5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "inline_invlist.c"
84 #include "unicode_constants.h"
86 #define RF_tainted 1 /* tainted information used? e.g. locale */
87 #define RF_warned 2 /* warned about big count? */
89 #define RF_utf8 8 /* Pattern contains multibyte chars? */
91 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
97 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
125 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
126 if (!CAT2(PL_utf8_,class)) { \
128 ENTER; save_re_context(); \
129 ok=CAT2(is_utf8_,class)((const U8*)str); \
130 PERL_UNUSED_VAR(ok); \
131 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
132 /* Doesn't do an assert to verify that is correct */
133 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
134 if (!CAT2(PL_utf8_,class)) { \
135 bool throw_away PERL_UNUSED_DECL; \
136 ENTER; save_re_context(); \
137 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
140 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
141 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
142 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
144 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
145 /* No asserts are done for some of these, in case called on a */ \
146 /* Unicode version in which they map to nothing */ \
147 LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
148 LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
150 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
152 /* The actual code for CCC_TRY, which uses several variables from the routine
153 * it's callable from. It is designed to be the bulk of a case statement.
154 * FUNC is the macro or function to call on non-utf8 targets that indicate if
155 * nextchr matches the class.
156 * UTF8_TEST is the whole test string to use for utf8 targets
157 * LOAD is what to use to test, and if not present to load in the swash for the
159 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
161 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
162 * utf8 and a variant, load the swash if necessary and test using the utf8
163 * test. Advance to the next character if test is ok, otherwise fail; If not
164 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
165 * fails, or advance to the next character */
167 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
168 if (locinput >= PL_regeol) { \
171 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
172 LOAD_UTF8_CHARCLASS(CLASS, STR); \
173 if (POS_OR_NEG (UTF8_TEST)) { \
176 locinput += PL_utf8skip[nextchr]; \
177 nextchr = UCHARAT(locinput); \
180 if (POS_OR_NEG (FUNC(nextchr))) { \
183 nextchr = UCHARAT(++locinput); \
186 /* Handle the non-locale cases for a character class and its complement. It
187 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
188 * This is because that code fails when the test succeeds, so we want to have
189 * the test fail so that the code succeeds. The swash is stored in a
190 * predictable PL_ place */
191 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
194 _CCC_TRY_CODE( !, FUNC, \
195 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
196 (U8*)locinput, TRUE)), \
199 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
200 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
201 (U8*)locinput, TRUE)), \
204 /* Generate the case statements for both locale and non-locale character
205 * classes in regmatch for classes that don't have special unicode semantics.
206 * Locales don't use an immediate swash, but an intermediary special locale
207 * function that is called on the pointer to the current place in the input
208 * string. That function will resolve to needing the same swash. One might
209 * think that because we don't know what the locale will match, we shouldn't
210 * check with the swash loading function that it loaded properly; ie, that we
211 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
212 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
214 #define CCC_TRY(NAME, NNAME, FUNC, \
215 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
216 NAMEA, NNAMEA, FUNCA, \
219 PL_reg_flags |= RF_tainted; \
220 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
222 PL_reg_flags |= RF_tainted; \
223 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
226 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
229 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
230 nextchr = UCHARAT(++locinput); \
233 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
237 locinput += PL_utf8skip[nextchr]; \
238 nextchr = UCHARAT(locinput); \
241 nextchr = UCHARAT(++locinput); \
244 /* Generate the non-locale cases */ \
245 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
247 /* This is like CCC_TRY, but has an extra set of parameters for generating case
248 * statements to handle separate Unicode semantics nodes */
249 #define CCC_TRY_U(NAME, NNAME, FUNC, \
250 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
251 NAMEU, NNAMEU, FUNCU, \
252 NAMEA, NNAMEA, FUNCA, \
254 CCC_TRY(NAME, NNAME, FUNC, \
255 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
256 NAMEA, NNAMEA, FUNCA, \
258 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
260 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
262 /* for use after a quantifier and before an EXACT-like node -- japhy */
263 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
265 * NOTE that *nothing* that affects backtracking should be in here, specifically
266 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
267 * node that is in between two EXACT like nodes when ascertaining what the required
268 * "follow" character is. This should probably be moved to regex compile time
269 * although it may be done at run time beause of the REF possibility - more
270 * investigation required. -- demerphq
272 #define JUMPABLE(rn) ( \
274 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
276 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
277 OP(rn) == PLUS || OP(rn) == MINMOD || \
279 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
281 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
283 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
286 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
287 we don't need this definition. */
288 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
289 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
290 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
293 /* ... so we use this as its faster. */
294 #define IS_TEXT(rn) ( OP(rn)==EXACT )
295 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
296 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
297 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
302 Search for mandatory following text node; for lookahead, the text must
303 follow but for lookbehind (rn->flags != 0) we skip to the next step.
305 #define FIND_NEXT_IMPT(rn) STMT_START { \
306 while (JUMPABLE(rn)) { \
307 const OPCODE type = OP(rn); \
308 if (type == SUSPEND || PL_regkind[type] == CURLY) \
309 rn = NEXTOPER(NEXTOPER(rn)); \
310 else if (type == PLUS) \
312 else if (type == IFMATCH) \
313 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
314 else rn += NEXT_OFF(rn); \
319 static void restore_pos(pTHX_ void *arg);
321 #define REGCP_PAREN_ELEMS 3
322 #define REGCP_OTHER_ELEMS 3
323 #define REGCP_FRAME_ELEMS 1
324 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
325 * are needed for the regexp context stack bookkeeping. */
328 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
331 const int retval = PL_savestack_ix;
332 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
333 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
334 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
336 GET_RE_DEBUG_FLAGS_DECL;
338 PERL_ARGS_ASSERT_REGCPPUSH;
340 if (paren_elems_to_push < 0)
341 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
342 paren_elems_to_push);
344 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
345 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
346 " out of range (%lu-%ld)",
347 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
349 SSGROW(total_elems + REGCP_FRAME_ELEMS);
352 if ((int)PL_regsize > (int)parenfloor)
353 PerlIO_printf(Perl_debug_log,
354 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
359 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
360 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
361 SSPUSHINT(rex->offs[p].end);
362 SSPUSHINT(rex->offs[p].start);
363 SSPUSHINT(rex->offs[p].start_tmp);
364 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
365 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
367 (IV)rex->offs[p].start,
368 (IV)rex->offs[p].start_tmp,
372 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
373 SSPUSHINT(PL_regsize);
374 SSPUSHINT(rex->lastparen);
375 SSPUSHINT(rex->lastcloseparen);
376 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
381 /* These are needed since we do not localize EVAL nodes: */
382 #define REGCP_SET(cp) \
384 PerlIO_printf(Perl_debug_log, \
385 " Setting an EVAL scope, savestack=%"IVdf"\n", \
386 (IV)PL_savestack_ix)); \
389 #define REGCP_UNWIND(cp) \
391 if (cp != PL_savestack_ix) \
392 PerlIO_printf(Perl_debug_log, \
393 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
394 (IV)(cp), (IV)PL_savestack_ix)); \
397 #define UNWIND_PAREN(lp, lcp) \
398 for (n = rex->lastparen; n > lp; n--) \
399 rex->offs[n].end = -1; \
400 rex->lastparen = n; \
401 rex->lastcloseparen = lcp;
405 S_regcppop(pTHX_ regexp *rex)
410 GET_RE_DEBUG_FLAGS_DECL;
412 PERL_ARGS_ASSERT_REGCPPOP;
414 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
416 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
417 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
418 rex->lastcloseparen = SSPOPINT;
419 rex->lastparen = SSPOPINT;
420 PL_regsize = SSPOPINT;
422 i -= REGCP_OTHER_ELEMS;
423 /* Now restore the parentheses context. */
425 if (i || rex->lastparen + 1 <= rex->nparens)
426 PerlIO_printf(Perl_debug_log,
427 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
433 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
435 rex->offs[paren].start_tmp = SSPOPINT;
436 rex->offs[paren].start = SSPOPINT;
438 if (paren <= rex->lastparen)
439 rex->offs[paren].end = tmps;
440 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
441 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
443 (IV)rex->offs[paren].start,
444 (IV)rex->offs[paren].start_tmp,
445 (IV)rex->offs[paren].end,
446 (paren > rex->lastparen ? "(skipped)" : ""));
451 /* It would seem that the similar code in regtry()
452 * already takes care of this, and in fact it is in
453 * a better location to since this code can #if 0-ed out
454 * but the code in regtry() is needed or otherwise tests
455 * requiring null fields (pat.t#187 and split.t#{13,14}
456 * (as of patchlevel 7877) will fail. Then again,
457 * this code seems to be necessary or otherwise
458 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
459 * --jhi updated by dapm */
460 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
462 rex->offs[i].start = -1;
463 rex->offs[i].end = -1;
464 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
465 " \\%"UVuf": %s ..-1 undeffing\n",
467 (i > PL_regsize) ? "-1" : " "
473 /* restore the parens and associated vars at savestack position ix,
474 * but without popping the stack */
477 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
479 I32 tmpix = PL_savestack_ix;
480 PL_savestack_ix = ix;
482 PL_savestack_ix = tmpix;
485 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
488 * pregexec and friends
491 #ifndef PERL_IN_XSUB_RE
493 - pregexec - match a regexp against a string
496 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
497 char *strbeg, I32 minend, SV *screamer, U32 nosave)
498 /* stringarg: the point in the string at which to begin matching */
499 /* strend: pointer to null at end of string */
500 /* strbeg: real beginning of string */
501 /* minend: end of match must be >= minend bytes after stringarg. */
502 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
503 * itself is accessed via the pointers above */
504 /* nosave: For optimizations. */
506 PERL_ARGS_ASSERT_PREGEXEC;
509 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
510 nosave ? 0 : REXEC_COPY_STR);
515 * Need to implement the following flags for reg_anch:
517 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
519 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
520 * INTUIT_AUTORITATIVE_ML
521 * INTUIT_ONCE_NOML - Intuit can match in one location only.
524 * Another flag for this function: SECOND_TIME (so that float substrs
525 * with giant delta may be not rechecked).
528 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
530 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
531 Otherwise, only SvCUR(sv) is used to get strbeg. */
533 /* XXXX We assume that strpos is strbeg unless sv. */
535 /* XXXX Some places assume that there is a fixed substring.
536 An update may be needed if optimizer marks as "INTUITable"
537 RExen without fixed substrings. Similarly, it is assumed that
538 lengths of all the strings are no more than minlen, thus they
539 cannot come from lookahead.
540 (Or minlen should take into account lookahead.)
541 NOTE: Some of this comment is not correct. minlen does now take account
542 of lookahead/behind. Further research is required. -- demerphq
546 /* A failure to find a constant substring means that there is no need to make
547 an expensive call to REx engine, thus we celebrate a failure. Similarly,
548 finding a substring too deep into the string means that less calls to
549 regtry() should be needed.
551 REx compiler's optimizer found 4 possible hints:
552 a) Anchored substring;
554 c) Whether we are anchored (beginning-of-line or \G);
555 d) First node (of those at offset 0) which may distinguish positions;
556 We use a)b)d) and multiline-part of c), and try to find a position in the
557 string which does not contradict any of them.
560 /* Most of decisions we do here should have been done at compile time.
561 The nodes of the REx which we used for the search should have been
562 deleted from the finite automaton. */
565 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
566 char *strend, const U32 flags, re_scream_pos_data *data)
569 struct regexp *const prog = (struct regexp *)SvANY(rx);
571 /* Should be nonnegative! */
577 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
579 char *other_last = NULL; /* other substr checked before this */
580 char *check_at = NULL; /* check substr found at this pos */
581 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
582 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
583 RXi_GET_DECL(prog,progi);
585 const char * const i_strpos = strpos;
587 GET_RE_DEBUG_FLAGS_DECL;
589 PERL_ARGS_ASSERT_RE_INTUIT_START;
590 PERL_UNUSED_ARG(flags);
591 PERL_UNUSED_ARG(data);
593 RX_MATCH_UTF8_set(rx,utf8_target);
596 PL_reg_flags |= RF_utf8;
599 debug_start_match(rx, utf8_target, strpos, strend,
600 sv ? "Guessing start of match in sv for"
601 : "Guessing start of match in string for");
604 /* CHR_DIST() would be more correct here but it makes things slow. */
605 if (prog->minlen > strend - strpos) {
606 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
607 "String too short... [re_intuit_start]\n"));
611 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
614 if (!prog->check_utf8 && prog->check_substr)
615 to_utf8_substr(prog);
616 check = prog->check_utf8;
618 if (!prog->check_substr && prog->check_utf8)
619 to_byte_substr(prog);
620 check = prog->check_substr;
622 if (check == &PL_sv_undef) {
623 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
624 "Non-utf8 string cannot match utf8 check string\n"));
627 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
628 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
629 || ( (prog->extflags & RXf_ANCH_BOL)
630 && !multiline ) ); /* Check after \n? */
633 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
634 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
635 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
637 && (strpos != strbeg)) {
638 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
641 if (prog->check_offset_min == prog->check_offset_max &&
642 !(prog->extflags & RXf_CANY_SEEN)) {
643 /* Substring at constant offset from beg-of-str... */
646 s = HOP3c(strpos, prog->check_offset_min, strend);
649 slen = SvCUR(check); /* >= 1 */
651 if ( strend - s > slen || strend - s < slen - 1
652 || (strend - s == slen && strend[-1] != '\n')) {
653 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
656 /* Now should match s[0..slen-2] */
658 if (slen && (*SvPVX_const(check) != *s
660 && memNE(SvPVX_const(check), s, slen)))) {
662 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
666 else if (*SvPVX_const(check) != *s
667 || ((slen = SvCUR(check)) > 1
668 && memNE(SvPVX_const(check), s, slen)))
671 goto success_at_start;
674 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
676 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
677 end_shift = prog->check_end_shift;
680 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
681 - (SvTAIL(check) != 0);
682 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
684 if (end_shift < eshift)
688 else { /* Can match at random position */
691 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
692 end_shift = prog->check_end_shift;
694 /* end shift should be non negative here */
697 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
699 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
700 (IV)end_shift, RX_PRECOMP(prog));
704 /* Find a possible match in the region s..strend by looking for
705 the "check" substring in the region corrected by start/end_shift. */
708 I32 srch_start_shift = start_shift;
709 I32 srch_end_shift = end_shift;
712 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
713 srch_end_shift -= ((strbeg - s) - srch_start_shift);
714 srch_start_shift = strbeg - s;
716 DEBUG_OPTIMISE_MORE_r({
717 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
718 (IV)prog->check_offset_min,
719 (IV)srch_start_shift,
721 (IV)prog->check_end_shift);
724 if (prog->extflags & RXf_CANY_SEEN) {
725 start_point= (U8*)(s + srch_start_shift);
726 end_point= (U8*)(strend - srch_end_shift);
728 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
729 end_point= HOP3(strend, -srch_end_shift, strbeg);
731 DEBUG_OPTIMISE_MORE_r({
732 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
733 (int)(end_point - start_point),
734 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
738 s = fbm_instr( start_point, end_point,
739 check, multiline ? FBMrf_MULTILINE : 0);
741 /* Update the count-of-usability, remove useless subpatterns,
745 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
746 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
747 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
748 (s ? "Found" : "Did not find"),
749 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
750 ? "anchored" : "floating"),
753 (s ? " at offset " : "...\n") );
758 /* Finish the diagnostic message */
759 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
761 /* XXX dmq: first branch is for positive lookbehind...
762 Our check string is offset from the beginning of the pattern.
763 So we need to do any stclass tests offset forward from that
772 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
773 Start with the other substr.
774 XXXX no SCREAM optimization yet - and a very coarse implementation
775 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
776 *always* match. Probably should be marked during compile...
777 Probably it is right to do no SCREAM here...
780 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
781 : (prog->float_substr && prog->anchored_substr))
783 /* Take into account the "other" substring. */
784 /* XXXX May be hopelessly wrong for UTF... */
787 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
790 char * const last = HOP3c(s, -start_shift, strbeg);
792 char * const saved_s = s;
795 t = s - prog->check_offset_max;
796 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
798 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
803 t = HOP3c(t, prog->anchored_offset, strend);
804 if (t < other_last) /* These positions already checked */
806 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
809 /* XXXX It is not documented what units *_offsets are in.
810 We assume bytes, but this is clearly wrong.
811 Meaning this code needs to be carefully reviewed for errors.
815 /* On end-of-str: see comment below. */
816 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
817 if (must == &PL_sv_undef) {
819 DEBUG_r(must = prog->anchored_utf8); /* for debug */
824 HOP3(HOP3(last1, prog->anchored_offset, strend)
825 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
827 multiline ? FBMrf_MULTILINE : 0
830 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
831 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
832 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
833 (s ? "Found" : "Contradicts"),
834 quoted, RE_SV_TAIL(must));
839 if (last1 >= last2) {
840 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
841 ", giving up...\n"));
844 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
845 ", trying floating at offset %ld...\n",
846 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
847 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
848 s = HOP3c(last, 1, strend);
852 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
853 (long)(s - i_strpos)));
854 t = HOP3c(s, -prog->anchored_offset, strbeg);
855 other_last = HOP3c(s, 1, strend);
863 else { /* Take into account the floating substring. */
865 char * const saved_s = s;
868 t = HOP3c(s, -start_shift, strbeg);
870 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
871 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
872 last = HOP3c(t, prog->float_max_offset, strend);
873 s = HOP3c(t, prog->float_min_offset, strend);
876 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
877 must = utf8_target ? prog->float_utf8 : prog->float_substr;
878 /* fbm_instr() takes into account exact value of end-of-str
879 if the check is SvTAIL(ed). Since false positives are OK,
880 and end-of-str is not later than strend we are OK. */
881 if (must == &PL_sv_undef) {
883 DEBUG_r(must = prog->float_utf8); /* for debug message */
886 s = fbm_instr((unsigned char*)s,
887 (unsigned char*)last + SvCUR(must)
889 must, multiline ? FBMrf_MULTILINE : 0);
891 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
892 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
893 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
894 (s ? "Found" : "Contradicts"),
895 quoted, RE_SV_TAIL(must));
899 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
900 ", giving up...\n"));
903 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
904 ", trying anchored starting at offset %ld...\n",
905 (long)(saved_s + 1 - i_strpos)));
907 s = HOP3c(t, 1, strend);
911 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
912 (long)(s - i_strpos)));
913 other_last = s; /* Fix this later. --Hugo */
923 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
925 DEBUG_OPTIMISE_MORE_r(
926 PerlIO_printf(Perl_debug_log,
927 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
928 (IV)prog->check_offset_min,
929 (IV)prog->check_offset_max,
937 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
939 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
942 /* Fixed substring is found far enough so that the match
943 cannot start at strpos. */
945 if (ml_anch && t[-1] != '\n') {
946 /* Eventually fbm_*() should handle this, but often
947 anchored_offset is not 0, so this check will not be wasted. */
948 /* XXXX In the code below we prefer to look for "^" even in
949 presence of anchored substrings. And we search even
950 beyond the found float position. These pessimizations
951 are historical artefacts only. */
953 while (t < strend - prog->minlen) {
955 if (t < check_at - prog->check_offset_min) {
956 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
957 /* Since we moved from the found position,
958 we definitely contradict the found anchored
959 substr. Due to the above check we do not
960 contradict "check" substr.
961 Thus we can arrive here only if check substr
962 is float. Redo checking for "other"=="fixed".
965 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
966 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
967 goto do_other_anchored;
969 /* We don't contradict the found floating substring. */
970 /* XXXX Why not check for STCLASS? */
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
973 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
976 /* Position contradicts check-string */
977 /* XXXX probably better to look for check-string
978 than for "\n", so one should lower the limit for t? */
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
980 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
981 other_last = strpos = s = t + 1;
986 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
987 PL_colors[0], PL_colors[1]));
991 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
992 PL_colors[0], PL_colors[1]));
996 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
999 /* The found string does not prohibit matching at strpos,
1000 - no optimization of calling REx engine can be performed,
1001 unless it was an MBOL and we are not after MBOL,
1002 or a future STCLASS check will fail this. */
1004 /* Even in this situation we may use MBOL flag if strpos is offset
1005 wrt the start of the string. */
1006 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1007 && (strpos != strbeg) && strpos[-1] != '\n'
1008 /* May be due to an implicit anchor of m{.*foo} */
1009 && !(prog->intflags & PREGf_IMPLICIT))
1014 DEBUG_EXECUTE_r( if (ml_anch)
1015 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1016 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1019 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1021 prog->check_utf8 /* Could be deleted already */
1022 && --BmUSEFUL(prog->check_utf8) < 0
1023 && (prog->check_utf8 == prog->float_utf8)
1025 prog->check_substr /* Could be deleted already */
1026 && --BmUSEFUL(prog->check_substr) < 0
1027 && (prog->check_substr == prog->float_substr)
1030 /* If flags & SOMETHING - do not do it many times on the same match */
1031 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1032 /* XXX Does the destruction order has to change with utf8_target? */
1033 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1034 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1035 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1036 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1037 check = NULL; /* abort */
1039 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1040 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1041 if (prog->intflags & PREGf_IMPLICIT)
1042 prog->extflags &= ~RXf_ANCH_MBOL;
1043 /* XXXX This is a remnant of the old implementation. It
1044 looks wasteful, since now INTUIT can use many
1045 other heuristics. */
1046 prog->extflags &= ~RXf_USE_INTUIT;
1047 /* XXXX What other flags might need to be cleared in this branch? */
1053 /* Last resort... */
1054 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1055 /* trie stclasses are too expensive to use here, we are better off to
1056 leave it to regmatch itself */
1057 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1058 /* minlen == 0 is possible if regstclass is \b or \B,
1059 and the fixed substr is ''$.
1060 Since minlen is already taken into account, s+1 is before strend;
1061 accidentally, minlen >= 1 guaranties no false positives at s + 1
1062 even for \b or \B. But (minlen? 1 : 0) below assumes that
1063 regstclass does not come from lookahead... */
1064 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1065 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1066 const U8* const str = (U8*)STRING(progi->regstclass);
1067 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1068 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1071 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1072 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1073 else if (prog->float_substr || prog->float_utf8)
1074 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1078 if (checked_upto < s)
1080 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1081 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1084 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1089 const char *what = NULL;
1091 if (endpos == strend) {
1092 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1093 "Could not match STCLASS...\n") );
1096 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1097 "This position contradicts STCLASS...\n") );
1098 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1100 checked_upto = HOPBACKc(endpos, start_shift);
1101 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1102 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1103 /* Contradict one of substrings */
1104 if (prog->anchored_substr || prog->anchored_utf8) {
1105 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1106 DEBUG_EXECUTE_r( what = "anchored" );
1108 s = HOP3c(t, 1, strend);
1109 if (s + start_shift + end_shift > strend) {
1110 /* XXXX Should be taken into account earlier? */
1111 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1112 "Could not match STCLASS...\n") );
1117 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1118 "Looking for %s substr starting at offset %ld...\n",
1119 what, (long)(s + start_shift - i_strpos)) );
1122 /* Have both, check_string is floating */
1123 if (t + start_shift >= check_at) /* Contradicts floating=check */
1124 goto retry_floating_check;
1125 /* Recheck anchored substring, but not floating... */
1129 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1130 "Looking for anchored substr starting at offset %ld...\n",
1131 (long)(other_last - i_strpos)) );
1132 goto do_other_anchored;
1134 /* Another way we could have checked stclass at the
1135 current position only: */
1140 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1141 "Looking for /%s^%s/m starting at offset %ld...\n",
1142 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1145 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1147 /* Check is floating substring. */
1148 retry_floating_check:
1149 t = check_at - start_shift;
1150 DEBUG_EXECUTE_r( what = "floating" );
1151 goto hop_and_restart;
1154 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1155 "By STCLASS: moving %ld --> %ld\n",
1156 (long)(t - i_strpos), (long)(s - i_strpos))
1160 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1161 "Does not contradict STCLASS...\n");
1166 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1167 PL_colors[4], (check ? "Guessed" : "Giving up"),
1168 PL_colors[5], (long)(s - i_strpos)) );
1171 fail_finish: /* Substring not found */
1172 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1173 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1175 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1176 PL_colors[4], PL_colors[5]));
1180 #define DECL_TRIE_TYPE(scan) \
1181 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1182 trie_type = ((scan->flags == EXACT) \
1183 ? (utf8_target ? trie_utf8 : trie_plain) \
1184 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1186 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1187 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1189 switch (trie_type) { \
1190 case trie_utf8_fold: \
1191 if ( foldlen>0 ) { \
1192 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1197 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1198 len = UTF8SKIP(uc); \
1199 skiplen = UNISKIP( uvc ); \
1200 foldlen -= skiplen; \
1201 uscan = foldbuf + skiplen; \
1204 case trie_latin_utf8_fold: \
1205 if ( foldlen>0 ) { \
1206 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1212 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1213 skiplen = UNISKIP( uvc ); \
1214 foldlen -= skiplen; \
1215 uscan = foldbuf + skiplen; \
1219 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1226 charid = trie->charmap[ uvc ]; \
1230 if (widecharmap) { \
1231 SV** const svpp = hv_fetch(widecharmap, \
1232 (char*)&uvc, sizeof(UV), 0); \
1234 charid = (U16)SvIV(*svpp); \
1239 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1243 && (ln == 1 || folder(s, pat_string, ln)) \
1244 && (!reginfo || regtry(reginfo, &s)) ) \
1250 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1252 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1258 #define REXEC_FBC_SCAN(CoDe) \
1260 while (s < strend) { \
1266 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1267 REXEC_FBC_UTF8_SCAN( \
1269 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1278 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1281 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1290 #define REXEC_FBC_TRYIT \
1291 if ((!reginfo || regtry(reginfo, &s))) \
1294 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1295 if (utf8_target) { \
1296 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1299 REXEC_FBC_CLASS_SCAN(CoNd); \
1302 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1303 if (utf8_target) { \
1305 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1308 REXEC_FBC_CLASS_SCAN(CoNd); \
1311 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1312 PL_reg_flags |= RF_tainted; \
1313 if (utf8_target) { \
1314 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1317 REXEC_FBC_CLASS_SCAN(CoNd); \
1320 #define DUMP_EXEC_POS(li,s,doutf8) \
1321 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1324 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1325 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1326 tmp = TEST_NON_UTF8(tmp); \
1327 REXEC_FBC_UTF8_SCAN( \
1328 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1337 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1338 if (s == PL_bostr) { \
1342 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1343 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1346 LOAD_UTF8_CHARCLASS_ALNUM(); \
1347 REXEC_FBC_UTF8_SCAN( \
1348 if (tmp == ! (TeSt2_UtF8)) { \
1357 /* The only difference between the BOUND and NBOUND cases is that
1358 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1359 * NBOUND. This is accomplished by passing it in either the if or else clause,
1360 * with the other one being empty */
1361 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1362 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1364 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1365 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1367 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1368 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1370 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1371 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1374 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1375 * be passed in completely with the variable name being tested, which isn't
1376 * such a clean interface, but this is easier to read than it was before. We
1377 * are looking for the boundary (or non-boundary between a word and non-word
1378 * character. The utf8 and non-utf8 cases have the same logic, but the details
1379 * must be different. Find the "wordness" of the character just prior to this
1380 * one, and compare it with the wordness of this one. If they differ, we have
1381 * a boundary. At the beginning of the string, pretend that the previous
1382 * character was a new-line */
1383 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1384 if (utf8_target) { \
1387 else { /* Not utf8 */ \
1388 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1389 tmp = TEST_NON_UTF8(tmp); \
1391 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1400 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1403 /* We know what class REx starts with. Try to find this position... */
1404 /* if reginfo is NULL, its a dryrun */
1405 /* annoyingly all the vars in this routine have different names from their counterparts
1406 in regmatch. /grrr */
1409 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1410 const char *strend, regmatch_info *reginfo)
1413 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1414 char *pat_string; /* The pattern's exactish string */
1415 char *pat_end; /* ptr to end char of pat_string */
1416 re_fold_t folder; /* Function for computing non-utf8 folds */
1417 const U8 *fold_array; /* array for folding ords < 256 */
1424 I32 tmp = 1; /* Scratch variable? */
1425 const bool utf8_target = PL_reg_match_utf8;
1426 UV utf8_fold_flags = 0;
1427 RXi_GET_DECL(prog,progi);
1429 PERL_ARGS_ASSERT_FIND_BYCLASS;
1431 /* We know what class it must start with. */
1435 if (utf8_target || OP(c) == ANYOFV) {
1436 STRLEN inclasslen = strend - s;
1437 REXEC_FBC_UTF8_CLASS_SCAN(
1438 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1441 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1446 if (tmp && (!reginfo || regtry(reginfo, &s)))
1454 if (UTF_PATTERN || utf8_target) {
1455 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1456 goto do_exactf_utf8;
1458 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1459 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1460 goto do_exactf_non_utf8; /* isn't dealt with by these */
1465 /* regcomp.c already folded this if pattern is in UTF-8 */
1466 utf8_fold_flags = 0;
1467 goto do_exactf_utf8;
1469 fold_array = PL_fold;
1471 goto do_exactf_non_utf8;
1474 if (UTF_PATTERN || utf8_target) {
1475 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1476 goto do_exactf_utf8;
1478 fold_array = PL_fold_locale;
1479 folder = foldEQ_locale;
1480 goto do_exactf_non_utf8;
1484 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1486 goto do_exactf_utf8;
1488 case EXACTFU_TRICKYFOLD:
1490 if (UTF_PATTERN || utf8_target) {
1491 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1492 goto do_exactf_utf8;
1495 /* Any 'ss' in the pattern should have been replaced by regcomp,
1496 * so we don't have to worry here about this single special case
1497 * in the Latin1 range */
1498 fold_array = PL_fold_latin1;
1499 folder = foldEQ_latin1;
1503 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1504 are no glitches with fold-length differences
1505 between the target string and pattern */
1507 /* The idea in the non-utf8 EXACTF* cases is to first find the
1508 * first character of the EXACTF* node and then, if necessary,
1509 * case-insensitively compare the full text of the node. c1 is the
1510 * first character. c2 is its fold. This logic will not work for
1511 * Unicode semantics and the german sharp ss, which hence should
1512 * not be compiled into a node that gets here. */
1513 pat_string = STRING(c);
1514 ln = STR_LEN(c); /* length to match in octets/bytes */
1516 /* We know that we have to match at least 'ln' bytes (which is the
1517 * same as characters, since not utf8). If we have to match 3
1518 * characters, and there are only 2 availabe, we know without
1519 * trying that it will fail; so don't start a match past the
1520 * required minimum number from the far end */
1521 e = HOP3c(strend, -((I32)ln), s);
1523 if (!reginfo && e < s) {
1524 e = s; /* Due to minlen logic of intuit() */
1528 c2 = fold_array[c1];
1529 if (c1 == c2) { /* If char and fold are the same */
1530 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1533 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1542 /* If one of the operands is in utf8, we can't use the simpler
1543 * folding above, due to the fact that many different characters
1544 * can have the same fold, or portion of a fold, or different-
1546 pat_string = STRING(c);
1547 ln = STR_LEN(c); /* length to match in octets/bytes */
1548 pat_end = pat_string + ln;
1549 lnc = (UTF_PATTERN) /* length to match in characters */
1550 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1553 /* We have 'lnc' characters to match in the pattern, but because of
1554 * multi-character folding, each character in the target can match
1555 * up to 3 characters (Unicode guarantees it will never exceed
1556 * this) if it is utf8-encoded; and up to 2 if not (based on the
1557 * fact that the Latin 1 folds are already determined, and the
1558 * only multi-char fold in that range is the sharp-s folding to
1559 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1560 * string character. Adjust lnc accordingly, rounding up, so that
1561 * if we need to match at least 4+1/3 chars, that really is 5. */
1562 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1563 lnc = (lnc + expansion - 1) / expansion;
1565 /* As in the non-UTF8 case, if we have to match 3 characters, and
1566 * only 2 are left, it's guaranteed to fail, so don't start a
1567 * match that would require us to go beyond the end of the string
1569 e = HOP3c(strend, -((I32)lnc), s);
1571 if (!reginfo && e < s) {
1572 e = s; /* Due to minlen logic of intuit() */
1575 /* XXX Note that we could recalculate e to stop the loop earlier,
1576 * as the worst case expansion above will rarely be met, and as we
1577 * go along we would usually find that e moves further to the left.
1578 * This would happen only after we reached the point in the loop
1579 * where if there were no expansion we should fail. Unclear if
1580 * worth the expense */
1583 char *my_strend= (char *)strend;
1584 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1585 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1586 && (!reginfo || regtry(reginfo, &s)) )
1590 s += (utf8_target) ? UTF8SKIP(s) : 1;
1595 PL_reg_flags |= RF_tainted;
1596 FBC_BOUND(isALNUM_LC,
1597 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1598 isALNUM_LC_utf8((U8*)s));
1601 PL_reg_flags |= RF_tainted;
1602 FBC_NBOUND(isALNUM_LC,
1603 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1604 isALNUM_LC_utf8((U8*)s));
1607 FBC_BOUND(isWORDCHAR,
1609 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1612 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1614 isWORDCHAR_A((U8*)s));
1617 FBC_NBOUND(isWORDCHAR,
1619 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1622 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1624 isWORDCHAR_A((U8*)s));
1627 FBC_BOUND(isWORDCHAR_L1,
1629 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1632 FBC_NBOUND(isWORDCHAR_L1,
1634 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1637 REXEC_FBC_CSCAN_TAINT(
1638 isALNUM_LC_utf8((U8*)s),
1643 REXEC_FBC_CSCAN_PRELOAD(
1644 LOAD_UTF8_CHARCLASS_ALNUM(),
1645 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1646 isWORDCHAR_L1((U8) *s)
1650 REXEC_FBC_CSCAN_PRELOAD(
1651 LOAD_UTF8_CHARCLASS_ALNUM(),
1652 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1657 /* Don't need to worry about utf8, as it can match only a single
1658 * byte invariant character */
1659 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1662 REXEC_FBC_CSCAN_PRELOAD(
1663 LOAD_UTF8_CHARCLASS_ALNUM(),
1664 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1665 ! isWORDCHAR_L1((U8) *s)
1669 REXEC_FBC_CSCAN_PRELOAD(
1670 LOAD_UTF8_CHARCLASS_ALNUM(),
1671 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1682 REXEC_FBC_CSCAN_TAINT(
1683 !isALNUM_LC_utf8((U8*)s),
1688 REXEC_FBC_CSCAN_PRELOAD(
1689 LOAD_UTF8_CHARCLASS_SPACE(),
1690 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1695 REXEC_FBC_CSCAN_PRELOAD(
1696 LOAD_UTF8_CHARCLASS_SPACE(),
1697 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1702 /* Don't need to worry about utf8, as it can match only a single
1703 * byte invariant character */
1704 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1707 REXEC_FBC_CSCAN_TAINT(
1708 isSPACE_LC_utf8((U8*)s),
1713 REXEC_FBC_CSCAN_PRELOAD(
1714 LOAD_UTF8_CHARCLASS_SPACE(),
1715 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1716 ! isSPACE_L1((U8) *s)
1720 REXEC_FBC_CSCAN_PRELOAD(
1721 LOAD_UTF8_CHARCLASS_SPACE(),
1722 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1733 REXEC_FBC_CSCAN_TAINT(
1734 !isSPACE_LC_utf8((U8*)s),
1739 REXEC_FBC_CSCAN_PRELOAD(
1740 LOAD_UTF8_CHARCLASS_DIGIT(),
1741 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1746 /* Don't need to worry about utf8, as it can match only a single
1747 * byte invariant character */
1748 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1751 REXEC_FBC_CSCAN_TAINT(
1752 isDIGIT_LC_utf8((U8*)s),
1757 REXEC_FBC_CSCAN_PRELOAD(
1758 LOAD_UTF8_CHARCLASS_DIGIT(),
1759 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1770 REXEC_FBC_CSCAN_TAINT(
1771 !isDIGIT_LC_utf8((U8*)s),
1778 is_LNBREAK_latin1(s)
1790 !is_VERTWS_latin1(s)
1796 is_HORIZWS_latin1(s)
1801 !is_HORIZWS_utf8(s),
1802 !is_HORIZWS_latin1(s)
1806 /* Don't need to worry about utf8, as it can match only a single
1807 * byte invariant character. The flag in this node type is the
1808 * class number to pass to _generic_isCC() to build a mask for
1809 * searching in PL_charclass[] */
1810 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1814 !_generic_isCC_A(*s, FLAGS(c)),
1815 !_generic_isCC_A(*s, FLAGS(c))
1823 /* what trie are we using right now */
1825 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1827 = (reg_trie_data*)progi->data->data[ aho->trie ];
1828 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1830 const char *last_start = strend - trie->minlen;
1832 const char *real_start = s;
1834 STRLEN maxlen = trie->maxlen;
1836 U8 **points; /* map of where we were in the input string
1837 when reading a given char. For ASCII this
1838 is unnecessary overhead as the relationship
1839 is always 1:1, but for Unicode, especially
1840 case folded Unicode this is not true. */
1841 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1845 GET_RE_DEBUG_FLAGS_DECL;
1847 /* We can't just allocate points here. We need to wrap it in
1848 * an SV so it gets freed properly if there is a croak while
1849 * running the match */
1852 sv_points=newSV(maxlen * sizeof(U8 *));
1853 SvCUR_set(sv_points,
1854 maxlen * sizeof(U8 *));
1855 SvPOK_on(sv_points);
1856 sv_2mortal(sv_points);
1857 points=(U8**)SvPV_nolen(sv_points );
1858 if ( trie_type != trie_utf8_fold
1859 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1862 bitmap=(U8*)trie->bitmap;
1864 bitmap=(U8*)ANYOF_BITMAP(c);
1866 /* this is the Aho-Corasick algorithm modified a touch
1867 to include special handling for long "unknown char"
1868 sequences. The basic idea being that we use AC as long
1869 as we are dealing with a possible matching char, when
1870 we encounter an unknown char (and we have not encountered
1871 an accepting state) we scan forward until we find a legal
1873 AC matching is basically that of trie matching, except
1874 that when we encounter a failing transition, we fall back
1875 to the current states "fail state", and try the current char
1876 again, a process we repeat until we reach the root state,
1877 state 1, or a legal transition. If we fail on the root state
1878 then we can either terminate if we have reached an accepting
1879 state previously, or restart the entire process from the beginning
1883 while (s <= last_start) {
1884 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1892 U8 *uscan = (U8*)NULL;
1893 U8 *leftmost = NULL;
1895 U32 accepted_word= 0;
1899 while ( state && uc <= (U8*)strend ) {
1901 U32 word = aho->states[ state ].wordnum;
1905 DEBUG_TRIE_EXECUTE_r(
1906 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1907 dump_exec_pos( (char *)uc, c, strend, real_start,
1908 (char *)uc, utf8_target );
1909 PerlIO_printf( Perl_debug_log,
1910 " Scanning for legal start char...\n");
1914 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1918 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1924 if (uc >(U8*)last_start) break;
1928 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1929 if (!leftmost || lpos < leftmost) {
1930 DEBUG_r(accepted_word=word);
1936 points[pointpos++ % maxlen]= uc;
1937 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1938 uscan, len, uvc, charid, foldlen,
1940 DEBUG_TRIE_EXECUTE_r({
1941 dump_exec_pos( (char *)uc, c, strend, real_start,
1943 PerlIO_printf(Perl_debug_log,
1944 " Charid:%3u CP:%4"UVxf" ",
1950 word = aho->states[ state ].wordnum;
1952 base = aho->states[ state ].trans.base;
1954 DEBUG_TRIE_EXECUTE_r({
1956 dump_exec_pos( (char *)uc, c, strend, real_start,
1958 PerlIO_printf( Perl_debug_log,
1959 "%sState: %4"UVxf", word=%"UVxf,
1960 failed ? " Fail transition to " : "",
1961 (UV)state, (UV)word);
1967 ( ((offset = base + charid
1968 - 1 - trie->uniquecharcount)) >= 0)
1969 && ((U32)offset < trie->lasttrans)
1970 && trie->trans[offset].check == state
1971 && (tmp=trie->trans[offset].next))
1973 DEBUG_TRIE_EXECUTE_r(
1974 PerlIO_printf( Perl_debug_log," - legal\n"));
1979 DEBUG_TRIE_EXECUTE_r(
1980 PerlIO_printf( Perl_debug_log," - fail\n"));
1982 state = aho->fail[state];
1986 /* we must be accepting here */
1987 DEBUG_TRIE_EXECUTE_r(
1988 PerlIO_printf( Perl_debug_log," - accepting\n"));
1997 if (!state) state = 1;
2000 if ( aho->states[ state ].wordnum ) {
2001 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2002 if (!leftmost || lpos < leftmost) {
2003 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2008 s = (char*)leftmost;
2009 DEBUG_TRIE_EXECUTE_r({
2011 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2012 (UV)accepted_word, (IV)(s - real_start)
2015 if (!reginfo || regtry(reginfo, &s)) {
2021 DEBUG_TRIE_EXECUTE_r({
2022 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2025 DEBUG_TRIE_EXECUTE_r(
2026 PerlIO_printf( Perl_debug_log,"No match.\n"));
2035 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2045 - regexec_flags - match a regexp against a string
2048 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2049 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2050 /* stringarg: the point in the string at which to begin matching */
2051 /* strend: pointer to null at end of string */
2052 /* strbeg: real beginning of string */
2053 /* minend: end of match must be >= minend bytes after stringarg. */
2054 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2055 * itself is accessed via the pointers above */
2056 /* data: May be used for some additional optimizations.
2057 Currently its only used, with a U32 cast, for transmitting
2058 the ganch offset when doing a /g match. This will change */
2059 /* nosave: For optimizations. */
2063 struct regexp *const prog = (struct regexp *)SvANY(rx);
2064 /*register*/ char *s;
2066 /*register*/ char *startpos = stringarg;
2067 I32 minlen; /* must match at least this many chars */
2068 I32 dontbother = 0; /* how many characters not to try at end */
2069 I32 end_shift = 0; /* Same for the end. */ /* CC */
2070 I32 scream_pos = -1; /* Internal iterator of scream. */
2071 char *scream_olds = NULL;
2072 const bool utf8_target = cBOOL(DO_UTF8(sv));
2074 RXi_GET_DECL(prog,progi);
2075 regmatch_info reginfo; /* create some info to pass to regtry etc */
2076 regexp_paren_pair *swap = NULL;
2077 GET_RE_DEBUG_FLAGS_DECL;
2079 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2080 PERL_UNUSED_ARG(data);
2082 /* Be paranoid... */
2083 if (prog == NULL || startpos == NULL) {
2084 Perl_croak(aTHX_ "NULL regexp parameter");
2088 multiline = prog->extflags & RXf_PMf_MULTILINE;
2089 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2091 RX_MATCH_UTF8_set(rx, utf8_target);
2093 debug_start_match(rx, utf8_target, startpos, strend,
2097 minlen = prog->minlen;
2099 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2100 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2101 "String too short [regexec_flags]...\n"));
2106 /* Check validity of program. */
2107 if (UCHARAT(progi->program) != REG_MAGIC) {
2108 Perl_croak(aTHX_ "corrupted regexp program");
2112 PL_reg_state.re_state_eval_setup_done = FALSE;
2116 PL_reg_flags |= RF_utf8;
2118 /* Mark beginning of line for ^ and lookbehind. */
2119 reginfo.bol = startpos; /* XXX not used ??? */
2123 /* Mark end of line for $ (and such) */
2126 /* see how far we have to get to not match where we matched before */
2127 reginfo.till = startpos+minend;
2129 /* If there is a "must appear" string, look for it. */
2132 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2134 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2135 reginfo.ganch = startpos + prog->gofs;
2136 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2137 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2138 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2140 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2141 && mg->mg_len >= 0) {
2142 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2143 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2144 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2146 if (prog->extflags & RXf_ANCH_GPOS) {
2147 if (s > reginfo.ganch)
2149 s = reginfo.ganch - prog->gofs;
2150 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2151 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2157 reginfo.ganch = strbeg + PTR2UV(data);
2158 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2159 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2161 } else { /* pos() not defined */
2162 reginfo.ganch = strbeg;
2163 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2164 "GPOS: reginfo.ganch = strbeg\n"));
2167 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2168 /* We have to be careful. If the previous successful match
2169 was from this regex we don't want a subsequent partially
2170 successful match to clobber the old results.
2171 So when we detect this possibility we add a swap buffer
2172 to the re, and switch the buffer each match. If we fail
2173 we switch it back, otherwise we leave it swapped.
2176 /* do we need a save destructor here for eval dies? */
2177 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2178 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2179 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2185 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2186 re_scream_pos_data d;
2188 d.scream_olds = &scream_olds;
2189 d.scream_pos = &scream_pos;
2190 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2192 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2193 goto phooey; /* not present */
2199 /* Simplest case: anchored match need be tried only once. */
2200 /* [unless only anchor is BOL and multiline is set] */
2201 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2202 if (s == startpos && regtry(®info, &startpos))
2204 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2205 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2210 dontbother = minlen - 1;
2211 end = HOP3c(strend, -dontbother, strbeg) - 1;
2212 /* for multiline we only have to try after newlines */
2213 if (prog->check_substr || prog->check_utf8) {
2214 /* because of the goto we can not easily reuse the macros for bifurcating the
2215 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2218 goto after_try_utf8;
2220 if (regtry(®info, &s)) {
2227 if (prog->extflags & RXf_USE_INTUIT) {
2228 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2237 } /* end search for check string in unicode */
2239 if (s == startpos) {
2240 goto after_try_latin;
2243 if (regtry(®info, &s)) {
2250 if (prog->extflags & RXf_USE_INTUIT) {
2251 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2260 } /* end search for check string in latin*/
2261 } /* end search for check string */
2262 else { /* search for newline */
2264 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2267 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2268 while (s <= end) { /* note it could be possible to match at the end of the string */
2269 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2270 if (regtry(®info, &s))
2274 } /* end search for newline */
2275 } /* end anchored/multiline check string search */
2277 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2279 /* the warning about reginfo.ganch being used without initialization
2280 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2281 and we only enter this block when the same bit is set. */
2282 char *tmp_s = reginfo.ganch - prog->gofs;
2284 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2289 /* Messy cases: unanchored match. */
2290 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2291 /* we have /x+whatever/ */
2292 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2297 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2298 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2299 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2304 DEBUG_EXECUTE_r( did_match = 1 );
2305 if (regtry(®info, &s)) goto got_it;
2307 while (s < strend && *s == ch)
2315 DEBUG_EXECUTE_r( did_match = 1 );
2316 if (regtry(®info, &s)) goto got_it;
2318 while (s < strend && *s == ch)
2323 DEBUG_EXECUTE_r(if (!did_match)
2324 PerlIO_printf(Perl_debug_log,
2325 "Did not find anchored character...\n")
2328 else if (prog->anchored_substr != NULL
2329 || prog->anchored_utf8 != NULL
2330 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2331 && prog->float_max_offset < strend - s)) {
2336 char *last1; /* Last position checked before */
2340 if (prog->anchored_substr || prog->anchored_utf8) {
2341 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2342 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2343 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2344 back_max = back_min = prog->anchored_offset;
2346 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2347 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2348 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2349 back_max = prog->float_max_offset;
2350 back_min = prog->float_min_offset;
2354 if (must == &PL_sv_undef)
2355 /* could not downgrade utf8 check substring, so must fail */
2361 last = HOP3c(strend, /* Cannot start after this */
2362 -(I32)(CHR_SVLEN(must)
2363 - (SvTAIL(must) != 0) + back_min), strbeg);
2366 last1 = HOPc(s, -1);
2368 last1 = s - 1; /* bogus */
2370 /* XXXX check_substr already used to find "s", can optimize if
2371 check_substr==must. */
2373 dontbother = end_shift;
2374 strend = HOPc(strend, -dontbother);
2375 while ( (s <= last) &&
2376 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2377 (unsigned char*)strend, must,
2378 multiline ? FBMrf_MULTILINE : 0)) ) {
2379 DEBUG_EXECUTE_r( did_match = 1 );
2380 if (HOPc(s, -back_max) > last1) {
2381 last1 = HOPc(s, -back_min);
2382 s = HOPc(s, -back_max);
2385 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2387 last1 = HOPc(s, -back_min);
2391 while (s <= last1) {
2392 if (regtry(®info, &s))
2398 while (s <= last1) {
2399 if (regtry(®info, &s))
2405 DEBUG_EXECUTE_r(if (!did_match) {
2406 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2407 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2408 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2409 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2410 ? "anchored" : "floating"),
2411 quoted, RE_SV_TAIL(must));
2415 else if ( (c = progi->regstclass) ) {
2417 const OPCODE op = OP(progi->regstclass);
2418 /* don't bother with what can't match */
2419 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2420 strend = HOPc(strend, -(minlen - 1));
2423 SV * const prop = sv_newmortal();
2424 regprop(prog, prop, c);
2426 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2428 PerlIO_printf(Perl_debug_log,
2429 "Matching stclass %.*s against %s (%d bytes)\n",
2430 (int)SvCUR(prop), SvPVX_const(prop),
2431 quoted, (int)(strend - s));
2434 if (find_byclass(prog, c, s, strend, ®info))
2436 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2440 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2447 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2448 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2449 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2451 little = SvPV_const(float_real, len);
2452 if (SvTAIL(float_real)) {
2453 /* This means that float_real contains an artificial \n on the end
2454 * due to the presence of something like this: /foo$/
2455 * where we can match both "foo" and "foo\n" at the end of the string.
2456 * So we have to compare the end of the string first against the float_real
2457 * without the \n and then against the full float_real with the string.
2458 * We have to watch out for cases where the string might be smaller
2459 * than the float_real or the float_real without the \n.
2461 char *checkpos= strend - len;
2463 PerlIO_printf(Perl_debug_log,
2464 "%sChecking for float_real.%s\n",
2465 PL_colors[4], PL_colors[5]));
2466 if (checkpos + 1 < strbeg) {
2467 /* can't match, even if we remove the trailing \n string is too short to match */
2469 PerlIO_printf(Perl_debug_log,
2470 "%sString shorter than required trailing substring, cannot match.%s\n",
2471 PL_colors[4], PL_colors[5]));
2473 } else if (memEQ(checkpos + 1, little, len - 1)) {
2474 /* can match, the end of the string matches without the "\n" */
2475 last = checkpos + 1;
2476 } else if (checkpos < strbeg) {
2477 /* cant match, string is too short when the "\n" is included */
2479 PerlIO_printf(Perl_debug_log,
2480 "%sString does not contain required trailing substring, cannot match.%s\n",
2481 PL_colors[4], PL_colors[5]));
2483 } else if (!multiline) {
2484 /* non multiline match, so compare with the "\n" at the end of the string */
2485 if (memEQ(checkpos, little, len)) {
2489 PerlIO_printf(Perl_debug_log,
2490 "%sString does not contain required trailing substring, cannot match.%s\n",
2491 PL_colors[4], PL_colors[5]));
2495 /* multiline match, so we have to search for a place where the full string is located */
2501 last = rninstr(s, strend, little, little + len);
2503 last = strend; /* matching "$" */
2506 /* at one point this block contained a comment which was probably
2507 * incorrect, which said that this was a "should not happen" case.
2508 * Even if it was true when it was written I am pretty sure it is
2509 * not anymore, so I have removed the comment and replaced it with
2512 PerlIO_printf(Perl_debug_log,
2513 "String does not contain required substring, cannot match.\n"
2517 dontbother = strend - last + prog->float_min_offset;
2519 if (minlen && (dontbother < minlen))
2520 dontbother = minlen - 1;
2521 strend -= dontbother; /* this one's always in bytes! */
2522 /* We don't know much -- general case. */
2525 if (regtry(®info, &s))
2534 if (regtry(®info, &s))
2536 } while (s++ < strend);
2546 PerlIO_printf(Perl_debug_log,
2547 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2553 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2555 if (PL_reg_state.re_state_eval_setup_done)
2556 restore_pos(aTHX_ prog);
2557 if (RXp_PAREN_NAMES(prog))
2558 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2560 /* make sure $`, $&, $', and $digit will work later */
2561 if ( !(flags & REXEC_NOT_FIRST) ) {
2562 if (flags & REXEC_COPY_STR) {
2563 #ifdef PERL_OLD_COPY_ON_WRITE
2565 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2567 PerlIO_printf(Perl_debug_log,
2568 "Copy on write: regexp capture, type %d\n",
2571 RX_MATCH_COPY_FREE(rx);
2572 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2573 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2574 assert (SvPOKp(prog->saved_copy));
2575 prog->sublen = PL_regeol - strbeg;
2576 prog->suboffset = 0;
2577 prog->subcoffset = 0;
2582 I32 max = PL_regeol - strbeg;
2585 if ( (flags & REXEC_COPY_SKIP_POST)
2586 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2587 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2588 ) { /* don't copy $' part of string */
2591 /* calculate the right-most part of the string covered
2592 * by a capture. Due to look-ahead, this may be to
2593 * the right of $&, so we have to scan all captures */
2594 while (n <= prog->lastparen) {
2595 if (prog->offs[n].end > max)
2596 max = prog->offs[n].end;
2600 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2601 ? prog->offs[0].start
2603 assert(max >= 0 && max <= PL_regeol - strbeg);
2606 if ( (flags & REXEC_COPY_SKIP_PRE)
2607 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2608 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2609 ) { /* don't copy $` part of string */
2612 /* calculate the left-most part of the string covered
2613 * by a capture. Due to look-behind, this may be to
2614 * the left of $&, so we have to scan all captures */
2615 while (min && n <= prog->lastparen) {
2616 if ( prog->offs[n].start != -1
2617 && prog->offs[n].start < min)
2619 min = prog->offs[n].start;
2623 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2624 && min > prog->offs[0].end
2626 min = prog->offs[0].end;
2630 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2633 if (RX_MATCH_COPIED(rx)) {
2634 if (sublen > prog->sublen)
2636 (char*)saferealloc(prog->subbeg, sublen+1);
2639 prog->subbeg = (char*)safemalloc(sublen+1);
2640 Copy(strbeg + min, prog->subbeg, sublen, char);
2641 prog->subbeg[sublen] = '\0';
2642 prog->suboffset = min;
2643 prog->sublen = sublen;
2644 RX_MATCH_COPIED_on(rx);
2646 prog->subcoffset = prog->suboffset;
2647 if (prog->suboffset && utf8_target) {
2648 /* Convert byte offset to chars.
2649 * XXX ideally should only compute this if @-/@+
2650 * has been seen, a la PL_sawampersand ??? */
2652 /* If there's a direct correspondence between the
2653 * string which we're matching and the original SV,
2654 * then we can use the utf8 len cache associated with
2655 * the SV. In particular, it means that under //g,
2656 * sv_pos_b2u() will use the previously cached
2657 * position to speed up working out the new length of
2658 * subcoffset, rather than counting from the start of
2659 * the string each time. This stops
2660 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2661 * from going quadratic */
2662 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2663 sv_pos_b2u(sv, &(prog->subcoffset));
2665 prog->subcoffset = utf8_length((U8*)strbeg,
2666 (U8*)(strbeg+prog->suboffset));
2670 RX_MATCH_COPY_FREE(rx);
2671 prog->subbeg = strbeg;
2672 prog->suboffset = 0;
2673 prog->subcoffset = 0;
2674 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2681 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2682 PL_colors[4], PL_colors[5]));
2683 if (PL_reg_state.re_state_eval_setup_done)
2684 restore_pos(aTHX_ prog);
2686 /* we failed :-( roll it back */
2687 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2688 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2693 Safefree(prog->offs);
2701 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2702 * Do inc before dec, in case old and new rex are the same */
2703 #define SET_reg_curpm(Re2) \
2704 if (PL_reg_state.re_state_eval_setup_done) { \
2705 (void)ReREFCNT_inc(Re2); \
2706 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2707 PM_SETRE((PL_reg_curpm), (Re2)); \
2712 - regtry - try match at specific point
2714 STATIC I32 /* 0 failure, 1 success */
2715 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2719 REGEXP *const rx = reginfo->prog;
2720 regexp *const prog = (struct regexp *)SvANY(rx);
2722 RXi_GET_DECL(prog,progi);
2723 GET_RE_DEBUG_FLAGS_DECL;
2725 PERL_ARGS_ASSERT_REGTRY;
2727 reginfo->cutpoint=NULL;
2729 if ((prog->extflags & RXf_EVAL_SEEN)
2730 && !PL_reg_state.re_state_eval_setup_done)
2734 PL_reg_state.re_state_eval_setup_done = TRUE;
2736 /* Make $_ available to executed code. */
2737 if (reginfo->sv != DEFSV) {
2739 DEFSV_set(reginfo->sv);
2742 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2743 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2744 /* prepare for quick setting of pos */
2745 #ifdef PERL_OLD_COPY_ON_WRITE
2746 if (SvIsCOW(reginfo->sv))
2747 sv_force_normal_flags(reginfo->sv, 0);
2749 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2750 &PL_vtbl_mglob, NULL, 0);
2754 PL_reg_oldpos = mg->mg_len;
2755 SAVEDESTRUCTOR_X(restore_pos, prog);
2757 if (!PL_reg_curpm) {
2758 Newxz(PL_reg_curpm, 1, PMOP);
2761 SV* const repointer = &PL_sv_undef;
2762 /* this regexp is also owned by the new PL_reg_curpm, which
2763 will try to free it. */
2764 av_push(PL_regex_padav, repointer);
2765 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2766 PL_regex_pad = AvARRAY(PL_regex_padav);
2771 PL_reg_oldcurpm = PL_curpm;
2772 PL_curpm = PL_reg_curpm;
2773 if (RXp_MATCH_COPIED(prog)) {
2774 /* Here is a serious problem: we cannot rewrite subbeg,
2775 since it may be needed if this match fails. Thus
2776 $` inside (?{}) could fail... */
2777 PL_reg_oldsaved = prog->subbeg;
2778 PL_reg_oldsavedlen = prog->sublen;
2779 PL_reg_oldsavedoffset = prog->suboffset;
2780 PL_reg_oldsavedcoffset = prog->suboffset;
2781 #ifdef PERL_OLD_COPY_ON_WRITE
2782 PL_nrs = prog->saved_copy;
2784 RXp_MATCH_COPIED_off(prog);
2787 PL_reg_oldsaved = NULL;
2788 prog->subbeg = PL_bostr;
2789 prog->suboffset = 0;
2790 prog->subcoffset = 0;
2791 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2794 PL_reg_starttry = *startposp;
2796 prog->offs[0].start = *startposp - PL_bostr;
2797 prog->lastparen = 0;
2798 prog->lastcloseparen = 0;
2801 /* XXXX What this code is doing here?!!! There should be no need
2802 to do this again and again, prog->lastparen should take care of
2805 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2806 * Actually, the code in regcppop() (which Ilya may be meaning by
2807 * prog->lastparen), is not needed at all by the test suite
2808 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2809 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2810 * Meanwhile, this code *is* needed for the
2811 * above-mentioned test suite tests to succeed. The common theme
2812 * on those tests seems to be returning null fields from matches.
2813 * --jhi updated by dapm */
2815 if (prog->nparens) {
2816 regexp_paren_pair *pp = prog->offs;
2818 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2826 result = regmatch(reginfo, *startposp, progi->program + 1);
2828 prog->offs[0].end = result;
2831 if (reginfo->cutpoint)
2832 *startposp= reginfo->cutpoint;
2833 REGCP_UNWIND(lastcp);
2838 #define sayYES goto yes
2839 #define sayNO goto no
2840 #define sayNO_SILENT goto no_silent
2842 /* we dont use STMT_START/END here because it leads to
2843 "unreachable code" warnings, which are bogus, but distracting. */
2844 #define CACHEsayNO \
2845 if (ST.cache_mask) \
2846 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2849 /* this is used to determine how far from the left messages like
2850 'failed...' are printed. It should be set such that messages
2851 are inline with the regop output that created them.
2853 #define REPORT_CODE_OFF 32
2856 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2857 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2859 #define SLAB_FIRST(s) (&(s)->states[0])
2860 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2862 /* grab a new slab and return the first slot in it */
2864 STATIC regmatch_state *
2867 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2870 regmatch_slab *s = PL_regmatch_slab->next;
2872 Newx(s, 1, regmatch_slab);
2873 s->prev = PL_regmatch_slab;
2875 PL_regmatch_slab->next = s;
2877 PL_regmatch_slab = s;
2878 return SLAB_FIRST(s);
2882 /* push a new state then goto it */
2884 #define PUSH_STATE_GOTO(state, node, input) \
2885 pushinput = input; \
2887 st->resume_state = state; \
2890 /* push a new state with success backtracking, then goto it */
2892 #define PUSH_YES_STATE_GOTO(state, node, input) \
2893 pushinput = input; \
2895 st->resume_state = state; \
2896 goto push_yes_state;
2903 regmatch() - main matching routine
2905 This is basically one big switch statement in a loop. We execute an op,
2906 set 'next' to point the next op, and continue. If we come to a point which
2907 we may need to backtrack to on failure such as (A|B|C), we push a
2908 backtrack state onto the backtrack stack. On failure, we pop the top
2909 state, and re-enter the loop at the state indicated. If there are no more
2910 states to pop, we return failure.
2912 Sometimes we also need to backtrack on success; for example /A+/, where
2913 after successfully matching one A, we need to go back and try to
2914 match another one; similarly for lookahead assertions: if the assertion
2915 completes successfully, we backtrack to the state just before the assertion
2916 and then carry on. In these cases, the pushed state is marked as
2917 'backtrack on success too'. This marking is in fact done by a chain of
2918 pointers, each pointing to the previous 'yes' state. On success, we pop to
2919 the nearest yes state, discarding any intermediate failure-only states.
2920 Sometimes a yes state is pushed just to force some cleanup code to be
2921 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2922 it to free the inner regex.
2924 Note that failure backtracking rewinds the cursor position, while
2925 success backtracking leaves it alone.
2927 A pattern is complete when the END op is executed, while a subpattern
2928 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2929 ops trigger the "pop to last yes state if any, otherwise return true"
2932 A common convention in this function is to use A and B to refer to the two
2933 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2934 the subpattern to be matched possibly multiple times, while B is the entire
2935 rest of the pattern. Variable and state names reflect this convention.
2937 The states in the main switch are the union of ops and failure/success of
2938 substates associated with with that op. For example, IFMATCH is the op
2939 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2940 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2941 successfully matched A and IFMATCH_A_fail is a state saying that we have
2942 just failed to match A. Resume states always come in pairs. The backtrack
2943 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2944 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2945 on success or failure.
2947 The struct that holds a backtracking state is actually a big union, with
2948 one variant for each major type of op. The variable st points to the
2949 top-most backtrack struct. To make the code clearer, within each
2950 block of code we #define ST to alias the relevant union.
2952 Here's a concrete example of a (vastly oversimplified) IFMATCH
2958 #define ST st->u.ifmatch
2960 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2961 ST.foo = ...; // some state we wish to save
2963 // push a yes backtrack state with a resume value of
2964 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2966 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
2969 case IFMATCH_A: // we have successfully executed A; now continue with B
2971 bar = ST.foo; // do something with the preserved value
2974 case IFMATCH_A_fail: // A failed, so the assertion failed
2975 ...; // do some housekeeping, then ...
2976 sayNO; // propagate the failure
2983 For any old-timers reading this who are familiar with the old recursive
2984 approach, the code above is equivalent to:
2986 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2995 ...; // do some housekeeping, then ...
2996 sayNO; // propagate the failure
2999 The topmost backtrack state, pointed to by st, is usually free. If you
3000 want to claim it, populate any ST.foo fields in it with values you wish to
3001 save, then do one of
3003 PUSH_STATE_GOTO(resume_state, node, newinput);
3004 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3006 which sets that backtrack state's resume value to 'resume_state', pushes a
3007 new free entry to the top of the backtrack stack, then goes to 'node'.
3008 On backtracking, the free slot is popped, and the saved state becomes the
3009 new free state. An ST.foo field in this new top state can be temporarily
3010 accessed to retrieve values, but once the main loop is re-entered, it
3011 becomes available for reuse.
3013 Note that the depth of the backtrack stack constantly increases during the
3014 left-to-right execution of the pattern, rather than going up and down with
3015 the pattern nesting. For example the stack is at its maximum at Z at the
3016 end of the pattern, rather than at X in the following:
3018 /(((X)+)+)+....(Y)+....Z/
3020 The only exceptions to this are lookahead/behind assertions and the cut,
3021 (?>A), which pop all the backtrack states associated with A before
3024 Backtrack state structs are allocated in slabs of about 4K in size.
3025 PL_regmatch_state and st always point to the currently active state,
3026 and PL_regmatch_slab points to the slab currently containing
3027 PL_regmatch_state. The first time regmatch() is called, the first slab is
3028 allocated, and is never freed until interpreter destruction. When the slab
3029 is full, a new one is allocated and chained to the end. At exit from
3030 regmatch(), slabs allocated since entry are freed.
3035 #define DEBUG_STATE_pp(pp) \
3037 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3038 PerlIO_printf(Perl_debug_log, \
3039 " %*s"pp" %s%s%s%s%s\n", \
3041 PL_reg_name[st->resume_state], \
3042 ((st==yes_state||st==mark_state) ? "[" : ""), \
3043 ((st==yes_state) ? "Y" : ""), \
3044 ((st==mark_state) ? "M" : ""), \
3045 ((st==yes_state||st==mark_state) ? "]" : "") \
3050 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3055 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3056 const char *start, const char *end, const char *blurb)
3058 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3060 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3065 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3066 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3068 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3069 start, end - start, 60);
3071 PerlIO_printf(Perl_debug_log,
3072 "%s%s REx%s %s against %s\n",
3073 PL_colors[4], blurb, PL_colors[5], s0, s1);
3075 if (utf8_target||utf8_pat)
3076 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3077 utf8_pat ? "pattern" : "",
3078 utf8_pat && utf8_target ? " and " : "",
3079 utf8_target ? "string" : ""
3085 S_dump_exec_pos(pTHX_ const char *locinput,
3086 const regnode *scan,
3087 const char *loc_regeol,
3088 const char *loc_bostr,
3089 const char *loc_reg_starttry,
3090 const bool utf8_target)
3092 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3093 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3094 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3095 /* The part of the string before starttry has one color
3096 (pref0_len chars), between starttry and current
3097 position another one (pref_len - pref0_len chars),
3098 after the current position the third one.
3099 We assume that pref0_len <= pref_len, otherwise we
3100 decrease pref0_len. */
3101 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3102 ? (5 + taill) - l : locinput - loc_bostr;
3105 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3107 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3109 pref0_len = pref_len - (locinput - loc_reg_starttry);
3110 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3111 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3112 ? (5 + taill) - pref_len : loc_regeol - locinput);
3113 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3117 if (pref0_len > pref_len)
3118 pref0_len = pref_len;
3120 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3122 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3123 (locinput - pref_len),pref0_len, 60, 4, 5);
3125 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3126 (locinput - pref_len + pref0_len),
3127 pref_len - pref0_len, 60, 2, 3);
3129 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3130 locinput, loc_regeol - locinput, 10, 0, 1);
3132 const STRLEN tlen=len0+len1+len2;
3133 PerlIO_printf(Perl_debug_log,
3134 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3135 (IV)(locinput - loc_bostr),
3138 (docolor ? "" : "> <"),
3140 (int)(tlen > 19 ? 0 : 19 - tlen),
3147 /* reg_check_named_buff_matched()
3148 * Checks to see if a named buffer has matched. The data array of
3149 * buffer numbers corresponding to the buffer is expected to reside
3150 * in the regexp->data->data array in the slot stored in the ARG() of
3151 * node involved. Note that this routine doesn't actually care about the
3152 * name, that information is not preserved from compilation to execution.
3153 * Returns the index of the leftmost defined buffer with the given name
3154 * or 0 if non of the buffers matched.
3157 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3160 RXi_GET_DECL(rex,rexi);
3161 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3162 I32 *nums=(I32*)SvPVX(sv_dat);
3164 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3166 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3167 if ((I32)rex->lastparen >= nums[n] &&
3168 rex->offs[nums[n]].end != -1)
3177 /* free all slabs above current one - called during LEAVE_SCOPE */
3180 S_clear_backtrack_stack(pTHX_ void *p)
3182 regmatch_slab *s = PL_regmatch_slab->next;
3187 PL_regmatch_slab->next = NULL;
3189 regmatch_slab * const osl = s;
3196 /* returns -1 on failure, $+[0] on success */
3198 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3200 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3204 const bool utf8_target = PL_reg_match_utf8;
3205 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3206 REGEXP *rex_sv = reginfo->prog;
3207 regexp *rex = (struct regexp *)SvANY(rex_sv);
3208 RXi_GET_DECL(rex,rexi);
3210 /* the current state. This is a cached copy of PL_regmatch_state */
3212 /* cache heavy used fields of st in registers */
3215 U32 n = 0; /* general value; init to avoid compiler warning */
3216 I32 ln = 0; /* len or last; init to avoid compiler warning */
3217 char *reginput = startpos;
3218 char *locinput = reginput;
3219 char *pushinput; /* where to continue after a PUSH */
3220 I32 nextchr; /* is always set to UCHARAT(locinput) */
3222 bool result = 0; /* return value of S_regmatch */
3223 int depth = 0; /* depth of backtrack stack */
3224 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3225 const U32 max_nochange_depth =
3226 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3227 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3228 regmatch_state *yes_state = NULL; /* state to pop to on success of
3230 /* mark_state piggy backs on the yes_state logic so that when we unwind
3231 the stack on success we can update the mark_state as we go */
3232 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3233 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3234 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3236 bool no_final = 0; /* prevent failure from backtracking? */
3237 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3238 char *startpoint = reginput;
3239 SV *popmark = NULL; /* are we looking for a mark? */
3240 SV *sv_commit = NULL; /* last mark name seen in failure */
3241 SV *sv_yes_mark = NULL; /* last mark name we have seen
3242 during a successful match */
3243 U32 lastopen = 0; /* last open we saw */
3244 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3245 SV* const oreplsv = GvSV(PL_replgv);
3246 /* these three flags are set by various ops to signal information to
3247 * the very next op. They have a useful lifetime of exactly one loop
3248 * iteration, and are not preserved or restored by state pushes/pops
3250 bool sw = 0; /* the condition value in (?(cond)a|b) */
3251 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3252 int logical = 0; /* the following EVAL is:
3256 or the following IFMATCH/UNLESSM is:
3257 false: plain (?=foo)
3258 true: used as a condition: (?(?=foo))
3260 PAD* last_pad = NULL;
3262 I32 gimme = G_SCALAR;
3263 CV *caller_cv = NULL; /* who called us */
3264 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3265 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3268 GET_RE_DEBUG_FLAGS_DECL;
3271 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3272 multicall_oldcatch = 0;
3273 multicall_cv = NULL;
3275 PERL_UNUSED_VAR(multicall_cop);
3276 PERL_UNUSED_VAR(newsp);
3279 PERL_ARGS_ASSERT_REGMATCH;
3281 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3282 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3284 /* on first ever call to regmatch, allocate first slab */
3285 if (!PL_regmatch_slab) {
3286 Newx(PL_regmatch_slab, 1, regmatch_slab);
3287 PL_regmatch_slab->prev = NULL;
3288 PL_regmatch_slab->next = NULL;
3289 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3292 oldsave = PL_savestack_ix;
3293 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3294 SAVEVPTR(PL_regmatch_slab);
3295 SAVEVPTR(PL_regmatch_state);
3297 /* grab next free state slot */
3298 st = ++PL_regmatch_state;
3299 if (st > SLAB_LAST(PL_regmatch_slab))
3300 st = PL_regmatch_state = S_push_slab(aTHX);
3302 /* Note that nextchr is a byte even in UTF */
3303 nextchr = UCHARAT(locinput);
3305 while (scan != NULL) {
3308 SV * const prop = sv_newmortal();
3309 regnode *rnext=regnext(scan);
3310 DUMP_EXEC_POS( locinput, scan, utf8_target );
3311 regprop(rex, prop, scan);
3313 PerlIO_printf(Perl_debug_log,
3314 "%3"IVdf":%*s%s(%"IVdf")\n",
3315 (IV)(scan - rexi->program), depth*2, "",
3317 (PL_regkind[OP(scan)] == END || !rnext) ?
3318 0 : (IV)(rnext - rexi->program));
3321 next = scan + NEXT_OFF(scan);
3324 state_num = OP(scan);
3328 switch (state_num) {
3330 if (locinput == PL_bostr)
3332 /* reginfo->till = reginfo->bol; */
3337 if (locinput == PL_bostr ||
3338 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3344 if (locinput == PL_bostr)
3348 if (locinput == reginfo->ganch)
3353 /* update the startpoint */
3354 st->u.keeper.val = rex->offs[0].start;
3355 reginput = locinput;
3356 rex->offs[0].start = locinput - PL_bostr;
3357 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3359 case KEEPS_next_fail:
3360 /* rollback the start point change */
3361 rex->offs[0].start = st->u.keeper.val;
3367 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3372 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3374 if (PL_regeol - locinput > 1)
3378 if (PL_regeol != locinput)
3382 if (!nextchr && locinput >= PL_regeol)
3385 locinput += PL_utf8skip[nextchr];
3386 if (locinput > PL_regeol)
3388 nextchr = UCHARAT(locinput);
3391 nextchr = UCHARAT(++locinput);
3394 if (!nextchr && locinput >= PL_regeol)
3396 nextchr = UCHARAT(++locinput);
3399 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3402 locinput += PL_utf8skip[nextchr];
3403 if (locinput > PL_regeol)
3405 nextchr = UCHARAT(locinput);
3408 nextchr = UCHARAT(++locinput);
3412 #define ST st->u.trie
3414 /* In this case the charclass data is available inline so
3415 we can fail fast without a lot of extra overhead.
3417 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3419 PerlIO_printf(Perl_debug_log,
3420 "%*s %sfailed to match trie start class...%s\n",
3421 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3424 assert(0); /* NOTREACHED */
3428 /* the basic plan of execution of the trie is:
3429 * At the beginning, run though all the states, and
3430 * find the longest-matching word. Also remember the position
3431 * of the shortest matching word. For example, this pattern:
3434 * when matched against the string "abcde", will generate
3435 * accept states for all words except 3, with the longest
3436 * matching word being 4, and the shortest being 1 (with
3437 * the position being after char 1 of the string).
3439 * Then for each matching word, in word order (i.e. 1,2,4,5),
3440 * we run the remainder of the pattern; on each try setting
3441 * the current position to the character following the word,
3442 * returning to try the next word on failure.
3444 * We avoid having to build a list of words at runtime by
3445 * using a compile-time structure, wordinfo[].prev, which
3446 * gives, for each word, the previous accepting word (if any).
3447 * In the case above it would contain the mappings 1->2, 2->0,
3448 * 3->0, 4->5, 5->1. We can use this table to generate, from
3449 * the longest word (4 above), a list of all words, by
3450 * following the list of prev pointers; this gives us the
3451 * unordered list 4,5,1,2. Then given the current word we have
3452 * just tried, we can go through the list and find the
3453 * next-biggest word to try (so if we just failed on word 2,
3454 * the next in the list is 4).
3456 * Since at runtime we don't record the matching position in
3457 * the string for each word, we have to work that out for
3458 * each word we're about to process. The wordinfo table holds
3459 * the character length of each word; given that we recorded
3460 * at the start: the position of the shortest word and its
3461 * length in chars, we just need to move the pointer the
3462 * difference between the two char lengths. Depending on
3463 * Unicode status and folding, that's cheap or expensive.
3465 * This algorithm is optimised for the case where are only a
3466 * small number of accept states, i.e. 0,1, or maybe 2.
3467 * With lots of accepts states, and having to try all of them,
3468 * it becomes quadratic on number of accept states to find all
3473 /* what type of TRIE am I? (utf8 makes this contextual) */
3474 DECL_TRIE_TYPE(scan);
3476 /* what trie are we using right now */
3477 reg_trie_data * const trie
3478 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3479 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3480 U32 state = trie->startstate;
3482 if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
3483 if (trie->states[ state ].wordnum) {
3485 PerlIO_printf(Perl_debug_log,
3486 "%*s %smatched empty string...%s\n",
3487 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3493 PerlIO_printf(Perl_debug_log,
3494 "%*s %sfailed to match trie start class...%s\n",
3495 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3502 U8 *uc = ( U8* )locinput;
3506 U8 *uscan = (U8*)NULL;
3507 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3508 U32 charcount = 0; /* how many input chars we have matched */
3509 U32 accepted = 0; /* have we seen any accepting states? */
3511 ST.jump = trie->jump;
3514 ST.longfold = FALSE; /* char longer if folded => it's harder */
3517 /* fully traverse the TRIE; note the position of the
3518 shortest accept state and the wordnum of the longest
3521 while ( state && uc <= (U8*)PL_regeol ) {
3522 U32 base = trie->states[ state ].trans.base;
3526 wordnum = trie->states[ state ].wordnum;
3528 if (wordnum) { /* it's an accept state */
3531 /* record first match position */
3533 ST.firstpos = (U8*)locinput;
3538 ST.firstchars = charcount;
3541 if (!ST.nextword || wordnum < ST.nextword)
3542 ST.nextword = wordnum;
3543 ST.topword = wordnum;
3546 DEBUG_TRIE_EXECUTE_r({
3547 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3548 PerlIO_printf( Perl_debug_log,
3549 "%*s %sState: %4"UVxf" Accepted: %c ",
3550 2+depth * 2, "", PL_colors[4],
3551 (UV)state, (accepted ? 'Y' : 'N'));
3554 /* read a char and goto next state */
3557 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3558 uscan, len, uvc, charid, foldlen,
3565 base + charid - 1 - trie->uniquecharcount)) >= 0)
3567 && ((U32)offset < trie->lasttrans)
3568 && trie->trans[offset].check == state)
3570 state = trie->trans[offset].next;
3581 DEBUG_TRIE_EXECUTE_r(
3582 PerlIO_printf( Perl_debug_log,
3583 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3584 charid, uvc, (UV)state, PL_colors[5] );
3590 /* calculate total number of accept states */
3595 w = trie->wordinfo[w].prev;
3598 ST.accepted = accepted;
3602 PerlIO_printf( Perl_debug_log,
3603 "%*s %sgot %"IVdf" possible matches%s\n",
3604 REPORT_CODE_OFF + depth * 2, "",
3605 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3607 goto trie_first_try; /* jump into the fail handler */
3609 assert(0); /* NOTREACHED */
3611 case TRIE_next_fail: /* we failed - try next alternative */
3615 REGCP_UNWIND(ST.cp);
3616 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3618 if (!--ST.accepted) {
3620 PerlIO_printf( Perl_debug_log,
3621 "%*s %sTRIE failed...%s\n",
3622 REPORT_CODE_OFF+depth*2, "",
3629 /* Find next-highest word to process. Note that this code
3630 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3633 U16 const nextword = ST.nextword;
3634 reg_trie_wordinfo * const wordinfo
3635 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3636 for (word=ST.topword; word; word=wordinfo[word].prev) {
3637 if (word > nextword && (!min || word < min))
3650 ST.lastparen = rex->lastparen;
3651 ST.lastcloseparen = rex->lastcloseparen;
3655 /* find start char of end of current word */
3657 U32 chars; /* how many chars to skip */
3658 reg_trie_data * const trie
3659 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3661 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3663 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3668 /* the hard option - fold each char in turn and find
3669 * its folded length (which may be different */
3670 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3678 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3686 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3691 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3707 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
3708 ? ST.jump[ST.nextword]
3712 PerlIO_printf( Perl_debug_log,
3713 "%*s %sTRIE matched word #%d, continuing%s\n",
3714 REPORT_CODE_OFF+depth*2, "",
3721 if (ST.accepted > 1 || has_cutgroup) {
3722 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
3723 assert(0); /* NOTREACHED */
3725 /* only one choice left - just continue */
3727 AV *const trie_words
3728 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3729 SV ** const tmp = av_fetch( trie_words,
3731 SV *sv= tmp ? sv_newmortal() : NULL;
3733 PerlIO_printf( Perl_debug_log,
3734 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3735 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3737 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3738 PL_colors[0], PL_colors[1],
3739 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3741 : "not compiled under -Dr",
3745 locinput = (char*)uc;
3746 nextchr = UCHARAT(locinput);
3747 continue; /* execute rest of RE */
3748 assert(0); /* NOTREACHED */
3753 char *s = STRING(scan);
3755 if (utf8_target != UTF_PATTERN) {
3756 /* The target and the pattern have differing utf8ness. */
3758 const char * const e = s + ln;
3761 /* The target is utf8, the pattern is not utf8. */
3766 if (NATIVE_TO_UNI(*(U8*)s) !=
3767 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3775 /* The target is not utf8, the pattern is utf8. */
3780 if (NATIVE_TO_UNI(*((U8*)l)) !=
3781 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3789 nextchr = UCHARAT(locinput);
3792 /* The target and the pattern have the same utf8ness. */
3793 /* Inline the first character, for speed. */
3794 if (UCHARAT(s) != nextchr)
3796 if (PL_regeol - locinput < ln)
3798 if (ln > 1 && memNE(s, locinput, ln))
3801 nextchr = UCHARAT(locinput);
3806 const U8 * fold_array;
3808 U32 fold_utf8_flags;
3810 PL_reg_flags |= RF_tainted;
3811 folder = foldEQ_locale;
3812 fold_array = PL_fold_locale;
3813 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3817 case EXACTFU_TRICKYFOLD:
3819 folder = foldEQ_latin1;
3820 fold_array = PL_fold_latin1;
3821 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3825 folder = foldEQ_latin1;
3826 fold_array = PL_fold_latin1;
3827 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3832 fold_array = PL_fold;
3833 fold_utf8_flags = 0;
3839 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
3840 /* Either target or the pattern are utf8, or has the issue where
3841 * the fold lengths may differ. */
3842 const char * const l = locinput;
3843 char *e = PL_regeol;