5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
56 **** Alterations to Henry's code are...
58 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
59 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
61 **** You may distribute under the terms of either the GNU General Public
62 **** License or the Artistic License, as specified in the README file.
64 * Beware that some of this code is subtly aware of the way operator
65 * precedence is structured in regular expressions. Serious changes in
66 * regular-expression syntax might require a total rethink.
69 #define PERL_IN_REGEXEC_C
72 #ifdef PERL_IN_XSUB_RE
78 #define RF_tainted 1 /* tainted information used? */
79 #define RF_warned 2 /* warned about big count? */
81 #define RF_utf8 8 /* Pattern contains multibyte chars? */
83 #define UTF ((PL_reg_flags & RF_utf8) != 0)
85 #define RS_init 1 /* eval environment created */
86 #define RS_set 2 /* replsv value is set */
92 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
98 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
99 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
101 #define HOPc(pos,off) \
102 (char *)(PL_reg_match_utf8 \
103 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
105 #define HOPBACKc(pos, off) \
106 (char*)(PL_reg_match_utf8\
107 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
108 : (pos - off >= PL_bostr) \
112 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
113 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
115 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
116 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
117 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
118 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
119 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
120 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
122 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
124 /* for use after a quantifier and before an EXACT-like node -- japhy */
125 #define JUMPABLE(rn) ( \
126 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
127 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
128 OP(rn) == PLUS || OP(rn) == MINMOD || \
129 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
132 #define HAS_TEXT(rn) ( \
133 PL_regkind[OP(rn)] == EXACT || PL_regkind[OP(rn)] == REF \
137 Search for mandatory following text node; for lookahead, the text must
138 follow but for lookbehind (rn->flags != 0) we skip to the next step.
140 #define FIND_NEXT_IMPT(rn) STMT_START { \
141 while (JUMPABLE(rn)) { \
142 const OPCODE type = OP(rn); \
143 if (type == SUSPEND || PL_regkind[type] == CURLY) \
144 rn = NEXTOPER(NEXTOPER(rn)); \
145 else if (type == PLUS) \
147 else if (type == IFMATCH) \
148 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
149 else rn += NEXT_OFF(rn); \
154 static void restore_pos(pTHX_ void *arg);
157 S_regcppush(pTHX_ I32 parenfloor)
160 const int retval = PL_savestack_ix;
161 #define REGCP_PAREN_ELEMS 4
162 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
164 GET_RE_DEBUG_FLAGS_DECL;
166 if (paren_elems_to_push < 0)
167 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
169 #define REGCP_OTHER_ELEMS 8
170 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
172 for (p = PL_regsize; p > parenfloor; p--) {
173 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
174 SSPUSHINT(PL_regendp[p]);
175 SSPUSHINT(PL_regstartp[p]);
176 SSPUSHPTR(PL_reg_start_tmp[p]);
178 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
179 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
180 (UV)p, (IV)PL_regstartp[p],
181 (IV)(PL_reg_start_tmp[p] - PL_bostr),
185 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
186 SSPUSHPTR(PL_regstartp);
187 SSPUSHPTR(PL_regendp);
188 SSPUSHINT(PL_regsize);
189 SSPUSHINT(*PL_reglastparen);
190 SSPUSHINT(*PL_reglastcloseparen);
191 SSPUSHPTR(PL_reginput);
192 #define REGCP_FRAME_ELEMS 2
193 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
194 * are needed for the regexp context stack bookkeeping. */
195 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
196 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
201 /* These are needed since we do not localize EVAL nodes: */
202 #define REGCP_SET(cp) \
204 PerlIO_printf(Perl_debug_log, \
205 " Setting an EVAL scope, savestack=%"IVdf"\n", \
206 (IV)PL_savestack_ix)); \
209 #define REGCP_UNWIND(cp) \
211 if (cp != PL_savestack_ix) \
212 PerlIO_printf(Perl_debug_log, \
213 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
214 (IV)(cp), (IV)PL_savestack_ix)); \
218 S_regcppop(pTHX_ const regexp *rex)
224 GET_RE_DEBUG_FLAGS_DECL;
226 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
228 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
229 i = SSPOPINT; /* Parentheses elements to pop. */
230 input = (char *) SSPOPPTR;
231 *PL_reglastcloseparen = SSPOPINT;
232 *PL_reglastparen = SSPOPINT;
233 PL_regsize = SSPOPINT;
234 PL_regendp=(I32 *) SSPOPPTR;
235 PL_regstartp=(I32 *) SSPOPPTR;
238 /* Now restore the parentheses context. */
239 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
240 i > 0; i -= REGCP_PAREN_ELEMS) {
242 U32 paren = (U32)SSPOPINT;
243 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
244 PL_regstartp[paren] = SSPOPINT;
246 if (paren <= *PL_reglastparen)
247 PL_regendp[paren] = tmps;
249 PerlIO_printf(Perl_debug_log,
250 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
251 (UV)paren, (IV)PL_regstartp[paren],
252 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
253 (IV)PL_regendp[paren],
254 (paren > *PL_reglastparen ? "(no)" : ""));
258 if (*PL_reglastparen + 1 <= rex->nparens) {
259 PerlIO_printf(Perl_debug_log,
260 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
261 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
265 /* It would seem that the similar code in regtry()
266 * already takes care of this, and in fact it is in
267 * a better location to since this code can #if 0-ed out
268 * but the code in regtry() is needed or otherwise tests
269 * requiring null fields (pat.t#187 and split.t#{13,14}
270 * (as of patchlevel 7877) will fail. Then again,
271 * this code seems to be necessary or otherwise
272 * building DynaLoader will fail:
273 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
275 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
277 PL_regstartp[i] = -1;
284 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
287 * pregexec and friends
290 #ifndef PERL_IN_XSUB_RE
292 - pregexec - match a regexp against a string
295 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
296 char *strbeg, I32 minend, SV *screamer, U32 nosave)
297 /* strend: pointer to null at end of string */
298 /* strbeg: real beginning of string */
299 /* minend: end of match must be >=minend after stringarg. */
300 /* nosave: For optimizations. */
303 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
304 nosave ? 0 : REXEC_COPY_STR);
309 * Need to implement the following flags for reg_anch:
311 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
313 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
314 * INTUIT_AUTORITATIVE_ML
315 * INTUIT_ONCE_NOML - Intuit can match in one location only.
318 * Another flag for this function: SECOND_TIME (so that float substrs
319 * with giant delta may be not rechecked).
322 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
324 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
325 Otherwise, only SvCUR(sv) is used to get strbeg. */
327 /* XXXX We assume that strpos is strbeg unless sv. */
329 /* XXXX Some places assume that there is a fixed substring.
330 An update may be needed if optimizer marks as "INTUITable"
331 RExen without fixed substrings. Similarly, it is assumed that
332 lengths of all the strings are no more than minlen, thus they
333 cannot come from lookahead.
334 (Or minlen should take into account lookahead.)
335 NOTE: Some of this comment is not correct. minlen does now take account
336 of lookahead/behind. Further research is required. -- demerphq
340 /* A failure to find a constant substring means that there is no need to make
341 an expensive call to REx engine, thus we celebrate a failure. Similarly,
342 finding a substring too deep into the string means that less calls to
343 regtry() should be needed.
345 REx compiler's optimizer found 4 possible hints:
346 a) Anchored substring;
348 c) Whether we are anchored (beginning-of-line or \G);
349 d) First node (of those at offset 0) which may distingush positions;
350 We use a)b)d) and multiline-part of c), and try to find a position in the
351 string which does not contradict any of them.
354 /* Most of decisions we do here should have been done at compile time.
355 The nodes of the REx which we used for the search should have been
356 deleted from the finite automaton. */
359 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
360 char *strend, U32 flags, re_scream_pos_data *data)
363 register I32 start_shift = 0;
364 /* Should be nonnegative! */
365 register I32 end_shift = 0;
370 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
372 register char *other_last = NULL; /* other substr checked before this */
373 char *check_at = NULL; /* check substr found at this pos */
374 const I32 multiline = prog->reganch & PMf_MULTILINE;
376 const char * const i_strpos = strpos;
379 GET_RE_DEBUG_FLAGS_DECL;
381 RX_MATCH_UTF8_set(prog,do_utf8);
383 if (prog->reganch & ROPT_UTF8) {
384 PL_reg_flags |= RF_utf8;
387 debug_start_match(prog, do_utf8, strpos, strend,
388 sv ? "Guessing start of match in sv for"
389 : "Guessing start of match in string for");
392 /* CHR_DIST() would be more correct here but it makes things slow. */
393 if (prog->minlen > strend - strpos) {
394 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
395 "String too short... [re_intuit_start]\n"));
399 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
402 if (!prog->check_utf8 && prog->check_substr)
403 to_utf8_substr(prog);
404 check = prog->check_utf8;
406 if (!prog->check_substr && prog->check_utf8)
407 to_byte_substr(prog);
408 check = prog->check_substr;
410 if (check == &PL_sv_undef) {
411 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
412 "Non-utf8 string cannot match utf8 check string\n"));
415 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
416 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
417 || ( (prog->reganch & ROPT_ANCH_BOL)
418 && !multiline ) ); /* Check after \n? */
421 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
422 | ROPT_IMPLICIT)) /* not a real BOL */
423 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
425 && (strpos != strbeg)) {
426 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
429 if (prog->check_offset_min == prog->check_offset_max &&
430 !(prog->reganch & ROPT_CANY_SEEN)) {
431 /* Substring at constant offset from beg-of-str... */
434 s = HOP3c(strpos, prog->check_offset_min, strend);
437 slen = SvCUR(check); /* >= 1 */
439 if ( strend - s > slen || strend - s < slen - 1
440 || (strend - s == slen && strend[-1] != '\n')) {
441 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
444 /* Now should match s[0..slen-2] */
446 if (slen && (*SvPVX_const(check) != *s
448 && memNE(SvPVX_const(check), s, slen)))) {
450 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
454 else if (*SvPVX_const(check) != *s
455 || ((slen = SvCUR(check)) > 1
456 && memNE(SvPVX_const(check), s, slen)))
459 goto success_at_start;
462 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
464 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
465 end_shift = prog->check_end_shift;
468 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
469 - (SvTAIL(check) != 0);
470 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
472 if (end_shift < eshift)
476 else { /* Can match at random position */
479 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
480 end_shift = prog->check_end_shift;
482 /* end shift should be non negative here */
485 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
487 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
488 (IV)end_shift, prog->precomp);
492 /* Find a possible match in the region s..strend by looking for
493 the "check" substring in the region corrected by start/end_shift. */
496 I32 srch_start_shift = start_shift;
497 I32 srch_end_shift = end_shift;
498 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
499 srch_end_shift -= ((strbeg - s) - srch_start_shift);
500 srch_start_shift = strbeg - s;
502 DEBUG_OPTIMISE_MORE_r({
503 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
504 (IV)prog->check_offset_min,
505 (IV)srch_start_shift,
507 (IV)prog->check_end_shift);
510 if (flags & REXEC_SCREAM) {
511 I32 p = -1; /* Internal iterator of scream. */
512 I32 * const pp = data ? data->scream_pos : &p;
514 if (PL_screamfirst[BmRARE(check)] >= 0
515 || ( BmRARE(check) == '\n'
516 && (BmPREVIOUS(check) == SvCUR(check) - 1)
518 s = screaminstr(sv, check,
519 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
522 /* we may be pointing at the wrong string */
523 if (s && RX_MATCH_COPIED(prog))
524 s = strbeg + (s - SvPVX_const(sv));
526 *data->scream_olds = s;
531 if (prog->reganch & ROPT_CANY_SEEN) {
532 start_point= (U8*)(s + srch_start_shift);
533 end_point= (U8*)(strend - srch_end_shift);
535 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
536 end_point= HOP3(strend, -srch_end_shift, strbeg);
538 DEBUG_OPTIMISE_MORE_r({
539 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
540 (int)(end_point - start_point),
541 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
545 s = fbm_instr( start_point, end_point,
546 check, multiline ? FBMrf_MULTILINE : 0);
549 /* Update the count-of-usability, remove useless subpatterns,
553 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
554 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
555 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
556 (s ? "Found" : "Did not find"),
557 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
558 ? "anchored" : "floating"),
561 (s ? " at offset " : "...\n") );
566 /* Finish the diagnostic message */
567 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
569 /* XXX dmq: first branch is for positive lookbehind...
570 Our check string is offset from the beginning of the pattern.
571 So we need to do any stclass tests offset forward from that
580 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
581 Start with the other substr.
582 XXXX no SCREAM optimization yet - and a very coarse implementation
583 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
584 *always* match. Probably should be marked during compile...
585 Probably it is right to do no SCREAM here...
588 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
589 : (prog->float_substr && prog->anchored_substr))
591 /* Take into account the "other" substring. */
592 /* XXXX May be hopelessly wrong for UTF... */
595 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
598 char * const last = HOP3c(s, -start_shift, strbeg);
600 char * const saved_s = s;
603 t = s - prog->check_offset_max;
604 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
606 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
611 t = HOP3c(t, prog->anchored_offset, strend);
612 if (t < other_last) /* These positions already checked */
614 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
617 /* XXXX It is not documented what units *_offsets are in.
618 We assume bytes, but this is clearly wrong.
619 Meaning this code needs to be carefully reviewed for errors.
623 /* On end-of-str: see comment below. */
624 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
625 if (must == &PL_sv_undef) {
627 DEBUG_r(must = prog->anchored_utf8); /* for debug */
632 HOP3(HOP3(last1, prog->anchored_offset, strend)
633 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
635 multiline ? FBMrf_MULTILINE : 0
638 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
639 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
640 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
641 (s ? "Found" : "Contradicts"),
642 quoted, RE_SV_TAIL(must));
647 if (last1 >= last2) {
648 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
649 ", giving up...\n"));
652 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
653 ", trying floating at offset %ld...\n",
654 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
655 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
656 s = HOP3c(last, 1, strend);
660 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
661 (long)(s - i_strpos)));
662 t = HOP3c(s, -prog->anchored_offset, strbeg);
663 other_last = HOP3c(s, 1, strend);
671 else { /* Take into account the floating substring. */
673 char * const saved_s = s;
676 t = HOP3c(s, -start_shift, strbeg);
678 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
679 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
680 last = HOP3c(t, prog->float_max_offset, strend);
681 s = HOP3c(t, prog->float_min_offset, strend);
684 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
685 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
686 /* fbm_instr() takes into account exact value of end-of-str
687 if the check is SvTAIL(ed). Since false positives are OK,
688 and end-of-str is not later than strend we are OK. */
689 if (must == &PL_sv_undef) {
691 DEBUG_r(must = prog->float_utf8); /* for debug message */
694 s = fbm_instr((unsigned char*)s,
695 (unsigned char*)last + SvCUR(must)
697 must, multiline ? FBMrf_MULTILINE : 0);
699 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
700 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
701 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
702 (s ? "Found" : "Contradicts"),
703 quoted, RE_SV_TAIL(must));
707 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
708 ", giving up...\n"));
711 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
712 ", trying anchored starting at offset %ld...\n",
713 (long)(saved_s + 1 - i_strpos)));
715 s = HOP3c(t, 1, strend);
719 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
720 (long)(s - i_strpos)));
721 other_last = s; /* Fix this later. --Hugo */
731 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
733 DEBUG_OPTIMISE_MORE_r(
734 PerlIO_printf(Perl_debug_log,
735 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
736 (IV)prog->check_offset_min,
737 (IV)prog->check_offset_max,
745 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
747 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
750 /* Fixed substring is found far enough so that the match
751 cannot start at strpos. */
753 if (ml_anch && t[-1] != '\n') {
754 /* Eventually fbm_*() should handle this, but often
755 anchored_offset is not 0, so this check will not be wasted. */
756 /* XXXX In the code below we prefer to look for "^" even in
757 presence of anchored substrings. And we search even
758 beyond the found float position. These pessimizations
759 are historical artefacts only. */
761 while (t < strend - prog->minlen) {
763 if (t < check_at - prog->check_offset_min) {
764 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
765 /* Since we moved from the found position,
766 we definitely contradict the found anchored
767 substr. Due to the above check we do not
768 contradict "check" substr.
769 Thus we can arrive here only if check substr
770 is float. Redo checking for "other"=="fixed".
773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
774 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
775 goto do_other_anchored;
777 /* We don't contradict the found floating substring. */
778 /* XXXX Why not check for STCLASS? */
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
781 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
784 /* Position contradicts check-string */
785 /* XXXX probably better to look for check-string
786 than for "\n", so one should lower the limit for t? */
787 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
788 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
789 other_last = strpos = s = t + 1;
794 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
795 PL_colors[0], PL_colors[1]));
799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
800 PL_colors[0], PL_colors[1]));
804 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
807 /* The found string does not prohibit matching at strpos,
808 - no optimization of calling REx engine can be performed,
809 unless it was an MBOL and we are not after MBOL,
810 or a future STCLASS check will fail this. */
812 /* Even in this situation we may use MBOL flag if strpos is offset
813 wrt the start of the string. */
814 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
815 && (strpos != strbeg) && strpos[-1] != '\n'
816 /* May be due to an implicit anchor of m{.*foo} */
817 && !(prog->reganch & ROPT_IMPLICIT))
822 DEBUG_EXECUTE_r( if (ml_anch)
823 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
824 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
827 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
829 prog->check_utf8 /* Could be deleted already */
830 && --BmUSEFUL(prog->check_utf8) < 0
831 && (prog->check_utf8 == prog->float_utf8)
833 prog->check_substr /* Could be deleted already */
834 && --BmUSEFUL(prog->check_substr) < 0
835 && (prog->check_substr == prog->float_substr)
838 /* If flags & SOMETHING - do not do it many times on the same match */
839 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
840 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
841 if (do_utf8 ? prog->check_substr : prog->check_utf8)
842 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
843 prog->check_substr = prog->check_utf8 = NULL; /* disable */
844 prog->float_substr = prog->float_utf8 = NULL; /* clear */
845 check = NULL; /* abort */
847 /* XXXX This is a remnant of the old implementation. It
848 looks wasteful, since now INTUIT can use many
850 prog->reganch &= ~RE_USE_INTUIT;
857 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
858 /* trie stclasses are too expensive to use here, we are better off to
859 leave it to regmatch itself */
860 if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) {
861 /* minlen == 0 is possible if regstclass is \b or \B,
862 and the fixed substr is ''$.
863 Since minlen is already taken into account, s+1 is before strend;
864 accidentally, minlen >= 1 guaranties no false positives at s + 1
865 even for \b or \B. But (minlen? 1 : 0) below assumes that
866 regstclass does not come from lookahead... */
867 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
868 This leaves EXACTF only, which is dealt with in find_byclass(). */
869 const U8* const str = (U8*)STRING(prog->regstclass);
870 const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT
871 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
874 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
875 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
876 else if (prog->float_substr || prog->float_utf8)
877 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
881 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n",
882 (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg));
885 s = find_byclass(prog, prog->regstclass, s, endpos, NULL);
888 const char *what = NULL;
890 if (endpos == strend) {
891 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
892 "Could not match STCLASS...\n") );
895 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
896 "This position contradicts STCLASS...\n") );
897 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
899 /* Contradict one of substrings */
900 if (prog->anchored_substr || prog->anchored_utf8) {
901 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
902 DEBUG_EXECUTE_r( what = "anchored" );
904 s = HOP3c(t, 1, strend);
905 if (s + start_shift + end_shift > strend) {
906 /* XXXX Should be taken into account earlier? */
907 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
908 "Could not match STCLASS...\n") );
913 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
914 "Looking for %s substr starting at offset %ld...\n",
915 what, (long)(s + start_shift - i_strpos)) );
918 /* Have both, check_string is floating */
919 if (t + start_shift >= check_at) /* Contradicts floating=check */
920 goto retry_floating_check;
921 /* Recheck anchored substring, but not floating... */
925 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
926 "Looking for anchored substr starting at offset %ld...\n",
927 (long)(other_last - i_strpos)) );
928 goto do_other_anchored;
930 /* Another way we could have checked stclass at the
931 current position only: */
936 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
937 "Looking for /%s^%s/m starting at offset %ld...\n",
938 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
941 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
943 /* Check is floating subtring. */
944 retry_floating_check:
945 t = check_at - start_shift;
946 DEBUG_EXECUTE_r( what = "floating" );
947 goto hop_and_restart;
950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
951 "By STCLASS: moving %ld --> %ld\n",
952 (long)(t - i_strpos), (long)(s - i_strpos))
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
957 "Does not contradict STCLASS...\n");
962 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
963 PL_colors[4], (check ? "Guessed" : "Giving up"),
964 PL_colors[5], (long)(s - i_strpos)) );
967 fail_finish: /* Substring not found */
968 if (prog->check_substr || prog->check_utf8) /* could be removed already */
969 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
972 PL_colors[4], PL_colors[5]));
978 #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \
979 foldlen, foldbuf, uniflags) STMT_START { \
980 switch (trie_type) { \
981 case trie_utf8_fold: \
983 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
988 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
989 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
990 foldlen -= UNISKIP( uvc ); \
991 uscan = foldbuf + UNISKIP( uvc ); \
995 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1003 charid = trie->charmap[ uvc ]; \
1007 if (trie->widecharmap) { \
1008 SV** const svpp = hv_fetch(trie->widecharmap, \
1009 (char*)&uvc, sizeof(UV), 0); \
1011 charid = (U16)SvIV(*svpp); \
1016 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1019 ibcmp_utf8(s, NULL, 0, do_utf8, \
1020 m, NULL, ln, (bool)UTF)) \
1021 && (!reginfo || regtry(reginfo, &s)) ) \
1024 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1025 uvchr_to_utf8(tmpbuf, c); \
1026 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1028 && (f == c1 || f == c2) \
1029 && (ln == foldlen || \
1030 !ibcmp_utf8((char *) foldbuf, \
1031 NULL, foldlen, do_utf8, \
1033 NULL, ln, (bool)UTF)) \
1034 && (!reginfo || regtry(reginfo, &s)) ) \
1039 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1043 && (ln == 1 || !(OP(c) == EXACTF \
1045 : ibcmp_locale(s, m, ln))) \
1046 && (!reginfo || regtry(reginfo, &s)) ) \
1052 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1054 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1060 #define REXEC_FBC_SCAN(CoDe) \
1062 while (s < strend) { \
1068 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1069 REXEC_FBC_UTF8_SCAN( \
1071 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1080 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1083 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1092 #define REXEC_FBC_TRYIT \
1093 if ((!reginfo || regtry(reginfo, &s))) \
1096 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1099 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1102 REXEC_FBC_CLASS_SCAN(CoNd); \
1106 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1107 PL_reg_flags |= RF_tainted; \
1109 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1112 REXEC_FBC_CLASS_SCAN(CoNd); \
1116 #define DUMP_EXEC_POS(li,s,doutf8) \
1117 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1119 /* We know what class REx starts with. Try to find this position... */
1120 /* if reginfo is NULL, its a dryrun */
1121 /* annoyingly all the vars in this routine have different names from their counterparts
1122 in regmatch. /grrr */
1125 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1126 const char *strend, regmatch_info *reginfo)
1129 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
1133 register STRLEN uskip;
1137 register I32 tmp = 1; /* Scratch variable? */
1138 register const bool do_utf8 = PL_reg_match_utf8;
1140 /* We know what class it must start with. */
1144 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1145 !UTF8_IS_INVARIANT((U8)s[0]) ?
1146 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1147 REGINCLASS(prog, c, (U8*)s));
1150 while (s < strend) {
1153 if (REGINCLASS(prog, c, (U8*)s) ||
1154 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1155 /* The assignment of 2 is intentional:
1156 * for the folded sharp s, the skip is 2. */
1157 (skip = SHARP_S_SKIP))) {
1158 if (tmp && (!reginfo || regtry(reginfo, &s)))
1171 if (tmp && (!reginfo || regtry(reginfo, &s)))
1179 ln = STR_LEN(c); /* length to match in octets/bytes */
1180 lnc = (I32) ln; /* length to match in characters */
1182 STRLEN ulen1, ulen2;
1184 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1185 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1186 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1188 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1189 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1191 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1193 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1196 while (sm < ((U8 *) m + ln)) {
1211 c2 = PL_fold_locale[c1];
1213 e = HOP3c(strend, -((I32)lnc), s);
1215 if (!reginfo && e < s)
1216 e = s; /* Due to minlen logic of intuit() */
1218 /* The idea in the EXACTF* cases is to first find the
1219 * first character of the EXACTF* node and then, if
1220 * necessary, case-insensitively compare the full
1221 * text of the node. The c1 and c2 are the first
1222 * characters (though in Unicode it gets a bit
1223 * more complicated because there are more cases
1224 * than just upper and lower: one needs to use
1225 * the so-called folding case for case-insensitive
1226 * matching (called "loose matching" in Unicode).
1227 * ibcmp_utf8() will do just that. */
1231 U8 tmpbuf [UTF8_MAXBYTES+1];
1232 STRLEN len, foldlen;
1233 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1235 /* Upper and lower of 1st char are equal -
1236 * probably not a "letter". */
1238 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1240 REXEC_FBC_EXACTISH_CHECK(c == c1);
1245 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1248 /* Handle some of the three Greek sigmas cases.
1249 * Note that not all the possible combinations
1250 * are handled here: some of them are handled
1251 * by the standard folding rules, and some of
1252 * them (the character class or ANYOF cases)
1253 * are handled during compiletime in
1254 * regexec.c:S_regclass(). */
1255 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1256 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1257 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1259 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1265 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1267 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1271 PL_reg_flags |= RF_tainted;
1278 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1279 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1281 tmp = ((OP(c) == BOUND ?
1282 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1283 LOAD_UTF8_CHARCLASS_ALNUM();
1284 REXEC_FBC_UTF8_SCAN(
1285 if (tmp == !(OP(c) == BOUND ?
1286 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1287 isALNUM_LC_utf8((U8*)s)))
1295 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1296 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1299 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1305 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1309 PL_reg_flags |= RF_tainted;
1316 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1317 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1319 tmp = ((OP(c) == NBOUND ?
1320 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1321 LOAD_UTF8_CHARCLASS_ALNUM();
1322 REXEC_FBC_UTF8_SCAN(
1323 if (tmp == !(OP(c) == NBOUND ?
1324 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1325 isALNUM_LC_utf8((U8*)s)))
1327 else REXEC_FBC_TRYIT;
1331 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1332 tmp = ((OP(c) == NBOUND ?
1333 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1336 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1338 else REXEC_FBC_TRYIT;
1341 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1345 REXEC_FBC_CSCAN_PRELOAD(
1346 LOAD_UTF8_CHARCLASS_ALNUM(),
1347 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1351 REXEC_FBC_CSCAN_TAINT(
1352 isALNUM_LC_utf8((U8*)s),
1356 REXEC_FBC_CSCAN_PRELOAD(
1357 LOAD_UTF8_CHARCLASS_ALNUM(),
1358 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1362 REXEC_FBC_CSCAN_TAINT(
1363 !isALNUM_LC_utf8((U8*)s),
1367 REXEC_FBC_CSCAN_PRELOAD(
1368 LOAD_UTF8_CHARCLASS_SPACE(),
1369 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1373 REXEC_FBC_CSCAN_TAINT(
1374 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1378 REXEC_FBC_CSCAN_PRELOAD(
1379 LOAD_UTF8_CHARCLASS_SPACE(),
1380 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1384 REXEC_FBC_CSCAN_TAINT(
1385 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1389 REXEC_FBC_CSCAN_PRELOAD(
1390 LOAD_UTF8_CHARCLASS_DIGIT(),
1391 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1395 REXEC_FBC_CSCAN_TAINT(
1396 isDIGIT_LC_utf8((U8*)s),
1400 REXEC_FBC_CSCAN_PRELOAD(
1401 LOAD_UTF8_CHARCLASS_DIGIT(),
1402 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1406 REXEC_FBC_CSCAN_TAINT(
1407 !isDIGIT_LC_utf8((U8*)s),
1413 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1414 trie_type = do_utf8 ?
1415 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1417 /* what trie are we using right now */
1419 = (reg_ac_data*)prog->data->data[ ARG( c ) ];
1420 reg_trie_data *trie=aho->trie;
1422 const char *last_start = strend - trie->minlen;
1424 const char *real_start = s;
1426 STRLEN maxlen = trie->maxlen;
1428 U8 **points; /* map of where we were in the input string
1429 when reading a given char. For ASCII this
1430 is unnecessary overhead as the relationship
1431 is always 1:1, but for unicode, especially
1432 case folded unicode this is not true. */
1433 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1437 GET_RE_DEBUG_FLAGS_DECL;
1439 /* We can't just allocate points here. We need to wrap it in
1440 * an SV so it gets freed properly if there is a croak while
1441 * running the match */
1444 sv_points=newSV(maxlen * sizeof(U8 *));
1445 SvCUR_set(sv_points,
1446 maxlen * sizeof(U8 *));
1447 SvPOK_on(sv_points);
1448 sv_2mortal(sv_points);
1449 points=(U8**)SvPV_nolen(sv_points );
1450 if ( trie_type != trie_utf8_fold
1451 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1454 bitmap=(U8*)trie->bitmap;
1456 bitmap=(U8*)ANYOF_BITMAP(c);
1458 /* this is the Aho-Corasick algorithm modified a touch
1459 to include special handling for long "unknown char"
1460 sequences. The basic idea being that we use AC as long
1461 as we are dealing with a possible matching char, when
1462 we encounter an unknown char (and we have not encountered
1463 an accepting state) we scan forward until we find a legal
1465 AC matching is basically that of trie matching, except
1466 that when we encounter a failing transition, we fall back
1467 to the current states "fail state", and try the current char
1468 again, a process we repeat until we reach the root state,
1469 state 1, or a legal transition. If we fail on the root state
1470 then we can either terminate if we have reached an accepting
1471 state previously, or restart the entire process from the beginning
1475 while (s <= last_start) {
1476 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1484 U8 *uscan = (U8*)NULL;
1485 U8 *leftmost = NULL;
1487 U32 accepted_word= 0;
1491 while ( state && uc <= (U8*)strend ) {
1493 U32 word = aho->states[ state ].wordnum;
1497 DEBUG_TRIE_EXECUTE_r(
1498 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1499 dump_exec_pos( (char *)uc, c, strend, real_start,
1500 (char *)uc, do_utf8 );
1501 PerlIO_printf( Perl_debug_log,
1502 " Scanning for legal start char...\n");
1505 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1510 if (uc >(U8*)last_start) break;
1514 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1515 if (!leftmost || lpos < leftmost) {
1516 DEBUG_r(accepted_word=word);
1522 points[pointpos++ % maxlen]= uc;
1523 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
1524 uvc, charid, foldlen, foldbuf, uniflags);
1525 DEBUG_TRIE_EXECUTE_r({
1526 dump_exec_pos( (char *)uc, c, strend, real_start,
1528 PerlIO_printf(Perl_debug_log,
1529 " Charid:%3u CP:%4"UVxf" ",
1535 word = aho->states[ state ].wordnum;
1537 base = aho->states[ state ].trans.base;
1539 DEBUG_TRIE_EXECUTE_r({
1541 dump_exec_pos( (char *)uc, c, strend, real_start,
1543 PerlIO_printf( Perl_debug_log,
1544 "%sState: %4"UVxf", word=%"UVxf,
1545 failed ? " Fail transition to " : "",
1546 (UV)state, (UV)word);
1551 (base + charid > trie->uniquecharcount )
1552 && (base + charid - 1 - trie->uniquecharcount
1554 && trie->trans[base + charid - 1 -
1555 trie->uniquecharcount].check == state
1556 && (tmp=trie->trans[base + charid - 1 -
1557 trie->uniquecharcount ].next))
1559 DEBUG_TRIE_EXECUTE_r(
1560 PerlIO_printf( Perl_debug_log," - legal\n"));
1565 DEBUG_TRIE_EXECUTE_r(
1566 PerlIO_printf( Perl_debug_log," - fail\n"));
1568 state = aho->fail[state];
1572 /* we must be accepting here */
1573 DEBUG_TRIE_EXECUTE_r(
1574 PerlIO_printf( Perl_debug_log," - accepting\n"));
1583 if (!state) state = 1;
1586 if ( aho->states[ state ].wordnum ) {
1587 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1588 if (!leftmost || lpos < leftmost) {
1589 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1594 s = (char*)leftmost;
1595 DEBUG_TRIE_EXECUTE_r({
1597 Perl_debug_log,"Matches word #%"UVxf" at position %d. Trying full pattern...\n",
1598 (UV)accepted_word, s - real_start
1601 if (!reginfo || regtry(reginfo, &s)) {
1607 DEBUG_TRIE_EXECUTE_r({
1608 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1611 DEBUG_TRIE_EXECUTE_r(
1612 PerlIO_printf( Perl_debug_log,"No match.\n"));
1621 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1630 - regexec_flags - match a regexp against a string
1633 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1634 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1635 /* strend: pointer to null at end of string */
1636 /* strbeg: real beginning of string */
1637 /* minend: end of match must be >=minend after stringarg. */
1638 /* data: May be used for some additional optimizations. */
1639 /* nosave: For optimizations. */
1642 /*register*/ char *s;
1643 register regnode *c;
1644 /*register*/ char *startpos = stringarg;
1645 I32 minlen; /* must match at least this many chars */
1646 I32 dontbother = 0; /* how many characters not to try at end */
1647 I32 end_shift = 0; /* Same for the end. */ /* CC */
1648 I32 scream_pos = -1; /* Internal iterator of scream. */
1649 char *scream_olds = NULL;
1650 SV* const oreplsv = GvSV(PL_replgv);
1651 const bool do_utf8 = (bool)DO_UTF8(sv);
1654 regmatch_info reginfo; /* create some info to pass to regtry etc */
1656 GET_RE_DEBUG_FLAGS_DECL;
1658 PERL_UNUSED_ARG(data);
1660 /* Be paranoid... */
1661 if (prog == NULL || startpos == NULL) {
1662 Perl_croak(aTHX_ "NULL regexp parameter");
1666 multiline = prog->reganch & PMf_MULTILINE;
1667 reginfo.prog = prog;
1669 RX_MATCH_UTF8_set(prog, do_utf8);
1671 debug_start_match(prog, do_utf8, startpos, strend,
1675 minlen = prog->minlen;
1677 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1678 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1679 "String too short [regexec_flags]...\n"));
1684 /* Check validity of program. */
1685 if (UCHARAT(prog->program) != REG_MAGIC) {
1686 Perl_croak(aTHX_ "corrupted regexp program");
1690 PL_reg_eval_set = 0;
1693 if (prog->reganch & ROPT_UTF8)
1694 PL_reg_flags |= RF_utf8;
1696 /* Mark beginning of line for ^ and lookbehind. */
1697 reginfo.bol = startpos; /* XXX not used ??? */
1701 /* Mark end of line for $ (and such) */
1704 /* see how far we have to get to not match where we matched before */
1705 reginfo.till = startpos+minend;
1707 /* If there is a "must appear" string, look for it. */
1710 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to set reginfo->ganch */
1713 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1714 reginfo.ganch = startpos;
1715 else if (sv && SvTYPE(sv) >= SVt_PVMG
1717 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1718 && mg->mg_len >= 0) {
1719 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1720 if (prog->reganch & ROPT_ANCH_GPOS) {
1721 if (s > reginfo.ganch)
1726 else /* pos() not defined */
1727 reginfo.ganch = strbeg;
1729 if (PL_curpm && (PM_GETRE(PL_curpm) == prog)) {
1732 /* We have to be careful. If the previous successful match
1733 was from this regex we don't want a subsequent paritally
1734 successful match to clobber the old results.
1735 So when we detect this possibility we add a swap buffer
1736 to the re, and switch the buffer each match. If we fail
1737 we switch it back, otherwise we leave it swapped.
1739 Newxz(prog->swap, 1, regexp_paren_ofs);
1740 /* no need to copy these */
1741 Newxz(prog->swap->startp, prog->nparens + 1, I32);
1742 Newxz(prog->swap->endp, prog->nparens + 1, I32);
1744 t = prog->swap->startp;
1745 prog->swap->startp = prog->startp;
1747 t = prog->swap->endp;
1748 prog->swap->endp = prog->endp;
1751 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1752 re_scream_pos_data d;
1754 d.scream_olds = &scream_olds;
1755 d.scream_pos = &scream_pos;
1756 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1758 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1759 goto phooey; /* not present */
1765 /* Simplest case: anchored match need be tried only once. */
1766 /* [unless only anchor is BOL and multiline is set] */
1767 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1768 if (s == startpos && regtry(®info, &startpos))
1770 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1771 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1776 dontbother = minlen - 1;
1777 end = HOP3c(strend, -dontbother, strbeg) - 1;
1778 /* for multiline we only have to try after newlines */
1779 if (prog->check_substr || prog->check_utf8) {
1783 if (regtry(®info, &s))
1788 if (prog->reganch & RE_USE_INTUIT) {
1789 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1800 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1801 if (regtry(®info, &s))
1808 } else if (ROPT_GPOS_CHECK == (prog->reganch & ROPT_GPOS_CHECK))
1810 /* the warning about reginfo.ganch being used without intialization
1811 is bogus -- we set it above, when prog->reganch & ROPT_GPOS_SEEN
1812 and we only enter this block when the same bit is set. */
1813 if (regtry(®info, ®info.ganch))
1818 /* Messy cases: unanchored match. */
1819 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1820 /* we have /x+whatever/ */
1821 /* it must be a one character string (XXXX Except UTF?) */
1826 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1827 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1828 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1833 DEBUG_EXECUTE_r( did_match = 1 );
1834 if (regtry(®info, &s)) goto got_it;
1836 while (s < strend && *s == ch)
1844 DEBUG_EXECUTE_r( did_match = 1 );
1845 if (regtry(®info, &s)) goto got_it;
1847 while (s < strend && *s == ch)
1852 DEBUG_EXECUTE_r(if (!did_match)
1853 PerlIO_printf(Perl_debug_log,
1854 "Did not find anchored character...\n")
1857 else if (prog->anchored_substr != NULL
1858 || prog->anchored_utf8 != NULL
1859 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1860 && prog->float_max_offset < strend - s)) {
1865 char *last1; /* Last position checked before */
1869 if (prog->anchored_substr || prog->anchored_utf8) {
1870 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1871 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1872 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1873 back_max = back_min = prog->anchored_offset;
1875 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1876 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1877 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1878 back_max = prog->float_max_offset;
1879 back_min = prog->float_min_offset;
1883 if (must == &PL_sv_undef)
1884 /* could not downgrade utf8 check substring, so must fail */
1890 last = HOP3c(strend, /* Cannot start after this */
1891 -(I32)(CHR_SVLEN(must)
1892 - (SvTAIL(must) != 0) + back_min), strbeg);
1895 last1 = HOPc(s, -1);
1897 last1 = s - 1; /* bogus */
1899 /* XXXX check_substr already used to find "s", can optimize if
1900 check_substr==must. */
1902 dontbother = end_shift;
1903 strend = HOPc(strend, -dontbother);
1904 while ( (s <= last) &&
1905 ((flags & REXEC_SCREAM)
1906 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1907 end_shift, &scream_pos, 0))
1908 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1909 (unsigned char*)strend, must,
1910 multiline ? FBMrf_MULTILINE : 0))) ) {
1911 /* we may be pointing at the wrong string */
1912 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1913 s = strbeg + (s - SvPVX_const(sv));
1914 DEBUG_EXECUTE_r( did_match = 1 );
1915 if (HOPc(s, -back_max) > last1) {
1916 last1 = HOPc(s, -back_min);
1917 s = HOPc(s, -back_max);
1920 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1922 last1 = HOPc(s, -back_min);
1926 while (s <= last1) {
1927 if (regtry(®info, &s))
1933 while (s <= last1) {
1934 if (regtry(®info, &s))
1940 DEBUG_EXECUTE_r(if (!did_match) {
1941 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1942 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1943 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1944 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1945 ? "anchored" : "floating"),
1946 quoted, RE_SV_TAIL(must));
1950 else if ( (c = prog->regstclass) ) {
1952 const OPCODE op = OP(prog->regstclass);
1953 /* don't bother with what can't match */
1954 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1955 strend = HOPc(strend, -(minlen - 1));
1958 SV * const prop = sv_newmortal();
1959 regprop(prog, prop, c);
1961 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1963 PerlIO_printf(Perl_debug_log,
1964 "Matching stclass %.*s against %s (%d chars)\n",
1965 (int)SvCUR(prop), SvPVX_const(prop),
1966 quoted, (int)(strend - s));
1969 if (find_byclass(prog, c, s, strend, ®info))
1971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
1975 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1980 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1981 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1982 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1984 if (flags & REXEC_SCREAM) {
1985 last = screaminstr(sv, float_real, s - strbeg,
1986 end_shift, &scream_pos, 1); /* last one */
1988 last = scream_olds; /* Only one occurrence. */
1989 /* we may be pointing at the wrong string */
1990 else if (RX_MATCH_COPIED(prog))
1991 s = strbeg + (s - SvPVX_const(sv));
1995 const char * const little = SvPV_const(float_real, len);
1997 if (SvTAIL(float_real)) {
1998 if (memEQ(strend - len + 1, little, len - 1))
1999 last = strend - len + 1;
2000 else if (!multiline)
2001 last = memEQ(strend - len, little, len)
2002 ? strend - len : NULL;
2008 last = rninstr(s, strend, little, little + len);
2010 last = strend; /* matching "$" */
2015 PerlIO_printf(Perl_debug_log,
2016 "%sCan't trim the tail, match fails (should not happen)%s\n",
2017 PL_colors[4], PL_colors[5]));
2018 goto phooey; /* Should not happen! */
2020 dontbother = strend - last + prog->float_min_offset;
2022 if (minlen && (dontbother < minlen))
2023 dontbother = minlen - 1;
2024 strend -= dontbother; /* this one's always in bytes! */
2025 /* We don't know much -- general case. */
2028 if (regtry(®info, &s))
2037 if (regtry(®info, &s))
2039 } while (s++ < strend);
2047 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2049 if (PL_reg_eval_set) {
2050 /* Preserve the current value of $^R */
2051 if (oreplsv != GvSV(PL_replgv))
2052 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2053 restored, the value remains
2055 restore_pos(aTHX_ prog);
2057 if (prog->paren_names)
2058 (void)hv_iterinit(prog->paren_names);
2060 /* make sure $`, $&, $', and $digit will work later */
2061 if ( !(flags & REXEC_NOT_FIRST) ) {
2062 RX_MATCH_COPY_FREE(prog);
2063 if (flags & REXEC_COPY_STR) {
2064 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2065 #ifdef PERL_OLD_COPY_ON_WRITE
2067 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2069 PerlIO_printf(Perl_debug_log,
2070 "Copy on write: regexp capture, type %d\n",
2073 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2074 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2075 assert (SvPOKp(prog->saved_copy));
2079 RX_MATCH_COPIED_on(prog);
2080 s = savepvn(strbeg, i);
2086 prog->subbeg = strbeg;
2087 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2094 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2095 PL_colors[4], PL_colors[5]));
2096 if (PL_reg_eval_set)
2097 restore_pos(aTHX_ prog);
2099 /* we failed :-( roll it back */
2101 t = prog->swap->startp;
2102 prog->swap->startp = prog->startp;
2104 t = prog->swap->endp;
2105 prog->swap->endp = prog->endp;
2113 - regtry - try match at specific point
2115 STATIC I32 /* 0 failure, 1 success */
2116 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2122 regexp *prog = reginfo->prog;
2123 GET_RE_DEBUG_FLAGS_DECL;
2124 reginfo->cutpoint=NULL;
2126 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2129 PL_reg_eval_set = RS_init;
2130 DEBUG_EXECUTE_r(DEBUG_s(
2131 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2132 (IV)(PL_stack_sp - PL_stack_base));
2135 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2136 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2138 /* Apparently this is not needed, judging by wantarray. */
2139 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2140 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2143 /* Make $_ available to executed code. */
2144 if (reginfo->sv != DEFSV) {
2146 DEFSV = reginfo->sv;
2149 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2150 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2151 /* prepare for quick setting of pos */
2152 #ifdef PERL_OLD_COPY_ON_WRITE
2153 if (SvIsCOW(reginfo->sv))
2154 sv_force_normal_flags(reginfo->sv, 0);
2156 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2157 &PL_vtbl_mglob, NULL, 0);
2161 PL_reg_oldpos = mg->mg_len;
2162 SAVEDESTRUCTOR_X(restore_pos, prog);
2164 if (!PL_reg_curpm) {
2165 Newxz(PL_reg_curpm, 1, PMOP);
2168 SV* const repointer = newSViv(0);
2169 /* so we know which PL_regex_padav element is PL_reg_curpm */
2170 SvFLAGS(repointer) |= SVf_BREAK;
2171 av_push(PL_regex_padav,repointer);
2172 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2173 PL_regex_pad = AvARRAY(PL_regex_padav);
2177 PM_SETRE(PL_reg_curpm, prog);
2178 PL_reg_oldcurpm = PL_curpm;
2179 PL_curpm = PL_reg_curpm;
2180 if (RX_MATCH_COPIED(prog)) {
2181 /* Here is a serious problem: we cannot rewrite subbeg,
2182 since it may be needed if this match fails. Thus
2183 $` inside (?{}) could fail... */
2184 PL_reg_oldsaved = prog->subbeg;
2185 PL_reg_oldsavedlen = prog->sublen;
2186 #ifdef PERL_OLD_COPY_ON_WRITE
2187 PL_nrs = prog->saved_copy;
2189 RX_MATCH_COPIED_off(prog);
2192 PL_reg_oldsaved = NULL;
2193 prog->subbeg = PL_bostr;
2194 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2196 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2197 prog->startp[0] = *startpos - PL_bostr;
2198 PL_reginput = *startpos;
2199 PL_reglastparen = &prog->lastparen;
2200 PL_reglastcloseparen = &prog->lastcloseparen;
2201 prog->lastparen = 0;
2202 prog->lastcloseparen = 0;
2204 PL_regstartp = prog->startp;
2205 PL_regendp = prog->endp;
2206 if (PL_reg_start_tmpl <= prog->nparens) {
2207 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2208 if(PL_reg_start_tmp)
2209 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2211 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2214 /* XXXX What this code is doing here?!!! There should be no need
2215 to do this again and again, PL_reglastparen should take care of
2218 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2219 * Actually, the code in regcppop() (which Ilya may be meaning by
2220 * PL_reglastparen), is not needed at all by the test suite
2221 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2222 * enough, for building DynaLoader, or otherwise this
2223 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2224 * will happen. Meanwhile, this code *is* needed for the
2225 * above-mentioned test suite tests to succeed. The common theme
2226 * on those tests seems to be returning null fields from matches.
2231 if (prog->nparens) {
2233 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2240 if (regmatch(reginfo, prog->program + 1)) {
2241 PL_regendp[0] = PL_reginput - PL_bostr;
2244 if (reginfo->cutpoint)
2245 *startpos= reginfo->cutpoint;
2246 REGCP_UNWIND(lastcp);
2251 #define sayYES goto yes
2252 #define sayNO goto no
2253 #define sayNO_SILENT goto no_silent
2255 /* we dont use STMT_START/END here because it leads to
2256 "unreachable code" warnings, which are bogus, but distracting. */
2257 #define CACHEsayNO \
2258 if (ST.cache_mask) \
2259 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2262 /* this is used to determine how far from the left messages like
2263 'failed...' are printed. It should be set such that messages
2264 are inline with the regop output that created them.
2266 #define REPORT_CODE_OFF 32
2269 /* Make sure there is a test for this +1 options in re_tests */
2270 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2272 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2273 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2275 #define SLAB_FIRST(s) (&(s)->states[0])
2276 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2278 /* grab a new slab and return the first slot in it */
2280 STATIC regmatch_state *
2283 #if PERL_VERSION < 9
2286 regmatch_slab *s = PL_regmatch_slab->next;
2288 Newx(s, 1, regmatch_slab);
2289 s->prev = PL_regmatch_slab;
2291 PL_regmatch_slab->next = s;
2293 PL_regmatch_slab = s;
2294 return SLAB_FIRST(s);
2298 /* push a new state then goto it */
2300 #define PUSH_STATE_GOTO(state, node) \
2302 st->resume_state = state; \
2305 /* push a new state with success backtracking, then goto it */
2307 #define PUSH_YES_STATE_GOTO(state, node) \
2309 st->resume_state = state; \
2310 goto push_yes_state;
2316 regmatch() - main matching routine
2318 This is basically one big switch statement in a loop. We execute an op,
2319 set 'next' to point the next op, and continue. If we come to a point which
2320 we may need to backtrack to on failure such as (A|B|C), we push a
2321 backtrack state onto the backtrack stack. On failure, we pop the top
2322 state, and re-enter the loop at the state indicated. If there are no more
2323 states to pop, we return failure.
2325 Sometimes we also need to backtrack on success; for example /A+/, where
2326 after successfully matching one A, we need to go back and try to
2327 match another one; similarly for lookahead assertions: if the assertion
2328 completes successfully, we backtrack to the state just before the assertion
2329 and then carry on. In these cases, the pushed state is marked as
2330 'backtrack on success too'. This marking is in fact done by a chain of
2331 pointers, each pointing to the previous 'yes' state. On success, we pop to
2332 the nearest yes state, discarding any intermediate failure-only states.
2333 Sometimes a yes state is pushed just to force some cleanup code to be
2334 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2335 it to free the inner regex.
2337 Note that failure backtracking rewinds the cursor position, while
2338 success backtracking leaves it alone.
2340 A pattern is complete when the END op is executed, while a subpattern
2341 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2342 ops trigger the "pop to last yes state if any, otherwise return true"
2345 A common convention in this function is to use A and B to refer to the two
2346 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2347 the subpattern to be matched possibly multiple times, while B is the entire
2348 rest of the pattern. Variable and state names reflect this convention.
2350 The states in the main switch are the union of ops and failure/success of
2351 substates associated with with that op. For example, IFMATCH is the op
2352 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2353 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2354 successfully matched A and IFMATCH_A_fail is a state saying that we have
2355 just failed to match A. Resume states always come in pairs. The backtrack
2356 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2357 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2358 on success or failure.
2360 The struct that holds a backtracking state is actually a big union, with
2361 one variant for each major type of op. The variable st points to the
2362 top-most backtrack struct. To make the code clearer, within each
2363 block of code we #define ST to alias the relevant union.
2365 Here's a concrete example of a (vastly oversimplified) IFMATCH
2371 #define ST st->u.ifmatch
2373 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2374 ST.foo = ...; // some state we wish to save
2376 // push a yes backtrack state with a resume value of
2377 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2379 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2382 case IFMATCH_A: // we have successfully executed A; now continue with B
2384 bar = ST.foo; // do something with the preserved value
2387 case IFMATCH_A_fail: // A failed, so the assertion failed
2388 ...; // do some housekeeping, then ...
2389 sayNO; // propagate the failure
2396 For any old-timers reading this who are familiar with the old recursive
2397 approach, the code above is equivalent to:
2399 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2408 ...; // do some housekeeping, then ...
2409 sayNO; // propagate the failure
2412 The topmost backtrack state, pointed to by st, is usually free. If you
2413 want to claim it, populate any ST.foo fields in it with values you wish to
2414 save, then do one of
2416 PUSH_STATE_GOTO(resume_state, node);
2417 PUSH_YES_STATE_GOTO(resume_state, node);
2419 which sets that backtrack state's resume value to 'resume_state', pushes a
2420 new free entry to the top of the backtrack stack, then goes to 'node'.
2421 On backtracking, the free slot is popped, and the saved state becomes the
2422 new free state. An ST.foo field in this new top state can be temporarily
2423 accessed to retrieve values, but once the main loop is re-entered, it
2424 becomes available for reuse.
2426 Note that the depth of the backtrack stack constantly increases during the
2427 left-to-right execution of the pattern, rather than going up and down with
2428 the pattern nesting. For example the stack is at its maximum at Z at the
2429 end of the pattern, rather than at X in the following:
2431 /(((X)+)+)+....(Y)+....Z/
2433 The only exceptions to this are lookahead/behind assertions and the cut,
2434 (?>A), which pop all the backtrack states associated with A before
2437 Bascktrack state structs are allocated in slabs of about 4K in size.
2438 PL_regmatch_state and st always point to the currently active state,
2439 and PL_regmatch_slab points to the slab currently containing
2440 PL_regmatch_state. The first time regmatch() is called, the first slab is
2441 allocated, and is never freed until interpreter destruction. When the slab
2442 is full, a new one is allocated and chained to the end. At exit from
2443 regmatch(), slabs allocated since entry are freed.
2448 #define DEBUG_STATE_pp(pp) \
2450 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2451 PerlIO_printf(Perl_debug_log, \
2452 " %*s"pp" %s%s%s%s%s\n", \
2454 reg_name[st->resume_state], \
2455 ((st==yes_state||st==mark_state) ? "[" : ""), \
2456 ((st==yes_state) ? "Y" : ""), \
2457 ((st==mark_state) ? "M" : ""), \
2458 ((st==yes_state||st==mark_state) ? "]" : "") \
2463 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2468 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2469 const char *start, const char *end, const char *blurb)
2471 const bool utf8_pat= prog->reganch & ROPT_UTF8 ? 1 : 0;
2475 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2476 prog->precomp, prog->prelen, 60);
2478 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2479 start, end - start, 60);
2481 PerlIO_printf(Perl_debug_log,
2482 "%s%s REx%s %s against %s\n",
2483 PL_colors[4], blurb, PL_colors[5], s0, s1);
2485 if (do_utf8||utf8_pat)
2486 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2487 utf8_pat ? "pattern" : "",
2488 utf8_pat && do_utf8 ? " and " : "",
2489 do_utf8 ? "string" : ""
2495 S_dump_exec_pos(pTHX_ const char *locinput,
2496 const regnode *scan,
2497 const char *loc_regeol,
2498 const char *loc_bostr,
2499 const char *loc_reg_starttry,
2502 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2503 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2504 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2505 /* The part of the string before starttry has one color
2506 (pref0_len chars), between starttry and current
2507 position another one (pref_len - pref0_len chars),
2508 after the current position the third one.
2509 We assume that pref0_len <= pref_len, otherwise we
2510 decrease pref0_len. */
2511 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2512 ? (5 + taill) - l : locinput - loc_bostr;
2515 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2517 pref0_len = pref_len - (locinput - loc_reg_starttry);
2518 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2519 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2520 ? (5 + taill) - pref_len : loc_regeol - locinput);
2521 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2525 if (pref0_len > pref_len)
2526 pref0_len = pref_len;
2528 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2530 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2531 (locinput - pref_len),pref0_len, 60, 4, 5);
2533 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2534 (locinput - pref_len + pref0_len),
2535 pref_len - pref0_len, 60, 2, 3);
2537 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2538 locinput, loc_regeol - locinput, 10, 0, 1);
2540 const STRLEN tlen=len0+len1+len2;
2541 PerlIO_printf(Perl_debug_log,
2542 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2543 (IV)(locinput - loc_bostr),
2546 (docolor ? "" : "> <"),
2548 (int)(tlen > 19 ? 0 : 19 - tlen),
2555 /* reg_check_named_buff_matched()
2556 * Checks to see if a named buffer has matched. The data array of
2557 * buffer numbers corresponding to the buffer is expected to reside
2558 * in the regexp->data->data array in the slot stored in the ARG() of
2559 * node involved. Note that this routine doesn't actually care about the
2560 * name, that information is not preserved from compilation to execution.
2561 * Returns the index of the leftmost defined buffer with the given name
2562 * or 0 if non of the buffers matched.
2565 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2567 SV *sv_dat=(SV*)rex->data->data[ ARG( scan ) ];
2568 I32 *nums=(I32*)SvPVX(sv_dat);
2569 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2570 if ((I32)*PL_reglastparen >= nums[n] &&
2571 PL_regendp[nums[n]] != -1)
2579 STATIC I32 /* 0 failure, 1 success */
2580 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2582 #if PERL_VERSION < 9
2586 register const bool do_utf8 = PL_reg_match_utf8;
2587 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2589 regexp *rex = reginfo->prog;
2591 regmatch_slab *orig_slab;
2592 regmatch_state *orig_state;
2594 /* the current state. This is a cached copy of PL_regmatch_state */
2595 register regmatch_state *st;
2597 /* cache heavy used fields of st in registers */
2598 register regnode *scan;
2599 register regnode *next;
2600 register U32 n = 0; /* general value; init to avoid compiler warning */
2601 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2602 register char *locinput = PL_reginput;
2603 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2605 bool result = 0; /* return value of S_regmatch */
2606 int depth = 0; /* depth of backtrack stack */
2607 int nochange_depth = 0; /* depth of GOSUB recursion with nochange*/
2608 regmatch_state *yes_state = NULL; /* state to pop to on success of
2610 /* mark_state piggy backs on the yes_state logic so that when we unwind
2611 the stack on success we can update the mark_state as we go */
2612 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2614 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2615 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2617 bool no_final = 0; /* prevent failure from backtracking? */
2618 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2619 char *startpoint = PL_reginput;
2620 SV *popmark = NULL; /* are we looking for a mark? */
2621 SV *sv_commit = NULL; /* last mark name seen in failure */
2622 SV *sv_yes_mark = NULL; /* last mark name we have seen
2623 during a successfull match */
2624 U32 lastopen = 0; /* last open we saw */
2625 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2627 /* these three flags are set by various ops to signal information to
2628 * the very next op. They have a useful lifetime of exactly one loop
2629 * iteration, and are not preserved or restored by state pushes/pops
2631 bool sw = 0; /* the condition value in (?(cond)a|b) */
2632 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2633 int logical = 0; /* the following EVAL is:
2637 or the following IFMATCH/UNLESSM is:
2638 false: plain (?=foo)
2639 true: used as a condition: (?(?=foo))
2643 GET_RE_DEBUG_FLAGS_DECL;
2647 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2649 /* on first ever call to regmatch, allocate first slab */
2650 if (!PL_regmatch_slab) {
2651 Newx(PL_regmatch_slab, 1, regmatch_slab);
2652 PL_regmatch_slab->prev = NULL;
2653 PL_regmatch_slab->next = NULL;
2654 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2657 /* remember current high-water mark for exit */
2658 /* XXX this should be done with SAVE* instead */
2659 orig_slab = PL_regmatch_slab;
2660 orig_state = PL_regmatch_state;
2662 /* grab next free state slot */
2663 st = ++PL_regmatch_state;
2664 if (st > SLAB_LAST(PL_regmatch_slab))
2665 st = PL_regmatch_state = S_push_slab(aTHX);
2667 /* Note that nextchr is a byte even in UTF */
2668 nextchr = UCHARAT(locinput);
2670 while (scan != NULL) {
2673 SV * const prop = sv_newmortal();
2674 regnode *rnext=regnext(scan);
2675 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2676 regprop(rex, prop, scan);
2678 PerlIO_printf(Perl_debug_log,
2679 "%3"IVdf":%*s%s(%"IVdf")\n",
2680 (IV)(scan - rex->program), depth*2, "",
2682 (PL_regkind[OP(scan)] == END || !rnext) ?
2683 0 : (IV)(rnext - rex->program));
2686 next = scan + NEXT_OFF(scan);
2689 state_num = OP(scan);
2692 switch (state_num) {
2694 if (locinput == PL_bostr)
2696 /* reginfo->till = reginfo->bol; */
2701 if (locinput == PL_bostr ||
2702 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2708 if (locinput == PL_bostr)
2712 if (locinput == reginfo->ganch)
2718 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2723 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2725 if (PL_regeol - locinput > 1)
2729 if (PL_regeol != locinput)
2733 if (!nextchr && locinput >= PL_regeol)
2736 locinput += PL_utf8skip[nextchr];
2737 if (locinput > PL_regeol)
2739 nextchr = UCHARAT(locinput);
2742 nextchr = UCHARAT(++locinput);
2745 if (!nextchr && locinput >= PL_regeol)
2747 nextchr = UCHARAT(++locinput);
2750 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2753 locinput += PL_utf8skip[nextchr];
2754 if (locinput > PL_regeol)
2756 nextchr = UCHARAT(locinput);
2759 nextchr = UCHARAT(++locinput);
2763 #define ST st->u.trie
2765 /* In this case the charclass data is available inline so
2766 we can fail fast without a lot of extra overhead.
2768 if (scan->flags == EXACT || !do_utf8) {
2769 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2771 PerlIO_printf(Perl_debug_log,
2772 "%*s %sfailed to match trie start class...%s\n",
2773 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2782 /* what type of TRIE am I? (utf8 makes this contextual) */
2783 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2784 trie_type = do_utf8 ?
2785 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2788 /* what trie are we using right now */
2789 reg_trie_data * const trie
2790 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2791 U32 state = trie->startstate;
2793 if (trie->bitmap && trie_type != trie_utf8_fold &&
2794 !TRIE_BITMAP_TEST(trie,*locinput)
2796 if (trie->states[ state ].wordnum) {
2798 PerlIO_printf(Perl_debug_log,
2799 "%*s %smatched empty string...%s\n",
2800 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2805 PerlIO_printf(Perl_debug_log,
2806 "%*s %sfailed to match trie start class...%s\n",
2807 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2814 U8 *uc = ( U8* )locinput;
2818 U8 *uscan = (U8*)NULL;
2820 SV *sv_accept_buff = NULL;
2821 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2823 ST.accepted = 0; /* how many accepting states we have seen */
2825 ST.jump = trie->jump;
2829 traverse the TRIE keeping track of all accepting states
2830 we transition through until we get to a failing node.
2833 while ( state && uc <= (U8*)PL_regeol ) {
2834 U32 base = trie->states[ state ].trans.base;
2837 /* We use charid to hold the wordnum as we don't use it
2838 for charid until after we have done the wordnum logic.
2839 We define an alias just so that the wordnum logic reads
2842 #define got_wordnum charid
2843 got_wordnum = trie->states[ state ].wordnum;
2845 if ( got_wordnum ) {
2846 if ( ! ST.accepted ) {
2849 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2850 sv_accept_buff=newSV(bufflen *
2851 sizeof(reg_trie_accepted) - 1);
2852 SvCUR_set(sv_accept_buff, 0);
2853 SvPOK_on(sv_accept_buff);
2854 sv_2mortal(sv_accept_buff);
2857 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2860 if (ST.accepted >= bufflen) {
2862 ST.accept_buff =(reg_trie_accepted*)
2863 SvGROW(sv_accept_buff,
2864 bufflen * sizeof(reg_trie_accepted));
2866 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2867 + sizeof(reg_trie_accepted));
2870 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2871 ST.accept_buff[ST.accepted].endpos = uc;
2873 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2877 DEBUG_TRIE_EXECUTE_r({
2878 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2879 PerlIO_printf( Perl_debug_log,
2880 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2881 2+depth * 2, "", PL_colors[4],
2882 (UV)state, (UV)ST.accepted );
2886 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
2887 uvc, charid, foldlen, foldbuf, uniflags);
2890 (base + charid > trie->uniquecharcount )
2891 && (base + charid - 1 - trie->uniquecharcount
2893 && trie->trans[base + charid - 1 -
2894 trie->uniquecharcount].check == state)
2896 state = trie->trans[base + charid - 1 -
2897 trie->uniquecharcount ].next;
2908 DEBUG_TRIE_EXECUTE_r(
2909 PerlIO_printf( Perl_debug_log,
2910 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2911 charid, uvc, (UV)state, PL_colors[5] );
2918 PerlIO_printf( Perl_debug_log,
2919 "%*s %sgot %"IVdf" possible matches%s\n",
2920 REPORT_CODE_OFF + depth * 2, "",
2921 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2926 case TRIE_next_fail: /* we failed - try next alterative */
2931 if ( ST.accepted == 1 ) {
2932 /* only one choice left - just continue */
2934 reg_trie_data * const trie
2935 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2936 SV ** const tmp = av_fetch( trie->words,
2937 ST.accept_buff[ 0 ].wordnum-1, 0 );
2938 PerlIO_printf( Perl_debug_log,
2939 "%*s %sonly one match left: #%d <%s>%s\n",
2940 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2941 ST.accept_buff[ 0 ].wordnum,
2942 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2945 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2946 /* in this case we free tmps/leave before we call regmatch
2947 as we wont be using accept_buff again. */
2949 locinput = PL_reginput;
2950 nextchr = UCHARAT(locinput);
2951 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
2954 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
2955 if (!has_cutgroup) {
2960 PUSH_YES_STATE_GOTO(TRIE_next, scan);
2963 continue; /* execute rest of RE */
2966 if (!ST.accepted-- ) {
2968 PerlIO_printf( Perl_debug_log,
2969 "%*s %sTRIE failed...%s\n",
2970 REPORT_CODE_OFF+depth*2, "",
2980 There are at least two accepting states left. Presumably
2981 the number of accepting states is going to be low,
2982 typically two. So we simply scan through to find the one
2983 with lowest wordnum. Once we find it, we swap the last
2984 state into its place and decrement the size. We then try to
2985 match the rest of the pattern at the point where the word
2986 ends. If we succeed, control just continues along the
2987 regex; if we fail we return here to try the next accepting
2994 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
2995 DEBUG_TRIE_EXECUTE_r(
2996 PerlIO_printf( Perl_debug_log,
2997 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2998 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
2999 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3000 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3003 if (ST.accept_buff[cur].wordnum <
3004 ST.accept_buff[best].wordnum)
3009 reg_trie_data * const trie
3010 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
3011 SV ** const tmp = av_fetch( trie->words,
3012 ST.accept_buff[ best ].wordnum - 1, 0 );
3013 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3015 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3016 PerlIO_printf( Perl_debug_log,
3017 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3018 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3019 ST.accept_buff[best].wordnum,
3020 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
3021 REG_NODE_NUM(nextop),
3025 if ( best<ST.accepted ) {
3026 reg_trie_accepted tmp = ST.accept_buff[ best ];
3027 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3028 ST.accept_buff[ ST.accepted ] = tmp;
3031 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3032 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3036 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3040 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3043 PUSH_STATE_GOTO(TRIE_next, scan);
3056 char *s = STRING(scan);
3058 if (do_utf8 != UTF) {
3059 /* The target and the pattern have differing utf8ness. */
3061 const char * const e = s + ln;
3064 /* The target is utf8, the pattern is not utf8. */
3069 if (NATIVE_TO_UNI(*(U8*)s) !=
3070 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3078 /* The target is not utf8, the pattern is utf8. */
3083 if (NATIVE_TO_UNI(*((U8*)l)) !=
3084 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3092 nextchr = UCHARAT(locinput);
3095 /* The target and the pattern have the same utf8ness. */
3096 /* Inline the first character, for speed. */
3097 if (UCHARAT(s) != nextchr)
3099 if (PL_regeol - locinput < ln)
3101 if (ln > 1 && memNE(s, locinput, ln))
3104 nextchr = UCHARAT(locinput);
3108 PL_reg_flags |= RF_tainted;
3111 char * const s = STRING(scan);
3114 if (do_utf8 || UTF) {
3115 /* Either target or the pattern are utf8. */
3116 const char * const l = locinput;
3117 char *e = PL_regeol;
3119 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3120 l, &e, 0, do_utf8)) {
3121 /* One more case for the sharp s:
3122 * pack("U0U*", 0xDF) =~ /ss/i,
3123 * the 0xC3 0x9F are the UTF-8
3124 * byte sequence for the U+00DF. */
3126 toLOWER(s[0]) == 's' &&
3128 toLOWER(s[1]) == 's' &&
3135 nextchr = UCHARAT(locinput);
3139 /* Neither the target and the pattern are utf8. */
3141 /* Inline the first character, for speed. */
3142 if (UCHARAT(s) != nextchr &&
3143 UCHARAT(s) != ((OP(scan) == EXACTF)
3144 ? PL_fold : PL_fold_locale)[nextchr])
3146 if (PL_regeol - locinput < ln)
3148 if (ln > 1 && (OP(scan) == EXACTF
3149 ? ibcmp(s, locinput, ln)
3150 : ibcmp_locale(s, locinput, ln)))
3153 nextchr = UCHARAT(locinput);
3158 STRLEN inclasslen = PL_regeol - locinput;
3160 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3162 if (locinput >= PL_regeol)
3164 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3165 nextchr = UCHARAT(locinput);
3170 nextchr = UCHARAT(locinput);
3171 if (!REGINCLASS(rex, scan, (U8*)locinput))
3173 if (!nextchr && locinput >= PL_regeol)
3175 nextchr = UCHARAT(++locinput);
3179 /* If we might have the case of the German sharp s
3180 * in a casefolding Unicode character class. */
3182 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3183 locinput += SHARP_S_SKIP;
3184 nextchr = UCHARAT(locinput);
3190 PL_reg_flags |= RF_tainted;
3196 LOAD_UTF8_CHARCLASS_ALNUM();
3197 if (!(OP(scan) == ALNUM
3198 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3199 : isALNUM_LC_utf8((U8*)locinput)))
3203 locinput += PL_utf8skip[nextchr];
3204 nextchr = UCHARAT(locinput);
3207 if (!(OP(scan) == ALNUM
3208 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3210 nextchr = UCHARAT(++locinput);
3213 PL_reg_flags |= RF_tainted;
3216 if (!nextchr && locinput >= PL_regeol)
3219 LOAD_UTF8_CHARCLASS_ALNUM();
3220 if (OP(scan) == NALNUM
3221 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3222 : isALNUM_LC_utf8((U8*)locinput))
3226 locinput += PL_utf8skip[nextchr];
3227 nextchr = UCHARAT(locinput);
3230 if (OP(scan) == NALNUM
3231 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3233 nextchr = UCHARAT(++locinput);
3237 PL_reg_flags |= RF_tainted;
3241 /* was last char in word? */
3243 if (locinput == PL_bostr)
3246 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3248 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3250 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3251 ln = isALNUM_uni(ln);
3252 LOAD_UTF8_CHARCLASS_ALNUM();
3253 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3256 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3257 n = isALNUM_LC_utf8((U8*)locinput);
3261 ln = (locinput != PL_bostr) ?
3262 UCHARAT(locinput - 1) : '\n';
3263 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3265 n = isALNUM(nextchr);
3268 ln = isALNUM_LC(ln);
3269 n = isALNUM_LC(nextchr);
3272 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3273 OP(scan) == BOUNDL))
3277 PL_reg_flags |= RF_tainted;
3283 if (UTF8_IS_CONTINUED(nextchr)) {
3284 LOAD_UTF8_CHARCLASS_SPACE();
3285 if (!(OP(scan) == SPACE
3286 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3287 : isSPACE_LC_utf8((U8*)locinput)))
3291 locinput += PL_utf8skip[nextchr];
3292 nextchr = UCHARAT(locinput);
3295 if (!(OP(scan) == SPACE
3296 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3298 nextchr = UCHARAT(++locinput);
3301 if (!(OP(scan) == SPACE
3302 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3304 nextchr = UCHARAT(++locinput);
3308 PL_reg_flags |= RF_tainted;
3311 if (!nextchr && locinput >= PL_regeol)
3314 LOAD_UTF8_CHARCLASS_SPACE();
3315 if (OP(scan) == NSPACE
3316 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3317 : isSPACE_LC_utf8((U8*)locinput))
3321 locinput += PL_utf8skip[nextchr];
3322 nextchr = UCHARAT(locinput);
3325 if (OP(scan) == NSPACE
3326 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3328 nextchr = UCHARAT(++locinput);
3331 PL_reg_flags |= RF_tainted;
3337 LOAD_UTF8_CHARCLASS_DIGIT();
3338 if (!(OP(scan) == DIGIT
3339 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3340 : isDIGIT_LC_utf8((U8*)locinput)))
3344 locinput += PL_utf8skip[nextchr];
3345 nextchr = UCHARAT(locinput);
3348 if (!(OP(scan) == DIGIT
3349 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3351 nextchr = UCHARAT(++locinput);
3354 PL_reg_flags |= RF_tainted;
3357 if (!nextchr && locinput >= PL_regeol)
3360 LOAD_UTF8_CHARCLASS_DIGIT();
3361 if (OP(scan) == NDIGIT
3362 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3363 : isDIGIT_LC_utf8((U8*)locinput))
3367 locinput += PL_utf8skip[nextchr];
3368 nextchr = UCHARAT(locinput);
3371 if (OP(scan) == NDIGIT
3372 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3374 nextchr = UCHARAT(++locinput);
3377 if (locinput >= PL_regeol)
3380 LOAD_UTF8_CHARCLASS_MARK();
3381 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3383 locinput += PL_utf8skip[nextchr];
3384 while (locinput < PL_regeol &&
3385 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3386 locinput += UTF8SKIP(locinput);
3387 if (locinput > PL_regeol)
3392 nextchr = UCHARAT(locinput);
3399 PL_reg_flags |= RF_tainted;
3404 n = reg_check_named_buff_matched(rex,scan);
3407 type = REF + ( type - NREF );
3414 PL_reg_flags |= RF_tainted;
3418 n = ARG(scan); /* which paren pair */
3421 ln = PL_regstartp[n];
3422 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3423 if (*PL_reglastparen < n || ln == -1)
3424 sayNO; /* Do not match unless seen CLOSEn. */
3425 if (ln == PL_regendp[n])
3429 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3431 const char *e = PL_bostr + PL_regendp[n];
3433 * Note that we can't do the "other character" lookup trick as
3434 * in the 8-bit case (no pun intended) because in Unicode we
3435 * have to map both upper and title case to lower case.
3439 STRLEN ulen1, ulen2;
3440 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3441 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3445 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3446 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3447 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3454 nextchr = UCHARAT(locinput);
3458 /* Inline the first character, for speed. */
3459 if (UCHARAT(s) != nextchr &&
3461 (UCHARAT(s) != (type == REFF
3462 ? PL_fold : PL_fold_locale)[nextchr])))
3464 ln = PL_regendp[n] - ln;
3465 if (locinput + ln > PL_regeol)
3467 if (ln > 1 && (type == REF
3468 ? memNE(s, locinput, ln)
3470 ? ibcmp(s, locinput, ln)
3471 : ibcmp_locale(s, locinput, ln))))
3474 nextchr = UCHARAT(locinput);
3484 #define ST st->u.eval
3488 regnode *startpoint;
3491 case GOSUB: /* /(...(?1))/ */
3492 if (cur_eval && cur_eval->locinput==locinput) {
3493 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3494 Perl_croak(aTHX_ "Infinite recursion in regex");
3495 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3497 "Pattern subroutine nesting without pos change"
3498 " exceeded limit in regex");
3503 (void)ReREFCNT_inc(rex);
3504 if (OP(scan)==GOSUB) {
3505 startpoint = scan + ARG2L(scan);
3506 ST.close_paren = ARG(scan);
3508 startpoint = re->program+1;
3511 goto eval_recurse_doit;
3513 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3514 if (cur_eval && cur_eval->locinput==locinput) {
3515 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3516 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3521 /* execute the code in the {...} */
3523 SV ** const before = SP;
3524 OP_4tree * const oop = PL_op;
3525 COP * const ocurcop = PL_curcop;
3529 PL_op = (OP_4tree*)rex->data->data[n];
3530 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3531 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3532 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3533 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3535 CALLRUNOPS(aTHX); /* Scalar context. */
3538 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3545 PAD_RESTORE_LOCAL(old_comppad);
3546 PL_curcop = ocurcop;
3549 sv_setsv(save_scalar(PL_replgv), ret);
3553 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3556 /* extract RE object from returned value; compiling if
3561 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3562 mg = mg_find(sv, PERL_MAGIC_qr);
3563 else if (SvSMAGICAL(ret)) {
3564 if (SvGMAGICAL(ret))
3565 sv_unmagic(ret, PERL_MAGIC_qr);
3567 mg = mg_find(ret, PERL_MAGIC_qr);
3571 re = (regexp *)mg->mg_obj;
3572 (void)ReREFCNT_inc(re);
3576 const char * const t = SvPV_const(ret, len);
3578 const I32 osize = PL_regsize;
3581 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3582 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3584 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3586 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3592 debug_start_match(re, do_utf8, locinput, PL_regeol,
3593 "Matching embedded");
3595 startpoint = re->program + 1;
3596 ST.close_paren = 0; /* only used for GOSUB */
3597 /* borrowed from regtry */
3598 if (PL_reg_start_tmpl <= re->nparens) {
3599 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3600 if(PL_reg_start_tmp)
3601 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3603 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3606 eval_recurse_doit: /* Share code with GOSUB below this line */
3607 /* run the pattern returned from (??{...}) */
3608 ST.cp = regcppush(0); /* Save *all* the positions. */
3609 REGCP_SET(ST.lastcp);
3611 PL_regstartp = re->startp; /* essentially NOOP on GOSUB */
3612 PL_regendp = re->endp; /* essentially NOOP on GOSUB */
3614 *PL_reglastparen = 0;
3615 *PL_reglastcloseparen = 0;
3616 PL_reginput = locinput;
3619 /* XXXX This is too dramatic a measure... */
3622 ST.toggle_reg_flags = PL_reg_flags;
3623 if (re->reganch & ROPT_UTF8)
3624 PL_reg_flags |= RF_utf8;
3626 PL_reg_flags &= ~RF_utf8;
3627 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3630 ST.prev_curlyx = cur_curlyx;
3634 ST.prev_eval = cur_eval;
3636 /* now continue from first node in postoned RE */
3637 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3640 /* logical is 1, /(?(?{...})X|Y)/ */
3641 sw = (bool)SvTRUE(ret);
3646 case EVAL_AB: /* cleanup after a successful (??{A})B */
3647 /* note: this is called twice; first after popping B, then A */
3648 PL_reg_flags ^= ST.toggle_reg_flags;
3652 cur_eval = ST.prev_eval;
3653 cur_curlyx = ST.prev_curlyx;
3654 /* XXXX This is too dramatic a measure... */
3659 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3660 /* note: this is called twice; first after popping B, then A */
3661 PL_reg_flags ^= ST.toggle_reg_flags;
3664 PL_reginput = locinput;
3665 REGCP_UNWIND(ST.lastcp);
3667 cur_eval = ST.prev_eval;
3668 cur_curlyx = ST.prev_curlyx;
3669 /* XXXX This is too dramatic a measure... */
3675 n = ARG(scan); /* which paren pair */
3676 PL_reg_start_tmp[n] = locinput;
3682 n = ARG(scan); /* which paren pair */
3683 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3684 PL_regendp[n] = locinput - PL_bostr;
3685 /*if (n > PL_regsize)
3687 if (n > *PL_reglastparen)
3688 *PL_reglastparen = n;
3689 *PL_reglastcloseparen = n;
3690 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3698 cursor && OP(cursor)!=END;
3699 cursor=regnext(cursor))
3701 if ( OP(cursor)==CLOSE ){
3703 if ( n <= lastopen ) {
3704 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3705 PL_regendp[n] = locinput - PL_bostr;
3706 /*if (n > PL_regsize)
3708 if (n > *PL_reglastparen)
3709 *PL_reglastparen = n;
3710 *PL_reglastcloseparen = n;
3711 if ( n == ARG(scan) || (cur_eval &&
3712 cur_eval->u.eval.close_paren == n))
3721 n = ARG(scan); /* which paren pair */
3722 sw = (bool)(*PL_reglastparen >= n && PL_regendp[n] != -1);
3725 /* reg_check_named_buff_matched returns 0 for no match */
3726 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3730 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3736 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3738 next = NEXTOPER(NEXTOPER(scan));
3740 next = scan + ARG(scan);
3741 if (OP(next) == IFTHEN) /* Fake one. */
3742 next = NEXTOPER(NEXTOPER(next));
3746 logical = scan->flags;
3749 /*******************************************************************
3751 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3752 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3753 STAR/PLUS/CURLY/CURLYN are used instead.)
3755 A*B is compiled as <CURLYX><A><WHILEM><B>
3757 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3758 state, which contains the current count, initialised to -1. It also sets
3759 cur_curlyx to point to this state, with any previous value saved in the
3762 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3763 since the pattern may possibly match zero times (i.e. it's a while {} loop
3764 rather than a do {} while loop).
3766 Each entry to WHILEM represents a successful match of A. The count in the
3767 CURLYX block is incremented, another WHILEM state is pushed, and execution
3768 passes to A or B depending on greediness and the current count.
3770 For example, if matching against the string a1a2a3b (where the aN are
3771 substrings that match /A/), then the match progresses as follows: (the
3772 pushed states are interspersed with the bits of strings matched so far):
3775 <CURLYX cnt=0><WHILEM>
3776 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3777 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3778 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3779 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3781 (Contrast this with something like CURLYM, which maintains only a single
3785 a1 <CURLYM cnt=1> a2
3786 a1 a2 <CURLYM cnt=2> a3
3787 a1 a2 a3 <CURLYM cnt=3> b
3790 Each WHILEM state block marks a point to backtrack to upon partial failure
3791 of A or B, and also contains some minor state data related to that
3792 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3793 overall state, such as the count, and pointers to the A and B ops.
3795 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3796 must always point to the *current* CURLYX block, the rules are:
3798 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3799 and set cur_curlyx to point the new block.
3801 When popping the CURLYX block after a successful or unsuccessful match,
3802 restore the previous cur_curlyx.
3804 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3805 to the outer one saved in the CURLYX block.
3807 When popping the WHILEM block after a successful or unsuccessful B match,
3808 restore the previous cur_curlyx.
3810 Here's an example for the pattern (AI* BI)*BO
3811 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3814 curlyx backtrack stack
3815 ------ ---------------
3817 CO <CO prev=NULL> <WO>
3818 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3819 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3820 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3822 At this point the pattern succeeds, and we work back down the stack to
3823 clean up, restoring as we go:
3825 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3826 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3827 CO <CO prev=NULL> <WO>
3830 *******************************************************************/
3832 #define ST st->u.curlyx
3834 case CURLYX: /* start of /A*B/ (for complex A) */
3836 /* No need to save/restore up to this paren */
3837 I32 parenfloor = scan->flags;
3839 assert(next); /* keep Coverity happy */
3840 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3843 /* XXXX Probably it is better to teach regpush to support
3844 parenfloor > PL_regsize... */
3845 if (parenfloor > (I32)*PL_reglastparen)
3846 parenfloor = *PL_reglastparen; /* Pessimization... */
3848 ST.prev_curlyx= cur_curlyx;
3850 ST.cp = PL_savestack_ix;
3852 /* these fields contain the state of the current curly.
3853 * they are accessed by subsequent WHILEMs */
3854 ST.parenfloor = parenfloor;
3855 ST.min = ARG1(scan);
3856 ST.max = ARG2(scan);
3857 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3861 ST.count = -1; /* this will be updated by WHILEM */
3862 ST.lastloc = NULL; /* this will be updated by WHILEM */
3864 PL_reginput = locinput;
3865 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
3869 case CURLYX_end: /* just finished matching all of A*B */
3870 if (PL_reg_eval_set){
3871 SV *pres= GvSV(PL_replgv);
3874 sv_setsv(GvSV(PL_replgv), pres);
3879 cur_curlyx = ST.prev_curlyx;
3883 case CURLYX_end_fail: /* just failed to match all of A*B */
3885 cur_curlyx = ST.prev_curlyx;
3891 #define ST st->u.whilem
3893 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
3895 /* see the discussion above about CURLYX/WHILEM */
3897 assert(cur_curlyx); /* keep Coverity happy */
3898 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
3899 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
3900 ST.cache_offset = 0;
3903 PL_reginput = locinput;
3905 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3906 "%*s whilem: matched %ld out of %ld..%ld\n",
3907 REPORT_CODE_OFF+depth*2, "", (long)n,
3908 (long)cur_curlyx->u.curlyx.min,
3909 (long)cur_curlyx->u.curlyx.max)
3912 /* First just match a string of min A's. */
3914 if (n < cur_curlyx->u.curlyx.min) {
3915 cur_curlyx->u.curlyx.lastloc = locinput;
3916 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
3920 /* If degenerate A matches "", assume A done. */
3922 if (locinput == cur_curlyx->u.curlyx.lastloc) {
3923 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3924 "%*s whilem: empty match detected, trying continuation...\n",
3925 REPORT_CODE_OFF+depth*2, "")
3927 goto do_whilem_B_max;
3930 /* super-linear cache processing */
3934 if (!PL_reg_maxiter) {
3935 /* start the countdown: Postpone detection until we
3936 * know the match is not *that* much linear. */
3937 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3938 /* possible overflow for long strings and many CURLYX's */
3939 if (PL_reg_maxiter < 0)
3940 PL_reg_maxiter = I32_MAX;
3941 PL_reg_leftiter = PL_reg_maxiter;
3944 if (PL_reg_leftiter-- == 0) {
3945 /* initialise cache */
3946 const I32 size = (PL_reg_maxiter + 7)/8;
3947 if (PL_reg_poscache) {
3948 if ((I32)PL_reg_poscache_size < size) {
3949 Renew(PL_reg_poscache, size, char);
3950 PL_reg_poscache_size = size;
3952 Zero(PL_reg_poscache, size, char);
3955 PL_reg_poscache_size = size;
3956 Newxz(PL_reg_poscache, size, char);
3958 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3959 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
3960 PL_colors[4], PL_colors[5])
3964 if (PL_reg_leftiter < 0) {
3965 /* have we already failed at this position? */
3967 offset = (scan->flags & 0xf) - 1
3968 + (locinput - PL_bostr) * (scan->flags>>4);
3969 mask = 1 << (offset % 8);
3971 if (PL_reg_poscache[offset] & mask) {
3972 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3973 "%*s whilem: (cache) already tried at this position...\n",
3974 REPORT_CODE_OFF+depth*2, "")
3976 sayNO; /* cache records failure */
3978 ST.cache_offset = offset;
3979 ST.cache_mask = mask;
3983 /* Prefer B over A for minimal matching. */
3985 if (cur_curlyx->u.curlyx.minmod) {
3986 ST.save_curlyx = cur_curlyx;
3987 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
3988 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
3989 REGCP_SET(ST.lastcp);
3990 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
3994 /* Prefer A over B for maximal matching. */
3996 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
3997 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3998 cur_curlyx->u.curlyx.lastloc = locinput;
3999 REGCP_SET(ST.lastcp);
4000 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
4003 goto do_whilem_B_max;
4007 case WHILEM_B_min: /* just matched B in a minimal match */
4008 case WHILEM_B_max: /* just matched B in a maximal match */
4009 cur_curlyx = ST.save_curlyx;
4013 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4014 cur_curlyx = ST.save_curlyx;
4015 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4016 cur_curlyx->u.curlyx.count--;
4020 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4021 REGCP_UNWIND(ST.lastcp);
4024 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4025 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4026 cur_curlyx->u.curlyx.count--;
4030 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4031 REGCP_UNWIND(ST.lastcp);
4032 regcppop(rex); /* Restore some previous $<digit>s? */
4033 PL_reginput = locinput;
4034 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4035 "%*s whilem: failed, trying continuation...\n",
4036 REPORT_CODE_OFF+depth*2, "")
4039 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4040 && ckWARN(WARN_REGEXP)
4041 && !(PL_reg_flags & RF_warned))
4043 PL_reg_flags |= RF_warned;
4044 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4045 "Complex regular subexpression recursion",
4050 ST.save_curlyx = cur_curlyx;
4051 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4052 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4055 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4056 cur_curlyx = ST.save_curlyx;
4057 REGCP_UNWIND(ST.lastcp);
4060 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
4061 /* Maximum greed exceeded */
4062 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4063 && ckWARN(WARN_REGEXP)
4064 && !(PL_reg_flags & RF_warned))
4066 PL_reg_flags |= RF_warned;
4067 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4068 "%s limit (%d) exceeded",
4069 "Complex regular subexpression recursion",
4072 cur_curlyx->u.curlyx.count--;
4076 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4077 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4079 /* Try grabbing another A and see if it helps. */
4080 PL_reginput = locinput;
4081 cur_curlyx->u.curlyx.lastloc = locinput;
4082 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4083 REGCP_SET(ST.lastcp);
4084 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
4088 #define ST st->u.branch
4090 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4091 next = scan + ARG(scan);
4094 scan = NEXTOPER(scan);
4097 case BRANCH: /* /(...|A|...)/ */
4098 scan = NEXTOPER(scan); /* scan now points to inner node */
4099 if ((!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
4102 /* last branch; skip state push and jump direct to node */
4105 ST.lastparen = *PL_reglastparen;
4106 ST.next_branch = next;
4108 PL_reginput = locinput;
4110 /* Now go into the branch */
4112 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4114 PUSH_STATE_GOTO(BRANCH_next, scan);
4118 PL_reginput = locinput;
4119 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4120 (SV*)rex->data->data[ ARG( scan ) ];
4121 PUSH_STATE_GOTO(CUTGROUP_next,next);
4123 case CUTGROUP_next_fail:
4126 if (st->u.mark.mark_name)
4127 sv_commit = st->u.mark.mark_name;
4133 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4138 REGCP_UNWIND(ST.cp);
4139 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4141 *PL_reglastparen = n;
4142 /*dmq: *PL_reglastcloseparen = n; */
4143 scan = ST.next_branch;
4144 /* no more branches? */
4145 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4147 PerlIO_printf( Perl_debug_log,
4148 "%*s %sBRANCH failed...%s\n",
4149 REPORT_CODE_OFF+depth*2, "",
4155 continue; /* execute next BRANCH[J] op */
4163 #define ST st->u.curlym
4165 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4167 /* This is an optimisation of CURLYX that enables us to push
4168 * only a single backtracking state, no matter now many matches
4169 * there are in {m,n}. It relies on the pattern being constant
4170 * length, with no parens to influence future backrefs
4174 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4176 /* if paren positive, emulate an OPEN/CLOSE around A */
4178 U32 paren = ST.me->flags;
4179 if (paren > PL_regsize)
4181 if (paren > *PL_reglastparen)
4182 *PL_reglastparen = paren;
4183 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4191 ST.c1 = CHRTEST_UNINIT;
4194 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4197 curlym_do_A: /* execute the A in /A{m,n}B/ */
4198 PL_reginput = locinput;
4199 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4202 case CURLYM_A: /* we've just matched an A */
4203 locinput = st->locinput;
4204 nextchr = UCHARAT(locinput);
4207 /* after first match, determine A's length: u.curlym.alen */
4208 if (ST.count == 1) {
4209 if (PL_reg_match_utf8) {
4211 while (s < PL_reginput) {
4217 ST.alen = PL_reginput - locinput;
4220 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4223 PerlIO_printf(Perl_debug_log,
4224 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4225 (int)(REPORT_CODE_OFF+(depth*2)), "",
4226 (IV) ST.count, (IV)ST.alen)
4229 locinput = PL_reginput;
4231 if (cur_eval && cur_eval->u.eval.close_paren &&
4232 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4235 if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
4236 goto curlym_do_A; /* try to match another A */
4237 goto curlym_do_B; /* try to match B */
4239 case CURLYM_A_fail: /* just failed to match an A */
4240 REGCP_UNWIND(ST.cp);
4242 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4243 || (cur_eval && cur_eval->u.eval.close_paren &&
4244 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4247 curlym_do_B: /* execute the B in /A{m,n}B/ */
4248 PL_reginput = locinput;
4249 if (ST.c1 == CHRTEST_UNINIT) {
4250 /* calculate c1 and c2 for possible match of 1st char
4251 * following curly */
4252 ST.c1 = ST.c2 = CHRTEST_VOID;
4253 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4254 regnode *text_node = ST.B;
4255 if (! HAS_TEXT(text_node))
4256 FIND_NEXT_IMPT(text_node);
4257 if (HAS_TEXT(text_node)
4258 && PL_regkind[OP(text_node)] != REF)
4260 ST.c1 = (U8)*STRING(text_node);
4262 (OP(text_node) == EXACTF || OP(text_node) == REFF)
4264 : (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4265 ? PL_fold_locale[ST.c1]
4272 PerlIO_printf(Perl_debug_log,
4273 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4274 (int)(REPORT_CODE_OFF+(depth*2)),
4277 if (ST.c1 != CHRTEST_VOID
4278 && UCHARAT(PL_reginput) != ST.c1
4279 && UCHARAT(PL_reginput) != ST.c2)
4281 /* simulate B failing */
4282 state_num = CURLYM_B_fail;
4283 goto reenter_switch;
4287 /* mark current A as captured */
4288 I32 paren = ST.me->flags;
4291 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4292 PL_regendp[paren] = PL_reginput - PL_bostr;
4293 /*dmq: *PL_reglastcloseparen = paren; */
4296 PL_regendp[paren] = -1;
4297 if (cur_eval && cur_eval->u.eval.close_paren &&
4298 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4307 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4310 case CURLYM_B_fail: /* just failed to match a B */
4311 REGCP_UNWIND(ST.cp);
4313 if (ST.count == ARG2(ST.me) /* max */)
4315 goto curlym_do_A; /* try to match a further A */
4317 /* backtrack one A */
4318 if (ST.count == ARG1(ST.me) /* min */)
4321 locinput = HOPc(locinput, -ST.alen);
4322 goto curlym_do_B; /* try to match B */
4325 #define ST st->u.curly
4327 #define CURLY_SETPAREN(paren, success) \
4330 PL_regstartp[paren] = HOPc(locinput, -1) - PL_bostr; \
4331 PL_regendp[paren] = locinput - PL_bostr; \
4332 *PL_reglastcloseparen = paren; \
4335 PL_regendp[paren] = -1; \
4338 case STAR: /* /A*B/ where A is width 1 */
4342 scan = NEXTOPER(scan);
4344 case PLUS: /* /A+B/ where A is width 1 */
4348 scan = NEXTOPER(scan);
4350 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4351 ST.paren = scan->flags; /* Which paren to set */
4352 if (ST.paren > PL_regsize)
4353 PL_regsize = ST.paren;
4354 if (ST.paren > *PL_reglastparen)
4355 *PL_reglastparen = ST.paren;
4356 ST.min = ARG1(scan); /* min to match */
4357 ST.max = ARG2(scan); /* max to match */
4358 if (cur_eval && cur_eval->u.eval.close_paren &&
4359 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4363 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4365 case CURLY: /* /A{m,n}B/ where A is width 1 */
4367 ST.min = ARG1(scan); /* min to match */
4368 ST.max = ARG2(scan); /* max to match */
4369 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4372 * Lookahead to avoid useless match attempts
4373 * when we know what character comes next.
4375 * Used to only do .*x and .*?x, but now it allows
4376 * for )'s, ('s and (?{ ... })'s to be in the way
4377 * of the quantifier and the EXACT-like node. -- japhy
4380 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4382 if (HAS_TEXT(next) || JUMPABLE(next)) {
4384 regnode *text_node = next;
4386 if (! HAS_TEXT(text_node))
4387 FIND_NEXT_IMPT(text_node);
4389 if (! HAS_TEXT(text_node))
4390 ST.c1 = ST.c2 = CHRTEST_VOID;
4392 if (PL_regkind[OP(text_node)] == REF) {
4393 ST.c1 = ST.c2 = CHRTEST_VOID;
4394 goto assume_ok_easy;
4397 s = (U8*)STRING(text_node);
4401 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
4402 ST.c2 = PL_fold[ST.c1];
4403 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4404 ST.c2 = PL_fold_locale[ST.c1];
4407 if (OP(text_node) == EXACTF || OP(text_node) == REFF) {
4408 STRLEN ulen1, ulen2;
4409 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4410 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4412 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4413 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4415 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4417 0 : UTF8_ALLOW_ANY);
4418 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4420 0 : UTF8_ALLOW_ANY);
4422 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4424 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4429 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4436 ST.c1 = ST.c2 = CHRTEST_VOID;
4441 PL_reginput = locinput;
4444 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4447 locinput = PL_reginput;
4449 if (ST.c1 == CHRTEST_VOID)
4450 goto curly_try_B_min;
4452 ST.oldloc = locinput;
4454 /* set ST.maxpos to the furthest point along the
4455 * string that could possibly match */
4456 if (ST.max == REG_INFTY) {
4457 ST.maxpos = PL_regeol - 1;
4459 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4463 int m = ST.max - ST.min;
4464 for (ST.maxpos = locinput;
4465 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4466 ST.maxpos += UTF8SKIP(ST.maxpos);
4469 ST.maxpos = locinput + ST.max - ST.min;
4470 if (ST.maxpos >= PL_regeol)
4471 ST.maxpos = PL_regeol - 1;
4473 goto curly_try_B_min_known;
4477 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4478 locinput = PL_reginput;
4479 if (ST.count < ST.min)
4481 if ((ST.count > ST.min)
4482 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4484 /* A{m,n} must come at the end of the string, there's
4485 * no point in backing off ... */
4487 /* ...except that $ and \Z can match before *and* after
4488 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4489 We may back off by one in this case. */
4490 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4494 goto curly_try_B_max;
4499 case CURLY_B_min_known_fail:
4500 /* failed to find B in a non-greedy match where c1,c2 valid */
4501 if (ST.paren && ST.count)
4502 PL_regendp[ST.paren] = -1;
4504 PL_reginput = locinput; /* Could be reset... */
4505 REGCP_UNWIND(ST.cp);
4506 /* Couldn't or didn't -- move forward. */
4507 ST.oldloc = locinput;
4509 locinput += UTF8SKIP(locinput);
4513 curly_try_B_min_known:
4514 /* find the next place where 'B' could work, then call B */
4518 n = (ST.oldloc == locinput) ? 0 : 1;
4519 if (ST.c1 == ST.c2) {
4521 /* set n to utf8_distance(oldloc, locinput) */
4522 while (locinput <= ST.maxpos &&
4523 utf8n_to_uvchr((U8*)locinput,
4524 UTF8_MAXBYTES, &len,
4525 uniflags) != (UV)ST.c1) {
4531 /* set n to utf8_distance(oldloc, locinput) */
4532 while (locinput <= ST.maxpos) {
4534 const UV c = utf8n_to_uvchr((U8*)locinput,
4535 UTF8_MAXBYTES, &len,
4537 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4545 if (ST.c1 == ST.c2) {
4546 while (locinput <= ST.maxpos &&
4547 UCHARAT(locinput) != ST.c1)
4551 while (locinput <= ST.maxpos
4552 && UCHARAT(locinput) != ST.c1
4553 && UCHARAT(locinput) != ST.c2)
4556 n = locinput - ST.oldloc;
4558 if (locinput > ST.maxpos)
4560 /* PL_reginput == oldloc now */
4563 if (regrepeat(rex, ST.A, n, depth) < n)
4566 PL_reginput = locinput;
4567 CURLY_SETPAREN(ST.paren, ST.count);
4568 if (cur_eval && cur_eval->u.eval.close_paren &&
4569 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4572 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4577 case CURLY_B_min_fail:
4578 /* failed to find B in a non-greedy match where c1,c2 invalid */
4579 if (ST.paren && ST.count)
4580 PL_regendp[ST.paren] = -1;
4582 REGCP_UNWIND(ST.cp);
4583 /* failed -- move forward one */
4584 PL_reginput = locinput;
4585 if (regrepeat(rex, ST.A, 1, depth)) {
4587 locinput = PL_reginput;
4588 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4589 ST.count > 0)) /* count overflow ? */
4592 CURLY_SETPAREN(ST.paren, ST.count);
4593 if (cur_eval && cur_eval->u.eval.close_paren &&
4594 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4597 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4605 /* a successful greedy match: now try to match B */
4606 if (cur_eval && cur_eval->u.eval.close_paren &&
4607 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4612 if (ST.c1 != CHRTEST_VOID)
4613 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4614 UTF8_MAXBYTES, 0, uniflags)
4615 : (UV) UCHARAT(PL_reginput);
4616 /* If it could work, try it. */
4617 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4618 CURLY_SETPAREN(ST.paren, ST.count);
4619 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4624 case CURLY_B_max_fail:
4625 /* failed to find B in a greedy match */
4626 if (ST.paren && ST.count)
4627 PL_regendp[ST.paren] = -1;
4629 REGCP_UNWIND(ST.cp);
4631 if (--ST.count < ST.min)
4633 PL_reginput = locinput = HOPc(locinput, -1);
4634 goto curly_try_B_max;
4641 /* we've just finished A in /(??{A})B/; now continue with B */
4645 st->u.eval.toggle_reg_flags
4646 = cur_eval->u.eval.toggle_reg_flags;
4647 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4649 st->u.eval.prev_rex = rex; /* inner */
4650 rex = cur_eval->u.eval.prev_rex; /* outer */
4651 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4653 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4654 REGCP_SET(st->u.eval.lastcp);
4655 PL_reginput = locinput;
4657 /* Restore parens of the outer rex without popping the
4659 tmpix = PL_savestack_ix;
4660 PL_savestack_ix = cur_eval->u.eval.lastcp;
4662 PL_savestack_ix = tmpix;
4664 st->u.eval.prev_eval = cur_eval;
4665 cur_eval = cur_eval->u.eval.prev_eval;
4667 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
4668 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
4669 PUSH_YES_STATE_GOTO(EVAL_AB,
4670 st->u.eval.prev_eval->u.eval.B); /* match B */
4673 if (locinput < reginfo->till) {
4674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4675 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4677 (long)(locinput - PL_reg_starttry),
4678 (long)(reginfo->till - PL_reg_starttry),
4680 sayNO_SILENT; /* Cannot match: too short. */
4682 PL_reginput = locinput; /* put where regtry can find it */
4683 sayYES; /* Success! */
4685 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4687 PerlIO_printf(Perl_debug_log,
4688 "%*s %ssubpattern success...%s\n",
4689 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4690 PL_reginput = locinput; /* put where regtry can find it */
4691 sayYES; /* Success! */
4694 #define ST st->u.ifmatch
4696 case SUSPEND: /* (?>A) */
4698 PL_reginput = locinput;
4701 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4703 goto ifmatch_trivial_fail_test;
4705 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4707 ifmatch_trivial_fail_test:
4709 char * const s = HOPBACKc(locinput, scan->flags);
4714 sw = 1 - (bool)ST.wanted;
4718 next = scan + ARG(scan);
4726 PL_reginput = locinput;
4730 ST.logical = logical;
4731 /* execute body of (?...A) */
4732 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4735 case IFMATCH_A_fail: /* body of (?...A) failed */
4736 ST.wanted = !ST.wanted;
4739 case IFMATCH_A: /* body of (?...A) succeeded */
4741 sw = (bool)ST.wanted;
4743 else if (!ST.wanted)
4746 if (OP(ST.me) == SUSPEND)
4747 locinput = PL_reginput;
4749 locinput = PL_reginput = st->locinput;
4750 nextchr = UCHARAT(locinput);
4752 scan = ST.me + ARG(ST.me);
4755 continue; /* execute B */
4760 next = scan + ARG(scan);
4765 reginfo->cutpoint = PL_regeol;
4768 PL_reginput = locinput;
4770 sv_yes_mark = sv_commit = (SV*)rex->data->data[ ARG( scan ) ];
4771 PUSH_STATE_GOTO(COMMIT_next,next);
4773 case COMMIT_next_fail:
4780 #define ST st->u.mark
4782 ST.prev_mark = mark_state;
4783 ST.mark_name = sv_commit = sv_yes_mark
4784 = (SV*)rex->data->data[ ARG( scan ) ];
4786 ST.mark_loc = PL_reginput = locinput;
4787 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
4789 case MARKPOINT_next:
4790 mark_state = ST.prev_mark;
4793 case MARKPOINT_next_fail:
4794 if (popmark && sv_eq(ST.mark_name,popmark))
4796 if (ST.mark_loc > startpoint)
4797 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
4798 popmark = NULL; /* we found our mark */
4799 sv_commit = ST.mark_name;
4802 PerlIO_printf(Perl_debug_log,
4803 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
4804 REPORT_CODE_OFF+depth*2, "",
4805 PL_colors[4], sv_commit, PL_colors[5]);
4808 mark_state = ST.prev_mark;
4809 sv_yes_mark = mark_state ?
4810 mark_state->u.mark.mark_name : NULL;
4814 PL_reginput = locinput;
4816 /* (*CUT) : if we fail we cut here*/
4817 ST.mark_name = NULL;
4818 ST.mark_loc = locinput;
4819 PUSH_STATE_GOTO(SKIP_next,next);
4821 /* (*CUT:NAME) : if there is a (*MARK:NAME) fail where it was,
4822 otherwise do nothing. Meaning we need to scan
4824 regmatch_state *cur = mark_state;
4825 SV *find = (SV*)rex->data->data[ ARG( scan ) ];
4828 if ( sv_eq( cur->u.mark.mark_name,
4831 ST.mark_name = find;
4832 PUSH_STATE_GOTO( SKIP_next, next );
4834 cur = cur->u.mark.prev_mark;
4837 /* Didn't find our (*MARK:NAME) so ignore this (*CUT:NAME) */
4839 case SKIP_next_fail:
4841 /* (*CUT:NAME) - Set up to search for the name as we
4842 collapse the stack*/
4843 popmark = ST.mark_name;
4845 /* (*CUT) - No name, we cut here.*/
4846 if (ST.mark_loc > startpoint)
4847 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
4848 /* but we set sv_commit to latest mark_name if there
4849 is one so they can test to see how things lead to this
4852 sv_commit=mark_state->u.mark.mark_name;
4860 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
4861 PTR2UV(scan), OP(scan));
4862 Perl_croak(aTHX_ "regexp memory corruption");
4866 /* switch break jumps here */
4867 scan = next; /* prepare to execute the next op and ... */
4868 continue; /* ... jump back to the top, reusing st */
4872 /* push a state that backtracks on success */
4873 st->u.yes.prev_yes_state = yes_state;
4877 /* push a new regex state, then continue at scan */
4879 regmatch_state *newst;
4882 regmatch_state *cur = st;
4883 regmatch_state *curyes = yes_state;
4885 regmatch_slab *slab = PL_regmatch_slab;
4886 for (;curd > -1;cur--,curd--) {
4887 if (cur < SLAB_FIRST(slab)) {
4889 cur = SLAB_LAST(slab);
4891 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
4892 REPORT_CODE_OFF + 2 + depth * 2,"",
4893 curd, reg_name[cur->resume_state],
4894 (curyes == cur) ? "yes" : ""
4897 curyes = cur->u.yes.prev_yes_state;
4900 DEBUG_STATE_pp("push")
4903 st->locinput = locinput;
4905 if (newst > SLAB_LAST(PL_regmatch_slab))
4906 newst = S_push_slab(aTHX);
4907 PL_regmatch_state = newst;
4909 locinput = PL_reginput;
4910 nextchr = UCHARAT(locinput);
4918 * We get here only if there's trouble -- normally "case END" is
4919 * the terminating point.
4921 Perl_croak(aTHX_ "corrupted regexp pointers");
4927 /* we have successfully completed a subexpression, but we must now
4928 * pop to the state marked by yes_state and continue from there */
4929 assert(st != yes_state);
4931 while (st != yes_state) {
4933 if (st < SLAB_FIRST(PL_regmatch_slab)) {
4934 PL_regmatch_slab = PL_regmatch_slab->prev;
4935 st = SLAB_LAST(PL_regmatch_slab);
4939 DEBUG_STATE_pp("pop (no final)");
4941 DEBUG_STATE_pp("pop (yes)");
4947 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
4948 || yes_state > SLAB_LAST(PL_regmatch_slab))
4950 /* not in this slab, pop slab */
4951 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
4952 PL_regmatch_slab = PL_regmatch_slab->prev;
4953 st = SLAB_LAST(PL_regmatch_slab);
4955 depth -= (st - yes_state);
4958 yes_state = st->u.yes.prev_yes_state;
4959 PL_regmatch_state = st;
4962 locinput= st->locinput;
4963 nextchr = UCHARAT(locinput);
4965 state_num = st->resume_state + no_final;
4966 goto reenter_switch;
4969 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
4970 PL_colors[4], PL_colors[5]));
4977 PerlIO_printf(Perl_debug_log,
4978 "%*s %sfailed...%s\n",
4979 REPORT_CODE_OFF+depth*2, "",
4980 PL_colors[4], PL_colors[5])
4992 /* there's a previous state to backtrack to */
4994 if (st < SLAB_FIRST(PL_regmatch_slab)) {
4995 PL_regmatch_slab = PL_regmatch_slab->prev;
4996 st = SLAB_LAST(PL_regmatch_slab);
4998 PL_regmatch_state = st;
4999 locinput= st->locinput;
5000 nextchr = UCHARAT(locinput);
5002 DEBUG_STATE_pp("pop");
5004 if (yes_state == st)
5005 yes_state = st->u.yes.prev_yes_state;
5007 state_num = st->resume_state + 1; /* failure = success + 1 */
5008 goto reenter_switch;
5013 if (rex->reganch & ROPT_VERBARG_SEEN) {
5014 SV *sv_err = get_sv("REGERROR", 1);
5015 SV *sv_mrk = get_sv("REGMARK", 1);
5017 sv_commit = &PL_sv_no;
5019 sv_yes_mark = &PL_sv_yes;
5022 sv_commit = &PL_sv_yes;
5023 sv_yes_mark = &PL_sv_no;
5025 sv_setsv(sv_err, sv_commit);
5026 sv_setsv(sv_mrk, sv_yes_mark);
5028 /* restore original high-water mark */
5029 PL_regmatch_slab = orig_slab;
5030 PL_regmatch_state = orig_state;
5032 /* free all slabs above current one */
5033 if (orig_slab->next) {
5034 regmatch_slab *sl = orig_slab->next;
5035 orig_slab->next = NULL;
5037 regmatch_slab * const osl = sl;
5047 - regrepeat - repeatedly match something simple, report how many
5050 * [This routine now assumes that it will only match on things of length 1.
5051 * That was true before, but now we assume scan - reginput is the count,
5052 * rather than incrementing count on every character. [Er, except utf8.]]
5055 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5058 register char *scan;
5060 register char *loceol = PL_regeol;
5061 register I32 hardcount = 0;
5062 register bool do_utf8 = PL_reg_match_utf8;
5065 if (max == REG_INFTY)
5067 else if (max < loceol - scan)
5068 loceol = scan + max;
5073 while (scan < loceol && hardcount < max && *scan != '\n') {
5074 scan += UTF8SKIP(scan);
5078 while (scan < loceol && *scan != '\n')
5085 while (scan < loceol && hardcount < max) {
5086 scan += UTF8SKIP(scan);
5096 case EXACT: /* length of string is 1 */
5098 while (scan < loceol && UCHARAT(scan) == c)
5101 case EXACTF: /* length of string is 1 */
5103 while (scan < loceol &&
5104 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5107 case EXACTFL: /* length of string is 1 */
5108 PL_reg_flags |= RF_tainted;
5110 while (scan < loceol &&
5111 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5117 while (hardcount < max && scan < loceol &&
5118 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
5119 scan += UTF8SKIP(scan);
5123 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5130 LOAD_UTF8_CHARCLASS_ALNUM();
5131 while (hardcount < max && scan < loceol &&
5132 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5133 scan += UTF8SKIP(scan);
5137 while (scan < loceol && isALNUM(*scan))
5142 PL_reg_flags |= RF_tainted;
5145 while (hardcount < max && scan < loceol &&
5146 isALNUM_LC_utf8((U8*)scan)) {
5147 scan += UTF8SKIP(scan);
5151 while (scan < loceol && isALNUM_LC(*scan))
5158 LOAD_UTF8_CHARCLASS_ALNUM();
5159 while (hardcount < max && scan < loceol &&
5160 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5161 scan += UTF8SKIP(scan);
5165 while (scan < loceol && !isALNUM(*scan))
5170 PL_reg_flags |= RF_tainted;
5173 while (hardcount < max && scan < loceol &&
5174 !isALNUM_LC_utf8((U8*)scan)) {
5175 scan += UTF8SKIP(scan);
5179 while (scan < loceol && !isALNUM_LC(*scan))
5186 LOAD_UTF8_CHARCLASS_SPACE();
5187 while (hardcount < max && scan < loceol &&
5189 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5190 scan += UTF8SKIP(scan);
5194 while (scan < loceol && isSPACE(*scan))
5199 PL_reg_flags |= RF_tainted;
5202 while (hardcount < max && scan < loceol &&
5203 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5204 scan += UTF8SKIP(scan);
5208 while (scan < loceol && isSPACE_LC(*scan))
5215 LOAD_UTF8_CHARCLASS_SPACE();
5216 while (hardcount < max && scan < loceol &&
5218 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5219 scan += UTF8SKIP(scan);
5223 while (scan < loceol && !isSPACE(*scan))
5228 PL_reg_flags |= RF_tainted;
5231 while (hardcount < max && scan < loceol &&
5232 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5233 scan += UTF8SKIP(scan);
5237 while (scan < loceol && !isSPACE_LC(*scan))
5244 LOAD_UTF8_CHARCLASS_DIGIT();
5245 while (hardcount < max && scan < loceol &&
5246 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5247 scan += UTF8SKIP(scan);
5251 while (scan < loceol && isDIGIT(*scan))
5258 LOAD_UTF8_CHARCLASS_DIGIT();
5259 while (hardcount < max && scan < loceol &&
5260 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5261 scan += UTF8SKIP(scan);
5265 while (scan < loceol && !isDIGIT(*scan))
5269 default: /* Called on something of 0 width. */
5270 break; /* So match right here or not at all. */
5276 c = scan - PL_reginput;
5280 GET_RE_DEBUG_FLAGS_DECL;
5282 SV * const prop = sv_newmortal();
5283 regprop(prog, prop, p);
5284 PerlIO_printf(Perl_debug_log,
5285 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5286 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5294 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5296 - regclass_swash - prepare the utf8 swash
5300 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5306 const struct reg_data * const data = prog ? prog->data : NULL;
5308 if (data && data->count) {
5309 const U32 n = ARG(node);
5311 if (data->what[n] == 's') {
5312 SV * const rv = (SV*)data->data[n];
5313 AV * const av = (AV*)SvRV((SV*)rv);
5314 SV **const ary = AvARRAY(av);
5317 /* See the end of regcomp.c:S_regclass() for
5318 * documentation of these array elements. */
5321 a = SvROK(ary[1]) ? &ary[1] : 0;
5322 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
5326 else if (si && doinit) {
5327 sw = swash_init("utf8", "", si, 1, 0);
5328 (void)av_store(av, 1, sw);
5345 - reginclass - determine if a character falls into a character class
5347 The n is the ANYOF regnode, the p is the target string, lenp
5348 is pointer to the maximum length of how far to go in the p
5349 (if the lenp is zero, UTF8SKIP(p) is used),
5350 do_utf8 tells whether the target string is in UTF-8.
5355 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5358 const char flags = ANYOF_FLAGS(n);
5364 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5365 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5366 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5367 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5368 if (len == (STRLEN)-1)
5369 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5372 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5373 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5376 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5377 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5380 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5384 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5387 if (swash_fetch(sw, p, do_utf8))
5389 else if (flags & ANYOF_FOLD) {
5390 if (!match && lenp && av) {
5392 for (i = 0; i <= av_len(av); i++) {
5393 SV* const sv = *av_fetch(av, i, FALSE);
5395 const char * const s = SvPV_const(sv, len);
5397 if (len <= plen && memEQ(s, (char*)p, len)) {
5405 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5408 to_utf8_fold(p, tmpbuf, &tmplen);
5409 if (swash_fetch(sw, tmpbuf, do_utf8))
5415 if (match && lenp && *lenp == 0)
5416 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5418 if (!match && c < 256) {
5419 if (ANYOF_BITMAP_TEST(n, c))
5421 else if (flags & ANYOF_FOLD) {
5424 if (flags & ANYOF_LOCALE) {
5425 PL_reg_flags |= RF_tainted;
5426 f = PL_fold_locale[c];
5430 if (f != c && ANYOF_BITMAP_TEST(n, f))
5434 if (!match && (flags & ANYOF_CLASS)) {
5435 PL_reg_flags |= RF_tainted;
5437 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5438 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5439 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5440 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5441 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5442 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5443 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5444 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5445 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5446 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5447 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5448 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5449 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5450 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5451 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5452 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5453 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5454 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5455 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5456 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5457 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5458 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5459 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5460 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5461 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5462 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5463 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5464 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5465 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5466 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5467 ) /* How's that for a conditional? */
5474 return (flags & ANYOF_INVERT) ? !match : match;
5478 S_reghop3(U8 *s, I32 off, const U8* lim)
5482 while (off-- && s < lim) {
5483 /* XXX could check well-formedness here */
5488 while (off++ && s > lim) {
5490 if (UTF8_IS_CONTINUED(*s)) {
5491 while (s > lim && UTF8_IS_CONTINUATION(*s))
5494 /* XXX could check well-formedness here */
5501 /* there are a bunch of places where we use two reghop3's that should
5502 be replaced with this routine. but since thats not done yet
5503 we ifdef it out - dmq
5506 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5510 while (off-- && s < rlim) {
5511 /* XXX could check well-formedness here */
5516 while (off++ && s > llim) {
5518 if (UTF8_IS_CONTINUED(*s)) {
5519 while (s > llim && UTF8_IS_CONTINUATION(*s))
5522 /* XXX could check well-formedness here */
5530 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5534 while (off-- && s < lim) {
5535 /* XXX could check well-formedness here */
5542 while (off++ && s > lim) {
5544 if (UTF8_IS_CONTINUED(*s)) {
5545 while (s > lim && UTF8_IS_CONTINUATION(*s))
5548 /* XXX could check well-formedness here */
5557 restore_pos(pTHX_ void *arg)
5560 regexp * const rex = (regexp *)arg;
5561 if (PL_reg_eval_set) {
5562 if (PL_reg_oldsaved) {
5563 rex->subbeg = PL_reg_oldsaved;
5564 rex->sublen = PL_reg_oldsavedlen;
5565 #ifdef PERL_OLD_COPY_ON_WRITE
5566 rex->saved_copy = PL_nrs;
5568 RX_MATCH_COPIED_on(rex);
5570 PL_reg_magic->mg_len = PL_reg_oldpos;
5571 PL_reg_eval_set = 0;
5572 PL_curpm = PL_reg_oldcurpm;
5577 S_to_utf8_substr(pTHX_ register regexp *prog)
5579 if (prog->float_substr && !prog->float_utf8) {
5580 SV* const sv = newSVsv(prog->float_substr);
5581 prog->float_utf8 = sv;
5582 sv_utf8_upgrade(sv);
5583 if (SvTAIL(prog->float_substr))
5585 if (prog->float_substr == prog->check_substr)
5586 prog->check_utf8 = sv;
5588 if (prog->anchored_substr && !prog->anchored_utf8) {
5589 SV* const sv = newSVsv(prog->anchored_substr);
5590 prog->anchored_utf8 = sv;
5591 sv_utf8_upgrade(sv);
5592 if (SvTAIL(prog->anchored_substr))
5594 if (prog->anchored_substr == prog->check_substr)
5595 prog->check_utf8 = sv;
5600 S_to_byte_substr(pTHX_ register regexp *prog)
5603 if (prog->float_utf8 && !prog->float_substr) {
5604 SV* sv = newSVsv(prog->float_utf8);
5605 prog->float_substr = sv;
5606 if (sv_utf8_downgrade(sv, TRUE)) {
5607 if (SvTAIL(prog->float_utf8))
5611 prog->float_substr = sv = &PL_sv_undef;
5613 if (prog->float_utf8 == prog->check_utf8)
5614 prog->check_substr = sv;
5616 if (prog->anchored_utf8 && !prog->anchored_substr) {
5617 SV* sv = newSVsv(prog->anchored_utf8);
5618 prog->anchored_substr = sv;
5619 if (sv_utf8_downgrade(sv, TRUE)) {
5620 if (SvTAIL(prog->anchored_utf8))
5624 prog->anchored_substr = sv = &PL_sv_undef;
5626 if (prog->anchored_utf8 == prog->check_utf8)
5627 prog->check_substr = sv;
5633 * c-indentation-style: bsd
5635 * indent-tabs-mode: t
5638 * ex: set ts=8 sts=4 sw=4 noet: