5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
56 **** Alterations to Henry's code are...
58 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
59 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
61 **** You may distribute under the terms of either the GNU General Public
62 **** License or the Artistic License, as specified in the README file.
64 * Beware that some of this code is subtly aware of the way operator
65 * precedence is structured in regular expressions. Serious changes in
66 * regular-expression syntax might require a total rethink.
69 #define PERL_IN_REGEXEC_C
72 #ifdef PERL_IN_XSUB_RE
78 #define RF_tainted 1 /* tainted information used? */
79 #define RF_warned 2 /* warned about big count? */
81 #define RF_utf8 8 /* Pattern contains multibyte chars? */
83 #define UTF ((PL_reg_flags & RF_utf8) != 0)
85 #define RS_init 1 /* eval environment created */
86 #define RS_set 2 /* replsv value is set */
92 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
98 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
99 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
101 #define HOPc(pos,off) \
102 (char *)(PL_reg_match_utf8 \
103 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
105 #define HOPBACKc(pos, off) \
106 (char*)(PL_reg_match_utf8\
107 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
108 : (pos - off >= PL_bostr) \
112 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
113 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
115 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
116 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
117 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
118 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
119 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
120 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
122 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
124 /* for use after a quantifier and before an EXACT-like node -- japhy */
125 #define JUMPABLE(rn) ( \
126 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
127 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
128 OP(rn) == PLUS || OP(rn) == MINMOD || \
129 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
132 #define HAS_TEXT(rn) ( \
133 PL_regkind[OP(rn)] == EXACT || PL_regkind[OP(rn)] == REF \
137 Search for mandatory following text node; for lookahead, the text must
138 follow but for lookbehind (rn->flags != 0) we skip to the next step.
140 #define FIND_NEXT_IMPT(rn) STMT_START { \
141 while (JUMPABLE(rn)) { \
142 const OPCODE type = OP(rn); \
143 if (type == SUSPEND || PL_regkind[type] == CURLY) \
144 rn = NEXTOPER(NEXTOPER(rn)); \
145 else if (type == PLUS) \
147 else if (type == IFMATCH) \
148 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
149 else rn += NEXT_OFF(rn); \
154 static void restore_pos(pTHX_ void *arg);
157 S_regcppush(pTHX_ I32 parenfloor)
160 const int retval = PL_savestack_ix;
161 #define REGCP_PAREN_ELEMS 4
162 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
164 GET_RE_DEBUG_FLAGS_DECL;
166 if (paren_elems_to_push < 0)
167 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
169 #define REGCP_OTHER_ELEMS 8
170 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
172 for (p = PL_regsize; p > parenfloor; p--) {
173 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
174 SSPUSHINT(PL_regendp[p]);
175 SSPUSHINT(PL_regstartp[p]);
176 SSPUSHPTR(PL_reg_start_tmp[p]);
178 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
179 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
180 (UV)p, (IV)PL_regstartp[p],
181 (IV)(PL_reg_start_tmp[p] - PL_bostr),
185 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
186 SSPUSHPTR(PL_regstartp);
187 SSPUSHPTR(PL_regendp);
188 SSPUSHINT(PL_regsize);
189 SSPUSHINT(*PL_reglastparen);
190 SSPUSHINT(*PL_reglastcloseparen);
191 SSPUSHPTR(PL_reginput);
192 #define REGCP_FRAME_ELEMS 2
193 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
194 * are needed for the regexp context stack bookkeeping. */
195 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
196 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
201 /* These are needed since we do not localize EVAL nodes: */
202 #define REGCP_SET(cp) \
204 PerlIO_printf(Perl_debug_log, \
205 " Setting an EVAL scope, savestack=%"IVdf"\n", \
206 (IV)PL_savestack_ix)); \
209 #define REGCP_UNWIND(cp) \
211 if (cp != PL_savestack_ix) \
212 PerlIO_printf(Perl_debug_log, \
213 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
214 (IV)(cp), (IV)PL_savestack_ix)); \
218 S_regcppop(pTHX_ const regexp *rex)
224 GET_RE_DEBUG_FLAGS_DECL;
226 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
228 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
229 i = SSPOPINT; /* Parentheses elements to pop. */
230 input = (char *) SSPOPPTR;
231 *PL_reglastcloseparen = SSPOPINT;
232 *PL_reglastparen = SSPOPINT;
233 PL_regsize = SSPOPINT;
234 PL_regendp=(I32 *) SSPOPPTR;
235 PL_regstartp=(I32 *) SSPOPPTR;
238 /* Now restore the parentheses context. */
239 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
240 i > 0; i -= REGCP_PAREN_ELEMS) {
242 U32 paren = (U32)SSPOPINT;
243 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
244 PL_regstartp[paren] = SSPOPINT;
246 if (paren <= *PL_reglastparen)
247 PL_regendp[paren] = tmps;
249 PerlIO_printf(Perl_debug_log,
250 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
251 (UV)paren, (IV)PL_regstartp[paren],
252 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
253 (IV)PL_regendp[paren],
254 (paren > *PL_reglastparen ? "(no)" : ""));
258 if (*PL_reglastparen + 1 <= rex->nparens) {
259 PerlIO_printf(Perl_debug_log,
260 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
261 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
265 /* It would seem that the similar code in regtry()
266 * already takes care of this, and in fact it is in
267 * a better location to since this code can #if 0-ed out
268 * but the code in regtry() is needed or otherwise tests
269 * requiring null fields (pat.t#187 and split.t#{13,14}
270 * (as of patchlevel 7877) will fail. Then again,
271 * this code seems to be necessary or otherwise
272 * building DynaLoader will fail:
273 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
275 for (i = *PL_reglastparen + 1; (U32)i <= rex->nparens; i++) {
277 PL_regstartp[i] = -1;
284 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
287 * pregexec and friends
290 #ifndef PERL_IN_XSUB_RE
292 - pregexec - match a regexp against a string
295 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
296 char *strbeg, I32 minend, SV *screamer, U32 nosave)
297 /* strend: pointer to null at end of string */
298 /* strbeg: real beginning of string */
299 /* minend: end of match must be >=minend after stringarg. */
300 /* nosave: For optimizations. */
303 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
304 nosave ? 0 : REXEC_COPY_STR);
309 * Need to implement the following flags for reg_anch:
311 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
313 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
314 * INTUIT_AUTORITATIVE_ML
315 * INTUIT_ONCE_NOML - Intuit can match in one location only.
318 * Another flag for this function: SECOND_TIME (so that float substrs
319 * with giant delta may be not rechecked).
322 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
324 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
325 Otherwise, only SvCUR(sv) is used to get strbeg. */
327 /* XXXX We assume that strpos is strbeg unless sv. */
329 /* XXXX Some places assume that there is a fixed substring.
330 An update may be needed if optimizer marks as "INTUITable"
331 RExen without fixed substrings. Similarly, it is assumed that
332 lengths of all the strings are no more than minlen, thus they
333 cannot come from lookahead.
334 (Or minlen should take into account lookahead.)
335 NOTE: Some of this comment is not correct. minlen does now take account
336 of lookahead/behind. Further research is required. -- demerphq
340 /* A failure to find a constant substring means that there is no need to make
341 an expensive call to REx engine, thus we celebrate a failure. Similarly,
342 finding a substring too deep into the string means that less calls to
343 regtry() should be needed.
345 REx compiler's optimizer found 4 possible hints:
346 a) Anchored substring;
348 c) Whether we are anchored (beginning-of-line or \G);
349 d) First node (of those at offset 0) which may distingush positions;
350 We use a)b)d) and multiline-part of c), and try to find a position in the
351 string which does not contradict any of them.
354 /* Most of decisions we do here should have been done at compile time.
355 The nodes of the REx which we used for the search should have been
356 deleted from the finite automaton. */
359 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
360 char *strend, U32 flags, re_scream_pos_data *data)
363 register I32 start_shift = 0;
364 /* Should be nonnegative! */
365 register I32 end_shift = 0;
370 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
372 register char *other_last = NULL; /* other substr checked before this */
373 char *check_at = NULL; /* check substr found at this pos */
374 const I32 multiline = prog->reganch & PMf_MULTILINE;
376 const char * const i_strpos = strpos;
379 GET_RE_DEBUG_FLAGS_DECL;
381 RX_MATCH_UTF8_set(prog,do_utf8);
383 if (prog->reganch & ROPT_UTF8) {
384 PL_reg_flags |= RF_utf8;
387 debug_start_match(prog, do_utf8, strpos, strend,
388 sv ? "Guessing start of match in sv for"
389 : "Guessing start of match in string for");
392 /* CHR_DIST() would be more correct here but it makes things slow. */
393 if (prog->minlen > strend - strpos) {
394 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
395 "String too short... [re_intuit_start]\n"));
399 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
402 if (!prog->check_utf8 && prog->check_substr)
403 to_utf8_substr(prog);
404 check = prog->check_utf8;
406 if (!prog->check_substr && prog->check_utf8)
407 to_byte_substr(prog);
408 check = prog->check_substr;
410 if (check == &PL_sv_undef) {
411 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
412 "Non-utf8 string cannot match utf8 check string\n"));
415 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
416 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
417 || ( (prog->reganch & ROPT_ANCH_BOL)
418 && !multiline ) ); /* Check after \n? */
421 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
422 | ROPT_IMPLICIT)) /* not a real BOL */
423 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
425 && (strpos != strbeg)) {
426 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
429 if (prog->check_offset_min == prog->check_offset_max &&
430 !(prog->reganch & ROPT_CANY_SEEN)) {
431 /* Substring at constant offset from beg-of-str... */
434 s = HOP3c(strpos, prog->check_offset_min, strend);
437 slen = SvCUR(check); /* >= 1 */
439 if ( strend - s > slen || strend - s < slen - 1
440 || (strend - s == slen && strend[-1] != '\n')) {
441 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
444 /* Now should match s[0..slen-2] */
446 if (slen && (*SvPVX_const(check) != *s
448 && memNE(SvPVX_const(check), s, slen)))) {
450 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
454 else if (*SvPVX_const(check) != *s
455 || ((slen = SvCUR(check)) > 1
456 && memNE(SvPVX_const(check), s, slen)))
459 goto success_at_start;
462 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
464 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
465 end_shift = prog->check_end_shift;
468 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
469 - (SvTAIL(check) != 0);
470 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
472 if (end_shift < eshift)
476 else { /* Can match at random position */
479 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
480 end_shift = prog->check_end_shift;
482 /* end shift should be non negative here */
485 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
487 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
488 (IV)end_shift, prog->precomp);
492 /* Find a possible match in the region s..strend by looking for
493 the "check" substring in the region corrected by start/end_shift. */
496 I32 srch_start_shift = start_shift;
497 I32 srch_end_shift = end_shift;
498 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
499 srch_end_shift -= ((strbeg - s) - srch_start_shift);
500 srch_start_shift = strbeg - s;
502 DEBUG_OPTIMISE_MORE_r({
503 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
504 (IV)prog->check_offset_min,
505 (IV)srch_start_shift,
507 (IV)prog->check_end_shift);
510 if (flags & REXEC_SCREAM) {
511 I32 p = -1; /* Internal iterator of scream. */
512 I32 * const pp = data ? data->scream_pos : &p;
514 if (PL_screamfirst[BmRARE(check)] >= 0
515 || ( BmRARE(check) == '\n'
516 && (BmPREVIOUS(check) == SvCUR(check) - 1)
518 s = screaminstr(sv, check,
519 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
522 /* we may be pointing at the wrong string */
523 if (s && RX_MATCH_COPIED(prog))
524 s = strbeg + (s - SvPVX_const(sv));
526 *data->scream_olds = s;
531 if (prog->reganch & ROPT_CANY_SEEN) {
532 start_point= (U8*)(s + srch_start_shift);
533 end_point= (U8*)(strend - srch_end_shift);
535 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
536 end_point= HOP3(strend, -srch_end_shift, strbeg);
538 DEBUG_OPTIMISE_MORE_r({
539 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
540 (int)(end_point - start_point),
541 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
545 s = fbm_instr( start_point, end_point,
546 check, multiline ? FBMrf_MULTILINE : 0);
549 /* Update the count-of-usability, remove useless subpatterns,
553 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
554 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
555 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
556 (s ? "Found" : "Did not find"),
557 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
558 ? "anchored" : "floating"),
561 (s ? " at offset " : "...\n") );
566 /* Finish the diagnostic message */
567 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
569 /* XXX dmq: first branch is for positive lookbehind...
570 Our check string is offset from the beginning of the pattern.
571 So we need to do any stclass tests offset forward from that
580 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
581 Start with the other substr.
582 XXXX no SCREAM optimization yet - and a very coarse implementation
583 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
584 *always* match. Probably should be marked during compile...
585 Probably it is right to do no SCREAM here...
588 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
589 : (prog->float_substr && prog->anchored_substr))
591 /* Take into account the "other" substring. */
592 /* XXXX May be hopelessly wrong for UTF... */
595 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
598 char * const last = HOP3c(s, -start_shift, strbeg);
600 char * const saved_s = s;
603 t = s - prog->check_offset_max;
604 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
606 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
611 t = HOP3c(t, prog->anchored_offset, strend);
612 if (t < other_last) /* These positions already checked */
614 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
617 /* XXXX It is not documented what units *_offsets are in.
618 We assume bytes, but this is clearly wrong.
619 Meaning this code needs to be carefully reviewed for errors.
623 /* On end-of-str: see comment below. */
624 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
625 if (must == &PL_sv_undef) {
627 DEBUG_r(must = prog->anchored_utf8); /* for debug */
632 HOP3(HOP3(last1, prog->anchored_offset, strend)
633 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
635 multiline ? FBMrf_MULTILINE : 0
638 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
639 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
640 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
641 (s ? "Found" : "Contradicts"),
642 quoted, RE_SV_TAIL(must));
647 if (last1 >= last2) {
648 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
649 ", giving up...\n"));
652 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
653 ", trying floating at offset %ld...\n",
654 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
655 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
656 s = HOP3c(last, 1, strend);
660 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
661 (long)(s - i_strpos)));
662 t = HOP3c(s, -prog->anchored_offset, strbeg);
663 other_last = HOP3c(s, 1, strend);
671 else { /* Take into account the floating substring. */
673 char * const saved_s = s;
676 t = HOP3c(s, -start_shift, strbeg);
678 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
679 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
680 last = HOP3c(t, prog->float_max_offset, strend);
681 s = HOP3c(t, prog->float_min_offset, strend);
684 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
685 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
686 /* fbm_instr() takes into account exact value of end-of-str
687 if the check is SvTAIL(ed). Since false positives are OK,
688 and end-of-str is not later than strend we are OK. */
689 if (must == &PL_sv_undef) {
691 DEBUG_r(must = prog->float_utf8); /* for debug message */
694 s = fbm_instr((unsigned char*)s,
695 (unsigned char*)last + SvCUR(must)
697 must, multiline ? FBMrf_MULTILINE : 0);
699 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
700 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
701 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
702 (s ? "Found" : "Contradicts"),
703 quoted, RE_SV_TAIL(must));
707 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
708 ", giving up...\n"));
711 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
712 ", trying anchored starting at offset %ld...\n",
713 (long)(saved_s + 1 - i_strpos)));
715 s = HOP3c(t, 1, strend);
719 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
720 (long)(s - i_strpos)));
721 other_last = s; /* Fix this later. --Hugo */
731 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
733 DEBUG_OPTIMISE_MORE_r(
734 PerlIO_printf(Perl_debug_log,
735 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
736 (IV)prog->check_offset_min,
737 (IV)prog->check_offset_max,
745 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
747 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
750 /* Fixed substring is found far enough so that the match
751 cannot start at strpos. */
753 if (ml_anch && t[-1] != '\n') {
754 /* Eventually fbm_*() should handle this, but often
755 anchored_offset is not 0, so this check will not be wasted. */
756 /* XXXX In the code below we prefer to look for "^" even in
757 presence of anchored substrings. And we search even
758 beyond the found float position. These pessimizations
759 are historical artefacts only. */
761 while (t < strend - prog->minlen) {
763 if (t < check_at - prog->check_offset_min) {
764 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
765 /* Since we moved from the found position,
766 we definitely contradict the found anchored
767 substr. Due to the above check we do not
768 contradict "check" substr.
769 Thus we can arrive here only if check substr
770 is float. Redo checking for "other"=="fixed".
773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
774 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
775 goto do_other_anchored;
777 /* We don't contradict the found floating substring. */
778 /* XXXX Why not check for STCLASS? */
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
781 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
784 /* Position contradicts check-string */
785 /* XXXX probably better to look for check-string
786 than for "\n", so one should lower the limit for t? */
787 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
788 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
789 other_last = strpos = s = t + 1;
794 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
795 PL_colors[0], PL_colors[1]));
799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
800 PL_colors[0], PL_colors[1]));
804 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
807 /* The found string does not prohibit matching at strpos,
808 - no optimization of calling REx engine can be performed,
809 unless it was an MBOL and we are not after MBOL,
810 or a future STCLASS check will fail this. */
812 /* Even in this situation we may use MBOL flag if strpos is offset
813 wrt the start of the string. */
814 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
815 && (strpos != strbeg) && strpos[-1] != '\n'
816 /* May be due to an implicit anchor of m{.*foo} */
817 && !(prog->reganch & ROPT_IMPLICIT))
822 DEBUG_EXECUTE_r( if (ml_anch)
823 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
824 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
827 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
829 prog->check_utf8 /* Could be deleted already */
830 && --BmUSEFUL(prog->check_utf8) < 0
831 && (prog->check_utf8 == prog->float_utf8)
833 prog->check_substr /* Could be deleted already */
834 && --BmUSEFUL(prog->check_substr) < 0
835 && (prog->check_substr == prog->float_substr)
838 /* If flags & SOMETHING - do not do it many times on the same match */
839 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
840 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
841 if (do_utf8 ? prog->check_substr : prog->check_utf8)
842 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
843 prog->check_substr = prog->check_utf8 = NULL; /* disable */
844 prog->float_substr = prog->float_utf8 = NULL; /* clear */
845 check = NULL; /* abort */
847 /* XXXX This is a remnant of the old implementation. It
848 looks wasteful, since now INTUIT can use many
850 prog->reganch &= ~RE_USE_INTUIT;
857 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
858 /* trie stclasses are too expensive to use here, we are better off to
859 leave it to regmatch itself */
860 if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) {
861 /* minlen == 0 is possible if regstclass is \b or \B,
862 and the fixed substr is ''$.
863 Since minlen is already taken into account, s+1 is before strend;
864 accidentally, minlen >= 1 guaranties no false positives at s + 1
865 even for \b or \B. But (minlen? 1 : 0) below assumes that
866 regstclass does not come from lookahead... */
867 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
868 This leaves EXACTF only, which is dealt with in find_byclass(). */
869 const U8* const str = (U8*)STRING(prog->regstclass);
870 const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT
871 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
874 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
875 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
876 else if (prog->float_substr || prog->float_utf8)
877 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
881 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n",
882 (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg));
885 s = find_byclass(prog, prog->regstclass, s, endpos, NULL);
888 const char *what = NULL;
890 if (endpos == strend) {
891 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
892 "Could not match STCLASS...\n") );
895 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
896 "This position contradicts STCLASS...\n") );
897 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
899 /* Contradict one of substrings */
900 if (prog->anchored_substr || prog->anchored_utf8) {
901 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
902 DEBUG_EXECUTE_r( what = "anchored" );
904 s = HOP3c(t, 1, strend);
905 if (s + start_shift + end_shift > strend) {
906 /* XXXX Should be taken into account earlier? */
907 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
908 "Could not match STCLASS...\n") );
913 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
914 "Looking for %s substr starting at offset %ld...\n",
915 what, (long)(s + start_shift - i_strpos)) );
918 /* Have both, check_string is floating */
919 if (t + start_shift >= check_at) /* Contradicts floating=check */
920 goto retry_floating_check;
921 /* Recheck anchored substring, but not floating... */
925 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
926 "Looking for anchored substr starting at offset %ld...\n",
927 (long)(other_last - i_strpos)) );
928 goto do_other_anchored;
930 /* Another way we could have checked stclass at the
931 current position only: */
936 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
937 "Looking for /%s^%s/m starting at offset %ld...\n",
938 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
941 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
943 /* Check is floating subtring. */
944 retry_floating_check:
945 t = check_at - start_shift;
946 DEBUG_EXECUTE_r( what = "floating" );
947 goto hop_and_restart;
950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
951 "By STCLASS: moving %ld --> %ld\n",
952 (long)(t - i_strpos), (long)(s - i_strpos))
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
957 "Does not contradict STCLASS...\n");
962 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
963 PL_colors[4], (check ? "Guessed" : "Giving up"),
964 PL_colors[5], (long)(s - i_strpos)) );
967 fail_finish: /* Substring not found */
968 if (prog->check_substr || prog->check_utf8) /* could be removed already */
969 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
972 PL_colors[4], PL_colors[5]));
978 #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \
979 foldlen, foldbuf, uniflags) STMT_START { \
980 switch (trie_type) { \
981 case trie_utf8_fold: \
983 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
988 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
989 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
990 foldlen -= UNISKIP( uvc ); \
991 uscan = foldbuf + UNISKIP( uvc ); \
995 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1003 charid = trie->charmap[ uvc ]; \
1007 if (trie->widecharmap) { \
1008 SV** const svpp = hv_fetch(trie->widecharmap, \
1009 (char*)&uvc, sizeof(UV), 0); \
1011 charid = (U16)SvIV(*svpp); \
1016 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1019 ibcmp_utf8(s, NULL, 0, do_utf8, \
1020 m, NULL, ln, (bool)UTF)) \
1021 && (!reginfo || regtry(reginfo, s)) ) \
1024 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1025 uvchr_to_utf8(tmpbuf, c); \
1026 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1028 && (f == c1 || f == c2) \
1029 && (ln == foldlen || \
1030 !ibcmp_utf8((char *) foldbuf, \
1031 NULL, foldlen, do_utf8, \
1033 NULL, ln, (bool)UTF)) \
1034 && (!reginfo || regtry(reginfo, s)) ) \
1039 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1043 && (ln == 1 || !(OP(c) == EXACTF \
1045 : ibcmp_locale(s, m, ln))) \
1046 && (!reginfo || regtry(reginfo, s)) ) \
1052 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1054 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1060 #define REXEC_FBC_SCAN(CoDe) \
1062 while (s < strend) { \
1068 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1069 REXEC_FBC_UTF8_SCAN( \
1071 if (tmp && (!reginfo || regtry(reginfo, s))) \
1080 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1083 if (tmp && (!reginfo || regtry(reginfo, s))) \
1092 #define REXEC_FBC_TRYIT \
1093 if ((!reginfo || regtry(reginfo, s))) \
1096 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1099 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1102 REXEC_FBC_CLASS_SCAN(CoNd); \
1106 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1107 PL_reg_flags |= RF_tainted; \
1109 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1112 REXEC_FBC_CLASS_SCAN(CoNd); \
1116 #define DUMP_EXEC_POS(li,s,doutf8) \
1117 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1119 /* We know what class REx starts with. Try to find this position... */
1120 /* if reginfo is NULL, its a dryrun */
1121 /* annoyingly all the vars in this routine have different names from their counterparts
1122 in regmatch. /grrr */
1125 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1126 const char *strend, const regmatch_info *reginfo)
1129 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
1133 register STRLEN uskip;
1137 register I32 tmp = 1; /* Scratch variable? */
1138 register const bool do_utf8 = PL_reg_match_utf8;
1140 /* We know what class it must start with. */
1144 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1145 !UTF8_IS_INVARIANT((U8)s[0]) ?
1146 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1147 REGINCLASS(prog, c, (U8*)s));
1150 while (s < strend) {
1153 if (REGINCLASS(prog, c, (U8*)s) ||
1154 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1155 /* The assignment of 2 is intentional:
1156 * for the folded sharp s, the skip is 2. */
1157 (skip = SHARP_S_SKIP))) {
1158 if (tmp && (!reginfo || regtry(reginfo, s)))
1171 if (tmp && (!reginfo || regtry(reginfo, s)))
1179 ln = STR_LEN(c); /* length to match in octets/bytes */
1180 lnc = (I32) ln; /* length to match in characters */
1182 STRLEN ulen1, ulen2;
1184 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1185 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1186 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1188 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1189 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1191 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1193 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1196 while (sm < ((U8 *) m + ln)) {
1211 c2 = PL_fold_locale[c1];
1213 e = HOP3c(strend, -((I32)lnc), s);
1215 if (!reginfo && e < s)
1216 e = s; /* Due to minlen logic of intuit() */
1218 /* The idea in the EXACTF* cases is to first find the
1219 * first character of the EXACTF* node and then, if
1220 * necessary, case-insensitively compare the full
1221 * text of the node. The c1 and c2 are the first
1222 * characters (though in Unicode it gets a bit
1223 * more complicated because there are more cases
1224 * than just upper and lower: one needs to use
1225 * the so-called folding case for case-insensitive
1226 * matching (called "loose matching" in Unicode).
1227 * ibcmp_utf8() will do just that. */
1231 U8 tmpbuf [UTF8_MAXBYTES+1];
1232 STRLEN len, foldlen;
1233 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1235 /* Upper and lower of 1st char are equal -
1236 * probably not a "letter". */
1238 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1240 REXEC_FBC_EXACTISH_CHECK(c == c1);
1245 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1248 /* Handle some of the three Greek sigmas cases.
1249 * Note that not all the possible combinations
1250 * are handled here: some of them are handled
1251 * by the standard folding rules, and some of
1252 * them (the character class or ANYOF cases)
1253 * are handled during compiletime in
1254 * regexec.c:S_regclass(). */
1255 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1256 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1257 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1259 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1265 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1267 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1271 PL_reg_flags |= RF_tainted;
1278 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1279 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1281 tmp = ((OP(c) == BOUND ?
1282 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1283 LOAD_UTF8_CHARCLASS_ALNUM();
1284 REXEC_FBC_UTF8_SCAN(
1285 if (tmp == !(OP(c) == BOUND ?
1286 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1287 isALNUM_LC_utf8((U8*)s)))
1295 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1296 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1299 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1305 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, s)))
1309 PL_reg_flags |= RF_tainted;
1316 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1317 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1319 tmp = ((OP(c) == NBOUND ?
1320 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1321 LOAD_UTF8_CHARCLASS_ALNUM();
1322 REXEC_FBC_UTF8_SCAN(
1323 if (tmp == !(OP(c) == NBOUND ?
1324 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1325 isALNUM_LC_utf8((U8*)s)))
1327 else REXEC_FBC_TRYIT;
1331 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1332 tmp = ((OP(c) == NBOUND ?
1333 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1336 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1338 else REXEC_FBC_TRYIT;
1341 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, s)))
1345 REXEC_FBC_CSCAN_PRELOAD(
1346 LOAD_UTF8_CHARCLASS_ALNUM(),
1347 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1351 REXEC_FBC_CSCAN_TAINT(
1352 isALNUM_LC_utf8((U8*)s),
1356 REXEC_FBC_CSCAN_PRELOAD(
1357 LOAD_UTF8_CHARCLASS_ALNUM(),
1358 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1362 REXEC_FBC_CSCAN_TAINT(
1363 !isALNUM_LC_utf8((U8*)s),
1367 REXEC_FBC_CSCAN_PRELOAD(
1368 LOAD_UTF8_CHARCLASS_SPACE(),
1369 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1373 REXEC_FBC_CSCAN_TAINT(
1374 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1378 REXEC_FBC_CSCAN_PRELOAD(
1379 LOAD_UTF8_CHARCLASS_SPACE(),
1380 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1384 REXEC_FBC_CSCAN_TAINT(
1385 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1389 REXEC_FBC_CSCAN_PRELOAD(
1390 LOAD_UTF8_CHARCLASS_DIGIT(),
1391 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1395 REXEC_FBC_CSCAN_TAINT(
1396 isDIGIT_LC_utf8((U8*)s),
1400 REXEC_FBC_CSCAN_PRELOAD(
1401 LOAD_UTF8_CHARCLASS_DIGIT(),
1402 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1406 REXEC_FBC_CSCAN_TAINT(
1407 !isDIGIT_LC_utf8((U8*)s),
1413 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1414 trie_type = do_utf8 ?
1415 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1417 /* what trie are we using right now */
1419 = (reg_ac_data*)prog->data->data[ ARG( c ) ];
1420 reg_trie_data *trie=aho->trie;
1422 const char *last_start = strend - trie->minlen;
1424 const char *real_start = s;
1426 STRLEN maxlen = trie->maxlen;
1428 U8 **points; /* map of where we were in the input string
1429 when reading a given char. For ASCII this
1430 is unnecessary overhead as the relationship
1431 is always 1:1, but for unicode, especially
1432 case folded unicode this is not true. */
1433 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1437 GET_RE_DEBUG_FLAGS_DECL;
1439 /* We can't just allocate points here. We need to wrap it in
1440 * an SV so it gets freed properly if there is a croak while
1441 * running the match */
1444 sv_points=newSV(maxlen * sizeof(U8 *));
1445 SvCUR_set(sv_points,
1446 maxlen * sizeof(U8 *));
1447 SvPOK_on(sv_points);
1448 sv_2mortal(sv_points);
1449 points=(U8**)SvPV_nolen(sv_points );
1450 if ( trie_type != trie_utf8_fold
1451 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1454 bitmap=(U8*)trie->bitmap;
1456 bitmap=(U8*)ANYOF_BITMAP(c);
1458 /* this is the Aho-Corasick algorithm modified a touch
1459 to include special handling for long "unknown char"
1460 sequences. The basic idea being that we use AC as long
1461 as we are dealing with a possible matching char, when
1462 we encounter an unknown char (and we have not encountered
1463 an accepting state) we scan forward until we find a legal
1465 AC matching is basically that of trie matching, except
1466 that when we encounter a failing transition, we fall back
1467 to the current states "fail state", and try the current char
1468 again, a process we repeat until we reach the root state,
1469 state 1, or a legal transition. If we fail on the root state
1470 then we can either terminate if we have reached an accepting
1471 state previously, or restart the entire process from the beginning
1475 while (s <= last_start) {
1476 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1484 U8 *uscan = (U8*)NULL;
1485 U8 *leftmost = NULL;
1487 U32 accepted_word= 0;
1491 while ( state && uc <= (U8*)strend ) {
1493 U32 word = aho->states[ state ].wordnum;
1497 DEBUG_TRIE_EXECUTE_r(
1498 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1499 dump_exec_pos( (char *)uc, c, strend, real_start,
1500 (char *)uc, do_utf8 );
1501 PerlIO_printf( Perl_debug_log,
1502 " Scanning for legal start char...\n");
1505 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1510 if (uc >(U8*)last_start) break;
1514 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1515 if (!leftmost || lpos < leftmost) {
1516 DEBUG_r(accepted_word=word);
1522 points[pointpos++ % maxlen]= uc;
1523 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
1524 uvc, charid, foldlen, foldbuf, uniflags);
1525 DEBUG_TRIE_EXECUTE_r({
1526 dump_exec_pos( (char *)uc, c, strend, real_start,
1528 PerlIO_printf(Perl_debug_log,
1529 " Charid:%3u CP:%4"UVxf" ",
1535 word = aho->states[ state ].wordnum;
1537 base = aho->states[ state ].trans.base;
1539 DEBUG_TRIE_EXECUTE_r({
1541 dump_exec_pos( (char *)uc, c, strend, real_start,
1543 PerlIO_printf( Perl_debug_log,
1544 "%sState: %4"UVxf", word=%"UVxf,
1545 failed ? " Fail transition to " : "",
1546 (UV)state, (UV)word);
1551 (base + charid > trie->uniquecharcount )
1552 && (base + charid - 1 - trie->uniquecharcount
1554 && trie->trans[base + charid - 1 -
1555 trie->uniquecharcount].check == state
1556 && (tmp=trie->trans[base + charid - 1 -
1557 trie->uniquecharcount ].next))
1559 DEBUG_TRIE_EXECUTE_r(
1560 PerlIO_printf( Perl_debug_log," - legal\n"));
1565 DEBUG_TRIE_EXECUTE_r(
1566 PerlIO_printf( Perl_debug_log," - fail\n"));
1568 state = aho->fail[state];
1572 /* we must be accepting here */
1573 DEBUG_TRIE_EXECUTE_r(
1574 PerlIO_printf( Perl_debug_log," - accepting\n"));
1583 if (!state) state = 1;
1586 if ( aho->states[ state ].wordnum ) {
1587 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1588 if (!leftmost || lpos < leftmost) {
1589 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1594 s = (char*)leftmost;
1595 DEBUG_TRIE_EXECUTE_r({
1597 Perl_debug_log,"Matches word #%"UVxf" at position %d. Trying full pattern...\n",
1598 (UV)accepted_word, s - real_start
1601 if (!reginfo || regtry(reginfo, s)) {
1607 DEBUG_TRIE_EXECUTE_r({
1608 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1611 DEBUG_TRIE_EXECUTE_r(
1612 PerlIO_printf( Perl_debug_log,"No match.\n"));
1621 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1630 - regexec_flags - match a regexp against a string
1633 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1634 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1635 /* strend: pointer to null at end of string */
1636 /* strbeg: real beginning of string */
1637 /* minend: end of match must be >=minend after stringarg. */
1638 /* data: May be used for some additional optimizations. */
1639 /* nosave: For optimizations. */
1643 register regnode *c;
1644 register char *startpos = stringarg;
1645 I32 minlen; /* must match at least this many chars */
1646 I32 dontbother = 0; /* how many characters not to try at end */
1647 I32 end_shift = 0; /* Same for the end. */ /* CC */
1648 I32 scream_pos = -1; /* Internal iterator of scream. */
1649 char *scream_olds = NULL;
1650 SV* const oreplsv = GvSV(PL_replgv);
1651 const bool do_utf8 = (bool)DO_UTF8(sv);
1654 regmatch_info reginfo; /* create some info to pass to regtry etc */
1656 GET_RE_DEBUG_FLAGS_DECL;
1658 PERL_UNUSED_ARG(data);
1660 /* Be paranoid... */
1661 if (prog == NULL || startpos == NULL) {
1662 Perl_croak(aTHX_ "NULL regexp parameter");
1666 multiline = prog->reganch & PMf_MULTILINE;
1667 reginfo.prog = prog;
1669 RX_MATCH_UTF8_set(prog, do_utf8);
1671 debug_start_match(prog, do_utf8, startpos, strend,
1675 minlen = prog->minlen;
1677 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1678 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1679 "String too short [regexec_flags]...\n"));
1684 /* Check validity of program. */
1685 if (UCHARAT(prog->program) != REG_MAGIC) {
1686 Perl_croak(aTHX_ "corrupted regexp program");
1690 PL_reg_eval_set = 0;
1693 if (prog->reganch & ROPT_UTF8)
1694 PL_reg_flags |= RF_utf8;
1696 /* Mark beginning of line for ^ and lookbehind. */
1697 reginfo.bol = startpos; /* XXX not used ??? */
1701 /* Mark end of line for $ (and such) */
1704 /* see how far we have to get to not match where we matched before */
1705 reginfo.till = startpos+minend;
1707 /* If there is a "must appear" string, look for it. */
1710 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to set reginfo->ganch */
1713 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1714 reginfo.ganch = startpos;
1715 else if (sv && SvTYPE(sv) >= SVt_PVMG
1717 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1718 && mg->mg_len >= 0) {
1719 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1720 if (prog->reganch & ROPT_ANCH_GPOS) {
1721 if (s > reginfo.ganch)
1726 else /* pos() not defined */
1727 reginfo.ganch = strbeg;
1730 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1731 re_scream_pos_data d;
1733 d.scream_olds = &scream_olds;
1734 d.scream_pos = &scream_pos;
1735 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1737 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1738 goto phooey; /* not present */
1744 /* Simplest case: anchored match need be tried only once. */
1745 /* [unless only anchor is BOL and multiline is set] */
1746 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1747 if (s == startpos && regtry(®info, startpos))
1749 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1750 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1755 dontbother = minlen - 1;
1756 end = HOP3c(strend, -dontbother, strbeg) - 1;
1757 /* for multiline we only have to try after newlines */
1758 if (prog->check_substr || prog->check_utf8) {
1762 if (regtry(®info, s))
1767 if (prog->reganch & RE_USE_INTUIT) {
1768 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1779 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1780 if (regtry(®info, s))
1787 } else if (ROPT_GPOS_CHECK == (prog->reganch & ROPT_GPOS_CHECK))
1789 /* the warning about reginfo.ganch being used without intialization
1790 is bogus -- we set it above, when prog->reganch & ROPT_GPOS_SEEN
1791 and we only enter this block when the same bit is set. */
1792 if (regtry(®info, reginfo.ganch))
1797 /* Messy cases: unanchored match. */
1798 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1799 /* we have /x+whatever/ */
1800 /* it must be a one character string (XXXX Except UTF?) */
1805 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1806 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1807 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1812 DEBUG_EXECUTE_r( did_match = 1 );
1813 if (regtry(®info, s)) goto got_it;
1815 while (s < strend && *s == ch)
1823 DEBUG_EXECUTE_r( did_match = 1 );
1824 if (regtry(®info, s)) goto got_it;
1826 while (s < strend && *s == ch)
1831 DEBUG_EXECUTE_r(if (!did_match)
1832 PerlIO_printf(Perl_debug_log,
1833 "Did not find anchored character...\n")
1836 else if (prog->anchored_substr != NULL
1837 || prog->anchored_utf8 != NULL
1838 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1839 && prog->float_max_offset < strend - s)) {
1844 char *last1; /* Last position checked before */
1848 if (prog->anchored_substr || prog->anchored_utf8) {
1849 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1850 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1851 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1852 back_max = back_min = prog->anchored_offset;
1854 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1855 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1856 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1857 back_max = prog->float_max_offset;
1858 back_min = prog->float_min_offset;
1862 if (must == &PL_sv_undef)
1863 /* could not downgrade utf8 check substring, so must fail */
1869 last = HOP3c(strend, /* Cannot start after this */
1870 -(I32)(CHR_SVLEN(must)
1871 - (SvTAIL(must) != 0) + back_min), strbeg);
1874 last1 = HOPc(s, -1);
1876 last1 = s - 1; /* bogus */
1878 /* XXXX check_substr already used to find "s", can optimize if
1879 check_substr==must. */
1881 dontbother = end_shift;
1882 strend = HOPc(strend, -dontbother);
1883 while ( (s <= last) &&
1884 ((flags & REXEC_SCREAM)
1885 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1886 end_shift, &scream_pos, 0))
1887 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1888 (unsigned char*)strend, must,
1889 multiline ? FBMrf_MULTILINE : 0))) ) {
1890 /* we may be pointing at the wrong string */
1891 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1892 s = strbeg + (s - SvPVX_const(sv));
1893 DEBUG_EXECUTE_r( did_match = 1 );
1894 if (HOPc(s, -back_max) > last1) {
1895 last1 = HOPc(s, -back_min);
1896 s = HOPc(s, -back_max);
1899 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1901 last1 = HOPc(s, -back_min);
1905 while (s <= last1) {
1906 if (regtry(®info, s))
1912 while (s <= last1) {
1913 if (regtry(®info, s))
1919 DEBUG_EXECUTE_r(if (!did_match) {
1920 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1921 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1922 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1923 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1924 ? "anchored" : "floating"),
1925 quoted, RE_SV_TAIL(must));
1929 else if ( (c = prog->regstclass) ) {
1931 const OPCODE op = OP(prog->regstclass);
1932 /* don't bother with what can't match */
1933 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1934 strend = HOPc(strend, -(minlen - 1));
1937 SV * const prop = sv_newmortal();
1938 regprop(prog, prop, c);
1940 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1942 PerlIO_printf(Perl_debug_log,
1943 "Matching stclass %.*s against %s (%d chars)\n",
1944 (int)SvCUR(prop), SvPVX_const(prop),
1945 quoted, (int)(strend - s));
1948 if (find_byclass(prog, c, s, strend, ®info))
1950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
1954 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1959 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1960 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1961 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1963 if (flags & REXEC_SCREAM) {
1964 last = screaminstr(sv, float_real, s - strbeg,
1965 end_shift, &scream_pos, 1); /* last one */
1967 last = scream_olds; /* Only one occurrence. */
1968 /* we may be pointing at the wrong string */
1969 else if (RX_MATCH_COPIED(prog))
1970 s = strbeg + (s - SvPVX_const(sv));
1974 const char * const little = SvPV_const(float_real, len);
1976 if (SvTAIL(float_real)) {
1977 if (memEQ(strend - len + 1, little, len - 1))
1978 last = strend - len + 1;
1979 else if (!multiline)
1980 last = memEQ(strend - len, little, len)
1981 ? strend - len : NULL;
1987 last = rninstr(s, strend, little, little + len);
1989 last = strend; /* matching "$" */
1994 PerlIO_printf(Perl_debug_log,
1995 "%sCan't trim the tail, match fails (should not happen)%s\n",
1996 PL_colors[4], PL_colors[5]));
1997 goto phooey; /* Should not happen! */
1999 dontbother = strend - last + prog->float_min_offset;
2001 if (minlen && (dontbother < minlen))
2002 dontbother = minlen - 1;
2003 strend -= dontbother; /* this one's always in bytes! */
2004 /* We don't know much -- general case. */
2007 if (regtry(®info, s))
2016 if (regtry(®info, s))
2018 } while (s++ < strend);
2026 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2028 if (PL_reg_eval_set) {
2029 /* Preserve the current value of $^R */
2030 if (oreplsv != GvSV(PL_replgv))
2031 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2032 restored, the value remains
2034 restore_pos(aTHX_ prog);
2036 if (prog->paren_names)
2037 (void)hv_iterinit(prog->paren_names);
2039 /* make sure $`, $&, $', and $digit will work later */
2040 if ( !(flags & REXEC_NOT_FIRST) ) {
2041 RX_MATCH_COPY_FREE(prog);
2042 if (flags & REXEC_COPY_STR) {
2043 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2044 #ifdef PERL_OLD_COPY_ON_WRITE
2046 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2048 PerlIO_printf(Perl_debug_log,
2049 "Copy on write: regexp capture, type %d\n",
2052 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2053 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2054 assert (SvPOKp(prog->saved_copy));
2058 RX_MATCH_COPIED_on(prog);
2059 s = savepvn(strbeg, i);
2065 prog->subbeg = strbeg;
2066 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2073 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2074 PL_colors[4], PL_colors[5]));
2075 if (PL_reg_eval_set)
2076 restore_pos(aTHX_ prog);
2082 - regtry - try match at specific point
2084 STATIC I32 /* 0 failure, 1 success */
2085 S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos)
2091 regexp *prog = reginfo->prog;
2092 GET_RE_DEBUG_FLAGS_DECL;
2094 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2097 PL_reg_eval_set = RS_init;
2098 DEBUG_EXECUTE_r(DEBUG_s(
2099 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2100 (IV)(PL_stack_sp - PL_stack_base));
2103 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2104 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2106 /* Apparently this is not needed, judging by wantarray. */
2107 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2108 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2111 /* Make $_ available to executed code. */
2112 if (reginfo->sv != DEFSV) {
2114 DEFSV = reginfo->sv;
2117 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2118 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2119 /* prepare for quick setting of pos */
2120 #ifdef PERL_OLD_COPY_ON_WRITE
2122 sv_force_normal_flags(sv, 0);
2124 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2125 &PL_vtbl_mglob, NULL, 0);
2129 PL_reg_oldpos = mg->mg_len;
2130 SAVEDESTRUCTOR_X(restore_pos, prog);
2132 if (!PL_reg_curpm) {
2133 Newxz(PL_reg_curpm, 1, PMOP);
2136 SV* const repointer = newSViv(0);
2137 /* so we know which PL_regex_padav element is PL_reg_curpm */
2138 SvFLAGS(repointer) |= SVf_BREAK;
2139 av_push(PL_regex_padav,repointer);
2140 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2141 PL_regex_pad = AvARRAY(PL_regex_padav);
2145 PM_SETRE(PL_reg_curpm, prog);
2146 PL_reg_oldcurpm = PL_curpm;
2147 PL_curpm = PL_reg_curpm;
2148 if (RX_MATCH_COPIED(prog)) {
2149 /* Here is a serious problem: we cannot rewrite subbeg,
2150 since it may be needed if this match fails. Thus
2151 $` inside (?{}) could fail... */
2152 PL_reg_oldsaved = prog->subbeg;
2153 PL_reg_oldsavedlen = prog->sublen;
2154 #ifdef PERL_OLD_COPY_ON_WRITE
2155 PL_nrs = prog->saved_copy;
2157 RX_MATCH_COPIED_off(prog);
2160 PL_reg_oldsaved = NULL;
2161 prog->subbeg = PL_bostr;
2162 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2164 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2165 prog->startp[0] = startpos - PL_bostr;
2166 PL_reginput = startpos;
2167 PL_reglastparen = &prog->lastparen;
2168 PL_reglastcloseparen = &prog->lastcloseparen;
2169 prog->lastparen = 0;
2170 prog->lastcloseparen = 0;
2172 PL_regstartp = prog->startp;
2173 PL_regendp = prog->endp;
2174 if (PL_reg_start_tmpl <= prog->nparens) {
2175 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2176 if(PL_reg_start_tmp)
2177 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2179 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2182 /* XXXX What this code is doing here?!!! There should be no need
2183 to do this again and again, PL_reglastparen should take care of
2186 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2187 * Actually, the code in regcppop() (which Ilya may be meaning by
2188 * PL_reglastparen), is not needed at all by the test suite
2189 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2190 * enough, for building DynaLoader, or otherwise this
2191 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2192 * will happen. Meanwhile, this code *is* needed for the
2193 * above-mentioned test suite tests to succeed. The common theme
2194 * on those tests seems to be returning null fields from matches.
2199 if (prog->nparens) {
2201 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2208 if (regmatch(reginfo, prog->program + 1)) {
2209 prog->endp[0] = PL_reginput - PL_bostr;
2212 REGCP_UNWIND(lastcp);
2217 #define sayYES goto yes
2218 #define sayNO goto no
2219 #define sayNO_SILENT goto no_silent
2221 /* we dont use STMT_START/END here because it leads to
2222 "unreachable code" warnings, which are bogus, but distracting. */
2223 #define CACHEsayNO \
2224 if (ST.cache_mask) \
2225 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2228 /* this is used to determine how far from the left messages like
2229 'failed...' are printed. It should be set such that messages
2230 are inline with the regop output that created them.
2232 #define REPORT_CODE_OFF 32
2235 /* Make sure there is a test for this +1 options in re_tests */
2236 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2238 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2239 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2241 #define SLAB_FIRST(s) (&(s)->states[0])
2242 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2244 /* grab a new slab and return the first slot in it */
2246 STATIC regmatch_state *
2249 #if PERL_VERSION < 9
2252 regmatch_slab *s = PL_regmatch_slab->next;
2254 Newx(s, 1, regmatch_slab);
2255 s->prev = PL_regmatch_slab;
2257 PL_regmatch_slab->next = s;
2259 PL_regmatch_slab = s;
2260 return SLAB_FIRST(s);
2264 /* push a new state then goto it */
2266 #define PUSH_STATE_GOTO(state, node) \
2268 st->resume_state = state; \
2271 /* push a new state with success backtracking, then goto it */
2273 #define PUSH_YES_STATE_GOTO(state, node) \
2275 st->resume_state = state; \
2276 goto push_yes_state;
2282 regmatch() - main matching routine
2284 This is basically one big switch statement in a loop. We execute an op,
2285 set 'next' to point the next op, and continue. If we come to a point which
2286 we may need to backtrack to on failure such as (A|B|C), we push a
2287 backtrack state onto the backtrack stack. On failure, we pop the top
2288 state, and re-enter the loop at the state indicated. If there are no more
2289 states to pop, we return failure.
2291 Sometimes we also need to backtrack on success; for example /A+/, where
2292 after successfully matching one A, we need to go back and try to
2293 match another one; similarly for lookahead assertions: if the assertion
2294 completes successfully, we backtrack to the state just before the assertion
2295 and then carry on. In these cases, the pushed state is marked as
2296 'backtrack on success too'. This marking is in fact done by a chain of
2297 pointers, each pointing to the previous 'yes' state. On success, we pop to
2298 the nearest yes state, discarding any intermediate failure-only states.
2299 Sometimes a yes state is pushed just to force some cleanup code to be
2300 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2301 it to free the inner regex.
2303 Note that failure backtracking rewinds the cursor position, while
2304 success backtracking leaves it alone.
2306 A pattern is complete when the END op is executed, while a subpattern
2307 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2308 ops trigger the "pop to last yes state if any, otherwise return true"
2311 A common convention in this function is to use A and B to refer to the two
2312 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2313 the subpattern to be matched possibly multiple times, while B is the entire
2314 rest of the pattern. Variable and state names reflect this convention.
2316 The states in the main switch are the union of ops and failure/success of
2317 substates associated with with that op. For example, IFMATCH is the op
2318 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2319 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2320 successfully matched A and IFMATCH_A_fail is a state saying that we have
2321 just failed to match A. Resume states always come in pairs. The backtrack
2322 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2323 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2324 on success or failure.
2326 The struct that holds a backtracking state is actually a big union, with
2327 one variant for each major type of op. The variable st points to the
2328 top-most backtrack struct. To make the code clearer, within each
2329 block of code we #define ST to alias the relevant union.
2331 Here's a concrete example of a (vastly oversimplified) IFMATCH
2337 #define ST st->u.ifmatch
2339 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2340 ST.foo = ...; // some state we wish to save
2342 // push a yes backtrack state with a resume value of
2343 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2345 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2348 case IFMATCH_A: // we have successfully executed A; now continue with B
2350 bar = ST.foo; // do something with the preserved value
2353 case IFMATCH_A_fail: // A failed, so the assertion failed
2354 ...; // do some housekeeping, then ...
2355 sayNO; // propagate the failure
2362 For any old-timers reading this who are familiar with the old recursive
2363 approach, the code above is equivalent to:
2365 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2374 ...; // do some housekeeping, then ...
2375 sayNO; // propagate the failure
2378 The topmost backtrack state, pointed to by st, is usually free. If you
2379 want to claim it, populate any ST.foo fields in it with values you wish to
2380 save, then do one of
2382 PUSH_STATE_GOTO(resume_state, node);
2383 PUSH_YES_STATE_GOTO(resume_state, node);
2385 which sets that backtrack state's resume value to 'resume_state', pushes a
2386 new free entry to the top of the backtrack stack, then goes to 'node'.
2387 On backtracking, the free slot is popped, and the saved state becomes the
2388 new free state. An ST.foo field in this new top state can be temporarily
2389 accessed to retrieve values, but once the main loop is re-entered, it
2390 becomes available for reuse.
2392 Note that the depth of the backtrack stack constantly increases during the
2393 left-to-right execution of the pattern, rather than going up and down with
2394 the pattern nesting. For example the stack is at its maximum at Z at the
2395 end of the pattern, rather than at X in the following:
2397 /(((X)+)+)+....(Y)+....Z/
2399 The only exceptions to this are lookahead/behind assertions and the cut,
2400 (?>A), which pop all the backtrack states associated with A before
2403 Bascktrack state structs are allocated in slabs of about 4K in size.
2404 PL_regmatch_state and st always point to the currently active state,
2405 and PL_regmatch_slab points to the slab currently containing
2406 PL_regmatch_state. The first time regmatch() is called, the first slab is
2407 allocated, and is never freed until interpreter destruction. When the slab
2408 is full, a new one is allocated and chained to the end. At exit from
2409 regmatch(), slabs allocated since entry are freed.
2414 #define DEBUG_STATE_pp(pp) \
2416 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2417 PerlIO_printf(Perl_debug_log, \
2420 reg_name[st->resume_state] ); \
2424 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2429 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2430 const char *start, const char *end, const char *blurb)
2432 const bool utf8_pat= prog->reganch & ROPT_UTF8 ? 1 : 0;
2436 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2437 prog->precomp, prog->prelen, 60);
2439 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2440 start, end - start, 60);
2442 PerlIO_printf(Perl_debug_log,
2443 "%s%s REx%s %s against %s\n",
2444 PL_colors[4], blurb, PL_colors[5], s0, s1);
2446 if (do_utf8||utf8_pat)
2447 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2448 utf8_pat ? "pattern" : "",
2449 utf8_pat && do_utf8 ? " and " : "",
2450 do_utf8 ? "string" : ""
2456 S_dump_exec_pos(pTHX_ const char *locinput,
2457 const regnode *scan,
2458 const char *loc_regeol,
2459 const char *loc_bostr,
2460 const char *loc_reg_starttry,
2463 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2464 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2465 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2466 /* The part of the string before starttry has one color
2467 (pref0_len chars), between starttry and current
2468 position another one (pref_len - pref0_len chars),
2469 after the current position the third one.
2470 We assume that pref0_len <= pref_len, otherwise we
2471 decrease pref0_len. */
2472 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2473 ? (5 + taill) - l : locinput - loc_bostr;
2476 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2478 pref0_len = pref_len - (locinput - loc_reg_starttry);
2479 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2480 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2481 ? (5 + taill) - pref_len : loc_regeol - locinput);
2482 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2486 if (pref0_len > pref_len)
2487 pref0_len = pref_len;
2489 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2491 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2492 (locinput - pref_len),pref0_len, 60, 4, 5);
2494 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2495 (locinput - pref_len + pref0_len),
2496 pref_len - pref0_len, 60, 2, 3);
2498 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2499 locinput, loc_regeol - locinput, 10, 0, 1);
2501 const STRLEN tlen=len0+len1+len2;
2502 PerlIO_printf(Perl_debug_log,
2503 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2504 (IV)(locinput - loc_bostr),
2507 (docolor ? "" : "> <"),
2509 (int)(tlen > 19 ? 0 : 19 - tlen),
2516 /* reg_check_named_buff_matched()
2517 * Checks to see if a named buffer has matched. The data array of
2518 * buffer numbers corresponding to the buffer is expected to reside
2519 * in the regexp->data->data array in the slot stored in the ARG() of
2520 * node involved. Note that this routine doesn't actually care about the
2521 * name, that information is not preserved from compilation to execution.
2522 * Returns the index of the leftmost defined buffer with the given name
2523 * or 0 if non of the buffers matched.
2526 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2528 SV *sv_dat=(SV*)rex->data->data[ ARG( scan ) ];
2529 I32 *nums=(I32*)SvPVX(sv_dat);
2530 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2531 if ((I32)*PL_reglastparen >= nums[n] &&
2532 PL_regendp[nums[n]] != -1)
2540 STATIC I32 /* 0 failure, 1 success */
2541 S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog)
2543 #if PERL_VERSION < 9
2547 register const bool do_utf8 = PL_reg_match_utf8;
2548 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2550 regexp *rex = reginfo->prog;
2552 regmatch_slab *orig_slab;
2553 regmatch_state *orig_state;
2555 /* the current state. This is a cached copy of PL_regmatch_state */
2556 register regmatch_state *st;
2558 /* cache heavy used fields of st in registers */
2559 register regnode *scan;
2560 register regnode *next;
2561 register I32 n = 0; /* general value; init to avoid compiler warning */
2562 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2563 register char *locinput = PL_reginput;
2564 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2566 bool result = 0; /* return value of S_regmatch */
2567 int depth = 0; /* depth of backtrack stack */
2568 int nochange_depth = 0; /* depth of GOSUB recursion with nochange*/
2569 regmatch_state *yes_state = NULL; /* state to pop to on success of
2571 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2572 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2574 bool no_final = 0; /* if true then we dont backtrack on failure */
2576 /* these three flags are set by various ops to signal information to
2577 * the very next op. They have a useful lifetime of exactly one loop
2578 * iteration, and are not preserved or restored by state pushes/pops
2580 bool sw = 0; /* the condition value in (?(cond)a|b) */
2581 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2582 int logical = 0; /* the following EVAL is:
2586 or the following IFMATCH/UNLESSM is:
2587 false: plain (?=foo)
2588 true: used as a condition: (?(?=foo))
2592 GET_RE_DEBUG_FLAGS_DECL;
2595 /* on first ever call to regmatch, allocate first slab */
2596 if (!PL_regmatch_slab) {
2597 Newx(PL_regmatch_slab, 1, regmatch_slab);
2598 PL_regmatch_slab->prev = NULL;
2599 PL_regmatch_slab->next = NULL;
2600 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2603 /* remember current high-water mark for exit */
2604 /* XXX this should be done with SAVE* instead */
2605 orig_slab = PL_regmatch_slab;
2606 orig_state = PL_regmatch_state;
2608 /* grab next free state slot */
2609 st = ++PL_regmatch_state;
2610 if (st > SLAB_LAST(PL_regmatch_slab))
2611 st = PL_regmatch_state = S_push_slab(aTHX);
2613 /* Note that nextchr is a byte even in UTF */
2614 nextchr = UCHARAT(locinput);
2616 while (scan != NULL) {
2619 SV * const prop = sv_newmortal();
2620 regnode *rnext=regnext(scan);
2621 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2622 regprop(rex, prop, scan);
2624 PerlIO_printf(Perl_debug_log,
2625 "%3"IVdf":%*s%s(%"IVdf")\n",
2626 (IV)(scan - rex->program), depth*2, "",
2628 (PL_regkind[OP(scan)] == END || !rnext) ?
2629 0 : (IV)(rnext - rex->program));
2632 next = scan + NEXT_OFF(scan);
2635 state_num = OP(scan);
2638 switch (state_num) {
2640 if (locinput == PL_bostr)
2642 /* reginfo->till = reginfo->bol; */
2647 if (locinput == PL_bostr ||
2648 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2654 if (locinput == PL_bostr)
2658 if (locinput == reginfo->ganch)
2664 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2669 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2671 if (PL_regeol - locinput > 1)
2675 if (PL_regeol != locinput)
2679 if (!nextchr && locinput >= PL_regeol)
2682 locinput += PL_utf8skip[nextchr];
2683 if (locinput > PL_regeol)
2685 nextchr = UCHARAT(locinput);
2688 nextchr = UCHARAT(++locinput);
2691 if (!nextchr && locinput >= PL_regeol)
2693 nextchr = UCHARAT(++locinput);
2696 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2699 locinput += PL_utf8skip[nextchr];
2700 if (locinput > PL_regeol)
2702 nextchr = UCHARAT(locinput);
2705 nextchr = UCHARAT(++locinput);
2709 #define ST st->u.trie
2711 /* In this case the charclass data is available inline so
2712 we can fail fast without a lot of extra overhead.
2714 if (scan->flags == EXACT || !do_utf8) {
2715 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2717 PerlIO_printf(Perl_debug_log,
2718 "%*s %sfailed to match trie start class...%s\n",
2719 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2728 /* what type of TRIE am I? (utf8 makes this contextual) */
2729 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2730 trie_type = do_utf8 ?
2731 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2734 /* what trie are we using right now */
2735 reg_trie_data * const trie
2736 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2737 U32 state = trie->startstate;
2739 if (trie->bitmap && trie_type != trie_utf8_fold &&
2740 !TRIE_BITMAP_TEST(trie,*locinput)
2742 if (trie->states[ state ].wordnum) {
2744 PerlIO_printf(Perl_debug_log,
2745 "%*s %smatched empty string...%s\n",
2746 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2751 PerlIO_printf(Perl_debug_log,
2752 "%*s %sfailed to match trie start class...%s\n",
2753 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2760 U8 *uc = ( U8* )locinput;
2764 U8 *uscan = (U8*)NULL;
2766 SV *sv_accept_buff = NULL;
2767 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2769 ST.accepted = 0; /* how many accepting states we have seen */
2771 ST.jump = trie->jump;
2775 traverse the TRIE keeping track of all accepting states
2776 we transition through until we get to a failing node.
2779 while ( state && uc <= (U8*)PL_regeol ) {
2780 U32 base = trie->states[ state ].trans.base;
2783 /* We use charid to hold the wordnum as we don't use it
2784 for charid until after we have done the wordnum logic.
2785 We define an alias just so that the wordnum logic reads
2788 #define got_wordnum charid
2789 got_wordnum = trie->states[ state ].wordnum;
2791 if ( got_wordnum ) {
2792 if ( ! ST.accepted ) {
2795 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2796 sv_accept_buff=newSV(bufflen *
2797 sizeof(reg_trie_accepted) - 1);
2798 SvCUR_set(sv_accept_buff, 0);
2799 SvPOK_on(sv_accept_buff);
2800 sv_2mortal(sv_accept_buff);
2803 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2806 if (ST.accepted >= bufflen) {
2808 ST.accept_buff =(reg_trie_accepted*)
2809 SvGROW(sv_accept_buff,
2810 bufflen * sizeof(reg_trie_accepted));
2812 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2813 + sizeof(reg_trie_accepted));
2816 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2817 ST.accept_buff[ST.accepted].endpos = uc;
2819 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2823 DEBUG_TRIE_EXECUTE_r({
2824 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2825 PerlIO_printf( Perl_debug_log,
2826 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2827 2+depth * 2, "", PL_colors[4],
2828 (UV)state, (UV)ST.accepted );
2832 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
2833 uvc, charid, foldlen, foldbuf, uniflags);
2836 (base + charid > trie->uniquecharcount )
2837 && (base + charid - 1 - trie->uniquecharcount
2839 && trie->trans[base + charid - 1 -
2840 trie->uniquecharcount].check == state)
2842 state = trie->trans[base + charid - 1 -
2843 trie->uniquecharcount ].next;
2854 DEBUG_TRIE_EXECUTE_r(
2855 PerlIO_printf( Perl_debug_log,
2856 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2857 charid, uvc, (UV)state, PL_colors[5] );
2864 PerlIO_printf( Perl_debug_log,
2865 "%*s %sgot %"IVdf" possible matches%s\n",
2866 REPORT_CODE_OFF + depth * 2, "",
2867 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2873 case TRIE_next_fail: /* we failed - try next alterative */
2875 if ( ST.accepted == 1 ) {
2876 /* only one choice left - just continue */
2878 reg_trie_data * const trie
2879 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2880 SV ** const tmp = RX_DEBUG(reginfo->prog)
2881 ? av_fetch( trie->words, ST.accept_buff[ 0 ].wordnum-1, 0 )
2883 PerlIO_printf( Perl_debug_log,
2884 "%*s %sonly one match left: #%d <%s>%s\n",
2885 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2886 ST.accept_buff[ 0 ].wordnum,
2887 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2890 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2891 /* in this case we free tmps/leave before we call regmatch
2892 as we wont be using accept_buff again. */
2895 locinput = PL_reginput;
2896 nextchr = UCHARAT(locinput);
2898 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
2901 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
2903 continue; /* execute rest of RE */
2906 if (!ST.accepted-- ) {
2913 There are at least two accepting states left. Presumably
2914 the number of accepting states is going to be low,
2915 typically two. So we simply scan through to find the one
2916 with lowest wordnum. Once we find it, we swap the last
2917 state into its place and decrement the size. We then try to
2918 match the rest of the pattern at the point where the word
2919 ends. If we succeed, control just continues along the
2920 regex; if we fail we return here to try the next accepting
2927 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
2928 DEBUG_TRIE_EXECUTE_r(
2929 PerlIO_printf( Perl_debug_log,
2930 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2931 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
2932 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
2933 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
2936 if (ST.accept_buff[cur].wordnum <
2937 ST.accept_buff[best].wordnum)
2942 reg_trie_data * const trie
2943 = (reg_trie_data*)rex->data->data[ ARG(ST.me) ];
2944 SV ** const tmp = RX_DEBUG(reginfo->prog)
2945 ? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 )
2947 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
2949 ST.me + ST.jump[ST.accept_buff[best].wordnum];
2950 PerlIO_printf( Perl_debug_log,
2951 "%*s %strying alternation #%d <%s> at node #%d %s\n",
2952 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2953 ST.accept_buff[best].wordnum,
2954 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2955 REG_NODE_NUM(nextop),
2959 if ( best<ST.accepted ) {
2960 reg_trie_accepted tmp = ST.accept_buff[ best ];
2961 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
2962 ST.accept_buff[ ST.accepted ] = tmp;
2965 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
2966 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
2967 PUSH_STATE_GOTO(TRIE_next, ST.B);
2970 PUSH_STATE_GOTO(TRIE_next, ST.me + ST.jump[ST.accept_buff[best].wordnum]);
2980 char *s = STRING(scan);
2982 if (do_utf8 != UTF) {
2983 /* The target and the pattern have differing utf8ness. */
2985 const char * const e = s + ln;
2988 /* The target is utf8, the pattern is not utf8. */
2993 if (NATIVE_TO_UNI(*(U8*)s) !=
2994 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3002 /* The target is not utf8, the pattern is utf8. */
3007 if (NATIVE_TO_UNI(*((U8*)l)) !=
3008 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3016 nextchr = UCHARAT(locinput);
3019 /* The target and the pattern have the same utf8ness. */
3020 /* Inline the first character, for speed. */
3021 if (UCHARAT(s) != nextchr)
3023 if (PL_regeol - locinput < ln)
3025 if (ln > 1 && memNE(s, locinput, ln))
3028 nextchr = UCHARAT(locinput);
3032 PL_reg_flags |= RF_tainted;
3035 char * const s = STRING(scan);
3038 if (do_utf8 || UTF) {
3039 /* Either target or the pattern are utf8. */
3040 const char * const l = locinput;
3041 char *e = PL_regeol;
3043 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3044 l, &e, 0, do_utf8)) {
3045 /* One more case for the sharp s:
3046 * pack("U0U*", 0xDF) =~ /ss/i,
3047 * the 0xC3 0x9F are the UTF-8
3048 * byte sequence for the U+00DF. */
3050 toLOWER(s[0]) == 's' &&
3052 toLOWER(s[1]) == 's' &&
3059 nextchr = UCHARAT(locinput);
3063 /* Neither the target and the pattern are utf8. */
3065 /* Inline the first character, for speed. */
3066 if (UCHARAT(s) != nextchr &&
3067 UCHARAT(s) != ((OP(scan) == EXACTF)
3068 ? PL_fold : PL_fold_locale)[nextchr])
3070 if (PL_regeol - locinput < ln)
3072 if (ln > 1 && (OP(scan) == EXACTF
3073 ? ibcmp(s, locinput, ln)
3074 : ibcmp_locale(s, locinput, ln)))
3077 nextchr = UCHARAT(locinput);
3082 STRLEN inclasslen = PL_regeol - locinput;
3084 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3086 if (locinput >= PL_regeol)
3088 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3089 nextchr = UCHARAT(locinput);
3094 nextchr = UCHARAT(locinput);
3095 if (!REGINCLASS(rex, scan, (U8*)locinput))
3097 if (!nextchr && locinput >= PL_regeol)
3099 nextchr = UCHARAT(++locinput);
3103 /* If we might have the case of the German sharp s
3104 * in a casefolding Unicode character class. */
3106 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3107 locinput += SHARP_S_SKIP;
3108 nextchr = UCHARAT(locinput);
3114 PL_reg_flags |= RF_tainted;
3120 LOAD_UTF8_CHARCLASS_ALNUM();
3121 if (!(OP(scan) == ALNUM
3122 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3123 : isALNUM_LC_utf8((U8*)locinput)))
3127 locinput += PL_utf8skip[nextchr];
3128 nextchr = UCHARAT(locinput);
3131 if (!(OP(scan) == ALNUM
3132 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3134 nextchr = UCHARAT(++locinput);
3137 PL_reg_flags |= RF_tainted;
3140 if (!nextchr && locinput >= PL_regeol)
3143 LOAD_UTF8_CHARCLASS_ALNUM();
3144 if (OP(scan) == NALNUM
3145 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3146 : isALNUM_LC_utf8((U8*)locinput))
3150 locinput += PL_utf8skip[nextchr];
3151 nextchr = UCHARAT(locinput);
3154 if (OP(scan) == NALNUM
3155 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3157 nextchr = UCHARAT(++locinput);
3161 PL_reg_flags |= RF_tainted;
3165 /* was last char in word? */
3167 if (locinput == PL_bostr)
3170 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3172 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3174 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3175 ln = isALNUM_uni(ln);
3176 LOAD_UTF8_CHARCLASS_ALNUM();
3177 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3180 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3181 n = isALNUM_LC_utf8((U8*)locinput);
3185 ln = (locinput != PL_bostr) ?
3186 UCHARAT(locinput - 1) : '\n';
3187 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3189 n = isALNUM(nextchr);
3192 ln = isALNUM_LC(ln);
3193 n = isALNUM_LC(nextchr);
3196 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3197 OP(scan) == BOUNDL))
3201 PL_reg_flags |= RF_tainted;
3207 if (UTF8_IS_CONTINUED(nextchr)) {
3208 LOAD_UTF8_CHARCLASS_SPACE();
3209 if (!(OP(scan) == SPACE
3210 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3211 : isSPACE_LC_utf8((U8*)locinput)))
3215 locinput += PL_utf8skip[nextchr];
3216 nextchr = UCHARAT(locinput);
3219 if (!(OP(scan) == SPACE
3220 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3222 nextchr = UCHARAT(++locinput);
3225 if (!(OP(scan) == SPACE
3226 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3228 nextchr = UCHARAT(++locinput);
3232 PL_reg_flags |= RF_tainted;
3235 if (!nextchr && locinput >= PL_regeol)
3238 LOAD_UTF8_CHARCLASS_SPACE();
3239 if (OP(scan) == NSPACE
3240 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3241 : isSPACE_LC_utf8((U8*)locinput))
3245 locinput += PL_utf8skip[nextchr];
3246 nextchr = UCHARAT(locinput);
3249 if (OP(scan) == NSPACE
3250 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3252 nextchr = UCHARAT(++locinput);
3255 PL_reg_flags |= RF_tainted;
3261 LOAD_UTF8_CHARCLASS_DIGIT();
3262 if (!(OP(scan) == DIGIT
3263 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3264 : isDIGIT_LC_utf8((U8*)locinput)))
3268 locinput += PL_utf8skip[nextchr];
3269 nextchr = UCHARAT(locinput);
3272 if (!(OP(scan) == DIGIT
3273 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3275 nextchr = UCHARAT(++locinput);
3278 PL_reg_flags |= RF_tainted;
3281 if (!nextchr && locinput >= PL_regeol)
3284 LOAD_UTF8_CHARCLASS_DIGIT();
3285 if (OP(scan) == NDIGIT
3286 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3287 : isDIGIT_LC_utf8((U8*)locinput))
3291 locinput += PL_utf8skip[nextchr];
3292 nextchr = UCHARAT(locinput);
3295 if (OP(scan) == NDIGIT
3296 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3298 nextchr = UCHARAT(++locinput);
3301 if (locinput >= PL_regeol)
3304 LOAD_UTF8_CHARCLASS_MARK();
3305 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3307 locinput += PL_utf8skip[nextchr];
3308 while (locinput < PL_regeol &&
3309 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3310 locinput += UTF8SKIP(locinput);
3311 if (locinput > PL_regeol)
3316 nextchr = UCHARAT(locinput);
3323 PL_reg_flags |= RF_tainted;
3328 n = reg_check_named_buff_matched(rex,scan);
3331 type = REF + ( type - NREF );
3338 PL_reg_flags |= RF_tainted;
3342 n = ARG(scan); /* which paren pair */
3345 ln = PL_regstartp[n];
3346 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3347 if ((I32)*PL_reglastparen < n || ln == -1)
3348 sayNO; /* Do not match unless seen CLOSEn. */
3349 if (ln == PL_regendp[n])
3353 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3355 const char *e = PL_bostr + PL_regendp[n];
3357 * Note that we can't do the "other character" lookup trick as
3358 * in the 8-bit case (no pun intended) because in Unicode we
3359 * have to map both upper and title case to lower case.
3363 STRLEN ulen1, ulen2;
3364 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3365 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3369 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3370 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3371 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3378 nextchr = UCHARAT(locinput);
3382 /* Inline the first character, for speed. */
3383 if (UCHARAT(s) != nextchr &&
3385 (UCHARAT(s) != (type == REFF
3386 ? PL_fold : PL_fold_locale)[nextchr])))
3388 ln = PL_regendp[n] - ln;
3389 if (locinput + ln > PL_regeol)
3391 if (ln > 1 && (type == REF
3392 ? memNE(s, locinput, ln)
3394 ? ibcmp(s, locinput, ln)
3395 : ibcmp_locale(s, locinput, ln))))
3398 nextchr = UCHARAT(locinput);
3408 #define ST st->u.eval
3412 regnode *startpoint;
3415 case GOSUB: /* /(...(?1))/ */
3416 if (cur_eval && cur_eval->locinput==locinput) {
3417 if (cur_eval->u.eval.close_paren == ARG(scan))
3418 Perl_croak(aTHX_ "Infinite recursion in regex");
3419 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3421 "Pattern subroutine nesting without pos change"
3422 " exceeded limit in regex");
3427 (void)ReREFCNT_inc(rex);
3428 if (OP(scan)==GOSUB) {
3429 startpoint = scan + ARG2L(scan);
3430 ST.close_paren = ARG(scan);
3432 startpoint = re->program+1;
3435 goto eval_recurse_doit;
3437 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3438 if (cur_eval && cur_eval->locinput==locinput) {
3439 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3440 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3445 /* execute the code in the {...} */
3447 SV ** const before = SP;
3448 OP_4tree * const oop = PL_op;
3449 COP * const ocurcop = PL_curcop;
3453 PL_op = (OP_4tree*)rex->data->data[n];
3454 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3455 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3456 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3458 CALLRUNOPS(aTHX); /* Scalar context. */
3461 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3468 PAD_RESTORE_LOCAL(old_comppad);
3469 PL_curcop = ocurcop;
3472 sv_setsv(save_scalar(PL_replgv), ret);
3476 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3479 /* extract RE object from returned value; compiling if
3484 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3485 mg = mg_find(sv, PERL_MAGIC_qr);
3486 else if (SvSMAGICAL(ret)) {
3487 if (SvGMAGICAL(ret))
3488 sv_unmagic(ret, PERL_MAGIC_qr);
3490 mg = mg_find(ret, PERL_MAGIC_qr);
3494 re = (regexp *)mg->mg_obj;
3495 (void)ReREFCNT_inc(re);
3499 const char * const t = SvPV_const(ret, len);
3501 const I32 osize = PL_regsize;
3504 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3505 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3507 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3509 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3515 debug_start_match(re, do_utf8, locinput, PL_regeol,
3516 "Matching embedded");
3518 startpoint = re->program + 1;
3519 ST.close_paren = 0; /* only used for GOSUB */
3520 /* borrowed from regtry */
3521 if (PL_reg_start_tmpl <= re->nparens) {
3522 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3523 if(PL_reg_start_tmp)
3524 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3526 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3529 eval_recurse_doit: /* Share code with GOSUB below this line */
3530 /* run the pattern returned from (??{...}) */
3531 ST.cp = regcppush(0); /* Save *all* the positions. */
3532 REGCP_SET(ST.lastcp);
3534 PL_regstartp = re->startp; /* essentially NOOP on GOSUB */
3535 PL_regendp = re->endp; /* essentially NOOP on GOSUB */
3537 *PL_reglastparen = 0;
3538 *PL_reglastcloseparen = 0;
3539 PL_reginput = locinput;
3542 /* XXXX This is too dramatic a measure... */
3545 ST.toggle_reg_flags = PL_reg_flags;
3546 if (re->reganch & ROPT_UTF8)
3547 PL_reg_flags |= RF_utf8;
3549 PL_reg_flags &= ~RF_utf8;
3550 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3553 ST.prev_curlyx = cur_curlyx;
3557 ST.prev_eval = cur_eval;
3559 /* now continue from first node in postoned RE */
3560 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3563 /* logical is 1, /(?(?{...})X|Y)/ */
3564 sw = (bool)SvTRUE(ret);
3569 case EVAL_AB: /* cleanup after a successful (??{A})B */
3570 /* note: this is called twice; first after popping B, then A */
3571 PL_reg_flags ^= ST.toggle_reg_flags;
3575 cur_eval = ST.prev_eval;
3576 cur_curlyx = ST.prev_curlyx;
3577 /* XXXX This is too dramatic a measure... */
3582 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3583 /* note: this is called twice; first after popping B, then A */
3584 PL_reg_flags ^= ST.toggle_reg_flags;
3587 PL_reginput = locinput;
3588 REGCP_UNWIND(ST.lastcp);
3590 cur_eval = ST.prev_eval;
3591 cur_curlyx = ST.prev_curlyx;
3592 /* XXXX This is too dramatic a measure... */
3598 n = ARG(scan); /* which paren pair */
3599 PL_reg_start_tmp[n] = locinput;
3604 n = ARG(scan); /* which paren pair */
3605 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3606 PL_regendp[n] = locinput - PL_bostr;
3607 /*if (n > PL_regsize)
3609 if (n > (I32)*PL_reglastparen)
3610 *PL_reglastparen = n;
3611 *PL_reglastcloseparen = n;
3612 if (cur_eval && cur_eval->u.eval.close_paren == (U32)n) {
3617 n = ARG(scan); /* which paren pair */
3618 sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3621 /* reg_check_named_buff_matched returns 0 for no match */
3622 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3626 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == (U32)n));
3632 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3634 next = NEXTOPER(NEXTOPER(scan));
3636 next = scan + ARG(scan);
3637 if (OP(next) == IFTHEN) /* Fake one. */
3638 next = NEXTOPER(NEXTOPER(next));
3642 logical = scan->flags;
3645 /*******************************************************************
3647 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3648 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3649 STAR/PLUS/CURLY/CURLYN are used instead.)
3651 A*B is compiled as <CURLYX><A><WHILEM><B>
3653 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3654 state, which contains the current count, initialised to -1. It also sets
3655 cur_curlyx to point to this state, with any previous value saved in the
3658 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3659 since the pattern may possibly match zero times (i.e. it's a while {} loop
3660 rather than a do {} while loop).
3662 Each entry to WHILEM represents a successful match of A. The count in the
3663 CURLYX block is incremented, another WHILEM state is pushed, and execution
3664 passes to A or B depending on greediness and the current count.
3666 For example, if matching against the string a1a2a3b (where the aN are
3667 substrings that match /A/), then the match progresses as follows: (the
3668 pushed states are interspersed with the bits of strings matched so far):
3671 <CURLYX cnt=0><WHILEM>
3672 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3673 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3674 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3675 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3677 (Contrast this with something like CURLYM, which maintains only a single
3681 a1 <CURLYM cnt=1> a2
3682 a1 a2 <CURLYM cnt=2> a3
3683 a1 a2 a3 <CURLYM cnt=3> b
3686 Each WHILEM state block marks a point to backtrack to upon partial failure
3687 of A or B, and also contains some minor state data related to that
3688 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3689 overall state, such as the count, and pointers to the A and B ops.
3691 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3692 must always point to the *current* CURLYX block, the rules are:
3694 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3695 and set cur_curlyx to point the new block.
3697 When popping the CURLYX block after a successful or unsuccessful match,
3698 restore the previous cur_curlyx.
3700 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3701 to the outer one saved in the CURLYX block.
3703 When popping the WHILEM block after a successful or unsuccessful B match,
3704 restore the previous cur_curlyx.
3706 Here's an example for the pattern (AI* BI)*BO
3707 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3710 curlyx backtrack stack
3711 ------ ---------------
3713 CO <CO prev=NULL> <WO>
3714 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3715 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3716 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3718 At this point the pattern succeeds, and we work back down the stack to
3719 clean up, restoring as we go:
3721 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3722 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3723 CO <CO prev=NULL> <WO>
3726 *******************************************************************/
3728 #define ST st->u.curlyx
3730 case CURLYX: /* start of /A*B/ (for complex A) */
3732 /* No need to save/restore up to this paren */
3733 I32 parenfloor = scan->flags;
3735 assert(next); /* keep Coverity happy */
3736 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3739 /* XXXX Probably it is better to teach regpush to support
3740 parenfloor > PL_regsize... */
3741 if (parenfloor > (I32)*PL_reglastparen)
3742 parenfloor = *PL_reglastparen; /* Pessimization... */
3744 ST.prev_curlyx= cur_curlyx;
3746 ST.cp = PL_savestack_ix;
3748 /* these fields contain the state of the current curly.
3749 * they are accessed by subsequent WHILEMs */
3750 ST.parenfloor = parenfloor;
3751 ST.min = ARG1(scan);
3752 ST.max = ARG2(scan);
3753 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3757 ST.count = -1; /* this will be updated by WHILEM */
3758 ST.lastloc = NULL; /* this will be updated by WHILEM */
3760 PL_reginput = locinput;
3761 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
3765 case CURLYX_end: /* just finished matching all of A*B */
3767 cur_curlyx = ST.prev_curlyx;
3771 case CURLYX_end_fail: /* just failed to match all of A*B */
3773 cur_curlyx = ST.prev_curlyx;
3779 #define ST st->u.whilem
3781 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
3783 /* see the discussion above about CURLYX/WHILEM */
3785 assert(cur_curlyx); /* keep Coverity happy */
3786 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
3787 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
3788 ST.cache_offset = 0;
3791 PL_reginput = locinput;
3793 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3794 "%*s whilem: matched %ld out of %ld..%ld\n",
3795 REPORT_CODE_OFF+depth*2, "", (long)n,
3796 (long)cur_curlyx->u.curlyx.min,
3797 (long)cur_curlyx->u.curlyx.max)
3800 /* First just match a string of min A's. */
3802 if (n < cur_curlyx->u.curlyx.min) {
3803 cur_curlyx->u.curlyx.lastloc = locinput;
3804 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
3808 /* If degenerate A matches "", assume A done. */
3810 if (locinput == cur_curlyx->u.curlyx.lastloc) {
3811 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3812 "%*s whilem: empty match detected, trying continuation...\n",
3813 REPORT_CODE_OFF+depth*2, "")
3815 goto do_whilem_B_max;
3818 /* super-linear cache processing */
3822 if (!PL_reg_maxiter) {
3823 /* start the countdown: Postpone detection until we
3824 * know the match is not *that* much linear. */
3825 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3826 /* possible overflow for long strings and many CURLYX's */
3827 if (PL_reg_maxiter < 0)
3828 PL_reg_maxiter = I32_MAX;
3829 PL_reg_leftiter = PL_reg_maxiter;
3832 if (PL_reg_leftiter-- == 0) {
3833 /* initialise cache */
3834 const I32 size = (PL_reg_maxiter + 7)/8;
3835 if (PL_reg_poscache) {
3836 if ((I32)PL_reg_poscache_size < size) {
3837 Renew(PL_reg_poscache, size, char);
3838 PL_reg_poscache_size = size;
3840 Zero(PL_reg_poscache, size, char);
3843 PL_reg_poscache_size = size;
3844 Newxz(PL_reg_poscache, size, char);
3846 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3847 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
3848 PL_colors[4], PL_colors[5])
3852 if (PL_reg_leftiter < 0) {
3853 /* have we already failed at this position? */
3855 offset = (scan->flags & 0xf) - 1
3856 + (locinput - PL_bostr) * (scan->flags>>4);
3857 mask = 1 << (offset % 8);
3859 if (PL_reg_poscache[offset] & mask) {
3860 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3861 "%*s whilem: (cache) already tried at this position...\n",
3862 REPORT_CODE_OFF+depth*2, "")
3864 sayNO; /* cache records failure */
3866 ST.cache_offset = offset;
3867 ST.cache_mask = mask;
3871 /* Prefer B over A for minimal matching. */
3873 if (cur_curlyx->u.curlyx.minmod) {
3874 ST.save_curlyx = cur_curlyx;
3875 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
3876 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
3877 REGCP_SET(ST.lastcp);
3878 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
3882 /* Prefer A over B for maximal matching. */
3884 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
3885 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3886 cur_curlyx->u.curlyx.lastloc = locinput;
3887 REGCP_SET(ST.lastcp);
3888 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
3891 goto do_whilem_B_max;
3895 case WHILEM_B_min: /* just matched B in a minimal match */
3896 case WHILEM_B_max: /* just matched B in a maximal match */
3897 cur_curlyx = ST.save_curlyx;
3901 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
3902 cur_curlyx = ST.save_curlyx;
3903 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
3904 cur_curlyx->u.curlyx.count--;
3908 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
3909 REGCP_UNWIND(ST.lastcp);
3912 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
3913 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
3914 cur_curlyx->u.curlyx.count--;
3918 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
3919 REGCP_UNWIND(ST.lastcp);
3920 regcppop(rex); /* Restore some previous $<digit>s? */
3921 PL_reginput = locinput;
3922 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3923 "%*s whilem: failed, trying continuation...\n",
3924 REPORT_CODE_OFF+depth*2, "")
3927 if (cur_curlyx->u.curlyx.count >= REG_INFTY
3928 && ckWARN(WARN_REGEXP)
3929 && !(PL_reg_flags & RF_warned))
3931 PL_reg_flags |= RF_warned;
3932 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3933 "Complex regular subexpression recursion",
3938 ST.save_curlyx = cur_curlyx;
3939 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
3940 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
3943 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
3944 cur_curlyx = ST.save_curlyx;
3945 REGCP_UNWIND(ST.lastcp);
3948 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
3949 /* Maximum greed exceeded */
3950 if (cur_curlyx->u.curlyx.count >= REG_INFTY
3951 && ckWARN(WARN_REGEXP)
3952 && !(PL_reg_flags & RF_warned))
3954 PL_reg_flags |= RF_warned;
3955 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
3956 "%s limit (%d) exceeded",
3957 "Complex regular subexpression recursion",
3960 cur_curlyx->u.curlyx.count--;
3964 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3965 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
3967 /* Try grabbing another A and see if it helps. */
3968 PL_reginput = locinput;
3969 cur_curlyx->u.curlyx.lastloc = locinput;
3970 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
3971 REGCP_SET(ST.lastcp);
3972 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
3976 #define ST st->u.branch
3978 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
3979 next = scan + ARG(scan);
3982 scan = NEXTOPER(scan);
3985 case BRANCH: /* /(...|A|...)/ */
3986 scan = NEXTOPER(scan); /* scan now points to inner node */
3987 if (!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
3988 /* last branch; skip state push and jump direct to node */
3990 ST.lastparen = *PL_reglastparen;
3991 ST.next_branch = next;
3993 PL_reginput = locinput;
3995 /* Now go into the branch */
3996 PUSH_STATE_GOTO(BRANCH_next, scan);
3999 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4000 REGCP_UNWIND(ST.cp);
4001 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4003 *PL_reglastparen = n;
4004 /*dmq: *PL_reglastcloseparen = n; */
4005 scan = ST.next_branch;
4006 /* no more branches? */
4007 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ))
4009 continue; /* execute next BRANCH[J] op */
4017 #define ST st->u.curlym
4019 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4021 /* This is an optimisation of CURLYX that enables us to push
4022 * only a single backtracking state, no matter now many matches
4023 * there are in {m,n}. It relies on the pattern being constant
4024 * length, with no parens to influence future backrefs
4028 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4030 /* if paren positive, emulate an OPEN/CLOSE around A */
4032 I32 paren = ST.me->flags;
4033 if (paren > PL_regsize)
4035 if (paren > (I32)*PL_reglastparen)
4036 *PL_reglastparen = paren;
4037 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4045 ST.c1 = CHRTEST_UNINIT;
4048 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4051 curlym_do_A: /* execute the A in /A{m,n}B/ */
4052 PL_reginput = locinput;
4053 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4056 case CURLYM_A: /* we've just matched an A */
4057 locinput = st->locinput;
4058 nextchr = UCHARAT(locinput);
4061 /* after first match, determine A's length: u.curlym.alen */
4062 if (ST.count == 1) {
4063 if (PL_reg_match_utf8) {
4065 while (s < PL_reginput) {
4071 ST.alen = PL_reginput - locinput;
4074 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4077 PerlIO_printf(Perl_debug_log,
4078 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4079 (int)(REPORT_CODE_OFF+(depth*2)), "",
4080 (IV) ST.count, (IV)ST.alen)
4083 locinput = PL_reginput;
4085 if (cur_eval && cur_eval->u.eval.close_paren &&
4086 cur_eval->u.eval.close_paren == ST.me->flags)
4089 if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
4090 goto curlym_do_A; /* try to match another A */
4091 goto curlym_do_B; /* try to match B */
4093 case CURLYM_A_fail: /* just failed to match an A */
4094 REGCP_UNWIND(ST.cp);
4096 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4097 || (cur_eval && cur_eval->u.eval.close_paren &&
4098 cur_eval->u.eval.close_paren == ST.me->flags))
4101 curlym_do_B: /* execute the B in /A{m,n}B/ */
4102 PL_reginput = locinput;
4103 if (ST.c1 == CHRTEST_UNINIT) {
4104 /* calculate c1 and c2 for possible match of 1st char
4105 * following curly */
4106 ST.c1 = ST.c2 = CHRTEST_VOID;
4107 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4108 regnode *text_node = ST.B;
4109 if (! HAS_TEXT(text_node))
4110 FIND_NEXT_IMPT(text_node);
4111 if (HAS_TEXT(text_node)
4112 && PL_regkind[OP(text_node)] != REF)
4114 ST.c1 = (U8)*STRING(text_node);
4116 (OP(text_node) == EXACTF || OP(text_node) == REFF)
4118 : (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4119 ? PL_fold_locale[ST.c1]
4126 PerlIO_printf(Perl_debug_log,
4127 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4128 (int)(REPORT_CODE_OFF+(depth*2)),
4131 if (ST.c1 != CHRTEST_VOID
4132 && UCHARAT(PL_reginput) != ST.c1
4133 && UCHARAT(PL_reginput) != ST.c2)
4135 /* simulate B failing */
4136 state_num = CURLYM_B_fail;
4137 goto reenter_switch;
4141 /* mark current A as captured */
4142 I32 paren = ST.me->flags;
4145 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4146 PL_regendp[paren] = PL_reginput - PL_bostr;
4147 /*dmq: *PL_reglastcloseparen = paren; */
4150 PL_regendp[paren] = -1;
4151 if (cur_eval && cur_eval->u.eval.close_paren &&
4152 cur_eval->u.eval.close_paren == ST.me->flags)
4161 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4164 case CURLYM_B_fail: /* just failed to match a B */
4165 REGCP_UNWIND(ST.cp);
4167 if (ST.count == ARG2(ST.me) /* max */)
4169 goto curlym_do_A; /* try to match a further A */
4171 /* backtrack one A */
4172 if (ST.count == ARG1(ST.me) /* min */)
4175 locinput = HOPc(locinput, -ST.alen);
4176 goto curlym_do_B; /* try to match B */
4179 #define ST st->u.curly
4181 #define CURLY_SETPAREN(paren, success) \
4184 PL_regstartp[paren] = HOPc(locinput, -1) - PL_bostr; \
4185 PL_regendp[paren] = locinput - PL_bostr; \
4186 *PL_reglastcloseparen = paren; \
4189 PL_regendp[paren] = -1; \
4192 case STAR: /* /A*B/ where A is width 1 */
4196 scan = NEXTOPER(scan);
4198 case PLUS: /* /A+B/ where A is width 1 */
4202 scan = NEXTOPER(scan);
4204 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4205 ST.paren = scan->flags; /* Which paren to set */
4206 if (ST.paren > PL_regsize)
4207 PL_regsize = ST.paren;
4208 if (ST.paren > (I32)*PL_reglastparen)
4209 *PL_reglastparen = ST.paren;
4210 ST.min = ARG1(scan); /* min to match */
4211 ST.max = ARG2(scan); /* max to match */
4212 if (cur_eval && cur_eval->u.eval.close_paren &&
4213 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4217 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4219 case CURLY: /* /A{m,n}B/ where A is width 1 */
4221 ST.min = ARG1(scan); /* min to match */
4222 ST.max = ARG2(scan); /* max to match */
4223 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4226 * Lookahead to avoid useless match attempts
4227 * when we know what character comes next.
4229 * Used to only do .*x and .*?x, but now it allows
4230 * for )'s, ('s and (?{ ... })'s to be in the way
4231 * of the quantifier and the EXACT-like node. -- japhy
4234 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4236 if (HAS_TEXT(next) || JUMPABLE(next)) {
4238 regnode *text_node = next;
4240 if (! HAS_TEXT(text_node))
4241 FIND_NEXT_IMPT(text_node);
4243 if (! HAS_TEXT(text_node))
4244 ST.c1 = ST.c2 = CHRTEST_VOID;
4246 if (PL_regkind[OP(text_node)] == REF) {
4247 ST.c1 = ST.c2 = CHRTEST_VOID;
4248 goto assume_ok_easy;
4251 s = (U8*)STRING(text_node);
4255 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
4256 ST.c2 = PL_fold[ST.c1];
4257 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4258 ST.c2 = PL_fold_locale[ST.c1];
4261 if (OP(text_node) == EXACTF || OP(text_node) == REFF) {
4262 STRLEN ulen1, ulen2;
4263 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4264 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4266 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4267 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4269 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4271 0 : UTF8_ALLOW_ANY);
4272 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4274 0 : UTF8_ALLOW_ANY);
4276 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4278 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4283 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4290 ST.c1 = ST.c2 = CHRTEST_VOID;
4295 PL_reginput = locinput;
4298 if (ST.min && regrepeat(rex, ST.A, ST.min) < ST.min)
4301 locinput = PL_reginput;
4303 if (ST.c1 == CHRTEST_VOID)
4304 goto curly_try_B_min;
4306 ST.oldloc = locinput;
4308 /* set ST.maxpos to the furthest point along the
4309 * string that could possibly match */
4310 if (ST.max == REG_INFTY) {
4311 ST.maxpos = PL_regeol - 1;
4313 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4317 int m = ST.max - ST.min;
4318 for (ST.maxpos = locinput;
4319 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4320 ST.maxpos += UTF8SKIP(ST.maxpos);
4323 ST.maxpos = locinput + ST.max - ST.min;
4324 if (ST.maxpos >= PL_regeol)
4325 ST.maxpos = PL_regeol - 1;
4327 goto curly_try_B_min_known;
4331 ST.count = regrepeat(rex, ST.A, ST.max);
4332 locinput = PL_reginput;
4333 if (ST.count < ST.min)
4335 if ((ST.count > ST.min)
4336 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4338 /* A{m,n} must come at the end of the string, there's
4339 * no point in backing off ... */
4341 /* ...except that $ and \Z can match before *and* after
4342 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4343 We may back off by one in this case. */
4344 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4348 goto curly_try_B_max;
4353 case CURLY_B_min_known_fail:
4354 /* failed to find B in a non-greedy match where c1,c2 valid */
4355 if (ST.paren && ST.count)
4356 PL_regendp[ST.paren] = -1;
4358 PL_reginput = locinput; /* Could be reset... */
4359 REGCP_UNWIND(ST.cp);
4360 /* Couldn't or didn't -- move forward. */
4361 ST.oldloc = locinput;
4363 locinput += UTF8SKIP(locinput);
4367 curly_try_B_min_known:
4368 /* find the next place where 'B' could work, then call B */
4372 n = (ST.oldloc == locinput) ? 0 : 1;
4373 if (ST.c1 == ST.c2) {
4375 /* set n to utf8_distance(oldloc, locinput) */
4376 while (locinput <= ST.maxpos &&
4377 utf8n_to_uvchr((U8*)locinput,
4378 UTF8_MAXBYTES, &len,
4379 uniflags) != (UV)ST.c1) {
4385 /* set n to utf8_distance(oldloc, locinput) */
4386 while (locinput <= ST.maxpos) {
4388 const UV c = utf8n_to_uvchr((U8*)locinput,
4389 UTF8_MAXBYTES, &len,
4391 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4399 if (ST.c1 == ST.c2) {
4400 while (locinput <= ST.maxpos &&
4401 UCHARAT(locinput) != ST.c1)
4405 while (locinput <= ST.maxpos
4406 && UCHARAT(locinput) != ST.c1
4407 && UCHARAT(locinput) != ST.c2)
4410 n = locinput - ST.oldloc;
4412 if (locinput > ST.maxpos)
4414 /* PL_reginput == oldloc now */
4417 if (regrepeat(rex, ST.A, n) < n)
4420 PL_reginput = locinput;
4421 CURLY_SETPAREN(ST.paren, ST.count);
4422 if (cur_eval && cur_eval->u.eval.close_paren &&
4423 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4426 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4431 case CURLY_B_min_fail:
4432 /* failed to find B in a non-greedy match where c1,c2 invalid */
4433 if (ST.paren && ST.count)
4434 PL_regendp[ST.paren] = -1;
4436 REGCP_UNWIND(ST.cp);
4437 /* failed -- move forward one */
4438 PL_reginput = locinput;
4439 if (regrepeat(rex, ST.A, 1)) {
4441 locinput = PL_reginput;
4442 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4443 ST.count > 0)) /* count overflow ? */
4446 CURLY_SETPAREN(ST.paren, ST.count);
4447 if (cur_eval && cur_eval->u.eval.close_paren &&
4448 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4451 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4459 /* a successful greedy match: now try to match B */
4460 if (cur_eval && cur_eval->u.eval.close_paren &&
4461 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4466 if (ST.c1 != CHRTEST_VOID)
4467 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4468 UTF8_MAXBYTES, 0, uniflags)
4469 : (UV) UCHARAT(PL_reginput);
4470 /* If it could work, try it. */
4471 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4472 CURLY_SETPAREN(ST.paren, ST.count);
4473 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4478 case CURLY_B_max_fail:
4479 /* failed to find B in a greedy match */
4480 if (ST.paren && ST.count)
4481 PL_regendp[ST.paren] = -1;
4483 REGCP_UNWIND(ST.cp);
4485 if (--ST.count < ST.min)
4487 PL_reginput = locinput = HOPc(locinput, -1);
4488 goto curly_try_B_max;
4495 /* we've just finished A in /(??{A})B/; now continue with B */
4499 st->u.eval.toggle_reg_flags
4500 = cur_eval->u.eval.toggle_reg_flags;
4501 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4503 st->u.eval.prev_rex = rex; /* inner */
4504 rex = cur_eval->u.eval.prev_rex; /* outer */
4505 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4507 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4508 REGCP_SET(st->u.eval.lastcp);
4509 PL_reginput = locinput;
4511 /* Restore parens of the outer rex without popping the
4513 tmpix = PL_savestack_ix;
4514 PL_savestack_ix = cur_eval->u.eval.lastcp;
4516 PL_savestack_ix = tmpix;
4518 st->u.eval.prev_eval = cur_eval;
4519 cur_eval = cur_eval->u.eval.prev_eval;
4521 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %x\n",
4522 REPORT_CODE_OFF+depth*2, "",(int)cur_eval););
4523 PUSH_YES_STATE_GOTO(EVAL_AB,
4524 st->u.eval.prev_eval->u.eval.B); /* match B */
4527 if (locinput < reginfo->till) {
4528 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4529 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4531 (long)(locinput - PL_reg_starttry),
4532 (long)(reginfo->till - PL_reg_starttry),
4534 sayNO_SILENT; /* Cannot match: too short. */
4536 PL_reginput = locinput; /* put where regtry can find it */
4537 sayYES; /* Success! */
4539 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4541 PerlIO_printf(Perl_debug_log,
4542 "%*s %ssubpattern success...%s\n",
4543 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4544 PL_reginput = locinput; /* put where regtry can find it */
4545 sayYES; /* Success! */
4548 #define ST st->u.ifmatch
4550 case SUSPEND: /* (?>A) */
4552 PL_reginput = locinput;
4555 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4557 goto ifmatch_trivial_fail_test;
4559 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4561 ifmatch_trivial_fail_test:
4563 char * const s = HOPBACKc(locinput, scan->flags);
4568 sw = 1 - (bool)ST.wanted;
4572 next = scan + ARG(scan);
4580 PL_reginput = locinput;
4584 ST.logical = logical;
4585 /* execute body of (?...A) */
4586 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4589 case IFMATCH_A_fail: /* body of (?...A) failed */
4590 ST.wanted = !ST.wanted;
4593 case IFMATCH_A: /* body of (?...A) succeeded */
4595 sw = (bool)ST.wanted;
4597 else if (!ST.wanted)
4600 if (OP(ST.me) == SUSPEND)
4601 locinput = PL_reginput;
4603 locinput = PL_reginput = st->locinput;
4604 nextchr = UCHARAT(locinput);
4606 scan = ST.me + ARG(ST.me);
4609 continue; /* execute B */
4614 next = scan + ARG(scan);
4619 PUSH_STATE_GOTO(COMMIT_next,next);
4621 case COMMIT_next_fail:
4627 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
4628 PTR2UV(scan), OP(scan));
4629 Perl_croak(aTHX_ "regexp memory corruption");
4637 /* push a state that backtracks on success */
4638 st->u.yes.prev_yes_state = yes_state;
4642 /* push a new regex state, then continue at scan */
4644 regmatch_state *newst;
4646 DEBUG_STATE_pp("push");
4648 st->locinput = locinput;
4650 if (newst > SLAB_LAST(PL_regmatch_slab))
4651 newst = S_push_slab(aTHX);
4652 PL_regmatch_state = newst;
4654 locinput = PL_reginput;
4655 nextchr = UCHARAT(locinput);
4663 * We get here only if there's trouble -- normally "case END" is
4664 * the terminating point.
4666 Perl_croak(aTHX_ "corrupted regexp pointers");
4672 /* we have successfully completed a subexpression, but we must now
4673 * pop to the state marked by yes_state and continue from there */
4674 assert(st != yes_state);
4676 while (st != yes_state) {
4678 if (st < SLAB_FIRST(PL_regmatch_slab)) {
4679 PL_regmatch_slab = PL_regmatch_slab->prev;
4680 st = SLAB_LAST(PL_regmatch_slab);
4684 DEBUG_STATE_pp("pop (no final)");
4686 DEBUG_STATE_pp("pop (yes)");
4692 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
4693 || yes_state > SLAB_LAST(PL_regmatch_slab))
4695 /* not in this slab, pop slab */
4696 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
4697 PL_regmatch_slab = PL_regmatch_slab->prev;
4698 st = SLAB_LAST(PL_regmatch_slab);
4700 depth -= (st - yes_state);
4703 yes_state = st->u.yes.prev_yes_state;
4704 PL_regmatch_state = st;
4706 state_num = st->resume_state + no_final;
4707 goto reenter_switch;
4710 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
4711 PL_colors[4], PL_colors[5]));
4718 PerlIO_printf(Perl_debug_log,
4719 "%*s %sfailed...%s\n",
4720 REPORT_CODE_OFF+depth*2, "",
4721 PL_colors[4], PL_colors[5])
4733 /* there's a previous state to backtrack to */
4735 if (st < SLAB_FIRST(PL_regmatch_slab)) {
4736 PL_regmatch_slab = PL_regmatch_slab->prev;
4737 st = SLAB_LAST(PL_regmatch_slab);
4739 PL_regmatch_state = st;
4740 locinput= st->locinput;
4741 nextchr = UCHARAT(locinput);
4743 DEBUG_STATE_pp("pop");
4745 if (yes_state == st)
4746 yes_state = st->u.yes.prev_yes_state;
4748 state_num = st->resume_state + 1; /* failure = success + 1 */
4749 goto reenter_switch;
4755 /* restore original high-water mark */
4756 PL_regmatch_slab = orig_slab;
4757 PL_regmatch_state = orig_state;
4759 /* free all slabs above current one */
4760 if (orig_slab->next) {
4761 regmatch_slab *sl = orig_slab->next;
4762 orig_slab->next = NULL;
4764 regmatch_slab * const osl = sl;
4774 - regrepeat - repeatedly match something simple, report how many
4777 * [This routine now assumes that it will only match on things of length 1.
4778 * That was true before, but now we assume scan - reginput is the count,
4779 * rather than incrementing count on every character. [Er, except utf8.]]
4782 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max)
4785 register char *scan;
4787 register char *loceol = PL_regeol;
4788 register I32 hardcount = 0;
4789 register bool do_utf8 = PL_reg_match_utf8;
4792 if (max == REG_INFTY)
4794 else if (max < loceol - scan)
4795 loceol = scan + max;
4800 while (scan < loceol && hardcount < max && *scan != '\n') {
4801 scan += UTF8SKIP(scan);
4805 while (scan < loceol && *scan != '\n')
4812 while (scan < loceol && hardcount < max) {
4813 scan += UTF8SKIP(scan);
4823 case EXACT: /* length of string is 1 */
4825 while (scan < loceol && UCHARAT(scan) == c)
4828 case EXACTF: /* length of string is 1 */
4830 while (scan < loceol &&
4831 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
4834 case EXACTFL: /* length of string is 1 */
4835 PL_reg_flags |= RF_tainted;
4837 while (scan < loceol &&
4838 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
4844 while (hardcount < max && scan < loceol &&
4845 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
4846 scan += UTF8SKIP(scan);
4850 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
4857 LOAD_UTF8_CHARCLASS_ALNUM();
4858 while (hardcount < max && scan < loceol &&
4859 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4860 scan += UTF8SKIP(scan);
4864 while (scan < loceol && isALNUM(*scan))
4869 PL_reg_flags |= RF_tainted;
4872 while (hardcount < max && scan < loceol &&
4873 isALNUM_LC_utf8((U8*)scan)) {
4874 scan += UTF8SKIP(scan);
4878 while (scan < loceol && isALNUM_LC(*scan))
4885 LOAD_UTF8_CHARCLASS_ALNUM();
4886 while (hardcount < max && scan < loceol &&
4887 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4888 scan += UTF8SKIP(scan);
4892 while (scan < loceol && !isALNUM(*scan))
4897 PL_reg_flags |= RF_tainted;
4900 while (hardcount < max && scan < loceol &&
4901 !isALNUM_LC_utf8((U8*)scan)) {
4902 scan += UTF8SKIP(scan);
4906 while (scan < loceol && !isALNUM_LC(*scan))
4913 LOAD_UTF8_CHARCLASS_SPACE();
4914 while (hardcount < max && scan < loceol &&
4916 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4917 scan += UTF8SKIP(scan);
4921 while (scan < loceol && isSPACE(*scan))
4926 PL_reg_flags |= RF_tainted;
4929 while (hardcount < max && scan < loceol &&
4930 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4931 scan += UTF8SKIP(scan);
4935 while (scan < loceol && isSPACE_LC(*scan))
4942 LOAD_UTF8_CHARCLASS_SPACE();
4943 while (hardcount < max && scan < loceol &&
4945 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4946 scan += UTF8SKIP(scan);
4950 while (scan < loceol && !isSPACE(*scan))
4955 PL_reg_flags |= RF_tainted;
4958 while (hardcount < max && scan < loceol &&
4959 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4960 scan += UTF8SKIP(scan);
4964 while (scan < loceol && !isSPACE_LC(*scan))
4971 LOAD_UTF8_CHARCLASS_DIGIT();
4972 while (hardcount < max && scan < loceol &&
4973 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
4974 scan += UTF8SKIP(scan);
4978 while (scan < loceol && isDIGIT(*scan))
4985 LOAD_UTF8_CHARCLASS_DIGIT();
4986 while (hardcount < max && scan < loceol &&
4987 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
4988 scan += UTF8SKIP(scan);
4992 while (scan < loceol && !isDIGIT(*scan))
4996 default: /* Called on something of 0 width. */
4997 break; /* So match right here or not at all. */
5003 c = scan - PL_reginput;
5007 GET_RE_DEBUG_FLAGS_DECL;
5009 SV * const prop = sv_newmortal();
5010 regprop(prog, prop, p);
5011 PerlIO_printf(Perl_debug_log,
5012 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5013 REPORT_CODE_OFF+1, "", SvPVX_const(prop),(IV)c,(IV)max);
5021 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5023 - regclass_swash - prepare the utf8 swash
5027 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5033 const struct reg_data * const data = prog ? prog->data : NULL;
5035 if (data && data->count) {
5036 const U32 n = ARG(node);
5038 if (data->what[n] == 's') {
5039 SV * const rv = (SV*)data->data[n];
5040 AV * const av = (AV*)SvRV((SV*)rv);
5041 SV **const ary = AvARRAY(av);
5044 /* See the end of regcomp.c:S_regclass() for
5045 * documentation of these array elements. */
5048 a = SvROK(ary[1]) ? &ary[1] : 0;
5049 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
5053 else if (si && doinit) {
5054 sw = swash_init("utf8", "", si, 1, 0);
5055 (void)av_store(av, 1, sw);
5072 - reginclass - determine if a character falls into a character class
5074 The n is the ANYOF regnode, the p is the target string, lenp
5075 is pointer to the maximum length of how far to go in the p
5076 (if the lenp is zero, UTF8SKIP(p) is used),
5077 do_utf8 tells whether the target string is in UTF-8.
5082 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5085 const char flags = ANYOF_FLAGS(n);
5091 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5092 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5093 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5094 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5095 if (len == (STRLEN)-1)
5096 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5099 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5100 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5103 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5104 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5107 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5111 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5114 if (swash_fetch(sw, p, do_utf8))
5116 else if (flags & ANYOF_FOLD) {
5117 if (!match && lenp && av) {
5119 for (i = 0; i <= av_len(av); i++) {
5120 SV* const sv = *av_fetch(av, i, FALSE);
5122 const char * const s = SvPV_const(sv, len);
5124 if (len <= plen && memEQ(s, (char*)p, len)) {
5132 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5135 to_utf8_fold(p, tmpbuf, &tmplen);
5136 if (swash_fetch(sw, tmpbuf, do_utf8))
5142 if (match && lenp && *lenp == 0)
5143 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5145 if (!match && c < 256) {
5146 if (ANYOF_BITMAP_TEST(n, c))
5148 else if (flags & ANYOF_FOLD) {
5151 if (flags & ANYOF_LOCALE) {
5152 PL_reg_flags |= RF_tainted;
5153 f = PL_fold_locale[c];
5157 if (f != c && ANYOF_BITMAP_TEST(n, f))
5161 if (!match && (flags & ANYOF_CLASS)) {
5162 PL_reg_flags |= RF_tainted;
5164 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5165 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5166 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5167 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5168 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5169 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5170 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5171 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5172 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5173 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5174 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5175 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5176 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5177 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5178 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5179 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5180 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5181 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5182 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5183 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5184 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5185 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5186 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5187 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5188 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5189 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5190 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5191 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5192 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5193 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5194 ) /* How's that for a conditional? */
5201 return (flags & ANYOF_INVERT) ? !match : match;
5205 S_reghop3(U8 *s, I32 off, const U8* lim)
5209 while (off-- && s < lim) {
5210 /* XXX could check well-formedness here */
5215 while (off++ && s > lim) {
5217 if (UTF8_IS_CONTINUED(*s)) {
5218 while (s > lim && UTF8_IS_CONTINUATION(*s))
5221 /* XXX could check well-formedness here */
5228 /* there are a bunch of places where we use two reghop3's that should
5229 be replaced with this routine. but since thats not done yet
5230 we ifdef it out - dmq
5233 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5237 while (off-- && s < rlim) {
5238 /* XXX could check well-formedness here */
5243 while (off++ && s > llim) {
5245 if (UTF8_IS_CONTINUED(*s)) {
5246 while (s > llim && UTF8_IS_CONTINUATION(*s))
5249 /* XXX could check well-formedness here */
5257 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5261 while (off-- && s < lim) {
5262 /* XXX could check well-formedness here */
5269 while (off++ && s > lim) {
5271 if (UTF8_IS_CONTINUED(*s)) {
5272 while (s > lim && UTF8_IS_CONTINUATION(*s))
5275 /* XXX could check well-formedness here */
5284 restore_pos(pTHX_ void *arg)
5287 regexp * const rex = (regexp *)arg;
5288 if (PL_reg_eval_set) {
5289 if (PL_reg_oldsaved) {
5290 rex->subbeg = PL_reg_oldsaved;
5291 rex->sublen = PL_reg_oldsavedlen;
5292 #ifdef PERL_OLD_COPY_ON_WRITE
5293 rex->saved_copy = PL_nrs;
5295 RX_MATCH_COPIED_on(rex);
5297 PL_reg_magic->mg_len = PL_reg_oldpos;
5298 PL_reg_eval_set = 0;
5299 PL_curpm = PL_reg_oldcurpm;
5304 S_to_utf8_substr(pTHX_ register regexp *prog)
5306 if (prog->float_substr && !prog->float_utf8) {
5307 SV* const sv = newSVsv(prog->float_substr);
5308 prog->float_utf8 = sv;
5309 sv_utf8_upgrade(sv);
5310 if (SvTAIL(prog->float_substr))
5312 if (prog->float_substr == prog->check_substr)
5313 prog->check_utf8 = sv;
5315 if (prog->anchored_substr && !prog->anchored_utf8) {
5316 SV* const sv = newSVsv(prog->anchored_substr);
5317 prog->anchored_utf8 = sv;
5318 sv_utf8_upgrade(sv);
5319 if (SvTAIL(prog->anchored_substr))
5321 if (prog->anchored_substr == prog->check_substr)
5322 prog->check_utf8 = sv;
5327 S_to_byte_substr(pTHX_ register regexp *prog)
5330 if (prog->float_utf8 && !prog->float_substr) {
5331 SV* sv = newSVsv(prog->float_utf8);
5332 prog->float_substr = sv;
5333 if (sv_utf8_downgrade(sv, TRUE)) {
5334 if (SvTAIL(prog->float_utf8))
5338 prog->float_substr = sv = &PL_sv_undef;
5340 if (prog->float_utf8 == prog->check_utf8)
5341 prog->check_substr = sv;
5343 if (prog->anchored_utf8 && !prog->anchored_substr) {
5344 SV* sv = newSVsv(prog->anchored_utf8);
5345 prog->anchored_substr = sv;
5346 if (sv_utf8_downgrade(sv, TRUE)) {
5347 if (SvTAIL(prog->anchored_utf8))
5351 prog->anchored_substr = sv = &PL_sv_undef;
5353 if (prog->anchored_utf8 == prog->check_utf8)
5354 prog->check_substr = sv;
5360 * c-indentation-style: bsd
5362 * indent-tabs-mode: t
5365 * ex: set ts=8 sts=4 sw=4 noet: