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. */
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)) {
3629 /* One more case for the sharp s:
3630 * pack("U0U*", 0xDF) =~ /ss/i,
3631 * the 0xC3 0x9F are the UTF-8
3632 * byte sequence for the U+00DF. */
3634 if (!(utf8_target &&
3635 toLOWER(s[0]) == 's' &&
3637 toLOWER(s[1]) == 's' &&
3644 nextchr = UCHARAT(locinput);
3648 /* Neither the target nor the pattern are utf8 */
3649 if (UCHARAT(s) != nextchr &&
3650 UCHARAT(s) != fold_array[nextchr])
3654 if (PL_regeol - locinput < ln)
3656 if (ln > 1 && ! folder(s, locinput, ln))
3659 nextchr = UCHARAT(locinput);
3663 /* XXX Could improve efficiency by separating these all out using a
3664 * macro or in-line function. At that point regcomp.c would no longer
3665 * have to set the FLAGS fields of these */
3668 PL_reg_flags |= RF_tainted;
3676 /* was last char in word? */
3677 if (utf8_target && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET) {
3678 if (locinput == PL_bostr)
3681 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3683 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3685 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3686 ln = isALNUM_uni(ln);
3687 LOAD_UTF8_CHARCLASS_ALNUM();
3688 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3691 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3692 n = isALNUM_LC_utf8((U8*)locinput);
3697 /* Here the string isn't utf8, or is utf8 and only ascii
3698 * characters are to match \w. In the latter case looking at
3699 * the byte just prior to the current one may be just the final
3700 * byte of a multi-byte character. This is ok. There are two
3702 * 1) it is a single byte character, and then the test is doing
3703 * just what it's supposed to.
3704 * 2) it is a multi-byte character, in which case the final
3705 * byte is never mistakable for ASCII, and so the test
3706 * will say it is not a word character, which is the
3707 * correct answer. */
3708 ln = (locinput != PL_bostr) ?
3709 UCHARAT(locinput - 1) : '\n';
3710 switch (FLAGS(scan)) {
3711 case REGEX_UNICODE_CHARSET:
3712 ln = isWORDCHAR_L1(ln);
3713 n = isWORDCHAR_L1(nextchr);
3715 case REGEX_LOCALE_CHARSET:
3716 ln = isALNUM_LC(ln);
3717 n = isALNUM_LC(nextchr);
3719 case REGEX_DEPENDS_CHARSET:
3721 n = isALNUM(nextchr);
3723 case REGEX_ASCII_RESTRICTED_CHARSET:
3724 ln = isWORDCHAR_A(ln);
3725 n = isWORDCHAR_A(nextchr);
3728 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3732 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3734 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3739 if (utf8_target || state_num == ANYOFV) {
3740 STRLEN inclasslen = PL_regeol - locinput;
3741 if (locinput >= PL_regeol)
3744 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3746 locinput += inclasslen;
3747 nextchr = UCHARAT(locinput);
3752 nextchr = UCHARAT(locinput);
3753 if (!nextchr && locinput >= PL_regeol)
3755 if (!REGINCLASS(rex, scan, (U8*)locinput))
3757 nextchr = UCHARAT(++locinput);
3761 /* Special char classes - The defines start on line 129 or so */
3762 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3763 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3764 ALNUMU, NALNUMU, isWORDCHAR_L1,
3765 ALNUMA, NALNUMA, isWORDCHAR_A,
3768 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3769 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3770 SPACEU, NSPACEU, isSPACE_L1,
3771 SPACEA, NSPACEA, isSPACE_A,
3774 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3775 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3776 DIGITA, NDIGITA, isDIGIT_A,
3779 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3780 a Unicode extended Grapheme Cluster */
3781 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3782 extended Grapheme Cluster is:
3785 | Prepend* Begin Extend*
3788 Begin is (Hangul-syllable | ! Control)
3789 Extend is (Grapheme_Extend | Spacing_Mark)
3790 Control is [ GCB_Control CR LF ]
3792 The discussion below shows how the code for CLUMP is derived
3793 from this regex. Note that most of these concepts are from
3794 property values of the Grapheme Cluster Boundary (GCB) property.
3795 No code point can have multiple property values for a given
3796 property. Thus a code point in Prepend can't be in Control, but
3797 it must be in !Control. This is why Control above includes
3798 GCB_Control plus CR plus LF. The latter two are used in the GCB
3799 property separately, and so can't be in GCB_Control, even though
3800 they logically are controls. Control is not the same as gc=cc,
3801 but includes format and other characters as well.
3803 The Unicode definition of Hangul-syllable is:
3805 | (L* ( ( V | LV ) V* | LVT ) T*)
3808 Each of these is a value for the GCB property, and hence must be
3809 disjoint, so the order they are tested is immaterial, so the
3810 above can safely be changed to
3813 | (L* ( LVT | ( V | LV ) V*) T*)
3815 The last two terms can be combined like this:
3817 | (( LVT | ( V | LV ) V*) T*))
3819 And refactored into this:
3820 L* (L | LVT T* | V V* T* | LV V* T*)
3822 That means that if we have seen any L's at all we can quit
3823 there, but if the next character is a LVT, a V or and LV we
3826 There is a subtlety with Prepend* which showed up in testing.
3827 Note that the Begin, and only the Begin is required in:
3828 | Prepend* Begin Extend*
3829 Also, Begin contains '! Control'. A Prepend must be a '!
3830 Control', which means it must be a Begin. What it comes down to
3831 is that if we match Prepend* and then find no suitable Begin
3832 afterwards, that if we backtrack the last Prepend, that one will
3833 be a suitable Begin.
3836 if (locinput >= PL_regeol)
3838 if (! utf8_target) {
3840 /* Match either CR LF or '.', as all the other possibilities
3842 locinput++; /* Match the . or CR */
3844 && locinput < PL_regeol
3845 && UCHARAT(locinput) == '\n') locinput++;
3849 /* Utf8: See if is ( CR LF ); already know that locinput <
3850 * PL_regeol, so locinput+1 is in bounds */
3851 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3855 /* In case have to backtrack to beginning, then match '.' */
3856 char *starting = locinput;
3858 /* In case have to backtrack the last prepend */
3859 char *previous_prepend = 0;
3861 LOAD_UTF8_CHARCLASS_GCB();
3863 /* Match (prepend)* */
3864 while (locinput < PL_regeol
3865 && swash_fetch(PL_utf8_X_prepend,
3866 (U8*)locinput, utf8_target))
3868 previous_prepend = locinput;
3869 locinput += UTF8SKIP(locinput);
3872 /* As noted above, if we matched a prepend character, but
3873 * the next thing won't match, back off the last prepend we
3874 * matched, as it is guaranteed to match the begin */
3875 if (previous_prepend
3876 && (locinput >= PL_regeol
3877 || ! swash_fetch(PL_utf8_X_begin,
3878 (U8*)locinput, utf8_target)))
3880 locinput = previous_prepend;
3883 /* Note that here we know PL_regeol > locinput, as we
3884 * tested that upon input to this switch case, and if we
3885 * moved locinput forward, we tested the result just above
3886 * and it either passed, or we backed off so that it will
3888 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3890 /* Here did not match the required 'Begin' in the
3891 * second term. So just match the very first
3892 * character, the '.' of the final term of the regex */
3893 locinput = starting + UTF8SKIP(starting);
3896 /* Here is the beginning of a character that can have
3897 * an extender. It is either a hangul syllable, or a
3899 if (swash_fetch(PL_utf8_X_non_hangul,
3900 (U8*)locinput, utf8_target))
3903 /* Here not a Hangul syllable, must be a
3904 * ('! * Control') */
3905 locinput += UTF8SKIP(locinput);
3908 /* Here is a Hangul syllable. It can be composed
3909 * of several individual characters. One
3910 * possibility is T+ */
3911 if (swash_fetch(PL_utf8_X_T,
3912 (U8*)locinput, utf8_target))
3914 while (locinput < PL_regeol
3915 && swash_fetch(PL_utf8_X_T,
3916 (U8*)locinput, utf8_target))
3918 locinput += UTF8SKIP(locinput);
3922 /* Here, not T+, but is a Hangul. That means
3923 * it is one of the others: L, LV, LVT or V,
3925 * L* (L | LVT T* | V V* T* | LV V* T*) */
3928 while (locinput < PL_regeol
3929 && swash_fetch(PL_utf8_X_L,
3930 (U8*)locinput, utf8_target))
3932 locinput += UTF8SKIP(locinput);
3935 /* Here, have exhausted L*. If the next
3936 * character is not an LV, LVT nor V, it means
3937 * we had to have at least one L, so matches L+
3938 * in the original equation, we have a complete
3939 * hangul syllable. Are done. */
3941 if (locinput < PL_regeol
3942 && swash_fetch(PL_utf8_X_LV_LVT_V,
3943 (U8*)locinput, utf8_target))
3946 /* Otherwise keep going. Must be LV, LVT
3947 * or V. See if LVT */
3948 if (swash_fetch(PL_utf8_X_LVT,
3949 (U8*)locinput, utf8_target))
3951 locinput += UTF8SKIP(locinput);
3954 /* Must be V or LV. Take it, then
3956 locinput += UTF8SKIP(locinput);
3957 while (locinput < PL_regeol
3958 && swash_fetch(PL_utf8_X_V,
3959 (U8*)locinput, utf8_target))
3961 locinput += UTF8SKIP(locinput);
3965 /* And any of LV, LVT, or V can be followed
3967 while (locinput < PL_regeol
3968 && swash_fetch(PL_utf8_X_T,
3972 locinput += UTF8SKIP(locinput);
3978 /* Match any extender */
3979 while (locinput < PL_regeol
3980 && swash_fetch(PL_utf8_X_extend,
3981 (U8*)locinput, utf8_target))
3983 locinput += UTF8SKIP(locinput);
3987 if (locinput > PL_regeol) sayNO;
3989 nextchr = UCHARAT(locinput);
3993 { /* The capture buffer cases. The ones beginning with N for the
3994 named buffers just convert to the equivalent numbered and
3995 pretend they were called as the corresponding numbered buffer
3997 /* don't initialize these in the declaration, it makes C++
4002 const U8 *fold_array;
4005 PL_reg_flags |= RF_tainted;
4006 folder = foldEQ_locale;
4007 fold_array = PL_fold_locale;
4009 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4013 folder = foldEQ_latin1;
4014 fold_array = PL_fold_latin1;
4016 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4020 folder = foldEQ_latin1;
4021 fold_array = PL_fold_latin1;
4023 utf8_fold_flags = 0;
4028 fold_array = PL_fold;
4030 utf8_fold_flags = 0;
4037 utf8_fold_flags = 0;
4040 /* For the named back references, find the corresponding buffer
4042 n = reg_check_named_buff_matched(rex,scan);
4047 goto do_nref_ref_common;
4050 PL_reg_flags |= RF_tainted;
4051 folder = foldEQ_locale;
4052 fold_array = PL_fold_locale;
4053 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4057 folder = foldEQ_latin1;
4058 fold_array = PL_fold_latin1;
4059 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4063 folder = foldEQ_latin1;
4064 fold_array = PL_fold_latin1;
4065 utf8_fold_flags = 0;
4070 fold_array = PL_fold;
4071 utf8_fold_flags = 0;
4077 utf8_fold_flags = 0;
4081 n = ARG(scan); /* which paren pair */
4084 ln = PL_regoffs[n].start;
4085 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4086 if (*PL_reglastparen < n || ln == -1)
4087 sayNO; /* Do not match unless seen CLOSEn. */
4088 if (ln == PL_regoffs[n].end)
4092 if (type != REF /* REF can do byte comparison */
4093 && (utf8_target || type == REFFU))
4094 { /* XXX handle REFFL better */
4095 char * limit = PL_regeol;
4097 /* This call case insensitively compares the entire buffer
4098 * at s, with the current input starting at locinput, but
4099 * not going off the end given by PL_regeol, and returns in
4100 * limit upon success, how much of the current input was
4102 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4103 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4108 nextchr = UCHARAT(locinput);
4112 /* Not utf8: Inline the first character, for speed. */
4113 if (UCHARAT(s) != nextchr &&
4115 UCHARAT(s) != fold_array[nextchr]))
4117 ln = PL_regoffs[n].end - ln;
4118 if (locinput + ln > PL_regeol)
4120 if (ln > 1 && (type == REF
4121 ? memNE(s, locinput, ln)
4122 : ! folder(s, locinput, ln)))
4125 nextchr = UCHARAT(locinput);
4135 #define ST st->u.eval
4140 regexp_internal *rei;
4141 regnode *startpoint;
4144 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4145 if (cur_eval && cur_eval->locinput==locinput) {
4146 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4147 Perl_croak(aTHX_ "Infinite recursion in regex");
4148 if ( ++nochange_depth > max_nochange_depth )
4150 "Pattern subroutine nesting without pos change"
4151 " exceeded limit in regex");
4158 (void)ReREFCNT_inc(rex_sv);
4159 if (OP(scan)==GOSUB) {
4160 startpoint = scan + ARG2L(scan);
4161 ST.close_paren = ARG(scan);
4163 startpoint = rei->program+1;
4166 goto eval_recurse_doit;
4168 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4169 if (cur_eval && cur_eval->locinput==locinput) {
4170 if ( ++nochange_depth > max_nochange_depth )
4171 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4176 /* execute the code in the {...} */
4178 SV ** const before = SP;
4179 OP_4tree * const oop = PL_op;
4180 COP * const ocurcop = PL_curcop;
4182 char *saved_regeol = PL_regeol;
4183 struct re_save_state saved_state;
4185 /* To not corrupt the existing regex state while executing the
4186 * eval we would normally put it on the save stack, like with
4187 * save_re_context. However, re-evals have a weird scoping so we
4188 * can't just add ENTER/LEAVE here. With that, things like
4190 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4192 * would break, as they expect the localisation to be unwound
4193 * only when the re-engine backtracks through the bit that
4196 * What we do instead is just saving the state in a local c
4199 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4202 PL_op = (OP_4tree*)rexi->data->data[n];
4203 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4204 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4205 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4206 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4209 SV *sv_mrk = get_sv("REGMARK", 1);
4210 sv_setsv(sv_mrk, sv_yes_mark);
4213 CALLRUNOPS(aTHX); /* Scalar context. */
4216 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4222 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4225 PAD_RESTORE_LOCAL(old_comppad);
4226 PL_curcop = ocurcop;
4227 PL_regeol = saved_regeol;
4230 sv_setsv(save_scalar(PL_replgv), ret);
4234 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4237 /* extract RE object from returned value; compiling if
4243 SV *const sv = SvRV(ret);
4245 if (SvTYPE(sv) == SVt_REGEXP) {
4247 } else if (SvSMAGICAL(sv)) {
4248 mg = mg_find(sv, PERL_MAGIC_qr);
4251 } else if (SvTYPE(ret) == SVt_REGEXP) {
4253 } else if (SvSMAGICAL(ret)) {
4254 if (SvGMAGICAL(ret)) {
4255 /* I don't believe that there is ever qr magic
4257 assert(!mg_find(ret, PERL_MAGIC_qr));
4258 sv_unmagic(ret, PERL_MAGIC_qr);
4261 mg = mg_find(ret, PERL_MAGIC_qr);
4262 /* testing suggests mg only ends up non-NULL for
4263 scalars who were upgraded and compiled in the
4264 else block below. In turn, this is only
4265 triggered in the "postponed utf8 string" tests
4271 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4275 rx = reg_temp_copy(NULL, rx);
4279 const I32 osize = PL_regsize;
4282 assert (SvUTF8(ret));
4283 } else if (SvUTF8(ret)) {
4284 /* Not doing UTF-8, despite what the SV says. Is
4285 this only if we're trapped in use 'bytes'? */
4286 /* Make a copy of the octet sequence, but without
4287 the flag on, as the compiler now honours the
4288 SvUTF8 flag on ret. */
4290 const char *const p = SvPV(ret, len);
4291 ret = newSVpvn_flags(p, len, SVs_TEMP);
4293 rx = CALLREGCOMP(ret, pm_flags);
4295 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4297 /* This isn't a first class regexp. Instead, it's
4298 caching a regexp onto an existing, Perl visible
4300 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4305 re = (struct regexp *)SvANY(rx);
4307 RXp_MATCH_COPIED_off(re);
4308 re->subbeg = rex->subbeg;
4309 re->sublen = rex->sublen;
4312 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4313 "Matching embedded");
4315 startpoint = rei->program + 1;
4316 ST.close_paren = 0; /* only used for GOSUB */
4317 /* borrowed from regtry */
4318 if (PL_reg_start_tmpl <= re->nparens) {
4319 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4320 if(PL_reg_start_tmp)
4321 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4323 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4326 eval_recurse_doit: /* Share code with GOSUB below this line */
4327 /* run the pattern returned from (??{...}) */
4328 ST.cp = regcppush(0); /* Save *all* the positions. */
4329 REGCP_SET(ST.lastcp);
4331 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4333 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4334 PL_reglastparen = &re->lastparen;
4335 PL_reglastcloseparen = &re->lastcloseparen;
4337 re->lastcloseparen = 0;
4339 PL_reginput = locinput;
4342 /* XXXX This is too dramatic a measure... */
4345 ST.toggle_reg_flags = PL_reg_flags;
4347 PL_reg_flags |= RF_utf8;
4349 PL_reg_flags &= ~RF_utf8;
4350 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4352 ST.prev_rex = rex_sv;
4353 ST.prev_curlyx = cur_curlyx;
4354 SETREX(rex_sv,re_sv);
4359 ST.prev_eval = cur_eval;
4361 /* now continue from first node in postoned RE */
4362 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4365 /* logical is 1, /(?(?{...})X|Y)/ */
4366 sw = cBOOL(SvTRUE(ret));
4371 case EVAL_AB: /* cleanup after a successful (??{A})B */
4372 /* note: this is called twice; first after popping B, then A */
4373 PL_reg_flags ^= ST.toggle_reg_flags;
4374 ReREFCNT_dec(rex_sv);
4375 SETREX(rex_sv,ST.prev_rex);
4376 rex = (struct regexp *)SvANY(rex_sv);
4377 rexi = RXi_GET(rex);
4379 cur_eval = ST.prev_eval;
4380 cur_curlyx = ST.prev_curlyx;
4382 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4383 PL_reglastparen = &rex->lastparen;
4384 PL_reglastcloseparen = &rex->lastcloseparen;
4385 /* also update PL_regoffs */
4386 PL_regoffs = rex->offs;
4388 /* XXXX This is too dramatic a measure... */
4390 if ( nochange_depth )
4395 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4396 /* note: this is called twice; first after popping B, then A */
4397 PL_reg_flags ^= ST.toggle_reg_flags;
4398 ReREFCNT_dec(rex_sv);
4399 SETREX(rex_sv,ST.prev_rex);
4400 rex = (struct regexp *)SvANY(rex_sv);
4401 rexi = RXi_GET(rex);
4402 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4403 PL_reglastparen = &rex->lastparen;
4404 PL_reglastcloseparen = &rex->lastcloseparen;
4406 PL_reginput = locinput;
4407 REGCP_UNWIND(ST.lastcp);
4409 cur_eval = ST.prev_eval;
4410 cur_curlyx = ST.prev_curlyx;
4411 /* XXXX This is too dramatic a measure... */
4413 if ( nochange_depth )
4419 n = ARG(scan); /* which paren pair */
4420 PL_reg_start_tmp[n] = locinput;
4426 n = ARG(scan); /* which paren pair */
4427 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4428 PL_regoffs[n].end = locinput - PL_bostr;
4429 /*if (n > PL_regsize)
4431 if (n > *PL_reglastparen)
4432 *PL_reglastparen = n;
4433 *PL_reglastcloseparen = n;
4434 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4442 cursor && OP(cursor)!=END;
4443 cursor=regnext(cursor))
4445 if ( OP(cursor)==CLOSE ){
4447 if ( n <= lastopen ) {
4449 = PL_reg_start_tmp[n] - PL_bostr;
4450 PL_regoffs[n].end = locinput - PL_bostr;
4451 /*if (n > PL_regsize)
4453 if (n > *PL_reglastparen)
4454 *PL_reglastparen = n;
4455 *PL_reglastcloseparen = n;
4456 if ( n == ARG(scan) || (cur_eval &&
4457 cur_eval->u.eval.close_paren == n))
4466 n = ARG(scan); /* which paren pair */
4467 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4470 /* reg_check_named_buff_matched returns 0 for no match */
4471 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4475 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4481 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4483 next = NEXTOPER(NEXTOPER(scan));
4485 next = scan + ARG(scan);
4486 if (OP(next) == IFTHEN) /* Fake one. */
4487 next = NEXTOPER(NEXTOPER(next));
4491 logical = scan->flags;
4494 /*******************************************************************
4496 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4497 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4498 STAR/PLUS/CURLY/CURLYN are used instead.)
4500 A*B is compiled as <CURLYX><A><WHILEM><B>
4502 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4503 state, which contains the current count, initialised to -1. It also sets
4504 cur_curlyx to point to this state, with any previous value saved in the
4507 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4508 since the pattern may possibly match zero times (i.e. it's a while {} loop
4509 rather than a do {} while loop).
4511 Each entry to WHILEM represents a successful match of A. The count in the
4512 CURLYX block is incremented, another WHILEM state is pushed, and execution
4513 passes to A or B depending on greediness and the current count.
4515 For example, if matching against the string a1a2a3b (where the aN are
4516 substrings that match /A/), then the match progresses as follows: (the
4517 pushed states are interspersed with the bits of strings matched so far):
4520 <CURLYX cnt=0><WHILEM>
4521 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4522 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4523 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4524 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4526 (Contrast this with something like CURLYM, which maintains only a single
4530 a1 <CURLYM cnt=1> a2
4531 a1 a2 <CURLYM cnt=2> a3
4532 a1 a2 a3 <CURLYM cnt=3> b
4535 Each WHILEM state block marks a point to backtrack to upon partial failure
4536 of A or B, and also contains some minor state data related to that
4537 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4538 overall state, such as the count, and pointers to the A and B ops.
4540 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4541 must always point to the *current* CURLYX block, the rules are:
4543 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4544 and set cur_curlyx to point the new block.
4546 When popping the CURLYX block after a successful or unsuccessful match,
4547 restore the previous cur_curlyx.
4549 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4550 to the outer one saved in the CURLYX block.
4552 When popping the WHILEM block after a successful or unsuccessful B match,
4553 restore the previous cur_curlyx.
4555 Here's an example for the pattern (AI* BI)*BO
4556 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4559 curlyx backtrack stack
4560 ------ ---------------
4562 CO <CO prev=NULL> <WO>
4563 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4564 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4565 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4567 At this point the pattern succeeds, and we work back down the stack to
4568 clean up, restoring as we go:
4570 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4571 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4572 CO <CO prev=NULL> <WO>
4575 *******************************************************************/
4577 #define ST st->u.curlyx
4579 case CURLYX: /* start of /A*B/ (for complex A) */
4581 /* No need to save/restore up to this paren */
4582 I32 parenfloor = scan->flags;
4584 assert(next); /* keep Coverity happy */
4585 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4588 /* XXXX Probably it is better to teach regpush to support
4589 parenfloor > PL_regsize... */
4590 if (parenfloor > (I32)*PL_reglastparen)
4591 parenfloor = *PL_reglastparen; /* Pessimization... */
4593 ST.prev_curlyx= cur_curlyx;
4595 ST.cp = PL_savestack_ix;
4597 /* these fields contain the state of the current curly.
4598 * they are accessed by subsequent WHILEMs */
4599 ST.parenfloor = parenfloor;
4604 ST.count = -1; /* this will be updated by WHILEM */
4605 ST.lastloc = NULL; /* this will be updated by WHILEM */
4607 PL_reginput = locinput;
4608 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4612 case CURLYX_end: /* just finished matching all of A*B */
4613 cur_curlyx = ST.prev_curlyx;
4617 case CURLYX_end_fail: /* just failed to match all of A*B */
4619 cur_curlyx = ST.prev_curlyx;
4625 #define ST st->u.whilem
4627 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4629 /* see the discussion above about CURLYX/WHILEM */
4631 int min = ARG1(cur_curlyx->u.curlyx.me);
4632 int max = ARG2(cur_curlyx->u.curlyx.me);
4633 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4635 assert(cur_curlyx); /* keep Coverity happy */
4636 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4637 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4638 ST.cache_offset = 0;
4641 PL_reginput = locinput;
4643 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4644 "%*s whilem: matched %ld out of %d..%d\n",
4645 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4648 /* First just match a string of min A's. */
4651 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4652 cur_curlyx->u.curlyx.lastloc = locinput;
4653 REGCP_SET(ST.lastcp);
4655 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4659 /* If degenerate A matches "", assume A done. */
4661 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4662 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4663 "%*s whilem: empty match detected, trying continuation...\n",
4664 REPORT_CODE_OFF+depth*2, "")
4666 goto do_whilem_B_max;
4669 /* super-linear cache processing */
4673 if (!PL_reg_maxiter) {
4674 /* start the countdown: Postpone detection until we
4675 * know the match is not *that* much linear. */
4676 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4677 /* possible overflow for long strings and many CURLYX's */
4678 if (PL_reg_maxiter < 0)
4679 PL_reg_maxiter = I32_MAX;
4680 PL_reg_leftiter = PL_reg_maxiter;
4683 if (PL_reg_leftiter-- == 0) {
4684 /* initialise cache */
4685 const I32 size = (PL_reg_maxiter + 7)/8;
4686 if (PL_reg_poscache) {
4687 if ((I32)PL_reg_poscache_size < size) {
4688 Renew(PL_reg_poscache, size, char);
4689 PL_reg_poscache_size = size;
4691 Zero(PL_reg_poscache, size, char);
4694 PL_reg_poscache_size = size;
4695 Newxz(PL_reg_poscache, size, char);
4697 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4698 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4699 PL_colors[4], PL_colors[5])
4703 if (PL_reg_leftiter < 0) {
4704 /* have we already failed at this position? */
4706 offset = (scan->flags & 0xf) - 1
4707 + (locinput - PL_bostr) * (scan->flags>>4);
4708 mask = 1 << (offset % 8);
4710 if (PL_reg_poscache[offset] & mask) {
4711 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4712 "%*s whilem: (cache) already tried at this position...\n",
4713 REPORT_CODE_OFF+depth*2, "")
4715 sayNO; /* cache records failure */
4717 ST.cache_offset = offset;
4718 ST.cache_mask = mask;
4722 /* Prefer B over A for minimal matching. */
4724 if (cur_curlyx->u.curlyx.minmod) {
4725 ST.save_curlyx = cur_curlyx;
4726 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4727 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4728 REGCP_SET(ST.lastcp);
4729 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4733 /* Prefer A over B for maximal matching. */
4735 if (n < max) { /* More greed allowed? */
4736 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4737 cur_curlyx->u.curlyx.lastloc = locinput;
4738 REGCP_SET(ST.lastcp);
4739 PUSH_STATE_GOTO(WHILEM_A_max, A);
4742 goto do_whilem_B_max;
4746 case WHILEM_B_min: /* just matched B in a minimal match */
4747 case WHILEM_B_max: /* just matched B in a maximal match */
4748 cur_curlyx = ST.save_curlyx;
4752 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4753 cur_curlyx = ST.save_curlyx;
4754 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4755 cur_curlyx->u.curlyx.count--;
4759 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4761 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4762 REGCP_UNWIND(ST.lastcp);
4764 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4765 cur_curlyx->u.curlyx.count--;
4769 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4770 REGCP_UNWIND(ST.lastcp);
4771 regcppop(rex); /* Restore some previous $<digit>s? */
4772 PL_reginput = locinput;
4773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4774 "%*s whilem: failed, trying continuation...\n",
4775 REPORT_CODE_OFF+depth*2, "")
4778 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4779 && ckWARN(WARN_REGEXP)
4780 && !(PL_reg_flags & RF_warned))
4782 PL_reg_flags |= RF_warned;
4783 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4784 "Complex regular subexpression recursion",
4789 ST.save_curlyx = cur_curlyx;
4790 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4791 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4794 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4795 cur_curlyx = ST.save_curlyx;
4796 REGCP_UNWIND(ST.lastcp);
4799 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4800 /* Maximum greed exceeded */
4801 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4802 && ckWARN(WARN_REGEXP)
4803 && !(PL_reg_flags & RF_warned))
4805 PL_reg_flags |= RF_warned;
4806 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4807 "%s limit (%d) exceeded",
4808 "Complex regular subexpression recursion",
4811 cur_curlyx->u.curlyx.count--;
4815 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4816 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4818 /* Try grabbing another A and see if it helps. */
4819 PL_reginput = locinput;
4820 cur_curlyx->u.curlyx.lastloc = locinput;
4821 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4822 REGCP_SET(ST.lastcp);
4823 PUSH_STATE_GOTO(WHILEM_A_min,
4824 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4828 #define ST st->u.branch
4830 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4831 next = scan + ARG(scan);
4834 scan = NEXTOPER(scan);
4837 case BRANCH: /* /(...|A|...)/ */
4838 scan = NEXTOPER(scan); /* scan now points to inner node */
4839 ST.lastparen = *PL_reglastparen;
4840 ST.next_branch = next;
4842 PL_reginput = locinput;
4844 /* Now go into the branch */
4846 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4848 PUSH_STATE_GOTO(BRANCH_next, scan);
4852 PL_reginput = locinput;
4853 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4854 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4855 PUSH_STATE_GOTO(CUTGROUP_next,next);
4857 case CUTGROUP_next_fail:
4860 if (st->u.mark.mark_name)
4861 sv_commit = st->u.mark.mark_name;
4867 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4872 REGCP_UNWIND(ST.cp);
4873 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4874 PL_regoffs[n].end = -1;
4875 *PL_reglastparen = n;
4876 /*dmq: *PL_reglastcloseparen = n; */
4877 scan = ST.next_branch;
4878 /* no more branches? */
4879 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4881 PerlIO_printf( Perl_debug_log,
4882 "%*s %sBRANCH failed...%s\n",
4883 REPORT_CODE_OFF+depth*2, "",
4889 continue; /* execute next BRANCH[J] op */
4897 #define ST st->u.curlym
4899 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4901 /* This is an optimisation of CURLYX that enables us to push
4902 * only a single backtracking state, no matter how many matches
4903 * there are in {m,n}. It relies on the pattern being constant
4904 * length, with no parens to influence future backrefs
4908 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4910 /* if paren positive, emulate an OPEN/CLOSE around A */
4912 U32 paren = ST.me->flags;
4913 if (paren > PL_regsize)
4915 if (paren > *PL_reglastparen)
4916 *PL_reglastparen = paren;
4917 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4925 ST.c1 = CHRTEST_UNINIT;
4928 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4931 curlym_do_A: /* execute the A in /A{m,n}B/ */
4932 PL_reginput = locinput;
4933 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4936 case CURLYM_A: /* we've just matched an A */
4937 locinput = st->locinput;
4938 nextchr = UCHARAT(locinput);
4941 /* after first match, determine A's length: u.curlym.alen */
4942 if (ST.count == 1) {
4943 if (PL_reg_match_utf8) {
4945 while (s < PL_reginput) {
4951 ST.alen = PL_reginput - locinput;
4954 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4957 PerlIO_printf(Perl_debug_log,
4958 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4959 (int)(REPORT_CODE_OFF+(depth*2)), "",
4960 (IV) ST.count, (IV)ST.alen)
4963 locinput = PL_reginput;
4965 if (cur_eval && cur_eval->u.eval.close_paren &&
4966 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4970 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4971 if ( max == REG_INFTY || ST.count < max )
4972 goto curlym_do_A; /* try to match another A */
4974 goto curlym_do_B; /* try to match B */
4976 case CURLYM_A_fail: /* just failed to match an A */
4977 REGCP_UNWIND(ST.cp);
4979 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4980 || (cur_eval && cur_eval->u.eval.close_paren &&
4981 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4984 curlym_do_B: /* execute the B in /A{m,n}B/ */
4985 PL_reginput = locinput;
4986 if (ST.c1 == CHRTEST_UNINIT) {
4987 /* calculate c1 and c2 for possible match of 1st char
4988 * following curly */
4989 ST.c1 = ST.c2 = CHRTEST_VOID;
4990 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4991 regnode *text_node = ST.B;
4992 if (! HAS_TEXT(text_node))
4993 FIND_NEXT_IMPT(text_node);
4996 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4998 But the former is redundant in light of the latter.
5000 if this changes back then the macro for
5001 IS_TEXT and friends need to change.
5003 if (PL_regkind[OP(text_node)] == EXACT)
5006 ST.c1 = (U8)*STRING(text_node);
5007 switch (OP(text_node)) {
5008 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5010 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5011 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5012 default: ST.c2 = ST.c1;
5019 PerlIO_printf(Perl_debug_log,
5020 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5021 (int)(REPORT_CODE_OFF+(depth*2)),
5024 if (ST.c1 != CHRTEST_VOID
5025 && UCHARAT(PL_reginput) != ST.c1
5026 && UCHARAT(PL_reginput) != ST.c2)
5028 /* simulate B failing */
5030 PerlIO_printf(Perl_debug_log,
5031 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5032 (int)(REPORT_CODE_OFF+(depth*2)),"",
5035 state_num = CURLYM_B_fail;
5036 goto reenter_switch;
5040 /* mark current A as captured */
5041 I32 paren = ST.me->flags;
5043 PL_regoffs[paren].start
5044 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5045 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5046 /*dmq: *PL_reglastcloseparen = paren; */
5049 PL_regoffs[paren].end = -1;
5050 if (cur_eval && cur_eval->u.eval.close_paren &&
5051 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5060 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5063 case CURLYM_B_fail: /* just failed to match a B */
5064 REGCP_UNWIND(ST.cp);
5066 I32 max = ARG2(ST.me);
5067 if (max != REG_INFTY && ST.count == max)
5069 goto curlym_do_A; /* try to match a further A */
5071 /* backtrack one A */
5072 if (ST.count == ARG1(ST.me) /* min */)
5075 locinput = HOPc(locinput, -ST.alen);
5076 goto curlym_do_B; /* try to match B */
5079 #define ST st->u.curly
5081 #define CURLY_SETPAREN(paren, success) \
5084 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5085 PL_regoffs[paren].end = locinput - PL_bostr; \
5086 *PL_reglastcloseparen = paren; \
5089 PL_regoffs[paren].end = -1; \
5092 case STAR: /* /A*B/ where A is width 1 */
5096 scan = NEXTOPER(scan);
5098 case PLUS: /* /A+B/ where A is width 1 */
5102 scan = NEXTOPER(scan);
5104 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5105 ST.paren = scan->flags; /* Which paren to set */
5106 if (ST.paren > PL_regsize)
5107 PL_regsize = ST.paren;
5108 if (ST.paren > *PL_reglastparen)
5109 *PL_reglastparen = ST.paren;
5110 ST.min = ARG1(scan); /* min to match */
5111 ST.max = ARG2(scan); /* max to match */
5112 if (cur_eval && cur_eval->u.eval.close_paren &&
5113 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5117 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5119 case CURLY: /* /A{m,n}B/ where A is width 1 */
5121 ST.min = ARG1(scan); /* min to match */
5122 ST.max = ARG2(scan); /* max to match */
5123 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5126 * Lookahead to avoid useless match attempts
5127 * when we know what character comes next.
5129 * Used to only do .*x and .*?x, but now it allows
5130 * for )'s, ('s and (?{ ... })'s to be in the way
5131 * of the quantifier and the EXACT-like node. -- japhy
5134 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5136 if (HAS_TEXT(next) || JUMPABLE(next)) {
5138 regnode *text_node = next;
5140 if (! HAS_TEXT(text_node))
5141 FIND_NEXT_IMPT(text_node);
5143 if (! HAS_TEXT(text_node))
5144 ST.c1 = ST.c2 = CHRTEST_VOID;
5146 if ( PL_regkind[OP(text_node)] != EXACT ) {
5147 ST.c1 = ST.c2 = CHRTEST_VOID;
5148 goto assume_ok_easy;
5151 s = (U8*)STRING(text_node);
5153 /* Currently we only get here when
5155 PL_rekind[OP(text_node)] == EXACT
5157 if this changes back then the macro for IS_TEXT and
5158 friends need to change. */
5161 switch (OP(text_node)) {
5162 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5164 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5165 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5166 default: ST.c2 = ST.c1; break;
5169 else { /* UTF_PATTERN */
5170 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5171 STRLEN ulen1, ulen2;
5172 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5173 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5175 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5176 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5178 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5180 0 : UTF8_ALLOW_ANY);
5181 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5183 0 : UTF8_ALLOW_ANY);
5185 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5187 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5192 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5199 ST.c1 = ST.c2 = CHRTEST_VOID;
5204 PL_reginput = locinput;
5207 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5210 locinput = PL_reginput;
5212 if (ST.c1 == CHRTEST_VOID)
5213 goto curly_try_B_min;
5215 ST.oldloc = locinput;
5217 /* set ST.maxpos to the furthest point along the
5218 * string that could possibly match */
5219 if (ST.max == REG_INFTY) {
5220 ST.maxpos = PL_regeol - 1;
5222 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5225 else if (utf8_target) {
5226 int m = ST.max - ST.min;
5227 for (ST.maxpos = locinput;
5228 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5229 ST.maxpos += UTF8SKIP(ST.maxpos);
5232 ST.maxpos = locinput + ST.max - ST.min;
5233 if (ST.maxpos >= PL_regeol)
5234 ST.maxpos = PL_regeol - 1;
5236 goto curly_try_B_min_known;
5240 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5241 locinput = PL_reginput;
5242 if (ST.count < ST.min)
5244 if ((ST.count > ST.min)
5245 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5247 /* A{m,n} must come at the end of the string, there's
5248 * no point in backing off ... */
5250 /* ...except that $ and \Z can match before *and* after
5251 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5252 We may back off by one in this case. */
5253 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5257 goto curly_try_B_max;
5262 case CURLY_B_min_known_fail:
5263 /* failed to find B in a non-greedy match where c1,c2 valid */
5264 if (ST.paren && ST.count)
5265 PL_regoffs[ST.paren].end = -1;
5267 PL_reginput = locinput; /* Could be reset... */
5268 REGCP_UNWIND(ST.cp);
5269 /* Couldn't or didn't -- move forward. */
5270 ST.oldloc = locinput;
5272 locinput += UTF8SKIP(locinput);
5276 curly_try_B_min_known:
5277 /* find the next place where 'B' could work, then call B */
5281 n = (ST.oldloc == locinput) ? 0 : 1;
5282 if (ST.c1 == ST.c2) {
5284 /* set n to utf8_distance(oldloc, locinput) */
5285 while (locinput <= ST.maxpos &&
5286 utf8n_to_uvchr((U8*)locinput,
5287 UTF8_MAXBYTES, &len,
5288 uniflags) != (UV)ST.c1) {
5294 /* set n to utf8_distance(oldloc, locinput) */
5295 while (locinput <= ST.maxpos) {
5297 const UV c = utf8n_to_uvchr((U8*)locinput,
5298 UTF8_MAXBYTES, &len,
5300 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5308 if (ST.c1 == ST.c2) {
5309 while (locinput <= ST.maxpos &&
5310 UCHARAT(locinput) != ST.c1)
5314 while (locinput <= ST.maxpos
5315 && UCHARAT(locinput) != ST.c1
5316 && UCHARAT(locinput) != ST.c2)
5319 n = locinput - ST.oldloc;
5321 if (locinput > ST.maxpos)
5323 /* PL_reginput == oldloc now */
5326 if (regrepeat(rex, ST.A, n, depth) < n)
5329 PL_reginput = locinput;
5330 CURLY_SETPAREN(ST.paren, ST.count);
5331 if (cur_eval && cur_eval->u.eval.close_paren &&
5332 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5335 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5340 case CURLY_B_min_fail:
5341 /* failed to find B in a non-greedy match where c1,c2 invalid */
5342 if (ST.paren && ST.count)
5343 PL_regoffs[ST.paren].end = -1;
5345 REGCP_UNWIND(ST.cp);
5346 /* failed -- move forward one */
5347 PL_reginput = locinput;
5348 if (regrepeat(rex, ST.A, 1, depth)) {
5350 locinput = PL_reginput;
5351 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5352 ST.count > 0)) /* count overflow ? */
5355 CURLY_SETPAREN(ST.paren, ST.count);
5356 if (cur_eval && cur_eval->u.eval.close_paren &&
5357 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5360 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5368 /* a successful greedy match: now try to match B */
5369 if (cur_eval && cur_eval->u.eval.close_paren &&
5370 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5375 if (ST.c1 != CHRTEST_VOID)
5376 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5377 UTF8_MAXBYTES, 0, uniflags)
5378 : (UV) UCHARAT(PL_reginput);
5379 /* If it could work, try it. */
5380 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5381 CURLY_SETPAREN(ST.paren, ST.count);
5382 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5387 case CURLY_B_max_fail:
5388 /* failed to find B in a greedy match */
5389 if (ST.paren && ST.count)
5390 PL_regoffs[ST.paren].end = -1;
5392 REGCP_UNWIND(ST.cp);
5394 if (--ST.count < ST.min)
5396 PL_reginput = locinput = HOPc(locinput, -1);
5397 goto curly_try_B_max;
5404 /* we've just finished A in /(??{A})B/; now continue with B */
5406 st->u.eval.toggle_reg_flags
5407 = cur_eval->u.eval.toggle_reg_flags;
5408 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5410 st->u.eval.prev_rex = rex_sv; /* inner */
5411 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5412 rex = (struct regexp *)SvANY(rex_sv);
5413 rexi = RXi_GET(rex);
5414 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5415 ReREFCNT_inc(rex_sv);
5416 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5418 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5419 PL_reglastparen = &rex->lastparen;
5420 PL_reglastcloseparen = &rex->lastcloseparen;
5422 REGCP_SET(st->u.eval.lastcp);
5423 PL_reginput = locinput;
5425 /* Restore parens of the outer rex without popping the
5427 tmpix = PL_savestack_ix;
5428 PL_savestack_ix = cur_eval->u.eval.lastcp;
5430 PL_savestack_ix = tmpix;
5432 st->u.eval.prev_eval = cur_eval;
5433 cur_eval = cur_eval->u.eval.prev_eval;
5435 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5436 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5437 if ( nochange_depth )
5440 PUSH_YES_STATE_GOTO(EVAL_AB,
5441 st->u.eval.prev_eval->u.eval.B); /* match B */
5444 if (locinput < reginfo->till) {
5445 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5446 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5448 (long)(locinput - PL_reg_starttry),
5449 (long)(reginfo->till - PL_reg_starttry),
5452 sayNO_SILENT; /* Cannot match: too short. */
5454 PL_reginput = locinput; /* put where regtry can find it */
5455 sayYES; /* Success! */
5457 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5459 PerlIO_printf(Perl_debug_log,
5460 "%*s %ssubpattern success...%s\n",
5461 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5462 PL_reginput = locinput; /* put where regtry can find it */
5463 sayYES; /* Success! */
5466 #define ST st->u.ifmatch
5468 case SUSPEND: /* (?>A) */
5470 PL_reginput = locinput;
5473 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5475 goto ifmatch_trivial_fail_test;
5477 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5479 ifmatch_trivial_fail_test:
5481 char * const s = HOPBACKc(locinput, scan->flags);
5486 sw = 1 - cBOOL(ST.wanted);
5490 next = scan + ARG(scan);
5498 PL_reginput = locinput;
5502 ST.logical = logical;
5503 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5505 /* execute body of (?...A) */
5506 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5509 case IFMATCH_A_fail: /* body of (?...A) failed */
5510 ST.wanted = !ST.wanted;
5513 case IFMATCH_A: /* body of (?...A) succeeded */
5515 sw = cBOOL(ST.wanted);
5517 else if (!ST.wanted)
5520 if (OP(ST.me) == SUSPEND)
5521 locinput = PL_reginput;
5523 locinput = PL_reginput = st->locinput;
5524 nextchr = UCHARAT(locinput);
5526 scan = ST.me + ARG(ST.me);
5529 continue; /* execute B */
5534 next = scan + ARG(scan);
5539 reginfo->cutpoint = PL_regeol;
5542 PL_reginput = locinput;
5544 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5545 PUSH_STATE_GOTO(COMMIT_next,next);
5547 case COMMIT_next_fail:
5554 #define ST st->u.mark
5556 ST.prev_mark = mark_state;
5557 ST.mark_name = sv_commit = sv_yes_mark
5558 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5560 ST.mark_loc = PL_reginput = locinput;
5561 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5563 case MARKPOINT_next:
5564 mark_state = ST.prev_mark;
5567 case MARKPOINT_next_fail:
5568 if (popmark && sv_eq(ST.mark_name,popmark))
5570 if (ST.mark_loc > startpoint)
5571 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5572 popmark = NULL; /* we found our mark */
5573 sv_commit = ST.mark_name;
5576 PerlIO_printf(Perl_debug_log,
5577 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5578 REPORT_CODE_OFF+depth*2, "",
5579 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5582 mark_state = ST.prev_mark;
5583 sv_yes_mark = mark_state ?
5584 mark_state->u.mark.mark_name : NULL;
5588 PL_reginput = locinput;
5590 /* (*SKIP) : if we fail we cut here*/
5591 ST.mark_name = NULL;
5592 ST.mark_loc = locinput;
5593 PUSH_STATE_GOTO(SKIP_next,next);
5595 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5596 otherwise do nothing. Meaning we need to scan
5598 regmatch_state *cur = mark_state;
5599 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5602 if ( sv_eq( cur->u.mark.mark_name,
5605 ST.mark_name = find;
5606 PUSH_STATE_GOTO( SKIP_next, next );
5608 cur = cur->u.mark.prev_mark;
5611 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5613 case SKIP_next_fail:
5615 /* (*CUT:NAME) - Set up to search for the name as we
5616 collapse the stack*/
5617 popmark = ST.mark_name;
5619 /* (*CUT) - No name, we cut here.*/
5620 if (ST.mark_loc > startpoint)
5621 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5622 /* but we set sv_commit to latest mark_name if there
5623 is one so they can test to see how things lead to this
5626 sv_commit=mark_state->u.mark.mark_name;
5634 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5636 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5639 U8 folded[UTF8_MAXBYTES_CASE+1];
5641 const char * const l = locinput;
5642 char *e = PL_regeol;
5643 to_uni_fold(n, folded, &foldlen);
5645 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5646 l, &e, 0, utf8_target)) {
5651 nextchr = UCHARAT(locinput);
5654 if ((n=is_LNBREAK(locinput,utf8_target))) {
5656 nextchr = UCHARAT(locinput);
5661 #define CASE_CLASS(nAmE) \
5663 if ((n=is_##nAmE(locinput,utf8_target))) { \
5665 nextchr = UCHARAT(locinput); \
5670 if ((n=is_##nAmE(locinput,utf8_target))) { \
5673 locinput += UTF8SKIP(locinput); \
5674 nextchr = UCHARAT(locinput); \
5679 CASE_CLASS(HORIZWS);
5683 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5684 PTR2UV(scan), OP(scan));
5685 Perl_croak(aTHX_ "regexp memory corruption");
5689 /* switch break jumps here */
5690 scan = next; /* prepare to execute the next op and ... */
5691 continue; /* ... jump back to the top, reusing st */
5695 /* push a state that backtracks on success */
5696 st->u.yes.prev_yes_state = yes_state;
5700 /* push a new regex state, then continue at scan */
5702 regmatch_state *newst;
5705 regmatch_state *cur = st;
5706 regmatch_state *curyes = yes_state;
5708 regmatch_slab *slab = PL_regmatch_slab;
5709 for (;curd > -1;cur--,curd--) {
5710 if (cur < SLAB_FIRST(slab)) {
5712 cur = SLAB_LAST(slab);
5714 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5715 REPORT_CODE_OFF + 2 + depth * 2,"",
5716 curd, PL_reg_name[cur->resume_state],
5717 (curyes == cur) ? "yes" : ""
5720 curyes = cur->u.yes.prev_yes_state;
5723 DEBUG_STATE_pp("push")
5726 st->locinput = locinput;
5728 if (newst > SLAB_LAST(PL_regmatch_slab))
5729 newst = S_push_slab(aTHX);
5730 PL_regmatch_state = newst;
5732 locinput = PL_reginput;
5733 nextchr = UCHARAT(locinput);
5741 * We get here only if there's trouble -- normally "case END" is
5742 * the terminating point.
5744 Perl_croak(aTHX_ "corrupted regexp pointers");
5750 /* we have successfully completed a subexpression, but we must now
5751 * pop to the state marked by yes_state and continue from there */
5752 assert(st != yes_state);
5754 while (st != yes_state) {
5756 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5757 PL_regmatch_slab = PL_regmatch_slab->prev;
5758 st = SLAB_LAST(PL_regmatch_slab);
5762 DEBUG_STATE_pp("pop (no final)");
5764 DEBUG_STATE_pp("pop (yes)");
5770 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5771 || yes_state > SLAB_LAST(PL_regmatch_slab))
5773 /* not in this slab, pop slab */
5774 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5775 PL_regmatch_slab = PL_regmatch_slab->prev;
5776 st = SLAB_LAST(PL_regmatch_slab);
5778 depth -= (st - yes_state);
5781 yes_state = st->u.yes.prev_yes_state;
5782 PL_regmatch_state = st;
5785 locinput= st->locinput;
5786 nextchr = UCHARAT(locinput);
5788 state_num = st->resume_state + no_final;
5789 goto reenter_switch;
5792 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5793 PL_colors[4], PL_colors[5]));
5795 if (PL_reg_eval_set) {
5796 /* each successfully executed (?{...}) block does the equivalent of
5797 * local $^R = do {...}
5798 * When popping the save stack, all these locals would be undone;
5799 * bypass this by setting the outermost saved $^R to the latest
5801 if (oreplsv != GvSV(PL_replgv))
5802 sv_setsv(oreplsv, GvSV(PL_replgv));
5809 PerlIO_printf(Perl_debug_log,
5810 "%*s %sfailed...%s\n",
5811 REPORT_CODE_OFF+depth*2, "",
5812 PL_colors[4], PL_colors[5])
5824 /* there's a previous state to backtrack to */
5826 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5827 PL_regmatch_slab = PL_regmatch_slab->prev;
5828 st = SLAB_LAST(PL_regmatch_slab);
5830 PL_regmatch_state = st;
5831 locinput= st->locinput;
5832 nextchr = UCHARAT(locinput);
5834 DEBUG_STATE_pp("pop");
5836 if (yes_state == st)
5837 yes_state = st->u.yes.prev_yes_state;
5839 state_num = st->resume_state + 1; /* failure = success + 1 */
5840 goto reenter_switch;
5845 if (rex->intflags & PREGf_VERBARG_SEEN) {
5846 SV *sv_err = get_sv("REGERROR", 1);
5847 SV *sv_mrk = get_sv("REGMARK", 1);
5849 sv_commit = &PL_sv_no;
5851 sv_yes_mark = &PL_sv_yes;
5854 sv_commit = &PL_sv_yes;
5855 sv_yes_mark = &PL_sv_no;
5857 sv_setsv(sv_err, sv_commit);
5858 sv_setsv(sv_mrk, sv_yes_mark);
5861 /* clean up; in particular, free all slabs above current one */
5862 LEAVE_SCOPE(oldsave);
5868 - regrepeat - repeatedly match something simple, report how many
5871 * [This routine now assumes that it will only match on things of length 1.
5872 * That was true before, but now we assume scan - reginput is the count,
5873 * rather than incrementing count on every character. [Er, except utf8.]]
5876 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5879 register char *scan;
5881 register char *loceol = PL_regeol;
5882 register I32 hardcount = 0;
5883 register bool utf8_target = PL_reg_match_utf8;
5886 PERL_UNUSED_ARG(depth);
5889 PERL_ARGS_ASSERT_REGREPEAT;
5892 if (max == REG_INFTY)
5894 else if (max < loceol - scan)
5895 loceol = scan + max;
5900 while (scan < loceol && hardcount < max && *scan != '\n') {
5901 scan += UTF8SKIP(scan);
5905 while (scan < loceol && *scan != '\n')
5912 while (scan < loceol && hardcount < max) {
5913 scan += UTF8SKIP(scan);
5924 /* To get here, EXACTish nodes must have *byte* length == 1. That
5925 * means they match only characters in the string that can be expressed
5926 * as a single byte. For non-utf8 strings, that means a simple match.
5927 * For utf8 strings, the character matched must be an invariant, or
5928 * downgradable to a single byte. The pattern's utf8ness is
5929 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5930 * it is, it's an invariant */
5933 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5935 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5936 while (scan < loceol && UCHARAT(scan) == c) {
5942 /* Here, the string is utf8, and the pattern char is different
5943 * in utf8 than not, so can't compare them directly. Outside the
5944 * loop, find find the two utf8 bytes that represent c, and then
5945 * look for those in sequence in the utf8 string */
5946 U8 high = UTF8_TWO_BYTE_HI(c);
5947 U8 low = UTF8_TWO_BYTE_LO(c);
5950 while (hardcount < max
5951 && scan + 1 < loceol
5952 && UCHARAT(scan) == high
5953 && UCHARAT(scan + 1) == low)
5961 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5965 PL_reg_flags |= RF_tainted;
5966 utf8_flags = FOLDEQ_UTF8_LOCALE;
5973 /* The comments for the EXACT case above apply as well to these fold
5978 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5980 if (utf8_target) { /* Use full Unicode fold matching */
5981 char *tmpeol = loceol;
5982 while (hardcount < max
5983 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
5984 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
5991 /* XXX Note that the above handles properly the German sharp s in
5992 * the pattern matching ss in the string. But it doesn't handle
5993 * properly cases where the string contains say 'LIGATURE ff' and
5994 * the pattern is 'f+'. This would require, say, a new function or
5995 * revised interface to foldEQ_utf8(), in which the maximum number
5996 * of characters to match could be passed and it would return how
5997 * many actually did. This is just one of many cases where
5998 * multi-char folds don't work properly, and so the fix is being
6004 /* Here, the string isn't utf8 and c is a single byte; and either
6005 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6006 * doesn't affect c. Can just do simple comparisons for exact or
6009 case EXACTF: folded = PL_fold[c]; break;
6011 case EXACTFU: folded = PL_fold_latin1[c]; break;
6012 case EXACTFL: folded = PL_fold_locale[c]; break;
6013 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6015 while (scan < loceol &&
6016 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6024 if (utf8_target || OP(p) == ANYOFV) {
6027 inclasslen = loceol - scan;
6028 while (hardcount < max
6029 && ((inclasslen = loceol - scan) > 0)
6030 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6036 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6044 LOAD_UTF8_CHARCLASS_ALNUM();
6045 while (hardcount < max && scan < loceol &&
6046 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6048 scan += UTF8SKIP(scan);
6052 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6060 while (scan < loceol && isALNUM((U8) *scan)) {
6065 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6070 PL_reg_flags |= RF_tainted;
6073 while (hardcount < max && scan < loceol &&
6074 isALNUM_LC_utf8((U8*)scan)) {
6075 scan += UTF8SKIP(scan);
6079 while (scan < loceol && isALNUM_LC(*scan))
6089 LOAD_UTF8_CHARCLASS_ALNUM();
6090 while (hardcount < max && scan < loceol &&
6091 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6093 scan += UTF8SKIP(scan);
6097 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6104 goto utf8_Nwordchar;
6105 while (scan < loceol && ! isALNUM((U8) *scan)) {
6111 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6112 scan += UTF8SKIP(scan);
6116 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6122 PL_reg_flags |= RF_tainted;
6125 while (hardcount < max && scan < loceol &&
6126 !isALNUM_LC_utf8((U8*)scan)) {
6127 scan += UTF8SKIP(scan);
6131 while (scan < loceol && !isALNUM_LC(*scan))
6141 LOAD_UTF8_CHARCLASS_SPACE();
6142 while (hardcount < max && scan < loceol &&
6144 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6146 scan += UTF8SKIP(scan);
6152 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6161 while (scan < loceol && isSPACE((U8) *scan)) {
6166 while (scan < loceol && isSPACE_A((U8) *scan)) {
6171 PL_reg_flags |= RF_tainted;
6174 while (hardcount < max && scan < loceol &&
6175 isSPACE_LC_utf8((U8*)scan)) {
6176 scan += UTF8SKIP(scan);
6180 while (scan < loceol && isSPACE_LC(*scan))
6190 LOAD_UTF8_CHARCLASS_SPACE();
6191 while (hardcount < max && scan < loceol &&
6193 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6195 scan += UTF8SKIP(scan);
6201 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6210 while (scan < loceol && ! isSPACE((U8) *scan)) {
6216 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6217 scan += UTF8SKIP(scan);
6221 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6227 PL_reg_flags |= RF_tainted;
6230 while (hardcount < max && scan < loceol &&
6231 !isSPACE_LC_utf8((U8*)scan)) {
6232 scan += UTF8SKIP(scan);
6236 while (scan < loceol && !isSPACE_LC(*scan))
6243 LOAD_UTF8_CHARCLASS_DIGIT();
6244 while (hardcount < max && scan < loceol &&
6245 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6246 scan += UTF8SKIP(scan);
6250 while (scan < loceol && isDIGIT(*scan))
6255 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6260 PL_reg_flags |= RF_tainted;
6263 while (hardcount < max && scan < loceol &&
6264 isDIGIT_LC_utf8((U8*)scan)) {
6265 scan += UTF8SKIP(scan);
6269 while (scan < loceol && isDIGIT_LC(*scan))
6276 LOAD_UTF8_CHARCLASS_DIGIT();
6277 while (hardcount < max && scan < loceol &&
6278 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6279 scan += UTF8SKIP(scan);
6283 while (scan < loceol && !isDIGIT(*scan))
6289 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6290 scan += UTF8SKIP(scan);
6294 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6300 PL_reg_flags |= RF_tainted;
6303 while (hardcount < max && scan < loceol &&
6304 !isDIGIT_LC_utf8((U8*)scan)) {
6305 scan += UTF8SKIP(scan);
6309 while (scan < loceol && !isDIGIT_LC(*scan))
6316 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6322 LNBREAK can match two latin chars, which is ok,
6323 because we have a null terminated string, but we
6324 have to use hardcount in this situation
6326 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6335 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6340 while (scan < loceol && is_HORIZWS_latin1(scan))
6347 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6348 scan += UTF8SKIP(scan);
6352 while (scan < loceol && !is_HORIZWS_latin1(scan))
6360 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6365 while (scan < loceol && is_VERTWS_latin1(scan))
6373 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6374 scan += UTF8SKIP(scan);
6378 while (scan < loceol && !is_VERTWS_latin1(scan))
6384 default: /* Called on something of 0 width. */
6385 break; /* So match right here or not at all. */
6391 c = scan - PL_reginput;
6395 GET_RE_DEBUG_FLAGS_DECL;
6397 SV * const prop = sv_newmortal();
6398 regprop(prog, prop, p);
6399 PerlIO_printf(Perl_debug_log,
6400 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6401 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6409 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6411 - regclass_swash - prepare the utf8 swash
6415 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6421 RXi_GET_DECL(prog,progi);
6422 const struct reg_data * const data = prog ? progi->data : NULL;
6424 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6426 assert(ANYOF_NONBITMAP(node));
6428 if (data && data->count) {
6429 const U32 n = ARG(node);
6431 if (data->what[n] == 's') {
6432 SV * const rv = MUTABLE_SV(data->data[n]);
6433 AV * const av = MUTABLE_AV(SvRV(rv));
6434 SV **const ary = AvARRAY(av);
6437 /* See the end of regcomp.c:S_regclass() for
6438 * documentation of these array elements. */
6441 a = SvROK(ary[1]) ? &ary[1] : NULL;
6442 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6446 else if (si && doinit) {
6447 sw = swash_init("utf8", "", si, 1, 0);
6448 (void)av_store(av, 1, sw);
6465 - reginclass - determine if a character falls into a character class
6467 n is the ANYOF regnode
6468 p is the target string
6469 lenp is pointer to the maximum number of bytes of how far to go in p
6470 (This is assumed wthout checking to always be at least the current
6472 utf8_target tells whether p is in UTF-8.
6474 Returns true if matched; false otherwise. If lenp is not NULL, on return
6475 from a successful match, the value it points to will be updated to how many
6476 bytes in p were matched. If there was no match, the value is undefined,
6477 possibly changed from the input.
6479 Note that this can be a synthetic start class, a combination of various
6480 nodes, so things you think might be mutually exclusive, such as locale,
6481 aren't. It can match both locale and non-locale
6486 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6489 const char flags = ANYOF_FLAGS(n);
6495 PERL_ARGS_ASSERT_REGINCLASS;
6497 /* If c is not already the code point, get it */
6498 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6499 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6500 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6501 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6502 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6503 * UTF8_ALLOW_FFFF */
6504 if (c_len == (STRLEN)-1)
6505 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6511 /* Use passed in max length, or one character if none passed in or less
6512 * than one character. And assume will match just one character. This is
6513 * overwritten later if matched more. */
6515 maxlen = (*lenp > c_len) ? *lenp : c_len;
6523 /* If this character is potentially in the bitmap, check it */
6525 if (ANYOF_BITMAP_TEST(n, c))
6527 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6534 else if (flags & ANYOF_LOCALE) {
6535 PL_reg_flags |= RF_tainted;
6537 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6538 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6542 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6543 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6544 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6545 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6546 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6547 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6548 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6549 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6550 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6551 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6552 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6553 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6554 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6555 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6556 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6557 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6558 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6559 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6560 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6561 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6562 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6563 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6564 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6565 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6566 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6567 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6568 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6569 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6570 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6571 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6572 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6573 ) /* How's that for a conditional? */
6580 /* If the bitmap didn't (or couldn't) match, and something outside the
6581 * bitmap could match, try that. Locale nodes specifiy completely the
6582 * behavior of code points in the bit map (otherwise, a utf8 target would
6583 * cause them to be treated as Unicode and not locale), except in
6584 * the very unlikely event when this node is a synthetic start class, which
6585 * could be a combination of locale and non-locale nodes. So allow locale
6586 * to match for the synthetic start class, which will give a false
6587 * positive that will be resolved when the match is done again as not part
6588 * of the synthetic start class */
6590 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6591 match = TRUE; /* Everything above 255 matches */
6593 else if (ANYOF_NONBITMAP(n)
6594 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6597 || (! (flags & ANYOF_LOCALE))
6598 || (flags & ANYOF_IS_SYNTHETIC)))))
6601 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6609 /* Not utf8. Convert as much of the string as available up
6610 * to the limit of how far the (single) character in the
6611 * pattern can possibly match (no need to go further). If
6612 * the node is a straight ANYOF or not folding, it can't
6613 * match more than one. Otherwise, It can match up to how
6614 * far a single char can fold to. Since not utf8, each
6615 * character is a single byte, so the max it can be in
6616 * bytes is the same as the max it can be in characters */
6617 STRLEN len = (OP(n) == ANYOF
6618 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6620 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6622 : UTF8_MAX_FOLD_CHAR_EXPAND;
6623 utf8_p = bytes_to_utf8(p, &len);
6626 if (swash_fetch(sw, utf8_p, TRUE))
6628 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6630 /* Here, we need to test if the fold of the target string
6631 * matches. The non-multi char folds have all been moved to
6632 * the compilation phase, and the multi-char folds have
6633 * been stored by regcomp into 'av'; we linearly check to
6634 * see if any match the target string (folded). We know
6635 * that the originals were each one character, but we don't
6636 * currently know how many characters/bytes each folded to,
6637 * except we do know that there are small limits imposed by
6638 * Unicode. XXX A performance enhancement would be to have
6639 * regcomp.c store the max number of chars/bytes that are
6640 * in an av entry, as, say the 0th element. Even better
6641 * would be to have a hash of the few characters that can
6642 * start a multi-char fold to the max number of chars of
6645 * If there is a match, we will need to advance (if lenp is
6646 * specified) the match pointer in the target string. But
6647 * what we are comparing here isn't that string directly,
6648 * but its fold, whose length may differ from the original.
6649 * As we go along in constructing the fold, therefore, we
6650 * create a map so that we know how many bytes in the
6651 * source to advance given that we have matched a certain
6652 * number of bytes in the fold. This map is stored in
6653 * 'map_fold_len_back'. Let n mean the number of bytes in
6654 * the fold of the first character that we are folding.
6655 * Then map_fold_len_back[n] is set to the number of bytes
6656 * in that first character. Similarly let m be the
6657 * corresponding number for the second character to be
6658 * folded. Then map_fold_len_back[n+m] is set to the
6659 * number of bytes occupied by the first two source
6660 * characters. ... */
6661 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6662 U8 folded[UTF8_MAXBYTES_CASE+1];
6663 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6664 STRLEN total_foldlen = 0; /* num bytes in fold of all
6667 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6669 /* Here, only need to fold the first char of the target
6670 * string. It the source wasn't utf8, is 1 byte long */
6671 to_utf8_fold(utf8_p, folded, &foldlen);
6672 total_foldlen = foldlen;
6673 map_fold_len_back[foldlen] = (utf8_target)
6679 /* Here, need to fold more than the first char. Do so
6680 * up to the limits */
6681 U8* source_ptr = utf8_p; /* The source for the fold
6684 U8* folded_ptr = folded;
6685 U8* e = utf8_p + maxlen; /* Can't go beyond last
6686 available byte in the
6690 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6694 /* Fold the next character */
6695 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6696 STRLEN this_char_foldlen;
6697 to_utf8_fold(source_ptr,
6699 &this_char_foldlen);
6701 /* Bail if it would exceed the byte limit for
6702 * folding a single char. */
6703 if (this_char_foldlen + folded_ptr - folded >
6709 /* Add the fold of this character */
6710 Copy(this_char_folded,
6714 source_ptr += UTF8SKIP(source_ptr);
6715 folded_ptr += this_char_foldlen;
6716 total_foldlen = folded_ptr - folded;
6718 /* Create map from the number of bytes in the fold
6719 * back to the number of bytes in the source. If
6720 * the source isn't utf8, the byte count is just
6721 * the number of characters so far */
6722 map_fold_len_back[total_foldlen]
6724 ? source_ptr - utf8_p
6731 /* Do the linear search to see if the fold is in the list
6732 * of multi-char folds. */
6735 for (i = 0; i <= av_len(av); i++) {
6736 SV* const sv = *av_fetch(av, i, FALSE);
6738 const char * const s = SvPV_const(sv, len);
6740 if (len <= total_foldlen && memEQ(s,
6745 /* Advance the target string ptr to account for
6746 * this fold, but have to translate from the
6747 * folded length to the corresponding source
6750 *lenp = map_fold_len_back[len];
6751 assert(*lenp != 0); /* Otherwise will loop */
6760 /* If we allocated a string above, free it */
6761 if (! utf8_target) Safefree(utf8_p);
6766 return (flags & ANYOF_INVERT) ? !match : match;
6770 S_reghop3(U8 *s, I32 off, const U8* lim)
6774 PERL_ARGS_ASSERT_REGHOP3;
6777 while (off-- && s < lim) {
6778 /* XXX could check well-formedness here */
6783 while (off++ && s > lim) {
6785 if (UTF8_IS_CONTINUED(*s)) {
6786 while (s > lim && UTF8_IS_CONTINUATION(*s))
6789 /* XXX could check well-formedness here */
6796 /* there are a bunch of places where we use two reghop3's that should
6797 be replaced with this routine. but since thats not done yet
6798 we ifdef it out - dmq
6801 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6805 PERL_ARGS_ASSERT_REGHOP4;
6808 while (off-- && s < rlim) {
6809 /* XXX could check well-formedness here */
6814 while (off++ && s > llim) {
6816 if (UTF8_IS_CONTINUED(*s)) {
6817 while (s > llim && UTF8_IS_CONTINUATION(*s))
6820 /* XXX could check well-formedness here */
6828 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6832 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6835 while (off-- && s < lim) {
6836 /* XXX could check well-formedness here */
6843 while (off++ && s > lim) {
6845 if (UTF8_IS_CONTINUED(*s)) {
6846 while (s > lim && UTF8_IS_CONTINUATION(*s))
6849 /* XXX could check well-formedness here */
6858 restore_pos(pTHX_ void *arg)
6861 regexp * const rex = (regexp *)arg;
6862 if (PL_reg_eval_set) {
6863 if (PL_reg_oldsaved) {
6864 rex->subbeg = PL_reg_oldsaved;
6865 rex->sublen = PL_reg_oldsavedlen;
6866 #ifdef PERL_OLD_COPY_ON_WRITE
6867 rex->saved_copy = PL_nrs;
6869 RXp_MATCH_COPIED_on(rex);
6871 PL_reg_magic->mg_len = PL_reg_oldpos;
6872 PL_reg_eval_set = 0;
6873 PL_curpm = PL_reg_oldcurpm;
6878 S_to_utf8_substr(pTHX_ register regexp *prog)
6882 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6885 if (prog->substrs->data[i].substr
6886 && !prog->substrs->data[i].utf8_substr) {
6887 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6888 prog->substrs->data[i].utf8_substr = sv;
6889 sv_utf8_upgrade(sv);
6890 if (SvVALID(prog->substrs->data[i].substr)) {
6891 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6892 if (flags & FBMcf_TAIL) {
6893 /* Trim the trailing \n that fbm_compile added last
6895 SvCUR_set(sv, SvCUR(sv) - 1);
6896 /* Whilst this makes the SV technically "invalid" (as its
6897 buffer is no longer followed by "\0") when fbm_compile()
6898 adds the "\n" back, a "\0" is restored. */
6900 fbm_compile(sv, flags);
6902 if (prog->substrs->data[i].substr == prog->check_substr)
6903 prog->check_utf8 = sv;
6909 S_to_byte_substr(pTHX_ register regexp *prog)
6914 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6917 if (prog->substrs->data[i].utf8_substr
6918 && !prog->substrs->data[i].substr) {
6919 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6920 if (sv_utf8_downgrade(sv, TRUE)) {
6921 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6923 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6924 if (flags & FBMcf_TAIL) {
6925 /* Trim the trailing \n that fbm_compile added last
6927 SvCUR_set(sv, SvCUR(sv) - 1);
6929 fbm_compile(sv, flags);
6935 prog->substrs->data[i].substr = sv;
6936 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6937 prog->check_substr = sv;
6944 * c-indentation-style: bsd
6946 * indent-tabs-mode: t
6949 * ex: set ts=8 sts=4 sw=4 noet: