5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
56 **** Alterations to Henry's code are...
58 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
59 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
61 **** You may distribute under the terms of either the GNU General Public
62 **** License or the Artistic License, as specified in the README file.
64 * Beware that some of this code is subtly aware of the way operator
65 * precedence is structured in regular expressions. Serious changes in
66 * regular-expression syntax might require a total rethink.
69 #define PERL_IN_REGEXEC_C
72 #ifdef PERL_IN_XSUB_RE
78 #define RF_tainted 1 /* tainted information used? */
79 #define RF_warned 2 /* warned about big count? */
81 #define RF_utf8 8 /* Pattern contains multibyte chars? */
83 #define UTF ((PL_reg_flags & RF_utf8) != 0)
85 #define RS_init 1 /* eval environment created */
86 #define RS_set 2 /* replsv value is set */
92 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
98 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
99 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
101 #define HOPc(pos,off) \
102 (char *)(PL_reg_match_utf8 \
103 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
105 #define HOPBACKc(pos, off) \
106 (char*)(PL_reg_match_utf8\
107 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
108 : (pos - off >= PL_bostr) \
112 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
113 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
115 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
116 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
117 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
118 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
119 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
120 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
122 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
124 /* for use after a quantifier and before an EXACT-like node -- japhy */
125 #define JUMPABLE(rn) ( \
126 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
127 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
128 OP(rn) == PLUS || OP(rn) == MINMOD || \
129 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
132 #define HAS_TEXT(rn) ( \
133 PL_regkind[OP(rn)] == EXACT || PL_regkind[OP(rn)] == REF \
137 Search for mandatory following text node; for lookahead, the text must
138 follow but for lookbehind (rn->flags != 0) we skip to the next step.
140 #define FIND_NEXT_IMPT(rn) STMT_START { \
141 while (JUMPABLE(rn)) { \
142 const OPCODE type = OP(rn); \
143 if (type == SUSPEND || PL_regkind[type] == CURLY) \
144 rn = NEXTOPER(NEXTOPER(rn)); \
145 else if (type == PLUS) \
147 else if (type == IFMATCH) \
148 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
149 else rn += NEXT_OFF(rn); \
154 static void restore_pos(pTHX_ void *arg);
157 S_regcppush(pTHX_ I32 parenfloor)
160 const int retval = PL_savestack_ix;
161 #define REGCP_PAREN_ELEMS 4
162 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
164 GET_RE_DEBUG_FLAGS_DECL;
166 if (paren_elems_to_push < 0)
167 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
169 #define REGCP_OTHER_ELEMS 8
170 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
172 for (p = PL_regsize; p > parenfloor; p--) {
173 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
174 SSPUSHINT(PL_regendp[p]);
175 SSPUSHINT(PL_regstartp[p]);
176 SSPUSHPTR(PL_reg_start_tmp[p]);
178 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
179 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
180 (UV)p, (IV)PL_regstartp[p],
181 (IV)(PL_reg_start_tmp[p] - PL_bostr),
185 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
186 SSPUSHPTR(PL_regstartp);
187 SSPUSHPTR(PL_regendp);
188 SSPUSHINT(PL_regsize);
189 SSPUSHINT(*PL_reglastparen);
190 SSPUSHINT(*PL_reglastcloseparen);
191 SSPUSHPTR(PL_reginput);
192 #define REGCP_FRAME_ELEMS 2
193 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
194 * are needed for the regexp context stack bookkeeping. */
195 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
196 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
201 /* These are needed since we do not localize EVAL nodes: */
202 #define REGCP_SET(cp) \
204 PerlIO_printf(Perl_debug_log, \
205 " Setting an EVAL scope, savestack=%"IVdf"\n", \
206 (IV)PL_savestack_ix)); \
209 #define REGCP_UNWIND(cp) \
211 if (cp != PL_savestack_ix) \
212 PerlIO_printf(Perl_debug_log, \
213 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
214 (IV)(cp), (IV)PL_savestack_ix)); \
218 S_regcppop(pTHX_ const regexp *rex)
224 GET_RE_DEBUG_FLAGS_DECL;
226 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
228 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
229 i = SSPOPINT; /* Parentheses elements to pop. */
230 input = (char *) SSPOPPTR;
231 *PL_reglastcloseparen = SSPOPINT;
232 *PL_reglastparen = SSPOPINT;
233 PL_regsize = SSPOPINT;
234 PL_regendp=(I32 *) SSPOPPTR;
235 PL_regstartp=(I32 *) SSPOPPTR;
238 /* Now restore the parentheses context. */
239 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
240 i > 0; i -= REGCP_PAREN_ELEMS) {
242 U32 paren = (U32)SSPOPINT;
243 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
244 PL_regstartp[paren] = SSPOPINT;
246 if (paren <= *PL_reglastparen)
247 PL_regendp[paren] = tmps;
249 PerlIO_printf(Perl_debug_log,
250 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
251 (UV)paren, (IV)PL_regstartp[paren],
252 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
253 (IV)PL_regendp[paren],
254 (paren > *PL_reglastparen ? "(no)" : ""));
258 if (*PL_reglastparen + 1 <= rex->nparens) {
259 PerlIO_printf(Perl_debug_log,
260 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
261 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
265 /* It would seem that the similar code in regtry()
266 * already takes care of this, and in fact it is in
267 * a better location to since this code can #if 0-ed out
268 * but the code in regtry() is needed or otherwise tests
269 * requiring null fields (pat.t#187 and split.t#{13,14}
270 * (as of patchlevel 7877) will fail. Then again,
271 * this code seems to be necessary or otherwise
272 * building DynaLoader will fail:
273 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
275 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
277 PL_regstartp[i] = -1;
284 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
287 * pregexec and friends
290 #ifndef PERL_IN_XSUB_RE
292 - pregexec - match a regexp against a string
295 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
296 char *strbeg, I32 minend, SV *screamer, U32 nosave)
297 /* strend: pointer to null at end of string */
298 /* strbeg: real beginning of string */
299 /* minend: end of match must be >=minend after stringarg. */
300 /* nosave: For optimizations. */
303 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
304 nosave ? 0 : REXEC_COPY_STR);
309 * Need to implement the following flags for reg_anch:
311 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
313 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
314 * INTUIT_AUTORITATIVE_ML
315 * INTUIT_ONCE_NOML - Intuit can match in one location only.
318 * Another flag for this function: SECOND_TIME (so that float substrs
319 * with giant delta may be not rechecked).
322 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
324 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
325 Otherwise, only SvCUR(sv) is used to get strbeg. */
327 /* XXXX We assume that strpos is strbeg unless sv. */
329 /* XXXX Some places assume that there is a fixed substring.
330 An update may be needed if optimizer marks as "INTUITable"
331 RExen without fixed substrings. Similarly, it is assumed that
332 lengths of all the strings are no more than minlen, thus they
333 cannot come from lookahead.
334 (Or minlen should take into account lookahead.)
335 NOTE: Some of this comment is not correct. minlen does now take account
336 of lookahead/behind. Further research is required. -- demerphq
340 /* A failure to find a constant substring means that there is no need to make
341 an expensive call to REx engine, thus we celebrate a failure. Similarly,
342 finding a substring too deep into the string means that less calls to
343 regtry() should be needed.
345 REx compiler's optimizer found 4 possible hints:
346 a) Anchored substring;
348 c) Whether we are anchored (beginning-of-line or \G);
349 d) First node (of those at offset 0) which may distingush positions;
350 We use a)b)d) and multiline-part of c), and try to find a position in the
351 string which does not contradict any of them.
354 /* Most of decisions we do here should have been done at compile time.
355 The nodes of the REx which we used for the search should have been
356 deleted from the finite automaton. */
359 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
360 char *strend, U32 flags, re_scream_pos_data *data)
363 register I32 start_shift = 0;
364 /* Should be nonnegative! */
365 register I32 end_shift = 0;
370 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
372 register char *other_last = NULL; /* other substr checked before this */
373 char *check_at = NULL; /* check substr found at this pos */
374 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
375 RXi_GET_DECL(prog,progi);
377 const char * const i_strpos = strpos;
380 GET_RE_DEBUG_FLAGS_DECL;
382 RX_MATCH_UTF8_set(prog,do_utf8);
384 if (prog->extflags & RXf_UTF8) {
385 PL_reg_flags |= RF_utf8;
388 debug_start_match(prog, do_utf8, strpos, strend,
389 sv ? "Guessing start of match in sv for"
390 : "Guessing start of match in string for");
393 /* CHR_DIST() would be more correct here but it makes things slow. */
394 if (prog->minlen > strend - strpos) {
395 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
396 "String too short... [re_intuit_start]\n"));
400 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
403 if (!prog->check_utf8 && prog->check_substr)
404 to_utf8_substr(prog);
405 check = prog->check_utf8;
407 if (!prog->check_substr && prog->check_utf8)
408 to_byte_substr(prog);
409 check = prog->check_substr;
411 if (check == &PL_sv_undef) {
412 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
413 "Non-utf8 string cannot match utf8 check string\n"));
416 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
417 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
418 || ( (prog->extflags & RXf_ANCH_BOL)
419 && !multiline ) ); /* Check after \n? */
422 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
423 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
424 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
426 && (strpos != strbeg)) {
427 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
430 if (prog->check_offset_min == prog->check_offset_max &&
431 !(prog->extflags & RXf_CANY_SEEN)) {
432 /* Substring at constant offset from beg-of-str... */
435 s = HOP3c(strpos, prog->check_offset_min, strend);
438 slen = SvCUR(check); /* >= 1 */
440 if ( strend - s > slen || strend - s < slen - 1
441 || (strend - s == slen && strend[-1] != '\n')) {
442 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
445 /* Now should match s[0..slen-2] */
447 if (slen && (*SvPVX_const(check) != *s
449 && memNE(SvPVX_const(check), s, slen)))) {
451 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
455 else if (*SvPVX_const(check) != *s
456 || ((slen = SvCUR(check)) > 1
457 && memNE(SvPVX_const(check), s, slen)))
460 goto success_at_start;
463 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
465 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
466 end_shift = prog->check_end_shift;
469 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
470 - (SvTAIL(check) != 0);
471 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
473 if (end_shift < eshift)
477 else { /* Can match at random position */
480 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
481 end_shift = prog->check_end_shift;
483 /* end shift should be non negative here */
486 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
488 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
489 (IV)end_shift, prog->precomp);
493 /* Find a possible match in the region s..strend by looking for
494 the "check" substring in the region corrected by start/end_shift. */
497 I32 srch_start_shift = start_shift;
498 I32 srch_end_shift = end_shift;
499 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
500 srch_end_shift -= ((strbeg - s) - srch_start_shift);
501 srch_start_shift = strbeg - s;
503 DEBUG_OPTIMISE_MORE_r({
504 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
505 (IV)prog->check_offset_min,
506 (IV)srch_start_shift,
508 (IV)prog->check_end_shift);
511 if (flags & REXEC_SCREAM) {
512 I32 p = -1; /* Internal iterator of scream. */
513 I32 * const pp = data ? data->scream_pos : &p;
515 if (PL_screamfirst[BmRARE(check)] >= 0
516 || ( BmRARE(check) == '\n'
517 && (BmPREVIOUS(check) == SvCUR(check) - 1)
519 s = screaminstr(sv, check,
520 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
523 /* we may be pointing at the wrong string */
524 if (s && RX_MATCH_COPIED(prog))
525 s = strbeg + (s - SvPVX_const(sv));
527 *data->scream_olds = s;
532 if (prog->extflags & RXf_CANY_SEEN) {
533 start_point= (U8*)(s + srch_start_shift);
534 end_point= (U8*)(strend - srch_end_shift);
536 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
537 end_point= HOP3(strend, -srch_end_shift, strbeg);
539 DEBUG_OPTIMISE_MORE_r({
540 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
541 (int)(end_point - start_point),
542 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
546 s = fbm_instr( start_point, end_point,
547 check, multiline ? FBMrf_MULTILINE : 0);
550 /* Update the count-of-usability, remove useless subpatterns,
554 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
555 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
556 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
557 (s ? "Found" : "Did not find"),
558 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
559 ? "anchored" : "floating"),
562 (s ? " at offset " : "...\n") );
567 /* Finish the diagnostic message */
568 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
570 /* XXX dmq: first branch is for positive lookbehind...
571 Our check string is offset from the beginning of the pattern.
572 So we need to do any stclass tests offset forward from that
581 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
582 Start with the other substr.
583 XXXX no SCREAM optimization yet - and a very coarse implementation
584 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
585 *always* match. Probably should be marked during compile...
586 Probably it is right to do no SCREAM here...
589 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
590 : (prog->float_substr && prog->anchored_substr))
592 /* Take into account the "other" substring. */
593 /* XXXX May be hopelessly wrong for UTF... */
596 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
599 char * const last = HOP3c(s, -start_shift, strbeg);
601 char * const saved_s = s;
604 t = s - prog->check_offset_max;
605 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
607 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
612 t = HOP3c(t, prog->anchored_offset, strend);
613 if (t < other_last) /* These positions already checked */
615 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
618 /* XXXX It is not documented what units *_offsets are in.
619 We assume bytes, but this is clearly wrong.
620 Meaning this code needs to be carefully reviewed for errors.
624 /* On end-of-str: see comment below. */
625 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
626 if (must == &PL_sv_undef) {
628 DEBUG_r(must = prog->anchored_utf8); /* for debug */
633 HOP3(HOP3(last1, prog->anchored_offset, strend)
634 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
636 multiline ? FBMrf_MULTILINE : 0
639 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
640 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
641 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
642 (s ? "Found" : "Contradicts"),
643 quoted, RE_SV_TAIL(must));
648 if (last1 >= last2) {
649 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
650 ", giving up...\n"));
653 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
654 ", trying floating at offset %ld...\n",
655 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
656 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
657 s = HOP3c(last, 1, strend);
661 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
662 (long)(s - i_strpos)));
663 t = HOP3c(s, -prog->anchored_offset, strbeg);
664 other_last = HOP3c(s, 1, strend);
672 else { /* Take into account the floating substring. */
674 char * const saved_s = s;
677 t = HOP3c(s, -start_shift, strbeg);
679 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
680 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
681 last = HOP3c(t, prog->float_max_offset, strend);
682 s = HOP3c(t, prog->float_min_offset, strend);
685 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
686 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
687 /* fbm_instr() takes into account exact value of end-of-str
688 if the check is SvTAIL(ed). Since false positives are OK,
689 and end-of-str is not later than strend we are OK. */
690 if (must == &PL_sv_undef) {
692 DEBUG_r(must = prog->float_utf8); /* for debug message */
695 s = fbm_instr((unsigned char*)s,
696 (unsigned char*)last + SvCUR(must)
698 must, multiline ? FBMrf_MULTILINE : 0);
700 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
701 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
702 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
703 (s ? "Found" : "Contradicts"),
704 quoted, RE_SV_TAIL(must));
708 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
709 ", giving up...\n"));
712 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
713 ", trying anchored starting at offset %ld...\n",
714 (long)(saved_s + 1 - i_strpos)));
716 s = HOP3c(t, 1, strend);
720 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
721 (long)(s - i_strpos)));
722 other_last = s; /* Fix this later. --Hugo */
732 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
734 DEBUG_OPTIMISE_MORE_r(
735 PerlIO_printf(Perl_debug_log,
736 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
737 (IV)prog->check_offset_min,
738 (IV)prog->check_offset_max,
746 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
748 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
751 /* Fixed substring is found far enough so that the match
752 cannot start at strpos. */
754 if (ml_anch && t[-1] != '\n') {
755 /* Eventually fbm_*() should handle this, but often
756 anchored_offset is not 0, so this check will not be wasted. */
757 /* XXXX In the code below we prefer to look for "^" even in
758 presence of anchored substrings. And we search even
759 beyond the found float position. These pessimizations
760 are historical artefacts only. */
762 while (t < strend - prog->minlen) {
764 if (t < check_at - prog->check_offset_min) {
765 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
766 /* Since we moved from the found position,
767 we definitely contradict the found anchored
768 substr. Due to the above check we do not
769 contradict "check" substr.
770 Thus we can arrive here only if check substr
771 is float. Redo checking for "other"=="fixed".
774 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
775 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
776 goto do_other_anchored;
778 /* We don't contradict the found floating substring. */
779 /* XXXX Why not check for STCLASS? */
781 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
782 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
785 /* Position contradicts check-string */
786 /* XXXX probably better to look for check-string
787 than for "\n", so one should lower the limit for t? */
788 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
789 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
790 other_last = strpos = s = t + 1;
795 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
796 PL_colors[0], PL_colors[1]));
800 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
801 PL_colors[0], PL_colors[1]));
805 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
808 /* The found string does not prohibit matching at strpos,
809 - no optimization of calling REx engine can be performed,
810 unless it was an MBOL and we are not after MBOL,
811 or a future STCLASS check will fail this. */
813 /* Even in this situation we may use MBOL flag if strpos is offset
814 wrt the start of the string. */
815 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
816 && (strpos != strbeg) && strpos[-1] != '\n'
817 /* May be due to an implicit anchor of m{.*foo} */
818 && !(prog->intflags & PREGf_IMPLICIT))
823 DEBUG_EXECUTE_r( if (ml_anch)
824 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
825 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
828 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
830 prog->check_utf8 /* Could be deleted already */
831 && --BmUSEFUL(prog->check_utf8) < 0
832 && (prog->check_utf8 == prog->float_utf8)
834 prog->check_substr /* Could be deleted already */
835 && --BmUSEFUL(prog->check_substr) < 0
836 && (prog->check_substr == prog->float_substr)
839 /* If flags & SOMETHING - do not do it many times on the same match */
840 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
841 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
842 if (do_utf8 ? prog->check_substr : prog->check_utf8)
843 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
844 prog->check_substr = prog->check_utf8 = NULL; /* disable */
845 prog->float_substr = prog->float_utf8 = NULL; /* clear */
846 check = NULL; /* abort */
848 /* XXXX This is a remnant of the old implementation. It
849 looks wasteful, since now INTUIT can use many
851 prog->extflags &= ~RXf_USE_INTUIT;
858 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
859 /* trie stclasses are too expensive to use here, we are better off to
860 leave it to regmatch itself */
861 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
862 /* minlen == 0 is possible if regstclass is \b or \B,
863 and the fixed substr is ''$.
864 Since minlen is already taken into account, s+1 is before strend;
865 accidentally, minlen >= 1 guaranties no false positives at s + 1
866 even for \b or \B. But (minlen? 1 : 0) below assumes that
867 regstclass does not come from lookahead... */
868 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
869 This leaves EXACTF only, which is dealt with in find_byclass(). */
870 const U8* const str = (U8*)STRING(progi->regstclass);
871 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
872 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
875 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
876 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
877 else if (prog->float_substr || prog->float_utf8)
878 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
882 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
883 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
886 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
889 const char *what = NULL;
891 if (endpos == strend) {
892 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
893 "Could not match STCLASS...\n") );
896 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
897 "This position contradicts STCLASS...\n") );
898 if ((prog->extflags & RXf_ANCH) && !ml_anch)
900 /* Contradict one of substrings */
901 if (prog->anchored_substr || prog->anchored_utf8) {
902 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
903 DEBUG_EXECUTE_r( what = "anchored" );
905 s = HOP3c(t, 1, strend);
906 if (s + start_shift + end_shift > strend) {
907 /* XXXX Should be taken into account earlier? */
908 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
909 "Could not match STCLASS...\n") );
914 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
915 "Looking for %s substr starting at offset %ld...\n",
916 what, (long)(s + start_shift - i_strpos)) );
919 /* Have both, check_string is floating */
920 if (t + start_shift >= check_at) /* Contradicts floating=check */
921 goto retry_floating_check;
922 /* Recheck anchored substring, but not floating... */
926 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
927 "Looking for anchored substr starting at offset %ld...\n",
928 (long)(other_last - i_strpos)) );
929 goto do_other_anchored;
931 /* Another way we could have checked stclass at the
932 current position only: */
937 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
938 "Looking for /%s^%s/m starting at offset %ld...\n",
939 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
942 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
944 /* Check is floating subtring. */
945 retry_floating_check:
946 t = check_at - start_shift;
947 DEBUG_EXECUTE_r( what = "floating" );
948 goto hop_and_restart;
951 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
952 "By STCLASS: moving %ld --> %ld\n",
953 (long)(t - i_strpos), (long)(s - i_strpos))
957 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
958 "Does not contradict STCLASS...\n");
963 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
964 PL_colors[4], (check ? "Guessed" : "Giving up"),
965 PL_colors[5], (long)(s - i_strpos)) );
968 fail_finish: /* Substring not found */
969 if (prog->check_substr || prog->check_utf8) /* could be removed already */
970 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
973 PL_colors[4], PL_colors[5]));
979 #define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \
980 foldlen, foldbuf, uniflags) STMT_START { \
981 switch (trie_type) { \
982 case trie_utf8_fold: \
984 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
989 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
990 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
991 foldlen -= UNISKIP( uvc ); \
992 uscan = foldbuf + UNISKIP( uvc ); \
996 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1004 charid = trie->charmap[ uvc ]; \
1008 if (trie->widecharmap) { \
1009 SV** const svpp = hv_fetch(trie->widecharmap, \
1010 (char*)&uvc, sizeof(UV), 0); \
1012 charid = (U16)SvIV(*svpp); \
1017 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1020 ibcmp_utf8(s, NULL, 0, do_utf8, \
1021 m, NULL, ln, (bool)UTF)) \
1022 && (!reginfo || regtry(reginfo, &s)) ) \
1025 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1026 uvchr_to_utf8(tmpbuf, c); \
1027 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1029 && (f == c1 || f == c2) \
1030 && (ln == foldlen || \
1031 !ibcmp_utf8((char *) foldbuf, \
1032 NULL, foldlen, do_utf8, \
1034 NULL, ln, (bool)UTF)) \
1035 && (!reginfo || regtry(reginfo, &s)) ) \
1040 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1044 && (ln == 1 || !(OP(c) == EXACTF \
1046 : ibcmp_locale(s, m, ln))) \
1047 && (!reginfo || regtry(reginfo, &s)) ) \
1053 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1055 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1061 #define REXEC_FBC_SCAN(CoDe) \
1063 while (s < strend) { \
1069 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1070 REXEC_FBC_UTF8_SCAN( \
1072 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1081 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1084 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1093 #define REXEC_FBC_TRYIT \
1094 if ((!reginfo || regtry(reginfo, &s))) \
1097 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1100 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1103 REXEC_FBC_CLASS_SCAN(CoNd); \
1107 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1108 PL_reg_flags |= RF_tainted; \
1110 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1113 REXEC_FBC_CLASS_SCAN(CoNd); \
1117 #define DUMP_EXEC_POS(li,s,doutf8) \
1118 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1120 /* We know what class REx starts with. Try to find this position... */
1121 /* if reginfo is NULL, its a dryrun */
1122 /* annoyingly all the vars in this routine have different names from their counterparts
1123 in regmatch. /grrr */
1126 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1127 const char *strend, regmatch_info *reginfo)
1130 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1134 register STRLEN uskip;
1138 register I32 tmp = 1; /* Scratch variable? */
1139 register const bool do_utf8 = PL_reg_match_utf8;
1140 RXi_GET_DECL(prog,progi);
1142 /* We know what class it must start with. */
1146 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1147 !UTF8_IS_INVARIANT((U8)s[0]) ?
1148 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1149 REGINCLASS(prog, c, (U8*)s));
1152 while (s < strend) {
1155 if (REGINCLASS(prog, c, (U8*)s) ||
1156 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1157 /* The assignment of 2 is intentional:
1158 * for the folded sharp s, the skip is 2. */
1159 (skip = SHARP_S_SKIP))) {
1160 if (tmp && (!reginfo || regtry(reginfo, &s)))
1173 if (tmp && (!reginfo || regtry(reginfo, &s)))
1181 ln = STR_LEN(c); /* length to match in octets/bytes */
1182 lnc = (I32) ln; /* length to match in characters */
1184 STRLEN ulen1, ulen2;
1186 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1187 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1188 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1190 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1191 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1193 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1195 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1198 while (sm < ((U8 *) m + ln)) {
1213 c2 = PL_fold_locale[c1];
1215 e = HOP3c(strend, -((I32)lnc), s);
1217 if (!reginfo && e < s)
1218 e = s; /* Due to minlen logic of intuit() */
1220 /* The idea in the EXACTF* cases is to first find the
1221 * first character of the EXACTF* node and then, if
1222 * necessary, case-insensitively compare the full
1223 * text of the node. The c1 and c2 are the first
1224 * characters (though in Unicode it gets a bit
1225 * more complicated because there are more cases
1226 * than just upper and lower: one needs to use
1227 * the so-called folding case for case-insensitive
1228 * matching (called "loose matching" in Unicode).
1229 * ibcmp_utf8() will do just that. */
1233 U8 tmpbuf [UTF8_MAXBYTES+1];
1234 STRLEN len, foldlen;
1235 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1237 /* Upper and lower of 1st char are equal -
1238 * probably not a "letter". */
1240 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1242 REXEC_FBC_EXACTISH_CHECK(c == c1);
1247 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1250 /* Handle some of the three Greek sigmas cases.
1251 * Note that not all the possible combinations
1252 * are handled here: some of them are handled
1253 * by the standard folding rules, and some of
1254 * them (the character class or ANYOF cases)
1255 * are handled during compiletime in
1256 * regexec.c:S_regclass(). */
1257 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1258 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1259 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1261 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1267 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1269 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1273 PL_reg_flags |= RF_tainted;
1280 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1281 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1283 tmp = ((OP(c) == BOUND ?
1284 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1285 LOAD_UTF8_CHARCLASS_ALNUM();
1286 REXEC_FBC_UTF8_SCAN(
1287 if (tmp == !(OP(c) == BOUND ?
1288 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1289 isALNUM_LC_utf8((U8*)s)))
1297 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1298 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1301 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1307 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1311 PL_reg_flags |= RF_tainted;
1318 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1319 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1321 tmp = ((OP(c) == NBOUND ?
1322 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1323 LOAD_UTF8_CHARCLASS_ALNUM();
1324 REXEC_FBC_UTF8_SCAN(
1325 if (tmp == !(OP(c) == NBOUND ?
1326 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1327 isALNUM_LC_utf8((U8*)s)))
1329 else REXEC_FBC_TRYIT;
1333 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1334 tmp = ((OP(c) == NBOUND ?
1335 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1338 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1340 else REXEC_FBC_TRYIT;
1343 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1347 REXEC_FBC_CSCAN_PRELOAD(
1348 LOAD_UTF8_CHARCLASS_ALNUM(),
1349 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1353 REXEC_FBC_CSCAN_TAINT(
1354 isALNUM_LC_utf8((U8*)s),
1358 REXEC_FBC_CSCAN_PRELOAD(
1359 LOAD_UTF8_CHARCLASS_ALNUM(),
1360 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1364 REXEC_FBC_CSCAN_TAINT(
1365 !isALNUM_LC_utf8((U8*)s),
1369 REXEC_FBC_CSCAN_PRELOAD(
1370 LOAD_UTF8_CHARCLASS_SPACE(),
1371 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1375 REXEC_FBC_CSCAN_TAINT(
1376 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1380 REXEC_FBC_CSCAN_PRELOAD(
1381 LOAD_UTF8_CHARCLASS_SPACE(),
1382 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1386 REXEC_FBC_CSCAN_TAINT(
1387 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1391 REXEC_FBC_CSCAN_PRELOAD(
1392 LOAD_UTF8_CHARCLASS_DIGIT(),
1393 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1397 REXEC_FBC_CSCAN_TAINT(
1398 isDIGIT_LC_utf8((U8*)s),
1402 REXEC_FBC_CSCAN_PRELOAD(
1403 LOAD_UTF8_CHARCLASS_DIGIT(),
1404 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1408 REXEC_FBC_CSCAN_TAINT(
1409 !isDIGIT_LC_utf8((U8*)s),
1415 const enum { trie_plain, trie_utf8, trie_utf8_fold }
1416 trie_type = do_utf8 ?
1417 (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
1419 /* what trie are we using right now */
1421 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1423 = (reg_trie_data*)progi->data->data[ aho->trie ];
1425 const char *last_start = strend - trie->minlen;
1427 const char *real_start = s;
1429 STRLEN maxlen = trie->maxlen;
1431 U8 **points; /* map of where we were in the input string
1432 when reading a given char. For ASCII this
1433 is unnecessary overhead as the relationship
1434 is always 1:1, but for unicode, especially
1435 case folded unicode this is not true. */
1436 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1440 GET_RE_DEBUG_FLAGS_DECL;
1442 /* We can't just allocate points here. We need to wrap it in
1443 * an SV so it gets freed properly if there is a croak while
1444 * running the match */
1447 sv_points=newSV(maxlen * sizeof(U8 *));
1448 SvCUR_set(sv_points,
1449 maxlen * sizeof(U8 *));
1450 SvPOK_on(sv_points);
1451 sv_2mortal(sv_points);
1452 points=(U8**)SvPV_nolen(sv_points );
1453 if ( trie_type != trie_utf8_fold
1454 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1457 bitmap=(U8*)trie->bitmap;
1459 bitmap=(U8*)ANYOF_BITMAP(c);
1461 /* this is the Aho-Corasick algorithm modified a touch
1462 to include special handling for long "unknown char"
1463 sequences. The basic idea being that we use AC as long
1464 as we are dealing with a possible matching char, when
1465 we encounter an unknown char (and we have not encountered
1466 an accepting state) we scan forward until we find a legal
1468 AC matching is basically that of trie matching, except
1469 that when we encounter a failing transition, we fall back
1470 to the current states "fail state", and try the current char
1471 again, a process we repeat until we reach the root state,
1472 state 1, or a legal transition. If we fail on the root state
1473 then we can either terminate if we have reached an accepting
1474 state previously, or restart the entire process from the beginning
1478 while (s <= last_start) {
1479 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1487 U8 *uscan = (U8*)NULL;
1488 U8 *leftmost = NULL;
1490 U32 accepted_word= 0;
1494 while ( state && uc <= (U8*)strend ) {
1496 U32 word = aho->states[ state ].wordnum;
1500 DEBUG_TRIE_EXECUTE_r(
1501 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1502 dump_exec_pos( (char *)uc, c, strend, real_start,
1503 (char *)uc, do_utf8 );
1504 PerlIO_printf( Perl_debug_log,
1505 " Scanning for legal start char...\n");
1508 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1513 if (uc >(U8*)last_start) break;
1517 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1518 if (!leftmost || lpos < leftmost) {
1519 DEBUG_r(accepted_word=word);
1525 points[pointpos++ % maxlen]= uc;
1526 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
1527 uvc, charid, foldlen, foldbuf, uniflags);
1528 DEBUG_TRIE_EXECUTE_r({
1529 dump_exec_pos( (char *)uc, c, strend, real_start,
1531 PerlIO_printf(Perl_debug_log,
1532 " Charid:%3u CP:%4"UVxf" ",
1538 word = aho->states[ state ].wordnum;
1540 base = aho->states[ state ].trans.base;
1542 DEBUG_TRIE_EXECUTE_r({
1544 dump_exec_pos( (char *)uc, c, strend, real_start,
1546 PerlIO_printf( Perl_debug_log,
1547 "%sState: %4"UVxf", word=%"UVxf,
1548 failed ? " Fail transition to " : "",
1549 (UV)state, (UV)word);
1554 (base + charid > trie->uniquecharcount )
1555 && (base + charid - 1 - trie->uniquecharcount
1557 && trie->trans[base + charid - 1 -
1558 trie->uniquecharcount].check == state
1559 && (tmp=trie->trans[base + charid - 1 -
1560 trie->uniquecharcount ].next))
1562 DEBUG_TRIE_EXECUTE_r(
1563 PerlIO_printf( Perl_debug_log," - legal\n"));
1568 DEBUG_TRIE_EXECUTE_r(
1569 PerlIO_printf( Perl_debug_log," - fail\n"));
1571 state = aho->fail[state];
1575 /* we must be accepting here */
1576 DEBUG_TRIE_EXECUTE_r(
1577 PerlIO_printf( Perl_debug_log," - accepting\n"));
1586 if (!state) state = 1;
1589 if ( aho->states[ state ].wordnum ) {
1590 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1591 if (!leftmost || lpos < leftmost) {
1592 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1597 s = (char*)leftmost;
1598 DEBUG_TRIE_EXECUTE_r({
1600 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1601 (UV)accepted_word, (IV)(s - real_start)
1604 if (!reginfo || regtry(reginfo, &s)) {
1610 DEBUG_TRIE_EXECUTE_r({
1611 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1614 DEBUG_TRIE_EXECUTE_r(
1615 PerlIO_printf( Perl_debug_log,"No match.\n"));
1624 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1633 - regexec_flags - match a regexp against a string
1636 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1637 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1638 /* strend: pointer to null at end of string */
1639 /* strbeg: real beginning of string */
1640 /* minend: end of match must be >=minend after stringarg. */
1641 /* data: May be used for some additional optimizations.
1642 Currently its only used, with a U32 cast, for transmitting
1643 the ganch offset when doing a /g match. This will change */
1644 /* nosave: For optimizations. */
1647 /*register*/ char *s;
1648 register regnode *c;
1649 /*register*/ char *startpos = stringarg;
1650 I32 minlen; /* must match at least this many chars */
1651 I32 dontbother = 0; /* how many characters not to try at end */
1652 I32 end_shift = 0; /* Same for the end. */ /* CC */
1653 I32 scream_pos = -1; /* Internal iterator of scream. */
1654 char *scream_olds = NULL;
1655 SV* const oreplsv = GvSV(PL_replgv);
1656 const bool do_utf8 = (bool)DO_UTF8(sv);
1658 RXi_GET_DECL(prog,progi);
1659 regmatch_info reginfo; /* create some info to pass to regtry etc */
1661 GET_RE_DEBUG_FLAGS_DECL;
1663 PERL_UNUSED_ARG(data);
1665 /* Be paranoid... */
1666 if (prog == NULL || startpos == NULL) {
1667 Perl_croak(aTHX_ "NULL regexp parameter");
1671 multiline = prog->extflags & RXf_PMf_MULTILINE;
1672 reginfo.prog = prog;
1674 RX_MATCH_UTF8_set(prog, do_utf8);
1676 debug_start_match(prog, do_utf8, startpos, strend,
1680 minlen = prog->minlen;
1682 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1683 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1684 "String too short [regexec_flags]...\n"));
1689 /* Check validity of program. */
1690 if (UCHARAT(progi->program) != REG_MAGIC) {
1691 Perl_croak(aTHX_ "corrupted regexp program");
1695 PL_reg_eval_set = 0;
1698 if (prog->extflags & RXf_UTF8)
1699 PL_reg_flags |= RF_utf8;
1701 /* Mark beginning of line for ^ and lookbehind. */
1702 reginfo.bol = startpos; /* XXX not used ??? */
1706 /* Mark end of line for $ (and such) */
1709 /* see how far we have to get to not match where we matched before */
1710 reginfo.till = startpos+minend;
1712 /* If there is a "must appear" string, look for it. */
1715 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1718 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1719 reginfo.ganch = startpos + prog->gofs;
1720 else if (sv && SvTYPE(sv) >= SVt_PVMG
1722 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1723 && mg->mg_len >= 0) {
1724 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1725 if (prog->extflags & RXf_ANCH_GPOS) {
1726 if (s > reginfo.ganch)
1728 s = reginfo.ganch - prog->gofs;
1732 reginfo.ganch = strbeg + PTR2UV(data);
1733 } else /* pos() not defined */
1734 reginfo.ganch = strbeg;
1736 if (PL_curpm && (PM_GETRE(PL_curpm) == prog)) {
1739 /* We have to be careful. If the previous successful match
1740 was from this regex we don't want a subsequent paritally
1741 successful match to clobber the old results.
1742 So when we detect this possibility we add a swap buffer
1743 to the re, and switch the buffer each match. If we fail
1744 we switch it back, otherwise we leave it swapped.
1746 Newxz(progi->swap, 1, regexp_paren_ofs);
1747 /* no need to copy these */
1748 Newxz(progi->swap->startp, prog->nparens + 1, I32);
1749 Newxz(progi->swap->endp, prog->nparens + 1, I32);
1751 t = progi->swap->startp;
1752 progi->swap->startp = prog->startp;
1754 t = progi->swap->endp;
1755 progi->swap->endp = prog->endp;
1758 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1759 re_scream_pos_data d;
1761 d.scream_olds = &scream_olds;
1762 d.scream_pos = &scream_pos;
1763 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1765 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1766 goto phooey; /* not present */
1772 /* Simplest case: anchored match need be tried only once. */
1773 /* [unless only anchor is BOL and multiline is set] */
1774 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
1775 if (s == startpos && regtry(®info, &startpos))
1777 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
1778 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
1783 dontbother = minlen - 1;
1784 end = HOP3c(strend, -dontbother, strbeg) - 1;
1785 /* for multiline we only have to try after newlines */
1786 if (prog->check_substr || prog->check_utf8) {
1790 if (regtry(®info, &s))
1795 if (prog->extflags & RXf_USE_INTUIT) {
1796 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1807 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1808 if (regtry(®info, &s))
1815 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
1817 /* the warning about reginfo.ganch being used without intialization
1818 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
1819 and we only enter this block when the same bit is set. */
1820 char *tmp_s = reginfo.ganch - prog->gofs;
1821 if (regtry(®info, &tmp_s))
1826 /* Messy cases: unanchored match. */
1827 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
1828 /* we have /x+whatever/ */
1829 /* it must be a one character string (XXXX Except UTF?) */
1834 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1835 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1836 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1841 DEBUG_EXECUTE_r( did_match = 1 );
1842 if (regtry(®info, &s)) goto got_it;
1844 while (s < strend && *s == ch)
1852 DEBUG_EXECUTE_r( did_match = 1 );
1853 if (regtry(®info, &s)) goto got_it;
1855 while (s < strend && *s == ch)
1860 DEBUG_EXECUTE_r(if (!did_match)
1861 PerlIO_printf(Perl_debug_log,
1862 "Did not find anchored character...\n")
1865 else if (prog->anchored_substr != NULL
1866 || prog->anchored_utf8 != NULL
1867 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1868 && prog->float_max_offset < strend - s)) {
1873 char *last1; /* Last position checked before */
1877 if (prog->anchored_substr || prog->anchored_utf8) {
1878 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1879 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1880 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1881 back_max = back_min = prog->anchored_offset;
1883 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1884 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1885 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1886 back_max = prog->float_max_offset;
1887 back_min = prog->float_min_offset;
1891 if (must == &PL_sv_undef)
1892 /* could not downgrade utf8 check substring, so must fail */
1898 last = HOP3c(strend, /* Cannot start after this */
1899 -(I32)(CHR_SVLEN(must)
1900 - (SvTAIL(must) != 0) + back_min), strbeg);
1903 last1 = HOPc(s, -1);
1905 last1 = s - 1; /* bogus */
1907 /* XXXX check_substr already used to find "s", can optimize if
1908 check_substr==must. */
1910 dontbother = end_shift;
1911 strend = HOPc(strend, -dontbother);
1912 while ( (s <= last) &&
1913 ((flags & REXEC_SCREAM)
1914 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
1915 end_shift, &scream_pos, 0))
1916 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
1917 (unsigned char*)strend, must,
1918 multiline ? FBMrf_MULTILINE : 0))) ) {
1919 /* we may be pointing at the wrong string */
1920 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1921 s = strbeg + (s - SvPVX_const(sv));
1922 DEBUG_EXECUTE_r( did_match = 1 );
1923 if (HOPc(s, -back_max) > last1) {
1924 last1 = HOPc(s, -back_min);
1925 s = HOPc(s, -back_max);
1928 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1930 last1 = HOPc(s, -back_min);
1934 while (s <= last1) {
1935 if (regtry(®info, &s))
1941 while (s <= last1) {
1942 if (regtry(®info, &s))
1948 DEBUG_EXECUTE_r(if (!did_match) {
1949 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
1950 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1951 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
1952 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1953 ? "anchored" : "floating"),
1954 quoted, RE_SV_TAIL(must));
1958 else if ( (c = progi->regstclass) ) {
1960 const OPCODE op = OP(progi->regstclass);
1961 /* don't bother with what can't match */
1962 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
1963 strend = HOPc(strend, -(minlen - 1));
1966 SV * const prop = sv_newmortal();
1967 regprop(prog, prop, c);
1969 RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1),
1971 PerlIO_printf(Perl_debug_log,
1972 "Matching stclass %.*s against %s (%d chars)\n",
1973 (int)SvCUR(prop), SvPVX_const(prop),
1974 quoted, (int)(strend - s));
1977 if (find_byclass(prog, c, s, strend, ®info))
1979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
1983 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1988 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1989 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1990 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1992 if (flags & REXEC_SCREAM) {
1993 last = screaminstr(sv, float_real, s - strbeg,
1994 end_shift, &scream_pos, 1); /* last one */
1996 last = scream_olds; /* Only one occurrence. */
1997 /* we may be pointing at the wrong string */
1998 else if (RX_MATCH_COPIED(prog))
1999 s = strbeg + (s - SvPVX_const(sv));
2003 const char * const little = SvPV_const(float_real, len);
2005 if (SvTAIL(float_real)) {
2006 if (memEQ(strend - len + 1, little, len - 1))
2007 last = strend - len + 1;
2008 else if (!multiline)
2009 last = memEQ(strend - len, little, len)
2010 ? strend - len : NULL;
2016 last = rninstr(s, strend, little, little + len);
2018 last = strend; /* matching "$" */
2023 PerlIO_printf(Perl_debug_log,
2024 "%sCan't trim the tail, match fails (should not happen)%s\n",
2025 PL_colors[4], PL_colors[5]));
2026 goto phooey; /* Should not happen! */
2028 dontbother = strend - last + prog->float_min_offset;
2030 if (minlen && (dontbother < minlen))
2031 dontbother = minlen - 1;
2032 strend -= dontbother; /* this one's always in bytes! */
2033 /* We don't know much -- general case. */
2036 if (regtry(®info, &s))
2045 if (regtry(®info, &s))
2047 } while (s++ < strend);
2055 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2057 if (PL_reg_eval_set) {
2058 /* Preserve the current value of $^R */
2059 if (oreplsv != GvSV(PL_replgv))
2060 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2061 restored, the value remains
2063 restore_pos(aTHX_ prog);
2065 if (prog->paren_names)
2066 (void)hv_iterinit(prog->paren_names);
2068 /* make sure $`, $&, $', and $digit will work later */
2069 if ( !(flags & REXEC_NOT_FIRST) ) {
2070 RX_MATCH_COPY_FREE(prog);
2071 if (flags & REXEC_COPY_STR) {
2072 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2073 #ifdef PERL_OLD_COPY_ON_WRITE
2075 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2077 PerlIO_printf(Perl_debug_log,
2078 "Copy on write: regexp capture, type %d\n",
2081 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2082 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2083 assert (SvPOKp(prog->saved_copy));
2087 RX_MATCH_COPIED_on(prog);
2088 s = savepvn(strbeg, i);
2094 prog->subbeg = strbeg;
2095 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2102 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2103 PL_colors[4], PL_colors[5]));
2104 if (PL_reg_eval_set)
2105 restore_pos(aTHX_ prog);
2107 /* we failed :-( roll it back */
2109 t = progi->swap->startp;
2110 progi->swap->startp = prog->startp;
2112 t = progi->swap->endp;
2113 progi->swap->endp = prog->endp;
2121 - regtry - try match at specific point
2123 STATIC I32 /* 0 failure, 1 success */
2124 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2130 regexp *prog = reginfo->prog;
2131 RXi_GET_DECL(prog,progi);
2132 GET_RE_DEBUG_FLAGS_DECL;
2133 reginfo->cutpoint=NULL;
2135 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2138 PL_reg_eval_set = RS_init;
2139 DEBUG_EXECUTE_r(DEBUG_s(
2140 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2141 (IV)(PL_stack_sp - PL_stack_base));
2144 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2145 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2147 /* Apparently this is not needed, judging by wantarray. */
2148 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2149 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2152 /* Make $_ available to executed code. */
2153 if (reginfo->sv != DEFSV) {
2155 DEFSV = reginfo->sv;
2158 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2159 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2160 /* prepare for quick setting of pos */
2161 #ifdef PERL_OLD_COPY_ON_WRITE
2162 if (SvIsCOW(reginfo->sv))
2163 sv_force_normal_flags(reginfo->sv, 0);
2165 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2166 &PL_vtbl_mglob, NULL, 0);
2170 PL_reg_oldpos = mg->mg_len;
2171 SAVEDESTRUCTOR_X(restore_pos, prog);
2173 if (!PL_reg_curpm) {
2174 Newxz(PL_reg_curpm, 1, PMOP);
2177 SV* const repointer = newSViv(0);
2178 /* so we know which PL_regex_padav element is PL_reg_curpm */
2179 SvFLAGS(repointer) |= SVf_BREAK;
2180 av_push(PL_regex_padav,repointer);
2181 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2182 PL_regex_pad = AvARRAY(PL_regex_padav);
2186 PM_SETRE(PL_reg_curpm, prog);
2187 PL_reg_oldcurpm = PL_curpm;
2188 PL_curpm = PL_reg_curpm;
2189 if (RX_MATCH_COPIED(prog)) {
2190 /* Here is a serious problem: we cannot rewrite subbeg,
2191 since it may be needed if this match fails. Thus
2192 $` inside (?{}) could fail... */
2193 PL_reg_oldsaved = prog->subbeg;
2194 PL_reg_oldsavedlen = prog->sublen;
2195 #ifdef PERL_OLD_COPY_ON_WRITE
2196 PL_nrs = prog->saved_copy;
2198 RX_MATCH_COPIED_off(prog);
2201 PL_reg_oldsaved = NULL;
2202 prog->subbeg = PL_bostr;
2203 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2205 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2206 prog->startp[0] = *startpos - PL_bostr;
2207 PL_reginput = *startpos;
2208 PL_reglastparen = &prog->lastparen;
2209 PL_reglastcloseparen = &prog->lastcloseparen;
2210 prog->lastparen = 0;
2211 prog->lastcloseparen = 0;
2213 PL_regstartp = prog->startp;
2214 PL_regendp = prog->endp;
2215 if (PL_reg_start_tmpl <= prog->nparens) {
2216 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2217 if(PL_reg_start_tmp)
2218 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2220 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2223 /* XXXX What this code is doing here?!!! There should be no need
2224 to do this again and again, PL_reglastparen should take care of
2227 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2228 * Actually, the code in regcppop() (which Ilya may be meaning by
2229 * PL_reglastparen), is not needed at all by the test suite
2230 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2231 * enough, for building DynaLoader, or otherwise this
2232 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2233 * will happen. Meanwhile, this code *is* needed for the
2234 * above-mentioned test suite tests to succeed. The common theme
2235 * on those tests seems to be returning null fields from matches.
2240 if (prog->nparens) {
2242 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2249 if (regmatch(reginfo, progi->program + 1)) {
2250 PL_regendp[0] = PL_reginput - PL_bostr;
2253 if (reginfo->cutpoint)
2254 *startpos= reginfo->cutpoint;
2255 REGCP_UNWIND(lastcp);
2260 #define sayYES goto yes
2261 #define sayNO goto no
2262 #define sayNO_SILENT goto no_silent
2264 /* we dont use STMT_START/END here because it leads to
2265 "unreachable code" warnings, which are bogus, but distracting. */
2266 #define CACHEsayNO \
2267 if (ST.cache_mask) \
2268 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2271 /* this is used to determine how far from the left messages like
2272 'failed...' are printed. It should be set such that messages
2273 are inline with the regop output that created them.
2275 #define REPORT_CODE_OFF 32
2278 /* Make sure there is a test for this +1 options in re_tests */
2279 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2281 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2282 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2284 #define SLAB_FIRST(s) (&(s)->states[0])
2285 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2287 /* grab a new slab and return the first slot in it */
2289 STATIC regmatch_state *
2292 #if PERL_VERSION < 9
2295 regmatch_slab *s = PL_regmatch_slab->next;
2297 Newx(s, 1, regmatch_slab);
2298 s->prev = PL_regmatch_slab;
2300 PL_regmatch_slab->next = s;
2302 PL_regmatch_slab = s;
2303 return SLAB_FIRST(s);
2307 /* push a new state then goto it */
2309 #define PUSH_STATE_GOTO(state, node) \
2311 st->resume_state = state; \
2314 /* push a new state with success backtracking, then goto it */
2316 #define PUSH_YES_STATE_GOTO(state, node) \
2318 st->resume_state = state; \
2319 goto push_yes_state;
2325 regmatch() - main matching routine
2327 This is basically one big switch statement in a loop. We execute an op,
2328 set 'next' to point the next op, and continue. If we come to a point which
2329 we may need to backtrack to on failure such as (A|B|C), we push a
2330 backtrack state onto the backtrack stack. On failure, we pop the top
2331 state, and re-enter the loop at the state indicated. If there are no more
2332 states to pop, we return failure.
2334 Sometimes we also need to backtrack on success; for example /A+/, where
2335 after successfully matching one A, we need to go back and try to
2336 match another one; similarly for lookahead assertions: if the assertion
2337 completes successfully, we backtrack to the state just before the assertion
2338 and then carry on. In these cases, the pushed state is marked as
2339 'backtrack on success too'. This marking is in fact done by a chain of
2340 pointers, each pointing to the previous 'yes' state. On success, we pop to
2341 the nearest yes state, discarding any intermediate failure-only states.
2342 Sometimes a yes state is pushed just to force some cleanup code to be
2343 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2344 it to free the inner regex.
2346 Note that failure backtracking rewinds the cursor position, while
2347 success backtracking leaves it alone.
2349 A pattern is complete when the END op is executed, while a subpattern
2350 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2351 ops trigger the "pop to last yes state if any, otherwise return true"
2354 A common convention in this function is to use A and B to refer to the two
2355 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2356 the subpattern to be matched possibly multiple times, while B is the entire
2357 rest of the pattern. Variable and state names reflect this convention.
2359 The states in the main switch are the union of ops and failure/success of
2360 substates associated with with that op. For example, IFMATCH is the op
2361 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2362 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2363 successfully matched A and IFMATCH_A_fail is a state saying that we have
2364 just failed to match A. Resume states always come in pairs. The backtrack
2365 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2366 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2367 on success or failure.
2369 The struct that holds a backtracking state is actually a big union, with
2370 one variant for each major type of op. The variable st points to the
2371 top-most backtrack struct. To make the code clearer, within each
2372 block of code we #define ST to alias the relevant union.
2374 Here's a concrete example of a (vastly oversimplified) IFMATCH
2380 #define ST st->u.ifmatch
2382 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2383 ST.foo = ...; // some state we wish to save
2385 // push a yes backtrack state with a resume value of
2386 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2388 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2391 case IFMATCH_A: // we have successfully executed A; now continue with B
2393 bar = ST.foo; // do something with the preserved value
2396 case IFMATCH_A_fail: // A failed, so the assertion failed
2397 ...; // do some housekeeping, then ...
2398 sayNO; // propagate the failure
2405 For any old-timers reading this who are familiar with the old recursive
2406 approach, the code above is equivalent to:
2408 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2417 ...; // do some housekeeping, then ...
2418 sayNO; // propagate the failure
2421 The topmost backtrack state, pointed to by st, is usually free. If you
2422 want to claim it, populate any ST.foo fields in it with values you wish to
2423 save, then do one of
2425 PUSH_STATE_GOTO(resume_state, node);
2426 PUSH_YES_STATE_GOTO(resume_state, node);
2428 which sets that backtrack state's resume value to 'resume_state', pushes a
2429 new free entry to the top of the backtrack stack, then goes to 'node'.
2430 On backtracking, the free slot is popped, and the saved state becomes the
2431 new free state. An ST.foo field in this new top state can be temporarily
2432 accessed to retrieve values, but once the main loop is re-entered, it
2433 becomes available for reuse.
2435 Note that the depth of the backtrack stack constantly increases during the
2436 left-to-right execution of the pattern, rather than going up and down with
2437 the pattern nesting. For example the stack is at its maximum at Z at the
2438 end of the pattern, rather than at X in the following:
2440 /(((X)+)+)+....(Y)+....Z/
2442 The only exceptions to this are lookahead/behind assertions and the cut,
2443 (?>A), which pop all the backtrack states associated with A before
2446 Bascktrack state structs are allocated in slabs of about 4K in size.
2447 PL_regmatch_state and st always point to the currently active state,
2448 and PL_regmatch_slab points to the slab currently containing
2449 PL_regmatch_state. The first time regmatch() is called, the first slab is
2450 allocated, and is never freed until interpreter destruction. When the slab
2451 is full, a new one is allocated and chained to the end. At exit from
2452 regmatch(), slabs allocated since entry are freed.
2457 #define DEBUG_STATE_pp(pp) \
2459 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2460 PerlIO_printf(Perl_debug_log, \
2461 " %*s"pp" %s%s%s%s%s\n", \
2463 reg_name[st->resume_state], \
2464 ((st==yes_state||st==mark_state) ? "[" : ""), \
2465 ((st==yes_state) ? "Y" : ""), \
2466 ((st==mark_state) ? "M" : ""), \
2467 ((st==yes_state||st==mark_state) ? "]" : "") \
2472 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2477 S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
2478 const char *start, const char *end, const char *blurb)
2480 const bool utf8_pat= prog->extflags & RXf_UTF8 ? 1 : 0;
2484 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2485 prog->precomp, prog->prelen, 60);
2487 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2488 start, end - start, 60);
2490 PerlIO_printf(Perl_debug_log,
2491 "%s%s REx%s %s against %s\n",
2492 PL_colors[4], blurb, PL_colors[5], s0, s1);
2494 if (do_utf8||utf8_pat)
2495 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2496 utf8_pat ? "pattern" : "",
2497 utf8_pat && do_utf8 ? " and " : "",
2498 do_utf8 ? "string" : ""
2504 S_dump_exec_pos(pTHX_ const char *locinput,
2505 const regnode *scan,
2506 const char *loc_regeol,
2507 const char *loc_bostr,
2508 const char *loc_reg_starttry,
2511 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2512 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2513 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2514 /* The part of the string before starttry has one color
2515 (pref0_len chars), between starttry and current
2516 position another one (pref_len - pref0_len chars),
2517 after the current position the third one.
2518 We assume that pref0_len <= pref_len, otherwise we
2519 decrease pref0_len. */
2520 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2521 ? (5 + taill) - l : locinput - loc_bostr;
2524 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2526 pref0_len = pref_len - (locinput - loc_reg_starttry);
2527 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2528 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2529 ? (5 + taill) - pref_len : loc_regeol - locinput);
2530 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2534 if (pref0_len > pref_len)
2535 pref0_len = pref_len;
2537 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2539 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2540 (locinput - pref_len),pref0_len, 60, 4, 5);
2542 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2543 (locinput - pref_len + pref0_len),
2544 pref_len - pref0_len, 60, 2, 3);
2546 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2547 locinput, loc_regeol - locinput, 10, 0, 1);
2549 const STRLEN tlen=len0+len1+len2;
2550 PerlIO_printf(Perl_debug_log,
2551 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2552 (IV)(locinput - loc_bostr),
2555 (docolor ? "" : "> <"),
2557 (int)(tlen > 19 ? 0 : 19 - tlen),
2564 /* reg_check_named_buff_matched()
2565 * Checks to see if a named buffer has matched. The data array of
2566 * buffer numbers corresponding to the buffer is expected to reside
2567 * in the regexp->data->data array in the slot stored in the ARG() of
2568 * node involved. Note that this routine doesn't actually care about the
2569 * name, that information is not preserved from compilation to execution.
2570 * Returns the index of the leftmost defined buffer with the given name
2571 * or 0 if non of the buffers matched.
2574 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
2576 RXi_GET_DECL(rex,rexi);
2577 SV *sv_dat=(SV*)rexi->data->data[ ARG( scan ) ];
2578 I32 *nums=(I32*)SvPVX(sv_dat);
2579 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2580 if ((I32)*PL_reglastparen >= nums[n] &&
2581 PL_regendp[nums[n]] != -1)
2589 STATIC I32 /* 0 failure, 1 success */
2590 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2592 #if PERL_VERSION < 9
2596 register const bool do_utf8 = PL_reg_match_utf8;
2597 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2599 regexp *rex = reginfo->prog;
2600 RXi_GET_DECL(rex,rexi);
2602 regmatch_slab *orig_slab;
2603 regmatch_state *orig_state;
2605 /* the current state. This is a cached copy of PL_regmatch_state */
2606 register regmatch_state *st;
2608 /* cache heavy used fields of st in registers */
2609 register regnode *scan;
2610 register regnode *next;
2611 register U32 n = 0; /* general value; init to avoid compiler warning */
2612 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2613 register char *locinput = PL_reginput;
2614 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2616 bool result = 0; /* return value of S_regmatch */
2617 int depth = 0; /* depth of backtrack stack */
2618 int nochange_depth = 0; /* depth of GOSUB recursion with nochange*/
2619 regmatch_state *yes_state = NULL; /* state to pop to on success of
2621 /* mark_state piggy backs on the yes_state logic so that when we unwind
2622 the stack on success we can update the mark_state as we go */
2623 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2625 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2626 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2628 bool no_final = 0; /* prevent failure from backtracking? */
2629 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2630 char *startpoint = PL_reginput;
2631 SV *popmark = NULL; /* are we looking for a mark? */
2632 SV *sv_commit = NULL; /* last mark name seen in failure */
2633 SV *sv_yes_mark = NULL; /* last mark name we have seen
2634 during a successfull match */
2635 U32 lastopen = 0; /* last open we saw */
2636 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2639 /* these three flags are set by various ops to signal information to
2640 * the very next op. They have a useful lifetime of exactly one loop
2641 * iteration, and are not preserved or restored by state pushes/pops
2643 bool sw = 0; /* the condition value in (?(cond)a|b) */
2644 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2645 int logical = 0; /* the following EVAL is:
2649 or the following IFMATCH/UNLESSM is:
2650 false: plain (?=foo)
2651 true: used as a condition: (?(?=foo))
2655 GET_RE_DEBUG_FLAGS_DECL;
2659 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2661 /* on first ever call to regmatch, allocate first slab */
2662 if (!PL_regmatch_slab) {
2663 Newx(PL_regmatch_slab, 1, regmatch_slab);
2664 PL_regmatch_slab->prev = NULL;
2665 PL_regmatch_slab->next = NULL;
2666 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2669 /* remember current high-water mark for exit */
2670 /* XXX this should be done with SAVE* instead */
2671 orig_slab = PL_regmatch_slab;
2672 orig_state = PL_regmatch_state;
2674 /* grab next free state slot */
2675 st = ++PL_regmatch_state;
2676 if (st > SLAB_LAST(PL_regmatch_slab))
2677 st = PL_regmatch_state = S_push_slab(aTHX);
2679 /* Note that nextchr is a byte even in UTF */
2680 nextchr = UCHARAT(locinput);
2682 while (scan != NULL) {
2685 SV * const prop = sv_newmortal();
2686 regnode *rnext=regnext(scan);
2687 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2688 regprop(rex, prop, scan);
2690 PerlIO_printf(Perl_debug_log,
2691 "%3"IVdf":%*s%s(%"IVdf")\n",
2692 (IV)(scan - rexi->program), depth*2, "",
2694 (PL_regkind[OP(scan)] == END || !rnext) ?
2695 0 : (IV)(rnext - rexi->program));
2698 next = scan + NEXT_OFF(scan);
2701 state_num = OP(scan);
2704 switch (state_num) {
2706 if (locinput == PL_bostr)
2708 /* reginfo->till = reginfo->bol; */
2713 if (locinput == PL_bostr ||
2714 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2720 if (locinput == PL_bostr)
2724 if (locinput == reginfo->ganch)
2730 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2735 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2737 if (PL_regeol - locinput > 1)
2741 if (PL_regeol != locinput)
2745 if (!nextchr && locinput >= PL_regeol)
2748 locinput += PL_utf8skip[nextchr];
2749 if (locinput > PL_regeol)
2751 nextchr = UCHARAT(locinput);
2754 nextchr = UCHARAT(++locinput);
2757 if (!nextchr && locinput >= PL_regeol)
2759 nextchr = UCHARAT(++locinput);
2762 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2765 locinput += PL_utf8skip[nextchr];
2766 if (locinput > PL_regeol)
2768 nextchr = UCHARAT(locinput);
2771 nextchr = UCHARAT(++locinput);
2775 #define ST st->u.trie
2777 /* In this case the charclass data is available inline so
2778 we can fail fast without a lot of extra overhead.
2780 if (scan->flags == EXACT || !do_utf8) {
2781 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2783 PerlIO_printf(Perl_debug_log,
2784 "%*s %sfailed to match trie start class...%s\n",
2785 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2794 /* what type of TRIE am I? (utf8 makes this contextual) */
2795 const enum { trie_plain, trie_utf8, trie_utf8_fold }
2796 trie_type = do_utf8 ?
2797 (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
2800 /* what trie are we using right now */
2801 reg_trie_data * const trie
2802 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2803 U32 state = trie->startstate;
2805 if (trie->bitmap && trie_type != trie_utf8_fold &&
2806 !TRIE_BITMAP_TEST(trie,*locinput)
2808 if (trie->states[ state ].wordnum) {
2810 PerlIO_printf(Perl_debug_log,
2811 "%*s %smatched empty string...%s\n",
2812 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2817 PerlIO_printf(Perl_debug_log,
2818 "%*s %sfailed to match trie start class...%s\n",
2819 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2826 U8 *uc = ( U8* )locinput;
2830 U8 *uscan = (U8*)NULL;
2832 SV *sv_accept_buff = NULL;
2833 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2835 ST.accepted = 0; /* how many accepting states we have seen */
2837 ST.jump = trie->jump;
2841 traverse the TRIE keeping track of all accepting states
2842 we transition through until we get to a failing node.
2845 while ( state && uc <= (U8*)PL_regeol ) {
2846 U32 base = trie->states[ state ].trans.base;
2849 /* We use charid to hold the wordnum as we don't use it
2850 for charid until after we have done the wordnum logic.
2851 We define an alias just so that the wordnum logic reads
2854 #define got_wordnum charid
2855 got_wordnum = trie->states[ state ].wordnum;
2857 if ( got_wordnum ) {
2858 if ( ! ST.accepted ) {
2861 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2862 sv_accept_buff=newSV(bufflen *
2863 sizeof(reg_trie_accepted) - 1);
2864 SvCUR_set(sv_accept_buff, 0);
2865 SvPOK_on(sv_accept_buff);
2866 sv_2mortal(sv_accept_buff);
2869 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2872 if (ST.accepted >= bufflen) {
2874 ST.accept_buff =(reg_trie_accepted*)
2875 SvGROW(sv_accept_buff,
2876 bufflen * sizeof(reg_trie_accepted));
2878 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2879 + sizeof(reg_trie_accepted));
2882 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
2883 ST.accept_buff[ST.accepted].endpos = uc;
2885 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
2889 DEBUG_TRIE_EXECUTE_r({
2890 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
2891 PerlIO_printf( Perl_debug_log,
2892 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
2893 2+depth * 2, "", PL_colors[4],
2894 (UV)state, (UV)ST.accepted );
2898 REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len,
2899 uvc, charid, foldlen, foldbuf, uniflags);
2902 (base + charid > trie->uniquecharcount )
2903 && (base + charid - 1 - trie->uniquecharcount
2905 && trie->trans[base + charid - 1 -
2906 trie->uniquecharcount].check == state)
2908 state = trie->trans[base + charid - 1 -
2909 trie->uniquecharcount ].next;
2920 DEBUG_TRIE_EXECUTE_r(
2921 PerlIO_printf( Perl_debug_log,
2922 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
2923 charid, uvc, (UV)state, PL_colors[5] );
2930 PerlIO_printf( Perl_debug_log,
2931 "%*s %sgot %"IVdf" possible matches%s\n",
2932 REPORT_CODE_OFF + depth * 2, "",
2933 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
2938 case TRIE_next_fail: /* we failed - try next alterative */
2943 if ( ST.accepted == 1 ) {
2944 /* only one choice left - just continue */
2946 reg_trie_data * const trie
2947 = (reg_trie_data*)rexi->data->data[ ARG(ST.me) ];
2948 SV ** const tmp = av_fetch( trie->words,
2949 ST.accept_buff[ 0 ].wordnum-1, 0 );
2950 SV *sv= tmp ? sv_newmortal() : NULL;
2952 PerlIO_printf( Perl_debug_log,
2953 "%*s %sonly one match left: #%d <%s>%s\n",
2954 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
2955 ST.accept_buff[ 0 ].wordnum,
2956 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
2957 PL_colors[0], PL_colors[1],
2958 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2960 : "not compiled under -Dr",
2963 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
2964 /* in this case we free tmps/leave before we call regmatch
2965 as we wont be using accept_buff again. */
2967 locinput = PL_reginput;
2968 nextchr = UCHARAT(locinput);
2969 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
2972 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
2973 if (!has_cutgroup) {
2978 PUSH_YES_STATE_GOTO(TRIE_next, scan);
2981 continue; /* execute rest of RE */
2984 if (!ST.accepted-- ) {
2986 PerlIO_printf( Perl_debug_log,
2987 "%*s %sTRIE failed...%s\n",
2988 REPORT_CODE_OFF+depth*2, "",
2998 There are at least two accepting states left. Presumably
2999 the number of accepting states is going to be low,
3000 typically two. So we simply scan through to find the one
3001 with lowest wordnum. Once we find it, we swap the last
3002 state into its place and decrement the size. We then try to
3003 match the rest of the pattern at the point where the word
3004 ends. If we succeed, control just continues along the
3005 regex; if we fail we return here to try the next accepting
3012 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3013 DEBUG_TRIE_EXECUTE_r(
3014 PerlIO_printf( Perl_debug_log,
3015 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3016 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3017 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3018 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3021 if (ST.accept_buff[cur].wordnum <
3022 ST.accept_buff[best].wordnum)
3027 reg_trie_data * const trie
3028 = (reg_trie_data*)rexi->data->data[ ARG(ST.me) ];
3029 SV ** const tmp = av_fetch( trie->words,
3030 ST.accept_buff[ best ].wordnum - 1, 0 );
3031 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3033 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3034 SV *sv= tmp ? sv_newmortal() : NULL;
3036 PerlIO_printf( Perl_debug_log,
3037 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3038 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3039 ST.accept_buff[best].wordnum,
3040 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3041 PL_colors[0], PL_colors[1],
3042 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3043 ) : "not compiled under -Dr",
3044 REG_NODE_NUM(nextop),
3048 if ( best<ST.accepted ) {
3049 reg_trie_accepted tmp = ST.accept_buff[ best ];
3050 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3051 ST.accept_buff[ ST.accepted ] = tmp;
3054 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3055 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3059 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3063 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3066 PUSH_STATE_GOTO(TRIE_next, scan);
3079 char *s = STRING(scan);
3081 if (do_utf8 != UTF) {
3082 /* The target and the pattern have differing utf8ness. */
3084 const char * const e = s + ln;
3087 /* The target is utf8, the pattern is not utf8. */
3092 if (NATIVE_TO_UNI(*(U8*)s) !=
3093 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3101 /* The target is not utf8, the pattern is utf8. */
3106 if (NATIVE_TO_UNI(*((U8*)l)) !=
3107 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3115 nextchr = UCHARAT(locinput);
3118 /* The target and the pattern have the same utf8ness. */
3119 /* Inline the first character, for speed. */
3120 if (UCHARAT(s) != nextchr)
3122 if (PL_regeol - locinput < ln)
3124 if (ln > 1 && memNE(s, locinput, ln))
3127 nextchr = UCHARAT(locinput);
3131 PL_reg_flags |= RF_tainted;
3134 char * const s = STRING(scan);
3137 if (do_utf8 || UTF) {
3138 /* Either target or the pattern are utf8. */
3139 const char * const l = locinput;
3140 char *e = PL_regeol;
3142 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3143 l, &e, 0, do_utf8)) {
3144 /* One more case for the sharp s:
3145 * pack("U0U*", 0xDF) =~ /ss/i,
3146 * the 0xC3 0x9F are the UTF-8
3147 * byte sequence for the U+00DF. */
3149 toLOWER(s[0]) == 's' &&
3151 toLOWER(s[1]) == 's' &&
3158 nextchr = UCHARAT(locinput);
3162 /* Neither the target and the pattern are utf8. */
3164 /* Inline the first character, for speed. */
3165 if (UCHARAT(s) != nextchr &&
3166 UCHARAT(s) != ((OP(scan) == EXACTF)
3167 ? PL_fold : PL_fold_locale)[nextchr])
3169 if (PL_regeol - locinput < ln)
3171 if (ln > 1 && (OP(scan) == EXACTF
3172 ? ibcmp(s, locinput, ln)
3173 : ibcmp_locale(s, locinput, ln)))
3176 nextchr = UCHARAT(locinput);
3181 STRLEN inclasslen = PL_regeol - locinput;
3183 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3185 if (locinput >= PL_regeol)
3187 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3188 nextchr = UCHARAT(locinput);
3193 nextchr = UCHARAT(locinput);
3194 if (!REGINCLASS(rex, scan, (U8*)locinput))
3196 if (!nextchr && locinput >= PL_regeol)
3198 nextchr = UCHARAT(++locinput);
3202 /* If we might have the case of the German sharp s
3203 * in a casefolding Unicode character class. */
3205 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3206 locinput += SHARP_S_SKIP;
3207 nextchr = UCHARAT(locinput);
3213 PL_reg_flags |= RF_tainted;
3219 LOAD_UTF8_CHARCLASS_ALNUM();
3220 if (!(OP(scan) == ALNUM
3221 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3222 : isALNUM_LC_utf8((U8*)locinput)))
3226 locinput += PL_utf8skip[nextchr];
3227 nextchr = UCHARAT(locinput);
3230 if (!(OP(scan) == ALNUM
3231 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3233 nextchr = UCHARAT(++locinput);
3236 PL_reg_flags |= RF_tainted;
3239 if (!nextchr && locinput >= PL_regeol)
3242 LOAD_UTF8_CHARCLASS_ALNUM();
3243 if (OP(scan) == NALNUM
3244 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3245 : isALNUM_LC_utf8((U8*)locinput))
3249 locinput += PL_utf8skip[nextchr];
3250 nextchr = UCHARAT(locinput);
3253 if (OP(scan) == NALNUM
3254 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3256 nextchr = UCHARAT(++locinput);
3260 PL_reg_flags |= RF_tainted;
3264 /* was last char in word? */
3266 if (locinput == PL_bostr)
3269 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3271 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3273 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3274 ln = isALNUM_uni(ln);
3275 LOAD_UTF8_CHARCLASS_ALNUM();
3276 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3279 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3280 n = isALNUM_LC_utf8((U8*)locinput);
3284 ln = (locinput != PL_bostr) ?
3285 UCHARAT(locinput - 1) : '\n';
3286 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3288 n = isALNUM(nextchr);
3291 ln = isALNUM_LC(ln);
3292 n = isALNUM_LC(nextchr);
3295 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3296 OP(scan) == BOUNDL))
3300 PL_reg_flags |= RF_tainted;
3306 if (UTF8_IS_CONTINUED(nextchr)) {
3307 LOAD_UTF8_CHARCLASS_SPACE();
3308 if (!(OP(scan) == SPACE
3309 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3310 : isSPACE_LC_utf8((U8*)locinput)))
3314 locinput += PL_utf8skip[nextchr];
3315 nextchr = UCHARAT(locinput);
3318 if (!(OP(scan) == SPACE
3319 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3321 nextchr = UCHARAT(++locinput);
3324 if (!(OP(scan) == SPACE
3325 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3327 nextchr = UCHARAT(++locinput);
3331 PL_reg_flags |= RF_tainted;
3334 if (!nextchr && locinput >= PL_regeol)
3337 LOAD_UTF8_CHARCLASS_SPACE();
3338 if (OP(scan) == NSPACE
3339 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3340 : isSPACE_LC_utf8((U8*)locinput))
3344 locinput += PL_utf8skip[nextchr];
3345 nextchr = UCHARAT(locinput);
3348 if (OP(scan) == NSPACE
3349 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3351 nextchr = UCHARAT(++locinput);
3354 PL_reg_flags |= RF_tainted;
3360 LOAD_UTF8_CHARCLASS_DIGIT();
3361 if (!(OP(scan) == DIGIT
3362 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3363 : isDIGIT_LC_utf8((U8*)locinput)))
3367 locinput += PL_utf8skip[nextchr];
3368 nextchr = UCHARAT(locinput);
3371 if (!(OP(scan) == DIGIT
3372 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3374 nextchr = UCHARAT(++locinput);
3377 PL_reg_flags |= RF_tainted;
3380 if (!nextchr && locinput >= PL_regeol)
3383 LOAD_UTF8_CHARCLASS_DIGIT();
3384 if (OP(scan) == NDIGIT
3385 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3386 : isDIGIT_LC_utf8((U8*)locinput))
3390 locinput += PL_utf8skip[nextchr];
3391 nextchr = UCHARAT(locinput);
3394 if (OP(scan) == NDIGIT
3395 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3397 nextchr = UCHARAT(++locinput);
3400 if (locinput >= PL_regeol)
3403 LOAD_UTF8_CHARCLASS_MARK();
3404 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3406 locinput += PL_utf8skip[nextchr];
3407 while (locinput < PL_regeol &&
3408 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3409 locinput += UTF8SKIP(locinput);
3410 if (locinput > PL_regeol)
3415 nextchr = UCHARAT(locinput);
3422 PL_reg_flags |= RF_tainted;
3427 n = reg_check_named_buff_matched(rex,scan);
3430 type = REF + ( type - NREF );
3437 PL_reg_flags |= RF_tainted;
3441 n = ARG(scan); /* which paren pair */
3444 ln = PL_regstartp[n];
3445 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3446 if (*PL_reglastparen < n || ln == -1)
3447 sayNO; /* Do not match unless seen CLOSEn. */
3448 if (ln == PL_regendp[n])
3452 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3454 const char *e = PL_bostr + PL_regendp[n];
3456 * Note that we can't do the "other character" lookup trick as
3457 * in the 8-bit case (no pun intended) because in Unicode we
3458 * have to map both upper and title case to lower case.
3462 STRLEN ulen1, ulen2;
3463 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3464 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3468 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3469 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3470 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3477 nextchr = UCHARAT(locinput);
3481 /* Inline the first character, for speed. */
3482 if (UCHARAT(s) != nextchr &&
3484 (UCHARAT(s) != (type == REFF
3485 ? PL_fold : PL_fold_locale)[nextchr])))
3487 ln = PL_regendp[n] - ln;
3488 if (locinput + ln > PL_regeol)
3490 if (ln > 1 && (type == REF
3491 ? memNE(s, locinput, ln)
3493 ? ibcmp(s, locinput, ln)
3494 : ibcmp_locale(s, locinput, ln))))
3497 nextchr = UCHARAT(locinput);
3507 #define ST st->u.eval
3511 regexp_internal *rei;
3512 regnode *startpoint;
3515 case GOSUB: /* /(...(?1))/ */
3516 if (cur_eval && cur_eval->locinput==locinput) {
3517 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3518 Perl_croak(aTHX_ "Infinite recursion in regex");
3519 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3521 "Pattern subroutine nesting without pos change"
3522 " exceeded limit in regex");
3528 (void)ReREFCNT_inc(rex);
3529 if (OP(scan)==GOSUB) {
3530 startpoint = scan + ARG2L(scan);
3531 ST.close_paren = ARG(scan);
3533 startpoint = rei->program+1;
3536 goto eval_recurse_doit;
3538 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3539 if (cur_eval && cur_eval->locinput==locinput) {
3540 if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH )
3541 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3546 /* execute the code in the {...} */
3548 SV ** const before = SP;
3549 OP_4tree * const oop = PL_op;
3550 COP * const ocurcop = PL_curcop;
3554 PL_op = (OP_4tree*)rexi->data->data[n];
3555 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3556 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3557 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3558 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3560 CALLRUNOPS(aTHX); /* Scalar context. */
3563 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3570 PAD_RESTORE_LOCAL(old_comppad);
3571 PL_curcop = ocurcop;
3574 sv_setsv(save_scalar(PL_replgv), ret);
3578 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3581 /* extract RE object from returned value; compiling if
3586 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3587 mg = mg_find(sv, PERL_MAGIC_qr);
3588 else if (SvSMAGICAL(ret)) {
3589 if (SvGMAGICAL(ret))
3590 sv_unmagic(ret, PERL_MAGIC_qr);
3592 mg = mg_find(ret, PERL_MAGIC_qr);
3596 re = (regexp *)mg->mg_obj;
3597 (void)ReREFCNT_inc(re);
3601 const char * const t = SvPV_const(ret, len);
3603 const I32 osize = PL_regsize;
3606 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3607 re = CALLREGCOMP((char*)t, (char*)t + len, &pm);
3609 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3611 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3618 debug_start_match(re, do_utf8, locinput, PL_regeol,
3619 "Matching embedded");
3621 startpoint = rei->program + 1;
3622 ST.close_paren = 0; /* only used for GOSUB */
3623 /* borrowed from regtry */
3624 if (PL_reg_start_tmpl <= re->nparens) {
3625 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3626 if(PL_reg_start_tmp)
3627 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3629 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3632 eval_recurse_doit: /* Share code with GOSUB below this line */
3633 /* run the pattern returned from (??{...}) */
3634 ST.cp = regcppush(0); /* Save *all* the positions. */
3635 REGCP_SET(ST.lastcp);
3637 PL_regstartp = re->startp; /* essentially NOOP on GOSUB */
3638 PL_regendp = re->endp; /* essentially NOOP on GOSUB */
3640 *PL_reglastparen = 0;
3641 *PL_reglastcloseparen = 0;
3642 PL_reginput = locinput;
3645 /* XXXX This is too dramatic a measure... */
3648 ST.toggle_reg_flags = PL_reg_flags;
3649 if (re->extflags & RXf_UTF8)
3650 PL_reg_flags |= RF_utf8;
3652 PL_reg_flags &= ~RF_utf8;
3653 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3656 ST.prev_curlyx = cur_curlyx;
3661 ST.prev_eval = cur_eval;
3663 /* now continue from first node in postoned RE */
3664 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3667 /* logical is 1, /(?(?{...})X|Y)/ */
3668 sw = (bool)SvTRUE(ret);
3673 case EVAL_AB: /* cleanup after a successful (??{A})B */
3674 /* note: this is called twice; first after popping B, then A */
3675 PL_reg_flags ^= ST.toggle_reg_flags;
3678 rexi = RXi_GET(rex);
3680 cur_eval = ST.prev_eval;
3681 cur_curlyx = ST.prev_curlyx;
3682 /* XXXX This is too dramatic a measure... */
3687 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3688 /* note: this is called twice; first after popping B, then A */
3689 PL_reg_flags ^= ST.toggle_reg_flags;
3692 rexi = RXi_GET(rex);
3693 PL_reginput = locinput;
3694 REGCP_UNWIND(ST.lastcp);
3696 cur_eval = ST.prev_eval;
3697 cur_curlyx = ST.prev_curlyx;
3698 /* XXXX This is too dramatic a measure... */
3704 n = ARG(scan); /* which paren pair */
3705 PL_reg_start_tmp[n] = locinput;
3711 n = ARG(scan); /* which paren pair */
3712 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3713 PL_regendp[n] = locinput - PL_bostr;
3714 /*if (n > PL_regsize)
3716 if (n > *PL_reglastparen)
3717 *PL_reglastparen = n;
3718 *PL_reglastcloseparen = n;
3719 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3727 cursor && OP(cursor)!=END;
3728 cursor=regnext(cursor))
3730 if ( OP(cursor)==CLOSE ){
3732 if ( n <= lastopen ) {
3733 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3734 PL_regendp[n] = locinput - PL_bostr;
3735 /*if (n > PL_regsize)
3737 if (n > *PL_reglastparen)
3738 *PL_reglastparen = n;
3739 *PL_reglastcloseparen = n;
3740 if ( n == ARG(scan) || (cur_eval &&
3741 cur_eval->u.eval.close_paren == n))
3750 n = ARG(scan); /* which paren pair */
3751 sw = (bool)(*PL_reglastparen >= n && PL_regendp[n] != -1);
3754 /* reg_check_named_buff_matched returns 0 for no match */
3755 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3759 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3765 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3767 next = NEXTOPER(NEXTOPER(scan));
3769 next = scan + ARG(scan);
3770 if (OP(next) == IFTHEN) /* Fake one. */
3771 next = NEXTOPER(NEXTOPER(next));
3775 logical = scan->flags;
3778 /*******************************************************************
3780 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3781 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3782 STAR/PLUS/CURLY/CURLYN are used instead.)
3784 A*B is compiled as <CURLYX><A><WHILEM><B>
3786 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3787 state, which contains the current count, initialised to -1. It also sets
3788 cur_curlyx to point to this state, with any previous value saved in the
3791 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3792 since the pattern may possibly match zero times (i.e. it's a while {} loop
3793 rather than a do {} while loop).
3795 Each entry to WHILEM represents a successful match of A. The count in the
3796 CURLYX block is incremented, another WHILEM state is pushed, and execution
3797 passes to A or B depending on greediness and the current count.
3799 For example, if matching against the string a1a2a3b (where the aN are
3800 substrings that match /A/), then the match progresses as follows: (the
3801 pushed states are interspersed with the bits of strings matched so far):
3804 <CURLYX cnt=0><WHILEM>
3805 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3806 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3807 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3808 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3810 (Contrast this with something like CURLYM, which maintains only a single
3814 a1 <CURLYM cnt=1> a2
3815 a1 a2 <CURLYM cnt=2> a3
3816 a1 a2 a3 <CURLYM cnt=3> b
3819 Each WHILEM state block marks a point to backtrack to upon partial failure
3820 of A or B, and also contains some minor state data related to that
3821 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
3822 overall state, such as the count, and pointers to the A and B ops.
3824 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
3825 must always point to the *current* CURLYX block, the rules are:
3827 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
3828 and set cur_curlyx to point the new block.
3830 When popping the CURLYX block after a successful or unsuccessful match,
3831 restore the previous cur_curlyx.
3833 When WHILEM is about to execute B, save the current cur_curlyx, and set it
3834 to the outer one saved in the CURLYX block.
3836 When popping the WHILEM block after a successful or unsuccessful B match,
3837 restore the previous cur_curlyx.
3839 Here's an example for the pattern (AI* BI)*BO
3840 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
3843 curlyx backtrack stack
3844 ------ ---------------
3846 CO <CO prev=NULL> <WO>
3847 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3848 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3849 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
3851 At this point the pattern succeeds, and we work back down the stack to
3852 clean up, restoring as we go:
3854 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
3855 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
3856 CO <CO prev=NULL> <WO>
3859 *******************************************************************/
3861 #define ST st->u.curlyx
3863 case CURLYX: /* start of /A*B/ (for complex A) */
3865 /* No need to save/restore up to this paren */
3866 I32 parenfloor = scan->flags;
3868 assert(next); /* keep Coverity happy */
3869 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3872 /* XXXX Probably it is better to teach regpush to support
3873 parenfloor > PL_regsize... */
3874 if (parenfloor > (I32)*PL_reglastparen)
3875 parenfloor = *PL_reglastparen; /* Pessimization... */
3877 ST.prev_curlyx= cur_curlyx;
3879 ST.cp = PL_savestack_ix;
3881 /* these fields contain the state of the current curly.
3882 * they are accessed by subsequent WHILEMs */
3883 ST.parenfloor = parenfloor;
3884 ST.min = ARG1(scan);
3885 ST.max = ARG2(scan);
3886 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3890 ST.count = -1; /* this will be updated by WHILEM */
3891 ST.lastloc = NULL; /* this will be updated by WHILEM */
3893 PL_reginput = locinput;
3894 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
3898 case CURLYX_end: /* just finished matching all of A*B */
3899 if (PL_reg_eval_set){
3900 SV *pres= GvSV(PL_replgv);
3903 sv_setsv(GvSV(PL_replgv), pres);
3908 cur_curlyx = ST.prev_curlyx;
3912 case CURLYX_end_fail: /* just failed to match all of A*B */
3914 cur_curlyx = ST.prev_curlyx;
3920 #define ST st->u.whilem
3922 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
3924 /* see the discussion above about CURLYX/WHILEM */
3926 assert(cur_curlyx); /* keep Coverity happy */
3927 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
3928 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
3929 ST.cache_offset = 0;
3932 PL_reginput = locinput;
3934 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3935 "%*s whilem: matched %ld out of %ld..%ld\n",
3936 REPORT_CODE_OFF+depth*2, "", (long)n,
3937 (long)cur_curlyx->u.curlyx.min,
3938 (long)cur_curlyx->u.curlyx.max)
3941 /* First just match a string of min A's. */
3943 if (n < cur_curlyx->u.curlyx.min) {
3944 cur_curlyx->u.curlyx.lastloc = locinput;
3945 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
3949 /* If degenerate A matches "", assume A done. */
3951 if (locinput == cur_curlyx->u.curlyx.lastloc) {
3952 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3953 "%*s whilem: empty match detected, trying continuation...\n",
3954 REPORT_CODE_OFF+depth*2, "")
3956 goto do_whilem_B_max;
3959 /* super-linear cache processing */
3963 if (!PL_reg_maxiter) {
3964 /* start the countdown: Postpone detection until we
3965 * know the match is not *that* much linear. */
3966 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3967 /* possible overflow for long strings and many CURLYX's */
3968 if (PL_reg_maxiter < 0)
3969 PL_reg_maxiter = I32_MAX;
3970 PL_reg_leftiter = PL_reg_maxiter;
3973 if (PL_reg_leftiter-- == 0) {
3974 /* initialise cache */
3975 const I32 size = (PL_reg_maxiter + 7)/8;
3976 if (PL_reg_poscache) {
3977 if ((I32)PL_reg_poscache_size < size) {
3978 Renew(PL_reg_poscache, size, char);
3979 PL_reg_poscache_size = size;
3981 Zero(PL_reg_poscache, size, char);
3984 PL_reg_poscache_size = size;
3985 Newxz(PL_reg_poscache, size, char);
3987 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
3988 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
3989 PL_colors[4], PL_colors[5])
3993 if (PL_reg_leftiter < 0) {
3994 /* have we already failed at this position? */
3996 offset = (scan->flags & 0xf) - 1
3997 + (locinput - PL_bostr) * (scan->flags>>4);
3998 mask = 1 << (offset % 8);
4000 if (PL_reg_poscache[offset] & mask) {
4001 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4002 "%*s whilem: (cache) already tried at this position...\n",
4003 REPORT_CODE_OFF+depth*2, "")
4005 sayNO; /* cache records failure */
4007 ST.cache_offset = offset;
4008 ST.cache_mask = mask;
4012 /* Prefer B over A for minimal matching. */
4014 if (cur_curlyx->u.curlyx.minmod) {
4015 ST.save_curlyx = cur_curlyx;
4016 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4017 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4018 REGCP_SET(ST.lastcp);
4019 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4023 /* Prefer A over B for maximal matching. */
4025 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
4026 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4027 cur_curlyx->u.curlyx.lastloc = locinput;
4028 REGCP_SET(ST.lastcp);
4029 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
4032 goto do_whilem_B_max;
4036 case WHILEM_B_min: /* just matched B in a minimal match */
4037 case WHILEM_B_max: /* just matched B in a maximal match */
4038 cur_curlyx = ST.save_curlyx;
4042 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4043 cur_curlyx = ST.save_curlyx;
4044 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4045 cur_curlyx->u.curlyx.count--;
4049 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4050 REGCP_UNWIND(ST.lastcp);
4053 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4054 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4055 cur_curlyx->u.curlyx.count--;
4059 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4060 REGCP_UNWIND(ST.lastcp);
4061 regcppop(rex); /* Restore some previous $<digit>s? */
4062 PL_reginput = locinput;
4063 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4064 "%*s whilem: failed, trying continuation...\n",
4065 REPORT_CODE_OFF+depth*2, "")
4068 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4069 && ckWARN(WARN_REGEXP)
4070 && !(PL_reg_flags & RF_warned))
4072 PL_reg_flags |= RF_warned;
4073 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4074 "Complex regular subexpression recursion",
4079 ST.save_curlyx = cur_curlyx;
4080 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4081 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4084 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4085 cur_curlyx = ST.save_curlyx;
4086 REGCP_UNWIND(ST.lastcp);
4089 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
4090 /* Maximum greed exceeded */
4091 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4092 && ckWARN(WARN_REGEXP)
4093 && !(PL_reg_flags & RF_warned))
4095 PL_reg_flags |= RF_warned;
4096 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4097 "%s limit (%d) exceeded",
4098 "Complex regular subexpression recursion",
4101 cur_curlyx->u.curlyx.count--;
4105 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4106 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4108 /* Try grabbing another A and see if it helps. */
4109 PL_reginput = locinput;
4110 cur_curlyx->u.curlyx.lastloc = locinput;
4111 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4112 REGCP_SET(ST.lastcp);
4113 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
4117 #define ST st->u.branch
4119 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4120 next = scan + ARG(scan);
4123 scan = NEXTOPER(scan);
4126 case BRANCH: /* /(...|A|...)/ */
4127 scan = NEXTOPER(scan); /* scan now points to inner node */
4128 if ((!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
4131 /* last branch; skip state push and jump direct to node */
4134 ST.lastparen = *PL_reglastparen;
4135 ST.next_branch = next;
4137 PL_reginput = locinput;
4139 /* Now go into the branch */
4141 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4143 PUSH_STATE_GOTO(BRANCH_next, scan);
4147 PL_reginput = locinput;
4148 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4149 (SV*)rexi->data->data[ ARG( scan ) ];
4150 PUSH_STATE_GOTO(CUTGROUP_next,next);
4152 case CUTGROUP_next_fail:
4155 if (st->u.mark.mark_name)
4156 sv_commit = st->u.mark.mark_name;
4162 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4167 REGCP_UNWIND(ST.cp);
4168 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4170 *PL_reglastparen = n;
4171 /*dmq: *PL_reglastcloseparen = n; */
4172 scan = ST.next_branch;
4173 /* no more branches? */
4174 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4176 PerlIO_printf( Perl_debug_log,
4177 "%*s %sBRANCH failed...%s\n",
4178 REPORT_CODE_OFF+depth*2, "",
4184 continue; /* execute next BRANCH[J] op */
4192 #define ST st->u.curlym
4194 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4196 /* This is an optimisation of CURLYX that enables us to push
4197 * only a single backtracking state, no matter now many matches
4198 * there are in {m,n}. It relies on the pattern being constant
4199 * length, with no parens to influence future backrefs
4203 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4205 /* if paren positive, emulate an OPEN/CLOSE around A */
4207 U32 paren = ST.me->flags;
4208 if (paren > PL_regsize)
4210 if (paren > *PL_reglastparen)
4211 *PL_reglastparen = paren;
4212 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4220 ST.c1 = CHRTEST_UNINIT;
4223 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4226 curlym_do_A: /* execute the A in /A{m,n}B/ */
4227 PL_reginput = locinput;
4228 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4231 case CURLYM_A: /* we've just matched an A */
4232 locinput = st->locinput;
4233 nextchr = UCHARAT(locinput);
4236 /* after first match, determine A's length: u.curlym.alen */
4237 if (ST.count == 1) {
4238 if (PL_reg_match_utf8) {
4240 while (s < PL_reginput) {
4246 ST.alen = PL_reginput - locinput;
4249 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4252 PerlIO_printf(Perl_debug_log,
4253 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4254 (int)(REPORT_CODE_OFF+(depth*2)), "",
4255 (IV) ST.count, (IV)ST.alen)
4258 locinput = PL_reginput;
4260 if (cur_eval && cur_eval->u.eval.close_paren &&
4261 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4264 if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
4265 goto curlym_do_A; /* try to match another A */
4266 goto curlym_do_B; /* try to match B */
4268 case CURLYM_A_fail: /* just failed to match an A */
4269 REGCP_UNWIND(ST.cp);
4271 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4272 || (cur_eval && cur_eval->u.eval.close_paren &&
4273 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4276 curlym_do_B: /* execute the B in /A{m,n}B/ */
4277 PL_reginput = locinput;
4278 if (ST.c1 == CHRTEST_UNINIT) {
4279 /* calculate c1 and c2 for possible match of 1st char
4280 * following curly */
4281 ST.c1 = ST.c2 = CHRTEST_VOID;
4282 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4283 regnode *text_node = ST.B;
4284 if (! HAS_TEXT(text_node))
4285 FIND_NEXT_IMPT(text_node);
4286 if (HAS_TEXT(text_node)
4287 && PL_regkind[OP(text_node)] != REF)
4289 ST.c1 = (U8)*STRING(text_node);
4291 (OP(text_node) == EXACTF || OP(text_node) == REFF)
4293 : (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4294 ? PL_fold_locale[ST.c1]
4301 PerlIO_printf(Perl_debug_log,
4302 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4303 (int)(REPORT_CODE_OFF+(depth*2)),
4306 if (ST.c1 != CHRTEST_VOID
4307 && UCHARAT(PL_reginput) != ST.c1
4308 && UCHARAT(PL_reginput) != ST.c2)
4310 /* simulate B failing */
4311 state_num = CURLYM_B_fail;
4312 goto reenter_switch;
4316 /* mark current A as captured */
4317 I32 paren = ST.me->flags;
4320 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4321 PL_regendp[paren] = PL_reginput - PL_bostr;
4322 /*dmq: *PL_reglastcloseparen = paren; */
4325 PL_regendp[paren] = -1;
4326 if (cur_eval && cur_eval->u.eval.close_paren &&
4327 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4336 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4339 case CURLYM_B_fail: /* just failed to match a B */
4340 REGCP_UNWIND(ST.cp);
4342 if (ST.count == ARG2(ST.me) /* max */)
4344 goto curlym_do_A; /* try to match a further A */
4346 /* backtrack one A */
4347 if (ST.count == ARG1(ST.me) /* min */)
4350 locinput = HOPc(locinput, -ST.alen);
4351 goto curlym_do_B; /* try to match B */
4354 #define ST st->u.curly
4356 #define CURLY_SETPAREN(paren, success) \
4359 PL_regstartp[paren] = HOPc(locinput, -1) - PL_bostr; \
4360 PL_regendp[paren] = locinput - PL_bostr; \
4361 *PL_reglastcloseparen = paren; \
4364 PL_regendp[paren] = -1; \
4367 case STAR: /* /A*B/ where A is width 1 */
4371 scan = NEXTOPER(scan);
4373 case PLUS: /* /A+B/ where A is width 1 */
4377 scan = NEXTOPER(scan);
4379 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4380 ST.paren = scan->flags; /* Which paren to set */
4381 if (ST.paren > PL_regsize)
4382 PL_regsize = ST.paren;
4383 if (ST.paren > *PL_reglastparen)
4384 *PL_reglastparen = ST.paren;
4385 ST.min = ARG1(scan); /* min to match */
4386 ST.max = ARG2(scan); /* max to match */
4387 if (cur_eval && cur_eval->u.eval.close_paren &&
4388 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4392 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4394 case CURLY: /* /A{m,n}B/ where A is width 1 */
4396 ST.min = ARG1(scan); /* min to match */
4397 ST.max = ARG2(scan); /* max to match */
4398 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4401 * Lookahead to avoid useless match attempts
4402 * when we know what character comes next.
4404 * Used to only do .*x and .*?x, but now it allows
4405 * for )'s, ('s and (?{ ... })'s to be in the way
4406 * of the quantifier and the EXACT-like node. -- japhy
4409 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4411 if (HAS_TEXT(next) || JUMPABLE(next)) {
4413 regnode *text_node = next;
4415 if (! HAS_TEXT(text_node))
4416 FIND_NEXT_IMPT(text_node);
4418 if (! HAS_TEXT(text_node))
4419 ST.c1 = ST.c2 = CHRTEST_VOID;
4421 if (PL_regkind[OP(text_node)] == REF) {
4422 ST.c1 = ST.c2 = CHRTEST_VOID;
4423 goto assume_ok_easy;
4426 s = (U8*)STRING(text_node);
4430 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
4431 ST.c2 = PL_fold[ST.c1];
4432 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4433 ST.c2 = PL_fold_locale[ST.c1];
4436 if (OP(text_node) == EXACTF || OP(text_node) == REFF) {
4437 STRLEN ulen1, ulen2;
4438 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4439 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4441 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4442 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4444 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4446 0 : UTF8_ALLOW_ANY);
4447 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4449 0 : UTF8_ALLOW_ANY);
4451 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4453 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4458 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4465 ST.c1 = ST.c2 = CHRTEST_VOID;
4470 PL_reginput = locinput;
4473 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4476 locinput = PL_reginput;
4478 if (ST.c1 == CHRTEST_VOID)
4479 goto curly_try_B_min;
4481 ST.oldloc = locinput;
4483 /* set ST.maxpos to the furthest point along the
4484 * string that could possibly match */
4485 if (ST.max == REG_INFTY) {
4486 ST.maxpos = PL_regeol - 1;
4488 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4492 int m = ST.max - ST.min;
4493 for (ST.maxpos = locinput;
4494 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4495 ST.maxpos += UTF8SKIP(ST.maxpos);
4498 ST.maxpos = locinput + ST.max - ST.min;
4499 if (ST.maxpos >= PL_regeol)
4500 ST.maxpos = PL_regeol - 1;
4502 goto curly_try_B_min_known;
4506 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4507 locinput = PL_reginput;
4508 if (ST.count < ST.min)
4510 if ((ST.count > ST.min)
4511 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4513 /* A{m,n} must come at the end of the string, there's
4514 * no point in backing off ... */
4516 /* ...except that $ and \Z can match before *and* after
4517 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4518 We may back off by one in this case. */
4519 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4523 goto curly_try_B_max;
4528 case CURLY_B_min_known_fail:
4529 /* failed to find B in a non-greedy match where c1,c2 valid */
4530 if (ST.paren && ST.count)
4531 PL_regendp[ST.paren] = -1;
4533 PL_reginput = locinput; /* Could be reset... */
4534 REGCP_UNWIND(ST.cp);
4535 /* Couldn't or didn't -- move forward. */
4536 ST.oldloc = locinput;
4538 locinput += UTF8SKIP(locinput);
4542 curly_try_B_min_known:
4543 /* find the next place where 'B' could work, then call B */
4547 n = (ST.oldloc == locinput) ? 0 : 1;
4548 if (ST.c1 == ST.c2) {
4550 /* set n to utf8_distance(oldloc, locinput) */
4551 while (locinput <= ST.maxpos &&
4552 utf8n_to_uvchr((U8*)locinput,
4553 UTF8_MAXBYTES, &len,
4554 uniflags) != (UV)ST.c1) {
4560 /* set n to utf8_distance(oldloc, locinput) */
4561 while (locinput <= ST.maxpos) {
4563 const UV c = utf8n_to_uvchr((U8*)locinput,
4564 UTF8_MAXBYTES, &len,
4566 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4574 if (ST.c1 == ST.c2) {
4575 while (locinput <= ST.maxpos &&
4576 UCHARAT(locinput) != ST.c1)
4580 while (locinput <= ST.maxpos
4581 && UCHARAT(locinput) != ST.c1
4582 && UCHARAT(locinput) != ST.c2)
4585 n = locinput - ST.oldloc;
4587 if (locinput > ST.maxpos)
4589 /* PL_reginput == oldloc now */
4592 if (regrepeat(rex, ST.A, n, depth) < n)
4595 PL_reginput = locinput;
4596 CURLY_SETPAREN(ST.paren, ST.count);
4597 if (cur_eval && cur_eval->u.eval.close_paren &&
4598 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4601 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4606 case CURLY_B_min_fail:
4607 /* failed to find B in a non-greedy match where c1,c2 invalid */
4608 if (ST.paren && ST.count)
4609 PL_regendp[ST.paren] = -1;
4611 REGCP_UNWIND(ST.cp);
4612 /* failed -- move forward one */
4613 PL_reginput = locinput;
4614 if (regrepeat(rex, ST.A, 1, depth)) {
4616 locinput = PL_reginput;
4617 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4618 ST.count > 0)) /* count overflow ? */
4621 CURLY_SETPAREN(ST.paren, ST.count);
4622 if (cur_eval && cur_eval->u.eval.close_paren &&
4623 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4626 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4634 /* a successful greedy match: now try to match B */
4635 if (cur_eval && cur_eval->u.eval.close_paren &&
4636 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4641 if (ST.c1 != CHRTEST_VOID)
4642 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4643 UTF8_MAXBYTES, 0, uniflags)
4644 : (UV) UCHARAT(PL_reginput);
4645 /* If it could work, try it. */
4646 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4647 CURLY_SETPAREN(ST.paren, ST.count);
4648 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4653 case CURLY_B_max_fail:
4654 /* failed to find B in a greedy match */
4655 if (ST.paren && ST.count)
4656 PL_regendp[ST.paren] = -1;
4658 REGCP_UNWIND(ST.cp);
4660 if (--ST.count < ST.min)
4662 PL_reginput = locinput = HOPc(locinput, -1);
4663 goto curly_try_B_max;
4670 /* we've just finished A in /(??{A})B/; now continue with B */
4674 st->u.eval.toggle_reg_flags
4675 = cur_eval->u.eval.toggle_reg_flags;
4676 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4678 st->u.eval.prev_rex = rex; /* inner */
4679 rex = cur_eval->u.eval.prev_rex; /* outer */
4680 rexi = RXi_GET(rex);
4681 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4683 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4684 REGCP_SET(st->u.eval.lastcp);
4685 PL_reginput = locinput;
4687 /* Restore parens of the outer rex without popping the
4689 tmpix = PL_savestack_ix;
4690 PL_savestack_ix = cur_eval->u.eval.lastcp;
4692 PL_savestack_ix = tmpix;
4694 st->u.eval.prev_eval = cur_eval;
4695 cur_eval = cur_eval->u.eval.prev_eval;
4697 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
4698 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
4699 PUSH_YES_STATE_GOTO(EVAL_AB,
4700 st->u.eval.prev_eval->u.eval.B); /* match B */
4703 if (locinput < reginfo->till) {
4704 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4705 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4707 (long)(locinput - PL_reg_starttry),
4708 (long)(reginfo->till - PL_reg_starttry),
4711 sayNO_SILENT; /* Cannot match: too short. */
4713 PL_reginput = locinput; /* put where regtry can find it */
4714 sayYES; /* Success! */
4716 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4718 PerlIO_printf(Perl_debug_log,
4719 "%*s %ssubpattern success...%s\n",
4720 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4721 PL_reginput = locinput; /* put where regtry can find it */
4722 sayYES; /* Success! */
4725 #define ST st->u.ifmatch
4727 case SUSPEND: /* (?>A) */
4729 PL_reginput = locinput;
4732 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4734 goto ifmatch_trivial_fail_test;
4736 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4738 ifmatch_trivial_fail_test:
4740 char * const s = HOPBACKc(locinput, scan->flags);
4745 sw = 1 - (bool)ST.wanted;
4749 next = scan + ARG(scan);
4757 PL_reginput = locinput;
4761 ST.logical = logical;
4762 /* execute body of (?...A) */
4763 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4766 case IFMATCH_A_fail: /* body of (?...A) failed */
4767 ST.wanted = !ST.wanted;
4770 case IFMATCH_A: /* body of (?...A) succeeded */
4772 sw = (bool)ST.wanted;
4774 else if (!ST.wanted)
4777 if (OP(ST.me) == SUSPEND)
4778 locinput = PL_reginput;
4780 locinput = PL_reginput = st->locinput;
4781 nextchr = UCHARAT(locinput);
4783 scan = ST.me + ARG(ST.me);
4786 continue; /* execute B */
4791 next = scan + ARG(scan);
4796 reginfo->cutpoint = PL_regeol;
4799 PL_reginput = locinput;
4801 sv_yes_mark = sv_commit = (SV*)rexi->data->data[ ARG( scan ) ];
4802 PUSH_STATE_GOTO(COMMIT_next,next);
4804 case COMMIT_next_fail:
4811 #define ST st->u.mark
4813 ST.prev_mark = mark_state;
4814 ST.mark_name = sv_commit = sv_yes_mark
4815 = (SV*)rexi->data->data[ ARG( scan ) ];
4817 ST.mark_loc = PL_reginput = locinput;
4818 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
4820 case MARKPOINT_next:
4821 mark_state = ST.prev_mark;
4824 case MARKPOINT_next_fail:
4825 if (popmark && sv_eq(ST.mark_name,popmark))
4827 if (ST.mark_loc > startpoint)
4828 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
4829 popmark = NULL; /* we found our mark */
4830 sv_commit = ST.mark_name;
4833 PerlIO_printf(Perl_debug_log,
4834 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
4835 REPORT_CODE_OFF+depth*2, "",
4836 PL_colors[4], sv_commit, PL_colors[5]);
4839 mark_state = ST.prev_mark;
4840 sv_yes_mark = mark_state ?
4841 mark_state->u.mark.mark_name : NULL;
4845 PL_reginput = locinput;
4847 /* (*CUT) : if we fail we cut here*/
4848 ST.mark_name = NULL;
4849 ST.mark_loc = locinput;
4850 PUSH_STATE_GOTO(SKIP_next,next);
4852 /* (*CUT:NAME) : if there is a (*MARK:NAME) fail where it was,
4853 otherwise do nothing. Meaning we need to scan
4855 regmatch_state *cur = mark_state;
4856 SV *find = (SV*)rexi->data->data[ ARG( scan ) ];
4859 if ( sv_eq( cur->u.mark.mark_name,
4862 ST.mark_name = find;
4863 PUSH_STATE_GOTO( SKIP_next, next );
4865 cur = cur->u.mark.prev_mark;
4868 /* Didn't find our (*MARK:NAME) so ignore this (*CUT:NAME) */
4870 case SKIP_next_fail:
4872 /* (*CUT:NAME) - Set up to search for the name as we
4873 collapse the stack*/
4874 popmark = ST.mark_name;
4876 /* (*CUT) - No name, we cut here.*/
4877 if (ST.mark_loc > startpoint)
4878 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
4879 /* but we set sv_commit to latest mark_name if there
4880 is one so they can test to see how things lead to this
4883 sv_commit=mark_state->u.mark.mark_name;
4891 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
4892 PTR2UV(scan), OP(scan));
4893 Perl_croak(aTHX_ "regexp memory corruption");
4897 /* switch break jumps here */
4898 scan = next; /* prepare to execute the next op and ... */
4899 continue; /* ... jump back to the top, reusing st */
4903 /* push a state that backtracks on success */
4904 st->u.yes.prev_yes_state = yes_state;
4908 /* push a new regex state, then continue at scan */
4910 regmatch_state *newst;
4913 regmatch_state *cur = st;
4914 regmatch_state *curyes = yes_state;
4916 regmatch_slab *slab = PL_regmatch_slab;
4917 for (;curd > -1;cur--,curd--) {
4918 if (cur < SLAB_FIRST(slab)) {
4920 cur = SLAB_LAST(slab);
4922 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
4923 REPORT_CODE_OFF + 2 + depth * 2,"",
4924 curd, reg_name[cur->resume_state],
4925 (curyes == cur) ? "yes" : ""
4928 curyes = cur->u.yes.prev_yes_state;
4931 DEBUG_STATE_pp("push")
4934 st->locinput = locinput;
4936 if (newst > SLAB_LAST(PL_regmatch_slab))
4937 newst = S_push_slab(aTHX);
4938 PL_regmatch_state = newst;
4940 locinput = PL_reginput;
4941 nextchr = UCHARAT(locinput);
4949 * We get here only if there's trouble -- normally "case END" is
4950 * the terminating point.
4952 Perl_croak(aTHX_ "corrupted regexp pointers");
4958 /* we have successfully completed a subexpression, but we must now
4959 * pop to the state marked by yes_state and continue from there */
4960 assert(st != yes_state);
4962 while (st != yes_state) {
4964 if (st < SLAB_FIRST(PL_regmatch_slab)) {
4965 PL_regmatch_slab = PL_regmatch_slab->prev;
4966 st = SLAB_LAST(PL_regmatch_slab);
4970 DEBUG_STATE_pp("pop (no final)");
4972 DEBUG_STATE_pp("pop (yes)");
4978 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
4979 || yes_state > SLAB_LAST(PL_regmatch_slab))
4981 /* not in this slab, pop slab */
4982 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
4983 PL_regmatch_slab = PL_regmatch_slab->prev;
4984 st = SLAB_LAST(PL_regmatch_slab);
4986 depth -= (st - yes_state);
4989 yes_state = st->u.yes.prev_yes_state;
4990 PL_regmatch_state = st;
4993 locinput= st->locinput;
4994 nextchr = UCHARAT(locinput);
4996 state_num = st->resume_state + no_final;
4997 goto reenter_switch;
5000 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5001 PL_colors[4], PL_colors[5]));
5008 PerlIO_printf(Perl_debug_log,
5009 "%*s %sfailed...%s\n",
5010 REPORT_CODE_OFF+depth*2, "",
5011 PL_colors[4], PL_colors[5])
5023 /* there's a previous state to backtrack to */
5025 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5026 PL_regmatch_slab = PL_regmatch_slab->prev;
5027 st = SLAB_LAST(PL_regmatch_slab);
5029 PL_regmatch_state = st;
5030 locinput= st->locinput;
5031 nextchr = UCHARAT(locinput);
5033 DEBUG_STATE_pp("pop");
5035 if (yes_state == st)
5036 yes_state = st->u.yes.prev_yes_state;
5038 state_num = st->resume_state + 1; /* failure = success + 1 */
5039 goto reenter_switch;
5044 if (rex->intflags & PREGf_VERBARG_SEEN) {
5045 SV *sv_err = get_sv("REGERROR", 1);
5046 SV *sv_mrk = get_sv("REGMARK", 1);
5048 sv_commit = &PL_sv_no;
5050 sv_yes_mark = &PL_sv_yes;
5053 sv_commit = &PL_sv_yes;
5054 sv_yes_mark = &PL_sv_no;
5056 sv_setsv(sv_err, sv_commit);
5057 sv_setsv(sv_mrk, sv_yes_mark);
5059 /* restore original high-water mark */
5060 PL_regmatch_slab = orig_slab;
5061 PL_regmatch_state = orig_state;
5063 /* free all slabs above current one */
5064 if (orig_slab->next) {
5065 regmatch_slab *sl = orig_slab->next;
5066 orig_slab->next = NULL;
5068 regmatch_slab * const osl = sl;
5078 - regrepeat - repeatedly match something simple, report how many
5081 * [This routine now assumes that it will only match on things of length 1.
5082 * That was true before, but now we assume scan - reginput is the count,
5083 * rather than incrementing count on every character. [Er, except utf8.]]
5086 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5089 register char *scan;
5091 register char *loceol = PL_regeol;
5092 register I32 hardcount = 0;
5093 register bool do_utf8 = PL_reg_match_utf8;
5096 if (max == REG_INFTY)
5098 else if (max < loceol - scan)
5099 loceol = scan + max;
5104 while (scan < loceol && hardcount < max && *scan != '\n') {
5105 scan += UTF8SKIP(scan);
5109 while (scan < loceol && *scan != '\n')
5116 while (scan < loceol && hardcount < max) {
5117 scan += UTF8SKIP(scan);
5127 case EXACT: /* length of string is 1 */
5129 while (scan < loceol && UCHARAT(scan) == c)
5132 case EXACTF: /* length of string is 1 */
5134 while (scan < loceol &&
5135 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5138 case EXACTFL: /* length of string is 1 */
5139 PL_reg_flags |= RF_tainted;
5141 while (scan < loceol &&
5142 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5148 while (hardcount < max && scan < loceol &&
5149 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
5150 scan += UTF8SKIP(scan);
5154 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5161 LOAD_UTF8_CHARCLASS_ALNUM();
5162 while (hardcount < max && scan < loceol &&
5163 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5164 scan += UTF8SKIP(scan);
5168 while (scan < loceol && isALNUM(*scan))
5173 PL_reg_flags |= RF_tainted;
5176 while (hardcount < max && scan < loceol &&
5177 isALNUM_LC_utf8((U8*)scan)) {
5178 scan += UTF8SKIP(scan);
5182 while (scan < loceol && isALNUM_LC(*scan))
5189 LOAD_UTF8_CHARCLASS_ALNUM();
5190 while (hardcount < max && scan < loceol &&
5191 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5192 scan += UTF8SKIP(scan);
5196 while (scan < loceol && !isALNUM(*scan))
5201 PL_reg_flags |= RF_tainted;
5204 while (hardcount < max && scan < loceol &&
5205 !isALNUM_LC_utf8((U8*)scan)) {
5206 scan += UTF8SKIP(scan);
5210 while (scan < loceol && !isALNUM_LC(*scan))
5217 LOAD_UTF8_CHARCLASS_SPACE();
5218 while (hardcount < max && scan < loceol &&
5220 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5221 scan += UTF8SKIP(scan);
5225 while (scan < loceol && isSPACE(*scan))
5230 PL_reg_flags |= RF_tainted;
5233 while (hardcount < max && scan < loceol &&
5234 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5235 scan += UTF8SKIP(scan);
5239 while (scan < loceol && isSPACE_LC(*scan))
5246 LOAD_UTF8_CHARCLASS_SPACE();
5247 while (hardcount < max && scan < loceol &&
5249 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5250 scan += UTF8SKIP(scan);
5254 while (scan < loceol && !isSPACE(*scan))
5259 PL_reg_flags |= RF_tainted;
5262 while (hardcount < max && scan < loceol &&
5263 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5264 scan += UTF8SKIP(scan);
5268 while (scan < loceol && !isSPACE_LC(*scan))
5275 LOAD_UTF8_CHARCLASS_DIGIT();
5276 while (hardcount < max && scan < loceol &&
5277 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5278 scan += UTF8SKIP(scan);
5282 while (scan < loceol && isDIGIT(*scan))
5289 LOAD_UTF8_CHARCLASS_DIGIT();
5290 while (hardcount < max && scan < loceol &&
5291 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5292 scan += UTF8SKIP(scan);
5296 while (scan < loceol && !isDIGIT(*scan))
5300 default: /* Called on something of 0 width. */
5301 break; /* So match right here or not at all. */
5307 c = scan - PL_reginput;
5311 GET_RE_DEBUG_FLAGS_DECL;
5313 SV * const prop = sv_newmortal();
5314 regprop(prog, prop, p);
5315 PerlIO_printf(Perl_debug_log,
5316 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5317 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5325 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5327 - regclass_swash - prepare the utf8 swash
5331 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5337 RXi_GET_DECL(prog,progi);
5338 const struct reg_data * const data = prog ? progi->data : NULL;
5340 if (data && data->count) {
5341 const U32 n = ARG(node);
5343 if (data->what[n] == 's') {
5344 SV * const rv = (SV*)data->data[n];
5345 AV * const av = (AV*)SvRV((SV*)rv);
5346 SV **const ary = AvARRAY(av);
5349 /* See the end of regcomp.c:S_regclass() for
5350 * documentation of these array elements. */
5353 a = SvROK(ary[1]) ? &ary[1] : 0;
5354 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
5358 else if (si && doinit) {
5359 sw = swash_init("utf8", "", si, 1, 0);
5360 (void)av_store(av, 1, sw);
5377 - reginclass - determine if a character falls into a character class
5379 The n is the ANYOF regnode, the p is the target string, lenp
5380 is pointer to the maximum length of how far to go in the p
5381 (if the lenp is zero, UTF8SKIP(p) is used),
5382 do_utf8 tells whether the target string is in UTF-8.
5387 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5390 const char flags = ANYOF_FLAGS(n);
5396 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5397 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5398 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5399 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5400 if (len == (STRLEN)-1)
5401 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5404 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5405 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5408 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5409 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5412 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5416 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5419 if (swash_fetch(sw, p, do_utf8))
5421 else if (flags & ANYOF_FOLD) {
5422 if (!match && lenp && av) {
5424 for (i = 0; i <= av_len(av); i++) {
5425 SV* const sv = *av_fetch(av, i, FALSE);
5427 const char * const s = SvPV_const(sv, len);
5429 if (len <= plen && memEQ(s, (char*)p, len)) {
5437 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5440 to_utf8_fold(p, tmpbuf, &tmplen);
5441 if (swash_fetch(sw, tmpbuf, do_utf8))
5447 if (match && lenp && *lenp == 0)
5448 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5450 if (!match && c < 256) {
5451 if (ANYOF_BITMAP_TEST(n, c))
5453 else if (flags & ANYOF_FOLD) {
5456 if (flags & ANYOF_LOCALE) {
5457 PL_reg_flags |= RF_tainted;
5458 f = PL_fold_locale[c];
5462 if (f != c && ANYOF_BITMAP_TEST(n, f))
5466 if (!match && (flags & ANYOF_CLASS)) {
5467 PL_reg_flags |= RF_tainted;
5469 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5470 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5471 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5472 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5473 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5474 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5475 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5476 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5477 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5478 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5479 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5480 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5481 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5482 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5483 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5484 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5485 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5486 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5487 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5488 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5489 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5490 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5491 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5492 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5493 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5494 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5495 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5496 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5497 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5498 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5499 ) /* How's that for a conditional? */
5506 return (flags & ANYOF_INVERT) ? !match : match;
5510 S_reghop3(U8 *s, I32 off, const U8* lim)
5514 while (off-- && s < lim) {
5515 /* XXX could check well-formedness here */
5520 while (off++ && s > lim) {
5522 if (UTF8_IS_CONTINUED(*s)) {
5523 while (s > lim && UTF8_IS_CONTINUATION(*s))
5526 /* XXX could check well-formedness here */
5533 /* there are a bunch of places where we use two reghop3's that should
5534 be replaced with this routine. but since thats not done yet
5535 we ifdef it out - dmq
5538 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5542 while (off-- && s < rlim) {
5543 /* XXX could check well-formedness here */
5548 while (off++ && s > llim) {
5550 if (UTF8_IS_CONTINUED(*s)) {
5551 while (s > llim && UTF8_IS_CONTINUATION(*s))
5554 /* XXX could check well-formedness here */
5562 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5566 while (off-- && s < lim) {
5567 /* XXX could check well-formedness here */
5574 while (off++ && s > lim) {
5576 if (UTF8_IS_CONTINUED(*s)) {
5577 while (s > lim && UTF8_IS_CONTINUATION(*s))
5580 /* XXX could check well-formedness here */
5589 restore_pos(pTHX_ void *arg)
5592 regexp * const rex = (regexp *)arg;
5593 if (PL_reg_eval_set) {
5594 if (PL_reg_oldsaved) {
5595 rex->subbeg = PL_reg_oldsaved;
5596 rex->sublen = PL_reg_oldsavedlen;
5597 #ifdef PERL_OLD_COPY_ON_WRITE
5598 rex->saved_copy = PL_nrs;
5600 RX_MATCH_COPIED_on(rex);
5602 PL_reg_magic->mg_len = PL_reg_oldpos;
5603 PL_reg_eval_set = 0;
5604 PL_curpm = PL_reg_oldcurpm;
5609 S_to_utf8_substr(pTHX_ register regexp *prog)
5611 if (prog->float_substr && !prog->float_utf8) {
5612 SV* const sv = newSVsv(prog->float_substr);
5613 prog->float_utf8 = sv;
5614 sv_utf8_upgrade(sv);
5615 if (SvTAIL(prog->float_substr))
5617 if (prog->float_substr == prog->check_substr)
5618 prog->check_utf8 = sv;
5620 if (prog->anchored_substr && !prog->anchored_utf8) {
5621 SV* const sv = newSVsv(prog->anchored_substr);
5622 prog->anchored_utf8 = sv;
5623 sv_utf8_upgrade(sv);
5624 if (SvTAIL(prog->anchored_substr))
5626 if (prog->anchored_substr == prog->check_substr)
5627 prog->check_utf8 = sv;
5632 S_to_byte_substr(pTHX_ register regexp *prog)
5635 if (prog->float_utf8 && !prog->float_substr) {
5636 SV* sv = newSVsv(prog->float_utf8);
5637 prog->float_substr = sv;
5638 if (sv_utf8_downgrade(sv, TRUE)) {
5639 if (SvTAIL(prog->float_utf8))
5643 prog->float_substr = sv = &PL_sv_undef;
5645 if (prog->float_utf8 == prog->check_utf8)
5646 prog->check_substr = sv;
5648 if (prog->anchored_utf8 && !prog->anchored_substr) {
5649 SV* sv = newSVsv(prog->anchored_utf8);
5650 prog->anchored_substr = sv;
5651 if (sv_utf8_downgrade(sv, TRUE)) {
5652 if (SvTAIL(prog->anchored_utf8))
5656 prog->anchored_substr = sv = &PL_sv_undef;
5658 if (prog->anchored_utf8 == prog->check_utf8)
5659 prog->check_substr = sv;
5665 * c-indentation-style: bsd
5667 * indent-tabs-mode: t
5670 * ex: set ts=8 sts=4 sw=4 noet: