5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
56 **** Alterations to Henry's code are...
58 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
59 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
61 **** You may distribute under the terms of either the GNU General Public
62 **** License or the Artistic License, as specified in the README file.
64 * Beware that some of this code is subtly aware of the way operator
65 * precedence is structured in regular expressions. Serious changes in
66 * regular-expression syntax might require a total rethink.
69 #define PERL_IN_REGEXEC_C
72 #ifdef PERL_IN_XSUB_RE
78 #define RF_tainted 1 /* tainted information used? */
79 #define RF_warned 2 /* warned about big count? */
81 #define RF_utf8 8 /* Pattern contains multibyte chars? */
83 #define UTF ((PL_reg_flags & RF_utf8) != 0)
85 #define RS_init 1 /* eval environment created */
86 #define RS_set 2 /* replsv value is set */
92 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
98 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
99 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
101 #define HOPc(pos,off) \
102 (char *)(PL_reg_match_utf8 \
103 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
105 #define HOPBACKc(pos, off) \
106 (char*)(PL_reg_match_utf8\
107 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
108 : (pos - off >= PL_bostr) \
112 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
113 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
115 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
116 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
117 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
118 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
119 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
120 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
122 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
124 /* for use after a quantifier and before an EXACT-like node -- japhy */
125 #define JUMPABLE(rn) ( \
126 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
127 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
128 OP(rn) == PLUS || OP(rn) == MINMOD || \
129 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
132 #define HAS_TEXT(rn) ( \
133 PL_regkind[OP(rn)] == EXACT || PL_regkind[OP(rn)] == REF \
137 Search for mandatory following text node; for lookahead, the text must
138 follow but for lookbehind (rn->flags != 0) we skip to the next step.
140 #define FIND_NEXT_IMPT(rn) STMT_START { \
141 while (JUMPABLE(rn)) { \
142 const OPCODE type = OP(rn); \
143 if (type == SUSPEND || PL_regkind[type] == CURLY) \
144 rn = NEXTOPER(NEXTOPER(rn)); \
145 else if (type == PLUS) \
147 else if (type == IFMATCH) \
148 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
149 else rn += NEXT_OFF(rn); \
154 static void restore_pos(pTHX_ void *arg);
157 S_regcppush(pTHX_ I32 parenfloor)
160 const int retval = PL_savestack_ix;
161 #define REGCP_PAREN_ELEMS 4
162 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
164 GET_RE_DEBUG_FLAGS_DECL;
166 if (paren_elems_to_push < 0)
167 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
169 #define REGCP_OTHER_ELEMS 8
170 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
172 for (p = PL_regsize; p > parenfloor; p--) {
173 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
174 SSPUSHINT(PL_regendp[p]);
175 SSPUSHINT(PL_regstartp[p]);
176 SSPUSHPTR(PL_reg_start_tmp[p]);
178 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
179 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
180 (UV)p, (IV)PL_regstartp[p],
181 (IV)(PL_reg_start_tmp[p] - PL_bostr),
185 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
186 SSPUSHPTR(PL_regstartp);
187 SSPUSHPTR(PL_regendp);
188 SSPUSHINT(PL_regsize);
189 SSPUSHINT(*PL_reglastparen);
190 SSPUSHINT(*PL_reglastcloseparen);
191 SSPUSHPTR(PL_reginput);
192 #define REGCP_FRAME_ELEMS 2
193 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
194 * are needed for the regexp context stack bookkeeping. */
195 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
196 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
201 /* These are needed since we do not localize EVAL nodes: */
202 #define REGCP_SET(cp) \
204 PerlIO_printf(Perl_debug_log, \
205 " Setting an EVAL scope, savestack=%"IVdf"\n", \
206 (IV)PL_savestack_ix)); \
209 #define REGCP_UNWIND(cp) \
211 if (cp != PL_savestack_ix) \
212 PerlIO_printf(Perl_debug_log, \
213 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
214 (IV)(cp), (IV)PL_savestack_ix)); \
218 S_regcppop(pTHX_ const regexp *rex)
224 GET_RE_DEBUG_FLAGS_DECL;
226 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
228 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
229 i = SSPOPINT; /* Parentheses elements to pop. */
230 input = (char *) SSPOPPTR;
231 *PL_reglastcloseparen = SSPOPINT;
232 *PL_reglastparen = SSPOPINT;
233 PL_regsize = SSPOPINT;
234 PL_regendp=(I32 *) SSPOPPTR;
235 PL_regstartp=(I32 *) SSPOPPTR;
238 /* Now restore the parentheses context. */
239 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
240 i > 0; i -= REGCP_PAREN_ELEMS) {
242 U32 paren = (U32)SSPOPINT;
243 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
244 PL_regstartp[paren] = SSPOPINT;
246 if (paren <= *PL_reglastparen)
247 PL_regendp[paren] = tmps;
249 PerlIO_printf(Perl_debug_log,
250 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
251 (UV)paren, (IV)PL_regstartp[paren],
252 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
253 (IV)PL_regendp[paren],
254 (paren > *PL_reglastparen ? "(no)" : ""));
258 if (*PL_reglastparen + 1 <= rex->nparens) {
259 PerlIO_printf(Perl_debug_log,
260 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
261 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
265 /* It would seem that the similar code in regtry()
266 * already takes care of this, and in fact it is in
267 * a better location to since this code can #if 0-ed out
268 * but the code in regtry() is needed or otherwise tests
269 * requiring null fields (pat.t#187 and split.t#{13,14}
270 * (as of patchlevel 7877) will fail. Then again,
271 * this code seems to be necessary or otherwise
272 * building DynaLoader will fail:
273 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
275 for (i = *PL_reglastparen + 1; (U32)i <= rex->nparens; i++) {
277 PL_regstartp[i] = -1;
284 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
287 * pregexec and friends
290 #ifndef PERL_IN_XSUB_RE
292 - pregexec - match a regexp against a string
295 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
296 char *strbeg, I32 minend, SV *screamer, U32 nosave)
297 /* strend: pointer to null at end of string */
298 /* strbeg: real beginning of string */
299 /* minend: end of match must be >=minend after stringarg. */
300 /* nosave: For optimizations. */
303 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
304 nosave ? 0 : REXEC_COPY_STR);
309 * Need to implement the following flags for reg_anch:
311 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
313 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
314 * INTUIT_AUTORITATIVE_ML
315 * INTUIT_ONCE_NOML - Intuit can match in one location only.
318 * Another flag for this function: SECOND_TIME (so that float substrs
319 * with giant delta may be not rechecked).
322 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
324 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
325 Otherwise, only SvCUR(sv) is used to get strbeg. */
327 /* XXXX We assume that strpos is strbeg unless sv. */
329 /* XXXX Some places assume that there is a fixed substring.
330 An update may be needed if optimizer marks as "INTUITable"
331 RExen without fixed substrings. Similarly, it is assumed that
332 lengths of all the strings are no more than minlen, thus they
333 cannot come from lookahead.
334 (Or minlen should take into account lookahead.)
335 NOTE: Some of this comment is not correct. minlen does now take account
336 of lookahead/behind. Further research is required. -- demerphq
340 /* A failure to find a constant substring means that there is no need to make
341 an expensive call to REx engine, thus we celebrate a failure. Similarly,
342 finding a substring too deep into the string means that less calls to
343 regtry() should be needed.
345 REx compiler's optimizer found 4 possible hints:
346 a) Anchored substring;
348 c) Whether we are anchored (beginning-of-line or \G);
349 d) First node (of those at offset 0) which may distingush positions;
350 We use a)b)d) and multiline-part of c), and try to find a position in the
351 string which does not contradict any of them.
354 /* Most of decisions we do here should have been done at compile time.
355 The nodes of the REx which we used for the search should have been
356 deleted from the finite automaton. */
359 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
360 char *strend, U32 flags, re_scream_pos_data *data)
363 register I32 start_shift = 0;
364 /* Should be nonnegative! */
365 register I32 end_shift = 0;
370 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
372 register char *other_last = NULL; /* other substr checked before this */
373 char *check_at = NULL; /* check substr found at this pos */
374 const I32 multiline = prog->reganch & PMf_MULTILINE;
376 const char * const i_strpos = strpos;
379 GET_RE_DEBUG_FLAGS_DECL;
381 RX_MATCH_UTF8_set(prog,do_utf8);
383 if (prog->reganch & ROPT_UTF8) {
384 PL_reg_flags |= RF_utf8;
387 debug_start_match(prog, do_utf8, strpos, strend,
388 sv ? "Guessing start of match in sv for"
389 : "Guessing start of match in string for");
392 /* CHR_DIST() would be more correct here but it makes things slow. */
393 if (prog->minlen > strend - strpos) {
394 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
395 "String too short... [re_intuit_start]\n"));
399 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
402 if (!prog->check_utf8 && prog->check_substr)
403 to_utf8_substr(prog);
404 check = prog->check_utf8;
406 if (!prog->check_substr && prog->check_utf8)
407 to_byte_substr(prog);
408 check = prog->check_substr;
410 if (check == &PL_sv_undef) {
411 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
412 "Non-utf8 string cannot match utf8 check string\n"));
415 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
416 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
417 || ( (prog->reganch & ROPT_ANCH_BOL)
418 && !multiline ) ); /* Check after \n? */
421 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
422 | ROPT_IMPLICIT)) /* not a real BOL */
423 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
425 && (strpos != strbeg)) {
426 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
429 if (prog->check_offset_min == prog->check_offset_max &&
430 !(prog->reganch & ROPT_CANY_SEEN)) {
431 /* Substring at constant offset from beg-of-str... */
434 s = HOP3c(strpos, prog->check_offset_min, strend);
437 slen = SvCUR(check); /* >= 1 */
439 if ( strend - s > slen || strend - s < slen - 1
440 || (strend - s == slen && strend[-1] != '\n')) {
441 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
444 /* Now should match s[0..slen-2] */
446 if (slen && (*SvPVX_const(check) != *s
448 && memNE(SvPVX_const(check), s, slen)))) {
450 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
454 else if (*SvPVX_const(check) != *s
455 || ((slen = SvCUR(check)) > 1
456 && memNE(SvPVX_const(check), s, slen)))
459 goto success_at_start;
462 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
464 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
465 end_shift = prog->check_end_shift;
468 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
469 - (SvTAIL(check) != 0);
470 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
472 if (end_shift < eshift)
476 else { /* Can match at random position */
479 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
480 end_shift = prog->check_end_shift;
482 /* end shift should be non negative here */
485 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
487 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
488 (IV)end_shift, prog->precomp);
492 /* Find a possible match in the region s..strend by looking for
493 the "check" substring in the region corrected by start/end_shift. */
496 I32 srch_start_shift = start_shift;
497 I32 srch_end_shift = end_shift;
498 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
499 srch_end_shift -= ((strbeg - s) - srch_start_shift);
500 srch_start_shift = strbeg - s;
502 DEBUG_OPTIMISE_MORE_r({
503 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
504 (IV)prog->check_offset_min,
505 (IV)srch_start_shift,
507 (IV)prog->check_end_shift);
510 if (flags & REXEC_SCREAM) {
511 I32 p = -1; /* Internal iterator of scream. */
512 I32 * const pp = data ? data->scream_pos : &p;
514 if (PL_screamfirst[BmRARE(check)] >= 0
515 || ( BmRARE(check) == '\n'
516 && (BmPREVIOUS(check) == SvCUR(check) - 1)
518 s = screaminstr(sv, check,
519 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
522 /* we may be pointing at the wrong string */
523 if (s && RX_MATCH_COPIED(prog))
524 s = strbeg + (s - SvPVX_const(sv));
526 *data->scream_olds = s;
531 if (prog->reganch & ROPT_CANY_SEEN) {
532 start_point= (U8*)(s + srch_start_shift);
533 end_point= (U8*)(strend - srch_end_shift);
535 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
536 end_point= HOP3(strend, -srch_end_shift, strbeg);
538 DEBUG_OPTIMISE_MORE_r({
539 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
540 (int)(end_point - start_point),
541 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
545 s = fbm_instr( start_point, end_point,
546 check, multiline ? FBMrf_MULTILINE : 0);
549 /* Update the count-of-usability, remove useless subpatterns,
553 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
554 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
555 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
556 (s ? "Found" : "Did not find"),
557 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
558 ? "anchored" : "floating"),
561 (s ? " at offset " : "...\n") );
566 /* Finish the diagnostic message */
567 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
569 /* XXX dmq: first branch is for positive lookbehind...
570 Our check string is offset from the beginning of the pattern.
571 So we need to do any stclass tests offset forward from that
580 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
581 Start with the other substr.
582 XXXX no SCREAM optimization yet - and a very coarse implementation
583 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
584 *always* match. Probably should be marked during compile...
585 Probably it is right to do no SCREAM here...
588 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
589 : (prog->float_substr && prog->anchored_substr))
591 /* Take into account the "other" substring. */
592 /* XXXX May be hopelessly wrong for UTF... */
595 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
598 char * const last = HOP3c(s, -start_shift, strbeg);
600 char * const saved_s = s;
603 t = s - prog->check_offset_max;
604 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
606 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
611 t = HOP3c(t, prog->anchored_offset, strend);
612 if (t < other_last) /* These positions already checked */
614 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
617 /* XXXX It is not documented what units *_offsets are in.
618 We assume bytes, but this is clearly wrong.
619 Meaning this code needs to be carefully reviewed for errors.
623 /* On end-of-str: see comment below. */
624 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
625 if (must == &PL_sv_undef) {
627 DEBUG_r(must = prog->anchored_utf8); /* for debug */
632 HOP3(HOP3(last1, prog->anchored_offset, strend)
633 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
635 multiline ? FBMrf_MULTILINE : 0
638 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
639 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
640 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
641 (s ? "Found" : "Contradicts"),
642 quoted, RE_SV_TAIL(must));
647 if (last1 >= last2) {
648 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
649 ", giving up...\n"));
652 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
653 ", trying floating at offset %ld...\n",
654 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
655 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
656 s = HOP3c(last, 1, strend);
660 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
661 (long)(s - i_strpos)));
662 t = HOP3c(s, -prog->anchored_offset, strbeg);
663 other_last = HOP3c(s, 1, strend);
671 else { /* Take into account the floating substring. */
673 char * const saved_s = s;
676 t = HOP3c(s, -start_shift, strbeg);
678 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
679 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
680 last = HOP3c(t, prog->float_max_offset, strend);
681 s = HOP3c(t, prog->float_min_offset, strend);
684 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
685 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
686 /* fbm_instr() takes into account exact value of end-of-str
687 if the check is SvTAIL(ed). Since false positives are OK,
688 and end-of-str is not later than strend we are OK. */
689 if (must == &PL_sv_undef) {
691 DEBUG_r(must = prog->float_utf8); /* for debug message */
694 s = fbm_instr((unsigned char*)s,
695 (unsigned char*)last + SvCUR(must)
697 must, multiline ? FBMrf_MULTILINE : 0);
699 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
700 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
701 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
702 (s ? "Found" : "Contradicts"),
703 quoted, RE_SV_TAIL(must));
707 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
708 ", giving up...\n"));
711 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
712 ", trying anchored starting at offset %ld...\n",
713 (long)(saved_s + 1 - i_strpos)));
715 s = HOP3c(t, 1, strend);
719 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
720 (long)(s - i_strpos)));
721 other_last = s; /* Fix this later. --Hugo */
731 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
733 DEBUG_OPTIMISE_MORE_r(
734 PerlIO_printf(Perl_debug_log,
735 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
736 (IV)prog->check_offset_min,
737 (IV)prog->check_offset_max,
745 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
747 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
750 /* Fixed substring is found far enough so that the match
751 cannot start at strpos. */
753 if (ml_anch && t[-1] != '\n') {
754 /* Eventually fbm_*() should handle this, but often
755 anchored_offset is not 0, so this check will not be wasted. */
756 /* XXXX In the code below we prefer to look for "^" even in
757 presence of anchored substrings. And we search even
758 beyond the found float position. These pessimizations
759 are historical artefacts only. */
761 while (t < strend - prog->minlen) {
763 if (t < check_at - prog->check_offset_min) {
764 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
765 /* Since we moved from the found position,
766 we definitely contradict the found anchored
767 substr. Due to the above check we do not
768 contradict "check" substr.
769 Thus we can arrive here only if check substr
770 is float. Redo checking for "other"=="fixed".
773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
774 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
775 goto do_other_anchored;
777 /* We don't contradict the found floating substring. */
778 /* XXXX Why not check for STCLASS? */
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
781 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
784 /* Position contradicts check-string */
785 /* XXXX probably better to look for check-string
786 than for "\n", so one should lower the limit for t? */
787 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
788 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
789 other_last = strpos = s = t + 1;
794 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
795 PL_colors[0], PL_colors[1]));
799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
800 PL_colors[0], PL_colors[1]));
804 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
807 /* The found string does not prohibit matching at strpos,
808 - no optimization of calling REx engine can be performed,
809 unless it was an MBOL and we are not after MBOL,
810 or a future STCLASS check will fail this. */
812 /* Even in this situation we may use MBOL flag if strpos is offset
813 wrt the start of the string. */
814 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
815 && (strpos != strbeg) && strpos[-1] != '\n'
816 /* May be due to an implicit anchor of m{.*foo} */
817 && !(prog->reganch & ROPT_IMPLICIT))
822 DEBUG_EXECUTE_r( if (ml_anch)
823 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
824 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
827 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
829 prog->check_utf8 /* Could be deleted already */
830 && --BmUSEFUL(prog->check_utf8) < 0
831 && (prog->check_utf8 == prog->float_utf8)
833 prog->check_substr /* Could be deleted already */
834 && --BmUSEFUL(prog->check_substr) < 0
835 && (prog->check_substr == prog->float_substr)
838 /* If flags & SOMETHING - do not do it many times on the same match */
839 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
840 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
841 if (do_utf8 ? prog->check_substr : prog->check_utf8)
842 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
843 prog->check_substr = prog->check_utf8 = NULL; /* disable */
844 prog->float_substr = prog->float_utf8 = NULL; /* clear */
845 check = NULL; /* abort */
847 /* XXXX This is a remnant of the old implementation. It
848 looks wasteful, since now INTUIT can use many
850 prog->reganch &= ~RE_USE_INTUIT;
857 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
858 /* trie stclasses are too expensive to use here, we are better off to
859 leave it to regmatch itself */
860 if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) {
861 /* minlen == 0 is possible if regstclass is \b or \B,
862 and the fixed substr is ''$.
863 Since minlen is already taken into account, s+1 is before strend;
864 accidentally, minlen >= 1 guaranties no false positives at s + 1
865 even for \b or \B. But (minlen? 1 : 0) below assumes that
866 regstclass does not come from lookahead... */
867 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
868 This leaves EXACTF only, which is dealt with in find_byclass(). */
869 const U8* const str = (U8*)STRING(prog->regstclass);
870 const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT
871 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
874 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
875 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
876 else if (prog->float_substr || prog->float_utf8)
877 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
881 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n",
882 (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg));
885 s = find_byclass(prog, prog->regstclass, s, endpos, NULL);
888 const char *what = NULL;
890 if (endpos == strend) {
891 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
892 "Could not match STCLASS...\n") );
895 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
896 "This position contradicts STCLASS...\n") );
897 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
899 /* Contradict one of substrings */
900 if (prog->anchored_substr || prog->anchored_utf8) {
901 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
902 DEBUG_EXECUTE_r( what = "anchored" );
904 s = HOP3c(t, 1, strend);
905 if (s + start_shift + end_shift > strend) {
906 /* XXXX Should be taken into account earlier? */
907 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
908 "Could not match STCLASS...\n") );
913 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
914 "Looking for %s substr starting at offset %ld...\n",
915 what, (long)(s + start_shift - i_strpos)) );
918 /* Have both, check_string is floating */
919 if (t + start_shift >= check_at) /* Contradicts floating=check */
920 goto retry_floating_check;
921 /* Recheck anchored substring, but not floating... */
925 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
926 "Looking for anchored substr starting at offset %ld...\n",
927 (long)(other_last - i_strpos)) );
928 goto do_other_anchored;
930 /* Another way we could have checked stclass at the
931 current position only: */
936 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
937 "Looking for /%s^%s/m starting at offset %ld...\n",
938 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
941 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
943 /* Check is floating subtring. */
944 retry_floating_check:
945 t = check_at - start_shift;
946 DEBUG_EXECUTE_r( what = "floating" );
947 goto hop_and_restart;
950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
951 "By STCLASS: moving %ld --> %ld\n",
952 (long)(t - i_strpos), (long)(s - i_strpos))
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
957 "Does not contradict STCLASS...\n");
962 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
963 PL_colors[4], (check ? "Guessed" : "Giving up"),
964 PL_colors[5], (long)(s - i_strpos)) );
967 fail_finish: /* Substring not found */
968 if (prog->check_substr || prog->check_utf8) /* could be removed already */
969 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
972 PL_colors[4], PL_colors[5]));
978 #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \
979 foldlen, foldbuf, uniflags) STMT_START { \
980 switch (trie_type) { \
981 case trie_utf8_fold: \
983 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
988 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
989 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
990 foldlen -= UNISKIP( uvc ); \
991 uscan = foldbuf + UNISKIP( uvc ); \
995 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1003 charid = trie->charmap[ uvc ]; \
1007 if (trie->widecharmap) { \
1008 SV** const svpp = hv_fetch(trie->widecharmap, \
1009 (char*)&uvc, sizeof(UV), 0); \
1011 charid = (U16)SvIV(*svpp); \
1016 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1019 ibcmp_utf8(s, NULL, 0, do_utf8, \
1020 m, NULL, ln, (bool)UTF)) \
1021 && (!reginfo || regtry(reginfo, &s)) ) \
1024 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1025 uvchr_to_utf8(tmpbuf, c); \
1026 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1028 && (f == c1 || f == c2) \
1029 && (ln == foldlen || \
1030 !ibcmp_utf8((char *) foldbuf, \
1031 NULL, foldlen, do_utf8, \
1033 NULL, ln, (bool)UTF)) \
1034 && (!reginfo || regtry(reginfo, &s)) ) \
1039 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1043 && (ln == 1 || !(OP(c) == EXACTF \
1045 : ibcmp_locale(s, m, ln))) \
1046 && (!reginfo || regtry(reginfo, &s)) ) \
1052 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1054 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1060 #define REXEC_FBC_SCAN(CoDe) \
1062 while (s < strend) { \
1068 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1069 REXEC_FBC_UTF8_SCAN( \
1071 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1080 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1083 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1092 #define REXEC_FBC_TRYIT \
1093 if ((!reginfo || regtry(reginfo, &s))) \
1096 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1099 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1102 REXEC_FBC_CLASS_SCAN(CoNd); \
1106 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1107 PL_reg_flags |= RF_tainted; \
1109 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1112 REXEC_FBC_CLASS_SCAN(CoNd); \
1116 #define DUMP_EXEC_POS(li,s,doutf8) \
1117 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1119 /* We know what class REx starts with. Try to find this position... */
1120 /* if reginfo is NULL, its a dryrun */
1121 /* annoyingly all the vars in this routine have different names from their counterparts
1122 in regmatch. /grrr */
1125 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1126 const char *strend, 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;
1730 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1731 re_scream_pos_data d;
1733 d.scream_olds = &scream_olds;
1734 d.scream_pos = &scream_pos;
1735 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1737 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1738 goto phooey; /* not present */
1744 /* Simplest case: anchored match need be tried only once. */
1745 /* [unless only anchor is BOL and multiline is set] */
1746 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1747 if (s == startpos && regtry(®info, &startpos))
1749 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1750 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1755 dontbother = minlen - 1;
1756 end = HOP3c(strend, -dontbother, strbeg) - 1;
1757 /* for multiline we only have to try after newlines */
1758 if (prog->check_substr || prog->check_utf8) {
1762 if (regtry(®info, &s))
1767 if (prog->reganch & RE_USE_INTUIT) {
1768 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1779 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1780 if (regtry(®info, &s))
1787 } else if (ROPT_GPOS_CHECK == (prog->reganch & ROPT_GPOS_CHECK))
1789 /* the warning about reginfo.ganch being used without intialization
1790 is bogus -- we set it above, when prog->reganch & ROPT_GPOS_SEEN
1791 and we only enter this block when the same bit is set. */
1792 if (regtry(®info, ®info.ganch))
1797 /* Messy cases: unanchored match. */
1798 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1799 /* we have /x+whatever/ */
1800 /* it must be a one character string (XXXX Except UTF?) */
1805 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1806 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1807 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1812 DEBUG_EXECUTE_r( did_match = 1 );
1813 if (regtry(®info, &s)) goto got_it;
1815 while (s < strend && *s == ch)
1823 DEBUG_EXECUTE_r( did_match = 1 );
1824 if (regtry(®info, &s)) goto got_it;
1826 while (s < strend && *s == ch)
1831 DEBUG_EXECUTE_r(if (!did_match)
1832 PerlIO_printf(Perl_debug_log,
1833 "Did not find anchored character...\n")
1836 else if (prog->anchored_substr != NULL
1837 || prog->anchored_utf8 != NULL
1838 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1839 && prog->float_max_offset < strend - s)) {
1844 char *last1; /* Last position checked before */
1848 if (prog->anchored_substr || prog->anchored_utf8) {
1849 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1850 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1851 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1852 back_max = back_min = prog->anchored_offset;
1854 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1855 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1856 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1857 back_max = prog->float_max_offset;
1858 back_min = prog->float_min_offset;
1862 if (must == &PL_sv_undef)
1863 /* could not downgrade utf8 check substring, so must fail */
1869 last = HOP3c(strend, /* Cannot start after this */
1870 -(I32)(CHR_SVLEN(must)
1871 - (SvTAIL(must) != 0) + back_min), strbeg);
1874 last1 = HOPc(s, -1);
1876 last1 = s - 1; /* bogus */
1878 /* XXXX check_substr already used to find "s", can optimize if
1879 check_substr==must. */
1881 dontbother = end_shift;
1882 strend = HOPc(strend, -dontbother);
1883 while ( (s <= last) &&
1884 ((flags & REXEC_SCREAM)
1885 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1886 end_shift, &scream_pos, 0))
1887 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1888 (unsigned char*)strend, must,
1889 multiline ? FBMrf_MULTILINE : 0))) ) {
1890 /* we may be pointing at the wrong string */
1891 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1892 s = strbeg + (s - SvPVX_const(sv));
1893 DEBUG_EXECUTE_r( did_match = 1 );
1894 if (HOPc(s, -back_max) > last1) {
1895 last1 = HOPc(s, -back_min);
1896 s = HOPc(s, -back_max);
1899 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1901 last1 = HOPc(s, -back_min);
1905 while (s <= last1) {
1906 if (regtry(®info, &s))
1912 while (s <= last1) {
1913 if (regtry(®info, &s))
1919 DEBUG_EXECUTE_r(if (!did_match) {
1920 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1921 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1922 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1923 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1924 ? "anchored" : "floating"),
1925 quoted, RE_SV_TAIL(must));
1929 else if ( (c = prog->regstclass) ) {
1931 const OPCODE op = OP(prog->regstclass);
1932 /* don't bother with what can't match */
1933 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1934 strend = HOPc(strend, -(minlen - 1));
1937 SV * const prop = sv_newmortal();
1938 regprop(prog, prop, c);
1940 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1942 PerlIO_printf(Perl_debug_log,
1943 "Matching stclass %.*s against %s (%d chars)\n",
1944 (int)SvCUR(prop), SvPVX_const(prop),
1945 quoted, (int)(strend - s));
1948 if (find_byclass(prog, c, s, strend, ®info))
1950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
1954 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1959 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1960 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1961 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1963 if (flags & REXEC_SCREAM) {
1964 last = screaminstr(sv, float_real, s - strbeg,
1965 end_shift, &scream_pos, 1); /* last one */
1967 last = scream_olds; /* Only one occurrence. */
1968 /* we may be pointing at the wrong string */
1969 else if (RX_MATCH_COPIED(prog))
1970 s = strbeg + (s - SvPVX_const(sv));
1974 const char * const little = SvPV_const(float_real, len);
1976 if (SvTAIL(float_real)) {
1977 if (memEQ(strend - len + 1, little, len - 1))
1978 last = strend - len + 1;
1979 else if (!multiline)
1980 last = memEQ(strend - len, little, len)
1981 ? strend - len : NULL;
1987 last = rninstr(s, strend, little, little + len);
1989 last = strend; /* matching "$" */
1994 PerlIO_printf(Perl_debug_log,
1995 "%sCan't trim the tail, match fails (should not happen)%s\n",
1996 PL_colors[4], PL_colors[5]));
1997 goto phooey; /* Should not happen! */
1999 dontbother = strend - last + prog->float_min_offset;
2001 if (minlen && (dontbother < minlen))
2002 dontbother = minlen - 1;
2003 strend -= dontbother; /* this one's always in bytes! */
2004 /* We don't know much -- general case. */
2007 if (regtry(®info, &s))
2016 if (regtry(®info, &s))
2018 } while (s++ < strend);
2026 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2028 if (PL_reg_eval_set) {
2029 /* Preserve the current value of $^R */
2030 if (oreplsv != GvSV(PL_replgv))
2031 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2032 restored, the value remains
2034 restore_pos(aTHX_ prog);
2036 if (prog->paren_names)
2037 (void)hv_iterinit(prog->paren_names);
2039 /* make sure $`, $&, $', and $digit will work later */
2040 if ( !(flags & REXEC_NOT_FIRST) ) {
2041 RX_MATCH_COPY_FREE(prog);
2042 if (flags & REXEC_COPY_STR) {
2043 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2044 #ifdef PERL_OLD_COPY_ON_WRITE
2046 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2048 PerlIO_printf(Perl_debug_log,
2049 "Copy on write: regexp capture, type %d\n",
2052 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2053 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2054 assert (SvPOKp(prog->saved_copy));
2058 RX_MATCH_COPIED_on(prog);
2059 s = savepvn(strbeg, i);
2065 prog->subbeg = strbeg;
2066 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2073 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2074 PL_colors[4], PL_colors[5]));
2075 if (PL_reg_eval_set)
2076 restore_pos(aTHX_ prog);
2082 - regtry - try match at specific point
2084 STATIC I32 /* 0 failure, 1 success */
2085 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2091 regexp *prog = reginfo->prog;
2092 GET_RE_DEBUG_FLAGS_DECL;
2093 reginfo->cutpoint=NULL;
2095 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2098 PL_reg_eval_set = RS_init;
2099 DEBUG_EXECUTE_r(DEBUG_s(
2100 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2101 (IV)(PL_stack_sp - PL_stack_base));
2104 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2105 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2107 /* Apparently this is not needed, judging by wantarray. */
2108 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2109 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2112 /* Make $_ available to executed code. */
2113 if (reginfo->sv != DEFSV) {
2115 DEFSV = reginfo->sv;
2118 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2119 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2120 /* prepare for quick setting of pos */
2121 #ifdef PERL_OLD_COPY_ON_WRITE
2123 sv_force_normal_flags(sv, 0);
2125 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2126 &PL_vtbl_mglob, NULL, 0);
2130 PL_reg_oldpos = mg->mg_len;
2131 SAVEDESTRUCTOR_X(restore_pos, prog);
2133 if (!PL_reg_curpm) {
2134 Newxz(PL_reg_curpm, 1, PMOP);
2137 SV* const repointer = newSViv(0);
2138 /* so we know which PL_regex_padav element is PL_reg_curpm */
2139 SvFLAGS(repointer) |= SVf_BREAK;
2140 av_push(PL_regex_padav,repointer);
2141 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2142 PL_regex_pad = AvARRAY(PL_regex_padav);
2146 PM_SETRE(PL_reg_curpm, prog);
2147 PL_reg_oldcurpm = PL_curpm;
2148 PL_curpm = PL_reg_curpm;
2149 if (RX_MATCH_COPIED(prog)) {
2150 /* Here is a serious problem: we cannot rewrite subbeg,
2151 since it may be needed if this match fails. Thus
2152 $` inside (?{}) could fail... */
2153 PL_reg_oldsaved = prog->subbeg;
2154 PL_reg_oldsavedlen = prog->sublen;
2155 #ifdef PERL_OLD_COPY_ON_WRITE
2156 PL_nrs = prog->saved_copy;
2158 RX_MATCH_COPIED_off(prog);
2161 PL_reg_oldsaved = NULL;
2162 prog->subbeg = PL_bostr;
2163 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2165 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2166 prog->startp[0] = *startpos - PL_bostr;
2167 PL_reginput = *startpos;
2168 PL_reglastparen = &prog->lastparen;
2169 PL_reglastcloseparen = &prog->lastcloseparen;
2170 prog->lastparen = 0;
2171 prog->lastcloseparen = 0;
2173 PL_regstartp = prog->startp;
2174 PL_regendp = prog->endp;
2175 if (PL_reg_start_tmpl <= prog->nparens) {
2176 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2177 if(PL_reg_start_tmp)
2178 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2180 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2183 /* XXXX What this code is doing here?!!! There should be no need
2184 to do this again and again, PL_reglastparen should take care of
2187 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2188 * Actually, the code in regcppop() (which Ilya may be meaning by
2189 * PL_reglastparen), is not needed at all by the test suite
2190 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2191 * enough, for building DynaLoader, or otherwise this
2192 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2193 * will happen. Meanwhile, this code *is* needed for the
2194 * above-mentioned test suite tests to succeed. The common theme
2195 * on those tests seems to be returning null fields from matches.
2200 if (prog->nparens) {
2202 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2209 if (regmatch(reginfo, prog->program + 1)) {
2210 prog->endp[0] = PL_reginput - PL_bostr;
2213 if (reginfo->cutpoint)
2214 *startpos= reginfo->cutpoint;
2215 REGCP_UNWIND(lastcp);
2220 #define sayYES goto yes
2221 #define sayNO goto no
2222 #define sayNO_SILENT goto no_silent
2224 /* we dont use STMT_START/END here because it leads to
2225 "unreachable code" warnings, which are bogus, but distracting. */
2226 #define CACHEsayNO \
2227 if (ST.cache_mask) \
2228 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2231 /* this is used to determine how far from the left messages like
2232 'failed...' are printed. It should be set such that messages
2233 are inline with the regop output that created them.
2235 #define REPORT_CODE_OFF 32
2238 /* Make sure there is a test for this +1 options in re_tests */
2239 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2241 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2242 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2244 #define SLAB_FIRST(s) (&(s)->states[0])
2245 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2247 /* grab a new slab and return the first slot in it */
2249 STATIC regmatch_state *
2252 #if PERL_VERSION < 9
2255 regmatch_slab *s = PL_regmatch_slab->next;
2257 Newx(s, 1, regmatch_slab);
2258 s->prev = PL_regmatch_slab;
2260 PL_regmatch_slab->next = s;
2262 PL_regmatch_slab = s;
2263 return SLAB_FIRST(s);
2267 /* push a new state then goto it */
2269 #define PUSH_STATE_GOTO(state, node) \
2271 st->resume_state = state; \
2274 /* push a new state with success backtracking, then goto it */
2276 #define PUSH_YES_STATE_GOTO(state, node) \
2278 st->resume_state = state; \
2279 goto push_yes_state;
2285 regmatch() - main matching routine
2287 This is basically one big switch statement in a loop. We execute an op,
2288 set 'next' to point the next op, and continue. If we come to a point which
2289 we may need to backtrack to on failure such as (A|B|C), we push a
2290 backtrack state onto the backtrack stack. On failure, we pop the top
2291 state, and re-enter the loop at the state indicated. If there are no more
2292 states to pop, we return failure.
2294 Sometimes we also need to backtrack on success; for example /A+/, where
2295 after successfully matching one A, we need to go back and try to
2296 match another one; similarly for lookahead assertions: if the assertion
2297 completes successfully, we backtrack to the state just before the assertion
2298 and then carry on. In these cases, the pushed state is marked as
2299 'backtrack on success too'. This marking is in fact done by a chain of
2300 pointers, each pointing to the previous 'yes' state. On success, we pop to
2301 the nearest yes state, discarding any intermediate failure-only states.
2302 Sometimes a yes state is pushed just to force some cleanup code to be
2303 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2304 it to free the inner regex.
2306 Note that failure backtracking rewinds the cursor position, while
2307 success backtracking leaves it alone.
2309 A pattern is complete when the END op is executed, while a subpattern
2310 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2311 ops trigger the "pop to last yes state if any, otherwise return true"
2314 A common convention in this function is to use A and B to refer to the two
2315 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2316 the subpattern to be matched possibly multiple times, while B is the entire
2317 rest of the pattern. Variable and state names reflect this convention.
2319 The states in the main switch are the union of ops and failure/success of
2320 substates associated with with that op. For example, IFMATCH is the op
2321 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2322 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2323 successfully matched A and IFMATCH_A_fail is a state saying that we have
2324 just failed to match A. Resume states always come in pairs. The backtrack
2325 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2326 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2327 on success or failure.
2329 The struct that holds a backtracking state is actually a big union, with
2330 one variant for each major type of op. The variable st points to the
2331 top-most backtrack struct. To make the code clearer, within each
2332 block of code we #define ST to alias the relevant union.
2334 Here's a concrete example of a (vastly oversimplified) IFMATCH
2340 #define ST st->u.ifmatch
2342 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2343 ST.foo = ...; // some state we wish to save
2345 // push a yes backtrack state with a resume value of
2346 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2348 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2351 case IFMATCH_A: // we have successfully executed A; now continue with B
2353 bar = ST.foo; // do something with the preserved value
2356 case IFMATCH_A_fail: // A failed, so the assertion failed
2357 ...; // do some housekeeping, then ...
2358 sayNO; // propagate the failure
2365 For any old-timers reading this who are familiar with the old recursive
2366 approach, the code above is equivalent to:
2368 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2377 ...; // do some housekeeping, then ...
2378 sayNO; // propagate the failure
2381 The topmost backtrack state, pointed to by st, is usually free. If you
2382 want to claim it, populate any ST.foo fields in it with values you wish to
2383 save, then do one of
2385 PUSH_STATE_GOTO(resume_state, node);
2386 PUSH_YES_STATE_GOTO(resume_state, node);
2388 which sets that backtrack state's resume value to 'resume_state', pushes a
2389 new free entry to the top of the backtrack stack, then goes to 'node'.
2390 On backtracking, the free slot is popped, and the saved state becomes the
2391 new free state. An ST.foo field in this new top state can be temporarily
2392 accessed to retrieve values, but once the main loop is re-entered, it
2393 becomes available for reuse.
2395 Note that the depth of the backtrack stack constantly increases during the
2396 left-to-right execution of the pattern, rather than going up and down with
2397 the pattern nesting. For example the stack is at its maximum at Z at the
2398 end of the pattern, rather than at X in the following:
2400 /(((X)+)+)+....(Y)+....Z/
2402 The only exceptions to this are lookahead/behind assertions and the cut,
2403 (?>A), which pop all the backtrack states associated with A before
2406 Bascktrack state structs are allocated in slabs of about 4K in size.
2407 PL_regmatch_state and st always point to the currently active state,
2408 and PL_regmatch_slab points to the slab currently containing
2409 PL_regmatch_state. The first time regmatch() is called, the first slab is
2410 allocated, and is never freed until interpreter destruction. When the slab
2411 is full, a new one is allocated and chained to the end. At exit from
2412 regmatch(), slabs allocated since entry are freed.
2417 #define DEBUG_STATE_pp(pp) \
2419 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2420 PerlIO_printf(Perl_debug_log, \
2423 reg_name[st->resume_state] ); \
2427 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2432 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2433 const char *start, const char *end, const char *blurb)
2435 const bool utf8_pat= prog->reganch & ROPT_UTF8 ? 1 : 0;
2439 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2440 prog->precomp, prog->prelen, 60);
2442 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2443 start, end - start, 60);
2445 PerlIO_printf(Perl_debug_log,
2446 "%s%s REx%s %s against %s\n",
2447 PL_colors[4], blurb, PL_colors[5], s0, s1);
2449 if (do_utf8||utf8_pat)
2450 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2451 utf8_pat ? "pattern" : "",
2452 utf8_pat && do_utf8 ? " and " : "",
2453 do_utf8 ? "string" : ""
2459 S_dump_exec_pos(pTHX_ const char *locinput,
2460 const regnode *scan,
2461 const char *loc_regeol,
2462 const char *loc_bostr,
2463 const char *loc_reg_starttry,
2466 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2467 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2468 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2469 /* The part of the string before starttry has one color
2470 (pref0_len chars), between starttry and current
2471 position another one (pref_len - pref0_len chars),
2472 after the current position the third one.
2473 We assume that pref0_len <= pref_len, otherwise we
2474 decrease pref0_len. */
2475 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2476 ? (5 + taill) - l : locinput - loc_bostr;
2479 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2481 pref0_len = pref_len - (locinput - loc_reg_starttry);
2482 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2483 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2484 ? (5 + taill) - pref_len : loc_regeol - locinput);
2485 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2489 if (pref0_len > pref_len)
2490 pref0_len = pref_len;
2492 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2494 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2495 (locinput - pref_len),pref0_len, 60, 4, 5);
2497 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2498 (locinput - pref_len + pref0_len),
2499 pref_len - pref0_len, 60, 2, 3);
2501 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2502 locinput, loc_regeol - locinput, 10, 0, 1);
2504 const STRLEN tlen=len0+len1+len2;
2505 PerlIO_printf(Perl_debug_log,
2506 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2507 (IV)(locinput - loc_bostr),
2510 (docolor ? "" : "> <"),
2512 (int)(tlen > 19 ? 0 : 19 - tlen),
2519 /* reg_check_named_buff_matched()
2520 * Checks to see if a named buffer has matched. The data array of
2521 * buffer numbers corresponding to the buffer is expected to reside
2522 * in the regexp->data->data array in the slot stored in the ARG() of
2523 * node involved. Note that this routine doesn't actually care about the
2524 * name, that information is not preserved from compilation to execution.
2525 * Returns the index of the leftmost defined buffer with the given name
2526 * or 0 if non of the buffers matched.
2529 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2531 SV *sv_dat=(SV*)rex->data->data[ ARG( scan ) ];
2532 I32 *nums=(I32*)SvPVX(sv_dat);
2533 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2534 if ((I32)*PL_reglastparen >= nums[n] &&
2535 PL_regendp[nums[n]] != -1)
2543 STATIC I32 /* 0 failure, 1 success */
2544 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2546 #if PERL_VERSION < 9
2550 register const bool do_utf8 = PL_reg_match_utf8;
2551 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2553 regexp *rex = reginfo->prog;
2555 regmatch_slab *orig_slab;
2556 regmatch_state *orig_state;
2558 /* the current state. This is a cached copy of PL_regmatch_state */
2559 register regmatch_state *st;
2561 /* cache heavy used fields of st in registers */
2562 register regnode *scan;
2563 register regnode *next;
2564 register I32 n = 0; /* general value; init to avoid compiler warning */
2565 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2566 register char *locinput = PL_reginput;
2567 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2569 bool result = 0; /* return value of S_regmatch */
2570 int depth = 0; /* depth of backtrack stack */
2571 int nochange_depth = 0; /* depth of GOSUB recursion with nochange*/
2572 regmatch_state *yes_state = NULL; /* state to pop to on success of
2574 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2575 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2579 /* these three flags are set by various ops to signal information to
2580 * the very next op. They have a useful lifetime of exactly one loop
2581 * iteration, and are not preserved or restored by state pushes/pops
2583 bool sw = 0; /* the condition value in (?(cond)a|b) */
2584 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2585 int logical = 0; /* the following EVAL is:
2589 or the following IFMATCH/UNLESSM is:
2590 false: plain (?=foo)
2591 true: used as a condition: (?(?=foo))
2595 GET_RE_DEBUG_FLAGS_DECL;
2599 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2601 /* on first ever call to regmatch, allocate first slab */
2602 if (!PL_regmatch_slab) {
2603 Newx(PL_regmatch_slab, 1, regmatch_slab);
2604 PL_regmatch_slab->prev = NULL;
2605 PL_regmatch_slab->next = NULL;
2606 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2609 /* remember current high-water mark for exit */
2610 /* XXX this should be done with SAVE* instead */
2611 orig_slab = PL_regmatch_slab;
2612 orig_state = PL_regmatch_state;
2614 /* grab next free state slot */
2615 st = ++PL_regmatch_state;
2616 if (st > SLAB_LAST(PL_regmatch_slab))
2617 st = PL_regmatch_state = S_push_slab(aTHX);
2619 /* Note that nextchr is a byte even in UTF */
2620 nextchr = UCHARAT(locinput);
2622 while (scan != NULL) {
2625 SV * const prop = sv_newmortal();
2626 regnode *rnext=regnext(scan);
2627 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2628 regprop(rex, prop, scan);
2630 PerlIO_printf(Perl_debug_log,
2631 "%3"IVdf":%*s%s(%"IVdf")\n",
2632 (IV)(scan - rex->program), depth*2, "",
2634 (PL_regkind[OP(scan)] == END || !rnext) ?
2635 0 : (IV)(rnext - rex->program));
2638 next = scan + NEXT_OFF(scan);
2641 state_num = OP(scan);
2644 switch (state_num) {
2646 if (locinput == PL_bostr)
2648 /* reginfo->till = reginfo->bol; */
2653 if (locinput == PL_bostr ||
2654 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2660 if (locinput == PL_bostr)
2664 if (locinput == reginfo->ganch)
2670 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2675 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2677 if (PL_regeol - locinput > 1)
2681 if (PL_regeol != locinput)
2685 if (!nextchr && locinput >= PL_regeol)
2688 locinput += PL_utf8skip[nextchr];
2689 if (locinput > PL_regeol)
2691 nextchr = UCHARAT(locinput);
2694 nextchr = UCHARAT(++locinput);
2697 if (!nextchr && locinput >= PL_regeol)
2699 nextchr = UCHARAT(++locinput);
2702 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2705 locinput += PL_utf8skip[nextchr];
2706 if (locinput > PL_regeol)
2708 nextchr = UCHARAT(locinput);
2711 nextchr = UCHARAT(++locinput);
2715 #define ST st->u.trie
2717 /* In this case the charclass data is available inline so
2718 we can fail fast without a lot of extra overhead.
2720 if (scan->flags == EXACT || !do_utf8) {
2721 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2723 PerlIO_printf(Perl_debug_log,
2724 "%*s %sfailed to match trie start class...%s\n",
2725 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2734 /* what type of TRIE am I? (utf8 makes this contextual) */
2735 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2736 trie_type = do_utf8 ?
2737 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2740 /* what trie are we using right now */
2741 reg_trie_data * const trie
2742 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2743 U32 state = trie->startstate;
2745 if (trie->bitmap && trie_type != trie_utf8_fold &&
2746 !TRIE_BITMAP_TEST(trie,*locinput)
2748 if (trie->states[ state ].wordnum) {
2750 PerlIO_printf(Perl_debug_log,
2751 "%*s %smatched empty string...%s\n",
2752 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2757 PerlIO_printf(Perl_debug_log,
2758 "%*s %sfailed to match trie start class...%s\n",
2759 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2766 U8 *uc = ( U8* )locinput;
2770 U8 *uscan = (U8*)NULL;
2772 SV *sv_accept_buff = NULL;
2773 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2775 ST.accepted = 0; /* how many accepting states we have seen */
2777 ST.jump = trie->jump;
2781 traverse the TRIE keeping track of all accepting states
2782 we transition through until we get to a failing node.
2785 while ( state && uc <= (U8*)PL_regeol ) {
2786 U32 base = trie->states[ state ].trans.base;
2789 /* We use charid to hold the wordnum as we don't use it
2790 for charid until after we have done the wordnum logic.
2791 We define an alias just so that the wordnum logic reads
2794 #define got_wordnum charid
2795 got_wordnum = trie->states[ state ].wordnum;
2797 if ( got_wordnum ) {
2798 if ( ! ST.accepted ) {
2801 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2802 sv_accept_buff=newSV(bufflen *
2803 sizeof(reg_trie_accepted) - 1);
2804 SvCUR_set(sv_accept_buff, 0);
2805 SvPOK_on(sv_accept_buff);
2806 sv_2mortal(sv_accept_buff);
2809 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2812 if (ST.accepted >= bufflen) {
2814 ST.accept_buff =(reg_trie_accepted*)
2815 SvGROW(sv_accept_buff,
2816 bufflen * sizeof(reg_trie_accepted));
2818 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2819 + sizeof(reg_trie_accepted));
2822 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2823 ST.accept_buff[ST.accepted].endpos = uc;
2825 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2829 DEBUG_TRIE_EXECUTE_r({
2830 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2831 PerlIO_printf( Perl_debug_log,
2832 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2833 2+depth * 2, "", PL_colors[4],
2834 (UV)state, (UV)ST.accepted );
2838 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
2839 uvc, charid, foldlen, foldbuf, uniflags);
2842 (base + charid > trie->uniquecharcount )
2843 && (base + charid - 1 - trie->uniquecharcount
2845 && trie->trans[base + charid - 1 -
2846 trie->uniquecharcount].check == state)
2848 state = trie->trans[base + charid - 1 -
2849 trie->uniquecharcount ].next;
2860 DEBUG_TRIE_EXECUTE_r(
2861 PerlIO_printf( Perl_debug_log,
2862 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2863 charid, uvc, (UV)state, PL_colors[5] );
2870 PerlIO_printf( Perl_debug_log,
2871 "%*s %sgot %"IVdf" possible matches%s\n",
2872 REPORT_CODE_OFF + depth * 2, "",
2873 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2879 case TRIE_next_fail: /* we failed - try next alterative */
2881 if ( ST.accepted == 1 ) {
2882 /* only one choice left - just continue */
2884 reg_trie_data * const trie
2885 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2886 SV ** const tmp = RX_DEBUG(reginfo->prog)
2887 ? av_fetch( trie->words, ST.accept_buff[ 0 ].wordnum-1, 0 )
2889 PerlIO_printf( Perl_debug_log,
2890 "%*s %sonly one match left: #%d <%s>%s\n",
2891 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2892 ST.accept_buff[ 0 ].wordnum,
2893 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2896 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2897 /* in this case we free tmps/leave before we call regmatch
2898 as we wont be using accept_buff again. */
2901 locinput = PL_reginput;
2902 nextchr = UCHARAT(locinput);
2904 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
2907 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
2909 continue; /* execute rest of RE */
2912 if (!ST.accepted-- ) {
2919 There are at least two accepting states left. Presumably
2920 the number of accepting states is going to be low,
2921 typically two. So we simply scan through to find the one
2922 with lowest wordnum. Once we find it, we swap the last
2923 state into its place and decrement the size. We then try to
2924 match the rest of the pattern at the point where the word
2925 ends. If we succeed, control just continues along the
2926 regex; if we fail we return here to try the next accepting
2933 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
2934 DEBUG_TRIE_EXECUTE_r(
2935 PerlIO_printf( Perl_debug_log,
2936 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2937 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
2938 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
2939 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
2942 if (ST.accept_buff[cur].wordnum <
2943 ST.accept_buff[best].wordnum)
2948 reg_trie_data * const trie
2949 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2950 SV ** const tmp = RX_DEBUG(reginfo->prog)
2951 ? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 )
2953 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
2955 ST.me + ST.jump[ST.accept_buff[best].wordnum];
2956 PerlIO_printf( Perl_debug_log,
2957 "%*s %strying alternation #%d <%s> at node #%d %s\n",
2958 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2959 ST.accept_buff[best].wordnum,
2960 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2961 REG_NODE_NUM(nextop),
2965 if ( best<ST.accepted ) {
2966 reg_trie_accepted tmp = ST.accept_buff[ best ];
2967 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
2968 ST.accept_buff[ ST.accepted ] = tmp;
2971 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
2972 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
2973 PUSH_STATE_GOTO(TRIE_next, ST.B);
2976 PUSH_STATE_GOTO(TRIE_next, ST.me + ST.jump[ST.accept_buff[best].wordnum]);
2986 char *s = STRING(scan);
2988 if (do_utf8 != UTF) {
2989 /* The target and the pattern have differing utf8ness. */
2991 const char * const e = s + ln;
2994 /* The target is utf8, the pattern is not utf8. */
2999 if (NATIVE_TO_UNI(*(U8*)s) !=
3000 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3008 /* The target is not utf8, the pattern is utf8. */
3013 if (NATIVE_TO_UNI(*((U8*)l)) !=
3014 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3022 nextchr = UCHARAT(locinput);
3025 /* The target and the pattern have the same utf8ness. */
3026 /* Inline the first character, for speed. */
3027 if (UCHARAT(s) != nextchr)
3029 if (PL_regeol - locinput < ln)
3031 if (ln > 1 && memNE(s, locinput, ln))
3034 nextchr = UCHARAT(locinput);
3038 PL_reg_flags |= RF_tainted;
3041 char * const s = STRING(scan);
3044 if (do_utf8 || UTF) {
3045 /* Either target or the pattern are utf8. */
3046 const char * const l = locinput;
3047 char *e = PL_regeol;
3049 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3050 l, &e, 0, do_utf8)) {
3051 /* One more case for the sharp s:
3052 * pack("U0U*", 0xDF) =~ /ss/i,
3053 * the 0xC3 0x9F are the UTF-8
3054 * byte sequence for the U+00DF. */
3056 toLOWER(s[0]) == 's' &&
3058 toLOWER(s[1]) == 's' &&
3065 nextchr = UCHARAT(locinput);
3069 /* Neither the target and the pattern are utf8. */
3071 /* Inline the first character, for speed. */
3072 if (UCHARAT(s) != nextchr &&
3073 UCHARAT(s) != ((OP(scan) == EXACTF)
3074 ? PL_fold : PL_fold_locale)[nextchr])
3076 if (PL_regeol - locinput < ln)
3078 if (ln > 1 && (OP(scan) == EXACTF
3079 ? ibcmp(s, locinput, ln)
3080 : ibcmp_locale(s, locinput, ln)))
3083 nextchr = UCHARAT(locinput);
3088 STRLEN inclasslen = PL_regeol - locinput;
3090 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3092 if (locinput >= PL_regeol)
3094 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3095 nextchr = UCHARAT(locinput);
3100 nextchr = UCHARAT(locinput);
3101 if (!REGINCLASS(rex, scan, (U8*)locinput))
3103 if (!nextchr && locinput >= PL_regeol)
3105 nextchr = UCHARAT(++locinput);
3109 /* If we might have the case of the German sharp s
3110 * in a casefolding Unicode character class. */
3112 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3113 locinput += SHARP_S_SKIP;
3114 nextchr = UCHARAT(locinput);
3120 PL_reg_flags |= RF_tainted;
3126 LOAD_UTF8_CHARCLASS_ALNUM();
3127 if (!(OP(scan) == ALNUM
3128 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3129 : isALNUM_LC_utf8((U8*)locinput)))
3133 locinput += PL_utf8skip[nextchr];
3134 nextchr = UCHARAT(locinput);
3137 if (!(OP(scan) == ALNUM
3138 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3140 nextchr = UCHARAT(++locinput);
3143 PL_reg_flags |= RF_tainted;
3146 if (!nextchr && locinput >= PL_regeol)
3149 LOAD_UTF8_CHARCLASS_ALNUM();
3150 if (OP(scan) == NALNUM
3151 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3152 : isALNUM_LC_utf8((U8*)locinput))
3156 locinput += PL_utf8skip[nextchr];
3157 nextchr = UCHARAT(locinput);
3160 if (OP(scan) == NALNUM
3161 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3163 nextchr = UCHARAT(++locinput);
3167 PL_reg_flags |= RF_tainted;
3171 /* was last char in word? */
3173 if (locinput == PL_bostr)
3176 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3178 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3180 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3181 ln = isALNUM_uni(ln);
3182 LOAD_UTF8_CHARCLASS_ALNUM();
3183 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3186 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3187 n = isALNUM_LC_utf8((U8*)locinput);
3191 ln = (locinput != PL_bostr) ?
3192 UCHARAT(locinput - 1) : '\n';
3193 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3195 n = isALNUM(nextchr);
3198 ln = isALNUM_LC(ln);
3199 n = isALNUM_LC(nextchr);
3202 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3203 OP(scan) == BOUNDL))
3207 PL_reg_flags |= RF_tainted;
3213 if (UTF8_IS_CONTINUED(nextchr)) {
3214 LOAD_UTF8_CHARCLASS_SPACE();
3215 if (!(OP(scan) == SPACE
3216 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3217 : isSPACE_LC_utf8((U8*)locinput)))
3221 locinput += PL_utf8skip[nextchr];
3222 nextchr = UCHARAT(locinput);
3225 if (!(OP(scan) == SPACE
3226 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3228 nextchr = UCHARAT(++locinput);
3231 if (!(OP(scan) == SPACE
3232 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3234 nextchr = UCHARAT(++locinput);
3238 PL_reg_flags |= RF_tainted;
3241 if (!nextchr && locinput >= PL_regeol)
3244 LOAD_UTF8_CHARCLASS_SPACE();
3245 if (OP(scan) == NSPACE
3246 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3247 : isSPACE_LC_utf8((U8*)locinput))
3251 locinput += PL_utf8skip[nextchr];
3252 nextchr = UCHARAT(locinput);
3255 if (OP(scan) == NSPACE
3256 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3258 nextchr = UCHARAT(++locinput);
3261 PL_reg_flags |= RF_tainted;
3267 LOAD_UTF8_CHARCLASS_DIGIT();
3268 if (!(OP(scan) == DIGIT
3269 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3270 : isDIGIT_LC_utf8((U8*)locinput)))
3274 locinput += PL_utf8skip[nextchr];
3275 nextchr = UCHARAT(locinput);
3278 if (!(OP(scan) == DIGIT
3279 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3281 nextchr = UCHARAT(++locinput);
3284 PL_reg_flags |= RF_tainted;
3287 if (!nextchr && locinput >= PL_regeol)
3290 LOAD_UTF8_CHARCLASS_DIGIT();
3291 if (OP(scan) == NDIGIT
3292 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3293 : isDIGIT_LC_utf8((U8*)locinput))
3297 locinput += PL_utf8skip[nextchr];
3298 nextchr = UCHARAT(locinput);
3301 if (OP(scan) == NDIGIT
3302 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3304 nextchr = UCHARAT(++locinput);
3307 if (locinput >= PL_regeol)
3310 LOAD_UTF8_CHARCLASS_MARK();
3311 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3313 locinput += PL_utf8skip[nextchr];
3314 while (locinput < PL_regeol &&
3315 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3316 locinput += UTF8SKIP(locinput);
3317 if (locinput > PL_regeol)
3322 nextchr = UCHARAT(locinput);
3329 PL_reg_flags |= RF_tainted;
3334 n = reg_check_named_buff_matched(rex,scan);
3337 type = REF + ( type - NREF );
3344 PL_reg_flags |= RF_tainted;
3348 n = ARG(scan); /* which paren pair */
3351 ln = PL_regstartp[n];
3352 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3353 if ((I32)*PL_reglastparen < n || ln == -1)
3354 sayNO; /* Do not match unless seen CLOSEn. */
3355 if (ln == PL_regendp[n])
3359 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3361 const char *e = PL_bostr + PL_regendp[n];
3363 * Note that we can't do the "other character" lookup trick as
3364 * in the 8-bit case (no pun intended) because in Unicode we
3365 * have to map both upper and title case to lower case.
3369 STRLEN ulen1, ulen2;
3370 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3371 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3375 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3376 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3377 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3384 nextchr = UCHARAT(locinput);
3388 /* Inline the first character, for speed. */
3389 if (UCHARAT(s) != nextchr &&
3391 (UCHARAT(s) != (type == REFF
3392 ? PL_fold : PL_fold_locale)[nextchr])))
3394 ln = PL_regendp[n] - ln;
3395 if (locinput + ln > PL_regeol)
3397 if (ln > 1 && (type == REF
3398 ? memNE(s, locinput, ln)
3400 ? ibcmp(s, locinput, ln)
3401 : ibcmp_locale(s, locinput, ln))))
3404 nextchr = UCHARAT(locinput);
3414 #define ST st->u.eval
3418 regnode *startpoint;
3421 case GOSUB: /* /(...(?1))/ */
3422 if (cur_eval && cur_eval->locinput==locinput) {
3423 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3424 Perl_croak(aTHX_ "Infinite recursion in regex");
3425 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3427 "Pattern subroutine nesting without pos change"
3428 " exceeded limit in regex");
3433 (void)ReREFCNT_inc(rex);
3434 if (OP(scan)==GOSUB) {
3435 startpoint = scan + ARG2L(scan);
3436 ST.close_paren = ARG(scan);
3438 startpoint = re->program+1;
3441 goto eval_recurse_doit;
3443 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3444 if (cur_eval && cur_eval->locinput==locinput) {
3445 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3446 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3451 /* execute the code in the {...} */
3453 SV ** const before = SP;
3454 OP_4tree * const oop = PL_op;
3455 COP * const ocurcop = PL_curcop;
3459 PL_op = (OP_4tree*)rex->data->data[n];
3460 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3461 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3462 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3463 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3465 CALLRUNOPS(aTHX); /* Scalar context. */
3468 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3475 PAD_RESTORE_LOCAL(old_comppad);
3476 PL_curcop = ocurcop;
3479 sv_setsv(save_scalar(PL_replgv), ret);
3483 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3486 /* extract RE object from returned value; compiling if
3491 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3492 mg = mg_find(sv, PERL_MAGIC_qr);
3493 else if (SvSMAGICAL(ret)) {
3494 if (SvGMAGICAL(ret))
3495 sv_unmagic(ret, PERL_MAGIC_qr);
3497 mg = mg_find(ret, PERL_MAGIC_qr);
3501 re = (regexp *)mg->mg_obj;
3502 (void)ReREFCNT_inc(re);
3506 const char * const t = SvPV_const(ret, len);
3508 const I32 osize = PL_regsize;
3511 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3512 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3514 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3516 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3522 debug_start_match(re, do_utf8, locinput, PL_regeol,
3523 "Matching embedded");
3525 startpoint = re->program + 1;
3526 ST.close_paren = 0; /* only used for GOSUB */
3527 /* borrowed from regtry */
3528 if (PL_reg_start_tmpl <= re->nparens) {
3529 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3530 if(PL_reg_start_tmp)
3531 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3533 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3536 eval_recurse_doit: /* Share code with GOSUB below this line */
3537 /* run the pattern returned from (??{...}) */
3538 ST.cp = regcppush(0); /* Save *all* the positions. */
3539 REGCP_SET(ST.lastcp);
3541 PL_regstartp = re->startp; /* essentially NOOP on GOSUB */
3542 PL_regendp = re->endp; /* essentially NOOP on GOSUB */
3544 *PL_reglastparen = 0;
3545 *PL_reglastcloseparen = 0;
3546 PL_reginput = locinput;
3549 /* XXXX This is too dramatic a measure... */
3552 ST.toggle_reg_flags = PL_reg_flags;
3553 if (re->reganch & ROPT_UTF8)
3554 PL_reg_flags |= RF_utf8;
3556 PL_reg_flags &= ~RF_utf8;
3557 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3560 ST.prev_curlyx = cur_curlyx;
3564 ST.prev_eval = cur_eval;
3566 /* now continue from first node in postoned RE */
3567 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3570 /* logical is 1, /(?(?{...})X|Y)/ */
3571 sw = (bool)SvTRUE(ret);
3576 case EVAL_AB: /* cleanup after a successful (??{A})B */
3577 /* note: this is called twice; first after popping B, then A */
3578 PL_reg_flags ^= ST.toggle_reg_flags;
3582 cur_eval = ST.prev_eval;
3583 cur_curlyx = ST.prev_curlyx;
3584 /* XXXX This is too dramatic a measure... */
3589 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3590 /* note: this is called twice; first after popping B, then A */
3591 PL_reg_flags ^= ST.toggle_reg_flags;
3594 PL_reginput = locinput;
3595 REGCP_UNWIND(ST.lastcp);
3597 cur_eval = ST.prev_eval;
3598 cur_curlyx = ST.prev_curlyx;
3599 /* XXXX This is too dramatic a measure... */
3605 n = ARG(scan); /* which paren pair */
3606 PL_reg_start_tmp[n] = locinput;
3611 n = ARG(scan); /* which paren pair */
3612 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3613 PL_regendp[n] = locinput - PL_bostr;
3614 /*if (n > PL_regsize)
3616 if (n > (I32)*PL_reglastparen)
3617 *PL_reglastparen = n;
3618 *PL_reglastcloseparen = n;
3619 if (cur_eval && cur_eval->u.eval.close_paren == (U32)n) {
3624 n = ARG(scan); /* which paren pair */
3625 sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3628 /* reg_check_named_buff_matched returns 0 for no match */
3629 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3633 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == (U32)n));
3639 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3641 next = NEXTOPER(NEXTOPER(scan));
3643 next = scan + ARG(scan);
3644 if (OP(next) == IFTHEN) /* Fake one. */
3645 next = NEXTOPER(NEXTOPER(next));
3649 logical = scan->flags;
3652 /*******************************************************************
3654 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3655 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3656 STAR/PLUS/CURLY/CURLYN are used instead.)
3658 A*B is compiled as <CURLYX><A><WHILEM><B>
3660 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3661 state, which contains the current count, initialised to -1. It also sets
3662 cur_curlyx to point to this state, with any previous value saved in the
3665 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3666 since the pattern may possibly match zero times (i.e. it's a while {} loop
3667 rather than a do {} while loop).
3669 Each entry to WHILEM represents a successful match of A. The count in the
3670 CURLYX block is incremented, another WHILEM state is pushed, and execution
3671 passes to A or B depending on greediness and the current count.
3673 For example, if matching against the string a1a2a3b (where the aN are
3674 substrings that match /A/), then the match progresses as follows: (the
3675 pushed states are interspersed with the bits of strings matched so far):
3678 <CURLYX cnt=0><WHILEM>
3679 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3680 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3681 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3682 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3684 (Contrast this with something like CURLYM, which maintains only a single
3688 a1 <CURLYM cnt=1> a2
3689 a1 a2 <CURLYM cnt=2> a3
3690 a1 a2 a3 <CURLYM cnt=3> b
3693 Each WHILEM state block marks a point to backtrack to upon partial failure
3694 of A or B, and also contains some minor state data related to that
3695 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3696 overall state, such as the count, and pointers to the A and B ops.
3698 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3699 must always point to the *current* CURLYX block, the rules are:
3701 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3702 and set cur_curlyx to point the new block.
3704 When popping the CURLYX block after a successful or unsuccessful match,
3705 restore the previous cur_curlyx.
3707 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3708 to the outer one saved in the CURLYX block.
3710 When popping the WHILEM block after a successful or unsuccessful B match,
3711 restore the previous cur_curlyx.
3713 Here's an example for the pattern (AI* BI)*BO
3714 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3717 curlyx backtrack stack
3718 ------ ---------------
3720 CO <CO prev=NULL> <WO>
3721 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3722 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3723 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3725 At this point the pattern succeeds, and we work back down the stack to
3726 clean up, restoring as we go:
3728 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3729 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3730 CO <CO prev=NULL> <WO>
3733 *******************************************************************/
3735 #define ST st->u.curlyx
3737 case CURLYX: /* start of /A*B/ (for complex A) */
3739 /* No need to save/restore up to this paren */
3740 I32 parenfloor = scan->flags;
3742 assert(next); /* keep Coverity happy */
3743 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3746 /* XXXX Probably it is better to teach regpush to support
3747 parenfloor > PL_regsize... */
3748 if (parenfloor > (I32)*PL_reglastparen)
3749 parenfloor = *PL_reglastparen; /* Pessimization... */
3751 ST.prev_curlyx= cur_curlyx;
3753 ST.cp = PL_savestack_ix;
3755 /* these fields contain the state of the current curly.
3756 * they are accessed by subsequent WHILEMs */
3757 ST.parenfloor = parenfloor;
3758 ST.min = ARG1(scan);
3759 ST.max = ARG2(scan);
3760 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3764 ST.count = -1; /* this will be updated by WHILEM */
3765 ST.lastloc = NULL; /* this will be updated by WHILEM */
3767 PL_reginput = locinput;
3768 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
3772 case CURLYX_end: /* just finished matching all of A*B */
3774 cur_curlyx = ST.prev_curlyx;
3778 case CURLYX_end_fail: /* just failed to match all of A*B */
3780 cur_curlyx = ST.prev_curlyx;
3786 #define ST st->u.whilem
3788 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
3790 /* see the discussion above about CURLYX/WHILEM */
3792 assert(cur_curlyx); /* keep Coverity happy */
3793 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
3794 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
3795 ST.cache_offset = 0;
3798 PL_reginput = locinput;
3800 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3801 "%*s whilem: matched %ld out of %ld..%ld\n",
3802 REPORT_CODE_OFF+depth*2, "", (long)n,
3803 (long)cur_curlyx->u.curlyx.min,
3804 (long)cur_curlyx->u.curlyx.max)
3807 /* First just match a string of min A's. */
3809 if (n < cur_curlyx->u.curlyx.min) {
3810 cur_curlyx->u.curlyx.lastloc = locinput;
3811 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
3815 /* If degenerate A matches "", assume A done. */
3817 if (locinput == cur_curlyx->u.curlyx.lastloc) {
3818 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3819 "%*s whilem: empty match detected, trying continuation...\n",
3820 REPORT_CODE_OFF+depth*2, "")
3822 goto do_whilem_B_max;
3825 /* super-linear cache processing */
3829 if (!PL_reg_maxiter) {
3830 /* start the countdown: Postpone detection until we
3831 * know the match is not *that* much linear. */
3832 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3833 /* possible overflow for long strings and many CURLYX's */
3834 if (PL_reg_maxiter < 0)
3835 PL_reg_maxiter = I32_MAX;
3836 PL_reg_leftiter = PL_reg_maxiter;
3839 if (PL_reg_leftiter-- == 0) {
3840 /* initialise cache */
3841 const I32 size = (PL_reg_maxiter + 7)/8;
3842 if (PL_reg_poscache) {
3843 if ((I32)PL_reg_poscache_size < size) {
3844 Renew(PL_reg_poscache, size, char);
3845 PL_reg_poscache_size = size;
3847 Zero(PL_reg_poscache, size, char);
3850 PL_reg_poscache_size = size;
3851 Newxz(PL_reg_poscache, size, char);
3853 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3854 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
3855 PL_colors[4], PL_colors[5])
3859 if (PL_reg_leftiter < 0) {
3860 /* have we already failed at this position? */
3862 offset = (scan->flags & 0xf) - 1
3863 + (locinput - PL_bostr) * (scan->flags>>4);
3864 mask = 1 << (offset % 8);
3866 if (PL_reg_poscache[offset] & mask) {
3867 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3868 "%*s whilem: (cache) already tried at this position...\n",
3869 REPORT_CODE_OFF+depth*2, "")
3871 sayNO; /* cache records failure */
3873 ST.cache_offset = offset;
3874 ST.cache_mask = mask;
3878 /* Prefer B over A for minimal matching. */
3880 if (cur_curlyx->u.curlyx.minmod) {
3881 ST.save_curlyx = cur_curlyx;
3882 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
3883 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
3884 REGCP_SET(ST.lastcp);
3885 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
3889 /* Prefer A over B for maximal matching. */
3891 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
3892 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3893 cur_curlyx->u.curlyx.lastloc = locinput;
3894 REGCP_SET(ST.lastcp);
3895 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
3898 goto do_whilem_B_max;
3902 case WHILEM_B_min: /* just matched B in a minimal match */
3903 case WHILEM_B_max: /* just matched B in a maximal match */
3904 cur_curlyx = ST.save_curlyx;
3908 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
3909 cur_curlyx = ST.save_curlyx;
3910 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
3911 cur_curlyx->u.curlyx.count--;
3915 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
3916 REGCP_UNWIND(ST.lastcp);
3919 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
3920 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
3921 cur_curlyx->u.curlyx.count--;
3925 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
3926 REGCP_UNWIND(ST.lastcp);
3927 regcppop(rex); /* Restore some previous $<digit>s? */
3928 PL_reginput = locinput;
3929 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3930 "%*s whilem: failed, trying continuation...\n",
3931 REPORT_CODE_OFF+depth*2, "")
3934 if (cur_curlyx->u.curlyx.count >= REG_INFTY
3935 && ckWARN(WARN_REGEXP)
3936 && !(PL_reg_flags & RF_warned))
3938 PL_reg_flags |= RF_warned;
3939 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3940 "Complex regular subexpression recursion",
3945 ST.save_curlyx = cur_curlyx;
3946 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
3947 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
3950 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
3951 cur_curlyx = ST.save_curlyx;
3952 REGCP_UNWIND(ST.lastcp);
3955 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
3956 /* Maximum greed exceeded */
3957 if (cur_curlyx->u.curlyx.count >= REG_INFTY
3958 && ckWARN(WARN_REGEXP)
3959 && !(PL_reg_flags & RF_warned))
3961 PL_reg_flags |= RF_warned;
3962 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
3963 "%s limit (%d) exceeded",
3964 "Complex regular subexpression recursion",
3967 cur_curlyx->u.curlyx.count--;
3971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3972 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
3974 /* Try grabbing another A and see if it helps. */
3975 PL_reginput = locinput;
3976 cur_curlyx->u.curlyx.lastloc = locinput;
3977 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3978 REGCP_SET(ST.lastcp);
3979 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
3983 #define ST st->u.branch
3985 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
3986 next = scan + ARG(scan);
3989 scan = NEXTOPER(scan);
3992 case BRANCH: /* /(...|A|...)/ */
3993 scan = NEXTOPER(scan); /* scan now points to inner node */
3994 if (!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
3995 /* last branch; skip state push and jump direct to node */
3997 ST.lastparen = *PL_reglastparen;
3998 ST.next_branch = next;
4000 PL_reginput = locinput;
4002 /* Now go into the branch */
4003 PUSH_STATE_GOTO(BRANCH_next, scan);
4006 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4007 REGCP_UNWIND(ST.cp);
4008 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4010 *PL_reglastparen = n;
4011 /*dmq: *PL_reglastcloseparen = n; */
4012 scan = ST.next_branch;
4013 /* no more branches? */
4014 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ))
4016 continue; /* execute next BRANCH[J] op */
4024 #define ST st->u.curlym
4026 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4028 /* This is an optimisation of CURLYX that enables us to push
4029 * only a single backtracking state, no matter now many matches
4030 * there are in {m,n}. It relies on the pattern being constant
4031 * length, with no parens to influence future backrefs
4035 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4037 /* if paren positive, emulate an OPEN/CLOSE around A */
4039 I32 paren = ST.me->flags;
4040 if (paren > PL_regsize)
4042 if (paren > (I32)*PL_reglastparen)
4043 *PL_reglastparen = paren;
4044 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4052 ST.c1 = CHRTEST_UNINIT;
4055 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4058 curlym_do_A: /* execute the A in /A{m,n}B/ */
4059 PL_reginput = locinput;
4060 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4063 case CURLYM_A: /* we've just matched an A */
4064 locinput = st->locinput;
4065 nextchr = UCHARAT(locinput);
4068 /* after first match, determine A's length: u.curlym.alen */
4069 if (ST.count == 1) {
4070 if (PL_reg_match_utf8) {
4072 while (s < PL_reginput) {
4078 ST.alen = PL_reginput - locinput;
4081 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4084 PerlIO_printf(Perl_debug_log,
4085 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4086 (int)(REPORT_CODE_OFF+(depth*2)), "",
4087 (IV) ST.count, (IV)ST.alen)
4090 locinput = PL_reginput;
4092 if (cur_eval && cur_eval->u.eval.close_paren &&
4093 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4096 if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
4097 goto curlym_do_A; /* try to match another A */
4098 goto curlym_do_B; /* try to match B */
4100 case CURLYM_A_fail: /* just failed to match an A */
4101 REGCP_UNWIND(ST.cp);
4103 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4104 || (cur_eval && cur_eval->u.eval.close_paren &&
4105 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4108 curlym_do_B: /* execute the B in /A{m,n}B/ */
4109 PL_reginput = locinput;
4110 if (ST.c1 == CHRTEST_UNINIT) {
4111 /* calculate c1 and c2 for possible match of 1st char
4112 * following curly */
4113 ST.c1 = ST.c2 = CHRTEST_VOID;
4114 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4115 regnode *text_node = ST.B;
4116 if (! HAS_TEXT(text_node))
4117 FIND_NEXT_IMPT(text_node);
4118 if (HAS_TEXT(text_node)
4119 && PL_regkind[OP(text_node)] != REF)
4121 ST.c1 = (U8)*STRING(text_node);
4123 (OP(text_node) == EXACTF || OP(text_node) == REFF)
4125 : (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4126 ? PL_fold_locale[ST.c1]
4133 PerlIO_printf(Perl_debug_log,
4134 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4135 (int)(REPORT_CODE_OFF+(depth*2)),
4138 if (ST.c1 != CHRTEST_VOID
4139 && UCHARAT(PL_reginput) != ST.c1
4140 && UCHARAT(PL_reginput) != ST.c2)
4142 /* simulate B failing */
4143 state_num = CURLYM_B_fail;
4144 goto reenter_switch;
4148 /* mark current A as captured */
4149 I32 paren = ST.me->flags;
4152 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4153 PL_regendp[paren] = PL_reginput - PL_bostr;
4154 /*dmq: *PL_reglastcloseparen = paren; */
4157 PL_regendp[paren] = -1;
4158 if (cur_eval && cur_eval->u.eval.close_paren &&
4159 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4168 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4171 case CURLYM_B_fail: /* just failed to match a B */
4172 REGCP_UNWIND(ST.cp);
4174 if (ST.count == ARG2(ST.me) /* max */)
4176 goto curlym_do_A; /* try to match a further A */
4178 /* backtrack one A */
4179 if (ST.count == ARG1(ST.me) /* min */)
4182 locinput = HOPc(locinput, -ST.alen);
4183 goto curlym_do_B; /* try to match B */
4186 #define ST st->u.curly
4188 #define CURLY_SETPAREN(paren, success) \
4191 PL_regstartp[paren] = HOPc(locinput, -1) - PL_bostr; \
4192 PL_regendp[paren] = locinput - PL_bostr; \
4193 *PL_reglastcloseparen = paren; \
4196 PL_regendp[paren] = -1; \
4199 case STAR: /* /A*B/ where A is width 1 */
4203 scan = NEXTOPER(scan);
4205 case PLUS: /* /A+B/ where A is width 1 */
4209 scan = NEXTOPER(scan);
4211 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4212 ST.paren = scan->flags; /* Which paren to set */
4213 if (ST.paren > PL_regsize)
4214 PL_regsize = ST.paren;
4215 if (ST.paren > (I32)*PL_reglastparen)
4216 *PL_reglastparen = ST.paren;
4217 ST.min = ARG1(scan); /* min to match */
4218 ST.max = ARG2(scan); /* max to match */
4219 if (cur_eval && cur_eval->u.eval.close_paren &&
4220 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4224 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4226 case CURLY: /* /A{m,n}B/ where A is width 1 */
4228 ST.min = ARG1(scan); /* min to match */
4229 ST.max = ARG2(scan); /* max to match */
4230 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4233 * Lookahead to avoid useless match attempts
4234 * when we know what character comes next.
4236 * Used to only do .*x and .*?x, but now it allows
4237 * for )'s, ('s and (?{ ... })'s to be in the way
4238 * of the quantifier and the EXACT-like node. -- japhy
4241 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4243 if (HAS_TEXT(next) || JUMPABLE(next)) {
4245 regnode *text_node = next;
4247 if (! HAS_TEXT(text_node))
4248 FIND_NEXT_IMPT(text_node);
4250 if (! HAS_TEXT(text_node))
4251 ST.c1 = ST.c2 = CHRTEST_VOID;
4253 if (PL_regkind[OP(text_node)] == REF) {
4254 ST.c1 = ST.c2 = CHRTEST_VOID;
4255 goto assume_ok_easy;
4258 s = (U8*)STRING(text_node);
4262 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
4263 ST.c2 = PL_fold[ST.c1];
4264 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4265 ST.c2 = PL_fold_locale[ST.c1];
4268 if (OP(text_node) == EXACTF || OP(text_node) == REFF) {
4269 STRLEN ulen1, ulen2;
4270 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4271 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4273 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4274 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4276 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4278 0 : UTF8_ALLOW_ANY);
4279 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4281 0 : UTF8_ALLOW_ANY);
4283 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4285 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4290 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4297 ST.c1 = ST.c2 = CHRTEST_VOID;
4302 PL_reginput = locinput;
4305 if (ST.min && regrepeat(rex, ST.A, ST.min) < ST.min)
4308 locinput = PL_reginput;
4310 if (ST.c1 == CHRTEST_VOID)
4311 goto curly_try_B_min;
4313 ST.oldloc = locinput;
4315 /* set ST.maxpos to the furthest point along the
4316 * string that could possibly match */
4317 if (ST.max == REG_INFTY) {
4318 ST.maxpos = PL_regeol - 1;
4320 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4324 int m = ST.max - ST.min;
4325 for (ST.maxpos = locinput;
4326 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4327 ST.maxpos += UTF8SKIP(ST.maxpos);
4330 ST.maxpos = locinput + ST.max - ST.min;
4331 if (ST.maxpos >= PL_regeol)
4332 ST.maxpos = PL_regeol - 1;
4334 goto curly_try_B_min_known;
4338 ST.count = regrepeat(rex, ST.A, ST.max);
4339 locinput = PL_reginput;
4340 if (ST.count < ST.min)
4342 if ((ST.count > ST.min)
4343 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4345 /* A{m,n} must come at the end of the string, there's
4346 * no point in backing off ... */
4348 /* ...except that $ and \Z can match before *and* after
4349 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4350 We may back off by one in this case. */
4351 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4355 goto curly_try_B_max;
4360 case CURLY_B_min_known_fail:
4361 /* failed to find B in a non-greedy match where c1,c2 valid */
4362 if (ST.paren && ST.count)
4363 PL_regendp[ST.paren] = -1;
4365 PL_reginput = locinput; /* Could be reset... */
4366 REGCP_UNWIND(ST.cp);
4367 /* Couldn't or didn't -- move forward. */
4368 ST.oldloc = locinput;
4370 locinput += UTF8SKIP(locinput);
4374 curly_try_B_min_known:
4375 /* find the next place where 'B' could work, then call B */
4379 n = (ST.oldloc == locinput) ? 0 : 1;
4380 if (ST.c1 == ST.c2) {
4382 /* set n to utf8_distance(oldloc, locinput) */
4383 while (locinput <= ST.maxpos &&
4384 utf8n_to_uvchr((U8*)locinput,
4385 UTF8_MAXBYTES, &len,
4386 uniflags) != (UV)ST.c1) {
4392 /* set n to utf8_distance(oldloc, locinput) */
4393 while (locinput <= ST.maxpos) {
4395 const UV c = utf8n_to_uvchr((U8*)locinput,
4396 UTF8_MAXBYTES, &len,
4398 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4406 if (ST.c1 == ST.c2) {
4407 while (locinput <= ST.maxpos &&
4408 UCHARAT(locinput) != ST.c1)
4412 while (locinput <= ST.maxpos
4413 && UCHARAT(locinput) != ST.c1
4414 && UCHARAT(locinput) != ST.c2)
4417 n = locinput - ST.oldloc;
4419 if (locinput > ST.maxpos)
4421 /* PL_reginput == oldloc now */
4424 if (regrepeat(rex, ST.A, n) < n)
4427 PL_reginput = locinput;
4428 CURLY_SETPAREN(ST.paren, ST.count);
4429 if (cur_eval && cur_eval->u.eval.close_paren &&
4430 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4433 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4438 case CURLY_B_min_fail:
4439 /* failed to find B in a non-greedy match where c1,c2 invalid */
4440 if (ST.paren && ST.count)
4441 PL_regendp[ST.paren] = -1;
4443 REGCP_UNWIND(ST.cp);
4444 /* failed -- move forward one */
4445 PL_reginput = locinput;
4446 if (regrepeat(rex, ST.A, 1)) {
4448 locinput = PL_reginput;
4449 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4450 ST.count > 0)) /* count overflow ? */
4453 CURLY_SETPAREN(ST.paren, ST.count);
4454 if (cur_eval && cur_eval->u.eval.close_paren &&
4455 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4458 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4466 /* a successful greedy match: now try to match B */
4467 if (cur_eval && cur_eval->u.eval.close_paren &&
4468 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4473 if (ST.c1 != CHRTEST_VOID)
4474 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4475 UTF8_MAXBYTES, 0, uniflags)
4476 : (UV) UCHARAT(PL_reginput);
4477 /* If it could work, try it. */
4478 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4479 CURLY_SETPAREN(ST.paren, ST.count);
4480 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4485 case CURLY_B_max_fail:
4486 /* failed to find B in a greedy match */
4487 if (ST.paren && ST.count)
4488 PL_regendp[ST.paren] = -1;
4490 REGCP_UNWIND(ST.cp);
4492 if (--ST.count < ST.min)
4494 PL_reginput = locinput = HOPc(locinput, -1);
4495 goto curly_try_B_max;
4502 /* we've just finished A in /(??{A})B/; now continue with B */
4506 st->u.eval.toggle_reg_flags
4507 = cur_eval->u.eval.toggle_reg_flags;
4508 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4510 st->u.eval.prev_rex = rex; /* inner */
4511 rex = cur_eval->u.eval.prev_rex; /* outer */
4512 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4514 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4515 REGCP_SET(st->u.eval.lastcp);
4516 PL_reginput = locinput;
4518 /* Restore parens of the outer rex without popping the
4520 tmpix = PL_savestack_ix;
4521 PL_savestack_ix = cur_eval->u.eval.lastcp;
4523 PL_savestack_ix = tmpix;
4525 st->u.eval.prev_eval = cur_eval;
4526 cur_eval = cur_eval->u.eval.prev_eval;
4528 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %x\n",
4529 REPORT_CODE_OFF+depth*2, "",(int)cur_eval););
4530 PUSH_YES_STATE_GOTO(EVAL_AB,
4531 st->u.eval.prev_eval->u.eval.B); /* match B */
4534 if (locinput < reginfo->till) {
4535 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4536 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4538 (long)(locinput - PL_reg_starttry),
4539 (long)(reginfo->till - PL_reg_starttry),
4541 sayNO_SILENT; /* Cannot match: too short. */
4543 PL_reginput = locinput; /* put where regtry can find it */
4544 sayYES; /* Success! */
4546 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4548 PerlIO_printf(Perl_debug_log,
4549 "%*s %ssubpattern success...%s\n",
4550 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4551 PL_reginput = locinput; /* put where regtry can find it */
4552 sayYES; /* Success! */
4555 #define ST st->u.ifmatch
4557 case SUSPEND: /* (?>A) */
4559 PL_reginput = locinput;
4562 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4564 goto ifmatch_trivial_fail_test;
4566 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4568 ifmatch_trivial_fail_test:
4570 char * const s = HOPBACKc(locinput, scan->flags);
4575 sw = 1 - (bool)ST.wanted;
4579 next = scan + ARG(scan);
4587 PL_reginput = locinput;
4591 ST.logical = logical;
4592 /* execute body of (?...A) */
4593 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4596 case IFMATCH_A_fail: /* body of (?...A) failed */
4597 ST.wanted = !ST.wanted;
4600 case IFMATCH_A: /* body of (?...A) succeeded */
4602 sw = (bool)ST.wanted;
4604 else if (!ST.wanted)
4607 if (OP(ST.me) == SUSPEND)
4608 locinput = PL_reginput;
4610 locinput = PL_reginput = st->locinput;
4611 nextchr = UCHARAT(locinput);
4613 scan = ST.me + ARG(ST.me);
4616 continue; /* execute B */
4621 next = scan + ARG(scan);
4626 reginfo->cutpoint=PL_regeol;
4630 if ( locinput > reginfo->bol )
4631 reginfo->cutpoint = HOPBACKc(locinput, 1);
4635 PL_reginput = locinput;
4636 PUSH_STATE_GOTO(COMMIT_next,next);
4638 case COMMIT_next_fail:
4644 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
4645 PTR2UV(scan), OP(scan));
4646 Perl_croak(aTHX_ "regexp memory corruption");
4654 /* push a state that backtracks on success */
4655 st->u.yes.prev_yes_state = yes_state;
4659 /* push a new regex state, then continue at scan */
4661 regmatch_state *newst;
4664 regmatch_state *cur = st;
4665 regmatch_state *curyes = yes_state;
4667 regmatch_slab *slab = PL_regmatch_slab;
4668 for (;curd > -1;cur--,curd--) {
4669 if (cur < SLAB_FIRST(slab)) {
4671 cur = SLAB_LAST(slab);
4673 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
4674 REPORT_CODE_OFF + 2 + depth * 2,"",
4675 curd, reg_name[cur->resume_state],
4676 (curyes == cur) ? "yes" : ""
4679 curyes = cur->u.yes.prev_yes_state;
4682 DEBUG_STATE_pp("push")
4685 st->locinput = locinput;
4687 if (newst > SLAB_LAST(PL_regmatch_slab))
4688 newst = S_push_slab(aTHX);
4689 PL_regmatch_state = newst;
4691 locinput = PL_reginput;
4692 nextchr = UCHARAT(locinput);
4700 * We get here only if there's trouble -- normally "case END" is
4701 * the terminating point.
4703 Perl_croak(aTHX_ "corrupted regexp pointers");
4709 /* we have successfully completed a subexpression, but we must now
4710 * pop to the state marked by yes_state and continue from there */
4711 assert(st != yes_state);
4713 while (st != yes_state) {
4715 if (st < SLAB_FIRST(PL_regmatch_slab)) {
4716 PL_regmatch_slab = PL_regmatch_slab->prev;
4717 st = SLAB_LAST(PL_regmatch_slab);
4721 DEBUG_STATE_pp("pop (no final)");
4723 DEBUG_STATE_pp("pop (yes)");
4729 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
4730 || yes_state > SLAB_LAST(PL_regmatch_slab))
4732 /* not in this slab, pop slab */
4733 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
4734 PL_regmatch_slab = PL_regmatch_slab->prev;
4735 st = SLAB_LAST(PL_regmatch_slab);
4737 depth -= (st - yes_state);
4740 yes_state = st->u.yes.prev_yes_state;
4741 PL_regmatch_state = st;
4744 state_num = st->resume_state + no_final;
4745 goto reenter_switch;
4748 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
4749 PL_colors[4], PL_colors[5]));
4756 PerlIO_printf(Perl_debug_log,
4757 "%*s %sfailed...%s\n",
4758 REPORT_CODE_OFF+depth*2, "",
4759 PL_colors[4], PL_colors[5])
4771 /* there's a previous state to backtrack to */
4773 if (st < SLAB_FIRST(PL_regmatch_slab)) {
4774 PL_regmatch_slab = PL_regmatch_slab->prev;
4775 st = SLAB_LAST(PL_regmatch_slab);
4777 PL_regmatch_state = st;
4778 locinput= st->locinput;
4779 nextchr = UCHARAT(locinput);
4781 DEBUG_STATE_pp("pop");
4783 if (yes_state == st)
4784 yes_state = st->u.yes.prev_yes_state;
4786 state_num = st->resume_state + 1; /* failure = success + 1 */
4787 goto reenter_switch;
4793 /* restore original high-water mark */
4794 PL_regmatch_slab = orig_slab;
4795 PL_regmatch_state = orig_state;
4797 /* free all slabs above current one */
4798 if (orig_slab->next) {
4799 regmatch_slab *sl = orig_slab->next;
4800 orig_slab->next = NULL;
4802 regmatch_slab * const osl = sl;
4812 - regrepeat - repeatedly match something simple, report how many
4815 * [This routine now assumes that it will only match on things of length 1.
4816 * That was true before, but now we assume scan - reginput is the count,
4817 * rather than incrementing count on every character. [Er, except utf8.]]
4820 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max)
4823 register char *scan;
4825 register char *loceol = PL_regeol;
4826 register I32 hardcount = 0;
4827 register bool do_utf8 = PL_reg_match_utf8;
4830 if (max == REG_INFTY)
4832 else if (max < loceol - scan)
4833 loceol = scan + max;
4838 while (scan < loceol && hardcount < max && *scan != '\n') {
4839 scan += UTF8SKIP(scan);
4843 while (scan < loceol && *scan != '\n')
4850 while (scan < loceol && hardcount < max) {
4851 scan += UTF8SKIP(scan);
4861 case EXACT: /* length of string is 1 */
4863 while (scan < loceol && UCHARAT(scan) == c)
4866 case EXACTF: /* length of string is 1 */
4868 while (scan < loceol &&
4869 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
4872 case EXACTFL: /* length of string is 1 */
4873 PL_reg_flags |= RF_tainted;
4875 while (scan < loceol &&
4876 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
4882 while (hardcount < max && scan < loceol &&
4883 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
4884 scan += UTF8SKIP(scan);
4888 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
4895 LOAD_UTF8_CHARCLASS_ALNUM();
4896 while (hardcount < max && scan < loceol &&
4897 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4898 scan += UTF8SKIP(scan);
4902 while (scan < loceol && isALNUM(*scan))
4907 PL_reg_flags |= RF_tainted;
4910 while (hardcount < max && scan < loceol &&
4911 isALNUM_LC_utf8((U8*)scan)) {
4912 scan += UTF8SKIP(scan);
4916 while (scan < loceol && isALNUM_LC(*scan))
4923 LOAD_UTF8_CHARCLASS_ALNUM();
4924 while (hardcount < max && scan < loceol &&
4925 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4926 scan += UTF8SKIP(scan);
4930 while (scan < loceol && !isALNUM(*scan))
4935 PL_reg_flags |= RF_tainted;
4938 while (hardcount < max && scan < loceol &&
4939 !isALNUM_LC_utf8((U8*)scan)) {
4940 scan += UTF8SKIP(scan);
4944 while (scan < loceol && !isALNUM_LC(*scan))
4951 LOAD_UTF8_CHARCLASS_SPACE();
4952 while (hardcount < max && scan < loceol &&
4954 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4955 scan += UTF8SKIP(scan);
4959 while (scan < loceol && isSPACE(*scan))
4964 PL_reg_flags |= RF_tainted;
4967 while (hardcount < max && scan < loceol &&
4968 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4969 scan += UTF8SKIP(scan);
4973 while (scan < loceol && isSPACE_LC(*scan))
4980 LOAD_UTF8_CHARCLASS_SPACE();
4981 while (hardcount < max && scan < loceol &&
4983 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4984 scan += UTF8SKIP(scan);
4988 while (scan < loceol && !isSPACE(*scan))
4993 PL_reg_flags |= RF_tainted;
4996 while (hardcount < max && scan < loceol &&
4997 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4998 scan += UTF8SKIP(scan);
5002 while (scan < loceol && !isSPACE_LC(*scan))
5009 LOAD_UTF8_CHARCLASS_DIGIT();
5010 while (hardcount < max && scan < loceol &&
5011 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5012 scan += UTF8SKIP(scan);
5016 while (scan < loceol && isDIGIT(*scan))
5023 LOAD_UTF8_CHARCLASS_DIGIT();
5024 while (hardcount < max && scan < loceol &&
5025 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5026 scan += UTF8SKIP(scan);
5030 while (scan < loceol && !isDIGIT(*scan))
5034 default: /* Called on something of 0 width. */
5035 break; /* So match right here or not at all. */
5041 c = scan - PL_reginput;
5045 GET_RE_DEBUG_FLAGS_DECL;
5047 SV * const prop = sv_newmortal();
5048 regprop(prog, prop, p);
5049 PerlIO_printf(Perl_debug_log,
5050 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5051 REPORT_CODE_OFF+1, "", SvPVX_const(prop),(IV)c,(IV)max);
5059 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5061 - regclass_swash - prepare the utf8 swash
5065 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5071 const struct reg_data * const data = prog ? prog->data : NULL;
5073 if (data && data->count) {
5074 const U32 n = ARG(node);
5076 if (data->what[n] == 's') {
5077 SV * const rv = (SV*)data->data[n];
5078 AV * const av = (AV*)SvRV((SV*)rv);
5079 SV **const ary = AvARRAY(av);
5082 /* See the end of regcomp.c:S_regclass() for
5083 * documentation of these array elements. */
5086 a = SvROK(ary[1]) ? &ary[1] : 0;
5087 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
5091 else if (si && doinit) {
5092 sw = swash_init("utf8", "", si, 1, 0);
5093 (void)av_store(av, 1, sw);
5110 - reginclass - determine if a character falls into a character class
5112 The n is the ANYOF regnode, the p is the target string, lenp
5113 is pointer to the maximum length of how far to go in the p
5114 (if the lenp is zero, UTF8SKIP(p) is used),
5115 do_utf8 tells whether the target string is in UTF-8.
5120 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5123 const char flags = ANYOF_FLAGS(n);
5129 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5130 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5131 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5132 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5133 if (len == (STRLEN)-1)
5134 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5137 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5138 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5141 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5142 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5145 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5149 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5152 if (swash_fetch(sw, p, do_utf8))
5154 else if (flags & ANYOF_FOLD) {
5155 if (!match && lenp && av) {
5157 for (i = 0; i <= av_len(av); i++) {
5158 SV* const sv = *av_fetch(av, i, FALSE);
5160 const char * const s = SvPV_const(sv, len);
5162 if (len <= plen && memEQ(s, (char*)p, len)) {
5170 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5173 to_utf8_fold(p, tmpbuf, &tmplen);
5174 if (swash_fetch(sw, tmpbuf, do_utf8))
5180 if (match && lenp && *lenp == 0)
5181 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5183 if (!match && c < 256) {
5184 if (ANYOF_BITMAP_TEST(n, c))
5186 else if (flags & ANYOF_FOLD) {
5189 if (flags & ANYOF_LOCALE) {
5190 PL_reg_flags |= RF_tainted;
5191 f = PL_fold_locale[c];
5195 if (f != c && ANYOF_BITMAP_TEST(n, f))
5199 if (!match && (flags & ANYOF_CLASS)) {
5200 PL_reg_flags |= RF_tainted;
5202 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5203 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5204 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5205 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5206 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5207 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5208 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5209 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5210 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5211 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5212 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5213 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5214 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5215 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5216 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5217 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5218 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5219 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5220 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5221 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5222 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5223 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5224 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5225 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5226 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5227 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5228 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5229 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5230 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5231 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5232 ) /* How's that for a conditional? */
5239 return (flags & ANYOF_INVERT) ? !match : match;
5243 S_reghop3(U8 *s, I32 off, const U8* lim)
5247 while (off-- && s < lim) {
5248 /* XXX could check well-formedness here */
5253 while (off++ && s > lim) {
5255 if (UTF8_IS_CONTINUED(*s)) {
5256 while (s > lim && UTF8_IS_CONTINUATION(*s))
5259 /* XXX could check well-formedness here */
5266 /* there are a bunch of places where we use two reghop3's that should
5267 be replaced with this routine. but since thats not done yet
5268 we ifdef it out - dmq
5271 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5275 while (off-- && s < rlim) {
5276 /* XXX could check well-formedness here */
5281 while (off++ && s > llim) {
5283 if (UTF8_IS_CONTINUED(*s)) {
5284 while (s > llim && UTF8_IS_CONTINUATION(*s))
5287 /* XXX could check well-formedness here */
5295 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5299 while (off-- && s < lim) {
5300 /* XXX could check well-formedness here */
5307 while (off++ && s > lim) {
5309 if (UTF8_IS_CONTINUED(*s)) {
5310 while (s > lim && UTF8_IS_CONTINUATION(*s))
5313 /* XXX could check well-formedness here */
5322 restore_pos(pTHX_ void *arg)
5325 regexp * const rex = (regexp *)arg;
5326 if (PL_reg_eval_set) {
5327 if (PL_reg_oldsaved) {
5328 rex->subbeg = PL_reg_oldsaved;
5329 rex->sublen = PL_reg_oldsavedlen;
5330 #ifdef PERL_OLD_COPY_ON_WRITE
5331 rex->saved_copy = PL_nrs;
5333 RX_MATCH_COPIED_on(rex);
5335 PL_reg_magic->mg_len = PL_reg_oldpos;
5336 PL_reg_eval_set = 0;
5337 PL_curpm = PL_reg_oldcurpm;
5342 S_to_utf8_substr(pTHX_ register regexp *prog)
5344 if (prog->float_substr && !prog->float_utf8) {
5345 SV* const sv = newSVsv(prog->float_substr);
5346 prog->float_utf8 = sv;
5347 sv_utf8_upgrade(sv);
5348 if (SvTAIL(prog->float_substr))
5350 if (prog->float_substr == prog->check_substr)
5351 prog->check_utf8 = sv;
5353 if (prog->anchored_substr && !prog->anchored_utf8) {
5354 SV* const sv = newSVsv(prog->anchored_substr);
5355 prog->anchored_utf8 = sv;
5356 sv_utf8_upgrade(sv);
5357 if (SvTAIL(prog->anchored_substr))
5359 if (prog->anchored_substr == prog->check_substr)
5360 prog->check_utf8 = sv;
5365 S_to_byte_substr(pTHX_ register regexp *prog)
5368 if (prog->float_utf8 && !prog->float_substr) {
5369 SV* sv = newSVsv(prog->float_utf8);
5370 prog->float_substr = sv;
5371 if (sv_utf8_downgrade(sv, TRUE)) {
5372 if (SvTAIL(prog->float_utf8))
5376 prog->float_substr = sv = &PL_sv_undef;
5378 if (prog->float_utf8 == prog->check_utf8)
5379 prog->check_substr = sv;
5381 if (prog->anchored_utf8 && !prog->anchored_substr) {
5382 SV* sv = newSVsv(prog->anchored_utf8);
5383 prog->anchored_substr = sv;
5384 if (sv_utf8_downgrade(sv, TRUE)) {
5385 if (SvTAIL(prog->anchored_utf8))
5389 prog->anchored_substr = sv = &PL_sv_undef;
5391 if (prog->anchored_utf8 == prog->check_utf8)
5392 prog->check_substr = sv;
5398 * c-indentation-style: bsd
5400 * indent-tabs-mode: t
5403 * ex: set ts=8 sts=4 sw=4 noet: