5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
56 **** Alterations to Henry's code are...
58 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
59 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 by Larry Wall and others
61 **** You may distribute under the terms of either the GNU General Public
62 **** License or the Artistic License, as specified in the README file.
64 * Beware that some of this code is subtly aware of the way operator
65 * precedence is structured in regular expressions. Serious changes in
66 * regular-expression syntax might require a total rethink.
69 #define PERL_IN_REGEXEC_C
72 #ifdef PERL_IN_XSUB_RE
78 #define RF_tainted 1 /* tainted information used? */
79 #define RF_warned 2 /* warned about big count? */
81 #define RF_utf8 8 /* Pattern contains multibyte chars? */
83 #define UTF ((PL_reg_flags & RF_utf8) != 0)
85 #define RS_init 1 /* eval environment created */
86 #define RS_set 2 /* replsv value is set */
92 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
98 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
99 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
101 #define HOPc(pos,off) \
102 (char *)(PL_reg_match_utf8 \
103 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
105 #define HOPBACKc(pos, off) \
106 (char*)(PL_reg_match_utf8\
107 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
108 : (pos - off >= PL_bostr) \
112 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
113 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
115 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
116 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
117 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
118 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
119 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
120 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
122 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
124 /* for use after a quantifier and before an EXACT-like node -- japhy */
125 /* it would be nice to rework regcomp.sym to generate this stuff. sigh */
126 #define JUMPABLE(rn) ( \
128 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
130 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
131 OP(rn) == PLUS || OP(rn) == MINMOD || \
132 OP(rn) == KEEPS || (PL_regkind[OP(rn)] == VERB) || \
133 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
135 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
137 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
140 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
141 we don't need this definition. */
142 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
143 #define IS_TEXTF(rn) ( OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
144 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
147 /* ... so we use this as its faster. */
148 #define IS_TEXT(rn) ( OP(rn)==EXACT )
149 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
150 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
155 Search for mandatory following text node; for lookahead, the text must
156 follow but for lookbehind (rn->flags != 0) we skip to the next step.
158 #define FIND_NEXT_IMPT(rn) STMT_START { \
159 while (JUMPABLE(rn)) { \
160 const OPCODE type = OP(rn); \
161 if (type == SUSPEND || PL_regkind[type] == CURLY) \
162 rn = NEXTOPER(NEXTOPER(rn)); \
163 else if (type == PLUS) \
165 else if (type == IFMATCH) \
166 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
167 else rn += NEXT_OFF(rn); \
172 static void restore_pos(pTHX_ void *arg);
175 S_regcppush(pTHX_ I32 parenfloor)
178 const int retval = PL_savestack_ix;
179 #define REGCP_PAREN_ELEMS 4
180 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
182 GET_RE_DEBUG_FLAGS_DECL;
184 if (paren_elems_to_push < 0)
185 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
187 #define REGCP_OTHER_ELEMS 7
188 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
190 for (p = PL_regsize; p > parenfloor; p--) {
191 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
192 SSPUSHINT(PL_regoffs[p].end);
193 SSPUSHINT(PL_regoffs[p].start);
194 SSPUSHPTR(PL_reg_start_tmp[p]);
196 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
197 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
198 (UV)p, (IV)PL_regoffs[p].start,
199 (IV)(PL_reg_start_tmp[p] - PL_bostr),
200 (IV)PL_regoffs[p].end
203 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
204 SSPUSHPTR(PL_regoffs);
205 SSPUSHINT(PL_regsize);
206 SSPUSHINT(*PL_reglastparen);
207 SSPUSHINT(*PL_reglastcloseparen);
208 SSPUSHPTR(PL_reginput);
209 #define REGCP_FRAME_ELEMS 2
210 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
211 * are needed for the regexp context stack bookkeeping. */
212 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
213 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
218 /* These are needed since we do not localize EVAL nodes: */
219 #define REGCP_SET(cp) \
221 PerlIO_printf(Perl_debug_log, \
222 " Setting an EVAL scope, savestack=%"IVdf"\n", \
223 (IV)PL_savestack_ix)); \
226 #define REGCP_UNWIND(cp) \
228 if (cp != PL_savestack_ix) \
229 PerlIO_printf(Perl_debug_log, \
230 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
231 (IV)(cp), (IV)PL_savestack_ix)); \
235 S_regcppop(pTHX_ const regexp *rex)
241 GET_RE_DEBUG_FLAGS_DECL;
243 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
245 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
246 i = SSPOPINT; /* Parentheses elements to pop. */
247 input = (char *) SSPOPPTR;
248 *PL_reglastcloseparen = SSPOPINT;
249 *PL_reglastparen = SSPOPINT;
250 PL_regsize = SSPOPINT;
251 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
254 /* Now restore the parentheses context. */
255 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
256 i > 0; i -= REGCP_PAREN_ELEMS) {
258 U32 paren = (U32)SSPOPINT;
259 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
260 PL_regoffs[paren].start = SSPOPINT;
262 if (paren <= *PL_reglastparen)
263 PL_regoffs[paren].end = tmps;
265 PerlIO_printf(Perl_debug_log,
266 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
267 (UV)paren, (IV)PL_regoffs[paren].start,
268 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
269 (IV)PL_regoffs[paren].end,
270 (paren > *PL_reglastparen ? "(no)" : ""));
274 if (*PL_reglastparen + 1 <= rex->nparens) {
275 PerlIO_printf(Perl_debug_log,
276 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
277 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
281 /* It would seem that the similar code in regtry()
282 * already takes care of this, and in fact it is in
283 * a better location to since this code can #if 0-ed out
284 * but the code in regtry() is needed or otherwise tests
285 * requiring null fields (pat.t#187 and split.t#{13,14}
286 * (as of patchlevel 7877) will fail. Then again,
287 * this code seems to be necessary or otherwise
288 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
289 * --jhi updated by dapm */
290 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
292 PL_regoffs[i].start = -1;
293 PL_regoffs[i].end = -1;
299 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
302 * pregexec and friends
305 #ifndef PERL_IN_XSUB_RE
307 - pregexec - match a regexp against a string
310 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
311 char *strbeg, I32 minend, SV *screamer, U32 nosave)
312 /* strend: pointer to null at end of string */
313 /* strbeg: real beginning of string */
314 /* minend: end of match must be >=minend after stringarg. */
315 /* nosave: For optimizations. */
318 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
319 nosave ? 0 : REXEC_COPY_STR);
324 * Need to implement the following flags for reg_anch:
326 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
328 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
329 * INTUIT_AUTORITATIVE_ML
330 * INTUIT_ONCE_NOML - Intuit can match in one location only.
333 * Another flag for this function: SECOND_TIME (so that float substrs
334 * with giant delta may be not rechecked).
337 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
339 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
340 Otherwise, only SvCUR(sv) is used to get strbeg. */
342 /* XXXX We assume that strpos is strbeg unless sv. */
344 /* XXXX Some places assume that there is a fixed substring.
345 An update may be needed if optimizer marks as "INTUITable"
346 RExen without fixed substrings. Similarly, it is assumed that
347 lengths of all the strings are no more than minlen, thus they
348 cannot come from lookahead.
349 (Or minlen should take into account lookahead.)
350 NOTE: Some of this comment is not correct. minlen does now take account
351 of lookahead/behind. Further research is required. -- demerphq
355 /* A failure to find a constant substring means that there is no need to make
356 an expensive call to REx engine, thus we celebrate a failure. Similarly,
357 finding a substring too deep into the string means that less calls to
358 regtry() should be needed.
360 REx compiler's optimizer found 4 possible hints:
361 a) Anchored substring;
363 c) Whether we are anchored (beginning-of-line or \G);
364 d) First node (of those at offset 0) which may distingush positions;
365 We use a)b)d) and multiline-part of c), and try to find a position in the
366 string which does not contradict any of them.
369 /* Most of decisions we do here should have been done at compile time.
370 The nodes of the REx which we used for the search should have been
371 deleted from the finite automaton. */
374 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
375 char *strend, const U32 flags, re_scream_pos_data *data)
378 struct regexp *const prog = (struct regexp *)SvANY(rx);
379 register I32 start_shift = 0;
380 /* Should be nonnegative! */
381 register I32 end_shift = 0;
386 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
388 register char *other_last = NULL; /* other substr checked before this */
389 char *check_at = NULL; /* check substr found at this pos */
390 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
391 RXi_GET_DECL(prog,progi);
393 const char * const i_strpos = strpos;
396 GET_RE_DEBUG_FLAGS_DECL;
398 RX_MATCH_UTF8_set(rx,do_utf8);
401 PL_reg_flags |= RF_utf8;
404 debug_start_match(rx, do_utf8, strpos, strend,
405 sv ? "Guessing start of match in sv for"
406 : "Guessing start of match in string for");
409 /* CHR_DIST() would be more correct here but it makes things slow. */
410 if (prog->minlen > strend - strpos) {
411 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
412 "String too short... [re_intuit_start]\n"));
416 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
419 if (!prog->check_utf8 && prog->check_substr)
420 to_utf8_substr(prog);
421 check = prog->check_utf8;
423 if (!prog->check_substr && prog->check_utf8)
424 to_byte_substr(prog);
425 check = prog->check_substr;
427 if (check == &PL_sv_undef) {
428 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
429 "Non-utf8 string cannot match utf8 check string\n"));
432 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
433 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
434 || ( (prog->extflags & RXf_ANCH_BOL)
435 && !multiline ) ); /* Check after \n? */
438 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
439 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
440 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
442 && (strpos != strbeg)) {
443 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
446 if (prog->check_offset_min == prog->check_offset_max &&
447 !(prog->extflags & RXf_CANY_SEEN)) {
448 /* Substring at constant offset from beg-of-str... */
451 s = HOP3c(strpos, prog->check_offset_min, strend);
454 slen = SvCUR(check); /* >= 1 */
456 if ( strend - s > slen || strend - s < slen - 1
457 || (strend - s == slen && strend[-1] != '\n')) {
458 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
461 /* Now should match s[0..slen-2] */
463 if (slen && (*SvPVX_const(check) != *s
465 && memNE(SvPVX_const(check), s, slen)))) {
467 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
471 else if (*SvPVX_const(check) != *s
472 || ((slen = SvCUR(check)) > 1
473 && memNE(SvPVX_const(check), s, slen)))
476 goto success_at_start;
479 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
481 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
482 end_shift = prog->check_end_shift;
485 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
486 - (SvTAIL(check) != 0);
487 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
489 if (end_shift < eshift)
493 else { /* Can match at random position */
496 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
497 end_shift = prog->check_end_shift;
499 /* end shift should be non negative here */
502 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
504 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
505 (IV)end_shift, RX_PRECOMP(prog));
509 /* Find a possible match in the region s..strend by looking for
510 the "check" substring in the region corrected by start/end_shift. */
513 I32 srch_start_shift = start_shift;
514 I32 srch_end_shift = end_shift;
515 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
516 srch_end_shift -= ((strbeg - s) - srch_start_shift);
517 srch_start_shift = strbeg - s;
519 DEBUG_OPTIMISE_MORE_r({
520 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
521 (IV)prog->check_offset_min,
522 (IV)srch_start_shift,
524 (IV)prog->check_end_shift);
527 if (flags & REXEC_SCREAM) {
528 I32 p = -1; /* Internal iterator of scream. */
529 I32 * const pp = data ? data->scream_pos : &p;
531 if (PL_screamfirst[BmRARE(check)] >= 0
532 || ( BmRARE(check) == '\n'
533 && (BmPREVIOUS(check) == SvCUR(check) - 1)
535 s = screaminstr(sv, check,
536 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
539 /* we may be pointing at the wrong string */
540 if (s && RXp_MATCH_COPIED(prog))
541 s = strbeg + (s - SvPVX_const(sv));
543 *data->scream_olds = s;
548 if (prog->extflags & RXf_CANY_SEEN) {
549 start_point= (U8*)(s + srch_start_shift);
550 end_point= (U8*)(strend - srch_end_shift);
552 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
553 end_point= HOP3(strend, -srch_end_shift, strbeg);
555 DEBUG_OPTIMISE_MORE_r({
556 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
557 (int)(end_point - start_point),
558 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
562 s = fbm_instr( start_point, end_point,
563 check, multiline ? FBMrf_MULTILINE : 0);
566 /* Update the count-of-usability, remove useless subpatterns,
570 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
571 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
572 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
573 (s ? "Found" : "Did not find"),
574 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
575 ? "anchored" : "floating"),
578 (s ? " at offset " : "...\n") );
583 /* Finish the diagnostic message */
584 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
586 /* XXX dmq: first branch is for positive lookbehind...
587 Our check string is offset from the beginning of the pattern.
588 So we need to do any stclass tests offset forward from that
597 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
598 Start with the other substr.
599 XXXX no SCREAM optimization yet - and a very coarse implementation
600 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
601 *always* match. Probably should be marked during compile...
602 Probably it is right to do no SCREAM here...
605 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
606 : (prog->float_substr && prog->anchored_substr))
608 /* Take into account the "other" substring. */
609 /* XXXX May be hopelessly wrong for UTF... */
612 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
615 char * const last = HOP3c(s, -start_shift, strbeg);
617 char * const saved_s = s;
620 t = s - prog->check_offset_max;
621 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
623 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
628 t = HOP3c(t, prog->anchored_offset, strend);
629 if (t < other_last) /* These positions already checked */
631 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
634 /* XXXX It is not documented what units *_offsets are in.
635 We assume bytes, but this is clearly wrong.
636 Meaning this code needs to be carefully reviewed for errors.
640 /* On end-of-str: see comment below. */
641 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
642 if (must == &PL_sv_undef) {
644 DEBUG_r(must = prog->anchored_utf8); /* for debug */
649 HOP3(HOP3(last1, prog->anchored_offset, strend)
650 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
652 multiline ? FBMrf_MULTILINE : 0
655 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
656 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
657 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
658 (s ? "Found" : "Contradicts"),
659 quoted, RE_SV_TAIL(must));
664 if (last1 >= last2) {
665 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
666 ", giving up...\n"));
669 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
670 ", trying floating at offset %ld...\n",
671 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
672 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
673 s = HOP3c(last, 1, strend);
677 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
678 (long)(s - i_strpos)));
679 t = HOP3c(s, -prog->anchored_offset, strbeg);
680 other_last = HOP3c(s, 1, strend);
688 else { /* Take into account the floating substring. */
690 char * const saved_s = s;
693 t = HOP3c(s, -start_shift, strbeg);
695 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
696 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
697 last = HOP3c(t, prog->float_max_offset, strend);
698 s = HOP3c(t, prog->float_min_offset, strend);
701 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
702 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
703 /* fbm_instr() takes into account exact value of end-of-str
704 if the check is SvTAIL(ed). Since false positives are OK,
705 and end-of-str is not later than strend we are OK. */
706 if (must == &PL_sv_undef) {
708 DEBUG_r(must = prog->float_utf8); /* for debug message */
711 s = fbm_instr((unsigned char*)s,
712 (unsigned char*)last + SvCUR(must)
714 must, multiline ? FBMrf_MULTILINE : 0);
716 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
717 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
718 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
719 (s ? "Found" : "Contradicts"),
720 quoted, RE_SV_TAIL(must));
724 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
725 ", giving up...\n"));
728 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
729 ", trying anchored starting at offset %ld...\n",
730 (long)(saved_s + 1 - i_strpos)));
732 s = HOP3c(t, 1, strend);
736 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
737 (long)(s - i_strpos)));
738 other_last = s; /* Fix this later. --Hugo */
748 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
750 DEBUG_OPTIMISE_MORE_r(
751 PerlIO_printf(Perl_debug_log,
752 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
753 (IV)prog->check_offset_min,
754 (IV)prog->check_offset_max,
762 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
764 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
767 /* Fixed substring is found far enough so that the match
768 cannot start at strpos. */
770 if (ml_anch && t[-1] != '\n') {
771 /* Eventually fbm_*() should handle this, but often
772 anchored_offset is not 0, so this check will not be wasted. */
773 /* XXXX In the code below we prefer to look for "^" even in
774 presence of anchored substrings. And we search even
775 beyond the found float position. These pessimizations
776 are historical artefacts only. */
778 while (t < strend - prog->minlen) {
780 if (t < check_at - prog->check_offset_min) {
781 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
782 /* Since we moved from the found position,
783 we definitely contradict the found anchored
784 substr. Due to the above check we do not
785 contradict "check" substr.
786 Thus we can arrive here only if check substr
787 is float. Redo checking for "other"=="fixed".
790 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
791 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
792 goto do_other_anchored;
794 /* We don't contradict the found floating substring. */
795 /* XXXX Why not check for STCLASS? */
797 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
798 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
801 /* Position contradicts check-string */
802 /* XXXX probably better to look for check-string
803 than for "\n", so one should lower the limit for t? */
804 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
805 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
806 other_last = strpos = s = t + 1;
811 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
812 PL_colors[0], PL_colors[1]));
816 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
817 PL_colors[0], PL_colors[1]));
821 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
824 /* The found string does not prohibit matching at strpos,
825 - no optimization of calling REx engine can be performed,
826 unless it was an MBOL and we are not after MBOL,
827 or a future STCLASS check will fail this. */
829 /* Even in this situation we may use MBOL flag if strpos is offset
830 wrt the start of the string. */
831 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
832 && (strpos != strbeg) && strpos[-1] != '\n'
833 /* May be due to an implicit anchor of m{.*foo} */
834 && !(prog->intflags & PREGf_IMPLICIT))
839 DEBUG_EXECUTE_r( if (ml_anch)
840 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
841 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
844 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
846 prog->check_utf8 /* Could be deleted already */
847 && --BmUSEFUL(prog->check_utf8) < 0
848 && (prog->check_utf8 == prog->float_utf8)
850 prog->check_substr /* Could be deleted already */
851 && --BmUSEFUL(prog->check_substr) < 0
852 && (prog->check_substr == prog->float_substr)
855 /* If flags & SOMETHING - do not do it many times on the same match */
856 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
857 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
858 if (do_utf8 ? prog->check_substr : prog->check_utf8)
859 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
860 prog->check_substr = prog->check_utf8 = NULL; /* disable */
861 prog->float_substr = prog->float_utf8 = NULL; /* clear */
862 check = NULL; /* abort */
864 /* XXXX This is a remnant of the old implementation. It
865 looks wasteful, since now INTUIT can use many
867 prog->extflags &= ~RXf_USE_INTUIT;
874 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
875 /* trie stclasses are too expensive to use here, we are better off to
876 leave it to regmatch itself */
877 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
878 /* minlen == 0 is possible if regstclass is \b or \B,
879 and the fixed substr is ''$.
880 Since minlen is already taken into account, s+1 is before strend;
881 accidentally, minlen >= 1 guaranties no false positives at s + 1
882 even for \b or \B. But (minlen? 1 : 0) below assumes that
883 regstclass does not come from lookahead... */
884 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
885 This leaves EXACTF only, which is dealt with in find_byclass(). */
886 const U8* const str = (U8*)STRING(progi->regstclass);
887 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
888 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
891 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
892 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
893 else if (prog->float_substr || prog->float_utf8)
894 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
898 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
899 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
902 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
905 const char *what = NULL;
907 if (endpos == strend) {
908 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
909 "Could not match STCLASS...\n") );
912 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
913 "This position contradicts STCLASS...\n") );
914 if ((prog->extflags & RXf_ANCH) && !ml_anch)
916 /* Contradict one of substrings */
917 if (prog->anchored_substr || prog->anchored_utf8) {
918 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
919 DEBUG_EXECUTE_r( what = "anchored" );
921 s = HOP3c(t, 1, strend);
922 if (s + start_shift + end_shift > strend) {
923 /* XXXX Should be taken into account earlier? */
924 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
925 "Could not match STCLASS...\n") );
930 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
931 "Looking for %s substr starting at offset %ld...\n",
932 what, (long)(s + start_shift - i_strpos)) );
935 /* Have both, check_string is floating */
936 if (t + start_shift >= check_at) /* Contradicts floating=check */
937 goto retry_floating_check;
938 /* Recheck anchored substring, but not floating... */
942 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
943 "Looking for anchored substr starting at offset %ld...\n",
944 (long)(other_last - i_strpos)) );
945 goto do_other_anchored;
947 /* Another way we could have checked stclass at the
948 current position only: */
953 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
954 "Looking for /%s^%s/m starting at offset %ld...\n",
955 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
958 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
960 /* Check is floating subtring. */
961 retry_floating_check:
962 t = check_at - start_shift;
963 DEBUG_EXECUTE_r( what = "floating" );
964 goto hop_and_restart;
967 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
968 "By STCLASS: moving %ld --> %ld\n",
969 (long)(t - i_strpos), (long)(s - i_strpos))
973 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
974 "Does not contradict STCLASS...\n");
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
980 PL_colors[4], (check ? "Guessed" : "Giving up"),
981 PL_colors[5], (long)(s - i_strpos)) );
984 fail_finish: /* Substring not found */
985 if (prog->check_substr || prog->check_utf8) /* could be removed already */
986 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
988 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
989 PL_colors[4], PL_colors[5]));
993 #define DECL_TRIE_TYPE(scan) \
994 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
995 trie_type = (scan->flags != EXACT) \
996 ? (do_utf8 ? trie_utf8_fold : (UTF ? trie_latin_utf8_fold : trie_plain)) \
997 : (do_utf8 ? trie_utf8 : trie_plain)
999 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1000 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1001 switch (trie_type) { \
1002 case trie_utf8_fold: \
1003 if ( foldlen>0 ) { \
1004 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1009 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1010 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1011 foldlen -= UNISKIP( uvc ); \
1012 uscan = foldbuf + UNISKIP( uvc ); \
1015 case trie_latin_utf8_fold: \
1016 if ( foldlen>0 ) { \
1017 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1023 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1024 foldlen -= UNISKIP( uvc ); \
1025 uscan = foldbuf + UNISKIP( uvc ); \
1029 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1037 charid = trie->charmap[ uvc ]; \
1041 if (widecharmap) { \
1042 SV** const svpp = hv_fetch(widecharmap, \
1043 (char*)&uvc, sizeof(UV), 0); \
1045 charid = (U16)SvIV(*svpp); \
1050 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1052 char *my_strend= (char *)strend; \
1055 !ibcmp_utf8(s, &my_strend, 0, do_utf8, \
1056 m, NULL, ln, (bool)UTF)) \
1057 && (!reginfo || regtry(reginfo, &s)) ) \
1060 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1061 uvchr_to_utf8(tmpbuf, c); \
1062 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1064 && (f == c1 || f == c2) \
1066 !ibcmp_utf8(s, &my_strend, 0, do_utf8,\
1067 m, NULL, ln, (bool)UTF)) \
1068 && (!reginfo || regtry(reginfo, &s)) ) \
1074 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1078 && (ln == 1 || !(OP(c) == EXACTF \
1080 : ibcmp_locale(s, m, ln))) \
1081 && (!reginfo || regtry(reginfo, &s)) ) \
1087 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1089 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1095 #define REXEC_FBC_SCAN(CoDe) \
1097 while (s < strend) { \
1103 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1104 REXEC_FBC_UTF8_SCAN( \
1106 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1115 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1118 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1127 #define REXEC_FBC_TRYIT \
1128 if ((!reginfo || regtry(reginfo, &s))) \
1131 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1133 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1136 REXEC_FBC_CLASS_SCAN(CoNd); \
1140 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1143 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1146 REXEC_FBC_CLASS_SCAN(CoNd); \
1150 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1151 PL_reg_flags |= RF_tainted; \
1153 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1156 REXEC_FBC_CLASS_SCAN(CoNd); \
1160 #define DUMP_EXEC_POS(li,s,doutf8) \
1161 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1163 /* We know what class REx starts with. Try to find this position... */
1164 /* if reginfo is NULL, its a dryrun */
1165 /* annoyingly all the vars in this routine have different names from their counterparts
1166 in regmatch. /grrr */
1169 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1170 const char *strend, regmatch_info *reginfo)
1173 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1177 register STRLEN uskip;
1181 register I32 tmp = 1; /* Scratch variable? */
1182 register const bool do_utf8 = PL_reg_match_utf8;
1183 RXi_GET_DECL(prog,progi);
1185 /* We know what class it must start with. */
1189 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1190 !UTF8_IS_INVARIANT((U8)s[0]) ?
1191 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1192 REGINCLASS(prog, c, (U8*)s));
1195 while (s < strend) {
1198 if (REGINCLASS(prog, c, (U8*)s) ||
1199 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1200 /* The assignment of 2 is intentional:
1201 * for the folded sharp s, the skip is 2. */
1202 (skip = SHARP_S_SKIP))) {
1203 if (tmp && (!reginfo || regtry(reginfo, &s)))
1216 if (tmp && (!reginfo || regtry(reginfo, &s)))
1224 ln = STR_LEN(c); /* length to match in octets/bytes */
1225 lnc = (I32) ln; /* length to match in characters */
1227 STRLEN ulen1, ulen2;
1229 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1230 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1231 /* used by commented-out code below */
1232 /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
1234 /* XXX: Since the node will be case folded at compile
1235 time this logic is a little odd, although im not
1236 sure that its actually wrong. --dmq */
1238 c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1239 c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1241 /* XXX: This is kinda strange. to_utf8_XYZ returns the
1242 codepoint of the first character in the converted
1243 form, yet originally we did the extra step.
1244 No tests fail by commenting this code out however
1245 so Ive left it out. -- dmq.
1247 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1249 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1254 while (sm < ((U8 *) m + ln)) {
1269 c2 = PL_fold_locale[c1];
1271 e = HOP3c(strend, -((I32)lnc), s);
1273 if (!reginfo && e < s)
1274 e = s; /* Due to minlen logic of intuit() */
1276 /* The idea in the EXACTF* cases is to first find the
1277 * first character of the EXACTF* node and then, if
1278 * necessary, case-insensitively compare the full
1279 * text of the node. The c1 and c2 are the first
1280 * characters (though in Unicode it gets a bit
1281 * more complicated because there are more cases
1282 * than just upper and lower: one needs to use
1283 * the so-called folding case for case-insensitive
1284 * matching (called "loose matching" in Unicode).
1285 * ibcmp_utf8() will do just that. */
1287 if (do_utf8 || UTF) {
1289 U8 tmpbuf [UTF8_MAXBYTES+1];
1292 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1294 /* Upper and lower of 1st char are equal -
1295 * probably not a "letter". */
1298 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1303 REXEC_FBC_EXACTISH_CHECK(c == c1);
1309 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1315 /* Handle some of the three Greek sigmas cases.
1316 * Note that not all the possible combinations
1317 * are handled here: some of them are handled
1318 * by the standard folding rules, and some of
1319 * them (the character class or ANYOF cases)
1320 * are handled during compiletime in
1321 * regexec.c:S_regclass(). */
1322 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1323 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1324 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1326 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1331 /* Neither pattern nor string are UTF8 */
1333 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1335 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1339 PL_reg_flags |= RF_tainted;
1346 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1347 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1349 tmp = ((OP(c) == BOUND ?
1350 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1351 LOAD_UTF8_CHARCLASS_ALNUM();
1352 REXEC_FBC_UTF8_SCAN(
1353 if (tmp == !(OP(c) == BOUND ?
1354 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1355 isALNUM_LC_utf8((U8*)s)))
1363 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1364 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1367 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1373 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1377 PL_reg_flags |= RF_tainted;
1384 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1385 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1387 tmp = ((OP(c) == NBOUND ?
1388 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1389 LOAD_UTF8_CHARCLASS_ALNUM();
1390 REXEC_FBC_UTF8_SCAN(
1391 if (tmp == !(OP(c) == NBOUND ?
1392 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1393 isALNUM_LC_utf8((U8*)s)))
1395 else REXEC_FBC_TRYIT;
1399 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1400 tmp = ((OP(c) == NBOUND ?
1401 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1404 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1406 else REXEC_FBC_TRYIT;
1409 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1413 REXEC_FBC_CSCAN_PRELOAD(
1414 LOAD_UTF8_CHARCLASS_ALNUM(),
1415 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1419 REXEC_FBC_CSCAN_TAINT(
1420 isALNUM_LC_utf8((U8*)s),
1424 REXEC_FBC_CSCAN_PRELOAD(
1425 LOAD_UTF8_CHARCLASS_ALNUM(),
1426 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1430 REXEC_FBC_CSCAN_TAINT(
1431 !isALNUM_LC_utf8((U8*)s),
1435 REXEC_FBC_CSCAN_PRELOAD(
1436 LOAD_UTF8_CHARCLASS_SPACE(),
1437 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1441 REXEC_FBC_CSCAN_TAINT(
1442 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1446 REXEC_FBC_CSCAN_PRELOAD(
1447 LOAD_UTF8_CHARCLASS_SPACE(),
1448 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1452 REXEC_FBC_CSCAN_TAINT(
1453 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1457 REXEC_FBC_CSCAN_PRELOAD(
1458 LOAD_UTF8_CHARCLASS_DIGIT(),
1459 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1463 REXEC_FBC_CSCAN_TAINT(
1464 isDIGIT_LC_utf8((U8*)s),
1468 REXEC_FBC_CSCAN_PRELOAD(
1469 LOAD_UTF8_CHARCLASS_DIGIT(),
1470 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1474 REXEC_FBC_CSCAN_TAINT(
1475 !isDIGIT_LC_utf8((U8*)s),
1481 is_LNBREAK_latin1(s)
1491 !is_VERTWS_latin1(s)
1496 is_HORIZWS_latin1(s)
1500 !is_HORIZWS_utf8(s),
1501 !is_HORIZWS_latin1(s)
1507 /* what trie are we using right now */
1509 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1511 = (reg_trie_data*)progi->data->data[ aho->trie ];
1512 HV *widecharmap = (HV*) progi->data->data[ aho->trie + 1 ];
1514 const char *last_start = strend - trie->minlen;
1516 const char *real_start = s;
1518 STRLEN maxlen = trie->maxlen;
1520 U8 **points; /* map of where we were in the input string
1521 when reading a given char. For ASCII this
1522 is unnecessary overhead as the relationship
1523 is always 1:1, but for Unicode, especially
1524 case folded Unicode this is not true. */
1525 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1529 GET_RE_DEBUG_FLAGS_DECL;
1531 /* We can't just allocate points here. We need to wrap it in
1532 * an SV so it gets freed properly if there is a croak while
1533 * running the match */
1536 sv_points=newSV(maxlen * sizeof(U8 *));
1537 SvCUR_set(sv_points,
1538 maxlen * sizeof(U8 *));
1539 SvPOK_on(sv_points);
1540 sv_2mortal(sv_points);
1541 points=(U8**)SvPV_nolen(sv_points );
1542 if ( trie_type != trie_utf8_fold
1543 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1546 bitmap=(U8*)trie->bitmap;
1548 bitmap=(U8*)ANYOF_BITMAP(c);
1550 /* this is the Aho-Corasick algorithm modified a touch
1551 to include special handling for long "unknown char"
1552 sequences. The basic idea being that we use AC as long
1553 as we are dealing with a possible matching char, when
1554 we encounter an unknown char (and we have not encountered
1555 an accepting state) we scan forward until we find a legal
1557 AC matching is basically that of trie matching, except
1558 that when we encounter a failing transition, we fall back
1559 to the current states "fail state", and try the current char
1560 again, a process we repeat until we reach the root state,
1561 state 1, or a legal transition. If we fail on the root state
1562 then we can either terminate if we have reached an accepting
1563 state previously, or restart the entire process from the beginning
1567 while (s <= last_start) {
1568 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1576 U8 *uscan = (U8*)NULL;
1577 U8 *leftmost = NULL;
1579 U32 accepted_word= 0;
1583 while ( state && uc <= (U8*)strend ) {
1585 U32 word = aho->states[ state ].wordnum;
1589 DEBUG_TRIE_EXECUTE_r(
1590 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1591 dump_exec_pos( (char *)uc, c, strend, real_start,
1592 (char *)uc, do_utf8 );
1593 PerlIO_printf( Perl_debug_log,
1594 " Scanning for legal start char...\n");
1597 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1602 if (uc >(U8*)last_start) break;
1606 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1607 if (!leftmost || lpos < leftmost) {
1608 DEBUG_r(accepted_word=word);
1614 points[pointpos++ % maxlen]= uc;
1615 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1616 uscan, len, uvc, charid, foldlen,
1618 DEBUG_TRIE_EXECUTE_r({
1619 dump_exec_pos( (char *)uc, c, strend, real_start,
1621 PerlIO_printf(Perl_debug_log,
1622 " Charid:%3u CP:%4"UVxf" ",
1628 word = aho->states[ state ].wordnum;
1630 base = aho->states[ state ].trans.base;
1632 DEBUG_TRIE_EXECUTE_r({
1634 dump_exec_pos( (char *)uc, c, strend, real_start,
1636 PerlIO_printf( Perl_debug_log,
1637 "%sState: %4"UVxf", word=%"UVxf,
1638 failed ? " Fail transition to " : "",
1639 (UV)state, (UV)word);
1644 (base + charid > trie->uniquecharcount )
1645 && (base + charid - 1 - trie->uniquecharcount
1647 && trie->trans[base + charid - 1 -
1648 trie->uniquecharcount].check == state
1649 && (tmp=trie->trans[base + charid - 1 -
1650 trie->uniquecharcount ].next))
1652 DEBUG_TRIE_EXECUTE_r(
1653 PerlIO_printf( Perl_debug_log," - legal\n"));
1658 DEBUG_TRIE_EXECUTE_r(
1659 PerlIO_printf( Perl_debug_log," - fail\n"));
1661 state = aho->fail[state];
1665 /* we must be accepting here */
1666 DEBUG_TRIE_EXECUTE_r(
1667 PerlIO_printf( Perl_debug_log," - accepting\n"));
1676 if (!state) state = 1;
1679 if ( aho->states[ state ].wordnum ) {
1680 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1681 if (!leftmost || lpos < leftmost) {
1682 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1687 s = (char*)leftmost;
1688 DEBUG_TRIE_EXECUTE_r({
1690 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1691 (UV)accepted_word, (IV)(s - real_start)
1694 if (!reginfo || regtry(reginfo, &s)) {
1700 DEBUG_TRIE_EXECUTE_r({
1701 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1704 DEBUG_TRIE_EXECUTE_r(
1705 PerlIO_printf( Perl_debug_log,"No match.\n"));
1714 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1723 S_swap_match_buff (pTHX_ regexp *prog) {
1724 regexp_paren_pair *t;
1727 /* We have to be careful. If the previous successful match
1728 was from this regex we don't want a subsequent paritally
1729 successful match to clobber the old results.
1730 So when we detect this possibility we add a swap buffer
1731 to the re, and switch the buffer each match. If we fail
1732 we switch it back, otherwise we leave it swapped.
1734 Newxz(prog->swap, (prog->nparens + 1), regexp_paren_pair);
1737 prog->swap = prog->offs;
1743 - regexec_flags - match a regexp against a string
1746 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1747 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1748 /* strend: pointer to null at end of string */
1749 /* strbeg: real beginning of string */
1750 /* minend: end of match must be >=minend after stringarg. */
1751 /* data: May be used for some additional optimizations.
1752 Currently its only used, with a U32 cast, for transmitting
1753 the ganch offset when doing a /g match. This will change */
1754 /* nosave: For optimizations. */
1757 struct regexp *const prog = (struct regexp *)SvANY(rx);
1758 /*register*/ char *s;
1759 register regnode *c;
1760 /*register*/ char *startpos = stringarg;
1761 I32 minlen; /* must match at least this many chars */
1762 I32 dontbother = 0; /* how many characters not to try at end */
1763 I32 end_shift = 0; /* Same for the end. */ /* CC */
1764 I32 scream_pos = -1; /* Internal iterator of scream. */
1765 char *scream_olds = NULL;
1766 const bool do_utf8 = (bool)DO_UTF8(sv);
1768 RXi_GET_DECL(prog,progi);
1769 regmatch_info reginfo; /* create some info to pass to regtry etc */
1770 bool swap_on_fail = 0;
1772 GET_RE_DEBUG_FLAGS_DECL;
1774 PERL_UNUSED_ARG(data);
1776 /* Be paranoid... */
1777 if (prog == NULL || startpos == NULL) {
1778 Perl_croak(aTHX_ "NULL regexp parameter");
1782 multiline = prog->extflags & RXf_PMf_MULTILINE;
1783 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
1785 RX_MATCH_UTF8_set(rx, do_utf8);
1787 debug_start_match(rx, do_utf8, startpos, strend,
1791 minlen = prog->minlen;
1793 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1794 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1795 "String too short [regexec_flags]...\n"));
1800 /* Check validity of program. */
1801 if (UCHARAT(progi->program) != REG_MAGIC) {
1802 Perl_croak(aTHX_ "corrupted regexp program");
1806 PL_reg_eval_set = 0;
1810 PL_reg_flags |= RF_utf8;
1812 /* Mark beginning of line for ^ and lookbehind. */
1813 reginfo.bol = startpos; /* XXX not used ??? */
1817 /* Mark end of line for $ (and such) */
1820 /* see how far we have to get to not match where we matched before */
1821 reginfo.till = startpos+minend;
1823 /* If there is a "must appear" string, look for it. */
1826 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1829 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1830 reginfo.ganch = startpos + prog->gofs;
1831 else if (sv && SvTYPE(sv) >= SVt_PVMG
1833 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1834 && mg->mg_len >= 0) {
1835 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1836 if (prog->extflags & RXf_ANCH_GPOS) {
1837 if (s > reginfo.ganch)
1839 s = reginfo.ganch - prog->gofs;
1843 reginfo.ganch = strbeg + PTR2UV(data);
1844 } else /* pos() not defined */
1845 reginfo.ganch = strbeg;
1847 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
1849 swap_match_buff(prog); /* do we need a save destructor here for
1852 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1853 re_scream_pos_data d;
1855 d.scream_olds = &scream_olds;
1856 d.scream_pos = &scream_pos;
1857 s = re_intuit_start(rx, sv, s, strend, flags, &d);
1859 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1860 goto phooey; /* not present */
1866 /* Simplest case: anchored match need be tried only once. */
1867 /* [unless only anchor is BOL and multiline is set] */
1868 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
1869 if (s == startpos && regtry(®info, &startpos))
1871 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
1872 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
1877 dontbother = minlen - 1;
1878 end = HOP3c(strend, -dontbother, strbeg) - 1;
1879 /* for multiline we only have to try after newlines */
1880 if (prog->check_substr || prog->check_utf8) {
1884 if (regtry(®info, &s))
1889 if (prog->extflags & RXf_USE_INTUIT) {
1890 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
1901 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1902 if (regtry(®info, &s))
1909 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
1911 /* the warning about reginfo.ganch being used without intialization
1912 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
1913 and we only enter this block when the same bit is set. */
1914 char *tmp_s = reginfo.ganch - prog->gofs;
1915 if (regtry(®info, &tmp_s))
1920 /* Messy cases: unanchored match. */
1921 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
1922 /* we have /x+whatever/ */
1923 /* it must be a one character string (XXXX Except UTF?) */
1928 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1929 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1930 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1935 DEBUG_EXECUTE_r( did_match = 1 );
1936 if (regtry(®info, &s)) goto got_it;
1938 while (s < strend && *s == ch)
1946 DEBUG_EXECUTE_r( did_match = 1 );
1947 if (regtry(®info, &s)) goto got_it;
1949 while (s < strend && *s == ch)
1954 DEBUG_EXECUTE_r(if (!did_match)
1955 PerlIO_printf(Perl_debug_log,
1956 "Did not find anchored character...\n")
1959 else if (prog->anchored_substr != NULL
1960 || prog->anchored_utf8 != NULL
1961 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1962 && prog->float_max_offset < strend - s)) {
1967 char *last1; /* Last position checked before */
1971 if (prog->anchored_substr || prog->anchored_utf8) {
1972 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1973 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1974 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1975 back_max = back_min = prog->anchored_offset;
1977 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1978 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1979 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1980 back_max = prog->float_max_offset;
1981 back_min = prog->float_min_offset;
1985 if (must == &PL_sv_undef)
1986 /* could not downgrade utf8 check substring, so must fail */
1992 last = HOP3c(strend, /* Cannot start after this */
1993 -(I32)(CHR_SVLEN(must)
1994 - (SvTAIL(must) != 0) + back_min), strbeg);
1997 last1 = HOPc(s, -1);
1999 last1 = s - 1; /* bogus */
2001 /* XXXX check_substr already used to find "s", can optimize if
2002 check_substr==must. */
2004 dontbother = end_shift;
2005 strend = HOPc(strend, -dontbother);
2006 while ( (s <= last) &&
2007 ((flags & REXEC_SCREAM)
2008 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2009 end_shift, &scream_pos, 0))
2010 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2011 (unsigned char*)strend, must,
2012 multiline ? FBMrf_MULTILINE : 0))) ) {
2013 /* we may be pointing at the wrong string */
2014 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2015 s = strbeg + (s - SvPVX_const(sv));
2016 DEBUG_EXECUTE_r( did_match = 1 );
2017 if (HOPc(s, -back_max) > last1) {
2018 last1 = HOPc(s, -back_min);
2019 s = HOPc(s, -back_max);
2022 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2024 last1 = HOPc(s, -back_min);
2028 while (s <= last1) {
2029 if (regtry(®info, &s))
2035 while (s <= last1) {
2036 if (regtry(®info, &s))
2042 DEBUG_EXECUTE_r(if (!did_match) {
2043 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
2044 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2045 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2046 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2047 ? "anchored" : "floating"),
2048 quoted, RE_SV_TAIL(must));
2052 else if ( (c = progi->regstclass) ) {
2054 const OPCODE op = OP(progi->regstclass);
2055 /* don't bother with what can't match */
2056 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2057 strend = HOPc(strend, -(minlen - 1));
2060 SV * const prop = sv_newmortal();
2061 regprop(prog, prop, c);
2063 RE_PV_QUOTED_DECL(quoted,do_utf8,PERL_DEBUG_PAD_ZERO(1),
2065 PerlIO_printf(Perl_debug_log,
2066 "Matching stclass %.*s against %s (%d chars)\n",
2067 (int)SvCUR(prop), SvPVX_const(prop),
2068 quoted, (int)(strend - s));
2071 if (find_byclass(prog, c, s, strend, ®info))
2073 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2077 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2082 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2083 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2084 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
2086 if (flags & REXEC_SCREAM) {
2087 last = screaminstr(sv, float_real, s - strbeg,
2088 end_shift, &scream_pos, 1); /* last one */
2090 last = scream_olds; /* Only one occurrence. */
2091 /* we may be pointing at the wrong string */
2092 else if (RXp_MATCH_COPIED(prog))
2093 s = strbeg + (s - SvPVX_const(sv));
2097 const char * const little = SvPV_const(float_real, len);
2099 if (SvTAIL(float_real)) {
2100 if (memEQ(strend - len + 1, little, len - 1))
2101 last = strend - len + 1;
2102 else if (!multiline)
2103 last = memEQ(strend - len, little, len)
2104 ? strend - len : NULL;
2110 last = rninstr(s, strend, little, little + len);
2112 last = strend; /* matching "$" */
2117 PerlIO_printf(Perl_debug_log,
2118 "%sCan't trim the tail, match fails (should not happen)%s\n",
2119 PL_colors[4], PL_colors[5]));
2120 goto phooey; /* Should not happen! */
2122 dontbother = strend - last + prog->float_min_offset;
2124 if (minlen && (dontbother < minlen))
2125 dontbother = minlen - 1;
2126 strend -= dontbother; /* this one's always in bytes! */
2127 /* We don't know much -- general case. */
2130 if (regtry(®info, &s))
2139 if (regtry(®info, &s))
2141 } while (s++ < strend);
2149 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2151 if (PL_reg_eval_set)
2152 restore_pos(aTHX_ prog);
2153 if (RXp_PAREN_NAMES(prog))
2154 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2156 /* make sure $`, $&, $', and $digit will work later */
2157 if ( !(flags & REXEC_NOT_FIRST) ) {
2158 RX_MATCH_COPY_FREE(rx);
2159 if (flags & REXEC_COPY_STR) {
2160 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2161 #ifdef PERL_OLD_COPY_ON_WRITE
2163 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2165 PerlIO_printf(Perl_debug_log,
2166 "Copy on write: regexp capture, type %d\n",
2169 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2170 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2171 assert (SvPOKp(prog->saved_copy));
2175 RX_MATCH_COPIED_on(rx);
2176 s = savepvn(strbeg, i);
2182 prog->subbeg = strbeg;
2183 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2190 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2191 PL_colors[4], PL_colors[5]));
2192 if (PL_reg_eval_set)
2193 restore_pos(aTHX_ prog);
2195 /* we failed :-( roll it back */
2196 swap_match_buff(prog);
2203 - regtry - try match at specific point
2205 STATIC I32 /* 0 failure, 1 success */
2206 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2210 REGEXP *const rx = reginfo->prog;
2211 regexp *const prog = (struct regexp *)SvANY(rx);
2212 RXi_GET_DECL(prog,progi);
2213 GET_RE_DEBUG_FLAGS_DECL;
2214 reginfo->cutpoint=NULL;
2216 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2219 PL_reg_eval_set = RS_init;
2220 DEBUG_EXECUTE_r(DEBUG_s(
2221 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2222 (IV)(PL_stack_sp - PL_stack_base));
2225 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2226 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2228 /* Apparently this is not needed, judging by wantarray. */
2229 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2230 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2233 /* Make $_ available to executed code. */
2234 if (reginfo->sv != DEFSV) {
2236 DEFSV = reginfo->sv;
2239 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2240 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2241 /* prepare for quick setting of pos */
2242 #ifdef PERL_OLD_COPY_ON_WRITE
2243 if (SvIsCOW(reginfo->sv))
2244 sv_force_normal_flags(reginfo->sv, 0);
2246 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2247 &PL_vtbl_mglob, NULL, 0);
2251 PL_reg_oldpos = mg->mg_len;
2252 SAVEDESTRUCTOR_X(restore_pos, prog);
2254 if (!PL_reg_curpm) {
2255 Newxz(PL_reg_curpm, 1, PMOP);
2258 SV* const repointer = newSViv(0);
2259 /* so we know which PL_regex_padav element is PL_reg_curpm */
2260 SvFLAGS(repointer) |= SVf_BREAK;
2261 av_push(PL_regex_padav,repointer);
2262 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2263 PL_regex_pad = AvARRAY(PL_regex_padav);
2267 PM_SETRE(PL_reg_curpm, rx);
2268 PL_reg_oldcurpm = PL_curpm;
2269 PL_curpm = PL_reg_curpm;
2270 if (RXp_MATCH_COPIED(prog)) {
2271 /* Here is a serious problem: we cannot rewrite subbeg,
2272 since it may be needed if this match fails. Thus
2273 $` inside (?{}) could fail... */
2274 PL_reg_oldsaved = prog->subbeg;
2275 PL_reg_oldsavedlen = prog->sublen;
2276 #ifdef PERL_OLD_COPY_ON_WRITE
2277 PL_nrs = prog->saved_copy;
2279 RXp_MATCH_COPIED_off(prog);
2282 PL_reg_oldsaved = NULL;
2283 prog->subbeg = PL_bostr;
2284 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2286 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2287 prog->offs[0].start = *startpos - PL_bostr;
2288 PL_reginput = *startpos;
2289 PL_reglastparen = &prog->lastparen;
2290 PL_reglastcloseparen = &prog->lastcloseparen;
2291 prog->lastparen = 0;
2292 prog->lastcloseparen = 0;
2294 PL_regoffs = prog->offs;
2295 if (PL_reg_start_tmpl <= prog->nparens) {
2296 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2297 if(PL_reg_start_tmp)
2298 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2300 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2303 /* XXXX What this code is doing here?!!! There should be no need
2304 to do this again and again, PL_reglastparen should take care of
2307 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2308 * Actually, the code in regcppop() (which Ilya may be meaning by
2309 * PL_reglastparen), is not needed at all by the test suite
2310 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2311 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2312 * Meanwhile, this code *is* needed for the
2313 * above-mentioned test suite tests to succeed. The common theme
2314 * on those tests seems to be returning null fields from matches.
2315 * --jhi updated by dapm */
2317 if (prog->nparens) {
2318 regexp_paren_pair *pp = PL_regoffs;
2320 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2328 if (regmatch(reginfo, progi->program + 1)) {
2329 PL_regoffs[0].end = PL_reginput - PL_bostr;
2332 if (reginfo->cutpoint)
2333 *startpos= reginfo->cutpoint;
2334 REGCP_UNWIND(lastcp);
2339 #define sayYES goto yes
2340 #define sayNO goto no
2341 #define sayNO_SILENT goto no_silent
2343 /* we dont use STMT_START/END here because it leads to
2344 "unreachable code" warnings, which are bogus, but distracting. */
2345 #define CACHEsayNO \
2346 if (ST.cache_mask) \
2347 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2350 /* this is used to determine how far from the left messages like
2351 'failed...' are printed. It should be set such that messages
2352 are inline with the regop output that created them.
2354 #define REPORT_CODE_OFF 32
2357 /* Make sure there is a test for this +1 options in re_tests */
2358 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2360 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2361 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2363 #define SLAB_FIRST(s) (&(s)->states[0])
2364 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2366 /* grab a new slab and return the first slot in it */
2368 STATIC regmatch_state *
2371 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2374 regmatch_slab *s = PL_regmatch_slab->next;
2376 Newx(s, 1, regmatch_slab);
2377 s->prev = PL_regmatch_slab;
2379 PL_regmatch_slab->next = s;
2381 PL_regmatch_slab = s;
2382 return SLAB_FIRST(s);
2386 /* push a new state then goto it */
2388 #define PUSH_STATE_GOTO(state, node) \
2390 st->resume_state = state; \
2393 /* push a new state with success backtracking, then goto it */
2395 #define PUSH_YES_STATE_GOTO(state, node) \
2397 st->resume_state = state; \
2398 goto push_yes_state;
2404 regmatch() - main matching routine
2406 This is basically one big switch statement in a loop. We execute an op,
2407 set 'next' to point the next op, and continue. If we come to a point which
2408 we may need to backtrack to on failure such as (A|B|C), we push a
2409 backtrack state onto the backtrack stack. On failure, we pop the top
2410 state, and re-enter the loop at the state indicated. If there are no more
2411 states to pop, we return failure.
2413 Sometimes we also need to backtrack on success; for example /A+/, where
2414 after successfully matching one A, we need to go back and try to
2415 match another one; similarly for lookahead assertions: if the assertion
2416 completes successfully, we backtrack to the state just before the assertion
2417 and then carry on. In these cases, the pushed state is marked as
2418 'backtrack on success too'. This marking is in fact done by a chain of
2419 pointers, each pointing to the previous 'yes' state. On success, we pop to
2420 the nearest yes state, discarding any intermediate failure-only states.
2421 Sometimes a yes state is pushed just to force some cleanup code to be
2422 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2423 it to free the inner regex.
2425 Note that failure backtracking rewinds the cursor position, while
2426 success backtracking leaves it alone.
2428 A pattern is complete when the END op is executed, while a subpattern
2429 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2430 ops trigger the "pop to last yes state if any, otherwise return true"
2433 A common convention in this function is to use A and B to refer to the two
2434 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2435 the subpattern to be matched possibly multiple times, while B is the entire
2436 rest of the pattern. Variable and state names reflect this convention.
2438 The states in the main switch are the union of ops and failure/success of
2439 substates associated with with that op. For example, IFMATCH is the op
2440 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2441 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2442 successfully matched A and IFMATCH_A_fail is a state saying that we have
2443 just failed to match A. Resume states always come in pairs. The backtrack
2444 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2445 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2446 on success or failure.
2448 The struct that holds a backtracking state is actually a big union, with
2449 one variant for each major type of op. The variable st points to the
2450 top-most backtrack struct. To make the code clearer, within each
2451 block of code we #define ST to alias the relevant union.
2453 Here's a concrete example of a (vastly oversimplified) IFMATCH
2459 #define ST st->u.ifmatch
2461 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2462 ST.foo = ...; // some state we wish to save
2464 // push a yes backtrack state with a resume value of
2465 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2467 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2470 case IFMATCH_A: // we have successfully executed A; now continue with B
2472 bar = ST.foo; // do something with the preserved value
2475 case IFMATCH_A_fail: // A failed, so the assertion failed
2476 ...; // do some housekeeping, then ...
2477 sayNO; // propagate the failure
2484 For any old-timers reading this who are familiar with the old recursive
2485 approach, the code above is equivalent to:
2487 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2496 ...; // do some housekeeping, then ...
2497 sayNO; // propagate the failure
2500 The topmost backtrack state, pointed to by st, is usually free. If you
2501 want to claim it, populate any ST.foo fields in it with values you wish to
2502 save, then do one of
2504 PUSH_STATE_GOTO(resume_state, node);
2505 PUSH_YES_STATE_GOTO(resume_state, node);
2507 which sets that backtrack state's resume value to 'resume_state', pushes a
2508 new free entry to the top of the backtrack stack, then goes to 'node'.
2509 On backtracking, the free slot is popped, and the saved state becomes the
2510 new free state. An ST.foo field in this new top state can be temporarily
2511 accessed to retrieve values, but once the main loop is re-entered, it
2512 becomes available for reuse.
2514 Note that the depth of the backtrack stack constantly increases during the
2515 left-to-right execution of the pattern, rather than going up and down with
2516 the pattern nesting. For example the stack is at its maximum at Z at the
2517 end of the pattern, rather than at X in the following:
2519 /(((X)+)+)+....(Y)+....Z/
2521 The only exceptions to this are lookahead/behind assertions and the cut,
2522 (?>A), which pop all the backtrack states associated with A before
2525 Bascktrack state structs are allocated in slabs of about 4K in size.
2526 PL_regmatch_state and st always point to the currently active state,
2527 and PL_regmatch_slab points to the slab currently containing
2528 PL_regmatch_state. The first time regmatch() is called, the first slab is
2529 allocated, and is never freed until interpreter destruction. When the slab
2530 is full, a new one is allocated and chained to the end. At exit from
2531 regmatch(), slabs allocated since entry are freed.
2536 #define DEBUG_STATE_pp(pp) \
2538 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2539 PerlIO_printf(Perl_debug_log, \
2540 " %*s"pp" %s%s%s%s%s\n", \
2542 PL_reg_name[st->resume_state], \
2543 ((st==yes_state||st==mark_state) ? "[" : ""), \
2544 ((st==yes_state) ? "Y" : ""), \
2545 ((st==mark_state) ? "M" : ""), \
2546 ((st==yes_state||st==mark_state) ? "]" : "") \
2551 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2556 S_debug_start_match(pTHX_ const REGEXP *prog, const bool do_utf8,
2557 const char *start, const char *end, const char *blurb)
2559 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2563 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2564 RX_PRECOMP(prog), RX_PRELEN(prog), 60);
2566 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2567 start, end - start, 60);
2569 PerlIO_printf(Perl_debug_log,
2570 "%s%s REx%s %s against %s\n",
2571 PL_colors[4], blurb, PL_colors[5], s0, s1);
2573 if (do_utf8||utf8_pat)
2574 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2575 utf8_pat ? "pattern" : "",
2576 utf8_pat && do_utf8 ? " and " : "",
2577 do_utf8 ? "string" : ""
2583 S_dump_exec_pos(pTHX_ const char *locinput,
2584 const regnode *scan,
2585 const char *loc_regeol,
2586 const char *loc_bostr,
2587 const char *loc_reg_starttry,
2590 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2591 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2592 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2593 /* The part of the string before starttry has one color
2594 (pref0_len chars), between starttry and current
2595 position another one (pref_len - pref0_len chars),
2596 after the current position the third one.
2597 We assume that pref0_len <= pref_len, otherwise we
2598 decrease pref0_len. */
2599 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2600 ? (5 + taill) - l : locinput - loc_bostr;
2603 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2605 pref0_len = pref_len - (locinput - loc_reg_starttry);
2606 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2607 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2608 ? (5 + taill) - pref_len : loc_regeol - locinput);
2609 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2613 if (pref0_len > pref_len)
2614 pref0_len = pref_len;
2616 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2618 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2619 (locinput - pref_len),pref0_len, 60, 4, 5);
2621 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2622 (locinput - pref_len + pref0_len),
2623 pref_len - pref0_len, 60, 2, 3);
2625 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2626 locinput, loc_regeol - locinput, 10, 0, 1);
2628 const STRLEN tlen=len0+len1+len2;
2629 PerlIO_printf(Perl_debug_log,
2630 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2631 (IV)(locinput - loc_bostr),
2634 (docolor ? "" : "> <"),
2636 (int)(tlen > 19 ? 0 : 19 - tlen),
2643 /* reg_check_named_buff_matched()
2644 * Checks to see if a named buffer has matched. The data array of
2645 * buffer numbers corresponding to the buffer is expected to reside
2646 * in the regexp->data->data array in the slot stored in the ARG() of
2647 * node involved. Note that this routine doesn't actually care about the
2648 * name, that information is not preserved from compilation to execution.
2649 * Returns the index of the leftmost defined buffer with the given name
2650 * or 0 if non of the buffers matched.
2653 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2655 RXi_GET_DECL(rex,rexi);
2656 SV *sv_dat=(SV*)rexi->data->data[ ARG( scan ) ];
2657 I32 *nums=(I32*)SvPVX(sv_dat);
2658 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2659 if ((I32)*PL_reglastparen >= nums[n] &&
2660 PL_regoffs[nums[n]].end != -1)
2669 /* free all slabs above current one - called during LEAVE_SCOPE */
2672 S_clear_backtrack_stack(pTHX_ void *p)
2674 regmatch_slab *s = PL_regmatch_slab->next;
2679 PL_regmatch_slab->next = NULL;
2681 regmatch_slab * const osl = s;
2688 #define SETREX(Re1,Re2) \
2689 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2692 STATIC I32 /* 0 failure, 1 success */
2693 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2695 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2699 register const bool do_utf8 = PL_reg_match_utf8;
2700 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2702 REGEXP *rex_sv = reginfo->prog;
2703 regexp *rex = (struct regexp *)SvANY(rex_sv);
2704 RXi_GET_DECL(rex,rexi);
2708 /* the current state. This is a cached copy of PL_regmatch_state */
2709 register regmatch_state *st;
2711 /* cache heavy used fields of st in registers */
2712 register regnode *scan;
2713 register regnode *next;
2714 register U32 n = 0; /* general value; init to avoid compiler warning */
2715 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2716 register char *locinput = PL_reginput;
2717 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2719 bool result = 0; /* return value of S_regmatch */
2720 int depth = 0; /* depth of backtrack stack */
2721 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2722 const U32 max_nochange_depth =
2723 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2724 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2726 regmatch_state *yes_state = NULL; /* state to pop to on success of
2728 /* mark_state piggy backs on the yes_state logic so that when we unwind
2729 the stack on success we can update the mark_state as we go */
2730 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2732 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2733 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2735 bool no_final = 0; /* prevent failure from backtracking? */
2736 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2737 char *startpoint = PL_reginput;
2738 SV *popmark = NULL; /* are we looking for a mark? */
2739 SV *sv_commit = NULL; /* last mark name seen in failure */
2740 SV *sv_yes_mark = NULL; /* last mark name we have seen
2741 during a successfull match */
2742 U32 lastopen = 0; /* last open we saw */
2743 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2745 SV* const oreplsv = GvSV(PL_replgv);
2748 /* these three flags are set by various ops to signal information to
2749 * the very next op. They have a useful lifetime of exactly one loop
2750 * iteration, and are not preserved or restored by state pushes/pops
2752 bool sw = 0; /* the condition value in (?(cond)a|b) */
2753 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2754 int logical = 0; /* the following EVAL is:
2758 or the following IFMATCH/UNLESSM is:
2759 false: plain (?=foo)
2760 true: used as a condition: (?(?=foo))
2764 GET_RE_DEBUG_FLAGS_DECL;
2767 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
2768 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2770 /* on first ever call to regmatch, allocate first slab */
2771 if (!PL_regmatch_slab) {
2772 Newx(PL_regmatch_slab, 1, regmatch_slab);
2773 PL_regmatch_slab->prev = NULL;
2774 PL_regmatch_slab->next = NULL;
2775 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2778 oldsave = PL_savestack_ix;
2779 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
2780 SAVEVPTR(PL_regmatch_slab);
2781 SAVEVPTR(PL_regmatch_state);
2783 /* grab next free state slot */
2784 st = ++PL_regmatch_state;
2785 if (st > SLAB_LAST(PL_regmatch_slab))
2786 st = PL_regmatch_state = S_push_slab(aTHX);
2788 /* Note that nextchr is a byte even in UTF */
2789 nextchr = UCHARAT(locinput);
2791 while (scan != NULL) {
2794 SV * const prop = sv_newmortal();
2795 regnode *rnext=regnext(scan);
2796 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2797 regprop(rex, prop, scan);
2799 PerlIO_printf(Perl_debug_log,
2800 "%3"IVdf":%*s%s(%"IVdf")\n",
2801 (IV)(scan - rexi->program), depth*2, "",
2803 (PL_regkind[OP(scan)] == END || !rnext) ?
2804 0 : (IV)(rnext - rexi->program));
2807 next = scan + NEXT_OFF(scan);
2810 state_num = OP(scan);
2813 switch (state_num) {
2815 if (locinput == PL_bostr)
2817 /* reginfo->till = reginfo->bol; */
2822 if (locinput == PL_bostr ||
2823 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2829 if (locinput == PL_bostr)
2833 if (locinput == reginfo->ganch)
2838 /* update the startpoint */
2839 st->u.keeper.val = PL_regoffs[0].start;
2840 PL_reginput = locinput;
2841 PL_regoffs[0].start = locinput - PL_bostr;
2842 PUSH_STATE_GOTO(KEEPS_next, next);
2844 case KEEPS_next_fail:
2845 /* rollback the start point change */
2846 PL_regoffs[0].start = st->u.keeper.val;
2852 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2857 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2859 if (PL_regeol - locinput > 1)
2863 if (PL_regeol != locinput)
2867 if (!nextchr && locinput >= PL_regeol)
2870 locinput += PL_utf8skip[nextchr];
2871 if (locinput > PL_regeol)
2873 nextchr = UCHARAT(locinput);
2876 nextchr = UCHARAT(++locinput);
2879 if (!nextchr && locinput >= PL_regeol)
2881 nextchr = UCHARAT(++locinput);
2884 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2887 locinput += PL_utf8skip[nextchr];
2888 if (locinput > PL_regeol)
2890 nextchr = UCHARAT(locinput);
2893 nextchr = UCHARAT(++locinput);
2897 #define ST st->u.trie
2899 /* In this case the charclass data is available inline so
2900 we can fail fast without a lot of extra overhead.
2902 if (scan->flags == EXACT || !do_utf8) {
2903 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2905 PerlIO_printf(Perl_debug_log,
2906 "%*s %sfailed to match trie start class...%s\n",
2907 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2916 /* what type of TRIE am I? (utf8 makes this contextual) */
2917 DECL_TRIE_TYPE(scan);
2919 /* what trie are we using right now */
2920 reg_trie_data * const trie
2921 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2922 HV * widecharmap = (HV *)rexi->data->data[ ARG( scan ) + 1 ];
2923 U32 state = trie->startstate;
2925 if (trie->bitmap && trie_type != trie_utf8_fold &&
2926 !TRIE_BITMAP_TEST(trie,*locinput)
2928 if (trie->states[ state ].wordnum) {
2930 PerlIO_printf(Perl_debug_log,
2931 "%*s %smatched empty string...%s\n",
2932 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2937 PerlIO_printf(Perl_debug_log,
2938 "%*s %sfailed to match trie start class...%s\n",
2939 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2946 U8 *uc = ( U8* )locinput;
2950 U8 *uscan = (U8*)NULL;
2952 SV *sv_accept_buff = NULL;
2953 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2955 ST.accepted = 0; /* how many accepting states we have seen */
2957 ST.jump = trie->jump;
2960 traverse the TRIE keeping track of all accepting states
2961 we transition through until we get to a failing node.
2964 while ( state && uc <= (U8*)PL_regeol ) {
2965 U32 base = trie->states[ state ].trans.base;
2968 /* We use charid to hold the wordnum as we don't use it
2969 for charid until after we have done the wordnum logic.
2970 We define an alias just so that the wordnum logic reads
2973 #define got_wordnum charid
2974 got_wordnum = trie->states[ state ].wordnum;
2976 if ( got_wordnum ) {
2977 if ( ! ST.accepted ) {
2979 /* SAVETMPS; */ /* XXX is this necessary? dmq */
2980 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2981 sv_accept_buff=newSV(bufflen *
2982 sizeof(reg_trie_accepted) - 1);
2983 SvCUR_set(sv_accept_buff, 0);
2984 SvPOK_on(sv_accept_buff);
2985 sv_2mortal(sv_accept_buff);
2988 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2991 if (ST.accepted >= bufflen) {
2993 ST.accept_buff =(reg_trie_accepted*)
2994 SvGROW(sv_accept_buff,
2995 bufflen * sizeof(reg_trie_accepted));
2997 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2998 + sizeof(reg_trie_accepted));
3001 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
3002 ST.accept_buff[ST.accepted].endpos = uc;
3004 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
3008 DEBUG_TRIE_EXECUTE_r({
3009 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
3010 PerlIO_printf( Perl_debug_log,
3011 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
3012 2+depth * 2, "", PL_colors[4],
3013 (UV)state, (UV)ST.accepted );
3017 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3018 uscan, len, uvc, charid, foldlen,
3022 (base + charid > trie->uniquecharcount )
3023 && (base + charid - 1 - trie->uniquecharcount
3025 && trie->trans[base + charid - 1 -
3026 trie->uniquecharcount].check == state)
3028 state = trie->trans[base + charid - 1 -
3029 trie->uniquecharcount ].next;
3040 DEBUG_TRIE_EXECUTE_r(
3041 PerlIO_printf( Perl_debug_log,
3042 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3043 charid, uvc, (UV)state, PL_colors[5] );
3050 PerlIO_printf( Perl_debug_log,
3051 "%*s %sgot %"IVdf" possible matches%s\n",
3052 REPORT_CODE_OFF + depth * 2, "",
3053 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3056 goto trie_first_try; /* jump into the fail handler */
3058 case TRIE_next_fail: /* we failed - try next alterative */
3060 REGCP_UNWIND(ST.cp);
3061 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3062 PL_regoffs[n].end = -1;
3063 *PL_reglastparen = n;
3072 ST.lastparen = *PL_reglastparen;
3075 if ( ST.accepted == 1 ) {
3076 /* only one choice left - just continue */
3078 AV *const trie_words
3079 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
3080 SV ** const tmp = av_fetch( trie_words,
3081 ST.accept_buff[ 0 ].wordnum-1, 0 );
3082 SV *sv= tmp ? sv_newmortal() : NULL;
3084 PerlIO_printf( Perl_debug_log,
3085 "%*s %sonly one match left: #%d <%s>%s\n",
3086 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3087 ST.accept_buff[ 0 ].wordnum,
3088 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3089 PL_colors[0], PL_colors[1],
3090 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3092 : "not compiled under -Dr",
3095 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
3096 /* in this case we free tmps/leave before we call regmatch
3097 as we wont be using accept_buff again. */
3099 locinput = PL_reginput;
3100 nextchr = UCHARAT(locinput);
3101 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
3104 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
3105 if (!has_cutgroup) {
3110 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3113 continue; /* execute rest of RE */
3116 if ( !ST.accepted-- ) {
3118 PerlIO_printf( Perl_debug_log,
3119 "%*s %sTRIE failed...%s\n",
3120 REPORT_CODE_OFF+depth*2, "",
3131 There are at least two accepting states left. Presumably
3132 the number of accepting states is going to be low,
3133 typically two. So we simply scan through to find the one
3134 with lowest wordnum. Once we find it, we swap the last
3135 state into its place and decrement the size. We then try to
3136 match the rest of the pattern at the point where the word
3137 ends. If we succeed, control just continues along the
3138 regex; if we fail we return here to try the next accepting
3145 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3146 DEBUG_TRIE_EXECUTE_r(
3147 PerlIO_printf( Perl_debug_log,
3148 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3149 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3150 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3151 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3154 if (ST.accept_buff[cur].wordnum <
3155 ST.accept_buff[best].wordnum)
3160 AV *const trie_words
3161 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
3162 SV ** const tmp = av_fetch( trie_words,
3163 ST.accept_buff[ best ].wordnum - 1, 0 );
3164 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3166 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3167 SV *sv= tmp ? sv_newmortal() : NULL;
3169 PerlIO_printf( Perl_debug_log,
3170 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3171 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3172 ST.accept_buff[best].wordnum,
3173 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3174 PL_colors[0], PL_colors[1],
3175 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3176 ) : "not compiled under -Dr",
3177 REG_NODE_NUM(nextop),
3181 if ( best<ST.accepted ) {
3182 reg_trie_accepted tmp = ST.accept_buff[ best ];
3183 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3184 ST.accept_buff[ ST.accepted ] = tmp;
3187 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3188 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3191 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3193 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3204 char *s = STRING(scan);
3206 if (do_utf8 != UTF) {
3207 /* The target and the pattern have differing utf8ness. */
3209 const char * const e = s + ln;
3212 /* The target is utf8, the pattern is not utf8. */
3217 if (NATIVE_TO_UNI(*(U8*)s) !=
3218 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3226 /* The target is not utf8, the pattern is utf8. */
3231 if (NATIVE_TO_UNI(*((U8*)l)) !=
3232 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3240 nextchr = UCHARAT(locinput);
3243 /* The target and the pattern have the same utf8ness. */
3244 /* Inline the first character, for speed. */
3245 if (UCHARAT(s) != nextchr)
3247 if (PL_regeol - locinput < ln)
3249 if (ln > 1 && memNE(s, locinput, ln))
3252 nextchr = UCHARAT(locinput);
3256 PL_reg_flags |= RF_tainted;
3259 char * const s = STRING(scan);
3262 if (do_utf8 || UTF) {
3263 /* Either target or the pattern are utf8. */
3264 const char * const l = locinput;
3265 char *e = PL_regeol;
3267 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3268 l, &e, 0, do_utf8)) {
3269 /* One more case for the sharp s:
3270 * pack("U0U*", 0xDF) =~ /ss/i,
3271 * the 0xC3 0x9F are the UTF-8
3272 * byte sequence for the U+00DF. */
3275 toLOWER(s[0]) == 's' &&
3277 toLOWER(s[1]) == 's' &&
3284 nextchr = UCHARAT(locinput);
3288 /* Neither the target and the pattern are utf8. */
3290 /* Inline the first character, for speed. */
3291 if (UCHARAT(s) != nextchr &&
3292 UCHARAT(s) != ((OP(scan) == EXACTF)
3293 ? PL_fold : PL_fold_locale)[nextchr])
3295 if (PL_regeol - locinput < ln)
3297 if (ln > 1 && (OP(scan) == EXACTF
3298 ? ibcmp(s, locinput, ln)
3299 : ibcmp_locale(s, locinput, ln)))
3302 nextchr = UCHARAT(locinput);
3307 STRLEN inclasslen = PL_regeol - locinput;
3309 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3311 if (locinput >= PL_regeol)
3313 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3314 nextchr = UCHARAT(locinput);
3319 nextchr = UCHARAT(locinput);
3320 if (!REGINCLASS(rex, scan, (U8*)locinput))
3322 if (!nextchr && locinput >= PL_regeol)
3324 nextchr = UCHARAT(++locinput);
3328 /* If we might have the case of the German sharp s
3329 * in a casefolding Unicode character class. */
3331 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3332 locinput += SHARP_S_SKIP;
3333 nextchr = UCHARAT(locinput);
3339 PL_reg_flags |= RF_tainted;
3345 LOAD_UTF8_CHARCLASS_ALNUM();
3346 if (!(OP(scan) == ALNUM
3347 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3348 : isALNUM_LC_utf8((U8*)locinput)))
3352 locinput += PL_utf8skip[nextchr];
3353 nextchr = UCHARAT(locinput);
3356 if (!(OP(scan) == ALNUM
3357 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3359 nextchr = UCHARAT(++locinput);
3362 PL_reg_flags |= RF_tainted;
3365 if (!nextchr && locinput >= PL_regeol)
3368 LOAD_UTF8_CHARCLASS_ALNUM();
3369 if (OP(scan) == NALNUM
3370 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3371 : isALNUM_LC_utf8((U8*)locinput))
3375 locinput += PL_utf8skip[nextchr];
3376 nextchr = UCHARAT(locinput);
3379 if (OP(scan) == NALNUM
3380 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3382 nextchr = UCHARAT(++locinput);
3386 PL_reg_flags |= RF_tainted;
3390 /* was last char in word? */
3392 if (locinput == PL_bostr)
3395 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3397 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3399 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3400 ln = isALNUM_uni(ln);
3401 LOAD_UTF8_CHARCLASS_ALNUM();
3402 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3405 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3406 n = isALNUM_LC_utf8((U8*)locinput);
3410 ln = (locinput != PL_bostr) ?
3411 UCHARAT(locinput - 1) : '\n';
3412 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3414 n = isALNUM(nextchr);
3417 ln = isALNUM_LC(ln);
3418 n = isALNUM_LC(nextchr);
3421 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3422 OP(scan) == BOUNDL))
3426 PL_reg_flags |= RF_tainted;
3432 if (UTF8_IS_CONTINUED(nextchr)) {
3433 LOAD_UTF8_CHARCLASS_SPACE();
3434 if (!(OP(scan) == SPACE
3435 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3436 : isSPACE_LC_utf8((U8*)locinput)))
3440 locinput += PL_utf8skip[nextchr];
3441 nextchr = UCHARAT(locinput);
3444 if (!(OP(scan) == SPACE
3445 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3447 nextchr = UCHARAT(++locinput);
3450 if (!(OP(scan) == SPACE
3451 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3453 nextchr = UCHARAT(++locinput);
3457 PL_reg_flags |= RF_tainted;
3460 if (!nextchr && locinput >= PL_regeol)
3463 LOAD_UTF8_CHARCLASS_SPACE();
3464 if (OP(scan) == NSPACE
3465 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3466 : isSPACE_LC_utf8((U8*)locinput))
3470 locinput += PL_utf8skip[nextchr];
3471 nextchr = UCHARAT(locinput);
3474 if (OP(scan) == NSPACE
3475 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3477 nextchr = UCHARAT(++locinput);
3480 PL_reg_flags |= RF_tainted;
3486 LOAD_UTF8_CHARCLASS_DIGIT();
3487 if (!(OP(scan) == DIGIT
3488 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3489 : isDIGIT_LC_utf8((U8*)locinput)))
3493 locinput += PL_utf8skip[nextchr];
3494 nextchr = UCHARAT(locinput);
3497 if (!(OP(scan) == DIGIT
3498 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3500 nextchr = UCHARAT(++locinput);
3503 PL_reg_flags |= RF_tainted;
3506 if (!nextchr && locinput >= PL_regeol)
3509 LOAD_UTF8_CHARCLASS_DIGIT();
3510 if (OP(scan) == NDIGIT
3511 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3512 : isDIGIT_LC_utf8((U8*)locinput))
3516 locinput += PL_utf8skip[nextchr];
3517 nextchr = UCHARAT(locinput);
3520 if (OP(scan) == NDIGIT
3521 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3523 nextchr = UCHARAT(++locinput);
3526 if (locinput >= PL_regeol)
3529 LOAD_UTF8_CHARCLASS_MARK();
3530 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3532 locinput += PL_utf8skip[nextchr];
3533 while (locinput < PL_regeol &&
3534 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3535 locinput += UTF8SKIP(locinput);
3536 if (locinput > PL_regeol)
3541 nextchr = UCHARAT(locinput);
3548 PL_reg_flags |= RF_tainted;
3553 n = reg_check_named_buff_matched(rex,scan);
3556 type = REF + ( type - NREF );
3563 PL_reg_flags |= RF_tainted;
3567 n = ARG(scan); /* which paren pair */
3570 ln = PL_regoffs[n].start;
3571 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3572 if (*PL_reglastparen < n || ln == -1)
3573 sayNO; /* Do not match unless seen CLOSEn. */
3574 if (ln == PL_regoffs[n].end)
3578 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3580 const char *e = PL_bostr + PL_regoffs[n].end;
3582 * Note that we can't do the "other character" lookup trick as
3583 * in the 8-bit case (no pun intended) because in Unicode we
3584 * have to map both upper and title case to lower case.
3588 STRLEN ulen1, ulen2;
3589 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3590 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3594 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3595 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3596 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3603 nextchr = UCHARAT(locinput);
3607 /* Inline the first character, for speed. */
3608 if (UCHARAT(s) != nextchr &&
3610 (UCHARAT(s) != (type == REFF
3611 ? PL_fold : PL_fold_locale)[nextchr])))
3613 ln = PL_regoffs[n].end - ln;
3614 if (locinput + ln > PL_regeol)
3616 if (ln > 1 && (type == REF
3617 ? memNE(s, locinput, ln)
3619 ? ibcmp(s, locinput, ln)
3620 : ibcmp_locale(s, locinput, ln))))
3623 nextchr = UCHARAT(locinput);
3633 #define ST st->u.eval
3638 regexp_internal *rei;
3639 regnode *startpoint;
3642 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
3643 if (cur_eval && cur_eval->locinput==locinput) {
3644 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3645 Perl_croak(aTHX_ "Infinite recursion in regex");
3646 if ( ++nochange_depth > max_nochange_depth )
3648 "Pattern subroutine nesting without pos change"
3649 " exceeded limit in regex");
3656 (void)ReREFCNT_inc(rex_sv);
3657 if (OP(scan)==GOSUB) {
3658 startpoint = scan + ARG2L(scan);
3659 ST.close_paren = ARG(scan);
3661 startpoint = rei->program+1;
3664 goto eval_recurse_doit;
3666 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3667 if (cur_eval && cur_eval->locinput==locinput) {
3668 if ( ++nochange_depth > max_nochange_depth )
3669 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3674 /* execute the code in the {...} */
3676 SV ** const before = SP;
3677 OP_4tree * const oop = PL_op;
3678 COP * const ocurcop = PL_curcop;
3682 PL_op = (OP_4tree*)rexi->data->data[n];
3683 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3684 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3685 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3686 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
3689 SV *sv_mrk = get_sv("REGMARK", 1);
3690 sv_setsv(sv_mrk, sv_yes_mark);
3693 CALLRUNOPS(aTHX); /* Scalar context. */
3696 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3703 PAD_RESTORE_LOCAL(old_comppad);
3704 PL_curcop = ocurcop;
3707 sv_setsv(save_scalar(PL_replgv), ret);
3711 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3714 /* extract RE object from returned value; compiling if
3720 SV *const sv = SvRV(ret);
3722 if (SvTYPE(sv) == SVt_REGEXP) {
3724 } else if (SvSMAGICAL(sv)) {
3725 mg = mg_find(sv, PERL_MAGIC_qr);
3728 } else if (SvTYPE(ret) == SVt_REGEXP) {
3730 } else if (SvSMAGICAL(ret)) {
3731 if (SvGMAGICAL(ret)) {
3732 /* I don't believe that there is ever qr magic
3734 assert(!mg_find(ret, PERL_MAGIC_qr));
3735 sv_unmagic(ret, PERL_MAGIC_qr);
3738 mg = mg_find(ret, PERL_MAGIC_qr);
3739 /* testing suggests mg only ends up non-NULL for
3740 scalars who were upgraded and compiled in the
3741 else block below. In turn, this is only
3742 triggered in the "postponed utf8 string" tests
3748 rx = mg->mg_obj; /*XXX:dmq*/
3752 rx = reg_temp_copy(rx);
3756 const I32 osize = PL_regsize;
3759 assert (SvUTF8(ret));
3760 } else if (SvUTF8(ret)) {
3761 /* Not doing UTF-8, despite what the SV says. Is
3762 this only if we're trapped in use 'bytes'? */
3763 /* Make a copy of the octet sequence, but without
3764 the flag on, as the compiler now honours the
3765 SvUTF8 flag on ret. */
3767 const char *const p = SvPV(ret, len);
3768 ret = newSVpvn_flags(p, len, SVs_TEMP);
3770 rx = CALLREGCOMP(ret, pm_flags);
3772 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3774 /* This isn't a first class regexp. Instead, it's
3775 caching a regexp onto an existing, Perl visible
3777 sv_magic(ret, rx, PERL_MAGIC_qr, 0, 0);
3782 re = (struct regexp *)SvANY(rx);
3784 RXp_MATCH_COPIED_off(re);
3785 re->subbeg = rex->subbeg;
3786 re->sublen = rex->sublen;
3789 debug_start_match(re_sv, do_utf8, locinput, PL_regeol,
3790 "Matching embedded");
3792 startpoint = rei->program + 1;
3793 ST.close_paren = 0; /* only used for GOSUB */
3794 /* borrowed from regtry */
3795 if (PL_reg_start_tmpl <= re->nparens) {
3796 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3797 if(PL_reg_start_tmp)
3798 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3800 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3803 eval_recurse_doit: /* Share code with GOSUB below this line */
3804 /* run the pattern returned from (??{...}) */
3805 ST.cp = regcppush(0); /* Save *all* the positions. */
3806 REGCP_SET(ST.lastcp);
3808 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
3810 *PL_reglastparen = 0;
3811 *PL_reglastcloseparen = 0;
3812 PL_reginput = locinput;
3815 /* XXXX This is too dramatic a measure... */
3818 ST.toggle_reg_flags = PL_reg_flags;
3820 PL_reg_flags |= RF_utf8;
3822 PL_reg_flags &= ~RF_utf8;
3823 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3825 ST.prev_rex = rex_sv;
3826 ST.prev_curlyx = cur_curlyx;
3827 SETREX(rex_sv,re_sv);
3832 ST.prev_eval = cur_eval;
3834 /* now continue from first node in postoned RE */
3835 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3838 /* logical is 1, /(?(?{...})X|Y)/ */
3839 sw = (bool)SvTRUE(ret);
3844 case EVAL_AB: /* cleanup after a successful (??{A})B */
3845 /* note: this is called twice; first after popping B, then A */
3846 PL_reg_flags ^= ST.toggle_reg_flags;
3847 ReREFCNT_dec(rex_sv);
3848 SETREX(rex_sv,ST.prev_rex);
3849 rex = (struct regexp *)SvANY(rex_sv);
3850 rexi = RXi_GET(rex);
3852 cur_eval = ST.prev_eval;
3853 cur_curlyx = ST.prev_curlyx;
3854 /* XXXX This is too dramatic a measure... */
3856 if ( nochange_depth )
3861 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3862 /* note: this is called twice; first after popping B, then A */
3863 PL_reg_flags ^= ST.toggle_reg_flags;
3864 ReREFCNT_dec(rex_sv);
3865 SETREX(rex_sv,ST.prev_rex);
3866 rex = (struct regexp *)SvANY(rex_sv);
3867 rexi = RXi_GET(rex);
3868 PL_reginput = locinput;
3869 REGCP_UNWIND(ST.lastcp);
3871 cur_eval = ST.prev_eval;
3872 cur_curlyx = ST.prev_curlyx;
3873 /* XXXX This is too dramatic a measure... */
3875 if ( nochange_depth )
3881 n = ARG(scan); /* which paren pair */
3882 PL_reg_start_tmp[n] = locinput;
3888 n = ARG(scan); /* which paren pair */
3889 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
3890 PL_regoffs[n].end = locinput - PL_bostr;
3891 /*if (n > PL_regsize)
3893 if (n > *PL_reglastparen)
3894 *PL_reglastparen = n;
3895 *PL_reglastcloseparen = n;
3896 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3904 cursor && OP(cursor)!=END;
3905 cursor=regnext(cursor))
3907 if ( OP(cursor)==CLOSE ){
3909 if ( n <= lastopen ) {
3911 = PL_reg_start_tmp[n] - PL_bostr;
3912 PL_regoffs[n].end = locinput - PL_bostr;
3913 /*if (n > PL_regsize)
3915 if (n > *PL_reglastparen)
3916 *PL_reglastparen = n;
3917 *PL_reglastcloseparen = n;
3918 if ( n == ARG(scan) || (cur_eval &&
3919 cur_eval->u.eval.close_paren == n))
3928 n = ARG(scan); /* which paren pair */
3929 sw = (bool)(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
3932 /* reg_check_named_buff_matched returns 0 for no match */
3933 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3937 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3943 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3945 next = NEXTOPER(NEXTOPER(scan));
3947 next = scan + ARG(scan);
3948 if (OP(next) == IFTHEN) /* Fake one. */
3949 next = NEXTOPER(NEXTOPER(next));
3953 logical = scan->flags;
3956 /*******************************************************************
3958 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3959 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3960 STAR/PLUS/CURLY/CURLYN are used instead.)
3962 A*B is compiled as <CURLYX><A><WHILEM><B>
3964 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3965 state, which contains the current count, initialised to -1. It also sets
3966 cur_curlyx to point to this state, with any previous value saved in the
3969 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3970 since the pattern may possibly match zero times (i.e. it's a while {} loop
3971 rather than a do {} while loop).
3973 Each entry to WHILEM represents a successful match of A. The count in the
3974 CURLYX block is incremented, another WHILEM state is pushed, and execution
3975 passes to A or B depending on greediness and the current count.
3977 For example, if matching against the string a1a2a3b (where the aN are
3978 substrings that match /A/), then the match progresses as follows: (the
3979 pushed states are interspersed with the bits of strings matched so far):
3982 <CURLYX cnt=0><WHILEM>
3983 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3984 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3985 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3986 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3988 (Contrast this with something like CURLYM, which maintains only a single
3992 a1 <CURLYM cnt=1> a2
3993 a1 a2 <CURLYM cnt=2> a3
3994 a1 a2 a3 <CURLYM cnt=3> b
3997 Each WHILEM state block marks a point to backtrack to upon partial failure
3998 of A or B, and also contains some minor state data related to that
3999 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4000 overall state, such as the count, and pointers to the A and B ops.
4002 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4003 must always point to the *current* CURLYX block, the rules are:
4005 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4006 and set cur_curlyx to point the new block.
4008 When popping the CURLYX block after a successful or unsuccessful match,
4009 restore the previous cur_curlyx.
4011 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4012 to the outer one saved in the CURLYX block.
4014 When popping the WHILEM block after a successful or unsuccessful B match,
4015 restore the previous cur_curlyx.
4017 Here's an example for the pattern (AI* BI)*BO
4018 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4021 curlyx backtrack stack
4022 ------ ---------------
4024 CO <CO prev=NULL> <WO>
4025 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4026 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4027 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4029 At this point the pattern succeeds, and we work back down the stack to
4030 clean up, restoring as we go:
4032 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4033 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4034 CO <CO prev=NULL> <WO>
4037 *******************************************************************/
4039 #define ST st->u.curlyx
4041 case CURLYX: /* start of /A*B/ (for complex A) */
4043 /* No need to save/restore up to this paren */
4044 I32 parenfloor = scan->flags;
4046 assert(next); /* keep Coverity happy */
4047 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4050 /* XXXX Probably it is better to teach regpush to support
4051 parenfloor > PL_regsize... */
4052 if (parenfloor > (I32)*PL_reglastparen)
4053 parenfloor = *PL_reglastparen; /* Pessimization... */
4055 ST.prev_curlyx= cur_curlyx;
4057 ST.cp = PL_savestack_ix;
4059 /* these fields contain the state of the current curly.
4060 * they are accessed by subsequent WHILEMs */
4061 ST.parenfloor = parenfloor;
4062 ST.min = ARG1(scan);
4063 ST.max = ARG2(scan);
4064 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
4068 ST.count = -1; /* this will be updated by WHILEM */
4069 ST.lastloc = NULL; /* this will be updated by WHILEM */
4071 PL_reginput = locinput;
4072 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4076 case CURLYX_end: /* just finished matching all of A*B */
4077 cur_curlyx = ST.prev_curlyx;
4081 case CURLYX_end_fail: /* just failed to match all of A*B */
4083 cur_curlyx = ST.prev_curlyx;
4089 #define ST st->u.whilem
4091 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4093 /* see the discussion above about CURLYX/WHILEM */
4095 assert(cur_curlyx); /* keep Coverity happy */
4096 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4097 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4098 ST.cache_offset = 0;
4101 PL_reginput = locinput;
4103 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4104 "%*s whilem: matched %ld out of %ld..%ld\n",
4105 REPORT_CODE_OFF+depth*2, "", (long)n,
4106 (long)cur_curlyx->u.curlyx.min,
4107 (long)cur_curlyx->u.curlyx.max)
4110 /* First just match a string of min A's. */
4112 if (n < cur_curlyx->u.curlyx.min) {
4113 cur_curlyx->u.curlyx.lastloc = locinput;
4114 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
4118 /* If degenerate A matches "", assume A done. */
4120 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4121 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4122 "%*s whilem: empty match detected, trying continuation...\n",
4123 REPORT_CODE_OFF+depth*2, "")
4125 goto do_whilem_B_max;
4128 /* super-linear cache processing */
4132 if (!PL_reg_maxiter) {
4133 /* start the countdown: Postpone detection until we
4134 * know the match is not *that* much linear. */
4135 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4136 /* possible overflow for long strings and many CURLYX's */
4137 if (PL_reg_maxiter < 0)
4138 PL_reg_maxiter = I32_MAX;
4139 PL_reg_leftiter = PL_reg_maxiter;
4142 if (PL_reg_leftiter-- == 0) {
4143 /* initialise cache */
4144 const I32 size = (PL_reg_maxiter + 7)/8;
4145 if (PL_reg_poscache) {
4146 if ((I32)PL_reg_poscache_size < size) {
4147 Renew(PL_reg_poscache, size, char);
4148 PL_reg_poscache_size = size;
4150 Zero(PL_reg_poscache, size, char);
4153 PL_reg_poscache_size = size;
4154 Newxz(PL_reg_poscache, size, char);
4156 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4157 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4158 PL_colors[4], PL_colors[5])
4162 if (PL_reg_leftiter < 0) {
4163 /* have we already failed at this position? */
4165 offset = (scan->flags & 0xf) - 1
4166 + (locinput - PL_bostr) * (scan->flags>>4);
4167 mask = 1 << (offset % 8);
4169 if (PL_reg_poscache[offset] & mask) {
4170 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4171 "%*s whilem: (cache) already tried at this position...\n",
4172 REPORT_CODE_OFF+depth*2, "")
4174 sayNO; /* cache records failure */
4176 ST.cache_offset = offset;
4177 ST.cache_mask = mask;
4181 /* Prefer B over A for minimal matching. */
4183 if (cur_curlyx->u.curlyx.minmod) {
4184 ST.save_curlyx = cur_curlyx;
4185 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4186 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4187 REGCP_SET(ST.lastcp);
4188 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4192 /* Prefer A over B for maximal matching. */
4194 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
4195 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4196 cur_curlyx->u.curlyx.lastloc = locinput;
4197 REGCP_SET(ST.lastcp);
4198 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
4201 goto do_whilem_B_max;
4205 case WHILEM_B_min: /* just matched B in a minimal match */
4206 case WHILEM_B_max: /* just matched B in a maximal match */
4207 cur_curlyx = ST.save_curlyx;
4211 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4212 cur_curlyx = ST.save_curlyx;
4213 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4214 cur_curlyx->u.curlyx.count--;
4218 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4219 REGCP_UNWIND(ST.lastcp);
4222 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4223 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4224 cur_curlyx->u.curlyx.count--;
4228 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4229 REGCP_UNWIND(ST.lastcp);
4230 regcppop(rex); /* Restore some previous $<digit>s? */
4231 PL_reginput = locinput;
4232 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4233 "%*s whilem: failed, trying continuation...\n",
4234 REPORT_CODE_OFF+depth*2, "")
4237 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4238 && ckWARN(WARN_REGEXP)
4239 && !(PL_reg_flags & RF_warned))
4241 PL_reg_flags |= RF_warned;
4242 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4243 "Complex regular subexpression recursion",
4248 ST.save_curlyx = cur_curlyx;
4249 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4250 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4253 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4254 cur_curlyx = ST.save_curlyx;
4255 REGCP_UNWIND(ST.lastcp);
4258 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
4259 /* Maximum greed exceeded */
4260 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4261 && ckWARN(WARN_REGEXP)
4262 && !(PL_reg_flags & RF_warned))
4264 PL_reg_flags |= RF_warned;
4265 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4266 "%s limit (%d) exceeded",
4267 "Complex regular subexpression recursion",
4270 cur_curlyx->u.curlyx.count--;
4274 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4275 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4277 /* Try grabbing another A and see if it helps. */
4278 PL_reginput = locinput;
4279 cur_curlyx->u.curlyx.lastloc = locinput;
4280 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4281 REGCP_SET(ST.lastcp);
4282 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
4286 #define ST st->u.branch
4288 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4289 next = scan + ARG(scan);
4292 scan = NEXTOPER(scan);
4295 case BRANCH: /* /(...|A|...)/ */
4296 scan = NEXTOPER(scan); /* scan now points to inner node */
4297 ST.lastparen = *PL_reglastparen;
4298 ST.next_branch = next;
4300 PL_reginput = locinput;
4302 /* Now go into the branch */
4304 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4306 PUSH_STATE_GOTO(BRANCH_next, scan);
4310 PL_reginput = locinput;
4311 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4312 (SV*)rexi->data->data[ ARG( scan ) ];
4313 PUSH_STATE_GOTO(CUTGROUP_next,next);
4315 case CUTGROUP_next_fail:
4318 if (st->u.mark.mark_name)
4319 sv_commit = st->u.mark.mark_name;
4325 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4330 REGCP_UNWIND(ST.cp);
4331 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4332 PL_regoffs[n].end = -1;
4333 *PL_reglastparen = n;
4334 /*dmq: *PL_reglastcloseparen = n; */
4335 scan = ST.next_branch;
4336 /* no more branches? */
4337 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4339 PerlIO_printf( Perl_debug_log,
4340 "%*s %sBRANCH failed...%s\n",
4341 REPORT_CODE_OFF+depth*2, "",
4347 continue; /* execute next BRANCH[J] op */
4355 #define ST st->u.curlym
4357 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4359 /* This is an optimisation of CURLYX that enables us to push
4360 * only a single backtracking state, no matter now many matches
4361 * there are in {m,n}. It relies on the pattern being constant
4362 * length, with no parens to influence future backrefs
4366 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4368 /* if paren positive, emulate an OPEN/CLOSE around A */
4370 U32 paren = ST.me->flags;
4371 if (paren > PL_regsize)
4373 if (paren > *PL_reglastparen)
4374 *PL_reglastparen = paren;
4375 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4383 ST.c1 = CHRTEST_UNINIT;
4386 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4389 curlym_do_A: /* execute the A in /A{m,n}B/ */
4390 PL_reginput = locinput;
4391 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4394 case CURLYM_A: /* we've just matched an A */
4395 locinput = st->locinput;
4396 nextchr = UCHARAT(locinput);
4399 /* after first match, determine A's length: u.curlym.alen */
4400 if (ST.count == 1) {
4401 if (PL_reg_match_utf8) {
4403 while (s < PL_reginput) {
4409 ST.alen = PL_reginput - locinput;
4412 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4415 PerlIO_printf(Perl_debug_log,
4416 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4417 (int)(REPORT_CODE_OFF+(depth*2)), "",
4418 (IV) ST.count, (IV)ST.alen)
4421 locinput = PL_reginput;
4423 if (cur_eval && cur_eval->u.eval.close_paren &&
4424 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4427 if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
4428 goto curlym_do_A; /* try to match another A */
4429 goto curlym_do_B; /* try to match B */
4431 case CURLYM_A_fail: /* just failed to match an A */
4432 REGCP_UNWIND(ST.cp);
4434 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4435 || (cur_eval && cur_eval->u.eval.close_paren &&
4436 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4439 curlym_do_B: /* execute the B in /A{m,n}B/ */
4440 PL_reginput = locinput;
4441 if (ST.c1 == CHRTEST_UNINIT) {
4442 /* calculate c1 and c2 for possible match of 1st char
4443 * following curly */
4444 ST.c1 = ST.c2 = CHRTEST_VOID;
4445 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4446 regnode *text_node = ST.B;
4447 if (! HAS_TEXT(text_node))
4448 FIND_NEXT_IMPT(text_node);
4451 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4453 But the former is redundant in light of the latter.
4455 if this changes back then the macro for
4456 IS_TEXT and friends need to change.
4458 if (PL_regkind[OP(text_node)] == EXACT)
4461 ST.c1 = (U8)*STRING(text_node);
4463 (IS_TEXTF(text_node))
4465 : (IS_TEXTFL(text_node))
4466 ? PL_fold_locale[ST.c1]
4473 PerlIO_printf(Perl_debug_log,
4474 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4475 (int)(REPORT_CODE_OFF+(depth*2)),
4478 if (ST.c1 != CHRTEST_VOID
4479 && UCHARAT(PL_reginput) != ST.c1
4480 && UCHARAT(PL_reginput) != ST.c2)
4482 /* simulate B failing */
4484 PerlIO_printf(Perl_debug_log,
4485 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
4486 (int)(REPORT_CODE_OFF+(depth*2)),"",
4489 state_num = CURLYM_B_fail;
4490 goto reenter_switch;
4494 /* mark current A as captured */
4495 I32 paren = ST.me->flags;
4497 PL_regoffs[paren].start
4498 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4499 PL_regoffs[paren].end = PL_reginput - PL_bostr;
4500 /*dmq: *PL_reglastcloseparen = paren; */
4503 PL_regoffs[paren].end = -1;
4504 if (cur_eval && cur_eval->u.eval.close_paren &&
4505 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4514 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4517 case CURLYM_B_fail: /* just failed to match a B */
4518 REGCP_UNWIND(ST.cp);
4520 if (ST.count == ARG2(ST.me) /* max */)
4522 goto curlym_do_A; /* try to match a further A */
4524 /* backtrack one A */
4525 if (ST.count == ARG1(ST.me) /* min */)
4528 locinput = HOPc(locinput, -ST.alen);
4529 goto curlym_do_B; /* try to match B */
4532 #define ST st->u.curly
4534 #define CURLY_SETPAREN(paren, success) \
4537 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
4538 PL_regoffs[paren].end = locinput - PL_bostr; \
4539 *PL_reglastcloseparen = paren; \
4542 PL_regoffs[paren].end = -1; \
4545 case STAR: /* /A*B/ where A is width 1 */
4549 scan = NEXTOPER(scan);
4551 case PLUS: /* /A+B/ where A is width 1 */
4555 scan = NEXTOPER(scan);
4557 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4558 ST.paren = scan->flags; /* Which paren to set */
4559 if (ST.paren > PL_regsize)
4560 PL_regsize = ST.paren;
4561 if (ST.paren > *PL_reglastparen)
4562 *PL_reglastparen = ST.paren;
4563 ST.min = ARG1(scan); /* min to match */
4564 ST.max = ARG2(scan); /* max to match */
4565 if (cur_eval && cur_eval->u.eval.close_paren &&
4566 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4570 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4572 case CURLY: /* /A{m,n}B/ where A is width 1 */
4574 ST.min = ARG1(scan); /* min to match */
4575 ST.max = ARG2(scan); /* max to match */
4576 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4579 * Lookahead to avoid useless match attempts
4580 * when we know what character comes next.
4582 * Used to only do .*x and .*?x, but now it allows
4583 * for )'s, ('s and (?{ ... })'s to be in the way
4584 * of the quantifier and the EXACT-like node. -- japhy
4587 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4589 if (HAS_TEXT(next) || JUMPABLE(next)) {
4591 regnode *text_node = next;
4593 if (! HAS_TEXT(text_node))
4594 FIND_NEXT_IMPT(text_node);
4596 if (! HAS_TEXT(text_node))
4597 ST.c1 = ST.c2 = CHRTEST_VOID;
4599 if ( PL_regkind[OP(text_node)] != EXACT ) {
4600 ST.c1 = ST.c2 = CHRTEST_VOID;
4601 goto assume_ok_easy;
4604 s = (U8*)STRING(text_node);
4606 /* Currently we only get here when
4608 PL_rekind[OP(text_node)] == EXACT
4610 if this changes back then the macro for IS_TEXT and
4611 friends need to change. */
4614 if (IS_TEXTF(text_node))
4615 ST.c2 = PL_fold[ST.c1];
4616 else if (IS_TEXTFL(text_node))
4617 ST.c2 = PL_fold_locale[ST.c1];
4620 if (IS_TEXTF(text_node)) {
4621 STRLEN ulen1, ulen2;
4622 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4623 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4625 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4626 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4628 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4630 0 : UTF8_ALLOW_ANY);
4631 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4633 0 : UTF8_ALLOW_ANY);
4635 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4637 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4642 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4649 ST.c1 = ST.c2 = CHRTEST_VOID;
4654 PL_reginput = locinput;
4657 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4660 locinput = PL_reginput;
4662 if (ST.c1 == CHRTEST_VOID)
4663 goto curly_try_B_min;
4665 ST.oldloc = locinput;
4667 /* set ST.maxpos to the furthest point along the
4668 * string that could possibly match */
4669 if (ST.max == REG_INFTY) {
4670 ST.maxpos = PL_regeol - 1;
4672 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4676 int m = ST.max - ST.min;
4677 for (ST.maxpos = locinput;
4678 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4679 ST.maxpos += UTF8SKIP(ST.maxpos);
4682 ST.maxpos = locinput + ST.max - ST.min;
4683 if (ST.maxpos >= PL_regeol)
4684 ST.maxpos = PL_regeol - 1;
4686 goto curly_try_B_min_known;
4690 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4691 locinput = PL_reginput;
4692 if (ST.count < ST.min)
4694 if ((ST.count > ST.min)
4695 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4697 /* A{m,n} must come at the end of the string, there's
4698 * no point in backing off ... */
4700 /* ...except that $ and \Z can match before *and* after
4701 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4702 We may back off by one in this case. */
4703 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4707 goto curly_try_B_max;
4712 case CURLY_B_min_known_fail:
4713 /* failed to find B in a non-greedy match where c1,c2 valid */
4714 if (ST.paren && ST.count)
4715 PL_regoffs[ST.paren].end = -1;
4717 PL_reginput = locinput; /* Could be reset... */
4718 REGCP_UNWIND(ST.cp);
4719 /* Couldn't or didn't -- move forward. */
4720 ST.oldloc = locinput;
4722 locinput += UTF8SKIP(locinput);
4726 curly_try_B_min_known:
4727 /* find the next place where 'B' could work, then call B */
4731 n = (ST.oldloc == locinput) ? 0 : 1;
4732 if (ST.c1 == ST.c2) {
4734 /* set n to utf8_distance(oldloc, locinput) */
4735 while (locinput <= ST.maxpos &&
4736 utf8n_to_uvchr((U8*)locinput,
4737 UTF8_MAXBYTES, &len,
4738 uniflags) != (UV)ST.c1) {
4744 /* set n to utf8_distance(oldloc, locinput) */
4745 while (locinput <= ST.maxpos) {
4747 const UV c = utf8n_to_uvchr((U8*)locinput,
4748 UTF8_MAXBYTES, &len,
4750 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4758 if (ST.c1 == ST.c2) {
4759 while (locinput <= ST.maxpos &&
4760 UCHARAT(locinput) != ST.c1)
4764 while (locinput <= ST.maxpos
4765 && UCHARAT(locinput) != ST.c1
4766 && UCHARAT(locinput) != ST.c2)
4769 n = locinput - ST.oldloc;
4771 if (locinput > ST.maxpos)
4773 /* PL_reginput == oldloc now */
4776 if (regrepeat(rex, ST.A, n, depth) < n)
4779 PL_reginput = locinput;
4780 CURLY_SETPAREN(ST.paren, ST.count);
4781 if (cur_eval && cur_eval->u.eval.close_paren &&
4782 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4785 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4790 case CURLY_B_min_fail:
4791 /* failed to find B in a non-greedy match where c1,c2 invalid */
4792 if (ST.paren && ST.count)
4793 PL_regoffs[ST.paren].end = -1;
4795 REGCP_UNWIND(ST.cp);
4796 /* failed -- move forward one */
4797 PL_reginput = locinput;
4798 if (regrepeat(rex, ST.A, 1, depth)) {
4800 locinput = PL_reginput;
4801 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4802 ST.count > 0)) /* count overflow ? */
4805 CURLY_SETPAREN(ST.paren, ST.count);
4806 if (cur_eval && cur_eval->u.eval.close_paren &&
4807 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4810 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4818 /* a successful greedy match: now try to match B */
4819 if (cur_eval && cur_eval->u.eval.close_paren &&
4820 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4825 if (ST.c1 != CHRTEST_VOID)
4826 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4827 UTF8_MAXBYTES, 0, uniflags)
4828 : (UV) UCHARAT(PL_reginput);
4829 /* If it could work, try it. */
4830 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4831 CURLY_SETPAREN(ST.paren, ST.count);
4832 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4837 case CURLY_B_max_fail:
4838 /* failed to find B in a greedy match */
4839 if (ST.paren && ST.count)
4840 PL_regoffs[ST.paren].end = -1;
4842 REGCP_UNWIND(ST.cp);
4844 if (--ST.count < ST.min)
4846 PL_reginput = locinput = HOPc(locinput, -1);
4847 goto curly_try_B_max;
4854 /* we've just finished A in /(??{A})B/; now continue with B */
4856 st->u.eval.toggle_reg_flags
4857 = cur_eval->u.eval.toggle_reg_flags;
4858 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4860 st->u.eval.prev_rex = rex_sv; /* inner */
4861 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
4862 rex = (struct regexp *)SvANY(rex_sv);
4863 rexi = RXi_GET(rex);
4864 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4865 ReREFCNT_inc(rex_sv);
4866 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4867 REGCP_SET(st->u.eval.lastcp);
4868 PL_reginput = locinput;
4870 /* Restore parens of the outer rex without popping the
4872 tmpix = PL_savestack_ix;
4873 PL_savestack_ix = cur_eval->u.eval.lastcp;
4875 PL_savestack_ix = tmpix;
4877 st->u.eval.prev_eval = cur_eval;
4878 cur_eval = cur_eval->u.eval.prev_eval;
4880 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
4881 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
4882 if ( nochange_depth )
4885 PUSH_YES_STATE_GOTO(EVAL_AB,
4886 st->u.eval.prev_eval->u.eval.B); /* match B */
4889 if (locinput < reginfo->till) {
4890 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4891 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4893 (long)(locinput - PL_reg_starttry),
4894 (long)(reginfo->till - PL_reg_starttry),
4897 sayNO_SILENT; /* Cannot match: too short. */
4899 PL_reginput = locinput; /* put where regtry can find it */
4900 sayYES; /* Success! */
4902 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4904 PerlIO_printf(Perl_debug_log,
4905 "%*s %ssubpattern success...%s\n",
4906 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4907 PL_reginput = locinput; /* put where regtry can find it */
4908 sayYES; /* Success! */
4911 #define ST st->u.ifmatch
4913 case SUSPEND: /* (?>A) */
4915 PL_reginput = locinput;
4918 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4920 goto ifmatch_trivial_fail_test;
4922 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4924 ifmatch_trivial_fail_test:
4926 char * const s = HOPBACKc(locinput, scan->flags);
4931 sw = 1 - (bool)ST.wanted;
4935 next = scan + ARG(scan);
4943 PL_reginput = locinput;
4947 ST.logical = logical;
4948 /* execute body of (?...A) */
4949 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4952 case IFMATCH_A_fail: /* body of (?...A) failed */
4953 ST.wanted = !ST.wanted;
4956 case IFMATCH_A: /* body of (?...A) succeeded */
4958 sw = (bool)ST.wanted;
4960 else if (!ST.wanted)
4963 if (OP(ST.me) == SUSPEND)
4964 locinput = PL_reginput;
4966 locinput = PL_reginput = st->locinput;
4967 nextchr = UCHARAT(locinput);
4969 scan = ST.me + ARG(ST.me);
4972 continue; /* execute B */
4977 next = scan + ARG(scan);
4982 reginfo->cutpoint = PL_regeol;
4985 PL_reginput = locinput;
4987 sv_yes_mark = sv_commit = (SV*)rexi->data->data[ ARG( scan ) ];
4988 PUSH_STATE_GOTO(COMMIT_next,next);
4990 case COMMIT_next_fail:
4997 #define ST st->u.mark
4999 ST.prev_mark = mark_state;
5000 ST.mark_name = sv_commit = sv_yes_mark
5001 = (SV*)rexi->data->data[ ARG( scan ) ];
5003 ST.mark_loc = PL_reginput = locinput;
5004 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5006 case MARKPOINT_next:
5007 mark_state = ST.prev_mark;
5010 case MARKPOINT_next_fail:
5011 if (popmark && sv_eq(ST.mark_name,popmark))
5013 if (ST.mark_loc > startpoint)
5014 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5015 popmark = NULL; /* we found our mark */
5016 sv_commit = ST.mark_name;
5019 PerlIO_printf(Perl_debug_log,
5020 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5021 REPORT_CODE_OFF+depth*2, "",
5022 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5025 mark_state = ST.prev_mark;
5026 sv_yes_mark = mark_state ?
5027 mark_state->u.mark.mark_name : NULL;
5031 PL_reginput = locinput;
5033 /* (*SKIP) : if we fail we cut here*/
5034 ST.mark_name = NULL;
5035 ST.mark_loc = locinput;
5036 PUSH_STATE_GOTO(SKIP_next,next);
5038 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5039 otherwise do nothing. Meaning we need to scan
5041 regmatch_state *cur = mark_state;
5042 SV *find = (SV*)rexi->data->data[ ARG( scan ) ];
5045 if ( sv_eq( cur->u.mark.mark_name,
5048 ST.mark_name = find;
5049 PUSH_STATE_GOTO( SKIP_next, next );
5051 cur = cur->u.mark.prev_mark;
5054 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5056 case SKIP_next_fail:
5058 /* (*CUT:NAME) - Set up to search for the name as we
5059 collapse the stack*/
5060 popmark = ST.mark_name;
5062 /* (*CUT) - No name, we cut here.*/
5063 if (ST.mark_loc > startpoint)
5064 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5065 /* but we set sv_commit to latest mark_name if there
5066 is one so they can test to see how things lead to this
5069 sv_commit=mark_state->u.mark.mark_name;
5077 if ( n == (U32)what_len_TRICKYFOLD(locinput,do_utf8,ln) ) {
5079 } else if ( 0xDF == n && !do_utf8 && !UTF ) {
5082 U8 folded[UTF8_MAXBYTES_CASE+1];
5084 const char * const l = locinput;
5085 char *e = PL_regeol;
5086 to_uni_fold(n, folded, &foldlen);
5088 if (ibcmp_utf8((const char*) folded, 0, foldlen, 1,
5089 l, &e, 0, do_utf8)) {
5094 nextchr = UCHARAT(locinput);
5097 if ((n=is_LNBREAK(locinput,do_utf8))) {
5099 nextchr = UCHARAT(locinput);
5104 #define CASE_CLASS(nAmE) \
5106 if ((n=is_##nAmE(locinput,do_utf8))) { \
5108 nextchr = UCHARAT(locinput); \
5113 if ((n=is_##nAmE(locinput,do_utf8))) { \
5116 locinput += UTF8SKIP(locinput); \
5117 nextchr = UCHARAT(locinput); \
5122 CASE_CLASS(HORIZWS);
5126 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5127 PTR2UV(scan), OP(scan));
5128 Perl_croak(aTHX_ "regexp memory corruption");
5132 /* switch break jumps here */
5133 scan = next; /* prepare to execute the next op and ... */
5134 continue; /* ... jump back to the top, reusing st */
5138 /* push a state that backtracks on success */
5139 st->u.yes.prev_yes_state = yes_state;
5143 /* push a new regex state, then continue at scan */
5145 regmatch_state *newst;
5148 regmatch_state *cur = st;
5149 regmatch_state *curyes = yes_state;
5151 regmatch_slab *slab = PL_regmatch_slab;
5152 for (;curd > -1;cur--,curd--) {
5153 if (cur < SLAB_FIRST(slab)) {
5155 cur = SLAB_LAST(slab);
5157 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5158 REPORT_CODE_OFF + 2 + depth * 2,"",
5159 curd, PL_reg_name[cur->resume_state],
5160 (curyes == cur) ? "yes" : ""
5163 curyes = cur->u.yes.prev_yes_state;
5166 DEBUG_STATE_pp("push")
5169 st->locinput = locinput;
5171 if (newst > SLAB_LAST(PL_regmatch_slab))
5172 newst = S_push_slab(aTHX);
5173 PL_regmatch_state = newst;
5175 locinput = PL_reginput;
5176 nextchr = UCHARAT(locinput);
5184 * We get here only if there's trouble -- normally "case END" is
5185 * the terminating point.
5187 Perl_croak(aTHX_ "corrupted regexp pointers");
5193 /* we have successfully completed a subexpression, but we must now
5194 * pop to the state marked by yes_state and continue from there */
5195 assert(st != yes_state);
5197 while (st != yes_state) {
5199 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5200 PL_regmatch_slab = PL_regmatch_slab->prev;
5201 st = SLAB_LAST(PL_regmatch_slab);
5205 DEBUG_STATE_pp("pop (no final)");
5207 DEBUG_STATE_pp("pop (yes)");
5213 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5214 || yes_state > SLAB_LAST(PL_regmatch_slab))
5216 /* not in this slab, pop slab */
5217 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5218 PL_regmatch_slab = PL_regmatch_slab->prev;
5219 st = SLAB_LAST(PL_regmatch_slab);
5221 depth -= (st - yes_state);
5224 yes_state = st->u.yes.prev_yes_state;
5225 PL_regmatch_state = st;
5228 locinput= st->locinput;
5229 nextchr = UCHARAT(locinput);
5231 state_num = st->resume_state + no_final;
5232 goto reenter_switch;
5235 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5236 PL_colors[4], PL_colors[5]));
5238 if (PL_reg_eval_set) {
5239 /* each successfully executed (?{...}) block does the equivalent of
5240 * local $^R = do {...}
5241 * When popping the save stack, all these locals would be undone;
5242 * bypass this by setting the outermost saved $^R to the latest
5244 if (oreplsv != GvSV(PL_replgv))
5245 sv_setsv(oreplsv, GvSV(PL_replgv));
5252 PerlIO_printf(Perl_debug_log,
5253 "%*s %sfailed...%s\n",
5254 REPORT_CODE_OFF+depth*2, "",
5255 PL_colors[4], PL_colors[5])
5267 /* there's a previous state to backtrack to */
5269 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5270 PL_regmatch_slab = PL_regmatch_slab->prev;
5271 st = SLAB_LAST(PL_regmatch_slab);
5273 PL_regmatch_state = st;
5274 locinput= st->locinput;
5275 nextchr = UCHARAT(locinput);
5277 DEBUG_STATE_pp("pop");
5279 if (yes_state == st)
5280 yes_state = st->u.yes.prev_yes_state;
5282 state_num = st->resume_state + 1; /* failure = success + 1 */
5283 goto reenter_switch;
5288 if (rex->intflags & PREGf_VERBARG_SEEN) {
5289 SV *sv_err = get_sv("REGERROR", 1);
5290 SV *sv_mrk = get_sv("REGMARK", 1);
5292 sv_commit = &PL_sv_no;
5294 sv_yes_mark = &PL_sv_yes;
5297 sv_commit = &PL_sv_yes;
5298 sv_yes_mark = &PL_sv_no;
5300 sv_setsv(sv_err, sv_commit);
5301 sv_setsv(sv_mrk, sv_yes_mark);
5304 /* clean up; in particular, free all slabs above current one */
5305 LEAVE_SCOPE(oldsave);
5311 - regrepeat - repeatedly match something simple, report how many
5314 * [This routine now assumes that it will only match on things of length 1.
5315 * That was true before, but now we assume scan - reginput is the count,
5316 * rather than incrementing count on every character. [Er, except utf8.]]
5319 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5322 register char *scan;
5324 register char *loceol = PL_regeol;
5325 register I32 hardcount = 0;
5326 register bool do_utf8 = PL_reg_match_utf8;
5328 PERL_UNUSED_ARG(depth);
5332 if (max == REG_INFTY)
5334 else if (max < loceol - scan)
5335 loceol = scan + max;
5340 while (scan < loceol && hardcount < max && *scan != '\n') {
5341 scan += UTF8SKIP(scan);
5345 while (scan < loceol && *scan != '\n')
5352 while (scan < loceol && hardcount < max) {
5353 scan += UTF8SKIP(scan);
5363 case EXACT: /* length of string is 1 */
5365 while (scan < loceol && UCHARAT(scan) == c)
5368 case EXACTF: /* length of string is 1 */
5370 while (scan < loceol &&
5371 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5374 case EXACTFL: /* length of string is 1 */
5375 PL_reg_flags |= RF_tainted;
5377 while (scan < loceol &&
5378 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5384 while (hardcount < max && scan < loceol &&
5385 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
5386 scan += UTF8SKIP(scan);
5390 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5397 LOAD_UTF8_CHARCLASS_ALNUM();
5398 while (hardcount < max && scan < loceol &&
5399 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5400 scan += UTF8SKIP(scan);
5404 while (scan < loceol && isALNUM(*scan))
5409 PL_reg_flags |= RF_tainted;
5412 while (hardcount < max && scan < loceol &&
5413 isALNUM_LC_utf8((U8*)scan)) {
5414 scan += UTF8SKIP(scan);
5418 while (scan < loceol && isALNUM_LC(*scan))
5425 LOAD_UTF8_CHARCLASS_ALNUM();
5426 while (hardcount < max && scan < loceol &&
5427 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5428 scan += UTF8SKIP(scan);
5432 while (scan < loceol && !isALNUM(*scan))
5437 PL_reg_flags |= RF_tainted;
5440 while (hardcount < max && scan < loceol &&
5441 !isALNUM_LC_utf8((U8*)scan)) {
5442 scan += UTF8SKIP(scan);
5446 while (scan < loceol && !isALNUM_LC(*scan))
5453 LOAD_UTF8_CHARCLASS_SPACE();
5454 while (hardcount < max && scan < loceol &&
5456 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5457 scan += UTF8SKIP(scan);
5461 while (scan < loceol && isSPACE(*scan))
5466 PL_reg_flags |= RF_tainted;
5469 while (hardcount < max && scan < loceol &&
5470 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5471 scan += UTF8SKIP(scan);
5475 while (scan < loceol && isSPACE_LC(*scan))
5482 LOAD_UTF8_CHARCLASS_SPACE();
5483 while (hardcount < max && scan < loceol &&
5485 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5486 scan += UTF8SKIP(scan);
5490 while (scan < loceol && !isSPACE(*scan))
5495 PL_reg_flags |= RF_tainted;
5498 while (hardcount < max && scan < loceol &&
5499 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5500 scan += UTF8SKIP(scan);
5504 while (scan < loceol && !isSPACE_LC(*scan))
5511 LOAD_UTF8_CHARCLASS_DIGIT();
5512 while (hardcount < max && scan < loceol &&
5513 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5514 scan += UTF8SKIP(scan);
5518 while (scan < loceol && isDIGIT(*scan))
5525 LOAD_UTF8_CHARCLASS_DIGIT();
5526 while (hardcount < max && scan < loceol &&
5527 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5528 scan += UTF8SKIP(scan);
5532 while (scan < loceol && !isDIGIT(*scan))
5538 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
5544 LNBREAK can match two latin chars, which is ok,
5545 because we have a null terminated string, but we
5546 have to use hardcount in this situation
5548 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
5557 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
5562 while (scan < loceol && is_HORIZWS_latin1(scan))
5569 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
5570 scan += UTF8SKIP(scan);
5574 while (scan < loceol && !is_HORIZWS_latin1(scan))
5582 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
5587 while (scan < loceol && is_VERTWS_latin1(scan))
5595 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
5596 scan += UTF8SKIP(scan);
5600 while (scan < loceol && !is_VERTWS_latin1(scan))
5606 default: /* Called on something of 0 width. */
5607 break; /* So match right here or not at all. */
5613 c = scan - PL_reginput;
5617 GET_RE_DEBUG_FLAGS_DECL;
5619 SV * const prop = sv_newmortal();
5620 regprop(prog, prop, p);
5621 PerlIO_printf(Perl_debug_log,
5622 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5623 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5631 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5633 - regclass_swash - prepare the utf8 swash
5637 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5643 RXi_GET_DECL(prog,progi);
5644 const struct reg_data * const data = prog ? progi->data : NULL;
5646 if (data && data->count) {
5647 const U32 n = ARG(node);
5649 if (data->what[n] == 's') {
5650 SV * const rv = (SV*)data->data[n];
5651 AV * const av = (AV*)SvRV((SV*)rv);
5652 SV **const ary = AvARRAY(av);
5655 /* See the end of regcomp.c:S_regclass() for
5656 * documentation of these array elements. */
5659 a = SvROK(ary[1]) ? &ary[1] : NULL;
5660 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
5664 else if (si && doinit) {
5665 sw = swash_init("utf8", "", si, 1, 0);
5666 (void)av_store(av, 1, sw);
5683 - reginclass - determine if a character falls into a character class
5685 The n is the ANYOF regnode, the p is the target string, lenp
5686 is pointer to the maximum length of how far to go in the p
5687 (if the lenp is zero, UTF8SKIP(p) is used),
5688 do_utf8 tells whether the target string is in UTF-8.
5693 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5696 const char flags = ANYOF_FLAGS(n);
5702 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5703 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5704 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5705 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5706 if (len == (STRLEN)-1)
5707 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5710 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5711 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5714 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5715 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5718 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5722 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5725 if (swash_fetch(sw, p, do_utf8))
5727 else if (flags & ANYOF_FOLD) {
5728 if (!match && lenp && av) {
5730 for (i = 0; i <= av_len(av); i++) {
5731 SV* const sv = *av_fetch(av, i, FALSE);
5733 const char * const s = SvPV_const(sv, len);
5735 if (len <= plen && memEQ(s, (char*)p, len)) {
5743 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5746 to_utf8_fold(p, tmpbuf, &tmplen);
5747 if (swash_fetch(sw, tmpbuf, do_utf8))
5753 if (match && lenp && *lenp == 0)
5754 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5756 if (!match && c < 256) {
5757 if (ANYOF_BITMAP_TEST(n, c))
5759 else if (flags & ANYOF_FOLD) {
5762 if (flags & ANYOF_LOCALE) {
5763 PL_reg_flags |= RF_tainted;
5764 f = PL_fold_locale[c];
5768 if (f != c && ANYOF_BITMAP_TEST(n, f))
5772 if (!match && (flags & ANYOF_CLASS)) {
5773 PL_reg_flags |= RF_tainted;
5775 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5776 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5777 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5778 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5779 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5780 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5781 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5782 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5783 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5784 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5785 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5786 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5787 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5788 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5789 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5790 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5791 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5792 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5793 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5794 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5795 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5796 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5797 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5798 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5799 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5800 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5801 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5802 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5803 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5804 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5805 ) /* How's that for a conditional? */
5812 return (flags & ANYOF_INVERT) ? !match : match;
5816 S_reghop3(U8 *s, I32 off, const U8* lim)
5820 while (off-- && s < lim) {
5821 /* XXX could check well-formedness here */
5826 while (off++ && s > lim) {
5828 if (UTF8_IS_CONTINUED(*s)) {
5829 while (s > lim && UTF8_IS_CONTINUATION(*s))
5832 /* XXX could check well-formedness here */
5839 /* there are a bunch of places where we use two reghop3's that should
5840 be replaced with this routine. but since thats not done yet
5841 we ifdef it out - dmq
5844 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5848 while (off-- && s < rlim) {
5849 /* XXX could check well-formedness here */
5854 while (off++ && s > llim) {
5856 if (UTF8_IS_CONTINUED(*s)) {
5857 while (s > llim && UTF8_IS_CONTINUATION(*s))
5860 /* XXX could check well-formedness here */
5868 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5872 while (off-- && s < lim) {
5873 /* XXX could check well-formedness here */
5880 while (off++ && s > lim) {
5882 if (UTF8_IS_CONTINUED(*s)) {
5883 while (s > lim && UTF8_IS_CONTINUATION(*s))
5886 /* XXX could check well-formedness here */
5895 restore_pos(pTHX_ void *arg)
5898 regexp * const rex = (regexp *)arg;
5899 if (PL_reg_eval_set) {
5900 if (PL_reg_oldsaved) {
5901 rex->subbeg = PL_reg_oldsaved;
5902 rex->sublen = PL_reg_oldsavedlen;
5903 #ifdef PERL_OLD_COPY_ON_WRITE
5904 rex->saved_copy = PL_nrs;
5906 RXp_MATCH_COPIED_on(rex);
5908 PL_reg_magic->mg_len = PL_reg_oldpos;
5909 PL_reg_eval_set = 0;
5910 PL_curpm = PL_reg_oldcurpm;
5915 S_to_utf8_substr(pTHX_ register regexp *prog)
5919 if (prog->substrs->data[i].substr
5920 && !prog->substrs->data[i].utf8_substr) {
5921 SV* const sv = newSVsv(prog->substrs->data[i].substr);
5922 prog->substrs->data[i].utf8_substr = sv;
5923 sv_utf8_upgrade(sv);
5924 if (SvVALID(prog->substrs->data[i].substr)) {
5925 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
5926 if (flags & FBMcf_TAIL) {
5927 /* Trim the trailing \n that fbm_compile added last
5929 SvCUR_set(sv, SvCUR(sv) - 1);
5930 /* Whilst this makes the SV technically "invalid" (as its
5931 buffer is no longer followed by "\0") when fbm_compile()
5932 adds the "\n" back, a "\0" is restored. */
5934 fbm_compile(sv, flags);
5936 if (prog->substrs->data[i].substr == prog->check_substr)
5937 prog->check_utf8 = sv;
5943 S_to_byte_substr(pTHX_ register regexp *prog)
5948 if (prog->substrs->data[i].utf8_substr
5949 && !prog->substrs->data[i].substr) {
5950 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
5951 if (sv_utf8_downgrade(sv, TRUE)) {
5952 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
5954 = BmFLAGS(prog->substrs->data[i].utf8_substr);
5955 if (flags & FBMcf_TAIL) {
5956 /* Trim the trailing \n that fbm_compile added last
5958 SvCUR_set(sv, SvCUR(sv) - 1);
5960 fbm_compile(sv, flags);
5966 prog->substrs->data[i].substr = sv;
5967 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
5968 prog->check_substr = sv;
5975 * c-indentation-style: bsd
5977 * indent-tabs-mode: t
5980 * ex: set ts=8 sts=4 sw=4 noet: