5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
56 **** Alterations to Henry's code are...
58 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
59 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
61 **** You may distribute under the terms of either the GNU General Public
62 **** License or the Artistic License, as specified in the README file.
64 * Beware that some of this code is subtly aware of the way operator
65 * precedence is structured in regular expressions. Serious changes in
66 * regular-expression syntax might require a total rethink.
69 #define PERL_IN_REGEXEC_C
72 #ifdef PERL_IN_XSUB_RE
78 #define RF_tainted 1 /* tainted information used? */
79 #define RF_warned 2 /* warned about big count? */
80 #define RF_evaled 4 /* Did an EVAL with setting? */
81 #define RF_utf8 8 /* Pattern contains multibyte chars? */
83 #define UTF ((PL_reg_flags & RF_utf8) != 0)
85 #define RS_init 1 /* eval environment created */
86 #define RS_set 2 /* replsv value is set */
92 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
98 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
99 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
101 #define HOPc(pos,off) \
102 (char *)(PL_reg_match_utf8 \
103 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
105 #define HOPBACKc(pos, off) \
106 (char*)(PL_reg_match_utf8\
107 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
108 : (pos - off >= PL_bostr) \
112 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
113 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
115 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
116 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
117 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
118 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
119 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
120 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
122 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
124 /* for use after a quantifier and before an EXACT-like node -- japhy */
125 #define JUMPABLE(rn) ( \
126 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
127 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
128 OP(rn) == PLUS || OP(rn) == MINMOD || \
129 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
132 #define HAS_TEXT(rn) ( \
133 PL_regkind[OP(rn)] == EXACT || PL_regkind[OP(rn)] == REF \
137 Search for mandatory following text node; for lookahead, the text must
138 follow but for lookbehind (rn->flags != 0) we skip to the next step.
140 #define FIND_NEXT_IMPT(rn) STMT_START { \
141 while (JUMPABLE(rn)) { \
142 const OPCODE type = OP(rn); \
143 if (type == SUSPEND || PL_regkind[type] == CURLY) \
144 rn = NEXTOPER(NEXTOPER(rn)); \
145 else if (type == PLUS) \
147 else if (type == IFMATCH) \
148 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
149 else rn += NEXT_OFF(rn); \
153 static void restore_pos(pTHX_ void *arg);
156 S_regcppush(pTHX_ I32 parenfloor)
159 const int retval = PL_savestack_ix;
160 #define REGCP_PAREN_ELEMS 4
161 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
163 GET_RE_DEBUG_FLAGS_DECL;
165 if (paren_elems_to_push < 0)
166 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
168 #define REGCP_OTHER_ELEMS 6
169 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
170 for (p = PL_regsize; p > parenfloor; p--) {
171 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
172 SSPUSHINT(PL_regendp[p]);
173 SSPUSHINT(PL_regstartp[p]);
174 SSPUSHPTR(PL_reg_start_tmp[p]);
176 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
177 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
178 (UV)p, (IV)PL_regstartp[p],
179 (IV)(PL_reg_start_tmp[p] - PL_bostr),
183 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
184 SSPUSHINT(PL_regsize);
185 SSPUSHINT(*PL_reglastparen);
186 SSPUSHINT(*PL_reglastcloseparen);
187 SSPUSHPTR(PL_reginput);
188 #define REGCP_FRAME_ELEMS 2
189 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
190 * are needed for the regexp context stack bookkeeping. */
191 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
192 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
197 /* These are needed since we do not localize EVAL nodes: */
198 #define REGCP_SET(cp) \
200 PerlIO_printf(Perl_debug_log, \
201 " Setting an EVAL scope, savestack=%"IVdf"\n", \
202 (IV)PL_savestack_ix)); \
205 #define REGCP_UNWIND(cp) \
207 if (cp != PL_savestack_ix) \
208 PerlIO_printf(Perl_debug_log, \
209 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
210 (IV)(cp), (IV)PL_savestack_ix)); \
214 S_regcppop(pTHX_ const regexp *rex)
220 GET_RE_DEBUG_FLAGS_DECL;
222 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
224 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
225 i = SSPOPINT; /* Parentheses elements to pop. */
226 input = (char *) SSPOPPTR;
227 *PL_reglastcloseparen = SSPOPINT;
228 *PL_reglastparen = SSPOPINT;
229 PL_regsize = SSPOPINT;
231 /* Now restore the parentheses context. */
232 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
233 i > 0; i -= REGCP_PAREN_ELEMS) {
235 U32 paren = (U32)SSPOPINT;
236 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
237 PL_regstartp[paren] = SSPOPINT;
239 if (paren <= *PL_reglastparen)
240 PL_regendp[paren] = tmps;
242 PerlIO_printf(Perl_debug_log,
243 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
244 (UV)paren, (IV)PL_regstartp[paren],
245 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
246 (IV)PL_regendp[paren],
247 (paren > *PL_reglastparen ? "(no)" : ""));
251 if (*PL_reglastparen + 1 <= rex->nparens) {
252 PerlIO_printf(Perl_debug_log,
253 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
254 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
258 /* It would seem that the similar code in regtry()
259 * already takes care of this, and in fact it is in
260 * a better location to since this code can #if 0-ed out
261 * but the code in regtry() is needed or otherwise tests
262 * requiring null fields (pat.t#187 and split.t#{13,14}
263 * (as of patchlevel 7877) will fail. Then again,
264 * this code seems to be necessary or otherwise
265 * building DynaLoader will fail:
266 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
268 for (i = *PL_reglastparen + 1; (U32)i <= rex->nparens; i++) {
270 PL_regstartp[i] = -1;
277 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
280 * pregexec and friends
283 #ifndef PERL_IN_XSUB_RE
285 - pregexec - match a regexp against a string
288 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
289 char *strbeg, I32 minend, SV *screamer, U32 nosave)
290 /* strend: pointer to null at end of string */
291 /* strbeg: real beginning of string */
292 /* minend: end of match must be >=minend after stringarg. */
293 /* nosave: For optimizations. */
296 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
297 nosave ? 0 : REXEC_COPY_STR);
302 * Need to implement the following flags for reg_anch:
304 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
306 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
307 * INTUIT_AUTORITATIVE_ML
308 * INTUIT_ONCE_NOML - Intuit can match in one location only.
311 * Another flag for this function: SECOND_TIME (so that float substrs
312 * with giant delta may be not rechecked).
315 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
317 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
318 Otherwise, only SvCUR(sv) is used to get strbeg. */
320 /* XXXX We assume that strpos is strbeg unless sv. */
322 /* XXXX Some places assume that there is a fixed substring.
323 An update may be needed if optimizer marks as "INTUITable"
324 RExen without fixed substrings. Similarly, it is assumed that
325 lengths of all the strings are no more than minlen, thus they
326 cannot come from lookahead.
327 (Or minlen should take into account lookahead.) */
329 /* A failure to find a constant substring means that there is no need to make
330 an expensive call to REx engine, thus we celebrate a failure. Similarly,
331 finding a substring too deep into the string means that less calls to
332 regtry() should be needed.
334 REx compiler's optimizer found 4 possible hints:
335 a) Anchored substring;
337 c) Whether we are anchored (beginning-of-line or \G);
338 d) First node (of those at offset 0) which may distingush positions;
339 We use a)b)d) and multiline-part of c), and try to find a position in the
340 string which does not contradict any of them.
343 /* Most of decisions we do here should have been done at compile time.
344 The nodes of the REx which we used for the search should have been
345 deleted from the finite automaton. */
348 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
349 char *strend, U32 flags, re_scream_pos_data *data)
352 register I32 start_shift = 0;
353 /* Should be nonnegative! */
354 register I32 end_shift = 0;
359 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
361 register char *other_last = NULL; /* other substr checked before this */
362 char *check_at = NULL; /* check substr found at this pos */
363 const I32 multiline = prog->reganch & PMf_MULTILINE;
365 const char * const i_strpos = strpos;
368 GET_RE_DEBUG_FLAGS_DECL;
370 RX_MATCH_UTF8_set(prog,do_utf8);
372 if (prog->reganch & ROPT_UTF8) {
373 PL_reg_flags |= RF_utf8;
376 debug_start_match(prog, do_utf8, strpos, strend,
377 sv ? "Guessing start of match in sv for"
378 : "Guessing start of match in string for");
381 /* CHR_DIST() would be more correct here but it makes things slow. */
382 if (prog->minlen > strend - strpos) {
383 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
384 "String too short... [re_intuit_start]\n"));
388 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
391 if (!prog->check_utf8 && prog->check_substr)
392 to_utf8_substr(prog);
393 check = prog->check_utf8;
395 if (!prog->check_substr && prog->check_utf8)
396 to_byte_substr(prog);
397 check = prog->check_substr;
399 if (check == &PL_sv_undef) {
400 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
401 "Non-utf8 string cannot match utf8 check string\n"));
404 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
405 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
406 || ( (prog->reganch & ROPT_ANCH_BOL)
407 && !multiline ) ); /* Check after \n? */
410 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
411 | ROPT_IMPLICIT)) /* not a real BOL */
412 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
414 && (strpos != strbeg)) {
415 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
418 if (prog->check_offset_min == prog->check_offset_max &&
419 !(prog->reganch & ROPT_CANY_SEEN)) {
420 /* Substring at constant offset from beg-of-str... */
423 s = HOP3c(strpos, prog->check_offset_min, strend);
426 slen = SvCUR(check); /* >= 1 */
428 if ( strend - s > slen || strend - s < slen - 1
429 || (strend - s == slen && strend[-1] != '\n')) {
430 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
433 /* Now should match s[0..slen-2] */
435 if (slen && (*SvPVX_const(check) != *s
437 && memNE(SvPVX_const(check), s, slen)))) {
439 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
443 else if (*SvPVX_const(check) != *s
444 || ((slen = SvCUR(check)) > 1
445 && memNE(SvPVX_const(check), s, slen)))
448 goto success_at_start;
451 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
453 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
454 end_shift = prog->check_end_shift;
457 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
458 - (SvTAIL(check) != 0);
459 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
461 if (end_shift < eshift)
465 else { /* Can match at random position */
468 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
469 end_shift = prog->check_end_shift;
471 /* end shift should be non negative here */
474 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
476 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
477 (IV)end_shift, prog->precomp);
481 /* Find a possible match in the region s..strend by looking for
482 the "check" substring in the region corrected by start/end_shift. */
485 I32 srch_start_shift = start_shift;
486 I32 srch_end_shift = end_shift;
487 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
488 srch_end_shift -= ((strbeg - s) - srch_start_shift);
489 srch_start_shift = strbeg - s;
492 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
493 (IV)prog->check_offset_min,
494 (IV)srch_start_shift,
496 (IV)prog->check_end_shift);
499 if (flags & REXEC_SCREAM) {
500 I32 p = -1; /* Internal iterator of scream. */
501 I32 * const pp = data ? data->scream_pos : &p;
503 if (PL_screamfirst[BmRARE(check)] >= 0
504 || ( BmRARE(check) == '\n'
505 && (BmPREVIOUS(check) == SvCUR(check) - 1)
507 s = screaminstr(sv, check,
508 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
511 /* we may be pointing at the wrong string */
512 if (s && RX_MATCH_COPIED(prog))
513 s = strbeg + (s - SvPVX_const(sv));
515 *data->scream_olds = s;
520 if (prog->reganch & ROPT_CANY_SEEN) {
521 start_point= (U8*)(s + srch_start_shift);
522 end_point= (U8*)(strend - srch_end_shift);
524 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
525 end_point= HOP3(strend, -srch_end_shift, strbeg);
528 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
529 (int)(end_point - start_point),
530 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
534 s = fbm_instr( start_point, end_point,
535 check, multiline ? FBMrf_MULTILINE : 0);
538 /* Update the count-of-usability, remove useless subpatterns,
542 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
543 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
544 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
545 (s ? "Found" : "Did not find"),
546 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
547 ? "anchored" : "floating"),
550 (s ? " at offset " : "...\n") );
555 /* Finish the diagnostic message */
556 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
558 /* XXX dmq: first branch is for positive lookbehind...
559 Our check string is offset from the beginning of the pattern.
560 So we need to do any stclass tests offset forward from that
569 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
570 Start with the other substr.
571 XXXX no SCREAM optimization yet - and a very coarse implementation
572 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
573 *always* match. Probably should be marked during compile...
574 Probably it is right to do no SCREAM here...
577 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
578 : (prog->float_substr && prog->anchored_substr))
580 /* Take into account the "other" substring. */
581 /* XXXX May be hopelessly wrong for UTF... */
584 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
587 char * const last = HOP3c(s, -start_shift, strbeg);
589 char * const saved_s = s;
592 t = s - prog->check_offset_max;
593 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
595 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
600 t = HOP3c(t, prog->anchored_offset, strend);
601 if (t < other_last) /* These positions already checked */
603 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
606 /* XXXX It is not documented what units *_offsets are in.
607 We assume bytes, but this is clearly wrong.
608 Meaning this code needs to be carefully reviewed for errors.
612 /* On end-of-str: see comment below. */
613 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
614 if (must == &PL_sv_undef) {
616 DEBUG_r(must = prog->anchored_utf8); /* for debug */
621 HOP3(HOP3(last1, prog->anchored_offset, strend)
622 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
624 multiline ? FBMrf_MULTILINE : 0
627 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
628 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
629 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
630 (s ? "Found" : "Contradicts"),
631 quoted, RE_SV_TAIL(must));
636 if (last1 >= last2) {
637 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
638 ", giving up...\n"));
641 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
642 ", trying floating at offset %ld...\n",
643 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
644 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
645 s = HOP3c(last, 1, strend);
649 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
650 (long)(s - i_strpos)));
651 t = HOP3c(s, -prog->anchored_offset, strbeg);
652 other_last = HOP3c(s, 1, strend);
660 else { /* Take into account the floating substring. */
662 char * const saved_s = s;
665 t = HOP3c(s, -start_shift, strbeg);
667 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
668 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
669 last = HOP3c(t, prog->float_max_offset, strend);
670 s = HOP3c(t, prog->float_min_offset, strend);
673 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
674 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
675 /* fbm_instr() takes into account exact value of end-of-str
676 if the check is SvTAIL(ed). Since false positives are OK,
677 and end-of-str is not later than strend we are OK. */
678 if (must == &PL_sv_undef) {
680 DEBUG_r(must = prog->float_utf8); /* for debug message */
683 s = fbm_instr((unsigned char*)s,
684 (unsigned char*)last + SvCUR(must)
686 must, multiline ? FBMrf_MULTILINE : 0);
688 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
689 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
690 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
691 (s ? "Found" : "Contradicts"),
692 quoted, RE_SV_TAIL(must));
696 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
697 ", giving up...\n"));
700 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
701 ", trying anchored starting at offset %ld...\n",
702 (long)(saved_s + 1 - i_strpos)));
704 s = HOP3c(t, 1, strend);
708 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
709 (long)(s - i_strpos)));
710 other_last = s; /* Fix this later. --Hugo */
720 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
723 PerlIO_printf(Perl_debug_log,
724 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
725 (IV)prog->check_offset_min,
726 (IV)prog->check_offset_max,
734 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
736 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
739 /* Fixed substring is found far enough so that the match
740 cannot start at strpos. */
742 if (ml_anch && t[-1] != '\n') {
743 /* Eventually fbm_*() should handle this, but often
744 anchored_offset is not 0, so this check will not be wasted. */
745 /* XXXX In the code below we prefer to look for "^" even in
746 presence of anchored substrings. And we search even
747 beyond the found float position. These pessimizations
748 are historical artefacts only. */
750 while (t < strend - prog->minlen) {
752 if (t < check_at - prog->check_offset_min) {
753 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
754 /* Since we moved from the found position,
755 we definitely contradict the found anchored
756 substr. Due to the above check we do not
757 contradict "check" substr.
758 Thus we can arrive here only if check substr
759 is float. Redo checking for "other"=="fixed".
762 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
763 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
764 goto do_other_anchored;
766 /* We don't contradict the found floating substring. */
767 /* XXXX Why not check for STCLASS? */
769 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
770 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
773 /* Position contradicts check-string */
774 /* XXXX probably better to look for check-string
775 than for "\n", so one should lower the limit for t? */
776 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
777 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
778 other_last = strpos = s = t + 1;
783 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
784 PL_colors[0], PL_colors[1]));
788 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
789 PL_colors[0], PL_colors[1]));
793 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
796 /* The found string does not prohibit matching at strpos,
797 - no optimization of calling REx engine can be performed,
798 unless it was an MBOL and we are not after MBOL,
799 or a future STCLASS check will fail this. */
801 /* Even in this situation we may use MBOL flag if strpos is offset
802 wrt the start of the string. */
803 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
804 && (strpos != strbeg) && strpos[-1] != '\n'
805 /* May be due to an implicit anchor of m{.*foo} */
806 && !(prog->reganch & ROPT_IMPLICIT))
811 DEBUG_EXECUTE_r( if (ml_anch)
812 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
813 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
816 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
818 prog->check_utf8 /* Could be deleted already */
819 && --BmUSEFUL(prog->check_utf8) < 0
820 && (prog->check_utf8 == prog->float_utf8)
822 prog->check_substr /* Could be deleted already */
823 && --BmUSEFUL(prog->check_substr) < 0
824 && (prog->check_substr == prog->float_substr)
827 /* If flags & SOMETHING - do not do it many times on the same match */
828 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
829 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
830 if (do_utf8 ? prog->check_substr : prog->check_utf8)
831 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
832 prog->check_substr = prog->check_utf8 = NULL; /* disable */
833 prog->float_substr = prog->float_utf8 = NULL; /* clear */
834 check = NULL; /* abort */
836 /* XXXX This is a remnant of the old implementation. It
837 looks wasteful, since now INTUIT can use many
839 prog->reganch &= ~RE_USE_INTUIT;
846 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
847 /* trie stclasses are too expensive to use here, we are better off to
848 leave it to regmatch itself */
849 if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) {
850 /* minlen == 0 is possible if regstclass is \b or \B,
851 and the fixed substr is ''$.
852 Since minlen is already taken into account, s+1 is before strend;
853 accidentally, minlen >= 1 guaranties no false positives at s + 1
854 even for \b or \B. But (minlen? 1 : 0) below assumes that
855 regstclass does not come from lookahead... */
856 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
857 This leaves EXACTF only, which is dealt with in find_byclass(). */
858 const U8* const str = (U8*)STRING(prog->regstclass);
859 const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT
860 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
863 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
864 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
865 else if (prog->float_substr || prog->float_utf8)
866 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
870 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n",
871 (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg));
874 s = find_byclass(prog, prog->regstclass, s, endpos, NULL);
877 const char *what = NULL;
879 if (endpos == strend) {
880 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
881 "Could not match STCLASS...\n") );
884 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
885 "This position contradicts STCLASS...\n") );
886 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
888 /* Contradict one of substrings */
889 if (prog->anchored_substr || prog->anchored_utf8) {
890 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
891 DEBUG_EXECUTE_r( what = "anchored" );
893 s = HOP3c(t, 1, strend);
894 if (s + start_shift + end_shift > strend) {
895 /* XXXX Should be taken into account earlier? */
896 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
897 "Could not match STCLASS...\n") );
902 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
903 "Looking for %s substr starting at offset %ld...\n",
904 what, (long)(s + start_shift - i_strpos)) );
907 /* Have both, check_string is floating */
908 if (t + start_shift >= check_at) /* Contradicts floating=check */
909 goto retry_floating_check;
910 /* Recheck anchored substring, but not floating... */
914 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
915 "Looking for anchored substr starting at offset %ld...\n",
916 (long)(other_last - i_strpos)) );
917 goto do_other_anchored;
919 /* Another way we could have checked stclass at the
920 current position only: */
925 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
926 "Looking for /%s^%s/m starting at offset %ld...\n",
927 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
930 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
932 /* Check is floating subtring. */
933 retry_floating_check:
934 t = check_at - start_shift;
935 DEBUG_EXECUTE_r( what = "floating" );
936 goto hop_and_restart;
939 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
940 "By STCLASS: moving %ld --> %ld\n",
941 (long)(t - i_strpos), (long)(s - i_strpos))
945 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
946 "Does not contradict STCLASS...\n");
951 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
952 PL_colors[4], (check ? "Guessed" : "Giving up"),
953 PL_colors[5], (long)(s - i_strpos)) );
956 fail_finish: /* Substring not found */
957 if (prog->check_substr || prog->check_utf8) /* could be removed already */
958 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
960 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
961 PL_colors[4], PL_colors[5]));
967 #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \
968 foldlen, foldbuf, uniflags) STMT_START { \
969 switch (trie_type) { \
970 case trie_utf8_fold: \
972 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
977 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
978 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
979 foldlen -= UNISKIP( uvc ); \
980 uscan = foldbuf + UNISKIP( uvc ); \
984 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
992 charid = trie->charmap[ uvc ]; \
996 if (trie->widecharmap) { \
997 SV** const svpp = hv_fetch(trie->widecharmap, \
998 (char*)&uvc, sizeof(UV), 0); \
1000 charid = (U16)SvIV(*svpp); \
1005 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1008 ibcmp_utf8(s, NULL, 0, do_utf8, \
1009 m, NULL, ln, (bool)UTF)) \
1010 && (!reginfo || regtry(reginfo, s)) ) \
1013 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1014 uvchr_to_utf8(tmpbuf, c); \
1015 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1017 && (f == c1 || f == c2) \
1018 && (ln == foldlen || \
1019 !ibcmp_utf8((char *) foldbuf, \
1020 NULL, foldlen, do_utf8, \
1022 NULL, ln, (bool)UTF)) \
1023 && (!reginfo || regtry(reginfo, s)) ) \
1028 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1032 && (ln == 1 || !(OP(c) == EXACTF \
1034 : ibcmp_locale(s, m, ln))) \
1035 && (!reginfo || regtry(reginfo, s)) ) \
1041 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1043 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1049 #define REXEC_FBC_SCAN(CoDe) \
1051 while (s < strend) { \
1057 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1058 REXEC_FBC_UTF8_SCAN( \
1060 if (tmp && (!reginfo || regtry(reginfo, s))) \
1069 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1072 if (tmp && (!reginfo || regtry(reginfo, s))) \
1081 #define REXEC_FBC_TRYIT \
1082 if ((!reginfo || regtry(reginfo, s))) \
1085 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1088 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1091 REXEC_FBC_CLASS_SCAN(CoNd); \
1095 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1096 PL_reg_flags |= RF_tainted; \
1098 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1101 REXEC_FBC_CLASS_SCAN(CoNd); \
1105 #define DUMP_EXEC_POS(li,s,doutf8) \
1106 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1108 /* We know what class REx starts with. Try to find this position... */
1109 /* if reginfo is NULL, its a dryrun */
1110 /* annoyingly all the vars in this routine have different names from their counterparts
1111 in regmatch. /grrr */
1114 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1115 const char *strend, const regmatch_info *reginfo)
1118 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
1122 register STRLEN uskip;
1126 register I32 tmp = 1; /* Scratch variable? */
1127 register const bool do_utf8 = PL_reg_match_utf8;
1129 /* We know what class it must start with. */
1133 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1134 !UTF8_IS_INVARIANT((U8)s[0]) ?
1135 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1136 REGINCLASS(prog, c, (U8*)s));
1139 while (s < strend) {
1142 if (REGINCLASS(prog, c, (U8*)s) ||
1143 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1144 /* The assignment of 2 is intentional:
1145 * for the folded sharp s, the skip is 2. */
1146 (skip = SHARP_S_SKIP))) {
1147 if (tmp && (!reginfo || regtry(reginfo, s)))
1160 if (tmp && (!reginfo || regtry(reginfo, s)))
1168 ln = STR_LEN(c); /* length to match in octets/bytes */
1169 lnc = (I32) ln; /* length to match in characters */
1171 STRLEN ulen1, ulen2;
1173 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1174 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1175 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1177 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1178 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1180 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1182 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1185 while (sm < ((U8 *) m + ln)) {
1200 c2 = PL_fold_locale[c1];
1202 e = HOP3c(strend, -((I32)lnc), s);
1204 if (!reginfo && e < s)
1205 e = s; /* Due to minlen logic of intuit() */
1207 /* The idea in the EXACTF* cases is to first find the
1208 * first character of the EXACTF* node and then, if
1209 * necessary, case-insensitively compare the full
1210 * text of the node. The c1 and c2 are the first
1211 * characters (though in Unicode it gets a bit
1212 * more complicated because there are more cases
1213 * than just upper and lower: one needs to use
1214 * the so-called folding case for case-insensitive
1215 * matching (called "loose matching" in Unicode).
1216 * ibcmp_utf8() will do just that. */
1220 U8 tmpbuf [UTF8_MAXBYTES+1];
1221 STRLEN len, foldlen;
1222 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1224 /* Upper and lower of 1st char are equal -
1225 * probably not a "letter". */
1227 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1229 REXEC_FBC_EXACTISH_CHECK(c == c1);
1234 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1237 /* Handle some of the three Greek sigmas cases.
1238 * Note that not all the possible combinations
1239 * are handled here: some of them are handled
1240 * by the standard folding rules, and some of
1241 * them (the character class or ANYOF cases)
1242 * are handled during compiletime in
1243 * regexec.c:S_regclass(). */
1244 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1245 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1246 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1248 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1254 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1256 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1260 PL_reg_flags |= RF_tainted;
1267 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1268 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1270 tmp = ((OP(c) == BOUND ?
1271 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1272 LOAD_UTF8_CHARCLASS_ALNUM();
1273 REXEC_FBC_UTF8_SCAN(
1274 if (tmp == !(OP(c) == BOUND ?
1275 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1276 isALNUM_LC_utf8((U8*)s)))
1284 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1285 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1288 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1294 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, s)))
1298 PL_reg_flags |= RF_tainted;
1305 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1306 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1308 tmp = ((OP(c) == NBOUND ?
1309 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1310 LOAD_UTF8_CHARCLASS_ALNUM();
1311 REXEC_FBC_UTF8_SCAN(
1312 if (tmp == !(OP(c) == NBOUND ?
1313 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1314 isALNUM_LC_utf8((U8*)s)))
1316 else REXEC_FBC_TRYIT;
1320 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1321 tmp = ((OP(c) == NBOUND ?
1322 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1325 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1327 else REXEC_FBC_TRYIT;
1330 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, s)))
1334 REXEC_FBC_CSCAN_PRELOAD(
1335 LOAD_UTF8_CHARCLASS_ALNUM(),
1336 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1340 REXEC_FBC_CSCAN_TAINT(
1341 isALNUM_LC_utf8((U8*)s),
1345 REXEC_FBC_CSCAN_PRELOAD(
1346 LOAD_UTF8_CHARCLASS_ALNUM(),
1347 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1351 REXEC_FBC_CSCAN_TAINT(
1352 !isALNUM_LC_utf8((U8*)s),
1356 REXEC_FBC_CSCAN_PRELOAD(
1357 LOAD_UTF8_CHARCLASS_SPACE(),
1358 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1362 REXEC_FBC_CSCAN_TAINT(
1363 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1367 REXEC_FBC_CSCAN_PRELOAD(
1368 LOAD_UTF8_CHARCLASS_SPACE(),
1369 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1373 REXEC_FBC_CSCAN_TAINT(
1374 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1378 REXEC_FBC_CSCAN_PRELOAD(
1379 LOAD_UTF8_CHARCLASS_DIGIT(),
1380 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1384 REXEC_FBC_CSCAN_TAINT(
1385 isDIGIT_LC_utf8((U8*)s),
1389 REXEC_FBC_CSCAN_PRELOAD(
1390 LOAD_UTF8_CHARCLASS_DIGIT(),
1391 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1395 REXEC_FBC_CSCAN_TAINT(
1396 !isDIGIT_LC_utf8((U8*)s),
1402 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1403 trie_type = do_utf8 ?
1404 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1406 /* what trie are we using right now */
1408 = (reg_ac_data*)prog->data->data[ ARG( c ) ];
1409 reg_trie_data *trie=aho->trie;
1411 const char *last_start = strend - trie->minlen;
1413 const char *real_start = s;
1415 STRLEN maxlen = trie->maxlen;
1417 U8 **points; /* map of where we were in the input string
1418 when reading a given char. For ASCII this
1419 is unnecessary overhead as the relationship
1420 is always 1:1, but for unicode, especially
1421 case folded unicode this is not true. */
1422 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1426 GET_RE_DEBUG_FLAGS_DECL;
1428 /* We can't just allocate points here. We need to wrap it in
1429 * an SV so it gets freed properly if there is a croak while
1430 * running the match */
1433 sv_points=newSV(maxlen * sizeof(U8 *));
1434 SvCUR_set(sv_points,
1435 maxlen * sizeof(U8 *));
1436 SvPOK_on(sv_points);
1437 sv_2mortal(sv_points);
1438 points=(U8**)SvPV_nolen(sv_points );
1439 if ( trie_type != trie_utf8_fold
1440 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1443 bitmap=(U8*)trie->bitmap;
1445 bitmap=(U8*)ANYOF_BITMAP(c);
1447 /* this is the Aho-Corasick algorithm modified a touch
1448 to include special handling for long "unknown char"
1449 sequences. The basic idea being that we use AC as long
1450 as we are dealing with a possible matching char, when
1451 we encounter an unknown char (and we have not encountered
1452 an accepting state) we scan forward until we find a legal
1454 AC matching is basically that of trie matching, except
1455 that when we encounter a failing transition, we fall back
1456 to the current states "fail state", and try the current char
1457 again, a process we repeat until we reach the root state,
1458 state 1, or a legal transition. If we fail on the root state
1459 then we can either terminate if we have reached an accepting
1460 state previously, or restart the entire process from the beginning
1464 while (s <= last_start) {
1465 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1473 U8 *uscan = (U8*)NULL;
1474 U8 *leftmost = NULL;
1476 U32 accepted_word= 0;
1480 while ( state && uc <= (U8*)strend ) {
1482 U32 word = aho->states[ state ].wordnum;
1486 DEBUG_TRIE_EXECUTE_r(
1487 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1488 dump_exec_pos( (char *)uc, c, strend, real_start,
1489 (char *)uc, do_utf8 );
1490 PerlIO_printf( Perl_debug_log,
1491 " Scanning for legal start char...\n");
1494 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1499 if (uc >(U8*)last_start) break;
1503 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1504 if (!leftmost || lpos < leftmost) {
1505 DEBUG_r(accepted_word=word);
1511 points[pointpos++ % maxlen]= uc;
1512 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
1513 uvc, charid, foldlen, foldbuf, uniflags);
1514 DEBUG_TRIE_EXECUTE_r({
1515 dump_exec_pos( (char *)uc, c, strend, real_start,
1517 PerlIO_printf(Perl_debug_log,
1518 " Charid:%3u CP:%4"UVxf" ",
1524 word = aho->states[ state ].wordnum;
1526 base = aho->states[ state ].trans.base;
1528 DEBUG_TRIE_EXECUTE_r({
1530 dump_exec_pos( (char *)uc, c, strend, real_start,
1532 PerlIO_printf( Perl_debug_log,
1533 "%sState: %4"UVxf", word=%"UVxf,
1534 failed ? " Fail transition to " : "",
1535 (UV)state, (UV)word);
1540 (base + charid > trie->uniquecharcount )
1541 && (base + charid - 1 - trie->uniquecharcount
1543 && trie->trans[base + charid - 1 -
1544 trie->uniquecharcount].check == state
1545 && (tmp=trie->trans[base + charid - 1 -
1546 trie->uniquecharcount ].next))
1548 DEBUG_TRIE_EXECUTE_r(
1549 PerlIO_printf( Perl_debug_log," - legal\n"));
1554 DEBUG_TRIE_EXECUTE_r(
1555 PerlIO_printf( Perl_debug_log," - fail\n"));
1557 state = aho->fail[state];
1561 /* we must be accepting here */
1562 DEBUG_TRIE_EXECUTE_r(
1563 PerlIO_printf( Perl_debug_log," - accepting\n"));
1572 if (!state) state = 1;
1575 if ( aho->states[ state ].wordnum ) {
1576 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1577 if (!leftmost || lpos < leftmost) {
1578 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1583 s = (char*)leftmost;
1584 DEBUG_TRIE_EXECUTE_r({
1586 Perl_debug_log,"Matches word #%"UVxf" at position %d. Trying full pattern...\n",
1587 (UV)accepted_word, s - real_start
1590 if (!reginfo || regtry(reginfo, s)) {
1596 DEBUG_TRIE_EXECUTE_r({
1597 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1600 DEBUG_TRIE_EXECUTE_r(
1601 PerlIO_printf( Perl_debug_log,"No match.\n"));
1610 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1619 - regexec_flags - match a regexp against a string
1622 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1623 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1624 /* strend: pointer to null at end of string */
1625 /* strbeg: real beginning of string */
1626 /* minend: end of match must be >=minend after stringarg. */
1627 /* data: May be used for some additional optimizations. */
1628 /* nosave: For optimizations. */
1632 register regnode *c;
1633 register char *startpos = stringarg;
1634 I32 minlen; /* must match at least this many chars */
1635 I32 dontbother = 0; /* how many characters not to try at end */
1636 I32 end_shift = 0; /* Same for the end. */ /* CC */
1637 I32 scream_pos = -1; /* Internal iterator of scream. */
1638 char *scream_olds = NULL;
1639 SV* const oreplsv = GvSV(PL_replgv);
1640 const bool do_utf8 = DO_UTF8(sv);
1643 regmatch_info reginfo; /* create some info to pass to regtry etc */
1645 GET_RE_DEBUG_FLAGS_DECL;
1647 PERL_UNUSED_ARG(data);
1649 /* Be paranoid... */
1650 if (prog == NULL || startpos == NULL) {
1651 Perl_croak(aTHX_ "NULL regexp parameter");
1655 multiline = prog->reganch & PMf_MULTILINE;
1656 reginfo.prog = prog;
1658 RX_MATCH_UTF8_set(prog, do_utf8);
1660 debug_start_match(prog, do_utf8, startpos, strend,
1664 minlen = prog->minlen;
1666 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1667 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1668 "String too short [regexec_flags]...\n"));
1673 /* Check validity of program. */
1674 if (UCHARAT(prog->program) != REG_MAGIC) {
1675 Perl_croak(aTHX_ "corrupted regexp program");
1679 PL_reg_eval_set = 0;
1682 if (prog->reganch & ROPT_UTF8)
1683 PL_reg_flags |= RF_utf8;
1685 /* Mark beginning of line for ^ and lookbehind. */
1686 reginfo.bol = startpos; /* XXX not used ??? */
1690 /* Mark end of line for $ (and such) */
1693 /* see how far we have to get to not match where we matched before */
1694 reginfo.till = startpos+minend;
1696 /* If there is a "must appear" string, look for it. */
1699 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to set reginfo->ganch */
1702 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1703 reginfo.ganch = startpos;
1704 else if (sv && SvTYPE(sv) >= SVt_PVMG
1706 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1707 && mg->mg_len >= 0) {
1708 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1709 if (prog->reganch & ROPT_ANCH_GPOS) {
1710 if (s > reginfo.ganch)
1715 else /* pos() not defined */
1716 reginfo.ganch = strbeg;
1719 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1720 re_scream_pos_data d;
1722 d.scream_olds = &scream_olds;
1723 d.scream_pos = &scream_pos;
1724 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1726 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1727 goto phooey; /* not present */
1733 /* Simplest case: anchored match need be tried only once. */
1734 /* [unless only anchor is BOL and multiline is set] */
1735 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1736 if (s == startpos && regtry(®info, startpos))
1738 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1739 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1744 dontbother = minlen - 1;
1745 end = HOP3c(strend, -dontbother, strbeg) - 1;
1746 /* for multiline we only have to try after newlines */
1747 if (prog->check_substr || prog->check_utf8) {
1751 if (regtry(®info, s))
1756 if (prog->reganch & RE_USE_INTUIT) {
1757 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1768 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1769 if (regtry(®info, s))
1776 } else if (prog->reganch & ROPT_ANCH_GPOS) {
1777 if (regtry(®info, reginfo.ganch))
1782 /* Messy cases: unanchored match. */
1783 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1784 /* we have /x+whatever/ */
1785 /* it must be a one character string (XXXX Except UTF?) */
1790 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1791 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1792 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1797 DEBUG_EXECUTE_r( did_match = 1 );
1798 if (regtry(®info, s)) goto got_it;
1800 while (s < strend && *s == ch)
1808 DEBUG_EXECUTE_r( did_match = 1 );
1809 if (regtry(®info, s)) goto got_it;
1811 while (s < strend && *s == ch)
1816 DEBUG_EXECUTE_r(if (!did_match)
1817 PerlIO_printf(Perl_debug_log,
1818 "Did not find anchored character...\n")
1821 else if (prog->anchored_substr != NULL
1822 || prog->anchored_utf8 != NULL
1823 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1824 && prog->float_max_offset < strend - s)) {
1829 char *last1; /* Last position checked before */
1833 if (prog->anchored_substr || prog->anchored_utf8) {
1834 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1835 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1836 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1837 back_max = back_min = prog->anchored_offset;
1839 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1840 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1841 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1842 back_max = prog->float_max_offset;
1843 back_min = prog->float_min_offset;
1847 if (must == &PL_sv_undef)
1848 /* could not downgrade utf8 check substring, so must fail */
1854 last = HOP3c(strend, /* Cannot start after this */
1855 -(I32)(CHR_SVLEN(must)
1856 - (SvTAIL(must) != 0) + back_min), strbeg);
1859 last1 = HOPc(s, -1);
1861 last1 = s - 1; /* bogus */
1863 /* XXXX check_substr already used to find "s", can optimize if
1864 check_substr==must. */
1866 dontbother = end_shift;
1867 strend = HOPc(strend, -dontbother);
1868 while ( (s <= last) &&
1869 ((flags & REXEC_SCREAM)
1870 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1871 end_shift, &scream_pos, 0))
1872 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1873 (unsigned char*)strend, must,
1874 multiline ? FBMrf_MULTILINE : 0))) ) {
1875 /* we may be pointing at the wrong string */
1876 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1877 s = strbeg + (s - SvPVX_const(sv));
1878 DEBUG_EXECUTE_r( did_match = 1 );
1879 if (HOPc(s, -back_max) > last1) {
1880 last1 = HOPc(s, -back_min);
1881 s = HOPc(s, -back_max);
1884 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1886 last1 = HOPc(s, -back_min);
1890 while (s <= last1) {
1891 if (regtry(®info, s))
1897 while (s <= last1) {
1898 if (regtry(®info, s))
1904 DEBUG_EXECUTE_r(if (!did_match) {
1905 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1906 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1907 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1908 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1909 ? "anchored" : "floating"),
1910 quoted, RE_SV_TAIL(must));
1914 else if ( (c = prog->regstclass) ) {
1916 const OPCODE op = OP(prog->regstclass);
1917 /* don't bother with what can't match */
1918 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1919 strend = HOPc(strend, -(minlen - 1));
1922 SV * const prop = sv_newmortal();
1923 regprop(prog, prop, c);
1925 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1927 PerlIO_printf(Perl_debug_log,
1928 "Matching stclass %.*s against %s (%d chars)\n",
1929 (int)SvCUR(prop), SvPVX_const(prop),
1930 quoted, (int)(strend - s));
1933 if (find_byclass(prog, c, s, strend, ®info))
1935 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
1939 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1944 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1945 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1946 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1948 if (flags & REXEC_SCREAM) {
1949 last = screaminstr(sv, float_real, s - strbeg,
1950 end_shift, &scream_pos, 1); /* last one */
1952 last = scream_olds; /* Only one occurrence. */
1953 /* we may be pointing at the wrong string */
1954 else if (RX_MATCH_COPIED(prog))
1955 s = strbeg + (s - SvPVX_const(sv));
1959 const char * const little = SvPV_const(float_real, len);
1961 if (SvTAIL(float_real)) {
1962 if (memEQ(strend - len + 1, little, len - 1))
1963 last = strend - len + 1;
1964 else if (!multiline)
1965 last = memEQ(strend - len, little, len)
1966 ? strend - len : NULL;
1972 last = rninstr(s, strend, little, little + len);
1974 last = strend; /* matching "$" */
1978 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1979 "%sCan't trim the tail, match fails (should not happen)%s\n",
1980 PL_colors[4], PL_colors[5]));
1981 goto phooey; /* Should not happen! */
1983 dontbother = strend - last + prog->float_min_offset;
1985 if (minlen && (dontbother < minlen))
1986 dontbother = minlen - 1;
1987 strend -= dontbother; /* this one's always in bytes! */
1988 /* We don't know much -- general case. */
1991 if (regtry(®info, s))
2000 if (regtry(®info, s))
2002 } while (s++ < strend);
2010 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2012 if (PL_reg_eval_set) {
2013 /* Preserve the current value of $^R */
2014 if (oreplsv != GvSV(PL_replgv))
2015 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2016 restored, the value remains
2018 restore_pos(aTHX_ prog);
2021 /* make sure $`, $&, $', and $digit will work later */
2022 if ( !(flags & REXEC_NOT_FIRST) ) {
2023 RX_MATCH_COPY_FREE(prog);
2024 if (flags & REXEC_COPY_STR) {
2025 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2026 #ifdef PERL_OLD_COPY_ON_WRITE
2028 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2030 PerlIO_printf(Perl_debug_log,
2031 "Copy on write: regexp capture, type %d\n",
2034 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2035 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2036 assert (SvPOKp(prog->saved_copy));
2040 RX_MATCH_COPIED_on(prog);
2041 s = savepvn(strbeg, i);
2047 prog->subbeg = strbeg;
2048 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2055 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2056 PL_colors[4], PL_colors[5]));
2057 if (PL_reg_eval_set)
2058 restore_pos(aTHX_ prog);
2063 - regtry - try match at specific point
2065 STATIC I32 /* 0 failure, 1 success */
2066 S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos)
2072 regexp *prog = reginfo->prog;
2073 GET_RE_DEBUG_FLAGS_DECL;
2076 PL_regindent = 0; /* XXXX Not good when matches are reenterable... */
2078 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2081 PL_reg_eval_set = RS_init;
2082 DEBUG_EXECUTE_r(DEBUG_s(
2083 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2084 (IV)(PL_stack_sp - PL_stack_base));
2086 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2087 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2088 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2090 /* Apparently this is not needed, judging by wantarray. */
2091 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2092 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2095 /* Make $_ available to executed code. */
2096 if (reginfo->sv != DEFSV) {
2098 DEFSV = reginfo->sv;
2101 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2102 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2103 /* prepare for quick setting of pos */
2104 #ifdef PERL_OLD_COPY_ON_WRITE
2106 sv_force_normal_flags(sv, 0);
2108 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2109 &PL_vtbl_mglob, NULL, 0);
2113 PL_reg_oldpos = mg->mg_len;
2114 SAVEDESTRUCTOR_X(restore_pos, prog);
2116 if (!PL_reg_curpm) {
2117 Newxz(PL_reg_curpm, 1, PMOP);
2120 SV* const repointer = newSViv(0);
2121 /* so we know which PL_regex_padav element is PL_reg_curpm */
2122 SvFLAGS(repointer) |= SVf_BREAK;
2123 av_push(PL_regex_padav,repointer);
2124 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2125 PL_regex_pad = AvARRAY(PL_regex_padav);
2129 PM_SETRE(PL_reg_curpm, prog);
2130 PL_reg_oldcurpm = PL_curpm;
2131 PL_curpm = PL_reg_curpm;
2132 if (RX_MATCH_COPIED(prog)) {
2133 /* Here is a serious problem: we cannot rewrite subbeg,
2134 since it may be needed if this match fails. Thus
2135 $` inside (?{}) could fail... */
2136 PL_reg_oldsaved = prog->subbeg;
2137 PL_reg_oldsavedlen = prog->sublen;
2138 #ifdef PERL_OLD_COPY_ON_WRITE
2139 PL_nrs = prog->saved_copy;
2141 RX_MATCH_COPIED_off(prog);
2144 PL_reg_oldsaved = NULL;
2145 prog->subbeg = PL_bostr;
2146 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2148 prog->startp[0] = startpos - PL_bostr;
2149 PL_reginput = startpos;
2150 PL_regstartp = prog->startp;
2151 PL_regendp = prog->endp;
2152 PL_reglastparen = &prog->lastparen;
2153 PL_reglastcloseparen = &prog->lastcloseparen;
2154 prog->lastparen = 0;
2155 prog->lastcloseparen = 0;
2157 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2158 if (PL_reg_start_tmpl <= prog->nparens) {
2159 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2160 if(PL_reg_start_tmp)
2161 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2163 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2166 /* XXXX What this code is doing here?!!! There should be no need
2167 to do this again and again, PL_reglastparen should take care of
2170 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2171 * Actually, the code in regcppop() (which Ilya may be meaning by
2172 * PL_reglastparen), is not needed at all by the test suite
2173 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2174 * enough, for building DynaLoader, or otherwise this
2175 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2176 * will happen. Meanwhile, this code *is* needed for the
2177 * above-mentioned test suite tests to succeed. The common theme
2178 * on those tests seems to be returning null fields from matches.
2183 if (prog->nparens) {
2185 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2192 if (regmatch(reginfo, prog->program + 1)) {
2193 prog->endp[0] = PL_reginput - PL_bostr;
2196 REGCP_UNWIND(lastcp);
2201 #define sayYES goto yes
2202 #define sayNO goto no
2203 #define sayNO_ANYOF goto no_anyof
2204 #define sayYES_FINAL goto yes_final
2205 #define sayNO_FINAL goto no_final
2206 #define sayNO_SILENT goto do_no
2207 #define saySAME(x) if (x) goto yes; else goto no
2209 #define CACHEsayNO STMT_START { \
2210 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) \
2211 PL_reg_poscache[st->u.whilem.cache_offset] |= \
2212 (1<<st->u.whilem.cache_bit); \
2217 /* this is used to determine how far from the left messages like
2218 'failed...' are printed. Currently 29 makes these messages line
2219 up with the opcode they refer to. Earlier perls used 25 which
2220 left these messages outdented making reviewing a debug output
2223 #define REPORT_CODE_OFF 29
2226 /* Make sure there is a test for this +1 options in re_tests */
2227 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2229 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2230 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2232 #define SLAB_FIRST(s) (&(s)->states[0])
2233 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2235 /* grab a new slab and return the first slot in it */
2237 STATIC regmatch_state *
2240 #if PERL_VERSION < 9
2243 regmatch_slab *s = PL_regmatch_slab->next;
2245 Newx(s, 1, regmatch_slab);
2246 s->prev = PL_regmatch_slab;
2248 PL_regmatch_slab->next = s;
2250 PL_regmatch_slab = s;
2251 return SLAB_FIRST(s);
2254 /* simulate a recursive call to regmatch */
2256 #define REGMATCH(ns, where) \
2259 st->resume_state = resume_##where; \
2260 goto start_recurse; \
2261 resume_point_##where:
2263 /* push a new state then goto it */
2265 #define PUSH_STATE_GOTO(state, node) \
2267 st->resume_state = state; \
2270 /* push a new state with success backtracking, then goto it */
2272 #define PUSH_YES_STATE_GOTO(state, node) \
2274 st->resume_state = state; \
2275 goto push_yes_state;
2280 - regmatch - main matching routine
2282 * Conceptually the strategy is simple: check to see whether the current
2283 * node matches, call self recursively to see whether the rest matches,
2284 * and then act accordingly. In practice we make some effort to avoid
2285 * recursion, in particular by going through "ordinary" nodes (that don't
2286 * need to know whether the rest of the match failed) by a loop instead of
2289 /* [lwall] I've hoisted the register declarations to the outer block in order to
2290 * maybe save a little bit of pushing and popping on the stack. It also takes
2291 * advantage of machines that use a register save mask on subroutine entry.
2293 * This function used to be heavily recursive, but since this had the
2294 * effect of blowing the CPU stack on complex regexes, it has been
2295 * restructured to be iterative, and to save state onto the heap rather
2296 * than the stack. Essentially whereever regmatch() used to be called, it
2297 * pushes the current state, notes where to return, then jumps back into
2300 * Originally the structure of this function used to look something like
2305 while (scan != NULL) {
2306 a++; // do stuff with a and b
2312 if (regmatch(...)) // recurse
2322 * Now it looks something like this:
2330 regmatch_state *st = new();
2332 st->a++; // do stuff with a and b
2334 while (scan != NULL) {
2342 st->resume_state = resume_FOO;
2343 goto start_recurse; // recurse
2352 st = new(); push a new state
2353 st->a = 1; st->b = 2;
2360 switch (resume_state) {
2362 goto resume_point_FOO;
2369 * WARNING: this means that any line in this function that contains a
2370 * REGMATCH() or TRYPAREN() is actually simulating a recursive call to
2371 * regmatch() using gotos instead. Thus the values of any local variables
2372 * not saved in the regmatch_state structure will have been lost when
2373 * execution resumes on the next line .
2375 * States (ie the st pointer) are allocated in slabs of about 4K in size.
2376 * PL_regmatch_state always points to the currently active state, and
2377 * PL_regmatch_slab points to the slab currently containing PL_regmatch_state.
2378 * The first time regmatch is called, the first slab is allocated, and is
2379 * never freed until interpreter desctruction. When the slab is full,
2380 * a new one is allocated chained to the end. At exit from regmatch, slabs
2381 * allocated since entry are freed.
2384 /* *** every FOO_fail should = FOO+1 */
2385 #define TRIE_next (REGNODE_MAX+1)
2386 #define TRIE_next_fail (REGNODE_MAX+2)
2387 #define EVAL_A (REGNODE_MAX+3)
2388 #define EVAL_A_fail (REGNODE_MAX+4)
2389 #define resume_CURLYX (REGNODE_MAX+5)
2390 #define resume_WHILEM1 (REGNODE_MAX+6)
2391 #define resume_WHILEM2 (REGNODE_MAX+7)
2392 #define resume_WHILEM3 (REGNODE_MAX+8)
2393 #define resume_WHILEM4 (REGNODE_MAX+9)
2394 #define resume_WHILEM5 (REGNODE_MAX+10)
2395 #define resume_WHILEM6 (REGNODE_MAX+11)
2396 #define BRANCH_next (REGNODE_MAX+12)
2397 #define BRANCH_next_fail (REGNODE_MAX+13)
2398 #define CURLYM_A (REGNODE_MAX+14)
2399 #define CURLYM_A_fail (REGNODE_MAX+15)
2400 #define CURLYM_B (REGNODE_MAX+16)
2401 #define CURLYM_B_fail (REGNODE_MAX+17)
2402 #define IFMATCH_A (REGNODE_MAX+18)
2403 #define IFMATCH_A_fail (REGNODE_MAX+19)
2404 #define CURLY_B_min_known (REGNODE_MAX+20)
2405 #define CURLY_B_min_known_fail (REGNODE_MAX+21)
2406 #define CURLY_B_min (REGNODE_MAX+22)
2407 #define CURLY_B_min_fail (REGNODE_MAX+23)
2408 #define CURLY_B_max (REGNODE_MAX+24)
2409 #define CURLY_B_max_fail (REGNODE_MAX+25)
2412 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2416 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2417 const char *start, const char *end, const char *blurb)
2419 const bool utf8_pat= prog->reganch & ROPT_UTF8 ? 1 : 0;
2423 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2424 prog->precomp, prog->prelen, 60);
2426 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2427 start, end - start, 60);
2429 PerlIO_printf(Perl_debug_log,
2430 "%s%s REx%s %s against %s\n",
2431 PL_colors[4], blurb, PL_colors[5], s0, s1);
2433 if (do_utf8||utf8_pat)
2434 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2435 utf8_pat ? "pattern" : "",
2436 utf8_pat && do_utf8 ? " and " : "",
2437 do_utf8 ? "string" : ""
2443 S_dump_exec_pos(pTHX_ const char *locinput,
2444 const regnode *scan,
2445 const char *loc_regeol,
2446 const char *loc_bostr,
2447 const char *loc_reg_starttry,
2450 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2451 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2452 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2453 /* The part of the string before starttry has one color
2454 (pref0_len chars), between starttry and current
2455 position another one (pref_len - pref0_len chars),
2456 after the current position the third one.
2457 We assume that pref0_len <= pref_len, otherwise we
2458 decrease pref0_len. */
2459 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2460 ? (5 + taill) - l : locinput - loc_bostr;
2463 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2465 pref0_len = pref_len - (locinput - loc_reg_starttry);
2466 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2467 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2468 ? (5 + taill) - pref_len : loc_regeol - locinput);
2469 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2473 if (pref0_len > pref_len)
2474 pref0_len = pref_len;
2476 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2478 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2479 (locinput - pref_len),pref0_len, 60, 4, 5);
2481 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2482 (locinput - pref_len + pref0_len),
2483 pref_len - pref0_len, 60, 2, 3);
2485 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2486 locinput, loc_regeol - locinput, 10, 0, 1);
2488 const STRLEN tlen=len0+len1+len2;
2489 PerlIO_printf(Perl_debug_log,
2490 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2491 (IV)(locinput - loc_bostr),
2494 (docolor ? "" : "> <"),
2496 tlen > 19 ? 0 : 19 - tlen,
2503 STATIC I32 /* 0 failure, 1 success */
2504 S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog)
2506 #if PERL_VERSION < 9
2510 register const bool do_utf8 = PL_reg_match_utf8;
2511 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2513 regexp *rex = reginfo->prog;
2515 regmatch_slab *orig_slab;
2516 regmatch_state *orig_state;
2518 /* the current state. This is a cached copy of PL_regmatch_state */
2519 register regmatch_state *st;
2521 /* cache heavy used fields of st in registers */
2522 register regnode *scan;
2523 register regnode *next;
2524 register I32 n = 0; /* initialize to shut up compiler warning */
2525 register char *locinput = PL_reginput;
2527 /* these variables are NOT saved during a recusive RFEGMATCH: */
2528 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2529 bool result = 0; /* return value of S_regmatch */
2530 int depth = 0; /* depth of recursion */
2531 regmatch_state *yes_state = NULL; /* state to pop to on success of
2533 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2539 GET_RE_DEBUG_FLAGS_DECL;
2543 /* on first ever call to regmatch, allocate first slab */
2544 if (!PL_regmatch_slab) {
2545 Newx(PL_regmatch_slab, 1, regmatch_slab);
2546 PL_regmatch_slab->prev = NULL;
2547 PL_regmatch_slab->next = NULL;
2548 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2551 /* remember current high-water mark for exit */
2552 /* XXX this should be done with SAVE* instead */
2553 orig_slab = PL_regmatch_slab;
2554 orig_state = PL_regmatch_state;
2556 /* grab next free state slot */
2557 st = ++PL_regmatch_state;
2558 if (st > SLAB_LAST(PL_regmatch_slab))
2559 st = PL_regmatch_state = S_push_slab(aTHX);
2566 /* Note that nextchr is a byte even in UTF */
2567 nextchr = UCHARAT(locinput);
2569 while (scan != NULL) {
2572 SV * const prop = sv_newmortal();
2573 regnode *rnext=regnext(scan);
2574 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2575 regprop(rex, prop, scan);
2577 PerlIO_printf(Perl_debug_log,
2578 "%3"IVdf":%*s%s(%"IVdf")\n",
2579 (IV)(scan - rex->program), PL_regindent*2, "",
2581 (PL_regkind[OP(scan)] == END || !rnext) ?
2582 0 : (IV)(rnext - rex->program));
2585 next = scan + NEXT_OFF(scan);
2588 state_num = OP(scan);
2591 switch (state_num) {
2593 if (locinput == PL_bostr)
2595 /* reginfo->till = reginfo->bol; */
2600 if (locinput == PL_bostr ||
2601 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2607 if (locinput == PL_bostr)
2611 if (locinput == reginfo->ganch)
2617 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2622 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2624 if (PL_regeol - locinput > 1)
2628 if (PL_regeol != locinput)
2632 if (!nextchr && locinput >= PL_regeol)
2635 locinput += PL_utf8skip[nextchr];
2636 if (locinput > PL_regeol)
2638 nextchr = UCHARAT(locinput);
2641 nextchr = UCHARAT(++locinput);
2644 if (!nextchr && locinput >= PL_regeol)
2646 nextchr = UCHARAT(++locinput);
2649 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2652 locinput += PL_utf8skip[nextchr];
2653 if (locinput > PL_regeol)
2655 nextchr = UCHARAT(locinput);
2658 nextchr = UCHARAT(++locinput);
2662 #define ST st->u.trie
2664 /* In this case the charclass data is available inline so
2665 we can fail fast without a lot of extra overhead.
2667 if (scan->flags == EXACT || !do_utf8) {
2668 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2670 PerlIO_printf(Perl_debug_log,
2671 "%*s %sfailed to match trie start class...%s\n",
2672 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])
2681 /* what type of TRIE am I? (utf8 makes this contextual) */
2682 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2683 trie_type = do_utf8 ?
2684 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2687 /* what trie are we using right now */
2688 reg_trie_data * const trie
2689 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2690 U32 state = trie->startstate;
2692 if (trie->bitmap && trie_type != trie_utf8_fold &&
2693 !TRIE_BITMAP_TEST(trie,*locinput)
2695 if (trie->states[ state ].wordnum) {
2697 PerlIO_printf(Perl_debug_log,
2698 "%*s %smatched empty string...%s\n",
2699 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])
2704 PerlIO_printf(Perl_debug_log,
2705 "%*s %sfailed to match trie start class...%s\n",
2706 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])
2713 U8 *uc = ( U8* )locinput;
2717 U8 *uscan = (U8*)NULL;
2719 SV *sv_accept_buff = NULL;
2720 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2722 ST.accepted = 0; /* how many accepting states we have seen */
2724 ST.jump = trie->jump;
2733 traverse the TRIE keeping track of all accepting states
2734 we transition through until we get to a failing node.
2737 while ( state && uc <= (U8*)PL_regeol ) {
2738 U32 base = trie->states[ state ].trans.base;
2741 /* We use charid to hold the wordnum as we don't use it
2742 for charid until after we have done the wordnum logic.
2743 We define an alias just so that the wordnum logic reads
2746 #define got_wordnum charid
2747 got_wordnum = trie->states[ state ].wordnum;
2749 if ( got_wordnum ) {
2750 if ( ! ST.accepted ) {
2753 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2754 sv_accept_buff=newSV(bufflen *
2755 sizeof(reg_trie_accepted) - 1);
2756 SvCUR_set(sv_accept_buff, 0);
2757 SvPOK_on(sv_accept_buff);
2758 sv_2mortal(sv_accept_buff);
2761 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2764 if (ST.accepted >= bufflen) {
2766 ST.accept_buff =(reg_trie_accepted*)
2767 SvGROW(sv_accept_buff,
2768 bufflen * sizeof(reg_trie_accepted));
2770 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2771 + sizeof(reg_trie_accepted));
2774 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2775 ST.accept_buff[ST.accepted].endpos = uc;
2777 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2781 DEBUG_TRIE_EXECUTE_r({
2782 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2783 PerlIO_printf( Perl_debug_log,
2784 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2785 2+PL_regindent * 2, "", PL_colors[4],
2786 (UV)state, (UV)ST.accepted );
2790 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
2791 uvc, charid, foldlen, foldbuf, uniflags);
2794 (base + charid > trie->uniquecharcount )
2795 && (base + charid - 1 - trie->uniquecharcount
2797 && trie->trans[base + charid - 1 -
2798 trie->uniquecharcount].check == state)
2800 state = trie->trans[base + charid - 1 -
2801 trie->uniquecharcount ].next;
2812 DEBUG_TRIE_EXECUTE_r(
2813 PerlIO_printf( Perl_debug_log,
2814 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2815 charid, uvc, (UV)state, PL_colors[5] );
2822 PerlIO_printf( Perl_debug_log,
2823 "%*s %sgot %"IVdf" possible matches%s\n",
2824 REPORT_CODE_OFF + PL_regindent * 2, "",
2825 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2831 case TRIE_next_fail: /* we failed - try next alterative */
2833 if ( ST.accepted == 1 ) {
2834 /* only one choice left - just continue */
2836 reg_trie_data * const trie
2837 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2838 SV ** const tmp = RX_DEBUG(reginfo->prog)
2839 ? av_fetch( trie->words, ST.accept_buff[ 0 ].wordnum-1, 0 )
2841 PerlIO_printf( Perl_debug_log,
2842 "%*s %sonly one match left: #%d <%s>%s\n",
2843 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2844 ST.accept_buff[ 0 ].wordnum,
2845 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2848 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2849 /* in this case we free tmps/leave before we call regmatch
2850 as we wont be using accept_buff again. */
2853 locinput = PL_reginput;
2854 nextchr = UCHARAT(locinput);
2859 scan = ST.B - ST.jump[ST.accept_buff[0].wordnum];
2861 continue; /* execute rest of RE */
2864 if (!ST.accepted-- ) {
2871 There are at least two accepting states left. Presumably
2872 the number of accepting states is going to be low,
2873 typically two. So we simply scan through to find the one
2874 with lowest wordnum. Once we find it, we swap the last
2875 state into its place and decrement the size. We then try to
2876 match the rest of the pattern at the point where the word
2877 ends. If we succeed, control just continues along the
2878 regex; if we fail we return here to try the next accepting
2885 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
2886 DEBUG_TRIE_EXECUTE_r(
2887 PerlIO_printf( Perl_debug_log,
2888 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2889 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2890 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
2891 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
2894 if (ST.accept_buff[cur].wordnum <
2895 ST.accept_buff[best].wordnum)
2900 reg_trie_data * const trie
2901 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2902 SV ** const tmp = RX_DEBUG(reginfo->prog)
2903 ? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 )
2905 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at node #%d %s\n",
2906 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2907 ST.accept_buff[best].wordnum,
2908 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", REG_NODE_NUM(scan),
2912 if ( best<ST.accepted ) {
2913 reg_trie_accepted tmp = ST.accept_buff[ best ];
2914 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
2915 ST.accept_buff[ ST.accepted ] = tmp;
2918 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
2920 PUSH_STATE_GOTO(TRIE_next, ST.B);
2923 PUSH_STATE_GOTO(TRIE_next, ST.B - ST.jump[ST.accept_buff[best].wordnum]);
2933 char *s = STRING(scan);
2934 st->ln = STR_LEN(scan);
2935 if (do_utf8 != UTF) {
2936 /* The target and the pattern have differing utf8ness. */
2938 const char * const e = s + st->ln;
2941 /* The target is utf8, the pattern is not utf8. */
2946 if (NATIVE_TO_UNI(*(U8*)s) !=
2947 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2955 /* The target is not utf8, the pattern is utf8. */
2960 if (NATIVE_TO_UNI(*((U8*)l)) !=
2961 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2969 nextchr = UCHARAT(locinput);
2972 /* The target and the pattern have the same utf8ness. */
2973 /* Inline the first character, for speed. */
2974 if (UCHARAT(s) != nextchr)
2976 if (PL_regeol - locinput < st->ln)
2978 if (st->ln > 1 && memNE(s, locinput, st->ln))
2981 nextchr = UCHARAT(locinput);
2985 PL_reg_flags |= RF_tainted;
2988 char * const s = STRING(scan);
2989 st->ln = STR_LEN(scan);
2991 if (do_utf8 || UTF) {
2992 /* Either target or the pattern are utf8. */
2993 const char * const l = locinput;
2994 char *e = PL_regeol;
2996 if (ibcmp_utf8(s, 0, st->ln, (bool)UTF,
2997 l, &e, 0, do_utf8)) {
2998 /* One more case for the sharp s:
2999 * pack("U0U*", 0xDF) =~ /ss/i,
3000 * the 0xC3 0x9F are the UTF-8
3001 * byte sequence for the U+00DF. */
3003 toLOWER(s[0]) == 's' &&
3005 toLOWER(s[1]) == 's' &&
3012 nextchr = UCHARAT(locinput);
3016 /* Neither the target and the pattern are utf8. */
3018 /* Inline the first character, for speed. */
3019 if (UCHARAT(s) != nextchr &&
3020 UCHARAT(s) != ((OP(scan) == EXACTF)
3021 ? PL_fold : PL_fold_locale)[nextchr])
3023 if (PL_regeol - locinput < st->ln)
3025 if (st->ln > 1 && (OP(scan) == EXACTF
3026 ? ibcmp(s, locinput, st->ln)
3027 : ibcmp_locale(s, locinput, st->ln)))
3030 nextchr = UCHARAT(locinput);
3035 STRLEN inclasslen = PL_regeol - locinput;
3037 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3039 if (locinput >= PL_regeol)
3041 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3042 nextchr = UCHARAT(locinput);
3047 nextchr = UCHARAT(locinput);
3048 if (!REGINCLASS(rex, scan, (U8*)locinput))
3050 if (!nextchr && locinput >= PL_regeol)
3052 nextchr = UCHARAT(++locinput);
3056 /* If we might have the case of the German sharp s
3057 * in a casefolding Unicode character class. */
3059 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3060 locinput += SHARP_S_SKIP;
3061 nextchr = UCHARAT(locinput);
3067 PL_reg_flags |= RF_tainted;
3073 LOAD_UTF8_CHARCLASS_ALNUM();
3074 if (!(OP(scan) == ALNUM
3075 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3076 : isALNUM_LC_utf8((U8*)locinput)))
3080 locinput += PL_utf8skip[nextchr];
3081 nextchr = UCHARAT(locinput);
3084 if (!(OP(scan) == ALNUM
3085 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3087 nextchr = UCHARAT(++locinput);
3090 PL_reg_flags |= RF_tainted;
3093 if (!nextchr && locinput >= PL_regeol)
3096 LOAD_UTF8_CHARCLASS_ALNUM();
3097 if (OP(scan) == NALNUM
3098 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3099 : isALNUM_LC_utf8((U8*)locinput))
3103 locinput += PL_utf8skip[nextchr];
3104 nextchr = UCHARAT(locinput);
3107 if (OP(scan) == NALNUM
3108 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3110 nextchr = UCHARAT(++locinput);
3114 PL_reg_flags |= RF_tainted;
3118 /* was last char in word? */
3120 if (locinput == PL_bostr)
3123 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3125 st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3127 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3128 st->ln = isALNUM_uni(st->ln);
3129 LOAD_UTF8_CHARCLASS_ALNUM();
3130 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3133 st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln));
3134 n = isALNUM_LC_utf8((U8*)locinput);
3138 st->ln = (locinput != PL_bostr) ?
3139 UCHARAT(locinput - 1) : '\n';
3140 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3141 st->ln = isALNUM(st->ln);
3142 n = isALNUM(nextchr);
3145 st->ln = isALNUM_LC(st->ln);
3146 n = isALNUM_LC(nextchr);
3149 if (((!st->ln) == (!n)) == (OP(scan) == BOUND ||
3150 OP(scan) == BOUNDL))
3154 PL_reg_flags |= RF_tainted;
3160 if (UTF8_IS_CONTINUED(nextchr)) {
3161 LOAD_UTF8_CHARCLASS_SPACE();
3162 if (!(OP(scan) == SPACE
3163 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3164 : isSPACE_LC_utf8((U8*)locinput)))
3168 locinput += PL_utf8skip[nextchr];
3169 nextchr = UCHARAT(locinput);
3172 if (!(OP(scan) == SPACE
3173 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3175 nextchr = UCHARAT(++locinput);
3178 if (!(OP(scan) == SPACE
3179 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3181 nextchr = UCHARAT(++locinput);
3185 PL_reg_flags |= RF_tainted;
3188 if (!nextchr && locinput >= PL_regeol)
3191 LOAD_UTF8_CHARCLASS_SPACE();
3192 if (OP(scan) == NSPACE
3193 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3194 : isSPACE_LC_utf8((U8*)locinput))
3198 locinput += PL_utf8skip[nextchr];
3199 nextchr = UCHARAT(locinput);
3202 if (OP(scan) == NSPACE
3203 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3205 nextchr = UCHARAT(++locinput);
3208 PL_reg_flags |= RF_tainted;
3214 LOAD_UTF8_CHARCLASS_DIGIT();
3215 if (!(OP(scan) == DIGIT
3216 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3217 : isDIGIT_LC_utf8((U8*)locinput)))
3221 locinput += PL_utf8skip[nextchr];
3222 nextchr = UCHARAT(locinput);
3225 if (!(OP(scan) == DIGIT
3226 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3228 nextchr = UCHARAT(++locinput);
3231 PL_reg_flags |= RF_tainted;
3234 if (!nextchr && locinput >= PL_regeol)
3237 LOAD_UTF8_CHARCLASS_DIGIT();
3238 if (OP(scan) == NDIGIT
3239 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3240 : isDIGIT_LC_utf8((U8*)locinput))
3244 locinput += PL_utf8skip[nextchr];
3245 nextchr = UCHARAT(locinput);
3248 if (OP(scan) == NDIGIT
3249 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3251 nextchr = UCHARAT(++locinput);
3254 if (locinput >= PL_regeol)
3257 LOAD_UTF8_CHARCLASS_MARK();
3258 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3260 locinput += PL_utf8skip[nextchr];
3261 while (locinput < PL_regeol &&
3262 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3263 locinput += UTF8SKIP(locinput);
3264 if (locinput > PL_regeol)
3269 nextchr = UCHARAT(locinput);
3272 PL_reg_flags |= RF_tainted;
3277 n = ARG(scan); /* which paren pair */
3278 st->ln = PL_regstartp[n];
3279 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3280 if ((I32)*PL_reglastparen < n || st->ln == -1)
3281 sayNO; /* Do not match unless seen CLOSEn. */
3282 if (st->ln == PL_regendp[n])
3285 s = PL_bostr + st->ln;
3286 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3288 const char *e = PL_bostr + PL_regendp[n];
3290 * Note that we can't do the "other character" lookup trick as
3291 * in the 8-bit case (no pun intended) because in Unicode we
3292 * have to map both upper and title case to lower case.
3294 if (OP(scan) == REFF) {
3296 STRLEN ulen1, ulen2;
3297 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3298 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3302 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3303 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3304 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3311 nextchr = UCHARAT(locinput);
3315 /* Inline the first character, for speed. */
3316 if (UCHARAT(s) != nextchr &&
3318 (UCHARAT(s) != ((OP(scan) == REFF
3319 ? PL_fold : PL_fold_locale)[nextchr]))))
3321 st->ln = PL_regendp[n] - st->ln;
3322 if (locinput + st->ln > PL_regeol)
3324 if (st->ln > 1 && (OP(scan) == REF
3325 ? memNE(s, locinput, st->ln)
3327 ? ibcmp(s, locinput, st->ln)
3328 : ibcmp_locale(s, locinput, st->ln))))
3331 nextchr = UCHARAT(locinput);
3342 #define ST st->u.eval
3344 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3348 /* execute the code in the {...} */
3350 SV ** const before = SP;
3351 OP_4tree * const oop = PL_op;
3352 COP * const ocurcop = PL_curcop;
3356 PL_op = (OP_4tree*)rex->data->data[n];
3357 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3358 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3359 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3361 CALLRUNOPS(aTHX); /* Scalar context. */
3364 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3371 PAD_RESTORE_LOCAL(old_comppad);
3372 PL_curcop = ocurcop;
3375 sv_setsv(save_scalar(PL_replgv), ret);
3379 if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */
3382 /* extract RE object from returned value; compiling if
3387 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3388 mg = mg_find(sv, PERL_MAGIC_qr);
3389 else if (SvSMAGICAL(ret)) {
3390 if (SvGMAGICAL(ret))
3391 sv_unmagic(ret, PERL_MAGIC_qr);
3393 mg = mg_find(ret, PERL_MAGIC_qr);
3397 re = (regexp *)mg->mg_obj;
3398 (void)ReREFCNT_inc(re);
3402 const char * const t = SvPV_const(ret, len);
3404 const I32 osize = PL_regsize;
3407 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3408 re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm);
3410 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3412 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3418 /* run the pattern returned from (??{...}) */
3419 ST.cp = regcppush(0); /* Save *all* the positions. */
3420 REGCP_SET(ST.lastcp);
3421 *PL_reglastparen = 0;
3422 *PL_reglastcloseparen = 0;
3423 PL_reginput = locinput;
3425 /* XXXX This is too dramatic a measure... */
3429 ST.toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3430 ((re->reganch & ROPT_UTF8) != 0);
3431 if (ST.toggleutf) PL_reg_flags ^= RF_utf8;
3437 debug_start_match(re, do_utf8, locinput, PL_regeol,
3438 "Matching embedded");
3440 /* now continue from first node in postoned RE */
3441 PUSH_YES_STATE_GOTO(EVAL_A, re->program + 1);
3444 /* /(?(?{...})X|Y)/ */
3445 st->sw = SvTRUE(ret);
3450 case EVAL_A: /* successfully ran inner rex (??{rex}) */
3452 PL_reg_flags ^= RF_utf8;
3455 /* XXXX This is too dramatic a measure... */
3457 /* Restore parens of the caller without popping the
3460 const I32 tmp = PL_savestack_ix;
3461 PL_savestack_ix = ST.lastcp;
3463 PL_savestack_ix = tmp;
3465 PL_reginput = locinput;
3466 /* continue at the node following the (??{...}) */
3469 case EVAL_A_fail: /* unsuccessfully ran inner rex (??{rex}) */
3470 /* Restore state to the outer re then re-throw the failure */
3472 PL_reg_flags ^= RF_utf8;
3476 /* XXXX This is too dramatic a measure... */
3479 PL_reginput = locinput;
3480 REGCP_UNWIND(ST.lastcp);
3487 n = ARG(scan); /* which paren pair */
3488 PL_reg_start_tmp[n] = locinput;
3493 n = ARG(scan); /* which paren pair */
3494 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3495 PL_regendp[n] = locinput - PL_bostr;
3496 if (n > (I32)*PL_reglastparen)
3497 *PL_reglastparen = n;
3498 *PL_reglastcloseparen = n;
3501 n = ARG(scan); /* which paren pair */
3502 st->sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3505 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3507 next = NEXTOPER(NEXTOPER(scan));
3509 next = scan + ARG(scan);
3510 if (OP(next) == IFTHEN) /* Fake one. */
3511 next = NEXTOPER(NEXTOPER(next));
3515 st->logical = scan->flags;
3517 /*******************************************************************
3518 cur_curlyx points to the regmatch_state associated with the most recent CURLYX.
3519 This struct contains info about the innermost (...)* loop (an
3520 "infoblock"), and a pointer to the next outer cur_curlyx.
3522 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3524 1) After matching Y, regnode for CURLYX is processed;
3526 2) This regnode populates cur_curlyx, and calls regmatch() recursively
3527 with the starting point at WHILEM node;
3529 3) Each hit of WHILEM node tries to match A and Z (in the order
3530 depending on the current iteration, min/max of {min,max} and
3531 greediness). The information about where are nodes for "A"
3532 and "Z" is read from cur_curlyx, as is info on how many times "A"
3533 was already matched, and greediness.
3535 4) After A matches, the same WHILEM node is hit again.
3537 5) Each time WHILEM is hit, cur_curlyx is the infoblock created by CURLYX
3538 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3539 resets cur_curlyx, since this Y(A)*Z can be a part of some other loop:
3540 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3541 of the external loop.
3543 Currently present infoblocks form a tree with a stem formed by cur_curlyx
3544 and whatever it mentions via ->next, and additional attached trees
3545 corresponding to temporarily unset infoblocks as in "5" above.
3547 In the following picture, infoblocks for outer loop of
3548 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3549 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3550 infoblocks are drawn below the "reset" infoblock.
3552 In fact in the picture below we do not show failed matches for Z and T
3553 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3554 more obvious *why* one needs to *temporary* unset infoblocks.]
3556 Matched REx position InfoBlocks Comment
3560 Y A)*?Z)*?T x <- O <- I
3561 YA )*?Z)*?T x <- O <- I
3562 YA A)*?Z)*?T x <- O <- I
3563 YAA )*?Z)*?T x <- O <- I
3564 YAA Z)*?T x <- O # Temporary unset I
3567 YAAZ Y(A)*?Z)*?T x <- O
3570 YAAZY (A)*?Z)*?T x <- O
3573 YAAZY A)*?Z)*?T x <- O <- I
3576 YAAZYA )*?Z)*?T x <- O <- I
3579 YAAZYA Z)*?T x <- O # Temporary unset I
3585 YAAZYAZ T x # Temporary unset O
3592 *******************************************************************/
3595 /* No need to save/restore up to this paren */
3596 parenfloor = scan->flags;
3600 CURLYX and WHILEM are always paired: they're the moral
3601 equivalent of pp_enteriter anbd pp_iter.
3603 The only time next could be null is if the node tree is
3604 corrupt. This was mentioned on p5p a few days ago.
3606 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3607 So we'll assert that this is true:
3610 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3612 /* XXXX Probably it is better to teach regpush to support
3613 parenfloor > PL_regsize... */
3614 if (parenfloor > (I32)*PL_reglastparen)
3615 parenfloor = *PL_reglastparen; /* Pessimization... */
3617 st->u.curlyx.cp = PL_savestack_ix;
3618 st->u.curlyx.outercc = cur_curlyx;
3620 /* these fields contain the state of the current curly.
3621 * they are accessed by subsequent WHILEMs;
3622 * cur and lastloc are also updated by WHILEM */
3623 st->u.curlyx.parenfloor = parenfloor;
3624 st->u.curlyx.cur = -1; /* this will be updated by WHILEM */
3625 st->u.curlyx.min = ARG1(scan);
3626 st->u.curlyx.max = ARG2(scan);
3627 st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3628 st->u.curlyx.lastloc = 0;
3629 /* st->next and st->minmod are also read by WHILEM */
3631 PL_reginput = locinput;
3632 REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */
3633 /*** all unsaved local vars undefined at this point */
3634 regcpblow(st->u.curlyx.cp);
3635 cur_curlyx = st->u.curlyx.outercc;
3641 * This is really hard to understand, because after we match
3642 * what we're trying to match, we must make sure the rest of
3643 * the REx is going to match for sure, and to do that we have
3644 * to go back UP the parse tree by recursing ever deeper. And
3645 * if it fails, we have to reset our parent's current state
3646 * that we can try again after backing off.
3651 cur_curlyx gets initialised by CURLYX ready for use by WHILEM.
3652 So again, unless somethings been corrupted, cur_curlyx cannot
3653 be null at that point in WHILEM.
3655 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3656 So we'll assert that this is true:
3659 st->u.whilem.lastloc = cur_curlyx->u.curlyx.lastloc; /* Detection of 0-len. */
3660 st->u.whilem.cache_offset = 0;
3661 st->u.whilem.cache_bit = 0;
3663 n = cur_curlyx->u.curlyx.cur + 1; /* how many we know we matched */
3664 PL_reginput = locinput;
3667 PerlIO_printf(Perl_debug_log,
3668 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3669 REPORT_CODE_OFF+PL_regindent*2, "",
3670 (long)n, (long)cur_curlyx->u.curlyx.min,
3671 (long)cur_curlyx->u.curlyx.max,
3675 /* If degenerate scan matches "", assume scan done. */
3677 if (locinput == cur_curlyx->u.curlyx.lastloc && n >=
3678 cur_curlyx->u.curlyx.min)
3680 st->u.whilem.savecc = cur_curlyx;
3681 cur_curlyx = cur_curlyx->u.curlyx.outercc;
3683 st->ln = cur_curlyx->u.curlyx.cur;
3685 PerlIO_printf(Perl_debug_log,
3686 "%*s empty match detected, try continuation...\n",
3687 REPORT_CODE_OFF+PL_regindent*2, "")
3689 REGMATCH(st->u.whilem.savecc->next, WHILEM1);
3690 /*** all unsaved local vars undefined at this point */
3691 cur_curlyx = st->u.whilem.savecc;
3694 if (cur_curlyx->u.curlyx.outercc)
3695 cur_curlyx->u.curlyx.outercc->u.curlyx.cur = st->ln;
3699 /* First just match a string of min scans. */
3701 if (n < cur_curlyx->u.curlyx.min) {
3702 cur_curlyx->u.curlyx.cur = n;
3703 cur_curlyx->u.curlyx.lastloc = locinput;
3704 REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM2);
3705 /*** all unsaved local vars undefined at this point */
3708 cur_curlyx->u.curlyx.cur = n - 1;
3709 cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
3714 /* Check whether we already were at this position.
3715 Postpone detection until we know the match is not
3716 *that* much linear. */
3717 if (!PL_reg_maxiter) {
3718 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3719 /* possible overflow for long strings and many CURLYX's */
3720 if (PL_reg_maxiter < 0)
3721 PL_reg_maxiter = I32_MAX;
3722 PL_reg_leftiter = PL_reg_maxiter;
3724 if (PL_reg_leftiter-- == 0) {
3725 const I32 size = (PL_reg_maxiter + 7)/8;
3726 if (PL_reg_poscache) {
3727 if ((I32)PL_reg_poscache_size < size) {
3728 Renew(PL_reg_poscache, size, char);
3729 PL_reg_poscache_size = size;
3731 Zero(PL_reg_poscache, size, char);
3734 PL_reg_poscache_size = size;
3735 Newxz(PL_reg_poscache, size, char);
3738 PerlIO_printf(Perl_debug_log,
3739 "%sDetected a super-linear match, switching on caching%s...\n",
3740 PL_colors[4], PL_colors[5])
3743 if (PL_reg_leftiter < 0) {
3744 st->u.whilem.cache_offset = locinput - PL_bostr;
3746 st->u.whilem.cache_offset = (scan->flags & 0xf) - 1
3747 + st->u.whilem.cache_offset * (scan->flags>>4);
3748 st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8;
3749 st->u.whilem.cache_offset /= 8;
3750 if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<<st->u.whilem.cache_bit)) {
3752 PerlIO_printf(Perl_debug_log,
3753 "%*s already tried at this position...\n",
3754 REPORT_CODE_OFF+PL_regindent*2, "")
3756 sayNO; /* cache records failure */
3761 /* Prefer next over scan for minimal matching. */
3763 if (cur_curlyx->minmod) {
3764 st->u.whilem.savecc = cur_curlyx;
3765 cur_curlyx = cur_curlyx->u.curlyx.outercc;
3767 st->ln = cur_curlyx->u.curlyx.cur;
3768 st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor);
3769 REGCP_SET(st->u.whilem.lastcp);
3770 REGMATCH(st->u.whilem.savecc->next, WHILEM3);
3771 /*** all unsaved local vars undefined at this point */
3772 cur_curlyx = st->u.whilem.savecc;
3774 regcpblow(st->u.whilem.cp);
3775 sayYES; /* All done. */
3777 REGCP_UNWIND(st->u.whilem.lastcp);
3779 if (cur_curlyx->u.curlyx.outercc)
3780 cur_curlyx->u.curlyx.outercc->u.curlyx.cur = st->ln;
3782 if (n >= cur_curlyx->u.curlyx.max) { /* Maximum greed exceeded? */
3783 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3784 && !(PL_reg_flags & RF_warned)) {
3785 PL_reg_flags |= RF_warned;
3786 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3787 "Complex regular subexpression recursion",
3794 PerlIO_printf(Perl_debug_log,
3795 "%*s trying longer...\n",
3796 REPORT_CODE_OFF+PL_regindent*2, "")
3798 /* Try scanning more and see if it helps. */
3799 PL_reginput = locinput;
3800 cur_curlyx->u.curlyx.cur = n;
3801 cur_curlyx->u.curlyx.lastloc = locinput;
3802 st->u.whilem.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3803 REGCP_SET(st->u.whilem.lastcp);
3804 REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM4);
3805 /*** all unsaved local vars undefined at this point */
3807 regcpblow(st->u.whilem.cp);
3810 REGCP_UNWIND(st->u.whilem.lastcp);
3812 cur_curlyx->u.curlyx.cur = n - 1;
3813 cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
3817 /* Prefer scan over next for maximal matching. */
3819 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
3820 st->u.whilem.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3821 cur_curlyx->u.curlyx.cur = n;
3822 cur_curlyx->u.curlyx.lastloc = locinput;
3823 REGCP_SET(st->u.whilem.lastcp);
3824 REGMATCH(cur_curlyx->u.curlyx.scan, WHILEM5);
3825 /*** all unsaved local vars undefined at this point */
3827 regcpblow(st->u.whilem.cp);
3830 REGCP_UNWIND(st->u.whilem.lastcp);
3831 regcppop(rex); /* Restore some previous $<digit>s? */
3832 PL_reginput = locinput;
3834 PerlIO_printf(Perl_debug_log,
3835 "%*s failed, try continuation...\n",
3836 REPORT_CODE_OFF+PL_regindent*2, "")
3839 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3840 && !(PL_reg_flags & RF_warned)) {
3841 PL_reg_flags |= RF_warned;
3842 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3843 "Complex regular subexpression recursion",
3847 /* Failed deeper matches of scan, so see if this one works. */
3848 st->u.whilem.savecc = cur_curlyx;
3849 cur_curlyx = cur_curlyx->u.curlyx.outercc;
3851 st->ln = cur_curlyx->u.curlyx.cur;
3852 REGMATCH(st->u.whilem.savecc->next, WHILEM6);
3853 /*** all unsaved local vars undefined at this point */
3854 cur_curlyx = st->u.whilem.savecc;
3857 if (cur_curlyx->u.curlyx.outercc)
3858 cur_curlyx->u.curlyx.outercc->u.curlyx.cur = st->ln;
3859 cur_curlyx->u.curlyx.cur = n - 1;
3860 cur_curlyx->u.curlyx.lastloc = st->u.whilem.lastloc;
3866 #define ST st->u.branch
3868 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
3869 next = scan + ARG(scan);
3872 scan = NEXTOPER(scan);
3875 case BRANCH: /* /(...|A|...)/ */
3876 scan = NEXTOPER(scan); /* scan now points to inner node */
3877 if (!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
3878 /* last branch; skip state push and jump direct to node */
3880 ST.lastparen = *PL_reglastparen;
3881 ST.next_branch = next;
3883 PL_reginput = locinput;
3885 /* Now go into the branch */
3886 PUSH_STATE_GOTO(BRANCH_next, scan);
3889 case BRANCH_next_fail: /* that branch failed; try the next, if any */
3890 REGCP_UNWIND(ST.cp);
3891 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3893 *PL_reglastparen = n;
3894 scan = ST.next_branch;
3895 /* no more branches? */
3896 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ))
3898 continue; /* execute next BRANCH[J] op */
3906 #define ST st->u.curlym
3908 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
3910 /* This is an optimisation of CURLYX that enables us to push
3911 * only a single backtracking state, no matter now many matches