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 const bool do_utf8 = (bool)DO_UTF8(sv);
1695 RXi_GET_DECL(prog,progi);
1696 regmatch_info reginfo; /* create some info to pass to regtry etc */
1697 bool swap_on_fail = 0;
1699 GET_RE_DEBUG_FLAGS_DECL;
1701 PERL_UNUSED_ARG(data);
1703 /* Be paranoid... */
1704 if (prog == NULL || startpos == NULL) {
1705 Perl_croak(aTHX_ "NULL regexp parameter");
1709 multiline = prog->extflags & RXf_PMf_MULTILINE;
1710 reginfo.prog = prog;
1712 RX_MATCH_UTF8_set(prog, do_utf8);
1714 debug_start_match(prog, do_utf8, startpos, strend,
1718 minlen = prog->minlen;
1720 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1721 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1722 "String too short [regexec_flags]...\n"));
1727 /* Check validity of program. */
1728 if (UCHARAT(progi->program) != REG_MAGIC) {
1729 Perl_croak(aTHX_ "corrupted regexp program");
1733 PL_reg_eval_set = 0;
1736 if (prog->extflags & RXf_UTF8)
1737 PL_reg_flags |= RF_utf8;
1739 /* Mark beginning of line for ^ and lookbehind. */
1740 reginfo.bol = startpos; /* XXX not used ??? */
1744 /* Mark end of line for $ (and such) */
1747 /* see how far we have to get to not match where we matched before */
1748 reginfo.till = startpos+minend;
1750 /* If there is a "must appear" string, look for it. */
1753 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1756 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1757 reginfo.ganch = startpos + prog->gofs;
1758 else if (sv && SvTYPE(sv) >= SVt_PVMG
1760 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1761 && mg->mg_len >= 0) {
1762 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1763 if (prog->extflags & RXf_ANCH_GPOS) {
1764 if (s > reginfo.ganch)
1766 s = reginfo.ganch - prog->gofs;
1770 reginfo.ganch = strbeg + PTR2UV(data);
1771 } else /* pos() not defined */
1772 reginfo.ganch = strbeg;
1774 if (PL_curpm && (PM_GETRE(PL_curpm) == prog)) {
1776 swap_match_buff(prog); /* do we need a save destructor here for
1779 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1780 re_scream_pos_data d;
1782 d.scream_olds = &scream_olds;
1783 d.scream_pos = &scream_pos;
1784 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1786 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1787 goto phooey; /* not present */
1793 /* Simplest case: anchored match need be tried only once. */
1794 /* [unless only anchor is BOL and multiline is set] */
1795 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
1796 if (s == startpos && regtry(®info, &startpos))
1798 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
1799 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
1804 dontbother = minlen - 1;
1805 end = HOP3c(strend, -dontbother, strbeg) - 1;
1806 /* for multiline we only have to try after newlines */
1807 if (prog->check_substr || prog->check_utf8) {
1811 if (regtry(®info, &s))
1816 if (prog->extflags & RXf_USE_INTUIT) {
1817 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1828 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1829 if (regtry(®info, &s))
1836 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
1838 /* the warning about reginfo.ganch being used without intialization
1839 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
1840 and we only enter this block when the same bit is set. */
1841 char *tmp_s = reginfo.ganch - prog->gofs;
1842 if (regtry(®info, &tmp_s))
1847 /* Messy cases: unanchored match. */
1848 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
1849 /* we have /x+whatever/ */
1850 /* it must be a one character string (XXXX Except UTF?) */
1855 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1856 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1857 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1862 DEBUG_EXECUTE_r( did_match = 1 );
1863 if (regtry(®info, &s)) goto got_it;
1865 while (s < strend && *s == ch)
1873 DEBUG_EXECUTE_r( did_match = 1 );
1874 if (regtry(®info, &s)) goto got_it;
1876 while (s < strend && *s == ch)
1881 DEBUG_EXECUTE_r(if (!did_match)
1882 PerlIO_printf(Perl_debug_log,
1883 "Did not find anchored character...\n")
1886 else if (prog->anchored_substr != NULL
1887 || prog->anchored_utf8 != NULL
1888 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1889 && prog->float_max_offset < strend - s)) {
1894 char *last1; /* Last position checked before */
1898 if (prog->anchored_substr || prog->anchored_utf8) {
1899 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1900 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1901 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1902 back_max = back_min = prog->anchored_offset;
1904 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1905 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1906 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1907 back_max = prog->float_max_offset;
1908 back_min = prog->float_min_offset;
1912 if (must == &PL_sv_undef)
1913 /* could not downgrade utf8 check substring, so must fail */
1919 last = HOP3c(strend, /* Cannot start after this */
1920 -(I32)(CHR_SVLEN(must)
1921 - (SvTAIL(must) != 0) + back_min), strbeg);
1924 last1 = HOPc(s, -1);
1926 last1 = s - 1; /* bogus */
1928 /* XXXX check_substr already used to find "s", can optimize if
1929 check_substr==must. */
1931 dontbother = end_shift;
1932 strend = HOPc(strend, -dontbother);
1933 while ( (s <= last) &&
1934 ((flags & REXEC_SCREAM)
1935 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1936 end_shift, &scream_pos, 0))
1937 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1938 (unsigned char*)strend, must,
1939 multiline ? FBMrf_MULTILINE : 0))) ) {
1940 /* we may be pointing at the wrong string */
1941 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1942 s = strbeg + (s - SvPVX_const(sv));
1943 DEBUG_EXECUTE_r( did_match = 1 );
1944 if (HOPc(s, -back_max) > last1) {
1945 last1 = HOPc(s, -back_min);
1946 s = HOPc(s, -back_max);
1949 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1951 last1 = HOPc(s, -back_min);
1955 while (s <= last1) {
1956 if (regtry(®info, &s))
1962 while (s <= last1) {
1963 if (regtry(®info, &s))
1969 DEBUG_EXECUTE_r(if (!did_match) {
1970 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1971 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1972 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1973 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1974 ? "anchored" : "floating"),
1975 quoted, RE_SV_TAIL(must));
1979 else if ( (c = progi->regstclass) ) {
1981 const OPCODE op = OP(progi->regstclass);
1982 /* don't bother with what can't match */
1983 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1984 strend = HOPc(strend, -(minlen - 1));
1987 SV * const prop = sv_newmortal();
1988 regprop(prog, prop, c);
1990 RE_PV_QUOTED_DECL(quoted,do_utf8,PERL_DEBUG_PAD_ZERO(1),
1992 PerlIO_printf(Perl_debug_log,
1993 "Matching stclass %.*s against %s (%d chars)\n",
1994 (int)SvCUR(prop), SvPVX_const(prop),
1995 quoted, (int)(strend - s));
1998 if (find_byclass(prog, c, s, strend, ®info))
2000 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2004 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2009 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2010 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2011 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
2013 if (flags & REXEC_SCREAM) {
2014 last = screaminstr(sv, float_real, s - strbeg,
2015 end_shift, &scream_pos, 1); /* last one */
2017 last = scream_olds; /* Only one occurrence. */
2018 /* we may be pointing at the wrong string */
2019 else if (RX_MATCH_COPIED(prog))
2020 s = strbeg + (s - SvPVX_const(sv));
2024 const char * const little = SvPV_const(float_real, len);
2026 if (SvTAIL(float_real)) {
2027 if (memEQ(strend - len + 1, little, len - 1))
2028 last = strend - len + 1;
2029 else if (!multiline)
2030 last = memEQ(strend - len, little, len)
2031 ? strend - len : NULL;
2037 last = rninstr(s, strend, little, little + len);
2039 last = strend; /* matching "$" */
2044 PerlIO_printf(Perl_debug_log,
2045 "%sCan't trim the tail, match fails (should not happen)%s\n",
2046 PL_colors[4], PL_colors[5]));
2047 goto phooey; /* Should not happen! */
2049 dontbother = strend - last + prog->float_min_offset;
2051 if (minlen && (dontbother < minlen))
2052 dontbother = minlen - 1;
2053 strend -= dontbother; /* this one's always in bytes! */
2054 /* We don't know much -- general case. */
2057 if (regtry(®info, &s))
2066 if (regtry(®info, &s))
2068 } while (s++ < strend);
2076 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2078 if (PL_reg_eval_set)
2079 restore_pos(aTHX_ prog);
2080 if (prog->paren_names)
2081 (void)hv_iterinit(prog->paren_names);
2083 /* make sure $`, $&, $', and $digit will work later */
2084 if ( !(flags & REXEC_NOT_FIRST) ) {
2085 RX_MATCH_COPY_FREE(prog);
2086 if (flags & REXEC_COPY_STR) {
2087 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2088 #ifdef PERL_OLD_COPY_ON_WRITE
2090 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2092 PerlIO_printf(Perl_debug_log,
2093 "Copy on write: regexp capture, type %d\n",
2096 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2097 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2098 assert (SvPOKp(prog->saved_copy));
2102 RX_MATCH_COPIED_on(prog);
2103 s = savepvn(strbeg, i);
2109 prog->subbeg = strbeg;
2110 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2117 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2118 PL_colors[4], PL_colors[5]));
2119 if (PL_reg_eval_set)
2120 restore_pos(aTHX_ prog);
2122 /* we failed :-( roll it back */
2123 swap_match_buff(prog);
2130 - regtry - try match at specific point
2132 STATIC I32 /* 0 failure, 1 success */
2133 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2137 regexp *prog = reginfo->prog;
2138 RXi_GET_DECL(prog,progi);
2139 GET_RE_DEBUG_FLAGS_DECL;
2140 reginfo->cutpoint=NULL;
2142 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2145 PL_reg_eval_set = RS_init;
2146 DEBUG_EXECUTE_r(DEBUG_s(
2147 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2148 (IV)(PL_stack_sp - PL_stack_base));
2151 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2152 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2154 /* Apparently this is not needed, judging by wantarray. */
2155 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2156 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2159 /* Make $_ available to executed code. */
2160 if (reginfo->sv != DEFSV) {
2162 DEFSV = reginfo->sv;
2165 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2166 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2167 /* prepare for quick setting of pos */
2168 #ifdef PERL_OLD_COPY_ON_WRITE
2169 if (SvIsCOW(reginfo->sv))
2170 sv_force_normal_flags(reginfo->sv, 0);
2172 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2173 &PL_vtbl_mglob, NULL, 0);
2177 PL_reg_oldpos = mg->mg_len;
2178 SAVEDESTRUCTOR_X(restore_pos, prog);
2180 if (!PL_reg_curpm) {
2181 Newxz(PL_reg_curpm, 1, PMOP);
2184 SV* const repointer = newSViv(0);
2185 /* so we know which PL_regex_padav element is PL_reg_curpm */
2186 SvFLAGS(repointer) |= SVf_BREAK;
2187 av_push(PL_regex_padav,repointer);
2188 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2189 PL_regex_pad = AvARRAY(PL_regex_padav);
2193 PM_SETRE(PL_reg_curpm, prog);
2194 PL_reg_oldcurpm = PL_curpm;
2195 PL_curpm = PL_reg_curpm;
2196 if (RX_MATCH_COPIED(prog)) {
2197 /* Here is a serious problem: we cannot rewrite subbeg,
2198 since it may be needed if this match fails. Thus
2199 $` inside (?{}) could fail... */
2200 PL_reg_oldsaved = prog->subbeg;
2201 PL_reg_oldsavedlen = prog->sublen;
2202 #ifdef PERL_OLD_COPY_ON_WRITE
2203 PL_nrs = prog->saved_copy;
2205 RX_MATCH_COPIED_off(prog);
2208 PL_reg_oldsaved = NULL;
2209 prog->subbeg = PL_bostr;
2210 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2212 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2213 prog->offs[0].start = *startpos - PL_bostr;
2214 PL_reginput = *startpos;
2215 PL_reglastparen = &prog->lastparen;
2216 PL_reglastcloseparen = &prog->lastcloseparen;
2217 prog->lastparen = 0;
2218 prog->lastcloseparen = 0;
2220 PL_regoffs = prog->offs;
2221 if (PL_reg_start_tmpl <= prog->nparens) {
2222 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2223 if(PL_reg_start_tmp)
2224 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2226 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2229 /* XXXX What this code is doing here?!!! There should be no need
2230 to do this again and again, PL_reglastparen should take care of
2233 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2234 * Actually, the code in regcppop() (which Ilya may be meaning by
2235 * PL_reglastparen), is not needed at all by the test suite
2236 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2237 * enough, for building DynaLoader, or otherwise this
2238 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2239 * will happen. Meanwhile, this code *is* needed for the
2240 * above-mentioned test suite tests to succeed. The common theme
2241 * on those tests seems to be returning null fields from matches.
2244 if (prog->nparens) {
2245 regexp_paren_pair *pp = PL_regoffs;
2247 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2255 if (regmatch(reginfo, progi->program + 1)) {
2256 PL_regoffs[0].end = PL_reginput - PL_bostr;
2259 if (reginfo->cutpoint)
2260 *startpos= reginfo->cutpoint;
2261 REGCP_UNWIND(lastcp);
2266 #define sayYES goto yes
2267 #define sayNO goto no
2268 #define sayNO_SILENT goto no_silent
2270 /* we dont use STMT_START/END here because it leads to
2271 "unreachable code" warnings, which are bogus, but distracting. */
2272 #define CACHEsayNO \
2273 if (ST.cache_mask) \
2274 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2277 /* this is used to determine how far from the left messages like
2278 'failed...' are printed. It should be set such that messages
2279 are inline with the regop output that created them.
2281 #define REPORT_CODE_OFF 32
2284 /* Make sure there is a test for this +1 options in re_tests */
2285 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2287 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2288 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2290 #define SLAB_FIRST(s) (&(s)->states[0])
2291 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2293 /* grab a new slab and return the first slot in it */
2295 STATIC regmatch_state *
2298 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2301 regmatch_slab *s = PL_regmatch_slab->next;
2303 Newx(s, 1, regmatch_slab);
2304 s->prev = PL_regmatch_slab;
2306 PL_regmatch_slab->next = s;
2308 PL_regmatch_slab = s;
2309 return SLAB_FIRST(s);
2313 /* push a new state then goto it */
2315 #define PUSH_STATE_GOTO(state, node) \
2317 st->resume_state = state; \
2320 /* push a new state with success backtracking, then goto it */
2322 #define PUSH_YES_STATE_GOTO(state, node) \
2324 st->resume_state = state; \
2325 goto push_yes_state;
2331 regmatch() - main matching routine
2333 This is basically one big switch statement in a loop. We execute an op,
2334 set 'next' to point the next op, and continue. If we come to a point which
2335 we may need to backtrack to on failure such as (A|B|C), we push a
2336 backtrack state onto the backtrack stack. On failure, we pop the top
2337 state, and re-enter the loop at the state indicated. If there are no more
2338 states to pop, we return failure.
2340 Sometimes we also need to backtrack on success; for example /A+/, where
2341 after successfully matching one A, we need to go back and try to
2342 match another one; similarly for lookahead assertions: if the assertion
2343 completes successfully, we backtrack to the state just before the assertion
2344 and then carry on. In these cases, the pushed state is marked as
2345 'backtrack on success too'. This marking is in fact done by a chain of
2346 pointers, each pointing to the previous 'yes' state. On success, we pop to
2347 the nearest yes state, discarding any intermediate failure-only states.
2348 Sometimes a yes state is pushed just to force some cleanup code to be
2349 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2350 it to free the inner regex.
2352 Note that failure backtracking rewinds the cursor position, while
2353 success backtracking leaves it alone.
2355 A pattern is complete when the END op is executed, while a subpattern
2356 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2357 ops trigger the "pop to last yes state if any, otherwise return true"
2360 A common convention in this function is to use A and B to refer to the two
2361 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2362 the subpattern to be matched possibly multiple times, while B is the entire
2363 rest of the pattern. Variable and state names reflect this convention.
2365 The states in the main switch are the union of ops and failure/success of
2366 substates associated with with that op. For example, IFMATCH is the op
2367 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2368 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2369 successfully matched A and IFMATCH_A_fail is a state saying that we have
2370 just failed to match A. Resume states always come in pairs. The backtrack
2371 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2372 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2373 on success or failure.
2375 The struct that holds a backtracking state is actually a big union, with
2376 one variant for each major type of op. The variable st points to the
2377 top-most backtrack struct. To make the code clearer, within each
2378 block of code we #define ST to alias the relevant union.
2380 Here's a concrete example of a (vastly oversimplified) IFMATCH
2386 #define ST st->u.ifmatch
2388 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2389 ST.foo = ...; // some state we wish to save
2391 // push a yes backtrack state with a resume value of
2392 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2394 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2397 case IFMATCH_A: // we have successfully executed A; now continue with B
2399 bar = ST.foo; // do something with the preserved value
2402 case IFMATCH_A_fail: // A failed, so the assertion failed
2403 ...; // do some housekeeping, then ...
2404 sayNO; // propagate the failure
2411 For any old-timers reading this who are familiar with the old recursive
2412 approach, the code above is equivalent to:
2414 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2423 ...; // do some housekeeping, then ...
2424 sayNO; // propagate the failure
2427 The topmost backtrack state, pointed to by st, is usually free. If you
2428 want to claim it, populate any ST.foo fields in it with values you wish to
2429 save, then do one of
2431 PUSH_STATE_GOTO(resume_state, node);
2432 PUSH_YES_STATE_GOTO(resume_state, node);
2434 which sets that backtrack state's resume value to 'resume_state', pushes a
2435 new free entry to the top of the backtrack stack, then goes to 'node'.
2436 On backtracking, the free slot is popped, and the saved state becomes the
2437 new free state. An ST.foo field in this new top state can be temporarily
2438 accessed to retrieve values, but once the main loop is re-entered, it
2439 becomes available for reuse.
2441 Note that the depth of the backtrack stack constantly increases during the
2442 left-to-right execution of the pattern, rather than going up and down with
2443 the pattern nesting. For example the stack is at its maximum at Z at the
2444 end of the pattern, rather than at X in the following:
2446 /(((X)+)+)+....(Y)+....Z/
2448 The only exceptions to this are lookahead/behind assertions and the cut,
2449 (?>A), which pop all the backtrack states associated with A before
2452 Bascktrack state structs are allocated in slabs of about 4K in size.
2453 PL_regmatch_state and st always point to the currently active state,
2454 and PL_regmatch_slab points to the slab currently containing
2455 PL_regmatch_state. The first time regmatch() is called, the first slab is
2456 allocated, and is never freed until interpreter destruction. When the slab
2457 is full, a new one is allocated and chained to the end. At exit from
2458 regmatch(), slabs allocated since entry are freed.
2463 #define DEBUG_STATE_pp(pp) \
2465 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2466 PerlIO_printf(Perl_debug_log, \
2467 " %*s"pp" %s%s%s%s%s\n", \
2469 PL_reg_name[st->resume_state], \
2470 ((st==yes_state||st==mark_state) ? "[" : ""), \
2471 ((st==yes_state) ? "Y" : ""), \
2472 ((st==mark_state) ? "M" : ""), \
2473 ((st==yes_state||st==mark_state) ? "]" : "") \
2478 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2483 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2484 const char *start, const char *end, const char *blurb)
2486 const bool utf8_pat= prog->extflags & RXf_UTF8 ? 1 : 0;
2490 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2491 prog->precomp, prog->prelen, 60);
2493 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2494 start, end - start, 60);
2496 PerlIO_printf(Perl_debug_log,
2497 "%s%s REx%s %s against %s\n",
2498 PL_colors[4], blurb, PL_colors[5], s0, s1);
2500 if (do_utf8||utf8_pat)
2501 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2502 utf8_pat ? "pattern" : "",
2503 utf8_pat && do_utf8 ? " and " : "",
2504 do_utf8 ? "string" : ""
2510 S_dump_exec_pos(pTHX_ const char *locinput,
2511 const regnode *scan,
2512 const char *loc_regeol,
2513 const char *loc_bostr,
2514 const char *loc_reg_starttry,
2517 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2518 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2519 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2520 /* The part of the string before starttry has one color
2521 (pref0_len chars), between starttry and current
2522 position another one (pref_len - pref0_len chars),
2523 after the current position the third one.
2524 We assume that pref0_len <= pref_len, otherwise we
2525 decrease pref0_len. */
2526 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2527 ? (5 + taill) - l : locinput - loc_bostr;
2530 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2532 pref0_len = pref_len - (locinput - loc_reg_starttry);
2533 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2534 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2535 ? (5 + taill) - pref_len : loc_regeol - locinput);
2536 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2540 if (pref0_len > pref_len)
2541 pref0_len = pref_len;
2543 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2545 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2546 (locinput - pref_len),pref0_len, 60, 4, 5);
2548 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2549 (locinput - pref_len + pref0_len),
2550 pref_len - pref0_len, 60, 2, 3);
2552 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2553 locinput, loc_regeol - locinput, 10, 0, 1);
2555 const STRLEN tlen=len0+len1+len2;
2556 PerlIO_printf(Perl_debug_log,
2557 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2558 (IV)(locinput - loc_bostr),
2561 (docolor ? "" : "> <"),
2563 (int)(tlen > 19 ? 0 : 19 - tlen),
2570 /* reg_check_named_buff_matched()
2571 * Checks to see if a named buffer has matched. The data array of
2572 * buffer numbers corresponding to the buffer is expected to reside
2573 * in the regexp->data->data array in the slot stored in the ARG() of
2574 * node involved. Note that this routine doesn't actually care about the
2575 * name, that information is not preserved from compilation to execution.
2576 * Returns the index of the leftmost defined buffer with the given name
2577 * or 0 if non of the buffers matched.
2580 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2582 RXi_GET_DECL(rex,rexi);
2583 SV *sv_dat=(SV*)rexi->data->data[ ARG( scan ) ];
2584 I32 *nums=(I32*)SvPVX(sv_dat);
2585 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2586 if ((I32)*PL_reglastparen >= nums[n] &&
2587 PL_regoffs[nums[n]].end != -1)
2596 /* free all slabs above current one - called during LEAVE_SCOPE */
2599 S_clear_backtrack_stack(pTHX_ void *p)
2601 regmatch_slab *s = PL_regmatch_slab->next;
2606 PL_regmatch_slab->next = NULL;
2608 regmatch_slab * const osl = s;
2615 #define SETREX(Re1,Re2) \
2616 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2619 STATIC I32 /* 0 failure, 1 success */
2620 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2622 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2626 register const bool do_utf8 = PL_reg_match_utf8;
2627 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2629 regexp *rex = reginfo->prog;
2630 RXi_GET_DECL(rex,rexi);
2634 /* the current state. This is a cached copy of PL_regmatch_state */
2635 register regmatch_state *st;
2637 /* cache heavy used fields of st in registers */
2638 register regnode *scan;
2639 register regnode *next;
2640 register U32 n = 0; /* general value; init to avoid compiler warning */
2641 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2642 register char *locinput = PL_reginput;
2643 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2645 bool result = 0; /* return value of S_regmatch */
2646 int depth = 0; /* depth of backtrack stack */
2647 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2648 const U32 max_nochange_depth =
2649 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2650 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2652 regmatch_state *yes_state = NULL; /* state to pop to on success of
2654 /* mark_state piggy backs on the yes_state logic so that when we unwind
2655 the stack on success we can update the mark_state as we go */
2656 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2658 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2659 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2661 bool no_final = 0; /* prevent failure from backtracking? */
2662 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2663 char *startpoint = PL_reginput;
2664 SV *popmark = NULL; /* are we looking for a mark? */
2665 SV *sv_commit = NULL; /* last mark name seen in failure */
2666 SV *sv_yes_mark = NULL; /* last mark name we have seen
2667 during a successfull match */
2668 U32 lastopen = 0; /* last open we saw */
2669 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2671 SV* const oreplsv = GvSV(PL_replgv);
2674 /* these three flags are set by various ops to signal information to
2675 * the very next op. They have a useful lifetime of exactly one loop
2676 * iteration, and are not preserved or restored by state pushes/pops
2678 bool sw = 0; /* the condition value in (?(cond)a|b) */
2679 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2680 int logical = 0; /* the following EVAL is:
2684 or the following IFMATCH/UNLESSM is:
2685 false: plain (?=foo)
2686 true: used as a condition: (?(?=foo))
2690 GET_RE_DEBUG_FLAGS_DECL;
2693 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
2694 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2696 /* on first ever call to regmatch, allocate first slab */
2697 if (!PL_regmatch_slab) {
2698 Newx(PL_regmatch_slab, 1, regmatch_slab);
2699 PL_regmatch_slab->prev = NULL;
2700 PL_regmatch_slab->next = NULL;
2701 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2704 oldsave = PL_savestack_ix;
2705 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
2706 SAVEVPTR(PL_regmatch_slab);
2707 SAVEVPTR(PL_regmatch_state);
2709 /* grab next free state slot */
2710 st = ++PL_regmatch_state;
2711 if (st > SLAB_LAST(PL_regmatch_slab))
2712 st = PL_regmatch_state = S_push_slab(aTHX);
2714 /* Note that nextchr is a byte even in UTF */
2715 nextchr = UCHARAT(locinput);
2717 while (scan != NULL) {
2720 SV * const prop = sv_newmortal();
2721 regnode *rnext=regnext(scan);
2722 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2723 regprop(rex, prop, scan);
2725 PerlIO_printf(Perl_debug_log,
2726 "%3"IVdf":%*s%s(%"IVdf")\n",
2727 (IV)(scan - rexi->program), depth*2, "",
2729 (PL_regkind[OP(scan)] == END || !rnext) ?
2730 0 : (IV)(rnext - rexi->program));
2733 next = scan + NEXT_OFF(scan);
2736 state_num = OP(scan);
2739 switch (state_num) {
2741 if (locinput == PL_bostr)
2743 /* reginfo->till = reginfo->bol; */
2748 if (locinput == PL_bostr ||
2749 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2755 if (locinput == PL_bostr)
2759 if (locinput == reginfo->ganch)
2764 /* update the startpoint */
2765 st->u.keeper.val = PL_regoffs[0].start;
2766 PL_reginput = locinput;
2767 PL_regoffs[0].start = locinput - PL_bostr;
2768 PUSH_STATE_GOTO(KEEPS_next, next);
2770 case KEEPS_next_fail:
2771 /* rollback the start point change */
2772 PL_regoffs[0].start = st->u.keeper.val;
2778 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2783 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2785 if (PL_regeol - locinput > 1)
2789 if (PL_regeol != locinput)
2793 if (!nextchr && locinput >= PL_regeol)
2796 locinput += PL_utf8skip[nextchr];
2797 if (locinput > PL_regeol)
2799 nextchr = UCHARAT(locinput);
2802 nextchr = UCHARAT(++locinput);
2805 if (!nextchr && locinput >= PL_regeol)
2807 nextchr = UCHARAT(++locinput);
2810 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2813 locinput += PL_utf8skip[nextchr];
2814 if (locinput > PL_regeol)
2816 nextchr = UCHARAT(locinput);
2819 nextchr = UCHARAT(++locinput);
2823 #define ST st->u.trie
2825 /* In this case the charclass data is available inline so
2826 we can fail fast without a lot of extra overhead.
2828 if (scan->flags == EXACT || !do_utf8) {
2829 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2831 PerlIO_printf(Perl_debug_log,
2832 "%*s %sfailed to match trie start class...%s\n",
2833 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2842 /* what type of TRIE am I? (utf8 makes this contextual) */
2843 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2844 trie_type = do_utf8 ?
2845 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2848 /* what trie are we using right now */
2849 reg_trie_data * const trie
2850 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2851 HV * widecharmap = (HV *)rexi->data->data[ ARG( scan ) + 1 ];
2852 U32 state = trie->startstate;
2854 if (trie->bitmap && trie_type != trie_utf8_fold &&
2855 !TRIE_BITMAP_TEST(trie,*locinput)
2857 if (trie->states[ state ].wordnum) {
2859 PerlIO_printf(Perl_debug_log,
2860 "%*s %smatched empty string...%s\n",
2861 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2866 PerlIO_printf(Perl_debug_log,
2867 "%*s %sfailed to match trie start class...%s\n",
2868 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2875 U8 *uc = ( U8* )locinput;
2879 U8 *uscan = (U8*)NULL;
2881 SV *sv_accept_buff = NULL;
2882 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2884 ST.accepted = 0; /* how many accepting states we have seen */
2886 ST.jump = trie->jump;
2889 traverse the TRIE keeping track of all accepting states
2890 we transition through until we get to a failing node.
2893 while ( state && uc <= (U8*)PL_regeol ) {
2894 U32 base = trie->states[ state ].trans.base;
2897 /* We use charid to hold the wordnum as we don't use it
2898 for charid until after we have done the wordnum logic.
2899 We define an alias just so that the wordnum logic reads
2902 #define got_wordnum charid
2903 got_wordnum = trie->states[ state ].wordnum;
2905 if ( got_wordnum ) {
2906 if ( ! ST.accepted ) {
2909 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2910 sv_accept_buff=newSV(bufflen *
2911 sizeof(reg_trie_accepted) - 1);
2912 SvCUR_set(sv_accept_buff, 0);
2913 SvPOK_on(sv_accept_buff);
2914 sv_2mortal(sv_accept_buff);
2917 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2920 if (ST.accepted >= bufflen) {
2922 ST.accept_buff =(reg_trie_accepted*)
2923 SvGROW(sv_accept_buff,
2924 bufflen * sizeof(reg_trie_accepted));
2926 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2927 + sizeof(reg_trie_accepted));
2930 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2931 ST.accept_buff[ST.accepted].endpos = uc;
2933 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2937 DEBUG_TRIE_EXECUTE_r({
2938 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2939 PerlIO_printf( Perl_debug_log,
2940 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2941 2+depth * 2, "", PL_colors[4],
2942 (UV)state, (UV)ST.accepted );
2946 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
2947 uscan, len, uvc, charid, foldlen,
2951 (base + charid > trie->uniquecharcount )
2952 && (base + charid - 1 - trie->uniquecharcount
2954 && trie->trans[base + charid - 1 -
2955 trie->uniquecharcount].check == state)
2957 state = trie->trans[base + charid - 1 -
2958 trie->uniquecharcount ].next;
2969 DEBUG_TRIE_EXECUTE_r(
2970 PerlIO_printf( Perl_debug_log,
2971 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2972 charid, uvc, (UV)state, PL_colors[5] );
2979 PerlIO_printf( Perl_debug_log,
2980 "%*s %sgot %"IVdf" possible matches%s\n",
2981 REPORT_CODE_OFF + depth * 2, "",
2982 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2985 goto trie_first_try; /* jump into the fail handler */
2987 case TRIE_next_fail: /* we failed - try next alterative */
2989 REGCP_UNWIND(ST.cp);
2990 for (n = *PL_reglastparen; n > ST.lastparen; n--)
2991 PL_regoffs[n].end = -1;
2992 *PL_reglastparen = n;
3001 ST.lastparen = *PL_reglastparen;
3004 if ( ST.accepted == 1 ) {
3005 /* only one choice left - just continue */
3007 AV *const trie_words
3008 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
3009 SV ** const tmp = av_fetch( trie_words,
3010 ST.accept_buff[ 0 ].wordnum-1, 0 );
3011 SV *sv= tmp ? sv_newmortal() : NULL;
3013 PerlIO_printf( Perl_debug_log,
3014 "%*s %sonly one match left: #%d <%s>%s\n",
3015 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3016 ST.accept_buff[ 0 ].wordnum,
3017 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3018 PL_colors[0], PL_colors[1],
3019 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3021 : "not compiled under -Dr",
3024 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
3025 /* in this case we free tmps/leave before we call regmatch
3026 as we wont be using accept_buff again. */
3028 locinput = PL_reginput;
3029 nextchr = UCHARAT(locinput);
3030 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
3033 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
3034 if (!has_cutgroup) {
3039 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3042 continue; /* execute rest of RE */
3045 if ( !ST.accepted-- ) {
3047 PerlIO_printf( Perl_debug_log,
3048 "%*s %sTRIE failed...%s\n",
3049 REPORT_CODE_OFF+depth*2, "",
3060 There are at least two accepting states left. Presumably
3061 the number of accepting states is going to be low,
3062 typically two. So we simply scan through to find the one
3063 with lowest wordnum. Once we find it, we swap the last
3064 state into its place and decrement the size. We then try to
3065 match the rest of the pattern at the point where the word
3066 ends. If we succeed, control just continues along the
3067 regex; if we fail we return here to try the next accepting
3074 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3075 DEBUG_TRIE_EXECUTE_r(
3076 PerlIO_printf( Perl_debug_log,
3077 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3078 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3079 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3080 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3083 if (ST.accept_buff[cur].wordnum <
3084 ST.accept_buff[best].wordnum)
3089 AV *const trie_words
3090 = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
3091 SV ** const tmp = av_fetch( trie_words,
3092 ST.accept_buff[ best ].wordnum - 1, 0 );
3093 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3095 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3096 SV *sv= tmp ? sv_newmortal() : NULL;
3098 PerlIO_printf( Perl_debug_log,
3099 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3100 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3101 ST.accept_buff[best].wordnum,
3102 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3103 PL_colors[0], PL_colors[1],
3104 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3105 ) : "not compiled under -Dr",
3106 REG_NODE_NUM(nextop),
3110 if ( best<ST.accepted ) {
3111 reg_trie_accepted tmp = ST.accept_buff[ best ];
3112 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3113 ST.accept_buff[ ST.accepted ] = tmp;
3116 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3117 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3121 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3125 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3128 PUSH_STATE_GOTO(TRIE_next, scan);
3141 char *s = STRING(scan);
3143 if (do_utf8 != UTF) {
3144 /* The target and the pattern have differing utf8ness. */
3146 const char * const e = s + ln;
3149 /* The target is utf8, the pattern is not utf8. */
3154 if (NATIVE_TO_UNI(*(U8*)s) !=
3155 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3163 /* The target is not utf8, the pattern is utf8. */
3168 if (NATIVE_TO_UNI(*((U8*)l)) !=
3169 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3177 nextchr = UCHARAT(locinput);
3180 /* The target and the pattern have the same utf8ness. */
3181 /* Inline the first character, for speed. */
3182 if (UCHARAT(s) != nextchr)
3184 if (PL_regeol - locinput < ln)
3186 if (ln > 1 && memNE(s, locinput, ln))
3189 nextchr = UCHARAT(locinput);
3193 PL_reg_flags |= RF_tainted;
3196 char * const s = STRING(scan);
3199 if (do_utf8 || UTF) {
3200 /* Either target or the pattern are utf8. */
3201 const char * const l = locinput;
3202 char *e = PL_regeol;
3204 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3205 l, &e, 0, do_utf8)) {
3206 /* One more case for the sharp s:
3207 * pack("U0U*", 0xDF) =~ /ss/i,
3208 * the 0xC3 0x9F are the UTF-8
3209 * byte sequence for the U+00DF. */
3211 toLOWER(s[0]) == 's' &&
3213 toLOWER(s[1]) == 's' &&
3220 nextchr = UCHARAT(locinput);
3224 /* Neither the target and the pattern are utf8. */
3226 /* Inline the first character, for speed. */
3227 if (UCHARAT(s) != nextchr &&
3228 UCHARAT(s) != ((OP(scan) == EXACTF)
3229 ? PL_fold : PL_fold_locale)[nextchr])
3231 if (PL_regeol - locinput < ln)
3233 if (ln > 1 && (OP(scan) == EXACTF
3234 ? ibcmp(s, locinput, ln)
3235 : ibcmp_locale(s, locinput, ln)))
3238 nextchr = UCHARAT(locinput);
3243 STRLEN inclasslen = PL_regeol - locinput;
3245 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3247 if (locinput >= PL_regeol)
3249 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3250 nextchr = UCHARAT(locinput);
3255 nextchr = UCHARAT(locinput);
3256 if (!REGINCLASS(rex, scan, (U8*)locinput))
3258 if (!nextchr && locinput >= PL_regeol)
3260 nextchr = UCHARAT(++locinput);
3264 /* If we might have the case of the German sharp s
3265 * in a casefolding Unicode character class. */
3267 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3268 locinput += SHARP_S_SKIP;
3269 nextchr = UCHARAT(locinput);
3275 PL_reg_flags |= RF_tainted;
3281 LOAD_UTF8_CHARCLASS_ALNUM();
3282 if (!(OP(scan) == ALNUM
3283 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3284 : isALNUM_LC_utf8((U8*)locinput)))
3288 locinput += PL_utf8skip[nextchr];
3289 nextchr = UCHARAT(locinput);
3292 if (!(OP(scan) == ALNUM
3293 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3295 nextchr = UCHARAT(++locinput);
3298 PL_reg_flags |= RF_tainted;
3301 if (!nextchr && locinput >= PL_regeol)
3304 LOAD_UTF8_CHARCLASS_ALNUM();
3305 if (OP(scan) == NALNUM
3306 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3307 : isALNUM_LC_utf8((U8*)locinput))
3311 locinput += PL_utf8skip[nextchr];
3312 nextchr = UCHARAT(locinput);
3315 if (OP(scan) == NALNUM
3316 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3318 nextchr = UCHARAT(++locinput);
3322 PL_reg_flags |= RF_tainted;
3326 /* was last char in word? */
3328 if (locinput == PL_bostr)
3331 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3333 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3335 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3336 ln = isALNUM_uni(ln);
3337 LOAD_UTF8_CHARCLASS_ALNUM();
3338 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3341 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3342 n = isALNUM_LC_utf8((U8*)locinput);
3346 ln = (locinput != PL_bostr) ?
3347 UCHARAT(locinput - 1) : '\n';
3348 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3350 n = isALNUM(nextchr);
3353 ln = isALNUM_LC(ln);
3354 n = isALNUM_LC(nextchr);
3357 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3358 OP(scan) == BOUNDL))
3362 PL_reg_flags |= RF_tainted;
3368 if (UTF8_IS_CONTINUED(nextchr)) {
3369 LOAD_UTF8_CHARCLASS_SPACE();
3370 if (!(OP(scan) == SPACE
3371 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3372 : isSPACE_LC_utf8((U8*)locinput)))
3376 locinput += PL_utf8skip[nextchr];
3377 nextchr = UCHARAT(locinput);
3380 if (!(OP(scan) == SPACE
3381 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3383 nextchr = UCHARAT(++locinput);
3386 if (!(OP(scan) == SPACE
3387 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3389 nextchr = UCHARAT(++locinput);
3393 PL_reg_flags |= RF_tainted;
3396 if (!nextchr && locinput >= PL_regeol)
3399 LOAD_UTF8_CHARCLASS_SPACE();
3400 if (OP(scan) == NSPACE
3401 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3402 : isSPACE_LC_utf8((U8*)locinput))
3406 locinput += PL_utf8skip[nextchr];
3407 nextchr = UCHARAT(locinput);
3410 if (OP(scan) == NSPACE
3411 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3413 nextchr = UCHARAT(++locinput);
3416 PL_reg_flags |= RF_tainted;
3422 LOAD_UTF8_CHARCLASS_DIGIT();
3423 if (!(OP(scan) == DIGIT
3424 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3425 : isDIGIT_LC_utf8((U8*)locinput)))
3429 locinput += PL_utf8skip[nextchr];
3430 nextchr = UCHARAT(locinput);
3433 if (!(OP(scan) == DIGIT
3434 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3436 nextchr = UCHARAT(++locinput);
3439 PL_reg_flags |= RF_tainted;
3442 if (!nextchr && locinput >= PL_regeol)
3445 LOAD_UTF8_CHARCLASS_DIGIT();
3446 if (OP(scan) == NDIGIT
3447 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3448 : isDIGIT_LC_utf8((U8*)locinput))
3452 locinput += PL_utf8skip[nextchr];
3453 nextchr = UCHARAT(locinput);
3456 if (OP(scan) == NDIGIT
3457 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3459 nextchr = UCHARAT(++locinput);
3462 if (locinput >= PL_regeol)
3465 LOAD_UTF8_CHARCLASS_MARK();
3466 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3468 locinput += PL_utf8skip[nextchr];
3469 while (locinput < PL_regeol &&
3470 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3471 locinput += UTF8SKIP(locinput);
3472 if (locinput > PL_regeol)
3477 nextchr = UCHARAT(locinput);
3484 PL_reg_flags |= RF_tainted;
3489 n = reg_check_named_buff_matched(rex,scan);
3492 type = REF + ( type - NREF );
3499 PL_reg_flags |= RF_tainted;
3503 n = ARG(scan); /* which paren pair */
3506 ln = PL_regoffs[n].start;
3507 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3508 if (*PL_reglastparen < n || ln == -1)
3509 sayNO; /* Do not match unless seen CLOSEn. */
3510 if (ln == PL_regoffs[n].end)
3514 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3516 const char *e = PL_bostr + PL_regoffs[n].end;
3518 * Note that we can't do the "other character" lookup trick as
3519 * in the 8-bit case (no pun intended) because in Unicode we
3520 * have to map both upper and title case to lower case.
3524 STRLEN ulen1, ulen2;
3525 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3526 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3530 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3531 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3532 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3539 nextchr = UCHARAT(locinput);
3543 /* Inline the first character, for speed. */
3544 if (UCHARAT(s) != nextchr &&
3546 (UCHARAT(s) != (type == REFF
3547 ? PL_fold : PL_fold_locale)[nextchr])))
3549 ln = PL_regoffs[n].end - ln;
3550 if (locinput + ln > PL_regeol)
3552 if (ln > 1 && (type == REF
3553 ? memNE(s, locinput, ln)
3555 ? ibcmp(s, locinput, ln)
3556 : ibcmp_locale(s, locinput, ln))))
3559 nextchr = UCHARAT(locinput);
3569 #define ST st->u.eval
3573 regexp_internal *rei;
3574 regnode *startpoint;
3577 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
3578 if (cur_eval && cur_eval->locinput==locinput) {
3579 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3580 Perl_croak(aTHX_ "Infinite recursion in regex");
3581 if ( ++nochange_depth > max_nochange_depth )
3583 "Pattern subroutine nesting without pos change"
3584 " exceeded limit in regex");
3590 (void)ReREFCNT_inc(rex);
3591 if (OP(scan)==GOSUB) {
3592 startpoint = scan + ARG2L(scan);
3593 ST.close_paren = ARG(scan);
3595 startpoint = rei->program+1;
3598 goto eval_recurse_doit;
3600 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3601 if (cur_eval && cur_eval->locinput==locinput) {
3602 if ( ++nochange_depth > max_nochange_depth )
3603 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3608 /* execute the code in the {...} */
3610 SV ** const before = SP;
3611 OP_4tree * const oop = PL_op;
3612 COP * const ocurcop = PL_curcop;
3616 PL_op = (OP_4tree*)rexi->data->data[n];
3617 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3618 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3619 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3620 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
3623 SV *sv_mrk = get_sv("REGMARK", 1);
3624 sv_setsv(sv_mrk, sv_yes_mark);
3627 CALLRUNOPS(aTHX); /* Scalar context. */
3630 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3637 PAD_RESTORE_LOCAL(old_comppad);
3638 PL_curcop = ocurcop;
3641 sv_setsv(save_scalar(PL_replgv), ret);
3645 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3648 /* extract RE object from returned value; compiling if
3653 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3654 mg = mg_find(sv, PERL_MAGIC_qr);
3655 else if (SvSMAGICAL(ret)) {
3656 if (SvGMAGICAL(ret))
3657 sv_unmagic(ret, PERL_MAGIC_qr);
3659 mg = mg_find(ret, PERL_MAGIC_qr);
3663 re = reg_temp_copy((regexp *)mg->mg_obj); /*XXX:dmq*/
3667 const char * const t = SvPV_const(ret, len);
3669 const I32 osize = PL_regsize;
3672 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3673 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3675 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3677 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3682 RX_MATCH_COPIED_off(re);
3683 re->subbeg = rex->subbeg;
3684 re->sublen = rex->sublen;
3687 debug_start_match(re, do_utf8, locinput, PL_regeol,
3688 "Matching embedded");
3690 startpoint = rei->program + 1;
3691 ST.close_paren = 0; /* only used for GOSUB */
3692 /* borrowed from regtry */
3693 if (PL_reg_start_tmpl <= re->nparens) {
3694 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3695 if(PL_reg_start_tmp)
3696 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3698 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3701 eval_recurse_doit: /* Share code with GOSUB below this line */
3702 /* run the pattern returned from (??{...}) */
3703 ST.cp = regcppush(0); /* Save *all* the positions. */
3704 REGCP_SET(ST.lastcp);
3706 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
3708 *PL_reglastparen = 0;
3709 *PL_reglastcloseparen = 0;
3710 PL_reginput = locinput;
3713 /* XXXX This is too dramatic a measure... */
3716 ST.toggle_reg_flags = PL_reg_flags;
3717 if (re->extflags & RXf_UTF8)
3718 PL_reg_flags |= RF_utf8;
3720 PL_reg_flags &= ~RF_utf8;
3721 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3724 ST.prev_curlyx = cur_curlyx;
3729 ST.prev_eval = cur_eval;
3731 /* now continue from first node in postoned RE */
3732 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3735 /* logical is 1, /(?(?{...})X|Y)/ */
3736 sw = (bool)SvTRUE(ret);
3741 case EVAL_AB: /* cleanup after a successful (??{A})B */
3742 /* note: this is called twice; first after popping B, then A */
3743 PL_reg_flags ^= ST.toggle_reg_flags;
3745 SETREX(rex,ST.prev_rex);
3746 rexi = RXi_GET(rex);
3748 cur_eval = ST.prev_eval;
3749 cur_curlyx = ST.prev_curlyx;
3750 /* XXXX This is too dramatic a measure... */
3752 if ( nochange_depth )
3757 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3758 /* note: this is called twice; first after popping B, then A */
3759 PL_reg_flags ^= ST.toggle_reg_flags;
3761 SETREX(rex,ST.prev_rex);
3762 rexi = RXi_GET(rex);
3763 PL_reginput = locinput;
3764 REGCP_UNWIND(ST.lastcp);
3766 cur_eval = ST.prev_eval;
3767 cur_curlyx = ST.prev_curlyx;
3768 /* XXXX This is too dramatic a measure... */
3770 if ( nochange_depth )
3776 n = ARG(scan); /* which paren pair */
3777 PL_reg_start_tmp[n] = locinput;
3783 n = ARG(scan); /* which paren pair */
3784 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
3785 PL_regoffs[n].end = locinput - PL_bostr;
3786 /*if (n > PL_regsize)
3788 if (n > *PL_reglastparen)
3789 *PL_reglastparen = n;
3790 *PL_reglastcloseparen = n;
3791 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3799 cursor && OP(cursor)!=END;
3800 cursor=regnext(cursor))
3802 if ( OP(cursor)==CLOSE ){
3804 if ( n <= lastopen ) {
3806 = PL_reg_start_tmp[n] - PL_bostr;
3807 PL_regoffs[n].end = locinput - PL_bostr;
3808 /*if (n > PL_regsize)
3810 if (n > *PL_reglastparen)
3811 *PL_reglastparen = n;
3812 *PL_reglastcloseparen = n;
3813 if ( n == ARG(scan) || (cur_eval &&
3814 cur_eval->u.eval.close_paren == n))
3823 n = ARG(scan); /* which paren pair */
3824 sw = (bool)(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
3827 /* reg_check_named_buff_matched returns 0 for no match */
3828 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3832 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3838 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3840 next = NEXTOPER(NEXTOPER(scan));
3842 next = scan + ARG(scan);
3843 if (OP(next) == IFTHEN) /* Fake one. */
3844 next = NEXTOPER(NEXTOPER(next));
3848 logical = scan->flags;
3851 /*******************************************************************
3853 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3854 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3855 STAR/PLUS/CURLY/CURLYN are used instead.)
3857 A*B is compiled as <CURLYX><A><WHILEM><B>
3859 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3860 state, which contains the current count, initialised to -1. It also sets
3861 cur_curlyx to point to this state, with any previous value saved in the
3864 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3865 since the pattern may possibly match zero times (i.e. it's a while {} loop
3866 rather than a do {} while loop).
3868 Each entry to WHILEM represents a successful match of A. The count in the
3869 CURLYX block is incremented, another WHILEM state is pushed, and execution
3870 passes to A or B depending on greediness and the current count.
3872 For example, if matching against the string a1a2a3b (where the aN are
3873 substrings that match /A/), then the match progresses as follows: (the
3874 pushed states are interspersed with the bits of strings matched so far):
3877 <CURLYX cnt=0><WHILEM>
3878 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3879 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3880 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3881 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3883 (Contrast this with something like CURLYM, which maintains only a single
3887 a1 <CURLYM cnt=1> a2
3888 a1 a2 <CURLYM cnt=2> a3
3889 a1 a2 a3 <CURLYM cnt=3> b
3892 Each WHILEM state block marks a point to backtrack to upon partial failure
3893 of A or B, and also contains some minor state data related to that
3894 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3895 overall state, such as the count, and pointers to the A and B ops.
3897 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3898 must always point to the *current* CURLYX block, the rules are:
3900 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3901 and set cur_curlyx to point the new block.
3903 When popping the CURLYX block after a successful or unsuccessful match,
3904 restore the previous cur_curlyx.
3906 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3907 to the outer one saved in the CURLYX block.
3909 When popping the WHILEM block after a successful or unsuccessful B match,
3910 restore the previous cur_curlyx.
3912 Here's an example for the pattern (AI* BI)*BO
3913 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3916 curlyx backtrack stack
3917 ------ ---------------
3919 CO <CO prev=NULL> <WO>
3920 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3921 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3922 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3924 At this point the pattern succeeds, and we work back down the stack to
3925 clean up, restoring as we go:
3927 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3928 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3929 CO <CO prev=NULL> <WO>
3932 *******************************************************************/
3934 #define ST st->u.curlyx
3936 case CURLYX: /* start of /A*B/ (for complex A) */
3938 /* No need to save/restore up to this paren */
3939 I32 parenfloor = scan->flags;
3941 assert(next); /* keep Coverity happy */
3942 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3945 /* XXXX Probably it is better to teach regpush to support
3946 parenfloor > PL_regsize... */
3947 if (parenfloor > (I32)*PL_reglastparen)
3948 parenfloor = *PL_reglastparen; /* Pessimization... */
3950 ST.prev_curlyx= cur_curlyx;
3952 ST.cp = PL_savestack_ix;
3954 /* these fields contain the state of the current curly.
3955 * they are accessed by subsequent WHILEMs */
3956 ST.parenfloor = parenfloor;
3957 ST.min = ARG1(scan);
3958 ST.max = ARG2(scan);
3959 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;