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
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
103 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
104 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
106 #define HOPc(pos,off) \
107 (char *)(PL_reg_match_utf8 \
108 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
110 #define HOPBACKc(pos, off) \
111 (char*)(PL_reg_match_utf8\
112 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
113 : (pos - off >= PL_bostr) \
117 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
118 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
120 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
121 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
122 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
123 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
124 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
125 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
127 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
129 /* for use after a quantifier and before an EXACT-like node -- japhy */
130 /* it would be nice to rework regcomp.sym to generate this stuff. sigh */
131 #define JUMPABLE(rn) ( \
133 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
135 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
136 OP(rn) == PLUS || OP(rn) == MINMOD || \
137 OP(rn) == KEEPS || (PL_regkind[OP(rn)] == VERB) || \
138 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
140 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
142 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
145 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
146 we don't need this definition. */
147 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
148 #define IS_TEXTF(rn) ( OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
149 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
152 /* ... so we use this as its faster. */
153 #define IS_TEXT(rn) ( OP(rn)==EXACT )
154 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
155 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
160 Search for mandatory following text node; for lookahead, the text must
161 follow but for lookbehind (rn->flags != 0) we skip to the next step.
163 #define FIND_NEXT_IMPT(rn) STMT_START { \
164 while (JUMPABLE(rn)) { \
165 const OPCODE type = OP(rn); \
166 if (type == SUSPEND || PL_regkind[type] == CURLY) \
167 rn = NEXTOPER(NEXTOPER(rn)); \
168 else if (type == PLUS) \
170 else if (type == IFMATCH) \
171 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
172 else rn += NEXT_OFF(rn); \
177 static void restore_pos(pTHX_ void *arg);
180 S_regcppush(pTHX_ I32 parenfloor)
183 const int retval = PL_savestack_ix;
184 #define REGCP_PAREN_ELEMS 4
185 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
187 GET_RE_DEBUG_FLAGS_DECL;
189 if (paren_elems_to_push < 0)
190 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
192 #define REGCP_OTHER_ELEMS 7
193 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
195 for (p = PL_regsize; p > parenfloor; p--) {
196 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
197 SSPUSHINT(PL_regoffs[p].end);
198 SSPUSHINT(PL_regoffs[p].start);
199 SSPUSHPTR(PL_reg_start_tmp[p]);
201 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
202 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
203 (UV)p, (IV)PL_regoffs[p].start,
204 (IV)(PL_reg_start_tmp[p] - PL_bostr),
205 (IV)PL_regoffs[p].end
208 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
209 SSPUSHPTR(PL_regoffs);
210 SSPUSHINT(PL_regsize);
211 SSPUSHINT(*PL_reglastparen);
212 SSPUSHINT(*PL_reglastcloseparen);
213 SSPUSHPTR(PL_reginput);
214 #define REGCP_FRAME_ELEMS 2
215 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
216 * are needed for the regexp context stack bookkeeping. */
217 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
218 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
223 /* These are needed since we do not localize EVAL nodes: */
224 #define REGCP_SET(cp) \
226 PerlIO_printf(Perl_debug_log, \
227 " Setting an EVAL scope, savestack=%"IVdf"\n", \
228 (IV)PL_savestack_ix)); \
231 #define REGCP_UNWIND(cp) \
233 if (cp != PL_savestack_ix) \
234 PerlIO_printf(Perl_debug_log, \
235 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
236 (IV)(cp), (IV)PL_savestack_ix)); \
240 S_regcppop(pTHX_ const regexp *rex)
245 GET_RE_DEBUG_FLAGS_DECL;
247 PERL_ARGS_ASSERT_REGCPPOP;
249 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
251 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
252 i = SSPOPINT; /* Parentheses elements to pop. */
253 input = (char *) SSPOPPTR;
254 *PL_reglastcloseparen = SSPOPINT;
255 *PL_reglastparen = SSPOPINT;
256 PL_regsize = SSPOPINT;
257 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
260 /* Now restore the parentheses context. */
261 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
262 i > 0; i -= REGCP_PAREN_ELEMS) {
264 U32 paren = (U32)SSPOPINT;
265 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
266 PL_regoffs[paren].start = SSPOPINT;
268 if (paren <= *PL_reglastparen)
269 PL_regoffs[paren].end = tmps;
271 PerlIO_printf(Perl_debug_log,
272 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
273 (UV)paren, (IV)PL_regoffs[paren].start,
274 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
275 (IV)PL_regoffs[paren].end,
276 (paren > *PL_reglastparen ? "(no)" : ""));
280 if (*PL_reglastparen + 1 <= rex->nparens) {
281 PerlIO_printf(Perl_debug_log,
282 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
283 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
287 /* It would seem that the similar code in regtry()
288 * already takes care of this, and in fact it is in
289 * a better location to since this code can #if 0-ed out
290 * but the code in regtry() is needed or otherwise tests
291 * requiring null fields (pat.t#187 and split.t#{13,14}
292 * (as of patchlevel 7877) will fail. Then again,
293 * this code seems to be necessary or otherwise
294 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
295 * --jhi updated by dapm */
296 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
298 PL_regoffs[i].start = -1;
299 PL_regoffs[i].end = -1;
305 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
308 * pregexec and friends
311 #ifndef PERL_IN_XSUB_RE
313 - pregexec - match a regexp against a string
316 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
317 char *strbeg, I32 minend, SV *screamer, U32 nosave)
318 /* strend: pointer to null at end of string */
319 /* strbeg: real beginning of string */
320 /* minend: end of match must be >=minend after stringarg. */
321 /* nosave: For optimizations. */
323 PERL_ARGS_ASSERT_PREGEXEC;
326 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
327 nosave ? 0 : REXEC_COPY_STR);
332 * Need to implement the following flags for reg_anch:
334 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
336 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
337 * INTUIT_AUTORITATIVE_ML
338 * INTUIT_ONCE_NOML - Intuit can match in one location only.
341 * Another flag for this function: SECOND_TIME (so that float substrs
342 * with giant delta may be not rechecked).
345 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
347 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
348 Otherwise, only SvCUR(sv) is used to get strbeg. */
350 /* XXXX We assume that strpos is strbeg unless sv. */
352 /* XXXX Some places assume that there is a fixed substring.
353 An update may be needed if optimizer marks as "INTUITable"
354 RExen without fixed substrings. Similarly, it is assumed that
355 lengths of all the strings are no more than minlen, thus they
356 cannot come from lookahead.
357 (Or minlen should take into account lookahead.)
358 NOTE: Some of this comment is not correct. minlen does now take account
359 of lookahead/behind. Further research is required. -- demerphq
363 /* A failure to find a constant substring means that there is no need to make
364 an expensive call to REx engine, thus we celebrate a failure. Similarly,
365 finding a substring too deep into the string means that less calls to
366 regtry() should be needed.
368 REx compiler's optimizer found 4 possible hints:
369 a) Anchored substring;
371 c) Whether we are anchored (beginning-of-line or \G);
372 d) First node (of those at offset 0) which may distingush positions;
373 We use a)b)d) and multiline-part of c), and try to find a position in the
374 string which does not contradict any of them.
377 /* Most of decisions we do here should have been done at compile time.
378 The nodes of the REx which we used for the search should have been
379 deleted from the finite automaton. */
382 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
383 char *strend, const U32 flags, re_scream_pos_data *data)
386 struct regexp *const prog = (struct regexp *)SvANY(rx);
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(rx,do_utf8);
410 PL_reg_flags |= RF_utf8;
413 debug_start_match(rx, 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 rx, 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 struct regexp *const prog = (struct regexp *)SvANY(rx);
1776 /*register*/ char *s;
1777 register regnode *c;
1778 /*register*/ char *startpos = stringarg;
1779 I32 minlen; /* must match at least this many chars */
1780 I32 dontbother = 0; /* how many characters not to try at end */
1781 I32 end_shift = 0; /* Same for the end. */ /* CC */
1782 I32 scream_pos = -1; /* Internal iterator of scream. */
1783 char *scream_olds = NULL;
1784 const bool do_utf8 = (bool)DO_UTF8(sv);
1786 RXi_GET_DECL(prog,progi);
1787 regmatch_info reginfo; /* create some info to pass to regtry etc */
1788 bool swap_on_fail = 0;
1789 GET_RE_DEBUG_FLAGS_DECL;
1791 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1792 PERL_UNUSED_ARG(data);
1794 /* Be paranoid... */
1795 if (prog == NULL || startpos == NULL) {
1796 Perl_croak(aTHX_ "NULL regexp parameter");
1800 multiline = prog->extflags & RXf_PMf_MULTILINE;
1801 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
1803 RX_MATCH_UTF8_set(rx, do_utf8);
1805 debug_start_match(rx, do_utf8, startpos, strend,
1809 minlen = prog->minlen;
1811 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1812 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1813 "String too short [regexec_flags]...\n"));
1818 /* Check validity of program. */
1819 if (UCHARAT(progi->program) != REG_MAGIC) {
1820 Perl_croak(aTHX_ "corrupted regexp program");
1824 PL_reg_eval_set = 0;
1828 PL_reg_flags |= RF_utf8;
1830 /* Mark beginning of line for ^ and lookbehind. */
1831 reginfo.bol = startpos; /* XXX not used ??? */
1835 /* Mark end of line for $ (and such) */
1838 /* see how far we have to get to not match where we matched before */
1839 reginfo.till = startpos+minend;
1841 /* If there is a "must appear" string, look for it. */
1844 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
1847 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1848 reginfo.ganch = startpos + prog->gofs;
1849 else if (sv && SvTYPE(sv) >= SVt_PVMG
1851 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1852 && mg->mg_len >= 0) {
1853 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1854 if (prog->extflags & RXf_ANCH_GPOS) {
1855 if (s > reginfo.ganch)
1857 s = reginfo.ganch - prog->gofs;
1861 reginfo.ganch = strbeg + PTR2UV(data);
1862 } else /* pos() not defined */
1863 reginfo.ganch = strbeg;
1865 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
1867 swap_match_buff(prog); /* do we need a save destructor here for
1870 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1871 re_scream_pos_data d;
1873 d.scream_olds = &scream_olds;
1874 d.scream_pos = &scream_pos;
1875 s = re_intuit_start(rx, sv, s, strend, flags, &d);
1877 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1878 goto phooey; /* not present */
1884 /* Simplest case: anchored match need be tried only once. */
1885 /* [unless only anchor is BOL and multiline is set] */
1886 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
1887 if (s == startpos && regtry(®info, &startpos))
1889 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
1890 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
1895 dontbother = minlen - 1;
1896 end = HOP3c(strend, -dontbother, strbeg) - 1;
1897 /* for multiline we only have to try after newlines */
1898 if (prog->check_substr || prog->check_utf8) {
1902 if (regtry(®info, &s))
1907 if (prog->extflags & RXf_USE_INTUIT) {
1908 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
1919 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1920 if (regtry(®info, &s))
1927 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
1929 /* the warning about reginfo.ganch being used without intialization
1930 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
1931 and we only enter this block when the same bit is set. */
1932 char *tmp_s = reginfo.ganch - prog->gofs;
1933 if (regtry(®info, &tmp_s))
1938 /* Messy cases: unanchored match. */
1939 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
1940 /* we have /x+whatever/ */
1941 /* it must be a one character string (XXXX Except UTF?) */
1946 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1947 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1948 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1953 DEBUG_EXECUTE_r( did_match = 1 );
1954 if (regtry(®info, &s)) goto got_it;
1956 while (s < strend && *s == ch)
1964 DEBUG_EXECUTE_r( did_match = 1 );
1965 if (regtry(®info, &s)) goto got_it;
1967 while (s < strend && *s == ch)
1972 DEBUG_EXECUTE_r(if (!did_match)
1973 PerlIO_printf(Perl_debug_log,
1974 "Did not find anchored character...\n")
1977 else if (prog->anchored_substr != NULL
1978 || prog->anchored_utf8 != NULL
1979 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1980 && prog->float_max_offset < strend - s)) {
1985 char *last1; /* Last position checked before */
1989 if (prog->anchored_substr || prog->anchored_utf8) {
1990 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1991 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1992 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1993 back_max = back_min = prog->anchored_offset;
1995 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1996 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1997 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1998 back_max = prog->float_max_offset;
1999 back_min = prog->float_min_offset;
2003 if (must == &PL_sv_undef)
2004 /* could not downgrade utf8 check substring, so must fail */
2010 last = HOP3c(strend, /* Cannot start after this */
2011 -(I32)(CHR_SVLEN(must)
2012 - (SvTAIL(must) != 0) + back_min), strbeg);
2015 last1 = HOPc(s, -1);
2017 last1 = s - 1; /* bogus */
2019 /* XXXX check_substr already used to find "s", can optimize if
2020 check_substr==must. */
2022 dontbother = end_shift;
2023 strend = HOPc(strend, -dontbother);
2024 while ( (s <= last) &&
2025 ((flags & REXEC_SCREAM)
2026 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2027 end_shift, &scream_pos, 0))
2028 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2029 (unsigned char*)strend, must,
2030 multiline ? FBMrf_MULTILINE : 0))) ) {
2031 /* we may be pointing at the wrong string */
2032 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2033 s = strbeg + (s - SvPVX_const(sv));
2034 DEBUG_EXECUTE_r( did_match = 1 );
2035 if (HOPc(s, -back_max) > last1) {
2036 last1 = HOPc(s, -back_min);
2037 s = HOPc(s, -back_max);
2040 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2042 last1 = HOPc(s, -back_min);
2046 while (s <= last1) {
2047 if (regtry(®info, &s))
2053 while (s <= last1) {
2054 if (regtry(®info, &s))
2060 DEBUG_EXECUTE_r(if (!did_match) {
2061 RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0),
2062 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2063 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2064 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2065 ? "anchored" : "floating"),
2066 quoted, RE_SV_TAIL(must));
2070 else if ( (c = progi->regstclass) ) {
2072 const OPCODE op = OP(progi->regstclass);
2073 /* don't bother with what can't match */
2074 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2075 strend = HOPc(strend, -(minlen - 1));
2078 SV * const prop = sv_newmortal();
2079 regprop(prog, prop, c);
2081 RE_PV_QUOTED_DECL(quoted,do_utf8,PERL_DEBUG_PAD_ZERO(1),
2083 PerlIO_printf(Perl_debug_log,
2084 "Matching stclass %.*s against %s (%d chars)\n",
2085 (int)SvCUR(prop), SvPVX_const(prop),
2086 quoted, (int)(strend - s));
2089 if (find_byclass(prog, c, s, strend, ®info))
2091 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2095 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2100 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
2101 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
2102 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
2104 if (flags & REXEC_SCREAM) {
2105 last = screaminstr(sv, float_real, s - strbeg,
2106 end_shift, &scream_pos, 1); /* last one */
2108 last = scream_olds; /* Only one occurrence. */
2109 /* we may be pointing at the wrong string */
2110 else if (RXp_MATCH_COPIED(prog))
2111 s = strbeg + (s - SvPVX_const(sv));
2115 const char * const little = SvPV_const(float_real, len);
2117 if (SvTAIL(float_real)) {
2118 if (memEQ(strend - len + 1, little, len - 1))
2119 last = strend - len + 1;
2120 else if (!multiline)
2121 last = memEQ(strend - len, little, len)
2122 ? strend - len : NULL;
2128 last = rninstr(s, strend, little, little + len);
2130 last = strend; /* matching "$" */
2135 PerlIO_printf(Perl_debug_log,
2136 "%sCan't trim the tail, match fails (should not happen)%s\n",
2137 PL_colors[4], PL_colors[5]));
2138 goto phooey; /* Should not happen! */
2140 dontbother = strend - last + prog->float_min_offset;
2142 if (minlen && (dontbother < minlen))
2143 dontbother = minlen - 1;
2144 strend -= dontbother; /* this one's always in bytes! */
2145 /* We don't know much -- general case. */
2148 if (regtry(®info, &s))
2157 if (regtry(®info, &s))
2159 } while (s++ < strend);
2167 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2169 if (PL_reg_eval_set)
2170 restore_pos(aTHX_ prog);
2171 if (RXp_PAREN_NAMES(prog))
2172 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2174 /* make sure $`, $&, $', and $digit will work later */
2175 if ( !(flags & REXEC_NOT_FIRST) ) {
2176 RX_MATCH_COPY_FREE(rx);
2177 if (flags & REXEC_COPY_STR) {
2178 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2179 #ifdef PERL_OLD_COPY_ON_WRITE
2181 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2183 PerlIO_printf(Perl_debug_log,
2184 "Copy on write: regexp capture, type %d\n",
2187 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2188 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2189 assert (SvPOKp(prog->saved_copy));
2193 RX_MATCH_COPIED_on(rx);
2194 s = savepvn(strbeg, i);
2200 prog->subbeg = strbeg;
2201 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2208 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2209 PL_colors[4], PL_colors[5]));
2210 if (PL_reg_eval_set)
2211 restore_pos(aTHX_ prog);
2213 /* we failed :-( roll it back */
2214 swap_match_buff(prog);
2221 - regtry - try match at specific point
2223 STATIC I32 /* 0 failure, 1 success */
2224 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2228 REGEXP *const rx = reginfo->prog;
2229 regexp *const prog = (struct regexp *)SvANY(rx);
2230 RXi_GET_DECL(prog,progi);
2231 GET_RE_DEBUG_FLAGS_DECL;
2233 PERL_ARGS_ASSERT_REGTRY;
2235 reginfo->cutpoint=NULL;
2237 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2240 PL_reg_eval_set = RS_init;
2241 DEBUG_EXECUTE_r(DEBUG_s(
2242 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2243 (IV)(PL_stack_sp - PL_stack_base));
2246 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2247 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2249 /* Apparently this is not needed, judging by wantarray. */
2250 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2251 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2254 /* Make $_ available to executed code. */
2255 if (reginfo->sv != DEFSV) {
2257 DEFSV_set(reginfo->sv);
2260 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2261 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2262 /* prepare for quick setting of pos */
2263 #ifdef PERL_OLD_COPY_ON_WRITE
2264 if (SvIsCOW(reginfo->sv))
2265 sv_force_normal_flags(reginfo->sv, 0);
2267 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2268 &PL_vtbl_mglob, NULL, 0);
2272 PL_reg_oldpos = mg->mg_len;
2273 SAVEDESTRUCTOR_X(restore_pos, prog);
2275 if (!PL_reg_curpm) {
2276 Newxz(PL_reg_curpm, 1, PMOP);
2279 SV* const repointer = &PL_sv_undef;
2280 /* this regexp is also owned by the new PL_reg_curpm, which
2281 will try to free it. */
2282 av_push(PL_regex_padav, repointer);
2283 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2284 PL_regex_pad = AvARRAY(PL_regex_padav);
2289 /* It seems that non-ithreads works both with and without this code.
2290 So for efficiency reasons it seems best not to have the code
2291 compiled when it is not needed. */
2292 /* This is safe against NULLs: */
2293 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2294 /* PM_reg_curpm owns a reference to this regexp. */
2297 PM_SETRE(PL_reg_curpm, rx);
2298 PL_reg_oldcurpm = PL_curpm;
2299 PL_curpm = PL_reg_curpm;
2300 if (RXp_MATCH_COPIED(prog)) {
2301 /* Here is a serious problem: we cannot rewrite subbeg,
2302 since it may be needed if this match fails. Thus
2303 $` inside (?{}) could fail... */
2304 PL_reg_oldsaved = prog->subbeg;
2305 PL_reg_oldsavedlen = prog->sublen;
2306 #ifdef PERL_OLD_COPY_ON_WRITE
2307 PL_nrs = prog->saved_copy;
2309 RXp_MATCH_COPIED_off(prog);
2312 PL_reg_oldsaved = NULL;
2313 prog->subbeg = PL_bostr;
2314 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2316 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2317 prog->offs[0].start = *startpos - PL_bostr;
2318 PL_reginput = *startpos;
2319 PL_reglastparen = &prog->lastparen;
2320 PL_reglastcloseparen = &prog->lastcloseparen;
2321 prog->lastparen = 0;
2322 prog->lastcloseparen = 0;
2324 PL_regoffs = prog->offs;
2325 if (PL_reg_start_tmpl <= prog->nparens) {
2326 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2327 if(PL_reg_start_tmp)
2328 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2330 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2333 /* XXXX What this code is doing here?!!! There should be no need
2334 to do this again and again, PL_reglastparen should take care of
2337 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2338 * Actually, the code in regcppop() (which Ilya may be meaning by
2339 * PL_reglastparen), is not needed at all by the test suite
2340 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2341 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2342 * Meanwhile, this code *is* needed for the
2343 * above-mentioned test suite tests to succeed. The common theme
2344 * on those tests seems to be returning null fields from matches.
2345 * --jhi updated by dapm */
2347 if (prog->nparens) {
2348 regexp_paren_pair *pp = PL_regoffs;
2350 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2358 if (regmatch(reginfo, progi->program + 1)) {
2359 PL_regoffs[0].end = PL_reginput - PL_bostr;
2362 if (reginfo->cutpoint)
2363 *startpos= reginfo->cutpoint;
2364 REGCP_UNWIND(lastcp);
2369 #define sayYES goto yes
2370 #define sayNO goto no
2371 #define sayNO_SILENT goto no_silent
2373 /* we dont use STMT_START/END here because it leads to
2374 "unreachable code" warnings, which are bogus, but distracting. */
2375 #define CACHEsayNO \
2376 if (ST.cache_mask) \
2377 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2380 /* this is used to determine how far from the left messages like
2381 'failed...' are printed. It should be set such that messages
2382 are inline with the regop output that created them.
2384 #define REPORT_CODE_OFF 32
2387 /* Make sure there is a test for this +1 options in re_tests */
2388 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2390 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2391 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2393 #define SLAB_FIRST(s) (&(s)->states[0])
2394 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2396 /* grab a new slab and return the first slot in it */
2398 STATIC regmatch_state *
2401 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2404 regmatch_slab *s = PL_regmatch_slab->next;
2406 Newx(s, 1, regmatch_slab);
2407 s->prev = PL_regmatch_slab;
2409 PL_regmatch_slab->next = s;
2411 PL_regmatch_slab = s;
2412 return SLAB_FIRST(s);
2416 /* push a new state then goto it */
2418 #define PUSH_STATE_GOTO(state, node) \
2420 st->resume_state = state; \
2423 /* push a new state with success backtracking, then goto it */
2425 #define PUSH_YES_STATE_GOTO(state, node) \
2427 st->resume_state = state; \
2428 goto push_yes_state;
2434 regmatch() - main matching routine
2436 This is basically one big switch statement in a loop. We execute an op,
2437 set 'next' to point the next op, and continue. If we come to a point which
2438 we may need to backtrack to on failure such as (A|B|C), we push a
2439 backtrack state onto the backtrack stack. On failure, we pop the top
2440 state, and re-enter the loop at the state indicated. If there are no more
2441 states to pop, we return failure.
2443 Sometimes we also need to backtrack on success; for example /A+/, where
2444 after successfully matching one A, we need to go back and try to
2445 match another one; similarly for lookahead assertions: if the assertion
2446 completes successfully, we backtrack to the state just before the assertion
2447 and then carry on. In these cases, the pushed state is marked as
2448 'backtrack on success too'. This marking is in fact done by a chain of
2449 pointers, each pointing to the previous 'yes' state. On success, we pop to
2450 the nearest yes state, discarding any intermediate failure-only states.
2451 Sometimes a yes state is pushed just to force some cleanup code to be
2452 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2453 it to free the inner regex.
2455 Note that failure backtracking rewinds the cursor position, while
2456 success backtracking leaves it alone.
2458 A pattern is complete when the END op is executed, while a subpattern
2459 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2460 ops trigger the "pop to last yes state if any, otherwise return true"
2463 A common convention in this function is to use A and B to refer to the two
2464 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2465 the subpattern to be matched possibly multiple times, while B is the entire
2466 rest of the pattern. Variable and state names reflect this convention.
2468 The states in the main switch are the union of ops and failure/success of
2469 substates associated with with that op. For example, IFMATCH is the op
2470 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2471 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2472 successfully matched A and IFMATCH_A_fail is a state saying that we have
2473 just failed to match A. Resume states always come in pairs. The backtrack
2474 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2475 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2476 on success or failure.
2478 The struct that holds a backtracking state is actually a big union, with
2479 one variant for each major type of op. The variable st points to the
2480 top-most backtrack struct. To make the code clearer, within each
2481 block of code we #define ST to alias the relevant union.
2483 Here's a concrete example of a (vastly oversimplified) IFMATCH
2489 #define ST st->u.ifmatch
2491 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2492 ST.foo = ...; // some state we wish to save
2494 // push a yes backtrack state with a resume value of
2495 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2497 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2500 case IFMATCH_A: // we have successfully executed A; now continue with B
2502 bar = ST.foo; // do something with the preserved value
2505 case IFMATCH_A_fail: // A failed, so the assertion failed
2506 ...; // do some housekeeping, then ...
2507 sayNO; // propagate the failure
2514 For any old-timers reading this who are familiar with the old recursive
2515 approach, the code above is equivalent to:
2517 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2526 ...; // do some housekeeping, then ...
2527 sayNO; // propagate the failure
2530 The topmost backtrack state, pointed to by st, is usually free. If you
2531 want to claim it, populate any ST.foo fields in it with values you wish to
2532 save, then do one of
2534 PUSH_STATE_GOTO(resume_state, node);
2535 PUSH_YES_STATE_GOTO(resume_state, node);
2537 which sets that backtrack state's resume value to 'resume_state', pushes a
2538 new free entry to the top of the backtrack stack, then goes to 'node'.
2539 On backtracking, the free slot is popped, and the saved state becomes the
2540 new free state. An ST.foo field in this new top state can be temporarily
2541 accessed to retrieve values, but once the main loop is re-entered, it
2542 becomes available for reuse.
2544 Note that the depth of the backtrack stack constantly increases during the
2545 left-to-right execution of the pattern, rather than going up and down with
2546 the pattern nesting. For example the stack is at its maximum at Z at the
2547 end of the pattern, rather than at X in the following:
2549 /(((X)+)+)+....(Y)+....Z/
2551 The only exceptions to this are lookahead/behind assertions and the cut,
2552 (?>A), which pop all the backtrack states associated with A before
2555 Bascktrack state structs are allocated in slabs of about 4K in size.
2556 PL_regmatch_state and st always point to the currently active state,
2557 and PL_regmatch_slab points to the slab currently containing
2558 PL_regmatch_state. The first time regmatch() is called, the first slab is
2559 allocated, and is never freed until interpreter destruction. When the slab
2560 is full, a new one is allocated and chained to the end. At exit from
2561 regmatch(), slabs allocated since entry are freed.
2566 #define DEBUG_STATE_pp(pp) \
2568 DUMP_EXEC_POS(locinput, scan, do_utf8); \
2569 PerlIO_printf(Perl_debug_log, \
2570 " %*s"pp" %s%s%s%s%s\n", \
2572 PL_reg_name[st->resume_state], \
2573 ((st==yes_state||st==mark_state) ? "[" : ""), \
2574 ((st==yes_state) ? "Y" : ""), \
2575 ((st==mark_state) ? "M" : ""), \
2576 ((st==yes_state||st==mark_state) ? "]" : "") \
2581 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2586 S_debug_start_match(pTHX_ const REGEXP *prog, const bool do_utf8,
2587 const char *start, const char *end, const char *blurb)
2589 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2591 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2596 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2597 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2599 RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
2600 start, end - start, 60);
2602 PerlIO_printf(Perl_debug_log,
2603 "%s%s REx%s %s against %s\n",
2604 PL_colors[4], blurb, PL_colors[5], s0, s1);
2606 if (do_utf8||utf8_pat)
2607 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2608 utf8_pat ? "pattern" : "",
2609 utf8_pat && do_utf8 ? " and " : "",
2610 do_utf8 ? "string" : ""
2616 S_dump_exec_pos(pTHX_ const char *locinput,
2617 const regnode *scan,
2618 const char *loc_regeol,
2619 const char *loc_bostr,
2620 const char *loc_reg_starttry,
2623 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2624 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2625 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2626 /* The part of the string before starttry has one color
2627 (pref0_len chars), between starttry and current
2628 position another one (pref_len - pref0_len chars),
2629 after the current position the third one.
2630 We assume that pref0_len <= pref_len, otherwise we
2631 decrease pref0_len. */
2632 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2633 ? (5 + taill) - l : locinput - loc_bostr;
2636 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2638 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2640 pref0_len = pref_len - (locinput - loc_reg_starttry);
2641 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2642 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2643 ? (5 + taill) - pref_len : loc_regeol - locinput);
2644 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2648 if (pref0_len > pref_len)
2649 pref0_len = pref_len;
2651 const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0;
2653 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2654 (locinput - pref_len),pref0_len, 60, 4, 5);
2656 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2657 (locinput - pref_len + pref0_len),
2658 pref_len - pref0_len, 60, 2, 3);
2660 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2661 locinput, loc_regeol - locinput, 10, 0, 1);
2663 const STRLEN tlen=len0+len1+len2;
2664 PerlIO_printf(Perl_debug_log,
2665 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2666 (IV)(locinput - loc_bostr),
2669 (docolor ? "" : "> <"),
2671 (int)(tlen > 19 ? 0 : 19 - tlen),
2678 /* reg_check_named_buff_matched()
2679 * Checks to see if a named buffer has matched. The data array of
2680 * buffer numbers corresponding to the buffer is expected to reside
2681 * in the regexp->data->data array in the slot stored in the ARG() of
2682 * node involved. Note that this routine doesn't actually care about the
2683 * name, that information is not preserved from compilation to execution.
2684 * Returns the index of the leftmost defined buffer with the given name
2685 * or 0 if non of the buffers matched.
2688 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2691 RXi_GET_DECL(rex,rexi);
2692 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2693 I32 *nums=(I32*)SvPVX(sv_dat);
2695 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2697 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2698 if ((I32)*PL_reglastparen >= nums[n] &&
2699 PL_regoffs[nums[n]].end != -1)
2708 /* free all slabs above current one - called during LEAVE_SCOPE */
2711 S_clear_backtrack_stack(pTHX_ void *p)
2713 regmatch_slab *s = PL_regmatch_slab->next;
2718 PL_regmatch_slab->next = NULL;
2720 regmatch_slab * const osl = s;
2727 #define SETREX(Re1,Re2) \
2728 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2731 STATIC I32 /* 0 failure, 1 success */
2732 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2734 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2738 register const bool do_utf8 = PL_reg_match_utf8;
2739 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2740 REGEXP *rex_sv = reginfo->prog;
2741 regexp *rex = (struct regexp *)SvANY(rex_sv);
2742 RXi_GET_DECL(rex,rexi);
2744 /* the current state. This is a cached copy of PL_regmatch_state */
2745 register regmatch_state *st;
2746 /* cache heavy used fields of st in registers */
2747 register regnode *scan;
2748 register regnode *next;
2749 register U32 n = 0; /* general value; init to avoid compiler warning */
2750 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2751 register char *locinput = PL_reginput;
2752 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2754 bool result = 0; /* return value of S_regmatch */
2755 int depth = 0; /* depth of backtrack stack */
2756 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2757 const U32 max_nochange_depth =
2758 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2759 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2760 regmatch_state *yes_state = NULL; /* state to pop to on success of
2762 /* mark_state piggy backs on the yes_state logic so that when we unwind
2763 the stack on success we can update the mark_state as we go */
2764 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2765 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2766 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
2768 bool no_final = 0; /* prevent failure from backtracking? */
2769 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
2770 char *startpoint = PL_reginput;
2771 SV *popmark = NULL; /* are we looking for a mark? */
2772 SV *sv_commit = NULL; /* last mark name seen in failure */
2773 SV *sv_yes_mark = NULL; /* last mark name we have seen
2774 during a successfull match */
2775 U32 lastopen = 0; /* last open we saw */
2776 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
2777 SV* const oreplsv = GvSV(PL_replgv);
2778 /* these three flags are set by various ops to signal information to
2779 * the very next op. They have a useful lifetime of exactly one loop
2780 * iteration, and are not preserved or restored by state pushes/pops
2782 bool sw = 0; /* the condition value in (?(cond)a|b) */
2783 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
2784 int logical = 0; /* the following EVAL is:
2788 or the following IFMATCH/UNLESSM is:
2789 false: plain (?=foo)
2790 true: used as a condition: (?(?=foo))
2793 GET_RE_DEBUG_FLAGS_DECL;
2796 PERL_ARGS_ASSERT_REGMATCH;
2798 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
2799 PerlIO_printf(Perl_debug_log,"regmatch start\n");
2801 /* on first ever call to regmatch, allocate first slab */
2802 if (!PL_regmatch_slab) {
2803 Newx(PL_regmatch_slab, 1, regmatch_slab);
2804 PL_regmatch_slab->prev = NULL;
2805 PL_regmatch_slab->next = NULL;
2806 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2809 oldsave = PL_savestack_ix;
2810 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
2811 SAVEVPTR(PL_regmatch_slab);
2812 SAVEVPTR(PL_regmatch_state);
2814 /* grab next free state slot */
2815 st = ++PL_regmatch_state;
2816 if (st > SLAB_LAST(PL_regmatch_slab))
2817 st = PL_regmatch_state = S_push_slab(aTHX);
2819 /* Note that nextchr is a byte even in UTF */
2820 nextchr = UCHARAT(locinput);
2822 while (scan != NULL) {
2825 SV * const prop = sv_newmortal();
2826 regnode *rnext=regnext(scan);
2827 DUMP_EXEC_POS( locinput, scan, do_utf8 );
2828 regprop(rex, prop, scan);
2830 PerlIO_printf(Perl_debug_log,
2831 "%3"IVdf":%*s%s(%"IVdf")\n",
2832 (IV)(scan - rexi->program), depth*2, "",
2834 (PL_regkind[OP(scan)] == END || !rnext) ?
2835 0 : (IV)(rnext - rexi->program));
2838 next = scan + NEXT_OFF(scan);
2841 state_num = OP(scan);
2845 assert(PL_reglastparen == &rex->lastparen);
2846 assert(PL_reglastcloseparen == &rex->lastcloseparen);
2847 assert(PL_regoffs == rex->offs);
2849 switch (state_num) {
2851 if (locinput == PL_bostr)
2853 /* reginfo->till = reginfo->bol; */
2858 if (locinput == PL_bostr ||
2859 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2865 if (locinput == PL_bostr)
2869 if (locinput == reginfo->ganch)
2874 /* update the startpoint */
2875 st->u.keeper.val = PL_regoffs[0].start;
2876 PL_reginput = locinput;
2877 PL_regoffs[0].start = locinput - PL_bostr;
2878 PUSH_STATE_GOTO(KEEPS_next, next);
2880 case KEEPS_next_fail:
2881 /* rollback the start point change */
2882 PL_regoffs[0].start = st->u.keeper.val;
2888 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2893 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2895 if (PL_regeol - locinput > 1)
2899 if (PL_regeol != locinput)
2903 if (!nextchr && locinput >= PL_regeol)
2906 locinput += PL_utf8skip[nextchr];
2907 if (locinput > PL_regeol)
2909 nextchr = UCHARAT(locinput);
2912 nextchr = UCHARAT(++locinput);
2915 if (!nextchr && locinput >= PL_regeol)
2917 nextchr = UCHARAT(++locinput);
2920 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2923 locinput += PL_utf8skip[nextchr];
2924 if (locinput > PL_regeol)
2926 nextchr = UCHARAT(locinput);
2929 nextchr = UCHARAT(++locinput);
2933 #define ST st->u.trie
2935 /* In this case the charclass data is available inline so
2936 we can fail fast without a lot of extra overhead.
2938 if (scan->flags == EXACT || !do_utf8) {
2939 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2941 PerlIO_printf(Perl_debug_log,
2942 "%*s %sfailed to match trie start class...%s\n",
2943 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2952 /* what type of TRIE am I? (utf8 makes this contextual) */
2953 DECL_TRIE_TYPE(scan);
2955 /* what trie are we using right now */
2956 reg_trie_data * const trie
2957 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2958 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
2959 U32 state = trie->startstate;
2961 if (trie->bitmap && trie_type != trie_utf8_fold &&
2962 !TRIE_BITMAP_TEST(trie,*locinput)
2964 if (trie->states[ state ].wordnum) {
2966 PerlIO_printf(Perl_debug_log,
2967 "%*s %smatched empty string...%s\n",
2968 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2973 PerlIO_printf(Perl_debug_log,
2974 "%*s %sfailed to match trie start class...%s\n",
2975 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2982 U8 *uc = ( U8* )locinput;
2986 U8 *uscan = (U8*)NULL;
2988 SV *sv_accept_buff = NULL;
2989 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2991 ST.accepted = 0; /* how many accepting states we have seen */
2993 ST.jump = trie->jump;
2996 traverse the TRIE keeping track of all accepting states
2997 we transition through until we get to a failing node.
3000 while ( state && uc <= (U8*)PL_regeol ) {
3001 U32 base = trie->states[ state ].trans.base;
3004 /* We use charid to hold the wordnum as we don't use it
3005 for charid until after we have done the wordnum logic.
3006 We define an alias just so that the wordnum logic reads
3009 #define got_wordnum charid
3010 got_wordnum = trie->states[ state ].wordnum;
3012 if ( got_wordnum ) {
3013 if ( ! ST.accepted ) {
3015 SAVETMPS; /* XXX is this necessary? dmq */
3016 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
3017 sv_accept_buff=newSV(bufflen *
3018 sizeof(reg_trie_accepted) - 1);
3019 SvCUR_set(sv_accept_buff, 0);
3020 SvPOK_on(sv_accept_buff);
3021 sv_2mortal(sv_accept_buff);
3024 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
3027 if (ST.accepted >= bufflen) {
3029 ST.accept_buff =(reg_trie_accepted*)
3030 SvGROW(sv_accept_buff,
3031 bufflen * sizeof(reg_trie_accepted));
3033 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
3034 + sizeof(reg_trie_accepted));
3037 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
3038 ST.accept_buff[ST.accepted].endpos = uc;
3040 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
3044 DEBUG_TRIE_EXECUTE_r({
3045 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
3046 PerlIO_printf( Perl_debug_log,
3047 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
3048 2+depth * 2, "", PL_colors[4],
3049 (UV)state, (UV)ST.accepted );
3053 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3054 uscan, len, uvc, charid, foldlen,
3058 (base + charid > trie->uniquecharcount )
3059 && (base + charid - 1 - trie->uniquecharcount
3061 && trie->trans[base + charid - 1 -
3062 trie->uniquecharcount].check == state)
3064 state = trie->trans[base + charid - 1 -
3065 trie->uniquecharcount ].next;
3076 DEBUG_TRIE_EXECUTE_r(
3077 PerlIO_printf( Perl_debug_log,
3078 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3079 charid, uvc, (UV)state, PL_colors[5] );
3086 PerlIO_printf( Perl_debug_log,
3087 "%*s %sgot %"IVdf" possible matches%s\n",
3088 REPORT_CODE_OFF + depth * 2, "",
3089 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3092 goto trie_first_try; /* jump into the fail handler */
3094 case TRIE_next_fail: /* we failed - try next alterative */
3096 REGCP_UNWIND(ST.cp);
3097 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3098 PL_regoffs[n].end = -1;
3099 *PL_reglastparen = n;
3108 ST.lastparen = *PL_reglastparen;
3111 if ( ST.accepted == 1 ) {
3112 /* only one choice left - just continue */
3114 AV *const trie_words
3115 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3116 SV ** const tmp = av_fetch( trie_words,
3117 ST.accept_buff[ 0 ].wordnum-1, 0 );
3118 SV *sv= tmp ? sv_newmortal() : NULL;
3120 PerlIO_printf( Perl_debug_log,
3121 "%*s %sonly one match left: #%d <%s>%s\n",
3122 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3123 ST.accept_buff[ 0 ].wordnum,
3124 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3125 PL_colors[0], PL_colors[1],
3126 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3128 : "not compiled under -Dr",
3131 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
3132 /* in this case we free tmps/leave before we call regmatch
3133 as we wont be using accept_buff again. */
3135 locinput = PL_reginput;
3136 nextchr = UCHARAT(locinput);
3137 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
3140 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
3141 if (!has_cutgroup) {
3146 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3149 continue; /* execute rest of RE */
3152 if ( !ST.accepted-- ) {
3154 PerlIO_printf( Perl_debug_log,
3155 "%*s %sTRIE failed...%s\n",
3156 REPORT_CODE_OFF+depth*2, "",
3167 There are at least two accepting states left. Presumably
3168 the number of accepting states is going to be low,
3169 typically two. So we simply scan through to find the one
3170 with lowest wordnum. Once we find it, we swap the last
3171 state into its place and decrement the size. We then try to
3172 match the rest of the pattern at the point where the word
3173 ends. If we succeed, control just continues along the
3174 regex; if we fail we return here to try the next accepting
3181 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3182 DEBUG_TRIE_EXECUTE_r(
3183 PerlIO_printf( Perl_debug_log,
3184 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3185 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3186 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3187 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3190 if (ST.accept_buff[cur].wordnum <
3191 ST.accept_buff[best].wordnum)
3196 AV *const trie_words
3197 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3198 SV ** const tmp = av_fetch( trie_words,
3199 ST.accept_buff[ best ].wordnum - 1, 0 );
3200 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3202 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3203 SV *sv= tmp ? sv_newmortal() : NULL;
3205 PerlIO_printf( Perl_debug_log,
3206 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3207 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3208 ST.accept_buff[best].wordnum,
3209 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3210 PL_colors[0], PL_colors[1],
3211 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3212 ) : "not compiled under -Dr",
3213 REG_NODE_NUM(nextop),
3217 if ( best<ST.accepted ) {
3218 reg_trie_accepted tmp = ST.accept_buff[ best ];
3219 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3220 ST.accept_buff[ ST.accepted ] = tmp;
3223 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3224 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3227 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3229 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3234 /* we dont want to throw this away, see bug 57042*/
3235 if (oreplsv != GvSV(PL_replgv))
3236 sv_setsv(oreplsv, GvSV(PL_replgv));
3243 char *s = STRING(scan);
3245 if (do_utf8 != UTF) {
3246 /* The target and the pattern have differing utf8ness. */
3248 const char * const e = s + ln;
3251 /* The target is utf8, the pattern is not utf8. */
3256 if (NATIVE_TO_UNI(*(U8*)s) !=
3257 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3265 /* The target is not utf8, the pattern is utf8. */
3270 if (NATIVE_TO_UNI(*((U8*)l)) !=
3271 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3279 nextchr = UCHARAT(locinput);
3282 /* The target and the pattern have the same utf8ness. */
3283 /* Inline the first character, for speed. */
3284 if (UCHARAT(s) != nextchr)
3286 if (PL_regeol - locinput < ln)
3288 if (ln > 1 && memNE(s, locinput, ln))
3291 nextchr = UCHARAT(locinput);
3295 PL_reg_flags |= RF_tainted;
3298 char * const s = STRING(scan);
3301 if (do_utf8 || UTF) {
3302 /* Either target or the pattern are utf8. */
3303 const char * const l = locinput;
3304 char *e = PL_regeol;
3306 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3307 l, &e, 0, do_utf8)) {
3308 /* One more case for the sharp s:
3309 * pack("U0U*", 0xDF) =~ /ss/i,
3310 * the 0xC3 0x9F are the UTF-8
3311 * byte sequence for the U+00DF. */
3314 toLOWER(s[0]) == 's' &&
3316 toLOWER(s[1]) == 's' &&
3323 nextchr = UCHARAT(locinput);
3327 /* Neither the target and the pattern are utf8. */
3329 /* Inline the first character, for speed. */
3330 if (UCHARAT(s) != nextchr &&
3331 UCHARAT(s) != ((OP(scan) == EXACTF)
3332 ? PL_fold : PL_fold_locale)[nextchr])
3334 if (PL_regeol - locinput < ln)
3336 if (ln > 1 && (OP(scan) == EXACTF
3337 ? ibcmp(s, locinput, ln)
3338 : ibcmp_locale(s, locinput, ln)))
3341 nextchr = UCHARAT(locinput);
3346 STRLEN inclasslen = PL_regeol - locinput;
3348 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3350 if (locinput >= PL_regeol)
3352 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3353 nextchr = UCHARAT(locinput);
3358 nextchr = UCHARAT(locinput);
3359 if (!REGINCLASS(rex, scan, (U8*)locinput))
3361 if (!nextchr && locinput >= PL_regeol)
3363 nextchr = UCHARAT(++locinput);
3367 /* If we might have the case of the German sharp s
3368 * in a casefolding Unicode character class. */
3370 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3371 locinput += SHARP_S_SKIP;
3372 nextchr = UCHARAT(locinput);
3378 PL_reg_flags |= RF_tainted;
3384 LOAD_UTF8_CHARCLASS_ALNUM();
3385 if (!(OP(scan) == ALNUM
3386 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3387 : isALNUM_LC_utf8((U8*)locinput)))
3391 locinput += PL_utf8skip[nextchr];
3392 nextchr = UCHARAT(locinput);
3395 if (!(OP(scan) == ALNUM
3396 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3398 nextchr = UCHARAT(++locinput);
3401 PL_reg_flags |= RF_tainted;
3404 if (!nextchr && locinput >= PL_regeol)
3407 LOAD_UTF8_CHARCLASS_ALNUM();
3408 if (OP(scan) == NALNUM
3409 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3410 : isALNUM_LC_utf8((U8*)locinput))
3414 locinput += PL_utf8skip[nextchr];
3415 nextchr = UCHARAT(locinput);
3418 if (OP(scan) == NALNUM
3419 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3421 nextchr = UCHARAT(++locinput);
3425 PL_reg_flags |= RF_tainted;
3429 /* was last char in word? */
3431 if (locinput == PL_bostr)
3434 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3436 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3438 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3439 ln = isALNUM_uni(ln);
3440 LOAD_UTF8_CHARCLASS_ALNUM();
3441 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3444 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3445 n = isALNUM_LC_utf8((U8*)locinput);
3449 ln = (locinput != PL_bostr) ?
3450 UCHARAT(locinput - 1) : '\n';
3451 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3453 n = isALNUM(nextchr);
3456 ln = isALNUM_LC(ln);
3457 n = isALNUM_LC(nextchr);
3460 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3461 OP(scan) == BOUNDL))
3465 PL_reg_flags |= RF_tainted;
3471 if (UTF8_IS_CONTINUED(nextchr)) {
3472 LOAD_UTF8_CHARCLASS_SPACE();
3473 if (!(OP(scan) == SPACE
3474 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3475 : isSPACE_LC_utf8((U8*)locinput)))
3479 locinput += PL_utf8skip[nextchr];
3480 nextchr = UCHARAT(locinput);
3483 if (!(OP(scan) == SPACE
3484 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3486 nextchr = UCHARAT(++locinput);
3489 if (!(OP(scan) == SPACE
3490 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3492 nextchr = UCHARAT(++locinput);
3496 PL_reg_flags |= RF_tainted;
3499 if (!nextchr && locinput >= PL_regeol)
3502 LOAD_UTF8_CHARCLASS_SPACE();
3503 if (OP(scan) == NSPACE
3504 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3505 : isSPACE_LC_utf8((U8*)locinput))
3509 locinput += PL_utf8skip[nextchr];
3510 nextchr = UCHARAT(locinput);
3513 if (OP(scan) == NSPACE
3514 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3516 nextchr = UCHARAT(++locinput);
3519 PL_reg_flags |= RF_tainted;
3525 LOAD_UTF8_CHARCLASS_DIGIT();
3526 if (!(OP(scan) == DIGIT
3527 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3528 : isDIGIT_LC_utf8((U8*)locinput)))
3532 locinput += PL_utf8skip[nextchr];
3533 nextchr = UCHARAT(locinput);
3536 if (!(OP(scan) == DIGIT
3537 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3539 nextchr = UCHARAT(++locinput);
3542 PL_reg_flags |= RF_tainted;
3545 if (!nextchr && locinput >= PL_regeol)
3548 LOAD_UTF8_CHARCLASS_DIGIT();
3549 if (OP(scan) == NDIGIT
3550 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3551 : isDIGIT_LC_utf8((U8*)locinput))
3555 locinput += PL_utf8skip[nextchr];
3556 nextchr = UCHARAT(locinput);
3559 if (OP(scan) == NDIGIT
3560 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3562 nextchr = UCHARAT(++locinput);
3565 if (locinput >= PL_regeol)
3568 LOAD_UTF8_CHARCLASS_MARK();
3569 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3571 locinput += PL_utf8skip[nextchr];
3572 while (locinput < PL_regeol &&
3573 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3574 locinput += UTF8SKIP(locinput);
3575 if (locinput > PL_regeol)
3580 nextchr = UCHARAT(locinput);
3587 PL_reg_flags |= RF_tainted;
3592 n = reg_check_named_buff_matched(rex,scan);
3595 type = REF + ( type - NREF );
3602 PL_reg_flags |= RF_tainted;
3606 n = ARG(scan); /* which paren pair */
3609 ln = PL_regoffs[n].start;
3610 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3611 if (*PL_reglastparen < n || ln == -1)
3612 sayNO; /* Do not match unless seen CLOSEn. */
3613 if (ln == PL_regoffs[n].end)
3617 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3619 const char *e = PL_bostr + PL_regoffs[n].end;
3621 * Note that we can't do the "other character" lookup trick as
3622 * in the 8-bit case (no pun intended) because in Unicode we
3623 * have to map both upper and title case to lower case.
3627 STRLEN ulen1, ulen2;
3628 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3629 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3633 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3634 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3635 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3642 nextchr = UCHARAT(locinput);
3646 /* Inline the first character, for speed. */
3647 if (UCHARAT(s) != nextchr &&
3649 (UCHARAT(s) != (type == REFF
3650 ? PL_fold : PL_fold_locale)[nextchr])))
3652 ln = PL_regoffs[n].end - ln;
3653 if (locinput + ln > PL_regeol)
3655 if (ln > 1 && (type == REF
3656 ? memNE(s, locinput, ln)
3658 ? ibcmp(s, locinput, ln)
3659 : ibcmp_locale(s, locinput, ln))))
3662 nextchr = UCHARAT(locinput);
3672 #define ST st->u.eval
3677 regexp_internal *rei;
3678 regnode *startpoint;
3681 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
3682 if (cur_eval && cur_eval->locinput==locinput) {
3683 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3684 Perl_croak(aTHX_ "Infinite recursion in regex");
3685 if ( ++nochange_depth > max_nochange_depth )
3687 "Pattern subroutine nesting without pos change"
3688 " exceeded limit in regex");
3695 (void)ReREFCNT_inc(rex_sv);
3696 if (OP(scan)==GOSUB) {
3697 startpoint = scan + ARG2L(scan);
3698 ST.close_paren = ARG(scan);
3700 startpoint = rei->program+1;
3703 goto eval_recurse_doit;
3705 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3706 if (cur_eval && cur_eval->locinput==locinput) {
3707 if ( ++nochange_depth > max_nochange_depth )
3708 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3713 /* execute the code in the {...} */
3715 SV ** const before = SP;
3716 OP_4tree * const oop = PL_op;
3717 COP * const ocurcop = PL_curcop;
3719 char *saved_regeol = PL_regeol;
3722 PL_op = (OP_4tree*)rexi->data->data[n];
3723 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3724 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3725 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3726 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
3729 SV *sv_mrk = get_sv("REGMARK", 1);
3730 sv_setsv(sv_mrk, sv_yes_mark);
3733 CALLRUNOPS(aTHX); /* Scalar context. */
3736 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3743 PAD_RESTORE_LOCAL(old_comppad);
3744 PL_curcop = ocurcop;
3745 PL_regeol = saved_regeol;
3748 sv_setsv(save_scalar(PL_replgv), ret);
3752 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3755 /* extract RE object from returned value; compiling if
3761 SV *const sv = SvRV(ret);
3763 if (SvTYPE(sv) == SVt_REGEXP) {
3765 } else if (SvSMAGICAL(sv)) {
3766 mg = mg_find(sv, PERL_MAGIC_qr);
3769 } else if (SvTYPE(ret) == SVt_REGEXP) {
3771 } else if (SvSMAGICAL(ret)) {
3772 if (SvGMAGICAL(ret)) {
3773 /* I don't believe that there is ever qr magic
3775 assert(!mg_find(ret, PERL_MAGIC_qr));
3776 sv_unmagic(ret, PERL_MAGIC_qr);
3779 mg = mg_find(ret, PERL_MAGIC_qr);
3780 /* testing suggests mg only ends up non-NULL for
3781 scalars who were upgraded and compiled in the
3782 else block below. In turn, this is only
3783 triggered in the "postponed utf8 string" tests
3789 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
3793 rx = reg_temp_copy(rx);
3797 const I32 osize = PL_regsize;
3800 assert (SvUTF8(ret));
3801 } else if (SvUTF8(ret)) {
3802 /* Not doing UTF-8, despite what the SV says. Is
3803 this only if we're trapped in use 'bytes'? */
3804 /* Make a copy of the octet sequence, but without
3805 the flag on, as the compiler now honours the
3806 SvUTF8 flag on ret. */
3808 const char *const p = SvPV(ret, len);
3809 ret = newSVpvn_flags(p, len, SVs_TEMP);
3811 rx = CALLREGCOMP(ret, pm_flags);
3813 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3815 /* This isn't a first class regexp. Instead, it's
3816 caching a regexp onto an existing, Perl visible
3818 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
3823 re = (struct regexp *)SvANY(rx);
3825 RXp_MATCH_COPIED_off(re);
3826 re->subbeg = rex->subbeg;
3827 re->sublen = rex->sublen;
3830 debug_start_match(re_sv, do_utf8, locinput, PL_regeol,
3831 "Matching embedded");
3833 startpoint = rei->program + 1;
3834 ST.close_paren = 0; /* only used for GOSUB */
3835 /* borrowed from regtry */
3836 if (PL_reg_start_tmpl <= re->nparens) {
3837 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3838 if(PL_reg_start_tmp)
3839 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3841 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3844 eval_recurse_doit: /* Share code with GOSUB below this line */
3845 /* run the pattern returned from (??{...}) */
3846 ST.cp = regcppush(0); /* Save *all* the positions. */
3847 REGCP_SET(ST.lastcp);
3849 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
3851 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
3852 PL_reglastparen = &re->lastparen;
3853 PL_reglastcloseparen = &re->lastcloseparen;
3855 re->lastcloseparen = 0;
3857 PL_reginput = locinput;
3860 /* XXXX This is too dramatic a measure... */
3863 ST.toggle_reg_flags = PL_reg_flags;
3865 PL_reg_flags |= RF_utf8;
3867 PL_reg_flags &= ~RF_utf8;
3868 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3870 ST.prev_rex = rex_sv;
3871 ST.prev_curlyx = cur_curlyx;
3872 SETREX(rex_sv,re_sv);
3877 ST.prev_eval = cur_eval;
3879 /* now continue from first node in postoned RE */
3880 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3883 /* logical is 1, /(?(?{...})X|Y)/ */
3884 sw = (bool)SvTRUE(ret);
3889 case EVAL_AB: /* cleanup after a successful (??{A})B */
3890 /* note: this is called twice; first after popping B, then A */
3891 PL_reg_flags ^= ST.toggle_reg_flags;
3892 ReREFCNT_dec(rex_sv);
3893 SETREX(rex_sv,ST.prev_rex);
3894 rex = (struct regexp *)SvANY(rex_sv);
3895 rexi = RXi_GET(rex);
3897 cur_eval = ST.prev_eval;
3898 cur_curlyx = ST.prev_curlyx;
3900 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
3901 PL_reglastparen = &rex->lastparen;
3902 PL_reglastcloseparen = &rex->lastcloseparen;
3903 /* also update PL_regoffs */
3904 PL_regoffs = rex->offs;
3906 /* XXXX This is too dramatic a measure... */
3908 if ( nochange_depth )
3913 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3914 /* note: this is called twice; first after popping B, then A */
3915 PL_reg_flags ^= ST.toggle_reg_flags;
3916 ReREFCNT_dec(rex_sv);
3917 SETREX(rex_sv,ST.prev_rex);
3918 rex = (struct regexp *)SvANY(rex_sv);
3919 rexi = RXi_GET(rex);
3920 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
3921 PL_reglastparen = &rex->lastparen;
3922 PL_reglastcloseparen = &rex->lastcloseparen;
3924 PL_reginput = locinput;
3925 REGCP_UNWIND(ST.lastcp);
3927 cur_eval = ST.prev_eval;
3928 cur_curlyx = ST.prev_curlyx;
3929 /* XXXX This is too dramatic a measure... */
3931 if ( nochange_depth )
3937 n = ARG(scan); /* which paren pair */
3938 PL_reg_start_tmp[n] = locinput;
3944 n = ARG(scan); /* which paren pair */
3945 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
3946 PL_regoffs[n].end = locinput - PL_bostr;
3947 /*if (n > PL_regsize)
3949 if (n > *PL_reglastparen)
3950 *PL_reglastparen = n;
3951 *PL_reglastcloseparen = n;
3952 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3960 cursor && OP(cursor)!=END;
3961 cursor=regnext(cursor))
3963 if ( OP(cursor)==CLOSE ){
3965 if ( n <= lastopen ) {
3967 = PL_reg_start_tmp[n] - PL_bostr;
3968 PL_regoffs[n].end = locinput - PL_bostr;
3969 /*if (n > PL_regsize)
3971 if (n > *PL_reglastparen)
3972 *PL_reglastparen = n;
3973 *PL_reglastcloseparen = n;
3974 if ( n == ARG(scan) || (cur_eval &&
3975 cur_eval->u.eval.close_paren == n))
3984 n = ARG(scan); /* which paren pair */
3985 sw = (bool)(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
3988 /* reg_check_named_buff_matched returns 0 for no match */
3989 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3993 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3999 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4001 next = NEXTOPER(NEXTOPER(scan));
4003 next = scan + ARG(scan);
4004 if (OP(next) == IFTHEN) /* Fake one. */
4005 next = NEXTOPER(NEXTOPER(next));
4009 logical = scan->flags;
4012 /*******************************************************************
4014 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4015 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4016 STAR/PLUS/CURLY/CURLYN are used instead.)
4018 A*B is compiled as <CURLYX><A><WHILEM><B>
4020 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4021 state, which contains the current count, initialised to -1. It also sets
4022 cur_curlyx to point to this state, with any previous value saved in the
4025 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4026 since the pattern may possibly match zero times (i.e. it's a while {} loop
4027 rather than a do {} while loop).
4029 Each entry to WHILEM represents a successful match of A. The count in the
4030 CURLYX block is incremented, another WHILEM state is pushed, and execution
4031 passes to A or B depending on greediness and the current count.
4033 For example, if matching against the string a1a2a3b (where the aN are
4034 substrings that match /A/), then the match progresses as follows: (the
4035 pushed states are interspersed with the bits of strings matched so far):
4038 <CURLYX cnt=0><WHILEM>
4039 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4040 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4041 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4042 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4044 (Contrast this with something like CURLYM, which maintains only a single
4048 a1 <CURLYM cnt=1> a2
4049 a1 a2 <CURLYM cnt=2> a3
4050 a1 a2 a3 <CURLYM cnt=3> b
4053 Each WHILEM state block marks a point to backtrack to upon partial failure
4054 of A or B, and also contains some minor state data related to that
4055 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4056 overall state, such as the count, and pointers to the A and B ops.
4058 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4059 must always point to the *current* CURLYX block, the rules are:
4061 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4062 and set cur_curlyx to point the new block.
4064 When popping the CURLYX block after a successful or unsuccessful match,
4065 restore the previous cur_curlyx.
4067 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4068 to the outer one saved in the CURLYX block.
4070 When popping the WHILEM block after a successful or unsuccessful B match,
4071 restore the previous cur_curlyx.
4073 Here's an example for the pattern (AI* BI)*BO
4074 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4077 curlyx backtrack stack
4078 ------ ---------------
4080 CO <CO prev=NULL> <WO>
4081 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4082 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4083 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4085 At this point the pattern succeeds, and we work back down the stack to
4086 clean up, restoring as we go:
4088 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4089 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4090 CO <CO prev=NULL> <WO>
4093 *******************************************************************/
4095 #define ST st->u.curlyx
4097 case CURLYX: /* start of /A*B/ (for complex A) */
4099 /* No need to save/restore up to this paren */
4100 I32 parenfloor = scan->flags;
4102 assert(next); /* keep Coverity happy */
4103 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4106 /* XXXX Probably it is better to teach regpush to support
4107 parenfloor > PL_regsize... */
4108 if (parenfloor > (I32)*PL_reglastparen)
4109 parenfloor = *PL_reglastparen; /* Pessimization... */
4111 ST.prev_curlyx= cur_curlyx;
4113 ST.cp = PL_savestack_ix;
4115 /* these fields contain the state of the current curly.
4116 * they are accessed by subsequent WHILEMs */
4117 ST.parenfloor = parenfloor;
4118 ST.min = ARG1(scan);
4119 ST.max = ARG2(scan);
4120 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
4124 ST.count = -1; /* this will be updated by WHILEM */
4125 ST.lastloc = NULL; /* this will be updated by WHILEM */
4127 PL_reginput = locinput;
4128 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4132 case CURLYX_end: /* just finished matching all of A*B */
4133 cur_curlyx = ST.prev_curlyx;
4137 case CURLYX_end_fail: /* just failed to match all of A*B */
4139 cur_curlyx = ST.prev_curlyx;
4145 #define ST st->u.whilem
4147 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4149 /* see the discussion above about CURLYX/WHILEM */
4151 assert(cur_curlyx); /* keep Coverity happy */
4152 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4153 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4154 ST.cache_offset = 0;
4157 PL_reginput = locinput;
4159 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4160 "%*s whilem: matched %ld out of %ld..%ld\n",
4161 REPORT_CODE_OFF+depth*2, "", (long)n,
4162 (long)cur_curlyx->u.curlyx.min,
4163 (long)cur_curlyx->u.curlyx.max)
4166 /* First just match a string of min A's. */
4168 if (n < cur_curlyx->u.curlyx.min) {
4169 cur_curlyx->u.curlyx.lastloc = locinput;
4170 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
4174 /* If degenerate A matches "", assume A done. */
4176 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4177 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4178 "%*s whilem: empty match detected, trying continuation...\n",
4179 REPORT_CODE_OFF+depth*2, "")
4181 goto do_whilem_B_max;
4184 /* super-linear cache processing */
4188 if (!PL_reg_maxiter) {
4189 /* start the countdown: Postpone detection until we
4190 * know the match is not *that* much linear. */
4191 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4192 /* possible overflow for long strings and many CURLYX's */
4193 if (PL_reg_maxiter < 0)
4194 PL_reg_maxiter = I32_MAX;
4195 PL_reg_leftiter = PL_reg_maxiter;
4198 if (PL_reg_leftiter-- == 0) {
4199 /* initialise cache */
4200 const I32 size = (PL_reg_maxiter + 7)/8;
4201 if (PL_reg_poscache) {
4202 if ((I32)PL_reg_poscache_size < size) {
4203 Renew(PL_reg_poscache, size, char);
4204 PL_reg_poscache_size = size;
4206 Zero(PL_reg_poscache, size, char);
4209 PL_reg_poscache_size = size;
4210 Newxz(PL_reg_poscache, size, char);
4212 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4213 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4214 PL_colors[4], PL_colors[5])
4218 if (PL_reg_leftiter < 0) {
4219 /* have we already failed at this position? */
4221 offset = (scan->flags & 0xf) - 1
4222 + (locinput - PL_bostr) * (scan->flags>>4);
4223 mask = 1 << (offset % 8);
4225 if (PL_reg_poscache[offset] & mask) {
4226 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4227 "%*s whilem: (cache) already tried at this position...\n",
4228 REPORT_CODE_OFF+depth*2, "")
4230 sayNO; /* cache records failure */
4232 ST.cache_offset = offset;
4233 ST.cache_mask = mask;
4237 /* Prefer B over A for minimal matching. */
4239 if (cur_curlyx->u.curlyx.minmod) {
4240 ST.save_curlyx = cur_curlyx;
4241 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4242 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4243 REGCP_SET(ST.lastcp);
4244 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4248 /* Prefer A over B for maximal matching. */
4250 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
4251 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4252 cur_curlyx->u.curlyx.lastloc = locinput;
4253 REGCP_SET(ST.lastcp);
4254 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
4257 goto do_whilem_B_max;
4261 case WHILEM_B_min: /* just matched B in a minimal match */
4262 case WHILEM_B_max: /* just matched B in a maximal match */
4263 cur_curlyx = ST.save_curlyx;
4267 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4268 cur_curlyx = ST.save_curlyx;
4269 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4270 cur_curlyx->u.curlyx.count--;
4274 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4275 REGCP_UNWIND(ST.lastcp);
4278 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4279 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4280 cur_curlyx->u.curlyx.count--;
4284 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4285 REGCP_UNWIND(ST.lastcp);
4286 regcppop(rex); /* Restore some previous $<digit>s? */
4287 PL_reginput = locinput;
4288 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4289 "%*s whilem: failed, trying continuation...\n",
4290 REPORT_CODE_OFF+depth*2, "")
4293 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4294 && ckWARN(WARN_REGEXP)
4295 && !(PL_reg_flags & RF_warned))
4297 PL_reg_flags |= RF_warned;
4298 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4299 "Complex regular subexpression recursion",
4304 ST.save_curlyx = cur_curlyx;
4305 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4306 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4309 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4310 cur_curlyx = ST.save_curlyx;
4311 REGCP_UNWIND(ST.lastcp);
4314 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
4315 /* Maximum greed exceeded */
4316 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4317 && ckWARN(WARN_REGEXP)
4318 && !(PL_reg_flags & RF_warned))
4320 PL_reg_flags |= RF_warned;
4321 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4322 "%s limit (%d) exceeded",
4323 "Complex regular subexpression recursion",
4326 cur_curlyx->u.curlyx.count--;
4330 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4331 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4333 /* Try grabbing another A and see if it helps. */
4334 PL_reginput = locinput;
4335 cur_curlyx->u.curlyx.lastloc = locinput;
4336 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4337 REGCP_SET(ST.lastcp);
4338 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
4342 #define ST st->u.branch
4344 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4345 next = scan + ARG(scan);
4348 scan = NEXTOPER(scan);
4351 case BRANCH: /* /(...|A|...)/ */
4352 scan = NEXTOPER(scan); /* scan now points to inner node */
4353 ST.lastparen = *PL_reglastparen;
4354 ST.next_branch = next;
4356 PL_reginput = locinput;
4358 /* Now go into the branch */
4360 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4362 PUSH_STATE_GOTO(BRANCH_next, scan);
4366 PL_reginput = locinput;
4367 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4368 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4369 PUSH_STATE_GOTO(CUTGROUP_next,next);
4371 case CUTGROUP_next_fail:
4374 if (st->u.mark.mark_name)
4375 sv_commit = st->u.mark.mark_name;
4381 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4386 REGCP_UNWIND(ST.cp);
4387 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4388 PL_regoffs[n].end = -1;
4389 *PL_reglastparen = n;
4390 /*dmq: *PL_reglastcloseparen = n; */
4391 scan = ST.next_branch;
4392 /* no more branches? */
4393 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4395 PerlIO_printf( Perl_debug_log,
4396 "%*s %sBRANCH failed...%s\n",
4397 REPORT_CODE_OFF+depth*2, "",
4403 continue; /* execute next BRANCH[J] op */
4411 #define ST st->u.curlym
4413 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4415 /* This is an optimisation of CURLYX that enables us to push
4416 * only a single backtracking state, no matter how many matches
4417 * there are in {m,n}. It relies on the pattern being constant
4418 * length, with no parens to influence future backrefs
4422 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4424 /* if paren positive, emulate an OPEN/CLOSE around A */
4426 U32 paren = ST.me->flags;
4427 if (paren > PL_regsize)
4429 if (paren > *PL_reglastparen)
4430 *PL_reglastparen = paren;
4431 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4439 ST.c1 = CHRTEST_UNINIT;
4442 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4445 curlym_do_A: /* execute the A in /A{m,n}B/ */
4446 PL_reginput = locinput;
4447 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4450 case CURLYM_A: /* we've just matched an A */
4451 locinput = st->locinput;
4452 nextchr = UCHARAT(locinput);
4455 /* after first match, determine A's length: u.curlym.alen */
4456 if (ST.count == 1) {
4457 if (PL_reg_match_utf8) {
4459 while (s < PL_reginput) {
4465 ST.alen = PL_reginput - locinput;
4468 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4471 PerlIO_printf(Perl_debug_log,
4472 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4473 (int)(REPORT_CODE_OFF+(depth*2)), "",
4474 (IV) ST.count, (IV)ST.alen)
4477 locinput = PL_reginput;
4479 if (cur_eval && cur_eval->u.eval.close_paren &&
4480 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4484 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4485 if ( max == REG_INFTY || ST.count < max )
4486 goto curlym_do_A; /* try to match another A */
4488 goto curlym_do_B; /* try to match B */
4490 case CURLYM_A_fail: /* just failed to match an A */
4491 REGCP_UNWIND(ST.cp);
4493 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4494 || (cur_eval && cur_eval->u.eval.close_paren &&
4495 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4498 curlym_do_B: /* execute the B in /A{m,n}B/ */
4499 PL_reginput = locinput;
4500 if (ST.c1 == CHRTEST_UNINIT) {
4501 /* calculate c1 and c2 for possible match of 1st char
4502 * following curly */
4503 ST.c1 = ST.c2 = CHRTEST_VOID;
4504 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4505 regnode *text_node = ST.B;
4506 if (! HAS_TEXT(text_node))
4507 FIND_NEXT_IMPT(text_node);
4510 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4512 But the former is redundant in light of the latter.
4514 if this changes back then the macro for
4515 IS_TEXT and friends need to change.
4517 if (PL_regkind[OP(text_node)] == EXACT)
4520 ST.c1 = (U8)*STRING(text_node);
4522 (IS_TEXTF(text_node))
4524 : (IS_TEXTFL(text_node))
4525 ? PL_fold_locale[ST.c1]
4532 PerlIO_printf(Perl_debug_log,
4533 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4534 (int)(REPORT_CODE_OFF+(depth*2)),
4537 if (ST.c1 != CHRTEST_VOID
4538 && UCHARAT(PL_reginput) != ST.c1
4539 && UCHARAT(PL_reginput) != ST.c2)
4541 /* simulate B failing */
4543 PerlIO_printf(Perl_debug_log,
4544 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
4545 (int)(REPORT_CODE_OFF+(depth*2)),"",
4548 state_num = CURLYM_B_fail;
4549 goto reenter_switch;
4553 /* mark current A as captured */
4554 I32 paren = ST.me->flags;
4556 PL_regoffs[paren].start
4557 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4558 PL_regoffs[paren].end = PL_reginput - PL_bostr;
4559 /*dmq: *PL_reglastcloseparen = paren; */
4562 PL_regoffs[paren].end = -1;
4563 if (cur_eval && cur_eval->u.eval.close_paren &&
4564 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4573 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4576 case CURLYM_B_fail: /* just failed to match a B */
4577 REGCP_UNWIND(ST.cp);
4579 I32 max = ARG2(ST.me);
4580 if (max != REG_INFTY && ST.count == max)
4582 goto curlym_do_A; /* try to match a further A */
4584 /* backtrack one A */
4585 if (ST.count == ARG1(ST.me) /* min */)
4588 locinput = HOPc(locinput, -ST.alen);
4589 goto curlym_do_B; /* try to match B */
4592 #define ST st->u.curly
4594 #define CURLY_SETPAREN(paren, success) \
4597 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
4598 PL_regoffs[paren].end = locinput - PL_bostr; \
4599 *PL_reglastcloseparen = paren; \
4602 PL_regoffs[paren].end = -1; \
4605 case STAR: /* /A*B/ where A is width 1 */
4609 scan = NEXTOPER(scan);
4611 case PLUS: /* /A+B/ where A is width 1 */
4615 scan = NEXTOPER(scan);
4617 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4618 ST.paren = scan->flags; /* Which paren to set */
4619 if (ST.paren > PL_regsize)
4620 PL_regsize = ST.paren;
4621 if (ST.paren > *PL_reglastparen)
4622 *PL_reglastparen = ST.paren;
4623 ST.min = ARG1(scan); /* min to match */
4624 ST.max = ARG2(scan); /* max to match */
4625 if (cur_eval && cur_eval->u.eval.close_paren &&
4626 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4630 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4632 case CURLY: /* /A{m,n}B/ where A is width 1 */
4634 ST.min = ARG1(scan); /* min to match */
4635 ST.max = ARG2(scan); /* max to match */
4636 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4639 * Lookahead to avoid useless match attempts
4640 * when we know what character comes next.
4642 * Used to only do .*x and .*?x, but now it allows
4643 * for )'s, ('s and (?{ ... })'s to be in the way
4644 * of the quantifier and the EXACT-like node. -- japhy
4647 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4649 if (HAS_TEXT(next) || JUMPABLE(next)) {
4651 regnode *text_node = next;
4653 if (! HAS_TEXT(text_node))
4654 FIND_NEXT_IMPT(text_node);
4656 if (! HAS_TEXT(text_node))
4657 ST.c1 = ST.c2 = CHRTEST_VOID;
4659 if ( PL_regkind[OP(text_node)] != EXACT ) {
4660 ST.c1 = ST.c2 = CHRTEST_VOID;
4661 goto assume_ok_easy;
4664 s = (U8*)STRING(text_node);
4666 /* Currently we only get here when
4668 PL_rekind[OP(text_node)] == EXACT
4670 if this changes back then the macro for IS_TEXT and
4671 friends need to change. */
4674 if (IS_TEXTF(text_node))
4675 ST.c2 = PL_fold[ST.c1];
4676 else if (IS_TEXTFL(text_node))
4677 ST.c2 = PL_fold_locale[ST.c1];
4680 if (IS_TEXTF(text_node)) {
4681 STRLEN ulen1, ulen2;
4682 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4683 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4685 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4686 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4688 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4690 0 : UTF8_ALLOW_ANY);
4691 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4693 0 : UTF8_ALLOW_ANY);
4695 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4697 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4702 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4709 ST.c1 = ST.c2 = CHRTEST_VOID;
4714 PL_reginput = locinput;
4717 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4720 locinput = PL_reginput;
4722 if (ST.c1 == CHRTEST_VOID)
4723 goto curly_try_B_min;
4725 ST.oldloc = locinput;
4727 /* set ST.maxpos to the furthest point along the
4728 * string that could possibly match */
4729 if (ST.max == REG_INFTY) {
4730 ST.maxpos = PL_regeol - 1;
4732 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4736 int m = ST.max - ST.min;
4737 for (ST.maxpos = locinput;
4738 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4739 ST.maxpos += UTF8SKIP(ST.maxpos);
4742 ST.maxpos = locinput + ST.max - ST.min;
4743 if (ST.maxpos >= PL_regeol)
4744 ST.maxpos = PL_regeol - 1;
4746 goto curly_try_B_min_known;
4750 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4751 locinput = PL_reginput;
4752 if (ST.count < ST.min)
4754 if ((ST.count > ST.min)
4755 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4757 /* A{m,n} must come at the end of the string, there's
4758 * no point in backing off ... */
4760 /* ...except that $ and \Z can match before *and* after
4761 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4762 We may back off by one in this case. */
4763 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4767 goto curly_try_B_max;
4772 case CURLY_B_min_known_fail:
4773 /* failed to find B in a non-greedy match where c1,c2 valid */
4774 if (ST.paren && ST.count)
4775 PL_regoffs[ST.paren].end = -1;
4777 PL_reginput = locinput; /* Could be reset... */
4778 REGCP_UNWIND(ST.cp);
4779 /* Couldn't or didn't -- move forward. */
4780 ST.oldloc = locinput;
4782 locinput += UTF8SKIP(locinput);
4786 curly_try_B_min_known:
4787 /* find the next place where 'B' could work, then call B */
4791 n = (ST.oldloc == locinput) ? 0 : 1;
4792 if (ST.c1 == ST.c2) {
4794 /* set n to utf8_distance(oldloc, locinput) */
4795 while (locinput <= ST.maxpos &&
4796 utf8n_to_uvchr((U8*)locinput,
4797 UTF8_MAXBYTES, &len,
4798 uniflags) != (UV)ST.c1) {
4804 /* set n to utf8_distance(oldloc, locinput) */
4805 while (locinput <= ST.maxpos) {
4807 const UV c = utf8n_to_uvchr((U8*)locinput,
4808 UTF8_MAXBYTES, &len,
4810 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4818 if (ST.c1 == ST.c2) {
4819 while (locinput <= ST.maxpos &&
4820 UCHARAT(locinput) != ST.c1)
4824 while (locinput <= ST.maxpos
4825 && UCHARAT(locinput) != ST.c1
4826 && UCHARAT(locinput) != ST.c2)
4829 n = locinput - ST.oldloc;
4831 if (locinput > ST.maxpos)
4833 /* PL_reginput == oldloc now */
4836 if (regrepeat(rex, ST.A, n, depth) < n)
4839 PL_reginput = locinput;
4840 CURLY_SETPAREN(ST.paren, ST.count);
4841 if (cur_eval && cur_eval->u.eval.close_paren &&
4842 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4845 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4850 case CURLY_B_min_fail:
4851 /* failed to find B in a non-greedy match where c1,c2 invalid */
4852 if (ST.paren && ST.count)
4853 PL_regoffs[ST.paren].end = -1;
4855 REGCP_UNWIND(ST.cp);
4856 /* failed -- move forward one */
4857 PL_reginput = locinput;
4858 if (regrepeat(rex, ST.A, 1, depth)) {
4860 locinput = PL_reginput;
4861 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4862 ST.count > 0)) /* count overflow ? */
4865 CURLY_SETPAREN(ST.paren, ST.count);
4866 if (cur_eval && cur_eval->u.eval.close_paren &&
4867 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4870 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4878 /* a successful greedy match: now try to match B */
4879 if (cur_eval && cur_eval->u.eval.close_paren &&
4880 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4885 if (ST.c1 != CHRTEST_VOID)
4886 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4887 UTF8_MAXBYTES, 0, uniflags)
4888 : (UV) UCHARAT(PL_reginput);
4889 /* If it could work, try it. */
4890 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4891 CURLY_SETPAREN(ST.paren, ST.count);
4892 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4897 case CURLY_B_max_fail:
4898 /* failed to find B in a greedy match */
4899 if (ST.paren && ST.count)
4900 PL_regoffs[ST.paren].end = -1;
4902 REGCP_UNWIND(ST.cp);
4904 if (--ST.count < ST.min)
4906 PL_reginput = locinput = HOPc(locinput, -1);
4907 goto curly_try_B_max;
4914 /* we've just finished A in /(??{A})B/; now continue with B */
4916 st->u.eval.toggle_reg_flags
4917 = cur_eval->u.eval.toggle_reg_flags;
4918 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4920 st->u.eval.prev_rex = rex_sv; /* inner */
4921 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
4922 rex = (struct regexp *)SvANY(rex_sv);
4923 rexi = RXi_GET(rex);
4924 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4925 ReREFCNT_inc(rex_sv);
4926 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4928 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4929 PL_reglastparen = &rex->lastparen;
4930 PL_reglastcloseparen = &rex->lastcloseparen;
4932 REGCP_SET(st->u.eval.lastcp);
4933 PL_reginput = locinput;
4935 /* Restore parens of the outer rex without popping the
4937 tmpix = PL_savestack_ix;
4938 PL_savestack_ix = cur_eval->u.eval.lastcp;
4940 PL_savestack_ix = tmpix;
4942 st->u.eval.prev_eval = cur_eval;
4943 cur_eval = cur_eval->u.eval.prev_eval;
4945 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
4946 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
4947 if ( nochange_depth )
4950 PUSH_YES_STATE_GOTO(EVAL_AB,
4951 st->u.eval.prev_eval->u.eval.B); /* match B */
4954 if (locinput < reginfo->till) {
4955 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4956 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4958 (long)(locinput - PL_reg_starttry),
4959 (long)(reginfo->till - PL_reg_starttry),
4962 sayNO_SILENT; /* Cannot match: too short. */
4964 PL_reginput = locinput; /* put where regtry can find it */
4965 sayYES; /* Success! */
4967 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4969 PerlIO_printf(Perl_debug_log,
4970 "%*s %ssubpattern success...%s\n",
4971 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4972 PL_reginput = locinput; /* put where regtry can find it */
4973 sayYES; /* Success! */
4976 #define ST st->u.ifmatch
4978 case SUSPEND: /* (?>A) */
4980 PL_reginput = locinput;
4983 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4985 goto ifmatch_trivial_fail_test;
4987 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4989 ifmatch_trivial_fail_test:
4991 char * const s = HOPBACKc(locinput, scan->flags);
4996 sw = 1 - (bool)ST.wanted;
5000 next = scan + ARG(scan);
5008 PL_reginput = locinput;
5012 ST.logical = logical;
5013 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5015 /* execute body of (?...A) */
5016 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5019 case IFMATCH_A_fail: /* body of (?...A) failed */
5020 ST.wanted = !ST.wanted;
5023 case IFMATCH_A: /* body of (?...A) succeeded */
5025 sw = (bool)ST.wanted;
5027 else if (!ST.wanted)
5030 if (OP(ST.me) == SUSPEND)
5031 locinput = PL_reginput;
5033 locinput = PL_reginput = st->locinput;
5034 nextchr = UCHARAT(locinput);
5036 scan = ST.me + ARG(ST.me);
5039 continue; /* execute B */
5044 next = scan + ARG(scan);
5049 reginfo->cutpoint = PL_regeol;
5052 PL_reginput = locinput;
5054 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5055 PUSH_STATE_GOTO(COMMIT_next,next);
5057 case COMMIT_next_fail:
5064 #define ST st->u.mark
5066 ST.prev_mark = mark_state;
5067 ST.mark_name = sv_commit = sv_yes_mark
5068 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5070 ST.mark_loc = PL_reginput = locinput;
5071 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5073 case MARKPOINT_next:
5074 mark_state = ST.prev_mark;
5077 case MARKPOINT_next_fail:
5078 if (popmark && sv_eq(ST.mark_name,popmark))
5080 if (ST.mark_loc > startpoint)
5081 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5082 popmark = NULL; /* we found our mark */
5083 sv_commit = ST.mark_name;
5086 PerlIO_printf(Perl_debug_log,
5087 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5088 REPORT_CODE_OFF+depth*2, "",
5089 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5092 mark_state = ST.prev_mark;
5093 sv_yes_mark = mark_state ?
5094 mark_state->u.mark.mark_name : NULL;
5098 PL_reginput = locinput;
5100 /* (*SKIP) : if we fail we cut here*/
5101 ST.mark_name = NULL;
5102 ST.mark_loc = locinput;
5103 PUSH_STATE_GOTO(SKIP_next,next);
5105 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5106 otherwise do nothing. Meaning we need to scan
5108 regmatch_state *cur = mark_state;
5109 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5112 if ( sv_eq( cur->u.mark.mark_name,
5115 ST.mark_name = find;
5116 PUSH_STATE_GOTO( SKIP_next, next );
5118 cur = cur->u.mark.prev_mark;
5121 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5123 case SKIP_next_fail:
5125 /* (*CUT:NAME) - Set up to search for the name as we
5126 collapse the stack*/
5127 popmark = ST.mark_name;
5129 /* (*CUT) - No name, we cut here.*/
5130 if (ST.mark_loc > startpoint)
5131 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5132 /* but we set sv_commit to latest mark_name if there
5133 is one so they can test to see how things lead to this
5136 sv_commit=mark_state->u.mark.mark_name;
5144 if ( n == (U32)what_len_TRICKYFOLD(locinput,do_utf8,ln) ) {
5146 } else if ( 0xDF == n && !do_utf8 && !UTF ) {
5149 U8 folded[UTF8_MAXBYTES_CASE+1];
5151 const char * const l = locinput;
5152 char *e = PL_regeol;
5153 to_uni_fold(n, folded, &foldlen);
5155 if (ibcmp_utf8((const char*) folded, 0, foldlen, 1,
5156 l, &e, 0, do_utf8)) {
5161 nextchr = UCHARAT(locinput);
5164 if ((n=is_LNBREAK(locinput,do_utf8))) {
5166 nextchr = UCHARAT(locinput);
5171 #define CASE_CLASS(nAmE) \
5173 if ((n=is_##nAmE(locinput,do_utf8))) { \
5175 nextchr = UCHARAT(locinput); \
5180 if ((n=is_##nAmE(locinput,do_utf8))) { \
5183 locinput += UTF8SKIP(locinput); \
5184 nextchr = UCHARAT(locinput); \
5189 CASE_CLASS(HORIZWS);
5193 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5194 PTR2UV(scan), OP(scan));
5195 Perl_croak(aTHX_ "regexp memory corruption");
5199 /* switch break jumps here */
5200 scan = next; /* prepare to execute the next op and ... */
5201 continue; /* ... jump back to the top, reusing st */
5205 /* push a state that backtracks on success */
5206 st->u.yes.prev_yes_state = yes_state;
5210 /* push a new regex state, then continue at scan */
5212 regmatch_state *newst;
5215 regmatch_state *cur = st;
5216 regmatch_state *curyes = yes_state;
5218 regmatch_slab *slab = PL_regmatch_slab;
5219 for (;curd > -1;cur--,curd--) {
5220 if (cur < SLAB_FIRST(slab)) {
5222 cur = SLAB_LAST(slab);
5224 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5225 REPORT_CODE_OFF + 2 + depth * 2,"",
5226 curd, PL_reg_name[cur->resume_state],
5227 (curyes == cur) ? "yes" : ""
5230 curyes = cur->u.yes.prev_yes_state;
5233 DEBUG_STATE_pp("push")
5236 st->locinput = locinput;
5238 if (newst > SLAB_LAST(PL_regmatch_slab))
5239 newst = S_push_slab(aTHX);
5240 PL_regmatch_state = newst;
5242 locinput = PL_reginput;
5243 nextchr = UCHARAT(locinput);
5251 * We get here only if there's trouble -- normally "case END" is
5252 * the terminating point.
5254 Perl_croak(aTHX_ "corrupted regexp pointers");
5260 /* we have successfully completed a subexpression, but we must now
5261 * pop to the state marked by yes_state and continue from there */
5262 assert(st != yes_state);
5264 while (st != yes_state) {
5266 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5267 PL_regmatch_slab = PL_regmatch_slab->prev;
5268 st = SLAB_LAST(PL_regmatch_slab);
5272 DEBUG_STATE_pp("pop (no final)");
5274 DEBUG_STATE_pp("pop (yes)");
5280 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5281 || yes_state > SLAB_LAST(PL_regmatch_slab))
5283 /* not in this slab, pop slab */
5284 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5285 PL_regmatch_slab = PL_regmatch_slab->prev;
5286 st = SLAB_LAST(PL_regmatch_slab);
5288 depth -= (st - yes_state);
5291 yes_state = st->u.yes.prev_yes_state;
5292 PL_regmatch_state = st;
5295 locinput= st->locinput;
5296 nextchr = UCHARAT(locinput);
5298 state_num = st->resume_state + no_final;
5299 goto reenter_switch;
5302 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5303 PL_colors[4], PL_colors[5]));
5305 if (PL_reg_eval_set) {
5306 /* each successfully executed (?{...}) block does the equivalent of
5307 * local $^R = do {...}
5308 * When popping the save stack, all these locals would be undone;
5309 * bypass this by setting the outermost saved $^R to the latest
5311 if (oreplsv != GvSV(PL_replgv))
5312 sv_setsv(oreplsv, GvSV(PL_replgv));
5319 PerlIO_printf(Perl_debug_log,
5320 "%*s %sfailed...%s\n",
5321 REPORT_CODE_OFF+depth*2, "",
5322 PL_colors[4], PL_colors[5])
5334 /* there's a previous state to backtrack to */
5336 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5337 PL_regmatch_slab = PL_regmatch_slab->prev;
5338 st = SLAB_LAST(PL_regmatch_slab);
5340 PL_regmatch_state = st;
5341 locinput= st->locinput;
5342 nextchr = UCHARAT(locinput);
5344 DEBUG_STATE_pp("pop");
5346 if (yes_state == st)
5347 yes_state = st->u.yes.prev_yes_state;
5349 state_num = st->resume_state + 1; /* failure = success + 1 */
5350 goto reenter_switch;
5355 if (rex->intflags & PREGf_VERBARG_SEEN) {
5356 SV *sv_err = get_sv("REGERROR", 1);
5357 SV *sv_mrk = get_sv("REGMARK", 1);
5359 sv_commit = &PL_sv_no;
5361 sv_yes_mark = &PL_sv_yes;
5364 sv_commit = &PL_sv_yes;
5365 sv_yes_mark = &PL_sv_no;
5367 sv_setsv(sv_err, sv_commit);
5368 sv_setsv(sv_mrk, sv_yes_mark);
5371 /* clean up; in particular, free all slabs above current one */
5372 LEAVE_SCOPE(oldsave);
5378 - regrepeat - repeatedly match something simple, report how many
5381 * [This routine now assumes that it will only match on things of length 1.
5382 * That was true before, but now we assume scan - reginput is the count,
5383 * rather than incrementing count on every character. [Er, except utf8.]]
5386 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5389 register char *scan;
5391 register char *loceol = PL_regeol;
5392 register I32 hardcount = 0;
5393 register bool do_utf8 = PL_reg_match_utf8;
5395 PERL_UNUSED_ARG(depth);
5398 PERL_ARGS_ASSERT_REGREPEAT;
5401 if (max == REG_INFTY)
5403 else if (max < loceol - scan)
5404 loceol = scan + max;
5409 while (scan < loceol && hardcount < max && *scan != '\n') {
5410 scan += UTF8SKIP(scan);
5414 while (scan < loceol && *scan != '\n')
5421 while (scan < loceol && hardcount < max) {
5422 scan += UTF8SKIP(scan);
5432 case EXACT: /* length of string is 1 */
5434 while (scan < loceol && UCHARAT(scan) == c)
5437 case EXACTF: /* length of string is 1 */
5439 while (scan < loceol &&
5440 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5443 case EXACTFL: /* length of string is 1 */
5444 PL_reg_flags |= RF_tainted;
5446 while (scan < loceol &&
5447 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5453 while (hardcount < max && scan < loceol &&
5454 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
5455 scan += UTF8SKIP(scan);
5459 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5466 LOAD_UTF8_CHARCLASS_ALNUM();
5467 while (hardcount < max && scan < loceol &&
5468 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5469 scan += UTF8SKIP(scan);
5473 while (scan < loceol && isALNUM(*scan))
5478 PL_reg_flags |= RF_tainted;
5481 while (hardcount < max && scan < loceol &&
5482 isALNUM_LC_utf8((U8*)scan)) {
5483 scan += UTF8SKIP(scan);
5487 while (scan < loceol && isALNUM_LC(*scan))
5494 LOAD_UTF8_CHARCLASS_ALNUM();
5495 while (hardcount < max && scan < loceol &&
5496 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5497 scan += UTF8SKIP(scan);
5501 while (scan < loceol && !isALNUM(*scan))
5506 PL_reg_flags |= RF_tainted;
5509 while (hardcount < max && scan < loceol &&
5510 !isALNUM_LC_utf8((U8*)scan)) {
5511 scan += UTF8SKIP(scan);
5515 while (scan < loceol && !isALNUM_LC(*scan))
5522 LOAD_UTF8_CHARCLASS_SPACE();
5523 while (hardcount < max && scan < loceol &&
5525 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5526 scan += UTF8SKIP(scan);
5530 while (scan < loceol && isSPACE(*scan))
5535 PL_reg_flags |= RF_tainted;
5538 while (hardcount < max && scan < loceol &&
5539 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5540 scan += UTF8SKIP(scan);
5544 while (scan < loceol && isSPACE_LC(*scan))
5551 LOAD_UTF8_CHARCLASS_SPACE();
5552 while (hardcount < max && scan < loceol &&
5554 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5555 scan += UTF8SKIP(scan);
5559 while (scan < loceol && !isSPACE(*scan))
5564 PL_reg_flags |= RF_tainted;
5567 while (hardcount < max && scan < loceol &&
5568 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5569 scan += UTF8SKIP(scan);
5573 while (scan < loceol && !isSPACE_LC(*scan))
5580 LOAD_UTF8_CHARCLASS_DIGIT();
5581 while (hardcount < max && scan < loceol &&
5582 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5583 scan += UTF8SKIP(scan);
5587 while (scan < loceol && isDIGIT(*scan))
5594 LOAD_UTF8_CHARCLASS_DIGIT();
5595 while (hardcount < max && scan < loceol &&
5596 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5597 scan += UTF8SKIP(scan);
5601 while (scan < loceol && !isDIGIT(*scan))
5607 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
5613 LNBREAK can match two latin chars, which is ok,
5614 because we have a null terminated string, but we
5615 have to use hardcount in this situation
5617 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
5626 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
5631 while (scan < loceol && is_HORIZWS_latin1(scan))
5638 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
5639 scan += UTF8SKIP(scan);
5643 while (scan < loceol && !is_HORIZWS_latin1(scan))
5651 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
5656 while (scan < loceol && is_VERTWS_latin1(scan))
5664 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
5665 scan += UTF8SKIP(scan);
5669 while (scan < loceol && !is_VERTWS_latin1(scan))
5675 default: /* Called on something of 0 width. */
5676 break; /* So match right here or not at all. */
5682 c = scan - PL_reginput;
5686 GET_RE_DEBUG_FLAGS_DECL;
5688 SV * const prop = sv_newmortal();
5689 regprop(prog, prop, p);
5690 PerlIO_printf(Perl_debug_log,
5691 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5692 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5700 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5702 - regclass_swash - prepare the utf8 swash
5706 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5712 RXi_GET_DECL(prog,progi);
5713 const struct reg_data * const data = prog ? progi->data : NULL;
5715 PERL_ARGS_ASSERT_REGCLASS_SWASH;
5717 if (data && data->count) {
5718 const U32 n = ARG(node);
5720 if (data->what[n] == 's') {
5721 SV * const rv = MUTABLE_SV(data->data[n]);
5722 AV * const av = MUTABLE_AV(SvRV(rv));
5723 SV **const ary = AvARRAY(av);
5726 /* See the end of regcomp.c:S_regclass() for
5727 * documentation of these array elements. */
5730 a = SvROK(ary[1]) ? &ary[1] : NULL;
5731 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
5735 else if (si && doinit) {
5736 sw = swash_init("utf8", "", si, 1, 0);
5737 (void)av_store(av, 1, sw);
5754 - reginclass - determine if a character falls into a character class
5756 The n is the ANYOF regnode, the p is the target string, lenp
5757 is pointer to the maximum length of how far to go in the p
5758 (if the lenp is zero, UTF8SKIP(p) is used),
5759 do_utf8 tells whether the target string is in UTF-8.
5764 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5767 const char flags = ANYOF_FLAGS(n);
5773 PERL_ARGS_ASSERT_REGINCLASS;
5775 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5776 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5777 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5778 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5779 if (len == (STRLEN)-1)
5780 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5783 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5784 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5787 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5788 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5791 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5795 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5803 utf8_p = bytes_to_utf8(p, &len);
5805 if (swash_fetch(sw, utf8_p, 1))
5807 else if (flags & ANYOF_FOLD) {
5808 if (!match && lenp && av) {
5810 for (i = 0; i <= av_len(av); i++) {
5811 SV* const sv = *av_fetch(av, i, FALSE);
5813 const char * const s = SvPV_const(sv, len);
5814 if (len <= plen && memEQ(s, (char*)utf8_p, len)) {
5822 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5825 to_utf8_fold(utf8_p, tmpbuf, &tmplen);
5826 if (swash_fetch(sw, tmpbuf, 1))
5831 /* If we allocated a string above, free it */
5832 if (! do_utf8) Safefree(utf8_p);
5835 if (match && lenp && *lenp == 0)
5836 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5838 if (!match && c < 256) {
5839 if (ANYOF_BITMAP_TEST(n, c))
5841 else if (flags & ANYOF_FOLD) {
5844 if (flags & ANYOF_LOCALE) {
5845 PL_reg_flags |= RF_tainted;
5846 f = PL_fold_locale[c];
5850 if (f != c && ANYOF_BITMAP_TEST(n, f))
5854 if (!match && (flags & ANYOF_CLASS)) {
5855 PL_reg_flags |= RF_tainted;
5857 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5858 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5859 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5860 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5861 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5862 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5863 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5864 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5865 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5866 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5867 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5868 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5869 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5870 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5871 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5872 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5873 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5874 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5875 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5876 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5877 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5878 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5879 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5880 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5881 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5882 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5883 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5884 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5885 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5886 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5887 ) /* How's that for a conditional? */
5894 return (flags & ANYOF_INVERT) ? !match : match;
5898 S_reghop3(U8 *s, I32 off, const U8* lim)
5902 PERL_ARGS_ASSERT_REGHOP3;
5905 while (off-- && s < lim) {
5906 /* XXX could check well-formedness here */
5911 while (off++ && s > lim) {
5913 if (UTF8_IS_CONTINUED(*s)) {
5914 while (s > lim && UTF8_IS_CONTINUATION(*s))
5917 /* XXX could check well-formedness here */
5924 /* there are a bunch of places where we use two reghop3's that should
5925 be replaced with this routine. but since thats not done yet
5926 we ifdef it out - dmq
5929 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5933 PERL_ARGS_ASSERT_REGHOP4;
5936 while (off-- && s < rlim) {
5937 /* XXX could check well-formedness here */
5942 while (off++ && s > llim) {
5944 if (UTF8_IS_CONTINUED(*s)) {
5945 while (s > llim && UTF8_IS_CONTINUATION(*s))
5948 /* XXX could check well-formedness here */
5956 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5960 PERL_ARGS_ASSERT_REGHOPMAYBE3;
5963 while (off-- && s < lim) {
5964 /* XXX could check well-formedness here */
5971 while (off++ && s > lim) {
5973 if (UTF8_IS_CONTINUED(*s)) {
5974 while (s > lim && UTF8_IS_CONTINUATION(*s))
5977 /* XXX could check well-formedness here */
5986 restore_pos(pTHX_ void *arg)
5989 regexp * const rex = (regexp *)arg;
5990 if (PL_reg_eval_set) {
5991 if (PL_reg_oldsaved) {
5992 rex->subbeg = PL_reg_oldsaved;
5993 rex->sublen = PL_reg_oldsavedlen;
5994 #ifdef PERL_OLD_COPY_ON_WRITE
5995 rex->saved_copy = PL_nrs;
5997 RXp_MATCH_COPIED_on(rex);
5999 PL_reg_magic->mg_len = PL_reg_oldpos;
6000 PL_reg_eval_set = 0;
6001 PL_curpm = PL_reg_oldcurpm;
6006 S_to_utf8_substr(pTHX_ register regexp *prog)
6010 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6013 if (prog->substrs->data[i].substr
6014 && !prog->substrs->data[i].utf8_substr) {
6015 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6016 prog->substrs->data[i].utf8_substr = sv;
6017 sv_utf8_upgrade(sv);
6018 if (SvVALID(prog->substrs->data[i].substr)) {
6019 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6020 if (flags & FBMcf_TAIL) {
6021 /* Trim the trailing \n that fbm_compile added last
6023 SvCUR_set(sv, SvCUR(sv) - 1);
6024 /* Whilst this makes the SV technically "invalid" (as its
6025 buffer is no longer followed by "\0") when fbm_compile()
6026 adds the "\n" back, a "\0" is restored. */
6028 fbm_compile(sv, flags);
6030 if (prog->substrs->data[i].substr == prog->check_substr)
6031 prog->check_utf8 = sv;
6037 S_to_byte_substr(pTHX_ register regexp *prog)
6042 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6045 if (prog->substrs->data[i].utf8_substr
6046 && !prog->substrs->data[i].substr) {
6047 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6048 if (sv_utf8_downgrade(sv, TRUE)) {
6049 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6051 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6052 if (flags & FBMcf_TAIL) {
6053 /* Trim the trailing \n that fbm_compile added last
6055 SvCUR_set(sv, SvCUR(sv) - 1);
6057 fbm_compile(sv, flags);
6063 prog->substrs->data[i].substr = sv;
6064 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6065 prog->check_substr = sv;
6072 * c-indentation-style: bsd
6074 * indent-tabs-mode: t
6077 * ex: set ts=8 sts=4 sw=4 noet: