5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
19 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
20 * confused with the original package (see point 3 below). Thanks, Henry!
23 /* Additional note: this code is very heavily munged from Henry's version
24 * in places. In some spots I've traded clarity for efficiency, so don't
25 * blame Henry for some of the lack of readability.
28 /* The names of the functions have been changed from regcomp and
29 * regexec to pregcomp and pregexec in order to avoid conflicts
30 * with the POSIX routines of the same names.
33 #ifdef PERL_EXT_RE_BUILD
34 /* need to replace pregcomp et al, so enable that */
35 # ifndef PERL_IN_XSUB_RE
36 # define PERL_IN_XSUB_RE
38 /* need access to debugger hooks */
39 # if defined(PERL_EXT_RE_DEBUG) && !defined(DEBUGGING)
44 #ifdef PERL_IN_XSUB_RE
45 /* We *really* need to overwrite these symbols: */
46 # define Perl_regexec_flags my_regexec
47 # define Perl_regdump my_regdump
48 # define Perl_regprop my_regprop
49 # define Perl_re_intuit_start my_re_intuit_start
50 /* *These* symbols are masked to allow static link. */
51 # define Perl_pregexec my_pregexec
52 # define Perl_reginitcolors my_reginitcolors
53 # define Perl_regclass_swash my_regclass_swash
55 # define PERL_NO_GET_CONTEXT
59 * pregcomp and pregexec -- regsub and regerror are not used in perl
61 * Copyright (c) 1986 by University of Toronto.
62 * Written by Henry Spencer. Not derived from licensed software.
64 * Permission is granted to anyone to use this software for any
65 * purpose on any computer system, and to redistribute it freely,
66 * subject to the following restrictions:
68 * 1. The author is not responsible for the consequences of use of
69 * this software, no matter how awful, even if they arise
72 * 2. The origin of this software must not be misrepresented, either
73 * by explicit claim or by omission.
75 * 3. Altered versions must be plainly marked as such, and must not
76 * be misrepresented as being the original software.
78 **** Alterations to Henry's code are...
80 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
81 **** 2000, 2001, 2002, 2003, 2004, 2005, by Larry Wall and others
83 **** You may distribute under the terms of either the GNU General Public
84 **** License or the Artistic License, as specified in the README file.
86 * Beware that some of this code is subtly aware of the way operator
87 * precedence is structured in regular expressions. Serious changes in
88 * regular-expression syntax might require a total rethink.
91 #define PERL_IN_REGEXEC_C
96 #define RF_tainted 1 /* tainted information used? */
97 #define RF_warned 2 /* warned about big count? */
98 #define RF_evaled 4 /* Did an EVAL with setting? */
99 #define RF_utf8 8 /* String contains multibyte chars? */
101 #define UTF ((PL_reg_flags & RF_utf8) != 0)
103 #define RS_init 1 /* eval environment created */
104 #define RS_set 2 /* replsv value is set */
107 #define STATIC static
110 #define REGINCLASS(p,c) (ANYOF_FLAGS(p) ? reginclass(p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
116 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
117 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
119 #define reghop_c(pos,off) ((char*)reghop((U8*)pos, off))
120 #define reghopmaybe_c(pos,off) ((char*)reghopmaybe((U8*)pos, off))
121 #define HOP(pos,off) (PL_reg_match_utf8 ? reghop((U8*)pos, off) : (U8*)(pos + off))
122 #define HOPMAYBE(pos,off) (PL_reg_match_utf8 ? reghopmaybe((U8*)pos, off) : (U8*)(pos + off))
123 #define HOPc(pos,off) ((char*)HOP(pos,off))
124 #define HOPMAYBEc(pos,off) ((char*)HOPMAYBE(pos,off))
126 #define HOPBACK(pos, off) ( \
127 (PL_reg_match_utf8) \
128 ? reghopmaybe((U8*)pos, -off) \
129 : (pos - off >= PL_bostr) \
133 #define HOPBACKc(pos, off) (char*)HOPBACK(pos, off)
135 #define reghop3_c(pos,off,lim) ((char*)reghop3((U8*)pos, off, (U8*)lim))
136 #define reghopmaybe3_c(pos,off,lim) ((char*)reghopmaybe3((U8*)pos, off, (U8*)lim))
137 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off))
138 #define HOPMAYBE3(pos,off,lim) (PL_reg_match_utf8 ? reghopmaybe3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off))
139 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
140 #define HOPMAYBE3c(pos,off,lim) ((char*)HOPMAYBE3(pos,off,lim))
142 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
143 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
144 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
145 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
146 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
147 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
149 /* for use after a quantifier and before an EXACT-like node -- japhy */
150 #define JUMPABLE(rn) ( \
151 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
152 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
153 OP(rn) == PLUS || OP(rn) == MINMOD || \
154 (PL_regkind[(U8)OP(rn)] == CURLY && ARG1(rn) > 0) \
157 #define HAS_TEXT(rn) ( \
158 PL_regkind[(U8)OP(rn)] == EXACT || PL_regkind[(U8)OP(rn)] == REF \
162 Search for mandatory following text node; for lookahead, the text must
163 follow but for lookbehind (rn->flags != 0) we skip to the next step.
165 #define FIND_NEXT_IMPT(rn) STMT_START { \
166 while (JUMPABLE(rn)) \
167 if (OP(rn) == SUSPEND || PL_regkind[(U8)OP(rn)] == CURLY) \
168 rn = NEXTOPER(NEXTOPER(rn)); \
169 else if (OP(rn) == PLUS) \
171 else if (OP(rn) == IFMATCH) \
172 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
173 else rn += NEXT_OFF(rn); \
176 static void restore_pos(pTHX_ void *arg);
179 S_regcppush(pTHX_ I32 parenfloor)
181 const int retval = PL_savestack_ix;
182 #define REGCP_PAREN_ELEMS 4
183 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
186 if (paren_elems_to_push < 0)
187 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
189 #define REGCP_OTHER_ELEMS 6
190 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
191 for (p = PL_regsize; p > parenfloor; p--) {
192 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
193 SSPUSHINT(PL_regendp[p]);
194 SSPUSHINT(PL_regstartp[p]);
195 SSPUSHPTR(PL_reg_start_tmp[p]);
198 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
199 SSPUSHINT(PL_regsize);
200 SSPUSHINT(*PL_reglastparen);
201 SSPUSHINT(*PL_reglastcloseparen);
202 SSPUSHPTR(PL_reginput);
203 #define REGCP_FRAME_ELEMS 2
204 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
205 * are needed for the regexp context stack bookkeeping. */
206 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
207 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
212 /* These are needed since we do not localize EVAL nodes: */
213 # define REGCP_SET(cp) DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, \
214 " Setting an EVAL scope, savestack=%"IVdf"\n", \
215 (IV)PL_savestack_ix)); cp = PL_savestack_ix
217 # define REGCP_UNWIND(cp) DEBUG_EXECUTE_r(cp != PL_savestack_ix ? \
218 PerlIO_printf(Perl_debug_log, \
219 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
220 (IV)(cp), (IV)PL_savestack_ix) : 0); regcpblow(cp)
229 GET_RE_DEBUG_FLAGS_DECL;
231 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
233 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
234 i = SSPOPINT; /* Parentheses elements to pop. */
235 input = (char *) SSPOPPTR;
236 *PL_reglastcloseparen = SSPOPINT;
237 *PL_reglastparen = SSPOPINT;
238 PL_regsize = SSPOPINT;
240 /* Now restore the parentheses context. */
241 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
242 i > 0; i -= REGCP_PAREN_ELEMS) {
244 paren = (U32)SSPOPINT;
245 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
246 PL_regstartp[paren] = SSPOPINT;
248 if (paren <= *PL_reglastparen)
249 PL_regendp[paren] = tmps;
251 PerlIO_printf(Perl_debug_log,
252 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
253 (UV)paren, (IV)PL_regstartp[paren],
254 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
255 (IV)PL_regendp[paren],
256 (paren > *PL_reglastparen ? "(no)" : ""));
260 if ((I32)(*PL_reglastparen + 1) <= PL_regnpar) {
261 PerlIO_printf(Perl_debug_log,
262 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
263 (IV)(*PL_reglastparen + 1), (IV)PL_regnpar);
267 /* It would seem that the similar code in regtry()
268 * already takes care of this, and in fact it is in
269 * a better location to since this code can #if 0-ed out
270 * but the code in regtry() is needed or otherwise tests
271 * requiring null fields (pat.t#187 and split.t#{13,14}
272 * (as of patchlevel 7877) will fail. Then again,
273 * this code seems to be necessary or otherwise
274 * building DynaLoader will fail:
275 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
277 for (paren = *PL_reglastparen + 1; (I32)paren <= PL_regnpar; paren++) {
278 if ((I32)paren > PL_regsize)
279 PL_regstartp[paren] = -1;
280 PL_regendp[paren] = -1;
287 S_regcp_set_to(pTHX_ I32 ss)
289 const I32 tmp = PL_savestack_ix;
291 PL_savestack_ix = ss;
293 PL_savestack_ix = tmp;
297 typedef struct re_cc_state
301 struct re_cc_state *prev;
306 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
308 #define TRYPAREN(paren, n, input) { \
311 PL_regstartp[paren] = HOPc(input, -1) - PL_bostr; \
312 PL_regendp[paren] = input - PL_bostr; \
315 PL_regendp[paren] = -1; \
317 if (regmatch(next)) \
320 PL_regendp[paren] = -1; \
325 * pregexec and friends
329 - pregexec - match a regexp against a string
332 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
333 char *strbeg, I32 minend, SV *screamer, U32 nosave)
334 /* strend: pointer to null at end of string */
335 /* strbeg: real beginning of string */
336 /* minend: end of match must be >=minend after stringarg. */
337 /* nosave: For optimizations. */
340 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
341 nosave ? 0 : REXEC_COPY_STR);
345 S_cache_re(pTHX_ regexp *prog)
347 PL_regprecomp = prog->precomp; /* Needed for FAIL. */
349 PL_regprogram = prog->program;
351 PL_regnpar = prog->nparens;
352 PL_regdata = prog->data;
357 * Need to implement the following flags for reg_anch:
359 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
361 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
362 * INTUIT_AUTORITATIVE_ML
363 * INTUIT_ONCE_NOML - Intuit can match in one location only.
366 * Another flag for this function: SECOND_TIME (so that float substrs
367 * with giant delta may be not rechecked).
370 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
372 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
373 Otherwise, only SvCUR(sv) is used to get strbeg. */
375 /* XXXX We assume that strpos is strbeg unless sv. */
377 /* XXXX Some places assume that there is a fixed substring.
378 An update may be needed if optimizer marks as "INTUITable"
379 RExen without fixed substrings. Similarly, it is assumed that
380 lengths of all the strings are no more than minlen, thus they
381 cannot come from lookahead.
382 (Or minlen should take into account lookahead.) */
384 /* A failure to find a constant substring means that there is no need to make
385 an expensive call to REx engine, thus we celebrate a failure. Similarly,
386 finding a substring too deep into the string means that less calls to
387 regtry() should be needed.
389 REx compiler's optimizer found 4 possible hints:
390 a) Anchored substring;
392 c) Whether we are anchored (beginning-of-line or \G);
393 d) First node (of those at offset 0) which may distingush positions;
394 We use a)b)d) and multiline-part of c), and try to find a position in the
395 string which does not contradict any of them.
398 /* Most of decisions we do here should have been done at compile time.
399 The nodes of the REx which we used for the search should have been
400 deleted from the finite automaton. */
403 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
404 char *strend, U32 flags, re_scream_pos_data *data)
406 register I32 start_shift = 0;
407 /* Should be nonnegative! */
408 register I32 end_shift = 0;
413 const int do_utf8 = sv ? SvUTF8(sv) : 0; /* if no sv we have to assume bytes */
415 register char *other_last = Nullch; /* other substr checked before this */
416 char *check_at = Nullch; /* check substr found at this pos */
417 const I32 multiline = prog->reganch & PMf_MULTILINE;
419 const char * const i_strpos = strpos;
420 SV * const dsv = PERL_DEBUG_PAD_ZERO(0);
423 GET_RE_DEBUG_FLAGS_DECL;
425 RX_MATCH_UTF8_set(prog,do_utf8);
427 if (prog->reganch & ROPT_UTF8) {
428 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
429 "UTF-8 regex...\n"));
430 PL_reg_flags |= RF_utf8;
434 const char *s = PL_reg_match_utf8 ?
435 sv_uni_display(dsv, sv, 60, UNI_DISPLAY_REGEX) :
437 const int len = PL_reg_match_utf8 ?
438 strlen(s) : strend - strpos;
441 if (PL_reg_match_utf8)
442 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
443 "UTF-8 target...\n"));
444 PerlIO_printf(Perl_debug_log,
445 "%sGuessing start of match, REx%s \"%s%.60s%s%s\" against \"%s%.*s%s%s\"...\n",
446 PL_colors[4], PL_colors[5], PL_colors[0],
449 (strlen(prog->precomp) > 60 ? "..." : ""),
451 (int)(len > 60 ? 60 : len),
453 (len > 60 ? "..." : "")
457 /* CHR_DIST() would be more correct here but it makes things slow. */
458 if (prog->minlen > strend - strpos) {
459 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
460 "String too short... [re_intuit_start]\n"));
463 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
466 if (!prog->check_utf8 && prog->check_substr)
467 to_utf8_substr(prog);
468 check = prog->check_utf8;
470 if (!prog->check_substr && prog->check_utf8)
471 to_byte_substr(prog);
472 check = prog->check_substr;
474 if (check == &PL_sv_undef) {
475 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
476 "Non-utf string cannot match utf check string\n"));
479 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
480 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
481 || ( (prog->reganch & ROPT_ANCH_BOL)
482 && !multiline ) ); /* Check after \n? */
485 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
486 | ROPT_IMPLICIT)) /* not a real BOL */
487 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
489 && (strpos != strbeg)) {
490 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
493 if (prog->check_offset_min == prog->check_offset_max &&
494 !(prog->reganch & ROPT_CANY_SEEN)) {
495 /* Substring at constant offset from beg-of-str... */
498 s = HOP3c(strpos, prog->check_offset_min, strend);
500 slen = SvCUR(check); /* >= 1 */
502 if ( strend - s > slen || strend - s < slen - 1
503 || (strend - s == slen && strend[-1] != '\n')) {
504 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
507 /* Now should match s[0..slen-2] */
509 if (slen && (*SvPVX_const(check) != *s
511 && memNE(SvPVX_const(check), s, slen)))) {
513 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
517 else if (*SvPVX_const(check) != *s
518 || ((slen = SvCUR(check)) > 1
519 && memNE(SvPVX_const(check), s, slen)))
521 goto success_at_start;
524 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
526 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
527 end_shift = prog->minlen - start_shift -
528 CHR_SVLEN(check) + (SvTAIL(check) != 0);
530 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
531 - (SvTAIL(check) != 0);
532 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
534 if (end_shift < eshift)
538 else { /* Can match at random position */
541 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
542 /* Should be nonnegative! */
543 end_shift = prog->minlen - start_shift -
544 CHR_SVLEN(check) + (SvTAIL(check) != 0);
547 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
549 Perl_croak(aTHX_ "panic: end_shift");
553 /* Find a possible match in the region s..strend by looking for
554 the "check" substring in the region corrected by start/end_shift. */
555 if (flags & REXEC_SCREAM) {
556 I32 p = -1; /* Internal iterator of scream. */
557 I32 * const pp = data ? data->scream_pos : &p;
559 if (PL_screamfirst[BmRARE(check)] >= 0
560 || ( BmRARE(check) == '\n'
561 && (BmPREVIOUS(check) == SvCUR(check) - 1)
563 s = screaminstr(sv, check,
564 start_shift + (s - strbeg), end_shift, pp, 0);
567 /* we may be pointing at the wrong string */
568 if (s && RX_MATCH_COPIED(prog))
569 s = strbeg + (s - SvPVX_const(sv));
571 *data->scream_olds = s;
573 else if (prog->reganch & ROPT_CANY_SEEN)
574 s = fbm_instr((U8*)(s + start_shift),
575 (U8*)(strend - end_shift),
576 check, multiline ? FBMrf_MULTILINE : 0);
578 s = fbm_instr(HOP3(s, start_shift, strend),
579 HOP3(strend, -end_shift, strbeg),
580 check, multiline ? FBMrf_MULTILINE : 0);
582 /* Update the count-of-usability, remove useless subpatterns,
585 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s %s substr \"%s%.*s%s\"%s%s",
586 (s ? "Found" : "Did not find"),
587 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) ? "anchored" : "floating"),
589 (int)(SvCUR(check) - (SvTAIL(check)!=0)),
591 PL_colors[1], (SvTAIL(check) ? "$" : ""),
592 (s ? " at offset " : "...\n") ) );
599 /* Finish the diagnostic message */
600 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
602 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
603 Start with the other substr.
604 XXXX no SCREAM optimization yet - and a very coarse implementation
605 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
606 *always* match. Probably should be marked during compile...
607 Probably it is right to do no SCREAM here...
610 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) : (prog->float_substr && prog->anchored_substr)) {
611 /* Take into account the "other" substring. */
612 /* XXXX May be hopelessly wrong for UTF... */
615 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
618 char * const last = HOP3c(s, -start_shift, strbeg);
623 t = s - prog->check_offset_max;
624 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
626 || ((t = reghopmaybe3_c(s, -(prog->check_offset_max), strpos))
631 t = HOP3c(t, prog->anchored_offset, strend);
632 if (t < other_last) /* These positions already checked */
634 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
637 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
638 /* On end-of-str: see comment below. */
639 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
640 if (must == &PL_sv_undef) {
642 DEBUG_EXECUTE_r(must = prog->anchored_utf8); /* for debug */
647 HOP3(HOP3(last1, prog->anchored_offset, strend)
648 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
650 multiline ? FBMrf_MULTILINE : 0
652 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
653 "%s anchored substr \"%s%.*s%s\"%s",
654 (s ? "Found" : "Contradicts"),
657 - (SvTAIL(must)!=0)),
659 PL_colors[1], (SvTAIL(must) ? "$" : "")));
661 if (last1 >= last2) {
662 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
663 ", giving up...\n"));
666 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
667 ", trying floating at offset %ld...\n",
668 (long)(HOP3c(s1, 1, strend) - i_strpos)));
669 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
670 s = HOP3c(last, 1, strend);
674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
675 (long)(s - i_strpos)));
676 t = HOP3c(s, -prog->anchored_offset, strbeg);
677 other_last = HOP3c(s, 1, strend);
685 else { /* Take into account the floating substring. */
690 t = HOP3c(s, -start_shift, strbeg);
692 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
693 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
694 last = HOP3c(t, prog->float_max_offset, strend);
695 s = HOP3c(t, prog->float_min_offset, strend);
698 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
699 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
700 /* fbm_instr() takes into account exact value of end-of-str
701 if the check is SvTAIL(ed). Since false positives are OK,
702 and end-of-str is not later than strend we are OK. */
703 if (must == &PL_sv_undef) {
705 DEBUG_EXECUTE_r(must = prog->float_utf8); /* for debug message */
708 s = fbm_instr((unsigned char*)s,
709 (unsigned char*)last + SvCUR(must)
711 must, multiline ? FBMrf_MULTILINE : 0);
712 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s floating substr \"%s%.*s%s\"%s",
713 (s ? "Found" : "Contradicts"),
715 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
717 PL_colors[1], (SvTAIL(must) ? "$" : "")));
720 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
721 ", giving up...\n"));
724 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
725 ", trying anchored starting at offset %ld...\n",
726 (long)(s1 + 1 - i_strpos)));
728 s = HOP3c(t, 1, strend);
732 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
733 (long)(s - i_strpos)));
734 other_last = s; /* Fix this later. --Hugo */
743 t = s - prog->check_offset_max;
744 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
746 || ((t = reghopmaybe3_c(s, -prog->check_offset_max, strpos))
748 /* Fixed substring is found far enough so that the match
749 cannot start at strpos. */
751 if (ml_anch && t[-1] != '\n') {
752 /* Eventually fbm_*() should handle this, but often
753 anchored_offset is not 0, so this check will not be wasted. */
754 /* XXXX In the code below we prefer to look for "^" even in
755 presence of anchored substrings. And we search even
756 beyond the found float position. These pessimizations
757 are historical artefacts only. */
759 while (t < strend - prog->minlen) {
761 if (t < check_at - prog->check_offset_min) {
762 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
763 /* Since we moved from the found position,
764 we definitely contradict the found anchored
765 substr. Due to the above check we do not
766 contradict "check" substr.
767 Thus we can arrive here only if check substr
768 is float. Redo checking for "other"=="fixed".
771 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
772 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
773 goto do_other_anchored;
775 /* We don't contradict the found floating substring. */
776 /* XXXX Why not check for STCLASS? */
778 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
779 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
782 /* Position contradicts check-string */
783 /* XXXX probably better to look for check-string
784 than for "\n", so one should lower the limit for t? */
785 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
786 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
787 other_last = strpos = s = t + 1;
792 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
793 PL_colors[0], PL_colors[1]));
797 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
798 PL_colors[0], PL_colors[1]));
802 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
805 /* The found string does not prohibit matching at strpos,
806 - no optimization of calling REx engine can be performed,
807 unless it was an MBOL and we are not after MBOL,
808 or a future STCLASS check will fail this. */
810 /* Even in this situation we may use MBOL flag if strpos is offset
811 wrt the start of the string. */
812 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
813 && (strpos != strbeg) && strpos[-1] != '\n'
814 /* May be due to an implicit anchor of m{.*foo} */
815 && !(prog->reganch & ROPT_IMPLICIT))
820 DEBUG_EXECUTE_r( if (ml_anch)
821 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
822 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
825 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
827 prog->check_utf8 /* Could be deleted already */
828 && --BmUSEFUL(prog->check_utf8) < 0
829 && (prog->check_utf8 == prog->float_utf8)
831 prog->check_substr /* Could be deleted already */
832 && --BmUSEFUL(prog->check_substr) < 0
833 && (prog->check_substr == prog->float_substr)
836 /* If flags & SOMETHING - do not do it many times on the same match */
837 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
838 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
839 if (do_utf8 ? prog->check_substr : prog->check_utf8)
840 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
841 prog->check_substr = prog->check_utf8 = Nullsv; /* disable */
842 prog->float_substr = prog->float_utf8 = Nullsv; /* clear */
843 check = Nullsv; /* abort */
845 /* XXXX This is a remnant of the old implementation. It
846 looks wasteful, since now INTUIT can use many
848 prog->reganch &= ~RE_USE_INTUIT;
855 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
856 if (prog->regstclass) {
857 /* minlen == 0 is possible if regstclass is \b or \B,
858 and the fixed substr is ''$.
859 Since minlen is already taken into account, s+1 is before strend;
860 accidentally, minlen >= 1 guaranties no false positives at s + 1
861 even for \b or \B. But (minlen? 1 : 0) below assumes that
862 regstclass does not come from lookahead... */
863 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
864 This leaves EXACTF only, which is dealt with in find_byclass(). */
865 const U8* const str = (U8*)STRING(prog->regstclass);
866 const int cl_l = (PL_regkind[(U8)OP(prog->regstclass)] == EXACT
867 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
869 const char * const endpos = (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
870 ? HOP3c(s, (prog->minlen ? cl_l : 0), strend)
871 : (prog->float_substr || prog->float_utf8
872 ? HOP3c(HOP3c(check_at, -start_shift, strbeg),
878 s = find_byclass(prog, prog->regstclass, s, endpos, 1);
881 const char *what = 0;
883 if (endpos == strend) {
884 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
885 "Could not match STCLASS...\n") );
888 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
889 "This position contradicts STCLASS...\n") );
890 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
892 /* Contradict one of substrings */
893 if (prog->anchored_substr || prog->anchored_utf8) {
894 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
895 DEBUG_EXECUTE_r( what = "anchored" );
897 s = HOP3c(t, 1, strend);
898 if (s + start_shift + end_shift > strend) {
899 /* XXXX Should be taken into account earlier? */
900 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
901 "Could not match STCLASS...\n") );
906 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
907 "Looking for %s substr starting at offset %ld...\n",
908 what, (long)(s + start_shift - i_strpos)) );
911 /* Have both, check_string is floating */
912 if (t + start_shift >= check_at) /* Contradicts floating=check */
913 goto retry_floating_check;
914 /* Recheck anchored substring, but not floating... */
918 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
919 "Looking for anchored substr starting at offset %ld...\n",
920 (long)(other_last - i_strpos)) );
921 goto do_other_anchored;
923 /* Another way we could have checked stclass at the
924 current position only: */
929 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
930 "Looking for /%s^%s/m starting at offset %ld...\n",
931 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
934 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
936 /* Check is floating subtring. */
937 retry_floating_check:
938 t = check_at - start_shift;
939 DEBUG_EXECUTE_r( what = "floating" );
940 goto hop_and_restart;
943 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
944 "By STCLASS: moving %ld --> %ld\n",
945 (long)(t - i_strpos), (long)(s - i_strpos))
949 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
950 "Does not contradict STCLASS...\n");
955 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
956 PL_colors[4], (check ? "Guessed" : "Giving up"),
957 PL_colors[5], (long)(s - i_strpos)) );
960 fail_finish: /* Substring not found */
961 if (prog->check_substr || prog->check_utf8) /* could be removed already */
962 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
964 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
965 PL_colors[4], PL_colors[5]));
969 /* We know what class REx starts with. Try to find this position... */
971 S_find_byclass(pTHX_ regexp * prog, regnode *c, char *s, const char *strend, I32 norun)
974 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
978 register STRLEN uskip;
982 register I32 tmp = 1; /* Scratch variable? */
983 register const bool do_utf8 = PL_reg_match_utf8;
985 /* We know what class it must start with. */
989 while (s + (uskip = UTF8SKIP(s)) <= strend) {
990 if ((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
991 !UTF8_IS_INVARIANT((U8)s[0]) ?
992 reginclass(c, (U8*)s, 0, do_utf8) :
993 REGINCLASS(c, (U8*)s)) {
994 if (tmp && (norun || regtry(prog, s)))
1005 while (s < strend) {
1008 if (REGINCLASS(c, (U8*)s) ||
1009 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1010 /* The assignment of 2 is intentional:
1011 * for the folded sharp s, the skip is 2. */
1012 (skip = SHARP_S_SKIP))) {
1013 if (tmp && (norun || regtry(prog, s)))
1025 while (s < strend) {
1026 if (tmp && (norun || regtry(prog, s)))
1035 ln = STR_LEN(c); /* length to match in octets/bytes */
1036 lnc = (I32) ln; /* length to match in characters */
1038 STRLEN ulen1, ulen2;
1040 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1041 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1043 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1044 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1046 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1047 0, ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1048 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1049 0, ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1051 while (sm < ((U8 *) m + ln)) {
1066 c2 = PL_fold_locale[c1];
1068 e = HOP3c(strend, -((I32)lnc), s);
1071 e = s; /* Due to minlen logic of intuit() */
1073 /* The idea in the EXACTF* cases is to first find the
1074 * first character of the EXACTF* node and then, if
1075 * necessary, case-insensitively compare the full
1076 * text of the node. The c1 and c2 are the first
1077 * characters (though in Unicode it gets a bit
1078 * more complicated because there are more cases
1079 * than just upper and lower: one needs to use
1080 * the so-called folding case for case-insensitive
1081 * matching (called "loose matching" in Unicode).
1082 * ibcmp_utf8() will do just that. */
1086 U8 tmpbuf [UTF8_MAXBYTES+1];
1087 STRLEN len, foldlen;
1090 /* Upper and lower of 1st char are equal -
1091 * probably not a "letter". */
1093 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1095 0 : UTF8_ALLOW_ANY);
1098 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1099 m, (char **)0, ln, (bool)UTF))
1100 && (norun || regtry(prog, s)) )
1103 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1104 uvchr_to_utf8(tmpbuf, c);
1105 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1107 && (f == c1 || f == c2)
1108 && (ln == foldlen ||
1109 !ibcmp_utf8((char *) foldbuf,
1110 (char **)0, foldlen, do_utf8,
1112 (char **)0, ln, (bool)UTF))
1113 && (norun || regtry(prog, s)) )
1121 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1123 0 : UTF8_ALLOW_ANY);
1125 /* Handle some of the three Greek sigmas cases.
1126 * Note that not all the possible combinations
1127 * are handled here: some of them are handled
1128 * by the standard folding rules, and some of
1129 * them (the character class or ANYOF cases)
1130 * are handled during compiletime in
1131 * regexec.c:S_regclass(). */
1132 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1133 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1134 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1136 if ( (c == c1 || c == c2)
1138 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1139 m, (char **)0, ln, (bool)UTF))
1140 && (norun || regtry(prog, s)) )
1143 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1144 uvchr_to_utf8(tmpbuf, c);
1145 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1147 && (f == c1 || f == c2)
1148 && (ln == foldlen ||
1149 !ibcmp_utf8((char *) foldbuf,
1150 (char **)0, foldlen, do_utf8,
1152 (char **)0, ln, (bool)UTF))
1153 && (norun || regtry(prog, s)) )
1164 && (ln == 1 || !(OP(c) == EXACTF
1166 : ibcmp_locale(s, m, ln)))
1167 && (norun || regtry(prog, s)) )
1173 if ( (*(U8*)s == c1 || *(U8*)s == c2)
1174 && (ln == 1 || !(OP(c) == EXACTF
1176 : ibcmp_locale(s, m, ln)))
1177 && (norun || regtry(prog, s)) )
1184 PL_reg_flags |= RF_tainted;
1191 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1193 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1195 tmp = ((OP(c) == BOUND ?
1196 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1197 LOAD_UTF8_CHARCLASS_ALNUM();
1198 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1199 if (tmp == !(OP(c) == BOUND ?
1200 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1201 isALNUM_LC_utf8((U8*)s)))
1204 if ((norun || regtry(prog, s)))
1211 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1212 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1213 while (s < strend) {
1215 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1217 if ((norun || regtry(prog, s)))
1223 if ((!prog->minlen && tmp) && (norun || regtry(prog, s)))
1227 PL_reg_flags |= RF_tainted;
1234 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1236 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1238 tmp = ((OP(c) == NBOUND ?
1239 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1240 LOAD_UTF8_CHARCLASS_ALNUM();
1241 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1242 if (tmp == !(OP(c) == NBOUND ?
1243 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1244 isALNUM_LC_utf8((U8*)s)))
1246 else if ((norun || regtry(prog, s)))
1252 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1253 tmp = ((OP(c) == NBOUND ?
1254 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1255 while (s < strend) {
1257 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1259 else if ((norun || regtry(prog, s)))
1264 if ((!prog->minlen && !tmp) && (norun || regtry(prog, s)))
1269 LOAD_UTF8_CHARCLASS_ALNUM();
1270 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1271 if (swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1272 if (tmp && (norun || regtry(prog, s)))
1283 while (s < strend) {
1285 if (tmp && (norun || regtry(prog, s)))
1297 PL_reg_flags |= RF_tainted;
1299 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1300 if (isALNUM_LC_utf8((U8*)s)) {
1301 if (tmp && (norun || regtry(prog, s)))
1312 while (s < strend) {
1313 if (isALNUM_LC(*s)) {
1314 if (tmp && (norun || regtry(prog, s)))
1327 LOAD_UTF8_CHARCLASS_ALNUM();
1328 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1329 if (!swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1330 if (tmp && (norun || regtry(prog, s)))
1341 while (s < strend) {
1343 if (tmp && (norun || regtry(prog, s)))
1355 PL_reg_flags |= RF_tainted;
1357 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1358 if (!isALNUM_LC_utf8((U8*)s)) {
1359 if (tmp && (norun || regtry(prog, s)))
1370 while (s < strend) {
1371 if (!isALNUM_LC(*s)) {
1372 if (tmp && (norun || regtry(prog, s)))
1385 LOAD_UTF8_CHARCLASS_SPACE();
1386 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1387 if (*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)) {
1388 if (tmp && (norun || regtry(prog, s)))
1399 while (s < strend) {
1401 if (tmp && (norun || regtry(prog, s)))
1413 PL_reg_flags |= RF_tainted;
1415 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1416 if (*s == ' ' || isSPACE_LC_utf8((U8*)s)) {
1417 if (tmp && (norun || regtry(prog, s)))
1428 while (s < strend) {
1429 if (isSPACE_LC(*s)) {
1430 if (tmp && (norun || regtry(prog, s)))
1443 LOAD_UTF8_CHARCLASS_SPACE();
1444 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1445 if (!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8))) {
1446 if (tmp && (norun || regtry(prog, s)))
1457 while (s < strend) {
1459 if (tmp && (norun || regtry(prog, s)))
1471 PL_reg_flags |= RF_tainted;
1473 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1474 if (!(*s == ' ' || isSPACE_LC_utf8((U8*)s))) {
1475 if (tmp && (norun || regtry(prog, s)))
1486 while (s < strend) {
1487 if (!isSPACE_LC(*s)) {
1488 if (tmp && (norun || regtry(prog, s)))
1501 LOAD_UTF8_CHARCLASS_DIGIT();
1502 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1503 if (swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1504 if (tmp && (norun || regtry(prog, s)))
1515 while (s < strend) {
1517 if (tmp && (norun || regtry(prog, s)))
1529 PL_reg_flags |= RF_tainted;
1531 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1532 if (isDIGIT_LC_utf8((U8*)s)) {
1533 if (tmp && (norun || regtry(prog, s)))
1544 while (s < strend) {
1545 if (isDIGIT_LC(*s)) {
1546 if (tmp && (norun || regtry(prog, s)))
1559 LOAD_UTF8_CHARCLASS_DIGIT();
1560 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1561 if (!swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1562 if (tmp && (norun || regtry(prog, s)))
1573 while (s < strend) {
1575 if (tmp && (norun || regtry(prog, s)))
1587 PL_reg_flags |= RF_tainted;
1589 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1590 if (!isDIGIT_LC_utf8((U8*)s)) {
1591 if (tmp && (norun || regtry(prog, s)))
1602 while (s < strend) {
1603 if (!isDIGIT_LC(*s)) {
1604 if (tmp && (norun || regtry(prog, s)))
1616 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1625 - regexec_flags - match a regexp against a string
1628 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1629 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1630 /* strend: pointer to null at end of string */
1631 /* strbeg: real beginning of string */
1632 /* minend: end of match must be >=minend after stringarg. */
1633 /* data: May be used for some additional optimizations. */
1634 /* nosave: For optimizations. */
1637 register regnode *c;
1638 register char *startpos = stringarg;
1639 I32 minlen; /* must match at least this many chars */
1640 I32 dontbother = 0; /* how many characters not to try at end */
1641 I32 end_shift = 0; /* Same for the end. */ /* CC */
1642 I32 scream_pos = -1; /* Internal iterator of scream. */
1644 SV* oreplsv = GvSV(PL_replgv);
1645 const bool do_utf8 = DO_UTF8(sv);
1646 const I32 multiline = prog->reganch & PMf_MULTILINE;
1648 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
1649 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
1652 GET_RE_DEBUG_FLAGS_DECL;
1654 PERL_UNUSED_ARG(data);
1655 RX_MATCH_UTF8_set(prog,do_utf8);
1661 PL_regnarrate = DEBUG_r_TEST;
1664 /* Be paranoid... */
1665 if (prog == NULL || startpos == NULL) {
1666 Perl_croak(aTHX_ "NULL regexp parameter");
1670 minlen = prog->minlen;
1671 if (strend - startpos < minlen) {
1672 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1673 "String too short [regexec_flags]...\n"));
1677 /* Check validity of program. */
1678 if (UCHARAT(prog->program) != REG_MAGIC) {
1679 Perl_croak(aTHX_ "corrupted regexp program");
1683 PL_reg_eval_set = 0;
1686 if (prog->reganch & ROPT_UTF8)
1687 PL_reg_flags |= RF_utf8;
1689 /* Mark beginning of line for ^ and lookbehind. */
1690 PL_regbol = startpos;
1694 /* Mark end of line for $ (and such) */
1697 /* see how far we have to get to not match where we matched before */
1698 PL_regtill = startpos+minend;
1700 /* We start without call_cc context. */
1703 /* If there is a "must appear" string, look for it. */
1706 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to have PL_reg_ganch */
1709 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1710 PL_reg_ganch = startpos;
1711 else if (sv && SvTYPE(sv) >= SVt_PVMG
1713 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1714 && mg->mg_len >= 0) {
1715 PL_reg_ganch = strbeg + mg->mg_len; /* Defined pos() */
1716 if (prog->reganch & ROPT_ANCH_GPOS) {
1717 if (s > PL_reg_ganch)
1722 else /* pos() not defined */
1723 PL_reg_ganch = strbeg;
1726 if (!(flags & REXEC_CHECKED) && (prog->check_substr != Nullsv || prog->check_utf8 != Nullsv)) {
1727 re_scream_pos_data d;
1729 d.scream_olds = &scream_olds;
1730 d.scream_pos = &scream_pos;
1731 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1733 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1734 goto phooey; /* not present */
1739 const char * const s0 = UTF
1740 ? pv_uni_display(dsv0, (U8*)prog->precomp, prog->prelen, 60,
1743 const int len0 = UTF ? SvCUR(dsv0) : prog->prelen;
1744 const char * const s1 = do_utf8 ? sv_uni_display(dsv1, sv, 60,
1745 UNI_DISPLAY_REGEX) : startpos;
1746 const int len1 = do_utf8 ? SvCUR(dsv1) : strend - startpos;
1749 PerlIO_printf(Perl_debug_log,
1750 "%sMatching REx%s \"%s%*.*s%s%s\" against \"%s%.*s%s%s\"\n",
1751 PL_colors[4], PL_colors[5], PL_colors[0],
1754 len0 > 60 ? "..." : "",
1756 (int)(len1 > 60 ? 60 : len1),
1758 (len1 > 60 ? "..." : "")
1762 /* Simplest case: anchored match need be tried only once. */
1763 /* [unless only anchor is BOL and multiline is set] */
1764 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1765 if (s == startpos && regtry(prog, startpos))
1767 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1768 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1773 dontbother = minlen - 1;
1774 end = HOP3c(strend, -dontbother, strbeg) - 1;
1775 /* for multiline we only have to try after newlines */
1776 if (prog->check_substr || prog->check_utf8) {
1780 if (regtry(prog, s))
1785 if (prog->reganch & RE_USE_INTUIT) {
1786 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1797 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1798 if (regtry(prog, s))
1805 } else if (prog->reganch & ROPT_ANCH_GPOS) {
1806 if (regtry(prog, PL_reg_ganch))
1811 /* Messy cases: unanchored match. */
1812 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1813 /* we have /x+whatever/ */
1814 /* it must be a one character string (XXXX Except UTF?) */
1819 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1820 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1821 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1824 while (s < strend) {
1826 DEBUG_EXECUTE_r( did_match = 1 );
1827 if (regtry(prog, s)) goto got_it;
1829 while (s < strend && *s == ch)
1836 while (s < strend) {
1838 DEBUG_EXECUTE_r( did_match = 1 );
1839 if (regtry(prog, s)) goto got_it;
1841 while (s < strend && *s == ch)
1847 DEBUG_EXECUTE_r(if (!did_match)
1848 PerlIO_printf(Perl_debug_log,
1849 "Did not find anchored character...\n")
1852 else if (prog->anchored_substr != Nullsv
1853 || prog->anchored_utf8 != Nullsv
1854 || ((prog->float_substr != Nullsv || prog->float_utf8 != Nullsv)
1855 && prog->float_max_offset < strend - s)) {
1860 char *last1; /* Last position checked before */
1864 if (prog->anchored_substr || prog->anchored_utf8) {
1865 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1866 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1867 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1868 back_max = back_min = prog->anchored_offset;
1870 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1871 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1872 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1873 back_max = prog->float_max_offset;
1874 back_min = prog->float_min_offset;
1876 if (must == &PL_sv_undef)
1877 /* could not downgrade utf8 check substring, so must fail */
1880 last = HOP3c(strend, /* Cannot start after this */
1881 -(I32)(CHR_SVLEN(must)
1882 - (SvTAIL(must) != 0) + back_min), strbeg);
1885 last1 = HOPc(s, -1);
1887 last1 = s - 1; /* bogus */
1889 /* XXXX check_substr already used to find "s", can optimize if
1890 check_substr==must. */
1892 dontbother = end_shift;
1893 strend = HOPc(strend, -dontbother);
1894 while ( (s <= last) &&
1895 ((flags & REXEC_SCREAM)
1896 ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg,
1897 end_shift, &scream_pos, 0))
1898 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend),
1899 (unsigned char*)strend, must,
1900 multiline ? FBMrf_MULTILINE : 0))) ) {
1901 /* we may be pointing at the wrong string */
1902 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1903 s = strbeg + (s - SvPVX_const(sv));
1904 DEBUG_EXECUTE_r( did_match = 1 );
1905 if (HOPc(s, -back_max) > last1) {
1906 last1 = HOPc(s, -back_min);
1907 s = HOPc(s, -back_max);
1910 char *t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1912 last1 = HOPc(s, -back_min);
1916 while (s <= last1) {
1917 if (regtry(prog, s))
1923 while (s <= last1) {
1924 if (regtry(prog, s))
1930 DEBUG_EXECUTE_r(if (!did_match)
1931 PerlIO_printf(Perl_debug_log,
1932 "Did not find %s substr \"%s%.*s%s\"%s...\n",
1933 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1934 ? "anchored" : "floating"),
1936 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
1938 PL_colors[1], (SvTAIL(must) ? "$" : ""))
1942 else if ((c = prog->regstclass)) {
1944 I32 op = (U8)OP(prog->regstclass);
1945 /* don't bother with what can't match */
1946 if (PL_regkind[op] != EXACT && op != CANY)
1947 strend = HOPc(strend, -(minlen - 1));
1950 SV *prop = sv_newmortal();
1958 pv_uni_display(dsv0, (U8*)SvPVX_const(prop), SvCUR(prop), 60,
1959 UNI_DISPLAY_REGEX) :
1961 len0 = UTF ? SvCUR(dsv0) : SvCUR(prop);
1963 sv_uni_display(dsv1, sv, 60, UNI_DISPLAY_REGEX) : s;
1964 len1 = UTF ? SvCUR(dsv1) : strend - s;
1965 PerlIO_printf(Perl_debug_log,
1966 "Matching stclass \"%*.*s\" against \"%*.*s\"\n",
1970 if (find_byclass(prog, c, s, strend, 0))
1972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass...\n"));
1976 if (prog->float_substr != Nullsv || prog->float_utf8 != Nullsv) {
1981 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1982 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1983 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1985 if (flags & REXEC_SCREAM) {
1986 last = screaminstr(sv, float_real, s - strbeg,
1987 end_shift, &scream_pos, 1); /* last one */
1989 last = scream_olds; /* Only one occurrence. */
1990 /* we may be pointing at the wrong string */
1991 else if (RX_MATCH_COPIED(prog))
1992 s = strbeg + (s - SvPVX_const(sv));
1996 const char * const little = SvPV_const(float_real, len);
1998 if (SvTAIL(float_real)) {
1999 if (memEQ(strend - len + 1, little, len - 1))
2000 last = strend - len + 1;
2001 else if (!multiline)
2002 last = memEQ(strend - len, little, len)
2003 ? strend - len : Nullch;
2009 last = rninstr(s, strend, little, little + len);
2011 last = strend; /* matching "$" */
2015 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2016 "%sCan't trim the tail, match fails (should not happen)%s\n",
2017 PL_colors[4], PL_colors[5]));
2018 goto phooey; /* Should not happen! */
2020 dontbother = strend - last + prog->float_min_offset;
2022 if (minlen && (dontbother < minlen))
2023 dontbother = minlen - 1;
2024 strend -= dontbother; /* this one's always in bytes! */
2025 /* We don't know much -- general case. */
2028 if (regtry(prog, s))
2037 if (regtry(prog, s))
2039 } while (s++ < strend);
2047 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2049 if (PL_reg_eval_set) {
2050 /* Preserve the current value of $^R */
2051 if (oreplsv != GvSV(PL_replgv))
2052 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2053 restored, the value remains
2055 restore_pos(aTHX_ 0);
2058 /* make sure $`, $&, $', and $digit will work later */
2059 if ( !(flags & REXEC_NOT_FIRST) ) {
2060 RX_MATCH_COPY_FREE(prog);
2061 if (flags & REXEC_COPY_STR) {
2062 I32 i = PL_regeol - startpos + (stringarg - strbeg);
2063 #ifdef PERL_OLD_COPY_ON_WRITE
2065 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2067 PerlIO_printf(Perl_debug_log,
2068 "Copy on write: regexp capture, type %d\n",
2071 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2072 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2073 assert (SvPOKp(prog->saved_copy));
2077 RX_MATCH_COPIED_on(prog);
2078 s = savepvn(strbeg, i);
2084 prog->subbeg = strbeg;
2085 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2092 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2093 PL_colors[4], PL_colors[5]));
2094 if (PL_reg_eval_set)
2095 restore_pos(aTHX_ 0);
2100 - regtry - try match at specific point
2102 STATIC I32 /* 0 failure, 1 success */
2103 S_regtry(pTHX_ regexp *prog, char *startpos)
2109 GET_RE_DEBUG_FLAGS_DECL;
2112 PL_regindent = 0; /* XXXX Not good when matches are reenterable... */
2114 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2117 PL_reg_eval_set = RS_init;
2118 DEBUG_EXECUTE_r(DEBUG_s(
2119 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2120 (IV)(PL_stack_sp - PL_stack_base));
2122 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2123 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2124 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2126 /* Apparently this is not needed, judging by wantarray. */
2127 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2128 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2131 /* Make $_ available to executed code. */
2132 if (PL_reg_sv != DEFSV) {
2137 if (!(SvTYPE(PL_reg_sv) >= SVt_PVMG && SvMAGIC(PL_reg_sv)
2138 && (mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global)))) {
2139 /* prepare for quick setting of pos */
2140 sv_magic(PL_reg_sv, (SV*)0,
2141 PERL_MAGIC_regex_global, Nullch, 0);
2142 mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global);
2146 PL_reg_oldpos = mg->mg_len;
2147 SAVEDESTRUCTOR_X(restore_pos, 0);
2149 if (!PL_reg_curpm) {
2150 Newxz(PL_reg_curpm, 1, PMOP);
2153 SV* repointer = newSViv(0);
2154 /* so we know which PL_regex_padav element is PL_reg_curpm */
2155 SvFLAGS(repointer) |= SVf_BREAK;
2156 av_push(PL_regex_padav,repointer);
2157 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2158 PL_regex_pad = AvARRAY(PL_regex_padav);
2162 PM_SETRE(PL_reg_curpm, prog);
2163 PL_reg_oldcurpm = PL_curpm;
2164 PL_curpm = PL_reg_curpm;
2165 if (RX_MATCH_COPIED(prog)) {
2166 /* Here is a serious problem: we cannot rewrite subbeg,
2167 since it may be needed if this match fails. Thus
2168 $` inside (?{}) could fail... */
2169 PL_reg_oldsaved = prog->subbeg;
2170 PL_reg_oldsavedlen = prog->sublen;
2171 #ifdef PERL_OLD_COPY_ON_WRITE
2172 PL_nrs = prog->saved_copy;
2174 RX_MATCH_COPIED_off(prog);
2177 PL_reg_oldsaved = Nullch;
2178 prog->subbeg = PL_bostr;
2179 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2181 prog->startp[0] = startpos - PL_bostr;
2182 PL_reginput = startpos;
2183 PL_regstartp = prog->startp;
2184 PL_regendp = prog->endp;
2185 PL_reglastparen = &prog->lastparen;
2186 PL_reglastcloseparen = &prog->lastcloseparen;
2187 prog->lastparen = 0;
2188 prog->lastcloseparen = 0;
2190 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2191 if (PL_reg_start_tmpl <= prog->nparens) {
2192 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2193 if(PL_reg_start_tmp)
2194 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2196 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2199 /* XXXX What this code is doing here?!!! There should be no need
2200 to do this again and again, PL_reglastparen should take care of
2203 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2204 * Actually, the code in regcppop() (which Ilya may be meaning by
2205 * PL_reglastparen), is not needed at all by the test suite
2206 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2207 * enough, for building DynaLoader, or otherwise this
2208 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2209 * will happen. Meanwhile, this code *is* needed for the
2210 * above-mentioned test suite tests to succeed. The common theme
2211 * on those tests seems to be returning null fields from matches.
2216 if (prog->nparens) {
2217 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2224 if (regmatch(prog->program + 1)) {
2225 prog->endp[0] = PL_reginput - PL_bostr;
2228 REGCP_UNWIND(lastcp);
2232 #define RE_UNWIND_BRANCH 1
2233 #define RE_UNWIND_BRANCHJ 2
2237 typedef struct { /* XX: makes sense to enlarge it... */
2241 } re_unwind_generic_t;
2254 } re_unwind_branch_t;
2256 typedef union re_unwind_t {
2258 re_unwind_generic_t generic;
2259 re_unwind_branch_t branch;
2262 #define sayYES goto yes
2263 #define sayNO goto no
2264 #define sayNO_ANYOF goto no_anyof
2265 #define sayYES_FINAL goto yes_final
2266 #define sayYES_LOUD goto yes_loud
2267 #define sayNO_FINAL goto no_final
2268 #define sayNO_SILENT goto do_no
2269 #define saySAME(x) if (x) goto yes; else goto no
2271 #define POSCACHE_SUCCESS 0 /* caching success rather than failure */
2272 #define POSCACHE_SEEN 1 /* we know what we're caching */
2273 #define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */
2274 #define CACHEsayYES STMT_START { \
2275 if (cache_offset | cache_bit) { \
2276 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2277 PL_reg_poscache[0] |= (1<<POSCACHE_SUCCESS) || (1<<POSCACHE_SEEN); \
2278 else if (!(PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2279 /* cache records failure, but this is success */ \
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[cache_offset] &= ~(1<<cache_bit); \
2290 #define CACHEsayNO STMT_START { \
2291 if (cache_offset | cache_bit) { \
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 */ \
2297 PerlIO_printf(Perl_debug_log, \
2298 "%*s (remove failure from success cache)\n", \
2299 REPORT_CODE_OFF+PL_regindent*2, "") \
2301 PL_reg_poscache[cache_offset] &= ~(1<<cache_bit); \
2307 /* this is used to determine how far from the left messages like
2308 'failed...' are printed. Currently 29 makes these messages line
2309 up with the opcode they refer to. Earlier perls used 25 which
2310 left these messages outdented making reviewing a debug output
2313 #define REPORT_CODE_OFF 29
2316 /* Make sure there is a test for this +1 options in re_tests */
2317 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2319 #define TRIE_CHECK_STATE_IS_ACCEPTING STMT_START { \
2320 if ( trie->states[ state ].wordnum ) { \
2321 if ( !accepted ) { \
2324 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN ; \
2325 sv_accept_buff=NEWSV( 1234, \
2326 bufflen * sizeof(reg_trie_accepted) - 1 ); \
2327 SvCUR_set( sv_accept_buff, sizeof(reg_trie_accepted) ); \
2328 SvPOK_on( sv_accept_buff ); \
2329 sv_2mortal( sv_accept_buff ); \
2330 accept_buff = (reg_trie_accepted*)SvPV_nolen( sv_accept_buff );\
2332 if ( accepted >= bufflen ) { \
2334 accept_buff =(reg_trie_accepted*)SvGROW( sv_accept_buff, \
2335 bufflen * sizeof(reg_trie_accepted) ); \
2337 SvCUR_set( sv_accept_buff,SvCUR( sv_accept_buff ) \
2338 + sizeof( reg_trie_accepted ) ); \
2340 accept_buff[ accepted ].wordnum = trie->states[ state ].wordnum; \
2341 accept_buff[ accepted ].endpos = uc; \
2345 #define TRIE_HANDLE_CHAR STMT_START { \
2346 if ( uvc < 256 ) { \
2347 charid = trie->charmap[ uvc ]; \
2350 if( trie->widecharmap ) { \
2351 SV** svpp = (SV**)NULL; \
2352 svpp = hv_fetch( trie->widecharmap, (char*)&uvc, \
2353 sizeof( UV ), 0 ); \
2355 charid = (U16)SvIV( *svpp ); \
2360 ( base + charid > trie->uniquecharcount ) && \
2361 ( base + charid - 1 - trie->uniquecharcount < trie->lasttrans) && \
2362 trie->trans[ base + charid - 1 - trie->uniquecharcount ].check == state ) \
2364 state = trie->trans[ base + charid - 1 - trie->uniquecharcount ].next; \
2372 - regmatch - main matching routine
2374 * Conceptually the strategy is simple: check to see whether the current
2375 * node matches, call self recursively to see whether the rest matches,
2376 * and then act accordingly. In practice we make some effort to avoid
2377 * recursion, in particular by going through "ordinary" nodes (that don't
2378 * need to know whether the rest of the match failed) by a loop instead of
2381 /* [lwall] I've hoisted the register declarations to the outer block in order to
2382 * maybe save a little bit of pushing and popping on the stack. It also takes
2383 * advantage of machines that use a register save mask on subroutine entry.
2385 STATIC I32 /* 0 failure, 1 success */
2386 S_regmatch(pTHX_ regnode *prog)
2389 register regnode *scan; /* Current node. */
2390 regnode *next; /* Next node. */
2391 regnode *inner; /* Next node in internal branch. */
2392 register I32 nextchr; /* renamed nextchr - nextchar colides with
2393 function of same name */
2394 register I32 n; /* no or next */
2395 register I32 ln = 0; /* len or last */
2396 register char *s = Nullch; /* operand or save */
2397 register char *locinput = PL_reginput;
2398 register I32 c1 = 0, c2 = 0, paren; /* case fold search, parenth */
2399 int minmod = 0, sw = 0, logical = 0;
2402 /* used by the trie code */
2403 SV *sv_accept_buff = 0; /* accepting states we have traversed */
2404 reg_trie_accepted *accept_buff = 0; /* "" */
2405 reg_trie_data *trie; /* what trie are we using right now */
2406 U32 accepted = 0; /* how many accepting states we have seen*/
2409 I32 firstcp = PL_savestack_ix;
2411 const register bool do_utf8 = PL_reg_match_utf8;
2413 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
2414 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
2415 SV *dsv2 = PERL_DEBUG_PAD_ZERO(2);
2417 SV *re_debug_flags = NULL;
2427 /* Note that nextchr is a byte even in UTF */
2428 nextchr = UCHARAT(locinput);
2430 while (scan != NULL) {
2433 SV *prop = sv_newmortal();
2434 const int docolor = *PL_colors[0];
2435 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2436 int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput);
2437 /* The part of the string before starttry has one color
2438 (pref0_len chars), between starttry and current
2439 position another one (pref_len - pref0_len chars),
2440 after the current position the third one.
2441 We assume that pref0_len <= pref_len, otherwise we
2442 decrease pref0_len. */
2443 int pref_len = (locinput - PL_bostr) > (5 + taill) - l
2444 ? (5 + taill) - l : locinput - PL_bostr;
2447 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2449 pref0_len = pref_len - (locinput - PL_reg_starttry);
2450 if (l + pref_len < (5 + taill) && l < PL_regeol - locinput)
2451 l = ( PL_regeol - locinput > (5 + taill) - pref_len
2452 ? (5 + taill) - pref_len : PL_regeol - locinput);
2453 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2457 if (pref0_len > pref_len)
2458 pref0_len = pref_len;
2459 regprop(prop, scan);
2461 const char * const s0 =
2462 do_utf8 && OP(scan) != CANY ?
2463 pv_uni_display(dsv0, (U8*)(locinput - pref_len),
2464 pref0_len, 60, UNI_DISPLAY_REGEX) :
2465 locinput - pref_len;
2466 const int len0 = do_utf8 ? strlen(s0) : pref0_len;
2467 const char * const s1 = do_utf8 && OP(scan) != CANY ?
2468 pv_uni_display(dsv1, (U8*)(locinput - pref_len + pref0_len),
2469 pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) :
2470 locinput - pref_len + pref0_len;
2471 const int len1 = do_utf8 ? strlen(s1) : pref_len - pref0_len;
2472 const char * const s2 = do_utf8 && OP(scan) != CANY ?
2473 pv_uni_display(dsv2, (U8*)locinput,
2474 PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) :
2476 const int len2 = do_utf8 ? strlen(s2) : l;
2477 PerlIO_printf(Perl_debug_log,
2478 "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|%3"IVdf":%*s%s\n",
2479 (IV)(locinput - PL_bostr),
2486 (docolor ? "" : "> <"),
2490 15 - l - pref_len + 1,
2492 (IV)(scan - PL_regprogram), PL_regindent*2, "",
2497 next = scan + NEXT_OFF(scan);
2503 if (locinput == PL_bostr)
2505 /* regtill = regbol; */
2510 if (locinput == PL_bostr ||
2511 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2517 if (locinput == PL_bostr)
2521 if (locinput == PL_reg_ganch)
2527 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2532 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2534 if (PL_regeol - locinput > 1)
2538 if (PL_regeol != locinput)
2542 if (!nextchr && locinput >= PL_regeol)
2545 locinput += PL_utf8skip[nextchr];
2546 if (locinput > PL_regeol)
2548 nextchr = UCHARAT(locinput);
2551 nextchr = UCHARAT(++locinput);
2554 if (!nextchr && locinput >= PL_regeol)
2556 nextchr = UCHARAT(++locinput);
2559 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2562 locinput += PL_utf8skip[nextchr];
2563 if (locinput > PL_regeol)
2565 nextchr = UCHARAT(locinput);
2568 nextchr = UCHARAT(++locinput);
2574 traverse the TRIE keeping track of all accepting states
2575 we transition through until we get to a failing node.
2577 we use two slightly different pieces of code to handle
2578 the traversal depending on whether its case sensitive or
2579 not. we reuse the accept code however. (this should probably
2580 be turned into a macro.)
2587 const U32 uniflags = ckWARN( WARN_UTF8 ) ? 0 : UTF8_ALLOW_ANY;
2588 U8 *uc = ( U8* )locinput;
2595 U8 *uscan = (U8*)NULL;
2599 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2601 while ( state && uc <= (U8*)PL_regeol ) {
2603 TRIE_CHECK_STATE_IS_ACCEPTING;
2605 base = trie->states[ state ].trans.base;
2607 DEBUG_TRIE_EXECUTE_r(
2608 PerlIO_printf( Perl_debug_log,
2609 "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ",
2610 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2611 (UV)state, (UV)base, (UV)accepted );
2618 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags );
2623 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2624 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2625 uvc = to_uni_fold( uvc, foldbuf, &foldlen );
2626 foldlen -= UNISKIP( uvc );
2627 uscan = foldbuf + UNISKIP( uvc );
2639 DEBUG_TRIE_EXECUTE_r(
2640 PerlIO_printf( Perl_debug_log,
2641 "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n",
2642 charid, uvc, (UV)state, PL_colors[5] );
2651 /* unreached codepoint: we jump into the middle of the next case
2652 from previous if blocks */
2655 const U32 uniflags = ckWARN( WARN_UTF8 ) ? 0 : UTF8_ALLOW_ANY;
2656 U8 *uc = (U8*)locinput;
2665 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2667 while ( state && uc <= (U8*)PL_regeol ) {
2669 TRIE_CHECK_STATE_IS_ACCEPTING;
2671 base = trie->states[ state ].trans.base;
2673 DEBUG_TRIE_EXECUTE_r(
2674 PerlIO_printf( Perl_debug_log,
2675 "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ",
2676 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2677 (UV)state, (UV)base, (UV)accepted );
2683 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2694 DEBUG_TRIE_EXECUTE_r(
2695 PerlIO_printf( Perl_debug_log,
2696 "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n",
2697 charid, uvc, (UV)state, PL_colors[5] );
2707 There was at least one accepting state that we
2708 transitioned through. Presumably the number of accepting
2709 states is going to be low, typically one or two. So we
2710 simply scan through to find the one with lowest wordnum.
2711 Once we find it, we swap the last state into its place
2712 and decrement the size. We then try to match the rest of
2713 the pattern at the point where the word ends, if we
2714 succeed then we end the loop, otherwise the loop
2715 eventually terminates once all of the accepting states
2722 if ( accepted == 1 ) {
2724 SV **tmp = av_fetch( trie->words, accept_buff[ 0 ].wordnum-1, 0 );
2725 PerlIO_printf( Perl_debug_log,
2726 "%*s %sonly one match : #%d <%s>%s\n",
2727 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2728 accept_buff[ 0 ].wordnum,
2729 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2732 PL_reginput = (char *)accept_buff[ 0 ].endpos;
2733 /* in this case we free tmps/leave before we call regmatch
2734 as we wont be using accept_buff again. */
2737 gotit = regmatch( scan + NEXT_OFF( scan ) );
2740 PerlIO_printf( Perl_debug_log,"%*s %sgot %"IVdf" possible matches%s\n",
2741 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], (IV)accepted,
2744 while ( !gotit && accepted-- ) {
2747 for( cur = 1 ; cur <= accepted ; cur++ ) {
2748 DEBUG_TRIE_EXECUTE_r(
2749 PerlIO_printf( Perl_debug_log,
2750 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2751 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2752 (IV)best, accept_buff[ best ].wordnum, (IV)cur,
2753 accept_buff[ cur ].wordnum, PL_colors[5] );
2756 if ( accept_buff[ cur ].wordnum < accept_buff[ best ].wordnum )
2760 SV **tmp = av_fetch( trie->words, accept_buff[ best ].wordnum - 1, 0 );
2761 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at 0x%p%s\n",
2762 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2763 accept_buff[best].wordnum,
2764 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",scan,
2767 if ( best<accepted ) {
2768 reg_trie_accepted tmp = accept_buff[ best ];
2769 accept_buff[ best ] = accept_buff[ accepted ];
2770 accept_buff[ accepted ] = tmp;
2773 PL_reginput = (char *)accept_buff[ best ].endpos;
2776 as far as I can tell we only need the SAVETMPS/FREETMPS
2777 for re's with EVAL in them but I'm leaving them in for
2778 all until I can be sure.
2781 gotit = regmatch( scan + NEXT_OFF( scan ) ) ;
2794 /* unreached codepoint */
2798 if (do_utf8 != UTF) {
2799 /* The target and the pattern have differing utf8ness. */
2801 const char *e = s + ln;
2804 /* The target is utf8, the pattern is not utf8. */
2809 if (NATIVE_TO_UNI(*(U8*)s) !=
2810 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2812 0 : UTF8_ALLOW_ANY))
2819 /* The target is not utf8, the pattern is utf8. */
2824 if (NATIVE_TO_UNI(*((U8*)l)) !=
2825 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2827 0 : UTF8_ALLOW_ANY))
2834 nextchr = UCHARAT(locinput);
2837 /* The target and the pattern have the same utf8ness. */
2838 /* Inline the first character, for speed. */
2839 if (UCHARAT(s) != nextchr)
2841 if (PL_regeol - locinput < ln)
2843 if (ln > 1 && memNE(s, locinput, ln))
2846 nextchr = UCHARAT(locinput);
2849 PL_reg_flags |= RF_tainted;
2855 if (do_utf8 || UTF) {
2856 /* Either target or the pattern are utf8. */
2858 char *e = PL_regeol;
2860 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
2861 l, &e, 0, do_utf8)) {
2862 /* One more case for the sharp s:
2863 * pack("U0U*", 0xDF) =~ /ss/i,
2864 * the 0xC3 0x9F are the UTF-8
2865 * byte sequence for the U+00DF. */
2867 toLOWER(s[0]) == 's' &&
2869 toLOWER(s[1]) == 's' &&
2876 nextchr = UCHARAT(locinput);
2880 /* Neither the target and the pattern are utf8. */
2882 /* Inline the first character, for speed. */
2883 if (UCHARAT(s) != nextchr &&
2884 UCHARAT(s) != ((OP(scan) == EXACTF)
2885 ? PL_fold : PL_fold_locale)[nextchr])
2887 if (PL_regeol - locinput < ln)
2889 if (ln > 1 && (OP(scan) == EXACTF
2890 ? ibcmp(s, locinput, ln)
2891 : ibcmp_locale(s, locinput, ln)))
2894 nextchr = UCHARAT(locinput);
2898 STRLEN inclasslen = PL_regeol - locinput;
2900 if (!reginclass(scan, (U8*)locinput, &inclasslen, do_utf8))
2902 if (locinput >= PL_regeol)
2904 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
2905 nextchr = UCHARAT(locinput);
2910 nextchr = UCHARAT(locinput);
2911 if (!REGINCLASS(scan, (U8*)locinput))
2913 if (!nextchr && locinput >= PL_regeol)
2915 nextchr = UCHARAT(++locinput);
2919 /* If we might have the case of the German sharp s
2920 * in a casefolding Unicode character class. */
2922 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
2923 locinput += SHARP_S_SKIP;
2924 nextchr = UCHARAT(locinput);
2930 PL_reg_flags |= RF_tainted;
2936 LOAD_UTF8_CHARCLASS_ALNUM();
2937 if (!(OP(scan) == ALNUM
2938 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2939 : isALNUM_LC_utf8((U8*)locinput)))
2943 locinput += PL_utf8skip[nextchr];
2944 nextchr = UCHARAT(locinput);
2947 if (!(OP(scan) == ALNUM
2948 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
2950 nextchr = UCHARAT(++locinput);
2953 PL_reg_flags |= RF_tainted;
2956 if (!nextchr && locinput >= PL_regeol)
2959 LOAD_UTF8_CHARCLASS_ALNUM();
2960 if (OP(scan) == NALNUM
2961 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2962 : isALNUM_LC_utf8((U8*)locinput))
2966 locinput += PL_utf8skip[nextchr];
2967 nextchr = UCHARAT(locinput);
2970 if (OP(scan) == NALNUM
2971 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
2973 nextchr = UCHARAT(++locinput);
2977 PL_reg_flags |= RF_tainted;
2981 /* was last char in word? */
2983 if (locinput == PL_bostr)
2986 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
2988 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
2990 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
2991 ln = isALNUM_uni(ln);
2992 LOAD_UTF8_CHARCLASS_ALNUM();
2993 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
2996 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
2997 n = isALNUM_LC_utf8((U8*)locinput);
3001 ln = (locinput != PL_bostr) ?
3002 UCHARAT(locinput - 1) : '\n';
3003 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3005 n = isALNUM(nextchr);
3008 ln = isALNUM_LC(ln);
3009 n = isALNUM_LC(nextchr);
3012 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3013 OP(scan) == BOUNDL))
3017 PL_reg_flags |= RF_tainted;
3023 if (UTF8_IS_CONTINUED(nextchr)) {
3024 LOAD_UTF8_CHARCLASS_SPACE();
3025 if (!(OP(scan) == SPACE
3026 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3027 : isSPACE_LC_utf8((U8*)locinput)))
3031 locinput += PL_utf8skip[nextchr];
3032 nextchr = UCHARAT(locinput);
3035 if (!(OP(scan) == SPACE
3036 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3038 nextchr = UCHARAT(++locinput);
3041 if (!(OP(scan) == SPACE
3042 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3044 nextchr = UCHARAT(++locinput);
3048 PL_reg_flags |= RF_tainted;
3051 if (!nextchr && locinput >= PL_regeol)
3054 LOAD_UTF8_CHARCLASS_SPACE();
3055 if (OP(scan) == NSPACE
3056 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3057 : isSPACE_LC_utf8((U8*)locinput))
3061 locinput += PL_utf8skip[nextchr];
3062 nextchr = UCHARAT(locinput);
3065 if (OP(scan) == NSPACE
3066 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3068 nextchr = UCHARAT(++locinput);
3071 PL_reg_flags |= RF_tainted;
3077 LOAD_UTF8_CHARCLASS_DIGIT();
3078 if (!(OP(scan) == DIGIT
3079 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3080 : isDIGIT_LC_utf8((U8*)locinput)))
3084 locinput += PL_utf8skip[nextchr];
3085 nextchr = UCHARAT(locinput);
3088 if (!(OP(scan) == DIGIT
3089 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3091 nextchr = UCHARAT(++locinput);
3094 PL_reg_flags |= RF_tainted;
3097 if (!nextchr && locinput >= PL_regeol)
3100 LOAD_UTF8_CHARCLASS_DIGIT();
3101 if (OP(scan) == NDIGIT
3102 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3103 : isDIGIT_LC_utf8((U8*)locinput))
3107 locinput += PL_utf8skip[nextchr];
3108 nextchr = UCHARAT(locinput);
3111 if (OP(scan) == NDIGIT
3112 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3114 nextchr = UCHARAT(++locinput);
3117 if (locinput >= PL_regeol)
3120 LOAD_UTF8_CHARCLASS_MARK();
3121 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3123 locinput += PL_utf8skip[nextchr];
3124 while (locinput < PL_regeol &&
3125 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3126 locinput += UTF8SKIP(locinput);
3127 if (locinput > PL_regeol)
3132 nextchr = UCHARAT(locinput);
3135 PL_reg_flags |= RF_tainted;
3139 n = ARG(scan); /* which paren pair */
3140 ln = PL_regstartp[n];
3141 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3142 if ((I32)*PL_reglastparen < n || ln == -1)
3143 sayNO; /* Do not match unless seen CLOSEn. */
3144 if (ln == PL_regendp[n])
3148 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3150 const char *e = PL_bostr + PL_regendp[n];
3152 * Note that we can't do the "other character" lookup trick as
3153 * in the 8-bit case (no pun intended) because in Unicode we
3154 * have to map both upper and title case to lower case.
3156 if (OP(scan) == REFF) {
3158 STRLEN ulen1, ulen2;
3159 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3160 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3164 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3165 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3166 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3173 nextchr = UCHARAT(locinput);
3177 /* Inline the first character, for speed. */
3178 if (UCHARAT(s) != nextchr &&
3180 (UCHARAT(s) != ((OP(scan) == REFF
3181 ? PL_fold : PL_fold_locale)[nextchr]))))
3183 ln = PL_regendp[n] - ln;
3184 if (locinput + ln > PL_regeol)
3186 if (ln > 1 && (OP(scan) == REF
3187 ? memNE(s, locinput, ln)
3189 ? ibcmp(s, locinput, ln)
3190 : ibcmp_locale(s, locinput, ln))))
3193 nextchr = UCHARAT(locinput);
3204 OP_4tree *oop = PL_op;
3205 COP *ocurcop = PL_curcop;
3208 struct regexp *oreg = PL_reg_re;
3211 PL_op = (OP_4tree*)PL_regdata->data[n];
3212 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3213 PAD_SAVE_LOCAL(old_comppad, (PAD*)PL_regdata->data[n + 2]);
3214 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3218 CALLRUNOPS(aTHX); /* Scalar context. */
3221 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3229 PAD_RESTORE_LOCAL(old_comppad);
3230 PL_curcop = ocurcop;
3232 if (logical == 2) { /* Postponed subexpression. */
3234 MAGIC *mg = Null(MAGIC*);
3236 CHECKPOINT cp, lastcp;
3240 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3241 mg = mg_find(sv, PERL_MAGIC_qr);
3242 else if (SvSMAGICAL(ret)) {
3243 if (SvGMAGICAL(ret))
3244 sv_unmagic(ret, PERL_MAGIC_qr);
3246 mg = mg_find(ret, PERL_MAGIC_qr);
3250 re = (regexp *)mg->mg_obj;
3251 (void)ReREFCNT_inc(re);
3255 const char *t = SvPV_const(ret, len);
3257 char * const oprecomp = PL_regprecomp;
3258 const I32 osize = PL_regsize;
3259 const I32 onpar = PL_regnpar;
3262 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3263 re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm);
3265 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3267 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3269 PL_regprecomp = oprecomp;
3274 PerlIO_printf(Perl_debug_log,
3275 "Entering embedded \"%s%.60s%s%s\"\n",
3279 (strlen(re->precomp) > 60 ? "..." : ""))
3282 state.prev = PL_reg_call_cc;
3283 state.cc = PL_regcc;
3284 state.re = PL_reg_re;
3288 cp = regcppush(0); /* Save *all* the positions. */
3291 state.ss = PL_savestack_ix;
3292 *PL_reglastparen = 0;
3293 *PL_reglastcloseparen = 0;
3294 PL_reg_call_cc = &state;
3295 PL_reginput = locinput;
3296 toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3297 ((re->reganch & ROPT_UTF8) != 0);
3298 if (toggleutf) PL_reg_flags ^= RF_utf8;
3300 /* XXXX This is too dramatic a measure... */
3303 if (regmatch(re->program + 1)) {
3304 /* Even though we succeeded, we need to restore
3305 global variables, since we may be wrapped inside
3306 SUSPEND, thus the match may be not finished yet. */
3308 /* XXXX Do this only if SUSPENDed? */
3309 PL_reg_call_cc = state.prev;
3310 PL_regcc = state.cc;
3311 PL_reg_re = state.re;
3312 cache_re(PL_reg_re);
3313 if (toggleutf) PL_reg_flags ^= RF_utf8;
3315 /* XXXX This is too dramatic a measure... */
3318 /* These are needed even if not SUSPEND. */
3324 REGCP_UNWIND(lastcp);
3326 PL_reg_call_cc = state.prev;
3327 PL_regcc = state.cc;
3328 PL_reg_re = state.re;
3329 cache_re(PL_reg_re);
3330 if (toggleutf) PL_reg_flags ^= RF_utf8;
3332 /* XXXX This is too dramatic a measure... */
3342 sv_setsv(save_scalar(PL_replgv), ret);
3348 n = ARG(scan); /* which paren pair */
3349 PL_reg_start_tmp[n] = locinput;
3354 n = ARG(scan); /* which paren pair */
3355 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3356 PL_regendp[n] = locinput - PL_bostr;
3357 if (n > (I32)*PL_reglastparen)
3358 *PL_reglastparen = n;
3359 *PL_reglastcloseparen = n;
3362 n = ARG(scan); /* which paren pair */
3363 sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3366 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3368 next = NEXTOPER(NEXTOPER(scan));
3370 next = scan + ARG(scan);
3371 if (OP(next) == IFTHEN) /* Fake one. */
3372 next = NEXTOPER(NEXTOPER(next));
3376 logical = scan->flags;
3378 /*******************************************************************
3379 PL_regcc contains infoblock about the innermost (...)* loop, and
3380 a pointer to the next outer infoblock.
3382 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3384 1) After matching X, regnode for CURLYX is processed;
3386 2) This regnode creates infoblock on the stack, and calls
3387 regmatch() recursively with the starting point at WHILEM node;
3389 3) Each hit of WHILEM node tries to match A and Z (in the order
3390 depending on the current iteration, min/max of {min,max} and
3391 greediness). The information about where are nodes for "A"
3392 and "Z" is read from the infoblock, as is info on how many times "A"
3393 was already matched, and greediness.
3395 4) After A matches, the same WHILEM node is hit again.
3397 5) Each time WHILEM is hit, PL_regcc is the infoblock created by CURLYX
3398 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3399 resets PL_regcc, since this Y(A)*Z can be a part of some other loop:
3400 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3401 of the external loop.
3403 Currently present infoblocks form a tree with a stem formed by PL_curcc
3404 and whatever it mentions via ->next, and additional attached trees
3405 corresponding to temporarily unset infoblocks as in "5" above.
3407 In the following picture infoblocks for outer loop of
3408 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3409 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3410 infoblocks are drawn below the "reset" infoblock.
3412 In fact in the picture below we do not show failed matches for Z and T
3413 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3414 more obvious *why* one needs to *temporary* unset infoblocks.]
3416 Matched REx position InfoBlocks Comment
3420 Y A)*?Z)*?T x <- O <- I
3421 YA )*?Z)*?T x <- O <- I
3422 YA A)*?Z)*?T x <- O <- I
3423 YAA )*?Z)*?T x <- O <- I
3424 YAA Z)*?T x <- O # Temporary unset I
3427 YAAZ Y(A)*?Z)*?T x <- O
3430 YAAZY (A)*?Z)*?T x <- O
3433 YAAZY A)*?Z)*?T x <- O <- I
3436 YAAZYA )*?Z)*?T x <- O <- I
3439 YAAZYA Z)*?T x <- O # Temporary unset I
3445 YAAZYAZ T x # Temporary unset O
3452 *******************************************************************/
3455 CHECKPOINT cp = PL_savestack_ix;
3456 /* No need to save/restore up to this paren */
3457 I32 parenfloor = scan->flags;
3459 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3461 cc.oldcc = PL_regcc;
3463 /* XXXX Probably it is better to teach regpush to support
3464 parenfloor > PL_regsize... */
3465 if (parenfloor > (I32)*PL_reglastparen)
3466 parenfloor = *PL_reglastparen; /* Pessimization... */
3467 cc.parenfloor = parenfloor;
3469 cc.min = ARG1(scan);
3470 cc.max = ARG2(scan);
3471 cc.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3475 PL_reginput = locinput;
3476 n = regmatch(PREVOPER(next)); /* start on the WHILEM */
3478 PL_regcc = cc.oldcc;
3484 * This is really hard to understand, because after we match
3485 * what we're trying to match, we must make sure the rest of
3486 * the REx is going to match for sure, and to do that we have
3487 * to go back UP the parse tree by recursing ever deeper. And
3488 * if it fails, we have to reset our parent's current state
3489 * that we can try again after backing off.
3492 CHECKPOINT cp, lastcp;
3493 CURCUR* cc = PL_regcc;
3494 char *lastloc = cc->lastloc; /* Detection of 0-len. */
3495 I32 cache_offset = 0, cache_bit = 0;
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 const 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 Newxz(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 cache_offset = locinput - PL_bostr;
3569 cache_offset = (scan->flags & 0xf) - 1 + POSCACHE_START
3570 + cache_offset * (scan->flags>>4);
3571 cache_bit = cache_offset % 8;
3573 if (PL_reg_poscache[cache_offset] & (1<<cache_bit)) {
3575 PerlIO_printf(Perl_debug_log,
3576 "%*s already tried at this position...\n",
3577 REPORT_CODE_OFF+PL_regindent*2, "")
3579 if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))
3580 /* cache records success */
3583 /* cache records failure */
3586 PL_reg_poscache[cache_offset] |= (1<<cache_bit);
3590 /* Prefer next over scan for minimal matching. */
3593 PL_regcc = cc->oldcc;
3596 cp = regcppush(cc->parenfloor);
3598 if (regmatch(cc->next)) {
3600 CACHEsayYES; /* All done. */
3602 REGCP_UNWIND(lastcp);
3608 if (n >= cc->max) { /* Maximum greed exceeded? */
3609 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3610 && !(PL_reg_flags & RF_warned)) {
3611 PL_reg_flags |= RF_warned;
3612 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3613 "Complex regular subexpression recursion",
3620 PerlIO_printf(Perl_debug_log,
3621 "%*s trying longer...\n",
3622 REPORT_CODE_OFF+PL_regindent*2, "")
3624 /* Try scanning more and see if it helps. */
3625 PL_reginput = locinput;
3627 cc->lastloc = locinput;
3628 cp = regcppush(cc->parenfloor);
3630 if (regmatch(cc->scan)) {
3634 REGCP_UNWIND(lastcp);
3637 cc->lastloc = lastloc;
3641 /* Prefer scan over next for maximal matching. */
3643 if (n < cc->max) { /* More greed allowed? */
3644 cp = regcppush(cc->parenfloor);
3646 cc->lastloc = locinput;
3648 if (regmatch(cc->scan)) {
3652 REGCP_UNWIND(lastcp);
3653 regcppop(); /* Restore some previous $<digit>s? */
3654 PL_reginput = locinput;
3656 PerlIO_printf(Perl_debug_log,
3657 "%*s failed, try continuation...\n",
3658 REPORT_CODE_OFF+PL_regindent*2, "")
3661 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3662 && !(PL_reg_flags & RF_warned)) {
3663 PL_reg_flags |= RF_warned;
3664 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3665 "Complex regular subexpression recursion",
3669 /* Failed deeper matches of scan, so see if this one works. */
3670 PL_regcc = cc->oldcc;
3673 if (regmatch(cc->next))
3679 cc->lastloc = lastloc;
3684 next = scan + ARG(scan);
3687 inner = NEXTOPER(NEXTOPER(scan));
3690 inner = NEXTOPER(scan);
3694 if (OP(next) != c1) /* No choice. */
3695 next = inner; /* Avoid recursion. */
3697 const I32 lastparen = *PL_reglastparen;
3699 re_unwind_branch_t *uw;
3701 /* Put unwinding data on stack */
3702 unwind1 = SSNEWt(1,re_unwind_branch_t);
3703 uw = SSPTRt(unwind1,re_unwind_branch_t);
3706 uw->type = ((c1 == BRANCH)
3708 : RE_UNWIND_BRANCHJ);
3709 uw->lastparen = lastparen;
3711 uw->locinput = locinput;
3712 uw->nextchr = nextchr;
3714 uw->regindent = ++PL_regindent;
3717 REGCP_SET(uw->lastcp);
3719 /* Now go into the first branch */
3732 /* We suppose that the next guy does not need
3733 backtracking: in particular, it is of constant non-zero length,
3734 and has no parenths to influence future backrefs. */
3735 ln = ARG1(scan); /* min to match */
3736 n = ARG2(scan); /* max to match */
3737 paren = scan->flags;
3739 if (paren > PL_regsize)
3741 if (paren > (I32)*PL_reglastparen)
3742 *PL_reglastparen = paren;
3744 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3746 scan += NEXT_OFF(scan); /* Skip former OPEN. */
3747 PL_reginput = locinput;
3750 if (ln && regrepeat_hard(scan, ln, &l) < ln)
3752 locinput = PL_reginput;
3753 if (HAS_TEXT(next) || JUMPABLE(next)) {
3754 regnode *text_node = next;
3756 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3758 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3760 if (PL_regkind[(U8)OP(text_node)] == REF) {
3764 else { c1 = (U8)*STRING(text_node); }
3765 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3767 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3768 c2 = PL_fold_locale[c1];
3777 while (n >= ln || (n == REG_INFTY && ln > 0)) { /* ln overflow ? */
3778 /* If it could work, try it. */
3780 UCHARAT(PL_reginput) == c1 ||
3781 UCHARAT(PL_reginput) == c2)
3785 PL_regstartp[paren] =
3786 HOPc(PL_reginput, -l) - PL_bostr;
3787 PL_regendp[paren] = PL_reginput - PL_bostr;
3790 PL_regendp[paren] = -1;
3794 REGCP_UNWIND(lastcp);
3796 /* Couldn't or didn't -- move forward. */
3797 PL_reginput = locinput;
3798 if (regrepeat_hard(scan, 1, &l)) {
3800 locinput = PL_reginput;
3807 n = regrepeat_hard(scan, n, &l);
3808 locinput = PL_reginput;
3810 PerlIO_printf(Perl_debug_log,
3811 "%*s matched %"IVdf" times, len=%"IVdf"...\n",
3812 (int)(REPORT_CODE_OFF+PL_regindent*2), "",
3816 if (HAS_TEXT(next) || JUMPABLE(next)) {
3817 regnode *text_node = next;
3819 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3821 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3823 if (PL_regkind[(U8)OP(text_node)] == REF) {
3827 else { c1 = (U8)*STRING(text_node); }
3829 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3831 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3832 c2 = PL_fold_locale[c1];
3843 /* If it could work, try it. */
3845 UCHARAT(PL_reginput) == c1 ||
3846 UCHARAT(PL_reginput) == c2)
3849 PerlIO_printf(Perl_debug_log,
3850 "%*s trying tail with n=%"IVdf"...\n",
3851 (int)(REPORT_CODE_OFF+PL_regindent*2), "", (IV)n)
3855 PL_regstartp[paren] = HOPc(PL_reginput, -l) - PL_bostr;
3856 PL_regendp[paren] = PL_reginput - PL_bostr;
3859 PL_regendp[paren] = -1;