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, 2007 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 OP(rn) == KEEPS || (PL_regkind[OP(rn)] == VERB) || \
130 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
132 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
134 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
137 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
138 we don't need this definition. */
139 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
140 #define IS_TEXTF(rn) ( OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
141 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
144 /* ... so we use this as its faster. */
145 #define IS_TEXT(rn) ( OP(rn)==EXACT )
146 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
147 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
152 Search for mandatory following text node; for lookahead, the text must
153 follow but for lookbehind (rn->flags != 0) we skip to the next step.
155 #define FIND_NEXT_IMPT(rn) STMT_START { \
156 while (JUMPABLE(rn)) { \
157 const OPCODE type = OP(rn); \
158 if (type == SUSPEND || PL_regkind[type] == CURLY) \
159 rn = NEXTOPER(NEXTOPER(rn)); \
160 else if (type == PLUS) \
162 else if (type == IFMATCH) \
163 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
164 else rn += NEXT_OFF(rn); \
169 static void restore_pos(pTHX_ void *arg);
172 S_regcppush(pTHX_ I32 parenfloor)
175 const int retval = PL_savestack_ix;
176 #define REGCP_PAREN_ELEMS 4
177 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
179 GET_RE_DEBUG_FLAGS_DECL;
181 if (paren_elems_to_push < 0)
182 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
184 #define REGCP_OTHER_ELEMS 8
185 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
187 for (p = PL_regsize; p > parenfloor; p--) {
188 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
189 SSPUSHINT(PL_regendp[p]);
190 SSPUSHINT(PL_regstartp[p]);
191 SSPUSHPTR(PL_reg_start_tmp[p]);
193 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
194 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
195 (UV)p, (IV)PL_regstartp[p],
196 (IV)(PL_reg_start_tmp[p] - PL_bostr),
200 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
201 SSPUSHPTR(PL_regstartp);
202 SSPUSHPTR(PL_regendp);
203 SSPUSHINT(PL_regsize);
204 SSPUSHINT(*PL_reglastparen);
205 SSPUSHINT(*PL_reglastcloseparen);
206 SSPUSHPTR(PL_reginput);
207 #define REGCP_FRAME_ELEMS 2
208 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
209 * are needed for the regexp context stack bookkeeping. */
210 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
211 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
216 /* These are needed since we do not localize EVAL nodes: */
217 #define REGCP_SET(cp) \
219 PerlIO_printf(Perl_debug_log, \
220 " Setting an EVAL scope, savestack=%"IVdf"\n", \
221 (IV)PL_savestack_ix)); \
224 #define REGCP_UNWIND(cp) \
226 if (cp != PL_savestack_ix) \
227 PerlIO_printf(Perl_debug_log, \
228 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
229 (IV)(cp), (IV)PL_savestack_ix)); \
233 S_regcppop(pTHX_ const regexp *rex)
239 GET_RE_DEBUG_FLAGS_DECL;
241 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
243 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
244 i = SSPOPINT; /* Parentheses elements to pop. */
245 input = (char *) SSPOPPTR;
246 *PL_reglastcloseparen = SSPOPINT;
247 *PL_reglastparen = SSPOPINT;
248 PL_regsize = SSPOPINT;
249 PL_regendp=(I32 *) SSPOPPTR;
250 PL_regstartp=(I32 *) SSPOPPTR;
253 /* Now restore the parentheses context. */
254 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
255 i > 0; i -= REGCP_PAREN_ELEMS) {
257 U32 paren = (U32)SSPOPINT;
258 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
259 PL_regstartp[paren] = SSPOPINT;
261 if (paren <= *PL_reglastparen)
262 PL_regendp[paren] = tmps;
264 PerlIO_printf(Perl_debug_log,
265 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
266 (UV)paren, (IV)PL_regstartp[paren],
267 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
268 (IV)PL_regendp[paren],
269 (paren > *PL_reglastparen ? "(no)" : ""));
273 if (*PL_reglastparen + 1 <= rex->nparens) {
274 PerlIO_printf(Perl_debug_log,
275 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
276 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
280 /* It would seem that the similar code in regtry()
281 * already takes care of this, and in fact it is in
282 * a better location to since this code can #if 0-ed out
283 * but the code in regtry() is needed or otherwise tests
284 * requiring null fields (pat.t#187 and split.t#{13,14}
285 * (as of patchlevel 7877) will fail. Then again,
286 * this code seems to be necessary or otherwise
287 * building DynaLoader will fail:
288 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
290 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
292 PL_regstartp[i] = -1;
299 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
302 * pregexec and friends
305 #ifndef PERL_IN_XSUB_RE
307 - pregexec - match a regexp against a string
310 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
311 char *strbeg, I32 minend, SV *screamer, U32 nosave)
312 /* strend: pointer to null at end of string */
313 /* strbeg: real beginning of string */
314 /* minend: end of match must be >=minend after stringarg. */
315 /* nosave: For optimizations. */
318 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
319 nosave ? 0 : REXEC_COPY_STR);
324 * Need to implement the following flags for reg_anch:
326 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
328 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
329 * INTUIT_AUTORITATIVE_ML
330 * INTUIT_ONCE_NOML - Intuit can match in one location only.
333 * Another flag for this function: SECOND_TIME (so that float substrs
334 * with giant delta may be not rechecked).
337 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
339 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
340 Otherwise, only SvCUR(sv) is used to get strbeg. */
342 /* XXXX We assume that strpos is strbeg unless sv. */
344 /* XXXX Some places assume that there is a fixed substring.
345 An update may be needed if optimizer marks as "INTUITable"
346 RExen without fixed substrings. Similarly, it is assumed that
347 lengths of all the strings are no more than minlen, thus they
348 cannot come from lookahead.
349 (Or minlen should take into account lookahead.)
350 NOTE: Some of this comment is not correct. minlen does now take account
351 of lookahead/behind. Further research is required. -- demerphq
355 /* A failure to find a constant substring means that there is no need to make
356 an expensive call to REx engine, thus we celebrate a failure. Similarly,
357 finding a substring too deep into the string means that less calls to
358 regtry() should be needed.
360 REx compiler's optimizer found 4 possible hints:
361 a) Anchored substring;
363 c) Whether we are anchored (beginning-of-line or \G);
364 d) First node (of those at offset 0) which may distingush positions;
365 We use a)b)d) and multiline-part of c), and try to find a position in the
366 string which does not contradict any of them.
369 /* Most of decisions we do here should have been done at compile time.
370 The nodes of the REx which we used for the search should have been
371 deleted from the finite automaton. */
374 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
375 char *strend, U32 flags, re_scream_pos_data *data)
378 register I32 start_shift = 0;
379 /* Should be nonnegative! */
380 register I32 end_shift = 0;
385 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
387 register char *other_last = NULL; /* other substr checked before this */
388 char *check_at = NULL; /* check substr found at this pos */
389 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
390 RXi_GET_DECL(prog,progi);
392 const char * const i_strpos = strpos;
395 GET_RE_DEBUG_FLAGS_DECL;
397 RX_MATCH_UTF8_set(prog,do_utf8);
399 if (prog->extflags & RXf_UTF8) {
400 PL_reg_flags |= RF_utf8;
403 debug_start_match(prog, do_utf8, strpos, strend,
404 sv ? "Guessing start of match in sv for"
405 : "Guessing start of match in string for");
408 /* CHR_DIST() would be more correct here but it makes things slow. */
409 if (prog->minlen > strend - strpos) {
410 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
411 "String too short... [re_intuit_start]\n"));
415 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
418 if (!prog->check_utf8 && prog->check_substr)
419 to_utf8_substr(prog);
420 check = prog->check_utf8;
422 if (!prog->check_substr && prog->check_utf8)
423 to_byte_substr(prog);
424 check = prog->check_substr;
426 if (check == &PL_sv_undef) {
427 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
428 "Non-utf8 string cannot match utf8 check string\n"));
431 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
432 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
433 || ( (prog->extflags & RXf_ANCH_BOL)
434 && !multiline ) ); /* Check after \n? */
437 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
438 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
439 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
441 && (strpos != strbeg)) {
442 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
445 if (prog->check_offset_min == prog->check_offset_max &&
446 !(prog->extflags & RXf_CANY_SEEN)) {
447 /* Substring at constant offset from beg-of-str... */
450 s = HOP3c(strpos, prog->check_offset_min, strend);
453 slen = SvCUR(check); /* >= 1 */
455 if ( strend - s > slen || strend - s < slen - 1
456 || (strend - s == slen && strend[-1] != '\n')) {
457 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
460 /* Now should match s[0..slen-2] */
462 if (slen && (*SvPVX_const(check) != *s
464 && memNE(SvPVX_const(check), s, slen)))) {
466 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
470 else if (*SvPVX_const(check) != *s
471 || ((slen = SvCUR(check)) > 1
472 && memNE(SvPVX_const(check), s, slen)))
475 goto success_at_start;
478 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
480 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
481 end_shift = prog->check_end_shift;
484 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
485 - (SvTAIL(check) != 0);
486 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
488 if (end_shift < eshift)
492 else { /* Can match at random position */
495 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
496 end_shift = prog->check_end_shift;
498 /* end shift should be non negative here */
501 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
503 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
504 (IV)end_shift, prog->precomp);
508 /* Find a possible match in the region s..strend by looking for
509 the "check" substring in the region corrected by start/end_shift. */
512 I32 srch_start_shift = start_shift;
513 I32 srch_end_shift = end_shift;
514 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
515 srch_end_shift -= ((strbeg - s) - srch_start_shift);
516 srch_start_shift = strbeg - s;
518 DEBUG_OPTIMISE_MORE_r({
519 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
520 (IV)prog->check_offset_min,
521 (IV)srch_start_shift,
523 (IV)prog->check_end_shift);
526 if (flags & REXEC_SCREAM) {
527 I32 p = -1; /* Internal iterator of scream. */
528 I32 * const pp = data ? data->scream_pos : &p;
530 if (PL_screamfirst[BmRARE(check)] >= 0
531 || ( BmRARE(check) == '\n'
532 && (BmPREVIOUS(check) == SvCUR(check) - 1)
534 s = screaminstr(sv, check,
535 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
538 /* we may be pointing at the wrong string */
539 if (s && RX_MATCH_COPIED(prog))
540 s = strbeg + (s - SvPVX_const(sv));
542 *data->scream_olds = s;
547 if (prog->extflags & RXf_CANY_SEEN) {
548 start_point= (U8*)(s + srch_start_shift);
549 end_point= (U8*)(strend - srch_end_shift);
551 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
552 end_point= HOP3(strend, -srch_end_shift, strbeg);
554 DEBUG_OPTIMISE_MORE_r({
555 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
556 (int)(end_point - start_point),
557 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
561 s = fbm_instr( start_point, end_point,
562 check, multiline ? FBMrf_MULTILINE : 0);
565 /* Update the count-of-usability, remove useless subpatterns,
569 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
570 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
571 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
572 (s ? "Found" : "Did not find"),
573 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
574 ? "anchored" : "floating"),
577 (s ? " at offset " : "...\n") );
582 /* Finish the diagnostic message */
583 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
585 /* XXX dmq: first branch is for positive lookbehind...
586 Our check string is offset from the beginning of the pattern.
587 So we need to do any stclass tests offset forward from that
596 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
597 Start with the other substr.
598 XXXX no SCREAM optimization yet - and a very coarse implementation
599 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
600 *always* match. Probably should be marked during compile...
601 Probably it is right to do no SCREAM here...
604 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
605 : (prog->float_substr && prog->anchored_substr))
607 /* Take into account the "other" substring. */
608 /* XXXX May be hopelessly wrong for UTF... */
611 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
614 char * const last = HOP3c(s, -start_shift, strbeg);
616 char * const saved_s = s;
619 t = s - prog->check_offset_max;
620 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
622 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
627 t = HOP3c(t, prog->anchored_offset, strend);
628 if (t < other_last) /* These positions already checked */
630 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
633 /* XXXX It is not documented what units *_offsets are in.
634 We assume bytes, but this is clearly wrong.
635 Meaning this code needs to be carefully reviewed for errors.
639 /* On end-of-str: see comment below. */
640 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
641 if (must == &PL_sv_undef) {
643 DEBUG_r(must = prog->anchored_utf8); /* for debug */
648 HOP3(HOP3(last1, prog->anchored_offset, strend)
649 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
651 multiline ? FBMrf_MULTILINE : 0
654 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
655 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
656 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
657 (s ? "Found" : "Contradicts"),
658 quoted, RE_SV_TAIL(must));
663 if (last1 >= last2) {
664 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
665 ", giving up...\n"));
668 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
669 ", trying floating at offset %ld...\n",
670 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
671 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
672 s = HOP3c(last, 1, strend);
676 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
677 (long)(s - i_strpos)));
678 t = HOP3c(s, -prog->anchored_offset, strbeg);
679 other_last = HOP3c(s, 1, strend);
687 else { /* Take into account the floating substring. */
689 char * const saved_s = s;
692 t = HOP3c(s, -start_shift, strbeg);
694 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
695 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
696 last = HOP3c(t, prog->float_max_offset, strend);
697 s = HOP3c(t, prog->float_min_offset, strend);
700 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
701 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
702 /* fbm_instr() takes into account exact value of end-of-str
703 if the check is SvTAIL(ed). Since false positives are OK,
704 and end-of-str is not later than strend we are OK. */
705 if (must == &PL_sv_undef) {
707 DEBUG_r(must = prog->float_utf8); /* for debug message */
710 s = fbm_instr((unsigned char*)s,
711 (unsigned char*)last + SvCUR(must)
713 must, multiline ? FBMrf_MULTILINE : 0);
715 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
716 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
717 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
718 (s ? "Found" : "Contradicts"),
719 quoted, RE_SV_TAIL(must));
723 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
724 ", giving up...\n"));
727 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
728 ", trying anchored starting at offset %ld...\n",
729 (long)(saved_s + 1 - i_strpos)));
731 s = HOP3c(t, 1, strend);
735 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
736 (long)(s - i_strpos)));
737 other_last = s; /* Fix this later. --Hugo */
747 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
749 DEBUG_OPTIMISE_MORE_r(
750 PerlIO_printf(Perl_debug_log,
751 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
752 (IV)prog->check_offset_min,
753 (IV)prog->check_offset_max,
761 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
763 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
766 /* Fixed substring is found far enough so that the match
767 cannot start at strpos. */
769 if (ml_anch && t[-1] != '\n') {
770 /* Eventually fbm_*() should handle this, but often
771 anchored_offset is not 0, so this check will not be wasted. */
772 /* XXXX In the code below we prefer to look for "^" even in
773 presence of anchored substrings. And we search even
774 beyond the found float position. These pessimizations
775 are historical artefacts only. */
777 while (t < strend - prog->minlen) {
779 if (t < check_at - prog->check_offset_min) {
780 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
781 /* Since we moved from the found position,
782 we definitely contradict the found anchored
783 substr. Due to the above check we do not
784 contradict "check" substr.
785 Thus we can arrive here only if check substr
786 is float. Redo checking for "other"=="fixed".
789 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
790 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
791 goto do_other_anchored;
793 /* We don't contradict the found floating substring. */
794 /* XXXX Why not check for STCLASS? */
796 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
797 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
800 /* Position contradicts check-string */
801 /* XXXX probably better to look for check-string
802 than for "\n", so one should lower the limit for t? */
803 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
804 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
805 other_last = strpos = s = t + 1;
810 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
811 PL_colors[0], PL_colors[1]));
815 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
816 PL_colors[0], PL_colors[1]));
820 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
823 /* The found string does not prohibit matching at strpos,
824 - no optimization of calling REx engine can be performed,
825 unless it was an MBOL and we are not after MBOL,
826 or a future STCLASS check will fail this. */
828 /* Even in this situation we may use MBOL flag if strpos is offset
829 wrt the start of the string. */
830 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
831 && (strpos != strbeg) && strpos[-1] != '\n'
832 /* May be due to an implicit anchor of m{.*foo} */
833 && !(prog->intflags & PREGf_IMPLICIT))
838 DEBUG_EXECUTE_r( if (ml_anch)
839 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
840 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
843 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
845 prog->check_utf8 /* Could be deleted already */
846 && --BmUSEFUL(prog->check_utf8) < 0
847 && (prog->check_utf8 == prog->float_utf8)
849 prog->check_substr /* Could be deleted already */
850 && --BmUSEFUL(prog->check_substr) < 0
851 && (prog->check_substr == prog->float_substr)
854 /* If flags & SOMETHING - do not do it many times on the same match */
855 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
856 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
857 if (do_utf8 ? prog->check_substr : prog->check_utf8)
858 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
859 prog->check_substr = prog->check_utf8 = NULL; /* disable */
860 prog->float_substr = prog->float_utf8 = NULL; /* clear */
861 check = NULL; /* abort */
863 /* XXXX This is a remnant of the old implementation. It
864 looks wasteful, since now INTUIT can use many
866 prog->extflags &= ~RXf_USE_INTUIT;
873 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
874 /* trie stclasses are too expensive to use here, we are better off to
875 leave it to regmatch itself */
876 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
877 /* minlen == 0 is possible if regstclass is \b or \B,
878 and the fixed substr is ''$.
879 Since minlen is already taken into account, s+1 is before strend;
880 accidentally, minlen >= 1 guaranties no false positives at s + 1
881 even for \b or \B. But (minlen? 1 : 0) below assumes that
882 regstclass does not come from lookahead... */
883 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
884 This leaves EXACTF only, which is dealt with in find_byclass(). */
885 const U8* const str = (U8*)STRING(progi->regstclass);
886 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
887 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
890 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
891 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
892 else if (prog->float_substr || prog->float_utf8)
893 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
897 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
898 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
901 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
904 const char *what = NULL;
906 if (endpos == strend) {
907 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
908 "Could not match STCLASS...\n") );
911 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
912 "This position contradicts STCLASS...\n") );
913 if ((prog->extflags & RXf_ANCH) && !ml_anch)
915 /* Contradict one of substrings */
916 if (prog->anchored_substr || prog->anchored_utf8) {
917 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
918 DEBUG_EXECUTE_r( what = "anchored" );
920 s = HOP3c(t, 1, strend);
921 if (s + start_shift + end_shift > strend) {
922 /* XXXX Should be taken into account earlier? */
923 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
924 "Could not match STCLASS...\n") );
929 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
930 "Looking for %s substr starting at offset %ld...\n",
931 what, (long)(s + start_shift - i_strpos)) );
934 /* Have both, check_string is floating */
935 if (t + start_shift >= check_at) /* Contradicts floating=check */
936 goto retry_floating_check;
937 /* Recheck anchored substring, but not floating... */
941 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
942 "Looking for anchored substr starting at offset %ld...\n",
943 (long)(other_last - i_strpos)) );
944 goto do_other_anchored;
946 /* Another way we could have checked stclass at the
947 current position only: */
952 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
953 "Looking for /%s^%s/m starting at offset %ld...\n",
954 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
957 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
959 /* Check is floating subtring. */
960 retry_floating_check:
961 t = check_at - start_shift;
962 DEBUG_EXECUTE_r( what = "floating" );
963 goto hop_and_restart;
966 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
967 "By STCLASS: moving %ld --> %ld\n",
968 (long)(t - i_strpos), (long)(s - i_strpos))
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
973 "Does not contradict STCLASS...\n");
978 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
979 PL_colors[4], (check ? "Guessed" : "Giving up"),
980 PL_colors[5], (long)(s - i_strpos)) );
983 fail_finish: /* Substring not found */
984 if (prog->check_substr || prog->check_utf8) /* could be removed already */
985 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
987 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
988 PL_colors[4], PL_colors[5]));
994 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
995 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
996 switch (trie_type) { \
997 case trie_utf8_fold: \
999 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1004 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1005 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1006 foldlen -= UNISKIP( uvc ); \
1007 uscan = foldbuf + UNISKIP( uvc ); \
1011 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1019 charid = trie->charmap[ uvc ]; \
1023 if (widecharmap) { \
1024 SV** const svpp = hv_fetch(widecharmap, \
1025 (char*)&uvc, sizeof(UV), 0); \
1027 charid = (U16)SvIV(*svpp); \
1032 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1035 ibcmp_utf8(s, NULL, 0, do_utf8, \
1036 m, NULL, ln, (bool)UTF)) \
1037 && (!reginfo || regtry(reginfo, &s)) ) \
1040 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1041 uvchr_to_utf8(tmpbuf, c); \
1042 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1044 && (f == c1 || f == c2) \
1045 && (ln == foldlen || \
1046 !ibcmp_utf8((char *) foldbuf, \
1047 NULL, foldlen, do_utf8, \
1049 NULL, ln, (bool)UTF)) \
1050 && (!reginfo || regtry(reginfo, &s)) ) \
1055 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1059 && (ln == 1 || !(OP(c) == EXACTF \
1061 : ibcmp_locale(s, m, ln))) \
1062 && (!reginfo || regtry(reginfo, &s)) ) \
1068 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1070 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1076 #define REXEC_FBC_SCAN(CoDe) \
1078 while (s < strend) { \
1084 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1085 REXEC_FBC_UTF8_SCAN( \
1087 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1096 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1099 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1108 #define REXEC_FBC_TRYIT \
1109 if ((!reginfo || regtry(reginfo, &s))) \
1112 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1115 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1118 REXEC_FBC_CLASS_SCAN(CoNd); \
1122 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1123 PL_reg_flags |= RF_tainted; \
1125 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1128 REXEC_FBC_CLASS_SCAN(CoNd); \
1132 #define DUMP_EXEC_POS(li,s,doutf8) \
1133 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1135 /* We know what class REx starts with. Try to find this position... */
1136 /* if reginfo is NULL, its a dryrun */
1137 /* annoyingly all the vars in this routine have different names from their counterparts
1138 in regmatch. /grrr */
1141 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1142 const char *strend, regmatch_info *reginfo)
1145 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1149 register STRLEN uskip;
1153 register I32 tmp = 1; /* Scratch variable? */
1154 register const bool do_utf8 = PL_reg_match_utf8;
1155 RXi_GET_DECL(prog,progi);
1157 /* We know what class it must start with. */
1161 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1162 !UTF8_IS_INVARIANT((U8)s[0]) ?
1163 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1164 REGINCLASS(prog, c, (U8*)s));
1167 while (s < strend) {
1170 if (REGINCLASS(prog, c, (U8*)s) ||
1171 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1172 /* The assignment of 2 is intentional:
1173 * for the folded sharp s, the skip is 2. */
1174 (skip = SHARP_S_SKIP))) {
1175 if (tmp && (!reginfo || regtry(reginfo, &s)))
1188 if (tmp && (!reginfo || regtry(reginfo, &s)))
1196 ln = STR_LEN(c); /* length to match in octets/bytes */
1197 lnc = (I32) ln; /* length to match in characters */
1199 STRLEN ulen1, ulen2;
1201 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1202 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1203 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1205 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1206 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1208 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1210 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1213 while (sm < ((U8 *) m + ln)) {
1228 c2 = PL_fold_locale[c1];
1230 e = HOP3c(strend, -((I32)lnc), s);
1232 if (!reginfo && e < s)
1233 e = s; /* Due to minlen logic of intuit() */
1235 /* The idea in the EXACTF* cases is to first find the
1236 * first character of the EXACTF* node and then, if
1237 * necessary, case-insensitively compare the full
1238 * text of the node. The c1 and c2 are the first
1239 * characters (though in Unicode it gets a bit
1240 * more complicated because there are more cases
1241 * than just upper and lower: one needs to use
1242 * the so-called folding case for case-insensitive
1243 * matching (called "loose matching" in Unicode).
1244 * ibcmp_utf8() will do just that. */
1248 U8 tmpbuf [UTF8_MAXBYTES+1];
1249 STRLEN len, foldlen;
1250 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1252 /* Upper and lower of 1st char are equal -
1253 * probably not a "letter". */
1255 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1257 REXEC_FBC_EXACTISH_CHECK(c == c1);
1262 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1265 /* Handle some of the three Greek sigmas cases.
1266 * Note that not all the possible combinations
1267 * are handled here: some of them are handled
1268 * by the standard folding rules, and some of
1269 * them (the character class or ANYOF cases)
1270 * are handled during compiletime in
1271 * regexec.c:S_regclass(). */
1272 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1273 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1274 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1276 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1282 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1284 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1288 PL_reg_flags |= RF_tainted;
1295 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1296 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1298 tmp = ((OP(c) == BOUND ?
1299 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1300 LOAD_UTF8_CHARCLASS_ALNUM();
1301 REXEC_FBC_UTF8_SCAN(
1302 if (tmp == !(OP(c) == BOUND ?
1303 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1304 isALNUM_LC_utf8((U8*)s)))
1312 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1313 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1316 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1322 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1326 PL_reg_flags |= RF_tainted;
1333 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1334 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1336 tmp = ((OP(c) == NBOUND ?
1337 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1338 LOAD_UTF8_CHARCLASS_ALNUM();
1339 REXEC_FBC_UTF8_SCAN(
1340 if (tmp == !(OP(c) == NBOUND ?
1341 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1342 isALNUM_LC_utf8((U8*)s)))
1344 else REXEC_FBC_TRYIT;
1348 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1349 tmp = ((OP(c) == NBOUND ?
1350 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1353 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1355 else REXEC_FBC_TRYIT;
1358 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1362 REXEC_FBC_CSCAN_PRELOAD(
1363 LOAD_UTF8_CHARCLASS_ALNUM(),
1364 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1368 REXEC_FBC_CSCAN_TAINT(
1369 isALNUM_LC_utf8((U8*)s),
1373 REXEC_FBC_CSCAN_PRELOAD(
1374 LOAD_UTF8_CHARCLASS_ALNUM(),
1375 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1379 REXEC_FBC_CSCAN_TAINT(
1380 !isALNUM_LC_utf8((U8*)s),
1384 REXEC_FBC_CSCAN_PRELOAD(
1385 LOAD_UTF8_CHARCLASS_SPACE(),
1386 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1390 REXEC_FBC_CSCAN_TAINT(
1391 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1395 REXEC_FBC_CSCAN_PRELOAD(
1396 LOAD_UTF8_CHARCLASS_SPACE(),
1397 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1401 REXEC_FBC_CSCAN_TAINT(
1402 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1406 REXEC_FBC_CSCAN_PRELOAD(
1407 LOAD_UTF8_CHARCLASS_DIGIT(),
1408 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1412 REXEC_FBC_CSCAN_TAINT(
1413 isDIGIT_LC_utf8((U8*)s),
1417 REXEC_FBC_CSCAN_PRELOAD(
1418 LOAD_UTF8_CHARCLASS_DIGIT(),
1419 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1423 REXEC_FBC_CSCAN_TAINT(
1424 !isDIGIT_LC_utf8((U8*)s),
1430 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1431 trie_type = do_utf8 ?
1432 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1434 /* what trie are we using right now */
1436 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1438 = (reg_trie_data*)progi->data->data[ aho->trie ];
1439 HV *widecharmap = (HV*) progi->data->data[ aho->trie + 1 ];
1441 const char *last_start = strend - trie->minlen;
1443 const char *real_start = s;
1445 STRLEN maxlen = trie->maxlen;
1447 U8 **points; /* map of where we were in the input string
1448 when reading a given char. For ASCII this
1449 is unnecessary overhead as the relationship
1450 is always 1:1, but for unicode, especially
1451 case folded unicode this is not true. */
1452 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1456 GET_RE_DEBUG_FLAGS_DECL;
1458 /* We can't just allocate points here. We need to wrap it in
1459 * an SV so it gets freed properly if there is a croak while
1460 * running the match */
1463 sv_points=newSV(maxlen * sizeof(U8 *));
1464 SvCUR_set(sv_points,
1465 maxlen * sizeof(U8 *));
1466 SvPOK_on(sv_points);
1467 sv_2mortal(sv_points);
1468 points=(U8**)SvPV_nolen(sv_points );
1469 if ( trie_type != trie_utf8_fold
1470 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1473 bitmap=(U8*)trie->bitmap;
1475 bitmap=(U8*)ANYOF_BITMAP(c);
1477 /* this is the Aho-Corasick algorithm modified a touch
1478 to include special handling for long "unknown char"
1479 sequences. The basic idea being that we use AC as long
1480 as we are dealing with a possible matching char, when
1481 we encounter an unknown char (and we have not encountered
1482 an accepting state) we scan forward until we find a legal
1484 AC matching is basically that of trie matching, except
1485 that when we encounter a failing transition, we fall back
1486 to the current states "fail state", and try the current char
1487 again, a process we repeat until we reach the root state,
1488 state 1, or a legal transition. If we fail on the root state
1489 then we can either terminate if we have reached an accepting
1490 state previously, or restart the entire process from the beginning
1494 while (s <= last_start) {
1495 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1503 U8 *uscan = (U8*)NULL;
1504 U8 *leftmost = NULL;
1506 U32 accepted_word= 0;
1510 while ( state && uc <= (U8*)strend ) {
1512 U32 word = aho->states[ state ].wordnum;
1516 DEBUG_TRIE_EXECUTE_r(
1517 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1518 dump_exec_pos( (char *)uc, c, strend, real_start,
1519 (char *)uc, do_utf8 );
1520 PerlIO_printf( Perl_debug_log,
1521 " Scanning for legal start char...\n");
1524 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1529 if (uc >(U8*)last_start) break;
1533 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1534 if (!leftmost || lpos < leftmost) {
1535 DEBUG_r(accepted_word=word);
1541 points[pointpos++ % maxlen]= uc;
1542 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1543 uscan, len, uvc, charid, foldlen,
1545 DEBUG_TRIE_EXECUTE_r({
1546 dump_exec_pos( (char *)uc, c, strend, real_start,
1548 PerlIO_printf(Perl_debug_log,
1549 " Charid:%3u CP:%4"UVxf" ",
1555 word = aho->states[ state ].wordnum;
1557 base = aho->states[ state ].trans.base;
1559 DEBUG_TRIE_EXECUTE_r({
1561 dump_exec_pos( (char *)uc, c, strend, real_start,
1563 PerlIO_printf( Perl_debug_log,
1564 "%sState: %4"UVxf", word=%"UVxf,
1565 failed ? " Fail transition to " : "",
1566 (UV)state, (UV)word);
1571 (base + charid > trie->uniquecharcount )
1572 && (base + charid - 1 - trie->uniquecharcount
1574 && trie->trans[base + charid - 1 -
1575 trie->uniquecharcount].check == state
1576 && (tmp=trie->trans[base + charid - 1 -
1577 trie->uniquecharcount ].next))
1579 DEBUG_TRIE_EXECUTE_r(
1580 PerlIO_printf( Perl_debug_log," - legal\n"));
1585 DEBUG_TRIE_EXECUTE_r(
1586 PerlIO_printf( Perl_debug_log," - fail\n"));
1588 state = aho->fail[state];
1592 /* we must be accepting here */
1593 DEBUG_TRIE_EXECUTE_r(
1594 PerlIO_printf( Perl_debug_log," - accepting\n"));
1603 if (!state) state = 1;
1606 if ( aho->states[ state ].wordnum ) {
1607 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1608 if (!leftmost || lpos < leftmost) {
1609 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1614 s = (char*)leftmost;
1615 DEBUG_TRIE_EXECUTE_r({
1617 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1618 (UV)accepted_word, (IV)(s - real_start)
1621 if (!reginfo || regtry(reginfo, &s)) {
1627 DEBUG_TRIE_EXECUTE_r({
1628 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1631 DEBUG_TRIE_EXECUTE_r(
1632 PerlIO_printf( Perl_debug_log,"No match.\n"));
1641 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1650 - regexec_flags - match a regexp against a string
1653 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1654 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1655 /* strend: pointer to null at end of string */
1656 /* strbeg: real beginning of string */
1657 /* minend: end of match must be >=minend after stringarg. */
1658 /* data: May be used for some additional optimizations.
1659 Currently its only used, with a U32 cast, for transmitting
1660 the ganch offset when doing a /g match. This will change */
1661 /* nosave: For optimizations. */
1664 /*register*/ char *s;
1665 register regnode *c;
1666 /*register*/ char *startpos = stringarg;
1667 I32 minlen; /* must match at least this many chars */
1668 I32 dontbother = 0; /* how many characters not to try at end */
1669 I32 end_shift = 0; /* Same for the end. */ /* CC */
1670 I32 scream_pos = -1; /* Internal iterator of scream. */
1671 char *scream_olds = NULL;
1672 SV* const oreplsv = GvSV(PL_replgv);
1673 const bool do_utf8 = (bool)DO_UTF8(sv);
1675 RXi_GET_DECL(prog,progi);
1676 regmatch_info reginfo; /* create some info to pass to regtry etc */
1678 GET_RE_DEBUG_FLAGS_DECL;
1680 PERL_UNUSED_ARG(data);
1682 /* Be paranoid... */
1683 if (prog == NULL || startpos == NULL) {
1684 Perl_croak(aTHX_ "NULL regexp parameter");
1688 multiline = prog->extflags & RXf_PMf_MULTILINE;
1689 reginfo.prog = prog;
1691 RX_MATCH_UTF8_set(prog, do_utf8);
1693 debug_start_match(prog, do_utf8, startpos, strend,
1697 minlen = prog->minlen;
1699 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1700 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1701 "String too short [regexec_flags]...\n"));
1706 /* Check validity of program. */
1707 if (UCHARAT(progi->program) != REG_MAGIC) {
1708 Perl_croak(aTHX_ "corrupted regexp program");
1712 PL_reg_eval_set = 0;
1715 if (prog->extflags & RXf_UTF8)
1716 PL_reg_flags |= RF_utf8;
1718 /* Mark beginning of line for ^ and lookbehind. */
1719 reginfo.bol = startpos; /* XXX not used ??? */
1723 /* Mark end of line for $ (and such) */
1726 /* see how far we have to get to not match where we matched before */
1727 reginfo.till = startpos+minend;
1729 /* If there is a "must appear" string, look for it. */
1732 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1735 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1736 reginfo.ganch = startpos + prog->gofs;
1737 else if (sv && SvTYPE(sv) >= SVt_PVMG
1739 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1740 && mg->mg_len >= 0) {
1741 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1742 if (prog->extflags & RXf_ANCH_GPOS) {
1743 if (s > reginfo.ganch)
1745 s = reginfo.ganch - prog->gofs;
1749 reginfo.ganch = strbeg + PTR2UV(data);
1750 } else /* pos() not defined */
1751 reginfo.ganch = strbeg;
1753 if (PL_curpm && (PM_GETRE(PL_curpm) == prog)) {
1756 /* We have to be careful. If the previous successful match
1757 was from this regex we don't want a subsequent paritally
1758 successful match to clobber the old results.
1759 So when we detect this possibility we add a swap buffer
1760 to the re, and switch the buffer each match. If we fail
1761 we switch it back, otherwise we leave it swapped.
1763 Newxz(progi->swap, 1, regexp_paren_ofs);
1764 /* no need to copy these */
1765 Newxz(progi->swap->startp, prog->nparens + 1, I32);
1766 Newxz(progi->swap->endp, prog->nparens + 1, I32);
1768 t = progi->swap->startp;
1769 progi->swap->startp = prog->startp;
1771 t = progi->swap->endp;
1772 progi->swap->endp = prog->endp;
1775 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1776 re_scream_pos_data d;
1778 d.scream_olds = &scream_olds;
1779 d.scream_pos = &scream_pos;
1780 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1782 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1783 goto phooey; /* not present */
1789 /* Simplest case: anchored match need be tried only once. */
1790 /* [unless only anchor is BOL and multiline is set] */
1791 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
1792 if (s == startpos && regtry(®info, &startpos))
1794 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
1795 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
1800 dontbother = minlen - 1;
1801 end = HOP3c(strend, -dontbother, strbeg) - 1;
1802 /* for multiline we only have to try after newlines */
1803 if (prog->check_substr || prog->check_utf8) {
1807 if (regtry(®info, &s))
1812 if (prog->extflags & RXf_USE_INTUIT) {
1813 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1824 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1825 if (regtry(®info, &s))
1832 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
1834 /* the warning about reginfo.ganch being used without intialization
1835 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
1836 and we only enter this block when the same bit is set. */
1837 char *tmp_s = reginfo.ganch - prog->gofs;
1838 if (regtry(®info, &tmp_s))
1843 /* Messy cases: unanchored match. */
1844 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
1845 /* we have /x+whatever/ */
1846 /* it must be a one character string (XXXX Except UTF?) */
1851 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1852 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1853 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1858 DEBUG_EXECUTE_r( did_match = 1 );
1859 if (regtry(®info, &s)) goto got_it;
1861 while (s < strend && *s == ch)
1869 DEBUG_EXECUTE_r( did_match = 1 );
1870 if (regtry(®info, &s)) goto got_it;
1872 while (s < strend && *s == ch)
1877 DEBUG_EXECUTE_r(if (!did_match)
1878 PerlIO_printf(Perl_debug_log,
1879 "Did not find anchored character...\n")
1882 else if (prog->anchored_substr != NULL
1883 || prog->anchored_utf8 != NULL
1884 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1885 && prog->float_max_offset < strend - s)) {
1890 char *last1; /* Last position checked before */
1894 if (prog->anchored_substr || prog->anchored_utf8) {
1895 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1896 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1897 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1898 back_max = back_min = prog->anchored_offset;
1900 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1901 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1902 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1903 back_max = prog->float_max_offset;
1904 back_min = prog->float_min_offset;
1908 if (must == &PL_sv_undef)
1909 /* could not downgrade utf8 check substring, so must fail */
1915 last = HOP3c(strend, /* Cannot start after this */
1916 -(I32)(CHR_SVLEN(must)
1917 - (SvTAIL(must) != 0) + back_min), strbeg);
1920 last1 = HOPc(s, -1);
1922 last1 = s - 1; /* bogus */
1924 /* XXXX check_substr already used to find "s", can optimize if
1925 check_substr==must. */
1927 dontbother = end_shift;
1928 strend = HOPc(strend, -dontbother);
1929 while ( (s <= last) &&
1930 ((flags & REXEC_SCREAM)
1931 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1932 end_shift, &scream_pos, 0))
1933 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1934 (unsigned char*)strend, must,
1935 multiline ? FBMrf_MULTILINE : 0))) ) {
1936 /* we may be pointing at the wrong string */
1937 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1938 s = strbeg + (s - SvPVX_const(sv));
1939 DEBUG_EXECUTE_r( did_match = 1 );
1940 if (HOPc(s, -back_max) > last1) {
1941 last1 = HOPc(s, -back_min);
1942 s = HOPc(s, -back_max);
1945 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1947 last1 = HOPc(s, -back_min);
1951 while (s <= last1) {
1952 if (regtry(®info, &s))
1958 while (s <= last1) {
1959 if (regtry(®info, &s))
1965 DEBUG_EXECUTE_r(if (!did_match) {
1966 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1967 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1968 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1969 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1970 ? "anchored" : "floating"),
1971 quoted, RE_SV_TAIL(must));
1975 else if ( (c = progi->regstclass) ) {
1977 const OPCODE op = OP(progi->regstclass);
1978 /* don't bother with what can't match */
1979 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1980 strend = HOPc(strend, -(minlen - 1));
1983 SV * const prop = sv_newmortal();
1984 regprop(prog, prop, c);
1986 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1988 PerlIO_printf(Perl_debug_log,
1989 "Matching stclass %.*s against %s (%d chars)\n",
1990 (int)SvCUR(prop), SvPVX_const(prop),
1991 quoted, (int)(strend - s));
1994 if (find_byclass(prog, c, s, strend, ®info))
1996 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2000 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2005 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2006 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2007 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
2009 if (flags & REXEC_SCREAM) {
2010 last = screaminstr(sv, float_real, s - strbeg,
2011 end_shift, &scream_pos, 1); /* last one */
2013 last = scream_olds; /* Only one occurrence. */
2014 /* we may be pointing at the wrong string */
2015 else if (RX_MATCH_COPIED(prog))
2016 s = strbeg + (s - SvPVX_const(sv));
2020 const char * const little = SvPV_const(float_real, len);
2022 if (SvTAIL(float_real)) {
2023 if (memEQ(strend - len + 1, little, len - 1))
2024 last = strend - len + 1;
2025 else if (!multiline)
2026 last = memEQ(strend - len, little, len)
2027 ? strend - len : NULL;
2033 last = rninstr(s, strend, little, little + len);
2035 last = strend; /* matching "$" */
2040 PerlIO_printf(Perl_debug_log,
2041 "%sCan't trim the tail, match fails (should not happen)%s\n",
2042 PL_colors[4], PL_colors[5]));
2043 goto phooey; /* Should not happen! */
2045 dontbother = strend - last + prog->float_min_offset;
2047 if (minlen && (dontbother < minlen))
2048 dontbother = minlen - 1;
2049 strend -= dontbother; /* this one's always in bytes! */
2050 /* We don't know much -- general case. */
2053 if (regtry(®info, &s))
2062 if (regtry(®info, &s))
2064 } while (s++ < strend);
2072 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2074 if (PL_reg_eval_set) {
2075 /* Preserve the current value of $^R */
2076 if (oreplsv != GvSV(PL_replgv))
2077 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2078 restored, the value remains
2080 restore_pos(aTHX_ prog);
2082 if (prog->paren_names)
2083 (void)hv_iterinit(prog->paren_names);
2085 /* make sure $`, $&, $', and $digit will work later */
2086 if ( !(flags & REXEC_NOT_FIRST) ) {
2087 RX_MATCH_COPY_FREE(prog);
2088 if (flags & REXEC_COPY_STR) {
2089 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2090 #ifdef PERL_OLD_COPY_ON_WRITE
2092 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2094 PerlIO_printf(Perl_debug_log,
2095 "Copy on write: regexp capture, type %d\n",
2098 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2099 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2100 assert (SvPOKp(prog->saved_copy));
2104 RX_MATCH_COPIED_on(prog);
2105 s = savepvn(strbeg, i);
2111 prog->subbeg = strbeg;
2112 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2119 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2120 PL_colors[4], PL_colors[5]));
2121 if (PL_reg_eval_set)
2122 restore_pos(aTHX_ prog);
2124 /* we failed :-( roll it back */
2126 t = progi->swap->startp;
2127 progi->swap->startp = prog->startp;
2129 t = progi->swap->endp;
2130 progi->swap->endp = prog->endp;
2140 - regtry - try match at specific point
2142 STATIC I32 /* 0 failure, 1 success */
2143 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2149 regexp *prog = reginfo->prog;
2150 RXi_GET_DECL(prog,progi);
2151 GET_RE_DEBUG_FLAGS_DECL;
2152 reginfo->cutpoint=NULL;
2154 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2157 PL_reg_eval_set = RS_init;
2158 DEBUG_EXECUTE_r(DEBUG_s(
2159 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2160 (IV)(PL_stack_sp - PL_stack_base));
2163 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2164 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2166 /* Apparently this is not needed, judging by wantarray. */
2167 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2168 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2171 /* Make $_ available to executed code. */
2172 if (reginfo->sv != DEFSV) {
2174 DEFSV = reginfo->sv;
2177 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2178 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2179 /* prepare for quick setting of pos */
2180 #ifdef PERL_OLD_COPY_ON_WRITE
2181 if (SvIsCOW(reginfo->sv))
2182 sv_force_normal_flags(reginfo->sv, 0);
2184 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2185 &PL_vtbl_mglob, NULL, 0);
2189 PL_reg_oldpos = mg->mg_len;
2190 SAVEDESTRUCTOR_X(restore_pos, prog);
2192 if (!PL_reg_curpm) {
2193 Newxz(PL_reg_curpm, 1, PMOP);
2196 SV* const repointer = newSViv(0);
2197 /* so we know which PL_regex_padav element is PL_reg_curpm */
2198 SvFLAGS(repointer) |= SVf_BREAK;
2199 av_push(PL_regex_padav,repointer);
2200 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2201 PL_regex_pad = AvARRAY(PL_regex_padav);
2205 PM_SETRE(PL_reg_curpm, prog);
2206 PL_reg_oldcurpm = PL_curpm;
2207 PL_curpm = PL_reg_curpm;
2208 if (RX_MATCH_COPIED(prog)) {
2209 /* Here is a serious problem: we cannot rewrite subbeg,
2210 since it may be needed if this match fails. Thus
2211 $` inside (?{}) could fail... */
2212 PL_reg_oldsaved = prog->subbeg;
2213 PL_reg_oldsavedlen = prog->sublen;
2214 #ifdef PERL_OLD_COPY_ON_WRITE
2215 PL_nrs = prog->saved_copy;
2217 RX_MATCH_COPIED_off(prog);
2220 PL_reg_oldsaved = NULL;
2221 prog->subbeg = PL_bostr;
2222 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2224 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2225 prog->startp[0] = *startpos - PL_bostr;
2226 PL_reginput = *startpos;
2227 PL_reglastparen = &prog->lastparen;
2228 PL_reglastcloseparen = &prog->lastcloseparen;
2229 prog->lastparen = 0;
2230 prog->lastcloseparen = 0;
2232 PL_regstartp = prog->startp;
2233 PL_regendp = prog->endp;
2234 if (PL_reg_start_tmpl <= prog->nparens) {
2235 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2236 if(PL_reg_start_tmp)
2237 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2239 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2242 /* XXXX What this code is doing here?!!! There should be no need
2243 to do this again and again, PL_reglastparen should take care of
2246 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2247 * Actually, the code in regcppop() (which Ilya may be meaning by
2248 * PL_reglastparen), is not needed at all by the test suite
2249 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2250 * enough, for building DynaLoader, or otherwise this
2251 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2252 * will happen. Meanwhile, this code *is* needed for the
2253 * above-mentioned test suite tests to succeed. The common theme
2254 * on those tests seems to be returning null fields from matches.
2259 if (prog->nparens) {
2261 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2268 if (regmatch(reginfo, progi->program + 1)) {
2269 PL_regendp[0] = PL_reginput - PL_bostr;
2272 if (reginfo->cutpoint)
2273 *startpos= reginfo->cutpoint;
2274 REGCP_UNWIND(lastcp);
2279 #define sayYES goto yes
2280 #define sayNO goto no
2281 #define sayNO_SILENT goto no_silent
2283 /* we dont use STMT_START/END here because it leads to
2284 "unreachable code" warnings, which are bogus, but distracting. */
2285 #define CACHEsayNO \
2286 if (ST.cache_mask) \
2287 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2290 /* this is used to determine how far from the left messages like
2291 'failed...' are printed. It should be set such that messages
2292 are inline with the regop output that created them.
2294 #define REPORT_CODE_OFF 32
2297 /* Make sure there is a test for this +1 options in re_tests */
2298 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2300 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2301 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2303 #define SLAB_FIRST(s) (&(s)->states[0])
2304 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2306 /* grab a new slab and return the first slot in it */
2308 STATIC regmatch_state *
2311 #if PERL_VERSION < 9
2314 regmatch_slab *s = PL_regmatch_slab->next;
2316 Newx(s, 1, regmatch_slab);
2317 s->prev = PL_regmatch_slab;
2319 PL_regmatch_slab->next = s;
2321 PL_regmatch_slab = s;
2322 return SLAB_FIRST(s);
2326 /* push a new state then goto it */
2328 #define PUSH_STATE_GOTO(state, node) \
2330 st->resume_state = state; \
2333 /* push a new state with success backtracking, then goto it */
2335 #define PUSH_YES_STATE_GOTO(state, node) \
2337 st->resume_state = state; \
2338 goto push_yes_state;
2344 regmatch() - main matching routine
2346 This is basically one big switch statement in a loop. We execute an op,
2347 set 'next' to point the next op, and continue. If we come to a point which
2348 we may need to backtrack to on failure such as (A|B|C), we push a
2349 backtrack state onto the backtrack stack. On failure, we pop the top
2350 state, and re-enter the loop at the state indicated. If there are no more
2351 states to pop, we return failure.
2353 Sometimes we also need to backtrack on success; for example /A+/, where
2354 after successfully matching one A, we need to go back and try to
2355 match another one; similarly for lookahead assertions: if the assertion
2356 completes successfully, we backtrack to the state just before the assertion
2357 and then carry on. In these cases, the pushed state is marked as
2358 'backtrack on success too'. This marking is in fact done by a chain of
2359 pointers, each pointing to the previous 'yes' state. On success, we pop to
2360 the nearest yes state, discarding any intermediate failure-only states.
2361 Sometimes a yes state is pushed just to force some cleanup code to be
2362 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2363 it to free the inner regex.
2365 Note that failure backtracking rewinds the cursor position, while
2366 success backtracking leaves it alone.
2368 A pattern is complete when the END op is executed, while a subpattern
2369 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2370 ops trigger the "pop to last yes state if any, otherwise return true"
2373 A common convention in this function is to use A and B to refer to the two
2374 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2375 the subpattern to be matched possibly multiple times, while B is the entire
2376 rest of the pattern. Variable and state names reflect this convention.
2378 The states in the main switch are the union of ops and failure/success of
2379 substates associated with with that op. For example, IFMATCH is the op
2380 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2381 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2382 successfully matched A and IFMATCH_A_fail is a state saying that we have
2383 just failed to match A. Resume states always come in pairs. The backtrack
2384 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2385 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2386 on success or failure.
2388 The struct that holds a backtracking state is actually a big union, with
2389 one variant for each major type of op. The variable st points to the
2390 top-most backtrack struct. To make the code clearer, within each
2391 block of code we #define ST to alias the relevant union.
2393 Here's a concrete example of a (vastly oversimplified) IFMATCH
2399 #define ST st->u.ifmatch
2401 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2402 ST.foo = ...; // some state we wish to save
2404 // push a yes backtrack state with a resume value of
2405 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2407 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2410 case IFMATCH_A: // we have successfully executed A; now continue with B
2412 bar = ST.foo; // do something with the preserved value
2415 case IFMATCH_A_fail: // A failed, so the assertion failed
2416 ...; // do some housekeeping, then ...
2417 sayNO; // propagate the failure
2424 For any old-timers reading this who are familiar with the old recursive
2425 approach, the code above is equivalent to:
2427 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2436 ...; // do some housekeeping, then ...
2437 sayNO; // propagate the failure
2440 The topmost backtrack state, pointed to by st, is usually free. If you
2441 want to claim it, populate any ST.foo fields in it with values you wish to
2442 save, then do one of
2444 PUSH_STATE_GOTO(resume_state, node);
2445 PUSH_YES_STATE_GOTO(resume_state, node);
2447 which sets that backtrack state's resume value to 'resume_state', pushes a
2448 new free entry to the top of the backtrack stack, then goes to 'node'.
2449 On backtracking, the free slot is popped, and the saved state becomes the
2450 new free state. An ST.foo field in this new top state can be temporarily
2451 accessed to retrieve values, but once the main loop is re-entered, it
2452 becomes available for reuse.
2454 Note that the depth of the backtrack stack constantly increases during the
2455 left-to-right execution of the pattern, rather than going up and down with
2456 the pattern nesting. For example the stack is at its maximum at Z at the
2457 end of the pattern, rather than at X in the following:
2459 /(((X)+)+)+....(Y)+....Z/
2461 The only exceptions to this are lookahead/behind assertions and the cut,
2462 (?>A), which pop all the backtrack states associated with A before
2465 Bascktrack state structs are allocated in slabs of about 4K in size.
2466 PL_regmatch_state and st always point to the currently active state,
2467 and PL_regmatch_slab points to the slab currently containing
2468 PL_regmatch_state. The first time regmatch() is called, the first slab is
2469 allocated, and is never freed until interpreter destruction. When the slab
2470 is full, a new one is allocated and chained to the end. At exit from
2471 regmatch(), slabs allocated since entry are freed.
2476 #define DEBUG_STATE_pp(pp) \
2478 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2479 PerlIO_printf(Perl_debug_log, \
2480 " %*s"pp" %s%s%s%s%s\n", \
2482 reg_name[st->resume_state], \
2483 ((st==yes_state||st==mark_state) ? "[" : ""), \
2484 ((st==yes_state) ? "Y" : ""), \
2485 ((st==mark_state) ? "M" : ""), \
2486 ((st==yes_state||st==mark_state) ? "]" : "") \
2491 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2496 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2497 const char *start, const char *end, const char *blurb)
2499 const bool utf8_pat= prog->extflags & RXf_UTF8 ? 1 : 0;
2503 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2504 prog->precomp, prog->prelen, 60);
2506 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2507 start, end - start, 60);
2509 PerlIO_printf(Perl_debug_log,
2510 "%s%s REx%s %s against %s\n",
2511 PL_colors[4], blurb, PL_colors[5], s0, s1);
2513 if (do_utf8||utf8_pat)
2514 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2515 utf8_pat ? "pattern" : "",
2516 utf8_pat && do_utf8 ? " and " : "",
2517 do_utf8 ? "string" : ""
2523 S_dump_exec_pos(pTHX_ const char *locinput,
2524 const regnode *scan,
2525 const char *loc_regeol,
2526 const char *loc_bostr,
2527 const char *loc_reg_starttry,
2530 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2531 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2532 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2533 /* The part of the string before starttry has one color
2534 (pref0_len chars), between starttry and current
2535 position another one (pref_len - pref0_len chars),
2536 after the current position the third one.
2537 We assume that pref0_len <= pref_len, otherwise we
2538 decrease pref0_len. */
2539 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2540 ? (5 + taill) - l : locinput - loc_bostr;
2543 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2545 pref0_len = pref_len - (locinput - loc_reg_starttry);
2546 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2547 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2548 ? (5 + taill) - pref_len : loc_regeol - locinput);
2549 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2553 if (pref0_len > pref_len)
2554 pref0_len = pref_len;
2556 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2558 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2559 (locinput - pref_len),pref0_len, 60, 4, 5);
2561 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2562 (locinput - pref_len + pref0_len),
2563 pref_len - pref0_len, 60, 2, 3);
2565 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2566 locinput, loc_regeol - locinput, 10, 0, 1);
2568 const STRLEN tlen=len0+len1+len2;
2569 PerlIO_printf(Perl_debug_log,
2570 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2571 (IV)(locinput - loc_bostr),
2574 (docolor ? "" : "> <"),
2576 (int)(tlen > 19 ? 0 : 19 - tlen),
2583 /* reg_check_named_buff_matched()
2584 * Checks to see if a named buffer has matched. The data array of
2585 * buffer numbers corresponding to the buffer is expected to reside
2586 * in the regexp->data->data array in the slot stored in the ARG() of
2587 * node involved. Note that this routine doesn't actually care about the
2588 * name, that information is not preserved from compilation to execution.
2589 * Returns the index of the leftmost defined buffer with the given name
2590 * or 0 if non of the buffers matched.
2593 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2595 RXi_GET_DECL(rex,rexi);
2596 SV *sv_dat=(SV*)rexi->data->data[ ARG( scan ) ];
2597 I32 *nums=(I32*)SvPVX(sv_dat);
2598 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2599 if ((I32)*PL_reglastparen >= nums[n] &&
2600 PL_regendp[nums[n]] != -1)
2608 STATIC I32 /* 0 failure, 1 success */
2609 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2611 #if PERL_VERSION < 9
2615 register const bool do_utf8 = PL_reg_match_utf8;
2616 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2618 regexp *rex = reginfo->prog;
2619 RXi_GET_DECL(rex,rexi);
2621 regmatch_slab *orig_slab;
2622 regmatch_state *orig_state;
2624 /* the current state. This is a cached copy of PL_regmatch_state */
2625 register regmatch_state *st;
2627 /* cache heavy used fields of st in registers */
2628 register regnode *scan;
2629 register regnode *next;
2630 register U32 n = 0; /* general value; init to avoid compiler warning */
2631 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2632 register char *locinput = PL_reginput;
2633 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2635 bool result = 0; /* return value of S_regmatch */
2636 int depth = 0; /* depth of backtrack stack */
2637 int nochange_depth = 0; /* depth of GOSUB recursion with nochange*/
2638 regmatch_state *yes_state = NULL; /* state to pop to on success of
2640 /* mark_state piggy backs on the yes_state logic so that when we unwind
2641 the stack on success we can update the mark_state as we go */
2642 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2644 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2645 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2647 bool no_final = 0; /* prevent failure from backtracking? */
2648 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2649 char *startpoint = PL_reginput;
2650 SV *popmark = NULL; /* are we looking for a mark? */
2651 SV *sv_commit = NULL; /* last mark name seen in failure */
2652 SV *sv_yes_mark = NULL; /* last mark name we have seen
2653 during a successfull match */
2654 U32 lastopen = 0; /* last open we saw */
2655 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2658 /* these three flags are set by various ops to signal information to
2659 * the very next op. They have a useful lifetime of exactly one loop
2660 * iteration, and are not preserved or restored by state pushes/pops
2662 bool sw = 0; /* the condition value in (?(cond)a|b) */
2663 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2664 int logical = 0; /* the following EVAL is:
2668 or the following IFMATCH/UNLESSM is:
2669 false: plain (?=foo)
2670 true: used as a condition: (?(?=foo))
2674 GET_RE_DEBUG_FLAGS_DECL;
2678 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2680 /* on first ever call to regmatch, allocate first slab */
2681 if (!PL_regmatch_slab) {
2682 Newx(PL_regmatch_slab, 1, regmatch_slab);
2683 PL_regmatch_slab->prev = NULL;
2684 PL_regmatch_slab->next = NULL;
2685 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2688 /* remember current high-water mark for exit */
2689 /* XXX this should be done with SAVE* instead */
2690 orig_slab = PL_regmatch_slab;
2691 orig_state = PL_regmatch_state;
2693 /* grab next free state slot */
2694 st = ++PL_regmatch_state;
2695 if (st > SLAB_LAST(PL_regmatch_slab))
2696 st = PL_regmatch_state = S_push_slab(aTHX);
2698 /* Note that nextchr is a byte even in UTF */
2699 nextchr = UCHARAT(locinput);
2701 while (scan != NULL) {
2704 SV * const prop = sv_newmortal();
2705 regnode *rnext=regnext(scan);
2706 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2707 regprop(rex, prop, scan);
2709 PerlIO_printf(Perl_debug_log,
2710 "%3"IVdf":%*s%s(%"IVdf")\n",
2711 (IV)(scan - rexi->program), depth*2, "",
2713 (PL_regkind[OP(scan)] == END || !rnext) ?
2714 0 : (IV)(rnext - rexi->program));
2717 next = scan + NEXT_OFF(scan);
2720 state_num = OP(scan);
2723 switch (state_num) {
2725 if (locinput == PL_bostr)
2727 /* reginfo->till = reginfo->bol; */
2732 if (locinput == PL_bostr ||
2733 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2739 if (locinput == PL_bostr)
2743 if (locinput == reginfo->ganch)
2748 /* update the startpoint */
2749 st->u.keeper.val = PL_regstartp[0];
2750 PL_reginput = locinput;
2751 PL_regstartp[0] = locinput - PL_bostr;
2752 PUSH_STATE_GOTO(KEEPS_next, next);
2754 case KEEPS_next_fail:
2755 /* rollback the start point change */
2756 PL_regstartp[0] = st->u.keeper.val;
2762 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2767 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2769 if (PL_regeol - locinput > 1)
2773 if (PL_regeol != locinput)
2777 if (!nextchr && locinput >= PL_regeol)
2780 locinput += PL_utf8skip[nextchr];
2781 if (locinput > PL_regeol)
2783 nextchr = UCHARAT(locinput);
2786 nextchr = UCHARAT(++locinput);
2789 if (!nextchr && locinput >= PL_regeol)
2791 nextchr = UCHARAT(++locinput);
2794 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2797 locinput += PL_utf8skip[nextchr];
2798 if (locinput > PL_regeol)
2800 nextchr = UCHARAT(locinput);
2803 nextchr = UCHARAT(++locinput);
2807 #define ST st->u.trie
2809 /* In this case the charclass data is available inline so
2810 we can fail fast without a lot of extra overhead.
2812 if (scan->flags == EXACT || !do_utf8) {
2813 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2815 PerlIO_printf(Perl_debug_log,
2816 "%*s %sfailed to match trie start class...%s\n",
2817 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2826 /* what type of TRIE am I? (utf8 makes this contextual) */
2827 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2828 trie_type = do_utf8 ?
2829 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2832 /* what trie are we using right now */
2833 reg_trie_data * const trie
2834 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2835 HV * widecharmap = (HV *)rexi->data->data[ ARG( scan ) + 1 ];
2836 U32 state = trie->startstate;
2838 if (trie->bitmap && trie_type != trie_utf8_fold &&
2839 !TRIE_BITMAP_TEST(trie,*locinput)
2841 if (trie->states[ state ].wordnum) {
2843 PerlIO_printf(Perl_debug_log,
2844 "%*s %smatched empty string...%s\n",
2845 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2850 PerlIO_printf(Perl_debug_log,
2851 "%*s %sfailed to match trie start class...%s\n",
2852 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2859 U8 *uc = ( U8* )locinput;
2863 U8 *uscan = (U8*)NULL;
2865 SV *sv_accept_buff = NULL;
2866 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2868 ST.accepted = 0; /* how many accepting states we have seen */
2870 ST.jump = trie->jump;
2874 traverse the TRIE keeping track of all accepting states
2875 we transition through until we get to a failing node.
2878 while ( state && uc <= (U8*)PL_regeol ) {
2879 U32 base = trie->states[ state ].trans.base;
2882 /* We use charid to hold the wordnum as we don't use it
2883 for charid until after we have done the wordnum logic.
2884 We define an alias just so that the wordnum logic reads
2887 #define got_wordnum charid
2888 got_wordnum = trie->states[ state ].wordnum;
2890 if ( got_wordnum ) {
2891 if ( ! ST.accepted ) {
2894 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2895 sv_accept_buff=newSV(bufflen *
2896 sizeof(reg_trie_accepted) - 1);
2897 SvCUR_set(sv_accept_buff, 0);
2898 SvPOK_on(sv_accept_buff);
2899 sv_2mortal(sv_accept_buff);
2902 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2905 if (ST.accepted >= bufflen) {
2907 ST.accept_buff =(reg_trie_accepted*)
2908 SvGROW(sv_accept_buff,
2909 bufflen * sizeof(reg_trie_accepted));
2911 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2912 + sizeof(reg_trie_accepted));
2915 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2916 ST.accept_buff[ST.accepted].endpos = uc;
2918 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2922 DEBUG_TRIE_EXECUTE_r({
2923 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2924 PerlIO_printf( Perl_debug_log,
2925 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2926 2+depth * 2, "", PL_colors[4],
2927 (UV)state, (UV)ST.accepted );
2931 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
2932 uscan, len, uvc, charid, foldlen,
2936 (base + charid > trie->uniquecharcount )
2937 && (base + charid - 1 - trie->uniquecharcount
2939 && trie->trans[base + charid - 1 -
2940 trie->uniquecharcount].check == state)
2942 state = trie->trans[base + charid - 1 -
2943 trie->uniquecharcount ].next;
2954 DEBUG_TRIE_EXECUTE_r(
2955 PerlIO_printf( Perl_debug_log,
2956 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2957 charid, uvc, (UV)state, PL_colors[5] );
2964 PerlIO_printf( Perl_debug_log,
2965 "%*s %sgot %"IVdf" possible matches%s\n",
2966 REPORT_CODE_OFF + depth * 2, "",
2967 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2972 case TRIE_next_fail: /* we failed - try next alterative */
2977 if ( ST.accepted == 1 ) {
2978 /* only one choice left - just continue */
2980 AV *const trie_words
2981 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
2982 SV ** const tmp = av_fetch( trie_words,
2983 ST.accept_buff[ 0 ].wordnum-1, 0 );
2984 SV *sv= tmp ? sv_newmortal() : NULL;
2986 PerlIO_printf( Perl_debug_log,
2987 "%*s %sonly one match left: #%d <%s>%s\n",
2988 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2989 ST.accept_buff[ 0 ].wordnum,
2990 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
2991 PL_colors[0], PL_colors[1],
2992 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2994 : "not compiled under -Dr",
2997 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2998 /* in this case we free tmps/leave before we call regmatch
2999 as we wont be using accept_buff again. */
3001 locinput = PL_reginput;
3002 nextchr = UCHARAT(locinput);
3003 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
3006 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
3007 if (!has_cutgroup) {
3012 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3015 continue; /* execute rest of RE */
3018 if (!ST.accepted-- ) {
3020 PerlIO_printf( Perl_debug_log,
3021 "%*s %sTRIE failed...%s\n",
3022 REPORT_CODE_OFF+depth*2, "",
3032 There are at least two accepting states left. Presumably
3033 the number of accepting states is going to be low,
3034 typically two. So we simply scan through to find the one
3035 with lowest wordnum. Once we find it, we swap the last
3036 state into its place and decrement the size. We then try to
3037 match the rest of the pattern at the point where the word
3038 ends. If we succeed, control just continues along the
3039 regex; if we fail we return here to try the next accepting
3046 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3047 DEBUG_TRIE_EXECUTE_r(
3048 PerlIO_printf( Perl_debug_log,
3049 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3050 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3051 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3052 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3055 if (ST.accept_buff[cur].wordnum <
3056 ST.accept_buff[best].wordnum)
3061 AV *const trie_words
3062 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
3063 SV ** const tmp = av_fetch( trie_words,
3064 ST.accept_buff[ best ].wordnum - 1, 0 );
3065 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3067 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3068 SV *sv= tmp ? sv_newmortal() : NULL;
3070 PerlIO_printf( Perl_debug_log,
3071 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3072 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3073 ST.accept_buff[best].wordnum,
3074 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3075 PL_colors[0], PL_colors[1],
3076 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3077 ) : "not compiled under -Dr",
3078 REG_NODE_NUM(nextop),
3082 if ( best<ST.accepted ) {
3083 reg_trie_accepted tmp = ST.accept_buff[ best ];
3084 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3085 ST.accept_buff[ ST.accepted ] = tmp;
3088 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3089 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3093 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3097 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3100 PUSH_STATE_GOTO(TRIE_next, scan);
3113 char *s = STRING(scan);
3115 if (do_utf8 != UTF) {
3116 /* The target and the pattern have differing utf8ness. */
3118 const char * const e = s + ln;
3121 /* The target is utf8, the pattern is not utf8. */
3126 if (NATIVE_TO_UNI(*(U8*)s) !=
3127 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3135 /* The target is not utf8, the pattern is utf8. */
3140 if (NATIVE_TO_UNI(*((U8*)l)) !=
3141 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3149 nextchr = UCHARAT(locinput);
3152 /* The target and the pattern have the same utf8ness. */
3153 /* Inline the first character, for speed. */
3154 if (UCHARAT(s) != nextchr)
3156 if (PL_regeol - locinput < ln)
3158 if (ln > 1 && memNE(s, locinput, ln))
3161 nextchr = UCHARAT(locinput);
3165 PL_reg_flags |= RF_tainted;
3168 char * const s = STRING(scan);
3171 if (do_utf8 || UTF) {
3172 /* Either target or the pattern are utf8. */
3173 const char * const l = locinput;
3174 char *e = PL_regeol;
3176 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3177 l, &e, 0, do_utf8)) {
3178 /* One more case for the sharp s:
3179 * pack("U0U*", 0xDF) =~ /ss/i,
3180 * the 0xC3 0x9F are the UTF-8
3181 * byte sequence for the U+00DF. */
3183 toLOWER(s[0]) == 's' &&
3185 toLOWER(s[1]) == 's' &&
3192 nextchr = UCHARAT(locinput);
3196 /* Neither the target and the pattern are utf8. */
3198 /* Inline the first character, for speed. */
3199 if (UCHARAT(s) != nextchr &&
3200 UCHARAT(s) != ((OP(scan) == EXACTF)
3201 ? PL_fold : PL_fold_locale)[nextchr])
3203 if (PL_regeol - locinput < ln)
3205 if (ln > 1 && (OP(scan) == EXACTF
3206 ? ibcmp(s, locinput, ln)
3207 : ibcmp_locale(s, locinput, ln)))
3210 nextchr = UCHARAT(locinput);
3215 STRLEN inclasslen = PL_regeol - locinput;
3217 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3219 if (locinput >= PL_regeol)
3221 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3222 nextchr = UCHARAT(locinput);
3227 nextchr = UCHARAT(locinput);
3228 if (!REGINCLASS(rex, scan, (U8*)locinput))
3230 if (!nextchr && locinput >= PL_regeol)
3232 nextchr = UCHARAT(++locinput);
3236 /* If we might have the case of the German sharp s
3237 * in a casefolding Unicode character class. */
3239 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3240 locinput += SHARP_S_SKIP;
3241 nextchr = UCHARAT(locinput);
3247 PL_reg_flags |= RF_tainted;
3253 LOAD_UTF8_CHARCLASS_ALNUM();
3254 if (!(OP(scan) == ALNUM
3255 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3256 : isALNUM_LC_utf8((U8*)locinput)))
3260 locinput += PL_utf8skip[nextchr];
3261 nextchr = UCHARAT(locinput);
3264 if (!(OP(scan) == ALNUM
3265 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3267 nextchr = UCHARAT(++locinput);
3270 PL_reg_flags |= RF_tainted;
3273 if (!nextchr && locinput >= PL_regeol)
3276 LOAD_UTF8_CHARCLASS_ALNUM();
3277 if (OP(scan) == NALNUM
3278 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3279 : isALNUM_LC_utf8((U8*)locinput))
3283 locinput += PL_utf8skip[nextchr];
3284 nextchr = UCHARAT(locinput);
3287 if (OP(scan) == NALNUM
3288 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3290 nextchr = UCHARAT(++locinput);
3294 PL_reg_flags |= RF_tainted;
3298 /* was last char in word? */
3300 if (locinput == PL_bostr)
3303 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3305 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3307 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3308 ln = isALNUM_uni(ln);
3309 LOAD_UTF8_CHARCLASS_ALNUM();
3310 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3313 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3314 n = isALNUM_LC_utf8((U8*)locinput);
3318 ln = (locinput != PL_bostr) ?
3319 UCHARAT(locinput - 1) : '\n';
3320 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3322 n = isALNUM(nextchr);
3325 ln = isALNUM_LC(ln);
3326 n = isALNUM_LC(nextchr);
3329 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3330 OP(scan) == BOUNDL))
3334 PL_reg_flags |= RF_tainted;
3340 if (UTF8_IS_CONTINUED(nextchr)) {
3341 LOAD_UTF8_CHARCLASS_SPACE();
3342 if (!(OP(scan) == SPACE
3343 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3344 : isSPACE_LC_utf8((U8*)locinput)))
3348 locinput += PL_utf8skip[nextchr];
3349 nextchr = UCHARAT(locinput);
3352 if (!(OP(scan) == SPACE
3353 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3355 nextchr = UCHARAT(++locinput);
3358 if (!(OP(scan) == SPACE
3359 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3361 nextchr = UCHARAT(++locinput);
3365 PL_reg_flags |= RF_tainted;
3368 if (!nextchr && locinput >= PL_regeol)
3371 LOAD_UTF8_CHARCLASS_SPACE();
3372 if (OP(scan) == NSPACE
3373 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3374 : isSPACE_LC_utf8((U8*)locinput))
3378 locinput += PL_utf8skip[nextchr];
3379 nextchr = UCHARAT(locinput);
3382 if (OP(scan) == NSPACE
3383 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3385 nextchr = UCHARAT(++locinput);
3388 PL_reg_flags |= RF_tainted;
3394 LOAD_UTF8_CHARCLASS_DIGIT();
3395 if (!(OP(scan) == DIGIT
3396 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3397 : isDIGIT_LC_utf8((U8*)locinput)))
3401 locinput += PL_utf8skip[nextchr];
3402 nextchr = UCHARAT(locinput);
3405 if (!(OP(scan) == DIGIT
3406 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3408 nextchr = UCHARAT(++locinput);
3411 PL_reg_flags |= RF_tainted;
3414 if (!nextchr && locinput >= PL_regeol)
3417 LOAD_UTF8_CHARCLASS_DIGIT();
3418 if (OP(scan) == NDIGIT
3419 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3420 : isDIGIT_LC_utf8((U8*)locinput))
3424 locinput += PL_utf8skip[nextchr];
3425 nextchr = UCHARAT(locinput);
3428 if (OP(scan) == NDIGIT
3429 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3431 nextchr = UCHARAT(++locinput);
3434 if (locinput >= PL_regeol)
3437 LOAD_UTF8_CHARCLASS_MARK();
3438 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3440 locinput += PL_utf8skip[nextchr];
3441 while (locinput < PL_regeol &&
3442 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3443 locinput += UTF8SKIP(locinput);
3444 if (locinput > PL_regeol)
3449 nextchr = UCHARAT(locinput);
3456 PL_reg_flags |= RF_tainted;
3461 n = reg_check_named_buff_matched(rex,scan);
3464 type = REF + ( type - NREF );
3471 PL_reg_flags |= RF_tainted;
3475 n = ARG(scan); /* which paren pair */
3478 ln = PL_regstartp[n];
3479 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3480 if (*PL_reglastparen < n || ln == -1)
3481 sayNO; /* Do not match unless seen CLOSEn. */
3482 if (ln == PL_regendp[n])
3486 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3488 const char *e = PL_bostr + PL_regendp[n];
3490 * Note that we can't do the "other character" lookup trick as
3491 * in the 8-bit case (no pun intended) because in Unicode we
3492 * have to map both upper and title case to lower case.
3496 STRLEN ulen1, ulen2;
3497 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3498 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3502 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3503 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3504 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3511 nextchr = UCHARAT(locinput);
3515 /* Inline the first character, for speed. */
3516 if (UCHARAT(s) != nextchr &&
3518 (UCHARAT(s) != (type == REFF
3519 ? PL_fold : PL_fold_locale)[nextchr])))
3521 ln = PL_regendp[n] - ln;
3522 if (locinput + ln > PL_regeol)
3524 if (ln > 1 && (type == REF
3525 ? memNE(s, locinput, ln)
3527 ? ibcmp(s, locinput, ln)
3528 : ibcmp_locale(s, locinput, ln))))
3531 nextchr = UCHARAT(locinput);
3541 #define ST st->u.eval
3545 regexp_internal *rei;
3546 regnode *startpoint;
3549 case GOSUB: /* /(...(?1))/ */
3550 if (cur_eval && cur_eval->locinput==locinput) {
3551 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3552 Perl_croak(aTHX_ "Infinite recursion in regex");
3553 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3555 "Pattern subroutine nesting without pos change"
3556 " exceeded limit in regex");
3562 (void)ReREFCNT_inc(rex);
3563 if (OP(scan)==GOSUB) {
3564 startpoint = scan + ARG2L(scan);
3565 ST.close_paren = ARG(scan);
3567 startpoint = rei->program+1;
3570 goto eval_recurse_doit;
3572 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3573 if (cur_eval && cur_eval->locinput==locinput) {
3574 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3575 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3579 { regexp *ocurpm = PM_GETRE(PL_curpm);
3580 char *osubbeg = rex->subbeg;
3581 STRLEN osublen = rex->sublen;
3583 /* execute the code in the {...} */
3585 SV ** const before = SP;
3586 OP_4tree * const oop = PL_op;
3587 COP * const ocurcop = PL_curcop;
3592 PL_op = (OP_4tree*)rexi->data->data[n];
3593 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3594 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3595 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3596 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3599 SV *sv_mrk = get_sv("REGMARK", 1);
3600 sv_setsv(sv_mrk, sv_yes_mark);
3602 /* make sure that $1 and friends are available with nested eval */
3603 PM_SETRE(PL_curpm,rex);
3604 rex->subbeg = ocurpm->subbeg;
3605 rex->sublen = ocurpm->sublen;
3607 CALLRUNOPS(aTHX); /* Scalar context. */
3610 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3617 PAD_RESTORE_LOCAL(old_comppad);
3618 PL_curcop = ocurcop;
3622 sv_setsv(save_scalar(PL_replgv), ret);
3626 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3629 /* extract RE object from returned value; compiling if
3634 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3635 mg = mg_find(sv, PERL_MAGIC_qr);
3636 else if (SvSMAGICAL(ret)) {
3637 if (SvGMAGICAL(ret))
3638 sv_unmagic(ret, PERL_MAGIC_qr);
3640 mg = mg_find(ret, PERL_MAGIC_qr);
3644 re = (regexp *)mg->mg_obj;
3645 (void)ReREFCNT_inc(re);
3649 const char * const t = SvPV_const(ret, len);
3651 const I32 osize = PL_regsize;
3654 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3655 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3657 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3659 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3666 /* restore PL_curpm after the eval */
3667 PM_SETRE(PL_curpm,ocurpm);
3668 rex->sublen = osublen;
3669 rex->subbeg = osubbeg;
3672 debug_start_match(re, do_utf8, locinput, PL_regeol,
3673 "Matching embedded");
3675 startpoint = rei->program + 1;
3676 ST.close_paren = 0; /* only used for GOSUB */
3677 /* borrowed from regtry */
3678 if (PL_reg_start_tmpl <= re->nparens) {
3679 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3680 if(PL_reg_start_tmp)
3681 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3683 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3687 eval_recurse_doit: /* Share code with GOSUB below this line */
3688 /* run the pattern returned from (??{...}) */
3689 ST.cp = regcppush(0); /* Save *all* the positions. */
3690 REGCP_SET(ST.lastcp);
3692 PL_regstartp = re->startp; /* essentially NOOP on GOSUB */
3693 PL_regendp = re->endp; /* essentially NOOP on GOSUB */
3695 *PL_reglastparen = 0;
3696 *PL_reglastcloseparen = 0;
3697 PL_reginput = locinput;
3700 /* XXXX This is too dramatic a measure... */
3703 ST.toggle_reg_flags = PL_reg_flags;
3704 if (re->extflags & RXf_UTF8)
3705 PL_reg_flags |= RF_utf8;
3707 PL_reg_flags &= ~RF_utf8;
3708 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3711 ST.prev_curlyx = cur_curlyx;
3716 ST.prev_eval = cur_eval;
3718 /* now continue from first node in postoned RE */
3719 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3722 /* restore PL_curpm after the eval */
3723 PM_SETRE(PL_curpm,ocurpm);
3724 rex->sublen = osublen;
3725 rex->subbeg = osubbeg;
3727 /* logical is 1, /(?(?{...})X|Y)/ */
3728 sw = (bool)SvTRUE(ret);
3733 case EVAL_AB: /* cleanup after a successful (??{A})B */
3734 /* note: this is called twice; first after popping B, then A */
3735 PL_reg_flags ^= ST.toggle_reg_flags;
3738 rexi = RXi_GET(rex);
3740 cur_eval = ST.prev_eval;
3741 cur_curlyx = ST.prev_curlyx;
3742 /* XXXX This is too dramatic a measure... */
3747 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3748 /* note: this is called twice; first after popping B, then A */
3749 PL_reg_flags ^= ST.toggle_reg_flags;
3752 rexi = RXi_GET(rex);
3753 PL_reginput = locinput;
3754 REGCP_UNWIND(ST.lastcp);
3756 cur_eval = ST.prev_eval;
3757 cur_curlyx = ST.prev_curlyx;
3758 /* XXXX This is too dramatic a measure... */
3764 n = ARG(scan); /* which paren pair */
3765 PL_reg_start_tmp[n] = locinput;
3771 n = ARG(scan); /* which paren pair */
3772 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3773 PL_regendp[n] = locinput - PL_bostr;
3774 /*if (n > PL_regsize)
3776 if (n > *PL_reglastparen)
3777 *PL_reglastparen = n;
3778 *PL_reglastcloseparen = n;
3779 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3787 cursor && OP(cursor)!=END;
3788 cursor=regnext(cursor))
3790 if ( OP(cursor)==CLOSE ){
3792 if ( n <= lastopen ) {
3793 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3794 PL_regendp[n] = locinput - PL_bostr;
3795 /*if (n > PL_regsize)
3797 if (n > *PL_reglastparen)
3798 *PL_reglastparen = n;
3799 *PL_reglastcloseparen = n;
3800 if ( n == ARG(scan) || (cur_eval &&
3801 cur_eval->u.eval.close_paren == n))
3810 n = ARG(scan); /* which paren pair */
3811 sw = (bool)(*PL_reglastparen >= n && PL_regendp[n] != -1);
3814 /* reg_check_named_buff_matched returns 0 for no match */
3815 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3819 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3825 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3827 next = NEXTOPER(NEXTOPER(scan));
3829 next = scan + ARG(scan);
3830 if (OP(next) == IFTHEN) /* Fake one. */
3831 next = NEXTOPER(NEXTOPER(next));
3835 logical = scan->flags;
3838 /*******************************************************************
3840 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3841 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3842 STAR/PLUS/CURLY/CURLYN are used instead.)
3844 A*B is compiled as <CURLYX><A><WHILEM><B>
3846 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3847 state, which contains the current count, initialised to -1. It also sets
3848 cur_curlyx to point to this state, with any previous value saved in the
3851 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3852 since the pattern may possibly match zero times (i.e. it's a while {} loop
3853 rather than a do {} while loop).
3855 Each entry to WHILEM represents a successful match of A. The count in the
3856 CURLYX block is incremented, another WHILEM state is pushed, and execution
3857 passes to A or B depending on greediness and the current count.
3859 For example, if matching against the string a1a2a3b (where the aN are
3860 substrings that match /A/), then the match progresses as follows: (the
3861 pushed states are interspersed with the bits of strings matched so far):
3864 <CURLYX cnt=0><WHILEM>
3865 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3866 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3867 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3868 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3870 (Contrast this with something like CURLYM, which maintains only a single
3874 a1 <CURLYM cnt=1> a2
3875 a1 a2 <CURLYM cnt=2> a3
3876 a1 a2 a3 <CURLYM cnt=3> b
3879 Each WHILEM state block marks a point to backtrack to upon partial failure
3880 of A or B, and also contains some minor state data related to that
3881 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3882 overall state, such as the count, and pointers to the A and B ops.
3884 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3885 must always point to the *current* CURLYX block, the rules are:
3887 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3888 and set cur_curlyx to point the new block.
3890 When popping the CURLYX block after a successful or unsuccessful match,
3891 restore the previous cur_curlyx.
3893 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3894 to the outer one saved in the CURLYX block.
3896 When popping the WHILEM block after a successful or unsuccessful B match,
3897 restore the previous cur_curlyx.
3899 Here's an example for the pattern (AI* BI)*BO
3900 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3903 curlyx backtrack stack
3904 ------ ---------------
3906 CO <CO prev=NULL> <WO>
3907 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3908 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3909 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3911 At this point the pattern succeeds, and we work back down the stack to
3912 clean up, restoring as we go:
3914 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3915 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3916 CO <CO prev=NULL> <WO>
3919 *******************************************************************/
3921 #define ST st->u.curlyx
3923 case CURLYX: /* start of /A*B/ (for complex A) */
3925 /* No need to save/restore up to this paren */
3926 I32 parenfloor = scan->flags;
3928 assert(next); /* keep Coverity happy */
3929 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3932 /* XXXX Probably it is better to teach regpush to support
3933 parenfloor > PL_regsize... */
3934 if (parenfloor > (I32)*PL_reglastparen)
3935 parenfloor = *PL_reglastparen; /* Pessimization... */
3937 ST.prev_curlyx= cur_curlyx;
3939 ST.cp = PL_savestack_ix;
3941 /* these fields contain the state of the current curly.
3942 * they are accessed by subsequent WHILEMs */
3943 ST.parenfloor = parenfloor;
3944 ST.min = ARG1(scan);
3945 ST.max = ARG2(scan);
3946 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3950 ST.count = -1; /* this will be updated by WHILEM */
3951 ST.lastloc = NULL; /* this will be updated by WHILEM */
3953 PL_reginput = locinput;
3954 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));