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);
2844 switch (state_num) {
2846 if (locinput == PL_bostr)
2848 /* reginfo->till = reginfo->bol; */
2853 if (locinput == PL_bostr ||
2854 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2860 if (locinput == PL_bostr)
2864 if (locinput == reginfo->ganch)
2869 /* update the startpoint */
2870 st->u.keeper.val = PL_regoffs[0].start;
2871 PL_reginput = locinput;
2872 PL_regoffs[0].start = locinput - PL_bostr;
2873 PUSH_STATE_GOTO(KEEPS_next, next);
2875 case KEEPS_next_fail:
2876 /* rollback the start point change */
2877 PL_regoffs[0].start = st->u.keeper.val;
2883 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2888 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2890 if (PL_regeol - locinput > 1)
2894 if (PL_regeol != locinput)
2898 if (!nextchr && locinput >= PL_regeol)
2901 locinput += PL_utf8skip[nextchr];
2902 if (locinput > PL_regeol)
2904 nextchr = UCHARAT(locinput);
2907 nextchr = UCHARAT(++locinput);
2910 if (!nextchr && locinput >= PL_regeol)
2912 nextchr = UCHARAT(++locinput);
2915 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2918 locinput += PL_utf8skip[nextchr];
2919 if (locinput > PL_regeol)
2921 nextchr = UCHARAT(locinput);
2924 nextchr = UCHARAT(++locinput);
2928 #define ST st->u.trie
2930 /* In this case the charclass data is available inline so
2931 we can fail fast without a lot of extra overhead.
2933 if (scan->flags == EXACT || !do_utf8) {
2934 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
2936 PerlIO_printf(Perl_debug_log,
2937 "%*s %sfailed to match trie start class...%s\n",
2938 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2947 /* what type of TRIE am I? (utf8 makes this contextual) */
2948 DECL_TRIE_TYPE(scan);
2950 /* what trie are we using right now */
2951 reg_trie_data * const trie
2952 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
2953 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
2954 U32 state = trie->startstate;
2956 if (trie->bitmap && trie_type != trie_utf8_fold &&
2957 !TRIE_BITMAP_TEST(trie,*locinput)
2959 if (trie->states[ state ].wordnum) {
2961 PerlIO_printf(Perl_debug_log,
2962 "%*s %smatched empty string...%s\n",
2963 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2968 PerlIO_printf(Perl_debug_log,
2969 "%*s %sfailed to match trie start class...%s\n",
2970 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
2977 U8 *uc = ( U8* )locinput;
2981 U8 *uscan = (U8*)NULL;
2983 SV *sv_accept_buff = NULL;
2984 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2986 ST.accepted = 0; /* how many accepting states we have seen */
2988 ST.jump = trie->jump;
2991 traverse the TRIE keeping track of all accepting states
2992 we transition through until we get to a failing node.
2995 while ( state && uc <= (U8*)PL_regeol ) {
2996 U32 base = trie->states[ state ].trans.base;
2999 /* We use charid to hold the wordnum as we don't use it
3000 for charid until after we have done the wordnum logic.
3001 We define an alias just so that the wordnum logic reads
3004 #define got_wordnum charid
3005 got_wordnum = trie->states[ state ].wordnum;
3007 if ( got_wordnum ) {
3008 if ( ! ST.accepted ) {
3010 SAVETMPS; /* XXX is this necessary? dmq */
3011 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
3012 sv_accept_buff=newSV(bufflen *
3013 sizeof(reg_trie_accepted) - 1);
3014 SvCUR_set(sv_accept_buff, 0);
3015 SvPOK_on(sv_accept_buff);
3016 sv_2mortal(sv_accept_buff);
3019 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
3022 if (ST.accepted >= bufflen) {
3024 ST.accept_buff =(reg_trie_accepted*)
3025 SvGROW(sv_accept_buff,
3026 bufflen * sizeof(reg_trie_accepted));
3028 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
3029 + sizeof(reg_trie_accepted));
3032 ST.accept_buff[ST.accepted].wordnum = got_wordnum;
3033 ST.accept_buff[ST.accepted].endpos = uc;
3035 } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum]));
3039 DEBUG_TRIE_EXECUTE_r({
3040 DUMP_EXEC_POS( (char *)uc, scan, do_utf8 );
3041 PerlIO_printf( Perl_debug_log,
3042 "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ",
3043 2+depth * 2, "", PL_colors[4],
3044 (UV)state, (UV)ST.accepted );
3048 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3049 uscan, len, uvc, charid, foldlen,
3053 (base + charid > trie->uniquecharcount )
3054 && (base + charid - 1 - trie->uniquecharcount
3056 && trie->trans[base + charid - 1 -
3057 trie->uniquecharcount].check == state)
3059 state = trie->trans[base + charid - 1 -
3060 trie->uniquecharcount ].next;
3071 DEBUG_TRIE_EXECUTE_r(
3072 PerlIO_printf( Perl_debug_log,
3073 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3074 charid, uvc, (UV)state, PL_colors[5] );
3081 PerlIO_printf( Perl_debug_log,
3082 "%*s %sgot %"IVdf" possible matches%s\n",
3083 REPORT_CODE_OFF + depth * 2, "",
3084 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3087 goto trie_first_try; /* jump into the fail handler */
3089 case TRIE_next_fail: /* we failed - try next alterative */
3091 REGCP_UNWIND(ST.cp);
3092 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3093 PL_regoffs[n].end = -1;
3094 *PL_reglastparen = n;
3103 ST.lastparen = *PL_reglastparen;
3106 if ( ST.accepted == 1 ) {
3107 /* only one choice left - just continue */
3109 AV *const trie_words
3110 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3111 SV ** const tmp = av_fetch( trie_words,
3112 ST.accept_buff[ 0 ].wordnum-1, 0 );
3113 SV *sv= tmp ? sv_newmortal() : NULL;
3115 PerlIO_printf( Perl_debug_log,
3116 "%*s %sonly one match left: #%d <%s>%s\n",
3117 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3118 ST.accept_buff[ 0 ].wordnum,
3119 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3120 PL_colors[0], PL_colors[1],
3121 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3123 : "not compiled under -Dr",
3126 PL_reginput = (char *)ST.accept_buff[ 0 ].endpos;
3127 /* in this case we free tmps/leave before we call regmatch
3128 as we wont be using accept_buff again. */
3130 locinput = PL_reginput;
3131 nextchr = UCHARAT(locinput);
3132 if ( !ST.jump || !ST.jump[ST.accept_buff[0].wordnum])
3135 scan = ST.me + ST.jump[ST.accept_buff[0].wordnum];
3136 if (!has_cutgroup) {
3141 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3144 continue; /* execute rest of RE */
3147 if ( !ST.accepted-- ) {
3149 PerlIO_printf( Perl_debug_log,
3150 "%*s %sTRIE failed...%s\n",
3151 REPORT_CODE_OFF+depth*2, "",
3162 There are at least two accepting states left. Presumably
3163 the number of accepting states is going to be low,
3164 typically two. So we simply scan through to find the one
3165 with lowest wordnum. Once we find it, we swap the last
3166 state into its place and decrement the size. We then try to
3167 match the rest of the pattern at the point where the word
3168 ends. If we succeed, control just continues along the
3169 regex; if we fail we return here to try the next accepting
3176 for( cur = 1 ; cur <= ST.accepted ; cur++ ) {
3177 DEBUG_TRIE_EXECUTE_r(
3178 PerlIO_printf( Perl_debug_log,
3179 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
3180 REPORT_CODE_OFF + depth * 2, "", PL_colors[4],
3181 (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur,
3182 ST.accept_buff[ cur ].wordnum, PL_colors[5] );
3185 if (ST.accept_buff[cur].wordnum <
3186 ST.accept_buff[best].wordnum)
3191 AV *const trie_words
3192 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3193 SV ** const tmp = av_fetch( trie_words,
3194 ST.accept_buff[ best ].wordnum - 1, 0 );
3195 regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
3197 ST.me + ST.jump[ST.accept_buff[best].wordnum];
3198 SV *sv= tmp ? sv_newmortal() : NULL;
3200 PerlIO_printf( Perl_debug_log,
3201 "%*s %strying alternation #%d <%s> at node #%d %s\n",
3202 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3203 ST.accept_buff[best].wordnum,
3204 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3205 PL_colors[0], PL_colors[1],
3206 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3207 ) : "not compiled under -Dr",
3208 REG_NODE_NUM(nextop),
3212 if ( best<ST.accepted ) {
3213 reg_trie_accepted tmp = ST.accept_buff[ best ];
3214 ST.accept_buff[ best ] = ST.accept_buff[ ST.accepted ];
3215 ST.accept_buff[ ST.accepted ] = tmp;
3218 PL_reginput = (char *)ST.accept_buff[ best ].endpos;
3219 if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
3222 scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
3224 PUSH_YES_STATE_GOTO(TRIE_next, scan);
3229 /* we dont want to throw this away, see bug 57042*/
3230 if (oreplsv != GvSV(PL_replgv))
3231 sv_setsv(oreplsv, GvSV(PL_replgv));
3238 char *s = STRING(scan);
3240 if (do_utf8 != UTF) {
3241 /* The target and the pattern have differing utf8ness. */
3243 const char * const e = s + ln;
3246 /* The target is utf8, the pattern is not utf8. */
3251 if (NATIVE_TO_UNI(*(U8*)s) !=
3252 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3260 /* The target is not utf8, the pattern is utf8. */
3265 if (NATIVE_TO_UNI(*((U8*)l)) !=
3266 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3274 nextchr = UCHARAT(locinput);
3277 /* The target and the pattern have the same utf8ness. */
3278 /* Inline the first character, for speed. */
3279 if (UCHARAT(s) != nextchr)
3281 if (PL_regeol - locinput < ln)
3283 if (ln > 1 && memNE(s, locinput, ln))
3286 nextchr = UCHARAT(locinput);
3290 PL_reg_flags |= RF_tainted;
3293 char * const s = STRING(scan);
3296 if (do_utf8 || UTF) {
3297 /* Either target or the pattern are utf8. */
3298 const char * const l = locinput;
3299 char *e = PL_regeol;
3301 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
3302 l, &e, 0, do_utf8)) {
3303 /* One more case for the sharp s:
3304 * pack("U0U*", 0xDF) =~ /ss/i,
3305 * the 0xC3 0x9F are the UTF-8
3306 * byte sequence for the U+00DF. */
3309 toLOWER(s[0]) == 's' &&
3311 toLOWER(s[1]) == 's' &&
3318 nextchr = UCHARAT(locinput);
3322 /* Neither the target and the pattern are utf8. */
3324 /* Inline the first character, for speed. */
3325 if (UCHARAT(s) != nextchr &&
3326 UCHARAT(s) != ((OP(scan) == EXACTF)
3327 ? PL_fold : PL_fold_locale)[nextchr])
3329 if (PL_regeol - locinput < ln)
3331 if (ln > 1 && (OP(scan) == EXACTF
3332 ? ibcmp(s, locinput, ln)
3333 : ibcmp_locale(s, locinput, ln)))
3336 nextchr = UCHARAT(locinput);
3341 STRLEN inclasslen = PL_regeol - locinput;
3343 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3345 if (locinput >= PL_regeol)
3347 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3348 nextchr = UCHARAT(locinput);
3353 nextchr = UCHARAT(locinput);
3354 if (!REGINCLASS(rex, scan, (U8*)locinput))
3356 if (!nextchr && locinput >= PL_regeol)
3358 nextchr = UCHARAT(++locinput);
3362 /* If we might have the case of the German sharp s
3363 * in a casefolding Unicode character class. */
3365 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3366 locinput += SHARP_S_SKIP;
3367 nextchr = UCHARAT(locinput);
3373 PL_reg_flags |= RF_tainted;
3379 LOAD_UTF8_CHARCLASS_ALNUM();
3380 if (!(OP(scan) == ALNUM
3381 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3382 : isALNUM_LC_utf8((U8*)locinput)))
3386 locinput += PL_utf8skip[nextchr];
3387 nextchr = UCHARAT(locinput);
3390 if (!(OP(scan) == ALNUM
3391 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3393 nextchr = UCHARAT(++locinput);
3396 PL_reg_flags |= RF_tainted;
3399 if (!nextchr && locinput >= PL_regeol)
3402 LOAD_UTF8_CHARCLASS_ALNUM();
3403 if (OP(scan) == NALNUM
3404 ? (bool)swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3405 : isALNUM_LC_utf8((U8*)locinput))
3409 locinput += PL_utf8skip[nextchr];
3410 nextchr = UCHARAT(locinput);
3413 if (OP(scan) == NALNUM
3414 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3416 nextchr = UCHARAT(++locinput);
3420 PL_reg_flags |= RF_tainted;
3424 /* was last char in word? */
3426 if (locinput == PL_bostr)
3429 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3431 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3433 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3434 ln = isALNUM_uni(ln);
3435 LOAD_UTF8_CHARCLASS_ALNUM();
3436 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3439 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3440 n = isALNUM_LC_utf8((U8*)locinput);
3444 ln = (locinput != PL_bostr) ?
3445 UCHARAT(locinput - 1) : '\n';
3446 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3448 n = isALNUM(nextchr);
3451 ln = isALNUM_LC(ln);
3452 n = isALNUM_LC(nextchr);
3455 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3456 OP(scan) == BOUNDL))
3460 PL_reg_flags |= RF_tainted;
3466 if (UTF8_IS_CONTINUED(nextchr)) {
3467 LOAD_UTF8_CHARCLASS_SPACE();
3468 if (!(OP(scan) == SPACE
3469 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3470 : isSPACE_LC_utf8((U8*)locinput)))
3474 locinput += PL_utf8skip[nextchr];
3475 nextchr = UCHARAT(locinput);
3478 if (!(OP(scan) == SPACE
3479 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3481 nextchr = UCHARAT(++locinput);
3484 if (!(OP(scan) == SPACE
3485 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3487 nextchr = UCHARAT(++locinput);
3491 PL_reg_flags |= RF_tainted;
3494 if (!nextchr && locinput >= PL_regeol)
3497 LOAD_UTF8_CHARCLASS_SPACE();
3498 if (OP(scan) == NSPACE
3499 ? (bool)swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3500 : isSPACE_LC_utf8((U8*)locinput))
3504 locinput += PL_utf8skip[nextchr];
3505 nextchr = UCHARAT(locinput);
3508 if (OP(scan) == NSPACE
3509 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3511 nextchr = UCHARAT(++locinput);
3514 PL_reg_flags |= RF_tainted;
3520 LOAD_UTF8_CHARCLASS_DIGIT();
3521 if (!(OP(scan) == DIGIT
3522 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3523 : isDIGIT_LC_utf8((U8*)locinput)))
3527 locinput += PL_utf8skip[nextchr];
3528 nextchr = UCHARAT(locinput);
3531 if (!(OP(scan) == DIGIT
3532 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3534 nextchr = UCHARAT(++locinput);
3537 PL_reg_flags |= RF_tainted;
3540 if (!nextchr && locinput >= PL_regeol)
3543 LOAD_UTF8_CHARCLASS_DIGIT();
3544 if (OP(scan) == NDIGIT
3545 ? (bool)swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3546 : isDIGIT_LC_utf8((U8*)locinput))
3550 locinput += PL_utf8skip[nextchr];
3551 nextchr = UCHARAT(locinput);
3554 if (OP(scan) == NDIGIT
3555 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3557 nextchr = UCHARAT(++locinput);
3560 if (locinput >= PL_regeol)
3563 LOAD_UTF8_CHARCLASS_MARK();
3564 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3566 locinput += PL_utf8skip[nextchr];
3567 while (locinput < PL_regeol &&
3568 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3569 locinput += UTF8SKIP(locinput);
3570 if (locinput > PL_regeol)
3575 nextchr = UCHARAT(locinput);
3582 PL_reg_flags |= RF_tainted;
3587 n = reg_check_named_buff_matched(rex,scan);
3590 type = REF + ( type - NREF );
3597 PL_reg_flags |= RF_tainted;
3601 n = ARG(scan); /* which paren pair */
3604 ln = PL_regoffs[n].start;
3605 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3606 if (*PL_reglastparen < n || ln == -1)
3607 sayNO; /* Do not match unless seen CLOSEn. */
3608 if (ln == PL_regoffs[n].end)
3612 if (do_utf8 && type != REF) { /* REF can do byte comparison */
3614 const char *e = PL_bostr + PL_regoffs[n].end;
3616 * Note that we can't do the "other character" lookup trick as
3617 * in the 8-bit case (no pun intended) because in Unicode we
3618 * have to map both upper and title case to lower case.
3622 STRLEN ulen1, ulen2;
3623 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3624 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3628 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3629 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3630 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3637 nextchr = UCHARAT(locinput);
3641 /* Inline the first character, for speed. */
3642 if (UCHARAT(s) != nextchr &&
3644 (UCHARAT(s) != (type == REFF
3645 ? PL_fold : PL_fold_locale)[nextchr])))
3647 ln = PL_regoffs[n].end - ln;
3648 if (locinput + ln > PL_regeol)
3650 if (ln > 1 && (type == REF
3651 ? memNE(s, locinput, ln)
3653 ? ibcmp(s, locinput, ln)
3654 : ibcmp_locale(s, locinput, ln))))
3657 nextchr = UCHARAT(locinput);
3667 #define ST st->u.eval
3672 regexp_internal *rei;
3673 regnode *startpoint;
3676 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
3677 if (cur_eval && cur_eval->locinput==locinput) {
3678 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
3679 Perl_croak(aTHX_ "Infinite recursion in regex");
3680 if ( ++nochange_depth > max_nochange_depth )
3682 "Pattern subroutine nesting without pos change"
3683 " exceeded limit in regex");
3690 (void)ReREFCNT_inc(rex_sv);
3691 if (OP(scan)==GOSUB) {
3692 startpoint = scan + ARG2L(scan);
3693 ST.close_paren = ARG(scan);
3695 startpoint = rei->program+1;
3698 goto eval_recurse_doit;
3700 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
3701 if (cur_eval && cur_eval->locinput==locinput) {
3702 if ( ++nochange_depth > max_nochange_depth )
3703 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
3708 /* execute the code in the {...} */
3710 SV ** const before = SP;
3711 OP_4tree * const oop = PL_op;
3712 COP * const ocurcop = PL_curcop;
3716 PL_op = (OP_4tree*)rexi->data->data[n];
3717 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
3718 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3719 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
3720 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
3723 SV *sv_mrk = get_sv("REGMARK", 1);
3724 sv_setsv(sv_mrk, sv_yes_mark);
3727 CALLRUNOPS(aTHX); /* Scalar context. */
3730 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3737 PAD_RESTORE_LOCAL(old_comppad);
3738 PL_curcop = ocurcop;
3741 sv_setsv(save_scalar(PL_replgv), ret);
3745 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
3748 /* extract RE object from returned value; compiling if
3754 SV *const sv = SvRV(ret);
3756 if (SvTYPE(sv) == SVt_REGEXP) {
3758 } else if (SvSMAGICAL(sv)) {
3759 mg = mg_find(sv, PERL_MAGIC_qr);
3762 } else if (SvTYPE(ret) == SVt_REGEXP) {
3764 } else if (SvSMAGICAL(ret)) {
3765 if (SvGMAGICAL(ret)) {
3766 /* I don't believe that there is ever qr magic
3768 assert(!mg_find(ret, PERL_MAGIC_qr));
3769 sv_unmagic(ret, PERL_MAGIC_qr);
3772 mg = mg_find(ret, PERL_MAGIC_qr);
3773 /* testing suggests mg only ends up non-NULL for
3774 scalars who were upgraded and compiled in the
3775 else block below. In turn, this is only
3776 triggered in the "postponed utf8 string" tests
3782 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
3786 rx = reg_temp_copy(rx);
3790 const I32 osize = PL_regsize;
3793 assert (SvUTF8(ret));
3794 } else if (SvUTF8(ret)) {
3795 /* Not doing UTF-8, despite what the SV says. Is
3796 this only if we're trapped in use 'bytes'? */
3797 /* Make a copy of the octet sequence, but without
3798 the flag on, as the compiler now honours the
3799 SvUTF8 flag on ret. */
3801 const char *const p = SvPV(ret, len);
3802 ret = newSVpvn_flags(p, len, SVs_TEMP);
3804 rx = CALLREGCOMP(ret, pm_flags);
3806 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3808 /* This isn't a first class regexp. Instead, it's
3809 caching a regexp onto an existing, Perl visible
3811 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
3816 re = (struct regexp *)SvANY(rx);
3818 RXp_MATCH_COPIED_off(re);
3819 re->subbeg = rex->subbeg;
3820 re->sublen = rex->sublen;
3823 debug_start_match(re_sv, do_utf8, locinput, PL_regeol,
3824 "Matching embedded");
3826 startpoint = rei->program + 1;
3827 ST.close_paren = 0; /* only used for GOSUB */
3828 /* borrowed from regtry */
3829 if (PL_reg_start_tmpl <= re->nparens) {
3830 PL_reg_start_tmpl = re->nparens*3/2 + 3;
3831 if(PL_reg_start_tmp)
3832 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3834 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
3837 eval_recurse_doit: /* Share code with GOSUB below this line */
3838 /* run the pattern returned from (??{...}) */
3839 ST.cp = regcppush(0); /* Save *all* the positions. */
3840 REGCP_SET(ST.lastcp);
3842 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
3844 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
3845 PL_reglastparen = &re->lastparen;
3846 PL_reglastcloseparen = &re->lastcloseparen;
3848 re->lastcloseparen = 0;
3850 PL_reginput = locinput;
3853 /* XXXX This is too dramatic a measure... */
3856 ST.toggle_reg_flags = PL_reg_flags;
3858 PL_reg_flags |= RF_utf8;
3860 PL_reg_flags &= ~RF_utf8;
3861 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
3863 ST.prev_rex = rex_sv;
3864 ST.prev_curlyx = cur_curlyx;
3865 SETREX(rex_sv,re_sv);
3870 ST.prev_eval = cur_eval;
3872 /* now continue from first node in postoned RE */
3873 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
3876 /* logical is 1, /(?(?{...})X|Y)/ */
3877 sw = (bool)SvTRUE(ret);
3882 case EVAL_AB: /* cleanup after a successful (??{A})B */
3883 /* note: this is called twice; first after popping B, then A */
3884 PL_reg_flags ^= ST.toggle_reg_flags;
3885 ReREFCNT_dec(rex_sv);
3886 SETREX(rex_sv,ST.prev_rex);
3887 rex = (struct regexp *)SvANY(rex_sv);
3888 rexi = RXi_GET(rex);
3890 cur_eval = ST.prev_eval;
3891 cur_curlyx = ST.prev_curlyx;
3893 PL_reglastparen = &rex->lastparen;
3894 PL_reglastcloseparen = &rex->lastcloseparen;
3896 /* XXXX This is too dramatic a measure... */
3898 if ( nochange_depth )
3903 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
3904 /* note: this is called twice; first after popping B, then A */
3905 PL_reg_flags ^= ST.toggle_reg_flags;
3906 ReREFCNT_dec(rex_sv);
3907 SETREX(rex_sv,ST.prev_rex);
3908 rex = (struct regexp *)SvANY(rex_sv);
3909 rexi = RXi_GET(rex);
3910 PL_reglastparen = &rex->lastparen;
3911 PL_reglastcloseparen = &rex->lastcloseparen;
3913 PL_reginput = locinput;
3914 REGCP_UNWIND(ST.lastcp);
3916 cur_eval = ST.prev_eval;
3917 cur_curlyx = ST.prev_curlyx;
3918 /* XXXX This is too dramatic a measure... */
3920 if ( nochange_depth )
3926 n = ARG(scan); /* which paren pair */
3927 PL_reg_start_tmp[n] = locinput;
3933 n = ARG(scan); /* which paren pair */
3934 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
3935 PL_regoffs[n].end = locinput - PL_bostr;
3936 /*if (n > PL_regsize)
3938 if (n > *PL_reglastparen)
3939 *PL_reglastparen = n;
3940 *PL_reglastcloseparen = n;
3941 if (cur_eval && cur_eval->u.eval.close_paren == n) {
3949 cursor && OP(cursor)!=END;
3950 cursor=regnext(cursor))
3952 if ( OP(cursor)==CLOSE ){
3954 if ( n <= lastopen ) {
3956 = PL_reg_start_tmp[n] - PL_bostr;
3957 PL_regoffs[n].end = locinput - PL_bostr;
3958 /*if (n > PL_regsize)
3960 if (n > *PL_reglastparen)
3961 *PL_reglastparen = n;
3962 *PL_reglastcloseparen = n;
3963 if ( n == ARG(scan) || (cur_eval &&
3964 cur_eval->u.eval.close_paren == n))
3973 n = ARG(scan); /* which paren pair */
3974 sw = (bool)(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
3977 /* reg_check_named_buff_matched returns 0 for no match */
3978 sw = (bool)(0 < reg_check_named_buff_matched(rex,scan));
3982 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
3988 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3990 next = NEXTOPER(NEXTOPER(scan));
3992 next = scan + ARG(scan);
3993 if (OP(next) == IFTHEN) /* Fake one. */
3994 next = NEXTOPER(NEXTOPER(next));
3998 logical = scan->flags;
4001 /*******************************************************************
4003 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4004 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4005 STAR/PLUS/CURLY/CURLYN are used instead.)
4007 A*B is compiled as <CURLYX><A><WHILEM><B>
4009 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4010 state, which contains the current count, initialised to -1. It also sets
4011 cur_curlyx to point to this state, with any previous value saved in the
4014 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4015 since the pattern may possibly match zero times (i.e. it's a while {} loop
4016 rather than a do {} while loop).
4018 Each entry to WHILEM represents a successful match of A. The count in the
4019 CURLYX block is incremented, another WHILEM state is pushed, and execution
4020 passes to A or B depending on greediness and the current count.
4022 For example, if matching against the string a1a2a3b (where the aN are
4023 substrings that match /A/), then the match progresses as follows: (the
4024 pushed states are interspersed with the bits of strings matched so far):
4027 <CURLYX cnt=0><WHILEM>
4028 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4029 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4030 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4031 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4033 (Contrast this with something like CURLYM, which maintains only a single
4037 a1 <CURLYM cnt=1> a2
4038 a1 a2 <CURLYM cnt=2> a3
4039 a1 a2 a3 <CURLYM cnt=3> b
4042 Each WHILEM state block marks a point to backtrack to upon partial failure
4043 of A or B, and also contains some minor state data related to that
4044 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4045 overall state, such as the count, and pointers to the A and B ops.
4047 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4048 must always point to the *current* CURLYX block, the rules are:
4050 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4051 and set cur_curlyx to point the new block.
4053 When popping the CURLYX block after a successful or unsuccessful match,
4054 restore the previous cur_curlyx.
4056 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4057 to the outer one saved in the CURLYX block.
4059 When popping the WHILEM block after a successful or unsuccessful B match,
4060 restore the previous cur_curlyx.
4062 Here's an example for the pattern (AI* BI)*BO
4063 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4066 curlyx backtrack stack
4067 ------ ---------------
4069 CO <CO prev=NULL> <WO>
4070 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4071 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4072 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4074 At this point the pattern succeeds, and we work back down the stack to
4075 clean up, restoring as we go:
4077 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4078 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4079 CO <CO prev=NULL> <WO>
4082 *******************************************************************/
4084 #define ST st->u.curlyx
4086 case CURLYX: /* start of /A*B/ (for complex A) */
4088 /* No need to save/restore up to this paren */
4089 I32 parenfloor = scan->flags;
4091 assert(next); /* keep Coverity happy */
4092 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4095 /* XXXX Probably it is better to teach regpush to support
4096 parenfloor > PL_regsize... */
4097 if (parenfloor > (I32)*PL_reglastparen)
4098 parenfloor = *PL_reglastparen; /* Pessimization... */
4100 ST.prev_curlyx= cur_curlyx;
4102 ST.cp = PL_savestack_ix;
4104 /* these fields contain the state of the current curly.
4105 * they are accessed by subsequent WHILEMs */
4106 ST.parenfloor = parenfloor;
4107 ST.min = ARG1(scan);
4108 ST.max = ARG2(scan);
4109 ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
4113 ST.count = -1; /* this will be updated by WHILEM */
4114 ST.lastloc = NULL; /* this will be updated by WHILEM */
4116 PL_reginput = locinput;
4117 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4121 case CURLYX_end: /* just finished matching all of A*B */
4122 cur_curlyx = ST.prev_curlyx;
4126 case CURLYX_end_fail: /* just failed to match all of A*B */
4128 cur_curlyx = ST.prev_curlyx;
4134 #define ST st->u.whilem
4136 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4138 /* see the discussion above about CURLYX/WHILEM */
4140 assert(cur_curlyx); /* keep Coverity happy */
4141 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4142 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4143 ST.cache_offset = 0;
4146 PL_reginput = locinput;
4148 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4149 "%*s whilem: matched %ld out of %ld..%ld\n",
4150 REPORT_CODE_OFF+depth*2, "", (long)n,
4151 (long)cur_curlyx->u.curlyx.min,
4152 (long)cur_curlyx->u.curlyx.max)
4155 /* First just match a string of min A's. */
4157 if (n < cur_curlyx->u.curlyx.min) {
4158 cur_curlyx->u.curlyx.lastloc = locinput;
4159 PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A);
4163 /* If degenerate A matches "", assume A done. */
4165 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4166 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4167 "%*s whilem: empty match detected, trying continuation...\n",
4168 REPORT_CODE_OFF+depth*2, "")
4170 goto do_whilem_B_max;
4173 /* super-linear cache processing */
4177 if (!PL_reg_maxiter) {
4178 /* start the countdown: Postpone detection until we
4179 * know the match is not *that* much linear. */
4180 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4181 /* possible overflow for long strings and many CURLYX's */
4182 if (PL_reg_maxiter < 0)
4183 PL_reg_maxiter = I32_MAX;
4184 PL_reg_leftiter = PL_reg_maxiter;
4187 if (PL_reg_leftiter-- == 0) {
4188 /* initialise cache */
4189 const I32 size = (PL_reg_maxiter + 7)/8;
4190 if (PL_reg_poscache) {
4191 if ((I32)PL_reg_poscache_size < size) {
4192 Renew(PL_reg_poscache, size, char);
4193 PL_reg_poscache_size = size;
4195 Zero(PL_reg_poscache, size, char);
4198 PL_reg_poscache_size = size;
4199 Newxz(PL_reg_poscache, size, char);
4201 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4202 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4203 PL_colors[4], PL_colors[5])
4207 if (PL_reg_leftiter < 0) {
4208 /* have we already failed at this position? */
4210 offset = (scan->flags & 0xf) - 1
4211 + (locinput - PL_bostr) * (scan->flags>>4);
4212 mask = 1 << (offset % 8);
4214 if (PL_reg_poscache[offset] & mask) {
4215 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4216 "%*s whilem: (cache) already tried at this position...\n",
4217 REPORT_CODE_OFF+depth*2, "")
4219 sayNO; /* cache records failure */
4221 ST.cache_offset = offset;
4222 ST.cache_mask = mask;
4226 /* Prefer B over A for minimal matching. */
4228 if (cur_curlyx->u.curlyx.minmod) {
4229 ST.save_curlyx = cur_curlyx;
4230 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4231 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4232 REGCP_SET(ST.lastcp);
4233 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4237 /* Prefer A over B for maximal matching. */
4239 if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */
4240 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4241 cur_curlyx->u.curlyx.lastloc = locinput;
4242 REGCP_SET(ST.lastcp);
4243 PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A);
4246 goto do_whilem_B_max;
4250 case WHILEM_B_min: /* just matched B in a minimal match */
4251 case WHILEM_B_max: /* just matched B in a maximal match */
4252 cur_curlyx = ST.save_curlyx;
4256 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4257 cur_curlyx = ST.save_curlyx;
4258 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4259 cur_curlyx->u.curlyx.count--;
4263 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4264 REGCP_UNWIND(ST.lastcp);
4267 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4268 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4269 cur_curlyx->u.curlyx.count--;
4273 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4274 REGCP_UNWIND(ST.lastcp);
4275 regcppop(rex); /* Restore some previous $<digit>s? */
4276 PL_reginput = locinput;
4277 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4278 "%*s whilem: failed, trying continuation...\n",
4279 REPORT_CODE_OFF+depth*2, "")
4282 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4283 && ckWARN(WARN_REGEXP)
4284 && !(PL_reg_flags & RF_warned))
4286 PL_reg_flags |= RF_warned;
4287 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4288 "Complex regular subexpression recursion",
4293 ST.save_curlyx = cur_curlyx;
4294 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4295 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4298 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4299 cur_curlyx = ST.save_curlyx;
4300 REGCP_UNWIND(ST.lastcp);
4303 if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) {
4304 /* Maximum greed exceeded */
4305 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4306 && ckWARN(WARN_REGEXP)
4307 && !(PL_reg_flags & RF_warned))
4309 PL_reg_flags |= RF_warned;
4310 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4311 "%s limit (%d) exceeded",
4312 "Complex regular subexpression recursion",
4315 cur_curlyx->u.curlyx.count--;
4319 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4320 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4322 /* Try grabbing another A and see if it helps. */
4323 PL_reginput = locinput;
4324 cur_curlyx->u.curlyx.lastloc = locinput;
4325 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4326 REGCP_SET(ST.lastcp);
4327 PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A);
4331 #define ST st->u.branch
4333 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4334 next = scan + ARG(scan);
4337 scan = NEXTOPER(scan);
4340 case BRANCH: /* /(...|A|...)/ */
4341 scan = NEXTOPER(scan); /* scan now points to inner node */
4342 ST.lastparen = *PL_reglastparen;
4343 ST.next_branch = next;
4345 PL_reginput = locinput;
4347 /* Now go into the branch */
4349 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4351 PUSH_STATE_GOTO(BRANCH_next, scan);
4355 PL_reginput = locinput;
4356 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4357 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4358 PUSH_STATE_GOTO(CUTGROUP_next,next);
4360 case CUTGROUP_next_fail:
4363 if (st->u.mark.mark_name)
4364 sv_commit = st->u.mark.mark_name;
4370 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4375 REGCP_UNWIND(ST.cp);
4376 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4377 PL_regoffs[n].end = -1;
4378 *PL_reglastparen = n;
4379 /*dmq: *PL_reglastcloseparen = n; */
4380 scan = ST.next_branch;
4381 /* no more branches? */
4382 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4384 PerlIO_printf( Perl_debug_log,
4385 "%*s %sBRANCH failed...%s\n",
4386 REPORT_CODE_OFF+depth*2, "",
4392 continue; /* execute next BRANCH[J] op */
4400 #define ST st->u.curlym
4402 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4404 /* This is an optimisation of CURLYX that enables us to push
4405 * only a single backtracking state, no matter now many matches
4406 * there are in {m,n}. It relies on the pattern being constant
4407 * length, with no parens to influence future backrefs
4411 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4413 /* if paren positive, emulate an OPEN/CLOSE around A */
4415 U32 paren = ST.me->flags;
4416 if (paren > PL_regsize)
4418 if (paren > *PL_reglastparen)
4419 *PL_reglastparen = paren;
4420 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4428 ST.c1 = CHRTEST_UNINIT;
4431 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4434 curlym_do_A: /* execute the A in /A{m,n}B/ */
4435 PL_reginput = locinput;
4436 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4439 case CURLYM_A: /* we've just matched an A */
4440 locinput = st->locinput;
4441 nextchr = UCHARAT(locinput);
4444 /* after first match, determine A's length: u.curlym.alen */
4445 if (ST.count == 1) {
4446 if (PL_reg_match_utf8) {
4448 while (s < PL_reginput) {
4454 ST.alen = PL_reginput - locinput;
4457 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4460 PerlIO_printf(Perl_debug_log,
4461 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4462 (int)(REPORT_CODE_OFF+(depth*2)), "",
4463 (IV) ST.count, (IV)ST.alen)
4466 locinput = PL_reginput;
4468 if (cur_eval && cur_eval->u.eval.close_paren &&
4469 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4472 if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
4473 goto curlym_do_A; /* try to match another A */
4474 goto curlym_do_B; /* try to match B */
4476 case CURLYM_A_fail: /* just failed to match an A */
4477 REGCP_UNWIND(ST.cp);
4479 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4480 || (cur_eval && cur_eval->u.eval.close_paren &&
4481 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4484 curlym_do_B: /* execute the B in /A{m,n}B/ */
4485 PL_reginput = locinput;
4486 if (ST.c1 == CHRTEST_UNINIT) {
4487 /* calculate c1 and c2 for possible match of 1st char
4488 * following curly */
4489 ST.c1 = ST.c2 = CHRTEST_VOID;
4490 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4491 regnode *text_node = ST.B;
4492 if (! HAS_TEXT(text_node))
4493 FIND_NEXT_IMPT(text_node);
4496 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4498 But the former is redundant in light of the latter.
4500 if this changes back then the macro for
4501 IS_TEXT and friends need to change.
4503 if (PL_regkind[OP(text_node)] == EXACT)
4506 ST.c1 = (U8)*STRING(text_node);
4508 (IS_TEXTF(text_node))
4510 : (IS_TEXTFL(text_node))
4511 ? PL_fold_locale[ST.c1]
4518 PerlIO_printf(Perl_debug_log,
4519 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4520 (int)(REPORT_CODE_OFF+(depth*2)),
4523 if (ST.c1 != CHRTEST_VOID
4524 && UCHARAT(PL_reginput) != ST.c1
4525 && UCHARAT(PL_reginput) != ST.c2)
4527 /* simulate B failing */
4529 PerlIO_printf(Perl_debug_log,
4530 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
4531 (int)(REPORT_CODE_OFF+(depth*2)),"",
4534 state_num = CURLYM_B_fail;
4535 goto reenter_switch;
4539 /* mark current A as captured */
4540 I32 paren = ST.me->flags;
4542 PL_regoffs[paren].start
4543 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4544 PL_regoffs[paren].end = PL_reginput - PL_bostr;
4545 /*dmq: *PL_reglastcloseparen = paren; */
4548 PL_regoffs[paren].end = -1;
4549 if (cur_eval && cur_eval->u.eval.close_paren &&
4550 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4559 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4562 case CURLYM_B_fail: /* just failed to match a B */
4563 REGCP_UNWIND(ST.cp);
4565 if (ST.count == ARG2(ST.me) /* max */)
4567 goto curlym_do_A; /* try to match a further A */
4569 /* backtrack one A */
4570 if (ST.count == ARG1(ST.me) /* min */)
4573 locinput = HOPc(locinput, -ST.alen);
4574 goto curlym_do_B; /* try to match B */
4577 #define ST st->u.curly
4579 #define CURLY_SETPAREN(paren, success) \
4582 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
4583 PL_regoffs[paren].end = locinput - PL_bostr; \
4584 *PL_reglastcloseparen = paren; \
4587 PL_regoffs[paren].end = -1; \
4590 case STAR: /* /A*B/ where A is width 1 */
4594 scan = NEXTOPER(scan);
4596 case PLUS: /* /A+B/ where A is width 1 */
4600 scan = NEXTOPER(scan);
4602 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
4603 ST.paren = scan->flags; /* Which paren to set */
4604 if (ST.paren > PL_regsize)
4605 PL_regsize = ST.paren;
4606 if (ST.paren > *PL_reglastparen)
4607 *PL_reglastparen = ST.paren;
4608 ST.min = ARG1(scan); /* min to match */
4609 ST.max = ARG2(scan); /* max to match */
4610 if (cur_eval && cur_eval->u.eval.close_paren &&
4611 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4615 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4617 case CURLY: /* /A{m,n}B/ where A is width 1 */
4619 ST.min = ARG1(scan); /* min to match */
4620 ST.max = ARG2(scan); /* max to match */
4621 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4624 * Lookahead to avoid useless match attempts
4625 * when we know what character comes next.
4627 * Used to only do .*x and .*?x, but now it allows
4628 * for )'s, ('s and (?{ ... })'s to be in the way
4629 * of the quantifier and the EXACT-like node. -- japhy
4632 if (ST.min > ST.max) /* XXX make this a compile-time check? */
4634 if (HAS_TEXT(next) || JUMPABLE(next)) {
4636 regnode *text_node = next;
4638 if (! HAS_TEXT(text_node))
4639 FIND_NEXT_IMPT(text_node);
4641 if (! HAS_TEXT(text_node))
4642 ST.c1 = ST.c2 = CHRTEST_VOID;
4644 if ( PL_regkind[OP(text_node)] != EXACT ) {
4645 ST.c1 = ST.c2 = CHRTEST_VOID;
4646 goto assume_ok_easy;
4649 s = (U8*)STRING(text_node);
4651 /* Currently we only get here when
4653 PL_rekind[OP(text_node)] == EXACT
4655 if this changes back then the macro for IS_TEXT and
4656 friends need to change. */
4659 if (IS_TEXTF(text_node))
4660 ST.c2 = PL_fold[ST.c1];
4661 else if (IS_TEXTFL(text_node))
4662 ST.c2 = PL_fold_locale[ST.c1];
4665 if (IS_TEXTF(text_node)) {
4666 STRLEN ulen1, ulen2;
4667 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4668 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4670 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4671 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4673 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
4675 0 : UTF8_ALLOW_ANY);
4676 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
4678 0 : UTF8_ALLOW_ANY);
4680 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4682 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4687 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4694 ST.c1 = ST.c2 = CHRTEST_VOID;
4699 PL_reginput = locinput;
4702 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
4705 locinput = PL_reginput;
4707 if (ST.c1 == CHRTEST_VOID)
4708 goto curly_try_B_min;
4710 ST.oldloc = locinput;
4712 /* set ST.maxpos to the furthest point along the
4713 * string that could possibly match */
4714 if (ST.max == REG_INFTY) {
4715 ST.maxpos = PL_regeol - 1;
4717 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
4721 int m = ST.max - ST.min;
4722 for (ST.maxpos = locinput;
4723 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
4724 ST.maxpos += UTF8SKIP(ST.maxpos);
4727 ST.maxpos = locinput + ST.max - ST.min;
4728 if (ST.maxpos >= PL_regeol)
4729 ST.maxpos = PL_regeol - 1;
4731 goto curly_try_B_min_known;
4735 ST.count = regrepeat(rex, ST.A, ST.max, depth);
4736 locinput = PL_reginput;
4737 if (ST.count < ST.min)
4739 if ((ST.count > ST.min)
4740 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
4742 /* A{m,n} must come at the end of the string, there's
4743 * no point in backing off ... */
4745 /* ...except that $ and \Z can match before *and* after
4746 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4747 We may back off by one in this case. */
4748 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
4752 goto curly_try_B_max;
4757 case CURLY_B_min_known_fail:
4758 /* failed to find B in a non-greedy match where c1,c2 valid */
4759 if (ST.paren && ST.count)
4760 PL_regoffs[ST.paren].end = -1;
4762 PL_reginput = locinput; /* Could be reset... */
4763 REGCP_UNWIND(ST.cp);
4764 /* Couldn't or didn't -- move forward. */
4765 ST.oldloc = locinput;
4767 locinput += UTF8SKIP(locinput);
4771 curly_try_B_min_known:
4772 /* find the next place where 'B' could work, then call B */
4776 n = (ST.oldloc == locinput) ? 0 : 1;
4777 if (ST.c1 == ST.c2) {
4779 /* set n to utf8_distance(oldloc, locinput) */
4780 while (locinput <= ST.maxpos &&
4781 utf8n_to_uvchr((U8*)locinput,
4782 UTF8_MAXBYTES, &len,
4783 uniflags) != (UV)ST.c1) {
4789 /* set n to utf8_distance(oldloc, locinput) */
4790 while (locinput <= ST.maxpos) {
4792 const UV c = utf8n_to_uvchr((U8*)locinput,
4793 UTF8_MAXBYTES, &len,
4795 if (c == (UV)ST.c1 || c == (UV)ST.c2)
4803 if (ST.c1 == ST.c2) {
4804 while (locinput <= ST.maxpos &&
4805 UCHARAT(locinput) != ST.c1)
4809 while (locinput <= ST.maxpos
4810 && UCHARAT(locinput) != ST.c1
4811 && UCHARAT(locinput) != ST.c2)
4814 n = locinput - ST.oldloc;
4816 if (locinput > ST.maxpos)
4818 /* PL_reginput == oldloc now */
4821 if (regrepeat(rex, ST.A, n, depth) < n)
4824 PL_reginput = locinput;
4825 CURLY_SETPAREN(ST.paren, ST.count);
4826 if (cur_eval && cur_eval->u.eval.close_paren &&
4827 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4830 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
4835 case CURLY_B_min_fail:
4836 /* failed to find B in a non-greedy match where c1,c2 invalid */
4837 if (ST.paren && ST.count)
4838 PL_regoffs[ST.paren].end = -1;
4840 REGCP_UNWIND(ST.cp);
4841 /* failed -- move forward one */
4842 PL_reginput = locinput;
4843 if (regrepeat(rex, ST.A, 1, depth)) {
4845 locinput = PL_reginput;
4846 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
4847 ST.count > 0)) /* count overflow ? */
4850 CURLY_SETPAREN(ST.paren, ST.count);
4851 if (cur_eval && cur_eval->u.eval.close_paren &&
4852 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4855 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
4863 /* a successful greedy match: now try to match B */
4864 if (cur_eval && cur_eval->u.eval.close_paren &&
4865 cur_eval->u.eval.close_paren == (U32)ST.paren) {
4870 if (ST.c1 != CHRTEST_VOID)
4871 c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput,
4872 UTF8_MAXBYTES, 0, uniflags)
4873 : (UV) UCHARAT(PL_reginput);
4874 /* If it could work, try it. */
4875 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
4876 CURLY_SETPAREN(ST.paren, ST.count);
4877 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
4882 case CURLY_B_max_fail:
4883 /* failed to find B in a greedy match */
4884 if (ST.paren && ST.count)
4885 PL_regoffs[ST.paren].end = -1;
4887 REGCP_UNWIND(ST.cp);
4889 if (--ST.count < ST.min)
4891 PL_reginput = locinput = HOPc(locinput, -1);
4892 goto curly_try_B_max;
4899 /* we've just finished A in /(??{A})B/; now continue with B */
4901 st->u.eval.toggle_reg_flags
4902 = cur_eval->u.eval.toggle_reg_flags;
4903 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
4905 st->u.eval.prev_rex = rex_sv; /* inner */
4906 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
4907 rex = (struct regexp *)SvANY(rex_sv);
4908 rexi = RXi_GET(rex);
4909 cur_curlyx = cur_eval->u.eval.prev_curlyx;
4910 ReREFCNT_inc(rex_sv);
4911 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
4912 REGCP_SET(st->u.eval.lastcp);
4913 PL_reginput = locinput;
4915 /* Restore parens of the outer rex without popping the
4917 tmpix = PL_savestack_ix;
4918 PL_savestack_ix = cur_eval->u.eval.lastcp;
4920 PL_savestack_ix = tmpix;
4922 st->u.eval.prev_eval = cur_eval;
4923 cur_eval = cur_eval->u.eval.prev_eval;
4925 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
4926 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
4927 if ( nochange_depth )
4930 PUSH_YES_STATE_GOTO(EVAL_AB,
4931 st->u.eval.prev_eval->u.eval.B); /* match B */
4934 if (locinput < reginfo->till) {
4935 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4936 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4938 (long)(locinput - PL_reg_starttry),
4939 (long)(reginfo->till - PL_reg_starttry),
4942 sayNO_SILENT; /* Cannot match: too short. */
4944 PL_reginput = locinput; /* put where regtry can find it */
4945 sayYES; /* Success! */
4947 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4949 PerlIO_printf(Perl_debug_log,
4950 "%*s %ssubpattern success...%s\n",
4951 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
4952 PL_reginput = locinput; /* put where regtry can find it */
4953 sayYES; /* Success! */
4956 #define ST st->u.ifmatch
4958 case SUSPEND: /* (?>A) */
4960 PL_reginput = locinput;
4963 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
4965 goto ifmatch_trivial_fail_test;
4967 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
4969 ifmatch_trivial_fail_test:
4971 char * const s = HOPBACKc(locinput, scan->flags);
4976 sw = 1 - (bool)ST.wanted;
4980 next = scan + ARG(scan);
4988 PL_reginput = locinput;
4992 ST.logical = logical;
4993 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
4995 /* execute body of (?...A) */
4996 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
4999 case IFMATCH_A_fail: /* body of (?...A) failed */
5000 ST.wanted = !ST.wanted;
5003 case IFMATCH_A: /* body of (?...A) succeeded */
5005 sw = (bool)ST.wanted;
5007 else if (!ST.wanted)
5010 if (OP(ST.me) == SUSPEND)
5011 locinput = PL_reginput;
5013 locinput = PL_reginput = st->locinput;
5014 nextchr = UCHARAT(locinput);
5016 scan = ST.me + ARG(ST.me);
5019 continue; /* execute B */
5024 next = scan + ARG(scan);
5029 reginfo->cutpoint = PL_regeol;
5032 PL_reginput = locinput;
5034 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5035 PUSH_STATE_GOTO(COMMIT_next,next);
5037 case COMMIT_next_fail:
5044 #define ST st->u.mark
5046 ST.prev_mark = mark_state;
5047 ST.mark_name = sv_commit = sv_yes_mark
5048 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5050 ST.mark_loc = PL_reginput = locinput;
5051 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5053 case MARKPOINT_next:
5054 mark_state = ST.prev_mark;
5057 case MARKPOINT_next_fail:
5058 if (popmark && sv_eq(ST.mark_name,popmark))
5060 if (ST.mark_loc > startpoint)
5061 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5062 popmark = NULL; /* we found our mark */
5063 sv_commit = ST.mark_name;
5066 PerlIO_printf(Perl_debug_log,
5067 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5068 REPORT_CODE_OFF+depth*2, "",
5069 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5072 mark_state = ST.prev_mark;
5073 sv_yes_mark = mark_state ?
5074 mark_state->u.mark.mark_name : NULL;
5078 PL_reginput = locinput;
5080 /* (*SKIP) : if we fail we cut here*/
5081 ST.mark_name = NULL;
5082 ST.mark_loc = locinput;
5083 PUSH_STATE_GOTO(SKIP_next,next);
5085 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5086 otherwise do nothing. Meaning we need to scan
5088 regmatch_state *cur = mark_state;
5089 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5092 if ( sv_eq( cur->u.mark.mark_name,
5095 ST.mark_name = find;
5096 PUSH_STATE_GOTO( SKIP_next, next );
5098 cur = cur->u.mark.prev_mark;
5101 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5103 case SKIP_next_fail:
5105 /* (*CUT:NAME) - Set up to search for the name as we
5106 collapse the stack*/
5107 popmark = ST.mark_name;
5109 /* (*CUT) - No name, we cut here.*/
5110 if (ST.mark_loc > startpoint)
5111 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5112 /* but we set sv_commit to latest mark_name if there
5113 is one so they can test to see how things lead to this
5116 sv_commit=mark_state->u.mark.mark_name;
5124 if ( n == (U32)what_len_TRICKYFOLD(locinput,do_utf8,ln) ) {
5126 } else if ( 0xDF == n && !do_utf8 && !UTF ) {
5129 U8 folded[UTF8_MAXBYTES_CASE+1];
5131 const char * const l = locinput;
5132 char *e = PL_regeol;
5133 to_uni_fold(n, folded, &foldlen);
5135 if (ibcmp_utf8((const char*) folded, 0, foldlen, 1,
5136 l, &e, 0, do_utf8)) {
5141 nextchr = UCHARAT(locinput);
5144 if ((n=is_LNBREAK(locinput,do_utf8))) {
5146 nextchr = UCHARAT(locinput);
5151 #define CASE_CLASS(nAmE) \
5153 if ((n=is_##nAmE(locinput,do_utf8))) { \
5155 nextchr = UCHARAT(locinput); \
5160 if ((n=is_##nAmE(locinput,do_utf8))) { \
5163 locinput += UTF8SKIP(locinput); \
5164 nextchr = UCHARAT(locinput); \
5169 CASE_CLASS(HORIZWS);
5173 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5174 PTR2UV(scan), OP(scan));
5175 Perl_croak(aTHX_ "regexp memory corruption");
5179 /* switch break jumps here */
5180 scan = next; /* prepare to execute the next op and ... */
5181 continue; /* ... jump back to the top, reusing st */
5185 /* push a state that backtracks on success */
5186 st->u.yes.prev_yes_state = yes_state;
5190 /* push a new regex state, then continue at scan */
5192 regmatch_state *newst;
5195 regmatch_state *cur = st;
5196 regmatch_state *curyes = yes_state;
5198 regmatch_slab *slab = PL_regmatch_slab;
5199 for (;curd > -1;cur--,curd--) {
5200 if (cur < SLAB_FIRST(slab)) {
5202 cur = SLAB_LAST(slab);
5204 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5205 REPORT_CODE_OFF + 2 + depth * 2,"",
5206 curd, PL_reg_name[cur->resume_state],
5207 (curyes == cur) ? "yes" : ""
5210 curyes = cur->u.yes.prev_yes_state;
5213 DEBUG_STATE_pp("push")
5216 st->locinput = locinput;
5218 if (newst > SLAB_LAST(PL_regmatch_slab))
5219 newst = S_push_slab(aTHX);
5220 PL_regmatch_state = newst;
5222 locinput = PL_reginput;
5223 nextchr = UCHARAT(locinput);
5231 * We get here only if there's trouble -- normally "case END" is
5232 * the terminating point.
5234 Perl_croak(aTHX_ "corrupted regexp pointers");
5240 /* we have successfully completed a subexpression, but we must now
5241 * pop to the state marked by yes_state and continue from there */
5242 assert(st != yes_state);
5244 while (st != yes_state) {
5246 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5247 PL_regmatch_slab = PL_regmatch_slab->prev;
5248 st = SLAB_LAST(PL_regmatch_slab);
5252 DEBUG_STATE_pp("pop (no final)");
5254 DEBUG_STATE_pp("pop (yes)");
5260 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5261 || yes_state > SLAB_LAST(PL_regmatch_slab))
5263 /* not in this slab, pop slab */
5264 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5265 PL_regmatch_slab = PL_regmatch_slab->prev;
5266 st = SLAB_LAST(PL_regmatch_slab);
5268 depth -= (st - yes_state);
5271 yes_state = st->u.yes.prev_yes_state;
5272 PL_regmatch_state = st;
5275 locinput= st->locinput;
5276 nextchr = UCHARAT(locinput);
5278 state_num = st->resume_state + no_final;
5279 goto reenter_switch;
5282 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5283 PL_colors[4], PL_colors[5]));
5285 if (PL_reg_eval_set) {
5286 /* each successfully executed (?{...}) block does the equivalent of
5287 * local $^R = do {...}
5288 * When popping the save stack, all these locals would be undone;
5289 * bypass this by setting the outermost saved $^R to the latest
5291 if (oreplsv != GvSV(PL_replgv))
5292 sv_setsv(oreplsv, GvSV(PL_replgv));
5299 PerlIO_printf(Perl_debug_log,
5300 "%*s %sfailed...%s\n",
5301 REPORT_CODE_OFF+depth*2, "",
5302 PL_colors[4], PL_colors[5])
5314 /* there's a previous state to backtrack to */
5316 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5317 PL_regmatch_slab = PL_regmatch_slab->prev;
5318 st = SLAB_LAST(PL_regmatch_slab);
5320 PL_regmatch_state = st;
5321 locinput= st->locinput;
5322 nextchr = UCHARAT(locinput);
5324 DEBUG_STATE_pp("pop");
5326 if (yes_state == st)
5327 yes_state = st->u.yes.prev_yes_state;
5329 state_num = st->resume_state + 1; /* failure = success + 1 */
5330 goto reenter_switch;
5335 if (rex->intflags & PREGf_VERBARG_SEEN) {
5336 SV *sv_err = get_sv("REGERROR", 1);
5337 SV *sv_mrk = get_sv("REGMARK", 1);
5339 sv_commit = &PL_sv_no;
5341 sv_yes_mark = &PL_sv_yes;
5344 sv_commit = &PL_sv_yes;
5345 sv_yes_mark = &PL_sv_no;
5347 sv_setsv(sv_err, sv_commit);
5348 sv_setsv(sv_mrk, sv_yes_mark);
5351 /* clean up; in particular, free all slabs above current one */
5352 LEAVE_SCOPE(oldsave);
5358 - regrepeat - repeatedly match something simple, report how many
5361 * [This routine now assumes that it will only match on things of length 1.
5362 * That was true before, but now we assume scan - reginput is the count,
5363 * rather than incrementing count on every character. [Er, except utf8.]]
5366 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5369 register char *scan;
5371 register char *loceol = PL_regeol;
5372 register I32 hardcount = 0;
5373 register bool do_utf8 = PL_reg_match_utf8;
5375 PERL_UNUSED_ARG(depth);
5378 PERL_ARGS_ASSERT_REGREPEAT;
5381 if (max == REG_INFTY)
5383 else if (max < loceol - scan)
5384 loceol = scan + max;
5389 while (scan < loceol && hardcount < max && *scan != '\n') {
5390 scan += UTF8SKIP(scan);
5394 while (scan < loceol && *scan != '\n')
5401 while (scan < loceol && hardcount < max) {
5402 scan += UTF8SKIP(scan);
5412 case EXACT: /* length of string is 1 */
5414 while (scan < loceol && UCHARAT(scan) == c)
5417 case EXACTF: /* length of string is 1 */
5419 while (scan < loceol &&
5420 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
5423 case EXACTFL: /* length of string is 1 */
5424 PL_reg_flags |= RF_tainted;
5426 while (scan < loceol &&
5427 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
5433 while (hardcount < max && scan < loceol &&
5434 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
5435 scan += UTF8SKIP(scan);
5439 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5446 LOAD_UTF8_CHARCLASS_ALNUM();
5447 while (hardcount < max && scan < loceol &&
5448 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5449 scan += UTF8SKIP(scan);
5453 while (scan < loceol && isALNUM(*scan))
5458 PL_reg_flags |= RF_tainted;
5461 while (hardcount < max && scan < loceol &&
5462 isALNUM_LC_utf8((U8*)scan)) {
5463 scan += UTF8SKIP(scan);
5467 while (scan < loceol && isALNUM_LC(*scan))
5474 LOAD_UTF8_CHARCLASS_ALNUM();
5475 while (hardcount < max && scan < loceol &&
5476 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
5477 scan += UTF8SKIP(scan);
5481 while (scan < loceol && !isALNUM(*scan))
5486 PL_reg_flags |= RF_tainted;
5489 while (hardcount < max && scan < loceol &&
5490 !isALNUM_LC_utf8((U8*)scan)) {
5491 scan += UTF8SKIP(scan);
5495 while (scan < loceol && !isALNUM_LC(*scan))
5502 LOAD_UTF8_CHARCLASS_SPACE();
5503 while (hardcount < max && scan < loceol &&
5505 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5506 scan += UTF8SKIP(scan);
5510 while (scan < loceol && isSPACE(*scan))
5515 PL_reg_flags |= RF_tainted;
5518 while (hardcount < max && scan < loceol &&
5519 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5520 scan += UTF8SKIP(scan);
5524 while (scan < loceol && isSPACE_LC(*scan))
5531 LOAD_UTF8_CHARCLASS_SPACE();
5532 while (hardcount < max && scan < loceol &&
5534 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
5535 scan += UTF8SKIP(scan);
5539 while (scan < loceol && !isSPACE(*scan))
5544 PL_reg_flags |= RF_tainted;
5547 while (hardcount < max && scan < loceol &&
5548 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
5549 scan += UTF8SKIP(scan);
5553 while (scan < loceol && !isSPACE_LC(*scan))
5560 LOAD_UTF8_CHARCLASS_DIGIT();
5561 while (hardcount < max && scan < loceol &&
5562 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5563 scan += UTF8SKIP(scan);
5567 while (scan < loceol && isDIGIT(*scan))
5574 LOAD_UTF8_CHARCLASS_DIGIT();
5575 while (hardcount < max && scan < loceol &&
5576 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
5577 scan += UTF8SKIP(scan);
5581 while (scan < loceol && !isDIGIT(*scan))
5587 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
5593 LNBREAK can match two latin chars, which is ok,
5594 because we have a null terminated string, but we
5595 have to use hardcount in this situation
5597 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
5606 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
5611 while (scan < loceol && is_HORIZWS_latin1(scan))
5618 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
5619 scan += UTF8SKIP(scan);
5623 while (scan < loceol && !is_HORIZWS_latin1(scan))
5631 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
5636 while (scan < loceol && is_VERTWS_latin1(scan))
5644 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
5645 scan += UTF8SKIP(scan);
5649 while (scan < loceol && !is_VERTWS_latin1(scan))
5655 default: /* Called on something of 0 width. */
5656 break; /* So match right here or not at all. */
5662 c = scan - PL_reginput;
5666 GET_RE_DEBUG_FLAGS_DECL;
5668 SV * const prop = sv_newmortal();
5669 regprop(prog, prop, p);
5670 PerlIO_printf(Perl_debug_log,
5671 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
5672 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
5680 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
5682 - regclass_swash - prepare the utf8 swash
5686 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
5692 RXi_GET_DECL(prog,progi);
5693 const struct reg_data * const data = prog ? progi->data : NULL;
5695 PERL_ARGS_ASSERT_REGCLASS_SWASH;
5697 if (data && data->count) {
5698 const U32 n = ARG(node);
5700 if (data->what[n] == 's') {
5701 SV * const rv = MUTABLE_SV(data->data[n]);
5702 AV * const av = MUTABLE_AV(SvRV(rv));
5703 SV **const ary = AvARRAY(av);
5706 /* See the end of regcomp.c:S_regclass() for
5707 * documentation of these array elements. */
5710 a = SvROK(ary[1]) ? &ary[1] : NULL;
5711 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
5715 else if (si && doinit) {
5716 sw = swash_init("utf8", "", si, 1, 0);
5717 (void)av_store(av, 1, sw);
5734 - reginclass - determine if a character falls into a character class
5736 The n is the ANYOF regnode, the p is the target string, lenp
5737 is pointer to the maximum length of how far to go in the p
5738 (if the lenp is zero, UTF8SKIP(p) is used),
5739 do_utf8 tells whether the target string is in UTF-8.
5744 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
5747 const char flags = ANYOF_FLAGS(n);
5753 PERL_ARGS_ASSERT_REGINCLASS;
5755 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5756 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5757 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5758 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5759 if (len == (STRLEN)-1)
5760 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5763 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5764 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5767 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5768 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5771 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5775 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5783 utf8_p = bytes_to_utf8(p, &len);
5785 if (swash_fetch(sw, utf8_p, 1))
5787 else if (flags & ANYOF_FOLD) {
5788 if (!match && lenp && av) {
5790 for (i = 0; i <= av_len(av); i++) {
5791 SV* const sv = *av_fetch(av, i, FALSE);
5793 const char * const s = SvPV_const(sv, len);
5794 if (len <= plen && memEQ(s, (char*)utf8_p, len)) {
5802 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5805 to_utf8_fold(utf8_p, tmpbuf, &tmplen);
5806 if (swash_fetch(sw, tmpbuf, 1))
5811 /* If we allocated a string above, free it */
5812 if (! do_utf8) Safefree(utf8_p);
5815 if (match && lenp && *lenp == 0)
5816 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5818 if (!match && c < 256) {
5819 if (ANYOF_BITMAP_TEST(n, c))
5821 else if (flags & ANYOF_FOLD) {
5824 if (flags & ANYOF_LOCALE) {
5825 PL_reg_flags |= RF_tainted;
5826 f = PL_fold_locale[c];
5830 if (f != c && ANYOF_BITMAP_TEST(n, f))
5834 if (!match && (flags & ANYOF_CLASS)) {
5835 PL_reg_flags |= RF_tainted;
5837 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5838 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5839 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5840 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5841 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5842 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5843 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5844 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5845 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5846 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5847 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5848 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5849 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5850 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5851 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5852 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5853 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5854 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5855 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5856 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5857 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5858 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5859 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5860 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5861 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5862 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5863 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5864 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5865 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5866 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5867 ) /* How's that for a conditional? */
5874 return (flags & ANYOF_INVERT) ? !match : match;
5878 S_reghop3(U8 *s, I32 off, const U8* lim)
5882 PERL_ARGS_ASSERT_REGHOP3;
5885 while (off-- && s < lim) {
5886 /* XXX could check well-formedness here */
5891 while (off++ && s > lim) {
5893 if (UTF8_IS_CONTINUED(*s)) {
5894 while (s > lim && UTF8_IS_CONTINUATION(*s))
5897 /* XXX could check well-formedness here */
5904 /* there are a bunch of places where we use two reghop3's that should
5905 be replaced with this routine. but since thats not done yet
5906 we ifdef it out - dmq
5909 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
5913 PERL_ARGS_ASSERT_REGHOP4;
5916 while (off-- && s < rlim) {
5917 /* XXX could check well-formedness here */
5922 while (off++ && s > llim) {
5924 if (UTF8_IS_CONTINUED(*s)) {
5925 while (s > llim && UTF8_IS_CONTINUATION(*s))
5928 /* XXX could check well-formedness here */
5936 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
5940 PERL_ARGS_ASSERT_REGHOPMAYBE3;
5943 while (off-- && s < lim) {
5944 /* XXX could check well-formedness here */
5951 while (off++ && s > lim) {
5953 if (UTF8_IS_CONTINUED(*s)) {
5954 while (s > lim && UTF8_IS_CONTINUATION(*s))
5957 /* XXX could check well-formedness here */
5966 restore_pos(pTHX_ void *arg)
5969 regexp * const rex = (regexp *)arg;
5970 if (PL_reg_eval_set) {
5971 if (PL_reg_oldsaved) {
5972 rex->subbeg = PL_reg_oldsaved;
5973 rex->sublen = PL_reg_oldsavedlen;
5974 #ifdef PERL_OLD_COPY_ON_WRITE
5975 rex->saved_copy = PL_nrs;
5977 RXp_MATCH_COPIED_on(rex);
5979 PL_reg_magic->mg_len = PL_reg_oldpos;
5980 PL_reg_eval_set = 0;
5981 PL_curpm = PL_reg_oldcurpm;
5986 S_to_utf8_substr(pTHX_ register regexp *prog)
5990 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
5993 if (prog->substrs->data[i].substr
5994 && !prog->substrs->data[i].utf8_substr) {
5995 SV* const sv = newSVsv(prog->substrs->data[i].substr);
5996 prog->substrs->data[i].utf8_substr = sv;
5997 sv_utf8_upgrade(sv);
5998 if (SvVALID(prog->substrs->data[i].substr)) {
5999 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6000 if (flags & FBMcf_TAIL) {
6001 /* Trim the trailing \n that fbm_compile added last
6003 SvCUR_set(sv, SvCUR(sv) - 1);
6004 /* Whilst this makes the SV technically "invalid" (as its
6005 buffer is no longer followed by "\0") when fbm_compile()
6006 adds the "\n" back, a "\0" is restored. */
6008 fbm_compile(sv, flags);
6010 if (prog->substrs->data[i].substr == prog->check_substr)
6011 prog->check_utf8 = sv;
6017 S_to_byte_substr(pTHX_ register regexp *prog)
6022 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6025 if (prog->substrs->data[i].utf8_substr
6026 && !prog->substrs->data[i].substr) {
6027 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6028 if (sv_utf8_downgrade(sv, TRUE)) {
6029 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6031 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6032 if (flags & FBMcf_TAIL) {
6033 /* Trim the trailing \n that fbm_compile added last
6035 SvCUR_set(sv, SvCUR(sv) - 1);
6037 fbm_compile(sv, flags);
6043 prog->substrs->data[i].substr = sv;
6044 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6045 prog->check_substr = sv;
6052 * c-indentation-style: bsd
6054 * indent-tabs-mode: t
6057 * ex: set ts=8 sts=4 sw=4 noet: