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 sayNO_FINAL goto no_final
2238 #define sayNO_SILENT goto do_no
2239 #define saySAME(x) if (x) goto yes; else goto no
2241 #define POSCACHE_SUCCESS 0 /* caching success rather than failure */
2242 #define POSCACHE_SEEN 1 /* we know what we're caching */
2243 #define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */
2245 #define CACHEsayYES STMT_START { \
2246 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \
2247 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) { \
2248 PL_reg_poscache[0] |= (1<<POSCACHE_SUCCESS) | (1<<POSCACHE_SEEN); \
2249 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2251 else if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS)) { \
2252 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2255 /* cache records failure, but this is success */ \
2257 PerlIO_printf(Perl_debug_log, \
2258 "%*s (remove success from failure cache)\n", \
2259 REPORT_CODE_OFF+PL_regindent*2, "") \
2261 PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<<st->u.whilem.cache_bit); \
2267 #define CACHEsayNO STMT_START { \
2268 if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \
2269 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) { \
2270 PL_reg_poscache[0] |= (1<<POSCACHE_SEEN); \
2271 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2273 else if (!(PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2274 PL_reg_poscache[st->u.whilem.cache_offset] |= (1<<st->u.whilem.cache_bit); \
2277 /* cache records success, but this is failure */ \
2279 PerlIO_printf(Perl_debug_log, \
2280 "%*s (remove failure from success cache)\n", \
2281 REPORT_CODE_OFF+PL_regindent*2, "") \
2283 PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<<st->u.whilem.cache_bit); \
2289 /* this is used to determine how far from the left messages like
2290 'failed...' are printed. Currently 29 makes these messages line
2291 up with the opcode they refer to. Earlier perls used 25 which
2292 left these messages outdented making reviewing a debug output
2295 #define REPORT_CODE_OFF 29
2298 /* Make sure there is a test for this +1 options in re_tests */
2299 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2301 /* this value indiciates that the c1/c2 "next char" test should be skipped */
2302 #define CHRTEST_VOID -1000
2304 #define SLAB_FIRST(s) (&(s)->states[0])
2305 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2307 /* grab a new slab and return the first slot in it */
2309 STATIC regmatch_state *
2312 regmatch_slab *s = PL_regmatch_slab->next;
2314 Newx(s, 1, regmatch_slab);
2315 s->prev = PL_regmatch_slab;
2317 PL_regmatch_slab->next = s;
2319 PL_regmatch_slab = s;
2320 return SLAB_FIRST(s);
2323 /* simulate a recursive call to regmatch */
2325 #define REGMATCH(ns, where) \
2328 st->resume_state = resume_##where; \
2329 goto start_recurse; \
2330 resume_point_##where:
2333 /* push a new regex state. Set newst to point to it */
2335 #define PUSH_STATE(newst, resume) \
2337 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "PUSH STATE(%d)\n", depth)); \
2341 st->locinput = locinput; \
2342 st->resume_state = resume; \
2344 if (newst > SLAB_LAST(PL_regmatch_slab)) \
2345 newst = S_push_slab(aTHX); \
2346 PL_regmatch_state = newst; \
2348 newst->minmod = 0; \
2350 newst->logical = 0; \
2351 newst->unwind = 0; \
2352 locinput = PL_reginput; \
2353 nextchr = UCHARAT(locinput);
2356 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "POP STATE(%d)\n", depth)); \
2359 if (st < SLAB_FIRST(PL_regmatch_slab)) { \
2360 PL_regmatch_slab = PL_regmatch_slab->prev; \
2361 st = SLAB_LAST(PL_regmatch_slab); \
2363 PL_regmatch_state = st; \
2367 locinput = st->locinput; \
2368 nextchr = UCHARAT(locinput);
2371 - regmatch - main matching routine
2373 * Conceptually the strategy is simple: check to see whether the current
2374 * node matches, call self recursively to see whether the rest matches,
2375 * and then act accordingly. In practice we make some effort to avoid
2376 * recursion, in particular by going through "ordinary" nodes (that don't
2377 * need to know whether the rest of the match failed) by a loop instead of
2380 /* [lwall] I've hoisted the register declarations to the outer block in order to
2381 * maybe save a little bit of pushing and popping on the stack. It also takes
2382 * advantage of machines that use a register save mask on subroutine entry.
2384 * This function used to be heavily recursive, but since this had the
2385 * effect of blowing the CPU stack on complex regexes, it has been
2386 * restructured to be iterative, and to save state onto the heap rather
2387 * than the stack. Essentially whereever regmatch() used to be called, it
2388 * pushes the current state, notes where to return, then jumps back into
2391 * Originally the structure of this function used to look something like
2396 while (scan != NULL) {
2397 a++; // do stuff with a and b
2403 if (regmatch(...)) // recurse
2413 * Now it looks something like this:
2421 regmatch_state *st = new();
2423 st->a++; // do stuff with a and b
2425 while (scan != NULL) {
2433 st->resume_state = resume_FOO;
2434 goto start_recurse; // recurse
2443 st = new(); push a new state
2444 st->a = 1; st->b = 2;
2451 switch (resume_state) {
2453 goto resume_point_FOO;
2460 * WARNING: this means that any line in this function that contains a
2461 * REGMATCH() or TRYPAREN() is actually simulating a recursive call to
2462 * regmatch() using gotos instead. Thus the values of any local variables
2463 * not saved in the regmatch_state structure will have been lost when
2464 * execution resumes on the next line .
2466 * States (ie the st pointer) are allocated in slabs of about 4K in size.
2467 * PL_regmatch_state always points to the currently active state, and
2468 * PL_regmatch_slab points to the slab currently containing PL_regmatch_state.
2469 * The first time regmatch is called, the first slab is allocated, and is
2470 * never freed until interpreter desctruction. When the slab is full,
2471 * a new one is allocated chained to the end. At exit from regmatch, slabs
2472 * allocated since entry are freed.
2476 STATIC I32 /* 0 failure, 1 success */
2477 S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog)
2480 register const bool do_utf8 = PL_reg_match_utf8;
2481 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2483 regexp *rex = reginfo->prog;
2485 regmatch_slab *orig_slab;
2486 regmatch_state *orig_state;
2488 /* the current state. This is a cached copy of PL_regmatch_state */
2489 register regmatch_state *st;
2491 /* cache heavy used fields of st in registers */
2492 register regnode *scan;
2493 register regnode *next;
2494 register I32 n = 0; /* initialize to shut up compiler warning */
2495 register char *locinput = PL_reginput;
2497 /* these variables are NOT saved during a recusive RFEGMATCH: */
2498 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2499 bool result; /* return value of S_regmatch */
2500 regnode *inner; /* Next node in internal branch. */
2501 int depth = 0; /* depth of recursion */
2502 regmatch_state *newst; /* when pushing a state, this is the new one */
2503 regmatch_state *yes_state = NULL; /* state to pop to on success of
2507 SV *re_debug_flags = NULL;
2512 /* on first ever call to regmatch, allocate first slab */
2513 if (!PL_regmatch_slab) {
2514 Newx(PL_regmatch_slab, 1, regmatch_slab);
2515 PL_regmatch_slab->prev = NULL;
2516 PL_regmatch_slab->next = NULL;
2517 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2520 /* remember current high-water mark for exit */
2521 /* XXX this should be done with SAVE* instead */
2522 orig_slab = PL_regmatch_slab;
2523 orig_state = PL_regmatch_state;
2525 /* grab next free state slot */
2526 st = ++PL_regmatch_state;
2527 if (st > SLAB_LAST(PL_regmatch_slab))
2528 st = PL_regmatch_state = S_push_slab(aTHX);
2535 /* Note that nextchr is a byte even in UTF */
2536 nextchr = UCHARAT(locinput);
2538 while (scan != NULL) {
2541 SV * const prop = sv_newmortal();
2542 const int docolor = *PL_colors[0];
2543 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2544 int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput);
2545 /* The part of the string before starttry has one color
2546 (pref0_len chars), between starttry and current
2547 position another one (pref_len - pref0_len chars),
2548 after the current position the third one.
2549 We assume that pref0_len <= pref_len, otherwise we
2550 decrease pref0_len. */
2551 int pref_len = (locinput - PL_bostr) > (5 + taill) - l
2552 ? (5 + taill) - l : locinput - PL_bostr;
2555 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2557 pref0_len = pref_len - (locinput - PL_reg_starttry);
2558 if (l + pref_len < (5 + taill) && l < PL_regeol - locinput)
2559 l = ( PL_regeol - locinput > (5 + taill) - pref_len
2560 ? (5 + taill) - pref_len : PL_regeol - locinput);
2561 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2565 if (pref0_len > pref_len)
2566 pref0_len = pref_len;
2567 regprop(rex, prop, scan);
2569 const char * const s0 =
2570 do_utf8 && OP(scan) != CANY ?
2571 pv_uni_display(PERL_DEBUG_PAD(0), (U8*)(locinput - pref_len),
2572 pref0_len, 60, UNI_DISPLAY_REGEX) :
2573 locinput - pref_len;
2574 const int len0 = do_utf8 ? strlen(s0) : pref0_len;
2575 const char * const s1 = do_utf8 && OP(scan) != CANY ?
2576 pv_uni_display(PERL_DEBUG_PAD(1),
2577 (U8*)(locinput - pref_len + pref0_len),
2578 pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) :
2579 locinput - pref_len + pref0_len;
2580 const int len1 = do_utf8 ? strlen(s1) : pref_len - pref0_len;
2581 const char * const s2 = do_utf8 && OP(scan) != CANY ?
2582 pv_uni_display(PERL_DEBUG_PAD(2), (U8*)locinput,
2583 PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) :
2585 const int len2 = do_utf8 ? strlen(s2) : l;
2586 PerlIO_printf(Perl_debug_log,
2587 "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|%3"IVdf":%*s%s\n",
2588 (IV)(locinput - PL_bostr),
2595 (docolor ? "" : "> <"),
2599 15 - l - pref_len + 1,
2601 (IV)(scan - rex->program), PL_regindent*2, "",
2606 next = scan + NEXT_OFF(scan);
2612 if (locinput == PL_bostr)
2614 /* reginfo->till = reginfo->bol; */
2619 if (locinput == PL_bostr ||
2620 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2626 if (locinput == PL_bostr)
2630 if (locinput == reginfo->ganch)
2636 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2641 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2643 if (PL_regeol - locinput > 1)
2647 if (PL_regeol != locinput)
2651 if (!nextchr && locinput >= PL_regeol)
2654 locinput += PL_utf8skip[nextchr];
2655 if (locinput > PL_regeol)
2657 nextchr = UCHARAT(locinput);
2660 nextchr = UCHARAT(++locinput);
2663 if (!nextchr && locinput >= PL_regeol)
2665 nextchr = UCHARAT(++locinput);
2668 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2671 locinput += PL_utf8skip[nextchr];
2672 if (locinput > PL_regeol)
2674 nextchr = UCHARAT(locinput);
2677 nextchr = UCHARAT(++locinput);
2683 traverse the TRIE keeping track of all accepting states
2684 we transition through until we get to a failing node.
2692 U8 *uc = ( U8* )locinput;
2699 U8 *uscan = (U8*)NULL;
2701 SV *sv_accept_buff = NULL;
2702 const enum { trie_plain, trie_utf8, trie_uft8_fold }
2703 trie_type = do_utf8 ?
2704 (OP(scan) == TRIE ? trie_utf8 : trie_uft8_fold)
2707 /* what trie are we using right now */
2709 = (reg_trie_data*)rex->data->data[ ARG( scan ) ];
2710 st->u.trie.accepted = 0; /* how many accepting states we have seen */
2713 while ( state && uc <= (U8*)PL_regeol ) {
2715 if (trie->states[ state ].wordnum) {
2716 if (!st->u.trie.accepted ) {
2719 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
2720 sv_accept_buff=newSV(bufflen *
2721 sizeof(reg_trie_accepted) - 1);
2722 SvCUR_set(sv_accept_buff,
2723 sizeof(reg_trie_accepted));
2724 SvPOK_on(sv_accept_buff);
2725 sv_2mortal(sv_accept_buff);
2726 st->u.trie.accept_buff =
2727 (reg_trie_accepted*)SvPV_nolen(sv_accept_buff );
2730 if (st->u.trie.accepted >= bufflen) {
2732 st->u.trie.accept_buff =(reg_trie_accepted*)
2733 SvGROW(sv_accept_buff,
2734 bufflen * sizeof(reg_trie_accepted));
2736 SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff)
2737 + sizeof(reg_trie_accepted));
2739 st->u.trie.accept_buff[st->u.trie.accepted].wordnum = trie->states[state].wordnum;
2740 st->u.trie.accept_buff[st->u.trie.accepted].endpos = uc;
2741 ++st->u.trie.accepted;
2744 base = trie->states[ state ].trans.base;
2746 DEBUG_TRIE_EXECUTE_r(
2747 PerlIO_printf( Perl_debug_log,
2748 "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ",
2749 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2750 (UV)state, (UV)base, (UV)st->u.trie.accepted );
2754 switch (trie_type) {
2755 case trie_uft8_fold:
2757 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags );
2762 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2763 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2764 uvc = to_uni_fold( uvc, foldbuf, &foldlen );
2765 foldlen -= UNISKIP( uvc );
2766 uscan = foldbuf + UNISKIP( uvc );
2770 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN,
2779 charid = trie->charmap[ uvc ];
2783 if (trie->widecharmap) {
2784 SV** svpp = (SV**)NULL;
2785 svpp = hv_fetch(trie->widecharmap,
2786 (char*)&uvc, sizeof(UV), 0);
2788 charid = (U16)SvIV(*svpp);
2793 (base + charid > trie->uniquecharcount )
2794 && (base + charid - 1 - trie->uniquecharcount
2796 && trie->trans[base + charid - 1 -
2797 trie->uniquecharcount].check == state)
2799 state = trie->trans[base + charid - 1 -
2800 trie->uniquecharcount ].next;
2811 DEBUG_TRIE_EXECUTE_r(
2812 PerlIO_printf( Perl_debug_log,
2813 "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n",
2814 charid, uvc, (UV)state, PL_colors[5] );
2817 if (!st->u.trie.accepted )
2821 There was at least one accepting state that we
2822 transitioned through. Presumably the number of accepting
2823 states is going to be low, typically one or two. So we
2824 simply scan through to find the one with lowest wordnum.
2825 Once we find it, we swap the last state into its place
2826 and decrement the size. We then try to match the rest of
2827 the pattern at the point where the word ends, if we
2828 succeed then we end the loop, otherwise the loop
2829 eventually terminates once all of the accepting states
2833 if ( st->u.trie.accepted == 1 ) {
2835 SV ** const tmp = av_fetch( trie->words, st->u.trie.accept_buff[ 0 ].wordnum-1, 0 );
2836 PerlIO_printf( Perl_debug_log,
2837 "%*s %sonly one match : #%d <%s>%s\n",
2838 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2839 st->u.trie.accept_buff[ 0 ].wordnum,
2840 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2843 PL_reginput = (char *)st->u.trie.accept_buff[ 0 ].endpos;
2844 /* in this case we free tmps/leave before we call regmatch
2845 as we wont be using accept_buff again. */
2848 REGMATCH(scan + NEXT_OFF(scan), TRIE1);
2849 /*** all unsaved local vars undefined at this point */
2852 PerlIO_printf( Perl_debug_log,"%*s %sgot %"IVdf" possible matches%s\n",
2853 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], (IV)st->u.trie.accepted,
2856 while ( !result && st->u.trie.accepted-- ) {
2859 for( cur = 1 ; cur <= st->u.trie.accepted ; cur++ ) {
2860 DEBUG_TRIE_EXECUTE_r(
2861 PerlIO_printf( Perl_debug_log,
2862 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2863 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2864 (IV)best, st->u.trie.accept_buff[ best ].wordnum, (IV)cur,
2865 st->u.trie.accept_buff[ cur ].wordnum, PL_colors[5] );
2868 if (st->u.trie.accept_buff[cur].wordnum <
2869 st->u.trie.accept_buff[best].wordnum)
2873 reg_trie_data * const trie = (reg_trie_data*)
2874 rex->data->data[ARG(scan)];
2875 SV ** const tmp = av_fetch( trie->words, st->u.trie.accept_buff[ best ].wordnum - 1, 0 );
2876 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at 0x%p%s\n",
2877 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2878 st->u.trie.accept_buff[best].wordnum,
2879 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",scan,
2882 if ( best<st->u.trie.accepted ) {
2883 reg_trie_accepted tmp = st->u.trie.accept_buff[ best ];
2884 st->u.trie.accept_buff[ best ] = st->u.trie.accept_buff[ st->u.trie.accepted ];
2885 st->u.trie.accept_buff[ st->u.trie.accepted ] = tmp;
2886 best = st->u.trie.accepted;
2888 PL_reginput = (char *)st->u.trie.accept_buff[ best ].endpos;
2891 as far as I can tell we only need the SAVETMPS/FREETMPS
2892 for re's with EVAL in them but I'm leaving them in for
2893 all until I can be sure.
2896 REGMATCH(scan + NEXT_OFF(scan), TRIE2);
2897 /*** all unsaved local vars undefined at this point */
2910 /* unreached codepoint */
2912 char *s = STRING(scan);
2913 st->ln = STR_LEN(scan);
2914 if (do_utf8 != UTF) {
2915 /* The target and the pattern have differing utf8ness. */
2917 const char *e = s + st->ln;
2920 /* The target is utf8, the pattern is not utf8. */
2925 if (NATIVE_TO_UNI(*(U8*)s) !=
2926 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2934 /* The target is not utf8, the pattern is utf8. */
2939 if (NATIVE_TO_UNI(*((U8*)l)) !=
2940 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2948 nextchr = UCHARAT(locinput);
2951 /* The target and the pattern have the same utf8ness. */
2952 /* Inline the first character, for speed. */
2953 if (UCHARAT(s) != nextchr)
2955 if (PL_regeol - locinput < st->ln)
2957 if (st->ln > 1 && memNE(s, locinput, st->ln))
2960 nextchr = UCHARAT(locinput);
2964 PL_reg_flags |= RF_tainted;
2967 char *s = STRING(scan);
2968 st->ln = STR_LEN(scan);
2970 if (do_utf8 || UTF) {
2971 /* Either target or the pattern are utf8. */
2973 char *e = PL_regeol;
2975 if (ibcmp_utf8(s, 0, st->ln, (bool)UTF,
2976 l, &e, 0, do_utf8)) {
2977 /* One more case for the sharp s:
2978 * pack("U0U*", 0xDF) =~ /ss/i,
2979 * the 0xC3 0x9F are the UTF-8
2980 * byte sequence for the U+00DF. */
2982 toLOWER(s[0]) == 's' &&
2984 toLOWER(s[1]) == 's' &&
2991 nextchr = UCHARAT(locinput);
2995 /* Neither the target and the pattern are utf8. */
2997 /* Inline the first character, for speed. */
2998 if (UCHARAT(s) != nextchr &&
2999 UCHARAT(s) != ((OP(scan) == EXACTF)
3000 ? PL_fold : PL_fold_locale)[nextchr])
3002 if (PL_regeol - locinput < st->ln)
3004 if (st->ln > 1 && (OP(scan) == EXACTF
3005 ? ibcmp(s, locinput, st->ln)
3006 : ibcmp_locale(s, locinput, st->ln)))
3009 nextchr = UCHARAT(locinput);
3014 STRLEN inclasslen = PL_regeol - locinput;
3016 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8))
3018 if (locinput >= PL_regeol)
3020 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
3021 nextchr = UCHARAT(locinput);
3026 nextchr = UCHARAT(locinput);
3027 if (!REGINCLASS(rex, scan, (U8*)locinput))
3029 if (!nextchr && locinput >= PL_regeol)
3031 nextchr = UCHARAT(++locinput);
3035 /* If we might have the case of the German sharp s
3036 * in a casefolding Unicode character class. */
3038 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3039 locinput += SHARP_S_SKIP;
3040 nextchr = UCHARAT(locinput);
3046 PL_reg_flags |= RF_tainted;
3052 LOAD_UTF8_CHARCLASS_ALNUM();
3053 if (!(OP(scan) == ALNUM
3054 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3055 : isALNUM_LC_utf8((U8*)locinput)))
3059 locinput += PL_utf8skip[nextchr];
3060 nextchr = UCHARAT(locinput);
3063 if (!(OP(scan) == ALNUM
3064 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
3066 nextchr = UCHARAT(++locinput);
3069 PL_reg_flags |= RF_tainted;
3072 if (!nextchr && locinput >= PL_regeol)
3075 LOAD_UTF8_CHARCLASS_ALNUM();
3076 if (OP(scan) == NALNUM
3077 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
3078 : isALNUM_LC_utf8((U8*)locinput))
3082 locinput += PL_utf8skip[nextchr];
3083 nextchr = UCHARAT(locinput);
3086 if (OP(scan) == NALNUM
3087 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
3089 nextchr = UCHARAT(++locinput);
3093 PL_reg_flags |= RF_tainted;
3097 /* was last char in word? */
3099 if (locinput == PL_bostr)
3102 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3104 st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3106 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3107 st->ln = isALNUM_uni(st->ln);
3108 LOAD_UTF8_CHARCLASS_ALNUM();
3109 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
3112 st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln));
3113 n = isALNUM_LC_utf8((U8*)locinput);
3117 st->ln = (locinput != PL_bostr) ?
3118 UCHARAT(locinput - 1) : '\n';
3119 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3120 st->ln = isALNUM(st->ln);
3121 n = isALNUM(nextchr);
3124 st->ln = isALNUM_LC(st->ln);
3125 n = isALNUM_LC(nextchr);
3128 if (((!st->ln) == (!n)) == (OP(scan) == BOUND ||
3129 OP(scan) == BOUNDL))
3133 PL_reg_flags |= RF_tainted;
3139 if (UTF8_IS_CONTINUED(nextchr)) {
3140 LOAD_UTF8_CHARCLASS_SPACE();
3141 if (!(OP(scan) == SPACE
3142 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3143 : isSPACE_LC_utf8((U8*)locinput)))
3147 locinput += PL_utf8skip[nextchr];
3148 nextchr = UCHARAT(locinput);
3151 if (!(OP(scan) == SPACE
3152 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3154 nextchr = UCHARAT(++locinput);
3157 if (!(OP(scan) == SPACE
3158 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3160 nextchr = UCHARAT(++locinput);
3164 PL_reg_flags |= RF_tainted;
3167 if (!nextchr && locinput >= PL_regeol)
3170 LOAD_UTF8_CHARCLASS_SPACE();
3171 if (OP(scan) == NSPACE
3172 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3173 : isSPACE_LC_utf8((U8*)locinput))
3177 locinput += PL_utf8skip[nextchr];
3178 nextchr = UCHARAT(locinput);
3181 if (OP(scan) == NSPACE
3182 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3184 nextchr = UCHARAT(++locinput);
3187 PL_reg_flags |= RF_tainted;
3193 LOAD_UTF8_CHARCLASS_DIGIT();
3194 if (!(OP(scan) == DIGIT
3195 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3196 : isDIGIT_LC_utf8((U8*)locinput)))
3200 locinput += PL_utf8skip[nextchr];
3201 nextchr = UCHARAT(locinput);
3204 if (!(OP(scan) == DIGIT
3205 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3207 nextchr = UCHARAT(++locinput);
3210 PL_reg_flags |= RF_tainted;
3213 if (!nextchr && locinput >= PL_regeol)
3216 LOAD_UTF8_CHARCLASS_DIGIT();
3217 if (OP(scan) == NDIGIT
3218 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3219 : isDIGIT_LC_utf8((U8*)locinput))
3223 locinput += PL_utf8skip[nextchr];
3224 nextchr = UCHARAT(locinput);
3227 if (OP(scan) == NDIGIT
3228 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3230 nextchr = UCHARAT(++locinput);
3233 if (locinput >= PL_regeol)
3236 LOAD_UTF8_CHARCLASS_MARK();
3237 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3239 locinput += PL_utf8skip[nextchr];
3240 while (locinput < PL_regeol &&
3241 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3242 locinput += UTF8SKIP(locinput);
3243 if (locinput > PL_regeol)
3248 nextchr = UCHARAT(locinput);
3251 PL_reg_flags |= RF_tainted;
3256 n = ARG(scan); /* which paren pair */
3257 st->ln = PL_regstartp[n];
3258 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3259 if ((I32)*PL_reglastparen < n || st->ln == -1)
3260 sayNO; /* Do not match unless seen CLOSEn. */
3261 if (st->ln == PL_regendp[n])
3264 s = PL_bostr + st->ln;
3265 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3267 const char *e = PL_bostr + PL_regendp[n];
3269 * Note that we can't do the "other character" lookup trick as
3270 * in the 8-bit case (no pun intended) because in Unicode we
3271 * have to map both upper and title case to lower case.
3273 if (OP(scan) == REFF) {
3275 STRLEN ulen1, ulen2;
3276 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3277 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3281 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3282 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3283 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3290 nextchr = UCHARAT(locinput);
3294 /* Inline the first character, for speed. */
3295 if (UCHARAT(s) != nextchr &&
3297 (UCHARAT(s) != ((OP(scan) == REFF
3298 ? PL_fold : PL_fold_locale)[nextchr]))))
3300 st->ln = PL_regendp[n] - st->ln;
3301 if (locinput + st->ln > PL_regeol)
3303 if (st->ln > 1 && (OP(scan) == REF
3304 ? memNE(s, locinput, st->ln)
3306 ? ibcmp(s, locinput, st->ln)
3307 : ibcmp_locale(s, locinput, st->ln))))
3310 nextchr = UCHARAT(locinput);
3323 /* execute the code in the {...} */
3325 SV ** const before = SP;
3326 OP_4tree * const oop = PL_op;
3327 COP * const ocurcop = PL_curcop;
3331 PL_op = (OP_4tree*)rex->data->data[n];
3332 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3333 PAD_SAVE_LOCAL(old_comppad, (PAD*)rex->data->data[n + 2]);
3334 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3336 CALLRUNOPS(aTHX); /* Scalar context. */
3339 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3346 PAD_RESTORE_LOCAL(old_comppad);
3347 PL_curcop = ocurcop;
3350 sv_setsv(save_scalar(PL_replgv), ret);
3354 if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */
3357 /* extract RE object from returned value; compiling if
3362 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3363 mg = mg_find(sv, PERL_MAGIC_qr);
3364 else if (SvSMAGICAL(ret)) {
3365 if (SvGMAGICAL(ret))
3366 sv_unmagic(ret, PERL_MAGIC_qr);
3368 mg = mg_find(ret, PERL_MAGIC_qr);
3372 re = (regexp *)mg->mg_obj;
3373 (void)ReREFCNT_inc(re);
3377 const char * const t = SvPV_const(ret, len);
3379 const I32 osize = PL_regsize;
3382 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3383 re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm);
3385 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3387 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3393 /* run the pattern returned from (??{...}) */
3396 PerlIO_printf(Perl_debug_log,
3397 "Entering embedded \"%s%.60s%s%s\"\n",
3401 (strlen(re->precomp) > 60 ? "..." : ""))
3404 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
3405 REGCP_SET(st->u.eval.lastcp);
3406 *PL_reglastparen = 0;
3407 *PL_reglastcloseparen = 0;
3408 PL_reginput = locinput;
3410 /* XXXX This is too dramatic a measure... */
3414 st->u.eval.toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3415 ((re->reganch & ROPT_UTF8) != 0);
3416 if (st->u.eval.toggleutf) PL_reg_flags ^= RF_utf8;
3417 st->u.eval.prev_rex = rex;
3420 /* resume to current state on success */
3421 st->u.yes.prev_yes_state = yes_state;
3423 PUSH_STATE(newst, resume_EVAL);
3426 /* now continue from first node in postoned RE */
3427 next = re->program + 1;
3431 /* /(?(?{...})X|Y)/ */
3432 st->sw = SvTRUE(ret);
3437 n = ARG(scan); /* which paren pair */
3438 PL_reg_start_tmp[n] = locinput;
3443 n = ARG(scan); /* which paren pair */
3444 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3445 PL_regendp[n] = locinput - PL_bostr;
3446 if (n > (I32)*PL_reglastparen)
3447 *PL_reglastparen = n;
3448 *PL_reglastcloseparen = n;
3451 n = ARG(scan); /* which paren pair */
3452 st->sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3455 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3457 next = NEXTOPER(NEXTOPER(scan));
3459 next = scan + ARG(scan);
3460 if (OP(next) == IFTHEN) /* Fake one. */
3461 next = NEXTOPER(NEXTOPER(next));
3465 st->logical = scan->flags;
3467 /*******************************************************************
3468 cc points to the regmatch_state associated with the most recent CURLYX.
3469 This struct contains info about the innermost (...)* loop (an
3470 "infoblock"), and a pointer to the next outer cc.
3472 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3474 1) After matching Y, regnode for CURLYX is processed;
3476 2) This regnode populates cc, and calls regmatch() recursively
3477 with the starting point at WHILEM node;
3479 3) Each hit of WHILEM node tries to match A and Z (in the order
3480 depending on the current iteration, min/max of {min,max} and
3481 greediness). The information about where are nodes for "A"
3482 and "Z" is read from cc, as is info on how many times "A"
3483 was already matched, and greediness.
3485 4) After A matches, the same WHILEM node is hit again.
3487 5) Each time WHILEM is hit, cc is the infoblock created by CURLYX
3488 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3489 resets cc, since this Y(A)*Z can be a part of some other loop:
3490 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3491 of the external loop.
3493 Currently present infoblocks form a tree with a stem formed by st->cc
3494 and whatever it mentions via ->next, and additional attached trees
3495 corresponding to temporarily unset infoblocks as in "5" above.
3497 In the following picture, infoblocks for outer loop of
3498 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3499 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3500 infoblocks are drawn below the "reset" infoblock.
3502 In fact in the picture below we do not show failed matches for Z and T
3503 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3504 more obvious *why* one needs to *temporary* unset infoblocks.]
3506 Matched REx position InfoBlocks Comment
3510 Y A)*?Z)*?T x <- O <- I
3511 YA )*?Z)*?T x <- O <- I
3512 YA A)*?Z)*?T x <- O <- I
3513 YAA )*?Z)*?T x <- O <- I
3514 YAA Z)*?T x <- O # Temporary unset I
3517 YAAZ Y(A)*?Z)*?T x <- O
3520 YAAZY (A)*?Z)*?T x <- O
3523 YAAZY A)*?Z)*?T x <- O <- I
3526 YAAZYA )*?Z)*?T x <- O <- I
3529 YAAZYA Z)*?T x <- O # Temporary unset I
3535 YAAZYAZ T x # Temporary unset O
3542 *******************************************************************/
3545 /* No need to save/restore up to this paren */
3546 I32 parenfloor = scan->flags;
3550 CURLYX and WHILEM are always paired: they're the moral
3551 equivalent of pp_enteriter anbd pp_iter.
3553 The only time next could be null is if the node tree is
3554 corrupt. This was mentioned on p5p a few days ago.
3556 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3557 So we'll assert that this is true:
3560 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3562 /* XXXX Probably it is better to teach regpush to support
3563 parenfloor > PL_regsize... */
3564 if (parenfloor > (I32)*PL_reglastparen)
3565 parenfloor = *PL_reglastparen; /* Pessimization... */
3567 st->u.curlyx.cp = PL_savestack_ix;
3568 st->u.curlyx.outercc = st->cc;
3570 /* these fields contain the state of the current curly.
3571 * they are accessed by subsequent WHILEMs;
3572 * cur and lastloc are also updated by WHILEM */
3573 st->u.curlyx.parenfloor = parenfloor;
3574 st->u.curlyx.cur = -1; /* this will be updated by WHILEM */
3575 st->u.curlyx.min = ARG1(scan);
3576 st->u.curlyx.max = ARG2(scan);
3577 st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3578 st->u.curlyx.lastloc = 0;
3579 /* st->next and st->minmod are also read by WHILEM */
3581 PL_reginput = locinput;
3582 REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */
3583 /*** all unsaved local vars undefined at this point */
3584 regcpblow(st->u.curlyx.cp);
3585 st->cc = st->u.curlyx.outercc;
3591 * This is really hard to understand, because after we match
3592 * what we're trying to match, we must make sure the rest of
3593 * the REx is going to match for sure, and to do that we have
3594 * to go back UP the parse tree by recursing ever deeper. And
3595 * if it fails, we have to reset our parent's current state
3596 * that we can try again after backing off.
3601 st->cc gets initialised by CURLYX ready for use by WHILEM.
3602 So again, unless somethings been corrupted, st->cc cannot
3603 be null at that point in WHILEM.
3605 See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
3606 So we'll assert that this is true:
3609 st->u.whilem.lastloc = st->cc->u.curlyx.lastloc; /* Detection of 0-len. */
3610 st->u.whilem.cache_offset = 0;
3611 st->u.whilem.cache_bit = 0;
3613 n = st->cc->u.curlyx.cur + 1; /* how many we know we matched */
3614 PL_reginput = locinput;
3617 PerlIO_printf(Perl_debug_log,
3618 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3619 REPORT_CODE_OFF+PL_regindent*2, "",
3620 (long)n, (long)st->cc->u.curlyx.min,
3621 (long)st->cc->u.curlyx.max, PTR2UV(st->cc))
3624 /* If degenerate scan matches "", assume scan done. */
3626 if (locinput == st->cc->u.curlyx.lastloc && n >= st->cc->u.curlyx.min) {
3627 st->u.whilem.savecc = st->cc;
3628 st->cc = st->cc->u.curlyx.outercc;
3630 st->ln = st->cc->u.curlyx.cur;
3632 PerlIO_printf(Perl_debug_log,
3633 "%*s empty match detected, try continuation...\n",
3634 REPORT_CODE_OFF+PL_regindent*2, "")
3636 REGMATCH(st->u.whilem.savecc->next, WHILEM1);
3637 /*** all unsaved local vars undefined at this point */
3638 st->cc = st->u.whilem.savecc;
3641 if (st->cc->u.curlyx.outercc)
3642 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3646 /* First just match a string of min scans. */
3648 if (n < st->cc->u.curlyx.min) {
3649 st->cc->u.curlyx.cur = n;
3650 st->cc->u.curlyx.lastloc = locinput;
3651 REGMATCH(st->cc->u.curlyx.scan, WHILEM2);
3652 /*** all unsaved local vars undefined at this point */
3655 st->cc->u.curlyx.cur = n - 1;
3656 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3661 /* Check whether we already were at this position.
3662 Postpone detection until we know the match is not
3663 *that* much linear. */
3664 if (!PL_reg_maxiter) {
3665 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3666 PL_reg_leftiter = PL_reg_maxiter;
3668 if (PL_reg_leftiter-- == 0) {
3669 const I32 size = (PL_reg_maxiter + 7 + POSCACHE_START)/8;
3670 if (PL_reg_poscache) {
3671 if ((I32)PL_reg_poscache_size < size) {
3672 Renew(PL_reg_poscache, size, char);
3673 PL_reg_poscache_size = size;
3675 Zero(PL_reg_poscache, size, char);
3678 PL_reg_poscache_size = size;
3679 Newxz(PL_reg_poscache, size, char);
3682 PerlIO_printf(Perl_debug_log,
3683 "%sDetected a super-linear match, switching on caching%s...\n",
3684 PL_colors[4], PL_colors[5])
3687 if (PL_reg_leftiter < 0) {
3688 st->u.whilem.cache_offset = locinput - PL_bostr;
3690 st->u.whilem.cache_offset = (scan->flags & 0xf) - 1 + POSCACHE_START
3691 + st->u.whilem.cache_offset * (scan->flags>>4);
3692 st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8;
3693 st->u.whilem.cache_offset /= 8;
3694 if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<<st->u.whilem.cache_bit)) {
3696 PerlIO_printf(Perl_debug_log,
3697 "%*s already tried at this position...\n",
3698 REPORT_CODE_OFF+PL_regindent*2, "")
3700 if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))
3701 /* cache records success */
3704 /* cache records failure */
3710 /* Prefer next over scan for minimal matching. */
3712 if (st->cc->minmod) {
3713 st->u.whilem.savecc = st->cc;
3714 st->cc = st->cc->u.curlyx.outercc;
3716 st->ln = st->cc->u.curlyx.cur;
3717 st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor);
3718 REGCP_SET(st->u.whilem.lastcp);
3719 REGMATCH(st->u.whilem.savecc->next, WHILEM3);
3720 /*** all unsaved local vars undefined at this point */
3721 st->cc = st->u.whilem.savecc;
3723 regcpblow(st->u.whilem.cp);
3724 CACHEsayYES; /* All done. */
3726 REGCP_UNWIND(st->u.whilem.lastcp);
3728 if (st->cc->u.curlyx.outercc)
3729 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3731 if (n >= st->cc->u.curlyx.max) { /* Maximum greed exceeded? */
3732 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3733 && !(PL_reg_flags & RF_warned)) {
3734 PL_reg_flags |= RF_warned;
3735 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3736 "Complex regular subexpression recursion",
3743 PerlIO_printf(Perl_debug_log,
3744 "%*s trying longer...\n",
3745 REPORT_CODE_OFF+PL_regindent*2, "")
3747 /* Try scanning more and see if it helps. */
3748 PL_reginput = locinput;
3749 st->cc->u.curlyx.cur = n;
3750 st->cc->u.curlyx.lastloc = locinput;
3751 st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor);
3752 REGCP_SET(st->u.whilem.lastcp);
3753 REGMATCH(st->cc->u.curlyx.scan, WHILEM4);
3754 /*** all unsaved local vars undefined at this point */
3756 regcpblow(st->u.whilem.cp);
3759 REGCP_UNWIND(st->u.whilem.lastcp);
3761 st->cc->u.curlyx.cur = n - 1;
3762 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3766 /* Prefer scan over next for maximal matching. */
3768 if (n < st->cc->u.curlyx.max) { /* More greed allowed? */
3769 st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor);
3770 st->cc->u.curlyx.cur = n;
3771 st->cc->u.curlyx.lastloc = locinput;
3772 REGCP_SET(st->u.whilem.lastcp);
3773 REGMATCH(st->cc->u.curlyx.scan, WHILEM5);
3774 /*** all unsaved local vars undefined at this point */
3776 regcpblow(st->u.whilem.cp);
3779 REGCP_UNWIND(st->u.whilem.lastcp);
3780 regcppop(rex); /* Restore some previous $<digit>s? */
3781 PL_reginput = locinput;
3783 PerlIO_printf(Perl_debug_log,
3784 "%*s failed, try continuation...\n",
3785 REPORT_CODE_OFF+PL_regindent*2, "")
3788 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3789 && !(PL_reg_flags & RF_warned)) {
3790 PL_reg_flags |= RF_warned;
3791 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3792 "Complex regular subexpression recursion",
3796 /* Failed deeper matches of scan, so see if this one works. */
3797 st->u.whilem.savecc = st->cc;
3798 st->cc = st->cc->u.curlyx.outercc;
3800 st->ln = st->cc->u.curlyx.cur;
3801 REGMATCH(st->u.whilem.savecc->next, WHILEM6);
3802 /*** all unsaved local vars undefined at this point */
3803 st->cc = st->u.whilem.savecc;
3806 if (st->cc->u.curlyx.outercc)
3807 st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln;
3808 st->cc->u.curlyx.cur = n - 1;
3809 st->cc->u.curlyx.lastloc = st->u.whilem.lastloc;
3814 next = scan + ARG(scan);
3817 inner = NEXTOPER(NEXTOPER(scan));
3820 inner = NEXTOPER(scan);
3825 if (!next || OP(next) != type) /* No choice. */
3826 next = inner; /* Avoid recursion. */
3828 const I32 lastparen = *PL_reglastparen;
3829 /* Put unwinding data on stack */
3830 const I32 unwind1 = SSNEWt(1,re_unwind_branch_t);
3831 re_unwind_branch_t * const uw = SSPTRt(unwind1,re_unwind_branch_t);
3833 uw->prev = st->unwind;
3834 st->unwind = unwind1;
3835 uw->type = ((type == BRANCH)
3837 : RE_UNWIND_BRANCHJ);
3838 uw->lastparen = lastparen;
3840 uw->locinput = locinput;
3841 uw->nextchr = nextchr;
3842 uw->minmod = st->minmod;
3844 uw->regindent = ++PL_regindent;
3847 REGCP_SET(uw->lastcp);
3849 /* Now go into the first branch */
3859 st->u.curlym.l = st->u.curlym.matches = 0;
3861 /* We suppose that the next guy does not need
3862 backtracking: in particular, it is of constant non-zero length,
3863 and has no parenths to influence future backrefs. */
3864 st->ln = ARG1(scan); /* min to match */
3865 n = ARG2(scan); /* max to match */
3866 st->u.curlym.paren = scan->flags;
3867 if (st->u.curlym.paren) {
3868 if (st->u.curlym.paren > PL_regsize)
3869 PL_regsize = st->u.curlym.paren;
3870 if (st->u.curlym.paren > (I32)*PL_reglastparen)
3871 *PL_reglastparen = st->u.curlym.paren;
3873 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3874 if (st->u.curlym.paren)
3875 scan += NEXT_OFF(scan); /* Skip former OPEN. */
3876 PL_reginput = locinput;
3877 st->u.curlym.maxwanted = st->minmod ? st->ln : n;
3878 while (PL_reginput < PL_regeol && st->u.curlym.matches < st->u.curlym.maxwanted) {
3879 /* resume to current state on success */
3880 st->u.yes.prev_yes_state = yes_state;
3882 REGMATCH(scan, CURLYM1);
3883 yes_state = st->u.yes.prev_yes_state;
3884 /*** all unsaved local vars undefined at this point */
3887 /* on first match, determine length, u.curlym.l */
3888 if (!st->u.curlym.matches++) {
3889 if (PL_reg_match_utf8) {
3891 while (s < PL_reginput) {
3897 st->u.curlym.l = PL_reginput - locinput;
3899 if (st->u.curlym.l == 0) {
3900 st->u.curlym.matches = st->u.curlym.maxwanted;
3904 locinput = PL_reginput;
3907 PL_reginput = locinput;
3908 if (st->u.curlym.matches < st->ln) {
3914 PerlIO_printf(Perl_debug_log,
3915 "%*s matched %"IVdf" times, len=%"IVdf"...\n",
3916 (int)(REPORT_CODE_OFF+PL_regindent*2), "",
3917 (IV) st->u.curlym.matches, (IV)st->u.curlym.l)
3920 /* calculate c1 and c1 for possible match of 1st char
3921 * following curly */
3922 st->u.curlym.c1 = st->u.curlym.c2 = CHRTEST_VOID;
3923 if (HAS_TEXT(next) || JUMPABLE(next)) {
3924 regnode *text_node = next;
3925 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3926 if (HAS_TEXT(text_node)
3927 && PL_regkind[(U8)OP(text_node)] != REF)
3929 st->u.curlym.c1 = (U8)*STRING(text_node);
3931 (OP(text_node) == EXACTF || OP(text_node) == REFF)
3932 ? PL_fold[st->u.curlym.c1]
3933 : (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3934 ? PL_fold_locale[st->u.curlym.c1]
3939 REGCP_SET(st->u.curlym.lastcp);
3941 st->u.curlym.minmod = st->minmod;
3943 while (st->u.curlym.matches >= st->ln
3944 && (st->u.curlym.matches <= n
3945 /* for REG_INFTY, ln could overflow to negative */
3946 || (n == REG_INFTY && st->u.curlym.matches >= 0)))
3948 /* If it could work, try it. */
3949 if (st->u.curlym.c1 == CHRTEST_VOID ||
3950 UCHARAT(PL_reginput) == st->u.curlym.c1 ||
3951 UCHARAT(PL_reginput) == st->u.curlym.c2)
3954 PerlIO_printf(Perl_debug_log,
3955 "%*s trying tail with matches=%"IVdf"...\n",
3956 (int)(REPORT_CODE_OFF+PL_regindent*2),
3957 "", (IV)st->u.curlym.matches)
3959 if (st->u.curlym.paren) {
3960 if (st->u.curlym.matches) {
3961 PL_regstartp[st->u.curlym.paren]
3962 = HOPc(PL_reginput, -st->u.curlym.l) - PL_bostr;
3963 PL_regendp[st->u.curlym.paren] = PL_reginput - PL_bostr;
3966 PL_regendp[st->u.curlym.paren] = -1;
3968 /* resume to current state on success */
3969 st->u.yes.prev_yes_state = yes_state;
3971 REGMATCH(next, CURLYM2);
3972 yes_state = st->u.yes.prev_yes_state;
3973 /*** all unsaved local vars undefined at this point */
3975 /* XXX tmp sayYES; */
3977 REGCP_UNWIND(st->u.curlym.lastcp);
3979 /* Couldn't or didn't -- move forward/backward. */
3980 if (st->u.curlym.minmod) {
3981 PL_reginput = locinput;
3982 /* resume to current state on success */
3983 st->u.yes.prev_yes_state = yes_state;
3985 REGMATCH(scan, CURLYM3);
3986 yes_state = st->u.yes.prev_yes_state;
3987 /*** all unsaved local vars undefined at this point */
3989 st->u.curlym.matches++;
3990 locinput = PL_reginput;
3996 st->u.curlym.matches--;
3997 locinput = HOPc(locinput, -st->u.curlym.l);
3998 PL_reginput = locinput;
4006 st->u.plus.paren = scan->flags; /* Which paren to set */
4007 if (st->u.plus.paren > PL_regsize)
4008 PL_regsize = st->u.plus.paren;
4009 if (st->u.plus.paren > (I32)*PL_reglastparen)
4010 *PL_reglastparen = st->u.plus.paren;
4011 st->ln = ARG1(scan); /* min to match */
4012 n = ARG2(scan); /* max to match */
4013 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
4016 st->u.plus.paren = 0;
4017 st->ln = ARG1(scan); /* min to match */
4018 n = ARG2(scan); /* max to match */
4019 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4024 scan = NEXTOPER(scan);
4025 st->u.plus.paren = 0;
4030 scan = NEXTOPER(scan);
4031 st->u.plus.paren = 0;
4034 * Lookahead to avoid useless match attempts
4035 * when we know what character comes next.
4039 * Used to only do .*x and .*?x, but now it allows
4040 * for )'s, ('s and (?{ ... })'s to be in the way
4041 * of the quantifier and the EXACT-like node. -- japhy
4044 if (HAS_TEXT(next) || JUMPABLE(next)) {
4046 regnode *text_node = next;
4048 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
4050 if (! HAS_TEXT(text_node))
4051 st->u.plus.c1 = st->u.plus.c2 = CHRTEST_VOID;
4053 if (PL_regkind[(U8)OP(text_node)] == REF) {
4054 st->u.plus.c1 = st->u.plus.c2 = CHRTEST_VOID;
4055 goto assume_ok_easy;
4057 else { s = (U8*)STRING(text_node); }
4060 st->u.plus.c2 = st->u.plus.c1 = *s;
4061 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
4062 st->u.plus.c2 = PL_fold[st->u.plus.c1];
4063 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
4064 st->u.plus.c2 = PL_fold_locale[st->u.plus.c1];
4067 if (OP(text_node) == EXACTF || OP(text_node) == REFF) {
4068 STRLEN ulen1, ulen2;
4069 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
4070 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
4072 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
4073 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
4075 st->u.plus.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
4077 st->u.plus.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
4081 st->u.plus.c2 = st->u.plus.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
4088 st->u.plus.c1 = st->u.plus.c2 = CHRTEST_VOID;
4090 PL_reginput = locinput;
4093 if (st->ln && regrepeat(rex, scan, st->ln) < st->ln)
4095 locinput = PL_reginput;
4096 REGCP_SET(st->u.plus.lastcp);
4097 if (st->u.plus.c1 != CHRTEST_VOID) {
4098 st->u.plus.old = locinput;
4099 st->u.plus.count = 0;
4101 if (n == REG_INFTY) {
4102 st->u.plus.e = PL_regeol - 1;
4104 while (UTF8_IS_CONTINUATION(*(U8*)st->u.plus.e))
4109 for (st->u.plus.e = locinput;
4110 m >0 && st->u.plus.e + UTF8SKIP(st->u.plus.e) <= PL_regeol; m--)
4111 st->u.plus.e += UTF8SKIP(st->u.plus.e);
4114 st->u.plus.e = locinput + n - st->ln;
4115 if (st->u.plus.e >= PL_regeol)
4116 st->u.plus.e = PL_regeol - 1;
4119 /* Find place 'next' could work */
4121 if (st->u.plus.c1 == st->u.plus.c2) {
4122 while (locinput <= st->u.plus.e &&
4123 UCHARAT(locinput) != st->u.plus.c1)
4126 while (locinput <= st->u.plus.e
4127 && UCHARAT(locinput) != st->u.plus.c1
4128 && UCHARAT(locinput) != st->u.plus.c2)
4131 st->u.plus.count = locinput - st->u.plus.old;
4134 if (st->u.plus.c1 == st->u.plus.c2) {
4136 /* count initialised to
4137 * utf8_distance(old, locinput) */
4138 while (locinput <= st->u.plus.e &&
4139 utf8n_to_uvchr((U8*)locinput,
4140 UTF8_MAXBYTES, &len,
4141 uniflags) != (UV)st->u.plus.c1) {
4146 /* count initialised to
4147 * utf8_distance(old, locinput) */
4148 while (locinput <= st->u.plus.e) {
4150 const UV c = utf8n_to_uvchr((U8*)locinput,
4151 UTF8_MAXBYTES, &len,
4153 if (c == (UV)st->u.plus.c1 || c == (UV)st->u.plus.c2)
4160 if (locinput > st->u.plus.e)
4162 /* PL_reginput == old now */
4163 if (locinput != st->u.plus.old) {
4164 st->ln = 1; /* Did some */
4165 if (regrepeat(rex, scan, st->u.plus.count) < st->u.plus.count)
4168 /* PL_reginput == locinput now */
4169 TRYPAREN(st->u.plus.paren, st->ln, locinput, PLUS1);
4170 /*** all unsaved local vars undefined at this point */
4171 PL_reginput = locinput; /* Could be reset... */
4172 REGCP_UNWIND(st->u.plus.lastcp);
4173 /* Couldn't or didn't -- move forward. */
4174 st->u.plus.old = locinput;
4176 locinput += UTF8SKIP(locinput);
4179 st->u.plus.count = 1;
4183 while (n >= st->ln || (n == REG_INFTY && st->ln > 0)) { /* ln overflow ? */
4185 if (st->u.plus.c1 != CHRTEST_VOID) {
4187 c = utf8n_to_uvchr((U8*)PL_reginput,
4191 c = UCHARAT(PL_reginput);
4192 /* If it could work, try it. */
4193 if (c == (UV)st->u.plus.c1 || c == (UV)st->u.plus.c2)
4195 TRYPAREN(st->u.plus.paren, st->ln, PL_reginput, PLUS2);
4196 /*** all unsaved local vars undefined at this point */
4197 REGCP_UNWIND(st->u.plus.lastcp);
4200 /* If it could work, try it. */
4201 else if (st->u.plus.c1 == CHRTEST_VOID)
4203 TRYPAREN(st->u.plus.paren, st->ln, PL_reginput, PLUS3);
4204 /*** all unsaved local vars undefined at this point */
4205 REGCP_UNWIND(st->u.plus.lastcp);
4207 /* Couldn't or didn't -- move forward. */
4208 PL_reginput = locinput;
4209 if (regrepeat(rex, scan, 1)) {
4211 locinput = PL_reginput;
4218 n = regrepeat(rex, scan, n);
4219 locinput = PL_reginput;
4220 if (st->ln < n && PL_regkind[(U8)OP(next)] == EOL &&
4221 (OP(next) != MEOL ||
4222 OP(next) == SEOL || OP(next) == EOS))
4224 st->ln = n; /* why back off? */
4225 /* ...because $ and \Z can match before *and* after
4226 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4227 We should back off by one in this case. */
4228 if (UCHARAT(PL_reginput - 1) == '\n' && OP(next) != EOS)
4231 REGCP_SET(st->u.plus.lastcp);
4234 while (n >= st->ln) {
4235 if (st->u.plus.c1 != CHRTEST_VOID) {
4237 c = utf8n_to_uvchr((U8*)PL_reginput,
4241 c = UCHARAT(PL_reginput);
4243 /* If it could work, try it. */
4244 if (st->u.plus.c1 == CHRTEST_VOID || c == (UV)st->u.plus.c1 || c == (UV)st->u.plus.c2)
4246 TRYPAREN(st->u.plus.paren, n, PL_reginput, PLUS4);
4247 /*** all unsaved local vars undefined at this point */
4248 REGCP_UNWIND(st->u.plus.lastcp);
4250 /* Couldn't or didn't -- back up. */
4252 PL_reginput = locinput = HOPc(locinput, -1);
4259 if (locinput < reginfo->till) {
4260 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4261 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4263 (long)(locinput - PL_reg_starttry),
4264 (long)(reginfo->till - PL_reg_starttry),
4266 sayNO_FINAL; /* Cannot match: too short. */
4268 PL_reginput = locinput; /* put where regtry can find it */
4269 sayYES_FINAL; /* Success! */
4271 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
4273 PerlIO_printf(Perl_debug_log,
4274 "%*s %ssubpattern success...%s\n",
4275 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5]));
4276 PL_reginput = locinput; /* put where regtry can find it */
4277 sayYES_FINAL; /* Success! */
4279 case SUSPEND: /* (?>FOO) */
4280 st->u.ifmatch.wanted = 1;
4281 PL_reginput = locinput;
4284 case UNLESSM: /* -ve lookaround: (?!FOO), or with flags, (?<!foo) */
4285 st->u.ifmatch.wanted = 0;
4286 goto ifmatch_trivial_fail_test;
4288 case IFMATCH: /* +ve lookaround: (?=FOO), or with flags, (?<=foo) */
4289 st->u.ifmatch.wanted = 1;
4290 ifmatch_trivial_fail_test:
4292 char * const s = HOPBACKc(locinput, scan->flags);
4297 st->sw = 1 - st->u.ifmatch.wanted;
4299 else if (st->u.ifmatch.wanted)
4301 next = scan + ARG(scan);
4309 PL_reginput = locinput;
4312 /* resume to current state on success */
4313 st->u.yes.prev_yes_state = yes_state;
4315 PUSH_STATE(newst, resume_IFMATCH);
4317 next = NEXTOPER(NEXTOPER(scan));
4321 next = scan + ARG(scan);
4326 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
4327 PTR2UV(scan), OP(scan));
4328 Perl_croak(aTHX_ "regexp memory corruption");
4336 /* simulate recursively calling regmatch(), but without actually
4337 * recursing - ie save the current state on the heap rather than on
4338 * the stack, then re-enter the loop. This avoids complex regexes
4339 * blowing the processor stack */
4343 /* push new state */
4344 regmatch_state *oldst = st;
4348 /* grab the next free state slot */
4350 if (st > SLAB_LAST(PL_regmatch_slab))
4351 st = S_push_slab(aTHX);
4352 PL_regmatch_state = st;
4356 oldst->locinput = locinput;
4359 locinput = PL_reginput;
4360 nextchr = UCHARAT(locinput);
4374 * We get here only if there's trouble -- normally "case END" is
4375 * the terminating point.
4377 Perl_croak(aTHX_ "corrupted regexp pointers");
4384 /* we have successfully completed a subexpression, but we must now
4385 * pop to the state marked by yes_state and continue from there */
4387 /*XXX tmp for CURLYM*/
4388 regmatch_slab * const oslab = PL_regmatch_slab;
4389 regmatch_state * const ost = st;
4390 regmatch_state * const oys = yes_state;
4393 assert(st != yes_state);
4394 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
4395 || yes_state > SLAB_LAST(PL_regmatch_slab))
4397 /* not in this slab, pop slab */
4398 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
4399 PL_regmatch_slab = PL_regmatch_slab->prev;
4400 st = SLAB_LAST(PL_regmatch_slab);
4402 depth -= (st - yes_state);
4403 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "POP STATE TO (%d)\n", depth));
4405 yes_state = st->u.yes.prev_yes_state;
4406 PL_regmatch_state = st;
4408 switch (st->resume_state) {
4410 if (st->u.eval.toggleutf)
4411 PL_reg_flags ^= RF_utf8;
4413 rex = st->u.eval.prev_rex;
4414 /* XXXX This is too dramatic a measure... */
4416 /* Restore parens of the caller without popping the
4419 const I32 tmp = PL_savestack_ix;
4420 PL_savestack_ix = st->u.eval.lastcp;
4422 PL_savestack_ix = tmp;
4424 PL_reginput = locinput;
4425 /* continue at the node following the (??{...}) */
4429 case resume_IFMATCH:
4432 st->sw = st->u.ifmatch.wanted;
4434 else if (!st->u.ifmatch.wanted)
4437 if (OP(st->scan) == SUSPEND)
4438 locinput = PL_reginput;
4440 locinput = PL_reginput = st->locinput;
4441 nextchr = UCHARAT(locinput);
4443 next = st->scan + ARG(st->scan);
4444 if (next == st->scan)
4448 /* XXX tmp don't handle yes_state yet */
4449 case resume_CURLYM1:
4450 case resume_CURLYM2:
4451 case resume_CURLYM3:
4452 PL_regmatch_slab =oslab;
4454 PL_regmatch_state = st;
4457 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "XXX revering a CURLYM\n"));
4461 Perl_croak(aTHX_ "unexpected yes reume state");
4465 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
4466 PL_colors[4], PL_colors[5]));
4473 /* XXX this is duplicate(ish) code to that in the do_no section.
4474 * eventually a yes should just pop the stack back to the current
4477 /* restore previous state and re-enter */
4480 switch (st->resume_state) {
4482 goto resume_point_TRIE1;
4484 goto resume_point_TRIE2;
4486 goto resume_point_CURLYX;
4487 case resume_WHILEM1:
4488 goto resume_point_WHILEM1;
4489 case resume_WHILEM2:
4490 goto resume_point_WHILEM2;
4491 case resume_WHILEM3:
4492 goto resume_point_WHILEM3;
4493 case resume_WHILEM4:
4494 goto resume_point_WHILEM4;
4495 case resume_WHILEM5:
4496 goto resume_point_WHILEM5;
4497 case resume_WHILEM6:
4498 goto resume_point_WHILEM6;
4499 case resume_CURLYM1:
4500 goto resume_point_CURLYM1;
4501 case resume_CURLYM2:
4502 goto resume_point_CURLYM2;
4503 case resume_CURLYM3:
4504 goto resume_point_CURLYM3;
4506 goto resume_point_PLUS1;
4508 goto resume_point_PLUS2;
4510 goto resume_point_PLUS3;
4512 goto resume_point_PLUS4;
4514 case resume_IFMATCH:
4517 Perl_croak(aTHX_ "regexp resume memory corruption");
4524 PerlIO_printf(Perl_debug_log,
4525 "%*s %sfailed...%s\n",
4526 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])
4532 re_unwind_t * const uw = SSPTRt(st->unwind,re_unwind_t);
4535 case RE_UNWIND_BRANCH:
4536 case RE_UNWIND_BRANCHJ:
4538 re_unwind_branch_t * const uwb = &(uw->branch);
4539 const I32 lastparen = uwb->lastparen;
4541 REGCP_UNWIND(uwb->lastcp);
4542 for (n = *PL_reglastparen; n > lastparen; n--)
4544 *PL_reglastparen = n;
4545 scan = next = uwb->next;
4546 st->minmod = uwb->minmod;
4548 OP(scan) != (uwb->type == RE_UNWIND_BRANCH
4549 ? BRANCH : BRANCHJ) ) { /* Failure */
4550 st->unwind = uwb->prev;
4556 /* Have more choice yet. Reuse the same uwb. */
4557 if ((n = (uwb->type == RE_UNWIND_BRANCH
4558 ? NEXT_OFF(next) : ARG(next))))
4561 next = NULL; /* XXXX Needn't unwinding in this case... */
4563 next = NEXTOPER(scan);
4564 if (uwb->type == RE_UNWIND_BRANCHJ)
4565 next = NEXTOPER(next);
4566 locinput = uwb->locinput;
4567 nextchr = uwb->nextchr;
4569 PL_regindent = uwb->regindent;
4576 Perl_croak(aTHX_ "regexp unwind memory corruption");
4587 /* there's a previous state to backtrack to */
4589 switch (st->resume_state) {
4591 goto resume_point_TRIE1;
4593 goto resume_point_TRIE2;
4595 /* we have failed an (??{...}). Restore state to the outer re
4596 * then re-throw the failure */
4597 if (st->u.eval.toggleutf)
4598 PL_reg_flags ^= RF_utf8;
4600 rex = st->u.eval.prev_rex;
4601 yes_state = st->u.yes.prev_yes_state;
4603 /* XXXX This is too dramatic a measure... */
4606 PL_reginput = locinput;
4607 REGCP_UNWIND(st->u.eval.lastcp);
4612 goto resume_point_CURLYX;
4613 case resume_WHILEM1:
4614 goto resume_point_WHILEM1;
4615 case resume_WHILEM2:
4616 goto resume_point_WHILEM2;
4617 case resume_WHILEM3:
4618 goto resume_point_WHILEM3;
4619 case resume_WHILEM4:
4620 goto resume_point_WHILEM4;
4621 case resume_WHILEM5:
4622 goto resume_point_WHILEM5;
4623 case resume_WHILEM6:
4624 goto resume_point_WHILEM6;
4625 case resume_CURLYM1:
4626 goto resume_point_CURLYM1;
4627 case resume_CURLYM2:
4628 goto resume_point_CURLYM2;
4629 case resume_CURLYM3:
4630 goto resume_point_CURLYM3;
4631 case resume_IFMATCH:
4632 yes_state = st->u.yes.prev_yes_state;
4635 st->sw = !st->u.ifmatch.wanted;
4637 else if (st->u.ifmatch.wanted)
4640 assert(OP(scan) != SUSPEND); /* XXX DAPM tmp */
4641 locinput = PL_reginput = st->locinput;
4642 nextchr = UCHARAT(locinput);
4643 next = scan + ARG(scan);
4649 goto resume_point_PLUS1;
4651 goto resume_point_PLUS2;
4653 goto resume_point_PLUS3;
4655 goto resume_point_PLUS4;
4657 Perl_croak(aTHX_ "regexp resume memory corruption");
4663 /* restore original high-water mark */
4664 PL_regmatch_slab = orig_slab;
4665 PL_regmatch_state = orig_state;
4667 /* free all slabs above current one */
4668 if (orig_slab->next) {
4669 regmatch_slab *sl = orig_slab->next;
4670 orig_slab->next = NULL;
4672 regmatch_slab * const osl = sl;
4683 - regrepeat - repeatedly match something simple, report how many
4686 * [This routine now assumes that it will only match on things of length 1.
4687 * That was true before, but now we assume scan - reginput is the count,
4688 * rather than incrementing count on every character. [Er, except utf8.]]
4691 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max)
4694 register char *scan;
4696 register char *loceol = PL_regeol;
4697 register I32 hardcount = 0;
4698 register bool do_utf8 = PL_reg_match_utf8;
4701 if (max == REG_INFTY)
4703 else if (max < loceol - scan)
4704 loceol = scan + max;
4709 while (scan < loceol && hardcount < max && *scan != '\n') {
4710 scan += UTF8SKIP(scan);
4714 while (scan < loceol && *scan != '\n')
4721 while (scan < loceol && hardcount < max) {
4722 scan += UTF8SKIP(scan);
4732 case EXACT: /* length of string is 1 */
4734 while (scan < loceol && UCHARAT(scan) == c)
4737 case EXACTF: /* length of string is 1 */
4739 while (scan < loceol &&
4740 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
4743 case EXACTFL: /* length of string is 1 */
4744 PL_reg_flags |= RF_tainted;
4746 while (scan < loceol &&
4747 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
4753 while (hardcount < max && scan < loceol &&
4754 reginclass(prog, p, (U8*)scan, 0, do_utf8)) {
4755 scan += UTF8SKIP(scan);
4759 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
4766 LOAD_UTF8_CHARCLASS_ALNUM();
4767 while (hardcount < max && scan < loceol &&
4768 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4769 scan += UTF8SKIP(scan);
4773 while (scan < loceol && isALNUM(*scan))
4778 PL_reg_flags |= RF_tainted;
4781 while (hardcount < max && scan < loceol &&
4782 isALNUM_LC_utf8((U8*)scan)) {
4783 scan += UTF8SKIP(scan);
4787 while (scan < loceol && isALNUM_LC(*scan))
4794 LOAD_UTF8_CHARCLASS_ALNUM();
4795 while (hardcount < max && scan < loceol &&
4796 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4797 scan += UTF8SKIP(scan);
4801 while (scan < loceol && !isALNUM(*scan))
4806 PL_reg_flags |= RF_tainted;
4809 while (hardcount < max && scan < loceol &&
4810 !isALNUM_LC_utf8((U8*)scan)) {
4811 scan += UTF8SKIP(scan);
4815 while (scan < loceol && !isALNUM_LC(*scan))
4822 LOAD_UTF8_CHARCLASS_SPACE();
4823 while (hardcount < max && scan < loceol &&
4825 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4826 scan += UTF8SKIP(scan);
4830 while (scan < loceol && isSPACE(*scan))
4835 PL_reg_flags |= RF_tainted;
4838 while (hardcount < max && scan < loceol &&
4839 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4840 scan += UTF8SKIP(scan);
4844 while (scan < loceol && isSPACE_LC(*scan))
4851 LOAD_UTF8_CHARCLASS_SPACE();
4852 while (hardcount < max && scan < loceol &&
4854 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4855 scan += UTF8SKIP(scan);
4859 while (scan < loceol && !isSPACE(*scan))
4864 PL_reg_flags |= RF_tainted;
4867 while (hardcount < max && scan < loceol &&
4868 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4869 scan += UTF8SKIP(scan);
4873 while (scan < loceol && !isSPACE_LC(*scan))
4880 LOAD_UTF8_CHARCLASS_DIGIT();
4881 while (hardcount < max && scan < loceol &&
4882 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
4883 scan += UTF8SKIP(scan);
4887 while (scan < loceol && isDIGIT(*scan))
4894 LOAD_UTF8_CHARCLASS_DIGIT();
4895 while (hardcount < max && scan < loceol &&
4896 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
4897 scan += UTF8SKIP(scan);
4901 while (scan < loceol && !isDIGIT(*scan))
4905 default: /* Called on something of 0 width. */
4906 break; /* So match right here or not at all. */
4912 c = scan - PL_reginput;
4916 SV *re_debug_flags = NULL;
4917 SV * const prop = sv_newmortal();
4920 regprop(prog, prop, p);
4921 PerlIO_printf(Perl_debug_log,
4922 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
4923 REPORT_CODE_OFF+1, "", SvPVX_const(prop),(IV)c,(IV)max);
4932 - regclass_swash - prepare the utf8 swash
4936 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
4942 const struct reg_data *data = prog ? prog->data : NULL;
4944 if (data && data->count) {
4945 const U32 n = ARG(node);
4947 if (data->what[n] == 's') {
4948 SV * const rv = (SV*)data->data[n];
4949 AV * const av = (AV*)SvRV((SV*)rv);
4950 SV **const ary = AvARRAY(av);
4953 /* See the end of regcomp.c:S_regclass() for
4954 * documentation of these array elements. */
4957 a = SvROK(ary[1]) ? &ary[1] : 0;
4958 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
4962 else if (si && doinit) {
4963 sw = swash_init("utf8", "", si, 1, 0);
4964 (void)av_store(av, 1, sw);
4980 - reginclass - determine if a character falls into a character class
4982 The n is the ANYOF regnode, the p is the target string, lenp
4983 is pointer to the maximum length of how far to go in the p
4984 (if the lenp is zero, UTF8SKIP(p) is used),
4985 do_utf8 tells whether the target string is in UTF-8.
4990 S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
4993 const char flags = ANYOF_FLAGS(n);
4999 if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
5000 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
5001 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
5002 /* see [perl #37836] for UTF8_ALLOW_ANYUV */
5003 if (len == (STRLEN)-1)
5004 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
5007 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
5008 if (do_utf8 || (flags & ANYOF_UNICODE)) {
5011 if (do_utf8 && !ANYOF_RUNTIME(n)) {
5012 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
5015 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
5019 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
5022 if (swash_fetch(sw, p, do_utf8))
5024 else if (flags & ANYOF_FOLD) {
5025 if (!match && lenp && av) {
5027 for (i = 0; i <= av_len(av); i++) {
5028 SV* const sv = *av_fetch(av, i, FALSE);
5030 const char * const s = SvPV_const(sv, len);
5032 if (len <= plen && memEQ(s, (char*)p, len)) {
5040 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5043 to_utf8_fold(p, tmpbuf, &tmplen);
5044 if (swash_fetch(sw, tmpbuf, do_utf8))
5050 if (match && lenp && *lenp == 0)
5051 *lenp = UNISKIP(NATIVE_TO_UNI(c));
5053 if (!match && c < 256) {
5054 if (ANYOF_BITMAP_TEST(n, c))
5056 else if (flags & ANYOF_FOLD) {
5059 if (flags & ANYOF_LOCALE) {
5060 PL_reg_flags |= RF_tainted;
5061 f = PL_fold_locale[c];
5065 if (f != c && ANYOF_BITMAP_TEST(n, f))
5069 if (!match && (flags & ANYOF_CLASS)) {
5070 PL_reg_flags |= RF_tainted;
5072 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
5073 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
5074 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
5075 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
5076 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
5077 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
5078 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
5079 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
5080 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
5081 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
5082 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
5083 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
5084 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
5085 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
5086 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
5087 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
5088 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
5089 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
5090 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
5091 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
5092 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
5093 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
5094 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
5095 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
5096 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
5097 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
5098 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
5099 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
5100 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
5101 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
5102 ) /* How's that for a conditional? */
5109 return (flags & ANYOF_INVERT) ? !match : match;
5113 S_reghop3(U8 *s, I32 off, U8* lim)
5117 while (off-- && s < lim) {
5118 /* XXX could check well-formedness here */
5126 if (UTF8_IS_CONTINUED(*s)) {
5127 while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s))
5130 /* XXX could check well-formedness here */
5138 S_reghopmaybe3(U8* s, I32 off, U8* lim)
5142 while (off-- && s < lim) {
5143 /* XXX could check well-formedness here */
5153 if (UTF8_IS_CONTINUED(*s)) {
5154 while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s))
5157 /* XXX could check well-formedness here */
5169 restore_pos(pTHX_ void *arg)
5172 regexp * const rex = (regexp *)arg;
5173 if (PL_reg_eval_set) {
5174 if (PL_reg_oldsaved) {
5175 rex->subbeg = PL_reg_oldsaved;
5176 rex->sublen = PL_reg_oldsavedlen;
5177 #ifdef PERL_OLD_COPY_ON_WRITE
5178 rex->saved_copy = PL_nrs;
5180 RX_MATCH_COPIED_on(rex);
5182 PL_reg_magic->mg_len = PL_reg_oldpos;
5183 PL_reg_eval_set = 0;
5184 PL_curpm = PL_reg_oldcurpm;
5189 S_to_utf8_substr(pTHX_ register regexp *prog)
5191 if (prog->float_substr && !prog->float_utf8) {
5192 SV* const sv = newSVsv(prog->float_substr);
5193 prog->float_utf8 = sv;
5194 sv_utf8_upgrade(sv);
5195 if (SvTAIL(prog->float_substr))
5197 if (prog->float_substr == prog->check_substr)
5198 prog->check_utf8 = sv;
5200 if (prog->anchored_substr && !prog->anchored_utf8) {
5201 SV* const sv = newSVsv(prog->anchored_substr);
5202 prog->anchored_utf8 = sv;
5203 sv_utf8_upgrade(sv);
5204 if (SvTAIL(prog->anchored_substr))
5206 if (prog->anchored_substr == prog->check_substr)
5207 prog->check_utf8 = sv;
5212 S_to_byte_substr(pTHX_ register regexp *prog)
5215 if (prog->float_utf8 && !prog->float_substr) {
5216 SV* sv = newSVsv(prog->float_utf8);
5217 prog->float_substr = sv;
5218 if (sv_utf8_downgrade(sv, TRUE)) {
5219 if (SvTAIL(prog->float_utf8))
5223 prog->float_substr = sv = &PL_sv_undef;
5225 if (prog->float_utf8 == prog->check_utf8)
5226 prog->check_substr = sv;
5228 if (prog->anchored_utf8 && !prog->anchored_substr) {
5229 SV* sv = newSVsv(prog->anchored_utf8);
5230 prog->anchored_substr = sv;
5231 if (sv_utf8_downgrade(sv, TRUE)) {
5232 if (SvTAIL(prog->anchored_utf8))
5236 prog->anchored_substr = sv = &PL_sv_undef;
5238 if (prog->anchored_utf8 == prog->check_utf8)
5239 prog->check_substr = sv;
5245 * c-indentation-style: bsd
5247 * indent-tabs-mode: t
5250 * ex: set ts=8 sts=4 sw=4 noet: