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
60 * pregcomp and pregexec -- regsub and regerror are not used in perl
62 * Copyright (c) 1986 by University of Toronto.
63 * Written by Henry Spencer. Not derived from licensed software.
65 * Permission is granted to anyone to use this software for any
66 * purpose on any computer system, and to redistribute it freely,
67 * subject to the following restrictions:
69 * 1. The author is not responsible for the consequences of use of
70 * this software, no matter how awful, even if they arise
73 * 2. The origin of this software must not be misrepresented, either
74 * by explicit claim or by omission.
76 * 3. Altered versions must be plainly marked as such, and must not
77 * be misrepresented as being the original software.
79 **** Alterations to Henry's code are...
81 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
82 **** 2000, 2001, 2002, 2003, 2004, 2005, by Larry Wall and others
84 **** You may distribute under the terms of either the GNU General Public
85 **** License or the Artistic License, as specified in the README file.
87 * Beware that some of this code is subtly aware of the way operator
88 * precedence is structured in regular expressions. Serious changes in
89 * regular-expression syntax might require a total rethink.
92 #define PERL_IN_REGEXEC_C
97 #define RF_tainted 1 /* tainted information used? */
98 #define RF_warned 2 /* warned about big count? */
99 #define RF_evaled 4 /* Did an EVAL with setting? */
100 #define RF_utf8 8 /* String contains multibyte chars? */
102 #define UTF ((PL_reg_flags & RF_utf8) != 0)
104 #define RS_init 1 /* eval environment created */
105 #define RS_set 2 /* replsv value is set */
108 #define STATIC static
111 #define REGINCLASS(p,c) (ANYOF_FLAGS(p) ? reginclass(p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
117 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
118 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
120 #define reghop_c(pos,off) ((char*)reghop((U8*)pos, off))
121 #define reghopmaybe_c(pos,off) ((char*)reghopmaybe((U8*)pos, off))
122 #define HOP(pos,off) (PL_reg_match_utf8 ? reghop((U8*)pos, off) : (U8*)(pos + off))
123 #define HOPMAYBE(pos,off) (PL_reg_match_utf8 ? reghopmaybe((U8*)pos, off) : (U8*)(pos + off))
124 #define HOPc(pos,off) ((char*)HOP(pos,off))
125 #define HOPMAYBEc(pos,off) ((char*)HOPMAYBE(pos,off))
127 #define HOPBACK(pos, off) ( \
128 (PL_reg_match_utf8) \
129 ? reghopmaybe((U8*)pos, -off) \
130 : (pos - off >= PL_bostr) \
134 #define HOPBACKc(pos, off) (char*)HOPBACK(pos, off)
136 #define reghop3_c(pos,off,lim) ((char*)reghop3((U8*)pos, off, (U8*)lim))
137 #define reghopmaybe3_c(pos,off,lim) ((char*)reghopmaybe3((U8*)pos, off, (U8*)lim))
138 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off))
139 #define HOPMAYBE3(pos,off,lim) (PL_reg_match_utf8 ? reghopmaybe3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off))
140 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
141 #define HOPMAYBE3c(pos,off,lim) ((char*)HOPMAYBE3(pos,off,lim))
143 #define LOAD_UTF8_CHARCLASS(a,b) STMT_START { if (!CAT2(PL_utf8_,a)) { ENTER; save_re_context(); (void)CAT2(is_utf8_, a)((const U8*)b); LEAVE; } } STMT_END
145 /* for use after a quantifier and before an EXACT-like node -- japhy */
146 #define JUMPABLE(rn) ( \
147 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
148 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
149 OP(rn) == PLUS || OP(rn) == MINMOD || \
150 (PL_regkind[(U8)OP(rn)] == CURLY && ARG1(rn) > 0) \
153 #define HAS_TEXT(rn) ( \
154 PL_regkind[(U8)OP(rn)] == EXACT || PL_regkind[(U8)OP(rn)] == REF \
158 Search for mandatory following text node; for lookahead, the text must
159 follow but for lookbehind (rn->flags != 0) we skip to the next step.
161 #define FIND_NEXT_IMPT(rn) STMT_START { \
162 while (JUMPABLE(rn)) \
163 if (OP(rn) == SUSPEND || PL_regkind[(U8)OP(rn)] == CURLY) \
164 rn = NEXTOPER(NEXTOPER(rn)); \
165 else if (OP(rn) == PLUS) \
167 else if (OP(rn) == IFMATCH) \
168 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
169 else rn += NEXT_OFF(rn); \
172 static void restore_pos(pTHX_ void *arg);
175 S_regcppush(pTHX_ I32 parenfloor)
177 int retval = PL_savestack_ix;
178 #define REGCP_PAREN_ELEMS 4
179 int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
182 if (paren_elems_to_push < 0)
183 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
185 #define REGCP_OTHER_ELEMS 6
186 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
187 for (p = PL_regsize; p > parenfloor; p--) {
188 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
189 SSPUSHINT(PL_regendp[p]);
190 SSPUSHINT(PL_regstartp[p]);
191 SSPUSHPTR(PL_reg_start_tmp[p]);
194 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
195 SSPUSHINT(PL_regsize);
196 SSPUSHINT(*PL_reglastparen);
197 SSPUSHINT(*PL_reglastcloseparen);
198 SSPUSHPTR(PL_reginput);
199 #define REGCP_FRAME_ELEMS 2
200 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
201 * are needed for the regexp context stack bookkeeping. */
202 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
203 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
208 /* These are needed since we do not localize EVAL nodes: */
209 # define REGCP_SET(cp) DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, \
210 " Setting an EVAL scope, savestack=%"IVdf"\n", \
211 (IV)PL_savestack_ix)); cp = PL_savestack_ix
213 # define REGCP_UNWIND(cp) DEBUG_EXECUTE_r(cp != PL_savestack_ix ? \
214 PerlIO_printf(Perl_debug_log, \
215 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
216 (IV)(cp), (IV)PL_savestack_ix) : 0); regcpblow(cp)
226 GET_RE_DEBUG_FLAGS_DECL;
228 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
230 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
231 i = SSPOPINT; /* Parentheses elements to pop. */
232 input = (char *) SSPOPPTR;
233 *PL_reglastcloseparen = SSPOPINT;
234 *PL_reglastparen = SSPOPINT;
235 PL_regsize = SSPOPINT;
237 /* Now restore the parentheses context. */
238 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
239 i > 0; i -= REGCP_PAREN_ELEMS) {
240 paren = (U32)SSPOPINT;
241 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
242 PL_regstartp[paren] = SSPOPINT;
244 if (paren <= *PL_reglastparen)
245 PL_regendp[paren] = tmps;
247 PerlIO_printf(Perl_debug_log,
248 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
249 (UV)paren, (IV)PL_regstartp[paren],
250 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
251 (IV)PL_regendp[paren],
252 (paren > *PL_reglastparen ? "(no)" : ""));
256 if ((I32)(*PL_reglastparen + 1) <= PL_regnpar) {
257 PerlIO_printf(Perl_debug_log,
258 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
259 (IV)(*PL_reglastparen + 1), (IV)PL_regnpar);
263 /* It would seem that the similar code in regtry()
264 * already takes care of this, and in fact it is in
265 * a better location to since this code can #if 0-ed out
266 * but the code in regtry() is needed or otherwise tests
267 * requiring null fields (pat.t#187 and split.t#{13,14}
268 * (as of patchlevel 7877) will fail. Then again,
269 * this code seems to be necessary or otherwise
270 * building DynaLoader will fail:
271 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
273 for (paren = *PL_reglastparen + 1; (I32)paren <= PL_regnpar; paren++) {
274 if ((I32)paren > PL_regsize)
275 PL_regstartp[paren] = -1;
276 PL_regendp[paren] = -1;
283 S_regcp_set_to(pTHX_ I32 ss)
285 I32 tmp = PL_savestack_ix;
287 PL_savestack_ix = ss;
289 PL_savestack_ix = tmp;
293 typedef struct re_cc_state
297 struct re_cc_state *prev;
302 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
304 #define TRYPAREN(paren, n, input) { \
307 PL_regstartp[paren] = HOPc(input, -1) - PL_bostr; \
308 PL_regendp[paren] = input - PL_bostr; \
311 PL_regendp[paren] = -1; \
313 if (regmatch(next)) \
316 PL_regendp[paren] = -1; \
321 * pregexec and friends
325 - pregexec - match a regexp against a string
328 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
329 char *strbeg, I32 minend, SV *screamer, U32 nosave)
330 /* strend: pointer to null at end of string */
331 /* strbeg: real beginning of string */
332 /* minend: end of match must be >=minend after stringarg. */
333 /* nosave: For optimizations. */
336 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
337 nosave ? 0 : REXEC_COPY_STR);
341 S_cache_re(pTHX_ regexp *prog)
343 PL_regprecomp = prog->precomp; /* Needed for FAIL. */
345 PL_regprogram = prog->program;
347 PL_regnpar = prog->nparens;
348 PL_regdata = prog->data;
353 * Need to implement the following flags for reg_anch:
355 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
357 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
358 * INTUIT_AUTORITATIVE_ML
359 * INTUIT_ONCE_NOML - Intuit can match in one location only.
362 * Another flag for this function: SECOND_TIME (so that float substrs
363 * with giant delta may be not rechecked).
366 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
368 /* If SCREAM, then SvPVX(sv) should be compatible with strpos and strend.
369 Otherwise, only SvCUR(sv) is used to get strbeg. */
371 /* XXXX We assume that strpos is strbeg unless sv. */
373 /* XXXX Some places assume that there is a fixed substring.
374 An update may be needed if optimizer marks as "INTUITable"
375 RExen without fixed substrings. Similarly, it is assumed that
376 lengths of all the strings are no more than minlen, thus they
377 cannot come from lookahead.
378 (Or minlen should take into account lookahead.) */
380 /* A failure to find a constant substring means that there is no need to make
381 an expensive call to REx engine, thus we celebrate a failure. Similarly,
382 finding a substring too deep into the string means that less calls to
383 regtry() should be needed.
385 REx compiler's optimizer found 4 possible hints:
386 a) Anchored substring;
388 c) Whether we are anchored (beginning-of-line or \G);
389 d) First node (of those at offset 0) which may distingush positions;
390 We use a)b)d) and multiline-part of c), and try to find a position in the
391 string which does not contradict any of them.
394 /* Most of decisions we do here should have been done at compile time.
395 The nodes of the REx which we used for the search should have been
396 deleted from the finite automaton. */
399 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
400 char *strend, U32 flags, re_scream_pos_data *data)
402 register I32 start_shift = 0;
403 /* Should be nonnegative! */
404 register I32 end_shift = 0;
409 int do_utf8 = sv ? SvUTF8(sv) : 0; /* if no sv we have to assume bytes */
411 register char *other_last = Nullch; /* other substr checked before this */
412 char *check_at = Nullch; /* check substr found at this pos */
413 I32 multiline = prog->reganch & PMf_MULTILINE;
415 char *i_strpos = strpos;
416 SV *dsv = PERL_DEBUG_PAD_ZERO(0);
419 GET_RE_DEBUG_FLAGS_DECL;
421 RX_MATCH_UTF8_set(prog,do_utf8);
423 if (prog->reganch & ROPT_UTF8) {
424 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
425 "UTF-8 regex...\n"));
426 PL_reg_flags |= RF_utf8;
430 char *s = PL_reg_match_utf8 ?
431 sv_uni_display(dsv, sv, 60, UNI_DISPLAY_REGEX) :
433 int len = PL_reg_match_utf8 ?
434 strlen(s) : strend - strpos;
437 if (PL_reg_match_utf8)
438 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
439 "UTF-8 target...\n"));
440 PerlIO_printf(Perl_debug_log,
441 "%sGuessing start of match, REx%s `%s%.60s%s%s' against `%s%.*s%s%s'...\n",
442 PL_colors[4],PL_colors[5],PL_colors[0],
445 (strlen(prog->precomp) > 60 ? "..." : ""),
447 (int)(len > 60 ? 60 : len),
449 (len > 60 ? "..." : "")
453 /* CHR_DIST() would be more correct here but it makes things slow. */
454 if (prog->minlen > strend - strpos) {
455 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
456 "String too short... [re_intuit_start]\n"));
459 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
462 if (!prog->check_utf8 && prog->check_substr)
463 to_utf8_substr(prog);
464 check = prog->check_utf8;
466 if (!prog->check_substr && prog->check_utf8)
467 to_byte_substr(prog);
468 check = prog->check_substr;
470 if (check == &PL_sv_undef) {
471 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
472 "Non-utf string cannot match utf check string\n"));
475 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
476 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
477 || ( (prog->reganch & ROPT_ANCH_BOL)
478 && !multiline ) ); /* Check after \n? */
481 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
482 | ROPT_IMPLICIT)) /* not a real BOL */
483 /* SvCUR is not set on references: SvRV and SvPVX overlap */
485 && (strpos != strbeg)) {
486 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
489 if (prog->check_offset_min == prog->check_offset_max &&
490 !(prog->reganch & ROPT_CANY_SEEN)) {
491 /* Substring at constant offset from beg-of-str... */
494 s = HOP3c(strpos, prog->check_offset_min, strend);
496 slen = SvCUR(check); /* >= 1 */
498 if ( strend - s > slen || strend - s < slen - 1
499 || (strend - s == slen && strend[-1] != '\n')) {
500 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
503 /* Now should match s[0..slen-2] */
505 if (slen && (*SvPVX(check) != *s
507 && memNE(SvPVX(check), s, slen)))) {
509 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
513 else if (*SvPVX(check) != *s
514 || ((slen = SvCUR(check)) > 1
515 && memNE(SvPVX(check), s, slen)))
517 goto success_at_start;
520 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
522 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
523 end_shift = prog->minlen - start_shift -
524 CHR_SVLEN(check) + (SvTAIL(check) != 0);
526 I32 end = prog->check_offset_max + CHR_SVLEN(check)
527 - (SvTAIL(check) != 0);
528 I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
530 if (end_shift < eshift)
534 else { /* Can match at random position */
537 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
538 /* Should be nonnegative! */
539 end_shift = prog->minlen - start_shift -
540 CHR_SVLEN(check) + (SvTAIL(check) != 0);
543 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
545 Perl_croak(aTHX_ "panic: end_shift");
549 /* Find a possible match in the region s..strend by looking for
550 the "check" substring in the region corrected by start/end_shift. */
551 if (flags & REXEC_SCREAM) {
552 I32 p = -1; /* Internal iterator of scream. */
553 I32 *pp = data ? data->scream_pos : &p;
555 if (PL_screamfirst[BmRARE(check)] >= 0
556 || ( BmRARE(check) == '\n'
557 && (BmPREVIOUS(check) == SvCUR(check) - 1)
559 s = screaminstr(sv, check,
560 start_shift + (s - strbeg), end_shift, pp, 0);
563 /* we may be pointing at the wrong string */
564 if (s && RX_MATCH_COPIED(prog))
565 s = strbeg + (s - SvPVX(sv));
567 *data->scream_olds = s;
569 else if (prog->reganch & ROPT_CANY_SEEN)
570 s = fbm_instr((U8*)(s + start_shift),
571 (U8*)(strend - end_shift),
572 check, multiline ? FBMrf_MULTILINE : 0);
574 s = fbm_instr(HOP3(s, start_shift, strend),
575 HOP3(strend, -end_shift, strbeg),
576 check, multiline ? FBMrf_MULTILINE : 0);
578 /* Update the count-of-usability, remove useless subpatterns,
581 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s %s substr `%s%.*s%s'%s%s",
582 (s ? "Found" : "Did not find"),
583 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) ? "anchored" : "floating"),
585 (int)(SvCUR(check) - (SvTAIL(check)!=0)),
587 PL_colors[1], (SvTAIL(check) ? "$" : ""),
588 (s ? " at offset " : "...\n") ) );
595 /* Finish the diagnostic message */
596 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
598 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
599 Start with the other substr.
600 XXXX no SCREAM optimization yet - and a very coarse implementation
601 XXXX /ttx+/ results in anchored=`ttx', floating=`x'. floating will
602 *always* match. Probably should be marked during compile...
603 Probably it is right to do no SCREAM here...
606 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) : (prog->float_substr && prog->anchored_substr)) {
607 /* Take into account the "other" substring. */
608 /* XXXX May be hopelessly wrong for UTF... */
611 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
614 char *last = HOP3c(s, -start_shift, strbeg), *last1, *last2;
618 t = s - prog->check_offset_max;
619 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
621 || ((t = reghopmaybe3_c(s, -(prog->check_offset_max), strpos))
626 t = HOP3c(t, prog->anchored_offset, strend);
627 if (t < other_last) /* These positions already checked */
629 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
632 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
633 /* On end-of-str: see comment below. */
634 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
635 if (must == &PL_sv_undef) {
637 DEBUG_EXECUTE_r(must = prog->anchored_utf8); /* for debug */
642 HOP3(HOP3(last1, prog->anchored_offset, strend)
643 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
645 multiline ? FBMrf_MULTILINE : 0
647 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
648 "%s anchored substr `%s%.*s%s'%s",
649 (s ? "Found" : "Contradicts"),
652 - (SvTAIL(must)!=0)),
654 PL_colors[1], (SvTAIL(must) ? "$" : "")));
656 if (last1 >= last2) {
657 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
658 ", giving up...\n"));
661 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
662 ", trying floating at offset %ld...\n",
663 (long)(HOP3c(s1, 1, strend) - i_strpos)));
664 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
665 s = HOP3c(last, 1, strend);
669 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
670 (long)(s - i_strpos)));
671 t = HOP3c(s, -prog->anchored_offset, strbeg);
672 other_last = HOP3c(s, 1, strend);
680 else { /* Take into account the floating substring. */
685 t = HOP3c(s, -start_shift, strbeg);
687 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
688 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
689 last = HOP3c(t, prog->float_max_offset, strend);
690 s = HOP3c(t, prog->float_min_offset, strend);
693 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
694 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
695 /* fbm_instr() takes into account exact value of end-of-str
696 if the check is SvTAIL(ed). Since false positives are OK,
697 and end-of-str is not later than strend we are OK. */
698 if (must == &PL_sv_undef) {
700 DEBUG_EXECUTE_r(must = prog->float_utf8); /* for debug message */
703 s = fbm_instr((unsigned char*)s,
704 (unsigned char*)last + SvCUR(must)
706 must, multiline ? FBMrf_MULTILINE : 0);
707 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s floating substr `%s%.*s%s'%s",
708 (s ? "Found" : "Contradicts"),
710 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
712 PL_colors[1], (SvTAIL(must) ? "$" : "")));
715 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
716 ", giving up...\n"));
719 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
720 ", trying anchored starting at offset %ld...\n",
721 (long)(s1 + 1 - i_strpos)));
723 s = HOP3c(t, 1, strend);
727 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
728 (long)(s - i_strpos)));
729 other_last = s; /* Fix this later. --Hugo */
738 t = s - prog->check_offset_max;
739 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
741 || ((t = reghopmaybe3_c(s, -prog->check_offset_max, strpos))
743 /* Fixed substring is found far enough so that the match
744 cannot start at strpos. */
746 if (ml_anch && t[-1] != '\n') {
747 /* Eventually fbm_*() should handle this, but often
748 anchored_offset is not 0, so this check will not be wasted. */
749 /* XXXX In the code below we prefer to look for "^" even in
750 presence of anchored substrings. And we search even
751 beyond the found float position. These pessimizations
752 are historical artefacts only. */
754 while (t < strend - prog->minlen) {
756 if (t < check_at - prog->check_offset_min) {
757 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
758 /* Since we moved from the found position,
759 we definitely contradict the found anchored
760 substr. Due to the above check we do not
761 contradict "check" substr.
762 Thus we can arrive here only if check substr
763 is float. Redo checking for "other"=="fixed".
766 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
767 PL_colors[0],PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
768 goto do_other_anchored;
770 /* We don't contradict the found floating substring. */
771 /* XXXX Why not check for STCLASS? */
773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
774 PL_colors[0],PL_colors[1], (long)(s - i_strpos)));
777 /* Position contradicts check-string */
778 /* XXXX probably better to look for check-string
779 than for "\n", so one should lower the limit for t? */
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
781 PL_colors[0],PL_colors[1], (long)(t + 1 - i_strpos)));
782 other_last = strpos = s = t + 1;
787 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
788 PL_colors[0],PL_colors[1]));
792 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
793 PL_colors[0],PL_colors[1]));
797 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
800 /* The found string does not prohibit matching at strpos,
801 - no optimization of calling REx engine can be performed,
802 unless it was an MBOL and we are not after MBOL,
803 or a future STCLASS check will fail this. */
805 /* Even in this situation we may use MBOL flag if strpos is offset
806 wrt the start of the string. */
807 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
808 && (strpos != strbeg) && strpos[-1] != '\n'
809 /* May be due to an implicit anchor of m{.*foo} */
810 && !(prog->reganch & ROPT_IMPLICIT))
815 DEBUG_EXECUTE_r( if (ml_anch)
816 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
817 (long)(strpos - i_strpos), PL_colors[0],PL_colors[1]);
820 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
822 prog->check_utf8 /* Could be deleted already */
823 && --BmUSEFUL(prog->check_utf8) < 0
824 && (prog->check_utf8 == prog->float_utf8)
826 prog->check_substr /* Could be deleted already */
827 && --BmUSEFUL(prog->check_substr) < 0
828 && (prog->check_substr == prog->float_substr)
831 /* If flags & SOMETHING - do not do it many times on the same match */
832 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
833 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
834 if (do_utf8 ? prog->check_substr : prog->check_utf8)
835 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
836 prog->check_substr = prog->check_utf8 = Nullsv; /* disable */
837 prog->float_substr = prog->float_utf8 = Nullsv; /* clear */
838 check = Nullsv; /* abort */
840 /* XXXX This is a remnant of the old implementation. It
841 looks wasteful, since now INTUIT can use many
843 prog->reganch &= ~RE_USE_INTUIT;
850 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
851 if (prog->regstclass) {
852 /* minlen == 0 is possible if regstclass is \b or \B,
853 and the fixed substr is ''$.
854 Since minlen is already taken into account, s+1 is before strend;
855 accidentally, minlen >= 1 guaranties no false positives at s + 1
856 even for \b or \B. But (minlen? 1 : 0) below assumes that
857 regstclass does not come from lookahead... */
858 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
859 This leaves EXACTF only, which is dealt with in find_byclass(). */
860 const U8* str = (U8*)STRING(prog->regstclass);
861 const int cl_l = (PL_regkind[(U8)OP(prog->regstclass)] == EXACT
862 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
864 char *endpos = (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
865 ? HOP3c(s, (prog->minlen ? cl_l : 0), strend)
866 : (prog->float_substr || prog->float_utf8
867 ? HOP3c(HOP3c(check_at, -start_shift, strbeg),
873 s = find_byclass(prog, prog->regstclass, s, endpos, 1);
876 const char *what = 0;
878 if (endpos == strend) {
879 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
880 "Could not match STCLASS...\n") );
883 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
884 "This position contradicts STCLASS...\n") );
885 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
887 /* Contradict one of substrings */
888 if (prog->anchored_substr || prog->anchored_utf8) {
889 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
890 DEBUG_EXECUTE_r( what = "anchored" );
892 s = HOP3c(t, 1, strend);
893 if (s + start_shift + end_shift > strend) {
894 /* XXXX Should be taken into account earlier? */
895 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
896 "Could not match STCLASS...\n") );
901 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
902 "Looking for %s substr starting at offset %ld...\n",
903 what, (long)(s + start_shift - i_strpos)) );
906 /* Have both, check_string is floating */
907 if (t + start_shift >= check_at) /* Contradicts floating=check */
908 goto retry_floating_check;
909 /* Recheck anchored substring, but not floating... */
913 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
914 "Looking for anchored substr starting at offset %ld...\n",
915 (long)(other_last - i_strpos)) );
916 goto do_other_anchored;
918 /* Another way we could have checked stclass at the
919 current position only: */
924 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
925 "Looking for /%s^%s/m starting at offset %ld...\n",
926 PL_colors[0],PL_colors[1], (long)(t - i_strpos)) );
929 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
931 /* Check is floating subtring. */
932 retry_floating_check:
933 t = check_at - start_shift;
934 DEBUG_EXECUTE_r( what = "floating" );
935 goto hop_and_restart;
938 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
939 "By STCLASS: moving %ld --> %ld\n",
940 (long)(t - i_strpos), (long)(s - i_strpos))
944 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
945 "Does not contradict STCLASS...\n");
950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
951 PL_colors[4], (check ? "Guessed" : "Giving up"),
952 PL_colors[5], (long)(s - i_strpos)) );
955 fail_finish: /* Substring not found */
956 if (prog->check_substr || prog->check_utf8) /* could be removed already */
957 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
959 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
960 PL_colors[4],PL_colors[5]));
964 /* We know what class REx starts with. Try to find this position... */
966 S_find_byclass(pTHX_ regexp * prog, regnode *c, char *s, char *strend, I32 norun)
968 I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
972 register STRLEN uskip;
976 register I32 tmp = 1; /* Scratch variable? */
977 register bool do_utf8 = PL_reg_match_utf8;
979 /* We know what class it must start with. */
983 while (s + (uskip = UTF8SKIP(s)) <= strend) {
984 if ((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
985 !UTF8_IS_INVARIANT((U8)s[0]) ?
986 reginclass(c, (U8*)s, 0, do_utf8) :
987 REGINCLASS(c, (U8*)s)) {
988 if (tmp && (norun || regtry(prog, s)))
1002 if (REGINCLASS(c, (U8*)s) ||
1003 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1004 /* The assignment of 2 is intentional:
1005 * for the folded sharp s, the skip is 2. */
1006 (skip = SHARP_S_SKIP))) {
1007 if (tmp && (norun || regtry(prog, s)))
1019 while (s < strend) {
1020 if (tmp && (norun || regtry(prog, s)))
1029 ln = STR_LEN(c); /* length to match in octets/bytes */
1030 lnc = (I32) ln; /* length to match in characters */
1032 STRLEN ulen1, ulen2;
1034 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1035 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1037 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1038 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1040 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1041 0, ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1042 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1043 0, ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1045 while (sm < ((U8 *) m + ln)) {
1060 c2 = PL_fold_locale[c1];
1062 e = HOP3c(strend, -((I32)lnc), s);
1065 e = s; /* Due to minlen logic of intuit() */
1067 /* The idea in the EXACTF* cases is to first find the
1068 * first character of the EXACTF* node and then, if
1069 * necessary, case-insensitively compare the full
1070 * text of the node. The c1 and c2 are the first
1071 * characters (though in Unicode it gets a bit
1072 * more complicated because there are more cases
1073 * than just upper and lower: one needs to use
1074 * the so-called folding case for case-insensitive
1075 * matching (called "loose matching" in Unicode).
1076 * ibcmp_utf8() will do just that. */
1080 U8 tmpbuf [UTF8_MAXBYTES+1];
1081 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1082 STRLEN len, foldlen;
1085 /* Upper and lower of 1st char are equal -
1086 * probably not a "letter". */
1088 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1090 0 : UTF8_ALLOW_ANY);
1093 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1094 m, (char **)0, ln, (bool)UTF))
1095 && (norun || regtry(prog, s)) )
1098 uvchr_to_utf8(tmpbuf, c);
1099 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1101 && (f == c1 || f == c2)
1102 && (ln == foldlen ||
1103 !ibcmp_utf8((char *) foldbuf,
1104 (char **)0, foldlen, do_utf8,
1106 (char **)0, ln, (bool)UTF))
1107 && (norun || regtry(prog, s)) )
1115 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1117 0 : UTF8_ALLOW_ANY);
1119 /* Handle some of the three Greek sigmas cases.
1120 * Note that not all the possible combinations
1121 * are handled here: some of them are handled
1122 * by the standard folding rules, and some of
1123 * them (the character class or ANYOF cases)
1124 * are handled during compiletime in
1125 * regexec.c:S_regclass(). */
1126 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1127 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1128 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1130 if ( (c == c1 || c == c2)
1132 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1133 m, (char **)0, ln, (bool)UTF))
1134 && (norun || regtry(prog, s)) )
1137 uvchr_to_utf8(tmpbuf, c);
1138 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1140 && (f == c1 || f == c2)
1141 && (ln == foldlen ||
1142 !ibcmp_utf8((char *) foldbuf,
1143 (char **)0, foldlen, do_utf8,
1145 (char **)0, ln, (bool)UTF))
1146 && (norun || regtry(prog, s)) )
1157 && (ln == 1 || !(OP(c) == EXACTF
1159 : ibcmp_locale(s, m, ln)))
1160 && (norun || regtry(prog, s)) )
1166 if ( (*(U8*)s == c1 || *(U8*)s == c2)
1167 && (ln == 1 || !(OP(c) == EXACTF
1169 : ibcmp_locale(s, m, ln)))
1170 && (norun || regtry(prog, s)) )
1177 PL_reg_flags |= RF_tainted;
1184 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1186 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1188 tmp = ((OP(c) == BOUND ?
1189 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1190 LOAD_UTF8_CHARCLASS(alnum,"a");
1191 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1192 if (tmp == !(OP(c) == BOUND ?
1193 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1194 isALNUM_LC_utf8((U8*)s)))
1197 if ((norun || regtry(prog, s)))
1204 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1205 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1206 while (s < strend) {
1208 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1210 if ((norun || regtry(prog, s)))
1216 if ((!prog->minlen && tmp) && (norun || regtry(prog, s)))
1220 PL_reg_flags |= RF_tainted;
1227 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1229 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1231 tmp = ((OP(c) == NBOUND ?
1232 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1233 LOAD_UTF8_CHARCLASS(alnum,"a");
1234 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1235 if (tmp == !(OP(c) == NBOUND ?
1236 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1237 isALNUM_LC_utf8((U8*)s)))
1239 else if ((norun || regtry(prog, s)))
1245 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1246 tmp = ((OP(c) == NBOUND ?
1247 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1248 while (s < strend) {
1250 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1252 else if ((norun || regtry(prog, s)))
1257 if ((!prog->minlen && !tmp) && (norun || regtry(prog, s)))
1262 LOAD_UTF8_CHARCLASS(alnum,"a");
1263 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1264 if (swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1265 if (tmp && (norun || regtry(prog, s)))
1276 while (s < strend) {
1278 if (tmp && (norun || regtry(prog, s)))
1290 PL_reg_flags |= RF_tainted;
1292 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1293 if (isALNUM_LC_utf8((U8*)s)) {
1294 if (tmp && (norun || regtry(prog, s)))
1305 while (s < strend) {
1306 if (isALNUM_LC(*s)) {
1307 if (tmp && (norun || regtry(prog, s)))
1320 LOAD_UTF8_CHARCLASS(alnum,"a");
1321 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1322 if (!swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1323 if (tmp && (norun || regtry(prog, s)))
1334 while (s < strend) {
1336 if (tmp && (norun || regtry(prog, s)))
1348 PL_reg_flags |= RF_tainted;
1350 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1351 if (!isALNUM_LC_utf8((U8*)s)) {
1352 if (tmp && (norun || regtry(prog, s)))
1363 while (s < strend) {
1364 if (!isALNUM_LC(*s)) {
1365 if (tmp && (norun || regtry(prog, s)))
1378 LOAD_UTF8_CHARCLASS(space," ");
1379 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1380 if (*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)) {
1381 if (tmp && (norun || regtry(prog, s)))
1392 while (s < strend) {
1394 if (tmp && (norun || regtry(prog, s)))
1406 PL_reg_flags |= RF_tainted;
1408 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1409 if (*s == ' ' || isSPACE_LC_utf8((U8*)s)) {
1410 if (tmp && (norun || regtry(prog, s)))
1421 while (s < strend) {
1422 if (isSPACE_LC(*s)) {
1423 if (tmp && (norun || regtry(prog, s)))
1436 LOAD_UTF8_CHARCLASS(space," ");
1437 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1438 if (!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8))) {
1439 if (tmp && (norun || regtry(prog, s)))
1450 while (s < strend) {
1452 if (tmp && (norun || regtry(prog, s)))
1464 PL_reg_flags |= RF_tainted;
1466 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1467 if (!(*s == ' ' || isSPACE_LC_utf8((U8*)s))) {
1468 if (tmp && (norun || regtry(prog, s)))
1479 while (s < strend) {
1480 if (!isSPACE_LC(*s)) {
1481 if (tmp && (norun || regtry(prog, s)))
1494 LOAD_UTF8_CHARCLASS(digit,"0");
1495 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1496 if (swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1497 if (tmp && (norun || regtry(prog, s)))
1508 while (s < strend) {
1510 if (tmp && (norun || regtry(prog, s)))
1522 PL_reg_flags |= RF_tainted;
1524 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1525 if (isDIGIT_LC_utf8((U8*)s)) {
1526 if (tmp && (norun || regtry(prog, s)))
1537 while (s < strend) {
1538 if (isDIGIT_LC(*s)) {
1539 if (tmp && (norun || regtry(prog, s)))
1552 LOAD_UTF8_CHARCLASS(digit,"0");
1553 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1554 if (!swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1555 if (tmp && (norun || regtry(prog, s)))
1566 while (s < strend) {
1568 if (tmp && (norun || regtry(prog, s)))
1580 PL_reg_flags |= RF_tainted;
1582 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1583 if (!isDIGIT_LC_utf8((U8*)s)) {
1584 if (tmp && (norun || regtry(prog, s)))
1595 while (s < strend) {
1596 if (!isDIGIT_LC(*s)) {
1597 if (tmp && (norun || regtry(prog, s)))
1609 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1618 - regexec_flags - match a regexp against a string
1621 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1622 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1623 /* strend: pointer to null at end of string */
1624 /* strbeg: real beginning of string */
1625 /* minend: end of match must be >=minend after stringarg. */
1626 /* data: May be used for some additional optimizations. */
1627 /* nosave: For optimizations. */
1630 register regnode *c;
1631 register char *startpos = stringarg;
1632 I32 minlen; /* must match at least this many chars */
1633 I32 dontbother = 0; /* how many characters not to try at end */
1634 /* I32 start_shift = 0; */ /* Offset of the start to find
1635 constant substr. */ /* CC */
1636 I32 end_shift = 0; /* Same for the end. */ /* CC */
1637 I32 scream_pos = -1; /* Internal iterator of scream. */
1639 SV* oreplsv = GvSV(PL_replgv);
1640 bool do_utf8 = DO_UTF8(sv);
1641 I32 multiline = prog->reganch & PMf_MULTILINE;
1643 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
1644 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
1647 GET_RE_DEBUG_FLAGS_DECL;
1649 RX_MATCH_UTF8_set(prog,do_utf8);
1655 PL_regnarrate = DEBUG_r_TEST;
1658 /* Be paranoid... */
1659 if (prog == NULL || startpos == NULL) {
1660 Perl_croak(aTHX_ "NULL regexp parameter");
1664 minlen = prog->minlen;
1665 if (strend - startpos < minlen) {
1666 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1667 "String too short [regexec_flags]...\n"));
1671 /* Check validity of program. */
1672 if (UCHARAT(prog->program) != REG_MAGIC) {
1673 Perl_croak(aTHX_ "corrupted regexp program");
1677 PL_reg_eval_set = 0;
1680 if (prog->reganch & ROPT_UTF8)
1681 PL_reg_flags |= RF_utf8;
1683 /* Mark beginning of line for ^ and lookbehind. */
1684 PL_regbol = startpos;
1688 /* Mark end of line for $ (and such) */
1691 /* see how far we have to get to not match where we matched before */
1692 PL_regtill = startpos+minend;
1694 /* We start without call_cc context. */
1697 /* If there is a "must appear" string, look for it. */
1700 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to have PL_reg_ganch */
1703 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1704 PL_reg_ganch = startpos;
1705 else if (sv && SvTYPE(sv) >= SVt_PVMG
1707 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1708 && mg->mg_len >= 0) {
1709 PL_reg_ganch = strbeg + mg->mg_len; /* Defined pos() */
1710 if (prog->reganch & ROPT_ANCH_GPOS) {
1711 if (s > PL_reg_ganch)
1716 else /* pos() not defined */
1717 PL_reg_ganch = strbeg;
1720 if (!(flags & REXEC_CHECKED) && (prog->check_substr != Nullsv || prog->check_utf8 != Nullsv)) {
1721 re_scream_pos_data d;
1723 d.scream_olds = &scream_olds;
1724 d.scream_pos = &scream_pos;
1725 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1727 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1728 goto phooey; /* not present */
1734 pv_uni_display(dsv0, (U8*)prog->precomp, prog->prelen, 60,
1735 UNI_DISPLAY_REGEX) :
1737 int len0 = UTF ? SvCUR(dsv0) : prog->prelen;
1738 char *s1 = do_utf8 ? sv_uni_display(dsv1, sv, 60,
1739 UNI_DISPLAY_REGEX) : startpos;
1740 int len1 = do_utf8 ? SvCUR(dsv1) : strend - startpos;
1743 PerlIO_printf(Perl_debug_log,
1744 "%sMatching REx%s `%s%*.*s%s%s' against `%s%.*s%s%s'\n",
1745 PL_colors[4],PL_colors[5],PL_colors[0],
1748 len0 > 60 ? "..." : "",
1750 (int)(len1 > 60 ? 60 : len1),
1752 (len1 > 60 ? "..." : "")
1756 /* Simplest case: anchored match need be tried only once. */
1757 /* [unless only anchor is BOL and multiline is set] */
1758 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1759 if (s == startpos && regtry(prog, startpos))
1761 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1762 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1767 dontbother = minlen - 1;
1768 end = HOP3c(strend, -dontbother, strbeg) - 1;
1769 /* for multiline we only have to try after newlines */
1770 if (prog->check_substr || prog->check_utf8) {
1774 if (regtry(prog, s))
1779 if (prog->reganch & RE_USE_INTUIT) {
1780 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1791 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1792 if (regtry(prog, s))
1799 } else if (prog->reganch & ROPT_ANCH_GPOS) {
1800 if (regtry(prog, PL_reg_ganch))
1805 /* Messy cases: unanchored match. */
1806 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1807 /* we have /x+whatever/ */
1808 /* it must be a one character string (XXXX Except UTF?) */
1813 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1814 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1815 ch = SvPVX(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1818 while (s < strend) {
1820 DEBUG_EXECUTE_r( did_match = 1 );
1821 if (regtry(prog, s)) goto got_it;
1823 while (s < strend && *s == ch)
1830 while (s < strend) {
1832 DEBUG_EXECUTE_r( did_match = 1 );
1833 if (regtry(prog, s)) goto got_it;
1835 while (s < strend && *s == ch)
1841 DEBUG_EXECUTE_r(if (!did_match)
1842 PerlIO_printf(Perl_debug_log,
1843 "Did not find anchored character...\n")
1847 else if (prog->anchored_substr != Nullsv
1848 || prog->anchored_utf8 != Nullsv
1849 || ((prog->float_substr != Nullsv || prog->float_utf8 != Nullsv)
1850 && prog->float_max_offset < strend - s)) {
1855 char *last1; /* Last position checked before */
1859 if (prog->anchored_substr || prog->anchored_utf8) {
1860 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1861 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1862 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1863 back_max = back_min = prog->anchored_offset;
1865 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1866 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1867 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1868 back_max = prog->float_max_offset;
1869 back_min = prog->float_min_offset;
1871 if (must == &PL_sv_undef)
1872 /* could not downgrade utf8 check substring, so must fail */
1875 last = HOP3c(strend, /* Cannot start after this */
1876 -(I32)(CHR_SVLEN(must)
1877 - (SvTAIL(must) != 0) + back_min), strbeg);
1880 last1 = HOPc(s, -1);
1882 last1 = s - 1; /* bogus */
1884 /* XXXX check_substr already used to find `s', can optimize if
1885 check_substr==must. */
1887 dontbother = end_shift;
1888 strend = HOPc(strend, -dontbother);
1889 while ( (s <= last) &&
1890 ((flags & REXEC_SCREAM)
1891 ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg,
1892 end_shift, &scream_pos, 0))
1893 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend),
1894 (unsigned char*)strend, must,
1895 multiline ? FBMrf_MULTILINE : 0))) ) {
1896 /* we may be pointing at the wrong string */
1897 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1898 s = strbeg + (s - SvPVX(sv));
1899 DEBUG_EXECUTE_r( did_match = 1 );
1900 if (HOPc(s, -back_max) > last1) {
1901 last1 = HOPc(s, -back_min);
1902 s = HOPc(s, -back_max);
1905 char *t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1907 last1 = HOPc(s, -back_min);
1911 while (s <= last1) {
1912 if (regtry(prog, s))
1918 while (s <= last1) {
1919 if (regtry(prog, s))
1925 DEBUG_EXECUTE_r(if (!did_match)
1926 PerlIO_printf(Perl_debug_log,
1927 "Did not find %s substr `%s%.*s%s'%s...\n",
1928 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1929 ? "anchored" : "floating"),
1931 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
1933 PL_colors[1], (SvTAIL(must) ? "$" : ""))
1937 else if ((c = prog->regstclass)) {
1939 I32 op = (U8)OP(prog->regstclass);
1940 /* don't bother with what can't match */
1941 if (PL_regkind[op] != EXACT && op != CANY)
1942 strend = HOPc(strend, -(minlen - 1));
1945 SV *prop = sv_newmortal();
1953 pv_uni_display(dsv0, (U8*)SvPVX(prop), SvCUR(prop), 60,
1954 UNI_DISPLAY_REGEX) :
1956 len0 = UTF ? SvCUR(dsv0) : SvCUR(prop);
1958 sv_uni_display(dsv1, sv, 60, UNI_DISPLAY_REGEX) : s;
1959 len1 = UTF ? SvCUR(dsv1) : strend - s;
1960 PerlIO_printf(Perl_debug_log,
1961 "Matching stclass `%*.*s' against `%*.*s'\n",
1965 if (find_byclass(prog, c, s, strend, 0))
1967 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass...\n"));
1971 if (prog->float_substr != Nullsv || prog->float_utf8 != Nullsv) {
1976 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1977 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1978 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1980 if (flags & REXEC_SCREAM) {
1981 last = screaminstr(sv, float_real, s - strbeg,
1982 end_shift, &scream_pos, 1); /* last one */
1984 last = scream_olds; /* Only one occurrence. */
1985 /* we may be pointing at the wrong string */
1986 else if (RX_MATCH_COPIED(prog))
1987 s = strbeg + (s - SvPVX(sv));
1991 const char * const little = SvPV(float_real, len);
1993 if (SvTAIL(float_real)) {
1994 if (memEQ(strend - len + 1, little, len - 1))
1995 last = strend - len + 1;
1996 else if (!multiline)
1997 last = memEQ(strend - len, little, len)
1998 ? strend - len : Nullch;
2004 last = rninstr(s, strend, little, little + len);
2006 last = strend; /* matching `$' */
2010 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2011 "%sCan't trim the tail, match fails (should not happen)%s\n",
2012 PL_colors[4],PL_colors[5]));
2013 goto phooey; /* Should not happen! */
2015 dontbother = strend - last + prog->float_min_offset;
2017 if (minlen && (dontbother < minlen))
2018 dontbother = minlen - 1;
2019 strend -= dontbother; /* this one's always in bytes! */
2020 /* We don't know much -- general case. */
2023 if (regtry(prog, s))
2032 if (regtry(prog, s))
2034 } while (s++ < strend);
2042 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2044 if (PL_reg_eval_set) {
2045 /* Preserve the current value of $^R */
2046 if (oreplsv != GvSV(PL_replgv))
2047 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2048 restored, the value remains
2050 restore_pos(aTHX_ 0);
2053 /* make sure $`, $&, $', and $digit will work later */
2054 if ( !(flags & REXEC_NOT_FIRST) ) {
2055 RX_MATCH_COPY_FREE(prog);
2056 if (flags & REXEC_COPY_STR) {
2057 I32 i = PL_regeol - startpos + (stringarg - strbeg);
2058 #ifdef PERL_COPY_ON_WRITE
2060 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2062 PerlIO_printf(Perl_debug_log,
2063 "Copy on write: regexp capture, type %d\n",
2066 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2067 prog->subbeg = SvPVX(prog->saved_copy);
2068 assert (SvPOKp(prog->saved_copy));
2072 RX_MATCH_COPIED_on(prog);
2073 s = savepvn(strbeg, i);
2079 prog->subbeg = strbeg;
2080 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2087 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2088 PL_colors[4],PL_colors[5]));
2089 if (PL_reg_eval_set)
2090 restore_pos(aTHX_ 0);
2095 - regtry - try match at specific point
2097 STATIC I32 /* 0 failure, 1 success */
2098 S_regtry(pTHX_ regexp *prog, char *startpos)
2104 GET_RE_DEBUG_FLAGS_DECL;
2107 PL_regindent = 0; /* XXXX Not good when matches are reenterable... */
2109 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2112 PL_reg_eval_set = RS_init;
2113 DEBUG_EXECUTE_r(DEBUG_s(
2114 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2115 (IV)(PL_stack_sp - PL_stack_base));
2117 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2118 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2119 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2121 /* Apparently this is not needed, judging by wantarray. */
2122 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2123 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2126 /* Make $_ available to executed code. */
2127 if (PL_reg_sv != DEFSV) {
2132 if (!(SvTYPE(PL_reg_sv) >= SVt_PVMG && SvMAGIC(PL_reg_sv)
2133 && (mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global)))) {
2134 /* prepare for quick setting of pos */
2135 sv_magic(PL_reg_sv, (SV*)0,
2136 PERL_MAGIC_regex_global, Nullch, 0);
2137 mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global);
2141 PL_reg_oldpos = mg->mg_len;
2142 SAVEDESTRUCTOR_X(restore_pos, 0);
2144 if (!PL_reg_curpm) {
2145 Newz(22,PL_reg_curpm, 1, PMOP);
2148 SV* repointer = newSViv(0);
2149 /* so we know which PL_regex_padav element is PL_reg_curpm */
2150 SvFLAGS(repointer) |= SVf_BREAK;
2151 av_push(PL_regex_padav,repointer);
2152 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2153 PL_regex_pad = AvARRAY(PL_regex_padav);
2157 PM_SETRE(PL_reg_curpm, prog);
2158 PL_reg_oldcurpm = PL_curpm;
2159 PL_curpm = PL_reg_curpm;
2160 if (RX_MATCH_COPIED(prog)) {
2161 /* Here is a serious problem: we cannot rewrite subbeg,
2162 since it may be needed if this match fails. Thus
2163 $` inside (?{}) could fail... */
2164 PL_reg_oldsaved = prog->subbeg;
2165 PL_reg_oldsavedlen = prog->sublen;
2166 #ifdef PERL_COPY_ON_WRITE
2167 PL_nrs = prog->saved_copy;
2169 RX_MATCH_COPIED_off(prog);
2172 PL_reg_oldsaved = Nullch;
2173 prog->subbeg = PL_bostr;
2174 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2176 prog->startp[0] = startpos - PL_bostr;
2177 PL_reginput = startpos;
2178 PL_regstartp = prog->startp;
2179 PL_regendp = prog->endp;
2180 PL_reglastparen = &prog->lastparen;
2181 PL_reglastcloseparen = &prog->lastcloseparen;
2182 prog->lastparen = 0;
2183 prog->lastcloseparen = 0;
2185 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2186 if (PL_reg_start_tmpl <= prog->nparens) {
2187 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2188 if(PL_reg_start_tmp)
2189 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2191 New(22,PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2194 /* XXXX What this code is doing here?!!! There should be no need
2195 to do this again and again, PL_reglastparen should take care of
2198 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2199 * Actually, the code in regcppop() (which Ilya may be meaning by
2200 * PL_reglastparen), is not needed at all by the test suite
2201 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2202 * enough, for building DynaLoader, or otherwise this
2203 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2204 * will happen. Meanwhile, this code *is* needed for the
2205 * above-mentioned test suite tests to succeed. The common theme
2206 * on those tests seems to be returning null fields from matches.
2211 if (prog->nparens) {
2212 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2219 if (regmatch(prog->program + 1)) {
2220 prog->endp[0] = PL_reginput - PL_bostr;
2223 REGCP_UNWIND(lastcp);
2227 #define RE_UNWIND_BRANCH 1
2228 #define RE_UNWIND_BRANCHJ 2
2232 typedef struct { /* XX: makes sense to enlarge it... */
2236 } re_unwind_generic_t;
2249 } re_unwind_branch_t;
2251 typedef union re_unwind_t {
2253 re_unwind_generic_t generic;
2254 re_unwind_branch_t branch;
2257 #define sayYES goto yes
2258 #define sayNO goto no
2259 #define sayNO_ANYOF goto no_anyof
2260 #define sayYES_FINAL goto yes_final
2261 #define sayYES_LOUD goto yes_loud
2262 #define sayNO_FINAL goto no_final
2263 #define sayNO_SILENT goto do_no
2264 #define saySAME(x) if (x) goto yes; else goto no
2266 #define POSCACHE_SUCCESS 0 /* caching success rather than failure */
2267 #define POSCACHE_SEEN 1 /* we know what we're caching */
2268 #define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */
2269 #define CACHEsayYES STMT_START { \
2270 if (cache_offset | cache_bit) { \
2271 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2272 PL_reg_poscache[0] |= (1<<POSCACHE_SUCCESS) || (1<<POSCACHE_SEEN); \
2273 else if (!(PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2274 /* cache records failure, but this is success */ \
2276 PerlIO_printf(Perl_debug_log, \
2277 "%*s (remove success from failure cache)\n", \
2278 REPORT_CODE_OFF+PL_regindent*2, "") \
2280 PL_reg_poscache[cache_offset] &= ~(1<<cache_bit); \
2285 #define CACHEsayNO STMT_START { \
2286 if (cache_offset | cache_bit) { \
2287 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2288 PL_reg_poscache[0] |= (1<<POSCACHE_SEEN); \
2289 else if ((PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2290 /* cache records success, but this is failure */ \
2292 PerlIO_printf(Perl_debug_log, \
2293 "%*s (remove failure from success cache)\n", \
2294 REPORT_CODE_OFF+PL_regindent*2, "") \
2296 PL_reg_poscache[cache_offset] &= ~(1<<cache_bit); \
2302 /* this is used to determine how far from the left messages like
2303 'failed...' are printed. Currently 29 makes these messages line
2304 up with the opcode they refer to. Earlier perls used 25 which
2305 left these messages outdented making reviewing a debug output
2308 #define REPORT_CODE_OFF 29
2311 /* Make sure there is a test for this +1 options in re_tests */
2312 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2314 #define TRIE_CHECK_STATE_IS_ACCEPTING STMT_START { \
2315 if ( trie->states[ state ].wordnum ) { \
2316 if ( !accepted ) { \
2319 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN ; \
2320 sv_accept_buff=NEWSV( 1234, \
2321 bufflen * sizeof(reg_trie_accepted) - 1 ); \
2322 SvCUR_set( sv_accept_buff, sizeof(reg_trie_accepted) ); \
2323 SvPOK_on( sv_accept_buff ); \
2324 sv_2mortal( sv_accept_buff ); \
2325 accept_buff = (reg_trie_accepted*)SvPV_nolen( sv_accept_buff );\
2327 if ( accepted >= bufflen ) { \
2329 accept_buff =(reg_trie_accepted*)SvGROW( sv_accept_buff, \
2330 bufflen * sizeof(reg_trie_accepted) ); \
2332 SvCUR_set( sv_accept_buff,SvCUR( sv_accept_buff ) \
2333 + sizeof( reg_trie_accepted ) ); \
2335 accept_buff[ accepted ].wordnum = trie->states[ state ].wordnum; \
2336 accept_buff[ accepted ].endpos = uc; \
2340 #define TRIE_HANDLE_CHAR STMT_START { \
2341 if ( uvc < 256 ) { \
2342 charid = trie->charmap[ uvc ]; \
2345 if( trie->widecharmap ) { \
2346 SV** svpp = (SV**)NULL; \
2347 svpp = hv_fetch( trie->widecharmap, (char*)&uvc, \
2348 sizeof( UV ), 0 ); \
2350 charid = (U16)SvIV( *svpp ); \
2355 ( base + charid > trie->uniquecharcount ) && \
2356 ( base + charid - 1 - trie->uniquecharcount < trie->lasttrans) && \
2357 trie->trans[ base + charid - 1 - trie->uniquecharcount ].check == state ) \
2359 state = trie->trans[ base + charid - 1 - trie->uniquecharcount ].next; \
2367 - regmatch - main matching routine
2369 * Conceptually the strategy is simple: check to see whether the current
2370 * node matches, call self recursively to see whether the rest matches,
2371 * and then act accordingly. In practice we make some effort to avoid
2372 * recursion, in particular by going through "ordinary" nodes (that don't
2373 * need to know whether the rest of the match failed) by a loop instead of
2376 /* [lwall] I've hoisted the register declarations to the outer block in order to
2377 * maybe save a little bit of pushing and popping on the stack. It also takes
2378 * advantage of machines that use a register save mask on subroutine entry.
2380 STATIC I32 /* 0 failure, 1 success */
2381 S_regmatch(pTHX_ regnode *prog)
2383 register regnode *scan; /* Current node. */
2384 regnode *next; /* Next node. */
2385 regnode *inner; /* Next node in internal branch. */
2386 register I32 nextchr; /* renamed nextchr - nextchar colides with
2387 function of same name */
2388 register I32 n; /* no or next */
2389 register I32 ln = 0; /* len or last */
2390 register char *s = Nullch; /* operand or save */
2391 register char *locinput = PL_reginput;
2392 register I32 c1 = 0, c2 = 0, paren; /* case fold search, parenth */
2393 int minmod = 0, sw = 0, logical = 0;
2396 /* used by the trie code */
2397 SV *sv_accept_buff = 0; /* accepting states we have traversed */
2398 reg_trie_accepted *accept_buff = 0; /* "" */
2399 reg_trie_data *trie; /* what trie are we using right now */
2400 U32 accepted = 0; /* how many accepting states we have seen*/
2403 I32 firstcp = PL_savestack_ix;
2405 register bool do_utf8 = PL_reg_match_utf8;
2407 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
2408 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
2409 SV *dsv2 = PERL_DEBUG_PAD_ZERO(2);
2411 SV *re_debug_flags = NULL;
2421 /* Note that nextchr is a byte even in UTF */
2422 nextchr = UCHARAT(locinput);
2424 while (scan != NULL) {
2427 SV *prop = sv_newmortal();
2428 int docolor = *PL_colors[0];
2429 int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2430 int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput);
2431 /* The part of the string before starttry has one color
2432 (pref0_len chars), between starttry and current
2433 position another one (pref_len - pref0_len chars),
2434 after the current position the third one.
2435 We assume that pref0_len <= pref_len, otherwise we
2436 decrease pref0_len. */
2437 int pref_len = (locinput - PL_bostr) > (5 + taill) - l
2438 ? (5 + taill) - l : locinput - PL_bostr;
2441 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2443 pref0_len = pref_len - (locinput - PL_reg_starttry);
2444 if (l + pref_len < (5 + taill) && l < PL_regeol - locinput)
2445 l = ( PL_regeol - locinput > (5 + taill) - pref_len
2446 ? (5 + taill) - pref_len : PL_regeol - locinput);
2447 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2451 if (pref0_len > pref_len)
2452 pref0_len = pref_len;
2453 regprop(prop, scan);
2456 do_utf8 && OP(scan) != CANY ?
2457 pv_uni_display(dsv0, (U8*)(locinput - pref_len),
2458 pref0_len, 60, UNI_DISPLAY_REGEX) :
2459 locinput - pref_len;
2460 int len0 = do_utf8 ? strlen(s0) : pref0_len;
2461 char *s1 = do_utf8 && OP(scan) != CANY ?
2462 pv_uni_display(dsv1, (U8*)(locinput - pref_len + pref0_len),
2463 pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) :
2464 locinput - pref_len + pref0_len;
2465 int len1 = do_utf8 ? strlen(s1) : pref_len - pref0_len;
2466 char *s2 = do_utf8 && OP(scan) != CANY ?
2467 pv_uni_display(dsv2, (U8*)locinput,
2468 PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) :
2470 int len2 = do_utf8 ? strlen(s2) : l;
2471 PerlIO_printf(Perl_debug_log,
2472 "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|%3"IVdf":%*s%s\n",
2473 (IV)(locinput - PL_bostr),
2480 (docolor ? "" : "> <"),
2484 15 - l - pref_len + 1,
2486 (IV)(scan - PL_regprogram), PL_regindent*2, "",
2491 next = scan + NEXT_OFF(scan);
2497 if (locinput == PL_bostr)
2499 /* regtill = regbol; */
2504 if (locinput == PL_bostr ||
2505 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2511 if (locinput == PL_bostr)
2515 if (locinput == PL_reg_ganch)
2521 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2526 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2528 if (PL_regeol - locinput > 1)
2532 if (PL_regeol != locinput)
2536 if (!nextchr && locinput >= PL_regeol)
2539 locinput += PL_utf8skip[nextchr];
2540 if (locinput > PL_regeol)
2542 nextchr = UCHARAT(locinput);
2545 nextchr = UCHARAT(++locinput);
2548 if (!nextchr && locinput >= PL_regeol)
2550 nextchr = UCHARAT(++locinput);
2553 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2556 locinput += PL_utf8skip[nextchr];
2557 if (locinput > PL_regeol)
2559 nextchr = UCHARAT(locinput);
2562 nextchr = UCHARAT(++locinput);
2568 traverse the TRIE keeping track of all accepting states
2569 we transition through until we get to a failing node.
2571 we use two slightly different pieces of code to handle
2572 the traversal depending on whether its case sensitive or
2573 not. we reuse the accept code however. (this should probably
2574 be turned into a macro.)
2581 U32 uniflags = ckWARN( WARN_UTF8 ) ? 0 : UTF8_ALLOW_ANY;
2582 U8 *uc = ( U8* )locinput;
2589 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2590 U8 *uscan = (U8*)NULL;
2594 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2596 while ( state && uc <= (U8*)PL_regeol ) {
2598 TRIE_CHECK_STATE_IS_ACCEPTING;
2600 base = trie->states[ state ].trans.base;
2602 DEBUG_TRIE_EXECUTE_r(
2603 PerlIO_printf( Perl_debug_log,
2604 "%*s %sState: %4x, Base: %4x Accepted: %4x ",
2605 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2606 state, base, accepted );
2611 if ( do_utf8 || UTF ) {
2613 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags );
2618 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2619 uvc = to_uni_fold( uvc, foldbuf, &foldlen );
2620 foldlen -= UNISKIP( uvc );
2621 uscan = foldbuf + UNISKIP( uvc );
2633 DEBUG_TRIE_EXECUTE_r(
2634 PerlIO_printf( Perl_debug_log,
2635 "Charid:%3x CV:%4x After State: %4x%s\n",
2636 charid, uvc, state, PL_colors[5] );
2645 /* unreached codepoint: we jump into the middle of the next case
2646 from previous if blocks */
2649 U32 uniflags = ckWARN( WARN_UTF8 ) ? 0 : UTF8_ALLOW_ANY;
2650 U8 *uc = (U8*)locinput;
2659 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2661 while ( state && uc <= (U8*)PL_regeol ) {
2663 TRIE_CHECK_STATE_IS_ACCEPTING;
2665 base = trie->states[ state ].trans.base;
2667 DEBUG_TRIE_EXECUTE_r(
2668 PerlIO_printf( Perl_debug_log,
2669 "%*s %sState: %4x, Base: %4x Accepted: %4x ",
2670 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2671 state, base, accepted );
2676 if ( do_utf8 || UTF ) {
2677 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2688 DEBUG_TRIE_EXECUTE_r(
2689 PerlIO_printf( Perl_debug_log,
2690 "Charid:%3x CV:%4x After State: %4x%s\n",
2691 charid, uvc, state, PL_colors[5] );
2701 There was at least one accepting state that we
2702 transitioned through. Presumably the number of accepting
2703 states is going to be low, typically one or two. So we
2704 simply scan through to find the one with lowest wordnum.
2705 Once we find it, we swap the last state into its place
2706 and decrement the size. We then try to match the rest of
2707 the pattern at the point where the word ends, if we
2708 succeed then we end the loop, otherwise the loop
2709 eventually terminates once all of the accepting states
2716 if ( accepted == 1 ) {
2718 SV **tmp = av_fetch( trie->words, accept_buff[ 0 ].wordnum-1, 0 );
2719 PerlIO_printf( Perl_debug_log,
2720 "%*s %sonly one match : #%d <%s>%s\n",
2721 REPORT_CODE_OFF+PL_regindent*2, "",PL_colors[4],
2722 accept_buff[ 0 ].wordnum,
2723 tmp ? SvPV_nolen( *tmp ) : "not compiled under -Dr",
2726 PL_reginput = (char *)accept_buff[ 0 ].endpos;
2727 /* in this case we free tmps/leave before we call regmatch
2728 as we wont be using accept_buff again. */
2731 gotit = regmatch( scan + NEXT_OFF( scan ) );
2734 PerlIO_printf( Perl_debug_log,"%*s %sgot %d possible matches%s\n",
2735 REPORT_CODE_OFF + PL_regindent * 2, "",PL_colors[4], accepted,
2738 while ( !gotit && accepted-- ) {
2741 for( cur = 1 ; cur <= accepted ; cur++ ) {
2742 DEBUG_TRIE_EXECUTE_r(
2743 PerlIO_printf( Perl_debug_log,
2744 "%*s %sgot %d (%d) as best, looking at %d (%d)%s\n",
2745 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2746 best, accept_buff[ best ].wordnum, cur,
2747 accept_buff[ cur ].wordnum, PL_colors[5] );
2750 if ( accept_buff[ cur ].wordnum < accept_buff[ best ].wordnum )
2754 SV **tmp = av_fetch( trie->words, accept_buff[ best ].wordnum - 1, 0 );
2755 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at 0x%p%s\n",
2756 REPORT_CODE_OFF+PL_regindent*2, "",PL_colors[4],
2757 accept_buff[best].wordnum,
2758 tmp ? SvPV_nolen( *tmp ) : "not compiled under -Dr",scan,
2761 if ( best<accepted ) {
2762 reg_trie_accepted tmp = accept_buff[ best ];
2763 accept_buff[ best ] = accept_buff[ accepted ];
2764 accept_buff[ accepted ] = tmp;
2767 PL_reginput = (char *)accept_buff[ best ].endpos;
2770 as far as I can tell we only need the SAVETMPS/FREETMPS
2771 for re's with EVAL in them but I'm leaving them in for
2772 all until I can be sure.
2775 gotit = regmatch( scan + NEXT_OFF( scan ) ) ;
2788 /* unreached codepoint */
2792 if (do_utf8 != UTF) {
2793 /* The target and the pattern have differing utf8ness. */
2799 /* The target is utf8, the pattern is not utf8. */
2803 if (NATIVE_TO_UNI(*(U8*)s) !=
2804 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2806 0 : UTF8_ALLOW_ANY))
2813 /* The target is not utf8, the pattern is utf8. */
2817 if (NATIVE_TO_UNI(*((U8*)l)) !=
2818 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2820 0 : UTF8_ALLOW_ANY))
2827 nextchr = UCHARAT(locinput);
2830 /* The target and the pattern have the same utf8ness. */
2831 /* Inline the first character, for speed. */
2832 if (UCHARAT(s) != nextchr)
2834 if (PL_regeol - locinput < ln)
2836 if (ln > 1 && memNE(s, locinput, ln))
2839 nextchr = UCHARAT(locinput);
2842 PL_reg_flags |= RF_tainted;
2848 if (do_utf8 || UTF) {
2849 /* Either target or the pattern are utf8. */
2851 char *e = PL_regeol;
2853 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
2854 l, &e, 0, do_utf8)) {
2855 /* One more case for the sharp s:
2856 * pack("U0U*", 0xDF) =~ /ss/i,
2857 * the 0xC3 0x9F are the UTF-8
2858 * byte sequence for the U+00DF. */
2860 toLOWER(s[0]) == 's' &&
2862 toLOWER(s[1]) == 's' &&
2869 nextchr = UCHARAT(locinput);
2873 /* Neither the target and the pattern are utf8. */
2875 /* Inline the first character, for speed. */
2876 if (UCHARAT(s) != nextchr &&
2877 UCHARAT(s) != ((OP(scan) == EXACTF)
2878 ? PL_fold : PL_fold_locale)[nextchr])
2880 if (PL_regeol - locinput < ln)
2882 if (ln > 1 && (OP(scan) == EXACTF
2883 ? ibcmp(s, locinput, ln)
2884 : ibcmp_locale(s, locinput, ln)))
2887 nextchr = UCHARAT(locinput);
2891 STRLEN inclasslen = PL_regeol - locinput;
2893 if (!reginclass(scan, (U8*)locinput, &inclasslen, do_utf8))
2895 if (locinput >= PL_regeol)
2897 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
2898 nextchr = UCHARAT(locinput);
2903 nextchr = UCHARAT(locinput);
2904 if (!REGINCLASS(scan, (U8*)locinput))
2906 if (!nextchr && locinput >= PL_regeol)
2908 nextchr = UCHARAT(++locinput);
2912 /* If we might have the case of the German sharp s
2913 * in a casefolding Unicode character class. */
2915 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
2916 locinput += SHARP_S_SKIP;
2917 nextchr = UCHARAT(locinput);
2923 PL_reg_flags |= RF_tainted;
2929 LOAD_UTF8_CHARCLASS(alnum,"a");
2930 if (!(OP(scan) == ALNUM
2931 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2932 : isALNUM_LC_utf8((U8*)locinput)))
2936 locinput += PL_utf8skip[nextchr];
2937 nextchr = UCHARAT(locinput);
2940 if (!(OP(scan) == ALNUM
2941 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
2943 nextchr = UCHARAT(++locinput);
2946 PL_reg_flags |= RF_tainted;
2949 if (!nextchr && locinput >= PL_regeol)
2952 LOAD_UTF8_CHARCLASS(alnum,"a");
2953 if (OP(scan) == NALNUM
2954 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2955 : isALNUM_LC_utf8((U8*)locinput))
2959 locinput += PL_utf8skip[nextchr];
2960 nextchr = UCHARAT(locinput);
2963 if (OP(scan) == NALNUM
2964 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
2966 nextchr = UCHARAT(++locinput);
2970 PL_reg_flags |= RF_tainted;
2974 /* was last char in word? */
2976 if (locinput == PL_bostr)
2979 U8 *r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
2981 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
2983 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
2984 ln = isALNUM_uni(ln);
2985 LOAD_UTF8_CHARCLASS(alnum,"a");
2986 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
2989 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
2990 n = isALNUM_LC_utf8((U8*)locinput);
2994 ln = (locinput != PL_bostr) ?
2995 UCHARAT(locinput - 1) : '\n';
2996 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
2998 n = isALNUM(nextchr);
3001 ln = isALNUM_LC(ln);
3002 n = isALNUM_LC(nextchr);
3005 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3006 OP(scan) == BOUNDL))
3010 PL_reg_flags |= RF_tainted;
3016 if (UTF8_IS_CONTINUED(nextchr)) {
3017 LOAD_UTF8_CHARCLASS(space," ");
3018 if (!(OP(scan) == SPACE
3019 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3020 : isSPACE_LC_utf8((U8*)locinput)))
3024 locinput += PL_utf8skip[nextchr];
3025 nextchr = UCHARAT(locinput);
3028 if (!(OP(scan) == SPACE
3029 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3031 nextchr = UCHARAT(++locinput);
3034 if (!(OP(scan) == SPACE
3035 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3037 nextchr = UCHARAT(++locinput);
3041 PL_reg_flags |= RF_tainted;
3044 if (!nextchr && locinput >= PL_regeol)
3047 LOAD_UTF8_CHARCLASS(space," ");
3048 if (OP(scan) == NSPACE
3049 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3050 : isSPACE_LC_utf8((U8*)locinput))
3054 locinput += PL_utf8skip[nextchr];
3055 nextchr = UCHARAT(locinput);
3058 if (OP(scan) == NSPACE
3059 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3061 nextchr = UCHARAT(++locinput);
3064 PL_reg_flags |= RF_tainted;
3070 LOAD_UTF8_CHARCLASS(digit,"0");
3071 if (!(OP(scan) == DIGIT
3072 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3073 : isDIGIT_LC_utf8((U8*)locinput)))
3077 locinput += PL_utf8skip[nextchr];
3078 nextchr = UCHARAT(locinput);
3081 if (!(OP(scan) == DIGIT
3082 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3084 nextchr = UCHARAT(++locinput);
3087 PL_reg_flags |= RF_tainted;
3090 if (!nextchr && locinput >= PL_regeol)
3093 LOAD_UTF8_CHARCLASS(digit,"0");
3094 if (OP(scan) == NDIGIT
3095 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3096 : isDIGIT_LC_utf8((U8*)locinput))
3100 locinput += PL_utf8skip[nextchr];
3101 nextchr = UCHARAT(locinput);
3104 if (OP(scan) == NDIGIT
3105 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3107 nextchr = UCHARAT(++locinput);
3110 if (locinput >= PL_regeol)
3113 LOAD_UTF8_CHARCLASS(mark,"~");
3114 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3116 locinput += PL_utf8skip[nextchr];
3117 while (locinput < PL_regeol &&
3118 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3119 locinput += UTF8SKIP(locinput);
3120 if (locinput > PL_regeol)
3125 nextchr = UCHARAT(locinput);
3128 PL_reg_flags |= RF_tainted;
3132 n = ARG(scan); /* which paren pair */
3133 ln = PL_regstartp[n];
3134 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3135 if ((I32)*PL_reglastparen < n || ln == -1)
3136 sayNO; /* Do not match unless seen CLOSEn. */
3137 if (ln == PL_regendp[n])
3141 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3143 char *e = PL_bostr + PL_regendp[n];
3145 * Note that we can't do the "other character" lookup trick as
3146 * in the 8-bit case (no pun intended) because in Unicode we
3147 * have to map both upper and title case to lower case.
3149 if (OP(scan) == REFF) {
3150 STRLEN ulen1, ulen2;
3151 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3152 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3156 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3157 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3158 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3165 nextchr = UCHARAT(locinput);
3169 /* Inline the first character, for speed. */
3170 if (UCHARAT(s) != nextchr &&
3172 (UCHARAT(s) != ((OP(scan) == REFF
3173 ? PL_fold : PL_fold_locale)[nextchr]))))
3175 ln = PL_regendp[n] - ln;
3176 if (locinput + ln > PL_regeol)
3178 if (ln > 1 && (OP(scan) == REF
3179 ? memNE(s, locinput, ln)
3181 ? ibcmp(s, locinput, ln)
3182 : ibcmp_locale(s, locinput, ln))))
3185 nextchr = UCHARAT(locinput);
3196 OP_4tree *oop = PL_op;
3197 COP *ocurcop = PL_curcop;
3200 struct regexp *oreg = PL_reg_re;
3203 PL_op = (OP_4tree*)PL_regdata->data[n];
3204 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3205 PAD_SAVE_LOCAL(old_comppad, (PAD*)PL_regdata->data[n + 2]);
3206 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3210 CALLRUNOPS(aTHX); /* Scalar context. */
3213 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3221 PAD_RESTORE_LOCAL(old_comppad);
3222 PL_curcop = ocurcop;
3224 if (logical == 2) { /* Postponed subexpression. */
3226 MAGIC *mg = Null(MAGIC*);
3228 CHECKPOINT cp, lastcp;
3232 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3233 mg = mg_find(sv, PERL_MAGIC_qr);
3234 else if (SvSMAGICAL(ret)) {
3235 if (SvGMAGICAL(ret))
3236 sv_unmagic(ret, PERL_MAGIC_qr);
3238 mg = mg_find(ret, PERL_MAGIC_qr);
3242 re = (regexp *)mg->mg_obj;
3243 (void)ReREFCNT_inc(re);
3247 char *t = SvPV(ret, len);
3249 char *oprecomp = PL_regprecomp;
3250 I32 osize = PL_regsize;
3251 I32 onpar = PL_regnpar;
3254 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3255 re = CALLREGCOMP(aTHX_ t, t + len, &pm);
3257 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3259 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3261 PL_regprecomp = oprecomp;
3266 PerlIO_printf(Perl_debug_log,
3267 "Entering embedded `%s%.60s%s%s'\n",
3271 (strlen(re->precomp) > 60 ? "..." : ""))
3274 state.prev = PL_reg_call_cc;
3275 state.cc = PL_regcc;
3276 state.re = PL_reg_re;
3280 cp = regcppush(0); /* Save *all* the positions. */
3283 state.ss = PL_savestack_ix;
3284 *PL_reglastparen = 0;
3285 *PL_reglastcloseparen = 0;
3286 PL_reg_call_cc = &state;
3287 PL_reginput = locinput;
3288 toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3289 ((re->reganch & ROPT_UTF8) != 0);
3290 if (toggleutf) PL_reg_flags ^= RF_utf8;
3292 /* XXXX This is too dramatic a measure... */
3295 if (regmatch(re->program + 1)) {
3296 /* Even though we succeeded, we need to restore
3297 global variables, since we may be wrapped inside
3298 SUSPEND, thus the match may be not finished yet. */
3300 /* XXXX Do this only if SUSPENDed? */
3301 PL_reg_call_cc = state.prev;
3302 PL_regcc = state.cc;
3303 PL_reg_re = state.re;
3304 cache_re(PL_reg_re);
3305 if (toggleutf) PL_reg_flags ^= RF_utf8;
3307 /* XXXX This is too dramatic a measure... */
3310 /* These are needed even if not SUSPEND. */
3316 REGCP_UNWIND(lastcp);
3318 PL_reg_call_cc = state.prev;
3319 PL_regcc = state.cc;
3320 PL_reg_re = state.re;
3321 cache_re(PL_reg_re);
3322 if (toggleutf) PL_reg_flags ^= RF_utf8;
3324 /* XXXX This is too dramatic a measure... */
3334 sv_setsv(save_scalar(PL_replgv), ret);
3340 n = ARG(scan); /* which paren pair */
3341 PL_reg_start_tmp[n] = locinput;
3346 n = ARG(scan); /* which paren pair */
3347 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3348 PL_regendp[n] = locinput - PL_bostr;
3349 if (n > (I32)*PL_reglastparen)
3350 *PL_reglastparen = n;
3351 *PL_reglastcloseparen = n;
3354 n = ARG(scan); /* which paren pair */
3355 sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3358 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3360 next = NEXTOPER(NEXTOPER(scan));
3362 next = scan + ARG(scan);
3363 if (OP(next) == IFTHEN) /* Fake one. */
3364 next = NEXTOPER(NEXTOPER(next));
3368 logical = scan->flags;
3370 /*******************************************************************
3371 PL_regcc contains infoblock about the innermost (...)* loop, and
3372 a pointer to the next outer infoblock.
3374 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3376 1) After matching X, regnode for CURLYX is processed;
3378 2) This regnode creates infoblock on the stack, and calls
3379 regmatch() recursively with the starting point at WHILEM node;
3381 3) Each hit of WHILEM node tries to match A and Z (in the order
3382 depending on the current iteration, min/max of {min,max} and
3383 greediness). The information about where are nodes for "A"
3384 and "Z" is read from the infoblock, as is info on how many times "A"
3385 was already matched, and greediness.
3387 4) After A matches, the same WHILEM node is hit again.
3389 5) Each time WHILEM is hit, PL_regcc is the infoblock created by CURLYX
3390 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3391 resets PL_regcc, since this Y(A)*Z can be a part of some other loop:
3392 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3393 of the external loop.
3395 Currently present infoblocks form a tree with a stem formed by PL_curcc
3396 and whatever it mentions via ->next, and additional attached trees
3397 corresponding to temporarily unset infoblocks as in "5" above.
3399 In the following picture infoblocks for outer loop of
3400 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3401 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3402 infoblocks are drawn below the "reset" infoblock.
3404 In fact in the picture below we do not show failed matches for Z and T
3405 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3406 more obvious *why* one needs to *temporary* unset infoblocks.]
3408 Matched REx position InfoBlocks Comment
3412 Y A)*?Z)*?T x <- O <- I
3413 YA )*?Z)*?T x <- O <- I
3414 YA A)*?Z)*?T x <- O <- I
3415 YAA )*?Z)*?T x <- O <- I
3416 YAA Z)*?T x <- O # Temporary unset I
3419 YAAZ Y(A)*?Z)*?T x <- O
3422 YAAZY (A)*?Z)*?T x <- O
3425 YAAZY A)*?Z)*?T x <- O <- I
3428 YAAZYA )*?Z)*?T x <- O <- I
3431 YAAZYA Z)*?T x <- O # Temporary unset I
3437 YAAZYAZ T x # Temporary unset O
3444 *******************************************************************/
3447 CHECKPOINT cp = PL_savestack_ix;
3448 /* No need to save/restore up to this paren */
3449 I32 parenfloor = scan->flags;
3451 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3453 cc.oldcc = PL_regcc;
3455 /* XXXX Probably it is better to teach regpush to support
3456 parenfloor > PL_regsize... */
3457 if (parenfloor > (I32)*PL_reglastparen)
3458 parenfloor = *PL_reglastparen; /* Pessimization... */
3459 cc.parenfloor = parenfloor;
3461 cc.min = ARG1(scan);
3462 cc.max = ARG2(scan);
3463 cc.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3467 PL_reginput = locinput;
3468 n = regmatch(PREVOPER(next)); /* start on the WHILEM */
3470 PL_regcc = cc.oldcc;
3476 * This is really hard to understand, because after we match
3477 * what we're trying to match, we must make sure the rest of
3478 * the REx is going to match for sure, and to do that we have
3479 * to go back UP the parse tree by recursing ever deeper. And
3480 * if it fails, we have to reset our parent's current state
3481 * that we can try again after backing off.
3484 CHECKPOINT cp, lastcp;
3485 CURCUR* cc = PL_regcc;
3486 char *lastloc = cc->lastloc; /* Detection of 0-len. */
3487 I32 cache_offset = 0, cache_bit = 0;
3489 n = cc->cur + 1; /* how many we know we matched */
3490 PL_reginput = locinput;
3493 PerlIO_printf(Perl_debug_log,
3494 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3495 REPORT_CODE_OFF+PL_regindent*2, "",
3496 (long)n, (long)cc->min,
3497 (long)cc->max, PTR2UV(cc))
3500 /* If degenerate scan matches "", assume scan done. */
3502 if (locinput == cc->lastloc && n >= cc->min) {
3503 PL_regcc = cc->oldcc;
3507 PerlIO_printf(Perl_debug_log,
3508 "%*s empty match detected, try continuation...\n",
3509 REPORT_CODE_OFF+PL_regindent*2, "")
3511 if (regmatch(cc->next))
3519 /* First just match a string of min scans. */
3523 cc->lastloc = locinput;
3524 if (regmatch(cc->scan))
3527 cc->lastloc = lastloc;
3532 /* Check whether we already were at this position.
3533 Postpone detection until we know the match is not
3534 *that* much linear. */
3535 if (!PL_reg_maxiter) {
3536 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3537 PL_reg_leftiter = PL_reg_maxiter;
3539 if (PL_reg_leftiter-- == 0) {
3540 I32 size = (PL_reg_maxiter + 7 + POSCACHE_START)/8;
3541 if (PL_reg_poscache) {
3542 if ((I32)PL_reg_poscache_size < size) {
3543 Renew(PL_reg_poscache, size, char);
3544 PL_reg_poscache_size = size;
3546 Zero(PL_reg_poscache, size, char);
3549 PL_reg_poscache_size = size;
3550 Newz(29, PL_reg_poscache, size, char);
3553 PerlIO_printf(Perl_debug_log,
3554 "%sDetected a super-linear match, switching on caching%s...\n",
3555 PL_colors[4], PL_colors[5])
3558 if (PL_reg_leftiter < 0) {
3559 cache_offset = locinput - PL_bostr;
3561 cache_offset = (scan->flags & 0xf) - 1 + POSCACHE_START
3562 + cache_offset * (scan->flags>>4);
3563 cache_bit = cache_offset % 8;
3565 if (PL_reg_poscache[cache_offset] & (1<<cache_bit)) {
3567 PerlIO_printf(Perl_debug_log,
3568 "%*s already tried at this position...\n",
3569 REPORT_CODE_OFF+PL_regindent*2, "")
3571 if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))
3572 /* cache records success */
3575 /* cache records failure */
3578 PL_reg_poscache[cache_offset] |= (1<<cache_bit);
3582 /* Prefer next over scan for minimal matching. */
3585 PL_regcc = cc->oldcc;
3588 cp = regcppush(cc->parenfloor);
3590 if (regmatch(cc->next)) {
3592 CACHEsayYES; /* All done. */
3594 REGCP_UNWIND(lastcp);
3600 if (n >= cc->max) { /* Maximum greed exceeded? */
3601 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3602 && !(PL_reg_flags & RF_warned)) {
3603 PL_reg_flags |= RF_warned;
3604 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3605 "Complex regular subexpression recursion",
3612 PerlIO_printf(Perl_debug_log,
3613 "%*s trying longer...\n",
3614 REPORT_CODE_OFF+PL_regindent*2, "")
3616 /* Try scanning more and see if it helps. */
3617 PL_reginput = locinput;
3619 cc->lastloc = locinput;
3620 cp = regcppush(cc->parenfloor);
3622 if (regmatch(cc->scan)) {
3626 REGCP_UNWIND(lastcp);
3629 cc->lastloc = lastloc;
3633 /* Prefer scan over next for maximal matching. */
3635 if (n < cc->max) { /* More greed allowed? */
3636 cp = regcppush(cc->parenfloor);
3638 cc->lastloc = locinput;
3640 if (regmatch(cc->scan)) {
3644 REGCP_UNWIND(lastcp);
3645 regcppop(); /* Restore some previous $<digit>s? */
3646 PL_reginput = locinput;
3648 PerlIO_printf(Perl_debug_log,
3649 "%*s failed, try continuation...\n",
3650 REPORT_CODE_OFF+PL_regindent*2, "")
3653 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3654 && !(PL_reg_flags & RF_warned)) {
3655 PL_reg_flags |= RF_warned;
3656 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3657 "Complex regular subexpression recursion",
3661 /* Failed deeper matches of scan, so see if this one works. */
3662 PL_regcc = cc->oldcc;
3665 if (regmatch(cc->next))
3671 cc->lastloc = lastloc;
3676 next = scan + ARG(scan);
3679 inner = NEXTOPER(NEXTOPER(scan));
3682 inner = NEXTOPER(scan);
3686 if (OP(next) != c1) /* No choice. */
3687 next = inner; /* Avoid recursion. */
3689 I32 lastparen = *PL_reglastparen;
3691 re_unwind_branch_t *uw;
3693 /* Put unwinding data on stack */
3694 unwind1 = SSNEWt(1,re_unwind_branch_t);
3695 uw = SSPTRt(unwind1,re_unwind_branch_t);
3698 uw->type = ((c1 == BRANCH)
3700 : RE_UNWIND_BRANCHJ);
3701 uw->lastparen = lastparen;
3703 uw->locinput = locinput;
3704 uw->nextchr = nextchr;
3706 uw->regindent = ++PL_regindent;
3709 REGCP_SET(uw->lastcp);
3711 /* Now go into the first branch */
3724 /* We suppose that the next guy does not need
3725 backtracking: in particular, it is of constant non-zero length,
3726 and has no parenths to influence future backrefs. */
3727 ln = ARG1(scan); /* min to match */
3728 n = ARG2(scan); /* max to match */
3729 paren = scan->flags;
3731 if (paren > PL_regsize)
3733 if (paren > (I32)*PL_reglastparen)
3734 *PL_reglastparen = paren;
3736 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3738 scan += NEXT_OFF(scan); /* Skip former OPEN. */
3739 PL_reginput = locinput;
3742 if (ln && regrepeat_hard(scan, ln, &l) < ln)
3744 locinput = PL_reginput;
3745 if (HAS_TEXT(next) || JUMPABLE(next)) {
3746 regnode *text_node = next;
3748 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3750 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3752 if (PL_regkind[(U8)OP(text_node)] == REF) {
3756 else { c1 = (U8)*STRING(text_node); }
3757 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3759 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3760 c2 = PL_fold_locale[c1];
3769 while (n >= ln || (n == REG_INFTY && ln > 0)) { /* ln overflow ? */
3770 /* If it could work, try it. */
3772 UCHARAT(PL_reginput) == c1 ||
3773 UCHARAT(PL_reginput) == c2)
3777 PL_regstartp[paren] =
3778 HOPc(PL_reginput, -l) - PL_bostr;
3779 PL_regendp[paren] = PL_reginput - PL_bostr;
3782 PL_regendp[paren] = -1;
3786 REGCP_UNWIND(lastcp);
3788 /* Couldn't or didn't -- move forward. */
3789 PL_reginput = locinput;
3790 if (regrepeat_hard(scan, 1, &l)) {
3792 locinput = PL_reginput;
3799 n = regrepeat_hard(scan, n, &l);
3800 locinput = PL_reginput;
3802 PerlIO_printf(Perl_debug_log,
3803 "%*s matched %"IVdf" times, len=%"IVdf"...\n",
3804 (int)(REPORT_CODE_OFF+PL_regindent*2), "",
3808 if (HAS_TEXT(next) || JUMPABLE(next)) {
3809 regnode *text_node = next;
3811 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3813 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3815 if (PL_regkind[(U8)OP(text_node)] == REF) {
3819 else { c1 = (U8)*STRING(text_node); }
3821 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3823 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3824 c2 = PL_fold_locale[c1];
3835 /* If it could work, try it. */
3837 UCHARAT(PL_reginput) == c1 ||
3838 UCHARAT(PL_reginput) == c2)
3841 PerlIO_printf(Perl_debug_log,
3842 "%*s trying tail with n=%"IVdf"...\n",
3843 (int)(REPORT_CODE_OFF+PL_regindent*2), "", (IV)n)
3847 PL_regstartp[paren] = HOPc(PL_reginput, -l) - PL_bostr;
3848 PL_regendp[paren] = PL_reginput - PL_bostr;
3851 PL_regendp[paren] = -1;
3855 REGCP_UNWIND(lastcp);
3857 /* Couldn't or didn't -- back up. */
3859 locinput = HOPc(locinput, -l);