PERL_IN_XSUB_RE
285 - pregexec - match a regexp against a string
288 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
289 char *strbeg, I32 minend, SV *screamer, U32 nosave)
290 /* strend: pointer to null at end of string */
291 /* strbeg: real beginning of string */
292 /* minend: end of match must be >=minend after stringarg. */
293 /* nosave: For optimizations. */
296 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
297 nosave ? 0 : REXEC_COPY_STR);
302 * Need to implement the following flags for reg_anch:
304 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
306 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
307 * INTUIT_AUTORITATIVE_ML
308 * INTUIT_ONCE_NOML - Intuit can match in one location only.
311 * Another flag for this function: SECOND_TIME (so that float substrs
312 * with giant delta may be not rechecked).
315 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
317 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
318 Otherwise, only SvCUR(sv) is used to get strbeg. */
320 /* XXXX We assume that strpos is strbeg unless sv. */
322 /* XXXX Some places assume that there is a fixed substring.
323 An update may be needed if optimizer marks as "INTUITable"
324 RExen without fixed substrings. Similarly, it is assumed that
325 lengths of all the strings are no more than minlen, thus they
326 cannot come from lookahead.
327 (Or minlen should take into account lookahead.) */
329 /* A failure to find a constant substring means that there is no need to make
330 an expensive call to REx engine, thus we celebrate a failure. Similarly,
331 finding a substring too deep into the string means that less calls to
332 regtry() should be needed.
334 REx compiler's optimizer found 4 possible hints:
335 a) Anchored substring;
337 c) Whether we are anchored (beginning-of-line or \G);
338 d) First node (of those at offset 0) which may distingush positions;
339 We use a)b)d) and multiline-part of c), and try to find a position in the
340 string which does not contradict any of them.
343 /* Most of decisions we do here should have been done at compile time.
344 The nodes of the REx which we used for the search should have been
345 deleted from the finite automaton. */
348 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
349 char *strend, U32 flags, re_scream_pos_data *data)
352 register I32 start_shift = 0;
353 /* Should be nonnegative! */
354 register I32 end_shift = 0;
359 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
361 register char *other_last = NULL; /* other substr checked before this */
362 char *check_at = NULL; /* check substr found at this pos */
363 const I32 multiline = prog->reganch & PMf_MULTILINE;
365 const char * const i_strpos = strpos;
368 GET_RE_DEBUG_FLAGS_DECL;
370 RX_MATCH_UTF8_set(prog,do_utf8);
372 if (prog->reganch & ROPT_UTF8) {
373 PL_reg_flags |= RF_utf8;
376 debug_start_match(prog, do_utf8, strpos, strend,
377 sv ? "Guessing start of match in sv for"
378 : "Guessing start of match in string for");
381 /* CHR_DIST() would be more correct here but it makes things slow. */
382 if (prog->minlen > strend - strpos) {
383 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
384 "String too short... [re_intuit_start]\n"));
388 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
391 if (!prog->check_utf8 && prog->check_substr)
392 to_utf8_substr(prog);
393 check = prog->check_utf8;
395 if (!prog->check_substr && prog->check_utf8)
396 to_byte_substr(prog);
397 check = prog->check_substr;
399 if (check == &PL_sv_undef) {
400 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
401 "Non-utf8 string cannot match utf8 check string\n"));
404 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
405 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
406 || ( (prog->reganch & ROPT_ANCH_BOL)
407 && !multiline ) ); /* Check after \n? */
410 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
411 | ROPT_IMPLICIT)) /* not a real BOL */
412 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
414 && (strpos != strbeg)) {
415 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
418 if (prog->check_offset_min == prog->check_offset_max &&
419 !(prog->reganch & ROPT_CANY_SEEN)) {
420 /* Substring at constant offset from beg-of-str... */
423 s = HOP3c(strpos, prog->check_offset_min, strend);
426 slen = SvCUR(check); /* >= 1 */
428 if ( strend - s > slen || strend - s < slen - 1
429 || (strend - s == slen && strend[-1] != '\n')) {
430 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
433 /* Now should match s[0..slen-2] */
435 if (slen && (*SvPVX_const(check) != *s
437 && memNE(SvPVX_const(check), s, slen)))) {
439 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
443 else if (*SvPVX_const(check) != *s
444 || ((slen = SvCUR(check)) > 1
445 && memNE(SvPVX_const(check), s, slen)))
448 goto success_at_start;
451 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
453 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
454 end_shift = prog->check_end_shift;
457 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
458 - (SvTAIL(check) != 0);
459 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
461 if (end_shift < eshift)
465 else { /* Can match at random position */
468 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
469 end_shift = prog->check_end_shift;
471 /* end shift should be non negative here */
474 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
476 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
477 (IV)end_shift, prog->precomp);
481 /* Find a possible match in the region s..strend by looking for
482 the "check" substring in the region corrected by start/end_shift. */
485 I32 srch_start_shift = start_shift;
486 I32 srch_end_shift = end_shift;
487 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
488 srch_end_shift -= ((strbeg - s) - srch_start_shift);
489 srch_start_shift = strbeg - s;
492 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
493 (IV)prog->check_offset_min,
494 (IV)srch_start_shift,
496 (IV)prog->check_end_shift);
499 if (flags & REXEC_SCREAM) {
500 I32 p = -1; /* Internal iterator of scream. */
501 I32 * const pp = data ? data->scream_pos : &p;
503 if (PL_screamfirst[BmRARE(check)] >= 0
504 || ( BmRARE(check) == '\n'
505 && (BmPREVIOUS(check) == SvCUR(check) - 1)
507 s = screaminstr(sv, check,
508 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
511 /* we may be pointing at the wrong string */
512 if (s && RX_MATCH_COPIED(prog))
513 s = strbeg + (s - SvPVX_const(sv));
515 *data->scream_olds = s;
520 if (prog->reganch & ROPT_CANY_SEEN) {
521 start_point= (U8*)(s + srch_start_shift);
522 end_point= (U8*)(strend - srch_end_shift);
524 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
525 end_point= HOP3(strend, -srch_end_shift, strbeg);
528 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
529 (int)(end_point - start_point),
530 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
534 s = fbm_instr( start_point, end_point,
535 check, multiline ? FBMrf_MULTILINE : 0);
538 /* Update the count-of-usability, remove useless subpatterns,
542 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
543 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
544 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
545 (s ? "Found" : "Did not find"),
546 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
547 ? "anchored" : "floating"),
550 (s ? " at offset " : "...\n") );
555 /* Finish the diagnostic message */
556 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
558 /* XXX dmq: first branch is for positive lookbehind...
559 Our check string is offset from the beginning of the pattern.
560 So we need to do any stclass tests offset forward from that
569 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
570 Start with the other substr.
571 XXXX no SCREAM optimization yet - and a very coarse implementation
572 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
573 *always* match. Probably should be marked during compile...
574 Probably it is right to do no SCREAM here...
577 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
578 : (prog->float_substr && prog->anchored_substr))
580 /* Take into account the "other" substring. */
581 /* XXXX May be hopelessly wrong for UTF... */
584 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
587 char * const last = HOP3c(s, -start_shift, strbeg);
589 char * const saved_s = s;
592 t = s - prog->check_offset_max;
593 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
595 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
600 t = HOP3c(t, prog->anchored_offset, strend);
601 if (t < other_last) /* These positions already checked */
603 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
606 /* XXXX It is not documented what units *_offsets are in.
607 We assume bytes, but this is clearly wrong.
608 Meaning this code needs to be carefully reviewed for errors.
612 /* On end-of-str: see comment below. */
613 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
614 if (must == &PL_sv_undef) {
616 DEBUG_r(must = prog->anchored_utf8); /* for debug */
621 HOP3(HOP3(last1, prog->anchored_offset, strend)
622 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
624 multiline ? FBMrf_MULTILINE : 0
627 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
628 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
629 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
630 (s ? "Found" : "Contradicts"),
631 quoted, RE_SV_TAIL(must));
636 if (last1 >= last2) {
637 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
638 ", giving up...\n"));
641 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
642 ", trying floating at offset %ld...\n",
643 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
644 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
645 s = HOP3c(last, 1, strend);
649 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
650 (long)(s - i_strpos)));
651 t = HOP3c(s, -prog->anchored_offset, strbeg);
652 other_last = HOP3c(s, 1, strend);
660 else { /* Take into account the floating substring. */
662 char * const saved_s = s;
665 t = HOP3c(s, -start_shift, strbeg);
667 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
668 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
669 last = HOP3c(t, prog->float_max_offset, strend);
670 s = HOP3c(t, prog->float_min_offset, strend);
673 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
674 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
675 /* fbm_instr() takes into account exact value of end-of-str
676 if the check is SvTAIL(ed). Since false positives are OK,
677 and end-of-str is not later than strend we are OK. */
678 if (must == &PL_sv_undef) {
680 DEBUG_r(must = prog->float_utf8); /* for debug message */
683 s = fbm_instr((unsigned char*)s,
684 (unsigned char*)last + SvCUR(must)
686 must, multiline ? FBMrf_MULTILINE : 0);
688 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
689 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
690 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
691 (s ? "Found" : "Contradicts"),
692 quoted, RE_SV_TAIL(must));
696 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
697 ", giving up...\n"));
700 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
701 ", trying anchored starting at offset %ld...\n",
702 (long)(saved_s + 1 - i_strpos)));
704 s = HOP3c(t, 1, strend);
708 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
709 (long)(s - i_strpos)));
710 other_last = s; /* Fix this later. --Hugo */
720 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
723 PerlIO_printf(Perl_debug_log,
724 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
725 (IV)prog->check_offset_min,
726 (IV)prog->check_offset_max,
734 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
736 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
739 /* Fixed substring is found far enough so that the match
740 cannot start at strpos. */
742 if (ml_anch && t[-1] != '\n') {
743 /* Eventually fbm_*() should handle this, but often
744 anchored_offset is not 0, so this check will not be wasted. */
745 /* XXXX In the code below we prefer to look for "^" even in
746 presence of anchored substrings. And we search even
747 beyond the found float position. These pessimizations
748 are historical artefacts only. */
750 while (t < strend - prog->minlen) {
752 if (t < check_at - prog->check_offset_min) {
753 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
754 /* Since we moved from the found position,
755 we definitely contradict the found anchored
756 substr. Due to the above check we do not
757 contradict "check" substr.
758 Thus we can arrive here only if check substr
759 is float. Redo checking for "other"=="fixed".
762 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
763 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
764 goto do_other_anchored;
766 /* We don't contradict the found floating substring. */
767 /* XXXX Why not check for STCLASS? */
769 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
770 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
773 /* Position contradicts check-string */
774 /* XXXX probably better to look for check-string
775 than for "\n", so one should lower the limit for t? */
776 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
777 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
778 other_last = strpos = s = t + 1;
783 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
784 PL_colors[0], PL_colors[1]));
788 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
789 PL_colors[0], PL_colors[1]));
793 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
796 /* The found string does not prohibit matching at strpos,
797 - no optimization of calling REx engine can be performed,
798 unless it was an MBOL and we are not after MBOL,
799 or a future STCLASS check will fail this. */
801 /* Even in this situation we may use MBOL flag if strpos is offset
802 wrt the start of the string. */
803 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
804 && (strpos != strbeg) && strpos[-1] != '\n'
805 /* May be due to an implicit anchor of m{.*foo} */
806 && !(prog->reganch & ROPT_IMPLICIT))
811 DEBUG_EXECUTE_r( if (ml_anch)
812 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
813 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
816 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
818 prog->check_utf8 /* Could be deleted already */
819 && --BmUSEFUL(prog->check_utf8) < 0
820 && (prog->check_utf8 == prog->float_utf8)
822 prog->check_substr /* Could be deleted already */
823 && --BmUSEFUL(prog->check_substr) < 0
824 && (prog->check_substr == prog->float_substr)
827 /* If flags & SOMETHING - do not do it many times on the same match */
828 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
829 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
830 if (do_utf8 ? prog->check_substr : prog->check_utf8)
831 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
832 prog->check_substr = prog->check_utf8 = NULL; /* disable */
833 prog->float_substr = prog->float_utf8 = NULL; /* clear */
834 check = NULL; /* abort */
836 /* XXXX This is a remnant of the old implementation. It
837 looks wasteful, since now INTUIT can use many
839 prog->reganch &= ~RE_USE_INTUIT;
846 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
847 /* trie stclasses are too expensive to use here, we are better off to
848 leave it to regmatch itself */
849 if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) {
850 /* minlen == 0 is possible if regstclass is \b or \B,
851 and the fixed substr is ''$.
852 Since minlen is already taken into account, s+1 is before strend;
853 accidentally, minlen >= 1 guaranties no false positives at s + 1
854 even for \b or \B. But (minlen? 1 : 0) below assumes that
855 regstclass does not come from lookahead... */
856 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
857 This leaves EXACTF only, which is dealt with in find_byclass(). */
858 const U8* const str = (U8*)STRING(prog->regstclass);
859 const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT
860 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
863 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
864 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
865 else if (prog->float_substr || prog->float_utf8)
866 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
870 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n",
871 (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg));
874 s = find_byclass(prog, prog->regstclass, s, endpos, NULL);
877 const char *what = NULL;
879 if (endpos == strend) {
880 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
881 "Could not match STCLASS...\n") );
884 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
885 "This position contradicts STCLASS...\n") );
886 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
888 /* Contradict one of substrings */
889 if (prog->anchored_substr || prog->anchored_utf8) {
890 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
891 DEBUG_EXECUTE_r( what = "anchored" );
893 s = HOP3c(t, 1, strend);
894 if (s + start_shift + end_shift > strend) {
895 /* XXXX Should be taken into account earlier? */
896 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
897 "Could not match STCLASS...\n") );
902 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
903 "Looking for %s substr starting at offset %ld...\n",
904 what, (long)(s + start_shift - i_strpos)) );
907 /* Have both, check_string is floating */
908 if (t + start_shift >= check_at) /* Contradicts floating=check */
909 goto retry_floating_check;
910 /* Recheck anchored substring, but not floating... */
914 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
915 "Looking for anchored substr starting at offset %ld...\n",
916 (long)(other_last - i_strpos)) );
917 goto do_other_anchored;
919 /* Another way we could have checked stclass at the
920 current position only: */
925 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
926 "Looking for /%s^%s/m starting at offset %ld...\n",
927 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
930 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
932 /* Check is floating subtring. */
933 retry_floating_check:
934 t = check_at - start_shift;
935 DEBUG_EXECUTE_r( what = "floating" );
936 goto hop_and_restart;
939 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
940 "By STCLASS: moving %ld --> %ld\n",
941 (long)(t - i_strpos), (long)(s - i_strpos))
945 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
946 "Does not contradict STCLASS...\n");
951 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
952 PL_colors[4], (check ? "Guessed" : "Giving up"),
953 PL_colors[5], (long)(s - i_strpos)) );
956 fail_finish: /* Substring not found */
957 if (prog->check_substr || prog->check_utf8) /* could be removed already */
958 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
960 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
961 PL_colors[4], PL_colors[5]));
967 #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \
968 foldlen, foldbuf, uniflags) STMT_START { \
969 switch (trie_type) { \
970 case trie_utf8_fold: \
972 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
977 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
978 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
979 foldlen -= UNISKIP( uvc ); \
980 uscan = foldbuf + UNISKIP( uvc ); \
984 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
992 charid = trie->charmap[ uvc ]; \
996 if (trie->widecharmap) { \
997 SV** const svpp = hv_fetch(trie->widecharmap, \
998 (char*)&uvc, sizeof(UV), 0); \
1000 charid = (U16)SvIV(*svpp); \
1005 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1008 ibcmp_utf8(s, NULL, 0, do_utf8, \
1009 m, NULL, ln, (bool)UTF)) \
1010 && (!reginfo || regtry(reginfo, s)) ) \
1013 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1014 uvchr_to_utf8(tmpbuf, c); \
1015 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1017 && (f == c1 || f == c2) \
1018 && (ln == foldlen || \
1019 !ibcmp_utf8((char *) foldbuf, \
1020 NULL, foldlen, do_utf8, \
1022 NULL, ln, (bool)UTF)) \
1023 && (!reginfo || regtry(reginfo, s)) ) \
1028 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1032 && (ln == 1 || !(OP(c) == EXACTF \
1034 : ibcmp_locale(s, m, ln))) \
1035 && (!reginfo || regtry(reginfo, s)) ) \
1041 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1043 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1049 #define REXEC_FBC_SCAN(CoDe) \
1051 while (s < strend) { \
1057 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1058 REXEC_FBC_UTF8_SCAN( \
1060 if (tmp && (!reginfo || regtry(reginfo, s))) \
1069 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1072 if (tmp && (!reginfo || regtry(reginfo, s))) \
1081 #define REXEC_FBC_TRYIT \
1082 if ((!reginfo || regtry(reginfo, s))) \
1085 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1088 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1091 REXEC_FBC_CLASS_SCAN(CoNd); \
1095 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1096 PL_reg_flags |= RF_tainted; \
1098 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1101 REXEC_FBC_CLASS_SCAN(CoNd); \
1105 #define DUMP_EXEC_POS(li,s,doutf8) \
1106 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1108 /* We know what class REx starts with. Try to find this position... */
1109 /* if reginfo is NULL, its a dryrun */
1110 /* annoyingly all the vars in this routine have different names from their counterparts
1111 in regmatch. /grrr */
1114 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1115 const char *strend, const regmatch_info *reginfo)
1118 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
1122 register STRLEN uskip;
1126 register I32 tmp = 1; /* Scratch variable? */
1127 register const bool do_utf8 = PL_reg_match_utf8;
1129 /* We know what class it must start with. */
1133 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1134 !UTF8_IS_INVARIANT((U8)s[0]) ?
1135 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1136 REGINCLASS(prog, c, (U8*)s));
1139 while (s < strend) {
1142 if (REGINCLASS(prog, c, (U8*)s) ||
1143 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1144 /* The assignment of 2 is intentional:
1145 * for the folded sharp s, the skip is 2. */
1146 (skip = SHARP_S_SKIP))) {
1147 if (tmp && (!reginfo || regtry(reginfo, s)))
1160 if (tmp && (!reginfo || regtry(reginfo, s)))
1168 ln = STR_LEN(c); /* length to match in octets/bytes */
1169 lnc = (I32) ln; /* length to match in characters */
1171 STRLEN ulen1, ulen2;
1173 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1174 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1175 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1177 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1178 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1180 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1182 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1185 while (sm < ((U8 *) m + ln)) {
1200 c2 = PL_fold_locale[c1];
1202 e = HOP3c(strend, -((I32)lnc), s);
1204 if (!reginfo && e < s)
1205 e = s; /* Due to minlen logic of intuit() */
1207 /* The idea in the EXACTF* cases is to first find the
1208 * first character of the EXACTF* node and then, if
1209 * necessary, case-insensitively compare the full
1210 * text of the node. The c1 and c2 are the first
1211 * characters (though in Unicode it gets a bit
1212 * more complicated because there are more cases
1213 * than just upper and lower: one needs to use
1214 * the so-called folding case for case-insensitive
1215 * matching (called "loose matching" in Unicode).
1216 * ibcmp_utf8() will do just that. */
1220 U8 tmpbuf [UTF8_MAXBYTES+1];
1221 STRLEN len, foldlen;
1222 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1224 /* Upper and lower of 1st char are equal -
1225 * probably not a "letter". */
1227 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1229 REXEC_FBC_EXACTISH_CHECK(c == c1);
1234 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1237 /* Handle some of the three Greek sigmas cases.
1238 * Note that not all the possible combinations
1239 * are handled here: some of them are handled
1240 * by the standard folding rules, and some of
1241 * them (the character class or ANYOF cases)
1242 * are handled during compiletime in
1243 * regexec.c:S_regclass(). */
1244 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1245 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1246 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1248 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1254 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1256 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1260 PL_reg_flags |= RF_tainted;
1267 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1268 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1270 tmp = ((OP(c) == BOUND ?
1271 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1272 LOAD_UTF8_CHARCLASS_ALNUM();
1273 REXEC_FBC_UTF8_SCAN(
1274 if (tmp == !(OP(c) == BOUND ?
1275 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1276 isALNUM_LC_utf8((U8*)s)))
1284 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1285 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1288 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1294 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, s)))
1298 PL_reg_flags |= RF_tainted;
1305 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1306 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1308 tmp = ((OP(c) == NBOUND ?
1309 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1310 LOAD_UTF8_CHARCLASS_ALNUM();
1311 REXEC_FBC_UTF8_SCAN(
1312 if (tmp == !(OP(c) == NBOUND ?
1313 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1314 isALNUM_LC_utf8((U8*)s)))
1316 else REXEC_FBC_TRYIT;
1320 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1321 tmp = ((OP(c) == NBOUND ?
1322 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1325 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1327 else REXEC_FBC_TRYIT;
1330 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, s)))
1334 REXEC_FBC_CSCAN_PRELOAD(
1335 LOAD_UTF8_CHARCLASS_ALNUM(),
1336 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1340 REXEC_FBC_CSCAN_TAINT(
1341 isALNUM_LC_utf8((U8*)s),
1345 REXEC_FBC_CSCAN_PRELOAD(
1346 LOAD_UTF8_CHARCLASS_ALNUM(),
1347 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1351 REXEC_FBC_CSCAN_TAINT(
1352 !isALNUM_LC_utf8((U8*)s),
1356 REXEC_FBC_CSCAN_PRELOAD(
1357 LOAD_UTF8_CHARCLASS_SPACE(),
1358 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1362 REXEC_FBC_CSCAN_TAINT(
1363 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1367 REXEC_FBC_CSCAN_PRELOAD(
1368 LOAD_UTF8_CHARCLASS_SPACE(),
1369 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1373 REXEC_FBC_CSCAN_TAINT(
1374 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1378 REXEC_FBC_CSCAN_PRELOAD(
1379 LOAD_UTF8_CHARCLASS_DIGIT(),
1380 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1384 REXEC_FBC_CSCAN_TAINT(
1385 isDIGIT_LC_utf8((U8*)s),
1389 REXEC_FBC_CSCAN_PRELOAD(
1390 LOAD_UTF8_CHARCLASS_DIGIT(),
1391 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1395 REXEC_FBC_CSCAN_TAINT(
1396 !isDIGIT_LC_utf8((U8*)s),
1402 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1403 trie_type = do_utf8 ?
1404 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1406 /* what trie are we using right now */
1408 = (reg_ac_data*)prog->data->data[ ARG( c ) ];
1409 reg_trie_data *trie=aho->trie;
1411 const char *last_start = strend - trie->minlen;
1413 const char *real_start = s;
1415 STRLEN maxlen = trie->maxlen;
1417 U8 **points; /* map of where we were in the input string
1418 when reading a given char. For ASCII this
1419 is unnecessary overhead as the relationship
1420 is always 1:1, but for unicode, especially
1421 case folded unicode this is not true. */
1422 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1426 GET_RE_DEBUG_FLAGS_DECL;
1428 /* We can't just allocate points here. We need to wrap it in
1429 * an SV so it gets freed properly if there is a croak while
1430 * running the match */
1433 sv_points=newSV(maxlen * sizeof(U8 *));
1434 SvCUR_set(sv_points,
1435 maxlen * sizeof(U8 *));
1436 SvPOK_on(sv_points);
1437 sv_2mortal(sv_points);
1438 points=(U8**)SvPV_nolen(sv_points );
1439 if ( trie_type != trie_utf8_fold
1440 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1443 bitmap=(U8*)trie->bitmap;
1445 bitmap=(U8*)ANYOF_BITMAP(c);
1447 /* this is the Aho-Corasick algorithm modified a touch
1448 to include special handling for long "unknown char"
1449 sequences. The basic idea being that we use AC as long
1450 as we are dealing with a possible matching char, when
1451 we encounter an unknown char (and we have not encountered
1452 an accepting state) we scan forward until we find a legal
1454 AC matching is basically that of trie matching, except
1455 that when we encounter a failing transition, we fall back
1456 to the current states "fail state", and try the current char
1457 again, a process we repeat until we reach the root state,
1458 state 1, or a legal transition. If we fail on the root state
1459 then we can either terminate if we have reached an accepting
1460 state previously, or restart the entire process from the beginning
1464 while (s <= last_start) {
1465 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1473 U8 *uscan = (U8*)NULL;
1474 U8 *leftmost = NULL;
1476 U32 accepted_word= 0;
1480 while ( state && uc <= (U8*)strend ) {
1482 U32 word = aho->states[ state ].wordnum;
1486 DEBUG_TRIE_EXECUTE_r(
1487 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1488 dump_exec_pos( (char *)uc, c, strend, real_start,
1489 (char *)uc, do_utf8 );
1490 PerlIO_printf( Perl_debug_log,
1491 " Scanning for legal start char...\n");
1494 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1499 if (uc >(U8*)last_start) break;
1503 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1504 if (!leftmost || lpos < leftmost) {
1505 DEBUG_r(accepted_word=word);
1511 points[pointpos++ % maxlen]= uc;
1512 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
1513 uvc, charid, foldlen, foldbuf, uniflags);
1514 DEBUG_TRIE_EXECUTE_r({
1515 dump_exec_pos( (char *)uc, c, strend, real_start,
1517 PerlIO_printf(Perl_debug_log,
1518 " Charid:%3u CP:%4"UVxf" ",
1524 word = aho->states[ state ].wordnum;
1526 base = aho->states[ state ].trans.base;
1528 DEBUG_TRIE_EXECUTE_r({
1530 dump_exec_pos( (char *)uc, c, strend, real_start,
1532 PerlIO_printf( Perl_debug_log,
1533 "%sState: %4"UVxf", word=%"UVxf,
1534 failed ? " Fail transition to " : "",
1535 (UV)state, (UV)word);
1540 (base + charid > trie->uniquecharcount )
1541 && (base + charid - 1 - trie->uniquecharcount
1543 && trie->trans[base + charid - 1 -
1544 trie->uniquecharcount].check == state
1545 && (tmp=trie->trans[base + charid - 1 -
1546 trie->uniquecharcount ].next))
1548 DEBUG_TRIE_EXECUTE_r(
1549 PerlIO_printf( Perl_debug_log," - legal\n"));
1554 DEBUG_TRIE_EXECUTE_r(
1555 PerlIO_printf( Perl_debug_log," - fail\n"));
1557 state = aho->fail[state];
1561 /* we must be accepting here */
1562 DEBUG_TRIE_EXECUTE_r(
1563 PerlIO_printf( Perl_debug_log," - accepting\n"));
1572 if (!state) state = 1;
1575 if ( aho->states[ state ].wordnum ) {
1576 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1577 if (!leftmost || lpos < leftmost) {
1578 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1583 s = (char*)leftmost;
1584 DEBUG_TRIE_EXECUTE_r({
1586 Perl_debug_log,"Matches word #%"UVxf" at position %d. Trying full pattern...\n",
1587 (UV)accepted_word, s - real_start
1590 if (!reginfo || regtry(reginfo, s)) {
1596 DEBUG_TRIE_EXECUTE_r({
1597 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1600 DEBUG_TRIE_EXECUTE_r(
1601 PerlIO_printf( Perl_debug_log,"No match.\n"));
1610 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1619 - regexec_flags - match a regexp against a string
1622 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1623 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1624 /* strend: pointer to null at end of string */
1625 /* strbeg: real beginning of string */
1626 /* minend: end of match must be >=minend after stringarg. */
1627 /* data: May be used for some additional optimizations. */
1628 /* nosave: For optimizations. */
1632 register regnode *c;
1633 register char *startpos = stringarg;
1634 I32 minlen; /* must match at least this many chars */
1635 I32 dontbother = 0; /* how many characters not to try at end */
1636 I32 end_shift = 0; /* Same for the end. */ /* CC */
1637 I32 scream_pos = -1; /* Internal iterator of scream. */
1638 char *scream_olds = NULL;
1639 SV* const oreplsv = GvSV(PL_replgv);
1640 const bool do_utf8 = (bool)DO_UTF8(sv);
1643 regmatch_info reginfo; /* create some info to pass to regtry etc */
1645 GET_RE_DEBUG_FLAGS_DECL;
1647 PERL_UNUSED_ARG(data);
1649 /* Be paranoid... */
1650 if (prog == NULL || startpos == NULL) {
1651 Perl_croak(aTHX_ "NULL regexp parameter");
1655 multiline = prog->reganch & PMf_MULTILINE;
1656 reginfo.prog = prog;
1658 RX_MATCH_UTF8_set(prog, do_utf8);
1660 debug_start_match(prog, do_utf8, startpos, strend,
1664 minlen = prog->minlen;
1666 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1667 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1668 "String too short [regexec_flags]...\n"));
1673 /* Check validity of program. */
1674 if (UCHARAT(prog->program) != REG_MAGIC) {
1675 Perl_croak(aTHX_ "corrupted regexp program");
1679 PL_reg_eval_set = 0;
1682 if (prog->reganch & ROPT_UTF8)
1683 PL_reg_flags |= RF_utf8;
1685 /* Mark beginning of line for ^ and lookbehind. */
1686 reginfo.bol = startpos; /* XXX not used ??? */
1690 /* Mark end of line for $ (and such) */
1693 /* see how far we have to get to not match where we matched before */
1694 reginfo.till = startpos+minend;
1696 /* If there is a "must appear" string, look for it. */
1699 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to set reginfo->ganch */
1702 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1703 reginfo.ganch = startpos;
1704 else if (sv && SvTYPE(sv) >= SVt_PVMG
1706 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1707 && mg->mg_len >= 0) {
1708 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1709 if (prog->reganch & ROPT_ANCH_GPOS) {
1710 if (s > reginfo.ganch)
1715 else /* pos() not defined */
1716 reginfo.ganch = strbeg;
1719 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1720 re_scream_pos_data d;
1722 d.scream_olds = &scream_olds;
1723 d.scream_pos = &scream_pos;
1724 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1726 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1727 goto phooey; /* not present */
1733 /* Simplest case: anchored match need be tried only once. */
1734 /* [unless only anchor is BOL and multiline is set] */
1735 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1736 if (s == startpos && regtry(®info, startpos))
1738 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1739 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1744 dontbother = minlen - 1;
1745 end = HOP3c(strend, -dontbother, strbeg) - 1;
1746 /* for multiline we only have to try after newlines */
1747 if (prog->check_substr || prog->check_utf8) {
1751 if (regtry(®info, s))
1756 if (prog->reganch & RE_USE_INTUIT) {
1757 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1768 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1769 if (regtry(®info, s))
1776 } else if (ROPT_GPOS_CHECK == (prog->reganch & ROPT_GPOS_CHECK))
1778 /* the warning about reginfo.ganch being used without intialization
1779 is bogus -- we set it above, when prog->reganch & ROPT_GPOS_SEEN
1780 and we only enter this block when the same bit is set. */
1781 if (regtry(®info, reginfo.ganch))
1786 /* Messy cases: unanchored match. */
1787 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1788 /* we have /x+whatever/ */
1789 /* it must be a one character string (XXXX Except UTF?) */
1794 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1795 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1796 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1801 DEBUG_EXECUTE_r( did_match = 1 );
1802 if (regtry(®info, s)) goto got_it;
1804 while (s < strend && *s == ch)
1812 DEBUG_EXECUTE_r( did_match = 1 );
1813 if (regtry(®info, s)) goto got_it;
1815 while (s < strend && *s == ch)
1820 DEBUG_EXECUTE_r(if (!did_match)
1821 PerlIO_printf(Perl_debug_log,
1822 "Did not find anchored character...\n")
1825 else if (prog->anchored_substr != NULL
1826 || prog->anchored_utf8 != NULL
1827 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1828 && prog->float_max_offset < strend - s)) {
1833 char *last1; /* Last position checked before */
1837 if (prog->anchored_substr || prog->anchored_utf8) {
1838 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1839 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1840 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1841 back_max = back_min = prog->anchored_offset;
1843 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1844 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1845 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1846 back_max = prog->float_max_offset;
1847 back_min = prog->float_min_offset;
1851 if (must == &PL_sv_undef)
1852 /* could not downgrade utf8 check substring, so must fail */
1858 last = HOP3c(strend, /* Cannot start after this */
1859 -(I32)(CHR_SVLEN(must)
1860 - (SvTAIL(must) != 0) + back_min), strbeg);
1863 last1 = HOPc(s, -1);
1865 last1 = s - 1; /* bogus */
1867 /* XXXX check_substr already used to find "s", can optimize if
1868 check_substr==must. */
1870 dontbother = end_shift;
1871 strend = HOPc(strend, -dontbother);
1872 while ( (s <= last) &&
1873 ((flags & REXEC_SCREAM)
1874 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1875 end_shift, &scream_pos, 0))
1876 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1877 (unsigned char*)strend, must,
1878 multiline ? FBMrf_MULTILINE : 0))) ) {
1879 /* we may be pointing at the wrong string */
1880 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1881 s = strbeg + (s - SvPVX_const(sv));
1882 DEBUG_EXECUTE_r( did_match = 1 );
1883 if (HOPc(s, -back_max) > last1) {
1884 last1 = HOPc(s, -back_min);
1885 s = HOPc(s, -back_max);
1888 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1890 last1 = HOPc(s, -back_min);
1894 while (s <= last1) {
1895 if (regtry(®info, s))
1901 while (s <= last1) {
1902 if (regtry(®info, s))
1908 DEBUG_EXECUTE_r(if (!did_match) {
1909 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1910 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1911 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1912 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1913 ? "anchored" : "floating"),
1914 quoted, RE_SV_TAIL(must));
1918 else if ( (c = prog->regstclass) ) {
1920 const OPCODE op = OP(prog->regstclass);
1921 /* don't bother with what can't match */
1922 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1923 strend = HOPc(strend, -(minlen - 1));
1926 SV * const prop = sv_newmortal();
1927 regprop(prog, prop, c);
1929 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1931 PerlIO_printf(Perl_debug_log,
1932 "Matching stclass %.*s against %s (%d chars)\n",
1933 (int)SvCUR(prop), SvPVX_const(prop),
1934 quoted, (int)(strend - s));
1937 if (find_byclass(prog, c, s, strend, ®info))
1939 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
1943 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1948 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1949 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1950 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1952 if (flags & REXEC_SCREAM) {
1953 last = screaminstr(sv, float_real, s - strbeg,
1954 end_shift, &scream_pos, 1); /* last one */
1956 last = scream_olds; /* Only one occurrence. */
1957 /* we may be pointing at the wrong string */
1958 else if (RX_MATCH_COPIED(prog))
1959 s = strbeg + (s - SvPVX_const(sv));
1963 const char * const little = SvPV_const(float_real, len);
1965 if (SvTAIL(float_real)) {
1966 if (memEQ(strend - len + 1, little, len - 1))
1967 last = strend - len + 1;
1968 else if (!multiline)
1969 last = memEQ(strend - len, little, len)
1970 ? strend - len : NULL;
1976 last = rninstr(s, strend, little, little + len);
1978 last = strend; /* matching "$" */
1982 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1983 "%sCan't trim the tail, match fails (should not happen)%s\n",
1984 PL_colors[4], PL_colors[5]));
1985 goto phooey; /* Should not happen! */
1987 dontbother = strend - last + prog->float_min_offset;
1989 if (minlen && (dontbother < minlen))
1990 dontbother = minlen - 1;
1991 strend -= dontbother; /* this one's always in bytes! */
1992 /* We don't know much -- general case. */
1995 if (regtry(®info, s))
2004 if (regtry(®info, s))
2006 } while (s++ < strend);
2014 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2016 if (PL_reg_eval_set) {
2017 /* Preserve the current value of $^R */
2018 if (oreplsv != GvSV(PL_replgv))
2019 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2020 restored, the value remains
2022 restore_pos(aTHX_ prog);
2025 /* make sure $`, $&, $', and $digit will work later */
2026 if ( !(flags & REXEC_NOT_FIRST) ) {
2027 RX_MATCH_COPY_FREE(prog);
2028 if (flags & REXEC_COPY_STR) {
2029 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2030 #ifdef PERL_OLD_COPY_ON_WRITE
2032 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2034 PerlIO_printf(Perl_debug_log,
2035 "Copy on write: regexp capture, type %d\n",
2038 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2039 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2040 assert (SvPOKp(prog->saved_copy));
2044 RX_MATCH_COPIED_on(prog);
2045 s = savepvn(strbeg, i);
2051 prog->subbeg = strbeg;
2052 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2059 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2060 PL_colors[4], PL_colors[5]));
2061 if (PL_reg_eval_set)
2062 restore_pos(aTHX_ prog);
2067 - regtry - try match at specific point
2069 STATIC I32 /* 0 failure, 1 success */
2070 S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos)
2076 regexp *prog = reginfo->prog;
2077 GET_RE_DEBUG_FLAGS_DECL;
2079 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2082 PL_reg_eval_set = RS_init;
2083 DEBUG_EXECUTE_r(DEBUG_s(
2084 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2085 (IV)(PL_stack_sp - PL_stack_base));
2087 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2088 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2089 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2091 /* Apparently this is not needed, judging by wantarray. */
2092 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2093 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2096 /* Make $_ available to executed code. */
2097 if (reginfo->sv != DEFSV) {
2099 DEFSV = reginfo->sv;
2102 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2103 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2104 /* prepare for quick setting of pos */
2105 #ifdef PERL_OLD_COPY_ON_WRITE
2107 sv_force_normal_flags(sv, 0);
2109 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2110 &PL_vtbl_mglob, NULL, 0);
2114 PL_reg_oldpos = mg->mg_len;
2115 SAVEDESTRUCTOR_X(restore_pos, prog);
2117 if (!PL_reg_curpm) {
2118 Newxz(PL_reg_curpm, 1, PMOP);
2121 SV* const repointer = newSViv(0);
2122 /* so we know which PL_regex_padav element is PL_reg_curpm */
2123 SvFLAGS(repointer) |= SVf_BREAK;
2124 av_push(PL_regex_padav,repointer);
2125 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2126 PL_regex_pad = AvARRAY(PL_regex_padav);
2130 PM_SETRE(PL_reg_curpm, prog);
2131 PL_reg_oldcurpm = PL_curpm;
2132 PL_curpm = PL_reg_curpm;
2133 if (RX_MATCH_COPIED(prog)) {
2134 /* Here is a serious problem: we cannot rewrite subbeg,
2135 since it may be needed if this match fails. Thus
2136 $` inside (?{}) could fail... */
2137 PL_reg_oldsaved = prog->subbeg;
2138 PL_reg_oldsavedlen = prog->sublen;
2139 #ifdef PERL_OLD_COPY_ON_WRITE
2140 PL_nrs = prog->saved_copy;
2142 RX_MATCH_COPIED_off(prog);
2145 PL_reg_oldsaved = NULL;
2146 prog->subbeg = PL_bostr;
2147 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2149 prog->startp[0] = startpos - PL_bostr;
2150 PL_reginput = startpos;
2151 PL_regstartp = prog->startp;
2152 PL_regendp = prog->endp;
2153 PL_reglastparen = &prog->lastparen;
2154 PL_reglastcloseparen = &prog->lastcloseparen;
2155 prog->lastparen = 0;
2156 prog->lastcloseparen = 0;
2158 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2159 if (PL_reg_start_tmpl <= prog->nparens) {
2160 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2161 if(PL_reg_start_tmp)
2162 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2164 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2167 /* XXXX What this code is doing here?!!! There should be no need
2168 to do this again and again, PL_reglastparen should take care of
2171 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2172 * Actually, the code in regcppop() (which Ilya may be meaning by
2173 * PL_reglastparen), is not needed at all by the test suite
2174 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2175 * enough, for building DynaLoader, or otherwise this
2176 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2177 * will happen. Meanwhile, this code *is* needed for the
2178 * above-mentioned test suite tests to succeed. The common theme
2179 * on those tests seems to be returning null fields from matches.
2184 if (prog->nparens) {
2186 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2193 if (regmatch(reginfo, prog->program + 1)) {
2194 prog->endp[0] = PL_reginput - PL_bostr;
2197 REGCP_UNWIND(lastcp);
2202 #define sayYES goto yes
2203 #define sayNO goto no
2204 #define sayNO_ANYOF goto no_anyof
2205 #define sayYES_FINAL goto yes_final
2206 #define sayNO_FINAL goto no_final
2207 #define sayNO_SILENT goto do_no
2208 #define saySAME(x) if (x) goto yes; else goto no
2210 /* we dont use STMT_START/END here because it leads to
2211 "unreachable code" warnings, which are bogus, but distracting. */
2212 #define CACHEsayNO \
2213 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) \
2214 PL_reg_poscache[st->u.whilem.cache_offset] |= \
2215 (1<<st->u.whilem.cache_bit); \
2218 /* this is used to determine how far from the left messages like
2219 'failed...' are printed. It should be set such that messages
2220 are inline with the regop output that created them.
2222 #define REPORT_CODE_OFF 32
2225 /* Make sure there is a test for this +1 options in re_tests */
2226 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2228 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2229 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2231 #define SLAB_FIRST(s) (&(s)->states[0])
2232 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2234 /* grab a new slab and return the first slot in it */
2236 STATIC regmatch_state *
2239 #if PERL_VERSION < 9
2242 regmatch_slab *s = PL_regmatch_slab->next;
2244 Newx(s, 1, regmatch_slab);
2245 s->prev = PL_regmatch_slab;
2247 PL_regmatch_slab->next = s;
2249 PL_regmatch_slab = s;
2250 return SLAB_FIRST(s);
2253 /* simulate a recursive call to regmatch */
2255 #define REGMATCH(ns, where) \
2258 st->resume_state = resume_##where; \
2259 goto start_recurse; \
2260 resume_point_##where:
2262 /* push a new state then goto it */
2264 #define PUSH_STATE_GOTO(state, node) \
2266 st->resume_state = state; \
2269 /* push a new state with success backtracking, then goto it */
2271 #define PUSH_YES_STATE_GOTO(state, node) \
2273 st->resume_state = state; \
2274 goto push_yes_state;
2279 - regmatch - main matching routine
2281 * Conceptually the strategy is simple: check to see whether the current
2282 * node matches, call self recursively to see whether the rest matches,
2283 * and then act accordingly. In practice we make some effort to avoid
2284 * recursion, in particular by going through "ordinary" nodes (that don't
2285 * need to know whether the rest of the match failed) by a loop instead of
2288 /* [lwall] I've hoisted the register declarations to the outer block in order to
2289 * maybe save a little bit of pushing and popping on the stack. It also takes
2290 * advantage of machines that use a register save mask on subroutine entry.
2292 * This function used to be heavily recursive, but since this had the
2293 * effect of blowing the CPU stack on complex regexes, it has been
2294 * restructured to be iterative, and to save state onto the heap rather
2295 * than the stack. Essentially whereever regmatch() used to be called, it
2296 * pushes the current state, notes where to return, then jumps back into
2299 * Originally the structure of this function used to look something like
2304 while (scan != NULL) {
2305 a++; // do stuff with a and b
2311 if (regmatch(...)) // recurse
2321 * Now it looks something like this:
2329 regmatch_state *st = new();
2331 st->a++; // do stuff with a and b
2333 while (scan != NULL) {
2341 st->resume_state = resume_FOO;
2342 goto start_recurse; // recurse
2351 st = new(); push a new state
2352 st->a = 1; st->b = 2;
2359 switch (resume_state) {
2361 goto resume_point_FOO;
2368 * WARNING: this means that any line in this function that contains a
2369 * REGMATCH() or TRYPAREN() is actually simulating a recursive call to
2370 * regmatch() using gotos instead. Thus the values of any local variables
2371 * not saved in the regmatch_state structure will have been lost when
2372 * execution resumes on the next line .
2374 * States (ie the st pointer) are allocated in slabs of about 4K in size.
2375 * PL_regmatch_state always points to the currently active state, and
2376 * PL_regmatch_slab points to the slab currently containing PL_regmatch_state.
2377 * The first time regmatch is called, the first slab is allocated, and is
2378 * never freed until interpreter desctruction. When the slab is full,
2379 * a new one is allocated chained to the end. At exit from regmatch, slabs
2380 * allocated since entry are freed.
2384 #define DEBUG_STATE_pp(pp) \
2386 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2387 PerlIO_printf(Perl_debug_log, \
2390 reg_name[st->resume_state] ); \
2394 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2399 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2400 const char *start, const char *end, const char *blurb)
2402 const bool utf8_pat= prog->reganch & ROPT_UTF8 ? 1 : 0;
2406 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2407 prog->precomp, prog->prelen, 60);
2409 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2410 start, end - start, 60);
2412 PerlIO_printf(Perl_debug_log,
2413 "%s%s REx%s %s against %s\n",
2414 PL_colors[4], blurb, PL_colors[5], s0, s1);
2416 if (do_utf8||utf8_pat)
2417 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2418 utf8_pat ? "pattern" : "",
2419 utf8_pat && do_utf8 ? " and " : "",
2420 do_utf8 ? "string" : ""
2426 S_dump_exec_pos(pTHX_ const char *locinput,
2427 const regnode *scan,
2428 const char *loc_regeol,
2429 const char *loc_bostr,
2430 const char *loc_reg_starttry,
2433 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2434 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2435 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2436 /* The part of the string before starttry has one color
2437 (pref0_len chars), between starttry and current
2438 position another one (pref_len - pref0_len chars),
2439 after the current position the third one.
2440 We assume that pref0_len <= pref_len, otherwise we
2441 decrease pref0_len. */
2442 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2443 ? (5 + taill) - l : locinput - loc_bostr;
2446 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2448 pref0_len = pref_len - (locinput - loc_reg_starttry);
2449 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2450 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2451 ? (5 + taill) - pref_len : loc_regeol - locinput);
2452 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2456 if (pref0_len > pref_len)
2457 pref0_len = pref_len;
2459 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2461 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2462 (locinput - pref_len),pref0_len, 60, 4, 5);
2464 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2465 (locinput - pref_len + pref0_len),
2466 pref_len - pref0_len, 60, 2, 3);
2468 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2469 locinput, loc_regeol - locinput, 10, 0, 1);
2471 const STRLEN tlen=len0+len1+len2;
2472 PerlIO_printf(Perl_debug_log,
2473 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2474 (IV)(locinput - loc_bostr),
2477 (docolor ? "" : "> <"),
2479 (int)(tlen > 19 ? 0 : 19 - tlen),
2486 STATIC I32 /* 0 failure, 1 success */
2487 S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog)
2489 #if PERL_VERSION < 9
2493 register const bool do_utf8 = PL_reg_match_utf8;
2494 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2496 regexp *rex = reginfo->prog;
2498 regmatch_slab *orig_slab;
2499 regmatch_state *orig_state;
2501 /* the current state. This is a cached copy of PL_regmatch_state */
2502 register regmatch_state *st;
2504 /* cache heavy used fields of st in registers */
2505 register regnode *scan;
2506 register regnode *next;
2507 register I32 n = 0; /* initialize to shut up compiler warning */
2508 register char *locinput = PL_reginput;
2510 /* these variables are NOT saved during a recusive RFEGMATCH: */
2511 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2512 bool result = 0; /* return value of S_regmatch */
2513 int depth = 0; /* depth of recursion */
2514 regmatch_state *yes_state = NULL; /* state to pop to on success of
2516 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2517 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2523 GET_RE_DEBUG_FLAGS_DECL;
2526 /* on first ever call to regmatch, allocate first slab */
2527 if (!PL_regmatch_slab) {
2528 Newx(PL_regmatch_slab, 1, regmatch_slab);
2529 PL_regmatch_slab->prev = NULL;
2530 PL_regmatch_slab->next = NULL;
2531 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2534 /* remember current high-water mark for exit */
2535 /* XXX this should be done with SAVE* instead */
2536 orig_slab = PL_regmatch_slab;
2537 orig_state = PL_regmatch_state;
2539 /* grab next free state slot */
2540 st = ++PL_regmatch_state;
2541 if (st > SLAB_LAST(PL_regmatch_slab))
2542 st = PL_regmatch_state = S_push_slab(aTHX);
2549 /* Note that nextchr is a byte even in UTF */
2550 nextchr = UCHARAT(locinput);
2552 while (scan != NULL) {
2555 SV * const prop = sv_newmortal();
2556 regnode *rnext=regnext(scan);
2557 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2558 regprop(rex, prop, scan);
2560 PerlIO_printf(Perl_debug_log,
2561 "%3"IVdf":%*s%s(%"IVdf")\n",
2562 (IV)(scan - rex->program), depth*2, "",
2564 (PL_regkind[OP(scan)] == END || !rnext) ?
2565 0 : (IV)(rnext - rex->program));
2568 next = scan + NEXT_OFF(scan);
2571 state_num = OP(scan);
2574 switch (state_num) {
2576 if (locinput == PL_bostr)
2578 /* reginfo->till = reginfo->bol; */
2583 if (locinput == PL_bostr ||
2584 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2590 if (locinput == PL_bostr)
2594 if (locinput == reginfo->ganch)
2600 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2605 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2607 if (PL_regeol - locinput > 1)
2611 if (PL_regeol != locinput)
2615 if (!nextchr && locinput >= PL_regeol)
2618 locinput += PL_utf8skip[nextchr];
2619 if (locinput > PL_regeol)
2621 nextchr = UCHARAT(locinput);
2624 nextchr = UCHARAT(++locinput);
2627 if (!nextchr && locinput >= PL_regeol)
2629 nextchr = UCHARAT(++locinput);
2632 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2635 locinput += PL_utf8skip[nextchr];
2636 if (locinput > PL_regeol)
2638 nextchr = UCHARAT(locinput);
2641 nextchr = UCHARAT(++locinput);
2645 #define ST st->u.trie
2647 /* In this case the charclass data is available inline so
2648 we can fail fast without a lot of extra overhead.
2650 if (scan->flags == EXACT || !do_utf8) {
2651 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2653 PerlIO_printf(Perl_debug_log,
2654 "%*s %sfailed to match trie start class...%s\n",
2655 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2664 /* what type of TRIE am I? (utf8 makes this contextual) */
2665 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2666 trie_type = do_utf8 ?
2667 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2670 /* what trie are we using right now */
2671 reg_trie_data * const trie
2672 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2673 U32 state = trie->startstate;
2675 if (trie->bitmap && trie_type != trie_utf8_fold &&
2676 !TRIE_BITMAP_TEST(trie,*locinput)
2678 if (trie->states[ state ].wordnum) {
2680 PerlIO_printf(Perl_debug_log,
2681 "%*s %smatched empty string...%s\n",
2682 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2687 PerlIO_printf(Perl_debug_log,
2688 "%*s %sfailed to match trie start class...%s\n",
2689 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2696 U8 *uc = ( U8* )locinput;
2700 U8 *uscan = (U8*)NULL;
2702 SV *sv_accept_buff = NULL;
2703 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2705 ST.accepted = 0; /* how many accepting states we have seen */
2707 ST.jump = trie->jump;
2716 traverse the TRIE keeping track of all accepting states
2717 we transition through until we get to a failing node.
2720 while ( state && uc <= (U8*)PL_regeol ) {
2721 U32 base = trie->states[ state ].trans.base;
2724 /* We use charid to hold the wordnum as we don't use it
2725 for charid until after we have done the wordnum logic.
2726 We define an alias just so that the wordnum logic reads
2729 #define got_wordnum charid
2730 got_wordnum = trie->states[ state ].wordnum;
2732 if ( got_wordnum ) {
2733 if ( ! ST.accepted ) {
2736 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2737 sv_accept_buff=newSV(bufflen *
2738 sizeof(reg_trie_accepted) - 1);
2739 SvCUR_set(sv_accept_buff, 0);
2740 SvPOK_on(sv_accept_buff);
2741 sv_2mortal(sv_accept_buff);
2744 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2747 if (ST.accepted >= bufflen) {
2749 ST.accept_buff =(reg_trie_accepted*)
2750 SvGROW(sv_accept_buff,
2751 bufflen * sizeof(reg_trie_accepted));
2753 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2754 + sizeof(reg_trie_accepted));
2757 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2758 ST.accept_buff[ST.accepted].endpos = uc;
2760 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2764 DEBUG_TRIE_EXECUTE_r({
2765 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2766 PerlIO_printf( Perl_debug_log,
2767 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2768 2+depth * 2, "", PL_colors[4],
2769 (UV)state, (UV)ST.accepted );
2773 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
2774 uvc, charid, foldlen, foldbuf, uniflags);
2777 (base + charid > trie->uniquecharcount )
2778 && (base + charid - 1 - trie->uniquecharcount
2780 && trie->trans[base + charid - 1 -
2781 trie->uniquecharcount].check == state)
2783 state = trie->trans[base + charid - 1 -
2784 trie->uniquecharcount ].next;
2795 DEBUG_TRIE_EXECUTE_r(
2796 PerlIO_printf( Perl_debug_log,
2797 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2798 charid, uvc, (UV)state, PL_colors[5] );
2805 PerlIO_printf( Perl_debug_log,
2806 "%*s %sgot %"IVdf" possible matches%s\n",
2807 REPORT_CODE_OFF + depth * 2, "",
2808 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2814 case TRIE_next_fail: /* we failed - try next alterative */
2816 if ( ST.accepted == 1 ) {
2817 /* only one choice left - just continue */
2819 reg_trie_data * const trie
2820 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2821 SV ** const tmp = RX_DEBUG(reginfo->prog)
2822 ? av_fetch( trie->words, ST.accept_buff[ 0 ].wordnum-1, 0 )
2824 PerlIO_printf( Perl_debug_log,
2825 "%*s %sonly one match left: #%d <%s>%s\n",
2826 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2827 ST.accept_buff[ 0 ].wordnum,
2828 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2831 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2832 /* in this case we free tmps/leave before we call regmatch
2833 as we wont be using accept_buff again. */
2836 locinput = PL_reginput;
2837 nextchr = UCHARAT(locinput);
2842 scan = ST.B - ST.jump[ST.accept_buff[0].wordnum];
2844 continue; /* execute rest of RE */
2847 if (!ST.accepted-- ) {
2854 There are at least two accepting states left. Presumably
2855 the number of accepting states is going to be low,
2856 typically two. So we simply scan through to find the one
2857 with lowest wordnum. Once we find it, we swap the last
2858 state into its place and decrement the size. We then try to
2859 match the rest of the pattern at the point where the word
2860 ends. If we succeed, control just continues along the
2861 regex; if we fail we return here to try the next accepting
2868 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
2869 DEBUG_TRIE_EXECUTE_r(
2870 PerlIO_printf( Perl_debug_log,
2871 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2872 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
2873 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
2874 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
2877 if (ST.accept_buff[cur].wordnum <
2878 ST.accept_buff[best].wordnum)
2883 reg_trie_data * const trie
2884 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2885 SV ** const tmp = RX_DEBUG(reginfo->prog)
2886 ? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 )
2888 regnode *nextop=!ST.jump ?
2890 ST.B - ST.jump[ST.accept_buff[best].wordnum];
2891 PerlIO_printf( Perl_debug_log,
2892 "%*s %strying alternation #%d <%s> at node #%d %s\n",
2893 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2894 ST.accept_buff[best].wordnum,
2895 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2896 REG_NODE_NUM(nextop),
2900 if ( best<ST.accepted ) {
2901 reg_trie_accepted tmp = ST.accept_buff[ best ];
2902 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
2903 ST.accept_buff[ ST.accepted ] = tmp;
2906 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
2908 PUSH_STATE_GOTO(TRIE_next, ST.B);
2911 PUSH_STATE_GOTO(TRIE_next, ST.B - ST.jump[ST.accept_buff[best].wordnum]);
2921 char *s = STRING(scan);
2922 st->ln = STR_LEN(scan);
2923 if (do_utf8 != UTF) {
2924 /* The target and the pattern have differing utf8ness. */
2926 const char * const e = s + st->ln;
2929 /* The target is utf8, the pattern is not utf8. */
2934 if (NATIVE_TO_UNI(*(U8*)s) !=
2935 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2943 /* The target is not utf8, the pattern is utf8. */
2948 if (NATIVE_TO_UNI(*((U8*)l)) !=
2949 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2957 nextchr = UCHARAT(locinput);
2960 /* The target and the pattern have the same utf8ness. */
2961 /* Inline the first character, for speed. */
2962 if (UCHARAT(s) != nextchr)
2964 if (PL_regeol - locinput < st->ln)
2966 if (st->ln > 1 && memNE(s, locinput, st->ln))
2969 nextchr = UCHARAT(locinput);
2973 PL_reg_flags |= RF_tainted;
2976 char * const s = STRING(scan);
2977 st->ln = STR_LEN(scan);
2979 if (do_utf8 || UTF) {
2980 /* Either target or the pattern are utf8. */
2981 const char * const l = locinput;
2982 char *e = PL_regeol;
2984 if (ibcmp_utf8(s, 0, st->ln, (bool)UTF,
2985 l, &e, 0, do_utf8)) {
2986 /* One more case for the sharp s:
2987 * pack("U0U*", 0xDF) =~ /ss/i,
2988 * the 0xC3 0x9F are the UTF-8
2989 * byte sequence for the U+00DF. */
2991 toLOWER(s[0]) == 's' &&
2993 toLOWER(s[1]) == 's' &&
3000 nextchr = UCHARAT(locinput);
3004 /* Neither the target and the pattern are utf8. */
3006 /* Inline the first character, for speed. */
3007 if (UCHARAT(s) != nextchr &&
3008 UCHARAT(s) != ((OP(scan) == EXACTF)
3009 ? PL_fold : PL_fold_locale)[nextchr])
3011 if (PL_regeol - locinput < st->ln)
3013 if (st->ln > 1 && (OP(scan) == EXACTF
3014 ? ibcmp(s, locinput, st->ln)
3015 : ibcmp_locale(s, locinput, st->ln)))
3018 nextchr = UCHARAT(locinput);
3023 STRLEN inclasslen = PL_regeol - locinput;
3025 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3027 if (locinput >= PL_regeol)
3029 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3030 nextchr = UCHARAT(locinput);
3035 nextchr = UCHARAT(locinput);
3036 if (!REGINCLASS(rex, scan, (U8*)locinput))
3038 if (!nextchr && locinput >= PL_regeol)
3040 nextchr = UCHARAT(++locinput);
3044 /* If we might have the case of the German sharp s
3045 * in a casefolding Unicode character class. */
3047 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3048 locinput += SHARP_S_SKIP;
3049 nextchr = UCHARAT(locinput);
3055 PL_reg_flags |= RF_tainted;
3061 LOAD_UTF8_CHARCLASS_ALNUM();
3062 if (!(OP(scan) == ALNUM
3063 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3064 : isALNUM_LC_utf8((U8*)locinput)))
3068 locinput += PL_utf8skip[nextchr];
3069 nextchr = UCHARAT(locinput);
3072 if (!(OP(scan) == ALNUM
3073 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3075 nextchr = UCHARAT(++locinput);
3078 PL_reg_flags |= RF_tainted;
3081 if (!nextchr && locinput >= PL_regeol)
3084 LOAD_UTF8_CHARCLASS_ALNUM();
3085 if (OP(scan) == NALNUM
3086 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3087 : isALNUM_LC_utf8((U8*)locinput))
3091 locinput += PL_utf8skip[nextchr];
3092 nextchr = UCHARAT(locinput);
3095 if (OP(scan) == NALNUM
3096 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3098 nextchr = UCHARAT(++locinput);
3102 PL_reg_flags |= RF_tainted;
3106 /* was last char in word? */
3108 if (locinput == PL_bostr)
3111 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3113 st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3115 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3116 st->ln = isALNUM_uni(st->ln);
3117 LOAD_UTF8_CHARCLASS_ALNUM();
3118 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3121 st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln));
3122 n = isALNUM_LC_utf8((U8*)locinput);
3126 st->ln = (locinput != PL_bostr) ?
3127 UCHARAT(locinput - 1) : '\n';
3128 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3129 st->ln = isALNUM(st->ln);
3130 n = isALNUM(nextchr);
3133 st->ln = isALNUM_LC(st->ln);
3134 n = isALNUM_LC(nextchr);
3137 if (((!st->ln) == (!n)) == (OP(scan) == BOUND ||
3138 OP(scan) == BOUNDL))
3142 PL_reg_flags |= RF_tainted;
3148 if (UTF8_IS_CONTINUED(nextchr)) {
3149 LOAD_UTF8_CHARCLASS_SPACE();
3150 if (!(OP(scan) == SPACE
3151 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3152 : isSPACE_LC_utf8((U8*)locinput)))
3156 locinput += PL_utf8skip[nextchr];
3157 nextchr = UCHARAT(locinput);
3160 if (!(OP(scan) == SPACE
3161 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3163 nextchr = UCHARAT(++locinput);
3166 if (!(OP(scan) == SPACE
3167 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3169 nextchr = UCHARAT(++locinput);
3173 PL_reg_flags |= RF_tainted;
3176 if (!nextchr && locinput >= PL_regeol)
3179 LOAD_UTF8_CHARCLASS_SPACE();
3180 if (OP(scan) == NSPACE
3181 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3182 : isSPACE_LC_utf8((U8*)locinput))
3186 locinput += PL_utf8skip[nextchr];
3187 nextchr = UCHARAT(locinput);
3190 if (OP(scan) == NSPACE
3191 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3193 nextchr = UCHARAT(++locinput);
3196 PL_reg_flags |= RF_tainted;
3202 LOAD_UTF8_CHARCLASS_DIGIT();
3203 if (!(OP(scan) == DIGIT
3204 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3205 : isDIGIT_LC_utf8((U8*)locinput)))
3209 locinput += PL_utf8skip[nextchr];
3210 nextchr = UCHARAT(locinput);
3213 if (!(OP(scan) == DIGIT
3214 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3216 nextchr = UCHARAT(++locinput);
3219 PL_reg_flags |= RF_tainted;
3222 if (!nextchr && locinput >= PL_regeol)
3225 LOAD_UTF8_CHARCLASS_DIGIT();
3226 if (OP(scan) == NDIGIT
3227 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3228 : isDIGIT_LC_utf8((U8*)locinput))
3232 locinput += PL_utf8skip[nextchr];
3233 nextchr = UCHARAT(locinput);
3236 if (OP(scan) == NDIGIT
3237 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3239 nextchr = UCHARAT(++locinput);
3242 if (locinput >= PL_regeol)
3245 LOAD_UTF8_CHARCLASS_MARK();
3246 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3248 locinput += PL_utf8skip[nextchr];
3249 while (locinput < PL_regeol &&
3250 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3251 locinput += UTF8SKIP(locinput);
3252 if (locinput > PL_regeol)
3257 nextchr = UCHARAT(locinput);
3260 PL_reg_flags |= RF_tainted;
3265 n = ARG(scan); /* which paren pair */
3266 st->ln = PL_regstartp[n];
3267 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3268 if ((I32)*PL_reglastparen < n || st->ln == -1)
3269 sayNO; /* Do not match unless seen CLOSEn. */
3270 if (st->ln == PL_regendp[n])
3273 s = PL_bostr + st->ln;
3274 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3276 const char *e = PL_bostr + PL_regendp[n];
3278 * Note that we can't do the "other character" lookup trick as
3279 * in the 8-bit case (no pun intended) because in Unicode we
3280 * have to map both upper and title case to lower case.
3282 if (OP(scan) == REFF) {
3284 STRLEN ulen1, ulen2;
3285 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3286 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3290 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3291 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3292 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3299 nextchr = UCHARAT(locinput);
3303 /* Inline the first character, for speed. */
3304 if (UCHARAT(s) != nextchr &&
3306 (UCHARAT(s) != ((OP(scan) == REFF
3307 ? PL_fold : PL_fold_locale)[nextchr]))))
3309 st->ln = PL_regendp[n] - st->ln;
3310 if (locinput + st->ln > PL_regeol)
3312 if (st->ln > 1 && (OP(scan) == REF
3313 ? memNE(s, locinput, st->ln)
3315 ? ibcmp(s, locinput, st->ln)
3316 : ibcmp_locale(s, locinput, st->ln))))
3319 nextchr = UCHARAT(locinput);
3330 #define ST st->u.eval
3332 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3336 /* execute the code in the {...} */
3338 SV ** const before = SP;
3339 OP_4tree * const oop = PL_op;
3340 COP * const ocurcop = PL_curcop;
3344 PL_op = (OP_4tree*)rex->data->data[n];
3345 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3346 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3347 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3349 CALLRUNOPS(aTHX); /* Scalar context. */
3352 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3359 PAD_RESTORE_LOCAL(old_comppad);
3360 PL_curcop = ocurcop;
3363 sv_setsv(save_scalar(PL_replgv), ret);
3367 if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */
3370 /* extract RE object from returned value; compiling if
3375 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3376 mg = mg_find(sv, PERL_MAGIC_qr);
3377 else if (SvSMAGICAL(ret)) {
3378 if (SvGMAGICAL(ret))
3379 sv_unmagic(ret, PERL_MAGIC_qr);
3381 mg = mg_find(ret, PERL_MAGIC_qr);
3385 re = (regexp *)mg->mg_obj;
3386 (void)ReREFCNT_inc(re);
3390 const char * const t = SvPV_const(ret, len);
3392 const I32 osize = PL_regsize;
3395 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3396 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3398 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3400 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3406 /* run the pattern returned from (??{...}) */
3407 ST.cp = regcppush(0); /* Save *all* the positions. */
3408 REGCP_SET(ST.lastcp);
3409 *PL_reglastparen = 0;
3410 *PL_reglastcloseparen = 0;
3411 PL_reginput = locinput;
3413 /* XXXX This is too dramatic a measure... */
3417 ST.toggle_reg_flags = PL_reg_flags;
3418 if (re->reganch & ROPT_UTF8)
3419 PL_reg_flags |= RF_utf8;
3421 PL_reg_flags &= ~RF_utf8;
3422 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3425 ST.prev_curlyx = cur_curlyx;
3429 ST.prev_eval = cur_eval;
3433 debug_start_match(re, do_utf8, locinput, PL_regeol,
3434 "Matching embedded");
3436 /* now continue from first node in postoned RE */
3437 PUSH_YES_STATE_GOTO(EVAL_AB, re->program + 1);
3440 /* /(?(?{...})X|Y)/ */
3441 st->sw = (bool)SvTRUE(ret);
3446 case EVAL_AB: /* cleanup after a successful (??{A})B */
3447 /* note: this is called twice; first after popping B, then A */
3448 PL_reg_flags ^= ST.toggle_reg_flags;
3452 cur_eval = ST.prev_eval;
3453 cur_curlyx = ST.prev_curlyx;
3454 /* XXXX This is too dramatic a measure... */
3459 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3460 /* note: this is called twice; first after popping B, then A */
3461 PL_reg_flags ^= ST.toggle_reg_flags;
3464 PL_reginput = locinput;
3465 REGCP_UNWIND(ST.lastcp);
3467 cur_eval = ST.prev_eval;
3468 cur_curlyx = ST.prev_curlyx;
3469 /* XXXX This is too dramatic a measure... */
3476 n = ARG(scan); /* which paren pair */
3477 PL_reg_start_tmp[n] = locinput;
3482 n = ARG(scan); /* which paren pair */
3483 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3484 PL_regendp[n] = locinput - PL_bostr;
3485 if (n > (I32)*PL_reglastparen)
3486 *PL_reglastparen = n;
3487 *PL_reglastcloseparen = n;
3490 n = ARG(scan); /* which paren pair */
3491 st->sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3494 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3496 next = NEXTOPER(NEXTOPER(scan));
3498 next = scan + ARG(scan);
3499 if (OP(next) == IFTHEN) /* Fake one. */
3500 next = NEXTOPER(NEXTOPER(next));
3504 st->logical = scan->flags;
3506 /*******************************************************************
3507 cur_curlyx points to the regmatch_state associated with the most recent CURLYX.
3508 This struct contains info about the innermost (...)* loop (an
3509 "infoblock"), and a pointer to the next outer cur_curlyx.
3511 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3513 1) After matching Y, regnode for CURLYX is processed;
3515 2) This regnode populates cur_curlyx, and calls regmatch() recursively
3516 with the starting point at WHILEM node;
3518 3) Each hit of WHILEM node tries to match A and Z (in the order
3519 depending on the current iteration, min/max of {min,max} and
3520 greediness). The information about where are nodes for "A"
3521 and "Z" is read from cur_curlyx, as is info on how many times "A"
3522 was already matched, and greediness.
3524 4) After A matches, the same WHILEM node is hit again.
3526 5) Each time WHILEM is hit, cur_curlyx is the infoblock created by CURLYX
3527 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3528 resets cur_curlyx, since this Y(A)*Z can be a part of some other loop:
3529 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3530 of the external loop.
3532 Currently present infoblocks form a tree with a stem formed by cur_curlyx
3533 and whatever it mentions via ->next, and additional attached trees
3534 corresponding to temporarily unset infoblocks as in "5" above.
3536 In the following picture, infoblocks for outer loop of
3537 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3538 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3539 infoblocks are drawn below the "reset" infoblock.
3541 In fact in the picture below we do not show failed matches for Z and T
3542 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3543 more obvious *why* one needs to *temporary* unset infoblocks.]
3545 Matched REx position InfoBlocks Comment
3549 Y A)*?Z)*?T x <- O <- I
3550 YA )*?Z)*?T x <- O <- I
3551 YA A)*?Z)*?T x <- O <- I
3552 YAA )*?Z)*?T x <- O <- I
3553 YAA Z)*?T x <- O # Temporary unset I
3556 YAAZ Y(A)*?Z)*?T x <- O
3559 YAAZY (A)*?Z)*?T x <- O
3562 YAAZY A)*?Z)*?T x <- O <- I
3565 YAAZYA )*?Z)*?T x <- O <- I
3568 YAAZYA Z)*?T x <- O # Temporary unset I
3574 YAAZYAZ T x # Temporary unset O
3581 *******************************************************************/
3584 /* No need to save/restore up to this paren */
3585 parenfloor = scan->flags;
3589 CURLYX and WHILEM are always paired: they're the moral
3590 equivalent of pp_enteriter anbd pp_iter.
3592 The only time next could be null is if the node tree is
3593 corrupt. This was mentioned on p5p a few days ago.
3595 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3596 So we'll assert that this is true:
3599 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3601 /* XXXX Probably it is better to teach regpush to support
3602 parenfloor > PL_regsize... */
3603 if (parenfloor > (I32)*PL_reglastparen)
3604 parenfloor = *PL_reglastparen; /* Pessimization... */
3606 st->u.curlyx.cp = PL_savestack_ix;
3607 st->u.curlyx.outercc = cur_curlyx;
3609 /* these fields contain the state of the current curly.
3610 * they are accessed by subsequent WHILEMs;
3611 * cur and lastloc are also updated by WHILEM */
3612 st->u.curlyx.parenfloor = parenfloor;
3613 st->u.curlyx.cur = -1; /* this will be updated by WHILEM */
3614 st->u.curlyx.min = ARG1(scan);
3615 st->u.curlyx.max = ARG2(scan);
3616 st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3617 st->u.curlyx.lastloc = 0;
3618 /* st->next and st->minmod are also read by WHILEM */
3620 PL_reginput = locinput;
3621 REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */
3622 /*** all unsaved local vars undefined at this point */
3623 regcpblow(st->u.curlyx.cp);
3624 cur_curlyx = st->u.curlyx.outercc;
3630 * This is really hard to understand, because after we match
3631 * what we're trying to match, we must make sure the rest of
3632 * the REx is going to match for sure, and to do that we have
3633 * to go back UP the parse tree by recursing ever deeper. And
3634 * if it fails, we have to reset our parent's current state
3635 * that we can try again after backing off.
3640 cur_curlyx gets initialised by CURLYX ready for use by WHILEM.
3641 So again, unless somethings been corrupted, cur_curlyx cannot
3642 be null at that point in WHILEM.
3644 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3645 So we'll assert that this is true:
3648 st->u.whilem.lastloc = cur_curlyx->u.curlyx.lastloc; /* Detection of 0-len. */
3649 st->u.whilem.cache_offset = 0;
3650 st->u.whilem.cache_bit = 0;
3652 n = cur_curlyx->u.curlyx.cur + 1; /* how many we know we matched */
3653 PL_reginput = locinput;
3656 PerlIO_printf(Perl_debug_log,
3657 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3658 REPORT_CODE_OFF+depth*2, "",
3659 (long)n, (long)cur_curlyx->u.curlyx.min,
3660 (long)cur_curlyx->u.curlyx.max,
3664 /* If degenerate scan matches "", assume scan done. */
3666 if (locinput == cur_curlyx->u.curlyx.lastloc && n >=
3667 cur_curlyx->u.curlyx.min)
3669 st->u.whilem.savecc = cur_curlyx;
3670 cur_curlyx = cur_curlyx->u.curlyx.outercc;
3672 st->ln = cur_curlyx->u.curlyx.cur;
3674 PerlIO_printf(Perl_debug_log,
3675 "%*s empty match detected, try continuation...\n",
3676 REPORT_CODE_OFF+depth*2, "")
3678 REGMATCH(st->u.whilem.savecc->next, WHILEM1);
3679 /*** all unsaved local vars undefined at this point */
3680 cur_curlyx = st->u.whilem.savecc;
3683 if (cur_curlyx->u.curlyx.outercc)
3684 cur_curlyx->u.curlyx.outercc->u.curlyx.cur = st->ln;
3688 /* First just match a string of min scans. */
3690 if (n < cur_curlyx->u.curlyx.min) {
3691 cur_curlyx->u.curlyx.cur = n;
3692 cur_curlyx->u.curlyx.lastloc = locinput;
3693 REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM2);
3694 /*** all unsaved local vars undefined at this point */
3697 cur_curlyx->u.curlyx.cur = n - 1;
3698 cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
3703 /* Check whether we already were at this position.
3704 Postpone detection until we know the match is not
3705 *that* much linear. */
3706 if (!PL_reg_maxiter) {
3707 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3708 /* possible overflow for long strings and many CURLYX's */
3709 if (PL_reg_maxiter < 0)
3710 PL_reg_maxiter = I32_MAX;
3711 PL_reg_leftiter = PL_reg_maxiter;
3713 if (PL_reg_leftiter-- == 0) {
3714 const I32 size = (PL_reg_maxiter + 7)/8;
3715 if (PL_reg_poscache) {
3716 if ((I32)PL_reg_poscache_size < size) {
3717 Renew(PL_reg_poscache, size, char);
3718 PL_reg_poscache_size = size;
3720 Zero(PL_reg_poscache, size, char);
3723 PL_reg_poscache_size = size;
3724 Newxz(PL_reg_poscache, size, char);
3727 PerlIO_printf(Perl_debug_log,
3728 "%sDetected a super-linear match, switching on caching%s...\n",
3729 PL_colors[4], PL_colors[5])
3732 if (PL_reg_leftiter < 0) {
3733 st->u.whilem.cache_offset = locinput - PL_bostr;
3735 st->u.whilem.cache_offset = (scan->flags & 0xf) - 1
3736 + st->u.whilem.cache_offset * (scan->flags>>4);
3737 st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8;
3738 st->u.whilem.cache_offset /= 8;
3739 if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<<st->u.whilem.cache_bit)) {
3741 PerlIO_printf(Perl_debug_log,
3742 "%*s already tried at this position...\n",
3743 REPORT_CODE_OFF+depth*2, "")
3745 sayNO; /* cache records failure */
3750 /* Prefer next over scan for minimal matching. */
3752 if (cur_curlyx->minmod) {
3753 st->u.whilem.savecc = cur_curlyx;
3754 cur_curlyx = cur_curlyx->u.curlyx.outercc;
3756 st->ln = cur_curlyx->u.curlyx.cur;
3757 st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor);
3758 REGCP_SET(st->u.whilem.lastcp);
3759 REGMATCH(st->u.whilem.savecc->next, WHILEM3);
3760 /*** all unsaved local vars undefined at this point */
3761 cur_curlyx = st->u.whilem.savecc;
3763 regcpblow(st->u.whilem.cp);
3764 sayYES; /* All done. */
3766 REGCP_UNWIND(st->u.whilem.lastcp);
3768 if (cur_curlyx->u.curlyx.outercc)
3769 cur_curlyx->u.curlyx.outercc->u.curlyx.cur = st->ln;
3771 if (n >= cur_curlyx->u.curlyx.max) { /* Maximum greed exceeded? */
3772 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3773 && !(PL_reg_flags & RF_warned)) {
3774 PL_reg_flags |= RF_warned;
3775 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3776 "Complex regular subexpression recursion",
3783 PerlIO_printf(Perl_debug_log,
3784 "%*s trying longer...\n",
3785 REPORT_CODE_OFF+depth*2, "")
3787 /* Try scanning more and see if it helps. */
3788 PL_reginput = locinput;
3789 cur_curlyx->u.curlyx.cur = n;
3790 cur_curlyx->u.curlyx.lastloc = locinput;
3791 st->u.whilem.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3792 REGCP_SET(st->u.whilem.lastcp);
3793 REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM4);
3794 /*** all unsaved local vars undefined at this point */
3796 regcpblow(st->u.whilem.cp);
3799 REGCP_UNWIND(st->u.whilem.lastcp);
3801 cur_curlyx->u.curlyx.cur = n - 1;
3802 cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
3806 /* Prefer scan over next for maximal matching. */
3808 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
3809 st->u.whilem.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3810 cur_curlyx->u.curlyx.cur = n;
3811 cur_curlyx->u.curlyx.lastloc = locinput;
3812 REGCP_SET(st->u.whilem.lastcp);
3813 REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM5);
3814 /*** all unsaved local vars undefined at this point */
3816 regcpblow(st->u.whilem.cp);
3819 REGCP_UNWIND(st->u.whilem.lastcp);
3820 regcppop(rex); /* Restore some previous $<digit>s? */
3821 PL_reginput = locinput;
3823 PerlIO_printf(Perl_debug_log,
3824 "%*s failed, try continuation...\n",
3825 REPORT_CODE_OFF+depth*2, "")
3828 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3829 && !(PL_reg_flags & RF_warned)) {
3830 PL_reg_flags |= RF_warned;
3831 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3832 "Complex regular subexpression recursion",
3836 /* Failed deeper matches of scan, so see if this one works. */
3837 st->u.whilem.savecc = cur_curlyx;
3838 cur_curlyx = cur_curlyx->u.curlyx.outercc;
3840 st->ln = cur_curlyx->u.curlyx.cur;
3841 REGMATCH(st->u.whilem.savecc->next, WHILEM6);
3842 /*** all unsaved local vars undefined at this point */
3843 cur_curlyx = st->u.whilem.savecc;
3846 if (cur_curlyx->u.curlyx.outercc)
3847 cur_curlyx->u.curlyx.outercc->u.curlyx.cur = st->ln;
3848 cur_curlyx->u.curlyx.cur = n - 1;
3849 cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
3855 #define ST st->u.branch
3857 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
3858 next = scan + ARG(scan);
3861 scan = NEXTOPER(scan);
3864 case BRANCH: /* /(...|A|...)/ */
3865 scan = NEXTOPER(scan); /* scan now points to inner node */
3866 if (!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
3867 /* last branch; skip state push and jump direct to node */
3869 ST.lastparen = *PL_reglastparen;
3870 ST.next_branch = next;
3872 PL_reginput = locinput;
3874 /* Now go into the branch */
3875 PUSH_STATE_GOTO(BRANCH_next, scan);
3878 case BRANCH_next_fail: /* that branch failed; try the next, if any */
3879 REGCP_UNWIND(ST.cp);
3880 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3882 *PL_reglastparen = n;
3883 scan = ST.next_branch;
3884 /* no more branches? */
3885 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ))
3887 continue; /* execute next BRANCH[J] op */
3895 #define ST st->u.curlym
3897 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
3899 /* This is an optimisation of CURLYX that enables us to push
3900 * only a single backtracking state, no matter now many matches
3901 * there are in {m,n}. It relies on the pattern being constant
3902 * length, with no parens to influence future backrefs