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); \
154 static void restore_pos(pTHX_ void *arg);
157 S_regcppush(pTHX_ I32 parenfloor)
160 const int retval = PL_savestack_ix;
161 #define REGCP_PAREN_ELEMS 4
162 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
164 GET_RE_DEBUG_FLAGS_DECL;
166 if (paren_elems_to_push < 0)
167 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
169 #define REGCP_OTHER_ELEMS 8
170 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
172 for (p = PL_regsize; p > parenfloor; p--) {
173 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
174 SSPUSHINT(PL_regendp[p]);
175 SSPUSHINT(PL_regstartp[p]);
176 SSPUSHPTR(PL_reg_start_tmp[p]);
178 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
179 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
180 (UV)p, (IV)PL_regstartp[p],
181 (IV)(PL_reg_start_tmp[p] - PL_bostr),
185 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
186 SSPUSHPTR(PL_regstartp);
187 SSPUSHPTR(PL_regendp);
188 SSPUSHINT(PL_regsize);
189 SSPUSHINT(*PL_reglastparen);
190 SSPUSHINT(*PL_reglastcloseparen);
191 SSPUSHPTR(PL_reginput);
192 #define REGCP_FRAME_ELEMS 2
193 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
194 * are needed for the regexp context stack bookkeeping. */
195 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
196 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
201 /* These are needed since we do not localize EVAL nodes: */
202 #define REGCP_SET(cp) \
204 PerlIO_printf(Perl_debug_log, \
205 " Setting an EVAL scope, savestack=%"IVdf"\n", \
206 (IV)PL_savestack_ix)); \
209 #define REGCP_UNWIND(cp) \
211 if (cp != PL_savestack_ix) \
212 PerlIO_printf(Perl_debug_log, \
213 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
214 (IV)(cp), (IV)PL_savestack_ix)); \
218 S_regcppop(pTHX_ const regexp *rex)
224 GET_RE_DEBUG_FLAGS_DECL;
226 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
228 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
229 i = SSPOPINT; /* Parentheses elements to pop. */
230 input = (char *) SSPOPPTR;
231 *PL_reglastcloseparen = SSPOPINT;
232 *PL_reglastparen = SSPOPINT;
233 PL_regsize = SSPOPINT;
234 PL_regendp=(I32 *) SSPOPPTR;
235 PL_regstartp=(I32 *) SSPOPPTR;
238 /* Now restore the parentheses context. */
239 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
240 i > 0; i -= REGCP_PAREN_ELEMS) {
242 U32 paren = (U32)SSPOPINT;
243 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
244 PL_regstartp[paren] = SSPOPINT;
246 if (paren <= *PL_reglastparen)
247 PL_regendp[paren] = tmps;
249 PerlIO_printf(Perl_debug_log,
250 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
251 (UV)paren, (IV)PL_regstartp[paren],
252 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
253 (IV)PL_regendp[paren],
254 (paren > *PL_reglastparen ? "(no)" : ""));
258 if (*PL_reglastparen + 1 <= rex->nparens) {
259 PerlIO_printf(Perl_debug_log,
260 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
261 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
265 /* It would seem that the similar code in regtry()
266 * already takes care of this, and in fact it is in
267 * a better location to since this code can #if 0-ed out
268 * but the code in regtry() is needed or otherwise tests
269 * requiring null fields (pat.t#187 and split.t#{13,14}
270 * (as of patchlevel 7877) will fail. Then again,
271 * this code seems to be necessary or otherwise
272 * building DynaLoader will fail:
273 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
275 for (i = *PL_reglastparen + 1; (U32)i <= rex->nparens; i++) {
277 PL_regstartp[i] = -1;
284 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
287 * pregexec and friends
290 #ifndef PERL_IN_XSUB_RE
292 - pregexec - match a regexp against a string
295 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
296 char *strbeg, I32 minend, SV *screamer, U32 nosave)
297 /* strend: pointer to null at end of string */
298 /* strbeg: real beginning of string */
299 /* minend: end of match must be >=minend after stringarg. */
300 /* nosave: For optimizations. */
303 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
304 nosave ? 0 : REXEC_COPY_STR);
309 * Need to implement the following flags for reg_anch:
311 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
313 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
314 * INTUIT_AUTORITATIVE_ML
315 * INTUIT_ONCE_NOML - Intuit can match in one location only.
318 * Another flag for this function: SECOND_TIME (so that float substrs
319 * with giant delta may be not rechecked).
322 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
324 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
325 Otherwise, only SvCUR(sv) is used to get strbeg. */
327 /* XXXX We assume that strpos is strbeg unless sv. */
329 /* XXXX Some places assume that there is a fixed substring.
330 An update may be needed if optimizer marks as "INTUITable"
331 RExen without fixed substrings. Similarly, it is assumed that
332 lengths of all the strings are no more than minlen, thus they
333 cannot come from lookahead.
334 (Or minlen should take into account lookahead.)
335 NOTE: Some of this comment is not correct. minlen does now take account
336 of lookahead/behind. Further research is required. -- demerphq
340 /* A failure to find a constant substring means that there is no need to make
341 an expensive call to REx engine, thus we celebrate a failure. Similarly,
342 finding a substring too deep into the string means that less calls to
343 regtry() should be needed.
345 REx compiler's optimizer found 4 possible hints:
346 a) Anchored substring;
348 c) Whether we are anchored (beginning-of-line or \G);
349 d) First node (of those at offset 0) which may distingush positions;
350 We use a)b)d) and multiline-part of c), and try to find a position in the
351 string which does not contradict any of them.
354 /* Most of decisions we do here should have been done at compile time.
355 The nodes of the REx which we used for the search should have been
356 deleted from the finite automaton. */
359 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
360 char *strend, U32 flags, re_scream_pos_data *data)
363 register I32 start_shift = 0;
364 /* Should be nonnegative! */
365 register I32 end_shift = 0;
370 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
372 register char *other_last = NULL; /* other substr checked before this */
373 char *check_at = NULL; /* check substr found at this pos */
374 const I32 multiline = prog->reganch & PMf_MULTILINE;
376 const char * const i_strpos = strpos;
379 GET_RE_DEBUG_FLAGS_DECL;
381 RX_MATCH_UTF8_set(prog,do_utf8);
383 if (prog->reganch & ROPT_UTF8) {
384 PL_reg_flags |= RF_utf8;
387 debug_start_match(prog, do_utf8, strpos, strend,
388 sv ? "Guessing start of match in sv for"
389 : "Guessing start of match in string for");
392 /* CHR_DIST() would be more correct here but it makes things slow. */
393 if (prog->minlen > strend - strpos) {
394 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
395 "String too short... [re_intuit_start]\n"));
399 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
402 if (!prog->check_utf8 && prog->check_substr)
403 to_utf8_substr(prog);
404 check = prog->check_utf8;
406 if (!prog->check_substr && prog->check_utf8)
407 to_byte_substr(prog);
408 check = prog->check_substr;
410 if (check == &PL_sv_undef) {
411 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
412 "Non-utf8 string cannot match utf8 check string\n"));
415 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
416 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
417 || ( (prog->reganch & ROPT_ANCH_BOL)
418 && !multiline ) ); /* Check after \n? */
421 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
422 | ROPT_IMPLICIT)) /* not a real BOL */
423 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
425 && (strpos != strbeg)) {
426 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
429 if (prog->check_offset_min == prog->check_offset_max &&
430 !(prog->reganch & ROPT_CANY_SEEN)) {
431 /* Substring at constant offset from beg-of-str... */
434 s = HOP3c(strpos, prog->check_offset_min, strend);
437 slen = SvCUR(check); /* >= 1 */
439 if ( strend - s > slen || strend - s < slen - 1
440 || (strend - s == slen && strend[-1] != '\n')) {
441 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
444 /* Now should match s[0..slen-2] */
446 if (slen && (*SvPVX_const(check) != *s
448 && memNE(SvPVX_const(check), s, slen)))) {
450 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
454 else if (*SvPVX_const(check) != *s
455 || ((slen = SvCUR(check)) > 1
456 && memNE(SvPVX_const(check), s, slen)))
459 goto success_at_start;
462 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
464 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
465 end_shift = prog->check_end_shift;
468 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
469 - (SvTAIL(check) != 0);
470 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
472 if (end_shift < eshift)
476 else { /* Can match at random position */
479 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
480 end_shift = prog->check_end_shift;
482 /* end shift should be non negative here */
485 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
487 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
488 (IV)end_shift, prog->precomp);
492 /* Find a possible match in the region s..strend by looking for
493 the "check" substring in the region corrected by start/end_shift. */
496 I32 srch_start_shift = start_shift;
497 I32 srch_end_shift = end_shift;
498 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
499 srch_end_shift -= ((strbeg - s) - srch_start_shift);
500 srch_start_shift = strbeg - s;
502 DEBUG_OPTIMISE_MORE_r({
503 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
504 (IV)prog->check_offset_min,
505 (IV)srch_start_shift,
507 (IV)prog->check_end_shift);
510 if (flags & REXEC_SCREAM) {
511 I32 p = -1; /* Internal iterator of scream. */
512 I32 * const pp = data ? data->scream_pos : &p;
514 if (PL_screamfirst[BmRARE(check)] >= 0
515 || ( BmRARE(check) == '\n'
516 && (BmPREVIOUS(check) == SvCUR(check) - 1)
518 s = screaminstr(sv, check,
519 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
522 /* we may be pointing at the wrong string */
523 if (s && RX_MATCH_COPIED(prog))
524 s = strbeg + (s - SvPVX_const(sv));
526 *data->scream_olds = s;
531 if (prog->reganch & ROPT_CANY_SEEN) {
532 start_point= (U8*)(s + srch_start_shift);
533 end_point= (U8*)(strend - srch_end_shift);
535 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
536 end_point= HOP3(strend, -srch_end_shift, strbeg);
538 DEBUG_OPTIMISE_MORE_r({
539 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
540 (int)(end_point - start_point),
541 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
545 s = fbm_instr( start_point, end_point,
546 check, multiline ? FBMrf_MULTILINE : 0);
549 /* Update the count-of-usability, remove useless subpatterns,
553 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
554 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
555 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
556 (s ? "Found" : "Did not find"),
557 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
558 ? "anchored" : "floating"),
561 (s ? " at offset " : "...\n") );
566 /* Finish the diagnostic message */
567 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
569 /* XXX dmq: first branch is for positive lookbehind...
570 Our check string is offset from the beginning of the pattern.
571 So we need to do any stclass tests offset forward from that
580 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
581 Start with the other substr.
582 XXXX no SCREAM optimization yet - and a very coarse implementation
583 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
584 *always* match. Probably should be marked during compile...
585 Probably it is right to do no SCREAM here...
588 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
589 : (prog->float_substr && prog->anchored_substr))
591 /* Take into account the "other" substring. */
592 /* XXXX May be hopelessly wrong for UTF... */
595 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
598 char * const last = HOP3c(s, -start_shift, strbeg);
600 char * const saved_s = s;
603 t = s - prog->check_offset_max;
604 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
606 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
611 t = HOP3c(t, prog->anchored_offset, strend);
612 if (t < other_last) /* These positions already checked */
614 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
617 /* XXXX It is not documented what units *_offsets are in.
618 We assume bytes, but this is clearly wrong.
619 Meaning this code needs to be carefully reviewed for errors.
623 /* On end-of-str: see comment below. */
624 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
625 if (must == &PL_sv_undef) {
627 DEBUG_r(must = prog->anchored_utf8); /* for debug */
632 HOP3(HOP3(last1, prog->anchored_offset, strend)
633 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
635 multiline ? FBMrf_MULTILINE : 0
638 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
639 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
640 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
641 (s ? "Found" : "Contradicts"),
642 quoted, RE_SV_TAIL(must));
647 if (last1 >= last2) {
648 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
649 ", giving up...\n"));
652 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
653 ", trying floating at offset %ld...\n",
654 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
655 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
656 s = HOP3c(last, 1, strend);
660 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
661 (long)(s - i_strpos)));
662 t = HOP3c(s, -prog->anchored_offset, strbeg);
663 other_last = HOP3c(s, 1, strend);
671 else { /* Take into account the floating substring. */
673 char * const saved_s = s;
676 t = HOP3c(s, -start_shift, strbeg);
678 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
679 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
680 last = HOP3c(t, prog->float_max_offset, strend);
681 s = HOP3c(t, prog->float_min_offset, strend);
684 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
685 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
686 /* fbm_instr() takes into account exact value of end-of-str
687 if the check is SvTAIL(ed). Since false positives are OK,
688 and end-of-str is not later than strend we are OK. */
689 if (must == &PL_sv_undef) {
691 DEBUG_r(must = prog->float_utf8); /* for debug message */
694 s = fbm_instr((unsigned char*)s,
695 (unsigned char*)last + SvCUR(must)
697 must, multiline ? FBMrf_MULTILINE : 0);
699 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
700 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
701 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
702 (s ? "Found" : "Contradicts"),
703 quoted, RE_SV_TAIL(must));
707 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
708 ", giving up...\n"));
711 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
712 ", trying anchored starting at offset %ld...\n",
713 (long)(saved_s + 1 - i_strpos)));
715 s = HOP3c(t, 1, strend);
719 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
720 (long)(s - i_strpos)));
721 other_last = s; /* Fix this later. --Hugo */
731 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
733 DEBUG_OPTIMISE_MORE_r(
734 PerlIO_printf(Perl_debug_log,
735 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
736 (IV)prog->check_offset_min,
737 (IV)prog->check_offset_max,
745 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
747 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
750 /* Fixed substring is found far enough so that the match
751 cannot start at strpos. */
753 if (ml_anch && t[-1] != '\n') {
754 /* Eventually fbm_*() should handle this, but often
755 anchored_offset is not 0, so this check will not be wasted. */
756 /* XXXX In the code below we prefer to look for "^" even in
757 presence of anchored substrings. And we search even
758 beyond the found float position. These pessimizations
759 are historical artefacts only. */
761 while (t < strend - prog->minlen) {
763 if (t < check_at - prog->check_offset_min) {
764 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
765 /* Since we moved from the found position,
766 we definitely contradict the found anchored
767 substr. Due to the above check we do not
768 contradict "check" substr.
769 Thus we can arrive here only if check substr
770 is float. Redo checking for "other"=="fixed".
773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
774 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
775 goto do_other_anchored;
777 /* We don't contradict the found floating substring. */
778 /* XXXX Why not check for STCLASS? */
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
781 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
784 /* Position contradicts check-string */
785 /* XXXX probably better to look for check-string
786 than for "\n", so one should lower the limit for t? */
787 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
788 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
789 other_last = strpos = s = t + 1;
794 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
795 PL_colors[0], PL_colors[1]));
799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
800 PL_colors[0], PL_colors[1]));
804 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
807 /* The found string does not prohibit matching at strpos,
808 - no optimization of calling REx engine can be performed,
809 unless it was an MBOL and we are not after MBOL,
810 or a future STCLASS check will fail this. */
812 /* Even in this situation we may use MBOL flag if strpos is offset
813 wrt the start of the string. */
814 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
815 && (strpos != strbeg) && strpos[-1] != '\n'
816 /* May be due to an implicit anchor of m{.*foo} */
817 && !(prog->reganch & ROPT_IMPLICIT))
822 DEBUG_EXECUTE_r( if (ml_anch)
823 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
824 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
827 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
829 prog->check_utf8 /* Could be deleted already */
830 && --BmUSEFUL(prog->check_utf8) < 0
831 && (prog->check_utf8 == prog->float_utf8)
833 prog->check_substr /* Could be deleted already */
834 && --BmUSEFUL(prog->check_substr) < 0
835 && (prog->check_substr == prog->float_substr)
838 /* If flags & SOMETHING - do not do it many times on the same match */
839 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
840 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
841 if (do_utf8 ? prog->check_substr : prog->check_utf8)
842 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
843 prog->check_substr = prog->check_utf8 = NULL; /* disable */
844 prog->float_substr = prog->float_utf8 = NULL; /* clear */
845 check = NULL; /* abort */
847 /* XXXX This is a remnant of the old implementation. It
848 looks wasteful, since now INTUIT can use many
850 prog->reganch &= ~RE_USE_INTUIT;
857 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
858 /* trie stclasses are too expensive to use here, we are better off to
859 leave it to regmatch itself */
860 if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) {
861 /* minlen == 0 is possible if regstclass is \b or \B,
862 and the fixed substr is ''$.
863 Since minlen is already taken into account, s+1 is before strend;
864 accidentally, minlen >= 1 guaranties no false positives at s + 1
865 even for \b or \B. But (minlen? 1 : 0) below assumes that
866 regstclass does not come from lookahead... */
867 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
868 This leaves EXACTF only, which is dealt with in find_byclass(). */
869 const U8* const str = (U8*)STRING(prog->regstclass);
870 const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT
871 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
874 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
875 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
876 else if (prog->float_substr || prog->float_utf8)
877 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
881 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n",
882 (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg));
885 s = find_byclass(prog, prog->regstclass, s, endpos, NULL);
888 const char *what = NULL;
890 if (endpos == strend) {
891 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
892 "Could not match STCLASS...\n") );
895 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
896 "This position contradicts STCLASS...\n") );
897 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
899 /* Contradict one of substrings */
900 if (prog->anchored_substr || prog->anchored_utf8) {
901 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
902 DEBUG_EXECUTE_r( what = "anchored" );
904 s = HOP3c(t, 1, strend);
905 if (s + start_shift + end_shift > strend) {
906 /* XXXX Should be taken into account earlier? */
907 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
908 "Could not match STCLASS...\n") );
913 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
914 "Looking for %s substr starting at offset %ld...\n",
915 what, (long)(s + start_shift - i_strpos)) );
918 /* Have both, check_string is floating */
919 if (t + start_shift >= check_at) /* Contradicts floating=check */
920 goto retry_floating_check;
921 /* Recheck anchored substring, but not floating... */
925 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
926 "Looking for anchored substr starting at offset %ld...\n",
927 (long)(other_last - i_strpos)) );
928 goto do_other_anchored;
930 /* Another way we could have checked stclass at the
931 current position only: */
936 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
937 "Looking for /%s^%s/m starting at offset %ld...\n",
938 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
941 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
943 /* Check is floating subtring. */
944 retry_floating_check:
945 t = check_at - start_shift;
946 DEBUG_EXECUTE_r( what = "floating" );
947 goto hop_and_restart;
950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
951 "By STCLASS: moving %ld --> %ld\n",
952 (long)(t - i_strpos), (long)(s - i_strpos))
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
957 "Does not contradict STCLASS...\n");
962 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
963 PL_colors[4], (check ? "Guessed" : "Giving up"),
964 PL_colors[5], (long)(s - i_strpos)) );
967 fail_finish: /* Substring not found */
968 if (prog->check_substr || prog->check_utf8) /* could be removed already */
969 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
972 PL_colors[4], PL_colors[5]));
978 #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \
979 foldlen, foldbuf, uniflags) STMT_START { \
980 switch (trie_type) { \
981 case trie_utf8_fold: \
983 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
988 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
989 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
990 foldlen -= UNISKIP( uvc ); \
991 uscan = foldbuf + UNISKIP( uvc ); \
995 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1003 charid = trie->charmap[ uvc ]; \
1007 if (trie->widecharmap) { \
1008 SV** const svpp = hv_fetch(trie->widecharmap, \
1009 (char*)&uvc, sizeof(UV), 0); \
1011 charid = (U16)SvIV(*svpp); \
1016 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1019 ibcmp_utf8(s, NULL, 0, do_utf8, \
1020 m, NULL, ln, (bool)UTF)) \
1021 && (!reginfo || regtry(reginfo, s)) ) \
1024 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1025 uvchr_to_utf8(tmpbuf, c); \
1026 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1028 && (f == c1 || f == c2) \
1029 && (ln == foldlen || \
1030 !ibcmp_utf8((char *) foldbuf, \
1031 NULL, foldlen, do_utf8, \
1033 NULL, ln, (bool)UTF)) \
1034 && (!reginfo || regtry(reginfo, s)) ) \
1039 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1043 && (ln == 1 || !(OP(c) == EXACTF \
1045 : ibcmp_locale(s, m, ln))) \
1046 && (!reginfo || regtry(reginfo, s)) ) \
1052 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1054 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1060 #define REXEC_FBC_SCAN(CoDe) \
1062 while (s < strend) { \
1068 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1069 REXEC_FBC_UTF8_SCAN( \
1071 if (tmp && (!reginfo || regtry(reginfo, s))) \
1080 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1083 if (tmp && (!reginfo || regtry(reginfo, s))) \
1092 #define REXEC_FBC_TRYIT \
1093 if ((!reginfo || regtry(reginfo, s))) \
1096 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1099 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1102 REXEC_FBC_CLASS_SCAN(CoNd); \
1106 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1107 PL_reg_flags |= RF_tainted; \
1109 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1112 REXEC_FBC_CLASS_SCAN(CoNd); \
1116 #define DUMP_EXEC_POS(li,s,doutf8) \
1117 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1119 /* We know what class REx starts with. Try to find this position... */
1120 /* if reginfo is NULL, its a dryrun */
1121 /* annoyingly all the vars in this routine have different names from their counterparts
1122 in regmatch. /grrr */
1125 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1126 const char *strend, const regmatch_info *reginfo)
1129 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
1133 register STRLEN uskip;
1137 register I32 tmp = 1; /* Scratch variable? */
1138 register const bool do_utf8 = PL_reg_match_utf8;
1140 /* We know what class it must start with. */
1144 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1145 !UTF8_IS_INVARIANT((U8)s[0]) ?
1146 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1147 REGINCLASS(prog, c, (U8*)s));
1150 while (s < strend) {
1153 if (REGINCLASS(prog, c, (U8*)s) ||
1154 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1155 /* The assignment of 2 is intentional:
1156 * for the folded sharp s, the skip is 2. */
1157 (skip = SHARP_S_SKIP))) {
1158 if (tmp && (!reginfo || regtry(reginfo, s)))
1171 if (tmp && (!reginfo || regtry(reginfo, s)))
1179 ln = STR_LEN(c); /* length to match in octets/bytes */
1180 lnc = (I32) ln; /* length to match in characters */
1182 STRLEN ulen1, ulen2;
1184 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1185 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1186 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1188 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1189 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1191 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1193 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1196 while (sm < ((U8 *) m + ln)) {
1211 c2 = PL_fold_locale[c1];
1213 e = HOP3c(strend, -((I32)lnc), s);
1215 if (!reginfo && e < s)
1216 e = s; /* Due to minlen logic of intuit() */
1218 /* The idea in the EXACTF* cases is to first find the
1219 * first character of the EXACTF* node and then, if
1220 * necessary, case-insensitively compare the full
1221 * text of the node. The c1 and c2 are the first
1222 * characters (though in Unicode it gets a bit
1223 * more complicated because there are more cases
1224 * than just upper and lower: one needs to use
1225 * the so-called folding case for case-insensitive
1226 * matching (called "loose matching" in Unicode).
1227 * ibcmp_utf8() will do just that. */
1231 U8 tmpbuf [UTF8_MAXBYTES+1];
1232 STRLEN len, foldlen;
1233 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1235 /* Upper and lower of 1st char are equal -
1236 * probably not a "letter". */
1238 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1240 REXEC_FBC_EXACTISH_CHECK(c == c1);
1245 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1248 /* Handle some of the three Greek sigmas cases.
1249 * Note that not all the possible combinations
1250 * are handled here: some of them are handled
1251 * by the standard folding rules, and some of
1252 * them (the character class or ANYOF cases)
1253 * are handled during compiletime in
1254 * regexec.c:S_regclass(). */
1255 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1256 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1257 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1259 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1265 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1267 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1271 PL_reg_flags |= RF_tainted;
1278 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1279 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1281 tmp = ((OP(c) == BOUND ?
1282 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1283 LOAD_UTF8_CHARCLASS_ALNUM();
1284 REXEC_FBC_UTF8_SCAN(
1285 if (tmp == !(OP(c) == BOUND ?
1286 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1287 isALNUM_LC_utf8((U8*)s)))
1295 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1296 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1299 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1305 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, s)))
1309 PL_reg_flags |= RF_tainted;
1316 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1317 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1319 tmp = ((OP(c) == NBOUND ?
1320 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1321 LOAD_UTF8_CHARCLASS_ALNUM();
1322 REXEC_FBC_UTF8_SCAN(
1323 if (tmp == !(OP(c) == NBOUND ?
1324 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1325 isALNUM_LC_utf8((U8*)s)))
1327 else REXEC_FBC_TRYIT;
1331 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1332 tmp = ((OP(c) == NBOUND ?
1333 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1336 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1338 else REXEC_FBC_TRYIT;
1341 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, 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_ALNUM(),
1358 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1362 REXEC_FBC_CSCAN_TAINT(
1363 !isALNUM_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_SPACE(),
1380 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1384 REXEC_FBC_CSCAN_TAINT(
1385 !(*s == ' ' || isSPACE_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),
1400 REXEC_FBC_CSCAN_PRELOAD(
1401 LOAD_UTF8_CHARCLASS_DIGIT(),
1402 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1406 REXEC_FBC_CSCAN_TAINT(
1407 !isDIGIT_LC_utf8((U8*)s),
1413 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1414 trie_type = do_utf8 ?
1415 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1417 /* what trie are we using right now */
1419 = (reg_ac_data*)prog->data->data[ ARG( c ) ];
1420 reg_trie_data *trie=aho->trie;
1422 const char *last_start = strend - trie->minlen;
1424 const char *real_start = s;
1426 STRLEN maxlen = trie->maxlen;
1428 U8 **points; /* map of where we were in the input string
1429 when reading a given char. For ASCII this
1430 is unnecessary overhead as the relationship
1431 is always 1:1, but for unicode, especially
1432 case folded unicode this is not true. */
1433 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1437 GET_RE_DEBUG_FLAGS_DECL;
1439 /* We can't just allocate points here. We need to wrap it in
1440 * an SV so it gets freed properly if there is a croak while
1441 * running the match */
1444 sv_points=newSV(maxlen * sizeof(U8 *));
1445 SvCUR_set(sv_points,
1446 maxlen * sizeof(U8 *));
1447 SvPOK_on(sv_points);
1448 sv_2mortal(sv_points);
1449 points=(U8**)SvPV_nolen(sv_points );
1450 if ( trie_type != trie_utf8_fold
1451 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1454 bitmap=(U8*)trie->bitmap;
1456 bitmap=(U8*)ANYOF_BITMAP(c);
1458 /* this is the Aho-Corasick algorithm modified a touch
1459 to include special handling for long "unknown char"
1460 sequences. The basic idea being that we use AC as long
1461 as we are dealing with a possible matching char, when
1462 we encounter an unknown char (and we have not encountered
1463 an accepting state) we scan forward until we find a legal
1465 AC matching is basically that of trie matching, except
1466 that when we encounter a failing transition, we fall back
1467 to the current states "fail state", and try the current char
1468 again, a process we repeat until we reach the root state,
1469 state 1, or a legal transition. If we fail on the root state
1470 then we can either terminate if we have reached an accepting
1471 state previously, or restart the entire process from the beginning
1475 while (s <= last_start) {
1476 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1484 U8 *uscan = (U8*)NULL;
1485 U8 *leftmost = NULL;
1487 U32 accepted_word= 0;
1491 while ( state && uc <= (U8*)strend ) {
1493 U32 word = aho->states[ state ].wordnum;
1497 DEBUG_TRIE_EXECUTE_r(
1498 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1499 dump_exec_pos( (char *)uc, c, strend, real_start,
1500 (char *)uc, do_utf8 );
1501 PerlIO_printf( Perl_debug_log,
1502 " Scanning for legal start char...\n");
1505 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1510 if (uc >(U8*)last_start) break;
1514 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1515 if (!leftmost || lpos < leftmost) {
1516 DEBUG_r(accepted_word=word);
1522 points[pointpos++ % maxlen]= uc;
1523 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
1524 uvc, charid, foldlen, foldbuf, uniflags);
1525 DEBUG_TRIE_EXECUTE_r({
1526 dump_exec_pos( (char *)uc, c, strend, real_start,
1528 PerlIO_printf(Perl_debug_log,
1529 " Charid:%3u CP:%4"UVxf" ",
1535 word = aho->states[ state ].wordnum;
1537 base = aho->states[ state ].trans.base;
1539 DEBUG_TRIE_EXECUTE_r({
1541 dump_exec_pos( (char *)uc, c, strend, real_start,
1543 PerlIO_printf( Perl_debug_log,
1544 "%sState: %4"UVxf", word=%"UVxf,
1545 failed ? " Fail transition to " : "",
1546 (UV)state, (UV)word);
1551 (base + charid > trie->uniquecharcount )
1552 && (base + charid - 1 - trie->uniquecharcount
1554 && trie->trans[base + charid - 1 -
1555 trie->uniquecharcount].check == state
1556 && (tmp=trie->trans[base + charid - 1 -
1557 trie->uniquecharcount ].next))
1559 DEBUG_TRIE_EXECUTE_r(
1560 PerlIO_printf( Perl_debug_log," - legal\n"));
1565 DEBUG_TRIE_EXECUTE_r(
1566 PerlIO_printf( Perl_debug_log," - fail\n"));
1568 state = aho->fail[state];
1572 /* we must be accepting here */
1573 DEBUG_TRIE_EXECUTE_r(
1574 PerlIO_printf( Perl_debug_log," - accepting\n"));
1583 if (!state) state = 1;
1586 if ( aho->states[ state ].wordnum ) {
1587 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1588 if (!leftmost || lpos < leftmost) {
1589 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1594 s = (char*)leftmost;
1595 DEBUG_TRIE_EXECUTE_r({
1597 Perl_debug_log,"Matches word #%"UVxf" at position %d. Trying full pattern...\n",
1598 (UV)accepted_word, s - real_start
1601 if (!reginfo || regtry(reginfo, s)) {
1607 DEBUG_TRIE_EXECUTE_r({
1608 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1611 DEBUG_TRIE_EXECUTE_r(
1612 PerlIO_printf( Perl_debug_log,"No match.\n"));
1621 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1630 - regexec_flags - match a regexp against a string
1633 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1634 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1635 /* strend: pointer to null at end of string */
1636 /* strbeg: real beginning of string */
1637 /* minend: end of match must be >=minend after stringarg. */
1638 /* data: May be used for some additional optimizations. */
1639 /* nosave: For optimizations. */
1643 register regnode *c;
1644 register char *startpos = stringarg;
1645 I32 minlen; /* must match at least this many chars */
1646 I32 dontbother = 0; /* how many characters not to try at end */
1647 I32 end_shift = 0; /* Same for the end. */ /* CC */
1648 I32 scream_pos = -1; /* Internal iterator of scream. */
1649 char *scream_olds = NULL;
1650 SV* const oreplsv = GvSV(PL_replgv);
1651 const bool do_utf8 = (bool)DO_UTF8(sv);
1654 regmatch_info reginfo; /* create some info to pass to regtry etc */
1656 GET_RE_DEBUG_FLAGS_DECL;
1658 PERL_UNUSED_ARG(data);
1660 /* Be paranoid... */
1661 if (prog == NULL || startpos == NULL) {
1662 Perl_croak(aTHX_ "NULL regexp parameter");
1666 multiline = prog->reganch & PMf_MULTILINE;
1667 reginfo.prog = prog;
1669 RX_MATCH_UTF8_set(prog, do_utf8);
1671 debug_start_match(prog, do_utf8, startpos, strend,
1675 minlen = prog->minlen;
1677 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1678 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1679 "String too short [regexec_flags]...\n"));
1684 /* Check validity of program. */
1685 if (UCHARAT(prog->program) != REG_MAGIC) {
1686 Perl_croak(aTHX_ "corrupted regexp program");
1690 PL_reg_eval_set = 0;
1693 if (prog->reganch & ROPT_UTF8)
1694 PL_reg_flags |= RF_utf8;
1696 /* Mark beginning of line for ^ and lookbehind. */
1697 reginfo.bol = startpos; /* XXX not used ??? */
1701 /* Mark end of line for $ (and such) */
1704 /* see how far we have to get to not match where we matched before */
1705 reginfo.till = startpos+minend;
1707 /* If there is a "must appear" string, look for it. */
1710 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to set reginfo->ganch */
1713 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1714 reginfo.ganch = startpos;
1715 else if (sv && SvTYPE(sv) >= SVt_PVMG
1717 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1718 && mg->mg_len >= 0) {
1719 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1720 if (prog->reganch & ROPT_ANCH_GPOS) {
1721 if (s > reginfo.ganch)
1726 else /* pos() not defined */
1727 reginfo.ganch = strbeg;
1730 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1731 re_scream_pos_data d;
1733 d.scream_olds = &scream_olds;
1734 d.scream_pos = &scream_pos;
1735 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1737 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1738 goto phooey; /* not present */
1744 /* Simplest case: anchored match need be tried only once. */
1745 /* [unless only anchor is BOL and multiline is set] */
1746 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1747 if (s == startpos && regtry(®info, startpos))
1749 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1750 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1755 dontbother = minlen - 1;
1756 end = HOP3c(strend, -dontbother, strbeg) - 1;
1757 /* for multiline we only have to try after newlines */
1758 if (prog->check_substr || prog->check_utf8) {
1762 if (regtry(®info, s))
1767 if (prog->reganch & RE_USE_INTUIT) {
1768 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1779 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1780 if (regtry(®info, s))
1787 } else if (ROPT_GPOS_CHECK == (prog->reganch & ROPT_GPOS_CHECK))
1789 /* the warning about reginfo.ganch being used without intialization
1790 is bogus -- we set it above, when prog->reganch & ROPT_GPOS_SEEN
1791 and we only enter this block when the same bit is set. */
1792 if (regtry(®info, reginfo.ganch))
1797 /* Messy cases: unanchored match. */
1798 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1799 /* we have /x+whatever/ */
1800 /* it must be a one character string (XXXX Except UTF?) */
1805 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1806 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1807 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1812 DEBUG_EXECUTE_r( did_match = 1 );
1813 if (regtry(®info, s)) goto got_it;
1815 while (s < strend && *s == ch)
1823 DEBUG_EXECUTE_r( did_match = 1 );
1824 if (regtry(®info, s)) goto got_it;
1826 while (s < strend && *s == ch)
1831 DEBUG_EXECUTE_r(if (!did_match)
1832 PerlIO_printf(Perl_debug_log,
1833 "Did not find anchored character...\n")
1836 else if (prog->anchored_substr != NULL
1837 || prog->anchored_utf8 != NULL
1838 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1839 && prog->float_max_offset < strend - s)) {
1844 char *last1; /* Last position checked before */
1848 if (prog->anchored_substr || prog->anchored_utf8) {
1849 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1850 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1851 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1852 back_max = back_min = prog->anchored_offset;
1854 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1855 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1856 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1857 back_max = prog->float_max_offset;
1858 back_min = prog->float_min_offset;
1862 if (must == &PL_sv_undef)
1863 /* could not downgrade utf8 check substring, so must fail */
1869 last = HOP3c(strend, /* Cannot start after this */
1870 -(I32)(CHR_SVLEN(must)
1871 - (SvTAIL(must) != 0) + back_min), strbeg);
1874 last1 = HOPc(s, -1);
1876 last1 = s - 1; /* bogus */
1878 /* XXXX check_substr already used to find "s", can optimize if
1879 check_substr==must. */
1881 dontbother = end_shift;
1882 strend = HOPc(strend, -dontbother);
1883 while ( (s <= last) &&
1884 ((flags & REXEC_SCREAM)
1885 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1886 end_shift, &scream_pos, 0))
1887 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1888 (unsigned char*)strend, must,
1889 multiline ? FBMrf_MULTILINE : 0))) ) {
1890 /* we may be pointing at the wrong string */
1891 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1892 s = strbeg + (s - SvPVX_const(sv));
1893 DEBUG_EXECUTE_r( did_match = 1 );
1894 if (HOPc(s, -back_max) > last1) {
1895 last1 = HOPc(s, -back_min);
1896 s = HOPc(s, -back_max);
1899 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1901 last1 = HOPc(s, -back_min);
1905 while (s <= last1) {
1906 if (regtry(®info, s))
1912 while (s <= last1) {
1913 if (regtry(®info, s))
1919 DEBUG_EXECUTE_r(if (!did_match) {
1920 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1921 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1922 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1923 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1924 ? "anchored" : "floating"),
1925 quoted, RE_SV_TAIL(must));
1929 else if ( (c = prog->regstclass) ) {
1931 const OPCODE op = OP(prog->regstclass);
1932 /* don't bother with what can't match */
1933 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1934 strend = HOPc(strend, -(minlen - 1));
1937 SV * const prop = sv_newmortal();
1938 regprop(prog, prop, c);
1940 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1942 PerlIO_printf(Perl_debug_log,
1943 "Matching stclass %.*s against %s (%d chars)\n",
1944 (int)SvCUR(prop), SvPVX_const(prop),
1945 quoted, (int)(strend - s));
1948 if (find_byclass(prog, c, s, strend, ®info))
1950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
1954 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1959 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1960 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1961 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1963 if (flags & REXEC_SCREAM) {
1964 last = screaminstr(sv, float_real, s - strbeg,
1965 end_shift, &scream_pos, 1); /* last one */
1967 last = scream_olds; /* Only one occurrence. */
1968 /* we may be pointing at the wrong string */
1969 else if (RX_MATCH_COPIED(prog))
1970 s = strbeg + (s - SvPVX_const(sv));
1974 const char * const little = SvPV_const(float_real, len);
1976 if (SvTAIL(float_real)) {
1977 if (memEQ(strend - len + 1, little, len - 1))
1978 last = strend - len + 1;
1979 else if (!multiline)
1980 last = memEQ(strend - len, little, len)
1981 ? strend - len : NULL;
1987 last = rninstr(s, strend, little, little + len);
1989 last = strend; /* matching "$" */
1994 PerlIO_printf(Perl_debug_log,
1995 "%sCan't trim the tail, match fails (should not happen)%s\n",
1996 PL_colors[4], PL_colors[5]));
1997 goto phooey; /* Should not happen! */
1999 dontbother = strend - last + prog->float_min_offset;
2001 if (minlen && (dontbother < minlen))
2002 dontbother = minlen - 1;
2003 strend -= dontbother; /* this one's always in bytes! */
2004 /* We don't know much -- general case. */
2007 if (regtry(®info, s))
2016 if (regtry(®info, s))
2018 } while (s++ < strend);
2026 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2028 if (PL_reg_eval_set) {
2029 /* Preserve the current value of $^R */
2030 if (oreplsv != GvSV(PL_replgv))
2031 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2032 restored, the value remains
2034 restore_pos(aTHX_ prog);
2036 if (prog->paren_names)
2037 (void)hv_iterinit(prog->paren_names);
2039 /* make sure $`, $&, $', and $digit will work later */
2040 if ( !(flags & REXEC_NOT_FIRST) ) {
2041 RX_MATCH_COPY_FREE(prog);
2042 if (flags & REXEC_COPY_STR) {
2043 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2044 #ifdef PERL_OLD_COPY_ON_WRITE
2046 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2048 PerlIO_printf(Perl_debug_log,
2049 "Copy on write: regexp capture, type %d\n",
2052 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2053 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2054 assert (SvPOKp(prog->saved_copy));
2058 RX_MATCH_COPIED_on(prog);
2059 s = savepvn(strbeg, i);
2065 prog->subbeg = strbeg;
2066 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2073 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2074 PL_colors[4], PL_colors[5]));
2075 if (PL_reg_eval_set)
2076 restore_pos(aTHX_ prog);
2082 - regtry - try match at specific point
2084 STATIC I32 /* 0 failure, 1 success */
2085 S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos)
2091 regexp *prog = reginfo->prog;
2092 GET_RE_DEBUG_FLAGS_DECL;
2094 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2097 PL_reg_eval_set = RS_init;
2098 DEBUG_EXECUTE_r(DEBUG_s(
2099 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2100 (IV)(PL_stack_sp - PL_stack_base));
2103 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2104 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2106 /* Apparently this is not needed, judging by wantarray. */
2107 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2108 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2111 /* Make $_ available to executed code. */
2112 if (reginfo->sv != DEFSV) {
2114 DEFSV = reginfo->sv;
2117 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2118 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2119 /* prepare for quick setting of pos */
2120 #ifdef PERL_OLD_COPY_ON_WRITE
2122 sv_force_normal_flags(sv, 0);
2124 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2125 &PL_vtbl_mglob, NULL, 0);
2129 PL_reg_oldpos = mg->mg_len;
2130 SAVEDESTRUCTOR_X(restore_pos, prog);
2132 if (!PL_reg_curpm) {
2133 Newxz(PL_reg_curpm, 1, PMOP);
2136 SV* const repointer = newSViv(0);
2137 /* so we know which PL_regex_padav element is PL_reg_curpm */
2138 SvFLAGS(repointer) |= SVf_BREAK;
2139 av_push(PL_regex_padav,repointer);
2140 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2141 PL_regex_pad = AvARRAY(PL_regex_padav);
2145 PM_SETRE(PL_reg_curpm, prog);
2146 PL_reg_oldcurpm = PL_curpm;
2147 PL_curpm = PL_reg_curpm;
2148 if (RX_MATCH_COPIED(prog)) {
2149 /* Here is a serious problem: we cannot rewrite subbeg,
2150 since it may be needed if this match fails. Thus
2151 $` inside (?{}) could fail... */
2152 PL_reg_oldsaved = prog->subbeg;
2153 PL_reg_oldsavedlen = prog->sublen;
2154 #ifdef PERL_OLD_COPY_ON_WRITE
2155 PL_nrs = prog->saved_copy;
2157 RX_MATCH_COPIED_off(prog);
2160 PL_reg_oldsaved = NULL;
2161 prog->subbeg = PL_bostr;
2162 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2164 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2165 prog->startp[0] = startpos - PL_bostr;
2166 PL_reginput = startpos;
2167 PL_reglastparen = &prog->lastparen;
2168 PL_reglastcloseparen = &prog->lastcloseparen;
2169 prog->lastparen = 0;
2170 prog->lastcloseparen = 0;
2172 PL_regstartp = prog->startp;
2173 PL_regendp = prog->endp;
2174 if (PL_reg_start_tmpl <= prog->nparens) {
2175 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2176 if(PL_reg_start_tmp)
2177 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2179 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2182 /* XXXX What this code is doing here?!!! There should be no need
2183 to do this again and again, PL_reglastparen should take care of
2186 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2187 * Actually, the code in regcppop() (which Ilya may be meaning by
2188 * PL_reglastparen), is not needed at all by the test suite
2189 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2190 * enough, for building DynaLoader, or otherwise this
2191 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2192 * will happen. Meanwhile, this code *is* needed for the
2193 * above-mentioned test suite tests to succeed. The common theme
2194 * on those tests seems to be returning null fields from matches.
2199 if (prog->nparens) {
2201 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2208 if (regmatch(reginfo, prog->program + 1)) {
2209 prog->endp[0] = PL_reginput - PL_bostr;
2212 REGCP_UNWIND(lastcp);
2217 #define sayYES goto yes
2218 #define sayNO goto no
2219 #define sayNO_SILENT goto no_silent
2221 /* we dont use STMT_START/END here because it leads to
2222 "unreachable code" warnings, which are bogus, but distracting. */
2223 #define CACHEsayNO \
2224 if (ST.cache_mask) \
2225 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2228 /* this is used to determine how far from the left messages like
2229 'failed...' are printed. It should be set such that messages
2230 are inline with the regop output that created them.
2232 #define REPORT_CODE_OFF 32
2235 /* Make sure there is a test for this +1 options in re_tests */
2236 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2238 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2239 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2241 #define SLAB_FIRST(s) (&(s)->states[0])
2242 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2244 /* grab a new slab and return the first slot in it */
2246 STATIC regmatch_state *
2249 #if PERL_VERSION < 9
2252 regmatch_slab *s = PL_regmatch_slab->next;
2254 Newx(s, 1, regmatch_slab);
2255 s->prev = PL_regmatch_slab;
2257 PL_regmatch_slab->next = s;
2259 PL_regmatch_slab = s;
2260 return SLAB_FIRST(s);
2264 /* push a new state then goto it */
2266 #define PUSH_STATE_GOTO(state, node) \
2268 st->resume_state = state; \
2271 /* push a new state with success backtracking, then goto it */
2273 #define PUSH_YES_STATE_GOTO(state, node) \
2275 st->resume_state = state; \
2276 goto push_yes_state;
2282 regmatch() - main matching routine
2284 This is basically one big switch statement in a loop. We execute an op,
2285 set 'next' to point the next op, and continue. If we come to a point which
2286 we may need to backtrack to on failure such as (A|B|C), we push a
2287 backtrack state onto the backtrack stack. On failure, we pop the top
2288 state, and re-enter the loop at the state indicated. If there are no more
2289 states to pop, we return failure.
2291 Sometimes we also need to backtrack on success; for example /A+/, where
2292 after successfully matching one A, we need to go back and try to
2293 match another one; similarly for lookahead assertions: if the assertion
2294 completes successfully, we backtrack to the state just before the assertion
2295 and then carry on. In these cases, the pushed state is marked as
2296 'backtrack on success too'. This marking is in fact done by a chain of
2297 pointers, each pointing to the previous 'yes' state. On success, we pop to
2298 the nearest yes state, discarding any intermediate failure-only states.
2299 Sometimes a yes state is pushed just to force some cleanup code to be
2300 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2301 it to free the inner regex.
2303 Note that failure backtracking rewinds the cursor position, while
2304 success backtracking leaves it alone.
2306 A pattern is complete when the END op is executed, while a subpattern
2307 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2308 ops trigger the "pop to last yes state if any, otherwise return true"
2311 A common convention in this function is to use A and B to refer to the two
2312 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2313 the subpattern to be matched possibly multiple times, while B is the entire
2314 rest of the pattern. Variable and state names reflect this convention.
2316 The states in the main switch are the union of ops and failure/success of
2317 substates associated with with that op. For example, IFMATCH is the op
2318 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2319 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2320 successfully matched A and IFMATCH_A_fail is a state saying that we have
2321 just failed to match A. Resume states always come in pairs. The backtrack
2322 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2323 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2324 on success or failure.
2326 The struct that holds a backtracking state is actually a big union, with
2327 one variant for each major type of op. The variable st points to the
2328 top-most backtrack struct. To make the code clearer, within each
2329 block of code we #define ST to alias the relevant union.
2331 Here's a concrete example of a (vastly oversimplified) IFMATCH
2337 #define ST st->u.ifmatch
2339 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2340 ST.foo = ...; // some state we wish to save
2342 // push a yes backtrack state with a resume value of
2343 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2345 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2348 case IFMATCH_A: // we have successfully executed A; now continue with B
2350 bar = ST.foo; // do something with the preserved value
2353 case IFMATCH_A_fail: // A failed, so the assertion failed
2354 ...; // do some housekeeping, then ...
2355 sayNO; // propagate the failure
2362 For any old-timers reading this who are familiar with the old recursive
2363 approach, the code above is equivalent to:
2365 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2374 ...; // do some housekeeping, then ...
2375 sayNO; // propagate the failure
2378 The topmost backtrack state, pointed to by st, is usually free. If you
2379 want to claim it, populate any ST.foo fields in it with values you wish to
2380 save, then do one of
2382 PUSH_STATE_GOTO(resume_state, node);
2383 PUSH_YES_STATE_GOTO(resume_state, node);
2385 which sets that backtrack state's resume value to 'resume_state', pushes a
2386 new free entry to the top of the backtrack stack, then goes to 'node'.
2387 On backtracking, the free slot is popped, and the saved state becomes the
2388 new free state. An ST.foo field in this new top state can be temporarily
2389 accessed to retrieve values, but once the main loop is re-entered, it
2390 becomes available for reuse.
2392 Note that the depth of the backtrack stack constantly increases during the
2393 left-to-right execution of the pattern, rather than going up and down with
2394 the pattern nesting. For example the stack is at its maximum at Z at the
2395 end of the pattern, rather than at X in the following:
2397 /(((X)+)+)+....(Y)+....Z/
2399 The only exceptions to this are lookahead/behind assertions and the cut,
2400 (?>A), which pop all the backtrack states associated with A before
2403 Bascktrack state structs are allocated in slabs of about 4K in size.
2404 PL_regmatch_state and st always point to the currently active state,
2405 and PL_regmatch_slab points to the slab currently containing
2406 PL_regmatch_state. The first time regmatch() is called, the first slab is
2407 allocated, and is never freed until interpreter destruction. When the slab
2408 is full, a new one is allocated and chained to the end. At exit from
2409 regmatch(), slabs allocated since entry are freed.
2414 #define DEBUG_STATE_pp(pp) \
2416 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2417 PerlIO_printf(Perl_debug_log, \
2420 reg_name[st->resume_state] ); \
2424 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2429 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2430 const char *start, const char *end, const char *blurb)
2432 const bool utf8_pat= prog->reganch & ROPT_UTF8 ? 1 : 0;
2436 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2437 prog->precomp, prog->prelen, 60);
2439 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2440 start, end - start, 60);
2442 PerlIO_printf(Perl_debug_log,
2443 "%s%s REx%s %s against %s\n",
2444 PL_colors[4], blurb, PL_colors[5], s0, s1);
2446 if (do_utf8||utf8_pat)
2447 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2448 utf8_pat ? "pattern" : "",
2449 utf8_pat && do_utf8 ? " and " : "",
2450 do_utf8 ? "string" : ""
2456 S_dump_exec_pos(pTHX_ const char *locinput,
2457 const regnode *scan,
2458 const char *loc_regeol,
2459 const char *loc_bostr,
2460 const char *loc_reg_starttry,
2463 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2464 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2465 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2466 /* The part of the string before starttry has one color
2467 (pref0_len chars), between starttry and current
2468 position another one (pref_len - pref0_len chars),
2469 after the current position the third one.
2470 We assume that pref0_len <= pref_len, otherwise we
2471 decrease pref0_len. */
2472 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2473 ? (5 + taill) - l : locinput - loc_bostr;
2476 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2478 pref0_len = pref_len - (locinput - loc_reg_starttry);
2479 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2480 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2481 ? (5 + taill) - pref_len : loc_regeol - locinput);
2482 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2486 if (pref0_len > pref_len)
2487 pref0_len = pref_len;
2489 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2491 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2492 (locinput - pref_len),pref0_len, 60, 4, 5);
2494 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2495 (locinput - pref_len + pref0_len),
2496 pref_len - pref0_len, 60, 2, 3);
2498 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2499 locinput, loc_regeol - locinput, 10, 0, 1);
2501 const STRLEN tlen=len0+len1+len2;
2502 PerlIO_printf(Perl_debug_log,
2503 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2504 (IV)(locinput - loc_bostr),
2507 (docolor ? "" : "> <"),
2509 (int)(tlen > 19 ? 0 : 19 - tlen),
2516 /* reg_check_named_buff_matched()
2517 * Checks to see if a named buffer has matched. The data array of
2518 * buffer numbers corresponding to the buffer is expected to reside
2519 * in the regexp->data->data array in the slot stored in the ARG() of
2520 * node involved. Note that this routine doesn't actually care about the
2521 * name, that information is not preserved from compilation to execution.
2522 * Returns the index of the leftmost defined buffer with the given name
2523 * or 0 if non of the buffers matched.
2526 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2528 SV *sv_dat=(SV*)rex->data->data[ ARG( scan ) ];
2529 I32 *nums=(I32*)SvPVX(sv_dat);
2530 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2531 if ((I32)*PL_reglastparen >= nums[n] &&
2532 PL_regendp[nums[n]] != -1)
2540 STATIC I32 /* 0 failure, 1 success */
2541 S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog)
2543 #if PERL_VERSION < 9
2547 register const bool do_utf8 = PL_reg_match_utf8;
2548 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2550 regexp *rex = reginfo->prog;
2552 regmatch_slab *orig_slab;
2553 regmatch_state *orig_state;
2555 /* the current state. This is a cached copy of PL_regmatch_state */
2556 register regmatch_state *st;
2558 /* cache heavy used fields of st in registers */
2559 register regnode *scan;
2560 register regnode *next;
2561 register I32 n = 0; /* general value; init to avoid compiler warning */
2562 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2563 register char *locinput = PL_reginput;
2564 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2566 bool result = 0; /* return value of S_regmatch */
2567 int depth = 0; /* depth of backtrack stack */
2568 int nochange_depth = 0; /* depth of GOSUB recursion with nochange*/
2569 regmatch_state *yes_state = NULL; /* state to pop to on success of
2571 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2572 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2575 /* these three flags are set by various ops to signal information to
2576 * the very next op. They have a useful lifetime of exactly one loop
2577 * iteration, and are not preserved or restored by state pushes/pops
2579 bool sw = 0; /* the condition value in (?(cond)a|b) */
2580 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2581 int logical = 0; /* the following EVAL is:
2585 or the following IFMATCH/UNLESSM is:
2586 false: plain (?=foo)
2587 true: used as a condition: (?(?=foo))
2591 GET_RE_DEBUG_FLAGS_DECL;
2594 /* on first ever call to regmatch, allocate first slab */
2595 if (!PL_regmatch_slab) {
2596 Newx(PL_regmatch_slab, 1, regmatch_slab);
2597 PL_regmatch_slab->prev = NULL;
2598 PL_regmatch_slab->next = NULL;
2599 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2602 /* remember current high-water mark for exit */
2603 /* XXX this should be done with SAVE* instead */
2604 orig_slab = PL_regmatch_slab;
2605 orig_state = PL_regmatch_state;
2607 /* grab next free state slot */
2608 st = ++PL_regmatch_state;
2609 if (st > SLAB_LAST(PL_regmatch_slab))
2610 st = PL_regmatch_state = S_push_slab(aTHX);
2612 /* Note that nextchr is a byte even in UTF */
2613 nextchr = UCHARAT(locinput);
2615 while (scan != NULL) {
2618 SV * const prop = sv_newmortal();
2619 regnode *rnext=regnext(scan);
2620 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2621 regprop(rex, prop, scan);
2623 PerlIO_printf(Perl_debug_log,
2624 "%3"IVdf":%*s%s(%"IVdf")\n",
2625 (IV)(scan - rex->program), depth*2, "",
2627 (PL_regkind[OP(scan)] == END || !rnext) ?
2628 0 : (IV)(rnext - rex->program));
2631 next = scan + NEXT_OFF(scan);
2634 state_num = OP(scan);
2637 switch (state_num) {
2639 if (locinput == PL_bostr)
2641 /* reginfo->till = reginfo->bol; */
2646 if (locinput == PL_bostr ||
2647 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2653 if (locinput == PL_bostr)
2657 if (locinput == reginfo->ganch)
2663 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2668 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2670 if (PL_regeol - locinput > 1)
2674 if (PL_regeol != locinput)
2678 if (!nextchr && locinput >= PL_regeol)
2681 locinput += PL_utf8skip[nextchr];
2682 if (locinput > PL_regeol)
2684 nextchr = UCHARAT(locinput);
2687 nextchr = UCHARAT(++locinput);
2690 if (!nextchr && locinput >= PL_regeol)
2692 nextchr = UCHARAT(++locinput);
2695 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2698 locinput += PL_utf8skip[nextchr];
2699 if (locinput > PL_regeol)
2701 nextchr = UCHARAT(locinput);
2704 nextchr = UCHARAT(++locinput);
2708 #define ST st->u.trie
2710 /* In this case the charclass data is available inline so
2711 we can fail fast without a lot of extra overhead.
2713 if (scan->flags == EXACT || !do_utf8) {
2714 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2716 PerlIO_printf(Perl_debug_log,
2717 "%*s %sfailed to match trie start class...%s\n",
2718 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2727 /* what type of TRIE am I? (utf8 makes this contextual) */
2728 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2729 trie_type = do_utf8 ?
2730 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2733 /* what trie are we using right now */
2734 reg_trie_data * const trie
2735 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2736 U32 state = trie->startstate;
2738 if (trie->bitmap && trie_type != trie_utf8_fold &&
2739 !TRIE_BITMAP_TEST(trie,*locinput)
2741 if (trie->states[ state ].wordnum) {
2743 PerlIO_printf(Perl_debug_log,
2744 "%*s %smatched empty string...%s\n",
2745 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2750 PerlIO_printf(Perl_debug_log,
2751 "%*s %sfailed to match trie start class...%s\n",
2752 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2759 U8 *uc = ( U8* )locinput;
2763 U8 *uscan = (U8*)NULL;
2765 SV *sv_accept_buff = NULL;
2766 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2768 ST.accepted = 0; /* how many accepting states we have seen */
2770 ST.jump = trie->jump;
2774 traverse the TRIE keeping track of all accepting states
2775 we transition through until we get to a failing node.
2778 while ( state && uc <= (U8*)PL_regeol ) {
2779 U32 base = trie->states[ state ].trans.base;
2782 /* We use charid to hold the wordnum as we don't use it
2783 for charid until after we have done the wordnum logic.
2784 We define an alias just so that the wordnum logic reads
2787 #define got_wordnum charid
2788 got_wordnum = trie->states[ state ].wordnum;
2790 if ( got_wordnum ) {
2791 if ( ! ST.accepted ) {
2794 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2795 sv_accept_buff=newSV(bufflen *
2796 sizeof(reg_trie_accepted) - 1);
2797 SvCUR_set(sv_accept_buff, 0);
2798 SvPOK_on(sv_accept_buff);
2799 sv_2mortal(sv_accept_buff);
2802 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2805 if (ST.accepted >= bufflen) {
2807 ST.accept_buff =(reg_trie_accepted*)
2808 SvGROW(sv_accept_buff,
2809 bufflen * sizeof(reg_trie_accepted));
2811 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2812 + sizeof(reg_trie_accepted));
2815 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2816 ST.accept_buff[ST.accepted].endpos = uc;
2818 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2822 DEBUG_TRIE_EXECUTE_r({
2823 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2824 PerlIO_printf( Perl_debug_log,
2825 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2826 2+depth * 2, "", PL_colors[4],
2827 (UV)state, (UV)ST.accepted );
2831 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
2832 uvc, charid, foldlen, foldbuf, uniflags);
2835 (base + charid > trie->uniquecharcount )
2836 && (base + charid - 1 - trie->uniquecharcount
2838 && trie->trans[base + charid - 1 -
2839 trie->uniquecharcount].check == state)
2841 state = trie->trans[base + charid - 1 -
2842 trie->uniquecharcount ].next;
2853 DEBUG_TRIE_EXECUTE_r(
2854 PerlIO_printf( Perl_debug_log,
2855 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2856 charid, uvc, (UV)state, PL_colors[5] );
2863 PerlIO_printf( Perl_debug_log,
2864 "%*s %sgot %"IVdf" possible matches%s\n",
2865 REPORT_CODE_OFF + depth * 2, "",
2866 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2872 case TRIE_next_fail: /* we failed - try next alterative */
2874 if ( ST.accepted == 1 ) {
2875 /* only one choice left - just continue */
2877 reg_trie_data * const trie
2878 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2879 SV ** const tmp = RX_DEBUG(reginfo->prog)
2880 ? av_fetch( trie->words, ST.accept_buff[ 0 ].wordnum-1, 0 )
2882 PerlIO_printf( Perl_debug_log,
2883 "%*s %sonly one match left: #%d <%s>%s\n",
2884 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2885 ST.accept_buff[ 0 ].wordnum,
2886 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2889 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2890 /* in this case we free tmps/leave before we call regmatch
2891 as we wont be using accept_buff again. */
2894 locinput = PL_reginput;
2895 nextchr = UCHARAT(locinput);
2897 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
2900 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
2902 continue; /* execute rest of RE */
2905 if (!ST.accepted-- ) {
2912 There are at least two accepting states left. Presumably
2913 the number of accepting states is going to be low,
2914 typically two. So we simply scan through to find the one
2915 with lowest wordnum. Once we find it, we swap the last
2916 state into its place and decrement the size. We then try to
2917 match the rest of the pattern at the point where the word
2918 ends. If we succeed, control just continues along the
2919 regex; if we fail we return here to try the next accepting
2926 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
2927 DEBUG_TRIE_EXECUTE_r(
2928 PerlIO_printf( Perl_debug_log,
2929 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2930 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
2931 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
2932 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
2935 if (ST.accept_buff[cur].wordnum <
2936 ST.accept_buff[best].wordnum)
2941 reg_trie_data * const trie
2942 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2943 SV ** const tmp = RX_DEBUG(reginfo->prog)
2944 ? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 )
2946 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
2948 ST.me + ST.jump[ST.accept_buff[best].wordnum];
2949 PerlIO_printf( Perl_debug_log,
2950 "%*s %strying alternation #%d <%s> at node #%d %s\n",
2951 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2952 ST.accept_buff[best].wordnum,
2953 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2954 REG_NODE_NUM(nextop),
2958 if ( best<ST.accepted ) {
2959 reg_trie_accepted tmp = ST.accept_buff[ best ];
2960 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
2961 ST.accept_buff[ ST.accepted ] = tmp;
2964 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
2965 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
2966 PUSH_STATE_GOTO(TRIE_next, ST.B);
2969 PUSH_STATE_GOTO(TRIE_next, ST.me + ST.jump[ST.accept_buff[best].wordnum]);
2979 char *s = STRING(scan);
2981 if (do_utf8 != UTF) {
2982 /* The target and the pattern have differing utf8ness. */
2984 const char * const e = s + ln;
2987 /* The target is utf8, the pattern is not utf8. */
2992 if (NATIVE_TO_UNI(*(U8*)s) !=
2993 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3001 /* The target is not utf8, the pattern is utf8. */
3006 if (NATIVE_TO_UNI(*((U8*)l)) !=
3007 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3015 nextchr = UCHARAT(locinput);
3018 /* The target and the pattern have the same utf8ness. */
3019 /* Inline the first character, for speed. */
3020 if (UCHARAT(s) != nextchr)
3022 if (PL_regeol - locinput < ln)
3024 if (ln > 1 && memNE(s, locinput, ln))
3027 nextchr = UCHARAT(locinput);
3031 PL_reg_flags |= RF_tainted;
3034 char * const s = STRING(scan);
3037 if (do_utf8 || UTF) {
3038 /* Either target or the pattern are utf8. */
3039 const char * const l = locinput;
3040 char *e = PL_regeol;
3042 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3043 l, &e, 0, do_utf8)) {
3044 /* One more case for the sharp s:
3045 * pack("U0U*", 0xDF) =~ /ss/i,
3046 * the 0xC3 0x9F are the UTF-8
3047 * byte sequence for the U+00DF. */
3049 toLOWER(s[0]) == 's' &&
3051 toLOWER(s[1]) == 's' &&
3058 nextchr = UCHARAT(locinput);
3062 /* Neither the target and the pattern are utf8. */
3064 /* Inline the first character, for speed. */
3065 if (UCHARAT(s) != nextchr &&
3066 UCHARAT(s) != ((OP(scan) == EXACTF)
3067 ? PL_fold : PL_fold_locale)[nextchr])
3069 if (PL_regeol - locinput < ln)
3071 if (ln > 1 && (OP(scan) == EXACTF
3072 ? ibcmp(s, locinput, ln)
3073 : ibcmp_locale(s, locinput, ln)))
3076 nextchr = UCHARAT(locinput);
3081 STRLEN inclasslen = PL_regeol - locinput;
3083 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3085 if (locinput >= PL_regeol)
3087 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3088 nextchr = UCHARAT(locinput);
3093 nextchr = UCHARAT(locinput);
3094 if (!REGINCLASS(rex, scan, (U8*)locinput))
3096 if (!nextchr && locinput >= PL_regeol)
3098 nextchr = UCHARAT(++locinput);
3102 /* If we might have the case of the German sharp s
3103 * in a casefolding Unicode character class. */
3105 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3106 locinput += SHARP_S_SKIP;
3107 nextchr = UCHARAT(locinput);
3113 PL_reg_flags |= RF_tainted;
3119 LOAD_UTF8_CHARCLASS_ALNUM();
3120 if (!(OP(scan) == ALNUM
3121 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3122 : isALNUM_LC_utf8((U8*)locinput)))
3126 locinput += PL_utf8skip[nextchr];
3127 nextchr = UCHARAT(locinput);
3130 if (!(OP(scan) == ALNUM
3131 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3133 nextchr = UCHARAT(++locinput);
3136 PL_reg_flags |= RF_tainted;
3139 if (!nextchr && locinput >= PL_regeol)
3142 LOAD_UTF8_CHARCLASS_ALNUM();
3143 if (OP(scan) == NALNUM
3144 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3145 : isALNUM_LC_utf8((U8*)locinput))
3149 locinput += PL_utf8skip[nextchr];
3150 nextchr = UCHARAT(locinput);
3153 if (OP(scan) == NALNUM
3154 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3156 nextchr = UCHARAT(++locinput);
3160 PL_reg_flags |= RF_tainted;
3164 /* was last char in word? */
3166 if (locinput == PL_bostr)
3169 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3171 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3173 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3174 ln = isALNUM_uni(ln);
3175 LOAD_UTF8_CHARCLASS_ALNUM();
3176 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3179 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3180 n = isALNUM_LC_utf8((U8*)locinput);
3184 ln = (locinput != PL_bostr) ?
3185 UCHARAT(locinput - 1) : '\n';
3186 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3188 n = isALNUM(nextchr);
3191 ln = isALNUM_LC(ln);
3192 n = isALNUM_LC(nextchr);
3195 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3196 OP(scan) == BOUNDL))
3200 PL_reg_flags |= RF_tainted;
3206 if (UTF8_IS_CONTINUED(nextchr)) {
3207 LOAD_UTF8_CHARCLASS_SPACE();
3208 if (!(OP(scan) == SPACE
3209 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3210 : isSPACE_LC_utf8((U8*)locinput)))
3214 locinput += PL_utf8skip[nextchr];
3215 nextchr = UCHARAT(locinput);
3218 if (!(OP(scan) == SPACE
3219 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3221 nextchr = UCHARAT(++locinput);
3224 if (!(OP(scan) == SPACE
3225 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3227 nextchr = UCHARAT(++locinput);
3231 PL_reg_flags |= RF_tainted;
3234 if (!nextchr && locinput >= PL_regeol)
3237 LOAD_UTF8_CHARCLASS_SPACE();
3238 if (OP(scan) == NSPACE
3239 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3240 : isSPACE_LC_utf8((U8*)locinput))
3244 locinput += PL_utf8skip[nextchr];
3245 nextchr = UCHARAT(locinput);
3248 if (OP(scan) == NSPACE
3249 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3251 nextchr = UCHARAT(++locinput);
3254 PL_reg_flags |= RF_tainted;
3260 LOAD_UTF8_CHARCLASS_DIGIT();
3261 if (!(OP(scan) == DIGIT
3262 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3263 : isDIGIT_LC_utf8((U8*)locinput)))
3267 locinput += PL_utf8skip[nextchr];
3268 nextchr = UCHARAT(locinput);
3271 if (!(OP(scan) == DIGIT
3272 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3274 nextchr = UCHARAT(++locinput);
3277 PL_reg_flags |= RF_tainted;
3280 if (!nextchr && locinput >= PL_regeol)
3283 LOAD_UTF8_CHARCLASS_DIGIT();
3284 if (OP(scan) == NDIGIT
3285 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3286 : isDIGIT_LC_utf8((U8*)locinput))
3290 locinput += PL_utf8skip[nextchr];
3291 nextchr = UCHARAT(locinput);
3294 if (OP(scan) == NDIGIT
3295 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3297 nextchr = UCHARAT(++locinput);
3300 if (locinput >= PL_regeol)
3303 LOAD_UTF8_CHARCLASS_MARK();
3304 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3306 locinput += PL_utf8skip[nextchr];
3307 while (locinput < PL_regeol &&
3308 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3309 locinput += UTF8SKIP(locinput);
3310 if (locinput > PL_regeol)
3315 nextchr = UCHARAT(locinput);
3322 PL_reg_flags |= RF_tainted;
3327 n = reg_check_named_buff_matched(rex,scan);
3330 type = REF + ( type - NREF );
3337 PL_reg_flags |= RF_tainted;
3341 n = ARG(scan); /* which paren pair */
3344 ln = PL_regstartp[n];
3345 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3346 if ((I32)*PL_reglastparen < n || ln == -1)
3347 sayNO; /* Do not match unless seen CLOSEn. */
3348 if (ln == PL_regendp[n])
3352 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3354 const char *e = PL_bostr + PL_regendp[n];
3356 * Note that we can't do the "other character" lookup trick as
3357 * in the 8-bit case (no pun intended) because in Unicode we
3358 * have to map both upper and title case to lower case.
3362 STRLEN ulen1, ulen2;
3363 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3364 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3368 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3369 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3370 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3377 nextchr = UCHARAT(locinput);
3381 /* Inline the first character, for speed. */
3382 if (UCHARAT(s) != nextchr &&
3384 (UCHARAT(s) != (type == REFF
3385 ? PL_fold : PL_fold_locale)[nextchr])))
3387 ln = PL_regendp[n] - ln;
3388 if (locinput + ln > PL_regeol)
3390 if (ln > 1 && (type == REF
3391 ? memNE(s, locinput, ln)
3393 ? ibcmp(s, locinput, ln)
3394 : ibcmp_locale(s, locinput, ln))))
3397 nextchr = UCHARAT(locinput);
3407 #define ST st->u.eval
3411 regnode *startpoint;
3414 case GOSUB: /* /(...(?1))/ */
3415 if (cur_eval && cur_eval->locinput==locinput) {
3416 if (cur_eval->u.eval.close_paren == ARG(scan))
3417 Perl_croak(aTHX_ "Infinite recursion in regex");
3418 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3420 "Pattern subroutine nesting without pos change"
3421 " exceeded limit in regex");
3426 (void)ReREFCNT_inc(rex);
3427 if (OP(scan)==GOSUB) {
3428 startpoint = scan + ARG2L(scan);
3429 ST.close_paren = ARG(scan);
3431 startpoint = re->program+1;
3434 goto eval_recurse_doit;
3436 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3437 if (cur_eval && cur_eval->locinput==locinput) {
3438 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3439 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3444 /* execute the code in the {...} */
3446 SV ** const before = SP;
3447 OP_4tree * const oop = PL_op;
3448 COP * const ocurcop = PL_curcop;
3452 PL_op = (OP_4tree*)rex->data->data[n];
3453 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3454 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3455 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3457 CALLRUNOPS(aTHX); /* Scalar context. */
3460 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3467 PAD_RESTORE_LOCAL(old_comppad);
3468 PL_curcop = ocurcop;
3471 sv_setsv(save_scalar(PL_replgv), ret);
3475 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3478 /* extract RE object from returned value; compiling if
3483 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3484 mg = mg_find(sv, PERL_MAGIC_qr);
3485 else if (SvSMAGICAL(ret)) {
3486 if (SvGMAGICAL(ret))
3487 sv_unmagic(ret, PERL_MAGIC_qr);
3489 mg = mg_find(ret, PERL_MAGIC_qr);
3493 re = (regexp *)mg->mg_obj;
3494 (void)ReREFCNT_inc(re);
3498 const char * const t = SvPV_const(ret, len);
3500 const I32 osize = PL_regsize;
3503 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3504 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3506 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3508 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3514 debug_start_match(re, do_utf8, locinput, PL_regeol,
3515 "Matching embedded");
3517 startpoint = re->program + 1;
3518 ST.close_paren = 0; /* only used for GOSUB */
3519 /* borrowed from regtry */
3520 if (PL_reg_start_tmpl <= re->nparens) {
3521 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3522 if(PL_reg_start_tmp)
3523 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3525 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3528 eval_recurse_doit: /* Share code with GOSUB below this line */
3529 /* run the pattern returned from (??{...}) */
3530 ST.cp = regcppush(0); /* Save *all* the positions. */
3531 REGCP_SET(ST.lastcp);
3533 PL_regstartp = re->startp; /* essentially NOOP on GOSUB */
3534 PL_regendp = re->endp; /* essentially NOOP on GOSUB */
3536 *PL_reglastparen = 0;
3537 *PL_reglastcloseparen = 0;
3538 PL_reginput = locinput;
3541 /* XXXX This is too dramatic a measure... */
3544 ST.toggle_reg_flags = PL_reg_flags;
3545 if (re->reganch & ROPT_UTF8)
3546 PL_reg_flags |= RF_utf8;
3548 PL_reg_flags &= ~RF_utf8;
3549 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3552 ST.prev_curlyx = cur_curlyx;
3556 ST.prev_eval = cur_eval;
3558 /* now continue from first node in postoned RE */
3559 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3562 /* logical is 1, /(?(?{...})X|Y)/ */
3563 sw = (bool)SvTRUE(ret);
3568 case EVAL_AB: /* cleanup after a successful (??{A})B */
3569 /* note: this is called twice; first after popping B, then A */
3570 PL_reg_flags ^= ST.toggle_reg_flags;
3574 cur_eval = ST.prev_eval;
3575 cur_curlyx = ST.prev_curlyx;
3576 /* XXXX This is too dramatic a measure... */
3581 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3582 /* note: this is called twice; first after popping B, then A */
3583 PL_reg_flags ^= ST.toggle_reg_flags;
3586 PL_reginput = locinput;
3587 REGCP_UNWIND(ST.lastcp);
3589 cur_eval = ST.prev_eval;
3590 cur_curlyx = ST.prev_curlyx;
3591 /* XXXX This is too dramatic a measure... */
3597 n = ARG(scan); /* which paren pair */
3598 PL_reg_start_tmp[n] = locinput;
3603 n = ARG(scan); /* which paren pair */
3604 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3605 PL_regendp[n] = locinput - PL_bostr;
3606 /*if (n > PL_regsize)
3608 if (n > (I32)*PL_reglastparen)
3609 *PL_reglastparen = n;
3610 *PL_reglastcloseparen = n;
3611 if (cur_eval && cur_eval->u.eval.close_paren == (U32)n) {
3616 n = ARG(scan); /* which paren pair */
3617 sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3620 /* reg_check_named_buff_matched returns 0 for no match */
3621 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3625 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == (U32)n));
3631 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3633 next = NEXTOPER(NEXTOPER(scan));
3635 next = scan + ARG(scan);
3636 if (OP(next) == IFTHEN) /* Fake one. */
3637 next = NEXTOPER(NEXTOPER(next));
3641 logical = scan->flags;
3644 /*******************************************************************
3646 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3647 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3648 STAR/PLUS/CURLY/CURLYN are used instead.)
3650 A*B is compiled as <CURLYX><A><WHILEM><B>
3652 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3653 state, which contains the current count, initialised to -1. It also sets
3654 cur_curlyx to point to this state, with any previous value saved in the
3657 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3658 since the pattern may possibly match zero times (i.e. it's a while {} loop
3659 rather than a do {} while loop).
3661 Each entry to WHILEM represents a successful match of A. The count in the
3662 CURLYX block is incremented, another WHILEM state is pushed, and execution
3663 passes to A or B depending on greediness and the current count.
3665 For example, if matching against the string a1a2a3b (where the aN are
3666 substrings that match /A/), then the match progresses as follows: (the
3667 pushed states are interspersed with the bits of strings matched so far):
3670 <CURLYX cnt=0><WHILEM>
3671 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3672 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3673 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3674 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3676 (Contrast this with something like CURLYM, which maintains only a single
3680 a1 <CURLYM cnt=1> a2
3681 a1 a2 <CURLYM cnt=2> a3
3682 a1 a2 a3 <CURLYM cnt=3> b
3685 Each WHILEM state block marks a point to backtrack to upon partial failure
3686 of A or B, and also contains some minor state data related to that
3687 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3688 overall state, such as the count, and pointers to the A and B ops.
3690 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3691 must always point to the *current* CURLYX block, the rules are:
3693 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3694 and set cur_curlyx to point the new block.
3696 When popping the CURLYX block after a successful or unsuccessful match,
3697 restore the previous cur_curlyx.
3699 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3700 to the outer one saved in the CURLYX block.
3702 When popping the WHILEM block after a successful or unsuccessful B match,
3703 restore the previous cur_curlyx.
3705 Here's an example for the pattern (AI* BI)*BO
3706 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3709 curlyx backtrack stack
3710 ------ ---------------
3712 CO <CO prev=NULL> <WO>
3713 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3714 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3715 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3717 At this point the pattern succeeds, and we work back down the stack to
3718 clean up, restoring as we go:
3720 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3721 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3722 CO <CO prev=NULL> <WO>
3725 *******************************************************************/
3727 #define ST st->u.curlyx
3729 case CURLYX: /* start of /A*B/ (for complex A) */
3731 /* No need to save/restore up to this paren */
3732 I32 parenfloor = scan->flags;
3734 assert(next); /* keep Coverity happy */
3735 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3738 /* XXXX Probably it is better to teach regpush to support
3739 parenfloor > PL_regsize... */
3740 if (parenfloor > (I32)*PL_reglastparen)
3741 parenfloor = *PL_reglastparen; /* Pessimization... */
3743 ST.prev_curlyx= cur_curlyx;
3745 ST.cp = PL_savestack_ix;
3747 /* these fields contain the state of the current curly.
3748 * they are accessed by subsequent WHILEMs */
3749 ST.parenfloor = parenfloor;
3750 ST.min = ARG1(scan);
3751 ST.max = ARG2(scan);
3752 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3756 ST.count = -1; /* this will be updated by WHILEM */
3757 ST.lastloc = NULL; /* this will be updated by WHILEM */
3759 PL_reginput = locinput;
3760 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
3764 case CURLYX_end: /* just finished matching all of A*B */
3766 cur_curlyx = ST.prev_curlyx;
3770 case CURLYX_end_fail: /* just failed to match all of A*B */
3772 cur_curlyx = ST.prev_curlyx;
3778 #define ST st->u.whilem
3780 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
3782 /* see the discussion above about CURLYX/WHILEM */
3784 assert(cur_curlyx); /* keep Coverity happy */
3785 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
3786 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
3787 ST.cache_offset = 0;
3790 PL_reginput = locinput;
3792 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3793 "%*s whilem: matched %ld out of %ld..%ld\n",
3794 REPORT_CODE_OFF+depth*2, "", (long)n,
3795 (long)cur_curlyx->u.curlyx.min,
3796 (long)cur_curlyx->u.curlyx.max)
3799 /* First just match a string of min A's. */
3801 if (n < cur_curlyx->u.curlyx.min) {
3802 cur_curlyx->u.curlyx.lastloc = locinput;
3803 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
3807 /* If degenerate A matches "", assume A done. */
3809 if (locinput == cur_curlyx->u.curlyx.lastloc) {
3810 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3811 "%*s whilem: empty match detected, trying continuation...\n",
3812 REPORT_CODE_OFF+depth*2, "")
3814 goto do_whilem_B_max;
3817 /* super-linear cache processing */
3821 if (!PL_reg_maxiter) {
3822 /* start the countdown: Postpone detection until we
3823 * know the match is not *that* much linear. */
3824 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3825 /* possible overflow for long strings and many CURLYX's */
3826 if (PL_reg_maxiter < 0)
3827 PL_reg_maxiter = I32_MAX;
3828 PL_reg_leftiter = PL_reg_maxiter;
3831 if (PL_reg_leftiter-- == 0) {
3832 /* initialise cache */
3833 const I32 size = (PL_reg_maxiter + 7)/8;
3834 if (PL_reg_poscache) {
3835 if ((I32)PL_reg_poscache_size < size) {
3836 Renew(PL_reg_poscache, size, char);
3837 PL_reg_poscache_size = size;
3839 Zero(PL_reg_poscache, size, char);
3842 PL_reg_poscache_size = size;
3843 Newxz(PL_reg_poscache, size, char);
3845 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3846 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
3847 PL_colors[4], PL_colors[5])
3851 if (PL_reg_leftiter < 0) {
3852 /* have we already failed at this position? */
3854 offset = (scan->flags & 0xf) - 1
3855 + (locinput - PL_bostr) * (scan->flags>>4);
3856 mask = 1 << (offset % 8);
3858 if (PL_reg_poscache[offset] & mask) {
3859 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3860 "%*s whilem: (cache) already tried at this position...\n",
3861 REPORT_CODE_OFF+depth*2, "")
3863 sayNO; /* cache records failure */
3865 ST.cache_offset = offset;
3866 ST.cache_mask = mask;
3870 /* Prefer B over A for minimal matching. */
3872 if (cur_curlyx->u.curlyx.minmod) {
3873 ST.save_curlyx = cur_curlyx;
3874 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
3875 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
3876 REGCP_SET(ST.lastcp);
3877 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
3881 /* Prefer A over B for maximal matching. */
3883 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
3884 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3885 cur_curlyx->u.curlyx.lastloc = locinput;
3886 REGCP_SET(ST.lastcp);
3887 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
3890 goto do_whilem_B_max;
3894 case WHILEM_B_min: /* just matched B in a minimal match */
3895 case WHILEM_B_max: /* just matched B in a maximal match */
3896 cur_curlyx = ST.save_curlyx;
3900 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
3901 cur_curlyx = ST.save_curlyx;
3902 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
3903 cur_curlyx->u.curlyx.count--;
3907 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
3908 REGCP_UNWIND(ST.lastcp);
3911 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
3912 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
3913 cur_curlyx->u.curlyx.count--;
3917 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
3918 REGCP_UNWIND(ST.lastcp);
3919 regcppop(rex); /* Restore some previous $<digit>s? */
3920 PL_reginput = locinput;
3921 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3922 "%*s whilem: failed, trying continuation...\n",
3923 REPORT_CODE_OFF+depth*2, "")
3926 if (cur_curlyx->u.curlyx.count >= REG_INFTY
3927 && ckWARN(WARN_REGEXP)
3928 && !(PL_reg_flags & RF_warned))
3930 PL_reg_flags |= RF_warned;
3931 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3932 "Complex regular subexpression recursion",
3937 ST.save_curlyx = cur_curlyx;
3938 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
3939 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
3942 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
3943 cur_curlyx = ST.save_curlyx;
3944 REGCP_UNWIND(ST.lastcp);
3947 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
3948 /* Maximum greed exceeded */
3949 if (cur_curlyx->u.curlyx.count >= REG_INFTY
3950 && ckWARN(WARN_REGEXP)
3951 && !(PL_reg_flags & RF_warned))
3953 PL_reg_flags |= RF_warned;
3954 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
3955 "%s limit (%d) exceeded",
3956 "Complex regular subexpression recursion",
3959 cur_curlyx->u.curlyx.count--;