5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
19 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
20 * confused with the original package (see point 3 below). Thanks, Henry!
23 /* Additional note: this code is very heavily munged from Henry's version
24 * in places. In some spots I've traded clarity for efficiency, so don't
25 * blame Henry for some of the lack of readability.
28 /* The names of the functions have been changed from regcomp and
29 * regexec to pregcomp and pregexec in order to avoid conflicts
30 * with the POSIX routines of the same names.
33 #ifdef PERL_EXT_RE_BUILD
34 /* need to replace pregcomp et al, so enable that */
35 # ifndef PERL_IN_XSUB_RE
36 # define PERL_IN_XSUB_RE
38 /* need access to debugger hooks */
39 # if defined(PERL_EXT_RE_DEBUG) && !defined(DEBUGGING)
44 #ifdef PERL_IN_XSUB_RE
45 /* We *really* need to overwrite these symbols: */
46 # define Perl_regexec_flags my_regexec
47 # define Perl_regdump my_regdump
48 # define Perl_regprop my_regprop
49 # define Perl_re_intuit_start my_re_intuit_start
50 /* *These* symbols are masked to allow static link. */
51 # define Perl_pregexec my_pregexec
52 # define Perl_reginitcolors my_reginitcolors
53 # define Perl_regclass_swash my_regclass_swash
55 # define PERL_NO_GET_CONTEXT
59 * pregcomp and pregexec -- regsub and regerror are not used in perl
61 * Copyright (c) 1986 by University of Toronto.
62 * Written by Henry Spencer. Not derived from licensed software.
64 * Permission is granted to anyone to use this software for any
65 * purpose on any computer system, and to redistribute it freely,
66 * subject to the following restrictions:
68 * 1. The author is not responsible for the consequences of use of
69 * this software, no matter how awful, even if they arise
72 * 2. The origin of this software must not be misrepresented, either
73 * by explicit claim or by omission.
75 * 3. Altered versions must be plainly marked as such, and must not
76 * be misrepresented as being the original software.
78 **** Alterations to Henry's code are...
80 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
81 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
83 **** You may distribute under the terms of either the GNU General Public
84 **** License or the Artistic License, as specified in the README file.
86 * Beware that some of this code is subtly aware of the way operator
87 * precedence is structured in regular expressions. Serious changes in
88 * regular-expression syntax might require a total rethink.
91 #define PERL_IN_REGEXEC_C
96 #define RF_tainted 1 /* tainted information used? */
97 #define RF_warned 2 /* warned about big count? */
98 #define RF_evaled 4 /* Did an EVAL with setting? */
99 #define RF_utf8 8 /* String contains multibyte chars? */
101 #define UTF ((PL_reg_flags & RF_utf8) != 0)
103 #define RS_init 1 /* eval environment created */
104 #define RS_set 2 /* replsv value is set */
107 #define STATIC static
110 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
116 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
117 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
119 #define HOPc(pos,off) ((char *)(PL_reg_match_utf8 \
120 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
122 #define HOPBACKc(pos, off) ((char*) \
123 ((PL_reg_match_utf8) \
124 ? reghopmaybe3((U8*)pos, -off, ((U8*)(off < 0 ? PL_regeol : PL_bostr))) \
125 : (pos - off >= PL_bostr) \
130 #define reghopmaybe3_c(pos,off,lim) ((char*)reghopmaybe3((U8*)pos, off, (U8*)lim))
131 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off))
132 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
134 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
135 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
136 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
137 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
138 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
139 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
141 /* for use after a quantifier and before an EXACT-like node -- japhy */
142 #define JUMPABLE(rn) ( \
143 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
144 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
145 OP(rn) == PLUS || OP(rn) == MINMOD || \
146 (PL_regkind[(U8)OP(rn)] == CURLY && ARG1(rn) > 0) \
149 #define HAS_TEXT(rn) ( \
150 PL_regkind[(U8)OP(rn)] == EXACT || PL_regkind[(U8)OP(rn)] == REF \
154 Search for mandatory following text node; for lookahead, the text must
155 follow but for lookbehind (rn->flags != 0) we skip to the next step.
157 #define FIND_NEXT_IMPT(rn) STMT_START { \
158 while (JUMPABLE(rn)) \
159 if (OP(rn) == SUSPEND || PL_regkind[(U8)OP(rn)] == CURLY) \
160 rn = NEXTOPER(NEXTOPER(rn)); \
161 else if (OP(rn) == PLUS) \
163 else if (OP(rn) == IFMATCH) \
164 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
165 else rn += NEXT_OFF(rn); \
168 static void restore_pos(pTHX_ void *arg);
171 S_regcppush(pTHX_ I32 parenfloor)
174 const int retval = PL_savestack_ix;
175 #define REGCP_PAREN_ELEMS 4
176 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
179 if (paren_elems_to_push < 0)
180 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
182 #define REGCP_OTHER_ELEMS 6
183 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
184 for (p = PL_regsize; p > parenfloor; p--) {
185 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
186 SSPUSHINT(PL_regendp[p]);
187 SSPUSHINT(PL_regstartp[p]);
188 SSPUSHPTR(PL_reg_start_tmp[p]);
191 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
192 SSPUSHINT(PL_regsize);
193 SSPUSHINT(*PL_reglastparen);
194 SSPUSHINT(*PL_reglastcloseparen);
195 SSPUSHPTR(PL_reginput);
196 #define REGCP_FRAME_ELEMS 2
197 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
198 * are needed for the regexp context stack bookkeeping. */
199 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
200 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
205 /* These are needed since we do not localize EVAL nodes: */
206 # define REGCP_SET(cp) DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, \
207 " Setting an EVAL scope, savestack=%"IVdf"\n", \
208 (IV)PL_savestack_ix)); cp = PL_savestack_ix
210 # define REGCP_UNWIND(cp) DEBUG_EXECUTE_r(cp != PL_savestack_ix ? \
211 PerlIO_printf(Perl_debug_log, \
212 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
213 (IV)(cp), (IV)PL_savestack_ix) : 0); regcpblow(cp)
216 S_regcppop(pTHX_ const regexp *rex)
222 GET_RE_DEBUG_FLAGS_DECL;
224 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
226 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
227 i = SSPOPINT; /* Parentheses elements to pop. */
228 input = (char *) SSPOPPTR;
229 *PL_reglastcloseparen = SSPOPINT;
230 *PL_reglastparen = SSPOPINT;
231 PL_regsize = SSPOPINT;
233 /* Now restore the parentheses context. */
234 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
235 i > 0; i -= REGCP_PAREN_ELEMS) {
237 U32 paren = (U32)SSPOPINT;
238 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
239 PL_regstartp[paren] = SSPOPINT;
241 if (paren <= *PL_reglastparen)
242 PL_regendp[paren] = tmps;
244 PerlIO_printf(Perl_debug_log,
245 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
246 (UV)paren, (IV)PL_regstartp[paren],
247 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
248 (IV)PL_regendp[paren],
249 (paren > *PL_reglastparen ? "(no)" : ""));
253 if ((I32)(*PL_reglastparen + 1) <= rex->nparens) {
254 PerlIO_printf(Perl_debug_log,
255 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
256 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
260 /* It would seem that the similar code in regtry()
261 * already takes care of this, and in fact it is in
262 * a better location to since this code can #if 0-ed out
263 * but the code in regtry() is needed or otherwise tests
264 * requiring null fields (pat.t#187 and split.t#{13,14}
265 * (as of patchlevel 7877) will fail. Then again,
266 * this code seems to be necessary or otherwise
267 * building DynaLoader will fail:
268 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
270 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
272 PL_regstartp[i] = -1;
279 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
281 #define TRYPAREN(paren, n, input, where) { \
284 PL_regstartp[paren] = HOPc(input, -1) - PL_bostr; \
285 PL_regendp[paren] = input - PL_bostr; \
288 PL_regendp[paren] = -1; \
290 REGMATCH(next, where); \
294 PL_regendp[paren] = -1; \
299 * pregexec and friends
303 - pregexec - match a regexp against a string
306 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
307 char *strbeg, I32 minend, SV *screamer, U32 nosave)
308 /* strend: pointer to null at end of string */
309 /* strbeg: real beginning of string */
310 /* minend: end of match must be >=minend after stringarg. */
311 /* nosave: For optimizations. */
314 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
315 nosave ? 0 : REXEC_COPY_STR);
320 * Need to implement the following flags for reg_anch:
322 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
324 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
325 * INTUIT_AUTORITATIVE_ML
326 * INTUIT_ONCE_NOML - Intuit can match in one location only.
329 * Another flag for this function: SECOND_TIME (so that float substrs
330 * with giant delta may be not rechecked).
333 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
335 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
336 Otherwise, only SvCUR(sv) is used to get strbeg. */
338 /* XXXX We assume that strpos is strbeg unless sv. */
340 /* XXXX Some places assume that there is a fixed substring.
341 An update may be needed if optimizer marks as "INTUITable"
342 RExen without fixed substrings. Similarly, it is assumed that
343 lengths of all the strings are no more than minlen, thus they
344 cannot come from lookahead.
345 (Or minlen should take into account lookahead.) */
347 /* A failure to find a constant substring means that there is no need to make
348 an expensive call to REx engine, thus we celebrate a failure. Similarly,
349 finding a substring too deep into the string means that less calls to
350 regtry() should be needed.
352 REx compiler's optimizer found 4 possible hints:
353 a) Anchored substring;
355 c) Whether we are anchored (beginning-of-line or \G);
356 d) First node (of those at offset 0) which may distingush positions;
357 We use a)b)d) and multiline-part of c), and try to find a position in the
358 string which does not contradict any of them.
361 /* Most of decisions we do here should have been done at compile time.
362 The nodes of the REx which we used for the search should have been
363 deleted from the finite automaton. */
366 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
367 char *strend, U32 flags, re_scream_pos_data *data)
370 register I32 start_shift = 0;
371 /* Should be nonnegative! */
372 register I32 end_shift = 0;
377 const int do_utf8 = sv ? SvUTF8(sv) : 0; /* if no sv we have to assume bytes */
379 register char *other_last = NULL; /* other substr checked before this */
380 char *check_at = NULL; /* check substr found at this pos */
381 const I32 multiline = prog->reganch & PMf_MULTILINE;
383 const char * const i_strpos = strpos;
384 SV * const dsv = PERL_DEBUG_PAD_ZERO(0);
387 GET_RE_DEBUG_FLAGS_DECL;
389 RX_MATCH_UTF8_set(prog,do_utf8);
391 if (prog->reganch & ROPT_UTF8) {
392 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
393 "UTF-8 regex...\n"));
394 PL_reg_flags |= RF_utf8;
398 const char *s = PL_reg_match_utf8 ?
399 sv_uni_display(dsv, sv, 60, UNI_DISPLAY_REGEX) :
401 const int len = PL_reg_match_utf8 ?
402 strlen(s) : strend - strpos;
405 if (PL_reg_match_utf8)
406 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
407 "UTF-8 target...\n"));
408 PerlIO_printf(Perl_debug_log,
409 "%sGuessing start of match, REx%s \"%s%.60s%s%s\" against \"%s%.*s%s%s\"...\n",
410 PL_colors[4], PL_colors[5], PL_colors[0],
413 (strlen(prog->precomp) > 60 ? "..." : ""),
415 (int)(len > 60 ? 60 : len),
417 (len > 60 ? "..." : "")
421 /* CHR_DIST() would be more correct here but it makes things slow. */
422 if (prog->minlen > strend - strpos) {
423 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
424 "String too short... [re_intuit_start]\n"));
427 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
430 if (!prog->check_utf8 && prog->check_substr)
431 to_utf8_substr(prog);
432 check = prog->check_utf8;
434 if (!prog->check_substr && prog->check_utf8)
435 to_byte_substr(prog);
436 check = prog->check_substr;
438 if (check == &PL_sv_undef) {
439 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
440 "Non-utf string cannot match utf check string\n"));
443 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
444 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
445 || ( (prog->reganch & ROPT_ANCH_BOL)
446 && !multiline ) ); /* Check after \n? */
449 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
450 | ROPT_IMPLICIT)) /* not a real BOL */
451 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
453 && (strpos != strbeg)) {
454 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
457 if (prog->check_offset_min == prog->check_offset_max &&
458 !(prog->reganch & ROPT_CANY_SEEN)) {
459 /* Substring at constant offset from beg-of-str... */
462 s = HOP3c(strpos, prog->check_offset_min, strend);
464 slen = SvCUR(check); /* >= 1 */
466 if ( strend - s > slen || strend - s < slen - 1
467 || (strend - s == slen && strend[-1] != '\n')) {
468 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
471 /* Now should match s[0..slen-2] */
473 if (slen && (*SvPVX_const(check) != *s
475 && memNE(SvPVX_const(check), s, slen)))) {
477 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
481 else if (*SvPVX_const(check) != *s
482 || ((slen = SvCUR(check)) > 1
483 && memNE(SvPVX_const(check), s, slen)))
486 goto success_at_start;
489 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
491 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
492 end_shift = prog->minlen - start_shift -
493 CHR_SVLEN(check) + (SvTAIL(check) != 0);
495 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
496 - (SvTAIL(check) != 0);
497 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
499 if (end_shift < eshift)
503 else { /* Can match at random position */
506 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
507 /* Should be nonnegative! */
508 end_shift = prog->minlen - start_shift -
509 CHR_SVLEN(check) + (SvTAIL(check) != 0);
512 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
514 Perl_croak(aTHX_ "panic: end_shift");
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. */
520 if (flags & REXEC_SCREAM) {
521 I32 p = -1; /* Internal iterator of scream. */
522 I32 * const pp = data ? data->scream_pos : &p;
524 if (PL_screamfirst[BmRARE(check)] >= 0
525 || ( BmRARE(check) == '\n'
526 && (BmPREVIOUS(check) == SvCUR(check) - 1)
528 s = screaminstr(sv, check,
529 start_shift + (s - strbeg), end_shift, pp, 0);
532 /* we may be pointing at the wrong string */
533 if (s && RX_MATCH_COPIED(prog))
534 s = strbeg + (s - SvPVX_const(sv));
536 *data->scream_olds = s;
538 else if (prog->reganch & ROPT_CANY_SEEN)
539 s = fbm_instr((U8*)(s + start_shift),
540 (U8*)(strend - end_shift),
541 check, multiline ? FBMrf_MULTILINE : 0);
543 s = fbm_instr(HOP3(s, start_shift, strend),
544 HOP3(strend, -end_shift, strbeg),
545 check, multiline ? FBMrf_MULTILINE : 0);
547 /* Update the count-of-usability, remove useless subpatterns,
550 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s %s substr \"%s%.*s%s\"%s%s",
551 (s ? "Found" : "Did not find"),
552 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) ? "anchored" : "floating"),
554 (int)(SvCUR(check) - (SvTAIL(check)!=0)),
556 PL_colors[1], (SvTAIL(check) ? "$" : ""),
557 (s ? " at offset " : "...\n") ) );
564 /* Finish the diagnostic message */
565 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
567 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
568 Start with the other substr.
569 XXXX no SCREAM optimization yet - and a very coarse implementation
570 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
571 *always* match. Probably should be marked during compile...
572 Probably it is right to do no SCREAM here...
575 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) : (prog->float_substr && prog->anchored_substr)) {
576 /* Take into account the "other" substring. */
577 /* XXXX May be hopelessly wrong for UTF... */
580 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
583 char * const last = HOP3c(s, -start_shift, strbeg);
588 t = s - prog->check_offset_max;
589 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
591 || ((t = reghopmaybe3_c(s, -(prog->check_offset_max), strpos))
596 t = HOP3c(t, prog->anchored_offset, strend);
597 if (t < other_last) /* These positions already checked */
599 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
602 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
603 /* On end-of-str: see comment below. */
604 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
605 if (must == &PL_sv_undef) {
607 DEBUG_EXECUTE_r(must = prog->anchored_utf8); /* for debug */
612 HOP3(HOP3(last1, prog->anchored_offset, strend)
613 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
615 multiline ? FBMrf_MULTILINE : 0
617 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
618 "%s anchored substr \"%s%.*s%s\"%s",
619 (s ? "Found" : "Contradicts"),
622 - (SvTAIL(must)!=0)),
624 PL_colors[1], (SvTAIL(must) ? "$" : "")));
626 if (last1 >= last2) {
627 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
628 ", giving up...\n"));
631 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
632 ", trying floating at offset %ld...\n",
633 (long)(HOP3c(s1, 1, strend) - i_strpos)));
634 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
635 s = HOP3c(last, 1, strend);
639 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
640 (long)(s - i_strpos)));
641 t = HOP3c(s, -prog->anchored_offset, strbeg);
642 other_last = HOP3c(s, 1, strend);
650 else { /* Take into account the floating substring. */
655 t = HOP3c(s, -start_shift, strbeg);
657 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
658 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
659 last = HOP3c(t, prog->float_max_offset, strend);
660 s = HOP3c(t, prog->float_min_offset, strend);
663 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
664 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
665 /* fbm_instr() takes into account exact value of end-of-str
666 if the check is SvTAIL(ed). Since false positives are OK,
667 and end-of-str is not later than strend we are OK. */
668 if (must == &PL_sv_undef) {
670 DEBUG_EXECUTE_r(must = prog->float_utf8); /* for debug message */
673 s = fbm_instr((unsigned char*)s,
674 (unsigned char*)last + SvCUR(must)
676 must, multiline ? FBMrf_MULTILINE : 0);
677 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s floating substr \"%s%.*s%s\"%s",
678 (s ? "Found" : "Contradicts"),
680 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
682 PL_colors[1], (SvTAIL(must) ? "$" : "")));
685 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
686 ", giving up...\n"));
689 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
690 ", trying anchored starting at offset %ld...\n",
691 (long)(s1 + 1 - i_strpos)));
693 s = HOP3c(t, 1, strend);
697 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
698 (long)(s - i_strpos)));
699 other_last = s; /* Fix this later. --Hugo */
708 t = s - prog->check_offset_max;
709 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
711 || ((t = reghopmaybe3_c(s, -prog->check_offset_max, strpos))
713 /* Fixed substring is found far enough so that the match
714 cannot start at strpos. */
716 if (ml_anch && t[-1] != '\n') {
717 /* Eventually fbm_*() should handle this, but often
718 anchored_offset is not 0, so this check will not be wasted. */
719 /* XXXX In the code below we prefer to look for "^" even in
720 presence of anchored substrings. And we search even
721 beyond the found float position. These pessimizations
722 are historical artefacts only. */
724 while (t < strend - prog->minlen) {
726 if (t < check_at - prog->check_offset_min) {
727 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
728 /* Since we moved from the found position,
729 we definitely contradict the found anchored
730 substr. Due to the above check we do not
731 contradict "check" substr.
732 Thus we can arrive here only if check substr
733 is float. Redo checking for "other"=="fixed".
736 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
737 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
738 goto do_other_anchored;
740 /* We don't contradict the found floating substring. */
741 /* XXXX Why not check for STCLASS? */
743 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
744 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
747 /* Position contradicts check-string */
748 /* XXXX probably better to look for check-string
749 than for "\n", so one should lower the limit for t? */
750 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
751 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
752 other_last = strpos = s = t + 1;
757 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
758 PL_colors[0], PL_colors[1]));
762 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
763 PL_colors[0], PL_colors[1]));
767 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
770 /* The found string does not prohibit matching at strpos,
771 - no optimization of calling REx engine can be performed,
772 unless it was an MBOL and we are not after MBOL,
773 or a future STCLASS check will fail this. */
775 /* Even in this situation we may use MBOL flag if strpos is offset
776 wrt the start of the string. */
777 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
778 && (strpos != strbeg) && strpos[-1] != '\n'
779 /* May be due to an implicit anchor of m{.*foo} */
780 && !(prog->reganch & ROPT_IMPLICIT))
785 DEBUG_EXECUTE_r( if (ml_anch)
786 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
787 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
790 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
792 prog->check_utf8 /* Could be deleted already */
793 && --BmUSEFUL(prog->check_utf8) < 0
794 && (prog->check_utf8 == prog->float_utf8)
796 prog->check_substr /* Could be deleted already */
797 && --BmUSEFUL(prog->check_substr) < 0
798 && (prog->check_substr == prog->float_substr)
801 /* If flags & SOMETHING - do not do it many times on the same match */
802 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
803 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
804 if (do_utf8 ? prog->check_substr : prog->check_utf8)
805 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
806 prog->check_substr = prog->check_utf8 = NULL; /* disable */
807 prog->float_substr = prog->float_utf8 = NULL; /* clear */
808 check = NULL; /* abort */
810 /* XXXX This is a remnant of the old implementation. It
811 looks wasteful, since now INTUIT can use many
813 prog->reganch &= ~RE_USE_INTUIT;
820 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
821 if (prog->regstclass) {
822 /* minlen == 0 is possible if regstclass is \b or \B,
823 and the fixed substr is ''$.
824 Since minlen is already taken into account, s+1 is before strend;
825 accidentally, minlen >= 1 guaranties no false positives at s + 1
826 even for \b or \B. But (minlen? 1 : 0) below assumes that
827 regstclass does not come from lookahead... */
828 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
829 This leaves EXACTF only, which is dealt with in find_byclass(). */
830 const U8* const str = (U8*)STRING(prog->regstclass);
831 const int cl_l = (PL_regkind[(U8)OP(prog->regstclass)] == EXACT
832 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
834 const char * const endpos = (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
835 ? HOP3c(s, (prog->minlen ? cl_l : 0), strend)
836 : (prog->float_substr || prog->float_utf8
837 ? HOP3c(HOP3c(check_at, -start_shift, strbeg),
842 s = find_byclass(prog, prog->regstclass, s, endpos, 1);
845 const char *what = NULL;
847 if (endpos == strend) {
848 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
849 "Could not match STCLASS...\n") );
852 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
853 "This position contradicts STCLASS...\n") );
854 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
856 /* Contradict one of substrings */
857 if (prog->anchored_substr || prog->anchored_utf8) {
858 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
859 DEBUG_EXECUTE_r( what = "anchored" );
861 s = HOP3c(t, 1, strend);
862 if (s + start_shift + end_shift > strend) {
863 /* XXXX Should be taken into account earlier? */
864 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
865 "Could not match STCLASS...\n") );
870 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
871 "Looking for %s substr starting at offset %ld...\n",
872 what, (long)(s + start_shift - i_strpos)) );
875 /* Have both, check_string is floating */
876 if (t + start_shift >= check_at) /* Contradicts floating=check */
877 goto retry_floating_check;
878 /* Recheck anchored substring, but not floating... */
882 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
883 "Looking for anchored substr starting at offset %ld...\n",
884 (long)(other_last - i_strpos)) );
885 goto do_other_anchored;
887 /* Another way we could have checked stclass at the
888 current position only: */
893 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
894 "Looking for /%s^%s/m starting at offset %ld...\n",
895 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
898 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
900 /* Check is floating subtring. */
901 retry_floating_check:
902 t = check_at - start_shift;
903 DEBUG_EXECUTE_r( what = "floating" );
904 goto hop_and_restart;
907 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
908 "By STCLASS: moving %ld --> %ld\n",
909 (long)(t - i_strpos), (long)(s - i_strpos))
913 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
914 "Does not contradict STCLASS...\n");
919 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
920 PL_colors[4], (check ? "Guessed" : "Giving up"),
921 PL_colors[5], (long)(s - i_strpos)) );
924 fail_finish: /* Substring not found */
925 if (prog->check_substr || prog->check_utf8) /* could be removed already */
926 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
928 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
929 PL_colors[4], PL_colors[5]));
933 /* We know what class REx starts with. Try to find this position... */
935 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, const char *strend, I32 norun)
938 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
942 register STRLEN uskip;
946 register I32 tmp = 1; /* Scratch variable? */
947 register const bool do_utf8 = PL_reg_match_utf8;
949 /* We know what class it must start with. */
953 while (s + (uskip = UTF8SKIP(s)) <= strend) {
954 if ((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
955 !UTF8_IS_INVARIANT((U8)s[0]) ?
956 reginclass(prog, c, (U8*)s, 0, do_utf8) :
957 REGINCLASS(prog, c, (U8*)s)) {
958 if (tmp && (norun || regtry(prog, s)))
972 if (REGINCLASS(prog, c, (U8*)s) ||
973 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
974 /* The assignment of 2 is intentional:
975 * for the folded sharp s, the skip is 2. */
976 (skip = SHARP_S_SKIP))) {
977 if (tmp && (norun || regtry(prog, s)))
990 if (tmp && (norun || regtry(prog, s)))
999 ln = STR_LEN(c); /* length to match in octets/bytes */
1000 lnc = (I32) ln; /* length to match in characters */
1002 STRLEN ulen1, ulen2;
1004 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1005 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1006 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1008 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1009 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1011 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1013 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1016 while (sm < ((U8 *) m + ln)) {
1031 c2 = PL_fold_locale[c1];
1033 e = HOP3c(strend, -((I32)lnc), s);
1036 e = s; /* Due to minlen logic of intuit() */
1038 /* The idea in the EXACTF* cases is to first find the
1039 * first character of the EXACTF* node and then, if
1040 * necessary, case-insensitively compare the full
1041 * text of the node. The c1 and c2 are the first
1042 * characters (though in Unicode it gets a bit
1043 * more complicated because there are more cases
1044 * than just upper and lower: one needs to use
1045 * the so-called folding case for case-insensitive
1046 * matching (called "loose matching" in Unicode).
1047 * ibcmp_utf8() will do just that. */
1051 U8 tmpbuf [UTF8_MAXBYTES+1];
1052 STRLEN len, foldlen;
1053 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1055 /* Upper and lower of 1st char are equal -
1056 * probably not a "letter". */
1058 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1062 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1063 m, (char **)0, ln, (bool)UTF))
1064 && (norun || regtry(prog, s)) )
1067 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1068 uvchr_to_utf8(tmpbuf, c);
1069 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1071 && (f == c1 || f == c2)
1072 && (ln == foldlen ||
1073 !ibcmp_utf8((char *) foldbuf,
1074 (char **)0, foldlen, do_utf8,
1076 (char **)0, ln, (bool)UTF))
1077 && (norun || regtry(prog, s)) )
1085 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1088 /* Handle some of the three Greek sigmas cases.
1089 * Note that not all the possible combinations
1090 * are handled here: some of them are handled
1091 * by the standard folding rules, and some of
1092 * them (the character class or ANYOF cases)
1093 * are handled during compiletime in
1094 * regexec.c:S_regclass(). */
1095 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1096 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1097 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1099 if ( (c == c1 || c == c2)
1101 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1102 m, (char **)0, ln, (bool)UTF))
1103 && (norun || regtry(prog, s)) )
1106 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1107 uvchr_to_utf8(tmpbuf, c);
1108 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1110 && (f == c1 || f == c2)
1111 && (ln == foldlen ||
1112 !ibcmp_utf8((char *) foldbuf,
1113 (char **)0, foldlen, do_utf8,
1115 (char **)0, ln, (bool)UTF))
1116 && (norun || regtry(prog, s)) )
1127 && (ln == 1 || !(OP(c) == EXACTF
1129 : ibcmp_locale(s, m, ln)))
1130 && (norun || regtry(prog, s)) )
1136 if ( (*(U8*)s == c1 || *(U8*)s == c2)
1137 && (ln == 1 || !(OP(c) == EXACTF
1139 : ibcmp_locale(s, m, ln)))
1140 && (norun || regtry(prog, s)) )
1147 PL_reg_flags |= RF_tainted;
1154 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1155 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1157 tmp = ((OP(c) == BOUND ?
1158 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1159 LOAD_UTF8_CHARCLASS_ALNUM();
1160 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1161 if (tmp == !(OP(c) == BOUND ?
1162 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1163 isALNUM_LC_utf8((U8*)s)))
1166 if ((norun || regtry(prog, s)))
1173 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1174 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1175 while (s < strend) {
1177 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1179 if ((norun || regtry(prog, s)))
1185 if ((!prog->minlen && tmp) && (norun || regtry(prog, s)))
1189 PL_reg_flags |= RF_tainted;
1196 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1197 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1199 tmp = ((OP(c) == NBOUND ?
1200 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1201 LOAD_UTF8_CHARCLASS_ALNUM();
1202 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1203 if (tmp == !(OP(c) == NBOUND ?
1204 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1205 isALNUM_LC_utf8((U8*)s)))
1207 else if ((norun || regtry(prog, s)))
1213 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1214 tmp = ((OP(c) == NBOUND ?
1215 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1216 while (s < strend) {
1218 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1220 else if ((norun || regtry(prog, s)))
1225 if ((!prog->minlen && !tmp) && (norun || regtry(prog, s)))
1230 LOAD_UTF8_CHARCLASS_ALNUM();
1231 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1232 if (swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1233 if (tmp && (norun || regtry(prog, s)))
1244 while (s < strend) {
1246 if (tmp && (norun || regtry(prog, s)))
1258 PL_reg_flags |= RF_tainted;
1260 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1261 if (isALNUM_LC_utf8((U8*)s)) {
1262 if (tmp && (norun || regtry(prog, s)))
1273 while (s < strend) {
1274 if (isALNUM_LC(*s)) {
1275 if (tmp && (norun || regtry(prog, s)))
1288 LOAD_UTF8_CHARCLASS_ALNUM();
1289 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1290 if (!swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1291 if (tmp && (norun || regtry(prog, s)))
1302 while (s < strend) {
1304 if (tmp && (norun || regtry(prog, s)))
1316 PL_reg_flags |= RF_tainted;
1318 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1319 if (!isALNUM_LC_utf8((U8*)s)) {
1320 if (tmp && (norun || regtry(prog, s)))
1331 while (s < strend) {
1332 if (!isALNUM_LC(*s)) {
1333 if (tmp && (norun || regtry(prog, s)))
1346 LOAD_UTF8_CHARCLASS_SPACE();
1347 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1348 if (*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)) {
1349 if (tmp && (norun || regtry(prog, s)))
1360 while (s < strend) {
1362 if (tmp && (norun || regtry(prog, s)))
1374 PL_reg_flags |= RF_tainted;
1376 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1377 if (*s == ' ' || isSPACE_LC_utf8((U8*)s)) {
1378 if (tmp && (norun || regtry(prog, s)))
1389 while (s < strend) {
1390 if (isSPACE_LC(*s)) {
1391 if (tmp && (norun || regtry(prog, s)))
1404 LOAD_UTF8_CHARCLASS_SPACE();
1405 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1406 if (!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8))) {
1407 if (tmp && (norun || regtry(prog, s)))
1418 while (s < strend) {
1420 if (tmp && (norun || regtry(prog, s)))
1432 PL_reg_flags |= RF_tainted;
1434 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1435 if (!(*s == ' ' || isSPACE_LC_utf8((U8*)s))) {
1436 if (tmp && (norun || regtry(prog, s)))
1447 while (s < strend) {
1448 if (!isSPACE_LC(*s)) {
1449 if (tmp && (norun || regtry(prog, s)))
1462 LOAD_UTF8_CHARCLASS_DIGIT();
1463 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1464 if (swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1465 if (tmp && (norun || regtry(prog, s)))
1476 while (s < strend) {
1478 if (tmp && (norun || regtry(prog, s)))
1490 PL_reg_flags |= RF_tainted;
1492 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1493 if (isDIGIT_LC_utf8((U8*)s)) {
1494 if (tmp && (norun || regtry(prog, s)))
1505 while (s < strend) {
1506 if (isDIGIT_LC(*s)) {
1507 if (tmp && (norun || regtry(prog, s)))
1520 LOAD_UTF8_CHARCLASS_DIGIT();
1521 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1522 if (!swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1523 if (tmp && (norun || regtry(prog, s)))
1534 while (s < strend) {
1536 if (tmp && (norun || regtry(prog, s)))
1548 PL_reg_flags |= RF_tainted;
1550 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1551 if (!isDIGIT_LC_utf8((U8*)s)) {
1552 if (tmp && (norun || regtry(prog, s)))
1563 while (s < strend) {
1564 if (!isDIGIT_LC(*s)) {
1565 if (tmp && (norun || regtry(prog, s)))
1577 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1586 - regexec_flags - match a regexp against a string
1589 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1590 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1591 /* strend: pointer to null at end of string */
1592 /* strbeg: real beginning of string */
1593 /* minend: end of match must be >=minend after stringarg. */
1594 /* data: May be used for some additional optimizations. */
1595 /* nosave: For optimizations. */
1599 register regnode *c;
1600 register char *startpos = stringarg;
1601 I32 minlen; /* must match at least this many chars */
1602 I32 dontbother = 0; /* how many characters not to try at end */
1603 I32 end_shift = 0; /* Same for the end. */ /* CC */
1604 I32 scream_pos = -1; /* Internal iterator of scream. */
1605 char *scream_olds = NULL;
1606 SV* oreplsv = GvSV(PL_replgv);
1607 const bool do_utf8 = DO_UTF8(sv);
1614 GET_RE_DEBUG_FLAGS_DECL;
1616 PERL_UNUSED_ARG(data);
1618 /* Be paranoid... */
1619 if (prog == NULL || startpos == NULL) {
1620 Perl_croak(aTHX_ "NULL regexp parameter");
1624 multiline = prog->reganch & PMf_MULTILINE;
1627 dsv0 = PERL_DEBUG_PAD_ZERO(0);
1628 dsv1 = PERL_DEBUG_PAD_ZERO(1);
1632 PL_regnarrate = DEBUG_r_TEST;
1635 RX_MATCH_UTF8_set(prog, do_utf8);
1637 minlen = prog->minlen;
1638 if (strend - startpos < minlen) {
1639 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1640 "String too short [regexec_flags]...\n"));
1644 /* Check validity of program. */
1645 if (UCHARAT(prog->program) != REG_MAGIC) {
1646 Perl_croak(aTHX_ "corrupted regexp program");
1650 PL_reg_eval_set = 0;
1653 if (prog->reganch & ROPT_UTF8)
1654 PL_reg_flags |= RF_utf8;
1656 /* Mark beginning of line for ^ and lookbehind. */
1657 PL_regbol = startpos;
1661 /* Mark end of line for $ (and such) */
1664 /* see how far we have to get to not match where we matched before */
1665 PL_regtill = startpos+minend;
1667 /* If there is a "must appear" string, look for it. */
1670 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to have PL_reg_ganch */
1673 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1674 PL_reg_ganch = startpos;
1675 else if (sv && SvTYPE(sv) >= SVt_PVMG
1677 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1678 && mg->mg_len >= 0) {
1679 PL_reg_ganch = strbeg + mg->mg_len; /* Defined pos() */
1680 if (prog->reganch & ROPT_ANCH_GPOS) {
1681 if (s > PL_reg_ganch)
1686 else /* pos() not defined */
1687 PL_reg_ganch = strbeg;
1690 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1691 re_scream_pos_data d;
1693 d.scream_olds = &scream_olds;
1694 d.scream_pos = &scream_pos;
1695 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1697 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1698 goto phooey; /* not present */
1703 const char * const s0 = UTF
1704 ? pv_uni_display(dsv0, (U8*)prog->precomp, prog->prelen, 60,
1707 const int len0 = UTF ? SvCUR(dsv0) : prog->prelen;
1708 const char * const s1 = do_utf8 ? sv_uni_display(dsv1, sv, 60,
1709 UNI_DISPLAY_REGEX) : startpos;
1710 const int len1 = do_utf8 ? SvCUR(dsv1) : strend - startpos;
1713 PerlIO_printf(Perl_debug_log,
1714 "%sMatching REx%s \"%s%*.*s%s%s\" against \"%s%.*s%s%s\"\n",
1715 PL_colors[4], PL_colors[5], PL_colors[0],
1718 len0 > 60 ? "..." : "",
1720 (int)(len1 > 60 ? 60 : len1),
1722 (len1 > 60 ? "..." : "")
1726 /* Simplest case: anchored match need be tried only once. */
1727 /* [unless only anchor is BOL and multiline is set] */
1728 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1729 if (s == startpos && regtry(prog, startpos))
1731 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1732 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1737 dontbother = minlen - 1;
1738 end = HOP3c(strend, -dontbother, strbeg) - 1;
1739 /* for multiline we only have to try after newlines */
1740 if (prog->check_substr || prog->check_utf8) {
1744 if (regtry(prog, s))
1749 if (prog->reganch & RE_USE_INTUIT) {
1750 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1761 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1762 if (regtry(prog, s))
1769 } else if (prog->reganch & ROPT_ANCH_GPOS) {
1770 if (regtry(prog, PL_reg_ganch))
1775 /* Messy cases: unanchored match. */
1776 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1777 /* we have /x+whatever/ */
1778 /* it must be a one character string (XXXX Except UTF?) */
1783 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1784 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1785 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1788 while (s < strend) {
1790 DEBUG_EXECUTE_r( did_match = 1 );
1791 if (regtry(prog, s)) goto got_it;
1793 while (s < strend && *s == ch)
1800 while (s < strend) {
1802 DEBUG_EXECUTE_r( did_match = 1 );
1803 if (regtry(prog, s)) goto got_it;
1805 while (s < strend && *s == ch)
1811 DEBUG_EXECUTE_r(if (!did_match)
1812 PerlIO_printf(Perl_debug_log,
1813 "Did not find anchored character...\n")
1816 else if (prog->anchored_substr != NULL
1817 || prog->anchored_utf8 != NULL
1818 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1819 && prog->float_max_offset < strend - s)) {
1824 char *last1; /* Last position checked before */
1828 if (prog->anchored_substr || prog->anchored_utf8) {
1829 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1830 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1831 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1832 back_max = back_min = prog->anchored_offset;
1834 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1835 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1836 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1837 back_max = prog->float_max_offset;
1838 back_min = prog->float_min_offset;
1840 if (must == &PL_sv_undef)
1841 /* could not downgrade utf8 check substring, so must fail */
1844 last = HOP3c(strend, /* Cannot start after this */
1845 -(I32)(CHR_SVLEN(must)
1846 - (SvTAIL(must) != 0) + back_min), strbeg);
1849 last1 = HOPc(s, -1);
1851 last1 = s - 1; /* bogus */
1853 /* XXXX check_substr already used to find "s", can optimize if
1854 check_substr==must. */
1856 dontbother = end_shift;
1857 strend = HOPc(strend, -dontbother);
1858 while ( (s <= last) &&
1859 ((flags & REXEC_SCREAM)
1860 ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg,
1861 end_shift, &scream_pos, 0))
1862 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend),
1863 (unsigned char*)strend, must,
1864 multiline ? FBMrf_MULTILINE : 0))) ) {
1865 /* we may be pointing at the wrong string */
1866 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1867 s = strbeg + (s - SvPVX_const(sv));
1868 DEBUG_EXECUTE_r( did_match = 1 );
1869 if (HOPc(s, -back_max) > last1) {
1870 last1 = HOPc(s, -back_min);
1871 s = HOPc(s, -back_max);
1874 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1876 last1 = HOPc(s, -back_min);
1880 while (s <= last1) {
1881 if (regtry(prog, s))
1887 while (s <= last1) {
1888 if (regtry(prog, s))
1894 DEBUG_EXECUTE_r(if (!did_match)
1895 PerlIO_printf(Perl_debug_log,
1896 "Did not find %s substr \"%s%.*s%s\"%s...\n",
1897 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1898 ? "anchored" : "floating"),
1900 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
1902 PL_colors[1], (SvTAIL(must) ? "$" : ""))
1906 else if ((c = prog->regstclass)) {
1908 I32 op = (U8)OP(prog->regstclass);
1909 /* don't bother with what can't match */
1910 if (PL_regkind[op] != EXACT && op != CANY)
1911 strend = HOPc(strend, -(minlen - 1));
1914 SV *prop = sv_newmortal();
1920 regprop(prog, prop, c);
1922 pv_uni_display(dsv0, (U8*)SvPVX_const(prop), SvCUR(prop), 60,
1923 UNI_DISPLAY_REGEX) :
1925 len0 = UTF ? SvCUR(dsv0) : SvCUR(prop);
1927 sv_uni_display(dsv1, sv, 60, UNI_DISPLAY_REGEX) : s;
1928 len1 = UTF ? SvCUR(dsv1) : strend - s;
1929 PerlIO_printf(Perl_debug_log,
1930 "Matching stclass \"%*.*s\" against \"%*.*s\"\n",
1934 if (find_byclass(prog, c, s, strend, 0))
1936 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass...\n"));
1940 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1945 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1946 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1947 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1949 if (flags & REXEC_SCREAM) {
1950 last = screaminstr(sv, float_real, s - strbeg,
1951 end_shift, &scream_pos, 1); /* last one */
1953 last = scream_olds; /* Only one occurrence. */
1954 /* we may be pointing at the wrong string */
1955 else if (RX_MATCH_COPIED(prog))
1956 s = strbeg + (s - SvPVX_const(sv));
1960 const char * const little = SvPV_const(float_real, len);
1962 if (SvTAIL(float_real)) {
1963 if (memEQ(strend - len + 1, little, len - 1))
1964 last = strend - len + 1;
1965 else if (!multiline)
1966 last = memEQ(strend - len, little, len)
1967 ? strend - len : NULL;
1973 last = rninstr(s, strend, little, little + len);
1975 last = strend; /* matching "$" */
1979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1980 "%sCan't trim the tail, match fails (should not happen)%s\n",
1981 PL_colors[4], PL_colors[5]));
1982 goto phooey; /* Should not happen! */
1984 dontbother = strend - last + prog->float_min_offset;
1986 if (minlen && (dontbother < minlen))
1987 dontbother = minlen - 1;
1988 strend -= dontbother; /* this one's always in bytes! */
1989 /* We don't know much -- general case. */
1992 if (regtry(prog, s))
2001 if (regtry(prog, s))
2003 } while (s++ < strend);
2011 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2013 if (PL_reg_eval_set) {
2014 /* Preserve the current value of $^R */
2015 if (oreplsv != GvSV(PL_replgv))
2016 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2017 restored, the value remains
2019 restore_pos(aTHX_ prog);
2022 /* make sure $`, $&, $', and $digit will work later */
2023 if ( !(flags & REXEC_NOT_FIRST) ) {
2024 RX_MATCH_COPY_FREE(prog);
2025 if (flags & REXEC_COPY_STR) {
2026 I32 i = PL_regeol - startpos + (stringarg - strbeg);
2027 #ifdef PERL_OLD_COPY_ON_WRITE
2029 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2031 PerlIO_printf(Perl_debug_log,
2032 "Copy on write: regexp capture, type %d\n",
2035 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2036 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2037 assert (SvPOKp(prog->saved_copy));
2041 RX_MATCH_COPIED_on(prog);
2042 s = savepvn(strbeg, i);
2048 prog->subbeg = strbeg;
2049 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2056 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2057 PL_colors[4], PL_colors[5]));
2058 if (PL_reg_eval_set)
2059 restore_pos(aTHX_ prog);
2064 - regtry - try match at specific point
2066 STATIC I32 /* 0 failure, 1 success */
2067 S_regtry(pTHX_ regexp *prog, char *startpos)
2073 GET_RE_DEBUG_FLAGS_DECL;
2076 PL_regindent = 0; /* XXXX Not good when matches are reenterable... */
2078 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2081 PL_reg_eval_set = RS_init;
2082 DEBUG_EXECUTE_r(DEBUG_s(
2083 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2084 (IV)(PL_stack_sp - PL_stack_base));
2086 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2087 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2088 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2090 /* Apparently this is not needed, judging by wantarray. */
2091 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2092 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2095 /* Make $_ available to executed code. */
2096 if (PL_reg_sv != DEFSV) {
2101 if (!(SvTYPE(PL_reg_sv) >= SVt_PVMG && SvMAGIC(PL_reg_sv)
2102 && (mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global)))) {
2103 /* prepare for quick setting of pos */
2104 #ifdef PERL_OLD_COPY_ON_WRITE
2106 sv_force_normal_flags(sv, 0);
2108 mg = sv_magicext(PL_reg_sv, (SV*)0, PERL_MAGIC_regex_global,
2109 &PL_vtbl_mglob, NULL, 0);
2113 PL_reg_oldpos = mg->mg_len;
2114 SAVEDESTRUCTOR_X(restore_pos, prog);
2116 if (!PL_reg_curpm) {
2117 Newxz(PL_reg_curpm, 1, PMOP);
2120 SV* repointer = newSViv(0);
2121 /* so we know which PL_regex_padav element is PL_reg_curpm */
2122 SvFLAGS(repointer) |= SVf_BREAK;
2123 av_push(PL_regex_padav,repointer);
2124 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2125 PL_regex_pad = AvARRAY(PL_regex_padav);
2129 PM_SETRE(PL_reg_curpm, prog);
2130 PL_reg_oldcurpm = PL_curpm;
2131 PL_curpm = PL_reg_curpm;
2132 if (RX_MATCH_COPIED(prog)) {
2133 /* Here is a serious problem: we cannot rewrite subbeg,
2134 since it may be needed if this match fails. Thus
2135 $` inside (?{}) could fail... */
2136 PL_reg_oldsaved = prog->subbeg;
2137 PL_reg_oldsavedlen = prog->sublen;
2138 #ifdef PERL_OLD_COPY_ON_WRITE
2139 PL_nrs = prog->saved_copy;
2141 RX_MATCH_COPIED_off(prog);
2144 PL_reg_oldsaved = NULL;
2145 prog->subbeg = PL_bostr;
2146 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2148 prog->startp[0] = startpos - PL_bostr;
2149 PL_reginput = startpos;
2150 PL_regstartp = prog->startp;
2151 PL_regendp = prog->endp;
2152 PL_reglastparen = &prog->lastparen;
2153 PL_reglastcloseparen = &prog->lastcloseparen;
2154 prog->lastparen = 0;
2155 prog->lastcloseparen = 0;
2157 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2158 if (PL_reg_start_tmpl <= prog->nparens) {
2159 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2160 if(PL_reg_start_tmp)
2161 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2163 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2166 /* XXXX What this code is doing here?!!! There should be no need
2167 to do this again and again, PL_reglastparen should take care of
2170 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2171 * Actually, the code in regcppop() (which Ilya may be meaning by
2172 * PL_reglastparen), is not needed at all by the test suite
2173 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2174 * enough, for building DynaLoader, or otherwise this
2175 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2176 * will happen. Meanwhile, this code *is* needed for the
2177 * above-mentioned test suite tests to succeed. The common theme
2178 * on those tests seems to be returning null fields from matches.
2183 if (prog->nparens) {
2185 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2192 if (regmatch(prog, prog->program + 1)) {
2193 prog->endp[0] = PL_reginput - PL_bostr;
2196 REGCP_UNWIND(lastcp);
2200 #define RE_UNWIND_BRANCH 1
2201 #define RE_UNWIND_BRANCHJ 2
2205 typedef struct { /* XX: makes sense to enlarge it... */
2209 } re_unwind_generic_t;
2223 } re_unwind_branch_t;
2225 typedef union re_unwind_t {
2227 re_unwind_generic_t generic;
2228 re_unwind_branch_t branch;
2231 #define sayYES goto yes
2232 #define sayNO goto no
2233 #define sayNO_ANYOF goto no_anyof
2234 #define sayYES_FINAL goto yes_final
2235 #define sayYES_LOUD goto yes_loud
2236 #define sayNO_FINAL goto no_final
2237 #define sayNO_SILENT goto do_no
2238 #define saySAME(x) if (x) goto yes; else goto no
2240 #define POSCACHE_SUCCESS 0 /* caching success rather than failure */
2241 #define POSCACHE_SEEN 1 /* we know what we're caching */
2242 #define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */
2243 #define CACHEsayYES STMT_START { \
2244 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \
2245 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2246 PL_reg_poscache[0] |= (1<<POSCACHE_SUCCESS) || (1<<POSCACHE_SEEN); \
2247 else if (!(PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2248 /* cache records failure, but this is success */ \
2250 PerlIO_printf(Perl_debug_log, \
2251 "%*s (remove success from failure cache)\n", \
2252 REPORT_CODE_OFF+PL_regindent*2, "") \
2254 PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<<st->u.whilem.cache_bit); \
2259 #define CACHEsayNO STMT_START { \
2260 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \
2261 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2262 PL_reg_poscache[0] |= (1<<POSCACHE_SEEN); \
2263 else if ((PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2264 /* cache records success, but this is failure */ \
2266 PerlIO_printf(Perl_debug_log, \
2267 "%*s (remove failure from success cache)\n", \
2268 REPORT_CODE_OFF+PL_regindent*2, "") \
2270 PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<<st->u.whilem.cache_bit); \
2276 /* this is used to determine how far from the left messages like
2277 'failed...' are printed. Currently 29 makes these messages line
2278 up with the opcode they refer to. Earlier perls used 25 which
2279 left these messages outdented making reviewing a debug output
2282 #define REPORT_CODE_OFF 29
2285 /* Make sure there is a test for this +1 options in re_tests */
2286 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2288 /* grab a new slab and return the first slot in it */
2290 STATIC regmatch_state *
2293 regmatch_slab *s = PL_regmatch_slab->next;
2295 Newx(s, 1, regmatch_slab);
2296 s->prev = PL_regmatch_slab;
2298 PL_regmatch_slab->next = s;
2300 PL_regmatch_slab = s;
2301 return &s->states[0];
2304 /* simulate a recursive call to regmatch */
2306 #define REGMATCH(ns, where) \
2309 st->resume_state = resume_##where; \
2310 goto start_recurse; \
2311 resume_point_##where:
2314 /* push a new regex state. Set newst to point to it */
2316 #define PUSH_STATE(newst, resume) \
2318 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "PUSH STATE(%d)\n", depth)); \
2322 st->locinput = locinput; \
2323 st->resume_state = resume; \
2325 if (newst > &(PL_regmatch_slab->states[PERL_REGMATCH_SLAB_SLOTS-1])) \
2326 newst = S_push_slab(aTHX); \
2327 PL_regmatch_state = newst; \
2329 newst->minmod = 0; \
2331 newst->logical = 0; \
2332 newst->unwind = 0; \
2333 locinput = PL_reginput; \
2334 nextchr = UCHARAT(locinput);
2337 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "POP STATE(%d)\n", depth)); \
2340 if (st < &PL_regmatch_slab->states[0]) { \
2341 PL_regmatch_slab = PL_regmatch_slab->prev; \
2342 st = &PL_regmatch_slab->states[PERL_REGMATCH_SLAB_SLOTS-1]; \
2344 PL_regmatch_state = st; \
2348 locinput = st->locinput; \
2349 nextchr = UCHARAT(locinput);
2352 - regmatch - main matching routine
2354 * Conceptually the strategy is simple: check to see whether the current
2355 * node matches, call self recursively to see whether the rest matches,
2356 * and then act accordingly. In practice we make some effort to avoid
2357 * recursion, in particular by going through "ordinary" nodes (that don't
2358 * need to know whether the rest of the match failed) by a loop instead of
2361 /* [lwall] I've hoisted the register declarations to the outer block in order to
2362 * maybe save a little bit of pushing and popping on the stack. It also takes
2363 * advantage of machines that use a register save mask on subroutine entry.
2365 * This function used to be heavily recursive, but since this had the
2366 * effect of blowing the CPU stack on complex regexes, it has been
2367 * restructured to be iterative, and to save state onto the heap rather
2368 * than the stack. Essentially whereever regmatch() used to be called, it
2369 * pushes the current state, notes where to return, then jumps back into
2372 * Originally the structure of this function used to look something like
2377 while (scan != NULL) {
2378 a++; // do stuff with a and b
2384 if (regmatch(...)) // recurse
2394 * Now it looks something like this:
2402 regmatch_state *st = new();
2404 st->a++; // do stuff with a and b
2406 while (scan != NULL) {
2414 st->resume_state = resume_FOO;
2415 goto start_recurse; // recurse
2424 st = new(); push a new state
2425 st->a = 1; st->b = 2;
2432 switch (resume_state) {
2434 goto resume_point_FOO;
2441 * WARNING: this means that any line in this function that contains a
2442 * REGMATCH() or TRYPAREN() is actually simulating a recursive call to
2443 * regmatch() using gotos instead. Thus the values of any local variables
2444 * not saved in the regmatch_state structure will have been lost when
2445 * execution resumes on the next line .
2447 * States (ie the st pointer) are allocated in slabs of about 4K in size.
2448 * PL_regmatch_state always points to the currently active state, and
2449 * PL_regmatch_slab points to the slab currently containing PL_regmatch_state.
2450 * The first time regmatch is called, the first slab is allocated, and is
2451 * never freed until interpreter desctruction. When the slab is full,
2452 * a new one is allocated chained to the end. At exit from regmatch, slabs
2453 * allocated since entry are freed.
2457 STATIC I32 /* 0 failure, 1 success */
2458 S_regmatch(pTHX_ regexp *rex, regnode *prog)
2461 register const bool do_utf8 = PL_reg_match_utf8;
2462 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2464 regmatch_slab *orig_slab;
2465 regmatch_state *orig_state;
2467 /* the current state. This is a cached copy of PL_regmatch_state */
2468 register regmatch_state *st;
2470 /* cache heavy used fields of st in registers */
2471 register regnode *scan;
2472 register regnode *next;
2473 register I32 n = 0; /* initialize to shut up compiler warning */
2474 register char *locinput = PL_reginput;
2476 /* these variables are NOT saved during a recusive RFEGMATCH: */
2477 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2478 bool result; /* return value of S_regmatch */
2479 regnode *inner; /* Next node in internal branch. */
2480 int depth = 0; /* depth of recursion */
2481 regmatch_state *newst; /* when pushing a state, this is the new one */
2482 regmatch_state *cur_eval = NULL; /* most recent (??{}) state */
2485 SV *re_debug_flags = NULL;
2490 /* on first ever call to regmatch, allocate first slab */
2491 if (!PL_regmatch_slab) {
2492 Newx(PL_regmatch_slab, 1, regmatch_slab);
2493 PL_regmatch_slab->prev = NULL;
2494 PL_regmatch_slab->next = NULL;
2495 PL_regmatch_state = &PL_regmatch_slab->states[0] - 1;
2498 /* remember current high-water mark for exit */
2499 /* XXX this should be done with SAVE* instead */
2500 orig_slab = PL_regmatch_slab;
2501 orig_state = PL_regmatch_state;
2503 /* grab next free state slot */
2504 st = ++PL_regmatch_state;
2505 if (st > &(PL_regmatch_slab->states[PERL_REGMATCH_SLAB_SLOTS-1]))
2506 st = PL_regmatch_state = S_push_slab(aTHX);
2513 /* Note that nextchr is a byte even in UTF */
2514 nextchr = UCHARAT(locinput);
2516 while (scan != NULL) {
2519 SV * const prop = sv_newmortal();
2520 const int docolor = *PL_colors[0];
2521 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2522 int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput);
2523 /* The part of the string before starttry has one color
2524 (pref0_len chars), between starttry and current
2525 position another one (pref_len - pref0_len chars),
2526 after the current position the third one.
2527 We assume that pref0_len <= pref_len, otherwise we
2528 decrease pref0_len. */
2529 int pref_len = (locinput - PL_bostr) > (5 + taill) - l
2530 ? (5 + taill) - l : locinput - PL_bostr;
2533 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2535 pref0_len = pref_len - (locinput - PL_reg_starttry);
2536 if (l + pref_len < (5 + taill) && l < PL_regeol - locinput)
2537 l = ( PL_regeol - locinput > (5 + taill) - pref_len
2538 ? (5 + taill) - pref_len : PL_regeol - locinput);
2539 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2543 if (pref0_len > pref_len)
2544 pref0_len = pref_len;
2545 regprop(rex, prop, scan);
2547 const char * const s0 =
2548 do_utf8 && OP(scan) != CANY ?
2549 pv_uni_display(PERL_DEBUG_PAD(0), (U8*)(locinput - pref_len),
2550 pref0_len, 60, UNI_DISPLAY_REGEX) :
2551 locinput - pref_len;
2552 const int len0 = do_utf8 ? strlen(s0) : pref0_len;
2553 const char * const s1 = do_utf8 && OP(scan) != CANY ?
2554 pv_uni_display(PERL_DEBUG_PAD(1),
2555 (U8*)(locinput - pref_len + pref0_len),
2556 pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) :
2557 locinput - pref_len + pref0_len;
2558 const int len1 = do_utf8 ? strlen(s1) : pref_len - pref0_len;
2559 const char * const s2 = do_utf8 && OP(scan) != CANY ?
2560 pv_uni_display(PERL_DEBUG_PAD(2), (U8*)locinput,
2561 PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) :
2563 const int len2 = do_utf8 ? strlen(s2) : l;
2564 PerlIO_printf(Perl_debug_log,
2565 "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|%3"IVdf":%*s%s\n",
2566 (IV)(locinput - PL_bostr),
2573 (docolor ? "" : "> <"),
2577 15 - l - pref_len + 1,
2579 (IV)(scan - rex->program), PL_regindent*2, "",
2584 next = scan + NEXT_OFF(scan);
2590 if (locinput == PL_bostr)
2592 /* regtill = regbol; */
2597 if (locinput == PL_bostr ||
2598 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2604 if (locinput == PL_bostr)
2608 if (locinput == PL_reg_ganch)
2614 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2619 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2621 if (PL_regeol - locinput > 1)
2625 if (PL_regeol != locinput)
2629 if (!nextchr && locinput >= PL_regeol)
2632 locinput += PL_utf8skip[nextchr];
2633 if (locinput > PL_regeol)
2635 nextchr = UCHARAT(locinput);
2638 nextchr = UCHARAT(++locinput);
2641 if (!nextchr && locinput >= PL_regeol)
2643 nextchr = UCHARAT(++locinput);
2646 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2649 locinput += PL_utf8skip[nextchr];
2650 if (locinput > PL_regeol)
2652 nextchr = UCHARAT(locinput);
2655 nextchr = UCHARAT(++locinput);
2661 traverse the TRIE keeping track of all accepting states
2662 we transition through until we get to a failing node.
2670 U8 *uc = ( U8* )locinput;
2677 U8 *uscan = (U8*)NULL;
2679 SV *sv_accept_buff = NULL;
2680 const enum { trie_plain, trie_utf8, trie_uft8_fold }
2681 trie_type = do_utf8 ?
2682 (OP(scan) == TRIE ? trie_utf8 : trie_uft8_fold)
2685 /* what trie are we using right now */
2687 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2688 st->u.trie.accepted = 0; /* how many accepting states we have seen */
2691 while ( state && uc <= (U8*)PL_regeol ) {
2693 if (trie->states[ state ].wordnum) {
2694 if (!st->u.trie.accepted ) {
2697 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2698 sv_accept_buff=newSV(bufflen *
2699 sizeof(reg_trie_accepted) - 1);
2700 SvCUR_set(sv_accept_buff,
2701 sizeof(reg_trie_accepted));
2702 SvPOK_on(sv_accept_buff);
2703 sv_2mortal(sv_accept_buff);
2704 st->u.trie.accept_buff =
2705 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2708 if (st->u.trie.accepted >= bufflen) {
2710 st->u.trie.accept_buff =(reg_trie_accepted*)
2711 SvGROW(sv_accept_buff,
2712 bufflen * sizeof(reg_trie_accepted));
2714 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2715 + sizeof(reg_trie_accepted));
2717 st->u.trie.accept_buff[st->u.trie.accepted].wordnum = trie->states[state].wordnum;
2718 st->u.trie.accept_buff[st->u.trie.accepted].endpos = uc;
2719 ++st->u.trie.accepted;
2722 base = trie->states[ state ].trans.base;
2724 DEBUG_TRIE_EXECUTE_r(
2725 PerlIO_printf( Perl_debug_log,
2726 "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ",
2727 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2728 (UV)state, (UV)base, (UV)st->u.trie.accepted );
2732 switch (trie_type) {
2733 case trie_uft8_fold:
2735 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags );
2740 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2741 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2742 uvc = to_uni_fold( uvc, foldbuf, &foldlen );
2743 foldlen -= UNISKIP( uvc );
2744 uscan = foldbuf + UNISKIP( uvc );
2748 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN,
2757 charid = trie->charmap[ uvc ];
2761 if (trie->widecharmap) {
2762 SV** svpp = (SV**)NULL;
2763 svpp = hv_fetch(trie->widecharmap,
2764 (char*)&uvc, sizeof(UV), 0);
2766 charid = (U16)SvIV(*svpp);
2771 (base + charid > trie->uniquecharcount )
2772 && (base + charid - 1 - trie->uniquecharcount
2774 && trie->trans[base + charid - 1 -
2775 trie->uniquecharcount].check == state)
2777 state = trie->trans[base + charid - 1 -
2778 trie->uniquecharcount ].next;
2789 DEBUG_TRIE_EXECUTE_r(
2790 PerlIO_printf( Perl_debug_log,
2791 "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n",
2792 charid, uvc, (UV)state, PL_colors[5] );
2795 if (!st->u.trie.accepted )
2799 There was at least one accepting state that we
2800 transitioned through. Presumably the number of accepting
2801 states is going to be low, typically one or two. So we
2802 simply scan through to find the one with lowest wordnum.
2803 Once we find it, we swap the last state into its place
2804 and decrement the size. We then try to match the rest of
2805 the pattern at the point where the word ends, if we
2806 succeed then we end the loop, otherwise the loop
2807 eventually terminates once all of the accepting states
2811 if ( st->u.trie.accepted == 1 ) {
2813 SV ** const tmp = av_fetch( trie->words, st->u.trie.accept_buff[ 0 ].wordnum-1, 0 );
2814 PerlIO_printf( Perl_debug_log,
2815 "%*s %sonly one match : #%d <%s>%s\n",
2816 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2817 st->u.trie.accept_buff[ 0 ].wordnum,
2818 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2821 PL_reginput = (char *)st->u.trie.accept_buff[ 0 ].endpos;
2822 /* in this case we free tmps/leave before we call regmatch
2823 as we wont be using accept_buff again. */
2826 REGMATCH(scan + NEXT_OFF(scan), TRIE1);
2827 /*** all unsaved local vars undefined at this point */
2830 PerlIO_printf( Perl_debug_log,"%*s %sgot %"IVdf" possible matches%s\n",
2831 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], (IV)st->u.trie.accepted,
2834 while ( !result && st->u.trie.accepted-- ) {
2837 for( cur = 1 ; cur <= st->u.trie.accepted ; cur++ ) {
2838 DEBUG_TRIE_EXECUTE_r(
2839 PerlIO_printf( Perl_debug_log,
2840 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2841 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2842 (IV)best, st->u.trie.accept_buff[ best ].wordnum, (IV)cur,
2843 st->u.trie.accept_buff[ cur ].wordnum, PL_colors[5] );
2846 if (st->u.trie.accept_buff[cur].wordnum <
2847 st->u.trie.accept_buff[best].wordnum)
2851 reg_trie_data * const trie = (reg_trie_data*)
2852 rex->data->data[ARG(scan)];
2853 SV ** const tmp = av_fetch( trie->words, st->u.trie.accept_buff[ best ].wordnum - 1, 0 );
2854 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at 0x%p%s\n",
2855 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2856 st->u.trie.accept_buff[best].wordnum,
2857 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",scan,
2860 if ( best<st->u.trie.accepted ) {
2861 reg_trie_accepted tmp = st->u.trie.accept_buff[ best ];
2862 st->u.trie.accept_buff[ best ] = st->u.trie.accept_buff[ st->u.trie.accepted ];
2863 st->u.trie.accept_buff[ st->u.trie.accepted ] = tmp;
2864 best = st->u.trie.accepted;
2866 PL_reginput = (char *)st->u.trie.accept_buff[ best ].endpos;
2869 as far as I can tell we only need the SAVETMPS/FREETMPS
2870 for re's with EVAL in them but I'm leaving them in for
2871 all until I can be sure.
2874 REGMATCH(scan + NEXT_OFF(scan), TRIE2);
2875 /*** all unsaved local vars undefined at this point */
2888 /* unreached codepoint */
2890 char *s = STRING(scan);
2891 st->ln = STR_LEN(scan);
2892 if (do_utf8 != UTF) {
2893 /* The target and the pattern have differing utf8ness. */
2895 const char *e = s + st->ln;
2898 /* The target is utf8, the pattern is not utf8. */
2903 if (NATIVE_TO_UNI(*(U8*)s) !=
2904 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2912 /* The target is not utf8, the pattern is utf8. */
2917 if (NATIVE_TO_UNI(*((U8*)l)) !=
2918 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2926 nextchr = UCHARAT(locinput);
2929 /* The target and the pattern have the same utf8ness. */
2930 /* Inline the first character, for speed. */
2931 if (UCHARAT(s) != nextchr)
2933 if (PL_regeol - locinput < st->ln)
2935 if (st->ln > 1 && memNE(s, locinput, st->ln))
2938 nextchr = UCHARAT(locinput);
2942 PL_reg_flags |= RF_tainted;
2945 char *s = STRING(scan);
2946 st->ln = STR_LEN(scan);
2948 if (do_utf8 || UTF) {
2949 /* Either target or the pattern are utf8. */
2951 char *e = PL_regeol;
2953 if (ibcmp_utf8(s, 0, st->ln, (bool)UTF,
2954 l, &e, 0, do_utf8)) {
2955 /* One more case for the sharp s:
2956 * pack("U0U*", 0xDF) =~ /ss/i,
2957 * the 0xC3 0x9F are the UTF-8
2958 * byte sequence for the U+00DF. */
2960 toLOWER(s[0]) == 's' &&
2962 toLOWER(s[1]) == 's' &&
2969 nextchr = UCHARAT(locinput);
2973 /* Neither the target and the pattern are utf8. */
2975 /* Inline the first character, for speed. */
2976 if (UCHARAT(s) != nextchr &&
2977 UCHARAT(s) != ((OP(scan) == EXACTF)
2978 ? PL_fold : PL_fold_locale)[nextchr])
2980 if (PL_regeol - locinput < st->ln)
2982 if (st->ln > 1 && (OP(scan) == EXACTF
2983 ? ibcmp(s, locinput, st->ln)
2984 : ibcmp_locale(s, locinput, st->ln)))
2987 nextchr = UCHARAT(locinput);
2992 STRLEN inclasslen = PL_regeol - locinput;
2994 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
2996 if (locinput >= PL_regeol)
2998 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
2999 nextchr = UCHARAT(locinput);
3004 nextchr = UCHARAT(locinput);
3005 if (!REGINCLASS(rex, scan, (U8*)locinput))
3007 if (!nextchr && locinput >= PL_regeol)
3009 nextchr = UCHARAT(++locinput);
3013 /* If we might have the case of the German sharp s
3014 * in a casefolding Unicode character class. */
3016 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3017 locinput += SHARP_S_SKIP;
3018 nextchr = UCHARAT(locinput);
3024 PL_reg_flags |= RF_tainted;
3030 LOAD_UTF8_CHARCLASS_ALNUM();
3031 if (!(OP(scan) == ALNUM
3032 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3033 : isALNUM_LC_utf8((U8*)locinput)))
3037 locinput += PL_utf8skip[nextchr];
3038 nextchr = UCHARAT(locinput);
3041 if (!(OP(scan) == ALNUM
3042 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3044 nextchr = UCHARAT(++locinput);
3047 PL_reg_flags |= RF_tainted;
3050 if (!nextchr && locinput >= PL_regeol)
3053 LOAD_UTF8_CHARCLASS_ALNUM();
3054 if (OP(scan) == NALNUM
3055 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3056 : isALNUM_LC_utf8((U8*)locinput))
3060 locinput += PL_utf8skip[nextchr];
3061 nextchr = UCHARAT(locinput);
3064 if (OP(scan) == NALNUM
3065 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3067 nextchr = UCHARAT(++locinput);
3071 PL_reg_flags |= RF_tainted;
3075 /* was last char in word? */
3077 if (locinput == PL_bostr)
3080 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3082 st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3084 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3085 st->ln = isALNUM_uni(st->ln);
3086 LOAD_UTF8_CHARCLASS_ALNUM();
3087 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3090 st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln));
3091 n = isALNUM_LC_utf8((U8*)locinput);
3095 st->ln = (locinput != PL_bostr) ?
3096 UCHARAT(locinput - 1) : '\n';
3097 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3098 st->ln = isALNUM(st->ln);
3099 n = isALNUM(nextchr);
3102 st->ln = isALNUM_LC(st->ln);
3103 n = isALNUM_LC(nextchr);
3106 if (((!st->ln) == (!n)) == (OP(scan) == BOUND ||
3107 OP(scan) == BOUNDL))
3111 PL_reg_flags |= RF_tainted;
3117 if (UTF8_IS_CONTINUED(nextchr)) {
3118 LOAD_UTF8_CHARCLASS_SPACE();
3119 if (!(OP(scan) == SPACE
3120 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3121 : isSPACE_LC_utf8((U8*)locinput)))
3125 locinput += PL_utf8skip[nextchr];
3126 nextchr = UCHARAT(locinput);
3129 if (!(OP(scan) == SPACE
3130 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3132 nextchr = UCHARAT(++locinput);
3135 if (!(OP(scan) == SPACE
3136 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3138 nextchr = UCHARAT(++locinput);
3142 PL_reg_flags |= RF_tainted;
3145 if (!nextchr && locinput >= PL_regeol)
3148 LOAD_UTF8_CHARCLASS_SPACE();
3149 if (OP(scan) == NSPACE
3150 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3151 : isSPACE_LC_utf8((U8*)locinput))
3155 locinput += PL_utf8skip[nextchr];
3156 nextchr = UCHARAT(locinput);
3159 if (OP(scan) == NSPACE
3160 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3162 nextchr = UCHARAT(++locinput);
3165 PL_reg_flags |= RF_tainted;
3171 LOAD_UTF8_CHARCLASS_DIGIT();
3172 if (!(OP(scan) == DIGIT
3173 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3174 : isDIGIT_LC_utf8((U8*)locinput)))
3178 locinput += PL_utf8skip[nextchr];
3179 nextchr = UCHARAT(locinput);
3182 if (!(OP(scan) == DIGIT
3183 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3185 nextchr = UCHARAT(++locinput);
3188 PL_reg_flags |= RF_tainted;
3191 if (!nextchr && locinput >= PL_regeol)
3194 LOAD_UTF8_CHARCLASS_DIGIT();
3195 if (OP(scan) == NDIGIT
3196 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3197 : isDIGIT_LC_utf8((U8*)locinput))
3201 locinput += PL_utf8skip[nextchr];
3202 nextchr = UCHARAT(locinput);
3205 if (OP(scan) == NDIGIT
3206 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3208 nextchr = UCHARAT(++locinput);
3211 if (locinput >= PL_regeol)
3214 LOAD_UTF8_CHARCLASS_MARK();
3215 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3217 locinput += PL_utf8skip[nextchr];
3218 while (locinput < PL_regeol &&
3219 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3220 locinput += UTF8SKIP(locinput);
3221 if (locinput > PL_regeol)
3226 nextchr = UCHARAT(locinput);
3229 PL_reg_flags |= RF_tainted;
3234 n = ARG(scan); /* which paren pair */
3235 st->ln = PL_regstartp[n];
3236 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3237 if ((I32)*PL_reglastparen < n || st->ln == -1)
3238 sayNO; /* Do not match unless seen CLOSEn. */
3239 if (st->ln == PL_regendp[n])
3242 s = PL_bostr + st->ln;
3243 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3245 const char *e = PL_bostr + PL_regendp[n];
3247 * Note that we can't do the "other character" lookup trick as
3248 * in the 8-bit case (no pun intended) because in Unicode we
3249 * have to map both upper and title case to lower case.
3251 if (OP(scan) == REFF) {
3253 STRLEN ulen1, ulen2;
3254 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3255 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3259 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3260 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3261 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3268 nextchr = UCHARAT(locinput);
3272 /* Inline the first character, for speed. */
3273 if (UCHARAT(s) != nextchr &&
3275 (UCHARAT(s) != ((OP(scan) == REFF
3276 ? PL_fold : PL_fold_locale)[nextchr]))))
3278 st->ln = PL_regendp[n] - st->ln;
3279 if (locinput + st->ln > PL_regeol)
3281 if (st->ln > 1 && (OP(scan) == REF
3282 ? memNE(s, locinput, st->ln)
3284 ? ibcmp(s, locinput, st->ln)
3285 : ibcmp_locale(s, locinput, st->ln))))
3288 nextchr = UCHARAT(locinput);
3301 /* execute the code in the {...} */
3303 SV ** const before = SP;
3304 OP_4tree * const oop = PL_op;
3305 COP * const ocurcop = PL_curcop;
3309 PL_op = (OP_4tree*)rex->data->data[n];
3310 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3311 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3312 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3314 CALLRUNOPS(aTHX); /* Scalar context. */
3317 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3324 PAD_RESTORE_LOCAL(old_comppad);
3325 PL_curcop = ocurcop;
3328 sv_setsv(save_scalar(PL_replgv), ret);
3332 if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */
3335 /* extract RE object from returned value; compiling if
3340 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3341 mg = mg_find(sv, PERL_MAGIC_qr);
3342 else if (SvSMAGICAL(ret)) {
3343 if (SvGMAGICAL(ret))
3344 sv_unmagic(ret, PERL_MAGIC_qr);
3346 mg = mg_find(ret, PERL_MAGIC_qr);
3350 re = (regexp *)mg->mg_obj;
3351 (void)ReREFCNT_inc(re);
3355 const char * const t = SvPV_const(ret, len);
3357 const I32 osize = PL_regsize;
3360 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3361 re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm);
3363 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3365 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3371 /* run the pattern returned from (??{...}) */
3374 PerlIO_printf(Perl_debug_log,
3375 "Entering embedded \"%s%.60s%s%s\"\n",
3379 (strlen(re->precomp) > 60 ? "..." : ""))
3382 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
3383 REGCP_SET(st->u.eval.lastcp);
3384 *PL_reglastparen = 0;
3385 *PL_reglastcloseparen = 0;
3386 PL_reginput = locinput;
3388 /* XXXX This is too dramatic a measure... */
3392 st->u.eval.toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3393 ((re->reganch & ROPT_UTF8) != 0);
3394 if (st->u.eval.toggleutf) PL_reg_flags ^= RF_utf8;
3395 st->u.eval.prev_rex = rex;
3398 st->u.eval.prev_eval = cur_eval;
3399 st->u.eval.prev_slab = PL_regmatch_slab;
3400 st->u.eval.depth = depth;
3402 PUSH_STATE(newst, resume_EVAL);
3405 /* now continue from first node in postoned RE */
3406 next = re->program + 1;
3410 /* /(?(?{...})X|Y)/ */
3411 st->sw = SvTRUE(ret);
3416 n = ARG(scan); /* which paren pair */
3417 PL_reg_start_tmp[n] = locinput;
3422 n = ARG(scan); /* which paren pair */
3423 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3424 PL_regendp[n] = locinput - PL_bostr;
3425 if (n > (I32)*PL_reglastparen)
3426 *PL_reglastparen = n;
3427 *PL_reglastcloseparen = n;
3430 n = ARG(scan); /* which paren pair */
3431 st->sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3434 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3436 next = NEXTOPER(NEXTOPER(scan));
3438 next = scan + ARG(scan);
3439 if (OP(next) == IFTHEN) /* Fake one. */
3440 next = NEXTOPER(NEXTOPER(next));
3444 st->logical = scan->flags;
3446 /*******************************************************************
3447 cc points to the regmatch_state associated with the most recent CURLYX.
3448 This struct contains info about the innermost (...)* loop (an
3449 "infoblock"), and a pointer to the next outer cc.
3451 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3453 1) After matching Y, regnode for CURLYX is processed;
3455 2) This regnode populates cc, and calls regmatch() recursively
3456 with the starting point at WHILEM node;
3458 3) Each hit of WHILEM node tries to match A and Z (in the order
3459 depending on the current iteration, min/max of {min,max} and
3460 greediness). The information about where are nodes for "A"
3461 and "Z" is read from cc, as is info on how many times "A"
3462 was already matched, and greediness.
3464 4) After A matches, the same WHILEM node is hit again.
3466 5) Each time WHILEM is hit, cc is the infoblock created by CURLYX
3467 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3468 resets cc, since this Y(A)*Z can be a part of some other loop:
3469 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3470 of the external loop.
3472 Currently present infoblocks form a tree with a stem formed by st->cc
3473 and whatever it mentions via ->next, and additional attached trees
3474 corresponding to temporarily unset infoblocks as in "5" above.
3476 In the following picture, infoblocks for outer loop of
3477 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3478 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3479 infoblocks are drawn below the "reset" infoblock.
3481 In fact in the picture below we do not show failed matches for Z and T
3482 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3483 more obvious *why* one needs to *temporary* unset infoblocks.]
3485 Matched REx position InfoBlocks Comment
3489 Y A)*?Z)*?T x <- O <- I
3490 YA )*?Z)*?T x <- O <- I
3491 YA A)*?Z)*?T x <- O <- I
3492 YAA )*?Z)*?T x <- O <- I
3493 YAA Z)*?T x <- O # Temporary unset I
3496 YAAZ Y(A)*?Z)*?T x <- O
3499 YAAZY (A)*?Z)*?T x <- O
3502 YAAZY A)*?Z)*?T x <- O <- I
3505 YAAZYA )*?Z)*?T x <- O <- I
3508 YAAZYA Z)*?T x <- O # Temporary unset I
3514 YAAZYAZ T x # Temporary unset O
3521 *******************************************************************/
3524 /* No need to save/restore up to this paren */
3525 I32 parenfloor = scan->flags;
3527 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3529 /* XXXX Probably it is better to teach regpush to support
3530 parenfloor > PL_regsize... */
3531 if (parenfloor > (I32)*PL_reglastparen)
3532 parenfloor = *PL_reglastparen; /* Pessimization... */
3534 st->u.curlyx.cp = PL_savestack_ix;
3535 st->u.curlyx.outercc = st->cc;
3537 /* these fields contain the state of the current curly.
3538 * they are accessed by subsequent WHILEMs;
3539 * cur and lastloc are also updated by WHILEM */
3540 st->u.curlyx.parenfloor = parenfloor;
3541 st->u.curlyx.cur = -1; /* this will be updated by WHILEM */
3542 st->u.curlyx.min = ARG1(scan);
3543 st->u.curlyx.max = ARG2(scan);
3544 st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3545 st->u.curlyx.lastloc = 0;
3546 /* st->next and st->minmod are also read by WHILEM */
3548 PL_reginput = locinput;
3549 REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */
3550 /*** all unsaved local vars undefined at this point */
3551 regcpblow(st->u.curlyx.cp);
3552 st->cc = st->u.curlyx.outercc;
3558 * This is really hard to understand, because after we match
3559 * what we're trying to match, we must make sure the rest of
3560 * the REx is going to match for sure, and to do that we have
3561 * to go back UP the parse tree by recursing ever deeper. And
3562 * if it fails, we have to reset our parent's current state
3563 * that we can try again after backing off.
3566 st->u.whilem.lastloc = st->cc->u.curlyx.lastloc; /* Detection of 0-len. */
3567 st->u.whilem.cache_offset = 0;
3568 st->u.whilem.cache_bit = 0;
3570 n = st->cc->u.curlyx.cur + 1; /* how many we know we matched */
3571 PL_reginput = locinput;
3574 PerlIO_printf(Perl_debug_log,
3575 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3576 REPORT_CODE_OFF+PL_regindent*2, "",
3577 (long)n, (long)st->cc->u.curlyx.min,
3578 (long)st->cc->u.curlyx.max, PTR2UV(st->cc))
3581 /* If degenerate scan matches "", assume scan done. */
3583 if (locinput == st->cc->u.curlyx.lastloc && n >= st->cc->u.curlyx.min) {
3584 st->u.whilem.savecc = st->cc;
3585 st->cc = st->cc->u.curlyx.outercc;
3587 st->ln = st->cc->u.curlyx.cur;
3589 PerlIO_printf(Perl_debug_log,
3590 "%*s empty match detected, try continuation...\n",
3591 REPORT_CODE_OFF+PL_regindent*2, "")
3593 REGMATCH(st->u.whilem.savecc->next, WHILEM1);
3594 /*** all unsaved local vars undefined at this point */
3595 st->cc = st->u.whilem.savecc;
3598 if (st->cc->u.curlyx.outercc)
3599 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3603 /* First just match a string of min scans. */
3605 if (n < st->cc->u.curlyx.min) {
3606 st->cc->u.curlyx.cur = n;
3607 st->cc->u.curlyx.lastloc = locinput;
3608 REGMATCH(st->cc->u.curlyx.scan, WHILEM2);
3609 /*** all unsaved local vars undefined at this point */
3612 st->cc->u.curlyx.cur = n - 1;
3613 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3618 /* Check whether we already were at this position.
3619 Postpone detection until we know the match is not
3620 *that* much linear. */
3621 if (!PL_reg_maxiter) {
3622 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3623 PL_reg_leftiter = PL_reg_maxiter;
3625 if (PL_reg_leftiter-- == 0) {
3626 const I32 size = (PL_reg_maxiter + 7 + POSCACHE_START)/8;
3627 if (PL_reg_poscache) {
3628 if ((I32)PL_reg_poscache_size < size) {
3629 Renew(PL_reg_poscache, size, char);
3630 PL_reg_poscache_size = size;
3632 Zero(PL_reg_poscache, size, char);
3635 PL_reg_poscache_size = size;
3636 Newxz(PL_reg_poscache, size, char);
3639 PerlIO_printf(Perl_debug_log,
3640 "%sDetected a super-linear match, switching on caching%s...\n",
3641 PL_colors[4], PL_colors[5])
3644 if (PL_reg_leftiter < 0) {
3645 st->u.whilem.cache_offset = locinput - PL_bostr;
3647 st->u.whilem.cache_offset = (scan->flags & 0xf) - 1 + POSCACHE_START
3648 + st->u.whilem.cache_offset * (scan->flags>>4);
3649 st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8;
3650 st->u.whilem.cache_offset /= 8;
3651 if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<<st->u.whilem.cache_bit)) {
3653 PerlIO_printf(Perl_debug_log,
3654 "%*s already tried at this position...\n",
3655 REPORT_CODE_OFF+PL_regindent*2, "")
3657 if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))
3658 /* cache records success */
3661 /* cache records failure */
3664 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit);
3668 /* Prefer next over scan for minimal matching. */
3670 if (st->cc->minmod) {
3671 st->u.whilem.savecc = st->cc;
3672 st->cc = st->cc->u.curlyx.outercc;
3674 st->ln = st->cc->u.curlyx.cur;
3675 st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor);
3676 REGCP_SET(st->u.whilem.lastcp);
3677 REGMATCH(st->u.whilem.savecc->next, WHILEM3);
3678 /*** all unsaved local vars undefined at this point */
3679 st->cc = st->u.whilem.savecc;
3681 regcpblow(st->u.whilem.cp);
3682 CACHEsayYES; /* All done. */
3684 REGCP_UNWIND(st->u.whilem.lastcp);
3686 if (st->cc->u.curlyx.outercc)
3687 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3689 if (n >= st->cc->u.curlyx.max) { /* Maximum greed exceeded? */
3690 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3691 && !(PL_reg_flags & RF_warned)) {
3692 PL_reg_flags |= RF_warned;
3693 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3694 "Complex regular subexpression recursion",
3701 PerlIO_printf(Perl_debug_log,
3702 "%*s trying longer...\n",
3703 REPORT_CODE_OFF+PL_regindent*2, "")
3705 /* Try scanning more and see if it helps. */
3706 PL_reginput = locinput;
3707 st->cc->u.curlyx.cur = n;
3708 st->cc->u.curlyx.lastloc = locinput;
3709 st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor);
3710 REGCP_SET(st->u.whilem.lastcp);
3711 REGMATCH(st->cc->u.curlyx.scan, WHILEM4);
3712 /*** all unsaved local vars undefined at this point */
3714 regcpblow(st->u.whilem.cp);
3717 REGCP_UNWIND(st->u.whilem.lastcp);
3719 st->cc->u.curlyx.cur = n - 1;
3720 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3724 /* Prefer scan over next for maximal matching. */
3726 if (n < st->cc->u.curlyx.max) { /* More greed allowed? */
3727 st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor);
3728 st->cc->u.curlyx.cur = n;
3729 st->cc->u.curlyx.lastloc = locinput;
3730 REGCP_SET(st->u.whilem.lastcp);
3731 REGMATCH(st->cc->u.curlyx.scan, WHILEM5);
3732 /*** all unsaved local vars undefined at this point */
3734 regcpblow(st->u.whilem.cp);
3737 REGCP_UNWIND(st->u.whilem.lastcp);
3738 regcppop(rex); /* Restore some previous $<digit>s? */
3739 PL_reginput = locinput;
3741 PerlIO_printf(Perl_debug_log,
3742 "%*s failed, try continuation...\n",
3743 REPORT_CODE_OFF+PL_regindent*2, "")
3746 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3747 && !(PL_reg_flags & RF_warned)) {
3748 PL_reg_flags |= RF_warned;
3749 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3750 "Complex regular subexpression recursion",
3754 /* Failed deeper matches of scan, so see if this one works. */
3755 st->u.whilem.savecc = st->cc;
3756 st->cc = st->cc->u.curlyx.outercc;
3758 st->ln = st->cc->u.curlyx.cur;
3759 REGMATCH(st->u.whilem.savecc->next, WHILEM6);
3760 /*** all unsaved local vars undefined at this point */
3761 st->cc = st->u.whilem.savecc;
3764 if (st->cc->u.curlyx.outercc)
3765 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3766 st->cc->u.curlyx.cur = n - 1;
3767 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3772 next = scan + ARG(scan);
3775 inner = NEXTOPER(NEXTOPER(scan));
3778 inner = NEXTOPER(scan);
3783 if (!next || OP(next) != type) /* No choice. */
3784 next = inner; /* Avoid recursion. */
3786 const I32 lastparen = *PL_reglastparen;
3787 /* Put unwinding data on stack */
3788 const I32 unwind1 = SSNEWt(1,re_unwind_branch_t);
3789 re_unwind_branch_t * const uw = SSPTRt(unwind1,re_unwind_branch_t);
3791 uw->prev = st->unwind;
3792 st->unwind = unwind1;
3793 uw->type = ((type == BRANCH)
3795 : RE_UNWIND_BRANCHJ);
3796 uw->lastparen = lastparen;
3798 uw->locinput = locinput;
3799 uw->nextchr = nextchr;
3800 uw->minmod = st->minmod;
3802 uw->regindent = ++PL_regindent;
3805 REGCP_SET(uw->lastcp);
3807 /* Now go into the first branch */
3817 st->u.curlym.l = st->u.curlym.matches = 0;
3819 /* We suppose that the next guy does not need
3820 backtracking: in particular, it is of constant non-zero length,
3821 and has no parenths to influence future backrefs. */
3822 st->ln = ARG1(scan); /* min to match */
3823 n = ARG2(scan); /* max to match */
3824 st->u.curlym.paren = scan->flags;
3825 if (st->u.curlym.paren) {
3826 if (st->u.curlym.paren > PL_regsize)
3827 PL_regsize = st->u.curlym.paren;
3828 if (st->u.curlym.paren > (I32)*PL_reglastparen)
3829 *PL_reglastparen = st->u.curlym.paren;
3831 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3832 if (st->u.curlym.paren)
3833 scan += NEXT_OFF(scan); /* Skip former OPEN. */
3834 PL_reginput = locinput;
3835 st->u.curlym.maxwanted = st->minmod ? st->ln : n;
3836 if (st->u.curlym.maxwanted) {
3837 while (PL_reginput < PL_regeol && st->u.curlym.matches < st->u.curlym.maxwanted) {
3838 REGMATCH(scan, CURLYM1);
3839 /*** all unsaved local vars undefined at this point */
3842 /* on first match, determine length, u.curlym.l */
3843 if (!st->u.curlym.matches++) {
3844 if (PL_reg_match_utf8) {
3846 while (s < PL_reginput) {
3852 st->u.curlym.l = PL_reginput - locinput;
3854 if (st->u.curlym.l == 0) {
3855 st->u.curlym.matches = st->u.curlym.maxwanted;
3859 locinput = PL_reginput;
3863 PL_reginput = locinput;
3867 if (st->ln && st->u.curlym.matches < st->ln)
3869 if (HAS_TEXT(next) || JUMPABLE(next)) {
3870 regnode *text_node = next;
3872 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3874 if (! HAS_TEXT(text_node)) st->u.curlym.c1 = st->u.curlym.c2 = -1000;
3876 if (PL_regkind[(U8)OP(text_node)] == REF) {
3877 st->u.curlym.c1 = st->u.curlym.c2 = -1000;
3880 else { st->u.curlym.c1 = (U8)*STRING(text_node); }
3881 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3882 st->u.curlym.c2 = PL_fold[st->u.curlym.c1];
3883 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3884 st->u.curlym.c2 = PL_fold_locale[st->u.curlym.c1];
3886 st->u.curlym.c2 = st->u.curlym.c1;
3890 st->u.curlym.c1 = st->u.curlym.c2 = -1000;
3892 REGCP_SET(st->u.curlym.lastcp);
3893 while (n >= st->ln || (n == REG_INFTY && st->ln > 0)) { /* ln overflow ? */
3894 /* If it could work, try it. */
3895 if (st->u.curlym.c1 == -1000 ||
3896 UCHARAT(PL_reginput) == st->u.curlym.c1 ||
3897 UCHARAT(PL_reginput) == st->u.curlym.c2)
3899 if (st->u.curlym.paren) {