5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
78 #ifdef PERL_IN_XSUB_RE
84 #define RF_tainted 1 /* tainted information used? */
85 #define RF_warned 2 /* warned about big count? */
87 #define RF_utf8 8 /* Pattern contains multibyte chars? */
89 #define UTF ((PL_reg_flags & RF_utf8) != 0)
91 #define RS_init 1 /* eval environment created */
92 #define RS_set 2 /* replsv value is set */
98 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
104 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
105 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
107 #define HOPc(pos,off) \
108 (char *)(PL_reg_match_utf8 \
109 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
111 #define HOPBACKc(pos, off) \
112 (char*)(PL_reg_match_utf8\
113 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
114 : (pos - off >= PL_bostr) \
118 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
119 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
121 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
122 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
123 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
124 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
125 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
126 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
128 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
130 /* for use after a quantifier and before an EXACT-like node -- japhy */
131 /* it would be nice to rework regcomp.sym to generate this stuff. sigh */
132 #define JUMPABLE(rn) ( \
134 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
136 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
137 OP(rn) == PLUS || OP(rn) == MINMOD || \
138 OP(rn) == KEEPS || (PL_regkind[OP(rn)] == VERB) || \
139 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
141 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
143 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
146 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
147 we don't need this definition. */
148 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
149 #define IS_TEXTF(rn) ( OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
150 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
153 /* ... so we use this as its faster. */
154 #define IS_TEXT(rn) ( OP(rn)==EXACT )
155 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
156 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
161 Search for mandatory following text node; for lookahead, the text must
162 follow but for lookbehind (rn->flags != 0) we skip to the next step.
164 #define FIND_NEXT_IMPT(rn) STMT_START { \
165 while (JUMPABLE(rn)) { \
166 const OPCODE type = OP(rn); \
167 if (type == SUSPEND || PL_regkind[type] == CURLY) \
168 rn = NEXTOPER(NEXTOPER(rn)); \
169 else if (type == PLUS) \
171 else if (type == IFMATCH) \
172 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
173 else rn += NEXT_OFF(rn); \
178 static void restore_pos(pTHX_ void *arg);
181 S_regcppush(pTHX_ I32 parenfloor)
184 const int retval = PL_savestack_ix;
185 #define REGCP_PAREN_ELEMS 4
186 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
188 GET_RE_DEBUG_FLAGS_DECL;
190 if (paren_elems_to_push < 0)
191 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
193 #define REGCP_OTHER_ELEMS 7
194 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
196 for (p = PL_regsize; p > parenfloor; p--) {
197 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
198 SSPUSHINT(PL_regoffs[p].end);
199 SSPUSHINT(PL_regoffs[p].start);
200 SSPUSHPTR(PL_reg_start_tmp[p]);
202 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
203 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
204 (UV)p, (IV)PL_regoffs[p].start,
205 (IV)(PL_reg_start_tmp[p] - PL_bostr),
206 (IV)PL_regoffs[p].end
209 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
210 SSPUSHPTR(PL_regoffs);
211 SSPUSHINT(PL_regsize);
212 SSPUSHINT(*PL_reglastparen);
213 SSPUSHINT(*PL_reglastcloseparen);
214 SSPUSHPTR(PL_reginput);
215 #define REGCP_FRAME_ELEMS 2
216 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
217 * are needed for the regexp context stack bookkeeping. */
218 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
219 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
224 /* These are needed since we do not localize EVAL nodes: */
225 #define REGCP_SET(cp) \
227 PerlIO_printf(Perl_debug_log, \
228 " Setting an EVAL scope, savestack=%"IVdf"\n", \
229 (IV)PL_savestack_ix)); \
232 #define REGCP_UNWIND(cp) \
234 if (cp != PL_savestack_ix) \
235 PerlIO_printf(Perl_debug_log, \
236 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
237 (IV)(cp), (IV)PL_savestack_ix)); \
241 S_regcppop(pTHX_ const regexp *rex)
246 GET_RE_DEBUG_FLAGS_DECL;
248 PERL_ARGS_ASSERT_REGCPPOP;
250 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
252 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
253 i = SSPOPINT; /* Parentheses elements to pop. */
254 input = (char *) SSPOPPTR;
255 *PL_reglastcloseparen = SSPOPINT;
256 *PL_reglastparen = SSPOPINT;
257 PL_regsize = SSPOPINT;
258 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
261 /* Now restore the parentheses context. */
262 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
263 i > 0; i -= REGCP_PAREN_ELEMS) {
265 U32 paren = (U32)SSPOPINT;
266 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
267 PL_regoffs[paren].start = SSPOPINT;
269 if (paren <= *PL_reglastparen)
270 PL_regoffs[paren].end = tmps;
272 PerlIO_printf(Perl_debug_log,
273 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
274 (UV)paren, (IV)PL_regoffs[paren].start,
275 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
276 (IV)PL_regoffs[paren].end,
277 (paren > *PL_reglastparen ? "(no)" : ""));
281 if (*PL_reglastparen + 1 <= rex->nparens) {
282 PerlIO_printf(Perl_debug_log,
283 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
284 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
288 /* It would seem that the similar code in regtry()
289 * already takes care of this, and in fact it is in
290 * a better location to since this code can #if 0-ed out
291 * but the code in regtry() is needed or otherwise tests
292 * requiring null fields (pat.t#187 and split.t#{13,14}
293 * (as of patchlevel 7877) will fail. Then again,
294 * this code seems to be necessary or otherwise
295 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
296 * --jhi updated by dapm */
297 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
299 PL_regoffs[i].start = -1;
300 PL_regoffs[i].end = -1;
306 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
309 * pregexec and friends
312 #ifndef PERL_IN_XSUB_RE
314 - pregexec - match a regexp against a string
317 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
318 char *strbeg, I32 minend, SV *screamer, U32 nosave)
319 /* strend: pointer to null at end of string */
320 /* strbeg: real beginning of string */
321 /* minend: end of match must be >=minend after stringarg. */
322 /* nosave: For optimizations. */
324 PERL_ARGS_ASSERT_PREGEXEC;
327 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
328 nosave ? 0 : REXEC_COPY_STR);
333 * Need to implement the following flags for reg_anch:
335 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
337 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
338 * INTUIT_AUTORITATIVE_ML
339 * INTUIT_ONCE_NOML - Intuit can match in one location only.
342 * Another flag for this function: SECOND_TIME (so that float substrs
343 * with giant delta may be not rechecked).
346 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
348 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
349 Otherwise, only SvCUR(sv) is used to get strbeg. */
351 /* XXXX We assume that strpos is strbeg unless sv. */
353 /* XXXX Some places assume that there is a fixed substring.
354 An update may be needed if optimizer marks as "INTUITable"
355 RExen without fixed substrings. Similarly, it is assumed that
356 lengths of all the strings are no more than minlen, thus they
357 cannot come from lookahead.
358 (Or minlen should take into account lookahead.)
359 NOTE: Some of this comment is not correct. minlen does now take account
360 of lookahead/behind. Further research is required. -- demerphq
364 /* A failure to find a constant substring means that there is no need to make
365 an expensive call to REx engine, thus we celebrate a failure. Similarly,
366 finding a substring too deep into the string means that less calls to
367 regtry() should be needed.
369 REx compiler's optimizer found 4 possible hints:
370 a) Anchored substring;
372 c) Whether we are anchored (beginning-of-line or \G);
373 d) First node (of those at offset 0) which may distingush positions;
374 We use a)b)d) and multiline-part of c), and try to find a position in the
375 string which does not contradict any of them.
378 /* Most of decisions we do here should have been done at compile time.
379 The nodes of the REx which we used for the search should have been
380 deleted from the finite automaton. */
383 Perl_re_intuit_start(pTHX_ REGEXP * const prog, SV *sv, char *strpos,
384 char *strend, const U32 flags, re_scream_pos_data *data)
387 register I32 start_shift = 0;
388 /* Should be nonnegative! */
389 register I32 end_shift = 0;
394 const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
396 register char *other_last = NULL; /* other substr checked before this */
397 char *check_at = NULL; /* check substr found at this pos */
398 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
399 RXi_GET_DECL(prog,progi);
401 const char * const i_strpos = strpos;
403 GET_RE_DEBUG_FLAGS_DECL;
405 PERL_ARGS_ASSERT_RE_INTUIT_START;
407 RX_MATCH_UTF8_set(prog,do_utf8);
410 PL_reg_flags |= RF_utf8;
413 debug_start_match(prog, do_utf8, strpos, strend,
414 sv ? "Guessing start of match in sv for"
415 : "Guessing start of match in string for");
418 /* CHR_DIST() would be more correct here but it makes things slow. */
419 if (prog->minlen > strend - strpos) {
420 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
421 "String too short... [re_intuit_start]\n"));
425 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
428 if (!prog->check_utf8 && prog->check_substr)
429 to_utf8_substr(prog);
430 check = prog->check_utf8;
432 if (!prog->check_substr && prog->check_utf8)
433 to_byte_substr(prog);
434 check = prog->check_substr;
436 if (check == &PL_sv_undef) {
437 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
438 "Non-utf8 string cannot match utf8 check string\n"));
441 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
442 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
443 || ( (prog->extflags & RXf_ANCH_BOL)
444 && !multiline ) ); /* Check after \n? */
447 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
448 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
449 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
451 && (strpos != strbeg)) {
452 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
455 if (prog->check_offset_min == prog->check_offset_max &&
456 !(prog->extflags & RXf_CANY_SEEN)) {
457 /* Substring at constant offset from beg-of-str... */
460 s = HOP3c(strpos, prog->check_offset_min, strend);
463 slen = SvCUR(check); /* >= 1 */
465 if ( strend - s > slen || strend - s < slen - 1
466 || (strend - s == slen && strend[-1] != '\n')) {
467 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
470 /* Now should match s[0..slen-2] */
472 if (slen && (*SvPVX_const(check) != *s
474 && memNE(SvPVX_const(check), s, slen)))) {
476 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
480 else if (*SvPVX_const(check) != *s
481 || ((slen = SvCUR(check)) > 1
482 && memNE(SvPVX_const(check), s, slen)))
485 goto success_at_start;
488 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
490 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
491 end_shift = prog->check_end_shift;
494 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
495 - (SvTAIL(check) != 0);
496 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
498 if (end_shift < eshift)
502 else { /* Can match at random position */
505 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
506 end_shift = prog->check_end_shift;
508 /* end shift should be non negative here */
511 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
513 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
514 (IV)end_shift, RX_PRECOMP(prog));
518 /* Find a possible match in the region s..strend by looking for
519 the "check" substring in the region corrected by start/end_shift. */
522 I32 srch_start_shift = start_shift;
523 I32 srch_end_shift = end_shift;
524 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
525 srch_end_shift -= ((strbeg - s) - srch_start_shift);
526 srch_start_shift = strbeg - s;
528 DEBUG_OPTIMISE_MORE_r({
529 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
530 (IV)prog->check_offset_min,
531 (IV)srch_start_shift,
533 (IV)prog->check_end_shift);
536 if (flags & REXEC_SCREAM) {
537 I32 p = -1; /* Internal iterator of scream. */
538 I32 * const pp = data ? data->scream_pos : &p;
540 if (PL_screamfirst[BmRARE(check)] >= 0
541 || ( BmRARE(check) == '\n'
542 && (BmPREVIOUS(check) == SvCUR(check) - 1)
544 s = screaminstr(sv, check,
545 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
548 /* we may be pointing at the wrong string */
549 if (s && RXp_MATCH_COPIED(prog))
550 s = strbeg + (s - SvPVX_const(sv));
552 *data->scream_olds = s;
557 if (prog->extflags & RXf_CANY_SEEN) {
558 start_point= (U8*)(s + srch_start_shift);
559 end_point= (U8*)(strend - srch_end_shift);
561 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
562 end_point= HOP3(strend, -srch_end_shift, strbeg);
564 DEBUG_OPTIMISE_MORE_r({
565 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
566 (int)(end_point - start_point),
567 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
571 s = fbm_instr( start_point, end_point,
572 check, multiline ? FBMrf_MULTILINE : 0);
575 /* Update the count-of-usability, remove useless subpatterns,
579 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
580 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
581 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
582 (s ? "Found" : "Did not find"),
583 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)
584 ? "anchored" : "floating"),
587 (s ? " at offset " : "...\n") );
592 /* Finish the diagnostic message */
593 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
595 /* XXX dmq: first branch is for positive lookbehind...
596 Our check string is offset from the beginning of the pattern.
597 So we need to do any stclass tests offset forward from that
606 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
607 Start with the other substr.
608 XXXX no SCREAM optimization yet - and a very coarse implementation
609 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
610 *always* match. Probably should be marked during compile...
611 Probably it is right to do no SCREAM here...
614 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8)
615 : (prog->float_substr && prog->anchored_substr))
617 /* Take into account the "other" substring. */
618 /* XXXX May be hopelessly wrong for UTF... */
621 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
624 char * const last = HOP3c(s, -start_shift, strbeg);
626 char * const saved_s = s;
629 t = s - prog->check_offset_max;
630 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
632 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
637 t = HOP3c(t, prog->anchored_offset, strend);
638 if (t < other_last) /* These positions already checked */
640 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
643 /* XXXX It is not documented what units *_offsets are in.
644 We assume bytes, but this is clearly wrong.
645 Meaning this code needs to be carefully reviewed for errors.
649 /* On end-of-str: see comment below. */
650 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
651 if (must == &PL_sv_undef) {
653 DEBUG_r(must = prog->anchored_utf8); /* for debug */
658 HOP3(HOP3(last1, prog->anchored_offset, strend)
659 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
661 multiline ? FBMrf_MULTILINE : 0
664 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
665 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
666 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
667 (s ? "Found" : "Contradicts"),
668 quoted, RE_SV_TAIL(must));
673 if (last1 >= last2) {
674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
675 ", giving up...\n"));
678 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
679 ", trying floating at offset %ld...\n",
680 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
681 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
682 s = HOP3c(last, 1, strend);
686 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
687 (long)(s - i_strpos)));
688 t = HOP3c(s, -prog->anchored_offset, strbeg);
689 other_last = HOP3c(s, 1, strend);
697 else { /* Take into account the floating substring. */
699 char * const saved_s = s;
702 t = HOP3c(s, -start_shift, strbeg);
704 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
705 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
706 last = HOP3c(t, prog->float_max_offset, strend);
707 s = HOP3c(t, prog->float_min_offset, strend);
710 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
711 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
712 /* fbm_instr() takes into account exact value of end-of-str
713 if the check is SvTAIL(ed). Since false positives are OK,
714 and end-of-str is not later than strend we are OK. */
715 if (must == &PL_sv_undef) {
717 DEBUG_r(must = prog->float_utf8); /* for debug message */
720 s = fbm_instr((unsigned char*)s,
721 (unsigned char*)last + SvCUR(must)
723 must, multiline ? FBMrf_MULTILINE : 0);
725 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
726 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
727 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
728 (s ? "Found" : "Contradicts"),
729 quoted, RE_SV_TAIL(must));
733 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
734 ", giving up...\n"));
737 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
738 ", trying anchored starting at offset %ld...\n",
739 (long)(saved_s + 1 - i_strpos)));
741 s = HOP3c(t, 1, strend);
745 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
746 (long)(s - i_strpos)));
747 other_last = s; /* Fix this later. --Hugo */
757 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
759 DEBUG_OPTIMISE_MORE_r(
760 PerlIO_printf(Perl_debug_log,
761 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
762 (IV)prog->check_offset_min,
763 (IV)prog->check_offset_max,
771 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
773 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
776 /* Fixed substring is found far enough so that the match
777 cannot start at strpos. */
779 if (ml_anch && t[-1] != '\n') {
780 /* Eventually fbm_*() should handle this, but often
781 anchored_offset is not 0, so this check will not be wasted. */
782 /* XXXX In the code below we prefer to look for "^" even in
783 presence of anchored substrings. And we search even
784 beyond the found float position. These pessimizations
785 are historical artefacts only. */
787 while (t < strend - prog->minlen) {
789 if (t < check_at - prog->check_offset_min) {
790 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
791 /* Since we moved from the found position,
792 we definitely contradict the found anchored
793 substr. Due to the above check we do not
794 contradict "check" substr.
795 Thus we can arrive here only if check substr
796 is float. Redo checking for "other"=="fixed".
799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
800 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
801 goto do_other_anchored;
803 /* We don't contradict the found floating substring. */
804 /* XXXX Why not check for STCLASS? */
806 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
807 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
810 /* Position contradicts check-string */
811 /* XXXX probably better to look for check-string
812 than for "\n", so one should lower the limit for t? */
813 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
814 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
815 other_last = strpos = s = t + 1;
820 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
821 PL_colors[0], PL_colors[1]));
825 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
826 PL_colors[0], PL_colors[1]));
830 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
833 /* The found string does not prohibit matching at strpos,
834 - no optimization of calling REx engine can be performed,
835 unless it was an MBOL and we are not after MBOL,
836 or a future STCLASS check will fail this. */
838 /* Even in this situation we may use MBOL flag if strpos is offset
839 wrt the start of the string. */
840 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
841 && (strpos != strbeg) && strpos[-1] != '\n'
842 /* May be due to an implicit anchor of m{.*foo} */
843 && !(prog->intflags & PREGf_IMPLICIT))
848 DEBUG_EXECUTE_r( if (ml_anch)
849 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
850 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
853 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
855 prog->check_utf8 /* Could be deleted already */
856 && --BmUSEFUL(prog->check_utf8) < 0
857 && (prog->check_utf8 == prog->float_utf8)
859 prog->check_substr /* Could be deleted already */
860 && --BmUSEFUL(prog->check_substr) < 0
861 && (prog->check_substr == prog->float_substr)
864 /* If flags & SOMETHING - do not do it many times on the same match */
865 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
866 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
867 if (do_utf8 ? prog->check_substr : prog->check_utf8)
868 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
869 prog->check_substr = prog->check_utf8 = NULL; /* disable */
870 prog->float_substr = prog->float_utf8 = NULL; /* clear */
871 check = NULL; /* abort */
873 /* XXXX This is a remnant of the old implementation. It
874 looks wasteful, since now INTUIT can use many
876 prog->extflags &= ~RXf_USE_INTUIT;
883 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
884 /* trie stclasses are too expensive to use here, we are better off to
885 leave it to regmatch itself */
886 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
887 /* minlen == 0 is possible if regstclass is \b or \B,
888 and the fixed substr is ''$.
889 Since minlen is already taken into account, s+1 is before strend;
890 accidentally, minlen >= 1 guaranties no false positives at s + 1
891 even for \b or \B. But (minlen? 1 : 0) below assumes that
892 regstclass does not come from lookahead... */
893 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
894 This leaves EXACTF only, which is dealt with in find_byclass(). */
895 const U8* const str = (U8*)STRING(progi->regstclass);
896 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
897 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
900 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
901 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
902 else if (prog->float_substr || prog->float_utf8)
903 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
907 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
908 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
911 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
914 const char *what = NULL;
916 if (endpos == strend) {
917 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
918 "Could not match STCLASS...\n") );
921 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
922 "This position contradicts STCLASS...\n") );
923 if ((prog->extflags & RXf_ANCH) && !ml_anch)
925 /* Contradict one of substrings */
926 if (prog->anchored_substr || prog->anchored_utf8) {
927 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
928 DEBUG_EXECUTE_r( what = "anchored" );
930 s = HOP3c(t, 1, strend);
931 if (s + start_shift + end_shift > strend) {
932 /* XXXX Should be taken into account earlier? */
933 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
934 "Could not match STCLASS...\n") );
939 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
940 "Looking for %s substr starting at offset %ld...\n",
941 what, (long)(s + start_shift - i_strpos)) );
944 /* Have both, check_string is floating */
945 if (t + start_shift >= check_at) /* Contradicts floating=check */
946 goto retry_floating_check;
947 /* Recheck anchored substring, but not floating... */
951 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
952 "Looking for anchored substr starting at offset %ld...\n",
953 (long)(other_last - i_strpos)) );
954 goto do_other_anchored;
956 /* Another way we could have checked stclass at the
957 current position only: */
962 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
963 "Looking for /%s^%s/m starting at offset %ld...\n",
964 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
967 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
969 /* Check is floating subtring. */
970 retry_floating_check:
971 t = check_at - start_shift;
972 DEBUG_EXECUTE_r( what = "floating" );
973 goto hop_and_restart;
976 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
977 "By STCLASS: moving %ld --> %ld\n",
978 (long)(t - i_strpos), (long)(s - i_strpos))
982 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
983 "Does not contradict STCLASS...\n");
988 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
989 PL_colors[4], (check ? "Guessed" : "Giving up"),
990 PL_colors[5], (long)(s - i_strpos)) );
993 fail_finish: /* Substring not found */
994 if (prog->check_substr || prog->check_utf8) /* could be removed already */
995 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
997 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
998 PL_colors[4], PL_colors[5]));
1002 #define DECL_TRIE_TYPE(scan) \
1003 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1004 trie_type = (scan->flags != EXACT) \
1005 ? (do_utf8 ? trie_utf8_fold : (UTF ? trie_latin_utf8_fold : trie_plain)) \
1006 : (do_utf8 ? trie_utf8 : trie_plain)
1008 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1009 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1010 UV uvc_unfolded = 0; \
1011 switch (trie_type) { \
1012 case trie_utf8_fold: \
1013 if ( foldlen>0 ) { \
1014 uvc_unfolded = uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1019 uvc_unfolded = uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1020 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1021 foldlen -= UNISKIP( uvc ); \
1022 uscan = foldbuf + UNISKIP( uvc ); \
1025 case trie_latin_utf8_fold: \
1026 if ( foldlen>0 ) { \
1027 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1033 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1034 foldlen -= UNISKIP( uvc ); \
1035 uscan = foldbuf + UNISKIP( uvc ); \
1039 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1047 charid = trie->charmap[ uvc ]; \
1051 if (widecharmap) { \
1052 SV** const svpp = hv_fetch(widecharmap, \
1053 (char*)&uvc, sizeof(UV), 0); \
1055 charid = (U16)SvIV(*svpp); \
1058 if (!charid && trie_type == trie_utf8_fold && !UTF) { \
1059 charid = trie->charmap[uvc_unfolded]; \
1063 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1065 char *my_strend= (char *)strend; \
1068 !ibcmp_utf8(s, &my_strend, 0, do_utf8, \
1069 m, NULL, ln, (bool)UTF)) \
1070 && (!reginfo || regtry(reginfo, &s)) ) \
1073 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1074 uvchr_to_utf8(tmpbuf, c); \
1075 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1077 && (f == c1 || f == c2) \
1079 !ibcmp_utf8(s, &my_strend, 0, do_utf8,\
1080 m, NULL, ln, (bool)UTF)) \
1081 && (!reginfo || regtry(reginfo, &s)) ) \
1087 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1091 && (ln == 1 || !(OP(c) == EXACTF \
1093 : ibcmp_locale(s, m, ln))) \
1094 && (!reginfo || regtry(reginfo, &s)) ) \
1100 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1102 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1108 #define REXEC_FBC_SCAN(CoDe) \
1110 while (s < strend) { \
1116 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1117 REXEC_FBC_UTF8_SCAN( \
1119 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1128 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1131 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1140 #define REXEC_FBC_TRYIT \
1141 if ((!reginfo || regtry(reginfo, &s))) \
1144 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1146 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1149 REXEC_FBC_CLASS_SCAN(CoNd); \
1153 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1156 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1159 REXEC_FBC_CLASS_SCAN(CoNd); \
1163 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1164 PL_reg_flags |= RF_tainted; \
1166 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1169 REXEC_FBC_CLASS_SCAN(CoNd); \
1173 #define DUMP_EXEC_POS(li,s,doutf8) \
1174 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1176 /* We know what class REx starts with. Try to find this position... */
1177 /* if reginfo is NULL, its a dryrun */
1178 /* annoyingly all the vars in this routine have different names from their counterparts
1179 in regmatch. /grrr */
1182 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1183 const char *strend, regmatch_info *reginfo)
1186 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1190 register STRLEN uskip;
1194 register I32 tmp = 1; /* Scratch variable? */
1195 register const bool do_utf8 = PL_reg_match_utf8;
1196 RXi_GET_DECL(prog,progi);
1198 PERL_ARGS_ASSERT_FIND_BYCLASS;
1200 /* We know what class it must start with. */
1204 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
1205 !UTF8_IS_INVARIANT((U8)s[0]) ?
1206 reginclass(prog, c, (U8*)s, 0, do_utf8) :
1207 REGINCLASS(prog, c, (U8*)s));
1210 while (s < strend) {
1213 if (REGINCLASS(prog, c, (U8*)s) ||
1214 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1215 /* The assignment of 2 is intentional:
1216 * for the folded sharp s, the skip is 2. */
1217 (skip = SHARP_S_SKIP))) {
1218 if (tmp && (!reginfo || regtry(reginfo, &s)))
1231 if (tmp && (!reginfo || regtry(reginfo, &s)))
1239 ln = STR_LEN(c); /* length to match in octets/bytes */
1240 lnc = (I32) ln; /* length to match in characters */
1242 STRLEN ulen1, ulen2;
1244 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1245 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1246 /* used by commented-out code below */
1247 /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
1249 /* XXX: Since the node will be case folded at compile
1250 time this logic is a little odd, although im not
1251 sure that its actually wrong. --dmq */
1253 c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1254 c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1256 /* XXX: This is kinda strange. to_utf8_XYZ returns the
1257 codepoint of the first character in the converted
1258 form, yet originally we did the extra step.
1259 No tests fail by commenting this code out however
1260 so Ive left it out. -- dmq.
1262 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1264 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1269 while (sm < ((U8 *) m + ln)) {
1284 c2 = PL_fold_locale[c1];
1286 e = HOP3c(strend, -((I32)lnc), s);
1288 if (!reginfo && e < s)
1289 e = s; /* Due to minlen logic of intuit() */
1291 /* The idea in the EXACTF* cases is to first find the
1292 * first character of the EXACTF* node and then, if
1293 * necessary, case-insensitively compare the full
1294 * text of the node. The c1 and c2 are the first
1295 * characters (though in Unicode it gets a bit
1296 * more complicated because there are more cases
1297 * than just upper and lower: one needs to use
1298 * the so-called folding case for case-insensitive
1299 * matching (called "loose matching" in Unicode).
1300 * ibcmp_utf8() will do just that. */
1302 if (do_utf8 || UTF) {
1304 U8 tmpbuf [UTF8_MAXBYTES+1];
1307 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1309 /* Upper and lower of 1st char are equal -
1310 * probably not a "letter". */
1313 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1318 REXEC_FBC_EXACTISH_CHECK(c == c1);
1324 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1330 /* Handle some of the three Greek sigmas cases.
1331 * Note that not all the possible combinations
1332 * are handled here: some of them are handled
1333 * by the standard folding rules, and some of
1334 * them (the character class or ANYOF cases)
1335 * are handled during compiletime in
1336 * regexec.c:S_regclass(). */
1337 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1338 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1339 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1341 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1346 /* Neither pattern nor string are UTF8 */
1348 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1350 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1354 PL_reg_flags |= RF_tainted;
1361 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1362 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1364 tmp = ((OP(c) == BOUND ?
1365 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1366 LOAD_UTF8_CHARCLASS_ALNUM();
1367 REXEC_FBC_UTF8_SCAN(
1368 if (tmp == !(OP(c) == BOUND ?
1369 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1370 isALNUM_LC_utf8((U8*)s)))
1378 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1379 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1382 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1388 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1392 PL_reg_flags |= RF_tainted;
1399 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1400 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1402 tmp = ((OP(c) == NBOUND ?
1403 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1404 LOAD_UTF8_CHARCLASS_ALNUM();
1405 REXEC_FBC_UTF8_SCAN(
1406 if (tmp == !(OP(c) == NBOUND ?
1407 (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1408 isALNUM_LC_utf8((U8*)s)))
1410 else REXEC_FBC_TRYIT;
1414 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1415 tmp = ((OP(c) == NBOUND ?
1416 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1419 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1421 else REXEC_FBC_TRYIT;
1424 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1428 REXEC_FBC_CSCAN_PRELOAD(
1429 LOAD_UTF8_CHARCLASS_ALNUM(),
1430 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1434 REXEC_FBC_CSCAN_TAINT(
1435 isALNUM_LC_utf8((U8*)s),
1439 REXEC_FBC_CSCAN_PRELOAD(
1440 LOAD_UTF8_CHARCLASS_ALNUM(),
1441 !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8),
1445 REXEC_FBC_CSCAN_TAINT(
1446 !isALNUM_LC_utf8((U8*)s),
1450 REXEC_FBC_CSCAN_PRELOAD(
1451 LOAD_UTF8_CHARCLASS_SPACE(),
1452 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8),
1456 REXEC_FBC_CSCAN_TAINT(
1457 *s == ' ' || isSPACE_LC_utf8((U8*)s),
1461 REXEC_FBC_CSCAN_PRELOAD(
1462 LOAD_UTF8_CHARCLASS_SPACE(),
1463 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)),
1467 REXEC_FBC_CSCAN_TAINT(
1468 !(*s == ' ' || isSPACE_LC_utf8((U8*)s)),
1472 REXEC_FBC_CSCAN_PRELOAD(
1473 LOAD_UTF8_CHARCLASS_DIGIT(),
1474 swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1478 REXEC_FBC_CSCAN_TAINT(
1479 isDIGIT_LC_utf8((U8*)s),
1483 REXEC_FBC_CSCAN_PRELOAD(
1484 LOAD_UTF8_CHARCLASS_DIGIT(),
1485 !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8),
1489 REXEC_FBC_CSCAN_TAINT(
1490 !isDIGIT_LC_utf8((U8*)s),
1496 is_LNBREAK_latin1(s)
1506 !is_VERTWS_latin1(s)
1511 is_HORIZWS_latin1(s)
1515 !is_HORIZWS_utf8(s),
1516 !is_HORIZWS_latin1(s)
1522 /* what trie are we using right now */
1524 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1526 = (reg_trie_data*)progi->data->data[ aho->trie ];
1527 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1529 const char *last_start = strend - trie->minlen;
1531 const char *real_start = s;
1533 STRLEN maxlen = trie->maxlen;
1535 U8 **points; /* map of where we were in the input string
1536 when reading a given char. For ASCII this
1537 is unnecessary overhead as the relationship
1538 is always 1:1, but for Unicode, especially
1539 case folded Unicode this is not true. */
1540 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1544 GET_RE_DEBUG_FLAGS_DECL;
1546 /* We can't just allocate points here. We need to wrap it in
1547 * an SV so it gets freed properly if there is a croak while
1548 * running the match */
1551 sv_points=newSV(maxlen * sizeof(U8 *));
1552 SvCUR_set(sv_points,
1553 maxlen * sizeof(U8 *));
1554 SvPOK_on(sv_points);
1555 sv_2mortal(sv_points);
1556 points=(U8**)SvPV_nolen(sv_points );
1557 if ( trie_type != trie_utf8_fold
1558 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1561 bitmap=(U8*)trie->bitmap;
1563 bitmap=(U8*)ANYOF_BITMAP(c);
1565 /* this is the Aho-Corasick algorithm modified a touch
1566 to include special handling for long "unknown char"
1567 sequences. The basic idea being that we use AC as long
1568 as we are dealing with a possible matching char, when
1569 we encounter an unknown char (and we have not encountered
1570 an accepting state) we scan forward until we find a legal
1572 AC matching is basically that of trie matching, except
1573 that when we encounter a failing transition, we fall back
1574 to the current states "fail state", and try the current char
1575 again, a process we repeat until we reach the root state,
1576 state 1, or a legal transition. If we fail on the root state
1577 then we can either terminate if we have reached an accepting
1578 state previously, or restart the entire process from the beginning
1582 while (s <= last_start) {
1583 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1591 U8 *uscan = (U8*)NULL;
1592 U8 *leftmost = NULL;
1594 U32 accepted_word= 0;
1598 while ( state && uc <= (U8*)strend ) {
1600 U32 word = aho->states[ state ].wordnum;
1604 DEBUG_TRIE_EXECUTE_r(
1605 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1606 dump_exec_pos( (char *)uc, c, strend, real_start,
1607 (char *)uc, do_utf8 );
1608 PerlIO_printf( Perl_debug_log,
1609 " Scanning for legal start char...\n");
1612 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1617 if (uc >(U8*)last_start) break;
1621 U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ];
1622 if (!leftmost || lpos < leftmost) {
1623 DEBUG_r(accepted_word=word);
1629 points[pointpos++ % maxlen]= uc;
1630 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1631 uscan, len, uvc, charid, foldlen,
1633 DEBUG_TRIE_EXECUTE_r({
1634 dump_exec_pos( (char *)uc, c, strend, real_start,
1636 PerlIO_printf(Perl_debug_log,
1637 " Charid:%3u CP:%4"UVxf" ",
1643 word = aho->states[ state ].wordnum;
1645 base = aho->states[ state ].trans.base;
1647 DEBUG_TRIE_EXECUTE_r({
1649 dump_exec_pos( (char *)uc, c, strend, real_start,
1651 PerlIO_printf( Perl_debug_log,
1652 "%sState: %4"UVxf", word=%"UVxf,
1653 failed ? " Fail transition to " : "",
1654 (UV)state, (UV)word);
1659 (base + charid > trie->uniquecharcount )
1660 && (base + charid - 1 - trie->uniquecharcount
1662 && trie->trans[base + charid - 1 -
1663 trie->uniquecharcount].check == state
1664 && (tmp=trie->trans[base + charid - 1 -
1665 trie->uniquecharcount ].next))
1667 DEBUG_TRIE_EXECUTE_r(
1668 PerlIO_printf( Perl_debug_log," - legal\n"));
1673 DEBUG_TRIE_EXECUTE_r(
1674 PerlIO_printf( Perl_debug_log," - fail\n"));
1676 state = aho->fail[state];
1680 /* we must be accepting here */
1681 DEBUG_TRIE_EXECUTE_r(
1682 PerlIO_printf( Perl_debug_log," - accepting\n"));
1691 if (!state) state = 1;
1694 if ( aho->states[ state ].wordnum ) {
1695 U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ];
1696 if (!leftmost || lpos < leftmost) {
1697 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1702 s = (char*)leftmost;
1703 DEBUG_TRIE_EXECUTE_r({
1705 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1706 (UV)accepted_word, (IV)(s - real_start)
1709 if (!reginfo || regtry(reginfo, &s)) {
1715 DEBUG_TRIE_EXECUTE_r({
1716 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1719 DEBUG_TRIE_EXECUTE_r(
1720 PerlIO_printf( Perl_debug_log,"No match.\n"));
1729 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1738 S_swap_match_buff (pTHX_ regexp *prog)
1740 regexp_paren_pair *t;
1742 PERL_ARGS_ASSERT_SWAP_MATCH_BUFF;
1745 /* We have to be careful. If the previous successful match
1746 was from this regex we don't want a subsequent paritally
1747 successful match to clobber the old results.
1748 So when we detect this possibility we add a swap buffer
1749 to the re, and switch the buffer each match. If we fail
1750 we switch it back, otherwise we leave it swapped.
1752 Newxz(prog->swap, (prog->nparens + 1), regexp_paren_pair);
1755 prog->swap = prog->offs;
1761 - regexec_flags - match a regexp against a string
1764 Perl_regexec_flags(pTHX_ REGEXP * const prog, char *stringarg, register char *strend,
1765 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1766 /* strend: pointer to null at end of string */
1767 /* strbeg: real beginning of string */
1768 /* minend: end of match must be >=minend after stringarg. */
1769 /* data: May be used for some additional optimizations.
1770 Currently its only used, with a U32 cast, for transmitting
1771 the ganch offset when doing a /g match. This will change */
1772 /* nosave: For optimizations. */
1775 /*register*/ char *s;
1776 register regnode *c;
1777 /*register*/ char *startpos = stringarg;
1778 I32 minlen; /* must match at least this many chars */
1779 I32 dontbother = 0; /* how many characters not to try at end */
1780 I32 end_shift = 0; /* Same for the end. */ /* CC */
1781 I32 scream_pos = -1; /* Internal iterator of scream. */
1782 char *scream_olds = NULL;
1783 const bool do_utf8 = (bool)DO_UTF8(sv);
1785 RXi_GET_DECL(prog,progi);
1786 regmatch_info reginfo; /* create some info to pass to regtry etc */
1787 bool swap_on_fail = 0;
1788 GET_RE_DEBUG_FLAGS_DECL;
1790 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1791 PERL_UNUSED_ARG(data);
1793 /* Be paranoid... */
1794 if (prog == NULL || startpos == NULL) {
1795 Perl_croak(aTHX_ "NULL regexp parameter");
1799 multiline = prog->extflags & RXf_PMf_MULTILINE;
1800 reginfo.prog = prog;
1802 RX_MATCH_UTF8_set(prog, do_utf8);
1804 debug_start_match(prog, do_utf8, startpos, strend,
1808 minlen = prog->minlen;
1810 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1811 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1812 "String too short [regexec_flags]...\n"));
1817 /* Check validity of program. */
1818 if (UCHARAT(progi->program) != REG_MAGIC) {
1819 Perl_croak(aTHX_ "corrupted regexp program");
1823 PL_reg_eval_set = 0;
1827 PL_reg_flags |= RF_utf8;
1829 /* Mark beginning of line for ^ and lookbehind. */
1830 reginfo.bol = startpos; /* XXX not used ??? */
1834 /* Mark end of line for $ (and such) */
1837 /* see how far we have to get to not match where we matched before */
1838 reginfo.till = startpos+minend;
1840 /* If there is a "must appear" string, look for it. */
1843 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1846 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1847 reginfo.ganch = startpos + prog->gofs;
1848 else if (sv && SvTYPE(sv) >= SVt_PVMG
1850 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1851 && mg->mg_len >= 0) {
1852 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1853 if (prog->extflags & RXf_ANCH_GPOS) {
1854 if (s > reginfo.ganch)
1856 s = reginfo.ganch - prog->gofs;
1860 reginfo.ganch = strbeg + PTR2UV(data);
1861 } else /* pos() not defined */
1862 reginfo.ganch = strbeg;
1864 if (PL_curpm && (PM_GETRE(PL_curpm) == prog)) {
1866 swap_match_buff(prog); /* do we need a save destructor here for
1869 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1870 re_scream_pos_data d;
1872 d.scream_olds = &scream_olds;
1873 d.scream_pos = &scream_pos;
1874 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1876 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1877 goto phooey; /* not present */
1883 /* Simplest case: anchored match need be tried only once. */
1884 /* [unless only anchor is BOL and multiline is set] */
1885 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
1886 if (s == startpos && regtry(®info, &startpos))
1888 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
1889 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
1894 dontbother = minlen - 1;
1895 end = HOP3c(strend, -dontbother, strbeg) - 1;
1896 /* for multiline we only have to try after newlines */
1897 if (prog->check_substr || prog->check_utf8) {
1901 if (regtry(®info, &s))
1906 if (prog->extflags & RXf_USE_INTUIT) {
1907 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1918 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1919 if (regtry(®info, &s))
1926 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
1928 /* the warning about reginfo.ganch being used without intialization
1929 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
1930 and we only enter this block when the same bit is set. */
1931 char *tmp_s = reginfo.ganch - prog->gofs;
1932 if (regtry(®info, &tmp_s))
1937 /* Messy cases: unanchored match. */
1938 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
1939 /* we have /x+whatever/ */
1940 /* it must be a one character string (XXXX Except UTF?) */
1945 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1946 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1947 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1952 DEBUG_EXECUTE_r( did_match = 1 );
1953 if (regtry(®info, &s)) goto got_it;
1955 while (s < strend && *s == ch)
1963 DEBUG_EXECUTE_r( did_match = 1 );
1964 if (regtry(®info, &s)) goto got_it;
1966 while (s < strend && *s == ch)
1971 DEBUG_EXECUTE_r(if (!did_match)
1972 PerlIO_printf(Perl_debug_log,
1973 "Did not find anchored character...\n")
1976 else if (prog->anchored_substr != NULL
1977 || prog->anchored_utf8 != NULL
1978 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1979 && prog->float_max_offset < strend - s)) {
1984 char *last1; /* Last position checked before */
1988 if (prog->anchored_substr || prog->anchored_utf8) {
1989 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1990 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1991 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1992 back_max = back_min = prog->anchored_offset;
1994 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1995 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1996 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1997 back_max = prog->float_max_offset;
1998 back_min = prog->float_min_offset;
2002 if (must == &PL_sv_undef)
2003 /* could not downgrade utf8 check substring, so must fail */
2009 last = HOP3c(strend, /* Cannot start after this */
2010 -(I32)(CHR_SVLEN(must)
2011 - (SvTAIL(must) != 0) + back_min), strbeg);
2014 last1 = HOPc(s, -1);
2016 last1 = s - 1; /* bogus */
2018 /* XXXX check_substr already used to find "s", can optimize if
2019 check_substr==must. */
2021 dontbother = end_shift;
2022 strend = HOPc(strend, -dontbother);
2023 while ( (s <= last) &&
2024 ((flags & REXEC_SCREAM)
2025 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2026 end_shift, &scream_pos, 0))
2027 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2028 (unsigned char*)strend, must,
2029 multiline ? FBMrf_MULTILINE : 0))) ) {
2030 /* we may be pointing at the wrong string */
2031 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2032 s = strbeg + (s - SvPVX_const(sv));
2033 DEBUG_EXECUTE_r( did_match = 1 );
2034 if (HOPc(s, -back_max) > last1) {
2035 last1 = HOPc(s, -back_min);
2036 s = HOPc(s, -back_max);
2039 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2041 last1 = HOPc(s, -back_min);
2045 while (s <= last1) {
2046 if (regtry(®info, &s))
2052 while (s <= last1) {
2053 if (regtry(®info, &s))
2059 DEBUG_EXECUTE_r(if (!did_match) {
2060 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
2061 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2062 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2063 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2064 ? "anchored" : "floating"),
2065 quoted, RE_SV_TAIL(must));
2069 else if ( (c = progi->regstclass) ) {
2071 const OPCODE op = OP(progi->regstclass);
2072 /* don't bother with what can't match */
2073 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2074 strend = HOPc(strend, -(minlen - 1));
2077 SV * const prop = sv_newmortal();
2078 regprop(prog, prop, c);
2080 RE_PV_QUOTED_DECL(quoted,do_utf8,PERL_DEBUG_PAD_ZERO(1),
2082 PerlIO_printf(Perl_debug_log,
2083 "Matching stclass %.*s against %s (%d chars)\n",
2084 (int)SvCUR(prop), SvPVX_const(prop),
2085 quoted, (int)(strend - s));
2088 if (find_byclass(prog, c, s, strend, ®info))
2090 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2094 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2099 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2100 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2101 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
2103 if (flags & REXEC_SCREAM) {
2104 last = screaminstr(sv, float_real, s - strbeg,
2105 end_shift, &scream_pos, 1); /* last one */
2107 last = scream_olds; /* Only one occurrence. */
2108 /* we may be pointing at the wrong string */
2109 else if (RXp_MATCH_COPIED(prog))
2110 s = strbeg + (s - SvPVX_const(sv));
2114 const char * const little = SvPV_const(float_real, len);
2116 if (SvTAIL(float_real)) {
2117 if (memEQ(strend - len + 1, little, len - 1))
2118 last = strend - len + 1;
2119 else if (!multiline)
2120 last = memEQ(strend - len, little, len)
2121 ? strend - len : NULL;
2127 last = rninstr(s, strend, little, little + len);
2129 last = strend; /* matching "$" */
2134 PerlIO_printf(Perl_debug_log,
2135 "%sCan't trim the tail, match fails (should not happen)%s\n",
2136 PL_colors[4], PL_colors[5]));
2137 goto phooey; /* Should not happen! */
2139 dontbother = strend - last + prog->float_min_offset;
2141 if (minlen && (dontbother < minlen))
2142 dontbother = minlen - 1;
2143 strend -= dontbother; /* this one's always in bytes! */
2144 /* We don't know much -- general case. */
2147 if (regtry(®info, &s))
2156 if (regtry(®info, &s))
2158 } while (s++ < strend);
2166 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2168 if (PL_reg_eval_set)
2169 restore_pos(aTHX_ prog);
2170 if (RXp_PAREN_NAMES(prog))
2171 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2173 /* make sure $`, $&, $', and $digit will work later */
2174 if ( !(flags & REXEC_NOT_FIRST) ) {
2175 RX_MATCH_COPY_FREE(prog);
2176 if (flags & REXEC_COPY_STR) {
2177 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2178 #ifdef PERL_OLD_COPY_ON_WRITE
2180 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2182 PerlIO_printf(Perl_debug_log,
2183 "Copy on write: regexp capture, type %d\n",
2186 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2187 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2188 assert (SvPOKp(prog->saved_copy));
2192 RX_MATCH_COPIED_on(prog);
2193 s = savepvn(strbeg, i);
2199 prog->subbeg = strbeg;
2200 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2207 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2208 PL_colors[4], PL_colors[5]));
2209 if (PL_reg_eval_set)
2210 restore_pos(aTHX_ prog);
2212 /* we failed :-( roll it back */
2213 swap_match_buff(prog);
2220 - regtry - try match at specific point
2222 STATIC I32 /* 0 failure, 1 success */
2223 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2227 regexp *prog = reginfo->prog;
2228 RXi_GET_DECL(prog,progi);
2229 GET_RE_DEBUG_FLAGS_DECL;
2231 PERL_ARGS_ASSERT_REGTRY;
2233 reginfo->cutpoint=NULL;
2235 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2238 PL_reg_eval_set = RS_init;
2239 DEBUG_EXECUTE_r(DEBUG_s(
2240 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2241 (IV)(PL_stack_sp - PL_stack_base));
2244 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2245 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2247 /* Apparently this is not needed, judging by wantarray. */
2248 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2249 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2252 /* Make $_ available to executed code. */
2253 if (reginfo->sv != DEFSV) {
2255 DEFSV_set(reginfo->sv);
2258 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2259 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2260 /* prepare for quick setting of pos */
2261 #ifdef PERL_OLD_COPY_ON_WRITE
2262 if (SvIsCOW(reginfo->sv))
2263 sv_force_normal_flags(reginfo->sv, 0);
2265 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2266 &PL_vtbl_mglob, NULL, 0);
2270 PL_reg_oldpos = mg->mg_len;
2271 SAVEDESTRUCTOR_X(restore_pos, prog);
2273 if (!PL_reg_curpm) {
2274 Newxz(PL_reg_curpm, 1, PMOP);
2277 SV* const repointer = newSViv(0);
2278 /* this regexp is also owned by the new PL_reg_curpm, which
2279 will try to free it. */
2280 av_push(PL_regex_padav,repointer);
2281 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2282 PL_regex_pad = AvARRAY(PL_regex_padav);
2287 /* It seems that non-ithreads works both with and without this code.
2288 So for efficiency reasons it seems best not to have the code
2289 compiled when it is not needed. */
2290 /* This is safe against NULLs: */
2291 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2292 /* PM_reg_curpm owns a reference to this regexp. */
2295 PM_SETRE(PL_reg_curpm, prog);
2296 PL_reg_oldcurpm = PL_curpm;
2297 PL_curpm = PL_reg_curpm;
2298 if (RXp_MATCH_COPIED(prog)) {
2299 /* Here is a serious problem: we cannot rewrite subbeg,
2300 since it may be needed if this match fails. Thus
2301 $` inside (?{}) could fail... */
2302 PL_reg_oldsaved = prog->subbeg;
2303 PL_reg_oldsavedlen = prog->sublen;
2304 #ifdef PERL_OLD_COPY_ON_WRITE
2305 PL_nrs = prog->saved_copy;
2307 RXp_MATCH_COPIED_off(prog);
2310 PL_reg_oldsaved = NULL;
2311 prog->subbeg = PL_bostr;
2312 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2314 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2315 prog->offs[0].start = *startpos - PL_bostr;
2316 PL_reginput = *startpos;
2317 PL_reglastparen = &prog->lastparen;
2318 PL_reglastcloseparen = &prog->lastcloseparen;
2319 prog->lastparen = 0;
2320 prog->lastcloseparen = 0;
2322 PL_regoffs = prog->offs;
2323 if (PL_reg_start_tmpl <= prog->nparens) {
2324 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2325 if(PL_reg_start_tmp)
2326 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2328 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2331 /* XXXX What this code is doing here?!!! There should be no need
2332 to do this again and again, PL_reglastparen should take care of
2335 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2336 * Actually, the code in regcppop() (which Ilya may be meaning by
2337 * PL_reglastparen), is not needed at all by the test suite
2338 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2339 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2340 * Meanwhile, this code *is* needed for the
2341 * above-mentioned test suite tests to succeed. The common theme
2342 * on those tests seems to be returning null fields from matches.
2343 * --jhi updated by dapm */
2345 if (prog->nparens) {
2346 regexp_paren_pair *pp = PL_regoffs;
2348 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2356 if (regmatch(reginfo, progi->program + 1)) {
2357 PL_regoffs[0].end = PL_reginput - PL_bostr;
2360 if (reginfo->cutpoint)
2361 *startpos= reginfo->cutpoint;
2362 REGCP_UNWIND(lastcp);
2367 #define sayYES goto yes
2368 #define sayNO goto no
2369 #define sayNO_SILENT goto no_silent
2371 /* we dont use STMT_START/END here because it leads to
2372 "unreachable code" warnings, which are bogus, but distracting. */
2373 #define CACHEsayNO \
2374 if (ST.cache_mask) \
2375 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2378 /* this is used to determine how far from the left messages like
2379 'failed...' are printed. It should be set such that messages
2380 are inline with the regop output that created them.
2382 #define REPORT_CODE_OFF 32
2385 /* Make sure there is a test for this +1 options in re_tests */
2386 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2388 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2389 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2391 #define SLAB_FIRST(s) (&(s)->states[0])
2392 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2394 /* grab a new slab and return the first slot in it */
2396 STATIC regmatch_state *
2399 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2402 regmatch_slab *s = PL_regmatch_slab->next;
2404 Newx(s, 1, regmatch_slab);
2405 s->prev = PL_regmatch_slab;
2407 PL_regmatch_slab->next = s;
2409 PL_regmatch_slab = s;
2410 return SLAB_FIRST(s);
2414 /* push a new state then goto it */
2416 #define PUSH_STATE_GOTO(state, node) \
2418 st->resume_state = state; \
2421 /* push a new state with success backtracking, then goto it */
2423 #define PUSH_YES_STATE_GOTO(state, node) \
2425 st->resume_state = state; \
2426 goto push_yes_state;
2432 regmatch() - main matching routine
2434 This is basically one big switch statement in a loop. We execute an op,
2435 set 'next' to point the next op, and continue. If we come to a point which
2436 we may need to backtrack to on failure such as (A|B|C), we push a
2437 backtrack state onto the backtrack stack. On failure, we pop the top
2438 state, and re-enter the loop at the state indicated. If there are no more
2439 states to pop, we return failure.
2441 Sometimes we also need to backtrack on success; for example /A+/, where
2442 after successfully matching one A, we need to go back and try to
2443 match another one; similarly for lookahead assertions: if the assertion
2444 completes successfully, we backtrack to the state just before the assertion
2445 and then carry on. In these cases, the pushed state is marked as
2446 'backtrack on success too'. This marking is in fact done by a chain of
2447 pointers, each pointing to the previous 'yes' state. On success, we pop to
2448 the nearest yes state, discarding any intermediate failure-only states.
2449 Sometimes a yes state is pushed just to force some cleanup code to be
2450 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2451 it to free the inner regex.
2453 Note that failure backtracking rewinds the cursor position, while
2454 success backtracking leaves it alone.
2456 A pattern is complete when the END op is executed, while a subpattern
2457 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2458 ops trigger the "pop to last yes state if any, otherwise return true"
2461 A common convention in this function is to use A and B to refer to the two
2462 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2463 the subpattern to be matched possibly multiple times, while B is the entire
2464 rest of the pattern. Variable and state names reflect this convention.
2466 The states in the main switch are the union of ops and failure/success of
2467 substates associated with with that op. For example, IFMATCH is the op
2468 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2469 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2470 successfully matched A and IFMATCH_A_fail is a state saying that we have
2471 just failed to match A. Resume states always come in pairs. The backtrack
2472 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2473 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2474 on success or failure.
2476 The struct that holds a backtracking state is actually a big union, with
2477 one variant for each major type of op. The variable st points to the
2478 top-most backtrack struct. To make the code clearer, within each
2479 block of code we #define ST to alias the relevant union.
2481 Here's a concrete example of a (vastly oversimplified) IFMATCH
2487 #define ST st->u.ifmatch
2489 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2490 ST.foo = ...; // some state we wish to save
2492 // push a yes backtrack state with a resume value of
2493 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2495 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2498 case IFMATCH_A: // we have successfully executed A; now continue with B
2500 bar = ST.foo; // do something with the preserved value
2503 case IFMATCH_A_fail: // A failed, so the assertion failed
2504 ...; // do some housekeeping, then ...
2505 sayNO; // propagate the failure
2512 For any old-timers reading this who are familiar with the old recursive
2513 approach, the code above is equivalent to:
2515 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2524 ...; // do some housekeeping, then ...
2525 sayNO; // propagate the failure
2528 The topmost backtrack state, pointed to by st, is usually free. If you
2529 want to claim it, populate any ST.foo fields in it with values you wish to
2530 save, then do one of
2532 PUSH_STATE_GOTO(resume_state, node);
2533 PUSH_YES_STATE_GOTO(resume_state, node);
2535 which sets that backtrack state's resume value to 'resume_state', pushes a
2536 new free entry to the top of the backtrack stack, then goes to 'node'.
2537 On backtracking, the free slot is popped, and the saved state becomes the
2538 new free state. An ST.foo field in this new top state can be temporarily
2539 accessed to retrieve values, but once the main loop is re-entered, it
2540 becomes available for reuse.
2542 Note that the depth of the backtrack stack constantly increases during the
2543 left-to-right execution of the pattern, rather than going up and down with
2544 the pattern nesting. For example the stack is at its maximum at Z at the
2545 end of the pattern, rather than at X in the following:
2547 /(((X)+)+)+....(Y)+....Z/
2549 The only exceptions to this are lookahead/behind assertions and the cut,
2550 (?>A), which pop all the backtrack states associated with A before
2553 Bascktrack state structs are allocated in slabs of about 4K in size.
2554 PL_regmatch_state and st always point to the currently active state,
2555 and PL_regmatch_slab points to the slab currently containing
2556 PL_regmatch_state. The first time regmatch() is called, the first slab is
2557 allocated, and is never freed until interpreter destruction. When the slab
2558 is full, a new one is allocated and chained to the end. At exit from
2559 regmatch(), slabs allocated since entry are freed.
2564 #define DEBUG_STATE_pp(pp) \
2566 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2567 PerlIO_printf(Perl_debug_log, \
2568 " %*s"pp" %s%s%s%s%s\n", \
2570 PL_reg_name[st->resume_state], \
2571 ((st==yes_state||st==mark_state) ? "[" : ""), \
2572 ((st==yes_state) ? "Y" : ""), \
2573 ((st==mark_state) ? "M" : ""), \
2574 ((st==yes_state||st==mark_state) ? "]" : "") \
2579 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2584 S_debug_start_match(pTHX_ const REGEXP *prog, const bool do_utf8,
2585 const char *start, const char *end, const char *blurb)
2587 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2589 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2594 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2595 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2597 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2598 start, end - start, 60);
2600 PerlIO_printf(Perl_debug_log,
2601 "%s%s REx%s %s against %s\n",
2602 PL_colors[4], blurb, PL_colors[5], s0, s1);
2604 if (do_utf8||utf8_pat)
2605 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2606 utf8_pat ? "pattern" : "",
2607 utf8_pat && do_utf8 ? " and " : "",
2608 do_utf8 ? "string" : ""
2614 S_dump_exec_pos(pTHX_ const char *locinput,
2615 const regnode *scan,
2616 const char *loc_regeol,
2617 const char *loc_bostr,
2618 const char *loc_reg_starttry,
2621 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2622 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2623 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2624 /* The part of the string before starttry has one color
2625 (pref0_len chars), between starttry and current
2626 position another one (pref_len - pref0_len chars),
2627 after the current position the third one.
2628 We assume that pref0_len <= pref_len, otherwise we
2629 decrease pref0_len. */
2630 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2631 ? (5 + taill) - l : locinput - loc_bostr;
2634 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2636 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2638 pref0_len = pref_len - (locinput - loc_reg_starttry);
2639 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2640 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2641 ? (5 + taill) - pref_len : loc_regeol - locinput);
2642 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2646 if (pref0_len > pref_len)
2647 pref0_len = pref_len;
2649 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2651 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2652 (locinput - pref_len),pref0_len, 60, 4, 5);
2654 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2655 (locinput - pref_len + pref0_len),
2656 pref_len - pref0_len, 60, 2, 3);
2658 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2659 locinput, loc_regeol - locinput, 10, 0, 1);
2661 const STRLEN tlen=len0+len1+len2;
2662 PerlIO_printf(Perl_debug_log,
2663 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2664 (IV)(locinput - loc_bostr),
2667 (docolor ? "" : "> <"),
2669 (int)(tlen > 19 ? 0 : 19 - tlen),
2676 /* reg_check_named_buff_matched()
2677 * Checks to see if a named buffer has matched. The data array of
2678 * buffer numbers corresponding to the buffer is expected to reside
2679 * in the regexp->data->data array in the slot stored in the ARG() of
2680 * node involved. Note that this routine doesn't actually care about the
2681 * name, that information is not preserved from compilation to execution.
2682 * Returns the index of the leftmost defined buffer with the given name
2683 * or 0 if non of the buffers matched.
2686 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2689 RXi_GET_DECL(rex,rexi);
2690 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2691 I32 *nums=(I32*)SvPVX(sv_dat);
2693 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2695 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2696 if ((I32)*PL_reglastparen >= nums[n] &&
2697 PL_regoffs[nums[n]].end != -1)
2706 /* free all slabs above current one - called during LEAVE_SCOPE */
2709 S_clear_backtrack_stack(pTHX_ void *p)
2711 regmatch_slab *s = PL_regmatch_slab->next;
2716 PL_regmatch_slab->next = NULL;
2718 regmatch_slab * const osl = s;
2725 #define SETREX(Re1,Re2) \
2726 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2729 STATIC I32 /* 0 failure, 1 success */
2730 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2732 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2736 register const bool do_utf8 = PL_reg_match_utf8;
2737 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2738 regexp *rex = reginfo->prog;
2739 RXi_GET_DECL(rex,rexi);
2741 /* the current state. This is a cached copy of PL_regmatch_state */
2742 register regmatch_state *st;
2743 /* cache heavy used fields of st in registers */
2744 register regnode *scan;
2745 register regnode *next;
2746 register U32 n = 0; /* general value; init to avoid compiler warning */
2747 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2748 register char *locinput = PL_reginput;
2749 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2751 bool result = 0; /* return value of S_regmatch */
2752 int depth = 0; /* depth of backtrack stack */
2753 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2754 const U32 max_nochange_depth =
2755 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2756 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2757 regmatch_state *yes_state = NULL; /* state to pop to on success of
2759 /* mark_state piggy backs on the yes_state logic so that when we unwind
2760 the stack on success we can update the mark_state as we go */
2761 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2762 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2763 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2765 bool no_final = 0; /* prevent failure from backtracking? */
2766 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2767 char *startpoint = PL_reginput;
2768 SV *popmark = NULL; /* are we looking for a mark? */
2769 SV *sv_commit = NULL; /* last mark name seen in failure */
2770 SV *sv_yes_mark = NULL; /* last mark name we have seen
2771 during a successfull match */
2772 U32 lastopen = 0; /* last open we saw */
2773 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2774 SV* const oreplsv = GvSV(PL_replgv);
2775 /* these three flags are set by various ops to signal information to
2776 * the very next op. They have a useful lifetime of exactly one loop
2777 * iteration, and are not preserved or restored by state pushes/pops
2779 bool sw = 0; /* the condition value in (?(cond)a|b) */
2780 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2781 int logical = 0; /* the following EVAL is:
2785 or the following IFMATCH/UNLESSM is:
2786 false: plain (?=foo)
2787 true: used as a condition: (?(?=foo))
2790 GET_RE_DEBUG_FLAGS_DECL;
2793 PERL_ARGS_ASSERT_REGMATCH;
2795 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
2796 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2798 /* on first ever call to regmatch, allocate first slab */
2799 if (!PL_regmatch_slab) {
2800 Newx(PL_regmatch_slab, 1, regmatch_slab);
2801 PL_regmatch_slab->prev = NULL;
2802 PL_regmatch_slab->next = NULL;
2803 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2806 oldsave = PL_savestack_ix;
2807 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
2808 SAVEVPTR(PL_regmatch_slab);
2809 SAVEVPTR(PL_regmatch_state);
2811 /* grab next free state slot */
2812 st = ++PL_regmatch_state;
2813 if (st > SLAB_LAST(PL_regmatch_slab))
2814 st = PL_regmatch_state = S_push_slab(aTHX);
2816 /* Note that nextchr is a byte even in UTF */
2817 nextchr = UCHARAT(locinput);
2819 while (scan != NULL) {
2822 SV * const prop = sv_newmortal();
2823 regnode *rnext=regnext(scan);
2824 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2825 regprop(rex, prop, scan);
2827 PerlIO_printf(Perl_debug_log,
2828 "%3"IVdf":%*s%s(%"IVdf")\n",
2829 (IV)(scan - rexi->program), depth*2, "",
2831 (PL_regkind[OP(scan)] == END || !rnext) ?
2832 0 : (IV)(rnext - rexi->program));
2835 next = scan + NEXT_OFF(scan);
2838 state_num = OP(scan);
2840 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
2843 assert(PL_reglastparen == &rex->lastparen);
2844 assert(PL_reglastcloseparen == &rex->lastcloseparen);
2845 assert(PL_regoffs == rex->offs);
2847 switch (state_num) {
2849 if (locinput == PL_bostr)
2851 /* reginfo->till = reginfo->bol; */
2856 if (locinput == PL_bostr ||
2857 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2863 if (locinput == PL_bostr)
2867 if (locinput == reginfo->ganch)
2872 /* update the startpoint */
2873 st->u.keeper.val = PL_regoffs[0].start;
2874 PL_reginput = locinput;
2875 PL_regoffs[0].start = locinput - PL_bostr;
2876 PUSH_STATE_GOTO(KEEPS_next, next);
2878 case KEEPS_next_fail:
2879 /* rollback the start point change */
2880 PL_regoffs[0].start = st->u.keeper.val;
2886 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2891 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2893 if (PL_regeol - locinput > 1)
2897 if (PL_regeol != locinput)
2901 if (!nextchr && locinput >= PL_regeol)
2904 locinput += PL_utf8skip[nextchr];
2905 if (locinput > PL_regeol)
2907 nextchr = UCHARAT(locinput);
2910 nextchr = UCHARAT(++locinput);
2913 if (!nextchr && locinput >= PL_regeol)
2915 nextchr = UCHARAT(++locinput);
2918 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2921 locinput += PL_utf8skip[nextchr];
2922 if (locinput > PL_regeol)
2924 nextchr = UCHARAT(locinput);
2927 nextchr = UCHARAT(++locinput);
2931 #define ST st->u.trie
2933 /* In this case the charclass data is available inline so
2934 we can fail fast without a lot of extra overhead.
2936 if (scan->flags == EXACT || !do_utf8) {
2937 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2939 PerlIO_printf(Perl_debug_log,
2940 "%*s %sfailed to match trie start class...%s\n",
2941 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2950 /* what type of TRIE am I? (utf8 makes this contextual) */
2951 DECL_TRIE_TYPE(scan);
2953 /* what trie are we using right now */
2954 reg_trie_data * const trie
2955 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2956 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
2957 U32 state = trie->startstate;
2959 if (trie->bitmap && trie_type != trie_utf8_fold &&
2960 !TRIE_BITMAP_TEST(trie,*locinput)
2962 if (trie->states[ state ].wordnum) {
2964 PerlIO_printf(Perl_debug_log,
2965 "%*s %smatched empty string...%s\n",
2966 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2971 PerlIO_printf(Perl_debug_log,
2972 "%*s %sfailed to match trie start class...%s\n",
2973 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2980 U8 *uc = ( U8* )locinput;
2984 U8 *uscan = (U8*)NULL;
2986 SV *sv_accept_buff = NULL;
2987 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2989 ST.accepted = 0; /* how many accepting states we have seen */
2991 ST.jump = trie->jump;
2994 traverse the TRIE keeping track of all accepting states
2995 we transition through until we get to a failing node.
2998 while ( state && uc <= (U8*)PL_regeol ) {
2999 U32 base = trie->states[ state ].trans.base;
3002 /* We use charid to hold the wordnum as we don't use it
3003 for charid until after we have done the wordnum logic.
3004 We define an alias just so that the wordnum logic reads
3007 #define got_wordnum charid
3008 got_wordnum = trie->states[ state ].wordnum;
3010 if ( got_wordnum ) {
3011 if ( ! ST.accepted ) {
3013 SAVETMPS; /* XXX is this necessary? dmq */
3014 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
3015 sv_accept_buff=newSV(bufflen *
3016 sizeof(reg_trie_accepted) - 1);
3017 SvCUR_set(sv_accept_buff, 0);
3018 SvPOK_on(sv_accept_buff);
3019 sv_2mortal(sv_accept_buff);
3022 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
3025 if (ST.accepted >= bufflen) {
3027 ST.accept_buff =(reg_trie_accepted*)
3028 SvGROW(sv_accept_buff,
3029 bufflen * sizeof(reg_trie_accepted));
3031 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
3032 + sizeof(reg_trie_accepted));
3035 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
3036 ST.accept_buff[ST.accepted].endpos = uc;
3038 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
3042 DEBUG_TRIE_EXECUTE_r({
3043 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
3044 PerlIO_printf( Perl_debug_log,
3045 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
3046 2+depth * 2, "", PL_colors[4],
3047 (UV)state, (UV)ST.accepted );
3051 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3052 uscan, len, uvc, charid, foldlen,
3056 (base + charid > trie->uniquecharcount )
3057 && (base + charid - 1 - trie->uniquecharcount
3059 && trie->trans[base + charid - 1 -
3060 trie->uniquecharcount].check == state)
3062 state = trie->trans[base + charid - 1 -
3063 trie->uniquecharcount ].next;
3074 DEBUG_TRIE_EXECUTE_r(
3075 PerlIO_printf( Perl_debug_log,
3076 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3077 charid, uvc, (UV)state, PL_colors[5] );
3084 PerlIO_printf( Perl_debug_log,
3085 "%*s %sgot %"IVdf" possible matches%s\n",
3086 REPORT_CODE_OFF + depth * 2, "",
3087 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3090 goto trie_first_try; /* jump into the fail handler */
3092 case TRIE_next_fail: /* we failed - try next alterative */
3094 REGCP_UNWIND(ST.cp);
3095 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3096 PL_regoffs[n].end = -1;
3097 *PL_reglastparen = n;
3106 ST.lastparen = *PL_reglastparen;
3109 if ( ST.accepted == 1 ) {
3110 /* only one choice left - just continue */
3112 AV *const trie_words
3113 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3114 SV ** const tmp = av_fetch( trie_words,
3115 ST.accept_buff[ 0 ].wordnum-1, 0 );
3116 SV *sv= tmp ? sv_newmortal() : NULL;
3118 PerlIO_printf( Perl_debug_log,
3119 "%*s %sonly one match left: #%d <%s>%s\n",
3120 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3121 ST.accept_buff[ 0 ].wordnum,
3122 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3123 PL_colors[0], PL_colors[1],
3124 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3126 : "not compiled under -Dr",
3129 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
3130 /* in this case we free tmps/leave before we call regmatch
3131 as we wont be using accept_buff again. */
3133 locinput = PL_reginput;
3134 nextchr = UCHARAT(locinput);
3135 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
3138 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
3139 if (!has_cutgroup) {
3144 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3147 continue; /* execute rest of RE */
3150 if ( !ST.accepted-- ) {
3152 PerlIO_printf( Perl_debug_log,
3153 "%*s %sTRIE failed...%s\n",
3154 REPORT_CODE_OFF+depth*2, "",
3165 There are at least two accepting states left. Presumably
3166 the number of accepting states is going to be low,
3167 typically two. So we simply scan through to find the one
3168 with lowest wordnum. Once we find it, we swap the last
3169 state into its place and decrement the size. We then try to
3170 match the rest of the pattern at the point where the word
3171 ends. If we succeed, control just continues along the
3172 regex; if we fail we return here to try the next accepting
3179 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3180 DEBUG_TRIE_EXECUTE_r(
3181 PerlIO_printf( Perl_debug_log,
3182 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3183 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3184 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3185 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3188 if (ST.accept_buff[cur].wordnum <
3189 ST.accept_buff[best].wordnum)
3194 AV *const trie_words
3195 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3196 SV ** const tmp = av_fetch( trie_words,
3197 ST.accept_buff[ best ].wordnum - 1, 0 );
3198 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3200 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3201 SV *sv= tmp ? sv_newmortal() : NULL;
3203 PerlIO_printf( Perl_debug_log,
3204 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3205 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3206 ST.accept_buff[best].wordnum,
3207 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3208 PL_colors[0], PL_colors[1],
3209 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3210 ) : "not compiled under -Dr",
3211 REG_NODE_NUM(nextop),
3215 if ( best<ST.accepted ) {
3216 reg_trie_accepted tmp = ST.accept_buff[ best ];
3217 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3218 ST.accept_buff[ ST.accepted ] = tmp;
3221 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3222 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3225 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3227 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3232 /* we dont want to throw this away, see bug 57042*/
3233 if (oreplsv != GvSV(PL_replgv))
3234 sv_setsv(oreplsv, GvSV(PL_replgv));
3241 char *s = STRING(scan);
3243 if (do_utf8 != UTF) {
3244 /* The target and the pattern have differing utf8ness. */
3246 const char * const e = s + ln;
3249 /* The target is utf8, the pattern is not utf8. */
3254 if (NATIVE_TO_UNI(*(U8*)s) !=
3255 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3263 /* The target is not utf8, the pattern is utf8. */
3268 if (NATIVE_TO_UNI(*((U8*)l)) !=
3269 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3277 nextchr = UCHARAT(locinput);
3280 /* The target and the pattern have the same utf8ness. */
3281 /* Inline the first character, for speed. */
3282 if (UCHARAT(s) != nextchr)
3284 if (PL_regeol - locinput < ln)
3286 if (ln > 1 && memNE(s, locinput, ln))
3289 nextchr = UCHARAT(locinput);
3293 PL_reg_flags |= RF_tainted;
3296 char * const s = STRING(scan);
3299 if (do_utf8 || UTF) {
3300 /* Either target or the pattern are utf8. */
3301 const char * const l = locinput;
3302 char *e = PL_regeol;
3304 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3305 l, &e, 0, do_utf8)) {
3306 /* One more case for the sharp s:
3307 * pack("U0U*", 0xDF) =~ /ss/i,
3308 * the 0xC3 0x9F are the UTF-8
3309 * byte sequence for the U+00DF. */
3312 toLOWER(s[0]) == 's' &&
3314 toLOWER(s[1]) == 's' &&
3321 nextchr = UCHARAT(locinput);
3325 /* Neither the target and the pattern are utf8. */
3327 /* Inline the first character, for speed. */
3328 if (UCHARAT(s) != nextchr &&
3329 UCHARAT(s) != ((OP(scan) == EXACTF)
3330 ? PL_fold : PL_fold_locale)[nextchr])
3332 if (PL_regeol - locinput < ln)
3334 if (ln > 1 && (OP(scan) == EXACTF
3335 ? ibcmp(s, locinput, ln)
3336 : ibcmp_locale(s, locinput, ln)))
3339 nextchr = UCHARAT(locinput);
3344 STRLEN inclasslen = PL_regeol - locinput;
3346 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3348 if (locinput >= PL_regeol)
3350 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3351 nextchr = UCHARAT(locinput);
3356 nextchr = UCHARAT(locinput);
3357 if (!REGINCLASS(rex, scan, (U8*)locinput))
3359 if (!nextchr && locinput >= PL_regeol)
3361 nextchr = UCHARAT(++locinput);
3365 /* If we might have the case of the German sharp s
3366 * in a casefolding Unicode character class. */
3368 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3369 locinput += SHARP_S_SKIP;
3370 nextchr = UCHARAT(locinput);
3376 PL_reg_flags |= RF_tainted;
3382 LOAD_UTF8_CHARCLASS_ALNUM();
3383 if (!(OP(scan) == ALNUM
3384 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3385 : isALNUM_LC_utf8((U8*)locinput)))
3389 locinput += PL_utf8skip[nextchr];
3390 nextchr = UCHARAT(locinput);
3393 if (!(OP(scan) == ALNUM
3394 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3396 nextchr = UCHARAT(++locinput);
3399 PL_reg_flags |= RF_tainted;
3402 if (!nextchr && locinput >= PL_regeol)
3405 LOAD_UTF8_CHARCLASS_ALNUM();
3406 if (OP(scan) == NALNUM
3407 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3408 : isALNUM_LC_utf8((U8*)locinput))
3412 locinput += PL_utf8skip[nextchr];
3413 nextchr = UCHARAT(locinput);
3416 if (OP(scan) == NALNUM
3417 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3419 nextchr = UCHARAT(++locinput);
3423 PL_reg_flags |= RF_tainted;
3427 /* was last char in word? */
3429 if (locinput == PL_bostr)
3432 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3434 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3436 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3437 ln = isALNUM_uni(ln);
3438 LOAD_UTF8_CHARCLASS_ALNUM();
3439 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3442 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3443 n = isALNUM_LC_utf8((U8*)locinput);
3447 ln = (locinput != PL_bostr) ?
3448 UCHARAT(locinput - 1) : '\n';
3449 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3451 n = isALNUM(nextchr);
3454 ln = isALNUM_LC(ln);
3455 n = isALNUM_LC(nextchr);
3458 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3459 OP(scan) == BOUNDL))
3463 PL_reg_flags |= RF_tainted;
3469 if (UTF8_IS_CONTINUED(nextchr)) {
3470 LOAD_UTF8_CHARCLASS_SPACE();
3471 if (!(OP(scan) == SPACE
3472 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3473 : isSPACE_LC_utf8((U8*)locinput)))
3477 locinput += PL_utf8skip[nextchr];
3478 nextchr = UCHARAT(locinput);
3481 if (!(OP(scan) == SPACE
3482 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3484 nextchr = UCHARAT(++locinput);
3487 if (!(OP(scan) == SPACE
3488 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3490 nextchr = UCHARAT(++locinput);
3494 PL_reg_flags |= RF_tainted;
3497 if (!nextchr && locinput >= PL_regeol)
3500 LOAD_UTF8_CHARCLASS_SPACE();
3501 if (OP(scan) == NSPACE
3502 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3503 : isSPACE_LC_utf8((U8*)locinput))
3507 locinput += PL_utf8skip[nextchr];
3508 nextchr = UCHARAT(locinput);
3511 if (OP(scan) == NSPACE
3512 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3514 nextchr = UCHARAT(++locinput);
3517 PL_reg_flags |= RF_tainted;
3523 LOAD_UTF8_CHARCLASS_DIGIT();
3524 if (!(OP(scan) == DIGIT
3525 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3526 : isDIGIT_LC_utf8((U8*)locinput)))
3530 locinput += PL_utf8skip[nextchr];
3531 nextchr = UCHARAT(locinput);
3534 if (!(OP(scan) == DIGIT
3535 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3537 nextchr = UCHARAT(++locinput);
3540 PL_reg_flags |= RF_tainted;
3543 if (!nextchr && locinput >= PL_regeol)
3546 LOAD_UTF8_CHARCLASS_DIGIT();
3547 if (OP(scan) == NDIGIT
3548 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3549 : isDIGIT_LC_utf8((U8*)locinput))
3553 locinput += PL_utf8skip[nextchr];
3554 nextchr = UCHARAT(locinput);
3557 if (OP(scan) == NDIGIT
3558 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3560 nextchr = UCHARAT(++locinput);
3563 if (locinput >= PL_regeol)
3566 LOAD_UTF8_CHARCLASS_MARK();
3567 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3569 locinput += PL_utf8skip[nextchr];
3570 while (locinput < PL_regeol &&
3571 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3572 locinput += UTF8SKIP(locinput);
3573 if (locinput > PL_regeol)
3578 nextchr = UCHARAT(locinput);
3585 PL_reg_flags |= RF_tainted;
3590 n = reg_check_named_buff_matched(rex,scan);
3593 type = REF + ( type - NREF );
3600 PL_reg_flags |= RF_tainted;
3604 n = ARG(scan); /* which paren pair */
3607 ln = PL_regoffs[n].start;
3608 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3609 if (*PL_reglastparen < n || ln == -1)
3610 sayNO; /* Do not match unless seen CLOSEn. */
3611 if (ln == PL_regoffs[n].end)
3615 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3617 const char *e = PL_bostr + PL_regoffs[n].end;
3619 * Note that we can't do the "other character" lookup trick as
3620 * in the 8-bit case (no pun intended) because in Unicode we
3621 * have to map both upper and title case to lower case.
3625 STRLEN ulen1, ulen2;
3626 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3627 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3631 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3632 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3633 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3640 nextchr = UCHARAT(locinput);
3644 /* Inline the first character, for speed. */
3645 if (UCHARAT(s) != nextchr &&
3647 (UCHARAT(s) != (type == REFF
3648 ? PL_fold : PL_fold_locale)[nextchr])))
3650 ln = PL_regoffs[n].end - ln;
3651 if (locinput + ln > PL_regeol)
3653 if (ln > 1 && (type == REF
3654 ? memNE(s, locinput, ln)
3656 ? ibcmp(s, locinput, ln)
3657 : ibcmp_locale(s, locinput, ln))))
3660 nextchr = UCHARAT(locinput);
3670 #define ST st->u.eval
3674 regexp_internal *rei;
3675 regnode *startpoint;
3678 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
3679 if (cur_eval && cur_eval->locinput==locinput) {
3680 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3681 Perl_croak(aTHX_ "Infinite recursion in regex");
3682 if ( ++nochange_depth > max_nochange_depth )
3684 "Pattern subroutine nesting without pos change"
3685 " exceeded limit in regex");
3691 (void)ReREFCNT_inc(rex);
3692 if (OP(scan)==GOSUB) {
3693 startpoint = scan + ARG2L(scan);
3694 ST.close_paren = ARG(scan);
3696 startpoint = rei->program+1;
3699 goto eval_recurse_doit;
3701 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3702 if (cur_eval && cur_eval->locinput==locinput) {
3703 if ( ++nochange_depth > max_nochange_depth )
3704 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3709 /* execute the code in the {...} */
3711 SV ** const before = SP;
3712 OP_4tree * const oop = PL_op;
3713 COP * const ocurcop = PL_curcop;
3717 PL_op = (OP_4tree*)rexi->data->data[n];
3718 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3719 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3720 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3721 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
3724 SV *sv_mrk = get_sv("REGMARK", 1);
3725 sv_setsv(sv_mrk, sv_yes_mark);
3728 CALLRUNOPS(aTHX); /* Scalar context. */
3731 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3738 PAD_RESTORE_LOCAL(old_comppad);
3739 PL_curcop = ocurcop;
3742 sv_setsv(save_scalar(PL_replgv), ret);
3746 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3749 /* extract RE object from returned value; compiling if
3754 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3755 mg = mg_find(sv, PERL_MAGIC_qr);
3756 else if (SvSMAGICAL(ret)) {
3757 if (SvGMAGICAL(ret))
3758 sv_unmagic(ret, PERL_MAGIC_qr);
3760 mg = mg_find(ret, PERL_MAGIC_qr);
3764 re = reg_temp_copy((regexp *)mg->mg_obj); /*XXX:dmq*/
3768 const I32 osize = PL_regsize;
3770 if (DO_UTF8(ret)) pm_flags |= RXf_UTF8;
3771 re = CALLREGCOMP(ret, pm_flags);
3773 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3775 sv_magic(ret,MUTABLE_SV(ReREFCNT_inc(re)),
3780 RXp_MATCH_COPIED_off(re);
3781 re->subbeg = rex->subbeg;
3782 re->sublen = rex->sublen;
3785 debug_start_match(re, do_utf8, locinput, PL_regeol,
3786 "Matching embedded");
3788 startpoint = rei->program + 1;
3789 ST.close_paren = 0; /* only used for GOSUB */
3790 /* borrowed from regtry */
3791 if (PL_reg_start_tmpl <= re->nparens) {
3792 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3793 if(PL_reg_start_tmp)
3794 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3796 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3799 eval_recurse_doit: /* Share code with GOSUB below this line */
3800 /* run the pattern returned from (??{...}) */
3801 ST.cp = regcppush(0); /* Save *all* the positions. */
3802 REGCP_SET(ST.lastcp);
3804 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
3806 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
3807 PL_reglastparen = &re->lastparen;
3808 PL_reglastcloseparen = &re->lastcloseparen;
3810 re->lastcloseparen = 0;
3812 PL_reginput = locinput;
3815 /* XXXX This is too dramatic a measure... */
3818 ST.toggle_reg_flags = PL_reg_flags;
3820 PL_reg_flags |= RF_utf8;
3822 PL_reg_flags &= ~RF_utf8;
3823 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3826 ST.prev_curlyx = cur_curlyx;
3831 ST.prev_eval = cur_eval;
3833 /* now continue from first node in postoned RE */
3834 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3837 /* logical is 1, /(?(?{...})X|Y)/ */
3838 sw = (bool)SvTRUE(ret);
3843 case EVAL_AB: /* cleanup after a successful (??{A})B */
3844 /* note: this is called twice; first after popping B, then A */
3845 PL_reg_flags ^= ST.toggle_reg_flags;
3847 SETREX(rex,ST.prev_rex);
3848 rexi = RXi_GET(rex);
3850 cur_eval = ST.prev_eval;
3851 cur_curlyx = ST.prev_curlyx;
3853 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
3854 PL_reglastparen = &rex->lastparen;
3855 PL_reglastcloseparen = &rex->lastcloseparen;
3856 /* also update PL_regoffs */
3857 PL_regoffs = rex->offs;
3859 /* XXXX This is too dramatic a measure... */
3861 if ( nochange_depth )
3866 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3867 /* note: this is called twice; first after popping B, then A */
3868 PL_reg_flags ^= ST.toggle_reg_flags;
3870 SETREX(rex,ST.prev_rex);
3871 rexi = RXi_GET(rex);
3872 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
3873 PL_reglastparen = &rex->lastparen;
3874 PL_reglastcloseparen = &rex->lastcloseparen;
3876 PL_reginput = locinput;
3877 REGCP_UNWIND(ST.lastcp);
3879 cur_eval = ST.prev_eval;
3880 cur_curlyx = ST.prev_curlyx;
3881 /* XXXX This is too dramatic a measure... */
3883 if ( nochange_depth )
3889 n = ARG(scan); /* which paren pair */
3890 PL_reg_start_tmp[n] = locinput;
3896 n = ARG(scan); /* which paren pair */
3897 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
3898 PL_regoffs[n].end = locinput - PL_bostr;
3899 /*if (n > PL_regsize)
3901 if (n > *PL_reglastparen)
3902 *PL_reglastparen = n;
3903 *PL_reglastcloseparen = n;
3904 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3912 cursor && OP(cursor)!=END;
3913 cursor=regnext(cursor))
3915 if ( OP(cursor)==CLOSE ){
3917 if ( n <= lastopen ) {
3919 = PL_reg_start_tmp[n] - PL_bostr;
3920 PL_regoffs[n].end = locinput - PL_bostr;
3921 /*if (n > PL_regsize)
3923 if (n > *PL_reglastparen)
3924 *PL_reglastparen = n;
3925 *PL_reglastcloseparen = n;
3926 if ( n == ARG(scan) || (cur_eval &&
3927 cur_eval->u.eval.close_paren == n))
3936 n = ARG(scan); /* which paren pair */
3937 sw = (bool)(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
3940 /* reg_check_named_buff_matched returns 0 for no match */
3941 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3945 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3951 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3953 next = NEXTOPER(NEXTOPER(scan));
3955 next = scan + ARG(scan);
3956 if (OP(next) == IFTHEN) /* Fake one. */
3957 next = NEXTOPER(NEXTOPER(next));
3961 logical = scan->flags;
3964 /*******************************************************************
3966 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
3967 pattern, where A and B are subpatterns. (For simple A, CURLYM or
3968 STAR/PLUS/CURLY/CURLYN are used instead.)
3970 A*B is compiled as <CURLYX><A><WHILEM><B>
3972 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
3973 state, which contains the current count, initialised to -1. It also sets
3974 cur_curlyx to point to this state, with any previous value saved in the
3977 CURLYX then jumps straight to the WHILEM op, rather than executing A,
3978 since the pattern may possibly match zero times (i.e. it's a while {} loop
3979 rather than a do {} while loop).
3981 Each entry to WHILEM represents a successful match of A. The count in the
3982 CURLYX block is incremented, another WHILEM state is pushed, and execution
3983 passes to A or B depending on greediness and the current count.
3985 For example, if matching against the string a1a2a3b (where the aN are
3986 substrings that match /A/), then the match progresses as follows: (the
3987 pushed states are interspersed with the bits of strings matched so far):
3990 <CURLYX cnt=0><WHILEM>
3991 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
3992 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
3993 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
3994 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
3996 (Contrast this with something like CURLYM, which maintains only a single
4000 a1 <CURLYM cnt=1> a2
4001 a1 a2 <CURLYM cnt=2> a3
4002 a1 a2 a3 <CURLYM cnt=3> b
4005 Each WHILEM state block marks a point to backtrack to upon partial failure
4006 of A or B, and also contains some minor state data related to that
4007 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4008 overall state, such as the count, and pointers to the A and B ops.
4010 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4011 must always point to the *current* CURLYX block, the rules are:
4013 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4014 and set cur_curlyx to point the new block.
4016 When popping the CURLYX block after a successful or unsuccessful match,
4017 restore the previous cur_curlyx.
4019 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4020 to the outer one saved in the CURLYX block.
4022 When popping the WHILEM block after a successful or unsuccessful B match,
4023 restore the previous cur_curlyx.
4025 Here's an example for the pattern (AI* BI)*BO
4026 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4029 curlyx backtrack stack
4030 ------ ---------------
4032 CO <CO prev=NULL> <WO>
4033 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4034 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4035 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4037 At this point the pattern succeeds, and we work back down the stack to
4038 clean up, restoring as we go:
4040 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4041 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4042 CO <CO prev=NULL> <WO>
4045 *******************************************************************/
4047 #define ST st->u.curlyx
4049 case CURLYX: /* start of /A*B/ (for complex A) */
4051 /* No need to save/restore up to this paren */
4052 I32 parenfloor = scan->flags;
4054 assert(next); /* keep Coverity happy */
4055 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4058 /* XXXX Probably it is better to teach regpush to support
4059 parenfloor > PL_regsize... */
4060 if (parenfloor > (I32)*PL_reglastparen)
4061 parenfloor = *PL_reglastparen; /* Pessimization... */
4063 ST.prev_curlyx= cur_curlyx;
4065 ST.cp = PL_savestack_ix;
4067 /* these fields contain the state of the current curly.
4068 * they are accessed by subsequent WHILEMs */
4069 ST.parenfloor = parenfloor;
4070 ST.min = ARG1(scan);
4071 ST.max = ARG2(scan);
4072 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
4076 ST.count = -1; /* this will be updated by WHILEM */
4077 ST.lastloc = NULL; /* this will be updated by WHILEM */
4079 PL_reginput = locinput;
4080 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4084 case CURLYX_end: /* just finished matching all of A*B */
4085 cur_curlyx = ST.prev_curlyx;
4089 case CURLYX_end_fail: /* just failed to match all of A*B */
4091 cur_curlyx = ST.prev_curlyx;
4097 #define ST st->u.whilem
4099 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4101 /* see the discussion above about CURLYX/WHILEM */
4103 assert(cur_curlyx); /* keep Coverity happy */
4104 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4105 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4106 ST.cache_offset = 0;
4109 PL_reginput = locinput;
4111 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4112 "%*s whilem: matched %ld out of %ld..%ld\n",
4113 REPORT_CODE_OFF+depth*2, "", (long)n,
4114 (long)cur_curlyx->u.curlyx.min,
4115 (long)cur_curlyx->u.curlyx.max)
4118 /* First just match a string of min A's. */
4120 if (n < cur_curlyx->u.curlyx.min) {
4121 cur_curlyx->u.curlyx.lastloc = locinput;
4122 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
4126 /* If degenerate A matches "", assume A done. */
4128 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4129 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4130 "%*s whilem: empty match detected, trying continuation...\n",
4131 REPORT_CODE_OFF+depth*2, "")
4133 goto do_whilem_B_max;
4136 /* super-linear cache processing */
4140 if (!PL_reg_maxiter) {
4141 /* start the countdown: Postpone detection until we
4142 * know the match is not *that* much linear. */
4143 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4144 /* possible overflow for long strings and many CURLYX's */
4145 if (PL_reg_maxiter < 0)
4146 PL_reg_maxiter = I32_MAX;
4147 PL_reg_leftiter = PL_reg_maxiter;
4150 if (PL_reg_leftiter-- == 0) {
4151 /* initialise cache */
4152 const I32 size = (PL_reg_maxiter + 7)/8;
4153 if (PL_reg_poscache) {
4154 if ((I32)PL_reg_poscache_size < size) {
4155 Renew(PL_reg_poscache, size, char);
4156 PL_reg_poscache_size = size;
4158 Zero(PL_reg_poscache, size, char);
4161 PL_reg_poscache_size = size;
4162 Newxz(PL_reg_poscache, size, char);
4164 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4165 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4166 PL_colors[4], PL_colors[5])
4170 if (PL_reg_leftiter < 0) {
4171 /* have we already failed at this position? */
4173 offset = (scan->flags & 0xf) - 1
4174 + (locinput - PL_bostr) * (scan->flags>>4);
4175 mask = 1 << (offset % 8);
4177 if (PL_reg_poscache[offset] & mask) {
4178 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4179 "%*s whilem: (cache) already tried at this position...\n",
4180 REPORT_CODE_OFF+depth*2, "")
4182 sayNO; /* cache records failure */
4184 ST.cache_offset = offset;
4185 ST.cache_mask = mask;
4189 /* Prefer B over A for minimal matching. */
4191 if (cur_curlyx->u.curlyx.minmod) {
4192 ST.save_curlyx = cur_curlyx;
4193 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4194 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4195 REGCP_SET(ST.lastcp);
4196 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4200 /* Prefer A over B for maximal matching. */
4202 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
4203 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4204 cur_curlyx->u.curlyx.lastloc = locinput;
4205 REGCP_SET(ST.lastcp);
4206 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
4209 goto do_whilem_B_max;
4213 case WHILEM_B_min: /* just matched B in a minimal match */
4214 case WHILEM_B_max: /* just matched B in a maximal match */
4215 cur_curlyx = ST.save_curlyx;
4219 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4220 cur_curlyx = ST.save_curlyx;
4221 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4222 cur_curlyx->u.curlyx.count--;
4226 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4227 REGCP_UNWIND(ST.lastcp);
4230 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4231 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4232 cur_curlyx->u.curlyx.count--;
4236 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4237 REGCP_UNWIND(ST.lastcp);
4238 regcppop(rex); /* Restore some previous $<digit>s? */
4239 PL_reginput = locinput;
4240 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4241 "%*s whilem: failed, trying continuation...\n",
4242 REPORT_CODE_OFF+depth*2, "")
4245 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4246 && ckWARN(WARN_REGEXP)
4247 && !(PL_reg_flags & RF_warned))
4249 PL_reg_flags |= RF_warned;
4250 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4251 "Complex regular subexpression recursion",
4256 ST.save_curlyx = cur_curlyx;
4257 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4258 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4261 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4262 cur_curlyx = ST.save_curlyx;
4263 REGCP_UNWIND(ST.lastcp);
4266 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
4267 /* Maximum greed exceeded */
4268 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4269 && ckWARN(WARN_REGEXP)
4270 && !(PL_reg_flags & RF_warned))
4272 PL_reg_flags |= RF_warned;
4273 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4274 "%s limit (%d) exceeded",
4275 "Complex regular subexpression recursion",
4278 cur_curlyx->u.curlyx.count--;
4282 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4283 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4285 /* Try grabbing another A and see if it helps. */
4286 PL_reginput = locinput;
4287 cur_curlyx->u.curlyx.lastloc = locinput;
4288 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4289 REGCP_SET(ST.lastcp);
4290 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
4294 #define ST st->u.branch
4296 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4297 next = scan + ARG(scan);
4300 scan = NEXTOPER(scan);
4303 case BRANCH: /* /(...|A|...)/ */
4304 scan = NEXTOPER(scan); /* scan now points to inner node */
4305 ST.lastparen = *PL_reglastparen;
4306 ST.next_branch = next;
4308 PL_reginput = locinput;
4310 /* Now go into the branch */
4312 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4314 PUSH_STATE_GOTO(BRANCH_next, scan);
4318 PL_reginput = locinput;
4319 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4320 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4321 PUSH_STATE_GOTO(CUTGROUP_next,next);
4323 case CUTGROUP_next_fail:
4326 if (st->u.mark.mark_name)
4327 sv_commit = st->u.mark.mark_name;
4333 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4338 REGCP_UNWIND(ST.cp);
4339 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4340 PL_regoffs[n].end = -1;
4341 *PL_reglastparen = n;
4342 /*dmq: *PL_reglastcloseparen = n; */
4343 scan = ST.next_branch;
4344 /* no more branches? */
4345 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4347 PerlIO_printf( Perl_debug_log,
4348 "%*s %sBRANCH failed...%s\n",
4349 REPORT_CODE_OFF+depth*2, "",
4355 continue; /* execute next BRANCH[J] op */
4363 #define ST st->u.curlym
4365 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4367 /* This is an optimisation of CURLYX that enables us to push
4368 * only a single backtracking state, no matter how many matches
4369 * there are in {m,n}. It relies on the pattern being constant
4370 * length, with no parens to influence future backrefs
4374 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4376 /* if paren positive, emulate an OPEN/CLOSE around A */
4378 U32 paren = ST.me->flags;
4379 if (paren > PL_regsize)
4381 if (paren > *PL_reglastparen)
4382 *PL_reglastparen = paren;
4383 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4391 ST.c1 = CHRTEST_UNINIT;
4394 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4397 curlym_do_A: /* execute the A in /A{m,n}B/ */
4398 PL_reginput = locinput;
4399 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4402 case CURLYM_A: /* we've just matched an A */
4403 locinput = st->locinput;
4404 nextchr = UCHARAT(locinput);
4407 /* after first match, determine A's length: u.curlym.alen */
4408 if (ST.count == 1) {
4409 if (PL_reg_match_utf8) {
4411 while (s < PL_reginput) {
4417 ST.alen = PL_reginput - locinput;
4420 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4423 PerlIO_printf(Perl_debug_log,
4424 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4425 (int)(REPORT_CODE_OFF+(depth*2)), "",
4426 (IV) ST.count, (IV)ST.alen)
4429 locinput = PL_reginput;
4431 if (cur_eval && cur_eval->u.eval.close_paren &&
4432 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4436 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4437 if ( max == REG_INFTY || ST.count < max )
4438 goto curlym_do_A; /* try to match another A */
4440 goto curlym_do_B; /* try to match B */
4442 case CURLYM_A_fail: /* just failed to match an A */
4443 REGCP_UNWIND(ST.cp);
4445 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4446 || (cur_eval && cur_eval->u.eval.close_paren &&
4447 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4450 curlym_do_B: /* execute the B in /A{m,n}B/ */
4451 PL_reginput = locinput;
4452 if (ST.c1 == CHRTEST_UNINIT) {
4453 /* calculate c1 and c2 for possible match of 1st char
4454 * following curly */
4455 ST.c1 = ST.c2 = CHRTEST_VOID;
4456 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4457 regnode *text_node = ST.B;
4458 if (! HAS_TEXT(text_node))
4459 FIND_NEXT_IMPT(text_node);
4462 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4464 But the former is redundant in light of the latter.
4466 if this changes back then the macro for
4467 IS_TEXT and friends need to change.
4469 if (PL_regkind[OP(text_node)] == EXACT)
4472 ST.c1 = (U8)*STRING(text_node);
4474 (IS_TEXTF(text_node))
4476 : (IS_TEXTFL(text_node))
4477 ? PL_fold_locale[ST.c1]
4484 PerlIO_printf(Perl_debug_log,
4485 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4486 (int)(REPORT_CODE_OFF+(depth*2)),
4489 if (ST.c1 != CHRTEST_VOID
4490 && UCHARAT(PL_reginput) != ST.c1
4491 && UCHARAT(PL_reginput) != ST.c2)
4493 /* simulate B failing */
4495 PerlIO_printf(Perl_debug_log,
4496 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
4497 (int)(REPORT_CODE_OFF+(depth*2)),"",
4500 state_num = CURLYM_B_fail;
4501 goto reenter_switch;
4505 /* mark current A as captured */
4506 I32 paren = ST.me->flags;
4508 PL_regoffs[paren].start
4509 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4510 PL_regoffs[paren].end = PL_reginput - PL_bostr;
4511 /*dmq: *PL_reglastcloseparen = paren; */
4514 PL_regoffs[paren].end = -1;
4515 if (cur_eval && cur_eval->u.eval.close_paren &&
4516 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4525 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4528 case CURLYM_B_fail: /* just failed to match a B */
4529 REGCP_UNWIND(ST.cp);
4531 I32 max = ARG2(ST.me);
4532 if (max != REG_INFTY && ST.count == max)
4534 goto curlym_do_A; /* try to match a further A */
4536 /* backtrack one A */
4537 if (ST.count == ARG1(ST.me) /* min */)
4540 locinput = HOPc(locinput, -ST.alen);
4541 goto curlym_do_B; /* try to match B */
4544 #define ST st->u.curly
4546 #define CURLY_SETPAREN(paren, success) \
4549 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
4550 PL_regoffs[paren].end = locinput - PL_bostr; \
4551 *PL_reglastcloseparen = paren; \
4554 PL_regoffs[paren].end = -1; \
4557 case STAR: /* /A*B/ where A is width 1 */
4561 scan = NEXTOPER(scan);
4563 case PLUS: /* /A+B/ where A is width 1 */
4567 scan = NEXTOPER(scan);
4569 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4570 ST.paren = scan->flags; /* Which paren to set */
4571 if (ST.paren > PL_regsize)
4572 PL_regsize = ST.paren;
4573 if (ST.paren > *PL_reglastparen)
4574 *PL_reglastparen = ST.paren;
4575 ST.min = ARG1(scan); /* min to match */
4576 ST.max = ARG2(scan); /* max to match */
4577 if (cur_eval && cur_eval->u.eval.close_paren &&
4578 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4582 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4584 case CURLY: /* /A{m,n}B/ where A is width 1 */
4586 ST.min = ARG1(scan); /* min to match */
4587 ST.max = ARG2(scan); /* max to match */
4588 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4591 * Lookahead to avoid useless match attempts
4592 * when we know what character comes next.
4594 * Used to only do .*x and .*?x, but now it allows
4595 * for )'s, ('s and (?{ ... })'s to be in the way
4596 * of the quantifier and the EXACT-like node. -- japhy
4599 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4601 if (HAS_TEXT(next) || JUMPABLE(next)) {
4603 regnode *text_node = next;
4605 if (! HAS_TEXT(text_node))
4606 FIND_NEXT_IMPT(text_node);
4608 if (! HAS_TEXT(text_node))
4609 ST.c1 = ST.c2 = CHRTEST_VOID;
4611 if ( PL_regkind[OP(text_node)] != EXACT ) {
4612 ST.c1 = ST.c2 = CHRTEST_VOID;
4613 goto assume_ok_easy;
4616 s = (U8*)STRING(text_node);
4618 /* Currently we only get here when
4620 PL_rekind[OP(text_node)] == EXACT
4622 if this changes back then the macro for IS_TEXT and
4623 friends need to change. */
4626 if (IS_TEXTF(text_node))
4627 ST.c2 = PL_fold[ST.c1];
4628 else if (IS_TEXTFL(text_node))
4629 ST.c2 = PL_fold_locale[ST.c1];
4632 if (IS_TEXTF(text_node)) {
4633 STRLEN ulen1, ulen2;
4634 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4635 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4637 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4638 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4640 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4642 0 : UTF8_ALLOW_ANY);
4643 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4645 0 : UTF8_ALLOW_ANY);
4647 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4649 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4654 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4661 ST.c1 = ST.c2 = CHRTEST_VOID;
4666 PL_reginput = locinput;
4669 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4672 locinput = PL_reginput;
4674 if (ST.c1 == CHRTEST_VOID)
4675 goto curly_try_B_min;
4677 ST.oldloc = locinput;
4679 /* set ST.maxpos to the furthest point along the
4680 * string that could possibly match */
4681 if (ST.max == REG_INFTY) {
4682 ST.maxpos = PL_regeol - 1;
4684 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4688 int m = ST.max - ST.min;
4689 for (ST.maxpos = locinput;
4690 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4691 ST.maxpos += UTF8SKIP(ST.maxpos);
4694 ST.maxpos = locinput + ST.max - ST.min;
4695 if (ST.maxpos >= PL_regeol)
4696 ST.maxpos = PL_regeol - 1;
4698 goto curly_try_B_min_known;
4702 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4703 locinput = PL_reginput;
4704 if (ST.count < ST.min)
4706 if ((ST.count > ST.min)
4707 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4709 /* A{m,n} must come at the end of the string, there's
4710 * no point in backing off ... */
4712 /* ...except that $ and \Z can match before *and* after
4713 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4714 We may back off by one in this case. */
4715 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4719 goto curly_try_B_max;
4724 case CURLY_B_min_known_fail:
4725 /* failed to find B in a non-greedy match where c1,c2 valid */
4726 if (ST.paren && ST.count)
4727 PL_regoffs[ST.paren].end = -1;
4729 PL_reginput = locinput; /* Could be reset... */
4730 REGCP_UNWIND(ST.cp);
4731 /* Couldn't or didn't -- move forward. */
4732 ST.oldloc = locinput;
4734 locinput += UTF8SKIP(locinput);
4738 curly_try_B_min_known:
4739 /* find the next place where 'B' could work, then call B */
4743 n = (ST.oldloc == locinput) ? 0 : 1;
4744 if (ST.c1 == ST.c2) {
4746 /* set n to utf8_distance(oldloc, locinput) */
4747 while (locinput <= ST.maxpos &&
4748 utf8n_to_uvchr((U8*)locinput,
4749 UTF8_MAXBYTES, &len,
4750 uniflags) != (UV)ST.c1) {
4756 /* set n to utf8_distance(oldloc, locinput) */
4757 while (locinput <= ST.maxpos) {
4759 const UV c = utf8n_to_uvchr((U8*)locinput,
4760 UTF8_MAXBYTES, &len,
4762 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4770 if (ST.c1 == ST.c2) {
4771 while (locinput <= ST.maxpos &&
4772 UCHARAT(locinput) != ST.c1)
4776 while (locinput <= ST.maxpos
4777 && UCHARAT(locinput) != ST.c1
4778 && UCHARAT(locinput) != ST.c2)
4781 n = locinput - ST.oldloc;
4783 if (locinput > ST.maxpos)
4785 /* PL_reginput == oldloc now */
4788 if (regrepeat(rex, ST.A, n, depth) < n)
4791 PL_reginput = locinput;
4792 CURLY_SETPAREN(ST.paren, ST.count);
4793 if (cur_eval && cur_eval->u.eval.close_paren &&
4794 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4797 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4802 case CURLY_B_min_fail:
4803 /* failed to find B in a non-greedy match where c1,c2 invalid */
4804 if (ST.paren && ST.count)
4805 PL_regoffs[ST.paren].end = -1;
4807 REGCP_UNWIND(ST.cp);
4808 /* failed -- move forward one */
4809 PL_reginput = locinput;
4810 if (regrepeat(rex, ST.A, 1, depth)) {
4812 locinput = PL_reginput;
4813 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4814 ST.count > 0)) /* count overflow ? */
4817 CURLY_SETPAREN(ST.paren, ST.count);
4818 if (cur_eval && cur_eval->u.eval.close_paren &&
4819 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4822 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4830 /* a successful greedy match: now try to match B */
4831 if (cur_eval && cur_eval->u.eval.close_paren &&
4832 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4837 if (ST.c1 != CHRTEST_VOID)
4838 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4839 UTF8_MAXBYTES, 0, uniflags)
4840 : (UV) UCHARAT(PL_reginput);
4841 /* If it could work, try it. */
4842 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4843 CURLY_SETPAREN(ST.paren, ST.count);
4844 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4849 case CURLY_B_max_fail:
4850 /* failed to find B in a greedy match */
4851 if (ST.paren && ST.count)
4852 PL_regoffs[ST.paren].end = -1;
4854 REGCP_UNWIND(ST.cp);
4856 if (--ST.count < ST.min)
4858 PL_reginput = locinput = HOPc(locinput, -1);
4859 goto curly_try_B_max;
4866 /* we've just finished A in /(??{A})B/; now continue with B */
4868 st->u.eval.toggle_reg_flags
4869 = cur_eval->u.eval.toggle_reg_flags;
4870 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4872 st->u.eval.prev_rex = rex; /* inner */
4873 SETREX(rex,cur_eval->u.eval.prev_rex);
4874 rexi = RXi_GET(rex);
4875 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4877 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4879 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4880 PL_reglastparen = &rex->lastparen;
4881 PL_reglastcloseparen = &rex->lastcloseparen;
4883 REGCP_SET(st->u.eval.lastcp);
4884 PL_reginput = locinput;
4886 /* Restore parens of the outer rex without popping the
4888 tmpix = PL_savestack_ix;
4889 PL_savestack_ix = cur_eval->u.eval.lastcp;
4891 PL_savestack_ix = tmpix;
4893 st->u.eval.prev_eval = cur_eval;
4894 cur_eval = cur_eval->u.eval.prev_eval;
4896 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
4897 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
4898 if ( nochange_depth )
4901 PUSH_YES_STATE_GOTO(EVAL_AB,
4902 st->u.eval.prev_eval->u.eval.B); /* match B */
4905 if (locinput < reginfo->till) {
4906 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4907 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4909 (long)(locinput - PL_reg_starttry),
4910 (long)(reginfo->till - PL_reg_starttry),
4913 sayNO_SILENT; /* Cannot match: too short. */
4915 PL_reginput = locinput; /* put where regtry can find it */
4916 sayYES; /* Success! */
4918 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4920 PerlIO_printf(Perl_debug_log,
4921 "%*s %ssubpattern success...%s\n",
4922 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4923 PL_reginput = locinput; /* put where regtry can find it */
4924 sayYES; /* Success! */
4927 #define ST st->u.ifmatch
4929 case SUSPEND: /* (?>A) */
4931 PL_reginput = locinput;
4934 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4936 goto ifmatch_trivial_fail_test;
4938 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4940 ifmatch_trivial_fail_test:
4942 char * const s = HOPBACKc(locinput, scan->flags);
4947 sw = 1 - (bool)ST.wanted;
4951 next = scan + ARG(scan);
4959 PL_reginput = locinput;
4963 ST.logical = logical;
4964 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
4966 /* execute body of (?...A) */
4967 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4970 case IFMATCH_A_fail: /* body of (?...A) failed */
4971 ST.wanted = !ST.wanted;
4974 case IFMATCH_A: /* body of (?...A) succeeded */
4976 sw = (bool)ST.wanted;
4978 else if (!ST.wanted)
4981 if (OP(ST.me) == SUSPEND)
4982 locinput = PL_reginput;
4984 locinput = PL_reginput = st->locinput;
4985 nextchr = UCHARAT(locinput);
4987 scan = ST.me + ARG(ST.me);
4990 continue; /* execute B */
4995 next = scan + ARG(scan);
5000 reginfo->cutpoint = PL_regeol;
5003 PL_reginput = locinput;
5005 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5006 PUSH_STATE_GOTO(COMMIT_next,next);
5008 case COMMIT_next_fail:
5015 #define ST st->u.mark
5017 ST.prev_mark = mark_state;
5018 ST.mark_name = sv_commit = sv_yes_mark
5019 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5021 ST.mark_loc = PL_reginput = locinput;
5022 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5024 case MARKPOINT_next:
5025 mark_state = ST.prev_mark;
5028 case MARKPOINT_next_fail:
5029 if (popmark && sv_eq(ST.mark_name,popmark))
5031 if (ST.mark_loc > startpoint)
5032 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5033 popmark = NULL; /* we found our mark */
5034 sv_commit = ST.mark_name;
5037 PerlIO_printf(Perl_debug_log,
5038 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5039 REPORT_CODE_OFF+depth*2, "",
5040 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5043 mark_state = ST.prev_mark;
5044 sv_yes_mark = mark_state ?
5045 mark_state->u.mark.mark_name : NULL;
5049 PL_reginput = locinput;
5051 /* (*SKIP) : if we fail we cut here*/
5052 ST.mark_name = NULL;
5053 ST.mark_loc = locinput;
5054 PUSH_STATE_GOTO(SKIP_next,next);
5056 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5057 otherwise do nothing. Meaning we need to scan
5059 regmatch_state *cur = mark_state;
5060 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5063 if ( sv_eq( cur->u.mark.mark_name,
5066 ST.mark_name = find;
5067 PUSH_STATE_GOTO( SKIP_next, next );
5069 cur = cur->u.mark.prev_mark;
5072 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5074 case SKIP_next_fail:
5076 /* (*CUT:NAME) - Set up to search for the name as we
5077 collapse the stack*/
5078 popmark = ST.mark_name;
5080 /* (*CUT) - No name, we cut here.*/
5081 if (ST.mark_loc > startpoint)
5082 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5083 /* but we set sv_commit to latest mark_name if there
5084 is one so they can test to see how things lead to this
5087 sv_commit=mark_state->u.mark.mark_name;
5095 if ( n == (U32)what_len_TRICKYFOLD(locinput,do_utf8,ln) ) {
5097 } else if ( 0xDF == n && !do_utf8 && !UTF ) {
5100 U8 folded[UTF8_MAXBYTES_CASE+1];
5102 const char * const l = locinput;
5103 char *e = PL_regeol;
5104 to_uni_fold(n, folded, &foldlen);
5106 if (ibcmp_utf8((const char*) folded, 0, foldlen, 1,
5107 l, &e, 0, do_utf8)) {
5112 nextchr = UCHARAT(locinput);
5115 if ((n=is_LNBREAK(locinput,do_utf8))) {
5117 nextchr = UCHARAT(locinput);
5122 #define CASE_CLASS(nAmE) \
5124 if ((n=is_##nAmE(locinput,do_utf8))) { \
5126 nextchr = UCHARAT(locinput); \
5131 if ((n=is_##nAmE(locinput,do_utf8))) { \
5134 locinput += UTF8SKIP(locinput); \
5135 nextchr = UCHARAT(locinput); \
5140 CASE_CLASS(HORIZWS);
5144 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5145 PTR2UV(scan), OP(scan));
5146 Perl_croak(aTHX_ "regexp memory corruption");
5150 /* switch break jumps here */
5151 scan = next; /* prepare to execute the next op and ... */
5152 continue; /* ... jump back to the top, reusing st */
5156 /* push a state that backtracks on success */
5157 st->u.yes.prev_yes_state = yes_state;
5161 /* push a new regex state, then continue at scan */
5163 regmatch_state *newst;
5166 regmatch_state *cur = st;
5167 regmatch_state *curyes = yes_state;
5169 regmatch_slab *slab = PL_regmatch_slab;
5170 for (;curd > -1;cur--,curd--) {
5171 if (cur < SLAB_FIRST(slab)) {
5173 cur = SLAB_LAST(slab);
5175 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5176 REPORT_CODE_OFF + 2 + depth * 2,"",
5177 curd, PL_reg_name[cur->resume_state],
5178 (curyes == cur) ? "yes" : ""
5181 curyes = cur->u.yes.prev_yes_state;
5184 DEBUG_STATE_pp("push")
5187 st->locinput = locinput;
5189 if (newst > SLAB_LAST(PL_regmatch_slab))
5190 newst = S_push_slab(aTHX);
5191 PL_regmatch_state = newst;
5193 locinput = PL_reginput;
5194 nextchr = UCHARAT(locinput);
5202 * We get here only if there's trouble -- normally "case END" is
5203 * the terminating point.
5205 Perl_croak(aTHX_ "corrupted regexp pointers");
5211 /* we have successfully completed a subexpression, but we must now
5212 * pop to the state marked by yes_state and continue from there */
5213 assert(st != yes_state);
5215 while (st != yes_state) {
5217 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5218 PL_regmatch_slab = PL_regmatch_slab->prev;
5219 st = SLAB_LAST(PL_regmatch_slab);
5223 DEBUG_STATE_pp("pop (no final)");
5225 DEBUG_STATE_pp("pop (yes)");
5231 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5232 || yes_state > SLAB_LAST(PL_regmatch_slab))
5234 /* not in this slab, pop slab */
5235 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5236 PL_regmatch_slab = PL_regmatch_slab->prev;
5237 st = SLAB_LAST(PL_regmatch_slab);
5239 depth -= (st - yes_state);
5242 yes_state = st->u.yes.prev_yes_state;
5243 PL_regmatch_state = st;
5246 locinput= st->locinput;
5247 nextchr = UCHARAT(locinput);
5249 state_num = st->resume_state + no_final;
5250 goto reenter_switch;
5253 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5254 PL_colors[4], PL_colors[5]));
5256 if (PL_reg_eval_set) {
5257 /* each successfully executed (?{...}) block does the equivalent of
5258 * local $^R = do {...}
5259 * When popping the save stack, all these locals would be undone;
5260 * bypass this by setting the outermost saved $^R to the latest
5262 if (oreplsv != GvSV(PL_replgv))
5263 sv_setsv(oreplsv, GvSV(PL_replgv));
5270 PerlIO_printf(Perl_debug_log,
5271 "%*s %sfailed...%s\n",
5272 REPORT_CODE_OFF+depth*2, "",
5273 PL_colors[4], PL_colors[5])
5285 /* there's a previous state to backtrack to */
5287 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5288 PL_regmatch_slab = PL_regmatch_slab->prev;
5289 st = SLAB_LAST(PL_regmatch_slab);
5291 PL_regmatch_state = st;
5292 locinput= st->locinput;
5293 nextchr = UCHARAT(locinput);
5295 DEBUG_STATE_pp("pop");
5297 if (yes_state == st)
5298 yes_state = st->u.yes.prev_yes_state;
5300 state_num = st->resume_state + 1; /* failure = success + 1 */
5301 goto reenter_switch;
5306 if (rex->intflags & PREGf_VERBARG_SEEN) {
5307 SV *sv_err = get_sv("REGERROR", 1);
5308 SV *sv_mrk = get_sv("REGMARK", 1);
5310 sv_commit = &PL_sv_no;
5312 sv_yes_mark = &PL_sv_yes;
5315 sv_commit = &PL_sv_yes;
5316 sv_yes_mark = &PL_sv_no;
5318 sv_setsv(sv_err, sv_commit);
5319 sv_setsv(sv_mrk, sv_yes_mark);
5322 /* clean up; in particular, free all slabs above current one */
5323 LEAVE_SCOPE(oldsave);
5329 - regrepeat - repeatedly match something simple, report how many
5332 * [This routine now assumes that it will only match on things of length 1.
5333 * That was true before, but now we assume scan - reginput is the count,
5334 * rather than incrementing count on every character. [Er, except utf8.]]
5337 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5340 register char *scan;
5342 register char *loceol = PL_regeol;
5343 register I32 hardcount = 0;
5344 register bool do_utf8 = PL_reg_match_utf8;
5346 PERL_UNUSED_ARG(depth);
5349 PERL_ARGS_ASSERT_REGREPEAT;
5352 if (max == REG_INFTY)
5354 else if (max < loceol - scan)
5355 loceol = scan + max;
5360 while (scan < loceol && hardcount < max && *scan != '\n') {
5361 scan += UTF8SKIP(scan);
5365 while (scan < loceol && *scan != '\n')
5372 while (scan < loceol && hardcount < max) {
5373 scan += UTF8SKIP(scan);
5383 case EXACT: /* length of string is 1 */
5385 while (scan < loceol && UCHARAT(scan) == c)
5388 case EXACTF: /* length of string is 1 */
5390 while (scan < loceol &&
5391 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5394 case EXACTFL: /* length of string is 1 */
5395 PL_reg_flags |= RF_tainted;
5397 while (scan < loceol &&
5398 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5404 while (hardcount < max && scan < loceol &&
5405 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
5406 scan += UTF8SKIP(scan);
5410 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5417 LOAD_UTF8_CHARCLASS_ALNUM();
5418 while (hardcount < max && scan < loceol &&
5419 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5420 scan += UTF8SKIP(scan);
5424 while (scan < loceol && isALNUM(*scan))
5429 PL_reg_flags |= RF_tainted;
5432 while (hardcount < max && scan < loceol &&
5433 isALNUM_LC_utf8((U8*)scan)) {
5434 scan += UTF8SKIP(scan);
5438 while (scan < loceol && isALNUM_LC(*scan))
5445 LOAD_UTF8_CHARCLASS_ALNUM();
5446 while (hardcount < max && scan < loceol &&
5447 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5448 scan += UTF8SKIP(scan);
5452 while (scan < loceol && !isALNUM(*scan))
5457 PL_reg_flags |= RF_tainted;
5460 while (hardcount < max && scan < loceol &&
5461 !isALNUM_LC_utf8((U8*)scan)) {
5462 scan += UTF8SKIP(scan);
5466 while (scan < loceol && !isALNUM_LC(*scan))
5473 LOAD_UTF8_CHARCLASS_SPACE();
5474 while (hardcount < max && scan < loceol &&
5476 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5477 scan += UTF8SKIP(scan);
5481 while (scan < loceol && isSPACE(*scan))
5486 PL_reg_flags |= RF_tainted;
5489 while (hardcount < max && scan < loceol &&
5490 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5491 scan += UTF8SKIP(scan);
5495 while (scan < loceol && isSPACE_LC(*scan))
5502 LOAD_UTF8_CHARCLASS_SPACE();
5503 while (hardcount < max && scan < loceol &&
5505 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5506 scan += UTF8SKIP(scan);
5510 while (scan < loceol && !isSPACE(*scan))
5515 PL_reg_flags |= RF_tainted;
5518 while (hardcount < max && scan < loceol &&
5519 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5520 scan += UTF8SKIP(scan);
5524 while (scan < loceol && !isSPACE_LC(*scan))
5531 LOAD_UTF8_CHARCLASS_DIGIT();
5532 while (hardcount < max && scan < loceol &&
5533 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5534 scan += UTF8SKIP(scan);
5538 while (scan < loceol && isDIGIT(*scan))
5545 LOAD_UTF8_CHARCLASS_DIGIT();
5546 while (hardcount < max && scan < loceol &&
5547 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5548 scan += UTF8SKIP(scan);
5552 while (scan < loceol && !isDIGIT(*scan))
5558 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
5564 LNBREAK can match two latin chars, which is ok,
5565 because we have a null terminated string, but we
5566 have to use hardcount in this situation
5568 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
5577 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
5582 while (scan < loceol && is_HORIZWS_latin1(scan))
5589 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
5590 scan += UTF8SKIP(scan);
5594 while (scan < loceol && !is_HORIZWS_latin1(scan))
5602 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
5607 while (scan < loceol && is_VERTWS_latin1(scan))
5615 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
5616 scan += UTF8SKIP(scan);
5620 while (scan < loceol && !is_VERTWS_latin1(scan))
5626 default: /* Called on something of 0 width. */
5627 break; /* So match right here or not at all. */
5633 c = scan - PL_reginput;
5637 GET_RE_DEBUG_FLAGS_DECL;
5639 SV * const prop = sv_newmortal();
5640 regprop(prog, prop, p);
5641 PerlIO_printf(Perl_debug_log,
5642 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5643 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5651 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5653 - regclass_swash - prepare the utf8 swash
5657 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5663 RXi_GET_DECL(prog,progi);
5664 const struct reg_data * const data = prog ? progi->data : NULL;
5666 PERL_ARGS_ASSERT_REGCLASS_SWASH;
5668 if (data && data->count) {
5669 const U32 n = ARG(node);
5671 if (data->what[n] == 's') {
5672 SV * const rv = MUTABLE_SV(data->data[n]);
5673 AV * const av = MUTABLE_AV(SvRV(rv));
5674 SV **const ary = AvARRAY(av);
5677 /* See the end of regcomp.c:S_regclass() for
5678 * documentation of these array elements. */
5681 a = SvROK(ary[1]) ? &ary[1] : NULL;
5682 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
5686 else if (si && doinit) {
5687 sw = swash_init("utf8", "", si, 1, 0);
5688 (void)av_store(av, 1, sw);
5705 - reginclass - determine if a character falls into a character class
5707 The n is the ANYOF regnode, the p is the target string, lenp
5708 is pointer to the maximum length of how far to go in the p
5709 (if the lenp is zero, UTF8SKIP(p) is used),
5710 do_utf8 tells whether the target string is in UTF-8.
5715 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5718 const char flags = ANYOF_FLAGS(n);
5724 PERL_ARGS_ASSERT_REGINCLASS;
5726 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5727 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5728 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5729 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5730 if (len == (STRLEN)-1)
5731 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5734 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5735 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5738 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5739 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5742 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5746 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5754 utf8_p = bytes_to_utf8(p, &len);
5756 if (swash_fetch(sw, utf8_p, 1))
5758 else if (flags & ANYOF_FOLD) {
5759 if (!match && lenp && av) {
5761 for (i = 0; i <= av_len(av); i++) {
5762 SV* const sv = *av_fetch(av, i, FALSE);
5764 const char * const s = SvPV_const(sv, len);
5765 if (len <= plen && memEQ(s, (char*)utf8_p, len)) {
5773 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5776 to_utf8_fold(utf8_p, tmpbuf, &tmplen);
5777 if (swash_fetch(sw, tmpbuf, 1))
5782 /* If we allocated a string above, free it */
5783 if (! do_utf8) Safefree(utf8_p);
5786 if (match && lenp && *lenp == 0)
5787 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5789 if (!match && c < 256) {
5790 if (ANYOF_BITMAP_TEST(n, c))
5792 else if (flags & ANYOF_FOLD) {
5795 if (flags & ANYOF_LOCALE) {
5796 PL_reg_flags |= RF_tainted;
5797 f = PL_fold_locale[c];
5801 if (f != c && ANYOF_BITMAP_TEST(n, f))
5805 if (!match && (flags & ANYOF_CLASS)) {
5806 PL_reg_flags |= RF_tainted;
5808 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5809 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5810 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5811 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5812 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5813 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5814 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5815 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5816 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5817 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5818 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5819 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5820 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5821 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5822 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5823 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5824 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5825 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5826 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5827 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5828 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5829 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5830 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5831 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5832 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5833 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5834 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5835 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5836 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5837 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5838 ) /* How's that for a conditional? */
5845 return (flags & ANYOF_INVERT) ? !match : match;
5849 S_reghop3(U8 *s, I32 off, const U8* lim)
5853 PERL_ARGS_ASSERT_REGHOP3;
5856 while (off-- && s < lim) {
5857 /* XXX could check well-formedness here */
5862 while (off++ && s > lim) {
5864 if (UTF8_IS_CONTINUED(*s)) {
5865 while (s > lim && UTF8_IS_CONTINUATION(*s))
5868 /* XXX could check well-formedness here */
5875 /* there are a bunch of places where we use two reghop3's that should
5876 be replaced with this routine. but since thats not done yet
5877 we ifdef it out - dmq
5880 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5884 PERL_ARGS_ASSERT_REGHOP4;
5887 while (off-- && s < rlim) {
5888 /* XXX could check well-formedness here */
5893 while (off++ && s > llim) {
5895 if (UTF8_IS_CONTINUED(*s)) {
5896 while (s > llim && UTF8_IS_CONTINUATION(*s))
5899 /* XXX could check well-formedness here */
5907 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5911 PERL_ARGS_ASSERT_REGHOPMAYBE3;
5914 while (off-- && s < lim) {
5915 /* XXX could check well-formedness here */
5922 while (off++ && s > lim) {
5924 if (UTF8_IS_CONTINUED(*s)) {
5925 while (s > lim && UTF8_IS_CONTINUATION(*s))
5928 /* XXX could check well-formedness here */
5937 restore_pos(pTHX_ void *arg)
5940 regexp * const rex = (regexp *)arg;
5941 if (PL_reg_eval_set) {
5942 if (PL_reg_oldsaved) {
5943 rex->subbeg = PL_reg_oldsaved;
5944 rex->sublen = PL_reg_oldsavedlen;
5945 #ifdef PERL_OLD_COPY_ON_WRITE
5946 rex->saved_copy = PL_nrs;
5948 RXp_MATCH_COPIED_on(rex);
5950 PL_reg_magic->mg_len = PL_reg_oldpos;
5951 PL_reg_eval_set = 0;
5952 PL_curpm = PL_reg_oldcurpm;
5957 S_to_utf8_substr(pTHX_ register regexp *prog)
5961 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
5964 if (prog->substrs->data[i].substr
5965 && !prog->substrs->data[i].utf8_substr) {
5966 SV* const sv = newSVsv(prog->substrs->data[i].substr);
5967 prog->substrs->data[i].utf8_substr = sv;
5968 sv_utf8_upgrade(sv);
5969 if (SvVALID(prog->substrs->data[i].substr)) {
5970 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
5971 if (flags & FBMcf_TAIL) {
5972 /* Trim the trailing \n that fbm_compile added last
5974 SvCUR_set(sv, SvCUR(sv) - 1);
5975 /* Whilst this makes the SV technically "invalid" (as its
5976 buffer is no longer followed by "\0") when fbm_compile()
5977 adds the "\n" back, a "\0" is restored. */
5979 fbm_compile(sv, flags);
5981 if (prog->substrs->data[i].substr == prog->check_substr)
5982 prog->check_utf8 = sv;
5988 S_to_byte_substr(pTHX_ register regexp *prog)
5993 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
5996 if (prog->substrs->data[i].utf8_substr
5997 && !prog->substrs->data[i].substr) {
5998 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
5999 if (sv_utf8_downgrade(sv, TRUE)) {
6000 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6002 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6003 if (flags & FBMcf_TAIL) {
6004 /* Trim the trailing \n that fbm_compile added last
6006 SvCUR_set(sv, SvCUR(sv) - 1);
6008 fbm_compile(sv, flags);
6014 prog->substrs->data[i].substr = sv;
6015 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6016 prog->check_substr = sv;
6023 * c-indentation-style: bsd
6025 * indent-tabs-mode: t
6028 * ex: set ts=8 sts=4 sw=4 noet: