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 U8* str = (U8*)STRING(prog->regstclass);
861 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),
870 char *startpos = strbeg;
874 s = find_byclass(prog, prog->regstclass, s, endpos, startpos, 1);
879 if (endpos == strend) {
880 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
881 "Could not match STCLASS...\n") );
884 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
885 "This position contradicts STCLASS...\n") );
886 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
888 /* Contradict one of substrings */
889 if (prog->anchored_substr || prog->anchored_utf8) {
890 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
891 DEBUG_EXECUTE_r( what = "anchored" );
893 s = HOP3c(t, 1, strend);
894 if (s + start_shift + end_shift > strend) {
895 /* XXXX Should be taken into account earlier? */
896 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
897 "Could not match STCLASS...\n") );
902 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
903 "Looking for %s substr starting at offset %ld...\n",
904 what, (long)(s + start_shift - i_strpos)) );
907 /* Have both, check_string is floating */
908 if (t + start_shift >= check_at) /* Contradicts floating=check */
909 goto retry_floating_check;
910 /* Recheck anchored substring, but not floating... */
914 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
915 "Looking for anchored substr starting at offset %ld...\n",
916 (long)(other_last - i_strpos)) );
917 goto do_other_anchored;
919 /* Another way we could have checked stclass at the
920 current position only: */
925 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
926 "Looking for /%s^%s/m starting at offset %ld...\n",
927 PL_colors[0],PL_colors[1], (long)(t - i_strpos)) );
930 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
932 /* Check is floating subtring. */
933 retry_floating_check:
934 t = check_at - start_shift;
935 DEBUG_EXECUTE_r( what = "floating" );
936 goto hop_and_restart;
939 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
940 "By STCLASS: moving %ld --> %ld\n",
941 (long)(t - i_strpos), (long)(s - i_strpos))
945 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
946 "Does not contradict STCLASS...\n");
951 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
952 PL_colors[4], (check ? "Guessed" : "Giving up"),
953 PL_colors[5], (long)(s - i_strpos)) );
956 fail_finish: /* Substring not found */
957 if (prog->check_substr || prog->check_utf8) /* could be removed already */
958 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
960 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
961 PL_colors[4],PL_colors[5]));
965 /* We know what class REx starts with. Try to find this position... */
967 S_find_byclass(pTHX_ regexp * prog, regnode *c, char *s, char *strend, char *startpos, I32 norun)
969 I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
973 register STRLEN uskip;
977 register I32 tmp = 1; /* Scratch variable? */
978 register bool do_utf8 = PL_reg_match_utf8;
980 /* We know what class it must start with. */
984 while (s + (uskip = UTF8SKIP(s)) <= strend) {
985 if ((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
986 !UTF8_IS_INVARIANT((U8)s[0]) ?
987 reginclass(c, (U8*)s, 0, do_utf8) :
988 REGINCLASS(c, (U8*)s)) {
989 if (tmp && (norun || regtry(prog, s)))
1000 while (s < strend) {
1003 if (REGINCLASS(c, (U8*)s) ||
1004 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1005 /* The assignment of 2 is intentional:
1006 * for the folded sharp s, the skip is 2. */
1007 (skip = SHARP_S_SKIP))) {
1008 if (tmp && (norun || regtry(prog, s)))
1020 while (s < strend) {
1021 if (tmp && (norun || regtry(prog, s)))
1030 ln = STR_LEN(c); /* length to match in octets/bytes */
1031 lnc = (I32) ln; /* length to match in characters */
1033 STRLEN ulen1, ulen2;
1035 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1036 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1038 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1039 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1041 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1042 0, ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1043 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1044 0, ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1046 while (sm < ((U8 *) m + ln)) {
1061 c2 = PL_fold_locale[c1];
1063 e = HOP3c(strend, -((I32)lnc), s);
1066 e = s; /* Due to minlen logic of intuit() */
1068 /* The idea in the EXACTF* cases is to first find the
1069 * first character of the EXACTF* node and then, if
1070 * necessary, case-insensitively compare the full
1071 * text of the node. The c1 and c2 are the first
1072 * characters (though in Unicode it gets a bit
1073 * more complicated because there are more cases
1074 * than just upper and lower: one needs to use
1075 * the so-called folding case for case-insensitive
1076 * matching (called "loose matching" in Unicode).
1077 * ibcmp_utf8() will do just that. */
1081 U8 tmpbuf [UTF8_MAXBYTES+1];
1082 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1083 STRLEN len, foldlen;
1086 /* Upper and lower of 1st char are equal -
1087 * probably not a "letter". */
1089 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1091 0 : UTF8_ALLOW_ANY);
1094 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1095 m, (char **)0, ln, (bool)UTF))
1096 && (norun || regtry(prog, s)) )
1099 uvchr_to_utf8(tmpbuf, c);
1100 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1102 && (f == c1 || f == c2)
1103 && (ln == foldlen ||
1104 !ibcmp_utf8((char *) foldbuf,
1105 (char **)0, foldlen, do_utf8,
1107 (char **)0, ln, (bool)UTF))
1108 && (norun || regtry(prog, s)) )
1116 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1118 0 : UTF8_ALLOW_ANY);
1120 /* Handle some of the three Greek sigmas cases.
1121 * Note that not all the possible combinations
1122 * are handled here: some of them are handled
1123 * by the standard folding rules, and some of
1124 * them (the character class or ANYOF cases)
1125 * are handled during compiletime in
1126 * regexec.c:S_regclass(). */
1127 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1128 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1129 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1131 if ( (c == c1 || c == c2)
1133 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1134 m, (char **)0, ln, (bool)UTF))
1135 && (norun || regtry(prog, s)) )
1138 uvchr_to_utf8(tmpbuf, c);
1139 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1141 && (f == c1 || f == c2)
1142 && (ln == foldlen ||
1143 !ibcmp_utf8((char *) foldbuf,
1144 (char **)0, foldlen, do_utf8,
1146 (char **)0, ln, (bool)UTF))
1147 && (norun || regtry(prog, s)) )
1158 && (ln == 1 || !(OP(c) == EXACTF
1160 : ibcmp_locale(s, m, ln)))
1161 && (norun || regtry(prog, s)) )
1167 if ( (*(U8*)s == c1 || *(U8*)s == c2)
1168 && (ln == 1 || !(OP(c) == EXACTF
1170 : ibcmp_locale(s, m, ln)))
1171 && (norun || regtry(prog, s)) )
1178 PL_reg_flags |= RF_tainted;
1185 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1187 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1189 tmp = ((OP(c) == BOUND ?
1190 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1191 LOAD_UTF8_CHARCLASS(alnum,"a");
1192 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1193 if (tmp == !(OP(c) == BOUND ?
1194 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1195 isALNUM_LC_utf8((U8*)s)))
1198 if ((norun || regtry(prog, s)))
1205 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1206 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1207 while (s < strend) {
1209 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1211 if ((norun || regtry(prog, s)))
1217 if ((!prog->minlen && tmp) && (norun || regtry(prog, s)))
1221 PL_reg_flags |= RF_tainted;
1228 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1230 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1232 tmp = ((OP(c) == NBOUND ?
1233 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1234 LOAD_UTF8_CHARCLASS(alnum,"a");
1235 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1236 if (tmp == !(OP(c) == NBOUND ?
1237 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1238 isALNUM_LC_utf8((U8*)s)))
1240 else if ((norun || regtry(prog, s)))
1246 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1247 tmp = ((OP(c) == NBOUND ?
1248 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1249 while (s < strend) {
1251 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1253 else if ((norun || regtry(prog, s)))
1258 if ((!prog->minlen && !tmp) && (norun || regtry(prog, s)))
1263 LOAD_UTF8_CHARCLASS(alnum,"a");
1264 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1265 if (swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1266 if (tmp && (norun || regtry(prog, s)))
1277 while (s < strend) {
1279 if (tmp && (norun || regtry(prog, s)))
1291 PL_reg_flags |= RF_tainted;
1293 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1294 if (isALNUM_LC_utf8((U8*)s)) {
1295 if (tmp && (norun || regtry(prog, s)))
1306 while (s < strend) {
1307 if (isALNUM_LC(*s)) {
1308 if (tmp && (norun || regtry(prog, s)))
1321 LOAD_UTF8_CHARCLASS(alnum,"a");
1322 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1323 if (!swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1324 if (tmp && (norun || regtry(prog, s)))
1335 while (s < strend) {
1337 if (tmp && (norun || regtry(prog, s)))
1349 PL_reg_flags |= RF_tainted;
1351 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1352 if (!isALNUM_LC_utf8((U8*)s)) {
1353 if (tmp && (norun || regtry(prog, s)))
1364 while (s < strend) {
1365 if (!isALNUM_LC(*s)) {
1366 if (tmp && (norun || regtry(prog, s)))
1379 LOAD_UTF8_CHARCLASS(space," ");
1380 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1381 if (*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)) {
1382 if (tmp && (norun || regtry(prog, s)))
1393 while (s < strend) {
1395 if (tmp && (norun || regtry(prog, s)))
1407 PL_reg_flags |= RF_tainted;
1409 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1410 if (*s == ' ' || isSPACE_LC_utf8((U8*)s)) {
1411 if (tmp && (norun || regtry(prog, s)))
1422 while (s < strend) {
1423 if (isSPACE_LC(*s)) {
1424 if (tmp && (norun || regtry(prog, s)))
1437 LOAD_UTF8_CHARCLASS(space," ");
1438 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1439 if (!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8))) {
1440 if (tmp && (norun || regtry(prog, s)))
1451 while (s < strend) {
1453 if (tmp && (norun || regtry(prog, s)))
1465 PL_reg_flags |= RF_tainted;
1467 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1468 if (!(*s == ' ' || isSPACE_LC_utf8((U8*)s))) {
1469 if (tmp && (norun || regtry(prog, s)))
1480 while (s < strend) {
1481 if (!isSPACE_LC(*s)) {
1482 if (tmp && (norun || regtry(prog, s)))
1495 LOAD_UTF8_CHARCLASS(digit,"0");
1496 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1497 if (swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1498 if (tmp && (norun || regtry(prog, s)))
1509 while (s < strend) {
1511 if (tmp && (norun || regtry(prog, s)))
1523 PL_reg_flags |= RF_tainted;
1525 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1526 if (isDIGIT_LC_utf8((U8*)s)) {
1527 if (tmp && (norun || regtry(prog, s)))
1538 while (s < strend) {
1539 if (isDIGIT_LC(*s)) {
1540 if (tmp && (norun || regtry(prog, s)))
1553 LOAD_UTF8_CHARCLASS(digit,"0");
1554 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1555 if (!swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1556 if (tmp && (norun || regtry(prog, s)))
1567 while (s < strend) {
1569 if (tmp && (norun || regtry(prog, s)))
1581 PL_reg_flags |= RF_tainted;
1583 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1584 if (!isDIGIT_LC_utf8((U8*)s)) {
1585 if (tmp && (norun || regtry(prog, s)))
1596 while (s < strend) {
1597 if (!isDIGIT_LC(*s)) {
1598 if (tmp && (norun || regtry(prog, s)))
1610 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1619 - regexec_flags - match a regexp against a string
1622 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1623 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1624 /* strend: pointer to null at end of string */
1625 /* strbeg: real beginning of string */
1626 /* minend: end of match must be >=minend after stringarg. */
1627 /* data: May be used for some additional optimizations. */
1628 /* nosave: For optimizations. */
1631 register regnode *c;
1632 register char *startpos = stringarg;
1633 I32 minlen; /* must match at least this many chars */
1634 I32 dontbother = 0; /* how many characters not to try at end */
1635 /* I32 start_shift = 0; */ /* Offset of the start to find
1636 constant substr. */ /* CC */
1637 I32 end_shift = 0; /* Same for the end. */ /* CC */
1638 I32 scream_pos = -1; /* Internal iterator of scream. */
1640 SV* oreplsv = GvSV(PL_replgv);
1641 bool do_utf8 = DO_UTF8(sv);
1642 I32 multiline = prog->reganch & PMf_MULTILINE;
1644 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
1645 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
1648 GET_RE_DEBUG_FLAGS_DECL;
1650 RX_MATCH_UTF8_set(prog,do_utf8);
1656 PL_regnarrate = DEBUG_r_TEST;
1659 /* Be paranoid... */
1660 if (prog == NULL || startpos == NULL) {
1661 Perl_croak(aTHX_ "NULL regexp parameter");
1665 minlen = prog->minlen;
1666 if (strend - startpos < minlen) {
1667 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1668 "String too short [regexec_flags]...\n"));
1672 /* Check validity of program. */
1673 if (UCHARAT(prog->program) != REG_MAGIC) {
1674 Perl_croak(aTHX_ "corrupted regexp program");
1678 PL_reg_eval_set = 0;
1681 if (prog->reganch & ROPT_UTF8)
1682 PL_reg_flags |= RF_utf8;
1684 /* Mark beginning of line for ^ and lookbehind. */
1685 PL_regbol = startpos;
1689 /* Mark end of line for $ (and such) */
1692 /* see how far we have to get to not match where we matched before */
1693 PL_regtill = startpos+minend;
1695 /* We start without call_cc context. */
1698 /* If there is a "must appear" string, look for it. */
1701 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to have PL_reg_ganch */
1704 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1705 PL_reg_ganch = startpos;
1706 else if (sv && SvTYPE(sv) >= SVt_PVMG
1708 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1709 && mg->mg_len >= 0) {
1710 PL_reg_ganch = strbeg + mg->mg_len; /* Defined pos() */
1711 if (prog->reganch & ROPT_ANCH_GPOS) {
1712 if (s > PL_reg_ganch)
1717 else /* pos() not defined */
1718 PL_reg_ganch = strbeg;
1721 if (!(flags & REXEC_CHECKED) && (prog->check_substr != Nullsv || prog->check_utf8 != Nullsv)) {
1722 re_scream_pos_data d;
1724 d.scream_olds = &scream_olds;
1725 d.scream_pos = &scream_pos;
1726 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1728 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1729 goto phooey; /* not present */
1735 pv_uni_display(dsv0, (U8*)prog->precomp, prog->prelen, 60,
1736 UNI_DISPLAY_REGEX) :
1738 int len0 = UTF ? SvCUR(dsv0) : prog->prelen;
1739 char *s1 = do_utf8 ? sv_uni_display(dsv1, sv, 60,
1740 UNI_DISPLAY_REGEX) : startpos;
1741 int len1 = do_utf8 ? SvCUR(dsv1) : strend - startpos;
1744 PerlIO_printf(Perl_debug_log,
1745 "%sMatching REx%s `%s%*.*s%s%s' against `%s%.*s%s%s'\n",
1746 PL_colors[4],PL_colors[5],PL_colors[0],
1749 len0 > 60 ? "..." : "",
1751 (int)(len1 > 60 ? 60 : len1),
1753 (len1 > 60 ? "..." : "")
1757 /* Simplest case: anchored match need be tried only once. */
1758 /* [unless only anchor is BOL and multiline is set] */
1759 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1760 if (s == startpos && regtry(prog, startpos))
1762 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1763 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1768 dontbother = minlen - 1;
1769 end = HOP3c(strend, -dontbother, strbeg) - 1;
1770 /* for multiline we only have to try after newlines */
1771 if (prog->check_substr || prog->check_utf8) {
1775 if (regtry(prog, s))
1780 if (prog->reganch & RE_USE_INTUIT) {
1781 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1792 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1793 if (regtry(prog, s))
1800 } else if (prog->reganch & ROPT_ANCH_GPOS) {
1801 if (regtry(prog, PL_reg_ganch))
1806 /* Messy cases: unanchored match. */
1807 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1808 /* we have /x+whatever/ */
1809 /* it must be a one character string (XXXX Except UTF?) */
1814 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1815 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1816 ch = SvPVX(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1819 while (s < strend) {
1821 DEBUG_EXECUTE_r( did_match = 1 );
1822 if (regtry(prog, s)) goto got_it;
1824 while (s < strend && *s == ch)
1831 while (s < strend) {
1833 DEBUG_EXECUTE_r( did_match = 1 );
1834 if (regtry(prog, s)) goto got_it;
1836 while (s < strend && *s == ch)
1842 DEBUG_EXECUTE_r(if (!did_match)
1843 PerlIO_printf(Perl_debug_log,
1844 "Did not find anchored character...\n")
1848 else if (prog->anchored_substr != Nullsv
1849 || prog->anchored_utf8 != Nullsv
1850 || ((prog->float_substr != Nullsv || prog->float_utf8 != Nullsv)
1851 && prog->float_max_offset < strend - s)) {
1856 char *last1; /* Last position checked before */
1860 if (prog->anchored_substr || prog->anchored_utf8) {
1861 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1862 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1863 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1864 back_max = back_min = prog->anchored_offset;
1866 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1867 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1868 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1869 back_max = prog->float_max_offset;
1870 back_min = prog->float_min_offset;
1872 if (must == &PL_sv_undef)
1873 /* could not downgrade utf8 check substring, so must fail */
1876 last = HOP3c(strend, /* Cannot start after this */
1877 -(I32)(CHR_SVLEN(must)
1878 - (SvTAIL(must) != 0) + back_min), strbeg);
1881 last1 = HOPc(s, -1);
1883 last1 = s - 1; /* bogus */
1885 /* XXXX check_substr already used to find `s', can optimize if
1886 check_substr==must. */
1888 dontbother = end_shift;
1889 strend = HOPc(strend, -dontbother);
1890 while ( (s <= last) &&
1891 ((flags & REXEC_SCREAM)
1892 ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg,
1893 end_shift, &scream_pos, 0))
1894 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend),
1895 (unsigned char*)strend, must,
1896 multiline ? FBMrf_MULTILINE : 0))) ) {
1897 /* we may be pointing at the wrong string */
1898 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1899 s = strbeg + (s - SvPVX(sv));
1900 DEBUG_EXECUTE_r( did_match = 1 );
1901 if (HOPc(s, -back_max) > last1) {
1902 last1 = HOPc(s, -back_min);
1903 s = HOPc(s, -back_max);
1906 char *t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1908 last1 = HOPc(s, -back_min);
1912 while (s <= last1) {
1913 if (regtry(prog, s))
1919 while (s <= last1) {
1920 if (regtry(prog, s))
1926 DEBUG_EXECUTE_r(if (!did_match)
1927 PerlIO_printf(Perl_debug_log,
1928 "Did not find %s substr `%s%.*s%s'%s...\n",
1929 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1930 ? "anchored" : "floating"),
1932 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
1934 PL_colors[1], (SvTAIL(must) ? "$" : ""))
1938 else if ((c = prog->regstclass)) {
1940 I32 op = (U8)OP(prog->regstclass);
1941 /* don't bother with what can't match */
1942 if (PL_regkind[op] != EXACT && op != CANY)
1943 strend = HOPc(strend, -(minlen - 1));
1946 SV *prop = sv_newmortal();
1954 pv_uni_display(dsv0, (U8*)SvPVX(prop), SvCUR(prop), 60,
1955 UNI_DISPLAY_REGEX) :
1957 len0 = UTF ? SvCUR(dsv0) : SvCUR(prop);
1959 sv_uni_display(dsv1, sv, 60, UNI_DISPLAY_REGEX) : s;
1960 len1 = UTF ? SvCUR(dsv1) : strend - s;
1961 PerlIO_printf(Perl_debug_log,
1962 "Matching stclass `%*.*s' against `%*.*s'\n",
1966 if (find_byclass(prog, c, s, strend, startpos, 0))
1968 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass...\n"));
1972 if (prog->float_substr != Nullsv || prog->float_utf8 != Nullsv) {
1977 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1978 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1979 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1981 if (flags & REXEC_SCREAM) {
1982 last = screaminstr(sv, float_real, s - strbeg,
1983 end_shift, &scream_pos, 1); /* last one */
1985 last = scream_olds; /* Only one occurrence. */
1986 /* we may be pointing at the wrong string */
1987 else if (RX_MATCH_COPIED(prog))
1988 s = strbeg + (s - SvPVX(sv));
1992 char *little = SvPV(float_real, len);
1994 if (SvTAIL(float_real)) {
1995 if (memEQ(strend - len + 1, little, len - 1))
1996 last = strend - len + 1;
1997 else if (!multiline)
1998 last = memEQ(strend - len, little, len)
1999 ? strend - len : Nullch;
2005 last = rninstr(s, strend, little, little + len);
2007 last = strend; /* matching `$' */
2011 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2012 "%sCan't trim the tail, match fails (should not happen)%s\n",
2013 PL_colors[4],PL_colors[5]));
2014 goto phooey; /* Should not happen! */
2016 dontbother = strend - last + prog->float_min_offset;
2018 if (minlen && (dontbother < minlen))
2019 dontbother = minlen - 1;
2020 strend -= dontbother; /* this one's always in bytes! */
2021 /* We don't know much -- general case. */
2024 if (regtry(prog, s))
2033 if (regtry(prog, s))
2035 } while (s++ < strend);
2043 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2045 if (PL_reg_eval_set) {
2046 /* Preserve the current value of $^R */
2047 if (oreplsv != GvSV(PL_replgv))
2048 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2049 restored, the value remains
2051 restore_pos(aTHX_ 0);
2054 /* make sure $`, $&, $', and $digit will work later */
2055 if ( !(flags & REXEC_NOT_FIRST) ) {
2056 RX_MATCH_COPY_FREE(prog);
2057 if (flags & REXEC_COPY_STR) {
2058 I32 i = PL_regeol - startpos + (stringarg - strbeg);
2059 #ifdef PERL_COPY_ON_WRITE
2061 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2063 PerlIO_printf(Perl_debug_log,
2064 "Copy on write: regexp capture, type %d\n",
2067 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2068 prog->subbeg = SvPVX(prog->saved_copy);
2069 assert (SvPOKp(prog->saved_copy));
2073 RX_MATCH_COPIED_on(prog);
2074 s = savepvn(strbeg, i);
2080 prog->subbeg = strbeg;
2081 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2088 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2089 PL_colors[4],PL_colors[5]));
2090 if (PL_reg_eval_set)
2091 restore_pos(aTHX_ 0);
2096 - regtry - try match at specific point
2098 STATIC I32 /* 0 failure, 1 success */
2099 S_regtry(pTHX_ regexp *prog, char *startpos)
2105 GET_RE_DEBUG_FLAGS_DECL;
2108 PL_regindent = 0; /* XXXX Not good when matches are reenterable... */
2110 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2113 PL_reg_eval_set = RS_init;
2114 DEBUG_EXECUTE_r(DEBUG_s(
2115 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2116 (IV)(PL_stack_sp - PL_stack_base));
2118 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2119 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2120 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2122 /* Apparently this is not needed, judging by wantarray. */
2123 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2124 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2127 /* Make $_ available to executed code. */
2128 if (PL_reg_sv != DEFSV) {
2133 if (!(SvTYPE(PL_reg_sv) >= SVt_PVMG && SvMAGIC(PL_reg_sv)
2134 && (mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global)))) {
2135 /* prepare for quick setting of pos */
2136 sv_magic(PL_reg_sv, (SV*)0,
2137 PERL_MAGIC_regex_global, Nullch, 0);
2138 mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global);
2142 PL_reg_oldpos = mg->mg_len;
2143 SAVEDESTRUCTOR_X(restore_pos, 0);
2145 if (!PL_reg_curpm) {
2146 Newz(22,PL_reg_curpm, 1, PMOP);
2149 SV* repointer = newSViv(0);
2150 /* so we know which PL_regex_padav element is PL_reg_curpm */
2151 SvFLAGS(repointer) |= SVf_BREAK;
2152 av_push(PL_regex_padav,repointer);
2153 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2154 PL_regex_pad = AvARRAY(PL_regex_padav);
2158 PM_SETRE(PL_reg_curpm, prog);
2159 PL_reg_oldcurpm = PL_curpm;
2160 PL_curpm = PL_reg_curpm;
2161 if (RX_MATCH_COPIED(prog)) {
2162 /* Here is a serious problem: we cannot rewrite subbeg,
2163 since it may be needed if this match fails. Thus
2164 $` inside (?{}) could fail... */
2165 PL_reg_oldsaved = prog->subbeg;
2166 PL_reg_oldsavedlen = prog->sublen;
2167 #ifdef PERL_COPY_ON_WRITE
2168 PL_nrs = prog->saved_copy;
2170 RX_MATCH_COPIED_off(prog);
2173 PL_reg_oldsaved = Nullch;
2174 prog->subbeg = PL_bostr;
2175 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2177 prog->startp[0] = startpos - PL_bostr;
2178 PL_reginput = startpos;
2179 PL_regstartp = prog->startp;
2180 PL_regendp = prog->endp;
2181 PL_reglastparen = &prog->lastparen;
2182 PL_reglastcloseparen = &prog->lastcloseparen;
2183 prog->lastparen = 0;
2184 prog->lastcloseparen = 0;
2186 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2187 if (PL_reg_start_tmpl <= prog->nparens) {
2188 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2189 if(PL_reg_start_tmp)
2190 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2192 New(22,PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2195 /* XXXX What this code is doing here?!!! There should be no need
2196 to do this again and again, PL_reglastparen should take care of
2199 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2200 * Actually, the code in regcppop() (which Ilya may be meaning by
2201 * PL_reglastparen), is not needed at all by the test suite
2202 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2203 * enough, for building DynaLoader, or otherwise this
2204 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2205 * will happen. Meanwhile, this code *is* needed for the
2206 * above-mentioned test suite tests to succeed. The common theme
2207 * on those tests seems to be returning null fields from matches.
2212 if (prog->nparens) {
2213 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2220 if (regmatch(prog->program + 1)) {
2221 prog->endp[0] = PL_reginput - PL_bostr;
2224 REGCP_UNWIND(lastcp);
2228 #define RE_UNWIND_BRANCH 1
2229 #define RE_UNWIND_BRANCHJ 2
2233 typedef struct { /* XX: makes sense to enlarge it... */
2237 } re_unwind_generic_t;
2250 } re_unwind_branch_t;
2252 typedef union re_unwind_t {
2254 re_unwind_generic_t generic;
2255 re_unwind_branch_t branch;
2258 #define sayYES goto yes
2259 #define sayNO goto no
2260 #define sayNO_ANYOF goto no_anyof
2261 #define sayYES_FINAL goto yes_final
2262 #define sayYES_LOUD goto yes_loud
2263 #define sayNO_FINAL goto no_final
2264 #define sayNO_SILENT goto do_no
2265 #define saySAME(x) if (x) goto yes; else goto no
2267 #define POSCACHE_SUCCESS 0 /* caching success rather than failure */
2268 #define POSCACHE_SEEN 1 /* we know what we're caching */
2269 #define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */
2270 #define CACHEsayYES STMT_START { \
2271 if (PL_reg_poscache) { \
2272 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2273 PL_reg_poscache[0] |= (1<<POSCACHE_SUCCESS) || (1<<POSCACHE_SEEN); \
2274 else if (!(PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2275 /* cache records failure, but this is success */ \
2276 I32 o = locinput - PL_bostr, b; \
2277 o = (scan->flags & 0xf) - 1 + POSCACHE_START + o * (scan->flags>>4); \
2281 PerlIO_printf(Perl_debug_log, \
2282 "%*s (remove success from failure cache)\n", \
2283 REPORT_CODE_OFF+PL_regindent*2, "") \
2285 PL_reg_poscache[o] &= ~(1<<b); \
2290 #define CACHEsayNO STMT_START { \
2291 if (PL_reg_poscache) { \
2292 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2293 PL_reg_poscache[0] |= (1<<POSCACHE_SEEN); \
2294 else if ((PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2295 /* cache records success, but this is failure */ \
2296 I32 o = locinput - PL_bostr, b; \
2297 o = (scan->flags & 0xf) - 1 + POSCACHE_START + o * (scan->flags>>4); \
2301 PerlIO_printf(Perl_debug_log, \
2302 "%*s (remove failure from success cache)\n", \
2303 REPORT_CODE_OFF+PL_regindent*2, "") \
2305 PL_reg_poscache[o] &= ~(1<<b); \
2311 /* this is used to determine how far from the left messages like
2312 'failed...' are printed. Currently 29 makes these messages line
2313 up with the opcode they refer to. Earlier perls used 25 which
2314 left these messages outdented making reviewing a debug output
2317 #define REPORT_CODE_OFF 29
2320 /* Make sure there is a test for this +1 options in re_tests */
2321 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2323 #define TRIE_CHECK_STATE_IS_ACCEPTING STMT_START { \
2324 if ( trie->states[ state ].wordnum ) { \
2325 if ( !accepted ) { \
2328 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN ; \
2329 sv_accept_buff=NEWSV( 1234, \
2330 bufflen * sizeof(reg_trie_accepted) - 1 ); \
2331 SvCUR_set( sv_accept_buff, sizeof(reg_trie_accepted) ); \
2332 SvPOK_on( sv_accept_buff ); \
2333 sv_2mortal( sv_accept_buff ); \
2334 accept_buff = (reg_trie_accepted*)SvPV_nolen( sv_accept_buff );\
2336 if ( accepted >= bufflen ) { \
2338 accept_buff =(reg_trie_accepted*)SvGROW( sv_accept_buff, \
2339 bufflen * sizeof(reg_trie_accepted) ); \
2341 SvCUR_set( sv_accept_buff,SvCUR( sv_accept_buff ) \
2342 + sizeof( reg_trie_accepted ) ); \
2344 accept_buff[ accepted ].wordnum = trie->states[ state ].wordnum; \
2345 accept_buff[ accepted ].endpos = uc; \
2349 #define TRIE_HANDLE_CHAR STMT_START { \
2350 if ( uvc < 256 ) { \
2351 charid = trie->charmap[ uvc ]; \
2354 if( trie->widecharmap ) { \
2355 SV** svpp = (SV**)NULL; \
2356 svpp = hv_fetch( trie->widecharmap, (char*)&uvc, \
2357 sizeof( UV ), 0 ); \
2359 charid = (U16)SvIV( *svpp ); \
2364 ( base + charid > trie->uniquecharcount ) && \
2365 ( base + charid - 1 - trie->uniquecharcount < trie->lasttrans) && \
2366 trie->trans[ base + charid - 1 - trie->uniquecharcount ].check == state ) \
2368 state = trie->trans[ base + charid - 1 - trie->uniquecharcount ].next; \
2376 - regmatch - main matching routine
2378 * Conceptually the strategy is simple: check to see whether the current
2379 * node matches, call self recursively to see whether the rest matches,
2380 * and then act accordingly. In practice we make some effort to avoid
2381 * recursion, in particular by going through "ordinary" nodes (that don't
2382 * need to know whether the rest of the match failed) by a loop instead of
2385 /* [lwall] I've hoisted the register declarations to the outer block in order to
2386 * maybe save a little bit of pushing and popping on the stack. It also takes
2387 * advantage of machines that use a register save mask on subroutine entry.
2389 STATIC I32 /* 0 failure, 1 success */
2390 S_regmatch(pTHX_ regnode *prog)
2392 register regnode *scan; /* Current node. */
2393 regnode *next; /* Next node. */
2394 regnode *inner; /* Next node in internal branch. */
2395 register I32 nextchr; /* renamed nextchr - nextchar colides with
2396 function of same name */
2397 register I32 n; /* no or next */
2398 register I32 ln = 0; /* len or last */
2399 register char *s = Nullch; /* operand or save */
2400 register char *locinput = PL_reginput;
2401 register I32 c1 = 0, c2 = 0, paren; /* case fold search, parenth */
2402 int minmod = 0, sw = 0, logical = 0;
2405 /* used by the trie code */
2406 SV *sv_accept_buff; /* accepting states we have traversed */
2407 reg_trie_accepted *accept_buff; /* "" */
2408 reg_trie_data *trie; /* what trie are we using right now */
2409 U32 accepted = 0; /* how many accepting states we have seen*/
2412 I32 firstcp = PL_savestack_ix;
2414 register bool do_utf8 = PL_reg_match_utf8;
2416 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
2417 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
2418 SV *dsv2 = PERL_DEBUG_PAD_ZERO(2);
2430 /* Note that nextchr is a byte even in UTF */
2431 nextchr = UCHARAT(locinput);
2433 while (scan != NULL) {
2436 SV *prop = sv_newmortal();
2437 int docolor = *PL_colors[0];
2438 int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2439 int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput);
2440 /* The part of the string before starttry has one color
2441 (pref0_len chars), between starttry and current
2442 position another one (pref_len - pref0_len chars),
2443 after the current position the third one.
2444 We assume that pref0_len <= pref_len, otherwise we
2445 decrease pref0_len. */
2446 int pref_len = (locinput - PL_bostr) > (5 + taill) - l
2447 ? (5 + taill) - l : locinput - PL_bostr;
2450 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2452 pref0_len = pref_len - (locinput - PL_reg_starttry);
2453 if (l + pref_len < (5 + taill) && l < PL_regeol - locinput)
2454 l = ( PL_regeol - locinput > (5 + taill) - pref_len
2455 ? (5 + taill) - pref_len : PL_regeol - locinput);
2456 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2460 if (pref0_len > pref_len)
2461 pref0_len = pref_len;
2462 regprop(prop, scan);
2465 do_utf8 && OP(scan) != CANY ?
2466 pv_uni_display(dsv0, (U8*)(locinput - pref_len),
2467 pref0_len, 60, UNI_DISPLAY_REGEX) :
2468 locinput - pref_len;
2469 int len0 = do_utf8 ? strlen(s0) : pref0_len;
2470 char *s1 = do_utf8 && OP(scan) != CANY ?
2471 pv_uni_display(dsv1, (U8*)(locinput - pref_len + pref0_len),
2472 pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) :
2473 locinput - pref_len + pref0_len;
2474 int len1 = do_utf8 ? strlen(s1) : pref_len - pref0_len;
2475 char *s2 = do_utf8 && OP(scan) != CANY ?
2476 pv_uni_display(dsv2, (U8*)locinput,
2477 PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) :
2479 int len2 = do_utf8 ? strlen(s2) : l;
2480 PerlIO_printf(Perl_debug_log,
2481 "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|%3"IVdf":%*s%s\n",
2482 (IV)(locinput - PL_bostr),
2489 (docolor ? "" : "> <"),
2493 15 - l - pref_len + 1,
2495 (IV)(scan - PL_regprogram), PL_regindent*2, "",
2500 next = scan + NEXT_OFF(scan);
2506 if (locinput == PL_bostr)
2508 /* regtill = regbol; */
2513 if (locinput == PL_bostr ||
2514 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2520 if (locinput == PL_bostr)
2524 if (locinput == PL_reg_ganch)
2530 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2535 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2537 if (PL_regeol - locinput > 1)
2541 if (PL_regeol != locinput)
2545 if (!nextchr && locinput >= PL_regeol)
2548 locinput += PL_utf8skip[nextchr];
2549 if (locinput > PL_regeol)
2551 nextchr = UCHARAT(locinput);
2554 nextchr = UCHARAT(++locinput);
2557 if (!nextchr && locinput >= PL_regeol)
2559 nextchr = UCHARAT(++locinput);
2562 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2565 locinput += PL_utf8skip[nextchr];
2566 if (locinput > PL_regeol)
2568 nextchr = UCHARAT(locinput);
2571 nextchr = UCHARAT(++locinput);
2577 traverse the TRIE keeping track of all accepting states
2578 we transition through until we get to a failing node.
2580 we use two slightly different pieces of code to handle
2581 the traversal depending on whether its case sensitive or
2582 not. we reuse the accept code however. (this should probably
2583 be turned into a macro.)
2590 U32 uniflags = ckWARN( WARN_UTF8 ) ? 0 : UTF8_ALLOW_ANY;
2591 U8 *uc = ( U8* )locinput;
2598 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2599 U8 *uscan = (U8*)NULL;
2603 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2605 while ( state && uc <= (U8*)PL_regeol ) {
2607 TRIE_CHECK_STATE_IS_ACCEPTING;
2609 base = trie->states[ state ].trans.base;
2611 DEBUG_TRIE_EXECUTE_r(
2612 PerlIO_printf( Perl_debug_log,
2613 "%*s %sState: %4x, Base: %4x Accepted: %4x ",
2614 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2615 state, base, accepted );
2620 if ( do_utf8 || UTF ) {
2622 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags );
2627 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2628 uvc = to_uni_fold( uvc, foldbuf, &foldlen );
2629 foldlen -= UNISKIP( uvc );
2630 uscan = foldbuf + UNISKIP( uvc );
2642 DEBUG_TRIE_EXECUTE_r(
2643 PerlIO_printf( Perl_debug_log,
2644 "Charid:%3x CV:%4x After State: %4x%s\n",
2645 charid, uvc, state, PL_colors[5] );
2654 /* unreached codepoint: we jump into the middle of the next case
2655 from previous if blocks */
2658 U32 uniflags = ckWARN( WARN_UTF8 ) ? 0 : UTF8_ALLOW_ANY;
2659 U8 *uc = (U8*)locinput;
2668 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2670 while ( state && uc <= (U8*)PL_regeol ) {
2672 TRIE_CHECK_STATE_IS_ACCEPTING;
2674 base = trie->states[ state ].trans.base;
2676 DEBUG_TRIE_EXECUTE_r(
2677 PerlIO_printf( Perl_debug_log,
2678 "%*s %sState: %4x, Base: %4x Accepted: %4x ",
2679 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2680 state, base, accepted );
2685 if ( do_utf8 || UTF ) {
2686 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2697 DEBUG_TRIE_EXECUTE_r(
2698 PerlIO_printf( Perl_debug_log,
2699 "Charid:%3x CV:%4x After State: %4x%s\n",
2700 charid, uvc, state, PL_colors[5] );
2710 There was at least one accepting state that we
2711 transitioned through. Presumably the number of accepting
2712 states is going to be low, typically one or two. So we
2713 simply scan through to find the one with lowest wordnum.
2714 Once we find it, we swap the last state into its place
2715 and decrement the size. We then try to match the rest of
2716 the pattern at the point where the word ends, if we
2717 succeed then we end the loop, otherwise the loop
2718 eventually terminates once all of the accepting states
2725 if ( accepted == 1 ) {
2727 SV **tmp = av_fetch( trie->words, accept_buff[ 0 ].wordnum-1, 0 );
2728 PerlIO_printf( Perl_debug_log,
2729 "%*s %sonly one match : #%d <%s>%s\n",
2730 REPORT_CODE_OFF+PL_regindent*2, "",PL_colors[4],
2731 accept_buff[ 0 ].wordnum,
2732 tmp ? SvPV_nolen( *tmp ) : "not compiled under -Dr",
2735 PL_reginput = (char *)accept_buff[ 0 ].endpos;
2736 /* in this case we free tmps/leave before we call regmatch
2737 as we wont be using accept_buff again. */
2740 gotit = regmatch( scan + NEXT_OFF( scan ) );
2743 PerlIO_printf( Perl_debug_log,"%*s %sgot %d possible matches%s\n",
2744 REPORT_CODE_OFF + PL_regindent * 2, "",PL_colors[4], accepted,
2747 while ( !gotit && accepted-- ) {
2750 for( cur = 1 ; cur <= accepted ; cur++ ) {
2751 DEBUG_TRIE_EXECUTE_r(
2752 PerlIO_printf( Perl_debug_log,
2753 "%*s %sgot %d (%d) as best, looking at %d (%d)%s\n",
2754 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2755 best, accept_buff[ best ].wordnum, cur,
2756 accept_buff[ cur ].wordnum, PL_colors[5] );
2759 if ( accept_buff[ cur ].wordnum < accept_buff[ best ].wordnum )
2763 SV **tmp = av_fetch( trie->words, accept_buff[ best ].wordnum - 1, 0 );
2764 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at 0x%p%s\n",
2765 REPORT_CODE_OFF+PL_regindent*2, "",PL_colors[4],
2766 accept_buff[best].wordnum,
2767 tmp ? SvPV_nolen( *tmp ) : "not compiled under -Dr",scan,
2770 if ( best<accepted ) {
2771 reg_trie_accepted tmp = accept_buff[ best ];
2772 accept_buff[ best ] = accept_buff[ accepted ];
2773 accept_buff[ accepted ] = tmp;
2776 PL_reginput = (char *)accept_buff[ best ].endpos;
2779 as far as I can tell we only need the SAVETMPS/FREETMPS
2780 for re's with EVAL in them but I'm leaving them in for
2781 all until I can be sure.
2784 gotit = regmatch( scan + NEXT_OFF( scan ) ) ;
2797 /* unreached codepoint */
2801 if (do_utf8 != UTF) {
2802 /* The target and the pattern have differing utf8ness. */
2808 /* The target is utf8, the pattern is not utf8. */
2812 if (NATIVE_TO_UNI(*(U8*)s) !=
2813 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2815 0 : UTF8_ALLOW_ANY))
2822 /* The target is not utf8, the pattern is utf8. */
2826 if (NATIVE_TO_UNI(*((U8*)l)) !=
2827 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2829 0 : UTF8_ALLOW_ANY))
2836 nextchr = UCHARAT(locinput);
2839 /* The target and the pattern have the same utf8ness. */
2840 /* Inline the first character, for speed. */
2841 if (UCHARAT(s) != nextchr)
2843 if (PL_regeol - locinput < ln)
2845 if (ln > 1 && memNE(s, locinput, ln))
2848 nextchr = UCHARAT(locinput);
2851 PL_reg_flags |= RF_tainted;
2857 if (do_utf8 || UTF) {
2858 /* Either target or the pattern are utf8. */
2860 char *e = PL_regeol;
2862 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
2863 l, &e, 0, do_utf8)) {
2864 /* One more case for the sharp s:
2865 * pack("U0U*", 0xDF) =~ /ss/i,
2866 * the 0xC3 0x9F are the UTF-8
2867 * byte sequence for the U+00DF. */
2869 toLOWER(s[0]) == 's' &&
2871 toLOWER(s[1]) == 's' &&
2878 nextchr = UCHARAT(locinput);
2882 /* Neither the target and the pattern are utf8. */
2884 /* Inline the first character, for speed. */
2885 if (UCHARAT(s) != nextchr &&
2886 UCHARAT(s) != ((OP(scan) == EXACTF)
2887 ? PL_fold : PL_fold_locale)[nextchr])
2889 if (PL_regeol - locinput < ln)
2891 if (ln > 1 && (OP(scan) == EXACTF
2892 ? ibcmp(s, locinput, ln)
2893 : ibcmp_locale(s, locinput, ln)))
2896 nextchr = UCHARAT(locinput);
2900 STRLEN inclasslen = PL_regeol - locinput;
2902 if (!reginclass(scan, (U8*)locinput, &inclasslen, do_utf8))
2904 if (locinput >= PL_regeol)
2906 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
2907 nextchr = UCHARAT(locinput);
2912 nextchr = UCHARAT(locinput);
2913 if (!REGINCLASS(scan, (U8*)locinput))
2915 if (!nextchr && locinput >= PL_regeol)
2917 nextchr = UCHARAT(++locinput);
2921 /* If we might have the case of the German sharp s
2922 * in a casefolding Unicode character class. */
2924 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
2925 locinput += SHARP_S_SKIP;
2926 nextchr = UCHARAT(locinput);
2932 PL_reg_flags |= RF_tainted;
2938 LOAD_UTF8_CHARCLASS(alnum,"a");
2939 if (!(OP(scan) == ALNUM
2940 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2941 : isALNUM_LC_utf8((U8*)locinput)))
2945 locinput += PL_utf8skip[nextchr];
2946 nextchr = UCHARAT(locinput);
2949 if (!(OP(scan) == ALNUM
2950 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
2952 nextchr = UCHARAT(++locinput);
2955 PL_reg_flags |= RF_tainted;
2958 if (!nextchr && locinput >= PL_regeol)
2961 LOAD_UTF8_CHARCLASS(alnum,"a");
2962 if (OP(scan) == NALNUM
2963 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2964 : isALNUM_LC_utf8((U8*)locinput))
2968 locinput += PL_utf8skip[nextchr];
2969 nextchr = UCHARAT(locinput);
2972 if (OP(scan) == NALNUM
2973 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
2975 nextchr = UCHARAT(++locinput);
2979 PL_reg_flags |= RF_tainted;
2983 /* was last char in word? */
2985 if (locinput == PL_bostr)
2988 U8 *r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
2990 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
2992 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
2993 ln = isALNUM_uni(ln);
2994 LOAD_UTF8_CHARCLASS(alnum,"a");
2995 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
2998 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
2999 n = isALNUM_LC_utf8((U8*)locinput);
3003 ln = (locinput != PL_bostr) ?
3004 UCHARAT(locinput - 1) : '\n';
3005 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3007 n = isALNUM(nextchr);
3010 ln = isALNUM_LC(ln);
3011 n = isALNUM_LC(nextchr);
3014 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3015 OP(scan) == BOUNDL))
3019 PL_reg_flags |= RF_tainted;
3025 if (UTF8_IS_CONTINUED(nextchr)) {
3026 LOAD_UTF8_CHARCLASS(space," ");
3027 if (!(OP(scan) == SPACE
3028 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3029 : isSPACE_LC_utf8((U8*)locinput)))
3033 locinput += PL_utf8skip[nextchr];
3034 nextchr = UCHARAT(locinput);
3037 if (!(OP(scan) == SPACE
3038 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3040 nextchr = UCHARAT(++locinput);
3043 if (!(OP(scan) == SPACE
3044 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3046 nextchr = UCHARAT(++locinput);
3050 PL_reg_flags |= RF_tainted;
3053 if (!nextchr && locinput >= PL_regeol)
3056 LOAD_UTF8_CHARCLASS(space," ");
3057 if (OP(scan) == NSPACE
3058 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3059 : isSPACE_LC_utf8((U8*)locinput))
3063 locinput += PL_utf8skip[nextchr];
3064 nextchr = UCHARAT(locinput);
3067 if (OP(scan) == NSPACE
3068 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3070 nextchr = UCHARAT(++locinput);
3073 PL_reg_flags |= RF_tainted;
3079 LOAD_UTF8_CHARCLASS(digit,"0");
3080 if (!(OP(scan) == DIGIT
3081 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3082 : isDIGIT_LC_utf8((U8*)locinput)))
3086 locinput += PL_utf8skip[nextchr];
3087 nextchr = UCHARAT(locinput);
3090 if (!(OP(scan) == DIGIT
3091 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3093 nextchr = UCHARAT(++locinput);
3096 PL_reg_flags |= RF_tainted;
3099 if (!nextchr && locinput >= PL_regeol)
3102 LOAD_UTF8_CHARCLASS(digit,"0");
3103 if (OP(scan) == NDIGIT
3104 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3105 : isDIGIT_LC_utf8((U8*)locinput))
3109 locinput += PL_utf8skip[nextchr];
3110 nextchr = UCHARAT(locinput);
3113 if (OP(scan) == NDIGIT
3114 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3116 nextchr = UCHARAT(++locinput);
3119 if (locinput >= PL_regeol)
3122 LOAD_UTF8_CHARCLASS(mark,"~");
3123 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3125 locinput += PL_utf8skip[nextchr];
3126 while (locinput < PL_regeol &&
3127 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3128 locinput += UTF8SKIP(locinput);
3129 if (locinput > PL_regeol)
3134 nextchr = UCHARAT(locinput);
3137 PL_reg_flags |= RF_tainted;
3141 n = ARG(scan); /* which paren pair */
3142 ln = PL_regstartp[n];
3143 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3144 if ((I32)*PL_reglastparen < n || ln == -1)
3145 sayNO; /* Do not match unless seen CLOSEn. */
3146 if (ln == PL_regendp[n])
3150 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3152 char *e = PL_bostr + PL_regendp[n];
3154 * Note that we can't do the "other character" lookup trick as
3155 * in the 8-bit case (no pun intended) because in Unicode we
3156 * have to map both upper and title case to lower case.
3158 if (OP(scan) == REFF) {
3159 STRLEN ulen1, ulen2;
3160 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3161 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3165 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3166 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3167 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3174 nextchr = UCHARAT(locinput);
3178 /* Inline the first character, for speed. */
3179 if (UCHARAT(s) != nextchr &&
3181 (UCHARAT(s) != ((OP(scan) == REFF
3182 ? PL_fold : PL_fold_locale)[nextchr]))))
3184 ln = PL_regendp[n] - ln;
3185 if (locinput + ln > PL_regeol)
3187 if (ln > 1 && (OP(scan) == REF
3188 ? memNE(s, locinput, ln)
3190 ? ibcmp(s, locinput, ln)
3191 : ibcmp_locale(s, locinput, ln))))
3194 nextchr = UCHARAT(locinput);
3205 OP_4tree *oop = PL_op;
3206 COP *ocurcop = PL_curcop;
3209 struct regexp *oreg = PL_reg_re;
3212 PL_op = (OP_4tree*)PL_regdata->data[n];
3213 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3214 PAD_SAVE_LOCAL(old_comppad, (PAD*)PL_regdata->data[n + 2]);
3215 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3219 CALLRUNOPS(aTHX); /* Scalar context. */
3222 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3230 PAD_RESTORE_LOCAL(old_comppad);
3231 PL_curcop = ocurcop;
3233 if (logical == 2) { /* Postponed subexpression. */
3235 MAGIC *mg = Null(MAGIC*);
3237 CHECKPOINT cp, lastcp;
3241 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3242 mg = mg_find(sv, PERL_MAGIC_qr);
3243 else if (SvSMAGICAL(ret)) {
3244 if (SvGMAGICAL(ret))
3245 sv_unmagic(ret, PERL_MAGIC_qr);
3247 mg = mg_find(ret, PERL_MAGIC_qr);
3251 re = (regexp *)mg->mg_obj;
3252 (void)ReREFCNT_inc(re);
3256 char *t = SvPV(ret, len);
3258 char *oprecomp = PL_regprecomp;
3259 I32 osize = PL_regsize;
3260 I32 onpar = PL_regnpar;
3263 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3264 re = CALLREGCOMP(aTHX_ t, t + len, &pm);
3266 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3268 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3270 PL_regprecomp = oprecomp;
3275 PerlIO_printf(Perl_debug_log,
3276 "Entering embedded `%s%.60s%s%s'\n",
3280 (strlen(re->precomp) > 60 ? "..." : ""))
3283 state.prev = PL_reg_call_cc;
3284 state.cc = PL_regcc;
3285 state.re = PL_reg_re;
3289 cp = regcppush(0); /* Save *all* the positions. */
3292 state.ss = PL_savestack_ix;
3293 *PL_reglastparen = 0;
3294 *PL_reglastcloseparen = 0;
3295 PL_reg_call_cc = &state;
3296 PL_reginput = locinput;
3297 toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3298 ((re->reganch & ROPT_UTF8) != 0);
3299 if (toggleutf) PL_reg_flags ^= RF_utf8;
3301 /* XXXX This is too dramatic a measure... */
3304 if (regmatch(re->program + 1)) {
3305 /* Even though we succeeded, we need to restore
3306 global variables, since we may be wrapped inside
3307 SUSPEND, thus the match may be not finished yet. */
3309 /* XXXX Do this only if SUSPENDed? */
3310 PL_reg_call_cc = state.prev;
3311 PL_regcc = state.cc;
3312 PL_reg_re = state.re;
3313 cache_re(PL_reg_re);
3314 if (toggleutf) PL_reg_flags ^= RF_utf8;
3316 /* XXXX This is too dramatic a measure... */
3319 /* These are needed even if not SUSPEND. */
3325 REGCP_UNWIND(lastcp);
3327 PL_reg_call_cc = state.prev;
3328 PL_regcc = state.cc;
3329 PL_reg_re = state.re;
3330 cache_re(PL_reg_re);
3331 if (toggleutf) PL_reg_flags ^= RF_utf8;
3333 /* XXXX This is too dramatic a measure... */
3343 sv_setsv(save_scalar(PL_replgv), ret);
3349 n = ARG(scan); /* which paren pair */
3350 PL_reg_start_tmp[n] = locinput;
3355 n = ARG(scan); /* which paren pair */
3356 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3357 PL_regendp[n] = locinput - PL_bostr;
3358 if (n > (I32)*PL_reglastparen)
3359 *PL_reglastparen = n;
3360 *PL_reglastcloseparen = n;
3363 n = ARG(scan); /* which paren pair */
3364 sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3367 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3369 next = NEXTOPER(NEXTOPER(scan));
3371 next = scan + ARG(scan);
3372 if (OP(next) == IFTHEN) /* Fake one. */
3373 next = NEXTOPER(NEXTOPER(next));
3377 logical = scan->flags;
3379 /*******************************************************************
3380 PL_regcc contains infoblock about the innermost (...)* loop, and
3381 a pointer to the next outer infoblock.
3383 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3385 1) After matching X, regnode for CURLYX is processed;
3387 2) This regnode creates infoblock on the stack, and calls
3388 regmatch() recursively with the starting point at WHILEM node;
3390 3) Each hit of WHILEM node tries to match A and Z (in the order
3391 depending on the current iteration, min/max of {min,max} and
3392 greediness). The information about where are nodes for "A"
3393 and "Z" is read from the infoblock, as is info on how many times "A"
3394 was already matched, and greediness.
3396 4) After A matches, the same WHILEM node is hit again.
3398 5) Each time WHILEM is hit, PL_regcc is the infoblock created by CURLYX
3399 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3400 resets PL_regcc, since this Y(A)*Z can be a part of some other loop:
3401 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3402 of the external loop.
3404 Currently present infoblocks form a tree with a stem formed by PL_curcc
3405 and whatever it mentions via ->next, and additional attached trees
3406 corresponding to temporarily unset infoblocks as in "5" above.
3408 In the following picture infoblocks for outer loop of
3409 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3410 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3411 infoblocks are drawn below the "reset" infoblock.
3413 In fact in the picture below we do not show failed matches for Z and T
3414 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3415 more obvious *why* one needs to *temporary* unset infoblocks.]
3417 Matched REx position InfoBlocks Comment
3421 Y A)*?Z)*?T x <- O <- I
3422 YA )*?Z)*?T x <- O <- I
3423 YA A)*?Z)*?T x <- O <- I
3424 YAA )*?Z)*?T x <- O <- I
3425 YAA Z)*?T x <- O # Temporary unset I
3428 YAAZ Y(A)*?Z)*?T x <- O
3431 YAAZY (A)*?Z)*?T x <- O
3434 YAAZY A)*?Z)*?T x <- O <- I
3437 YAAZYA )*?Z)*?T x <- O <- I
3440 YAAZYA Z)*?T x <- O # Temporary unset I
3446 YAAZYAZ T x # Temporary unset O
3453 *******************************************************************/
3456 CHECKPOINT cp = PL_savestack_ix;
3457 /* No need to save/restore up to this paren */
3458 I32 parenfloor = scan->flags;
3460 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3462 cc.oldcc = PL_regcc;
3464 /* XXXX Probably it is better to teach regpush to support
3465 parenfloor > PL_regsize... */
3466 if (parenfloor > (I32)*PL_reglastparen)
3467 parenfloor = *PL_reglastparen; /* Pessimization... */
3468 cc.parenfloor = parenfloor;
3470 cc.min = ARG1(scan);
3471 cc.max = ARG2(scan);
3472 cc.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3476 PL_reginput = locinput;
3477 n = regmatch(PREVOPER(next)); /* start on the WHILEM */
3479 PL_regcc = cc.oldcc;
3485 * This is really hard to understand, because after we match
3486 * what we're trying to match, we must make sure the rest of
3487 * the REx is going to match for sure, and to do that we have
3488 * to go back UP the parse tree by recursing ever deeper. And
3489 * if it fails, we have to reset our parent's current state
3490 * that we can try again after backing off.
3493 CHECKPOINT cp, lastcp;
3494 CURCUR* cc = PL_regcc;
3495 char *lastloc = cc->lastloc; /* Detection of 0-len. */
3497 n = cc->cur + 1; /* how many we know we matched */
3498 PL_reginput = locinput;
3501 PerlIO_printf(Perl_debug_log,
3502 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3503 REPORT_CODE_OFF+PL_regindent*2, "",
3504 (long)n, (long)cc->min,
3505 (long)cc->max, PTR2UV(cc))
3508 /* If degenerate scan matches "", assume scan done. */
3510 if (locinput == cc->lastloc && n >= cc->min) {
3511 PL_regcc = cc->oldcc;
3515 PerlIO_printf(Perl_debug_log,
3516 "%*s empty match detected, try continuation...\n",
3517 REPORT_CODE_OFF+PL_regindent*2, "")
3519 if (regmatch(cc->next))
3527 /* First just match a string of min scans. */
3531 cc->lastloc = locinput;
3532 if (regmatch(cc->scan))
3535 cc->lastloc = lastloc;
3540 /* Check whether we already were at this position.
3541 Postpone detection until we know the match is not
3542 *that* much linear. */
3543 if (!PL_reg_maxiter) {
3544 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3545 PL_reg_leftiter = PL_reg_maxiter;
3547 if (PL_reg_leftiter-- == 0) {
3548 I32 size = (PL_reg_maxiter + 7 + POSCACHE_START)/8;
3549 if (PL_reg_poscache) {
3550 if ((I32)PL_reg_poscache_size < size) {
3551 Renew(PL_reg_poscache, size, char);
3552 PL_reg_poscache_size = size;
3554 Zero(PL_reg_poscache, size, char);
3557 PL_reg_poscache_size = size;
3558 Newz(29, PL_reg_poscache, size, char);
3561 PerlIO_printf(Perl_debug_log,
3562 "%sDetected a super-linear match, switching on caching%s...\n",
3563 PL_colors[4], PL_colors[5])
3566 if (PL_reg_leftiter < 0) {
3567 I32 o = locinput - PL_bostr, b;
3569 o = (scan->flags & 0xf) - 1 + POSCACHE_START + o * (scan->flags>>4);
3572 if (PL_reg_poscache[o] & (1<<b)) {
3574 PerlIO_printf(Perl_debug_log,
3575 "%*s already tried at this position...\n",
3576 REPORT_CODE_OFF+PL_regindent*2, "")
3578 if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))
3579 /* cache records success */
3582 /* cache records failure */
3585 PL_reg_poscache[o] |= (1<<b);
3589 /* Prefer next over scan for minimal matching. */
3592 PL_regcc = cc->oldcc;
3595 cp = regcppush(cc->parenfloor);
3597 if (regmatch(cc->next)) {
3599 CACHEsayYES; /* All done. */
3601 REGCP_UNWIND(lastcp);
3607 if (n >= cc->max) { /* Maximum greed exceeded? */
3608 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3609 && !(PL_reg_flags & RF_warned)) {
3610 PL_reg_flags |= RF_warned;
3611 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3612 "Complex regular subexpression recursion",
3619 PerlIO_printf(Perl_debug_log,
3620 "%*s trying longer...\n",
3621 REPORT_CODE_OFF+PL_regindent*2, "")
3623 /* Try scanning more and see if it helps. */
3624 PL_reginput = locinput;
3626 cc->lastloc = locinput;
3627 cp = regcppush(cc->parenfloor);
3629 if (regmatch(cc->scan)) {
3633 REGCP_UNWIND(lastcp);
3636 cc->lastloc = lastloc;
3640 /* Prefer scan over next for maximal matching. */
3642 if (n < cc->max) { /* More greed allowed? */
3643 cp = regcppush(cc->parenfloor);
3645 cc->lastloc = locinput;
3647 if (regmatch(cc->scan)) {
3651 REGCP_UNWIND(lastcp);
3652 regcppop(); /* Restore some previous $<digit>s? */
3653 PL_reginput = locinput;
3655 PerlIO_printf(Perl_debug_log,
3656 "%*s failed, try continuation...\n",
3657 REPORT_CODE_OFF+PL_regindent*2, "")
3660 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3661 && !(PL_reg_flags & RF_warned)) {
3662 PL_reg_flags |= RF_warned;
3663 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3664 "Complex regular subexpression recursion",
3668 /* Failed deeper matches of scan, so see if this one works. */
3669 PL_regcc = cc->oldcc;
3672 if (regmatch(cc->next))
3678 cc->lastloc = lastloc;
3683 next = scan + ARG(scan);
3686 inner = NEXTOPER(NEXTOPER(scan));
3689 inner = NEXTOPER(scan);
3693 if (OP(next) != c1) /* No choice. */
3694 next = inner; /* Avoid recursion. */
3696 I32 lastparen = *PL_reglastparen;
3698 re_unwind_branch_t *uw;
3700 /* Put unwinding data on stack */
3701 unwind1 = SSNEWt(1,re_unwind_branch_t);
3702 uw = SSPTRt(unwind1,re_unwind_branch_t);
3705 uw->type = ((c1 == BRANCH)
3707 : RE_UNWIND_BRANCHJ);
3708 uw->lastparen = lastparen;
3710 uw->locinput = locinput;
3711 uw->nextchr = nextchr;
3713 uw->regindent = ++PL_regindent;
3716 REGCP_SET(uw->lastcp);
3718 /* Now go into the first branch */
3731 /* We suppose that the next guy does not need
3732 backtracking: in particular, it is of constant non-zero length,
3733 and has no parenths to influence future backrefs. */
3734 ln = ARG1(scan); /* min to match */
3735 n = ARG2(scan); /* max to match */
3736 paren = scan->flags;
3738 if (paren > PL_regsize)
3740 if (paren > (I32)*PL_reglastparen)
3741 *PL_reglastparen = paren;
3743 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3745 scan += NEXT_OFF(scan); /* Skip former OPEN. */
3746 PL_reginput = locinput;
3749 if (ln && regrepeat_hard(scan, ln, &l) < ln)
3751 locinput = PL_reginput;
3752 if (HAS_TEXT(next) || JUMPABLE(next)) {
3753 regnode *text_node = next;
3755 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3757 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3759 if (PL_regkind[(U8)OP(text_node)] == REF) {
3763 else { c1 = (U8)*STRING(text_node); }
3764 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3766 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3767 c2 = PL_fold_locale[c1];
3776 while (n >= ln || (n == REG_INFTY && ln > 0)) { /* ln overflow ? */
3777 /* If it could work, try it. */
3779 UCHARAT(PL_reginput) == c1 ||
3780 UCHARAT(PL_reginput) == c2)
3784 PL_regstartp[paren] =
3785 HOPc(PL_reginput, -l) - PL_bostr;
3786 PL_regendp[paren] = PL_reginput - PL_bostr;
3789 PL_regendp[paren] = -1;
3793 REGCP_UNWIND(lastcp);
3795 /* Couldn't or didn't -- move forward. */
3796 PL_reginput = locinput;
3797 if (regrepeat_hard(scan, 1, &l)) {
3799 locinput = PL_reginput;
3806 n = regrepeat_hard(scan, n, &l);
3807 locinput = PL_reginput;
3809 PerlIO_printf(Perl_debug_log,
3810 "%*s matched %"IVdf" times, len=%"IVdf"...\n",
3811 (int)(REPORT_CODE_OFF+PL_regindent*2), "",
3815 if (HAS_TEXT(next) || JUMPABLE(next)) {
3816 regnode *text_node = next;
3818 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3820 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3822 if (PL_regkind[(U8)OP(text_node)] == REF) {
3826 else { c1 = (U8)*STRING(text_node); }
3828 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3830 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3831 c2 = PL_fold_locale[c1];
3842 /* If it could work, try it. */
3844 UCHARAT(PL_reginput) == c1 ||
3845 UCHARAT(PL_reginput) == c2)
3848 PerlIO_printf(Perl_debug_log,
3849 "%*s trying tail with n=%"IVdf"...\n",
3850 (int)(REPORT_CODE_OFF+PL_regindent*2), "", (IV)n)
3854 PL_regstartp[paren] = HOPc(PL_reginput, -l) - PL_bostr;
3855 PL_regendp[paren] = PL_reginput - PL_bostr;
3858 PL_regendp[paren] = -1;