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, NULL);
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... */
934 /* if reginfo is NULL, its a dryrun */
937 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, const char
938 *strend, const regmatch_info *reginfo)
941 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
945 register STRLEN uskip;
949 register I32 tmp = 1; /* Scratch variable? */
950 register const bool do_utf8 = PL_reg_match_utf8;
952 /* We know what class it must start with. */
956 while (s + (uskip = UTF8SKIP(s)) <= strend) {
957 if ((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
958 !UTF8_IS_INVARIANT((U8)s[0]) ?
959 reginclass(prog, c, (U8*)s, 0, do_utf8) :
960 REGINCLASS(prog, c, (U8*)s)) {
961 if (tmp && (!reginfo || regtry(reginfo, s)))
975 if (REGINCLASS(prog, c, (U8*)s) ||
976 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
977 /* The assignment of 2 is intentional:
978 * for the folded sharp s, the skip is 2. */
979 (skip = SHARP_S_SKIP))) {
980 if (tmp && (!reginfo || regtry(reginfo, s)))
993 if (tmp && (!reginfo || regtry(reginfo, s)))
1002 ln = STR_LEN(c); /* length to match in octets/bytes */
1003 lnc = (I32) ln; /* length to match in characters */
1005 STRLEN ulen1, ulen2;
1007 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1008 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1009 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1011 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1012 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1014 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1016 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1019 while (sm < ((U8 *) m + ln)) {
1034 c2 = PL_fold_locale[c1];
1036 e = HOP3c(strend, -((I32)lnc), s);
1038 if (!reginfo && e < s)
1039 e = s; /* Due to minlen logic of intuit() */
1041 /* The idea in the EXACTF* cases is to first find the
1042 * first character of the EXACTF* node and then, if
1043 * necessary, case-insensitively compare the full
1044 * text of the node. The c1 and c2 are the first
1045 * characters (though in Unicode it gets a bit
1046 * more complicated because there are more cases
1047 * than just upper and lower: one needs to use
1048 * the so-called folding case for case-insensitive
1049 * matching (called "loose matching" in Unicode).
1050 * ibcmp_utf8() will do just that. */
1054 U8 tmpbuf [UTF8_MAXBYTES+1];
1055 STRLEN len, foldlen;
1056 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1058 /* Upper and lower of 1st char are equal -
1059 * probably not a "letter". */
1061 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1065 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1066 m, (char **)0, ln, (bool)UTF))
1067 && (!reginfo || regtry(reginfo, s)) )
1070 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1071 uvchr_to_utf8(tmpbuf, c);
1072 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1074 && (f == c1 || f == c2)
1075 && (ln == foldlen ||
1076 !ibcmp_utf8((char *) foldbuf,
1077 (char **)0, foldlen, do_utf8,
1079 (char **)0, ln, (bool)UTF))
1080 && (!reginfo || regtry(reginfo, s)) )
1088 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1091 /* Handle some of the three Greek sigmas cases.
1092 * Note that not all the possible combinations
1093 * are handled here: some of them are handled
1094 * by the standard folding rules, and some of
1095 * them (the character class or ANYOF cases)
1096 * are handled during compiletime in
1097 * regexec.c:S_regclass(). */
1098 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1099 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1100 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1102 if ( (c == c1 || c == c2)
1104 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1105 m, (char **)0, ln, (bool)UTF))
1106 && (!reginfo || regtry(reginfo, s)) )
1109 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1110 uvchr_to_utf8(tmpbuf, c);
1111 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1113 && (f == c1 || f == c2)
1114 && (ln == foldlen ||
1115 !ibcmp_utf8((char *) foldbuf,
1116 (char **)0, foldlen, do_utf8,
1118 (char **)0, ln, (bool)UTF))
1119 && (!reginfo || regtry(reginfo, s)) )
1130 && (ln == 1 || !(OP(c) == EXACTF
1132 : ibcmp_locale(s, m, ln)))
1133 && (!reginfo || regtry(reginfo, s)) )
1139 if ( (*(U8*)s == c1 || *(U8*)s == c2)
1140 && (ln == 1 || !(OP(c) == EXACTF
1142 : ibcmp_locale(s, m, ln)))
1143 && (!reginfo || regtry(reginfo, s)) )
1150 PL_reg_flags |= RF_tainted;
1157 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1158 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1160 tmp = ((OP(c) == BOUND ?
1161 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1162 LOAD_UTF8_CHARCLASS_ALNUM();
1163 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1164 if (tmp == !(OP(c) == BOUND ?
1165 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1166 isALNUM_LC_utf8((U8*)s)))
1169 if ((!reginfo || regtry(reginfo, s)))
1176 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1177 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1178 while (s < strend) {
1180 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1182 if ((!reginfo || regtry(reginfo, s)))
1188 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, s)))
1192 PL_reg_flags |= RF_tainted;
1199 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1200 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1202 tmp = ((OP(c) == NBOUND ?
1203 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1204 LOAD_UTF8_CHARCLASS_ALNUM();
1205 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1206 if (tmp == !(OP(c) == NBOUND ?
1207 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1208 isALNUM_LC_utf8((U8*)s)))
1210 else if ((!reginfo || regtry(reginfo, s)))
1216 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1217 tmp = ((OP(c) == NBOUND ?
1218 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1219 while (s < strend) {
1221 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1223 else if ((!reginfo || regtry(reginfo, s)))
1228 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, s)))
1233 LOAD_UTF8_CHARCLASS_ALNUM();
1234 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1235 if (swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1236 if (tmp && (!reginfo || regtry(reginfo, s)))
1247 while (s < strend) {
1249 if (tmp && (!reginfo || regtry(reginfo, s)))
1261 PL_reg_flags |= RF_tainted;
1263 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1264 if (isALNUM_LC_utf8((U8*)s)) {
1265 if (tmp && (!reginfo || regtry(reginfo, s)))
1276 while (s < strend) {
1277 if (isALNUM_LC(*s)) {
1278 if (tmp && (!reginfo || regtry(reginfo, s)))
1291 LOAD_UTF8_CHARCLASS_ALNUM();
1292 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1293 if (!swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1294 if (tmp && (!reginfo || regtry(reginfo, s)))
1305 while (s < strend) {
1307 if (tmp && (!reginfo || regtry(reginfo, s)))
1319 PL_reg_flags |= RF_tainted;
1321 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1322 if (!isALNUM_LC_utf8((U8*)s)) {
1323 if (tmp && (!reginfo || regtry(reginfo, s)))
1334 while (s < strend) {
1335 if (!isALNUM_LC(*s)) {
1336 if (tmp && (!reginfo || regtry(reginfo, s)))
1349 LOAD_UTF8_CHARCLASS_SPACE();
1350 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1351 if (*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)) {
1352 if (tmp && (!reginfo || regtry(reginfo, s)))
1363 while (s < strend) {
1365 if (tmp && (!reginfo || regtry(reginfo, s)))
1377 PL_reg_flags |= RF_tainted;
1379 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1380 if (*s == ' ' || isSPACE_LC_utf8((U8*)s)) {
1381 if (tmp && (!reginfo || regtry(reginfo, s)))
1392 while (s < strend) {
1393 if (isSPACE_LC(*s)) {
1394 if (tmp && (!reginfo || regtry(reginfo, s)))
1407 LOAD_UTF8_CHARCLASS_SPACE();
1408 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1409 if (!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8))) {
1410 if (tmp && (!reginfo || regtry(reginfo, s)))
1421 while (s < strend) {
1423 if (tmp && (!reginfo || regtry(reginfo, s)))
1435 PL_reg_flags |= RF_tainted;
1437 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1438 if (!(*s == ' ' || isSPACE_LC_utf8((U8*)s))) {
1439 if (tmp && (!reginfo || regtry(reginfo, s)))
1450 while (s < strend) {
1451 if (!isSPACE_LC(*s)) {
1452 if (tmp && (!reginfo || regtry(reginfo, s)))
1465 LOAD_UTF8_CHARCLASS_DIGIT();
1466 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1467 if (swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1468 if (tmp && (!reginfo || regtry(reginfo, s)))
1479 while (s < strend) {
1481 if (tmp && (!reginfo || regtry(reginfo, s)))
1493 PL_reg_flags |= RF_tainted;
1495 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1496 if (isDIGIT_LC_utf8((U8*)s)) {
1497 if (tmp && (!reginfo || regtry(reginfo, s)))
1508 while (s < strend) {
1509 if (isDIGIT_LC(*s)) {
1510 if (tmp && (!reginfo || regtry(reginfo, s)))
1523 LOAD_UTF8_CHARCLASS_DIGIT();
1524 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1525 if (!swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1526 if (tmp && (!reginfo || regtry(reginfo, s)))
1537 while (s < strend) {
1539 if (tmp && (!reginfo || regtry(reginfo, s)))
1551 PL_reg_flags |= RF_tainted;
1553 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1554 if (!isDIGIT_LC_utf8((U8*)s)) {
1555 if (tmp && (!reginfo || regtry(reginfo, s)))
1566 while (s < strend) {
1567 if (!isDIGIT_LC(*s)) {
1568 if (tmp && (!reginfo || regtry(reginfo, s)))
1580 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1589 - regexec_flags - match a regexp against a string
1592 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1593 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1594 /* strend: pointer to null at end of string */
1595 /* strbeg: real beginning of string */
1596 /* minend: end of match must be >=minend after stringarg. */
1597 /* data: May be used for some additional optimizations. */
1598 /* nosave: For optimizations. */
1602 register regnode *c;
1603 register char *startpos = stringarg;
1604 I32 minlen; /* must match at least this many chars */
1605 I32 dontbother = 0; /* how many characters not to try at end */
1606 I32 end_shift = 0; /* Same for the end. */ /* CC */
1607 I32 scream_pos = -1; /* Internal iterator of scream. */
1608 char *scream_olds = NULL;
1609 SV* oreplsv = GvSV(PL_replgv);
1610 const bool do_utf8 = DO_UTF8(sv);
1616 regmatch_info reginfo; /* create some info to pass to regtry etc */
1618 GET_RE_DEBUG_FLAGS_DECL;
1620 PERL_UNUSED_ARG(data);
1622 /* Be paranoid... */
1623 if (prog == NULL || startpos == NULL) {
1624 Perl_croak(aTHX_ "NULL regexp parameter");
1628 multiline = prog->reganch & PMf_MULTILINE;
1629 reginfo.prog = prog;
1632 dsv0 = PERL_DEBUG_PAD_ZERO(0);
1633 dsv1 = PERL_DEBUG_PAD_ZERO(1);
1636 RX_MATCH_UTF8_set(prog, do_utf8);
1638 minlen = prog->minlen;
1639 if (strend - startpos < minlen) {
1640 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1641 "String too short [regexec_flags]...\n"));
1645 /* Check validity of program. */
1646 if (UCHARAT(prog->program) != REG_MAGIC) {
1647 Perl_croak(aTHX_ "corrupted regexp program");
1651 PL_reg_eval_set = 0;
1654 if (prog->reganch & ROPT_UTF8)
1655 PL_reg_flags |= RF_utf8;
1657 /* Mark beginning of line for ^ and lookbehind. */
1658 reginfo.bol = startpos; /* XXX not used ??? */
1662 /* Mark end of line for $ (and such) */
1665 /* see how far we have to get to not match where we matched before */
1666 reginfo.till = startpos+minend;
1668 /* If there is a "must appear" string, look for it. */
1671 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to set reginfo->ganch */
1674 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1675 reginfo.ganch = startpos;
1676 else if (sv && SvTYPE(sv) >= SVt_PVMG
1678 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1679 && mg->mg_len >= 0) {
1680 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
1681 if (prog->reganch & ROPT_ANCH_GPOS) {
1682 if (s > reginfo.ganch)
1687 else /* pos() not defined */
1688 reginfo.ganch = strbeg;
1691 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
1692 re_scream_pos_data d;
1694 d.scream_olds = &scream_olds;
1695 d.scream_pos = &scream_pos;
1696 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1698 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1699 goto phooey; /* not present */
1704 const char * const s0 = UTF
1705 ? pv_uni_display(dsv0, (U8*)prog->precomp, prog->prelen, 60,
1708 const int len0 = UTF ? SvCUR(dsv0) : prog->prelen;
1709 const char * const s1 = do_utf8 ? sv_uni_display(dsv1, sv, 60,
1710 UNI_DISPLAY_REGEX) : startpos;
1711 const int len1 = do_utf8 ? SvCUR(dsv1) : strend - startpos;
1714 PerlIO_printf(Perl_debug_log,
1715 "%sMatching REx%s \"%s%*.*s%s%s\" against \"%s%.*s%s%s\"\n",
1716 PL_colors[4], PL_colors[5], PL_colors[0],
1719 len0 > 60 ? "..." : "",
1721 (int)(len1 > 60 ? 60 : len1),
1723 (len1 > 60 ? "..." : "")
1727 /* Simplest case: anchored match need be tried only once. */
1728 /* [unless only anchor is BOL and multiline is set] */
1729 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1730 if (s == startpos && regtry(®info, startpos))
1732 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1733 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1738 dontbother = minlen - 1;
1739 end = HOP3c(strend, -dontbother, strbeg) - 1;
1740 /* for multiline we only have to try after newlines */
1741 if (prog->check_substr || prog->check_utf8) {
1745 if (regtry(®info, s))
1750 if (prog->reganch & RE_USE_INTUIT) {
1751 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1762 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1763 if (regtry(®info, s))
1770 } else if (prog->reganch & ROPT_ANCH_GPOS) {
1771 if (regtry(®info, reginfo.ganch))
1776 /* Messy cases: unanchored match. */
1777 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1778 /* we have /x+whatever/ */
1779 /* it must be a one character string (XXXX Except UTF?) */
1784 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1785 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1786 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1789 while (s < strend) {
1791 DEBUG_EXECUTE_r( did_match = 1 );
1792 if (regtry(®info, s)) goto got_it;
1794 while (s < strend && *s == ch)
1801 while (s < strend) {
1803 DEBUG_EXECUTE_r( did_match = 1 );
1804 if (regtry(®info, s)) goto got_it;
1806 while (s < strend && *s == ch)
1812 DEBUG_EXECUTE_r(if (!did_match)
1813 PerlIO_printf(Perl_debug_log,
1814 "Did not find anchored character...\n")
1817 else if (prog->anchored_substr != NULL
1818 || prog->anchored_utf8 != NULL
1819 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
1820 && prog->float_max_offset < strend - s)) {
1825 char *last1; /* Last position checked before */
1829 if (prog->anchored_substr || prog->anchored_utf8) {
1830 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1831 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1832 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1833 back_max = back_min = prog->anchored_offset;
1835 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1836 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1837 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1838 back_max = prog->float_max_offset;
1839 back_min = prog->float_min_offset;
1841 if (must == &PL_sv_undef)
1842 /* could not downgrade utf8 check substring, so must fail */
1845 last = HOP3c(strend, /* Cannot start after this */
1846 -(I32)(CHR_SVLEN(must)
1847 - (SvTAIL(must) != 0) + back_min), strbeg);
1850 last1 = HOPc(s, -1);
1852 last1 = s - 1; /* bogus */
1854 /* XXXX check_substr already used to find "s", can optimize if
1855 check_substr==must. */
1857 dontbother = end_shift;
1858 strend = HOPc(strend, -dontbother);
1859 while ( (s <= last) &&
1860 ((flags & REXEC_SCREAM)
1861 ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg,
1862 end_shift, &scream_pos, 0))
1863 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend),
1864 (unsigned char*)strend, must,
1865 multiline ? FBMrf_MULTILINE : 0))) ) {
1866 /* we may be pointing at the wrong string */
1867 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1868 s = strbeg + (s - SvPVX_const(sv));
1869 DEBUG_EXECUTE_r( did_match = 1 );
1870 if (HOPc(s, -back_max) > last1) {
1871 last1 = HOPc(s, -back_min);
1872 s = HOPc(s, -back_max);
1875 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1877 last1 = HOPc(s, -back_min);
1881 while (s <= last1) {
1882 if (regtry(®info, s))
1888 while (s <= last1) {
1889 if (regtry(®info, s))
1895 DEBUG_EXECUTE_r(if (!did_match)
1896 PerlIO_printf(Perl_debug_log,
1897 "Did not find %s substr \"%s%.*s%s\"%s...\n",
1898 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1899 ? "anchored" : "floating"),
1901 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
1903 PL_colors[1], (SvTAIL(must) ? "$" : ""))
1907 else if ((c = prog->regstclass)) {
1909 I32 op = (U8)OP(prog->regstclass);
1910 /* don't bother with what can't match */
1911 if (PL_regkind[op] != EXACT && op != CANY)
1912 strend = HOPc(strend, -(minlen - 1));
1915 SV *prop = sv_newmortal();
1921 regprop(prog, prop, c);
1923 pv_uni_display(dsv0, (U8*)SvPVX_const(prop), SvCUR(prop), 60,
1924 UNI_DISPLAY_REGEX) :
1926 len0 = UTF ? SvCUR(dsv0) : SvCUR(prop);
1928 sv_uni_display(dsv1, sv, 60, UNI_DISPLAY_REGEX) : s;
1929 len1 = UTF ? SvCUR(dsv1) : strend - s;
1930 PerlIO_printf(Perl_debug_log,
1931 "Matching stclass \"%*.*s\" against \"%*.*s\"\n",
1935 if (find_byclass(prog, c, s, strend, ®info))
1937 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass...\n"));
1941 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
1946 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1947 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1948 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1950 if (flags & REXEC_SCREAM) {
1951 last = screaminstr(sv, float_real, s - strbeg,
1952 end_shift, &scream_pos, 1); /* last one */
1954 last = scream_olds; /* Only one occurrence. */
1955 /* we may be pointing at the wrong string */
1956 else if (RX_MATCH_COPIED(prog))
1957 s = strbeg + (s - SvPVX_const(sv));
1961 const char * const little = SvPV_const(float_real, len);
1963 if (SvTAIL(float_real)) {
1964 if (memEQ(strend - len + 1, little, len - 1))
1965 last = strend - len + 1;
1966 else if (!multiline)
1967 last = memEQ(strend - len, little, len)
1968 ? strend - len : NULL;
1974 last = rninstr(s, strend, little, little + len);
1976 last = strend; /* matching "$" */
1980 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1981 "%sCan't trim the tail, match fails (should not happen)%s\n",
1982 PL_colors[4], PL_colors[5]));
1983 goto phooey; /* Should not happen! */
1985 dontbother = strend - last + prog->float_min_offset;
1987 if (minlen && (dontbother < minlen))
1988 dontbother = minlen - 1;
1989 strend -= dontbother; /* this one's always in bytes! */
1990 /* We don't know much -- general case. */
1993 if (regtry(®info, s))
2002 if (regtry(®info, s))
2004 } while (s++ < strend);
2012 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2014 if (PL_reg_eval_set) {
2015 /* Preserve the current value of $^R */
2016 if (oreplsv != GvSV(PL_replgv))
2017 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2018 restored, the value remains
2020 restore_pos(aTHX_ prog);
2023 /* make sure $`, $&, $', and $digit will work later */
2024 if ( !(flags & REXEC_NOT_FIRST) ) {
2025 RX_MATCH_COPY_FREE(prog);
2026 if (flags & REXEC_COPY_STR) {
2027 I32 i = PL_regeol - startpos + (stringarg - strbeg);
2028 #ifdef PERL_OLD_COPY_ON_WRITE
2030 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2032 PerlIO_printf(Perl_debug_log,
2033 "Copy on write: regexp capture, type %d\n",
2036 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2037 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2038 assert (SvPOKp(prog->saved_copy));
2042 RX_MATCH_COPIED_on(prog);
2043 s = savepvn(strbeg, i);
2049 prog->subbeg = strbeg;
2050 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2057 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2058 PL_colors[4], PL_colors[5]));
2059 if (PL_reg_eval_set)
2060 restore_pos(aTHX_ prog);
2065 - regtry - try match at specific point
2067 STATIC I32 /* 0 failure, 1 success */
2068 S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos)
2074 regexp *prog = reginfo->prog;
2075 GET_RE_DEBUG_FLAGS_DECL;
2078 PL_regindent = 0; /* XXXX Not good when matches are reenterable... */
2080 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2083 PL_reg_eval_set = RS_init;
2084 DEBUG_EXECUTE_r(DEBUG_s(
2085 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2086 (IV)(PL_stack_sp - PL_stack_base));
2088 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2089 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2090 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2092 /* Apparently this is not needed, judging by wantarray. */
2093 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2094 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2097 /* Make $_ available to executed code. */
2098 if (reginfo->sv != DEFSV) {
2100 DEFSV = reginfo->sv;
2103 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2104 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2105 /* prepare for quick setting of pos */
2106 #ifdef PERL_OLD_COPY_ON_WRITE
2108 sv_force_normal_flags(sv, 0);
2110 mg = sv_magicext(reginfo->sv, (SV*)0, PERL_MAGIC_regex_global,
2111 &PL_vtbl_mglob, NULL, 0);
2115 PL_reg_oldpos = mg->mg_len;
2116 SAVEDESTRUCTOR_X(restore_pos, prog);
2118 if (!PL_reg_curpm) {
2119 Newxz(PL_reg_curpm, 1, PMOP);
2122 SV* repointer = newSViv(0);
2123 /* so we know which PL_regex_padav element is PL_reg_curpm */
2124 SvFLAGS(repointer) |= SVf_BREAK;
2125 av_push(PL_regex_padav,repointer);
2126 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2127 PL_regex_pad = AvARRAY(PL_regex_padav);
2131 PM_SETRE(PL_reg_curpm, prog);
2132 PL_reg_oldcurpm = PL_curpm;
2133 PL_curpm = PL_reg_curpm;
2134 if (RX_MATCH_COPIED(prog)) {
2135 /* Here is a serious problem: we cannot rewrite subbeg,
2136 since it may be needed if this match fails. Thus
2137 $` inside (?{}) could fail... */
2138 PL_reg_oldsaved = prog->subbeg;
2139 PL_reg_oldsavedlen = prog->sublen;
2140 #ifdef PERL_OLD_COPY_ON_WRITE
2141 PL_nrs = prog->saved_copy;
2143 RX_MATCH_COPIED_off(prog);
2146 PL_reg_oldsaved = NULL;
2147 prog->subbeg = PL_bostr;
2148 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2150 prog->startp[0] = startpos - PL_bostr;
2151 PL_reginput = startpos;
2152 PL_regstartp = prog->startp;
2153 PL_regendp = prog->endp;
2154 PL_reglastparen = &prog->lastparen;
2155 PL_reglastcloseparen = &prog->lastcloseparen;
2156 prog->lastparen = 0;
2157 prog->lastcloseparen = 0;
2159 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2160 if (PL_reg_start_tmpl <= prog->nparens) {
2161 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2162 if(PL_reg_start_tmp)
2163 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2165 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2168 /* XXXX What this code is doing here?!!! There should be no need
2169 to do this again and again, PL_reglastparen should take care of
2172 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2173 * Actually, the code in regcppop() (which Ilya may be meaning by
2174 * PL_reglastparen), is not needed at all by the test suite
2175 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2176 * enough, for building DynaLoader, or otherwise this
2177 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2178 * will happen. Meanwhile, this code *is* needed for the
2179 * above-mentioned test suite tests to succeed. The common theme
2180 * on those tests seems to be returning null fields from matches.
2185 if (prog->nparens) {
2187 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2194 if (regmatch(reginfo, prog->program + 1)) {
2195 prog->endp[0] = PL_reginput - PL_bostr;
2198 REGCP_UNWIND(lastcp);
2202 #define RE_UNWIND_BRANCH 1
2203 #define RE_UNWIND_BRANCHJ 2
2207 typedef struct { /* XX: makes sense to enlarge it... */
2211 } re_unwind_generic_t;
2225 } re_unwind_branch_t;
2227 typedef union re_unwind_t {
2229 re_unwind_generic_t generic;
2230 re_unwind_branch_t branch;
2233 #define sayYES goto yes
2234 #define sayNO goto no
2235 #define sayNO_ANYOF goto no_anyof
2236 #define sayYES_FINAL goto yes_final
2237 #define sayYES_LOUD goto yes_loud
2238 #define sayNO_FINAL goto no_final
2239 #define sayNO_SILENT goto do_no
2240 #define saySAME(x) if (x) goto yes; else goto no
2242 #define POSCACHE_SUCCESS 0 /* caching success rather than failure */
2243 #define POSCACHE_SEEN 1 /* we know what we're caching */
2244 #define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */
2246 #define CACHEsayYES STMT_START { \
2247 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \
2248 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) { \
2249 PL_reg_poscache[0] |= (1<<POSCACHE_SUCCESS) | (1<<POSCACHE_SEEN); \
2250 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2252 else if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS)) { \
2253 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2256 /* cache records failure, but this is success */ \
2258 PerlIO_printf(Perl_debug_log, \
2259 "%*s (remove success from failure cache)\n", \
2260 REPORT_CODE_OFF+PL_regindent*2, "") \
2262 PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<<st->u.whilem.cache_bit); \
2268 #define CACHEsayNO STMT_START { \
2269 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \
2270 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) { \
2271 PL_reg_poscache[0] |= (1<<POSCACHE_SEEN); \
2272 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2274 else if (!(PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2275 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2278 /* cache records success, but this is failure */ \
2280 PerlIO_printf(Perl_debug_log, \
2281 "%*s (remove failure from success cache)\n", \
2282 REPORT_CODE_OFF+PL_regindent*2, "") \
2284 PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<<st->u.whilem.cache_bit); \
2290 /* this is used to determine how far from the left messages like
2291 'failed...' are printed. Currently 29 makes these messages line
2292 up with the opcode they refer to. Earlier perls used 25 which
2293 left these messages outdented making reviewing a debug output
2296 #define REPORT_CODE_OFF 29
2299 /* Make sure there is a test for this +1 options in re_tests */
2300 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2302 #define SLAB_FIRST(s) (&(s)->states[0])
2303 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2305 /* grab a new slab and return the first slot in it */
2307 STATIC regmatch_state *
2310 regmatch_slab *s = PL_regmatch_slab->next;
2312 Newx(s, 1, regmatch_slab);
2313 s->prev = PL_regmatch_slab;
2315 PL_regmatch_slab->next = s;
2317 PL_regmatch_slab = s;
2318 return SLAB_FIRST(s);
2321 /* simulate a recursive call to regmatch */
2323 #define REGMATCH(ns, where) \
2326 st->resume_state = resume_##where; \
2327 goto start_recurse; \
2328 resume_point_##where:
2331 /* push a new regex state. Set newst to point to it */
2333 #define PUSH_STATE(newst, resume) \
2335 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "PUSH STATE(%d)\n", depth)); \
2339 st->locinput = locinput; \
2340 st->resume_state = resume; \
2342 if (newst > SLAB_LAST(PL_regmatch_slab)) \
2343 newst = S_push_slab(aTHX); \
2344 PL_regmatch_state = newst; \
2346 newst->minmod = 0; \
2348 newst->logical = 0; \
2349 newst->unwind = 0; \
2350 locinput = PL_reginput; \
2351 nextchr = UCHARAT(locinput);
2354 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "POP STATE(%d)\n", depth)); \
2357 if (st < SLAB_FIRST(PL_regmatch_slab)) { \
2358 PL_regmatch_slab = PL_regmatch_slab->prev; \
2359 st = SLAB_LAST(PL_regmatch_slab); \
2361 PL_regmatch_state = st; \
2365 locinput = st->locinput; \
2366 nextchr = UCHARAT(locinput);
2369 - regmatch - main matching routine
2371 * Conceptually the strategy is simple: check to see whether the current
2372 * node matches, call self recursively to see whether the rest matches,
2373 * and then act accordingly. In practice we make some effort to avoid
2374 * recursion, in particular by going through "ordinary" nodes (that don't
2375 * need to know whether the rest of the match failed) by a loop instead of
2378 /* [lwall] I've hoisted the register declarations to the outer block in order to
2379 * maybe save a little bit of pushing and popping on the stack. It also takes
2380 * advantage of machines that use a register save mask on subroutine entry.
2382 * This function used to be heavily recursive, but since this had the
2383 * effect of blowing the CPU stack on complex regexes, it has been
2384 * restructured to be iterative, and to save state onto the heap rather
2385 * than the stack. Essentially whereever regmatch() used to be called, it
2386 * pushes the current state, notes where to return, then jumps back into
2389 * Originally the structure of this function used to look something like
2394 while (scan != NULL) {
2395 a++; // do stuff with a and b
2401 if (regmatch(...)) // recurse
2411 * Now it looks something like this:
2419 regmatch_state *st = new();
2421 st->a++; // do stuff with a and b
2423 while (scan != NULL) {
2431 st->resume_state = resume_FOO;
2432 goto start_recurse; // recurse
2441 st = new(); push a new state
2442 st->a = 1; st->b = 2;
2449 switch (resume_state) {
2451 goto resume_point_FOO;
2458 * WARNING: this means that any line in this function that contains a
2459 * REGMATCH() or TRYPAREN() is actually simulating a recursive call to
2460 * regmatch() using gotos instead. Thus the values of any local variables
2461 * not saved in the regmatch_state structure will have been lost when
2462 * execution resumes on the next line .
2464 * States (ie the st pointer) are allocated in slabs of about 4K in size.
2465 * PL_regmatch_state always points to the currently active state, and
2466 * PL_regmatch_slab points to the slab currently containing PL_regmatch_state.
2467 * The first time regmatch is called, the first slab is allocated, and is
2468 * never freed until interpreter desctruction. When the slab is full,
2469 * a new one is allocated chained to the end. At exit from regmatch, slabs
2470 * allocated since entry are freed.
2474 STATIC I32 /* 0 failure, 1 success */
2475 S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog)
2478 register const bool do_utf8 = PL_reg_match_utf8;
2479 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2481 regexp *rex = reginfo->prog;
2483 regmatch_slab *orig_slab;
2484 regmatch_state *orig_state;
2486 /* the current state. This is a cached copy of PL_regmatch_state */
2487 register regmatch_state *st;
2489 /* cache heavy used fields of st in registers */
2490 register regnode *scan;
2491 register regnode *next;
2492 register I32 n = 0; /* initialize to shut up compiler warning */
2493 register char *locinput = PL_reginput;
2495 /* these variables are NOT saved during a recusive RFEGMATCH: */
2496 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2497 bool result; /* return value of S_regmatch */
2498 regnode *inner; /* Next node in internal branch. */
2499 int depth = 0; /* depth of recursion */
2500 regmatch_state *newst; /* when pushing a state, this is the new one */
2501 regmatch_state *yes_state = NULL; /* state to pop to on success of
2505 SV *re_debug_flags = NULL;
2510 /* on first ever call to regmatch, allocate first slab */
2511 if (!PL_regmatch_slab) {
2512 Newx(PL_regmatch_slab, 1, regmatch_slab);
2513 PL_regmatch_slab->prev = NULL;
2514 PL_regmatch_slab->next = NULL;
2515 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2518 /* remember current high-water mark for exit */
2519 /* XXX this should be done with SAVE* instead */
2520 orig_slab = PL_regmatch_slab;
2521 orig_state = PL_regmatch_state;
2523 /* grab next free state slot */
2524 st = ++PL_regmatch_state;
2525 if (st > SLAB_LAST(PL_regmatch_slab))
2526 st = PL_regmatch_state = S_push_slab(aTHX);
2533 /* Note that nextchr is a byte even in UTF */
2534 nextchr = UCHARAT(locinput);
2536 while (scan != NULL) {
2539 SV * const prop = sv_newmortal();
2540 const int docolor = *PL_colors[0];
2541 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2542 int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput);
2543 /* The part of the string before starttry has one color
2544 (pref0_len chars), between starttry and current
2545 position another one (pref_len - pref0_len chars),
2546 after the current position the third one.
2547 We assume that pref0_len <= pref_len, otherwise we
2548 decrease pref0_len. */
2549 int pref_len = (locinput - PL_bostr) > (5 + taill) - l
2550 ? (5 + taill) - l : locinput - PL_bostr;
2553 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2555 pref0_len = pref_len - (locinput - PL_reg_starttry);
2556 if (l + pref_len < (5 + taill) && l < PL_regeol - locinput)
2557 l = ( PL_regeol - locinput > (5 + taill) - pref_len
2558 ? (5 + taill) - pref_len : PL_regeol - locinput);
2559 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2563 if (pref0_len > pref_len)
2564 pref0_len = pref_len;
2565 regprop(rex, prop, scan);
2567 const char * const s0 =
2568 do_utf8 && OP(scan) != CANY ?
2569 pv_uni_display(PERL_DEBUG_PAD(0), (U8*)(locinput - pref_len),
2570 pref0_len, 60, UNI_DISPLAY_REGEX) :
2571 locinput - pref_len;
2572 const int len0 = do_utf8 ? strlen(s0) : pref0_len;
2573 const char * const s1 = do_utf8 && OP(scan) != CANY ?
2574 pv_uni_display(PERL_DEBUG_PAD(1),
2575 (U8*)(locinput - pref_len + pref0_len),
2576 pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) :
2577 locinput - pref_len + pref0_len;
2578 const int len1 = do_utf8 ? strlen(s1) : pref_len - pref0_len;
2579 const char * const s2 = do_utf8 && OP(scan) != CANY ?
2580 pv_uni_display(PERL_DEBUG_PAD(2), (U8*)locinput,
2581 PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) :
2583 const int len2 = do_utf8 ? strlen(s2) : l;
2584 PerlIO_printf(Perl_debug_log,
2585 "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|%3"IVdf":%*s%s\n",
2586 (IV)(locinput - PL_bostr),
2593 (docolor ? "" : "> <"),
2597 15 - l - pref_len + 1,
2599 (IV)(scan - rex->program), PL_regindent*2, "",
2604 next = scan + NEXT_OFF(scan);
2610 if (locinput == PL_bostr)
2612 /* reginfo->till = reginfo->bol; */
2617 if (locinput == PL_bostr ||
2618 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2624 if (locinput == PL_bostr)
2628 if (locinput == reginfo->ganch)
2634 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2639 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2641 if (PL_regeol - locinput > 1)
2645 if (PL_regeol != locinput)
2649 if (!nextchr && locinput >= PL_regeol)
2652 locinput += PL_utf8skip[nextchr];
2653 if (locinput > PL_regeol)
2655 nextchr = UCHARAT(locinput);
2658 nextchr = UCHARAT(++locinput);
2661 if (!nextchr && locinput >= PL_regeol)
2663 nextchr = UCHARAT(++locinput);
2666 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2669 locinput += PL_utf8skip[nextchr];
2670 if (locinput > PL_regeol)
2672 nextchr = UCHARAT(locinput);
2675 nextchr = UCHARAT(++locinput);
2681 traverse the TRIE keeping track of all accepting states
2682 we transition through until we get to a failing node.
2690 U8 *uc = ( U8* )locinput;
2697 U8 *uscan = (U8*)NULL;
2699 SV *sv_accept_buff = NULL;
2700 const enum { trie_plain, trie_utf8, trie_uft8_fold }
2701 trie_type = do_utf8 ?
2702 (OP(scan) == TRIE ? trie_utf8 : trie_uft8_fold)
2705 /* what trie are we using right now */
2707 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2708 st->u.trie.accepted = 0; /* how many accepting states we have seen */
2711 while ( state && uc <= (U8*)PL_regeol ) {
2713 if (trie->states[ state ].wordnum) {
2714 if (!st->u.trie.accepted ) {
2717 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2718 sv_accept_buff=newSV(bufflen *
2719 sizeof(reg_trie_accepted) - 1);
2720 SvCUR_set(sv_accept_buff,
2721 sizeof(reg_trie_accepted));
2722 SvPOK_on(sv_accept_buff);
2723 sv_2mortal(sv_accept_buff);
2724 st->u.trie.accept_buff =
2725 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2728 if (st->u.trie.accepted >= bufflen) {
2730 st->u.trie.accept_buff =(reg_trie_accepted*)
2731 SvGROW(sv_accept_buff,
2732 bufflen * sizeof(reg_trie_accepted));
2734 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2735 + sizeof(reg_trie_accepted));
2737 st->u.trie.accept_buff[st->u.trie.accepted].wordnum = trie->states[state].wordnum;
2738 st->u.trie.accept_buff[st->u.trie.accepted].endpos = uc;
2739 ++st->u.trie.accepted;
2742 base = trie->states[ state ].trans.base;
2744 DEBUG_TRIE_EXECUTE_r(
2745 PerlIO_printf( Perl_debug_log,
2746 "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ",
2747 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2748 (UV)state, (UV)base, (UV)st->u.trie.accepted );
2752 switch (trie_type) {
2753 case trie_uft8_fold:
2755 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags );
2760 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2761 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2762 uvc = to_uni_fold( uvc, foldbuf, &foldlen );
2763 foldlen -= UNISKIP( uvc );
2764 uscan = foldbuf + UNISKIP( uvc );
2768 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN,
2777 charid = trie->charmap[ uvc ];
2781 if (trie->widecharmap) {
2782 SV** svpp = (SV**)NULL;
2783 svpp = hv_fetch(trie->widecharmap,
2784 (char*)&uvc, sizeof(UV), 0);
2786 charid = (U16)SvIV(*svpp);
2791 (base + charid > trie->uniquecharcount )
2792 && (base + charid - 1 - trie->uniquecharcount
2794 && trie->trans[base + charid - 1 -
2795 trie->uniquecharcount].check == state)
2797 state = trie->trans[base + charid - 1 -
2798 trie->uniquecharcount ].next;
2809 DEBUG_TRIE_EXECUTE_r(
2810 PerlIO_printf( Perl_debug_log,
2811 "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n",
2812 charid, uvc, (UV)state, PL_colors[5] );
2815 if (!st->u.trie.accepted )
2819 There was at least one accepting state that we
2820 transitioned through. Presumably the number of accepting
2821 states is going to be low, typically one or two. So we
2822 simply scan through to find the one with lowest wordnum.
2823 Once we find it, we swap the last state into its place
2824 and decrement the size. We then try to match the rest of
2825 the pattern at the point where the word ends, if we
2826 succeed then we end the loop, otherwise the loop
2827 eventually terminates once all of the accepting states
2831 if ( st->u.trie.accepted == 1 ) {
2833 SV ** const tmp = av_fetch( trie->words, st->u.trie.accept_buff[ 0 ].wordnum-1, 0 );
2834 PerlIO_printf( Perl_debug_log,
2835 "%*s %sonly one match : #%d <%s>%s\n",
2836 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2837 st->u.trie.accept_buff[ 0 ].wordnum,
2838 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2841 PL_reginput = (char *)st->u.trie.accept_buff[ 0 ].endpos;
2842 /* in this case we free tmps/leave before we call regmatch
2843 as we wont be using accept_buff again. */
2846 REGMATCH(scan + NEXT_OFF(scan), TRIE1);
2847 /*** all unsaved local vars undefined at this point */
2850 PerlIO_printf( Perl_debug_log,"%*s %sgot %"IVdf" possible matches%s\n",
2851 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], (IV)st->u.trie.accepted,
2854 while ( !result && st->u.trie.accepted-- ) {
2857 for( cur = 1 ; cur <= st->u.trie.accepted ; cur++ ) {
2858 DEBUG_TRIE_EXECUTE_r(
2859 PerlIO_printf( Perl_debug_log,
2860 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2861 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2862 (IV)best, st->u.trie.accept_buff[ best ].wordnum, (IV)cur,
2863 st->u.trie.accept_buff[ cur ].wordnum, PL_colors[5] );
2866 if (st->u.trie.accept_buff[cur].wordnum <
2867 st->u.trie.accept_buff[best].wordnum)
2871 reg_trie_data * const trie = (reg_trie_data*)
2872 rex->data->data[ARG(scan)];
2873 SV ** const tmp = av_fetch( trie->words, st->u.trie.accept_buff[ best ].wordnum - 1, 0 );
2874 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at 0x%p%s\n",
2875 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2876 st->u.trie.accept_buff[best].wordnum,
2877 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",scan,
2880 if ( best<st->u.trie.accepted ) {
2881 reg_trie_accepted tmp = st->u.trie.accept_buff[ best ];
2882 st->u.trie.accept_buff[ best ] = st->u.trie.accept_buff[ st->u.trie.accepted ];
2883 st->u.trie.accept_buff[ st->u.trie.accepted ] = tmp;
2884 best = st->u.trie.accepted;
2886 PL_reginput = (char *)st->u.trie.accept_buff[ best ].endpos;
2889 as far as I can tell we only need the SAVETMPS/FREETMPS
2890 for re's with EVAL in them but I'm leaving them in for
2891 all until I can be sure.
2894 REGMATCH(scan + NEXT_OFF(scan), TRIE2);
2895 /*** all unsaved local vars undefined at this point */
2908 /* unreached codepoint */
2910 char *s = STRING(scan);
2911 st->ln = STR_LEN(scan);
2912 if (do_utf8 != UTF) {
2913 /* The target and the pattern have differing utf8ness. */
2915 const char *e = s + st->ln;
2918 /* The target is utf8, the pattern is not utf8. */
2923 if (NATIVE_TO_UNI(*(U8*)s) !=
2924 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2932 /* The target is not utf8, the pattern is utf8. */
2937 if (NATIVE_TO_UNI(*((U8*)l)) !=
2938 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2946 nextchr = UCHARAT(locinput);
2949 /* The target and the pattern have the same utf8ness. */
2950 /* Inline the first character, for speed. */
2951 if (UCHARAT(s) != nextchr)
2953 if (PL_regeol - locinput < st->ln)
2955 if (st->ln > 1 && memNE(s, locinput, st->ln))
2958 nextchr = UCHARAT(locinput);
2962 PL_reg_flags |= RF_tainted;
2965 char *s = STRING(scan);
2966 st->ln = STR_LEN(scan);
2968 if (do_utf8 || UTF) {
2969 /* Either target or the pattern are utf8. */
2971 char *e = PL_regeol;
2973 if (ibcmp_utf8(s, 0, st->ln, (bool)UTF,
2974 l, &e, 0, do_utf8)) {
2975 /* One more case for the sharp s:
2976 * pack("U0U*", 0xDF) =~ /ss/i,
2977 * the 0xC3 0x9F are the UTF-8
2978 * byte sequence for the U+00DF. */
2980 toLOWER(s[0]) == 's' &&
2982 toLOWER(s[1]) == 's' &&
2989 nextchr = UCHARAT(locinput);
2993 /* Neither the target and the pattern are utf8. */
2995 /* Inline the first character, for speed. */
2996 if (UCHARAT(s) != nextchr &&
2997 UCHARAT(s) != ((OP(scan) == EXACTF)
2998 ? PL_fold : PL_fold_locale)[nextchr])
3000 if (PL_regeol - locinput < st->ln)
3002 if (st->ln > 1 && (OP(scan) == EXACTF
3003 ? ibcmp(s, locinput, st->ln)
3004 : ibcmp_locale(s, locinput, st->ln)))
3007 nextchr = UCHARAT(locinput);
3012 STRLEN inclasslen = PL_regeol - locinput;
3014 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3016 if (locinput >= PL_regeol)
3018 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3019 nextchr = UCHARAT(locinput);
3024 nextchr = UCHARAT(locinput);
3025 if (!REGINCLASS(rex, scan, (U8*)locinput))
3027 if (!nextchr && locinput >= PL_regeol)
3029 nextchr = UCHARAT(++locinput);
3033 /* If we might have the case of the German sharp s
3034 * in a casefolding Unicode character class. */
3036 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3037 locinput += SHARP_S_SKIP;
3038 nextchr = UCHARAT(locinput);
3044 PL_reg_flags |= RF_tainted;
3050 LOAD_UTF8_CHARCLASS_ALNUM();
3051 if (!(OP(scan) == ALNUM
3052 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3053 : isALNUM_LC_utf8((U8*)locinput)))
3057 locinput += PL_utf8skip[nextchr];
3058 nextchr = UCHARAT(locinput);
3061 if (!(OP(scan) == ALNUM
3062 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3064 nextchr = UCHARAT(++locinput);
3067 PL_reg_flags |= RF_tainted;
3070 if (!nextchr && locinput >= PL_regeol)
3073 LOAD_UTF8_CHARCLASS_ALNUM();
3074 if (OP(scan) == NALNUM
3075 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3076 : isALNUM_LC_utf8((U8*)locinput))
3080 locinput += PL_utf8skip[nextchr];
3081 nextchr = UCHARAT(locinput);
3084 if (OP(scan) == NALNUM
3085 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3087 nextchr = UCHARAT(++locinput);
3091 PL_reg_flags |= RF_tainted;
3095 /* was last char in word? */
3097 if (locinput == PL_bostr)
3100 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3102 st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3104 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3105 st->ln = isALNUM_uni(st->ln);
3106 LOAD_UTF8_CHARCLASS_ALNUM();
3107 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3110 st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln));
3111 n = isALNUM_LC_utf8((U8*)locinput);
3115 st->ln = (locinput != PL_bostr) ?
3116 UCHARAT(locinput - 1) : '\n';
3117 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3118 st->ln = isALNUM(st->ln);
3119 n = isALNUM(nextchr);
3122 st->ln = isALNUM_LC(st->ln);
3123 n = isALNUM_LC(nextchr);
3126 if (((!st->ln) == (!n)) == (OP(scan) == BOUND ||
3127 OP(scan) == BOUNDL))
3131 PL_reg_flags |= RF_tainted;
3137 if (UTF8_IS_CONTINUED(nextchr)) {
3138 LOAD_UTF8_CHARCLASS_SPACE();
3139 if (!(OP(scan) == SPACE
3140 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3141 : isSPACE_LC_utf8((U8*)locinput)))
3145 locinput += PL_utf8skip[nextchr];
3146 nextchr = UCHARAT(locinput);
3149 if (!(OP(scan) == SPACE
3150 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3152 nextchr = UCHARAT(++locinput);
3155 if (!(OP(scan) == SPACE
3156 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3158 nextchr = UCHARAT(++locinput);
3162 PL_reg_flags |= RF_tainted;
3165 if (!nextchr && locinput >= PL_regeol)
3168 LOAD_UTF8_CHARCLASS_SPACE();
3169 if (OP(scan) == NSPACE
3170 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3171 : isSPACE_LC_utf8((U8*)locinput))
3175 locinput += PL_utf8skip[nextchr];
3176 nextchr = UCHARAT(locinput);
3179 if (OP(scan) == NSPACE
3180 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3182 nextchr = UCHARAT(++locinput);
3185 PL_reg_flags |= RF_tainted;
3191 LOAD_UTF8_CHARCLASS_DIGIT();
3192 if (!(OP(scan) == DIGIT
3193 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3194 : isDIGIT_LC_utf8((U8*)locinput)))
3198 locinput += PL_utf8skip[nextchr];
3199 nextchr = UCHARAT(locinput);
3202 if (!(OP(scan) == DIGIT
3203 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3205 nextchr = UCHARAT(++locinput);
3208 PL_reg_flags |= RF_tainted;
3211 if (!nextchr && locinput >= PL_regeol)
3214 LOAD_UTF8_CHARCLASS_DIGIT();
3215 if (OP(scan) == NDIGIT
3216 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3217 : isDIGIT_LC_utf8((U8*)locinput))
3221 locinput += PL_utf8skip[nextchr];
3222 nextchr = UCHARAT(locinput);
3225 if (OP(scan) == NDIGIT
3226 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3228 nextchr = UCHARAT(++locinput);
3231 if (locinput >= PL_regeol)
3234 LOAD_UTF8_CHARCLASS_MARK();
3235 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3237 locinput += PL_utf8skip[nextchr];
3238 while (locinput < PL_regeol &&
3239 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3240 locinput += UTF8SKIP(locinput);
3241 if (locinput > PL_regeol)
3246 nextchr = UCHARAT(locinput);
3249 PL_reg_flags |= RF_tainted;
3254 n = ARG(scan); /* which paren pair */
3255 st->ln = PL_regstartp[n];
3256 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3257 if ((I32)*PL_reglastparen < n || st->ln == -1)
3258 sayNO; /* Do not match unless seen CLOSEn. */
3259 if (st->ln == PL_regendp[n])
3262 s = PL_bostr + st->ln;
3263 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3265 const char *e = PL_bostr + PL_regendp[n];
3267 * Note that we can't do the "other character" lookup trick as
3268 * in the 8-bit case (no pun intended) because in Unicode we
3269 * have to map both upper and title case to lower case.
3271 if (OP(scan) == REFF) {
3273 STRLEN ulen1, ulen2;
3274 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3275 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3279 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3280 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3281 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3288 nextchr = UCHARAT(locinput);
3292 /* Inline the first character, for speed. */
3293 if (UCHARAT(s) != nextchr &&
3295 (UCHARAT(s) != ((OP(scan) == REFF
3296 ? PL_fold : PL_fold_locale)[nextchr]))))
3298 st->ln = PL_regendp[n] - st->ln;
3299 if (locinput + st->ln > PL_regeol)
3301 if (st->ln > 1 && (OP(scan) == REF
3302 ? memNE(s, locinput, st->ln)
3304 ? ibcmp(s, locinput, st->ln)
3305 : ibcmp_locale(s, locinput, st->ln))))
3308 nextchr = UCHARAT(locinput);
3321 /* execute the code in the {...} */
3323 SV ** const before = SP;
3324 OP_4tree * const oop = PL_op;
3325 COP * const ocurcop = PL_curcop;
3329 PL_op = (OP_4tree*)rex->data->data[n];
3330 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3331 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3332 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3334 CALLRUNOPS(aTHX); /* Scalar context. */
3337 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3344 PAD_RESTORE_LOCAL(old_comppad);
3345 PL_curcop = ocurcop;
3348 sv_setsv(save_scalar(PL_replgv), ret);
3352 if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */
3355 /* extract RE object from returned value; compiling if
3360 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3361 mg = mg_find(sv, PERL_MAGIC_qr);
3362 else if (SvSMAGICAL(ret)) {
3363 if (SvGMAGICAL(ret))
3364 sv_unmagic(ret, PERL_MAGIC_qr);
3366 mg = mg_find(ret, PERL_MAGIC_qr);
3370 re = (regexp *)mg->mg_obj;
3371 (void)ReREFCNT_inc(re);
3375 const char * const t = SvPV_const(ret, len);
3377 const I32 osize = PL_regsize;
3380 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3381 re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm);
3383 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3385 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3391 /* run the pattern returned from (??{...}) */
3394 PerlIO_printf(Perl_debug_log,
3395 "Entering embedded \"%s%.60s%s%s\"\n",
3399 (strlen(re->precomp) > 60 ? "..." : ""))
3402 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
3403 REGCP_SET(st->u.eval.lastcp);
3404 *PL_reglastparen = 0;
3405 *PL_reglastcloseparen = 0;
3406 PL_reginput = locinput;
3408 /* XXXX This is too dramatic a measure... */
3412 st->u.eval.toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3413 ((re->reganch & ROPT_UTF8) != 0);
3414 if (st->u.eval.toggleutf) PL_reg_flags ^= RF_utf8;
3415 st->u.eval.prev_rex = rex;
3418 /* resume to current state on success */
3419 st->u.yes.prev_yes_state = yes_state;
3421 PUSH_STATE(newst, resume_EVAL);
3424 /* now continue from first node in postoned RE */
3425 next = re->program + 1;
3429 /* /(?(?{...})X|Y)/ */
3430 st->sw = SvTRUE(ret);
3435 n = ARG(scan); /* which paren pair */
3436 PL_reg_start_tmp[n] = locinput;
3441 n = ARG(scan); /* which paren pair */
3442 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3443 PL_regendp[n] = locinput - PL_bostr;
3444 if (n > (I32)*PL_reglastparen)
3445 *PL_reglastparen = n;
3446 *PL_reglastcloseparen = n;
3449 n = ARG(scan); /* which paren pair */
3450 st->sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3453 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3455 next = NEXTOPER(NEXTOPER(scan));
3457 next = scan + ARG(scan);
3458 if (OP(next) == IFTHEN) /* Fake one. */
3459 next = NEXTOPER(NEXTOPER(next));
3463 st->logical = scan->flags;
3465 /*******************************************************************
3466 cc points to the regmatch_state associated with the most recent CURLYX.
3467 This struct contains info about the innermost (...)* loop (an
3468 "infoblock"), and a pointer to the next outer cc.
3470 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3472 1) After matching Y, regnode for CURLYX is processed;
3474 2) This regnode populates cc, and calls regmatch() recursively
3475 with the starting point at WHILEM node;
3477 3) Each hit of WHILEM node tries to match A and Z (in the order
3478 depending on the current iteration, min/max of {min,max} and
3479 greediness). The information about where are nodes for "A"
3480 and "Z" is read from cc, as is info on how many times "A"
3481 was already matched, and greediness.
3483 4) After A matches, the same WHILEM node is hit again.
3485 5) Each time WHILEM is hit, cc is the infoblock created by CURLYX
3486 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3487 resets cc, since this Y(A)*Z can be a part of some other loop:
3488 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3489 of the external loop.
3491 Currently present infoblocks form a tree with a stem formed by st->cc
3492 and whatever it mentions via ->next, and additional attached trees
3493 corresponding to temporarily unset infoblocks as in "5" above.
3495 In the following picture, infoblocks for outer loop of
3496 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3497 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3498 infoblocks are drawn below the "reset" infoblock.
3500 In fact in the picture below we do not show failed matches for Z and T
3501 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3502 more obvious *why* one needs to *temporary* unset infoblocks.]
3504 Matched REx position InfoBlocks Comment
3508 Y A)*?Z)*?T x <- O <- I
3509 YA )*?Z)*?T x <- O <- I
3510 YA A)*?Z)*?T x <- O <- I
3511 YAA )*?Z)*?T x <- O <- I
3512 YAA Z)*?T x <- O # Temporary unset I
3515 YAAZ Y(A)*?Z)*?T x <- O
3518 YAAZY (A)*?Z)*?T x <- O
3521 YAAZY A)*?Z)*?T x <- O <- I
3524 YAAZYA )*?Z)*?T x <- O <- I
3527 YAAZYA Z)*?T x <- O # Temporary unset I
3533 YAAZYAZ T x # Temporary unset O
3540 *******************************************************************/
3543 /* No need to save/restore up to this paren */
3544 I32 parenfloor = scan->flags;
3548 CURLYX and WHILEM are always paired: they're the moral
3549 equivalent of pp_enteriter anbd pp_iter.
3551 The only time next could be null is if the node tree is
3552 corrupt. This was mentioned on p5p a few days ago.
3554 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3555 So we'll assert that this is true:
3558 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3560 /* XXXX Probably it is better to teach regpush to support
3561 parenfloor > PL_regsize... */
3562 if (parenfloor > (I32)*PL_reglastparen)
3563 parenfloor = *PL_reglastparen; /* Pessimization... */
3565 st->u.curlyx.cp = PL_savestack_ix;
3566 st->u.curlyx.outercc = st->cc;
3568 /* these fields contain the state of the current curly.
3569 * they are accessed by subsequent WHILEMs;
3570 * cur and lastloc are also updated by WHILEM */
3571 st->u.curlyx.parenfloor = parenfloor;
3572 st->u.curlyx.cur = -1; /* this will be updated by WHILEM */
3573 st->u.curlyx.min = ARG1(scan);
3574 st->u.curlyx.max = ARG2(scan);
3575 st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3576 st->u.curlyx.lastloc = 0;
3577 /* st->next and st->minmod are also read by WHILEM */
3579 PL_reginput = locinput;
3580 REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */
3581 /*** all unsaved local vars undefined at this point */
3582 regcpblow(st->u.curlyx.cp);
3583 st->cc = st->u.curlyx.outercc;
3589 * This is really hard to understand, because after we match
3590 * what we're trying to match, we must make sure the rest of
3591 * the REx is going to match for sure, and to do that we have
3592 * to go back UP the parse tree by recursing ever deeper. And
3593 * if it fails, we have to reset our parent's current state
3594 * that we can try again after backing off.
3599 st->cc gets initialised by CURLYX ready for use by WHILEM.
3600 So again, unless somethings been corrupted, st->cc cannot
3601 be null at that point in WHILEM.
3603 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3604 So we'll assert that this is true:
3607 st->u.whilem.lastloc = st->cc->u.curlyx.lastloc; /* Detection of 0-len. */
3608 st->u.whilem.cache_offset = 0;
3609 st->u.whilem.cache_bit = 0;
3611 n = st->cc->u.curlyx.cur + 1; /* how many we know we matched */
3612 PL_reginput = locinput;
3615 PerlIO_printf(Perl_debug_log,
3616 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3617 REPORT_CODE_OFF+PL_regindent*2, "",
3618 (long)n, (long)st->cc->u.curlyx.min,
3619 (long)st->cc->u.curlyx.max, PTR2UV(st->cc))
3622 /* If degenerate scan matches "", assume scan done. */
3624 if (locinput == st->cc->u.curlyx.lastloc && n >= st->cc->u.curlyx.min) {
3625 st->u.whilem.savecc = st->cc;
3626 st->cc = st->cc->u.curlyx.outercc;
3628 st->ln = st->cc->u.curlyx.cur;
3630 PerlIO_printf(Perl_debug_log,
3631 "%*s empty match detected, try continuation...\n",
3632 REPORT_CODE_OFF+PL_regindent*2, "")
3634 REGMATCH(st->u.whilem.savecc->next, WHILEM1);
3635 /*** all unsaved local vars undefined at this point */
3636 st->cc = st->u.whilem.savecc;
3639 if (st->cc->u.curlyx.outercc)
3640 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3644 /* First just match a string of min scans. */
3646 if (n < st->cc->u.curlyx.min) {
3647 st->cc->u.curlyx.cur = n;
3648 st->cc->u.curlyx.lastloc = locinput;
3649 REGMATCH(st->cc->u.curlyx.scan, WHILEM2);
3650 /*** all unsaved local vars undefined at this point */
3653 st->cc->u.curlyx.cur = n - 1;
3654 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3659 /* Check whether we already were at this position.
3660 Postpone detection until we know the match is not
3661 *that* much linear. */
3662 if (!PL_reg_maxiter) {
3663 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3664 PL_reg_leftiter = PL_reg_maxiter;
3666 if (PL_reg_leftiter-- == 0) {
3667 const I32 size = (PL_reg_maxiter + 7 + POSCACHE_START)/8;
3668 if (PL_reg_poscache) {
3669 if ((I32)PL_reg_poscache_size < size) {
3670 Renew(PL_reg_poscache, size, char);
3671 PL_reg_poscache_size = size;
3673 Zero(PL_reg_poscache, size, char);
3676 PL_reg_poscache_size = size;
3677 Newxz(PL_reg_poscache, size, char);
3680 PerlIO_printf(Perl_debug_log,
3681 "%sDetected a super-linear match, switching on caching%s...\n",
3682 PL_colors[4], PL_colors[5])
3685 if (PL_reg_leftiter < 0) {
3686 st->u.whilem.cache_offset = locinput - PL_bostr;
3688 st->u.whilem.cache_offset = (scan->flags & 0xf) - 1 + POSCACHE_START
3689 + st->u.whilem.cache_offset * (scan->flags>>4);
3690 st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8;
3691 st->u.whilem.cache_offset /= 8;
3692 if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<<st->u.whilem.cache_bit)) {
3694 PerlIO_printf(Perl_debug_log,
3695 "%*s already tried at this position...\n",
3696 REPORT_CODE_OFF+PL_regindent*2, "")
3698 if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))
3699 /* cache records success */
3702 /* cache records failure */
3708 /* Prefer next over scan for minimal matching. */
3710 if (st->cc->minmod) {
3711 st->u.whilem.savecc = st->cc;
3712 st->cc = st->cc->u.curlyx.outercc;
3714 st->ln = st->cc->u.curlyx.cur;
3715 st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor);
3716 REGCP_SET(st->u.whilem.lastcp);
3717 REGMATCH(st->u.whilem.savecc->next, WHILEM3);
3718 /*** all unsaved local vars undefined at this point */
3719 st->cc = st->u.whilem.savecc;
3721 regcpblow(st->u.whilem.cp);
3722 CACHEsayYES; /* All done. */
3724 REGCP_UNWIND(st->u.whilem.lastcp);
3726 if (st->cc->u.curlyx.outercc)
3727 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3729 if (n >= st->cc->u.curlyx.max) { /* Maximum greed exceeded? */
3730 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3731 && !(PL_reg_flags & RF_warned)) {
3732 PL_reg_flags |= RF_warned;
3733 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3734 "Complex regular subexpression recursion",
3741 PerlIO_printf(Perl_debug_log,
3742 "%*s trying longer...\n",
3743 REPORT_CODE_OFF+PL_regindent*2, "")
3745 /* Try scanning more and see if it helps. */
3746 PL_reginput = locinput;
3747 st->cc->u.curlyx.cur = n;
3748 st->cc->u.curlyx.lastloc = locinput;
3749 st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor);
3750 REGCP_SET(st->u.whilem.lastcp);
3751 REGMATCH(st->cc->u.curlyx.scan, WHILEM4);
3752 /*** all unsaved local vars undefined at this point */
3754 regcpblow(st->u.whilem.cp);
3757 REGCP_UNWIND(st->u.whilem.lastcp);
3759 st->cc->u.curlyx.cur = n - 1;
3760 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3764 /* Prefer scan over next for maximal matching. */
3766 if (n < st->cc->u.curlyx.max) { /* More greed allowed? */
3767 st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor);
3768 st->cc->u.curlyx.cur = n;
3769 st->cc->u.curlyx.lastloc = locinput;
3770 REGCP_SET(st->u.whilem.lastcp);
3771 REGMATCH(st->cc->u.curlyx.scan, WHILEM5);
3772 /*** all unsaved local vars undefined at this point */
3774 regcpblow(st->u.whilem.cp);
3777 REGCP_UNWIND(st->u.whilem.lastcp);
3778 regcppop(rex); /* Restore some previous $<digit>s? */
3779 PL_reginput = locinput;
3781 PerlIO_printf(Perl_debug_log,
3782 "%*s failed, try continuation...\n",
3783 REPORT_CODE_OFF+PL_regindent*2, "")
3786 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3787 && !(PL_reg_flags & RF_warned)) {
3788 PL_reg_flags |= RF_warned;
3789 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3790 "Complex regular subexpression recursion",
3794 /* Failed deeper matches of scan, so see if this one works. */
3795 st->u.whilem.savecc = st->cc;
3796 st->cc = st->cc->u.curlyx.outercc;
3798 st->ln = st->cc->u.curlyx.cur;
3799 REGMATCH(st->u.whilem.savecc->next, WHILEM6);
3800 /*** all unsaved local vars undefined at this point */
3801 st->cc = st->u.whilem.savecc;
3804 if (st->cc->u.curlyx.outercc)
3805 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3806 st->cc->u.curlyx.cur = n - 1;
3807 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3812 next = scan + ARG(scan);
3815 inner = NEXTOPER(NEXTOPER(scan));
3818 inner = NEXTOPER(scan);
3823 if (!next || OP(next) != type) /* No choice. */
3824 next = inner; /* Avoid recursion. */
3826 const I32 lastparen = *PL_reglastparen;
3827 /* Put unwinding data on stack */
3828 const I32 unwind1 = SSNEWt(1,re_unwind_branch_t);
3829 re_unwind_branch_t * const uw = SSPTRt(unwind1,re_unwind_branch_t);
3831 uw->prev = st->unwind;
3832 st->unwind = unwind1;
3833 uw->type = ((type == BRANCH)
3835 : RE_UNWIND_BRANCHJ);
3836 uw->lastparen = lastparen;
3838 uw->locinput = locinput;
3839 uw->nextchr = nextchr;
3840 uw->minmod = st->minmod;
3842 uw->regindent = ++PL_regindent;
3845 REGCP_SET(uw->lastcp);
3847 /* Now go into the first branch */
3857 st->u.curlym.l = st->u.curlym.matches = 0;
3859 /* We suppose that the next guy does not need
3860 backtracking: in particular, it is of constant non-zero length,
3861 and has no parenths to influence future backrefs. */
3862 st->ln = ARG1(scan); /* min to match */
3863 n = ARG2(scan); /* max to match */
3864 st->u.curlym.paren = scan->flags;
3865 if (st->u.curlym.paren) {
3866 if (st->u.curlym.paren > PL_regsize)
3867 PL_regsize = st->u.curlym.paren;
3868 if (st->u.curlym.paren > (I32)*PL_reglastparen)
3869 *PL_reglastparen = st->u.curlym.paren;
3871 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3872 if (st->u.curlym.paren)
3873 scan += NEXT_OFF(scan); /* Skip former OPEN. */
3874 PL_reginput = locinput;
3875 st->u.curlym.maxwanted = st->minmod ? st->ln : n;
3876 if (st->u.curlym.maxwanted) {
3877 while (PL_reginput < PL_regeol && st->u.curlym.matches < st->u.curlym.maxwanted) {
3878 REGMATCH(scan, CURLYM1);
3879 /*** all unsaved local vars undefined at this point */
3882 /* on first match, determine length, u.curlym.l */
3883 if (!st->u.curlym.matches++) {
3884 if (PL_reg_match_utf8) {
3886 while (s < PL_reginput) {
3892 st->u.curlym.l = PL_reginput - locinput;
3894 if (st->u.curlym.l == 0) {
3895 st->u.curlym.matches = st->u.curlym.maxwanted;
3899 locinput = PL_reginput;
3903 PL_reginput = locinput;