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 * building DynaLoader will fail:
289 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
291 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
293 PL_regoffs[i].start = -1;
294 PL_regoffs[i].end = -1;
300 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
303 * pregexec and friends
306 #ifndef PERL_IN_XSUB_RE
308 - pregexec - match a regexp against a string
311 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
312 char *strbeg, I32 minend, SV *screamer, U32 nosave)
313 /* strend: pointer to null at end of string */
314 /* strbeg: real beginning of string */
315 /* minend: end of match must be >=minend after stringarg. */
316 /* nosave: For optimizations. */
319 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
320 nosave ? 0 : REXEC_COPY_STR);
325 * Need to implement the following flags for reg_anch:
327 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
329 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
330 * INTUIT_AUTORITATIVE_ML
331 * INTUIT_ONCE_NOML - Intuit can match in one location only.
334 * Another flag for this function: SECOND_TIME (so that float substrs
335 * with giant delta may be not rechecked).
338 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
340 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
341 Otherwise, only SvCUR(sv) is used to get strbeg. */
343 /* XXXX We assume that strpos is strbeg unless sv. */
345 /* XXXX Some places assume that there is a fixed substring.
346 An update may be needed if optimizer marks as "INTUITable"
347 RExen without fixed substrings. Similarly, it is assumed that
348 lengths of all the strings are no more than minlen, thus they
349 cannot come from lookahead.
350 (Or minlen should take into account lookahead.)
351 NOTE: Some of this comment is not correct. minlen does now take account
352 of lookahead/behind. Further research is required. -- demerphq
356 /* A failure to find a constant substring means that there is no need to make
357 an expensive call to REx engine, thus we celebrate a failure. Similarly,
358 finding a substring too deep into the string means that less calls to
359 regtry() should be needed.
361 REx compiler's optimizer found 4 possible hints:
362 a) Anchored substring;
364 c) Whether we are anchored (beginning-of-line or \G);
365 d) First node (of those at offset 0) which may distingush positions;
366 We use a)b)d) and multiline-part of c), and try to find a position in the
367 string which does not contradict any of them.
370 /* Most of decisions we do here should have been done at compile time.
371 The nodes of the REx which we used for the search should have been
372 deleted from the finite automaton. */
375 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
376 char *strend, U32 flags, re_scream_pos_data *data)
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(prog,do_utf8);
400 if (prog->extflags & RXf_UTF8) {
401 PL_reg_flags |= RF_utf8;
404 debug_start_match(prog, 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, prog->precomp);
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 && RX_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]));
995 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
996 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
997 switch (trie_type) { \
998 case trie_utf8_fold: \
1000 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1005 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1006 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1007 foldlen -= UNISKIP( uvc ); \
1008 uscan = foldbuf + UNISKIP( uvc ); \
1012 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1020 charid = trie->charmap[ uvc ]; \
1024 if (widecharmap) { \
1025 SV** const svpp = hv_fetch(widecharmap, \
1026 (char*)&uvc, sizeof(UV), 0); \
1028 charid = (U16)SvIV(*svpp); \
1033 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1036 ibcmp_utf8(s, NULL, 0, do_utf8, \
1037 m, NULL, ln, (bool)UTF)) \
1038 && (!reginfo || regtry(reginfo, &s)) ) \
1041 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1042 uvchr_to_utf8(tmpbuf, c); \
1043 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1045 && (f == c1 || f == c2) \
1046 && (ln == foldlen || \
1047 !ibcmp_utf8((char *) foldbuf, \
1048 NULL, foldlen, do_utf8, \
1050 NULL, ln, (bool)UTF)) \
1051 && (!reginfo || regtry(reginfo, &s)) ) \
1056 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1060 && (ln == 1 || !(OP(c) == EXACTF \
1062 : ibcmp_locale(s, m, ln))) \
1063 && (!reginfo || regtry(reginfo, &s)) ) \
1069 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1071 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1077 #define REXEC_FBC_SCAN(CoDe) \
1079 while (s < strend) { \
1085 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1086 REXEC_FBC_UTF8_SCAN( \
1088 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1097 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1100 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1109 #define REXEC_FBC_TRYIT \
1110 if ((!reginfo || regtry(reginfo, &s))) \
1113 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1116 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1119 REXEC_FBC_CLASS_SCAN(CoNd); \
1123 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1124 PL_reg_flags |= RF_tainted; \
1126 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1129 REXEC_FBC_CLASS_SCAN(CoNd); \
1133 #define DUMP_EXEC_POS(li,s,doutf8) \
1134 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1136 /* We know what class REx starts with. Try to find this position... */
1137 /* if reginfo is NULL, its a dryrun */
1138 /* annoyingly all the vars in this routine have different names from their counterparts
1139 in regmatch. /grrr */
1142 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1143 const char *strend, regmatch_info *reginfo)
1146 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1150 register STRLEN uskip;
1154 register I32 tmp = 1; /* Scratch variable? */
1155 register const bool do_utf8 = PL_reg_match_utf8;
1156 RXi_GET_DECL(prog,progi);
1158 /* We know what class it must start with. */
1162 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1163 !UTF8_IS_INVARIANT((U8)s[0]) ?
1164 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1165 REGINCLASS(prog, c, (U8*)s));
1168 while (s < strend) {
1171 if (REGINCLASS(prog, c, (U8*)s) ||
1172 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1173 /* The assignment of 2 is intentional:
1174 * for the folded sharp s, the skip is 2. */
1175 (skip = SHARP_S_SKIP))) {
1176 if (tmp && (!reginfo || regtry(reginfo, &s)))
1189 if (tmp && (!reginfo || regtry(reginfo, &s)))
1197 ln = STR_LEN(c); /* length to match in octets/bytes */
1198 lnc = (I32) ln; /* length to match in characters */
1200 STRLEN ulen1, ulen2;
1202 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1203 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1204 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1206 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1207 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1209 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1211 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1214 while (sm < ((U8 *) m + ln)) {
1229 c2 = PL_fold_locale[c1];
1231 e = HOP3c(strend, -((I32)lnc), s);
1233 if (!reginfo && e < s)
1234 e = s; /* Due to minlen logic of intuit() */
1236 /* The idea in the EXACTF* cases is to first find the
1237 * first character of the EXACTF* node and then, if
1238 * necessary, case-insensitively compare the full
1239 * text of the node. The c1 and c2 are the first
1240 * characters (though in Unicode it gets a bit
1241 * more complicated because there are more cases
1242 * than just upper and lower: one needs to use
1243 * the so-called folding case for case-insensitive
1244 * matching (called "loose matching" in Unicode).
1245 * ibcmp_utf8() will do just that. */
1249 U8 tmpbuf [UTF8_MAXBYTES+1];
1250 STRLEN len, foldlen;
1251 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1253 /* Upper and lower of 1st char are equal -
1254 * probably not a "letter". */
1256 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1258 REXEC_FBC_EXACTISH_CHECK(c == c1);
1263 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1266 /* Handle some of the three Greek sigmas cases.
1267 * Note that not all the possible combinations
1268 * are handled here: some of them are handled
1269 * by the standard folding rules, and some of
1270 * them (the character class or ANYOF cases)
1271 * are handled during compiletime in
1272 * regexec.c:S_regclass(). */
1273 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1274 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1275 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1277 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1283 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1285 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1289 PL_reg_flags |= RF_tainted;
1296 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1297 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1299 tmp = ((OP(c) == BOUND ?
1300 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1301 LOAD_UTF8_CHARCLASS_ALNUM();
1302 REXEC_FBC_UTF8_SCAN(
1303 if (tmp == !(OP(c) == BOUND ?
1304 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1305 isALNUM_LC_utf8((U8*)s)))
1313 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1314 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1317 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1323 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1327 PL_reg_flags |= RF_tainted;
1334 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1335 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1337 tmp = ((OP(c) == NBOUND ?
1338 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1339 LOAD_UTF8_CHARCLASS_ALNUM();
1340 REXEC_FBC_UTF8_SCAN(
1341 if (tmp == !(OP(c) == NBOUND ?
1342 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1343 isALNUM_LC_utf8((U8*)s)))
1345 else REXEC_FBC_TRYIT;
1349 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1350 tmp = ((OP(c) == NBOUND ?
1351 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1354 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1356 else REXEC_FBC_TRYIT;
1359 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1363 REXEC_FBC_CSCAN_PRELOAD(
1364 LOAD_UTF8_CHARCLASS_ALNUM(),
1365 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1369 REXEC_FBC_CSCAN_TAINT(
1370 isALNUM_LC_utf8((U8*)s),
1374 REXEC_FBC_CSCAN_PRELOAD(
1375 LOAD_UTF8_CHARCLASS_ALNUM(),
1376 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1380 REXEC_FBC_CSCAN_TAINT(
1381 !isALNUM_LC_utf8((U8*)s),
1385 REXEC_FBC_CSCAN_PRELOAD(
1386 LOAD_UTF8_CHARCLASS_SPACE(),
1387 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1391 REXEC_FBC_CSCAN_TAINT(
1392 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1396 REXEC_FBC_CSCAN_PRELOAD(
1397 LOAD_UTF8_CHARCLASS_SPACE(),
1398 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1402 REXEC_FBC_CSCAN_TAINT(
1403 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1407 REXEC_FBC_CSCAN_PRELOAD(
1408 LOAD_UTF8_CHARCLASS_DIGIT(),
1409 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1413 REXEC_FBC_CSCAN_TAINT(
1414 isDIGIT_LC_utf8((U8*)s),
1418 REXEC_FBC_CSCAN_PRELOAD(
1419 LOAD_UTF8_CHARCLASS_DIGIT(),
1420 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1424 REXEC_FBC_CSCAN_TAINT(
1425 !isDIGIT_LC_utf8((U8*)s),
1431 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1432 trie_type = do_utf8 ?
1433 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1435 /* what trie are we using right now */
1437 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1439 = (reg_trie_data*)progi->data->data[ aho->trie ];
1440 HV *widecharmap = (HV*) progi->data->data[ aho->trie + 1 ];
1442 const char *last_start = strend - trie->minlen;
1444 const char *real_start = s;
1446 STRLEN maxlen = trie->maxlen;
1448 U8 **points; /* map of where we were in the input string
1449 when reading a given char. For ASCII this
1450 is unnecessary overhead as the relationship
1451 is always 1:1, but for unicode, especially
1452 case folded unicode this is not true. */
1453 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1457 GET_RE_DEBUG_FLAGS_DECL;
1459 /* We can't just allocate points here. We need to wrap it in
1460 * an SV so it gets freed properly if there is a croak while
1461 * running the match */
1464 sv_points=newSV(maxlen * sizeof(U8 *));
1465 SvCUR_set(sv_points,
1466 maxlen * sizeof(U8 *));
1467 SvPOK_on(sv_points);
1468 sv_2mortal(sv_points);
1469 points=(U8**)SvPV_nolen(sv_points );
1470 if ( trie_type != trie_utf8_fold
1471 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1474 bitmap=(U8*)trie->bitmap;
1476 bitmap=(U8*)ANYOF_BITMAP(c);
1478 /* this is the Aho-Corasick algorithm modified a touch
1479 to include special handling for long "unknown char"
1480 sequences. The basic idea being that we use AC as long
1481 as we are dealing with a possible matching char, when
1482 we encounter an unknown char (and we have not encountered
1483 an accepting state) we scan forward until we find a legal
1485 AC matching is basically that of trie matching, except
1486 that when we encounter a failing transition, we fall back
1487 to the current states "fail state", and try the current char
1488 again, a process we repeat until we reach the root state,
1489 state 1, or a legal transition. If we fail on the root state
1490 then we can either terminate if we have reached an accepting
1491 state previously, or restart the entire process from the beginning
1495 while (s <= last_start) {
1496 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1504 U8 *uscan = (U8*)NULL;
1505 U8 *leftmost = NULL;
1507 U32 accepted_word= 0;
1511 while ( state && uc <= (U8*)strend ) {
1513 U32 word = aho->states[ state ].wordnum;
1517 DEBUG_TRIE_EXECUTE_r(
1518 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1519 dump_exec_pos( (char *)uc, c, strend, real_start,
1520 (char *)uc, do_utf8 );
1521 PerlIO_printf( Perl_debug_log,
1522 " Scanning for legal start char...\n");
1525 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1530 if (uc >(U8*)last_start) break;
1534 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1535 if (!leftmost || lpos < leftmost) {
1536 DEBUG_r(accepted_word=word);
1542 points[pointpos++ % maxlen]= uc;
1543 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1544 uscan, len, uvc, charid, foldlen,
1546 DEBUG_TRIE_EXECUTE_r({
1547 dump_exec_pos( (char *)uc, c, strend, real_start,
1549 PerlIO_printf(Perl_debug_log,
1550 " Charid:%3u CP:%4"UVxf" ",
1556 word = aho->states[ state ].wordnum;
1558 base = aho->states[ state ].trans.base;
1560 DEBUG_TRIE_EXECUTE_r({
1562 dump_exec_pos( (char *)uc, c, strend, real_start,
1564 PerlIO_printf( Perl_debug_log,
1565 "%sState: %4"UVxf", word=%"UVxf,
1566 failed ? " Fail transition to " : "",
1567 (UV)state, (UV)word);
1572 (base + charid > trie->uniquecharcount )
1573 && (base + charid - 1 - trie->uniquecharcount
1575 && trie->trans[base + charid - 1 -
1576 trie->uniquecharcount].check == state
1577 && (tmp=trie->trans[base + charid - 1 -
1578 trie->uniquecharcount ].next))
1580 DEBUG_TRIE_EXECUTE_r(
1581 PerlIO_printf( Perl_debug_log," - legal\n"));
1586 DEBUG_TRIE_EXECUTE_r(
1587 PerlIO_printf( Perl_debug_log," - fail\n"));
1589 state = aho->fail[state];
1593 /* we must be accepting here */
1594 DEBUG_TRIE_EXECUTE_r(
1595 PerlIO_printf( Perl_debug_log," - accepting\n"));
1604 if (!state) state = 1;
1607 if ( aho->states[ state ].wordnum ) {
1608 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1609 if (!leftmost || lpos < leftmost) {
1610 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1615 s = (char*)leftmost;
1616 DEBUG_TRIE_EXECUTE_r({
1618 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1619 (UV)accepted_word, (IV)(s - real_start)
1622 if (!reginfo || regtry(reginfo, &s)) {
1628 DEBUG_TRIE_EXECUTE_r({
1629 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1632 DEBUG_TRIE_EXECUTE_r(
1633 PerlIO_printf( Perl_debug_log,"No match.\n"));
1642 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1651 S_swap_match_buff (pTHX_ regexp *prog) {
1652 regexp_paren_pair *t;
1655 /* We have to be careful. If the previous successful match
1656 was from this regex we don't want a subsequent paritally
1657 successful match to clobber the old results.
1658 So when we detect this possibility we add a swap buffer
1659 to the re, and switch the buffer each match. If we fail
1660 we switch it back, otherwise we leave it swapped.
1662 Newxz(prog->swap, (prog->nparens + 1), regexp_paren_pair);
1665 prog->swap = prog->offs;
1671 - regexec_flags - match a regexp against a string
1674 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1675 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1676 /* strend: pointer to null at end of string */
1677 /* strbeg: real beginning of string */
1678 /* minend: end of match must be >=minend after stringarg. */
1679 /* data: May be used for some additional optimizations.
1680 Currently its only used, with a U32 cast, for transmitting
1681 the ganch offset when doing a /g match. This will change */
1682 /* nosave: For optimizations. */
1685 /*register*/ char *s;
1686 register regnode *c;
1687 /*register*/ char *startpos = stringarg;
1688 I32 minlen; /* must match at least this many chars */
1689 I32 dontbother = 0; /* how many characters not to try at end */
1690 I32 end_shift = 0; /* Same for the end. */ /* CC */
1691 I32 scream_pos = -1; /* Internal iterator of scream. */
1692 char *scream_olds = NULL;
1693 SV* const oreplsv = GvSV(PL_replgv);
1694 const bool do_utf8 = (bool)DO_UTF8(sv);
1696 RXi_GET_DECL(prog,progi);
1697 regmatch_info reginfo; /* create some info to pass to regtry etc */
1698 bool swap_on_fail = 0;
1700 GET_RE_DEBUG_FLAGS_DECL;
1702 PERL_UNUSED_ARG(data);
1704 /* Be paranoid... */
1705 if (prog == NULL || startpos == NULL) {
1706 Perl_croak(aTHX_ "NULL regexp parameter");
1710 multiline = prog->extflags & RXf_PMf_MULTILINE;
1711 reginfo.prog = prog;
1713 RX_MATCH_UTF8_set(prog, do_utf8);
1715 debug_start_match(prog, do_utf8, startpos, strend,
1719 minlen = prog->minlen;
1721 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1722 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1723 "String too short [regexec_flags]...\n"));
1728 /* Check validity of program. */
1729 if (UCHARAT(progi->program) != REG_MAGIC) {
1730 Perl_croak(aTHX_ "corrupted regexp program");
1734 PL_reg_eval_set = 0;
1737 if (prog->extflags & RXf_UTF8)
1738 PL_reg_flags |= RF_utf8;
1740 /* Mark beginning of line for ^ and lookbehind. */
1741 reginfo.bol = startpos; /* XXX not used ??? */
1745 /* Mark end of line for $ (and such) */
1748 /* see how far we have to get to not match where we matched before */
1749 reginfo.till = startpos+minend;
1751 /* If there is a "must appear" string, look for it. */
1754 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1757 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1758 reginfo.ganch = startpos + prog->gofs;
1759 else if (sv && SvTYPE(sv) >= SVt_PVMG
1761 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1762 && mg->mg_len >= 0) {
1763 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1764 if (prog->extflags & RXf_ANCH_GPOS) {
1765 if (s > reginfo.ganch)
1767 s = reginfo.ganch - prog->gofs;
1771 reginfo.ganch = strbeg + PTR2UV(data);
1772 } else /* pos() not defined */
1773 reginfo.ganch = strbeg;
1775 if (PL_curpm && (PM_GETRE(PL_curpm) == prog)) {
1777 swap_match_buff(prog); /* do we need a save destructor here for
1780 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1781 re_scream_pos_data d;
1783 d.scream_olds = &scream_olds;
1784 d.scream_pos = &scream_pos;
1785 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1787 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1788 goto phooey; /* not present */
1794 /* Simplest case: anchored match need be tried only once. */
1795 /* [unless only anchor is BOL and multiline is set] */
1796 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
1797 if (s == startpos && regtry(®info, &startpos))
1799 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
1800 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
1805 dontbother = minlen - 1;
1806 end = HOP3c(strend, -dontbother, strbeg) - 1;
1807 /* for multiline we only have to try after newlines */
1808 if (prog->check_substr || prog->check_utf8) {
1812 if (regtry(®info, &s))
1817 if (prog->extflags & RXf_USE_INTUIT) {
1818 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1829 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1830 if (regtry(®info, &s))
1837 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
1839 /* the warning about reginfo.ganch being used without intialization
1840 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
1841 and we only enter this block when the same bit is set. */
1842 char *tmp_s = reginfo.ganch - prog->gofs;
1843 if (regtry(®info, &tmp_s))
1848 /* Messy cases: unanchored match. */
1849 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
1850 /* we have /x+whatever/ */
1851 /* it must be a one character string (XXXX Except UTF?) */
1856 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1857 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1858 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1863 DEBUG_EXECUTE_r( did_match = 1 );
1864 if (regtry(®info, &s)) goto got_it;
1866 while (s < strend && *s == ch)
1874 DEBUG_EXECUTE_r( did_match = 1 );
1875 if (regtry(®info, &s)) goto got_it;
1877 while (s < strend && *s == ch)
1882 DEBUG_EXECUTE_r(if (!did_match)
1883 PerlIO_printf(Perl_debug_log,
1884 "Did not find anchored character...\n")
1887 else if (prog->anchored_substr != NULL
1888 || prog->anchored_utf8 != NULL
1889 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1890 && prog->float_max_offset < strend - s)) {
1895 char *last1; /* Last position checked before */
1899 if (prog->anchored_substr || prog->anchored_utf8) {
1900 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1901 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1902 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1903 back_max = back_min = prog->anchored_offset;
1905 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1906 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1907 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1908 back_max = prog->float_max_offset;
1909 back_min = prog->float_min_offset;
1913 if (must == &PL_sv_undef)
1914 /* could not downgrade utf8 check substring, so must fail */
1920 last = HOP3c(strend, /* Cannot start after this */
1921 -(I32)(CHR_SVLEN(must)
1922 - (SvTAIL(must) != 0) + back_min), strbeg);
1925 last1 = HOPc(s, -1);
1927 last1 = s - 1; /* bogus */
1929 /* XXXX check_substr already used to find "s", can optimize if
1930 check_substr==must. */
1932 dontbother = end_shift;
1933 strend = HOPc(strend, -dontbother);
1934 while ( (s <= last) &&
1935 ((flags & REXEC_SCREAM)
1936 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1937 end_shift, &scream_pos, 0))
1938 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1939 (unsigned char*)strend, must,
1940 multiline ? FBMrf_MULTILINE : 0))) ) {
1941 /* we may be pointing at the wrong string */
1942 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1943 s = strbeg + (s - SvPVX_const(sv));
1944 DEBUG_EXECUTE_r( did_match = 1 );
1945 if (HOPc(s, -back_max) > last1) {
1946 last1 = HOPc(s, -back_min);
1947 s = HOPc(s, -back_max);
1950 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1952 last1 = HOPc(s, -back_min);
1956 while (s <= last1) {
1957 if (regtry(®info, &s))
1963 while (s <= last1) {
1964 if (regtry(®info, &s))
1970 DEBUG_EXECUTE_r(if (!did_match) {
1971 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1972 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1973 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1974 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1975 ? "anchored" : "floating"),
1976 quoted, RE_SV_TAIL(must));
1980 else if ( (c = progi->regstclass) ) {
1982 const OPCODE op = OP(progi->regstclass);
1983 /* don't bother with what can't match */
1984 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1985 strend = HOPc(strend, -(minlen - 1));
1988 SV * const prop = sv_newmortal();
1989 regprop(prog, prop, c);
1991 RE_PV_QUOTED_DECL(quoted,do_utf8,PERL_DEBUG_PAD_ZERO(1),
1993 PerlIO_printf(Perl_debug_log,
1994 "Matching stclass %.*s against %s (%d chars)\n",
1995 (int)SvCUR(prop), SvPVX_const(prop),
1996 quoted, (int)(strend - s));
1999 if (find_byclass(prog, c, s, strend, ®info))
2001 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2005 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2010 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2011 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2012 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
2014 if (flags & REXEC_SCREAM) {
2015 last = screaminstr(sv, float_real, s - strbeg,
2016 end_shift, &scream_pos, 1); /* last one */
2018 last = scream_olds; /* Only one occurrence. */
2019 /* we may be pointing at the wrong string */
2020 else if (RX_MATCH_COPIED(prog))
2021 s = strbeg + (s - SvPVX_const(sv));
2025 const char * const little = SvPV_const(float_real, len);
2027 if (SvTAIL(float_real)) {
2028 if (memEQ(strend - len + 1, little, len - 1))
2029 last = strend - len + 1;
2030 else if (!multiline)
2031 last = memEQ(strend - len, little, len)
2032 ? strend - len : NULL;
2038 last = rninstr(s, strend, little, little + len);
2040 last = strend; /* matching "$" */
2045 PerlIO_printf(Perl_debug_log,
2046 "%sCan't trim the tail, match fails (should not happen)%s\n",
2047 PL_colors[4], PL_colors[5]));
2048 goto phooey; /* Should not happen! */
2050 dontbother = strend - last + prog->float_min_offset;
2052 if (minlen && (dontbother < minlen))
2053 dontbother = minlen - 1;
2054 strend -= dontbother; /* this one's always in bytes! */
2055 /* We don't know much -- general case. */
2058 if (regtry(®info, &s))
2067 if (regtry(®info, &s))
2069 } while (s++ < strend);
2077 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2079 if (PL_reg_eval_set) {
2080 /* Preserve the current value of $^R */
2081 if (oreplsv != GvSV(PL_replgv))
2082 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2083 restored, the value remains
2085 restore_pos(aTHX_ prog);
2087 if (prog->paren_names)
2088 (void)hv_iterinit(prog->paren_names);
2090 /* make sure $`, $&, $', and $digit will work later */
2091 if ( !(flags & REXEC_NOT_FIRST) ) {
2092 RX_MATCH_COPY_FREE(prog);
2093 if (flags & REXEC_COPY_STR) {
2094 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2095 #ifdef PERL_OLD_COPY_ON_WRITE
2097 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2099 PerlIO_printf(Perl_debug_log,
2100 "Copy on write: regexp capture, type %d\n",
2103 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2104 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2105 assert (SvPOKp(prog->saved_copy));
2109 RX_MATCH_COPIED_on(prog);
2110 s = savepvn(strbeg, i);
2116 prog->subbeg = strbeg;
2117 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2124 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2125 PL_colors[4], PL_colors[5]));
2126 if (PL_reg_eval_set)
2127 restore_pos(aTHX_ prog);
2129 /* we failed :-( roll it back */
2130 swap_match_buff(prog);
2137 - regtry - try match at specific point
2139 STATIC I32 /* 0 failure, 1 success */
2140 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2144 regexp *prog = reginfo->prog;
2145 RXi_GET_DECL(prog,progi);
2146 GET_RE_DEBUG_FLAGS_DECL;
2147 reginfo->cutpoint=NULL;
2149 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2152 PL_reg_eval_set = RS_init;
2153 DEBUG_EXECUTE_r(DEBUG_s(
2154 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2155 (IV)(PL_stack_sp - PL_stack_base));
2158 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2159 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2161 /* Apparently this is not needed, judging by wantarray. */
2162 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2163 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2166 /* Make $_ available to executed code. */
2167 if (reginfo->sv != DEFSV) {
2169 DEFSV = reginfo->sv;
2172 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2173 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2174 /* prepare for quick setting of pos */
2175 #ifdef PERL_OLD_COPY_ON_WRITE
2176 if (SvIsCOW(reginfo->sv))
2177 sv_force_normal_flags(reginfo->sv, 0);
2179 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2180 &PL_vtbl_mglob, NULL, 0);
2184 PL_reg_oldpos = mg->mg_len;
2185 SAVEDESTRUCTOR_X(restore_pos, prog);
2187 if (!PL_reg_curpm) {
2188 Newxz(PL_reg_curpm, 1, PMOP);
2191 SV* const repointer = newSViv(0);
2192 /* so we know which PL_regex_padav element is PL_reg_curpm */
2193 SvFLAGS(repointer) |= SVf_BREAK;
2194 av_push(PL_regex_padav,repointer);
2195 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2196 PL_regex_pad = AvARRAY(PL_regex_padav);
2200 PM_SETRE(PL_reg_curpm, prog);
2201 PL_reg_oldcurpm = PL_curpm;
2202 PL_curpm = PL_reg_curpm;
2203 if (RX_MATCH_COPIED(prog)) {
2204 /* Here is a serious problem: we cannot rewrite subbeg,
2205 since it may be needed if this match fails. Thus
2206 $` inside (?{}) could fail... */
2207 PL_reg_oldsaved = prog->subbeg;
2208 PL_reg_oldsavedlen = prog->sublen;
2209 #ifdef PERL_OLD_COPY_ON_WRITE
2210 PL_nrs = prog->saved_copy;
2212 RX_MATCH_COPIED_off(prog);
2215 PL_reg_oldsaved = NULL;
2216 prog->subbeg = PL_bostr;
2217 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2219 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2220 prog->offs[0].start = *startpos - PL_bostr;
2221 PL_reginput = *startpos;
2222 PL_reglastparen = &prog->lastparen;
2223 PL_reglastcloseparen = &prog->lastcloseparen;
2224 prog->lastparen = 0;
2225 prog->lastcloseparen = 0;
2227 PL_regoffs = prog->offs;
2228 if (PL_reg_start_tmpl <= prog->nparens) {
2229 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2230 if(PL_reg_start_tmp)
2231 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2233 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2236 /* XXXX What this code is doing here?!!! There should be no need
2237 to do this again and again, PL_reglastparen should take care of
2240 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2241 * Actually, the code in regcppop() (which Ilya may be meaning by
2242 * PL_reglastparen), is not needed at all by the test suite
2243 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2244 * enough, for building DynaLoader, or otherwise this
2245 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2246 * will happen. Meanwhile, this code *is* needed for the
2247 * above-mentioned test suite tests to succeed. The common theme
2248 * on those tests seems to be returning null fields from matches.
2251 if (prog->nparens) {
2252 regexp_paren_pair *pp = PL_regoffs;
2254 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2262 if (regmatch(reginfo, progi->program + 1)) {
2263 PL_regoffs[0].end = PL_reginput - PL_bostr;
2266 if (reginfo->cutpoint)
2267 *startpos= reginfo->cutpoint;
2268 REGCP_UNWIND(lastcp);
2273 #define sayYES goto yes
2274 #define sayNO goto no
2275 #define sayNO_SILENT goto no_silent
2277 /* we dont use STMT_START/END here because it leads to
2278 "unreachable code" warnings, which are bogus, but distracting. */
2279 #define CACHEsayNO \
2280 if (ST.cache_mask) \
2281 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2284 /* this is used to determine how far from the left messages like
2285 'failed...' are printed. It should be set such that messages
2286 are inline with the regop output that created them.
2288 #define REPORT_CODE_OFF 32
2291 /* Make sure there is a test for this +1 options in re_tests */
2292 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2294 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2295 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2297 #define SLAB_FIRST(s) (&(s)->states[0])
2298 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2300 /* grab a new slab and return the first slot in it */
2302 STATIC regmatch_state *
2305 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2308 regmatch_slab *s = PL_regmatch_slab->next;
2310 Newx(s, 1, regmatch_slab);
2311 s->prev = PL_regmatch_slab;
2313 PL_regmatch_slab->next = s;
2315 PL_regmatch_slab = s;
2316 return SLAB_FIRST(s);
2320 /* push a new state then goto it */
2322 #define PUSH_STATE_GOTO(state, node) \
2324 st->resume_state = state; \
2327 /* push a new state with success backtracking, then goto it */
2329 #define PUSH_YES_STATE_GOTO(state, node) \
2331 st->resume_state = state; \
2332 goto push_yes_state;
2338 regmatch() - main matching routine
2340 This is basically one big switch statement in a loop. We execute an op,
2341 set 'next' to point the next op, and continue. If we come to a point which
2342 we may need to backtrack to on failure such as (A|B|C), we push a
2343 backtrack state onto the backtrack stack. On failure, we pop the top
2344 state, and re-enter the loop at the state indicated. If there are no more
2345 states to pop, we return failure.
2347 Sometimes we also need to backtrack on success; for example /A+/, where
2348 after successfully matching one A, we need to go back and try to
2349 match another one; similarly for lookahead assertions: if the assertion
2350 completes successfully, we backtrack to the state just before the assertion
2351 and then carry on. In these cases, the pushed state is marked as
2352 'backtrack on success too'. This marking is in fact done by a chain of
2353 pointers, each pointing to the previous 'yes' state. On success, we pop to
2354 the nearest yes state, discarding any intermediate failure-only states.
2355 Sometimes a yes state is pushed just to force some cleanup code to be
2356 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2357 it to free the inner regex.
2359 Note that failure backtracking rewinds the cursor position, while
2360 success backtracking leaves it alone.
2362 A pattern is complete when the END op is executed, while a subpattern
2363 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2364 ops trigger the "pop to last yes state if any, otherwise return true"
2367 A common convention in this function is to use A and B to refer to the two
2368 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2369 the subpattern to be matched possibly multiple times, while B is the entire
2370 rest of the pattern. Variable and state names reflect this convention.
2372 The states in the main switch are the union of ops and failure/success of
2373 substates associated with with that op. For example, IFMATCH is the op
2374 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2375 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2376 successfully matched A and IFMATCH_A_fail is a state saying that we have
2377 just failed to match A. Resume states always come in pairs. The backtrack
2378 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2379 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2380 on success or failure.
2382 The struct that holds a backtracking state is actually a big union, with
2383 one variant for each major type of op. The variable st points to the
2384 top-most backtrack struct. To make the code clearer, within each
2385 block of code we #define ST to alias the relevant union.
2387 Here's a concrete example of a (vastly oversimplified) IFMATCH
2393 #define ST st->u.ifmatch
2395 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2396 ST.foo = ...; // some state we wish to save
2398 // push a yes backtrack state with a resume value of
2399 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2401 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2404 case IFMATCH_A: // we have successfully executed A; now continue with B
2406 bar = ST.foo; // do something with the preserved value
2409 case IFMATCH_A_fail: // A failed, so the assertion failed
2410 ...; // do some housekeeping, then ...
2411 sayNO; // propagate the failure
2418 For any old-timers reading this who are familiar with the old recursive
2419 approach, the code above is equivalent to:
2421 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2430 ...; // do some housekeeping, then ...
2431 sayNO; // propagate the failure
2434 The topmost backtrack state, pointed to by st, is usually free. If you
2435 want to claim it, populate any ST.foo fields in it with values you wish to
2436 save, then do one of
2438 PUSH_STATE_GOTO(resume_state, node);
2439 PUSH_YES_STATE_GOTO(resume_state, node);
2441 which sets that backtrack state's resume value to 'resume_state', pushes a
2442 new free entry to the top of the backtrack stack, then goes to 'node'.
2443 On backtracking, the free slot is popped, and the saved state becomes the
2444 new free state. An ST.foo field in this new top state can be temporarily
2445 accessed to retrieve values, but once the main loop is re-entered, it
2446 becomes available for reuse.
2448 Note that the depth of the backtrack stack constantly increases during the
2449 left-to-right execution of the pattern, rather than going up and down with
2450 the pattern nesting. For example the stack is at its maximum at Z at the
2451 end of the pattern, rather than at X in the following:
2453 /(((X)+)+)+....(Y)+....Z/
2455 The only exceptions to this are lookahead/behind assertions and the cut,
2456 (?>A), which pop all the backtrack states associated with A before
2459 Bascktrack state structs are allocated in slabs of about 4K in size.
2460 PL_regmatch_state and st always point to the currently active state,
2461 and PL_regmatch_slab points to the slab currently containing
2462 PL_regmatch_state. The first time regmatch() is called, the first slab is
2463 allocated, and is never freed until interpreter destruction. When the slab
2464 is full, a new one is allocated and chained to the end. At exit from
2465 regmatch(), slabs allocated since entry are freed.
2470 #define DEBUG_STATE_pp(pp) \
2472 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2473 PerlIO_printf(Perl_debug_log, \
2474 " %*s"pp" %s%s%s%s%s\n", \
2476 PL_reg_name[st->resume_state], \
2477 ((st==yes_state||st==mark_state) ? "[" : ""), \
2478 ((st==yes_state) ? "Y" : ""), \
2479 ((st==mark_state) ? "M" : ""), \
2480 ((st==yes_state||st==mark_state) ? "]" : "") \
2485 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2490 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2491 const char *start, const char *end, const char *blurb)
2493 const bool utf8_pat= prog->extflags & RXf_UTF8 ? 1 : 0;
2497 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2498 prog->precomp, prog->prelen, 60);
2500 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2501 start, end - start, 60);
2503 PerlIO_printf(Perl_debug_log,
2504 "%s%s REx%s %s against %s\n",
2505 PL_colors[4], blurb, PL_colors[5], s0, s1);
2507 if (do_utf8||utf8_pat)
2508 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2509 utf8_pat ? "pattern" : "",
2510 utf8_pat && do_utf8 ? " and " : "",
2511 do_utf8 ? "string" : ""
2517 S_dump_exec_pos(pTHX_ const char *locinput,
2518 const regnode *scan,
2519 const char *loc_regeol,
2520 const char *loc_bostr,
2521 const char *loc_reg_starttry,
2524 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2525 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2526 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2527 /* The part of the string before starttry has one color
2528 (pref0_len chars), between starttry and current
2529 position another one (pref_len - pref0_len chars),
2530 after the current position the third one.
2531 We assume that pref0_len <= pref_len, otherwise we
2532 decrease pref0_len. */
2533 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2534 ? (5 + taill) - l : locinput - loc_bostr;
2537 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2539 pref0_len = pref_len - (locinput - loc_reg_starttry);
2540 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2541 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2542 ? (5 + taill) - pref_len : loc_regeol - locinput);
2543 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2547 if (pref0_len > pref_len)
2548 pref0_len = pref_len;
2550 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2552 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2553 (locinput - pref_len),pref0_len, 60, 4, 5);
2555 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2556 (locinput - pref_len + pref0_len),
2557 pref_len - pref0_len, 60, 2, 3);
2559 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2560 locinput, loc_regeol - locinput, 10, 0, 1);
2562 const STRLEN tlen=len0+len1+len2;
2563 PerlIO_printf(Perl_debug_log,
2564 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2565 (IV)(locinput - loc_bostr),
2568 (docolor ? "" : "> <"),
2570 (int)(tlen > 19 ? 0 : 19 - tlen),
2577 /* reg_check_named_buff_matched()
2578 * Checks to see if a named buffer has matched. The data array of
2579 * buffer numbers corresponding to the buffer is expected to reside
2580 * in the regexp->data->data array in the slot stored in the ARG() of
2581 * node involved. Note that this routine doesn't actually care about the
2582 * name, that information is not preserved from compilation to execution.
2583 * Returns the index of the leftmost defined buffer with the given name
2584 * or 0 if non of the buffers matched.
2587 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2589 RXi_GET_DECL(rex,rexi);
2590 SV *sv_dat=(SV*)rexi->data->data[ ARG( scan ) ];
2591 I32 *nums=(I32*)SvPVX(sv_dat);
2592 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2593 if ((I32)*PL_reglastparen >= nums[n] &&
2594 PL_regoffs[nums[n]].end != -1)
2602 #define SETREX(Re1,Re2) \
2603 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2606 STATIC I32 /* 0 failure, 1 success */
2607 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2609 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2613 register const bool do_utf8 = PL_reg_match_utf8;
2614 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2616 regexp *rex = reginfo->prog;
2617 RXi_GET_DECL(rex,rexi);
2619 regmatch_slab *orig_slab;
2620 regmatch_state *orig_state;
2622 /* the current state. This is a cached copy of PL_regmatch_state */
2623 register regmatch_state *st;
2625 /* cache heavy used fields of st in registers */
2626 register regnode *scan;
2627 register regnode *next;
2628 register U32 n = 0; /* general value; init to avoid compiler warning */
2629 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2630 register char *locinput = PL_reginput;
2631 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2633 bool result = 0; /* return value of S_regmatch */
2634 int depth = 0; /* depth of backtrack stack */
2635 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2636 const U32 max_nochange_depth =
2637 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2638 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2640 regmatch_state *yes_state = NULL; /* state to pop to on success of
2642 /* mark_state piggy backs on the yes_state logic so that when we unwind
2643 the stack on success we can update the mark_state as we go */
2644 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2646 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2647 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2649 bool no_final = 0; /* prevent failure from backtracking? */
2650 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2651 char *startpoint = PL_reginput;
2652 SV *popmark = NULL; /* are we looking for a mark? */
2653 SV *sv_commit = NULL; /* last mark name seen in failure */
2654 SV *sv_yes_mark = NULL; /* last mark name we have seen
2655 during a successfull match */
2656 U32 lastopen = 0; /* last open we saw */
2657 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2660 /* these three flags are set by various ops to signal information to
2661 * the very next op. They have a useful lifetime of exactly one loop
2662 * iteration, and are not preserved or restored by state pushes/pops
2664 bool sw = 0; /* the condition value in (?(cond)a|b) */
2665 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2666 int logical = 0; /* the following EVAL is:
2670 or the following IFMATCH/UNLESSM is:
2671 false: plain (?=foo)
2672 true: used as a condition: (?(?=foo))
2676 GET_RE_DEBUG_FLAGS_DECL;
2679 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
2680 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2682 /* on first ever call to regmatch, allocate first slab */
2683 if (!PL_regmatch_slab) {
2684 Newx(PL_regmatch_slab, 1, regmatch_slab);
2685 PL_regmatch_slab->prev = NULL;
2686 PL_regmatch_slab->next = NULL;
2687 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2690 /* remember current high-water mark for exit */
2691 /* XXX this should be done with SAVE* instead */
2692 orig_slab = PL_regmatch_slab;
2693 orig_state = PL_regmatch_state;
2695 /* grab next free state slot */
2696 st = ++PL_regmatch_state;
2697 if (st > SLAB_LAST(PL_regmatch_slab))
2698 st = PL_regmatch_state = S_push_slab(aTHX);
2700 /* Note that nextchr is a byte even in UTF */
2701 nextchr = UCHARAT(locinput);
2703 while (scan != NULL) {
2706 SV * const prop = sv_newmortal();
2707 regnode *rnext=regnext(scan);
2708 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2709 regprop(rex, prop, scan);
2711 PerlIO_printf(Perl_debug_log,
2712 "%3"IVdf":%*s%s(%"IVdf")\n",
2713 (IV)(scan - rexi->program), depth*2, "",
2715 (PL_regkind[OP(scan)] == END || !rnext) ?
2716 0 : (IV)(rnext - rexi->program));
2719 next = scan + NEXT_OFF(scan);
2722 state_num = OP(scan);
2725 switch (state_num) {
2727 if (locinput == PL_bostr)
2729 /* reginfo->till = reginfo->bol; */
2734 if (locinput == PL_bostr ||
2735 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2741 if (locinput == PL_bostr)
2745 if (locinput == reginfo->ganch)
2750 /* update the startpoint */
2751 st->u.keeper.val = PL_regoffs[0].start;
2752 PL_reginput = locinput;
2753 PL_regoffs[0].start = locinput - PL_bostr;
2754 PUSH_STATE_GOTO(KEEPS_next, next);
2756 case KEEPS_next_fail:
2757 /* rollback the start point change */
2758 PL_regoffs[0].start = st->u.keeper.val;
2764 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2769 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2771 if (PL_regeol - locinput > 1)
2775 if (PL_regeol != locinput)
2779 if (!nextchr && locinput >= PL_regeol)
2782 locinput += PL_utf8skip[nextchr];
2783 if (locinput > PL_regeol)
2785 nextchr = UCHARAT(locinput);
2788 nextchr = UCHARAT(++locinput);
2791 if (!nextchr && locinput >= PL_regeol)
2793 nextchr = UCHARAT(++locinput);
2796 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2799 locinput += PL_utf8skip[nextchr];
2800 if (locinput > PL_regeol)
2802 nextchr = UCHARAT(locinput);
2805 nextchr = UCHARAT(++locinput);
2809 #define ST st->u.trie
2811 /* In this case the charclass data is available inline so
2812 we can fail fast without a lot of extra overhead.
2814 if (scan->flags == EXACT || !do_utf8) {
2815 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2817 PerlIO_printf(Perl_debug_log,
2818 "%*s %sfailed to match trie start class...%s\n",
2819 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2828 /* what type of TRIE am I? (utf8 makes this contextual) */
2829 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2830 trie_type = do_utf8 ?
2831 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2834 /* what trie are we using right now */
2835 reg_trie_data * const trie
2836 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2837 HV * widecharmap = (HV *)rexi->data->data[ ARG( scan ) + 1 ];
2838 U32 state = trie->startstate;
2840 if (trie->bitmap && trie_type != trie_utf8_fold &&
2841 !TRIE_BITMAP_TEST(trie,*locinput)
2843 if (trie->states[ state ].wordnum) {
2845 PerlIO_printf(Perl_debug_log,
2846 "%*s %smatched empty string...%s\n",
2847 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2852 PerlIO_printf(Perl_debug_log,
2853 "%*s %sfailed to match trie start class...%s\n",
2854 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2861 U8 *uc = ( U8* )locinput;
2865 U8 *uscan = (U8*)NULL;
2867 SV *sv_accept_buff = NULL;
2868 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2870 ST.accepted = 0; /* how many accepting states we have seen */
2872 ST.jump = trie->jump;
2875 traverse the TRIE keeping track of all accepting states
2876 we transition through until we get to a failing node.
2879 while ( state && uc <= (U8*)PL_regeol ) {
2880 U32 base = trie->states[ state ].trans.base;
2883 /* We use charid to hold the wordnum as we don't use it
2884 for charid until after we have done the wordnum logic.
2885 We define an alias just so that the wordnum logic reads
2888 #define got_wordnum charid
2889 got_wordnum = trie->states[ state ].wordnum;
2891 if ( got_wordnum ) {
2892 if ( ! ST.accepted ) {
2895 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2896 sv_accept_buff=newSV(bufflen *
2897 sizeof(reg_trie_accepted) - 1);
2898 SvCUR_set(sv_accept_buff, 0);
2899 SvPOK_on(sv_accept_buff);
2900 sv_2mortal(sv_accept_buff);
2903 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2906 if (ST.accepted >= bufflen) {
2908 ST.accept_buff =(reg_trie_accepted*)
2909 SvGROW(sv_accept_buff,
2910 bufflen * sizeof(reg_trie_accepted));
2912 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2913 + sizeof(reg_trie_accepted));
2916 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2917 ST.accept_buff[ST.accepted].endpos = uc;
2919 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2923 DEBUG_TRIE_EXECUTE_r({
2924 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2925 PerlIO_printf( Perl_debug_log,
2926 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2927 2+depth * 2, "", PL_colors[4],
2928 (UV)state, (UV)ST.accepted );
2932 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
2933 uscan, len, uvc, charid, foldlen,
2937 (base + charid > trie->uniquecharcount )
2938 && (base + charid - 1 - trie->uniquecharcount
2940 && trie->trans[base + charid - 1 -
2941 trie->uniquecharcount].check == state)
2943 state = trie->trans[base + charid - 1 -
2944 trie->uniquecharcount ].next;
2955 DEBUG_TRIE_EXECUTE_r(
2956 PerlIO_printf( Perl_debug_log,
2957 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2958 charid, uvc, (UV)state, PL_colors[5] );
2965 PerlIO_printf( Perl_debug_log,
2966 "%*s %sgot %"IVdf" possible matches%s\n",
2967 REPORT_CODE_OFF + depth * 2, "",
2968 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2971 goto trie_first_try; /* jump into the fail handler */
2973 case TRIE_next_fail: /* we failed - try next alterative */
2975 REGCP_UNWIND(ST.cp);
2976 for (n = *PL_reglastparen; n > ST.lastparen; n--)
2977 PL_regoffs[n].end = -1;
2978 *PL_reglastparen = n;
2987 ST.lastparen = *PL_reglastparen;
2990 if ( ST.accepted == 1 ) {
2991 /* only one choice left - just continue */
2993 AV *const trie_words
2994 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
2995 SV ** const tmp = av_fetch( trie_words,
2996 ST.accept_buff[ 0 ].wordnum-1, 0 );
2997 SV *sv= tmp ? sv_newmortal() : NULL;
2999 PerlIO_printf( Perl_debug_log,
3000 "%*s %sonly one match left: #%d <%s>%s\n",
3001 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3002 ST.accept_buff[ 0 ].wordnum,
3003 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3004 PL_colors[0], PL_colors[1],
3005 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3007 : "not compiled under -Dr",
3010 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
3011 /* in this case we free tmps/leave before we call regmatch
3012 as we wont be using accept_buff again. */
3014 locinput = PL_reginput;
3015 nextchr = UCHARAT(locinput);
3016 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
3019 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
3020 if (!has_cutgroup) {
3025 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3028 continue; /* execute rest of RE */
3031 if ( !ST.accepted-- ) {
3033 PerlIO_printf( Perl_debug_log,
3034 "%*s %sTRIE failed...%s\n",
3035 REPORT_CODE_OFF+depth*2, "",
3046 There are at least two accepting states left. Presumably
3047 the number of accepting states is going to be low,
3048 typically two. So we simply scan through to find the one
3049 with lowest wordnum. Once we find it, we swap the last
3050 state into its place and decrement the size. We then try to
3051 match the rest of the pattern at the point where the word
3052 ends. If we succeed, control just continues along the
3053 regex; if we fail we return here to try the next accepting
3060 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3061 DEBUG_TRIE_EXECUTE_r(
3062 PerlIO_printf( Perl_debug_log,
3063 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3064 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3065 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3066 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3069 if (ST.accept_buff[cur].wordnum <
3070 ST.accept_buff[best].wordnum)
3075 AV *const trie_words
3076 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
3077 SV ** const tmp = av_fetch( trie_words,
3078 ST.accept_buff[ best ].wordnum - 1, 0 );
3079 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3081 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3082 SV *sv= tmp ? sv_newmortal() : NULL;
3084 PerlIO_printf( Perl_debug_log,
3085 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3086 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3087 ST.accept_buff[best].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)
3091 ) : "not compiled under -Dr",
3092 REG_NODE_NUM(nextop),
3096 if ( best<ST.accepted ) {
3097 reg_trie_accepted tmp = ST.accept_buff[ best ];
3098 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3099 ST.accept_buff[ ST.accepted ] = tmp;
3102 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3103 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3107 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3111 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3114 PUSH_STATE_GOTO(TRIE_next, scan);
3127 char *s = STRING(scan);
3129 if (do_utf8 != UTF) {
3130 /* The target and the pattern have differing utf8ness. */
3132 const char * const e = s + ln;
3135 /* The target is utf8, the pattern is not utf8. */
3140 if (NATIVE_TO_UNI(*(U8*)s) !=
3141 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3149 /* The target is not utf8, the pattern is utf8. */
3154 if (NATIVE_TO_UNI(*((U8*)l)) !=
3155 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3163 nextchr = UCHARAT(locinput);
3166 /* The target and the pattern have the same utf8ness. */
3167 /* Inline the first character, for speed. */
3168 if (UCHARAT(s) != nextchr)
3170 if (PL_regeol - locinput < ln)
3172 if (ln > 1 && memNE(s, locinput, ln))
3175 nextchr = UCHARAT(locinput);
3179 PL_reg_flags |= RF_tainted;
3182 char * const s = STRING(scan);
3185 if (do_utf8 || UTF) {
3186 /* Either target or the pattern are utf8. */
3187 const char * const l = locinput;
3188 char *e = PL_regeol;
3190 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3191 l, &e, 0, do_utf8)) {
3192 /* One more case for the sharp s:
3193 * pack("U0U*", 0xDF) =~ /ss/i,
3194 * the 0xC3 0x9F are the UTF-8
3195 * byte sequence for the U+00DF. */
3197 toLOWER(s[0]) == 's' &&
3199 toLOWER(s[1]) == 's' &&
3206 nextchr = UCHARAT(locinput);
3210 /* Neither the target and the pattern are utf8. */
3212 /* Inline the first character, for speed. */
3213 if (UCHARAT(s) != nextchr &&
3214 UCHARAT(s) != ((OP(scan) == EXACTF)
3215 ? PL_fold : PL_fold_locale)[nextchr])
3217 if (PL_regeol - locinput < ln)
3219 if (ln > 1 && (OP(scan) == EXACTF
3220 ? ibcmp(s, locinput, ln)
3221 : ibcmp_locale(s, locinput, ln)))
3224 nextchr = UCHARAT(locinput);
3229 STRLEN inclasslen = PL_regeol - locinput;
3231 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3233 if (locinput >= PL_regeol)
3235 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3236 nextchr = UCHARAT(locinput);
3241 nextchr = UCHARAT(locinput);
3242 if (!REGINCLASS(rex, scan, (U8*)locinput))
3244 if (!nextchr && locinput >= PL_regeol)
3246 nextchr = UCHARAT(++locinput);
3250 /* If we might have the case of the German sharp s
3251 * in a casefolding Unicode character class. */
3253 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3254 locinput += SHARP_S_SKIP;
3255 nextchr = UCHARAT(locinput);
3261 PL_reg_flags |= RF_tainted;
3267 LOAD_UTF8_CHARCLASS_ALNUM();
3268 if (!(OP(scan) == ALNUM
3269 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3270 : isALNUM_LC_utf8((U8*)locinput)))
3274 locinput += PL_utf8skip[nextchr];
3275 nextchr = UCHARAT(locinput);
3278 if (!(OP(scan) == ALNUM
3279 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3281 nextchr = UCHARAT(++locinput);
3284 PL_reg_flags |= RF_tainted;
3287 if (!nextchr && locinput >= PL_regeol)
3290 LOAD_UTF8_CHARCLASS_ALNUM();
3291 if (OP(scan) == NALNUM
3292 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3293 : isALNUM_LC_utf8((U8*)locinput))
3297 locinput += PL_utf8skip[nextchr];
3298 nextchr = UCHARAT(locinput);
3301 if (OP(scan) == NALNUM
3302 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3304 nextchr = UCHARAT(++locinput);
3308 PL_reg_flags |= RF_tainted;
3312 /* was last char in word? */
3314 if (locinput == PL_bostr)
3317 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3319 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3321 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3322 ln = isALNUM_uni(ln);
3323 LOAD_UTF8_CHARCLASS_ALNUM();
3324 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3327 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3328 n = isALNUM_LC_utf8((U8*)locinput);
3332 ln = (locinput != PL_bostr) ?
3333 UCHARAT(locinput - 1) : '\n';
3334 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3336 n = isALNUM(nextchr);
3339 ln = isALNUM_LC(ln);
3340 n = isALNUM_LC(nextchr);
3343 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3344 OP(scan) == BOUNDL))
3348 PL_reg_flags |= RF_tainted;
3354 if (UTF8_IS_CONTINUED(nextchr)) {
3355 LOAD_UTF8_CHARCLASS_SPACE();
3356 if (!(OP(scan) == SPACE
3357 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3358 : isSPACE_LC_utf8((U8*)locinput)))
3362 locinput += PL_utf8skip[nextchr];
3363 nextchr = UCHARAT(locinput);
3366 if (!(OP(scan) == SPACE
3367 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3369 nextchr = UCHARAT(++locinput);
3372 if (!(OP(scan) == SPACE
3373 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3375 nextchr = UCHARAT(++locinput);
3379 PL_reg_flags |= RF_tainted;
3382 if (!nextchr && locinput >= PL_regeol)
3385 LOAD_UTF8_CHARCLASS_SPACE();
3386 if (OP(scan) == NSPACE
3387 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3388 : isSPACE_LC_utf8((U8*)locinput))
3392 locinput += PL_utf8skip[nextchr];
3393 nextchr = UCHARAT(locinput);
3396 if (OP(scan) == NSPACE
3397 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3399 nextchr = UCHARAT(++locinput);
3402 PL_reg_flags |= RF_tainted;
3408 LOAD_UTF8_CHARCLASS_DIGIT();
3409 if (!(OP(scan) == DIGIT
3410 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3411 : isDIGIT_LC_utf8((U8*)locinput)))
3415 locinput += PL_utf8skip[nextchr];
3416 nextchr = UCHARAT(locinput);
3419 if (!(OP(scan) == DIGIT
3420 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3422 nextchr = UCHARAT(++locinput);
3425 PL_reg_flags |= RF_tainted;
3428 if (!nextchr && locinput >= PL_regeol)
3431 LOAD_UTF8_CHARCLASS_DIGIT();
3432 if (OP(scan) == NDIGIT
3433 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3434 : isDIGIT_LC_utf8((U8*)locinput))
3438 locinput += PL_utf8skip[nextchr];
3439 nextchr = UCHARAT(locinput);
3442 if (OP(scan) == NDIGIT
3443 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3445 nextchr = UCHARAT(++locinput);
3448 if (locinput >= PL_regeol)
3451 LOAD_UTF8_CHARCLASS_MARK();
3452 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3454 locinput += PL_utf8skip[nextchr];
3455 while (locinput < PL_regeol &&
3456 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3457 locinput += UTF8SKIP(locinput);
3458 if (locinput > PL_regeol)
3463 nextchr = UCHARAT(locinput);
3470 PL_reg_flags |= RF_tainted;
3475 n = reg_check_named_buff_matched(rex,scan);
3478 type = REF + ( type - NREF );
3485 PL_reg_flags |= RF_tainted;
3489 n = ARG(scan); /* which paren pair */
3492 ln = PL_regoffs[n].start;
3493 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3494 if (*PL_reglastparen < n || ln == -1)
3495 sayNO; /* Do not match unless seen CLOSEn. */
3496 if (ln == PL_regoffs[n].end)
3500 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3502 const char *e = PL_bostr + PL_regoffs[n].end;
3504 * Note that we can't do the "other character" lookup trick as
3505 * in the 8-bit case (no pun intended) because in Unicode we
3506 * have to map both upper and title case to lower case.
3510 STRLEN ulen1, ulen2;
3511 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3512 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3516 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3517 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3518 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3525 nextchr = UCHARAT(locinput);
3529 /* Inline the first character, for speed. */
3530 if (UCHARAT(s) != nextchr &&
3532 (UCHARAT(s) != (type == REFF
3533 ? PL_fold : PL_fold_locale)[nextchr])))
3535 ln = PL_regoffs[n].end - ln;
3536 if (locinput + ln > PL_regeol)
3538 if (ln > 1 && (type == REF
3539 ? memNE(s, locinput, ln)
3541 ? ibcmp(s, locinput, ln)
3542 : ibcmp_locale(s, locinput, ln))))
3545 nextchr = UCHARAT(locinput);
3555 #define ST st->u.eval
3559 regexp_internal *rei;
3560 regnode *startpoint;
3563 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
3564 if (cur_eval && cur_eval->locinput==locinput) {
3565 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3566 Perl_croak(aTHX_ "Infinite recursion in regex");
3567 if ( ++nochange_depth > max_nochange_depth )
3569 "Pattern subroutine nesting without pos change"
3570 " exceeded limit in regex");
3576 (void)ReREFCNT_inc(rex);
3577 if (OP(scan)==GOSUB) {
3578 startpoint = scan + ARG2L(scan);
3579 ST.close_paren = ARG(scan);
3581 startpoint = rei->program+1;
3584 goto eval_recurse_doit;
3586 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3587 if (cur_eval && cur_eval->locinput==locinput) {
3588 if ( ++nochange_depth > max_nochange_depth )
3589 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3594 /* execute the code in the {...} */
3596 SV ** const before = SP;
3597 OP_4tree * const oop = PL_op;
3598 COP * const ocurcop = PL_curcop;
3602 PL_op = (OP_4tree*)rexi->data->data[n];
3603 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3604 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3605 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3606 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
3609 SV *sv_mrk = get_sv("REGMARK", 1);
3610 sv_setsv(sv_mrk, sv_yes_mark);
3613 CALLRUNOPS(aTHX); /* Scalar context. */
3616 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3623 PAD_RESTORE_LOCAL(old_comppad);
3624 PL_curcop = ocurcop;
3627 sv_setsv(save_scalar(PL_replgv), ret);
3631 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3634 /* extract RE object from returned value; compiling if
3639 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3640 mg = mg_find(sv, PERL_MAGIC_qr);
3641 else if (SvSMAGICAL(ret)) {
3642 if (SvGMAGICAL(ret))
3643 sv_unmagic(ret, PERL_MAGIC_qr);
3645 mg = mg_find(ret, PERL_MAGIC_qr);
3649 re = reg_temp_copy((regexp *)mg->mg_obj); /*XXX:dmq*/
3653 const char * const t = SvPV_const(ret, len);
3655 const I32 osize = PL_regsize;
3658 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3659 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3661 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3663 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3668 RX_MATCH_COPIED_off(re);
3669 re->subbeg = rex->subbeg;
3670 re->sublen = rex->sublen;
3673 debug_start_match(re, do_utf8, locinput, PL_regeol,
3674 "Matching embedded");
3676 startpoint = rei->program + 1;
3677 ST.close_paren = 0; /* only used for GOSUB */
3678 /* borrowed from regtry */
3679 if (PL_reg_start_tmpl <= re->nparens) {
3680 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3681 if(PL_reg_start_tmp)
3682 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3684 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3687 eval_recurse_doit: /* Share code with GOSUB below this line */
3688 /* run the pattern returned from (??{...}) */
3689 ST.cp = regcppush(0); /* Save *all* the positions. */
3690 REGCP_SET(ST.lastcp);
3692 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
3694 *PL_reglastparen = 0;
3695 *PL_reglastcloseparen = 0;
3696 PL_reginput = locinput;
3699 /* XXXX This is too dramatic a measure... */
3702 ST.toggle_reg_flags = PL_reg_flags;
3703 if (re->extflags & RXf_UTF8)
3704 PL_reg_flags |= RF_utf8;
3706 PL_reg_flags &= ~RF_utf8;
3707 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3710 ST.prev_curlyx = cur_curlyx;
3715 ST.prev_eval = cur_eval;
3717 /* now continue from first node in postoned RE */
3718 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3721 /* logical is 1, /(?(?{...})X|Y)/ */
3722 sw = (bool)SvTRUE(ret);
3727 case EVAL_AB: /* cleanup after a successful (??{A})B */
3728 /* note: this is called twice; first after popping B, then A */
3729 PL_reg_flags ^= ST.toggle_reg_flags;
3731 SETREX(rex,ST.prev_rex);
3732 rexi = RXi_GET(rex);
3734 cur_eval = ST.prev_eval;
3735 cur_curlyx = ST.prev_curlyx;
3736 /* XXXX This is too dramatic a measure... */
3738 if ( nochange_depth )
3743 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3744 /* note: this is called twice; first after popping B, then A */
3745 PL_reg_flags ^= ST.toggle_reg_flags;
3747 SETREX(rex,ST.prev_rex);
3748 rexi = RXi_GET(rex);
3749 PL_reginput = locinput;
3750 REGCP_UNWIND(ST.lastcp);
3752 cur_eval = ST.prev_eval;
3753 cur_curlyx = ST.prev_curlyx;
3754 /* XXXX This is too dramatic a measure... */
3756 if ( nochange_depth )
3762 n = ARG(scan); /* which paren pair */
3763 PL_reg_start_tmp[n] = locinput;
3769 n = ARG(scan); /* which paren pair */
3770 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
3771 PL_regoffs[n].end = locinput - PL_bostr;
3772 /*if (n > PL_regsize)
3774 if (n > *PL_reglastparen)
3775 *PL_reglastparen = n;
3776 *PL_reglastcloseparen = n;
3777 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3785 cursor && OP(cursor)!=END;
3786 cursor=regnext(cursor))
3788 if ( OP(cursor)==CLOSE ){
3790 if ( n <= lastopen ) {
3792 = PL_reg_start_tmp[n] - PL_bostr;
3793 PL_regoffs[n].end = locinput - PL_bostr;
3794 /*if (n > PL_regsize)
3796 if (n > *PL_reglastparen)
3797 *PL_reglastparen = n;
3798 *PL_reglastcloseparen = n;
3799 if ( n == ARG(scan) || (cur_eval &&
3800 cur_eval->u.eval.close_paren == n))
3809 n = ARG(scan); /* which paren pair */
3810 sw = (bool)(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
3813 /* reg_check_named_buff_matched returns 0 for no match */
3814 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3818 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3824 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3826 next = NEXTOPER(NEXTOPER(scan));
3828 next = scan + ARG(scan);
3829 if (OP(next) == IFTHEN) /* Fake one. */
3830 next = NEXTOPER(NEXTOPER(next));
3834 logical = scan->flags;
3837 /*******************************************************************
3839 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3840 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3841 STAR/PLUS/CURLY/CURLYN are used instead.)
3843 A*B is compiled as <CURLYX><A><WHILEM><B>
3845 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3846 state, which contains the current count, initialised to -1. It also sets
3847 cur_curlyx to point to this state, with any previous value saved in the
3850 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3851 since the pattern may possibly match zero times (i.e. it's a while {} loop
3852 rather than a do {} while loop).
3854 Each entry to WHILEM represents a successful match of A. The count in the
3855 CURLYX block is incremented, another WHILEM state is pushed, and execution
3856 passes to A or B depending on greediness and the current count.
3858 For example, if matching against the string a1a2a3b (where the aN are
3859 substrings that match /A/), then the match progresses as follows: (the
3860 pushed states are interspersed with the bits of strings matched so far):
3863 <CURLYX cnt=0><WHILEM>
3864 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3865 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3866 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3867 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3869 (Contrast this with something like CURLYM, which maintains only a single
3873 a1 <CURLYM cnt=1> a2
3874 a1 a2 <CURLYM cnt=2> a3
3875 a1 a2 a3 <CURLYM cnt=3> b
3878 Each WHILEM state block marks a point to backtrack to upon partial failure
3879 of A or B, and also contains some minor state data related to that
3880 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3881 overall state, such as the count, and pointers to the A and B ops.
3883 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3884 must always point to the *current* CURLYX block, the rules are:
3886 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3887 and set cur_curlyx to point the new block.
3889 When popping the CURLYX block after a successful or unsuccessful match,
3890 restore the previous cur_curlyx.
3892 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3893 to the outer one saved in the CURLYX block.
3895 When popping the WHILEM block after a successful or unsuccessful B match,
3896 restore the previous cur_curlyx.
3898 Here's an example for the pattern (AI* BI)*BO
3899 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3902 curlyx backtrack stack
3903 ------ ---------------
3905 CO <CO prev=NULL> <WO>
3906 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3907 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3908 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3910 At this point the pattern succeeds, and we work back down the stack to
3911 clean up, restoring as we go:
3913 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3914 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3915 CO <CO prev=NULL> <WO>
3918 *******************************************************************/
3920 #define ST st->u.curlyx
3922 case CURLYX: /* start of /A*B/ (for complex A) */
3924 /* No need to save/restore up to this paren */
3925 I32 parenfloor = scan->flags;
3927 assert(next); /* keep Coverity happy */
3928 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3931 /* XXXX Probably it is better to teach regpush to support
3932 parenfloor > PL_regsize... */
3933 if (parenfloor > (I32)*PL_reglastparen)
3934 parenfloor = *PL_reglastparen; /* Pessimization... */
3936 ST.prev_curlyx= cur_curlyx;
3938 ST.cp = PL_savestack_ix;
3940 /* these fields contain the state of the current curly.
3941 * they are accessed by subsequent WHILEMs */
3942 ST.parenfloor = parenfloor;
3943 ST.min = ARG1(scan);
3944 ST.max = ARG2(scan);
3945 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3949 ST.count = -1; /* this will be updated by WHILEM */
3950 ST.lastloc = NULL; /* this will be updated by WHILEM */
3952 PL_reginput = locinput;
3953 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
3957 case CURLYX_end: /* just finished matching all of A*B */
3958 if (PL_reg_eval_set){
3959 SV *pres= GvSV(PL_replgv);
3962 sv_setsv(GvSV(PL_replgv), pres);
3967 cur_curlyx = ST.prev_curlyx;
3971 case CURLYX_end_fail: /* just failed to match all of A*B */
3973 cur_curlyx = ST.prev_curlyx;
3979 #define ST st->u.whilem
3981 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
3983 /* see the discussion above about CURLYX/WHILEM */
3985 assert(cur_curlyx); /* keep Coverity happy */
3986 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
3987 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
3988 ST.cache_offset = 0;
3991 PL_reginput = locinput;
3993 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3994 "%*s whilem: matched %ld out of %ld..%ld\n",
3995 REPORT_CODE_OFF+depth*2, "", (long)n,
3996 (long)cur_curlyx->u.curlyx.min,
3997 (long)cur_curlyx->u.curlyx.max)
4000 /* First just match a string of min A's. */
4002 if (n < cur_curlyx->u.curlyx.min) {
4003 cur_curlyx->u.curlyx.lastloc = locinput;
4004 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
4008 /* If degenerate A matches "", assume A done. */
4010 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4011 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4012 "%*s whilem: empty match detected, trying continuation...\n",
4013 REPORT_CODE_OFF+depth*2, "")
4015 goto do_whilem_B_max;
4018 /* super-linear cache processing */
4022 if (!PL_reg_maxiter) {
4023 /* start the countdown: Postpone detection until we
4024 * know the match is not *that* much linear. */
4025 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4026 /* possible overflow for long strings and many CURLYX's */
4027 if (PL_reg_maxiter < 0)
4028 PL_reg_maxiter = I32_MAX;
4029 PL_reg_leftiter = PL_reg_maxiter;
4032 if (PL_reg_leftiter-- == 0) {
4033 /* initialise cache */
4034 const I32 size = (PL_reg_maxiter + 7)/8;
4035 if (PL_reg_poscache) {
4036 if ((I32)PL_reg_poscache_size < size) {
4037 Renew(PL_reg_poscache, size, char);
4038 PL_reg_poscache_size = size;
4040 Zero(PL_reg_poscache, size, char);
4043 PL_reg_poscache_size = size;
4044 Newxz(PL_reg_poscache, size, char);
4046 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4047 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4048 PL_colors[4], PL_colors[5])
4052 if (PL_reg_leftiter < 0) {
4053 /* have we already failed at this position? */
4055 offset = (scan->flags & 0xf) - 1
4056 + (locinput - PL_bostr) * (scan->flags>>4);
4057 mask = 1 << (offset % 8);
4059 if (PL_reg_poscache[offset] & mask) {
4060 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4061 "%*s whilem: (cache) already tried at this position...\n",
4062 REPORT_CODE_OFF+depth*2, "")
4064 sayNO; /* cache records failure */
4066 ST.cache_offset = offset;
4067 ST.cache_mask = mask;
4071 /* Prefer B over A for minimal matching. */
4073 if (cur_curlyx->u.curlyx.minmod) {
4074 ST.save_curlyx = cur_curlyx;
4075 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4076 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4077 REGCP_SET(ST.lastcp);
4078 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4082 /* Prefer A over B for maximal matching. */
4084 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
4085 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4086 cur_curlyx->u.curlyx.lastloc = locinput;
4087 REGCP_SET(ST.lastcp);
4088 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
4091 goto do_whilem_B_max;
4095 case WHILEM_B_min: /* just matched B in a minimal match */
4096 case WHILEM_B_max: /* just matched B in a maximal match */
4097 cur_curlyx = ST.save_curlyx;
4101 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4102 cur_curlyx = ST.save_curlyx;
4103 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4104 cur_curlyx->u.curlyx.count--;
4108 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4109 REGCP_UNWIND(ST.lastcp);
4112 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4113 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4114 cur_curlyx->u.curlyx.count--;
4118 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4119 REGCP_UNWIND(ST.lastcp);
4120 regcppop(rex); /* Restore some previous $<digit>s? */
4121 PL_reginput = locinput;
4122 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4123 "%*s whilem: failed, trying continuation...\n",
4124 REPORT_CODE_OFF+depth*2, "")
4127 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4128 && ckWARN(WARN_REGEXP)
4129 && !(PL_reg_flags & RF_warned))
4131 PL_reg_flags |= RF_warned;
4132 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4133 "Complex regular subexpression recursion",
4138 ST.save_curlyx = cur_curlyx;
4139 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4140 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4143 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4144 cur_curlyx = ST.save_curlyx;
4145 REGCP_UNWIND(ST.lastcp);
4148 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
4149 /* Maximum greed exceeded */
4150 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4151 && ckWARN(WARN_REGEXP)
4152 && !(PL_reg_flags & RF_warned))
4154 PL_reg_flags |= RF_warned;
4155 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4156 "%s limit (%d) exceeded",
4157 "Complex regular subexpression recursion",
4160 cur_curlyx->u.curlyx.count--;
4164 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4165 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4167 /* Try grabbing another A and see if it helps. */
4168 PL_reginput = locinput;
4169 cur_curlyx->u.curlyx.lastloc = locinput;
4170 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4171 REGCP_SET(ST.lastcp);
4172 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
4176 #define ST st->u.branch
4178 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4179 next = scan + ARG(scan);
4182 scan = NEXTOPER(scan);
4185 case BRANCH: /* /(...|A|...)/ */
4186 scan = NEXTOPER(scan); /* scan now points to inner node */
4187 if ((!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
4190 /* last branch; skip state push and jump direct to node */
4193 ST.lastparen = *PL_reglastparen;
4194 ST.next_branch = next;
4196 PL_reginput = locinput;
4198 /* Now go into the branch */
4200 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4202 PUSH_STATE_GOTO(BRANCH_next, scan);
4206 PL_reginput = locinput;
4207 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4208 (SV*)rexi->data->data[ ARG( scan ) ];
4209 PUSH_STATE_GOTO(CUTGROUP_next,next);
4211 case CUTGROUP_next_fail:
4214 if (st->u.mark.mark_name)
4215 sv_commit = st->u.mark.mark_name;
4221 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4226 REGCP_UNWIND(ST.cp);
4227 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4228 PL_regoffs[n].end = -1;
4229 *PL_reglastparen = n;
4230 /*dmq: *PL_reglastcloseparen = n; */
4231 scan = ST.next_branch;
4232 /* no more branches? */
4233 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4235 PerlIO_printf( Perl_debug_log,
4236 "%*s %sBRANCH failed...%s\n",
4237 REPORT_CODE_OFF+depth*2, "",
4243 continue; /* execute next BRANCH[J] op */
4251 #define ST st->u.curlym
4253 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4255 /* This is an optimisation of CURLYX that enables us to push
4256 * only a single backtracking state, no matter now many matches
4257 * there are in {m,n}. It relies on the pattern being constant
4258 * length, with no parens to influence future backrefs
4262 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4264 /* if paren positive, emulate an OPEN/CLOSE around A */
4266 U32 paren = ST.me->flags;
4267 if (paren > PL_regsize)
4269 if (paren > *PL_reglastparen)
4270 *PL_reglastparen = paren;
4271 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4279 ST.c1 = CHRTEST_UNINIT;
4282 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4285 curlym_do_A: /* execute the A in /A{m,n}B/ */
4286 PL_reginput = locinput;
4287 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4290 case CURLYM_A: /* we've just matched an A */
4291 locinput = st->locinput;
4292 nextchr = UCHARAT(locinput);
4295 /* after first match, determine A's length: u.curlym.alen */
4296 if (ST.count == 1) {
4297 if (PL_reg_match_utf8) {
4299 while (s < PL_reginput) {
4305 ST.alen = PL_reginput - locinput;
4308 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4311 PerlIO_printf(Perl_debug_log,
4312 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4313 (int)(REPORT_CODE_OFF+(depth*2)), "",
4314 (IV) ST.count, (IV)ST.alen)
4317 locinput = PL_reginput;
4319 if (cur_eval && cur_eval->u.eval.close_paren &&
4320 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4323 if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
4324 goto curlym_do_A; /* try to match another A */
4325 goto curlym_do_B; /* try to match B */
4327 case CURLYM_A_fail: /* just failed to match an A */
4328 REGCP_UNWIND(ST.cp);
4330 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4331 || (cur_eval && cur_eval->u.eval.close_paren &&
4332 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4335 curlym_do_B: /* execute the B in /A{m,n}B/ */
4336 PL_reginput = locinput;
4337 if (ST.c1 == CHRTEST_UNINIT) {
4338 /* calculate c1 and c2 for possible match of 1st char
4339 * following curly */
4340 ST.c1 = ST.c2 = CHRTEST_VOID;
4341 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4342 regnode *text_node = ST.B;
4343 if (! HAS_TEXT(text_node))
4344 FIND_NEXT_IMPT(text_node);
4347 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4349 But the former is redundant in light of the latter.
4351 if this changes back then the macro for
4352 IS_TEXT and friends need to change.
4354 if (PL_regkind[OP(text_node)] == EXACT)
4357 ST.c1 = (U8)*STRING(text_node);
4359 (IS_TEXTF(text_node))
4361 : (IS_TEXTFL(text_node))
4362 ? PL_fold_locale[ST.c1]
4369 PerlIO_printf(Perl_debug_log,
4370 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4371 (int)(REPORT_CODE_OFF+(depth*2)),
4374 if (ST.c1 != CHRTEST_VOID
4375 && UCHARAT(PL_reginput) != ST.c1
4376 && UCHARAT(PL_reginput) != ST.c2)
4378 /* simulate B failing */
4380 PerlIO_printf(Perl_debug_log,
4381 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
4382 (int)(REPORT_CODE_OFF+(depth*2)),"",
4385 state_num = CURLYM_B_fail;
4386 goto reenter_switch;
4390 /* mark current A as captured */
4391 I32 paren = ST.me->flags;
4393 PL_regoffs[paren].start
4394 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4395 PL_regoffs[paren].end = PL_reginput - PL_bostr;
4396 /*dmq: *PL_reglastcloseparen = paren; */
4399 PL_regoffs[paren].end = -1;
4400 if (cur_eval && cur_eval->u.eval.close_paren &&
4401 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4410 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4413 case CURLYM_B_fail: /* just failed to match a B */
4414 REGCP_UNWIND(ST.cp);
4416 if (ST.count == ARG2(ST.me) /* max */)
4418 goto curlym_do_A; /* try to match a further A */
4420 /* backtrack one A */
4421 if (ST.count == ARG1(ST.me) /* min */)
4424 locinput = HOPc(locinput, -ST.alen);
4425 goto curlym_do_B; /* try to match B */
4428 #define ST st->u.curly
4430 #define CURLY_SETPAREN(paren, success) \
4433 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
4434 PL_regoffs[paren].end = locinput - PL_bostr; \
4435 *PL_reglastcloseparen = paren; \
4438 PL_regoffs[paren].end = -1; \
4441 case STAR: /* /A*B/ where A is width 1 */
4445 scan = NEXTOPER(scan);
4447 case PLUS: /* /A+B/ where A is width 1 */
4451 scan = NEXTOPER(scan);
4453 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4454 ST.paren = scan->flags; /* Which paren to set */
4455 if (ST.paren > PL_regsize)
4456 PL_regsize = ST.paren;
4457 if (ST.paren > *PL_reglastparen)
4458 *PL_reglastparen = ST.paren;
4459 ST.min = ARG1(scan); /* min to match */
4460 ST.max = ARG2(scan); /* max to match */
4461 if (cur_eval && cur_eval->u.eval.close_paren &&
4462 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4466 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4468 case CURLY: /* /A{m,n}B/ where A is width 1 */
4470 ST.min = ARG1(scan); /* min to match */
4471 ST.max = ARG2(scan); /* max to match */
4472 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4475 * Lookahead to avoid useless match attempts
4476 * when we know what character comes next.
4478 * Used to only do .*x and .*?x, but now it allows
4479 * for )'s, ('s and (?{ ... })'s to be in the way
4480 * of the quantifier and the EXACT-like node. -- japhy
4483 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4485 if (HAS_TEXT(next) || JUMPABLE(next)) {
4487 regnode *text_node = next;
4489 if (! HAS_TEXT(text_node))
4490 FIND_NEXT_IMPT(text_node);
4492 if (! HAS_TEXT(text_node))
4493 ST.c1 = ST.c2 = CHRTEST_VOID;
4495 if ( PL_regkind[OP(text_node)] != EXACT ) {
4496 ST.c1 = ST.c2 = CHRTEST_VOID;
4497 goto assume_ok_easy;
4500 s = (U8*)STRING(text_node);
4502 /* Currently we only get here when
4504 PL_rekind[OP(text_node)] == EXACT
4506 if this changes back then the macro for IS_TEXT and
4507 friends need to change. */
4510 if (IS_TEXTF(text_node))
4511 ST.c2 = PL_fold[ST.c1];
4512 else if (IS_TEXTFL(text_node))
4513 ST.c2 = PL_fold_locale[ST.c1];
4516 if (IS_TEXTF(text_node)) {
4517 STRLEN ulen1, ulen2;
4518 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4519 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4521 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4522 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4524 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4526 0 : UTF8_ALLOW_ANY);
4527 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4529 0 : UTF8_ALLOW_ANY);
4531 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4533 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4538 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4545 ST.c1 = ST.c2 = CHRTEST_VOID;
4550 PL_reginput = locinput;
4553 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4556 locinput = PL_reginput;
4558 if (ST.c1 == CHRTEST_VOID)
4559 goto curly_try_B_min;
4561 ST.oldloc = locinput;
4563 /* set ST.maxpos to the furthest point along the
4564 * string that could possibly match */
4565 if (ST.max == REG_INFTY) {
4566 ST.maxpos = PL_regeol - 1;
4568 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4572 int m = ST.max - ST.min;
4573 for (ST.maxpos = locinput;
4574 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4575 ST.maxpos += UTF8SKIP(ST.maxpos);
4578 ST.maxpos = locinput + ST.max - ST.min;
4579 if (ST.maxpos >= PL_regeol)
4580 ST.maxpos = PL_regeol - 1;
4582 goto curly_try_B_min_known;
4586 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4587 locinput = PL_reginput;
4588 if (ST.count < ST.min)
4590 if ((ST.count > ST.min)
4591 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4593 /* A{m,n} must come at the end of the string, there's
4594 * no point in backing off ... */
4596 /* ...except that $ and \Z can match before *and* after
4597 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4598 We may back off by one in this case. */
4599 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4603 goto curly_try_B_max;
4608 case CURLY_B_min_known_fail:
4609 /* failed to find B in a non-greedy match where c1,c2 valid */
4610 if (ST.paren && ST.count)
4611 PL_regoffs[ST.paren].end = -1;
4613 PL_reginput = locinput; /* Could be reset... */
4614 REGCP_UNWIND(ST.cp);
4615 /* Couldn't or didn't -- move forward. */
4616 ST.oldloc = locinput;
4618 locinput += UTF8SKIP(locinput);
4622 curly_try_B_min_known:
4623 /* find the next place where 'B' could work, then call B */
4627 n = (ST.oldloc == locinput) ? 0 : 1;
4628 if (ST.c1 == ST.c2) {
4630 /* set n to utf8_distance(oldloc, locinput) */
4631 while (locinput <= ST.maxpos &&
4632 utf8n_to_uvchr((U8*)locinput,
4633 UTF8_MAXBYTES, &len,
4634 uniflags) != (UV)ST.c1) {
4640 /* set n to utf8_distance(oldloc, locinput) */
4641 while (locinput <= ST.maxpos) {
4643 const UV c = utf8n_to_uvchr((U8*)locinput,
4644 UTF8_MAXBYTES, &len,
4646 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4654 if (ST.c1 == ST.c2) {
4655 while (locinput <= ST.maxpos &&
4656 UCHARAT(locinput) != ST.c1)
4660 while (locinput <= ST.maxpos
4661 && UCHARAT(locinput) != ST.c1
4662 && UCHARAT(locinput) != ST.c2)
4665 n = locinput - ST.oldloc;
4667 if (locinput > ST.maxpos)
4669 /* PL_reginput == oldloc now */
4672 if (regrepeat(rex, ST.A, n, depth) < n)
4675 PL_reginput = locinput;
4676 CURLY_SETPAREN(ST.paren, ST.count);
4677 if (cur_eval && cur_eval->u.eval.close_paren &&
4678 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4681 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4686 case CURLY_B_min_fail:
4687 /* failed to find B in a non-greedy match where c1,c2 invalid */
4688 if (ST.paren && ST.count)
4689 PL_regoffs[ST.paren].end = -1;
4691 REGCP_UNWIND(ST.cp);
4692 /* failed -- move forward one */
4693 PL_reginput = locinput;
4694 if (regrepeat(rex, ST.A, 1, depth)) {
4696 locinput = PL_reginput;
4697 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4698 ST.count > 0)) /* count overflow ? */
4701 CURLY_SETPAREN(ST.paren, ST.count);
4702 if (cur_eval && cur_eval->u.eval.close_paren &&
4703 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4706 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4714 /* a successful greedy match: now try to match B */
4715 if (cur_eval && cur_eval->u.eval.close_paren &&
4716 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4721 if (ST.c1 != CHRTEST_VOID)
4722 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4723 UTF8_MAXBYTES, 0, uniflags)
4724 : (UV) UCHARAT(PL_reginput);
4725 /* If it could work, try it. */
4726 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4727 CURLY_SETPAREN(ST.paren, ST.count);
4728 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4733 case CURLY_B_max_fail:
4734 /* failed to find B in a greedy match */
4735 if (ST.paren && ST.count)
4736 PL_regoffs[ST.paren].end = -1;
4738 REGCP_UNWIND(ST.cp);
4740 if (--ST.count < ST.min)
4742 PL_reginput = locinput = HOPc(locinput, -1);
4743 goto curly_try_B_max;
4750 /* we've just finished A in /(??{A})B/; now continue with B */
4752 st->u.eval.toggle_reg_flags
4753 = cur_eval->u.eval.toggle_reg_flags;
4754 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4756 st->u.eval.prev_rex = rex; /* inner */
4757 SETREX(rex,cur_eval->u.eval.prev_rex);
4758 rexi = RXi_GET(rex);
4759 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4761 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4762 REGCP_SET(st->u.eval.lastcp);
4763 PL_reginput = locinput;
4765 /* Restore parens of the outer rex without popping the
4767 tmpix = PL_savestack_ix;
4768 PL_savestack_ix = cur_eval->u.eval.lastcp;
4770 PL_savestack_ix = tmpix;
4772 st->u.eval.prev_eval = cur_eval;
4773 cur_eval = cur_eval->u.eval.prev_eval;
4775 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
4776 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
4777 if ( nochange_depth )
4780 PUSH_YES_STATE_GOTO(EVAL_AB,
4781 st->u.eval.prev_eval->u.eval.B); /* match B */
4784 if (locinput < reginfo->till) {
4785 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4786 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4788 (long)(locinput - PL_reg_starttry),
4789 (long)(reginfo->till - PL_reg_starttry),
4792 sayNO_SILENT; /* Cannot match: too short. */
4794 PL_reginput = locinput; /* put where regtry can find it */
4795 sayYES; /* Success! */
4797 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4799 PerlIO_printf(Perl_debug_log,
4800 "%*s %ssubpattern success...%s\n",
4801 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4802 PL_reginput = locinput; /* put where regtry can find it */
4803 sayYES; /* Success! */
4806 #define ST st->u.ifmatch
4808 case SUSPEND: /* (?>A) */
4810 PL_reginput = locinput;
4813 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4815 goto ifmatch_trivial_fail_test;
4817 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4819 ifmatch_trivial_fail_test:
4821 char * const s = HOPBACKc(locinput, scan->flags);
4826 sw = 1 - (bool)ST.wanted;
4830 next = scan + ARG(scan);
4838 PL_reginput = locinput;
4842 ST.logical = logical;
4843 /* execute body of (?...A) */
4844 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4847 case IFMATCH_A_fail: /* body of (?...A) failed */
4848 ST.wanted = !ST.wanted;
4851 case IFMATCH_A: /* body of (?...A) succeeded */
4853 sw = (bool)ST.wanted;
4855 else if (!ST.wanted)
4858 if (OP(ST.me) == SUSPEND)
4859 locinput = PL_reginput;
4861 locinput = PL_reginput = st->locinput;
4862 nextchr = UCHARAT(locinput);
4864 scan = ST.me + ARG(ST.me);
4867 continue; /* execute B */
4872 next = scan + ARG(scan);
4877 reginfo->cutpoint = PL_regeol;
4880 PL_reginput = locinput;
4882 sv_yes_mark = sv_commit = (SV*)rexi->data->data[ ARG( scan ) ];
4883 PUSH_STATE_GOTO(COMMIT_next,next);
4885 case COMMIT_next_fail:
4892 #define ST st->u.mark
4894 ST.prev_mark = mark_state;
4895 ST.mark_name = sv_commit = sv_yes_mark
4896 = (SV*)rexi->data->data[ ARG( scan ) ];
4898 ST.mark_loc = PL_reginput = locinput;
4899 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
4901 case MARKPOINT_next:
4902 mark_state = ST.prev_mark;
4905 case MARKPOINT_next_fail:
4906 if (popmark && sv_eq(ST.mark_name,popmark))
4908 if (ST.mark_loc > startpoint)
4909 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
4910 popmark = NULL; /* we found our mark */
4911 sv_commit = ST.mark_name;
4914 PerlIO_printf(Perl_debug_log,
4915 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
4916 REPORT_CODE_OFF+depth*2, "",
4917 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
4920 mark_state = ST.prev_mark;
4921 sv_yes_mark = mark_state ?
4922 mark_state->u.mark.mark_name : NULL;
4926 PL_reginput = locinput;
4928 /* (*SKIP) : if we fail we cut here*/
4929 ST.mark_name = NULL;
4930 ST.mark_loc = locinput;
4931 PUSH_STATE_GOTO(SKIP_next,next);
4933 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
4934 otherwise do nothing. Meaning we need to scan
4936 regmatch_state *cur = mark_state;
4937 SV *find = (SV*)rexi->data->data[ ARG( scan ) ];
4940 if ( sv_eq( cur->u.mark.mark_name,
4943 ST.mark_name = find;
4944 PUSH_STATE_GOTO( SKIP_next, next );
4946 cur = cur->u.mark.prev_mark;
4949 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
4951 case SKIP_next_fail:
4953 /* (*CUT:NAME) - Set up to search for the name as we
4954 collapse the stack*/
4955 popmark = ST.mark_name;
4957 /* (*CUT) - No name, we cut here.*/
4958 if (ST.mark_loc > startpoint)
4959 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
4960 /* but we set sv_commit to latest mark_name if there
4961 is one so they can test to see how things lead to this
4964 sv_commit=mark_state->u.mark.mark_name;
4972 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
4973 PTR2UV(scan), OP(scan));
4974 Perl_croak(aTHX_ "regexp memory corruption");
4978 /* switch break jumps here */
4979 scan = next; /* prepare to execute the next op and ... */
4980 continue; /* ... jump back to the top, reusing st */
4984 /* push a state that backtracks on success */
4985 st->u.yes.prev_yes_state = yes_state;
4989 /* push a new regex state, then continue at scan */
4991 regmatch_state *newst;
4994 regmatch_state *cur = st;
4995 regmatch_state *curyes = yes_state;
4997 regmatch_slab *slab = PL_regmatch_slab;
4998 for (;curd > -1;cur--,curd--) {
4999 if (cur < SLAB_FIRST(slab)) {
5001 cur = SLAB_LAST(slab);
5003 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5004 REPORT_CODE_OFF + 2 + depth * 2,"",
5005 curd, PL_reg_name[cur->resume_state],
5006 (curyes == cur) ? "yes" : ""
5009 curyes = cur->u.yes.prev_yes_state;
5012 DEBUG_STATE_pp("push")
5015 st->locinput = locinput;
5017 if (newst > SLAB_LAST(PL_regmatch_slab))
5018 newst = S_push_slab(aTHX);
5019 PL_regmatch_state = newst;
5021 locinput = PL_reginput;
5022 nextchr = UCHARAT(locinput);
5030 * We get here only if there's trouble -- normally "case END" is
5031 * the terminating point.
5033 Perl_croak(aTHX_ "corrupted regexp pointers");
5039 /* we have successfully completed a subexpression, but we must now
5040 * pop to the state marked by yes_state and continue from there */
5041 assert(st != yes_state);
5043 while (st != yes_state) {
5045 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5046 PL_regmatch_slab = PL_regmatch_slab->prev;
5047 st = SLAB_LAST(PL_regmatch_slab);
5051 DEBUG_STATE_pp("pop (no final)");
5053 DEBUG_STATE_pp("pop (yes)");
5059 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5060 || yes_state > SLAB_LAST(PL_regmatch_slab))
5062 /* not in this slab, pop slab */
5063 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5064 PL_regmatch_slab = PL_regmatch_slab->prev;
5065 st = SLAB_LAST(PL_regmatch_slab);
5067 depth -= (st - yes_state);
5070 yes_state = st->u.yes.prev_yes_state;
5071 PL_regmatch_state = st;
5074 locinput= st->locinput;
5075 nextchr = UCHARAT(locinput);
5077 state_num = st->resume_state + no_final;
5078 goto reenter_switch;
5081 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5082 PL_colors[4], PL_colors[5]));
5089 PerlIO_printf(Perl_debug_log,
5090 "%*s %sfailed...%s\n",
5091 REPORT_CODE_OFF+depth*2, "",
5092 PL_colors[4], PL_colors[5])
5104 /* there's a previous state to backtrack to */
5106 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5107 PL_regmatch_slab = PL_regmatch_slab->prev;
5108 st = SLAB_LAST(PL_regmatch_slab);
5110 PL_regmatch_state = st;
5111 locinput= st->locinput;
5112 nextchr = UCHARAT(locinput);
5114 DEBUG_STATE_pp("pop");
5116 if (yes_state == st)
5117 yes_state = st->u.yes.prev_yes_state;
5119 state_num = st->resume_state + 1; /* failure = success + 1 */
5120 goto reenter_switch;
5125 if (rex->intflags & PREGf_VERBARG_SEEN) {
5126 SV *sv_err = get_sv("REGERROR", 1);
5127 SV *sv_mrk = get_sv("REGMARK", 1);
5129 sv_commit = &PL_sv_no;
5131 sv_yes_mark = &PL_sv_yes;
5134 sv_commit = &PL_sv_yes;
5135 sv_yes_mark = &PL_sv_no;
5137 sv_setsv(sv_err, sv_commit);
5138 sv_setsv(sv_mrk, sv_yes_mark);
5140 /* restore original high-water mark */
5141 PL_regmatch_slab = orig_slab;
5142 PL_regmatch_state = orig_state;
5144 /* free all slabs above current one */
5145 if (orig_slab->next) {
5146 regmatch_slab *sl = orig_slab->next;
5147 orig_slab->next = NULL;
5149 regmatch_slab * const osl = sl;
5159 - regrepeat - repeatedly match something simple, report how many
5162 * [This routine now assumes that it will only match on things of length 1.
5163 * That was true before, but now we assume scan - reginput is the count,
5164 * rather than incrementing count on every character. [Er, except utf8.]]
5167 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5170 register char *scan;
5172 register char *loceol = PL_regeol;
5173 register I32 hardcount = 0;
5174 register bool do_utf8 = PL_reg_match_utf8;
5176 PERL_UNUSED_ARG(depth);
5180 if (max == REG_INFTY)
5182 else if (max < loceol - scan)
5183 loceol = scan + max;
5188 while (scan < loceol && hardcount < max && *scan != '\n') {
5189 scan += UTF8SKIP(scan);
5193 while (scan < loceol && *scan != '\n')
5200 while (scan < loceol && hardcount < max) {
5201 scan += UTF8SKIP(scan);
5211 case EXACT: /* length of string is 1 */
5213 while (scan < loceol && UCHARAT(scan) == c)
5216 case EXACTF: /* length of string is 1 */
5218 while (scan < loceol &&
5219 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5222 case EXACTFL: /* length of string is 1 */
5223 PL_reg_flags |= RF_tainted;
5225 while (scan < loceol &&
5226 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5232 while (hardcount < max && scan < loceol &&
5233 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
5234 scan += UTF8SKIP(scan);
5238 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5245 LOAD_UTF8_CHARCLASS_ALNUM();
5246 while (hardcount < max && scan < loceol &&
5247 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5248 scan += UTF8SKIP(scan);
5252 while (scan < loceol && isALNUM(*scan))
5257 PL_reg_flags |= RF_tainted;
5260 while (hardcount < max && scan < loceol &&
5261 isALNUM_LC_utf8((U8*)scan)) {
5262 scan += UTF8SKIP(scan);
5266 while (scan < loceol && isALNUM_LC(*scan))
5273 LOAD_UTF8_CHARCLASS_ALNUM();
5274 while (hardcount < max && scan < loceol &&
5275 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5276 scan += UTF8SKIP(scan);
5280 while (scan < loceol && !isALNUM(*scan))
5285 PL_reg_flags |= RF_tainted;
5288 while (hardcount < max && scan < loceol &&
5289 !isALNUM_LC_utf8((U8*)scan)) {
5290 scan += UTF8SKIP(scan);
5294 while (scan < loceol && !isALNUM_LC(*scan))
5301 LOAD_UTF8_CHARCLASS_SPACE();
5302 while (hardcount < max && scan < loceol &&
5304 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5305 scan += UTF8SKIP(scan);
5309 while (scan < loceol && isSPACE(*scan))
5314 PL_reg_flags |= RF_tainted;
5317 while (hardcount < max && scan < loceol &&
5318 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5319 scan += UTF8SKIP(scan);
5323 while (scan < loceol && isSPACE_LC(*scan))
5330 LOAD_UTF8_CHARCLASS_SPACE();
5331 while (hardcount < max && scan < loceol &&
5333 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5334 scan += UTF8SKIP(scan);
5338 while (scan < loceol && !isSPACE(*scan))
5343 PL_reg_flags |= RF_tainted;
5346 while (hardcount < max && scan < loceol &&
5347 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5348 scan += UTF8SKIP(scan);
5352 while (scan < loceol && !isSPACE_LC(*scan))
5359 LOAD_UTF8_CHARCLASS_DIGIT();
5360 while (hardcount < max && scan < loceol &&
5361 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5362 scan += UTF8SKIP(scan);
5366 while (scan < loceol && isDIGIT(*scan))
5373 LOAD_UTF8_CHARCLASS_DIGIT();
5374 while (hardcount < max && scan < loceol &&
5375 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5376 scan += UTF8SKIP(scan);
5380 while (scan < loceol && !isDIGIT(*scan))
5384 default: /* Called on something of 0 width. */
5385 break; /* So match right here or not at all. */
5391 c = scan - PL_reginput;
5395 GET_RE_DEBUG_FLAGS_DECL;
5397 SV * const prop = sv_newmortal();
5398 regprop(prog, prop, p);
5399 PerlIO_printf(Perl_debug_log,
5400 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5401 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5409 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5411 - regclass_swash - prepare the utf8 swash
5415 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5421 RXi_GET_DECL(prog,progi);
5422 const struct reg_data * const data = prog ? progi->data : NULL;
5424 if (data && data->count) {
5425 const U32 n = ARG(node);
5427 if (data->what[n] == 's') {
5428 SV * const rv = (SV*)data->data[n];
5429 AV * const av = (AV*)SvRV((SV*)rv);
5430 SV **const ary = AvARRAY(av);
5433 /* See the end of regcomp.c:S_regclass() for
5434 * documentation of these array elements. */
5437 a = SvROK(ary[1]) ? &ary[1] : 0;
5438 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
5442 else if (si && doinit) {
5443 sw = swash_init("utf8", "", si, 1, 0);
5444 (void)av_store(av, 1, sw);
5461 - reginclass - determine if a character falls into a character class
5463 The n is the ANYOF regnode, the p is the target string, lenp
5464 is pointer to the maximum length of how far to go in the p
5465 (if the lenp is zero, UTF8SKIP(p) is used),
5466 do_utf8 tells whether the target string is in UTF-8.
5471 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5474 const char flags = ANYOF_FLAGS(n);
5480 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5481 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5482 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5483 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5484 if (len == (STRLEN)-1)
5485 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5488 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5489 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5492 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5493 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5496 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5500 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5503 if (swash_fetch(sw, p, do_utf8))
5505 else if (flags & ANYOF_FOLD) {
5506 if (!match && lenp && av) {
5508 for (i = 0; i <= av_len(av); i++) {
5509 SV* const sv = *av_fetch(av, i, FALSE);
5511 const char * const s = SvPV_const(sv, len);
5513 if (len <= plen && memEQ(s, (char*)p, len)) {
5521 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5524 to_utf8_fold(p, tmpbuf, &tmplen);
5525 if (swash_fetch(sw, tmpbuf, do_utf8))
5531 if (match && lenp && *lenp == 0)
5532 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5534 if (!match && c < 256) {
5535 if (ANYOF_BITMAP_TEST(n, c))
5537 else if (flags & ANYOF_FOLD) {
5540 if (flags & ANYOF_LOCALE) {
5541 PL_reg_flags |= RF_tainted;
5542 f = PL_fold_locale[c];
5546 if (f != c && ANYOF_BITMAP_TEST(n, f))
5550 if (!match && (flags & ANYOF_CLASS)) {
5551 PL_reg_flags |= RF_tainted;
5553 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5554 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5555 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5556 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5557 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5558 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5559 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5560 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5561 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5562 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5563 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5564 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5565 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5566 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5567 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5568 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5569 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5570 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5571 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5572 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5573 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5574 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5575 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5576 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5577 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5578 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5579 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5580 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5581 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5582 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5583 ) /* How's that for a conditional? */
5590 return (flags & ANYOF_INVERT) ? !match : match;
5594 S_reghop3(U8 *s, I32 off, const U8* lim)
5598 while (off-- && s < lim) {
5599 /* XXX could check well-formedness here */
5604 while (off++ && s > lim) {
5606 if (UTF8_IS_CONTINUED(*s)) {
5607 while (s > lim && UTF8_IS_CONTINUATION(*s))
5610 /* XXX could check well-formedness here */
5617 /* there are a bunch of places where we use two reghop3's that should
5618 be replaced with this routine. but since thats not done yet
5619 we ifdef it out - dmq
5622 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5626 while (off-- && s < rlim) {
5627 /* XXX could check well-formedness here */
5632 while (off++ && s > llim) {
5634 if (UTF8_IS_CONTINUED(*s)) {
5635 while (s > llim && UTF8_IS_CONTINUATION(*s))
5638 /* XXX could check well-formedness here */
5646 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5650 while (off-- && s < lim) {
5651 /* XXX could check well-formedness here */
5658 while (off++ && s > lim) {
5660 if (UTF8_IS_CONTINUED(*s)) {
5661 while (s > lim && UTF8_IS_CONTINUATION(*s))
5664 /* XXX could check well-formedness here */
5673 restore_pos(pTHX_ void *arg)
5676 regexp * const rex = (regexp *)arg;
5677 if (PL_reg_eval_set) {
5678 if (PL_reg_oldsaved) {
5679 rex->subbeg = PL_reg_oldsaved;
5680 rex->sublen = PL_reg_oldsavedlen;
5681 #ifdef PERL_OLD_COPY_ON_WRITE
5682 rex->saved_copy = PL_nrs;
5684 RX_MATCH_COPIED_on(rex);
5686 PL_reg_magic->mg_len = PL_reg_oldpos;
5687 PL_reg_eval_set = 0;
5688 PL_curpm = PL_reg_oldcurpm;
5693 S_to_utf8_substr(pTHX_ register regexp *prog)
5697 if (prog->substrs->data[i].substr
5698 && !prog->substrs->data[i].utf8_substr) {
5699 SV* const sv = newSVsv(prog->substrs->data[i].substr);
5700 prog->substrs->data[i].utf8_substr = sv;
5701 sv_utf8_upgrade(sv);
5702 if (SvVALID(prog->substrs->data[i].substr)) {
5703 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
5704 if (flags & FBMcf_TAIL) {
5705 /* Trim the trailing \n that fbm_compile added last
5707 SvCUR_set(sv, SvCUR(sv) - 1);
5708 /* Whilst this makes the SV technically "invalid" (as its
5709 buffer is no longer followed by "\0") when fbm_compile()
5710 adds the "\n" back, a "\0" is restored. */
5712 fbm_compile(sv, flags);
5714 if (prog->substrs->data[i].substr == prog->check_substr)
5715 prog->check_utf8 = sv;
5721 S_to_byte_substr(pTHX_ register regexp *prog)
5726 if (prog->substrs->data[i].utf8_substr
5727 && !prog->substrs->data[i].substr) {
5728 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
5729 if (sv_utf8_downgrade(sv, TRUE)) {
5730 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
5732 = BmFLAGS(prog->substrs->data[i].utf8_substr);
5733 if (flags & FBMcf_TAIL) {
5734 /* Trim the trailing \n that fbm_compile added last
5736 SvCUR_set(sv, SvCUR(sv) - 1);
5738 fbm_compile(sv, flags);
5744 prog->substrs->data[i].substr = sv;
5745 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
5746 prog->check_substr = sv;
5753 * c-indentation-style: bsd
5755 * indent-tabs-mode: t
5758 * ex: set ts=8 sts=4 sw=4 noet: