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 #define RF_tainted 1 /* tainted information used? e.g. locale */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
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)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
128 /* Doesn't do an assert to verify that is correct */
129 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
130 if (!CAT2(PL_utf8_,class)) { bool throw_away; ENTER; save_re_context(); throw_away = CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END
132 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
133 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
134 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
136 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
137 LOAD_UTF8_CHARCLASS(X_begin, " "); \
138 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
139 /* These are utf8 constants, and not utf-ebcdic constants, so the \
140 * assert should likely and hopefully fail on an EBCDIC machine */ \
141 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
143 /* No asserts are done for these, in case called on an early \
144 * Unicode version in which they map to nothing */ \
145 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
146 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
147 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
148 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
149 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
150 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
151 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
153 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
155 /* The actual code for CCC_TRY, which uses several variables from the routine
156 * it's callable from. It is designed to be the bulk of a case statement.
157 * FUNC is the macro or function to call on non-utf8 targets that indicate if
158 * nextchr matches the class.
159 * UTF8_TEST is the whole test string to use for utf8 targets
160 * LOAD is what to use to test, and if not present to load in the swash for the
162 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
164 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
165 * utf8 and a variant, load the swash if necessary and test using the utf8
166 * test. Advance to the next character if test is ok, otherwise fail; If not
167 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
168 * fails, or advance to the next character */
170 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
171 if (locinput >= PL_regeol) { \
174 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
175 LOAD_UTF8_CHARCLASS(CLASS, STR); \
176 if (POS_OR_NEG (UTF8_TEST)) { \
179 locinput += PL_utf8skip[nextchr]; \
180 nextchr = UCHARAT(locinput); \
183 if (POS_OR_NEG (FUNC(nextchr))) { \
186 nextchr = UCHARAT(++locinput); \
189 /* Handle the non-locale cases for a character class and its complement. It
190 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
191 * This is because that code fails when the test succeeds, so we want to have
192 * the test fail so that the code succeeds. The swash is stored in a
193 * predictable PL_ place */
194 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
197 _CCC_TRY_CODE( !, FUNC, \
198 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
199 (U8*)locinput, TRUE)), \
202 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
203 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
204 (U8*)locinput, TRUE)), \
207 /* Generate the case statements for both locale and non-locale character
208 * classes in regmatch for classes that don't have special unicode semantics.
209 * Locales don't use an immediate swash, but an intermediary special locale
210 * function that is called on the pointer to the current place in the input
211 * string. That function will resolve to needing the same swash. One might
212 * think that because we don't know what the locale will match, we shouldn't
213 * check with the swash loading function that it loaded properly; ie, that we
214 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
215 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
217 #define CCC_TRY(NAME, NNAME, FUNC, \
218 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
219 NAMEA, NNAMEA, FUNCA, \
222 PL_reg_flags |= RF_tainted; \
223 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
225 PL_reg_flags |= RF_tainted; \
226 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
229 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
232 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
233 nextchr = UCHARAT(++locinput); \
236 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
240 locinput += PL_utf8skip[nextchr]; \
241 nextchr = UCHARAT(locinput); \
244 nextchr = UCHARAT(++locinput); \
247 /* Generate the non-locale cases */ \
248 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
250 /* This is like CCC_TRY, but has an extra set of parameters for generating case
251 * statements to handle separate Unicode semantics nodes */
252 #define CCC_TRY_U(NAME, NNAME, FUNC, \
253 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
254 NAMEU, NNAMEU, FUNCU, \
255 NAMEA, NNAMEA, FUNCA, \
257 CCC_TRY(NAME, NNAME, FUNC, \
258 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
259 NAMEA, NNAMEA, FUNCA, \
261 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
263 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
265 /* for use after a quantifier and before an EXACT-like node -- japhy */
266 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
268 * NOTE that *nothing* that affects backtracking should be in here, specifically
269 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
270 * node that is in between two EXACT like nodes when ascertaining what the required
271 * "follow" character is. This should probably be moved to regex compile time
272 * although it may be done at run time beause of the REF possibility - more
273 * investigation required. -- demerphq
275 #define JUMPABLE(rn) ( \
277 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
279 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
280 OP(rn) == PLUS || OP(rn) == MINMOD || \
282 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
284 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
286 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
289 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
290 we don't need this definition. */
291 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
292 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
293 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
296 /* ... so we use this as its faster. */
297 #define IS_TEXT(rn) ( OP(rn)==EXACT )
298 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
299 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
300 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
305 Search for mandatory following text node; for lookahead, the text must
306 follow but for lookbehind (rn->flags != 0) we skip to the next step.
308 #define FIND_NEXT_IMPT(rn) STMT_START { \
309 while (JUMPABLE(rn)) { \
310 const OPCODE type = OP(rn); \
311 if (type == SUSPEND || PL_regkind[type] == CURLY) \
312 rn = NEXTOPER(NEXTOPER(rn)); \
313 else if (type == PLUS) \
315 else if (type == IFMATCH) \
316 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
317 else rn += NEXT_OFF(rn); \
322 static void restore_pos(pTHX_ void *arg);
324 #define REGCP_PAREN_ELEMS 4
325 #define REGCP_OTHER_ELEMS 5
326 #define REGCP_FRAME_ELEMS 1
327 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
328 * are needed for the regexp context stack bookkeeping. */
331 S_regcppush(pTHX_ I32 parenfloor)
334 const int retval = PL_savestack_ix;
335 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
336 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
337 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
339 GET_RE_DEBUG_FLAGS_DECL;
341 if (paren_elems_to_push < 0)
342 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
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);
351 for (p = PL_regsize; p > parenfloor; p--) {
352 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
353 SSPUSHINT(PL_regoffs[p].end);
354 SSPUSHINT(PL_regoffs[p].start);
355 SSPUSHPTR(PL_reg_start_tmp[p]);
357 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
358 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
359 (UV)p, (IV)PL_regoffs[p].start,
360 (IV)(PL_reg_start_tmp[p] - PL_bostr),
361 (IV)PL_regoffs[p].end
364 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
365 SSPUSHPTR(PL_regoffs);
366 SSPUSHINT(PL_regsize);
367 SSPUSHINT(*PL_reglastparen);
368 SSPUSHINT(*PL_reglastcloseparen);
369 SSPUSHPTR(PL_reginput);
370 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
375 /* These are needed since we do not localize EVAL nodes: */
376 #define REGCP_SET(cp) \
378 PerlIO_printf(Perl_debug_log, \
379 " Setting an EVAL scope, savestack=%"IVdf"\n", \
380 (IV)PL_savestack_ix)); \
383 #define REGCP_UNWIND(cp) \
385 if (cp != PL_savestack_ix) \
386 PerlIO_printf(Perl_debug_log, \
387 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
388 (IV)(cp), (IV)PL_savestack_ix)); \
392 S_regcppop(pTHX_ const regexp *rex)
397 GET_RE_DEBUG_FLAGS_DECL;
399 PERL_ARGS_ASSERT_REGCPPOP;
401 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
403 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
404 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
405 input = (char *) SSPOPPTR;
406 *PL_reglastcloseparen = SSPOPINT;
407 *PL_reglastparen = SSPOPINT;
408 PL_regsize = SSPOPINT;
409 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
411 i -= REGCP_OTHER_ELEMS;
412 /* Now restore the parentheses context. */
413 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
415 U32 paren = (U32)SSPOPINT;
416 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
417 PL_regoffs[paren].start = SSPOPINT;
419 if (paren <= *PL_reglastparen)
420 PL_regoffs[paren].end = tmps;
422 PerlIO_printf(Perl_debug_log,
423 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
424 (UV)paren, (IV)PL_regoffs[paren].start,
425 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
426 (IV)PL_regoffs[paren].end,
427 (paren > *PL_reglastparen ? "(no)" : ""));
431 if (*PL_reglastparen + 1 <= rex->nparens) {
432 PerlIO_printf(Perl_debug_log,
433 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
434 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
438 /* It would seem that the similar code in regtry()
439 * already takes care of this, and in fact it is in
440 * a better location to since this code can #if 0-ed out
441 * but the code in regtry() is needed or otherwise tests
442 * requiring null fields (pat.t#187 and split.t#{13,14}
443 * (as of patchlevel 7877) will fail. Then again,
444 * this code seems to be necessary or otherwise
445 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
446 * --jhi updated by dapm */
447 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
449 PL_regoffs[i].start = -1;
450 PL_regoffs[i].end = -1;
456 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
459 * pregexec and friends
462 #ifndef PERL_IN_XSUB_RE
464 - pregexec - match a regexp against a string
467 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
468 char *strbeg, I32 minend, SV *screamer, U32 nosave)
469 /* strend: pointer to null at end of string */
470 /* strbeg: real beginning of string */
471 /* minend: end of match must be >=minend after stringarg. */
472 /* nosave: For optimizations. */
474 PERL_ARGS_ASSERT_PREGEXEC;
477 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
478 nosave ? 0 : REXEC_COPY_STR);
483 * Need to implement the following flags for reg_anch:
485 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
487 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
488 * INTUIT_AUTORITATIVE_ML
489 * INTUIT_ONCE_NOML - Intuit can match in one location only.
492 * Another flag for this function: SECOND_TIME (so that float substrs
493 * with giant delta may be not rechecked).
496 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
498 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
499 Otherwise, only SvCUR(sv) is used to get strbeg. */
501 /* XXXX We assume that strpos is strbeg unless sv. */
503 /* XXXX Some places assume that there is a fixed substring.
504 An update may be needed if optimizer marks as "INTUITable"
505 RExen without fixed substrings. Similarly, it is assumed that
506 lengths of all the strings are no more than minlen, thus they
507 cannot come from lookahead.
508 (Or minlen should take into account lookahead.)
509 NOTE: Some of this comment is not correct. minlen does now take account
510 of lookahead/behind. Further research is required. -- demerphq
514 /* A failure to find a constant substring means that there is no need to make
515 an expensive call to REx engine, thus we celebrate a failure. Similarly,
516 finding a substring too deep into the string means that less calls to
517 regtry() should be needed.
519 REx compiler's optimizer found 4 possible hints:
520 a) Anchored substring;
522 c) Whether we are anchored (beginning-of-line or \G);
523 d) First node (of those at offset 0) which may distinguish positions;
524 We use a)b)d) and multiline-part of c), and try to find a position in the
525 string which does not contradict any of them.
528 /* Most of decisions we do here should have been done at compile time.
529 The nodes of the REx which we used for the search should have been
530 deleted from the finite automaton. */
533 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
534 char *strend, const U32 flags, re_scream_pos_data *data)
537 struct regexp *const prog = (struct regexp *)SvANY(rx);
538 register I32 start_shift = 0;
539 /* Should be nonnegative! */
540 register I32 end_shift = 0;
545 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
547 register char *other_last = NULL; /* other substr checked before this */
548 char *check_at = NULL; /* check substr found at this pos */
549 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
550 RXi_GET_DECL(prog,progi);
552 const char * const i_strpos = strpos;
554 GET_RE_DEBUG_FLAGS_DECL;
556 PERL_ARGS_ASSERT_RE_INTUIT_START;
558 RX_MATCH_UTF8_set(rx,utf8_target);
561 PL_reg_flags |= RF_utf8;
564 debug_start_match(rx, utf8_target, strpos, strend,
565 sv ? "Guessing start of match in sv for"
566 : "Guessing start of match in string for");
569 /* CHR_DIST() would be more correct here but it makes things slow. */
570 if (prog->minlen > strend - strpos) {
571 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
572 "String too short... [re_intuit_start]\n"));
576 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
579 if (!prog->check_utf8 && prog->check_substr)
580 to_utf8_substr(prog);
581 check = prog->check_utf8;
583 if (!prog->check_substr && prog->check_utf8)
584 to_byte_substr(prog);
585 check = prog->check_substr;
587 if (check == &PL_sv_undef) {
588 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
589 "Non-utf8 string cannot match utf8 check string\n"));
592 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
593 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
594 || ( (prog->extflags & RXf_ANCH_BOL)
595 && !multiline ) ); /* Check after \n? */
598 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
599 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
600 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
602 && (strpos != strbeg)) {
603 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
606 if (prog->check_offset_min == prog->check_offset_max &&
607 !(prog->extflags & RXf_CANY_SEEN)) {
608 /* Substring at constant offset from beg-of-str... */
611 s = HOP3c(strpos, prog->check_offset_min, strend);
614 slen = SvCUR(check); /* >= 1 */
616 if ( strend - s > slen || strend - s < slen - 1
617 || (strend - s == slen && strend[-1] != '\n')) {
618 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
621 /* Now should match s[0..slen-2] */
623 if (slen && (*SvPVX_const(check) != *s
625 && memNE(SvPVX_const(check), s, slen)))) {
627 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
631 else if (*SvPVX_const(check) != *s
632 || ((slen = SvCUR(check)) > 1
633 && memNE(SvPVX_const(check), s, slen)))
636 goto success_at_start;
639 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
641 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
642 end_shift = prog->check_end_shift;
645 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
646 - (SvTAIL(check) != 0);
647 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
649 if (end_shift < eshift)
653 else { /* Can match at random position */
656 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
657 end_shift = prog->check_end_shift;
659 /* end shift should be non negative here */
662 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
664 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
665 (IV)end_shift, RX_PRECOMP(prog));
669 /* Find a possible match in the region s..strend by looking for
670 the "check" substring in the region corrected by start/end_shift. */
673 I32 srch_start_shift = start_shift;
674 I32 srch_end_shift = end_shift;
675 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
676 srch_end_shift -= ((strbeg - s) - srch_start_shift);
677 srch_start_shift = strbeg - s;
679 DEBUG_OPTIMISE_MORE_r({
680 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
681 (IV)prog->check_offset_min,
682 (IV)srch_start_shift,
684 (IV)prog->check_end_shift);
687 if (flags & REXEC_SCREAM) {
688 I32 p = -1; /* Internal iterator of scream. */
689 I32 * const pp = data ? data->scream_pos : &p;
691 if (PL_screamfirst[BmRARE(check)] >= 0
692 || ( BmRARE(check) == '\n'
693 && (BmPREVIOUS(check) == SvCUR(check) - 1)
695 s = screaminstr(sv, check,
696 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
699 /* we may be pointing at the wrong string */
700 if (s && RXp_MATCH_COPIED(prog))
701 s = strbeg + (s - SvPVX_const(sv));
703 *data->scream_olds = s;
708 if (prog->extflags & RXf_CANY_SEEN) {
709 start_point= (U8*)(s + srch_start_shift);
710 end_point= (U8*)(strend - srch_end_shift);
712 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
713 end_point= HOP3(strend, -srch_end_shift, strbeg);
715 DEBUG_OPTIMISE_MORE_r({
716 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
717 (int)(end_point - start_point),
718 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
722 s = fbm_instr( start_point, end_point,
723 check, multiline ? FBMrf_MULTILINE : 0);
726 /* Update the count-of-usability, remove useless subpatterns,
730 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
731 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
732 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
733 (s ? "Found" : "Did not find"),
734 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
735 ? "anchored" : "floating"),
738 (s ? " at offset " : "...\n") );
743 /* Finish the diagnostic message */
744 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
746 /* XXX dmq: first branch is for positive lookbehind...
747 Our check string is offset from the beginning of the pattern.
748 So we need to do any stclass tests offset forward from that
757 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
758 Start with the other substr.
759 XXXX no SCREAM optimization yet - and a very coarse implementation
760 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
761 *always* match. Probably should be marked during compile...
762 Probably it is right to do no SCREAM here...
765 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
766 : (prog->float_substr && prog->anchored_substr))
768 /* Take into account the "other" substring. */
769 /* XXXX May be hopelessly wrong for UTF... */
772 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
775 char * const last = HOP3c(s, -start_shift, strbeg);
777 char * const saved_s = s;
780 t = s - prog->check_offset_max;
781 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
783 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
788 t = HOP3c(t, prog->anchored_offset, strend);
789 if (t < other_last) /* These positions already checked */
791 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
794 /* XXXX It is not documented what units *_offsets are in.
795 We assume bytes, but this is clearly wrong.
796 Meaning this code needs to be carefully reviewed for errors.
800 /* On end-of-str: see comment below. */
801 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
802 if (must == &PL_sv_undef) {
804 DEBUG_r(must = prog->anchored_utf8); /* for debug */
809 HOP3(HOP3(last1, prog->anchored_offset, strend)
810 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
812 multiline ? FBMrf_MULTILINE : 0
815 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
816 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
817 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
818 (s ? "Found" : "Contradicts"),
819 quoted, RE_SV_TAIL(must));
824 if (last1 >= last2) {
825 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
826 ", giving up...\n"));
829 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
830 ", trying floating at offset %ld...\n",
831 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
832 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
833 s = HOP3c(last, 1, strend);
837 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
838 (long)(s - i_strpos)));
839 t = HOP3c(s, -prog->anchored_offset, strbeg);
840 other_last = HOP3c(s, 1, strend);
848 else { /* Take into account the floating substring. */
850 char * const saved_s = s;
853 t = HOP3c(s, -start_shift, strbeg);
855 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
856 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
857 last = HOP3c(t, prog->float_max_offset, strend);
858 s = HOP3c(t, prog->float_min_offset, strend);
861 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
862 must = utf8_target ? prog->float_utf8 : prog->float_substr;
863 /* fbm_instr() takes into account exact value of end-of-str
864 if the check is SvTAIL(ed). Since false positives are OK,
865 and end-of-str is not later than strend we are OK. */
866 if (must == &PL_sv_undef) {
868 DEBUG_r(must = prog->float_utf8); /* for debug message */
871 s = fbm_instr((unsigned char*)s,
872 (unsigned char*)last + SvCUR(must)
874 must, multiline ? FBMrf_MULTILINE : 0);
876 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
877 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
878 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
879 (s ? "Found" : "Contradicts"),
880 quoted, RE_SV_TAIL(must));
884 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
885 ", giving up...\n"));
888 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
889 ", trying anchored starting at offset %ld...\n",
890 (long)(saved_s + 1 - i_strpos)));
892 s = HOP3c(t, 1, strend);
896 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
897 (long)(s - i_strpos)));
898 other_last = s; /* Fix this later. --Hugo */
908 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
910 DEBUG_OPTIMISE_MORE_r(
911 PerlIO_printf(Perl_debug_log,
912 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
913 (IV)prog->check_offset_min,
914 (IV)prog->check_offset_max,
922 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
924 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
927 /* Fixed substring is found far enough so that the match
928 cannot start at strpos. */
930 if (ml_anch && t[-1] != '\n') {
931 /* Eventually fbm_*() should handle this, but often
932 anchored_offset is not 0, so this check will not be wasted. */
933 /* XXXX In the code below we prefer to look for "^" even in
934 presence of anchored substrings. And we search even
935 beyond the found float position. These pessimizations
936 are historical artefacts only. */
938 while (t < strend - prog->minlen) {
940 if (t < check_at - prog->check_offset_min) {
941 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
942 /* Since we moved from the found position,
943 we definitely contradict the found anchored
944 substr. Due to the above check we do not
945 contradict "check" substr.
946 Thus we can arrive here only if check substr
947 is float. Redo checking for "other"=="fixed".
950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
951 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
952 goto do_other_anchored;
954 /* We don't contradict the found floating substring. */
955 /* XXXX Why not check for STCLASS? */
957 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
958 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
961 /* Position contradicts check-string */
962 /* XXXX probably better to look for check-string
963 than for "\n", so one should lower the limit for t? */
964 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
965 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
966 other_last = strpos = s = t + 1;
971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
972 PL_colors[0], PL_colors[1]));
976 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
977 PL_colors[0], PL_colors[1]));
981 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
984 /* The found string does not prohibit matching at strpos,
985 - no optimization of calling REx engine can be performed,
986 unless it was an MBOL and we are not after MBOL,
987 or a future STCLASS check will fail this. */
989 /* Even in this situation we may use MBOL flag if strpos is offset
990 wrt the start of the string. */
991 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
992 && (strpos != strbeg) && strpos[-1] != '\n'
993 /* May be due to an implicit anchor of m{.*foo} */
994 && !(prog->intflags & PREGf_IMPLICIT))
999 DEBUG_EXECUTE_r( if (ml_anch)
1000 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1001 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1004 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1006 prog->check_utf8 /* Could be deleted already */
1007 && --BmUSEFUL(prog->check_utf8) < 0
1008 && (prog->check_utf8 == prog->float_utf8)
1010 prog->check_substr /* Could be deleted already */
1011 && --BmUSEFUL(prog->check_substr) < 0
1012 && (prog->check_substr == prog->float_substr)
1015 /* If flags & SOMETHING - do not do it many times on the same match */
1016 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1017 /* XXX Does the destruction order has to change with utf8_target? */
1018 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1019 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1020 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1021 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1022 check = NULL; /* abort */
1024 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1025 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1026 if (prog->intflags & PREGf_IMPLICIT)
1027 prog->extflags &= ~RXf_ANCH_MBOL;
1028 /* XXXX This is a remnant of the old implementation. It
1029 looks wasteful, since now INTUIT can use many
1030 other heuristics. */
1031 prog->extflags &= ~RXf_USE_INTUIT;
1032 /* XXXX What other flags might need to be cleared in this branch? */
1038 /* Last resort... */
1039 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1040 /* trie stclasses are too expensive to use here, we are better off to
1041 leave it to regmatch itself */
1042 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1043 /* minlen == 0 is possible if regstclass is \b or \B,
1044 and the fixed substr is ''$.
1045 Since minlen is already taken into account, s+1 is before strend;
1046 accidentally, minlen >= 1 guaranties no false positives at s + 1
1047 even for \b or \B. But (minlen? 1 : 0) below assumes that
1048 regstclass does not come from lookahead... */
1049 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1050 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1051 const U8* const str = (U8*)STRING(progi->regstclass);
1052 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1053 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1056 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1057 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1058 else if (prog->float_substr || prog->float_utf8)
1059 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1063 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1064 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1067 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1070 const char *what = NULL;
1072 if (endpos == strend) {
1073 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1074 "Could not match STCLASS...\n") );
1077 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1078 "This position contradicts STCLASS...\n") );
1079 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1081 /* Contradict one of substrings */
1082 if (prog->anchored_substr || prog->anchored_utf8) {
1083 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1084 DEBUG_EXECUTE_r( what = "anchored" );
1086 s = HOP3c(t, 1, strend);
1087 if (s + start_shift + end_shift > strend) {
1088 /* XXXX Should be taken into account earlier? */
1089 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1090 "Could not match STCLASS...\n") );
1095 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1096 "Looking for %s substr starting at offset %ld...\n",
1097 what, (long)(s + start_shift - i_strpos)) );
1100 /* Have both, check_string is floating */
1101 if (t + start_shift >= check_at) /* Contradicts floating=check */
1102 goto retry_floating_check;
1103 /* Recheck anchored substring, but not floating... */
1107 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1108 "Looking for anchored substr starting at offset %ld...\n",
1109 (long)(other_last - i_strpos)) );
1110 goto do_other_anchored;
1112 /* Another way we could have checked stclass at the
1113 current position only: */
1118 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1119 "Looking for /%s^%s/m starting at offset %ld...\n",
1120 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1123 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1125 /* Check is floating substring. */
1126 retry_floating_check:
1127 t = check_at - start_shift;
1128 DEBUG_EXECUTE_r( what = "floating" );
1129 goto hop_and_restart;
1132 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1133 "By STCLASS: moving %ld --> %ld\n",
1134 (long)(t - i_strpos), (long)(s - i_strpos))
1138 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1139 "Does not contradict STCLASS...\n");
1144 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1145 PL_colors[4], (check ? "Guessed" : "Giving up"),
1146 PL_colors[5], (long)(s - i_strpos)) );
1149 fail_finish: /* Substring not found */
1150 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1151 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1153 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1154 PL_colors[4], PL_colors[5]));
1158 #define DECL_TRIE_TYPE(scan) \
1159 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1160 trie_type = (scan->flags != EXACT) \
1161 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1162 : (utf8_target ? trie_utf8 : trie_plain)
1164 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1165 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1166 switch (trie_type) { \
1167 case trie_utf8_fold: \
1168 if ( foldlen>0 ) { \
1169 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1174 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1175 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1176 foldlen -= UNISKIP( uvc ); \
1177 uscan = foldbuf + UNISKIP( uvc ); \
1180 case trie_latin_utf8_fold: \
1181 if ( foldlen>0 ) { \
1182 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1188 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1189 foldlen -= UNISKIP( uvc ); \
1190 uscan = foldbuf + UNISKIP( uvc ); \
1194 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1201 charid = trie->charmap[ uvc ]; \
1205 if (widecharmap) { \
1206 SV** const svpp = hv_fetch(widecharmap, \
1207 (char*)&uvc, sizeof(UV), 0); \
1209 charid = (U16)SvIV(*svpp); \
1214 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1218 && (ln == 1 || folder(s, pat_string, ln)) \
1219 && (!reginfo || regtry(reginfo, &s)) ) \
1225 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1227 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1233 #define REXEC_FBC_SCAN(CoDe) \
1235 while (s < strend) { \
1241 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1242 REXEC_FBC_UTF8_SCAN( \
1244 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1253 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1256 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1265 #define REXEC_FBC_TRYIT \
1266 if ((!reginfo || regtry(reginfo, &s))) \
1269 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1270 if (utf8_target) { \
1271 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1274 REXEC_FBC_CLASS_SCAN(CoNd); \
1277 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1278 if (utf8_target) { \
1280 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1283 REXEC_FBC_CLASS_SCAN(CoNd); \
1286 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1287 PL_reg_flags |= RF_tainted; \
1288 if (utf8_target) { \
1289 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1292 REXEC_FBC_CLASS_SCAN(CoNd); \
1295 #define DUMP_EXEC_POS(li,s,doutf8) \
1296 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1299 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1300 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1301 tmp = TEST_NON_UTF8(tmp); \
1302 REXEC_FBC_UTF8_SCAN( \
1303 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1312 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1313 if (s == PL_bostr) { \
1317 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1318 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1321 LOAD_UTF8_CHARCLASS_ALNUM(); \
1322 REXEC_FBC_UTF8_SCAN( \
1323 if (tmp == ! (TeSt2_UtF8)) { \
1332 /* The only difference between the BOUND and NBOUND cases is that
1333 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1334 * NBOUND. This is accomplished by passing it in either the if or else clause,
1335 * with the other one being empty */
1336 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1337 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1339 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1340 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1342 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1343 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1345 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1346 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1349 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1350 * be passed in completely with the variable name being tested, which isn't
1351 * such a clean interface, but this is easier to read than it was before. We
1352 * are looking for the boundary (or non-boundary between a word and non-word
1353 * character. The utf8 and non-utf8 cases have the same logic, but the details
1354 * must be different. Find the "wordness" of the character just prior to this
1355 * one, and compare it with the wordness of this one. If they differ, we have
1356 * a boundary. At the beginning of the string, pretend that the previous
1357 * character was a new-line */
1358 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1359 if (utf8_target) { \
1362 else { /* Not utf8 */ \
1363 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1364 tmp = TEST_NON_UTF8(tmp); \
1366 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1375 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1378 /* We know what class REx starts with. Try to find this position... */
1379 /* if reginfo is NULL, its a dryrun */
1380 /* annoyingly all the vars in this routine have different names from their counterparts
1381 in regmatch. /grrr */
1384 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1385 const char *strend, regmatch_info *reginfo)
1388 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1389 char *pat_string; /* The pattern's exactish string */
1390 char *pat_end; /* ptr to end char of pat_string */
1391 re_fold_t folder; /* Function for computing non-utf8 folds */
1392 const U8 *fold_array; /* array for folding ords < 256 */
1395 register STRLEN uskip;
1399 register I32 tmp = 1; /* Scratch variable? */
1400 register const bool utf8_target = PL_reg_match_utf8;
1401 UV utf8_fold_flags = 0;
1402 RXi_GET_DECL(prog,progi);
1404 PERL_ARGS_ASSERT_FIND_BYCLASS;
1406 /* We know what class it must start with. */
1410 if (utf8_target || OP(c) == ANYOFV) {
1411 STRLEN inclasslen = strend - s;
1412 REXEC_FBC_UTF8_CLASS_SCAN(
1413 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1416 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1421 if (tmp && (!reginfo || regtry(reginfo, &s)))
1429 if (UTF_PATTERN || utf8_target) {
1430 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1431 goto do_exactf_utf8;
1433 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1434 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1435 goto do_exactf_non_utf8; /* isn't dealt with by these */
1438 if (UTF_PATTERN || utf8_target) {
1439 utf8_fold_flags = 0;
1440 goto do_exactf_utf8;
1442 fold_array = PL_fold_latin1;
1443 folder = foldEQ_latin1;
1444 /* XXX This uses the full utf8 fold because if the pattern contains
1445 * 'ss' it could match LATIN_SMALL_LETTER SHARP_S in the string.
1446 * There could be a new node type, say EXACTFU_SS, which is
1447 * generated by regcomp only if there is an 'ss', and then every
1448 * other case could goto do_exactf_non_utf8;*/
1449 goto do_exactf_utf8;
1452 if (UTF_PATTERN || utf8_target) {
1453 utf8_fold_flags = 0;
1454 goto do_exactf_utf8;
1456 fold_array = PL_fold;
1458 goto do_exactf_non_utf8;
1461 if (UTF_PATTERN || utf8_target) {
1462 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1463 goto do_exactf_utf8;
1465 fold_array = PL_fold_locale;
1466 folder = foldEQ_locale;
1470 do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */
1472 /* The idea in the non-utf8 EXACTF* cases is to first find the
1473 * first character of the EXACTF* node and then, if necessary,
1474 * case-insensitively compare the full text of the node. c1 is the
1475 * first character. c2 is its fold. This logic will not work for
1476 * Unicode semantics and the german sharp ss, which hence should
1477 * not be compiled into a node that gets here. */
1478 pat_string = STRING(c);
1479 ln = STR_LEN(c); /* length to match in octets/bytes */
1481 e = HOP3c(strend, -((I32)ln), s);
1483 if (!reginfo && e < s) {
1484 e = s; /* Due to minlen logic of intuit() */
1488 c2 = fold_array[c1];
1489 if (c1 == c2) { /* If char and fold are the same */
1490 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1493 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1499 /* If one of the operands is in utf8, we can't use the simpler
1500 * folding above, due to the fact that many different characters
1501 * can have the same fold, or portion of a fold, or different-
1503 pat_string = STRING(c);
1504 ln = STR_LEN(c); /* length to match in octets/bytes */
1505 pat_end = pat_string + ln;
1506 lnc = (UTF_PATTERN) /* length to match in characters */
1507 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1510 e = HOP3c(strend, -((I32)lnc), s);
1512 if (!reginfo && e < s) {
1513 e = s; /* Due to minlen logic of intuit() */
1517 char *my_strend= (char *)strend;
1518 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1519 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1520 && (!reginfo || regtry(reginfo, &s)) )
1528 PL_reg_flags |= RF_tainted;
1529 FBC_BOUND(isALNUM_LC,
1530 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1531 isALNUM_LC_utf8((U8*)s));
1534 PL_reg_flags |= RF_tainted;
1535 FBC_NBOUND(isALNUM_LC,
1536 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1537 isALNUM_LC_utf8((U8*)s));
1540 FBC_BOUND(isWORDCHAR,
1542 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1545 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1547 isWORDCHAR_A((U8*)s));
1550 FBC_NBOUND(isWORDCHAR,
1552 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1555 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1557 isWORDCHAR_A((U8*)s));
1560 FBC_BOUND(isWORDCHAR_L1,
1562 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1565 FBC_NBOUND(isWORDCHAR_L1,
1567 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1570 REXEC_FBC_CSCAN_TAINT(
1571 isALNUM_LC_utf8((U8*)s),
1576 REXEC_FBC_CSCAN_PRELOAD(
1577 LOAD_UTF8_CHARCLASS_ALNUM(),
1578 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1579 isWORDCHAR_L1((U8) *s)
1583 REXEC_FBC_CSCAN_PRELOAD(
1584 LOAD_UTF8_CHARCLASS_ALNUM(),
1585 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1590 /* Don't need to worry about utf8, as it can match only a single
1591 * byte invariant character */
1592 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1595 REXEC_FBC_CSCAN_PRELOAD(
1596 LOAD_UTF8_CHARCLASS_ALNUM(),
1597 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1598 ! isWORDCHAR_L1((U8) *s)
1602 REXEC_FBC_CSCAN_PRELOAD(
1603 LOAD_UTF8_CHARCLASS_ALNUM(),
1604 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1615 REXEC_FBC_CSCAN_TAINT(
1616 !isALNUM_LC_utf8((U8*)s),
1621 REXEC_FBC_CSCAN_PRELOAD(
1622 LOAD_UTF8_CHARCLASS_SPACE(),
1623 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1628 REXEC_FBC_CSCAN_PRELOAD(
1629 LOAD_UTF8_CHARCLASS_SPACE(),
1630 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1635 /* Don't need to worry about utf8, as it can match only a single
1636 * byte invariant character */
1637 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1640 REXEC_FBC_CSCAN_TAINT(
1641 isSPACE_LC_utf8((U8*)s),
1646 REXEC_FBC_CSCAN_PRELOAD(
1647 LOAD_UTF8_CHARCLASS_SPACE(),
1648 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1649 ! isSPACE_L1((U8) *s)
1653 REXEC_FBC_CSCAN_PRELOAD(
1654 LOAD_UTF8_CHARCLASS_SPACE(),
1655 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1666 REXEC_FBC_CSCAN_TAINT(
1667 !isSPACE_LC_utf8((U8*)s),
1672 REXEC_FBC_CSCAN_PRELOAD(
1673 LOAD_UTF8_CHARCLASS_DIGIT(),
1674 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1679 /* Don't need to worry about utf8, as it can match only a single
1680 * byte invariant character */
1681 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1684 REXEC_FBC_CSCAN_TAINT(
1685 isDIGIT_LC_utf8((U8*)s),
1690 REXEC_FBC_CSCAN_PRELOAD(
1691 LOAD_UTF8_CHARCLASS_DIGIT(),
1692 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1703 REXEC_FBC_CSCAN_TAINT(
1704 !isDIGIT_LC_utf8((U8*)s),
1711 is_LNBREAK_latin1(s)
1723 !is_VERTWS_latin1(s)
1729 is_HORIZWS_latin1(s)
1734 !is_HORIZWS_utf8(s),
1735 !is_HORIZWS_latin1(s)
1742 /* what trie are we using right now */
1744 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1746 = (reg_trie_data*)progi->data->data[ aho->trie ];
1747 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1749 const char *last_start = strend - trie->minlen;
1751 const char *real_start = s;
1753 STRLEN maxlen = trie->maxlen;
1755 U8 **points; /* map of where we were in the input string
1756 when reading a given char. For ASCII this
1757 is unnecessary overhead as the relationship
1758 is always 1:1, but for Unicode, especially
1759 case folded Unicode this is not true. */
1760 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1764 GET_RE_DEBUG_FLAGS_DECL;
1766 /* We can't just allocate points here. We need to wrap it in
1767 * an SV so it gets freed properly if there is a croak while
1768 * running the match */
1771 sv_points=newSV(maxlen * sizeof(U8 *));
1772 SvCUR_set(sv_points,
1773 maxlen * sizeof(U8 *));
1774 SvPOK_on(sv_points);
1775 sv_2mortal(sv_points);
1776 points=(U8**)SvPV_nolen(sv_points );
1777 if ( trie_type != trie_utf8_fold
1778 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1781 bitmap=(U8*)trie->bitmap;
1783 bitmap=(U8*)ANYOF_BITMAP(c);
1785 /* this is the Aho-Corasick algorithm modified a touch
1786 to include special handling for long "unknown char"
1787 sequences. The basic idea being that we use AC as long
1788 as we are dealing with a possible matching char, when
1789 we encounter an unknown char (and we have not encountered
1790 an accepting state) we scan forward until we find a legal
1792 AC matching is basically that of trie matching, except
1793 that when we encounter a failing transition, we fall back
1794 to the current states "fail state", and try the current char
1795 again, a process we repeat until we reach the root state,
1796 state 1, or a legal transition. If we fail on the root state
1797 then we can either terminate if we have reached an accepting
1798 state previously, or restart the entire process from the beginning
1802 while (s <= last_start) {
1803 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1811 U8 *uscan = (U8*)NULL;
1812 U8 *leftmost = NULL;
1814 U32 accepted_word= 0;
1818 while ( state && uc <= (U8*)strend ) {
1820 U32 word = aho->states[ state ].wordnum;
1824 DEBUG_TRIE_EXECUTE_r(
1825 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1826 dump_exec_pos( (char *)uc, c, strend, real_start,
1827 (char *)uc, utf8_target );
1828 PerlIO_printf( Perl_debug_log,
1829 " Scanning for legal start char...\n");
1833 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1837 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1843 if (uc >(U8*)last_start) break;
1847 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1848 if (!leftmost || lpos < leftmost) {
1849 DEBUG_r(accepted_word=word);
1855 points[pointpos++ % maxlen]= uc;
1856 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1857 uscan, len, uvc, charid, foldlen,
1859 DEBUG_TRIE_EXECUTE_r({
1860 dump_exec_pos( (char *)uc, c, strend, real_start,
1862 PerlIO_printf(Perl_debug_log,
1863 " Charid:%3u CP:%4"UVxf" ",
1869 word = aho->states[ state ].wordnum;
1871 base = aho->states[ state ].trans.base;
1873 DEBUG_TRIE_EXECUTE_r({
1875 dump_exec_pos( (char *)uc, c, strend, real_start,
1877 PerlIO_printf( Perl_debug_log,
1878 "%sState: %4"UVxf", word=%"UVxf,
1879 failed ? " Fail transition to " : "",
1880 (UV)state, (UV)word);
1886 ( ((offset = base + charid
1887 - 1 - trie->uniquecharcount)) >= 0)
1888 && ((U32)offset < trie->lasttrans)
1889 && trie->trans[offset].check == state
1890 && (tmp=trie->trans[offset].next))
1892 DEBUG_TRIE_EXECUTE_r(
1893 PerlIO_printf( Perl_debug_log," - legal\n"));
1898 DEBUG_TRIE_EXECUTE_r(
1899 PerlIO_printf( Perl_debug_log," - fail\n"));
1901 state = aho->fail[state];
1905 /* we must be accepting here */
1906 DEBUG_TRIE_EXECUTE_r(
1907 PerlIO_printf( Perl_debug_log," - accepting\n"));
1916 if (!state) state = 1;
1919 if ( aho->states[ state ].wordnum ) {
1920 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1921 if (!leftmost || lpos < leftmost) {
1922 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1927 s = (char*)leftmost;
1928 DEBUG_TRIE_EXECUTE_r({
1930 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1931 (UV)accepted_word, (IV)(s - real_start)
1934 if (!reginfo || regtry(reginfo, &s)) {
1940 DEBUG_TRIE_EXECUTE_r({
1941 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1944 DEBUG_TRIE_EXECUTE_r(
1945 PerlIO_printf( Perl_debug_log,"No match.\n"));
1954 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1964 - regexec_flags - match a regexp against a string
1967 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1968 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1969 /* strend: pointer to null at end of string */
1970 /* strbeg: real beginning of string */
1971 /* minend: end of match must be >=minend after stringarg. */
1972 /* data: May be used for some additional optimizations.
1973 Currently its only used, with a U32 cast, for transmitting
1974 the ganch offset when doing a /g match. This will change */
1975 /* nosave: For optimizations. */
1978 struct regexp *const prog = (struct regexp *)SvANY(rx);
1979 /*register*/ char *s;
1980 register regnode *c;
1981 /*register*/ char *startpos = stringarg;
1982 I32 minlen; /* must match at least this many chars */
1983 I32 dontbother = 0; /* how many characters not to try at end */
1984 I32 end_shift = 0; /* Same for the end. */ /* CC */
1985 I32 scream_pos = -1; /* Internal iterator of scream. */
1986 char *scream_olds = NULL;
1987 const bool utf8_target = cBOOL(DO_UTF8(sv));
1989 RXi_GET_DECL(prog,progi);
1990 regmatch_info reginfo; /* create some info to pass to regtry etc */
1991 regexp_paren_pair *swap = NULL;
1992 GET_RE_DEBUG_FLAGS_DECL;
1994 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1995 PERL_UNUSED_ARG(data);
1997 /* Be paranoid... */
1998 if (prog == NULL || startpos == NULL) {
1999 Perl_croak(aTHX_ "NULL regexp parameter");
2003 multiline = prog->extflags & RXf_PMf_MULTILINE;
2004 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2006 RX_MATCH_UTF8_set(rx, utf8_target);
2008 debug_start_match(rx, utf8_target, startpos, strend,
2012 minlen = prog->minlen;
2014 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2015 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2016 "String too short [regexec_flags]...\n"));
2021 /* Check validity of program. */
2022 if (UCHARAT(progi->program) != REG_MAGIC) {
2023 Perl_croak(aTHX_ "corrupted regexp program");
2027 PL_reg_eval_set = 0;
2031 PL_reg_flags |= RF_utf8;
2033 /* Mark beginning of line for ^ and lookbehind. */
2034 reginfo.bol = startpos; /* XXX not used ??? */
2038 /* Mark end of line for $ (and such) */
2041 /* see how far we have to get to not match where we matched before */
2042 reginfo.till = startpos+minend;
2044 /* If there is a "must appear" string, look for it. */
2047 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2049 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2050 reginfo.ganch = startpos + prog->gofs;
2051 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2052 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2053 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2055 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2056 && mg->mg_len >= 0) {
2057 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2058 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2059 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2061 if (prog->extflags & RXf_ANCH_GPOS) {
2062 if (s > reginfo.ganch)
2064 s = reginfo.ganch - prog->gofs;
2065 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2066 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2072 reginfo.ganch = strbeg + PTR2UV(data);
2073 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2074 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2076 } else { /* pos() not defined */
2077 reginfo.ganch = strbeg;
2078 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2079 "GPOS: reginfo.ganch = strbeg\n"));
2082 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2083 /* We have to be careful. If the previous successful match
2084 was from this regex we don't want a subsequent partially
2085 successful match to clobber the old results.
2086 So when we detect this possibility we add a swap buffer
2087 to the re, and switch the buffer each match. If we fail
2088 we switch it back, otherwise we leave it swapped.
2091 /* do we need a save destructor here for eval dies? */
2092 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2094 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2095 re_scream_pos_data d;
2097 d.scream_olds = &scream_olds;
2098 d.scream_pos = &scream_pos;
2099 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2101 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2102 goto phooey; /* not present */
2108 /* Simplest case: anchored match need be tried only once. */
2109 /* [unless only anchor is BOL and multiline is set] */
2110 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2111 if (s == startpos && regtry(®info, &startpos))
2113 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2114 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2119 dontbother = minlen - 1;
2120 end = HOP3c(strend, -dontbother, strbeg) - 1;
2121 /* for multiline we only have to try after newlines */
2122 if (prog->check_substr || prog->check_utf8) {
2123 /* because of the goto we can not easily reuse the macros for bifurcating the
2124 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2127 goto after_try_utf8;
2129 if (regtry(®info, &s)) {
2136 if (prog->extflags & RXf_USE_INTUIT) {
2137 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2146 } /* end search for check string in unicode */
2148 if (s == startpos) {
2149 goto after_try_latin;
2152 if (regtry(®info, &s)) {
2159 if (prog->extflags & RXf_USE_INTUIT) {
2160 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2169 } /* end search for check string in latin*/
2170 } /* end search for check string */
2171 else { /* search for newline */
2173 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2176 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2178 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2179 if (regtry(®info, &s))
2183 } /* end search for newline */
2184 } /* end anchored/multiline check string search */
2186 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2188 /* the warning about reginfo.ganch being used without initialization
2189 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2190 and we only enter this block when the same bit is set. */
2191 char *tmp_s = reginfo.ganch - prog->gofs;
2193 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2198 /* Messy cases: unanchored match. */
2199 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2200 /* we have /x+whatever/ */
2201 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2206 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2207 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2208 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2213 DEBUG_EXECUTE_r( did_match = 1 );
2214 if (regtry(®info, &s)) goto got_it;
2216 while (s < strend && *s == ch)
2224 DEBUG_EXECUTE_r( did_match = 1 );
2225 if (regtry(®info, &s)) goto got_it;
2227 while (s < strend && *s == ch)
2232 DEBUG_EXECUTE_r(if (!did_match)
2233 PerlIO_printf(Perl_debug_log,
2234 "Did not find anchored character...\n")
2237 else if (prog->anchored_substr != NULL
2238 || prog->anchored_utf8 != NULL
2239 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2240 && prog->float_max_offset < strend - s)) {
2245 char *last1; /* Last position checked before */
2249 if (prog->anchored_substr || prog->anchored_utf8) {
2250 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2251 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2252 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2253 back_max = back_min = prog->anchored_offset;
2255 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2256 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2257 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2258 back_max = prog->float_max_offset;
2259 back_min = prog->float_min_offset;
2263 if (must == &PL_sv_undef)
2264 /* could not downgrade utf8 check substring, so must fail */
2270 last = HOP3c(strend, /* Cannot start after this */
2271 -(I32)(CHR_SVLEN(must)
2272 - (SvTAIL(must) != 0) + back_min), strbeg);
2275 last1 = HOPc(s, -1);
2277 last1 = s - 1; /* bogus */
2279 /* XXXX check_substr already used to find "s", can optimize if
2280 check_substr==must. */
2282 dontbother = end_shift;
2283 strend = HOPc(strend, -dontbother);
2284 while ( (s <= last) &&
2285 ((flags & REXEC_SCREAM)
2286 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2287 end_shift, &scream_pos, 0))
2288 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2289 (unsigned char*)strend, must,
2290 multiline ? FBMrf_MULTILINE : 0))) ) {
2291 /* we may be pointing at the wrong string */
2292 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2293 s = strbeg + (s - SvPVX_const(sv));
2294 DEBUG_EXECUTE_r( did_match = 1 );
2295 if (HOPc(s, -back_max) > last1) {
2296 last1 = HOPc(s, -back_min);
2297 s = HOPc(s, -back_max);
2300 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2302 last1 = HOPc(s, -back_min);
2306 while (s <= last1) {
2307 if (regtry(®info, &s))
2313 while (s <= last1) {
2314 if (regtry(®info, &s))
2320 DEBUG_EXECUTE_r(if (!did_match) {
2321 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2322 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2323 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2324 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2325 ? "anchored" : "floating"),
2326 quoted, RE_SV_TAIL(must));
2330 else if ( (c = progi->regstclass) ) {
2332 const OPCODE op = OP(progi->regstclass);
2333 /* don't bother with what can't match */
2334 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2335 strend = HOPc(strend, -(minlen - 1));
2338 SV * const prop = sv_newmortal();
2339 regprop(prog, prop, c);
2341 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2343 PerlIO_printf(Perl_debug_log,
2344 "Matching stclass %.*s against %s (%d bytes)\n",
2345 (int)SvCUR(prop), SvPVX_const(prop),
2346 quoted, (int)(strend - s));
2349 if (find_byclass(prog, c, s, strend, ®info))
2351 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2355 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2360 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2361 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2362 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2364 if (flags & REXEC_SCREAM) {
2365 last = screaminstr(sv, float_real, s - strbeg,
2366 end_shift, &scream_pos, 1); /* last one */
2368 last = scream_olds; /* Only one occurrence. */
2369 /* we may be pointing at the wrong string */
2370 else if (RXp_MATCH_COPIED(prog))
2371 s = strbeg + (s - SvPVX_const(sv));
2375 const char * const little = SvPV_const(float_real, len);
2377 if (SvTAIL(float_real)) {
2378 if (memEQ(strend - len + 1, little, len - 1))
2379 last = strend - len + 1;
2380 else if (!multiline)
2381 last = memEQ(strend - len, little, len)
2382 ? strend - len : NULL;
2388 last = rninstr(s, strend, little, little + len);
2390 last = strend; /* matching "$" */
2395 PerlIO_printf(Perl_debug_log,
2396 "%sCan't trim the tail, match fails (should not happen)%s\n",
2397 PL_colors[4], PL_colors[5]));
2398 goto phooey; /* Should not happen! */
2400 dontbother = strend - last + prog->float_min_offset;
2402 if (minlen && (dontbother < minlen))
2403 dontbother = minlen - 1;
2404 strend -= dontbother; /* this one's always in bytes! */
2405 /* We don't know much -- general case. */
2408 if (regtry(®info, &s))
2417 if (regtry(®info, &s))
2419 } while (s++ < strend);
2428 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2430 if (PL_reg_eval_set)
2431 restore_pos(aTHX_ prog);
2432 if (RXp_PAREN_NAMES(prog))
2433 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2435 /* make sure $`, $&, $', and $digit will work later */
2436 if ( !(flags & REXEC_NOT_FIRST) ) {
2437 RX_MATCH_COPY_FREE(rx);
2438 if (flags & REXEC_COPY_STR) {
2439 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2440 #ifdef PERL_OLD_COPY_ON_WRITE
2442 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2444 PerlIO_printf(Perl_debug_log,
2445 "Copy on write: regexp capture, type %d\n",
2448 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2449 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2450 assert (SvPOKp(prog->saved_copy));
2454 RX_MATCH_COPIED_on(rx);
2455 s = savepvn(strbeg, i);
2461 prog->subbeg = strbeg;
2462 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2469 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2470 PL_colors[4], PL_colors[5]));
2471 if (PL_reg_eval_set)
2472 restore_pos(aTHX_ prog);
2474 /* we failed :-( roll it back */
2475 Safefree(prog->offs);
2484 - regtry - try match at specific point
2486 STATIC I32 /* 0 failure, 1 success */
2487 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2491 REGEXP *const rx = reginfo->prog;
2492 regexp *const prog = (struct regexp *)SvANY(rx);
2493 RXi_GET_DECL(prog,progi);
2494 GET_RE_DEBUG_FLAGS_DECL;
2496 PERL_ARGS_ASSERT_REGTRY;
2498 reginfo->cutpoint=NULL;
2500 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2503 PL_reg_eval_set = RS_init;
2504 DEBUG_EXECUTE_r(DEBUG_s(
2505 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2506 (IV)(PL_stack_sp - PL_stack_base));
2509 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2510 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2512 /* Apparently this is not needed, judging by wantarray. */
2513 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2514 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2517 /* Make $_ available to executed code. */
2518 if (reginfo->sv != DEFSV) {
2520 DEFSV_set(reginfo->sv);
2523 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2524 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2525 /* prepare for quick setting of pos */
2526 #ifdef PERL_OLD_COPY_ON_WRITE
2527 if (SvIsCOW(reginfo->sv))
2528 sv_force_normal_flags(reginfo->sv, 0);
2530 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2531 &PL_vtbl_mglob, NULL, 0);
2535 PL_reg_oldpos = mg->mg_len;
2536 SAVEDESTRUCTOR_X(restore_pos, prog);
2538 if (!PL_reg_curpm) {
2539 Newxz(PL_reg_curpm, 1, PMOP);
2542 SV* const repointer = &PL_sv_undef;
2543 /* this regexp is also owned by the new PL_reg_curpm, which
2544 will try to free it. */
2545 av_push(PL_regex_padav, repointer);
2546 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2547 PL_regex_pad = AvARRAY(PL_regex_padav);
2552 /* It seems that non-ithreads works both with and without this code.
2553 So for efficiency reasons it seems best not to have the code
2554 compiled when it is not needed. */
2555 /* This is safe against NULLs: */
2556 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2557 /* PM_reg_curpm owns a reference to this regexp. */
2558 (void)ReREFCNT_inc(rx);
2560 PM_SETRE(PL_reg_curpm, rx);
2561 PL_reg_oldcurpm = PL_curpm;
2562 PL_curpm = PL_reg_curpm;
2563 if (RXp_MATCH_COPIED(prog)) {
2564 /* Here is a serious problem: we cannot rewrite subbeg,
2565 since it may be needed if this match fails. Thus
2566 $` inside (?{}) could fail... */
2567 PL_reg_oldsaved = prog->subbeg;
2568 PL_reg_oldsavedlen = prog->sublen;
2569 #ifdef PERL_OLD_COPY_ON_WRITE
2570 PL_nrs = prog->saved_copy;
2572 RXp_MATCH_COPIED_off(prog);
2575 PL_reg_oldsaved = NULL;
2576 prog->subbeg = PL_bostr;
2577 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2579 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2580 prog->offs[0].start = *startpos - PL_bostr;
2581 PL_reginput = *startpos;
2582 PL_reglastparen = &prog->lastparen;
2583 PL_reglastcloseparen = &prog->lastcloseparen;
2584 prog->lastparen = 0;
2585 prog->lastcloseparen = 0;
2587 PL_regoffs = prog->offs;
2588 if (PL_reg_start_tmpl <= prog->nparens) {
2589 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2590 if(PL_reg_start_tmp)
2591 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2593 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2596 /* XXXX What this code is doing here?!!! There should be no need
2597 to do this again and again, PL_reglastparen should take care of
2600 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2601 * Actually, the code in regcppop() (which Ilya may be meaning by
2602 * PL_reglastparen), is not needed at all by the test suite
2603 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2604 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2605 * Meanwhile, this code *is* needed for the
2606 * above-mentioned test suite tests to succeed. The common theme
2607 * on those tests seems to be returning null fields from matches.
2608 * --jhi updated by dapm */
2610 if (prog->nparens) {
2611 regexp_paren_pair *pp = PL_regoffs;
2613 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2621 if (regmatch(reginfo, progi->program + 1)) {
2622 PL_regoffs[0].end = PL_reginput - PL_bostr;
2625 if (reginfo->cutpoint)
2626 *startpos= reginfo->cutpoint;
2627 REGCP_UNWIND(lastcp);
2632 #define sayYES goto yes
2633 #define sayNO goto no
2634 #define sayNO_SILENT goto no_silent
2636 /* we dont use STMT_START/END here because it leads to
2637 "unreachable code" warnings, which are bogus, but distracting. */
2638 #define CACHEsayNO \
2639 if (ST.cache_mask) \
2640 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2643 /* this is used to determine how far from the left messages like
2644 'failed...' are printed. It should be set such that messages
2645 are inline with the regop output that created them.
2647 #define REPORT_CODE_OFF 32
2650 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2651 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2653 #define SLAB_FIRST(s) (&(s)->states[0])
2654 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2656 /* grab a new slab and return the first slot in it */
2658 STATIC regmatch_state *
2661 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2664 regmatch_slab *s = PL_regmatch_slab->next;
2666 Newx(s, 1, regmatch_slab);
2667 s->prev = PL_regmatch_slab;
2669 PL_regmatch_slab->next = s;
2671 PL_regmatch_slab = s;
2672 return SLAB_FIRST(s);
2676 /* push a new state then goto it */
2678 #define PUSH_STATE_GOTO(state, node) \
2680 st->resume_state = state; \
2683 /* push a new state with success backtracking, then goto it */
2685 #define PUSH_YES_STATE_GOTO(state, node) \
2687 st->resume_state = state; \
2688 goto push_yes_state;
2694 regmatch() - main matching routine
2696 This is basically one big switch statement in a loop. We execute an op,
2697 set 'next' to point the next op, and continue. If we come to a point which
2698 we may need to backtrack to on failure such as (A|B|C), we push a
2699 backtrack state onto the backtrack stack. On failure, we pop the top
2700 state, and re-enter the loop at the state indicated. If there are no more
2701 states to pop, we return failure.
2703 Sometimes we also need to backtrack on success; for example /A+/, where
2704 after successfully matching one A, we need to go back and try to
2705 match another one; similarly for lookahead assertions: if the assertion
2706 completes successfully, we backtrack to the state just before the assertion
2707 and then carry on. In these cases, the pushed state is marked as
2708 'backtrack on success too'. This marking is in fact done by a chain of
2709 pointers, each pointing to the previous 'yes' state. On success, we pop to
2710 the nearest yes state, discarding any intermediate failure-only states.
2711 Sometimes a yes state is pushed just to force some cleanup code to be
2712 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2713 it to free the inner regex.
2715 Note that failure backtracking rewinds the cursor position, while
2716 success backtracking leaves it alone.
2718 A pattern is complete when the END op is executed, while a subpattern
2719 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2720 ops trigger the "pop to last yes state if any, otherwise return true"
2723 A common convention in this function is to use A and B to refer to the two
2724 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2725 the subpattern to be matched possibly multiple times, while B is the entire
2726 rest of the pattern. Variable and state names reflect this convention.
2728 The states in the main switch are the union of ops and failure/success of
2729 substates associated with with that op. For example, IFMATCH is the op
2730 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2731 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2732 successfully matched A and IFMATCH_A_fail is a state saying that we have
2733 just failed to match A. Resume states always come in pairs. The backtrack
2734 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2735 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2736 on success or failure.
2738 The struct that holds a backtracking state is actually a big union, with
2739 one variant for each major type of op. The variable st points to the
2740 top-most backtrack struct. To make the code clearer, within each
2741 block of code we #define ST to alias the relevant union.
2743 Here's a concrete example of a (vastly oversimplified) IFMATCH
2749 #define ST st->u.ifmatch
2751 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2752 ST.foo = ...; // some state we wish to save
2754 // push a yes backtrack state with a resume value of
2755 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2757 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2760 case IFMATCH_A: // we have successfully executed A; now continue with B
2762 bar = ST.foo; // do something with the preserved value
2765 case IFMATCH_A_fail: // A failed, so the assertion failed
2766 ...; // do some housekeeping, then ...
2767 sayNO; // propagate the failure
2774 For any old-timers reading this who are familiar with the old recursive
2775 approach, the code above is equivalent to:
2777 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2786 ...; // do some housekeeping, then ...
2787 sayNO; // propagate the failure
2790 The topmost backtrack state, pointed to by st, is usually free. If you
2791 want to claim it, populate any ST.foo fields in it with values you wish to
2792 save, then do one of
2794 PUSH_STATE_GOTO(resume_state, node);
2795 PUSH_YES_STATE_GOTO(resume_state, node);
2797 which sets that backtrack state's resume value to 'resume_state', pushes a
2798 new free entry to the top of the backtrack stack, then goes to 'node'.
2799 On backtracking, the free slot is popped, and the saved state becomes the
2800 new free state. An ST.foo field in this new top state can be temporarily
2801 accessed to retrieve values, but once the main loop is re-entered, it
2802 becomes available for reuse.
2804 Note that the depth of the backtrack stack constantly increases during the
2805 left-to-right execution of the pattern, rather than going up and down with
2806 the pattern nesting. For example the stack is at its maximum at Z at the
2807 end of the pattern, rather than at X in the following:
2809 /(((X)+)+)+....(Y)+....Z/
2811 The only exceptions to this are lookahead/behind assertions and the cut,
2812 (?>A), which pop all the backtrack states associated with A before
2815 Backtrack state structs are allocated in slabs of about 4K in size.
2816 PL_regmatch_state and st always point to the currently active state,
2817 and PL_regmatch_slab points to the slab currently containing
2818 PL_regmatch_state. The first time regmatch() is called, the first slab is
2819 allocated, and is never freed until interpreter destruction. When the slab
2820 is full, a new one is allocated and chained to the end. At exit from
2821 regmatch(), slabs allocated since entry are freed.
2826 #define DEBUG_STATE_pp(pp) \
2828 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2829 PerlIO_printf(Perl_debug_log, \
2830 " %*s"pp" %s%s%s%s%s\n", \
2832 PL_reg_name[st->resume_state], \
2833 ((st==yes_state||st==mark_state) ? "[" : ""), \
2834 ((st==yes_state) ? "Y" : ""), \
2835 ((st==mark_state) ? "M" : ""), \
2836 ((st==yes_state||st==mark_state) ? "]" : "") \
2841 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2846 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2847 const char *start, const char *end, const char *blurb)
2849 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2851 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2856 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2857 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2859 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2860 start, end - start, 60);
2862 PerlIO_printf(Perl_debug_log,
2863 "%s%s REx%s %s against %s\n",
2864 PL_colors[4], blurb, PL_colors[5], s0, s1);
2866 if (utf8_target||utf8_pat)
2867 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2868 utf8_pat ? "pattern" : "",
2869 utf8_pat && utf8_target ? " and " : "",
2870 utf8_target ? "string" : ""
2876 S_dump_exec_pos(pTHX_ const char *locinput,
2877 const regnode *scan,
2878 const char *loc_regeol,
2879 const char *loc_bostr,
2880 const char *loc_reg_starttry,
2881 const bool utf8_target)
2883 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2884 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2885 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2886 /* The part of the string before starttry has one color
2887 (pref0_len chars), between starttry and current
2888 position another one (pref_len - pref0_len chars),
2889 after the current position the third one.
2890 We assume that pref0_len <= pref_len, otherwise we
2891 decrease pref0_len. */
2892 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2893 ? (5 + taill) - l : locinput - loc_bostr;
2896 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2898 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2900 pref0_len = pref_len - (locinput - loc_reg_starttry);
2901 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2902 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2903 ? (5 + taill) - pref_len : loc_regeol - locinput);
2904 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2908 if (pref0_len > pref_len)
2909 pref0_len = pref_len;
2911 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2913 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2914 (locinput - pref_len),pref0_len, 60, 4, 5);
2916 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2917 (locinput - pref_len + pref0_len),
2918 pref_len - pref0_len, 60, 2, 3);
2920 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2921 locinput, loc_regeol - locinput, 10, 0, 1);
2923 const STRLEN tlen=len0+len1+len2;
2924 PerlIO_printf(Perl_debug_log,
2925 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2926 (IV)(locinput - loc_bostr),
2929 (docolor ? "" : "> <"),
2931 (int)(tlen > 19 ? 0 : 19 - tlen),
2938 /* reg_check_named_buff_matched()
2939 * Checks to see if a named buffer has matched. The data array of
2940 * buffer numbers corresponding to the buffer is expected to reside
2941 * in the regexp->data->data array in the slot stored in the ARG() of
2942 * node involved. Note that this routine doesn't actually care about the
2943 * name, that information is not preserved from compilation to execution.
2944 * Returns the index of the leftmost defined buffer with the given name
2945 * or 0 if non of the buffers matched.
2948 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2951 RXi_GET_DECL(rex,rexi);
2952 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2953 I32 *nums=(I32*)SvPVX(sv_dat);
2955 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2957 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2958 if ((I32)*PL_reglastparen >= nums[n] &&
2959 PL_regoffs[nums[n]].end != -1)
2968 /* free all slabs above current one - called during LEAVE_SCOPE */
2971 S_clear_backtrack_stack(pTHX_ void *p)
2973 regmatch_slab *s = PL_regmatch_slab->next;
2978 PL_regmatch_slab->next = NULL;
2980 regmatch_slab * const osl = s;
2987 #define SETREX(Re1,Re2) \
2988 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2991 STATIC I32 /* 0 failure, 1 success */
2992 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2994 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2998 register const bool utf8_target = PL_reg_match_utf8;
2999 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3000 REGEXP *rex_sv = reginfo->prog;
3001 regexp *rex = (struct regexp *)SvANY(rex_sv);
3002 RXi_GET_DECL(rex,rexi);
3004 /* the current state. This is a cached copy of PL_regmatch_state */
3005 register regmatch_state *st;
3006 /* cache heavy used fields of st in registers */
3007 register regnode *scan;
3008 register regnode *next;
3009 register U32 n = 0; /* general value; init to avoid compiler warning */
3010 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3011 register char *locinput = PL_reginput;
3012 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3014 bool result = 0; /* return value of S_regmatch */
3015 int depth = 0; /* depth of backtrack stack */
3016 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3017 const U32 max_nochange_depth =
3018 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3019 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3020 regmatch_state *yes_state = NULL; /* state to pop to on success of
3022 /* mark_state piggy backs on the yes_state logic so that when we unwind
3023 the stack on success we can update the mark_state as we go */
3024 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3025 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3026 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3028 bool no_final = 0; /* prevent failure from backtracking? */
3029 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3030 char *startpoint = PL_reginput;
3031 SV *popmark = NULL; /* are we looking for a mark? */
3032 SV *sv_commit = NULL; /* last mark name seen in failure */
3033 SV *sv_yes_mark = NULL; /* last mark name we have seen
3034 during a successful match */
3035 U32 lastopen = 0; /* last open we saw */
3036 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3037 SV* const oreplsv = GvSV(PL_replgv);
3038 /* these three flags are set by various ops to signal information to
3039 * the very next op. They have a useful lifetime of exactly one loop
3040 * iteration, and are not preserved or restored by state pushes/pops
3042 bool sw = 0; /* the condition value in (?(cond)a|b) */
3043 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3044 int logical = 0; /* the following EVAL is:
3048 or the following IFMATCH/UNLESSM is:
3049 false: plain (?=foo)
3050 true: used as a condition: (?(?=foo))
3053 GET_RE_DEBUG_FLAGS_DECL;
3056 PERL_ARGS_ASSERT_REGMATCH;
3058 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3059 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3061 /* on first ever call to regmatch, allocate first slab */
3062 if (!PL_regmatch_slab) {
3063 Newx(PL_regmatch_slab, 1, regmatch_slab);
3064 PL_regmatch_slab->prev = NULL;
3065 PL_regmatch_slab->next = NULL;
3066 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3069 oldsave = PL_savestack_ix;
3070 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3071 SAVEVPTR(PL_regmatch_slab);
3072 SAVEVPTR(PL_regmatch_state);
3074 /* grab next free state slot */
3075 st = ++PL_regmatch_state;
3076 if (st > SLAB_LAST(PL_regmatch_slab))
3077 st = PL_regmatch_state = S_push_slab(aTHX);
3079 /* Note that nextchr is a byte even in UTF */
3080 nextchr = UCHARAT(locinput);
3082 while (scan != NULL) {
3085 SV * const prop = sv_newmortal();
3086 regnode *rnext=regnext(scan);
3087 DUMP_EXEC_POS( locinput, scan, utf8_target );
3088 regprop(rex, prop, scan);
3090 PerlIO_printf(Perl_debug_log,
3091 "%3"IVdf":%*s%s(%"IVdf")\n",
3092 (IV)(scan - rexi->program), depth*2, "",
3094 (PL_regkind[OP(scan)] == END || !rnext) ?
3095 0 : (IV)(rnext - rexi->program));
3098 next = scan + NEXT_OFF(scan);
3101 state_num = OP(scan);
3105 assert(PL_reglastparen == &rex->lastparen);
3106 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3107 assert(PL_regoffs == rex->offs);
3109 switch (state_num) {
3111 if (locinput == PL_bostr)
3113 /* reginfo->till = reginfo->bol; */
3118 if (locinput == PL_bostr ||
3119 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3125 if (locinput == PL_bostr)
3129 if (locinput == reginfo->ganch)
3134 /* update the startpoint */
3135 st->u.keeper.val = PL_regoffs[0].start;
3136 PL_reginput = locinput;
3137 PL_regoffs[0].start = locinput - PL_bostr;
3138 PUSH_STATE_GOTO(KEEPS_next, next);
3140 case KEEPS_next_fail:
3141 /* rollback the start point change */
3142 PL_regoffs[0].start = st->u.keeper.val;
3148 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3153 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3155 if (PL_regeol - locinput > 1)
3159 if (PL_regeol != locinput)
3163 if (!nextchr && locinput >= PL_regeol)
3166 locinput += PL_utf8skip[nextchr];
3167 if (locinput > PL_regeol)
3169 nextchr = UCHARAT(locinput);
3172 nextchr = UCHARAT(++locinput);
3175 if (!nextchr && locinput >= PL_regeol)
3177 nextchr = UCHARAT(++locinput);
3180 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3183 locinput += PL_utf8skip[nextchr];
3184 if (locinput > PL_regeol)
3186 nextchr = UCHARAT(locinput);
3189 nextchr = UCHARAT(++locinput);
3193 #define ST st->u.trie
3195 /* In this case the charclass data is available inline so
3196 we can fail fast without a lot of extra overhead.
3198 if (scan->flags == EXACT || !utf8_target) {
3199 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3201 PerlIO_printf(Perl_debug_log,
3202 "%*s %sfailed to match trie start class...%s\n",
3203 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3211 /* the basic plan of execution of the trie is:
3212 * At the beginning, run though all the states, and
3213 * find the longest-matching word. Also remember the position
3214 * of the shortest matching word. For example, this pattern:
3217 * when matched against the string "abcde", will generate
3218 * accept states for all words except 3, with the longest
3219 * matching word being 4, and the shortest being 1 (with
3220 * the position being after char 1 of the string).
3222 * Then for each matching word, in word order (i.e. 1,2,4,5),
3223 * we run the remainder of the pattern; on each try setting
3224 * the current position to the character following the word,
3225 * returning to try the next word on failure.
3227 * We avoid having to build a list of words at runtime by
3228 * using a compile-time structure, wordinfo[].prev, which
3229 * gives, for each word, the previous accepting word (if any).
3230 * In the case above it would contain the mappings 1->2, 2->0,
3231 * 3->0, 4->5, 5->1. We can use this table to generate, from
3232 * the longest word (4 above), a list of all words, by
3233 * following the list of prev pointers; this gives us the
3234 * unordered list 4,5,1,2. Then given the current word we have
3235 * just tried, we can go through the list and find the
3236 * next-biggest word to try (so if we just failed on word 2,
3237 * the next in the list is 4).
3239 * Since at runtime we don't record the matching position in
3240 * the string for each word, we have to work that out for
3241 * each word we're about to process. The wordinfo table holds
3242 * the character length of each word; given that we recorded
3243 * at the start: the position of the shortest word and its
3244 * length in chars, we just need to move the pointer the
3245 * difference between the two char lengths. Depending on
3246 * Unicode status and folding, that's cheap or expensive.
3248 * This algorithm is optimised for the case where are only a
3249 * small number of accept states, i.e. 0,1, or maybe 2.
3250 * With lots of accepts states, and having to try all of them,
3251 * it becomes quadratic on number of accept states to find all
3256 /* what type of TRIE am I? (utf8 makes this contextual) */
3257 DECL_TRIE_TYPE(scan);
3259 /* what trie are we using right now */
3260 reg_trie_data * const trie
3261 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3262 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3263 U32 state = trie->startstate;
3265 if (trie->bitmap && trie_type != trie_utf8_fold &&
3266 !TRIE_BITMAP_TEST(trie,*locinput)
3268 if (trie->states[ state ].wordnum) {
3270 PerlIO_printf(Perl_debug_log,
3271 "%*s %smatched empty string...%s\n",
3272 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3278 PerlIO_printf(Perl_debug_log,
3279 "%*s %sfailed to match trie start class...%s\n",
3280 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3287 U8 *uc = ( U8* )locinput;
3291 U8 *uscan = (U8*)NULL;
3292 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3293 U32 charcount = 0; /* how many input chars we have matched */
3294 U32 accepted = 0; /* have we seen any accepting states? */
3297 ST.jump = trie->jump;
3300 ST.longfold = FALSE; /* char longer if folded => it's harder */
3303 /* fully traverse the TRIE; note the position of the
3304 shortest accept state and the wordnum of the longest
3307 while ( state && uc <= (U8*)PL_regeol ) {
3308 U32 base = trie->states[ state ].trans.base;
3312 wordnum = trie->states[ state ].wordnum;
3314 if (wordnum) { /* it's an accept state */
3317 /* record first match position */
3319 ST.firstpos = (U8*)locinput;
3324 ST.firstchars = charcount;
3327 if (!ST.nextword || wordnum < ST.nextword)
3328 ST.nextword = wordnum;
3329 ST.topword = wordnum;
3332 DEBUG_TRIE_EXECUTE_r({
3333 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3334 PerlIO_printf( Perl_debug_log,
3335 "%*s %sState: %4"UVxf" Accepted: %c ",
3336 2+depth * 2, "", PL_colors[4],
3337 (UV)state, (accepted ? 'Y' : 'N'));
3340 /* read a char and goto next state */
3343 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3344 uscan, len, uvc, charid, foldlen,
3351 base + charid - 1 - trie->uniquecharcount)) >= 0)
3353 && ((U32)offset < trie->lasttrans)
3354 && trie->trans[offset].check == state)
3356 state = trie->trans[offset].next;
3367 DEBUG_TRIE_EXECUTE_r(
3368 PerlIO_printf( Perl_debug_log,
3369 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3370 charid, uvc, (UV)state, PL_colors[5] );
3376 /* calculate total number of accept states */
3381 w = trie->wordinfo[w].prev;
3384 ST.accepted = accepted;
3388 PerlIO_printf( Perl_debug_log,
3389 "%*s %sgot %"IVdf" possible matches%s\n",
3390 REPORT_CODE_OFF + depth * 2, "",
3391 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3393 goto trie_first_try; /* jump into the fail handler */
3397 case TRIE_next_fail: /* we failed - try next alternative */
3399 REGCP_UNWIND(ST.cp);
3400 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3401 PL_regoffs[n].end = -1;
3402 *PL_reglastparen = n;
3404 if (!--ST.accepted) {
3406 PerlIO_printf( Perl_debug_log,
3407 "%*s %sTRIE failed...%s\n",
3408 REPORT_CODE_OFF+depth*2, "",
3415 /* Find next-highest word to process. Note that this code
3416 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3417 register U16 min = 0;
3419 register U16 const nextword = ST.nextword;
3420 register reg_trie_wordinfo * const wordinfo
3421 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3422 for (word=ST.topword; word; word=wordinfo[word].prev) {
3423 if (word > nextword && (!min || word < min))
3436 ST.lastparen = *PL_reglastparen;
3440 /* find start char of end of current word */
3442 U32 chars; /* how many chars to skip */
3443 U8 *uc = ST.firstpos;
3444 reg_trie_data * const trie
3445 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3447 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3449 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3453 /* the hard option - fold each char in turn and find
3454 * its folded length (which may be different */
3455 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3463 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3471 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3476 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3490 PL_reginput = (char *)uc;
3493 scan = (ST.jump && ST.jump[ST.nextword])
3494 ? ST.me + ST.jump[ST.nextword]
3498 PerlIO_printf( Perl_debug_log,
3499 "%*s %sTRIE matched word #%d, continuing%s\n",
3500 REPORT_CODE_OFF+depth*2, "",
3507 if (ST.accepted > 1 || has_cutgroup) {
3508 PUSH_STATE_GOTO(TRIE_next, scan);
3511 /* only one choice left - just continue */
3513 AV *const trie_words
3514 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3515 SV ** const tmp = av_fetch( trie_words,
3517 SV *sv= tmp ? sv_newmortal() : NULL;
3519 PerlIO_printf( Perl_debug_log,
3520 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3521 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3523 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3524 PL_colors[0], PL_colors[1],
3525 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3527 : "not compiled under -Dr",
3531 locinput = PL_reginput;
3532 nextchr = UCHARAT(locinput);
3533 continue; /* execute rest of RE */
3538 char *s = STRING(scan);
3540 if (utf8_target != UTF_PATTERN) {
3541 /* The target and the pattern have differing utf8ness. */
3543 const char * const e = s + ln;
3546 /* The target is utf8, the pattern is not utf8. */
3551 if (NATIVE_TO_UNI(*(U8*)s) !=
3552 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3560 /* The target is not utf8, the pattern is utf8. */
3565 if (NATIVE_TO_UNI(*((U8*)l)) !=
3566 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3574 nextchr = UCHARAT(locinput);
3577 /* The target and the pattern have the same utf8ness. */
3578 /* Inline the first character, for speed. */
3579 if (UCHARAT(s) != nextchr)
3581 if (PL_regeol - locinput < ln)
3583 if (ln > 1 && memNE(s, locinput, ln))
3586 nextchr = UCHARAT(locinput);
3591 const U8 * fold_array;
3593 U32 fold_utf8_flags;
3595 PL_reg_flags |= RF_tainted;
3596 folder = foldEQ_locale;
3597 fold_array = PL_fold_locale;
3598 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3602 folder = foldEQ_latin1;
3603 fold_array = PL_fold_latin1;
3604 fold_utf8_flags = 0;
3608 folder = foldEQ_latin1;
3609 fold_array = PL_fold_latin1;
3610 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3615 fold_array = PL_fold;
3616 fold_utf8_flags = 0;
3622 if (utf8_target || UTF_PATTERN) {
3623 /* Either target or the pattern are utf8. */
3624 const char * const l = locinput;
3625 char *e = PL_regeol;
3627 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3628 l, &e, 0, utf8_target, fold_utf8_flags))
3633 nextchr = UCHARAT(locinput);
3637 /* Neither the target nor the pattern are utf8 */
3638 if (UCHARAT(s) != nextchr &&
3639 UCHARAT(s) != fold_array[nextchr])
3643 if (PL_regeol - locinput < ln)
3645 if (ln > 1 && ! folder(s, locinput, ln))
3648 nextchr = UCHARAT(locinput);
3652 /* XXX Could improve efficiency by separating these all out using a
3653 * macro or in-line function. At that point regcomp.c would no longer
3654 * have to set the FLAGS fields of these */
3657 PL_reg_flags |= RF_tainted;
3665 /* was last char in word? */
3667 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3668 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3670 if (locinput == PL_bostr)
3673 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3675 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3677 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3678 ln = isALNUM_uni(ln);
3679 LOAD_UTF8_CHARCLASS_ALNUM();
3680 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3683 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3684 n = isALNUM_LC_utf8((U8*)locinput);
3689 /* Here the string isn't utf8, or is utf8 and only ascii
3690 * characters are to match \w. In the latter case looking at
3691 * the byte just prior to the current one may be just the final
3692 * byte of a multi-byte character. This is ok. There are two
3694 * 1) it is a single byte character, and then the test is doing
3695 * just what it's supposed to.
3696 * 2) it is a multi-byte character, in which case the final
3697 * byte is never mistakable for ASCII, and so the test
3698 * will say it is not a word character, which is the
3699 * correct answer. */
3700 ln = (locinput != PL_bostr) ?
3701 UCHARAT(locinput - 1) : '\n';
3702 switch (FLAGS(scan)) {
3703 case REGEX_UNICODE_CHARSET:
3704 ln = isWORDCHAR_L1(ln);
3705 n = isWORDCHAR_L1(nextchr);
3707 case REGEX_LOCALE_CHARSET:
3708 ln = isALNUM_LC(ln);
3709 n = isALNUM_LC(nextchr);
3711 case REGEX_DEPENDS_CHARSET:
3713 n = isALNUM(nextchr);
3715 case REGEX_ASCII_RESTRICTED_CHARSET:
3716 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3717 ln = isWORDCHAR_A(ln);
3718 n = isWORDCHAR_A(nextchr);
3721 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3725 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3727 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3732 if (utf8_target || state_num == ANYOFV) {
3733 STRLEN inclasslen = PL_regeol - locinput;
3734 if (locinput >= PL_regeol)
3737 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3739 locinput += inclasslen;
3740 nextchr = UCHARAT(locinput);
3745 nextchr = UCHARAT(locinput);
3746 if (!nextchr && locinput >= PL_regeol)
3748 if (!REGINCLASS(rex, scan, (U8*)locinput))
3750 nextchr = UCHARAT(++locinput);
3754 /* Special char classes - The defines start on line 129 or so */
3755 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3756 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3757 ALNUMU, NALNUMU, isWORDCHAR_L1,
3758 ALNUMA, NALNUMA, isWORDCHAR_A,
3761 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3762 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3763 SPACEU, NSPACEU, isSPACE_L1,
3764 SPACEA, NSPACEA, isSPACE_A,
3767 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3768 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3769 DIGITA, NDIGITA, isDIGIT_A,
3772 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3773 a Unicode extended Grapheme Cluster */
3774 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3775 extended Grapheme Cluster is:
3778 | Prepend* Begin Extend*
3781 Begin is (Hangul-syllable | ! Control)
3782 Extend is (Grapheme_Extend | Spacing_Mark)
3783 Control is [ GCB_Control CR LF ]
3785 The discussion below shows how the code for CLUMP is derived
3786 from this regex. Note that most of these concepts are from
3787 property values of the Grapheme Cluster Boundary (GCB) property.
3788 No code point can have multiple property values for a given
3789 property. Thus a code point in Prepend can't be in Control, but
3790 it must be in !Control. This is why Control above includes
3791 GCB_Control plus CR plus LF. The latter two are used in the GCB
3792 property separately, and so can't be in GCB_Control, even though
3793 they logically are controls. Control is not the same as gc=cc,
3794 but includes format and other characters as well.
3796 The Unicode definition of Hangul-syllable is:
3798 | (L* ( ( V | LV ) V* | LVT ) T*)
3801 Each of these is a value for the GCB property, and hence must be
3802 disjoint, so the order they are tested is immaterial, so the
3803 above can safely be changed to
3806 | (L* ( LVT | ( V | LV ) V*) T*)
3808 The last two terms can be combined like this:
3810 | (( LVT | ( V | LV ) V*) T*))
3812 And refactored into this:
3813 L* (L | LVT T* | V V* T* | LV V* T*)
3815 That means that if we have seen any L's at all we can quit
3816 there, but if the next character is a LVT, a V or and LV we
3819 There is a subtlety with Prepend* which showed up in testing.
3820 Note that the Begin, and only the Begin is required in:
3821 | Prepend* Begin Extend*
3822 Also, Begin contains '! Control'. A Prepend must be a '!
3823 Control', which means it must be a Begin. What it comes down to
3824 is that if we match Prepend* and then find no suitable Begin
3825 afterwards, that if we backtrack the last Prepend, that one will
3826 be a suitable Begin.
3829 if (locinput >= PL_regeol)
3831 if (! utf8_target) {
3833 /* Match either CR LF or '.', as all the other possibilities
3835 locinput++; /* Match the . or CR */
3837 && locinput < PL_regeol
3838 && UCHARAT(locinput) == '\n') locinput++;
3842 /* Utf8: See if is ( CR LF ); already know that locinput <
3843 * PL_regeol, so locinput+1 is in bounds */
3844 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3848 /* In case have to backtrack to beginning, then match '.' */
3849 char *starting = locinput;
3851 /* In case have to backtrack the last prepend */
3852 char *previous_prepend = 0;
3854 LOAD_UTF8_CHARCLASS_GCB();
3856 /* Match (prepend)* */
3857 while (locinput < PL_regeol
3858 && swash_fetch(PL_utf8_X_prepend,
3859 (U8*)locinput, utf8_target))
3861 previous_prepend = locinput;
3862 locinput += UTF8SKIP(locinput);
3865 /* As noted above, if we matched a prepend character, but
3866 * the next thing won't match, back off the last prepend we
3867 * matched, as it is guaranteed to match the begin */
3868 if (previous_prepend
3869 && (locinput >= PL_regeol
3870 || ! swash_fetch(PL_utf8_X_begin,
3871 (U8*)locinput, utf8_target)))
3873 locinput = previous_prepend;
3876 /* Note that here we know PL_regeol > locinput, as we
3877 * tested that upon input to this switch case, and if we
3878 * moved locinput forward, we tested the result just above
3879 * and it either passed, or we backed off so that it will
3881 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3883 /* Here did not match the required 'Begin' in the
3884 * second term. So just match the very first
3885 * character, the '.' of the final term of the regex */
3886 locinput = starting + UTF8SKIP(starting);
3889 /* Here is the beginning of a character that can have
3890 * an extender. It is either a hangul syllable, or a
3892 if (swash_fetch(PL_utf8_X_non_hangul,
3893 (U8*)locinput, utf8_target))
3896 /* Here not a Hangul syllable, must be a
3897 * ('! * Control') */
3898 locinput += UTF8SKIP(locinput);
3901 /* Here is a Hangul syllable. It can be composed
3902 * of several individual characters. One
3903 * possibility is T+ */
3904 if (swash_fetch(PL_utf8_X_T,
3905 (U8*)locinput, utf8_target))
3907 while (locinput < PL_regeol
3908 && swash_fetch(PL_utf8_X_T,
3909 (U8*)locinput, utf8_target))
3911 locinput += UTF8SKIP(locinput);
3915 /* Here, not T+, but is a Hangul. That means
3916 * it is one of the others: L, LV, LVT or V,
3918 * L* (L | LVT T* | V V* T* | LV V* T*) */
3921 while (locinput < PL_regeol
3922 && swash_fetch(PL_utf8_X_L,
3923 (U8*)locinput, utf8_target))
3925 locinput += UTF8SKIP(locinput);
3928 /* Here, have exhausted L*. If the next
3929 * character is not an LV, LVT nor V, it means
3930 * we had to have at least one L, so matches L+
3931 * in the original equation, we have a complete
3932 * hangul syllable. Are done. */
3934 if (locinput < PL_regeol
3935 && swash_fetch(PL_utf8_X_LV_LVT_V,
3936 (U8*)locinput, utf8_target))
3939 /* Otherwise keep going. Must be LV, LVT
3940 * or V. See if LVT */
3941 if (swash_fetch(PL_utf8_X_LVT,
3942 (U8*)locinput, utf8_target))
3944 locinput += UTF8SKIP(locinput);
3947 /* Must be V or LV. Take it, then
3949 locinput += UTF8SKIP(locinput);
3950 while (locinput < PL_regeol
3951 && swash_fetch(PL_utf8_X_V,
3952 (U8*)locinput, utf8_target))
3954 locinput += UTF8SKIP(locinput);
3958 /* And any of LV, LVT, or V can be followed
3960 while (locinput < PL_regeol
3961 && swash_fetch(PL_utf8_X_T,
3965 locinput += UTF8SKIP(locinput);
3971 /* Match any extender */
3972 while (locinput < PL_regeol
3973 && swash_fetch(PL_utf8_X_extend,
3974 (U8*)locinput, utf8_target))
3976 locinput += UTF8SKIP(locinput);
3980 if (locinput > PL_regeol) sayNO;
3982 nextchr = UCHARAT(locinput);
3986 { /* The capture buffer cases. The ones beginning with N for the
3987 named buffers just convert to the equivalent numbered and
3988 pretend they were called as the corresponding numbered buffer
3990 /* don't initialize these in the declaration, it makes C++
3995 const U8 *fold_array;
3998 PL_reg_flags |= RF_tainted;
3999 folder = foldEQ_locale;
4000 fold_array = PL_fold_locale;
4002 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4006 folder = foldEQ_latin1;
4007 fold_array = PL_fold_latin1;
4009 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4013 folder = foldEQ_latin1;
4014 fold_array = PL_fold_latin1;
4016 utf8_fold_flags = 0;
4021 fold_array = PL_fold;
4023 utf8_fold_flags = 0;
4030 utf8_fold_flags = 0;
4033 /* For the named back references, find the corresponding buffer
4035 n = reg_check_named_buff_matched(rex,scan);
4040 goto do_nref_ref_common;
4043 PL_reg_flags |= RF_tainted;
4044 folder = foldEQ_locale;
4045 fold_array = PL_fold_locale;
4046 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4050 folder = foldEQ_latin1;
4051 fold_array = PL_fold_latin1;
4052 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4056 folder = foldEQ_latin1;
4057 fold_array = PL_fold_latin1;
4058 utf8_fold_flags = 0;
4063 fold_array = PL_fold;
4064 utf8_fold_flags = 0;
4070 utf8_fold_flags = 0;
4074 n = ARG(scan); /* which paren pair */
4077 ln = PL_regoffs[n].start;
4078 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4079 if (*PL_reglastparen < n || ln == -1)
4080 sayNO; /* Do not match unless seen CLOSEn. */
4081 if (ln == PL_regoffs[n].end)
4085 if (type != REF /* REF can do byte comparison */
4086 && (utf8_target || type == REFFU))
4087 { /* XXX handle REFFL better */
4088 char * limit = PL_regeol;
4090 /* This call case insensitively compares the entire buffer
4091 * at s, with the current input starting at locinput, but
4092 * not going off the end given by PL_regeol, and returns in
4093 * limit upon success, how much of the current input was
4095 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4096 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4101 nextchr = UCHARAT(locinput);
4105 /* Not utf8: Inline the first character, for speed. */
4106 if (UCHARAT(s) != nextchr &&
4108 UCHARAT(s) != fold_array[nextchr]))
4110 ln = PL_regoffs[n].end - ln;
4111 if (locinput + ln > PL_regeol)
4113 if (ln > 1 && (type == REF
4114 ? memNE(s, locinput, ln)
4115 : ! folder(s, locinput, ln)))
4118 nextchr = UCHARAT(locinput);
4128 #define ST st->u.eval
4133 regexp_internal *rei;
4134 regnode *startpoint;
4137 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4138 if (cur_eval && cur_eval->locinput==locinput) {
4139 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4140 Perl_croak(aTHX_ "Infinite recursion in regex");
4141 if ( ++nochange_depth > max_nochange_depth )
4143 "Pattern subroutine nesting without pos change"
4144 " exceeded limit in regex");
4151 (void)ReREFCNT_inc(rex_sv);
4152 if (OP(scan)==GOSUB) {
4153 startpoint = scan + ARG2L(scan);
4154 ST.close_paren = ARG(scan);
4156 startpoint = rei->program+1;
4159 goto eval_recurse_doit;
4161 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4162 if (cur_eval && cur_eval->locinput==locinput) {
4163 if ( ++nochange_depth > max_nochange_depth )
4164 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4169 /* execute the code in the {...} */
4171 SV ** const before = SP;
4172 OP_4tree * const oop = PL_op;
4173 COP * const ocurcop = PL_curcop;
4175 char *saved_regeol = PL_regeol;
4176 struct re_save_state saved_state;
4178 /* To not corrupt the existing regex state while executing the
4179 * eval we would normally put it on the save stack, like with
4180 * save_re_context. However, re-evals have a weird scoping so we
4181 * can't just add ENTER/LEAVE here. With that, things like
4183 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4185 * would break, as they expect the localisation to be unwound
4186 * only when the re-engine backtracks through the bit that
4189 * What we do instead is just saving the state in a local c
4192 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4195 PL_op = (OP_4tree*)rexi->data->data[n];
4196 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4197 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4198 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4199 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4202 SV *sv_mrk = get_sv("REGMARK", 1);
4203 sv_setsv(sv_mrk, sv_yes_mark);
4206 CALLRUNOPS(aTHX); /* Scalar context. */
4209 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4215 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4218 PAD_RESTORE_LOCAL(old_comppad);
4219 PL_curcop = ocurcop;
4220 PL_regeol = saved_regeol;
4223 sv_setsv(save_scalar(PL_replgv), ret);
4227 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4230 /* extract RE object from returned value; compiling if
4236 SV *const sv = SvRV(ret);
4238 if (SvTYPE(sv) == SVt_REGEXP) {
4240 } else if (SvSMAGICAL(sv)) {
4241 mg = mg_find(sv, PERL_MAGIC_qr);
4244 } else if (SvTYPE(ret) == SVt_REGEXP) {
4246 } else if (SvSMAGICAL(ret)) {
4247 if (SvGMAGICAL(ret)) {
4248 /* I don't believe that there is ever qr magic
4250 assert(!mg_find(ret, PERL_MAGIC_qr));
4251 sv_unmagic(ret, PERL_MAGIC_qr);
4254 mg = mg_find(ret, PERL_MAGIC_qr);
4255 /* testing suggests mg only ends up non-NULL for
4256 scalars who were upgraded and compiled in the
4257 else block below. In turn, this is only
4258 triggered in the "postponed utf8 string" tests
4264 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4268 rx = reg_temp_copy(NULL, rx);
4272 const I32 osize = PL_regsize;
4275 assert (SvUTF8(ret));
4276 } else if (SvUTF8(ret)) {
4277 /* Not doing UTF-8, despite what the SV says. Is
4278 this only if we're trapped in use 'bytes'? */
4279 /* Make a copy of the octet sequence, but without
4280 the flag on, as the compiler now honours the
4281 SvUTF8 flag on ret. */
4283 const char *const p = SvPV(ret, len);
4284 ret = newSVpvn_flags(p, len, SVs_TEMP);
4286 rx = CALLREGCOMP(ret, pm_flags);
4288 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4290 /* This isn't a first class regexp. Instead, it's
4291 caching a regexp onto an existing, Perl visible
4293 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4298 re = (struct regexp *)SvANY(rx);
4300 RXp_MATCH_COPIED_off(re);
4301 re->subbeg = rex->subbeg;
4302 re->sublen = rex->sublen;
4305 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4306 "Matching embedded");
4308 startpoint = rei->program + 1;
4309 ST.close_paren = 0; /* only used for GOSUB */
4310 /* borrowed from regtry */
4311 if (PL_reg_start_tmpl <= re->nparens) {
4312 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4313 if(PL_reg_start_tmp)
4314 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4316 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4319 eval_recurse_doit: /* Share code with GOSUB below this line */
4320 /* run the pattern returned from (??{...}) */
4321 ST.cp = regcppush(0); /* Save *all* the positions. */
4322 REGCP_SET(ST.lastcp);
4324 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4326 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4327 PL_reglastparen = &re->lastparen;
4328 PL_reglastcloseparen = &re->lastcloseparen;
4330 re->lastcloseparen = 0;
4332 PL_reginput = locinput;
4335 /* XXXX This is too dramatic a measure... */
4338 ST.toggle_reg_flags = PL_reg_flags;
4340 PL_reg_flags |= RF_utf8;
4342 PL_reg_flags &= ~RF_utf8;
4343 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4345 ST.prev_rex = rex_sv;
4346 ST.prev_curlyx = cur_curlyx;
4347 SETREX(rex_sv,re_sv);
4352 ST.prev_eval = cur_eval;
4354 /* now continue from first node in postoned RE */
4355 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4358 /* logical is 1, /(?(?{...})X|Y)/ */
4359 sw = cBOOL(SvTRUE(ret));
4364 case EVAL_AB: /* cleanup after a successful (??{A})B */
4365 /* note: this is called twice; first after popping B, then A */
4366 PL_reg_flags ^= ST.toggle_reg_flags;
4367 ReREFCNT_dec(rex_sv);
4368 SETREX(rex_sv,ST.prev_rex);
4369 rex = (struct regexp *)SvANY(rex_sv);
4370 rexi = RXi_GET(rex);
4372 cur_eval = ST.prev_eval;
4373 cur_curlyx = ST.prev_curlyx;
4375 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4376 PL_reglastparen = &rex->lastparen;
4377 PL_reglastcloseparen = &rex->lastcloseparen;
4378 /* also update PL_regoffs */
4379 PL_regoffs = rex->offs;
4381 /* XXXX This is too dramatic a measure... */
4383 if ( nochange_depth )
4388 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4389 /* note: this is called twice; first after popping B, then A */
4390 PL_reg_flags ^= ST.toggle_reg_flags;
4391 ReREFCNT_dec(rex_sv);
4392 SETREX(rex_sv,ST.prev_rex);
4393 rex = (struct regexp *)SvANY(rex_sv);
4394 rexi = RXi_GET(rex);
4395 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4396 PL_reglastparen = &rex->lastparen;
4397 PL_reglastcloseparen = &rex->lastcloseparen;
4399 PL_reginput = locinput;
4400 REGCP_UNWIND(ST.lastcp);
4402 cur_eval = ST.prev_eval;
4403 cur_curlyx = ST.prev_curlyx;
4404 /* XXXX This is too dramatic a measure... */
4406 if ( nochange_depth )
4412 n = ARG(scan); /* which paren pair */
4413 PL_reg_start_tmp[n] = locinput;
4419 n = ARG(scan); /* which paren pair */
4420 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4421 PL_regoffs[n].end = locinput - PL_bostr;
4422 /*if (n > PL_regsize)
4424 if (n > *PL_reglastparen)
4425 *PL_reglastparen = n;
4426 *PL_reglastcloseparen = n;
4427 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4435 cursor && OP(cursor)!=END;
4436 cursor=regnext(cursor))
4438 if ( OP(cursor)==CLOSE ){
4440 if ( n <= lastopen ) {
4442 = PL_reg_start_tmp[n] - PL_bostr;
4443 PL_regoffs[n].end = locinput - PL_bostr;
4444 /*if (n > PL_regsize)
4446 if (n > *PL_reglastparen)
4447 *PL_reglastparen = n;
4448 *PL_reglastcloseparen = n;
4449 if ( n == ARG(scan) || (cur_eval &&
4450 cur_eval->u.eval.close_paren == n))
4459 n = ARG(scan); /* which paren pair */
4460 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4463 /* reg_check_named_buff_matched returns 0 for no match */
4464 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4468 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4474 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4476 next = NEXTOPER(NEXTOPER(scan));
4478 next = scan + ARG(scan);
4479 if (OP(next) == IFTHEN) /* Fake one. */
4480 next = NEXTOPER(NEXTOPER(next));
4484 logical = scan->flags;
4487 /*******************************************************************
4489 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4490 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4491 STAR/PLUS/CURLY/CURLYN are used instead.)
4493 A*B is compiled as <CURLYX><A><WHILEM><B>
4495 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4496 state, which contains the current count, initialised to -1. It also sets
4497 cur_curlyx to point to this state, with any previous value saved in the
4500 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4501 since the pattern may possibly match zero times (i.e. it's a while {} loop
4502 rather than a do {} while loop).
4504 Each entry to WHILEM represents a successful match of A. The count in the
4505 CURLYX block is incremented, another WHILEM state is pushed, and execution
4506 passes to A or B depending on greediness and the current count.
4508 For example, if matching against the string a1a2a3b (where the aN are
4509 substrings that match /A/), then the match progresses as follows: (the
4510 pushed states are interspersed with the bits of strings matched so far):
4513 <CURLYX cnt=0><WHILEM>
4514 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4515 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4516 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4517 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4519 (Contrast this with something like CURLYM, which maintains only a single
4523 a1 <CURLYM cnt=1> a2
4524 a1 a2 <CURLYM cnt=2> a3
4525 a1 a2 a3 <CURLYM cnt=3> b
4528 Each WHILEM state block marks a point to backtrack to upon partial failure
4529 of A or B, and also contains some minor state data related to that
4530 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4531 overall state, such as the count, and pointers to the A and B ops.
4533 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4534 must always point to the *current* CURLYX block, the rules are:
4536 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4537 and set cur_curlyx to point the new block.
4539 When popping the CURLYX block after a successful or unsuccessful match,
4540 restore the previous cur_curlyx.
4542 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4543 to the outer one saved in the CURLYX block.
4545 When popping the WHILEM block after a successful or unsuccessful B match,
4546 restore the previous cur_curlyx.
4548 Here's an example for the pattern (AI* BI)*BO
4549 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4552 curlyx backtrack stack
4553 ------ ---------------
4555 CO <CO prev=NULL> <WO>
4556 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4557 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4558 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4560 At this point the pattern succeeds, and we work back down the stack to
4561 clean up, restoring as we go:
4563 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4564 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4565 CO <CO prev=NULL> <WO>
4568 *******************************************************************/
4570 #define ST st->u.curlyx
4572 case CURLYX: /* start of /A*B/ (for complex A) */
4574 /* No need to save/restore up to this paren */
4575 I32 parenfloor = scan->flags;
4577 assert(next); /* keep Coverity happy */
4578 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4581 /* XXXX Probably it is better to teach regpush to support
4582 parenfloor > PL_regsize... */
4583 if (parenfloor > (I32)*PL_reglastparen)
4584 parenfloor = *PL_reglastparen; /* Pessimization... */
4586 ST.prev_curlyx= cur_curlyx;
4588 ST.cp = PL_savestack_ix;
4590 /* these fields contain the state of the current curly.
4591 * they are accessed by subsequent WHILEMs */
4592 ST.parenfloor = parenfloor;
4597 ST.count = -1; /* this will be updated by WHILEM */
4598 ST.lastloc = NULL; /* this will be updated by WHILEM */
4600 PL_reginput = locinput;
4601 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4605 case CURLYX_end: /* just finished matching all of A*B */
4606 cur_curlyx = ST.prev_curlyx;
4610 case CURLYX_end_fail: /* just failed to match all of A*B */
4612 cur_curlyx = ST.prev_curlyx;
4618 #define ST st->u.whilem
4620 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4622 /* see the discussion above about CURLYX/WHILEM */
4624 int min = ARG1(cur_curlyx->u.curlyx.me);
4625 int max = ARG2(cur_curlyx->u.curlyx.me);
4626 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4628 assert(cur_curlyx); /* keep Coverity happy */
4629 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4630 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4631 ST.cache_offset = 0;
4634 PL_reginput = locinput;
4636 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4637 "%*s whilem: matched %ld out of %d..%d\n",
4638 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4641 /* First just match a string of min A's. */
4644 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4645 cur_curlyx->u.curlyx.lastloc = locinput;
4646 REGCP_SET(ST.lastcp);
4648 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4652 /* If degenerate A matches "", assume A done. */
4654 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4655 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4656 "%*s whilem: empty match detected, trying continuation...\n",
4657 REPORT_CODE_OFF+depth*2, "")
4659 goto do_whilem_B_max;
4662 /* super-linear cache processing */
4666 if (!PL_reg_maxiter) {
4667 /* start the countdown: Postpone detection until we
4668 * know the match is not *that* much linear. */
4669 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4670 /* possible overflow for long strings and many CURLYX's */
4671 if (PL_reg_maxiter < 0)
4672 PL_reg_maxiter = I32_MAX;
4673 PL_reg_leftiter = PL_reg_maxiter;
4676 if (PL_reg_leftiter-- == 0) {
4677 /* initialise cache */
4678 const I32 size = (PL_reg_maxiter + 7)/8;
4679 if (PL_reg_poscache) {
4680 if ((I32)PL_reg_poscache_size < size) {
4681 Renew(PL_reg_poscache, size, char);
4682 PL_reg_poscache_size = size;
4684 Zero(PL_reg_poscache, size, char);
4687 PL_reg_poscache_size = size;
4688 Newxz(PL_reg_poscache, size, char);
4690 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4691 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4692 PL_colors[4], PL_colors[5])
4696 if (PL_reg_leftiter < 0) {
4697 /* have we already failed at this position? */
4699 offset = (scan->flags & 0xf) - 1
4700 + (locinput - PL_bostr) * (scan->flags>>4);
4701 mask = 1 << (offset % 8);
4703 if (PL_reg_poscache[offset] & mask) {
4704 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4705 "%*s whilem: (cache) already tried at this position...\n",
4706 REPORT_CODE_OFF+depth*2, "")
4708 sayNO; /* cache records failure */
4710 ST.cache_offset = offset;
4711 ST.cache_mask = mask;
4715 /* Prefer B over A for minimal matching. */
4717 if (cur_curlyx->u.curlyx.minmod) {
4718 ST.save_curlyx = cur_curlyx;
4719 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4720 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4721 REGCP_SET(ST.lastcp);
4722 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4726 /* Prefer A over B for maximal matching. */
4728 if (n < max) { /* More greed allowed? */
4729 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4730 cur_curlyx->u.curlyx.lastloc = locinput;
4731 REGCP_SET(ST.lastcp);
4732 PUSH_STATE_GOTO(WHILEM_A_max, A);
4735 goto do_whilem_B_max;
4739 case WHILEM_B_min: /* just matched B in a minimal match */
4740 case WHILEM_B_max: /* just matched B in a maximal match */
4741 cur_curlyx = ST.save_curlyx;
4745 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4746 cur_curlyx = ST.save_curlyx;
4747 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4748 cur_curlyx->u.curlyx.count--;
4752 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4754 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4755 REGCP_UNWIND(ST.lastcp);
4757 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4758 cur_curlyx->u.curlyx.count--;
4762 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4763 REGCP_UNWIND(ST.lastcp);
4764 regcppop(rex); /* Restore some previous $<digit>s? */
4765 PL_reginput = locinput;
4766 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4767 "%*s whilem: failed, trying continuation...\n",
4768 REPORT_CODE_OFF+depth*2, "")
4771 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4772 && ckWARN(WARN_REGEXP)
4773 && !(PL_reg_flags & RF_warned))
4775 PL_reg_flags |= RF_warned;
4776 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4777 "Complex regular subexpression recursion",
4782 ST.save_curlyx = cur_curlyx;
4783 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4784 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4787 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4788 cur_curlyx = ST.save_curlyx;
4789 REGCP_UNWIND(ST.lastcp);
4792 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4793 /* Maximum greed exceeded */
4794 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4795 && ckWARN(WARN_REGEXP)
4796 && !(PL_reg_flags & RF_warned))
4798 PL_reg_flags |= RF_warned;
4799 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4800 "%s limit (%d) exceeded",
4801 "Complex regular subexpression recursion",
4804 cur_curlyx->u.curlyx.count--;
4808 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4809 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4811 /* Try grabbing another A and see if it helps. */
4812 PL_reginput = locinput;
4813 cur_curlyx->u.curlyx.lastloc = locinput;
4814 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4815 REGCP_SET(ST.lastcp);
4816 PUSH_STATE_GOTO(WHILEM_A_min,
4817 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4821 #define ST st->u.branch
4823 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4824 next = scan + ARG(scan);
4827 scan = NEXTOPER(scan);
4830 case BRANCH: /* /(...|A|...)/ */
4831 scan = NEXTOPER(scan); /* scan now points to inner node */
4832 ST.lastparen = *PL_reglastparen;
4833 ST.next_branch = next;
4835 PL_reginput = locinput;
4837 /* Now go into the branch */
4839 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4841 PUSH_STATE_GOTO(BRANCH_next, scan);
4845 PL_reginput = locinput;
4846 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4847 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4848 PUSH_STATE_GOTO(CUTGROUP_next,next);
4850 case CUTGROUP_next_fail:
4853 if (st->u.mark.mark_name)
4854 sv_commit = st->u.mark.mark_name;
4860 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4865 REGCP_UNWIND(ST.cp);
4866 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4867 PL_regoffs[n].end = -1;
4868 *PL_reglastparen = n;
4869 /*dmq: *PL_reglastcloseparen = n; */
4870 scan = ST.next_branch;
4871 /* no more branches? */
4872 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4874 PerlIO_printf( Perl_debug_log,
4875 "%*s %sBRANCH failed...%s\n",
4876 REPORT_CODE_OFF+depth*2, "",
4882 continue; /* execute next BRANCH[J] op */
4890 #define ST st->u.curlym
4892 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4894 /* This is an optimisation of CURLYX that enables us to push
4895 * only a single backtracking state, no matter how many matches
4896 * there are in {m,n}. It relies on the pattern being constant
4897 * length, with no parens to influence future backrefs
4901 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4903 /* if paren positive, emulate an OPEN/CLOSE around A */
4905 U32 paren = ST.me->flags;
4906 if (paren > PL_regsize)
4908 if (paren > *PL_reglastparen)
4909 *PL_reglastparen = paren;
4910 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4918 ST.c1 = CHRTEST_UNINIT;
4921 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4924 curlym_do_A: /* execute the A in /A{m,n}B/ */
4925 PL_reginput = locinput;
4926 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4929 case CURLYM_A: /* we've just matched an A */
4930 locinput = st->locinput;
4931 nextchr = UCHARAT(locinput);
4934 /* after first match, determine A's length: u.curlym.alen */
4935 if (ST.count == 1) {
4936 if (PL_reg_match_utf8) {
4938 while (s < PL_reginput) {
4944 ST.alen = PL_reginput - locinput;
4947 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4950 PerlIO_printf(Perl_debug_log,
4951 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4952 (int)(REPORT_CODE_OFF+(depth*2)), "",
4953 (IV) ST.count, (IV)ST.alen)
4956 locinput = PL_reginput;
4958 if (cur_eval && cur_eval->u.eval.close_paren &&
4959 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4963 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4964 if ( max == REG_INFTY || ST.count < max )
4965 goto curlym_do_A; /* try to match another A */
4967 goto curlym_do_B; /* try to match B */
4969 case CURLYM_A_fail: /* just failed to match an A */
4970 REGCP_UNWIND(ST.cp);
4972 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4973 || (cur_eval && cur_eval->u.eval.close_paren &&
4974 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4977 curlym_do_B: /* execute the B in /A{m,n}B/ */
4978 PL_reginput = locinput;
4979 if (ST.c1 == CHRTEST_UNINIT) {
4980 /* calculate c1 and c2 for possible match of 1st char
4981 * following curly */
4982 ST.c1 = ST.c2 = CHRTEST_VOID;
4983 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4984 regnode *text_node = ST.B;
4985 if (! HAS_TEXT(text_node))
4986 FIND_NEXT_IMPT(text_node);
4989 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4991 But the former is redundant in light of the latter.
4993 if this changes back then the macro for
4994 IS_TEXT and friends need to change.
4996 if (PL_regkind[OP(text_node)] == EXACT)
4999 ST.c1 = (U8)*STRING(text_node);
5000 switch (OP(text_node)) {
5001 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5003 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5004 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5005 default: ST.c2 = ST.c1;
5012 PerlIO_printf(Perl_debug_log,
5013 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5014 (int)(REPORT_CODE_OFF+(depth*2)),
5017 if (ST.c1 != CHRTEST_VOID
5018 && UCHARAT(PL_reginput) != ST.c1
5019 && UCHARAT(PL_reginput) != ST.c2)
5021 /* simulate B failing */
5023 PerlIO_printf(Perl_debug_log,
5024 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5025 (int)(REPORT_CODE_OFF+(depth*2)),"",
5028 state_num = CURLYM_B_fail;
5029 goto reenter_switch;
5033 /* mark current A as captured */
5034 I32 paren = ST.me->flags;
5036 PL_regoffs[paren].start
5037 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5038 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5039 /*dmq: *PL_reglastcloseparen = paren; */
5042 PL_regoffs[paren].end = -1;
5043 if (cur_eval && cur_eval->u.eval.close_paren &&
5044 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5053 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5056 case CURLYM_B_fail: /* just failed to match a B */
5057 REGCP_UNWIND(ST.cp);
5059 I32 max = ARG2(ST.me);
5060 if (max != REG_INFTY && ST.count == max)
5062 goto curlym_do_A; /* try to match a further A */
5064 /* backtrack one A */
5065 if (ST.count == ARG1(ST.me) /* min */)
5068 locinput = HOPc(locinput, -ST.alen);
5069 goto curlym_do_B; /* try to match B */
5072 #define ST st->u.curly
5074 #define CURLY_SETPAREN(paren, success) \
5077 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5078 PL_regoffs[paren].end = locinput - PL_bostr; \
5079 *PL_reglastcloseparen = paren; \
5082 PL_regoffs[paren].end = -1; \
5085 case STAR: /* /A*B/ where A is width 1 */
5089 scan = NEXTOPER(scan);
5091 case PLUS: /* /A+B/ where A is width 1 */
5095 scan = NEXTOPER(scan);
5097 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5098 ST.paren = scan->flags; /* Which paren to set */
5099 if (ST.paren > PL_regsize)
5100 PL_regsize = ST.paren;
5101 if (ST.paren > *PL_reglastparen)
5102 *PL_reglastparen = ST.paren;
5103 ST.min = ARG1(scan); /* min to match */
5104 ST.max = ARG2(scan); /* max to match */
5105 if (cur_eval && cur_eval->u.eval.close_paren &&
5106 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5110 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5112 case CURLY: /* /A{m,n}B/ where A is width 1 */
5114 ST.min = ARG1(scan); /* min to match */
5115 ST.max = ARG2(scan); /* max to match */
5116 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5119 * Lookahead to avoid useless match attempts
5120 * when we know what character comes next.
5122 * Used to only do .*x and .*?x, but now it allows
5123 * for )'s, ('s and (?{ ... })'s to be in the way
5124 * of the quantifier and the EXACT-like node. -- japhy
5127 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5129 if (HAS_TEXT(next) || JUMPABLE(next)) {
5131 regnode *text_node = next;
5133 if (! HAS_TEXT(text_node))
5134 FIND_NEXT_IMPT(text_node);
5136 if (! HAS_TEXT(text_node))
5137 ST.c1 = ST.c2 = CHRTEST_VOID;
5139 if ( PL_regkind[OP(text_node)] != EXACT ) {
5140 ST.c1 = ST.c2 = CHRTEST_VOID;
5141 goto assume_ok_easy;
5144 s = (U8*)STRING(text_node);
5146 /* Currently we only get here when
5148 PL_rekind[OP(text_node)] == EXACT
5150 if this changes back then the macro for IS_TEXT and
5151 friends need to change. */
5154 switch (OP(text_node)) {
5155 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5157 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5158 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5159 default: ST.c2 = ST.c1; break;
5162 else { /* UTF_PATTERN */
5163 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5164 STRLEN ulen1, ulen2;
5165 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5166 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5168 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5169 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5171 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5173 0 : UTF8_ALLOW_ANY);
5174 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5176 0 : UTF8_ALLOW_ANY);
5178 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5180 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5185 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5192 ST.c1 = ST.c2 = CHRTEST_VOID;
5197 PL_reginput = locinput;
5200 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5203 locinput = PL_reginput;
5205 if (ST.c1 == CHRTEST_VOID)
5206 goto curly_try_B_min;
5208 ST.oldloc = locinput;
5210 /* set ST.maxpos to the furthest point along the
5211 * string that could possibly match */
5212 if (ST.max == REG_INFTY) {
5213 ST.maxpos = PL_regeol - 1;
5215 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5218 else if (utf8_target) {
5219 int m = ST.max - ST.min;
5220 for (ST.maxpos = locinput;
5221 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5222 ST.maxpos += UTF8SKIP(ST.maxpos);
5225 ST.maxpos = locinput + ST.max - ST.min;
5226 if (ST.maxpos >= PL_regeol)
5227 ST.maxpos = PL_regeol - 1;
5229 goto curly_try_B_min_known;
5233 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5234 locinput = PL_reginput;
5235 if (ST.count < ST.min)
5237 if ((ST.count > ST.min)
5238 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5240 /* A{m,n} must come at the end of the string, there's
5241 * no point in backing off ... */
5243 /* ...except that $ and \Z can match before *and* after
5244 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5245 We may back off by one in this case. */
5246 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5250 goto curly_try_B_max;
5255 case CURLY_B_min_known_fail:
5256 /* failed to find B in a non-greedy match where c1,c2 valid */
5257 if (ST.paren && ST.count)
5258 PL_regoffs[ST.paren].end = -1;
5260 PL_reginput = locinput; /* Could be reset... */
5261 REGCP_UNWIND(ST.cp);
5262 /* Couldn't or didn't -- move forward. */
5263 ST.oldloc = locinput;
5265 locinput += UTF8SKIP(locinput);
5269 curly_try_B_min_known:
5270 /* find the next place where 'B' could work, then call B */
5274 n = (ST.oldloc == locinput) ? 0 : 1;
5275 if (ST.c1 == ST.c2) {
5277 /* set n to utf8_distance(oldloc, locinput) */
5278 while (locinput <= ST.maxpos &&
5279 utf8n_to_uvchr((U8*)locinput,
5280 UTF8_MAXBYTES, &len,
5281 uniflags) != (UV)ST.c1) {
5287 /* set n to utf8_distance(oldloc, locinput) */
5288 while (locinput <= ST.maxpos) {
5290 const UV c = utf8n_to_uvchr((U8*)locinput,
5291 UTF8_MAXBYTES, &len,
5293 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5301 if (ST.c1 == ST.c2) {
5302 while (locinput <= ST.maxpos &&
5303 UCHARAT(locinput) != ST.c1)
5307 while (locinput <= ST.maxpos
5308 && UCHARAT(locinput) != ST.c1
5309 && UCHARAT(locinput) != ST.c2)
5312 n = locinput - ST.oldloc;
5314 if (locinput > ST.maxpos)
5316 /* PL_reginput == oldloc now */
5319 if (regrepeat(rex, ST.A, n, depth) < n)
5322 PL_reginput = locinput;
5323 CURLY_SETPAREN(ST.paren, ST.count);
5324 if (cur_eval && cur_eval->u.eval.close_paren &&
5325 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5328 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5333 case CURLY_B_min_fail:
5334 /* failed to find B in a non-greedy match where c1,c2 invalid */
5335 if (ST.paren && ST.count)
5336 PL_regoffs[ST.paren].end = -1;
5338 REGCP_UNWIND(ST.cp);
5339 /* failed -- move forward one */
5340 PL_reginput = locinput;
5341 if (regrepeat(rex, ST.A, 1, depth)) {
5343 locinput = PL_reginput;
5344 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5345 ST.count > 0)) /* count overflow ? */
5348 CURLY_SETPAREN(ST.paren, ST.count);
5349 if (cur_eval && cur_eval->u.eval.close_paren &&
5350 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5353 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5361 /* a successful greedy match: now try to match B */
5362 if (cur_eval && cur_eval->u.eval.close_paren &&
5363 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5368 if (ST.c1 != CHRTEST_VOID)
5369 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5370 UTF8_MAXBYTES, 0, uniflags)
5371 : (UV) UCHARAT(PL_reginput);
5372 /* If it could work, try it. */
5373 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5374 CURLY_SETPAREN(ST.paren, ST.count);
5375 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5380 case CURLY_B_max_fail:
5381 /* failed to find B in a greedy match */
5382 if (ST.paren && ST.count)
5383 PL_regoffs[ST.paren].end = -1;
5385 REGCP_UNWIND(ST.cp);
5387 if (--ST.count < ST.min)
5389 PL_reginput = locinput = HOPc(locinput, -1);
5390 goto curly_try_B_max;
5397 /* we've just finished A in /(??{A})B/; now continue with B */
5399 st->u.eval.toggle_reg_flags
5400 = cur_eval->u.eval.toggle_reg_flags;
5401 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5403 st->u.eval.prev_rex = rex_sv; /* inner */
5404 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5405 rex = (struct regexp *)SvANY(rex_sv);
5406 rexi = RXi_GET(rex);
5407 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5408 (void)ReREFCNT_inc(rex_sv);
5409 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5411 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5412 PL_reglastparen = &rex->lastparen;
5413 PL_reglastcloseparen = &rex->lastcloseparen;
5415 REGCP_SET(st->u.eval.lastcp);
5416 PL_reginput = locinput;
5418 /* Restore parens of the outer rex without popping the
5420 tmpix = PL_savestack_ix;
5421 PL_savestack_ix = cur_eval->u.eval.lastcp;
5423 PL_savestack_ix = tmpix;
5425 st->u.eval.prev_eval = cur_eval;
5426 cur_eval = cur_eval->u.eval.prev_eval;
5428 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5429 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5430 if ( nochange_depth )
5433 PUSH_YES_STATE_GOTO(EVAL_AB,
5434 st->u.eval.prev_eval->u.eval.B); /* match B */
5437 if (locinput < reginfo->till) {
5438 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5439 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5441 (long)(locinput - PL_reg_starttry),
5442 (long)(reginfo->till - PL_reg_starttry),
5445 sayNO_SILENT; /* Cannot match: too short. */
5447 PL_reginput = locinput; /* put where regtry can find it */
5448 sayYES; /* Success! */
5450 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5452 PerlIO_printf(Perl_debug_log,
5453 "%*s %ssubpattern success...%s\n",
5454 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5455 PL_reginput = locinput; /* put where regtry can find it */
5456 sayYES; /* Success! */
5459 #define ST st->u.ifmatch
5461 case SUSPEND: /* (?>A) */
5463 PL_reginput = locinput;
5466 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5468 goto ifmatch_trivial_fail_test;
5470 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5472 ifmatch_trivial_fail_test:
5474 char * const s = HOPBACKc(locinput, scan->flags);
5479 sw = 1 - cBOOL(ST.wanted);
5483 next = scan + ARG(scan);
5491 PL_reginput = locinput;
5495 ST.logical = logical;
5496 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5498 /* execute body of (?...A) */
5499 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5502 case IFMATCH_A_fail: /* body of (?...A) failed */
5503 ST.wanted = !ST.wanted;
5506 case IFMATCH_A: /* body of (?...A) succeeded */
5508 sw = cBOOL(ST.wanted);
5510 else if (!ST.wanted)
5513 if (OP(ST.me) == SUSPEND)
5514 locinput = PL_reginput;
5516 locinput = PL_reginput = st->locinput;
5517 nextchr = UCHARAT(locinput);
5519 scan = ST.me + ARG(ST.me);
5522 continue; /* execute B */
5527 next = scan + ARG(scan);
5532 reginfo->cutpoint = PL_regeol;
5535 PL_reginput = locinput;
5537 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5538 PUSH_STATE_GOTO(COMMIT_next,next);
5540 case COMMIT_next_fail:
5547 #define ST st->u.mark
5549 ST.prev_mark = mark_state;
5550 ST.mark_name = sv_commit = sv_yes_mark
5551 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5553 ST.mark_loc = PL_reginput = locinput;
5554 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5556 case MARKPOINT_next:
5557 mark_state = ST.prev_mark;
5560 case MARKPOINT_next_fail:
5561 if (popmark && sv_eq(ST.mark_name,popmark))
5563 if (ST.mark_loc > startpoint)
5564 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5565 popmark = NULL; /* we found our mark */
5566 sv_commit = ST.mark_name;
5569 PerlIO_printf(Perl_debug_log,
5570 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5571 REPORT_CODE_OFF+depth*2, "",
5572 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5575 mark_state = ST.prev_mark;
5576 sv_yes_mark = mark_state ?
5577 mark_state->u.mark.mark_name : NULL;
5581 PL_reginput = locinput;
5583 /* (*SKIP) : if we fail we cut here*/
5584 ST.mark_name = NULL;
5585 ST.mark_loc = locinput;
5586 PUSH_STATE_GOTO(SKIP_next,next);
5588 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5589 otherwise do nothing. Meaning we need to scan
5591 regmatch_state *cur = mark_state;
5592 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5595 if ( sv_eq( cur->u.mark.mark_name,
5598 ST.mark_name = find;
5599 PUSH_STATE_GOTO( SKIP_next, next );
5601 cur = cur->u.mark.prev_mark;
5604 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5606 case SKIP_next_fail:
5608 /* (*CUT:NAME) - Set up to search for the name as we
5609 collapse the stack*/
5610 popmark = ST.mark_name;
5612 /* (*CUT) - No name, we cut here.*/
5613 if (ST.mark_loc > startpoint)
5614 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5615 /* but we set sv_commit to latest mark_name if there
5616 is one so they can test to see how things lead to this
5619 sv_commit=mark_state->u.mark.mark_name;
5627 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5629 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5632 U8 folded[UTF8_MAXBYTES_CASE+1];
5634 const char * const l = locinput;
5635 char *e = PL_regeol;
5636 to_uni_fold(n, folded, &foldlen);
5638 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5639 l, &e, 0, utf8_target)) {
5644 nextchr = UCHARAT(locinput);
5647 if ((n=is_LNBREAK(locinput,utf8_target))) {
5649 nextchr = UCHARAT(locinput);
5654 #define CASE_CLASS(nAmE) \
5656 if ((n=is_##nAmE(locinput,utf8_target))) { \
5658 nextchr = UCHARAT(locinput); \
5663 if ((n=is_##nAmE(locinput,utf8_target))) { \
5666 locinput += UTF8SKIP(locinput); \
5667 nextchr = UCHARAT(locinput); \
5672 CASE_CLASS(HORIZWS);
5676 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5677 PTR2UV(scan), OP(scan));
5678 Perl_croak(aTHX_ "regexp memory corruption");
5682 /* switch break jumps here */
5683 scan = next; /* prepare to execute the next op and ... */
5684 continue; /* ... jump back to the top, reusing st */
5688 /* push a state that backtracks on success */
5689 st->u.yes.prev_yes_state = yes_state;
5693 /* push a new regex state, then continue at scan */
5695 regmatch_state *newst;
5698 regmatch_state *cur = st;
5699 regmatch_state *curyes = yes_state;
5701 regmatch_slab *slab = PL_regmatch_slab;
5702 for (;curd > -1;cur--,curd--) {
5703 if (cur < SLAB_FIRST(slab)) {
5705 cur = SLAB_LAST(slab);
5707 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5708 REPORT_CODE_OFF + 2 + depth * 2,"",
5709 curd, PL_reg_name[cur->resume_state],
5710 (curyes == cur) ? "yes" : ""
5713 curyes = cur->u.yes.prev_yes_state;
5716 DEBUG_STATE_pp("push")
5719 st->locinput = locinput;
5721 if (newst > SLAB_LAST(PL_regmatch_slab))
5722 newst = S_push_slab(aTHX);
5723 PL_regmatch_state = newst;
5725 locinput = PL_reginput;
5726 nextchr = UCHARAT(locinput);
5734 * We get here only if there's trouble -- normally "case END" is
5735 * the terminating point.
5737 Perl_croak(aTHX_ "corrupted regexp pointers");
5743 /* we have successfully completed a subexpression, but we must now
5744 * pop to the state marked by yes_state and continue from there */
5745 assert(st != yes_state);
5747 while (st != yes_state) {
5749 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5750 PL_regmatch_slab = PL_regmatch_slab->prev;
5751 st = SLAB_LAST(PL_regmatch_slab);
5755 DEBUG_STATE_pp("pop (no final)");
5757 DEBUG_STATE_pp("pop (yes)");
5763 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5764 || yes_state > SLAB_LAST(PL_regmatch_slab))
5766 /* not in this slab, pop slab */
5767 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5768 PL_regmatch_slab = PL_regmatch_slab->prev;
5769 st = SLAB_LAST(PL_regmatch_slab);
5771 depth -= (st - yes_state);
5774 yes_state = st->u.yes.prev_yes_state;
5775 PL_regmatch_state = st;
5778 locinput= st->locinput;
5779 nextchr = UCHARAT(locinput);
5781 state_num = st->resume_state + no_final;
5782 goto reenter_switch;
5785 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5786 PL_colors[4], PL_colors[5]));
5788 if (PL_reg_eval_set) {
5789 /* each successfully executed (?{...}) block does the equivalent of
5790 * local $^R = do {...}
5791 * When popping the save stack, all these locals would be undone;
5792 * bypass this by setting the outermost saved $^R to the latest
5794 if (oreplsv != GvSV(PL_replgv))
5795 sv_setsv(oreplsv, GvSV(PL_replgv));
5802 PerlIO_printf(Perl_debug_log,
5803 "%*s %sfailed...%s\n",
5804 REPORT_CODE_OFF+depth*2, "",
5805 PL_colors[4], PL_colors[5])
5817 /* there's a previous state to backtrack to */
5819 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5820 PL_regmatch_slab = PL_regmatch_slab->prev;
5821 st = SLAB_LAST(PL_regmatch_slab);
5823 PL_regmatch_state = st;
5824 locinput= st->locinput;
5825 nextchr = UCHARAT(locinput);
5827 DEBUG_STATE_pp("pop");
5829 if (yes_state == st)
5830 yes_state = st->u.yes.prev_yes_state;
5832 state_num = st->resume_state + 1; /* failure = success + 1 */
5833 goto reenter_switch;
5838 if (rex->intflags & PREGf_VERBARG_SEEN) {
5839 SV *sv_err = get_sv("REGERROR", 1);
5840 SV *sv_mrk = get_sv("REGMARK", 1);
5842 sv_commit = &PL_sv_no;
5844 sv_yes_mark = &PL_sv_yes;
5847 sv_commit = &PL_sv_yes;
5848 sv_yes_mark = &PL_sv_no;
5850 sv_setsv(sv_err, sv_commit);
5851 sv_setsv(sv_mrk, sv_yes_mark);
5854 /* clean up; in particular, free all slabs above current one */
5855 LEAVE_SCOPE(oldsave);
5861 - regrepeat - repeatedly match something simple, report how many
5864 * [This routine now assumes that it will only match on things of length 1.
5865 * That was true before, but now we assume scan - reginput is the count,
5866 * rather than incrementing count on every character. [Er, except utf8.]]
5869 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5872 register char *scan;
5874 register char *loceol = PL_regeol;
5875 register I32 hardcount = 0;
5876 register bool utf8_target = PL_reg_match_utf8;
5879 PERL_UNUSED_ARG(depth);
5882 PERL_ARGS_ASSERT_REGREPEAT;
5885 if (max == REG_INFTY)
5887 else if (max < loceol - scan)
5888 loceol = scan + max;
5893 while (scan < loceol && hardcount < max && *scan != '\n') {
5894 scan += UTF8SKIP(scan);
5898 while (scan < loceol && *scan != '\n')
5905 while (scan < loceol && hardcount < max) {
5906 scan += UTF8SKIP(scan);
5917 /* To get here, EXACTish nodes must have *byte* length == 1. That
5918 * means they match only characters in the string that can be expressed
5919 * as a single byte. For non-utf8 strings, that means a simple match.
5920 * For utf8 strings, the character matched must be an invariant, or
5921 * downgradable to a single byte. The pattern's utf8ness is
5922 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5923 * it is, it's an invariant */
5926 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5928 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5929 while (scan < loceol && UCHARAT(scan) == c) {
5935 /* Here, the string is utf8, and the pattern char is different
5936 * in utf8 than not, so can't compare them directly. Outside the
5937 * loop, find find the two utf8 bytes that represent c, and then
5938 * look for those in sequence in the utf8 string */
5939 U8 high = UTF8_TWO_BYTE_HI(c);
5940 U8 low = UTF8_TWO_BYTE_LO(c);
5943 while (hardcount < max
5944 && scan + 1 < loceol
5945 && UCHARAT(scan) == high
5946 && UCHARAT(scan + 1) == low)
5954 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5958 PL_reg_flags |= RF_tainted;
5959 utf8_flags = FOLDEQ_UTF8_LOCALE;
5966 /* The comments for the EXACT case above apply as well to these fold
5971 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5973 if (utf8_target) { /* Use full Unicode fold matching */
5974 char *tmpeol = loceol;
5975 while (hardcount < max
5976 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
5977 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
5984 /* XXX Note that the above handles properly the German sharp s in
5985 * the pattern matching ss in the string. But it doesn't handle
5986 * properly cases where the string contains say 'LIGATURE ff' and
5987 * the pattern is 'f+'. This would require, say, a new function or
5988 * revised interface to foldEQ_utf8(), in which the maximum number
5989 * of characters to match could be passed and it would return how
5990 * many actually did. This is just one of many cases where
5991 * multi-char folds don't work properly, and so the fix is being
5997 /* Here, the string isn't utf8 and c is a single byte; and either
5998 * the pattern isn't utf8 or c is an invariant, so its utf8ness
5999 * doesn't affect c. Can just do simple comparisons for exact or
6002 case EXACTF: folded = PL_fold[c]; break;
6004 case EXACTFU: folded = PL_fold_latin1[c]; break;
6005 case EXACTFL: folded = PL_fold_locale[c]; break;
6006 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6008 while (scan < loceol &&
6009 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6017 if (utf8_target || OP(p) == ANYOFV) {
6020 inclasslen = loceol - scan;
6021 while (hardcount < max
6022 && ((inclasslen = loceol - scan) > 0)
6023 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6029 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6037 LOAD_UTF8_CHARCLASS_ALNUM();
6038 while (hardcount < max && scan < loceol &&
6039 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6041 scan += UTF8SKIP(scan);
6045 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6053 while (scan < loceol && isALNUM((U8) *scan)) {
6058 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6063 PL_reg_flags |= RF_tainted;
6066 while (hardcount < max && scan < loceol &&
6067 isALNUM_LC_utf8((U8*)scan)) {
6068 scan += UTF8SKIP(scan);
6072 while (scan < loceol && isALNUM_LC(*scan))
6082 LOAD_UTF8_CHARCLASS_ALNUM();
6083 while (hardcount < max && scan < loceol &&
6084 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6086 scan += UTF8SKIP(scan);
6090 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6097 goto utf8_Nwordchar;
6098 while (scan < loceol && ! isALNUM((U8) *scan)) {
6104 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6105 scan += UTF8SKIP(scan);
6109 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6115 PL_reg_flags |= RF_tainted;
6118 while (hardcount < max && scan < loceol &&
6119 !isALNUM_LC_utf8((U8*)scan)) {
6120 scan += UTF8SKIP(scan);
6124 while (scan < loceol && !isALNUM_LC(*scan))
6134 LOAD_UTF8_CHARCLASS_SPACE();
6135 while (hardcount < max && scan < loceol &&
6137 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6139 scan += UTF8SKIP(scan);
6145 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6154 while (scan < loceol && isSPACE((U8) *scan)) {
6159 while (scan < loceol && isSPACE_A((U8) *scan)) {
6164 PL_reg_flags |= RF_tainted;
6167 while (hardcount < max && scan < loceol &&
6168 isSPACE_LC_utf8((U8*)scan)) {
6169 scan += UTF8SKIP(scan);
6173 while (scan < loceol && isSPACE_LC(*scan))
6183 LOAD_UTF8_CHARCLASS_SPACE();
6184 while (hardcount < max && scan < loceol &&
6186 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6188 scan += UTF8SKIP(scan);
6194 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6203 while (scan < loceol && ! isSPACE((U8) *scan)) {
6209 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6210 scan += UTF8SKIP(scan);
6214 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6220 PL_reg_flags |= RF_tainted;
6223 while (hardcount < max && scan < loceol &&
6224 !isSPACE_LC_utf8((U8*)scan)) {
6225 scan += UTF8SKIP(scan);
6229 while (scan < loceol && !isSPACE_LC(*scan))
6236 LOAD_UTF8_CHARCLASS_DIGIT();
6237 while (hardcount < max && scan < loceol &&
6238 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6239 scan += UTF8SKIP(scan);
6243 while (scan < loceol && isDIGIT(*scan))
6248 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6253 PL_reg_flags |= RF_tainted;
6256 while (hardcount < max && scan < loceol &&
6257 isDIGIT_LC_utf8((U8*)scan)) {
6258 scan += UTF8SKIP(scan);
6262 while (scan < loceol && isDIGIT_LC(*scan))
6269 LOAD_UTF8_CHARCLASS_DIGIT();
6270 while (hardcount < max && scan < loceol &&
6271 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6272 scan += UTF8SKIP(scan);
6276 while (scan < loceol && !isDIGIT(*scan))
6282 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6283 scan += UTF8SKIP(scan);
6287 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6293 PL_reg_flags |= RF_tainted;
6296 while (hardcount < max && scan < loceol &&
6297 !isDIGIT_LC_utf8((U8*)scan)) {
6298 scan += UTF8SKIP(scan);
6302 while (scan < loceol && !isDIGIT_LC(*scan))
6309 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6315 LNBREAK can match two latin chars, which is ok,
6316 because we have a null terminated string, but we
6317 have to use hardcount in this situation
6319 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6328 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6333 while (scan < loceol && is_HORIZWS_latin1(scan))
6340 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6341 scan += UTF8SKIP(scan);
6345 while (scan < loceol && !is_HORIZWS_latin1(scan))
6353 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6358 while (scan < loceol && is_VERTWS_latin1(scan))
6366 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6367 scan += UTF8SKIP(scan);
6371 while (scan < loceol && !is_VERTWS_latin1(scan))
6377 default: /* Called on something of 0 width. */
6378 break; /* So match right here or not at all. */
6384 c = scan - PL_reginput;
6388 GET_RE_DEBUG_FLAGS_DECL;
6390 SV * const prop = sv_newmortal();
6391 regprop(prog, prop, p);
6392 PerlIO_printf(Perl_debug_log,
6393 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6394 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6402 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6404 - regclass_swash - prepare the utf8 swash
6408 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6414 RXi_GET_DECL(prog,progi);
6415 const struct reg_data * const data = prog ? progi->data : NULL;
6417 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6419 assert(ANYOF_NONBITMAP(node));
6421 if (data && data->count) {
6422 const U32 n = ARG(node);
6424 if (data->what[n] == 's') {
6425 SV * const rv = MUTABLE_SV(data->data[n]);
6426 AV * const av = MUTABLE_AV(SvRV(rv));
6427 SV **const ary = AvARRAY(av);
6430 /* See the end of regcomp.c:S_regclass() for
6431 * documentation of these array elements. */
6434 a = SvROK(ary[1]) ? &ary[1] : NULL;
6435 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6439 else if (si && doinit) {
6440 sw = swash_init("utf8", "", si, 1, 0);
6441 (void)av_store(av, 1, sw);
6458 - reginclass - determine if a character falls into a character class
6460 n is the ANYOF regnode
6461 p is the target string
6462 lenp is pointer to the maximum number of bytes of how far to go in p
6463 (This is assumed wthout checking to always be at least the current
6465 utf8_target tells whether p is in UTF-8.
6467 Returns true if matched; false otherwise. If lenp is not NULL, on return
6468 from a successful match, the value it points to will be updated to how many
6469 bytes in p were matched. If there was no match, the value is undefined,
6470 possibly changed from the input.
6472 Note that this can be a synthetic start class, a combination of various
6473 nodes, so things you think might be mutually exclusive, such as locale,
6474 aren't. It can match both locale and non-locale
6479 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6482 const char flags = ANYOF_FLAGS(n);
6488 PERL_ARGS_ASSERT_REGINCLASS;
6490 /* If c is not already the code point, get it */
6491 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6492 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6493 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6494 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6495 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6496 * UTF8_ALLOW_FFFF */
6497 if (c_len == (STRLEN)-1)
6498 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6504 /* Use passed in max length, or one character if none passed in or less
6505 * than one character. And assume will match just one character. This is
6506 * overwritten later if matched more. */
6508 maxlen = (*lenp > c_len) ? *lenp : c_len;
6516 /* If this character is potentially in the bitmap, check it */
6518 if (ANYOF_BITMAP_TEST(n, c))
6520 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6527 else if (flags & ANYOF_LOCALE) {
6528 PL_reg_flags |= RF_tainted;
6530 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6531 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6535 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6536 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6537 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6538 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6539 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6540 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6541 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6542 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6543 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6544 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6545 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6546 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6547 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6548 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6549 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6550 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6551 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6552 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6553 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6554 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6555 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6556 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6557 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6558 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6559 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6560 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6561 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6562 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6563 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6564 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6565 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6566 ) /* How's that for a conditional? */
6573 /* If the bitmap didn't (or couldn't) match, and something outside the
6574 * bitmap could match, try that. Locale nodes specifiy completely the
6575 * behavior of code points in the bit map (otherwise, a utf8 target would
6576 * cause them to be treated as Unicode and not locale), except in
6577 * the very unlikely event when this node is a synthetic start class, which
6578 * could be a combination of locale and non-locale nodes. So allow locale
6579 * to match for the synthetic start class, which will give a false
6580 * positive that will be resolved when the match is done again as not part
6581 * of the synthetic start class */
6583 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6584 match = TRUE; /* Everything above 255 matches */
6586 else if (ANYOF_NONBITMAP(n)
6587 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6590 || (! (flags & ANYOF_LOCALE))
6591 || (flags & ANYOF_IS_SYNTHETIC)))))
6594 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6602 /* Not utf8. Convert as much of the string as available up
6603 * to the limit of how far the (single) character in the
6604 * pattern can possibly match (no need to go further). If
6605 * the node is a straight ANYOF or not folding, it can't
6606 * match more than one. Otherwise, It can match up to how
6607 * far a single char can fold to. Since not utf8, each
6608 * character is a single byte, so the max it can be in
6609 * bytes is the same as the max it can be in characters */
6610 STRLEN len = (OP(n) == ANYOF
6611 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6613 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6615 : UTF8_MAX_FOLD_CHAR_EXPAND;
6616 utf8_p = bytes_to_utf8(p, &len);
6619 if (swash_fetch(sw, utf8_p, TRUE))
6621 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6623 /* Here, we need to test if the fold of the target string
6624 * matches. The non-multi char folds have all been moved to
6625 * the compilation phase, and the multi-char folds have
6626 * been stored by regcomp into 'av'; we linearly check to
6627 * see if any match the target string (folded). We know
6628 * that the originals were each one character, but we don't
6629 * currently know how many characters/bytes each folded to,
6630 * except we do know that there are small limits imposed by
6631 * Unicode. XXX A performance enhancement would be to have
6632 * regcomp.c store the max number of chars/bytes that are
6633 * in an av entry, as, say the 0th element. Even better
6634 * would be to have a hash of the few characters that can
6635 * start a multi-char fold to the max number of chars of
6638 * If there is a match, we will need to advance (if lenp is
6639 * specified) the match pointer in the target string. But
6640 * what we are comparing here isn't that string directly,
6641 * but its fold, whose length may differ from the original.
6642 * As we go along in constructing the fold, therefore, we
6643 * create a map so that we know how many bytes in the
6644 * source to advance given that we have matched a certain
6645 * number of bytes in the fold. This map is stored in
6646 * 'map_fold_len_back'. Let n mean the number of bytes in
6647 * the fold of the first character that we are folding.
6648 * Then map_fold_len_back[n] is set to the number of bytes
6649 * in that first character. Similarly let m be the
6650 * corresponding number for the second character to be
6651 * folded. Then map_fold_len_back[n+m] is set to the
6652 * number of bytes occupied by the first two source
6653 * characters. ... */
6654 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6655 U8 folded[UTF8_MAXBYTES_CASE+1];
6656 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6657 STRLEN total_foldlen = 0; /* num bytes in fold of all
6660 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6662 /* Here, only need to fold the first char of the target
6663 * string. It the source wasn't utf8, is 1 byte long */
6664 to_utf8_fold(utf8_p, folded, &foldlen);
6665 total_foldlen = foldlen;
6666 map_fold_len_back[foldlen] = (utf8_target)
6672 /* Here, need to fold more than the first char. Do so
6673 * up to the limits */
6674 U8* source_ptr = utf8_p; /* The source for the fold
6677 U8* folded_ptr = folded;
6678 U8* e = utf8_p + maxlen; /* Can't go beyond last
6679 available byte in the
6683 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6687 /* Fold the next character */
6688 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6689 STRLEN this_char_foldlen;
6690 to_utf8_fold(source_ptr,
6692 &this_char_foldlen);
6694 /* Bail if it would exceed the byte limit for
6695 * folding a single char. */
6696 if (this_char_foldlen + folded_ptr - folded >
6702 /* Add the fold of this character */
6703 Copy(this_char_folded,
6707 source_ptr += UTF8SKIP(source_ptr);
6708 folded_ptr += this_char_foldlen;
6709 total_foldlen = folded_ptr - folded;
6711 /* Create map from the number of bytes in the fold
6712 * back to the number of bytes in the source. If
6713 * the source isn't utf8, the byte count is just
6714 * the number of characters so far */
6715 map_fold_len_back[total_foldlen]
6717 ? source_ptr - utf8_p
6724 /* Do the linear search to see if the fold is in the list
6725 * of multi-char folds. */
6728 for (i = 0; i <= av_len(av); i++) {
6729 SV* const sv = *av_fetch(av, i, FALSE);
6731 const char * const s = SvPV_const(sv, len);
6733 if (len <= total_foldlen
6734 && memEQ(s, (char*)folded, len)
6736 /* If 0, means matched a partial char. See
6738 && map_fold_len_back[len])
6741 /* Advance the target string ptr to account for
6742 * this fold, but have to translate from the
6743 * folded length to the corresponding source
6746 *lenp = map_fold_len_back[len];
6755 /* If we allocated a string above, free it */
6756 if (! utf8_target) Safefree(utf8_p);
6761 return (flags & ANYOF_INVERT) ? !match : match;
6765 S_reghop3(U8 *s, I32 off, const U8* lim)
6769 PERL_ARGS_ASSERT_REGHOP3;
6772 while (off-- && s < lim) {
6773 /* XXX could check well-formedness here */
6778 while (off++ && s > lim) {
6780 if (UTF8_IS_CONTINUED(*s)) {
6781 while (s > lim && UTF8_IS_CONTINUATION(*s))
6784 /* XXX could check well-formedness here */
6791 /* there are a bunch of places where we use two reghop3's that should
6792 be replaced with this routine. but since thats not done yet
6793 we ifdef it out - dmq
6796 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6800 PERL_ARGS_ASSERT_REGHOP4;
6803 while (off-- && s < rlim) {
6804 /* XXX could check well-formedness here */
6809 while (off++ && s > llim) {
6811 if (UTF8_IS_CONTINUED(*s)) {
6812 while (s > llim && UTF8_IS_CONTINUATION(*s))
6815 /* XXX could check well-formedness here */
6823 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6827 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6830 while (off-- && s < lim) {
6831 /* XXX could check well-formedness here */
6838 while (off++ && s > lim) {
6840 if (UTF8_IS_CONTINUED(*s)) {
6841 while (s > lim && UTF8_IS_CONTINUATION(*s))
6844 /* XXX could check well-formedness here */
6853 restore_pos(pTHX_ void *arg)
6856 regexp * const rex = (regexp *)arg;
6857 if (PL_reg_eval_set) {
6858 if (PL_reg_oldsaved) {
6859 rex->subbeg = PL_reg_oldsaved;
6860 rex->sublen = PL_reg_oldsavedlen;
6861 #ifdef PERL_OLD_COPY_ON_WRITE
6862 rex->saved_copy = PL_nrs;
6864 RXp_MATCH_COPIED_on(rex);
6866 PL_reg_magic->mg_len = PL_reg_oldpos;
6867 PL_reg_eval_set = 0;
6868 PL_curpm = PL_reg_oldcurpm;
6873 S_to_utf8_substr(pTHX_ register regexp *prog)
6877 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6880 if (prog->substrs->data[i].substr
6881 && !prog->substrs->data[i].utf8_substr) {
6882 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6883 prog->substrs->data[i].utf8_substr = sv;
6884 sv_utf8_upgrade(sv);
6885 if (SvVALID(prog->substrs->data[i].substr)) {
6886 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6887 if (flags & FBMcf_TAIL) {
6888 /* Trim the trailing \n that fbm_compile added last
6890 SvCUR_set(sv, SvCUR(sv) - 1);
6891 /* Whilst this makes the SV technically "invalid" (as its
6892 buffer is no longer followed by "\0") when fbm_compile()
6893 adds the "\n" back, a "\0" is restored. */
6895 fbm_compile(sv, flags);
6897 if (prog->substrs->data[i].substr == prog->check_substr)
6898 prog->check_utf8 = sv;
6904 S_to_byte_substr(pTHX_ register regexp *prog)
6909 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6912 if (prog->substrs->data[i].utf8_substr
6913 && !prog->substrs->data[i].substr) {
6914 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6915 if (sv_utf8_downgrade(sv, TRUE)) {
6916 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6918 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6919 if (flags & FBMcf_TAIL) {
6920 /* Trim the trailing \n that fbm_compile added last
6922 SvCUR_set(sv, SvCUR(sv) - 1);
6924 fbm_compile(sv, flags);
6930 prog->substrs->data[i].substr = sv;
6931 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6932 prog->check_substr = sv;
6939 * c-indentation-style: bsd
6941 * indent-tabs-mode: t
6944 * ex: set ts=8 sts=4 sw=4 noet: