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 while (s < strend) {
1419 if (REGINCLASS(prog, c, (U8*)s)) {
1420 if (tmp && (!reginfo || regtry(reginfo, &s)))
1433 if (tmp && (!reginfo || regtry(reginfo, &s)))
1441 if (UTF_PATTERN || utf8_target) {
1442 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1443 goto do_exactf_utf8;
1445 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1446 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1447 goto do_exactf_non_utf8; /* isn't dealt with by these */
1450 if (UTF_PATTERN || utf8_target) {
1451 utf8_fold_flags = 0;
1452 goto do_exactf_utf8;
1454 fold_array = PL_fold_latin1;
1455 folder = foldEQ_latin1;
1456 /* XXX This uses the full utf8 fold because if the pattern contains
1457 * 'ss' it could match LATIN_SMALL_LETTER SHARP_S in the string.
1458 * There could be a new node type, say EXACTFU_SS, which is
1459 * generated by regcomp only if there is an 'ss', and then every
1460 * other case could goto do_exactf_non_utf8;*/
1461 goto do_exactf_utf8;
1464 if (UTF_PATTERN || utf8_target) {
1465 utf8_fold_flags = 0;
1466 goto do_exactf_utf8;
1468 fold_array = PL_fold;
1470 goto do_exactf_non_utf8;
1473 if (UTF_PATTERN || utf8_target) {
1474 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1475 goto do_exactf_utf8;
1477 fold_array = PL_fold_locale;
1478 folder = foldEQ_locale;
1482 do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */
1484 /* The idea in the non-utf8 EXACTF* cases is to first find the
1485 * first character of the EXACTF* node and then, if necessary,
1486 * case-insensitively compare the full text of the node. c1 is the
1487 * first character. c2 is its fold. This logic will not work for
1488 * Unicode semantics and the german sharp ss, which hence should
1489 * not be compiled into a node that gets here. */
1490 pat_string = STRING(c);
1491 ln = STR_LEN(c); /* length to match in octets/bytes */
1493 e = HOP3c(strend, -((I32)ln), s);
1495 if (!reginfo && e < s) {
1496 e = s; /* Due to minlen logic of intuit() */
1500 c2 = fold_array[c1];
1501 if (c1 == c2) { /* If char and fold are the same */
1502 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1505 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1511 /* If one of the operands is in utf8, we can't use the simpler
1512 * folding above, due to the fact that many different characters
1513 * can have the same fold, or portion of a fold, or different-
1515 pat_string = STRING(c);
1516 ln = STR_LEN(c); /* length to match in octets/bytes */
1517 pat_end = pat_string + ln;
1518 lnc = (UTF_PATTERN) /* length to match in characters */
1519 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1522 e = HOP3c(strend, -((I32)lnc), s);
1524 if (!reginfo && e < s) {
1525 e = s; /* Due to minlen logic of intuit() */
1529 char *my_strend= (char *)strend;
1530 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1531 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1532 && (!reginfo || regtry(reginfo, &s)) )
1540 PL_reg_flags |= RF_tainted;
1541 FBC_BOUND(isALNUM_LC,
1542 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1543 isALNUM_LC_utf8((U8*)s));
1546 PL_reg_flags |= RF_tainted;
1547 FBC_NBOUND(isALNUM_LC,
1548 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1549 isALNUM_LC_utf8((U8*)s));
1552 FBC_BOUND(isWORDCHAR,
1554 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1557 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1559 isWORDCHAR_A((U8*)s));
1562 FBC_NBOUND(isWORDCHAR,
1564 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1567 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1569 isWORDCHAR_A((U8*)s));
1572 FBC_BOUND(isWORDCHAR_L1,
1574 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1577 FBC_NBOUND(isWORDCHAR_L1,
1579 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1582 REXEC_FBC_CSCAN_TAINT(
1583 isALNUM_LC_utf8((U8*)s),
1588 REXEC_FBC_CSCAN_PRELOAD(
1589 LOAD_UTF8_CHARCLASS_ALNUM(),
1590 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1591 isWORDCHAR_L1((U8) *s)
1595 REXEC_FBC_CSCAN_PRELOAD(
1596 LOAD_UTF8_CHARCLASS_ALNUM(),
1597 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1602 /* Don't need to worry about utf8, as it can match only a single
1603 * byte invariant character */
1604 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1607 REXEC_FBC_CSCAN_PRELOAD(
1608 LOAD_UTF8_CHARCLASS_ALNUM(),
1609 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1610 ! isWORDCHAR_L1((U8) *s)
1614 REXEC_FBC_CSCAN_PRELOAD(
1615 LOAD_UTF8_CHARCLASS_ALNUM(),
1616 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1627 REXEC_FBC_CSCAN_TAINT(
1628 !isALNUM_LC_utf8((U8*)s),
1633 REXEC_FBC_CSCAN_PRELOAD(
1634 LOAD_UTF8_CHARCLASS_SPACE(),
1635 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1640 REXEC_FBC_CSCAN_PRELOAD(
1641 LOAD_UTF8_CHARCLASS_SPACE(),
1642 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1647 /* Don't need to worry about utf8, as it can match only a single
1648 * byte invariant character */
1649 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1652 REXEC_FBC_CSCAN_TAINT(
1653 isSPACE_LC_utf8((U8*)s),
1658 REXEC_FBC_CSCAN_PRELOAD(
1659 LOAD_UTF8_CHARCLASS_SPACE(),
1660 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1661 ! isSPACE_L1((U8) *s)
1665 REXEC_FBC_CSCAN_PRELOAD(
1666 LOAD_UTF8_CHARCLASS_SPACE(),
1667 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1678 REXEC_FBC_CSCAN_TAINT(
1679 !isSPACE_LC_utf8((U8*)s),
1684 REXEC_FBC_CSCAN_PRELOAD(
1685 LOAD_UTF8_CHARCLASS_DIGIT(),
1686 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1691 /* Don't need to worry about utf8, as it can match only a single
1692 * byte invariant character */
1693 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1696 REXEC_FBC_CSCAN_TAINT(
1697 isDIGIT_LC_utf8((U8*)s),
1702 REXEC_FBC_CSCAN_PRELOAD(
1703 LOAD_UTF8_CHARCLASS_DIGIT(),
1704 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1715 REXEC_FBC_CSCAN_TAINT(
1716 !isDIGIT_LC_utf8((U8*)s),
1723 is_LNBREAK_latin1(s)
1735 !is_VERTWS_latin1(s)
1741 is_HORIZWS_latin1(s)
1746 !is_HORIZWS_utf8(s),
1747 !is_HORIZWS_latin1(s)
1754 /* what trie are we using right now */
1756 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1758 = (reg_trie_data*)progi->data->data[ aho->trie ];
1759 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1761 const char *last_start = strend - trie->minlen;
1763 const char *real_start = s;
1765 STRLEN maxlen = trie->maxlen;
1767 U8 **points; /* map of where we were in the input string
1768 when reading a given char. For ASCII this
1769 is unnecessary overhead as the relationship
1770 is always 1:1, but for Unicode, especially
1771 case folded Unicode this is not true. */
1772 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1776 GET_RE_DEBUG_FLAGS_DECL;
1778 /* We can't just allocate points here. We need to wrap it in
1779 * an SV so it gets freed properly if there is a croak while
1780 * running the match */
1783 sv_points=newSV(maxlen * sizeof(U8 *));
1784 SvCUR_set(sv_points,
1785 maxlen * sizeof(U8 *));
1786 SvPOK_on(sv_points);
1787 sv_2mortal(sv_points);
1788 points=(U8**)SvPV_nolen(sv_points );
1789 if ( trie_type != trie_utf8_fold
1790 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1793 bitmap=(U8*)trie->bitmap;
1795 bitmap=(U8*)ANYOF_BITMAP(c);
1797 /* this is the Aho-Corasick algorithm modified a touch
1798 to include special handling for long "unknown char"
1799 sequences. The basic idea being that we use AC as long
1800 as we are dealing with a possible matching char, when
1801 we encounter an unknown char (and we have not encountered
1802 an accepting state) we scan forward until we find a legal
1804 AC matching is basically that of trie matching, except
1805 that when we encounter a failing transition, we fall back
1806 to the current states "fail state", and try the current char
1807 again, a process we repeat until we reach the root state,
1808 state 1, or a legal transition. If we fail on the root state
1809 then we can either terminate if we have reached an accepting
1810 state previously, or restart the entire process from the beginning
1814 while (s <= last_start) {
1815 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1823 U8 *uscan = (U8*)NULL;
1824 U8 *leftmost = NULL;
1826 U32 accepted_word= 0;
1830 while ( state && uc <= (U8*)strend ) {
1832 U32 word = aho->states[ state ].wordnum;
1836 DEBUG_TRIE_EXECUTE_r(
1837 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1838 dump_exec_pos( (char *)uc, c, strend, real_start,
1839 (char *)uc, utf8_target );
1840 PerlIO_printf( Perl_debug_log,
1841 " Scanning for legal start char...\n");
1845 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1849 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1855 if (uc >(U8*)last_start) break;
1859 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1860 if (!leftmost || lpos < leftmost) {
1861 DEBUG_r(accepted_word=word);
1867 points[pointpos++ % maxlen]= uc;
1868 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1869 uscan, len, uvc, charid, foldlen,
1871 DEBUG_TRIE_EXECUTE_r({
1872 dump_exec_pos( (char *)uc, c, strend, real_start,
1874 PerlIO_printf(Perl_debug_log,
1875 " Charid:%3u CP:%4"UVxf" ",
1881 word = aho->states[ state ].wordnum;
1883 base = aho->states[ state ].trans.base;
1885 DEBUG_TRIE_EXECUTE_r({
1887 dump_exec_pos( (char *)uc, c, strend, real_start,
1889 PerlIO_printf( Perl_debug_log,
1890 "%sState: %4"UVxf", word=%"UVxf,
1891 failed ? " Fail transition to " : "",
1892 (UV)state, (UV)word);
1898 ( ((offset = base + charid
1899 - 1 - trie->uniquecharcount)) >= 0)
1900 && ((U32)offset < trie->lasttrans)
1901 && trie->trans[offset].check == state
1902 && (tmp=trie->trans[offset].next))
1904 DEBUG_TRIE_EXECUTE_r(
1905 PerlIO_printf( Perl_debug_log," - legal\n"));
1910 DEBUG_TRIE_EXECUTE_r(
1911 PerlIO_printf( Perl_debug_log," - fail\n"));
1913 state = aho->fail[state];
1917 /* we must be accepting here */
1918 DEBUG_TRIE_EXECUTE_r(
1919 PerlIO_printf( Perl_debug_log," - accepting\n"));
1928 if (!state) state = 1;
1931 if ( aho->states[ state ].wordnum ) {
1932 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1933 if (!leftmost || lpos < leftmost) {
1934 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1939 s = (char*)leftmost;
1940 DEBUG_TRIE_EXECUTE_r({
1942 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1943 (UV)accepted_word, (IV)(s - real_start)
1946 if (!reginfo || regtry(reginfo, &s)) {
1952 DEBUG_TRIE_EXECUTE_r({
1953 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1956 DEBUG_TRIE_EXECUTE_r(
1957 PerlIO_printf( Perl_debug_log,"No match.\n"));
1966 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1976 - regexec_flags - match a regexp against a string
1979 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1980 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1981 /* strend: pointer to null at end of string */
1982 /* strbeg: real beginning of string */
1983 /* minend: end of match must be >=minend after stringarg. */
1984 /* data: May be used for some additional optimizations.
1985 Currently its only used, with a U32 cast, for transmitting
1986 the ganch offset when doing a /g match. This will change */
1987 /* nosave: For optimizations. */
1990 struct regexp *const prog = (struct regexp *)SvANY(rx);
1991 /*register*/ char *s;
1992 register regnode *c;
1993 /*register*/ char *startpos = stringarg;
1994 I32 minlen; /* must match at least this many chars */
1995 I32 dontbother = 0; /* how many characters not to try at end */
1996 I32 end_shift = 0; /* Same for the end. */ /* CC */
1997 I32 scream_pos = -1; /* Internal iterator of scream. */
1998 char *scream_olds = NULL;
1999 const bool utf8_target = cBOOL(DO_UTF8(sv));
2001 RXi_GET_DECL(prog,progi);
2002 regmatch_info reginfo; /* create some info to pass to regtry etc */
2003 regexp_paren_pair *swap = NULL;
2004 GET_RE_DEBUG_FLAGS_DECL;
2006 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2007 PERL_UNUSED_ARG(data);
2009 /* Be paranoid... */
2010 if (prog == NULL || startpos == NULL) {
2011 Perl_croak(aTHX_ "NULL regexp parameter");
2015 multiline = prog->extflags & RXf_PMf_MULTILINE;
2016 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2018 RX_MATCH_UTF8_set(rx, utf8_target);
2020 debug_start_match(rx, utf8_target, startpos, strend,
2024 minlen = prog->minlen;
2026 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2027 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2028 "String too short [regexec_flags]...\n"));
2033 /* Check validity of program. */
2034 if (UCHARAT(progi->program) != REG_MAGIC) {
2035 Perl_croak(aTHX_ "corrupted regexp program");
2039 PL_reg_eval_set = 0;
2043 PL_reg_flags |= RF_utf8;
2045 /* Mark beginning of line for ^ and lookbehind. */
2046 reginfo.bol = startpos; /* XXX not used ??? */
2050 /* Mark end of line for $ (and such) */
2053 /* see how far we have to get to not match where we matched before */
2054 reginfo.till = startpos+minend;
2056 /* If there is a "must appear" string, look for it. */
2059 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2061 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2062 reginfo.ganch = startpos + prog->gofs;
2063 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2064 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2065 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2067 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2068 && mg->mg_len >= 0) {
2069 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2070 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2071 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2073 if (prog->extflags & RXf_ANCH_GPOS) {
2074 if (s > reginfo.ganch)
2076 s = reginfo.ganch - prog->gofs;
2077 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2078 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2084 reginfo.ganch = strbeg + PTR2UV(data);
2085 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2086 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2088 } else { /* pos() not defined */
2089 reginfo.ganch = strbeg;
2090 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2091 "GPOS: reginfo.ganch = strbeg\n"));
2094 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2095 /* We have to be careful. If the previous successful match
2096 was from this regex we don't want a subsequent partially
2097 successful match to clobber the old results.
2098 So when we detect this possibility we add a swap buffer
2099 to the re, and switch the buffer each match. If we fail
2100 we switch it back, otherwise we leave it swapped.
2103 /* do we need a save destructor here for eval dies? */
2104 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2106 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2107 re_scream_pos_data d;
2109 d.scream_olds = &scream_olds;
2110 d.scream_pos = &scream_pos;
2111 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2113 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2114 goto phooey; /* not present */
2120 /* Simplest case: anchored match need be tried only once. */
2121 /* [unless only anchor is BOL and multiline is set] */
2122 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2123 if (s == startpos && regtry(®info, &startpos))
2125 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2126 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2131 dontbother = minlen - 1;
2132 end = HOP3c(strend, -dontbother, strbeg) - 1;
2133 /* for multiline we only have to try after newlines */
2134 if (prog->check_substr || prog->check_utf8) {
2135 /* because of the goto we can not easily reuse the macros for bifurcating the
2136 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2139 goto after_try_utf8;
2141 if (regtry(®info, &s)) {
2148 if (prog->extflags & RXf_USE_INTUIT) {
2149 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2158 } /* end search for check string in unicode */
2160 if (s == startpos) {
2161 goto after_try_latin;
2164 if (regtry(®info, &s)) {
2171 if (prog->extflags & RXf_USE_INTUIT) {
2172 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2181 } /* end search for check string in latin*/
2182 } /* end search for check string */
2183 else { /* search for newline */
2185 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2188 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2190 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2191 if (regtry(®info, &s))
2195 } /* end search for newline */
2196 } /* end anchored/multiline check string search */
2198 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2200 /* the warning about reginfo.ganch being used without initialization
2201 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2202 and we only enter this block when the same bit is set. */
2203 char *tmp_s = reginfo.ganch - prog->gofs;
2205 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2210 /* Messy cases: unanchored match. */
2211 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2212 /* we have /x+whatever/ */
2213 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2218 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2219 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2220 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2225 DEBUG_EXECUTE_r( did_match = 1 );
2226 if (regtry(®info, &s)) goto got_it;
2228 while (s < strend && *s == ch)
2236 DEBUG_EXECUTE_r( did_match = 1 );
2237 if (regtry(®info, &s)) goto got_it;
2239 while (s < strend && *s == ch)
2244 DEBUG_EXECUTE_r(if (!did_match)
2245 PerlIO_printf(Perl_debug_log,
2246 "Did not find anchored character...\n")
2249 else if (prog->anchored_substr != NULL
2250 || prog->anchored_utf8 != NULL
2251 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2252 && prog->float_max_offset < strend - s)) {
2257 char *last1; /* Last position checked before */
2261 if (prog->anchored_substr || prog->anchored_utf8) {
2262 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2263 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2264 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2265 back_max = back_min = prog->anchored_offset;
2267 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2268 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2269 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2270 back_max = prog->float_max_offset;
2271 back_min = prog->float_min_offset;
2275 if (must == &PL_sv_undef)
2276 /* could not downgrade utf8 check substring, so must fail */
2282 last = HOP3c(strend, /* Cannot start after this */
2283 -(I32)(CHR_SVLEN(must)
2284 - (SvTAIL(must) != 0) + back_min), strbeg);
2287 last1 = HOPc(s, -1);
2289 last1 = s - 1; /* bogus */
2291 /* XXXX check_substr already used to find "s", can optimize if
2292 check_substr==must. */
2294 dontbother = end_shift;
2295 strend = HOPc(strend, -dontbother);
2296 while ( (s <= last) &&
2297 ((flags & REXEC_SCREAM)
2298 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2299 end_shift, &scream_pos, 0))
2300 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2301 (unsigned char*)strend, must,
2302 multiline ? FBMrf_MULTILINE : 0))) ) {
2303 /* we may be pointing at the wrong string */
2304 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2305 s = strbeg + (s - SvPVX_const(sv));
2306 DEBUG_EXECUTE_r( did_match = 1 );
2307 if (HOPc(s, -back_max) > last1) {
2308 last1 = HOPc(s, -back_min);
2309 s = HOPc(s, -back_max);
2312 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2314 last1 = HOPc(s, -back_min);
2318 while (s <= last1) {
2319 if (regtry(®info, &s))
2325 while (s <= last1) {
2326 if (regtry(®info, &s))
2332 DEBUG_EXECUTE_r(if (!did_match) {
2333 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2334 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2335 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2336 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2337 ? "anchored" : "floating"),
2338 quoted, RE_SV_TAIL(must));
2342 else if ( (c = progi->regstclass) ) {
2344 const OPCODE op = OP(progi->regstclass);
2345 /* don't bother with what can't match */
2346 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2347 strend = HOPc(strend, -(minlen - 1));
2350 SV * const prop = sv_newmortal();
2351 regprop(prog, prop, c);
2353 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2355 PerlIO_printf(Perl_debug_log,
2356 "Matching stclass %.*s against %s (%d bytes)\n",
2357 (int)SvCUR(prop), SvPVX_const(prop),
2358 quoted, (int)(strend - s));
2361 if (find_byclass(prog, c, s, strend, ®info))
2363 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2367 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2372 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2373 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2374 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2376 if (flags & REXEC_SCREAM) {
2377 last = screaminstr(sv, float_real, s - strbeg,
2378 end_shift, &scream_pos, 1); /* last one */
2380 last = scream_olds; /* Only one occurrence. */
2381 /* we may be pointing at the wrong string */
2382 else if (RXp_MATCH_COPIED(prog))
2383 s = strbeg + (s - SvPVX_const(sv));
2387 const char * const little = SvPV_const(float_real, len);
2389 if (SvTAIL(float_real)) {
2390 if (memEQ(strend - len + 1, little, len - 1))
2391 last = strend - len + 1;
2392 else if (!multiline)
2393 last = memEQ(strend - len, little, len)
2394 ? strend - len : NULL;
2400 last = rninstr(s, strend, little, little + len);
2402 last = strend; /* matching "$" */
2407 PerlIO_printf(Perl_debug_log,
2408 "%sCan't trim the tail, match fails (should not happen)%s\n",
2409 PL_colors[4], PL_colors[5]));
2410 goto phooey; /* Should not happen! */
2412 dontbother = strend - last + prog->float_min_offset;
2414 if (minlen && (dontbother < minlen))
2415 dontbother = minlen - 1;
2416 strend -= dontbother; /* this one's always in bytes! */
2417 /* We don't know much -- general case. */
2420 if (regtry(®info, &s))
2429 if (regtry(®info, &s))
2431 } while (s++ < strend);
2440 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2442 if (PL_reg_eval_set)
2443 restore_pos(aTHX_ prog);
2444 if (RXp_PAREN_NAMES(prog))
2445 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2447 /* make sure $`, $&, $', and $digit will work later */
2448 if ( !(flags & REXEC_NOT_FIRST) ) {
2449 RX_MATCH_COPY_FREE(rx);
2450 if (flags & REXEC_COPY_STR) {
2451 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2452 #ifdef PERL_OLD_COPY_ON_WRITE
2454 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2456 PerlIO_printf(Perl_debug_log,
2457 "Copy on write: regexp capture, type %d\n",
2460 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2461 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2462 assert (SvPOKp(prog->saved_copy));
2466 RX_MATCH_COPIED_on(rx);
2467 s = savepvn(strbeg, i);
2473 prog->subbeg = strbeg;
2474 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2481 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2482 PL_colors[4], PL_colors[5]));
2483 if (PL_reg_eval_set)
2484 restore_pos(aTHX_ prog);
2486 /* we failed :-( roll it back */
2487 Safefree(prog->offs);
2496 - regtry - try match at specific point
2498 STATIC I32 /* 0 failure, 1 success */
2499 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2503 REGEXP *const rx = reginfo->prog;
2504 regexp *const prog = (struct regexp *)SvANY(rx);
2505 RXi_GET_DECL(prog,progi);
2506 GET_RE_DEBUG_FLAGS_DECL;
2508 PERL_ARGS_ASSERT_REGTRY;
2510 reginfo->cutpoint=NULL;
2512 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2515 PL_reg_eval_set = RS_init;
2516 DEBUG_EXECUTE_r(DEBUG_s(
2517 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2518 (IV)(PL_stack_sp - PL_stack_base));
2521 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2522 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2524 /* Apparently this is not needed, judging by wantarray. */
2525 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2526 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2529 /* Make $_ available to executed code. */
2530 if (reginfo->sv != DEFSV) {
2532 DEFSV_set(reginfo->sv);
2535 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2536 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2537 /* prepare for quick setting of pos */
2538 #ifdef PERL_OLD_COPY_ON_WRITE
2539 if (SvIsCOW(reginfo->sv))
2540 sv_force_normal_flags(reginfo->sv, 0);
2542 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2543 &PL_vtbl_mglob, NULL, 0);
2547 PL_reg_oldpos = mg->mg_len;
2548 SAVEDESTRUCTOR_X(restore_pos, prog);
2550 if (!PL_reg_curpm) {
2551 Newxz(PL_reg_curpm, 1, PMOP);
2554 SV* const repointer = &PL_sv_undef;
2555 /* this regexp is also owned by the new PL_reg_curpm, which
2556 will try to free it. */
2557 av_push(PL_regex_padav, repointer);
2558 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2559 PL_regex_pad = AvARRAY(PL_regex_padav);
2564 /* It seems that non-ithreads works both with and without this code.
2565 So for efficiency reasons it seems best not to have the code
2566 compiled when it is not needed. */
2567 /* This is safe against NULLs: */
2568 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2569 /* PM_reg_curpm owns a reference to this regexp. */
2572 PM_SETRE(PL_reg_curpm, rx);
2573 PL_reg_oldcurpm = PL_curpm;
2574 PL_curpm = PL_reg_curpm;
2575 if (RXp_MATCH_COPIED(prog)) {
2576 /* Here is a serious problem: we cannot rewrite subbeg,
2577 since it may be needed if this match fails. Thus
2578 $` inside (?{}) could fail... */
2579 PL_reg_oldsaved = prog->subbeg;
2580 PL_reg_oldsavedlen = prog->sublen;
2581 #ifdef PERL_OLD_COPY_ON_WRITE
2582 PL_nrs = prog->saved_copy;
2584 RXp_MATCH_COPIED_off(prog);
2587 PL_reg_oldsaved = NULL;
2588 prog->subbeg = PL_bostr;
2589 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2591 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2592 prog->offs[0].start = *startpos - PL_bostr;
2593 PL_reginput = *startpos;
2594 PL_reglastparen = &prog->lastparen;
2595 PL_reglastcloseparen = &prog->lastcloseparen;
2596 prog->lastparen = 0;
2597 prog->lastcloseparen = 0;
2599 PL_regoffs = prog->offs;
2600 if (PL_reg_start_tmpl <= prog->nparens) {
2601 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2602 if(PL_reg_start_tmp)
2603 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2605 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2608 /* XXXX What this code is doing here?!!! There should be no need
2609 to do this again and again, PL_reglastparen should take care of
2612 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2613 * Actually, the code in regcppop() (which Ilya may be meaning by
2614 * PL_reglastparen), is not needed at all by the test suite
2615 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2616 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2617 * Meanwhile, this code *is* needed for the
2618 * above-mentioned test suite tests to succeed. The common theme
2619 * on those tests seems to be returning null fields from matches.
2620 * --jhi updated by dapm */
2622 if (prog->nparens) {
2623 regexp_paren_pair *pp = PL_regoffs;
2625 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2633 if (regmatch(reginfo, progi->program + 1)) {
2634 PL_regoffs[0].end = PL_reginput - PL_bostr;
2637 if (reginfo->cutpoint)
2638 *startpos= reginfo->cutpoint;
2639 REGCP_UNWIND(lastcp);
2644 #define sayYES goto yes
2645 #define sayNO goto no
2646 #define sayNO_SILENT goto no_silent
2648 /* we dont use STMT_START/END here because it leads to
2649 "unreachable code" warnings, which are bogus, but distracting. */
2650 #define CACHEsayNO \
2651 if (ST.cache_mask) \
2652 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2655 /* this is used to determine how far from the left messages like
2656 'failed...' are printed. It should be set such that messages
2657 are inline with the regop output that created them.
2659 #define REPORT_CODE_OFF 32
2662 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2663 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2665 #define SLAB_FIRST(s) (&(s)->states[0])
2666 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2668 /* grab a new slab and return the first slot in it */
2670 STATIC regmatch_state *
2673 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2676 regmatch_slab *s = PL_regmatch_slab->next;
2678 Newx(s, 1, regmatch_slab);
2679 s->prev = PL_regmatch_slab;
2681 PL_regmatch_slab->next = s;
2683 PL_regmatch_slab = s;
2684 return SLAB_FIRST(s);
2688 /* push a new state then goto it */
2690 #define PUSH_STATE_GOTO(state, node) \
2692 st->resume_state = state; \
2695 /* push a new state with success backtracking, then goto it */
2697 #define PUSH_YES_STATE_GOTO(state, node) \
2699 st->resume_state = state; \
2700 goto push_yes_state;
2706 regmatch() - main matching routine
2708 This is basically one big switch statement in a loop. We execute an op,
2709 set 'next' to point the next op, and continue. If we come to a point which
2710 we may need to backtrack to on failure such as (A|B|C), we push a
2711 backtrack state onto the backtrack stack. On failure, we pop the top
2712 state, and re-enter the loop at the state indicated. If there are no more
2713 states to pop, we return failure.
2715 Sometimes we also need to backtrack on success; for example /A+/, where
2716 after successfully matching one A, we need to go back and try to
2717 match another one; similarly for lookahead assertions: if the assertion
2718 completes successfully, we backtrack to the state just before the assertion
2719 and then carry on. In these cases, the pushed state is marked as
2720 'backtrack on success too'. This marking is in fact done by a chain of
2721 pointers, each pointing to the previous 'yes' state. On success, we pop to
2722 the nearest yes state, discarding any intermediate failure-only states.
2723 Sometimes a yes state is pushed just to force some cleanup code to be
2724 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2725 it to free the inner regex.
2727 Note that failure backtracking rewinds the cursor position, while
2728 success backtracking leaves it alone.
2730 A pattern is complete when the END op is executed, while a subpattern
2731 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2732 ops trigger the "pop to last yes state if any, otherwise return true"
2735 A common convention in this function is to use A and B to refer to the two
2736 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2737 the subpattern to be matched possibly multiple times, while B is the entire
2738 rest of the pattern. Variable and state names reflect this convention.
2740 The states in the main switch are the union of ops and failure/success of
2741 substates associated with with that op. For example, IFMATCH is the op
2742 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2743 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2744 successfully matched A and IFMATCH_A_fail is a state saying that we have
2745 just failed to match A. Resume states always come in pairs. The backtrack
2746 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2747 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2748 on success or failure.
2750 The struct that holds a backtracking state is actually a big union, with
2751 one variant for each major type of op. The variable st points to the
2752 top-most backtrack struct. To make the code clearer, within each
2753 block of code we #define ST to alias the relevant union.
2755 Here's a concrete example of a (vastly oversimplified) IFMATCH
2761 #define ST st->u.ifmatch
2763 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2764 ST.foo = ...; // some state we wish to save
2766 // push a yes backtrack state with a resume value of
2767 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2769 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2772 case IFMATCH_A: // we have successfully executed A; now continue with B
2774 bar = ST.foo; // do something with the preserved value
2777 case IFMATCH_A_fail: // A failed, so the assertion failed
2778 ...; // do some housekeeping, then ...
2779 sayNO; // propagate the failure
2786 For any old-timers reading this who are familiar with the old recursive
2787 approach, the code above is equivalent to:
2789 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2798 ...; // do some housekeeping, then ...
2799 sayNO; // propagate the failure
2802 The topmost backtrack state, pointed to by st, is usually free. If you
2803 want to claim it, populate any ST.foo fields in it with values you wish to
2804 save, then do one of
2806 PUSH_STATE_GOTO(resume_state, node);
2807 PUSH_YES_STATE_GOTO(resume_state, node);
2809 which sets that backtrack state's resume value to 'resume_state', pushes a
2810 new free entry to the top of the backtrack stack, then goes to 'node'.
2811 On backtracking, the free slot is popped, and the saved state becomes the
2812 new free state. An ST.foo field in this new top state can be temporarily
2813 accessed to retrieve values, but once the main loop is re-entered, it
2814 becomes available for reuse.
2816 Note that the depth of the backtrack stack constantly increases during the
2817 left-to-right execution of the pattern, rather than going up and down with
2818 the pattern nesting. For example the stack is at its maximum at Z at the
2819 end of the pattern, rather than at X in the following:
2821 /(((X)+)+)+....(Y)+....Z/
2823 The only exceptions to this are lookahead/behind assertions and the cut,
2824 (?>A), which pop all the backtrack states associated with A before
2827 Backtrack state structs are allocated in slabs of about 4K in size.
2828 PL_regmatch_state and st always point to the currently active state,
2829 and PL_regmatch_slab points to the slab currently containing
2830 PL_regmatch_state. The first time regmatch() is called, the first slab is
2831 allocated, and is never freed until interpreter destruction. When the slab
2832 is full, a new one is allocated and chained to the end. At exit from
2833 regmatch(), slabs allocated since entry are freed.
2838 #define DEBUG_STATE_pp(pp) \
2840 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2841 PerlIO_printf(Perl_debug_log, \
2842 " %*s"pp" %s%s%s%s%s\n", \
2844 PL_reg_name[st->resume_state], \
2845 ((st==yes_state||st==mark_state) ? "[" : ""), \
2846 ((st==yes_state) ? "Y" : ""), \
2847 ((st==mark_state) ? "M" : ""), \
2848 ((st==yes_state||st==mark_state) ? "]" : "") \
2853 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2858 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2859 const char *start, const char *end, const char *blurb)
2861 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2863 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2868 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2869 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2871 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2872 start, end - start, 60);
2874 PerlIO_printf(Perl_debug_log,
2875 "%s%s REx%s %s against %s\n",
2876 PL_colors[4], blurb, PL_colors[5], s0, s1);
2878 if (utf8_target||utf8_pat)
2879 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2880 utf8_pat ? "pattern" : "",
2881 utf8_pat && utf8_target ? " and " : "",
2882 utf8_target ? "string" : ""
2888 S_dump_exec_pos(pTHX_ const char *locinput,
2889 const regnode *scan,
2890 const char *loc_regeol,
2891 const char *loc_bostr,
2892 const char *loc_reg_starttry,
2893 const bool utf8_target)
2895 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2896 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2897 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2898 /* The part of the string before starttry has one color
2899 (pref0_len chars), between starttry and current
2900 position another one (pref_len - pref0_len chars),
2901 after the current position the third one.
2902 We assume that pref0_len <= pref_len, otherwise we
2903 decrease pref0_len. */
2904 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2905 ? (5 + taill) - l : locinput - loc_bostr;
2908 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2910 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2912 pref0_len = pref_len - (locinput - loc_reg_starttry);
2913 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2914 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2915 ? (5 + taill) - pref_len : loc_regeol - locinput);
2916 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2920 if (pref0_len > pref_len)
2921 pref0_len = pref_len;
2923 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2925 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2926 (locinput - pref_len),pref0_len, 60, 4, 5);
2928 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2929 (locinput - pref_len + pref0_len),
2930 pref_len - pref0_len, 60, 2, 3);
2932 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2933 locinput, loc_regeol - locinput, 10, 0, 1);
2935 const STRLEN tlen=len0+len1+len2;
2936 PerlIO_printf(Perl_debug_log,
2937 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2938 (IV)(locinput - loc_bostr),
2941 (docolor ? "" : "> <"),
2943 (int)(tlen > 19 ? 0 : 19 - tlen),
2950 /* reg_check_named_buff_matched()
2951 * Checks to see if a named buffer has matched. The data array of
2952 * buffer numbers corresponding to the buffer is expected to reside
2953 * in the regexp->data->data array in the slot stored in the ARG() of
2954 * node involved. Note that this routine doesn't actually care about the
2955 * name, that information is not preserved from compilation to execution.
2956 * Returns the index of the leftmost defined buffer with the given name
2957 * or 0 if non of the buffers matched.
2960 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2963 RXi_GET_DECL(rex,rexi);
2964 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2965 I32 *nums=(I32*)SvPVX(sv_dat);
2967 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2969 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2970 if ((I32)*PL_reglastparen >= nums[n] &&
2971 PL_regoffs[nums[n]].end != -1)
2980 /* free all slabs above current one - called during LEAVE_SCOPE */
2983 S_clear_backtrack_stack(pTHX_ void *p)
2985 regmatch_slab *s = PL_regmatch_slab->next;
2990 PL_regmatch_slab->next = NULL;
2992 regmatch_slab * const osl = s;
2999 #define SETREX(Re1,Re2) \
3000 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3003 STATIC I32 /* 0 failure, 1 success */
3004 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3006 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3010 register const bool utf8_target = PL_reg_match_utf8;
3011 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3012 REGEXP *rex_sv = reginfo->prog;
3013 regexp *rex = (struct regexp *)SvANY(rex_sv);
3014 RXi_GET_DECL(rex,rexi);
3016 /* the current state. This is a cached copy of PL_regmatch_state */
3017 register regmatch_state *st;
3018 /* cache heavy used fields of st in registers */
3019 register regnode *scan;
3020 register regnode *next;
3021 register U32 n = 0; /* general value; init to avoid compiler warning */
3022 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3023 register char *locinput = PL_reginput;
3024 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3026 bool result = 0; /* return value of S_regmatch */
3027 int depth = 0; /* depth of backtrack stack */
3028 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3029 const U32 max_nochange_depth =
3030 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3031 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3032 regmatch_state *yes_state = NULL; /* state to pop to on success of
3034 /* mark_state piggy backs on the yes_state logic so that when we unwind
3035 the stack on success we can update the mark_state as we go */
3036 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3037 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3038 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3040 bool no_final = 0; /* prevent failure from backtracking? */
3041 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3042 char *startpoint = PL_reginput;
3043 SV *popmark = NULL; /* are we looking for a mark? */
3044 SV *sv_commit = NULL; /* last mark name seen in failure */
3045 SV *sv_yes_mark = NULL; /* last mark name we have seen
3046 during a successful match */
3047 U32 lastopen = 0; /* last open we saw */
3048 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3049 SV* const oreplsv = GvSV(PL_replgv);
3050 /* these three flags are set by various ops to signal information to
3051 * the very next op. They have a useful lifetime of exactly one loop
3052 * iteration, and are not preserved or restored by state pushes/pops
3054 bool sw = 0; /* the condition value in (?(cond)a|b) */
3055 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3056 int logical = 0; /* the following EVAL is:
3060 or the following IFMATCH/UNLESSM is:
3061 false: plain (?=foo)
3062 true: used as a condition: (?(?=foo))
3065 GET_RE_DEBUG_FLAGS_DECL;
3068 PERL_ARGS_ASSERT_REGMATCH;
3070 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3071 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3073 /* on first ever call to regmatch, allocate first slab */
3074 if (!PL_regmatch_slab) {
3075 Newx(PL_regmatch_slab, 1, regmatch_slab);
3076 PL_regmatch_slab->prev = NULL;
3077 PL_regmatch_slab->next = NULL;
3078 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3081 oldsave = PL_savestack_ix;
3082 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3083 SAVEVPTR(PL_regmatch_slab);
3084 SAVEVPTR(PL_regmatch_state);
3086 /* grab next free state slot */
3087 st = ++PL_regmatch_state;
3088 if (st > SLAB_LAST(PL_regmatch_slab))
3089 st = PL_regmatch_state = S_push_slab(aTHX);
3091 /* Note that nextchr is a byte even in UTF */
3092 nextchr = UCHARAT(locinput);
3094 while (scan != NULL) {
3097 SV * const prop = sv_newmortal();
3098 regnode *rnext=regnext(scan);
3099 DUMP_EXEC_POS( locinput, scan, utf8_target );
3100 regprop(rex, prop, scan);
3102 PerlIO_printf(Perl_debug_log,
3103 "%3"IVdf":%*s%s(%"IVdf")\n",
3104 (IV)(scan - rexi->program), depth*2, "",
3106 (PL_regkind[OP(scan)] == END || !rnext) ?
3107 0 : (IV)(rnext - rexi->program));
3110 next = scan + NEXT_OFF(scan);
3113 state_num = OP(scan);
3117 assert(PL_reglastparen == &rex->lastparen);
3118 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3119 assert(PL_regoffs == rex->offs);
3121 switch (state_num) {
3123 if (locinput == PL_bostr)
3125 /* reginfo->till = reginfo->bol; */
3130 if (locinput == PL_bostr ||
3131 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3137 if (locinput == PL_bostr)
3141 if (locinput == reginfo->ganch)
3146 /* update the startpoint */
3147 st->u.keeper.val = PL_regoffs[0].start;
3148 PL_reginput = locinput;
3149 PL_regoffs[0].start = locinput - PL_bostr;
3150 PUSH_STATE_GOTO(KEEPS_next, next);
3152 case KEEPS_next_fail:
3153 /* rollback the start point change */
3154 PL_regoffs[0].start = st->u.keeper.val;
3160 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3165 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3167 if (PL_regeol - locinput > 1)
3171 if (PL_regeol != locinput)
3175 if (!nextchr && locinput >= PL_regeol)
3178 locinput += PL_utf8skip[nextchr];
3179 if (locinput > PL_regeol)
3181 nextchr = UCHARAT(locinput);
3184 nextchr = UCHARAT(++locinput);
3187 if (!nextchr && locinput >= PL_regeol)
3189 nextchr = UCHARAT(++locinput);
3192 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3195 locinput += PL_utf8skip[nextchr];
3196 if (locinput > PL_regeol)
3198 nextchr = UCHARAT(locinput);
3201 nextchr = UCHARAT(++locinput);
3205 #define ST st->u.trie
3207 /* In this case the charclass data is available inline so
3208 we can fail fast without a lot of extra overhead.
3210 if (scan->flags == EXACT || !utf8_target) {
3211 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3213 PerlIO_printf(Perl_debug_log,
3214 "%*s %sfailed to match trie start class...%s\n",
3215 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3223 /* the basic plan of execution of the trie is:
3224 * At the beginning, run though all the states, and
3225 * find the longest-matching word. Also remember the position
3226 * of the shortest matching word. For example, this pattern:
3229 * when matched against the string "abcde", will generate
3230 * accept states for all words except 3, with the longest
3231 * matching word being 4, and the shortest being 1 (with
3232 * the position being after char 1 of the string).
3234 * Then for each matching word, in word order (i.e. 1,2,4,5),
3235 * we run the remainder of the pattern; on each try setting
3236 * the current position to the character following the word,
3237 * returning to try the next word on failure.
3239 * We avoid having to build a list of words at runtime by
3240 * using a compile-time structure, wordinfo[].prev, which
3241 * gives, for each word, the previous accepting word (if any).
3242 * In the case above it would contain the mappings 1->2, 2->0,
3243 * 3->0, 4->5, 5->1. We can use this table to generate, from
3244 * the longest word (4 above), a list of all words, by
3245 * following the list of prev pointers; this gives us the
3246 * unordered list 4,5,1,2. Then given the current word we have
3247 * just tried, we can go through the list and find the
3248 * next-biggest word to try (so if we just failed on word 2,
3249 * the next in the list is 4).
3251 * Since at runtime we don't record the matching position in
3252 * the string for each word, we have to work that out for
3253 * each word we're about to process. The wordinfo table holds
3254 * the character length of each word; given that we recorded
3255 * at the start: the position of the shortest word and its
3256 * length in chars, we just need to move the pointer the
3257 * difference between the two char lengths. Depending on
3258 * Unicode status and folding, that's cheap or expensive.
3260 * This algorithm is optimised for the case where are only a
3261 * small number of accept states, i.e. 0,1, or maybe 2.
3262 * With lots of accepts states, and having to try all of them,
3263 * it becomes quadratic on number of accept states to find all
3268 /* what type of TRIE am I? (utf8 makes this contextual) */
3269 DECL_TRIE_TYPE(scan);
3271 /* what trie are we using right now */
3272 reg_trie_data * const trie
3273 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3274 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3275 U32 state = trie->startstate;
3277 if (trie->bitmap && trie_type != trie_utf8_fold &&
3278 !TRIE_BITMAP_TEST(trie,*locinput)
3280 if (trie->states[ state ].wordnum) {
3282 PerlIO_printf(Perl_debug_log,
3283 "%*s %smatched empty string...%s\n",
3284 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3290 PerlIO_printf(Perl_debug_log,
3291 "%*s %sfailed to match trie start class...%s\n",
3292 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3299 U8 *uc = ( U8* )locinput;
3303 U8 *uscan = (U8*)NULL;
3304 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3305 U32 charcount = 0; /* how many input chars we have matched */
3306 U32 accepted = 0; /* have we seen any accepting states? */
3309 ST.jump = trie->jump;
3312 ST.longfold = FALSE; /* char longer if folded => it's harder */
3315 /* fully traverse the TRIE; note the position of the
3316 shortest accept state and the wordnum of the longest
3319 while ( state && uc <= (U8*)PL_regeol ) {
3320 U32 base = trie->states[ state ].trans.base;
3324 wordnum = trie->states[ state ].wordnum;
3326 if (wordnum) { /* it's an accept state */
3329 /* record first match position */
3331 ST.firstpos = (U8*)locinput;
3336 ST.firstchars = charcount;
3339 if (!ST.nextword || wordnum < ST.nextword)
3340 ST.nextword = wordnum;
3341 ST.topword = wordnum;
3344 DEBUG_TRIE_EXECUTE_r({
3345 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3346 PerlIO_printf( Perl_debug_log,
3347 "%*s %sState: %4"UVxf" Accepted: %c ",
3348 2+depth * 2, "", PL_colors[4],
3349 (UV)state, (accepted ? 'Y' : 'N'));
3352 /* read a char and goto next state */
3355 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3356 uscan, len, uvc, charid, foldlen,
3363 base + charid - 1 - trie->uniquecharcount)) >= 0)
3365 && ((U32)offset < trie->lasttrans)
3366 && trie->trans[offset].check == state)
3368 state = trie->trans[offset].next;
3379 DEBUG_TRIE_EXECUTE_r(
3380 PerlIO_printf( Perl_debug_log,
3381 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3382 charid, uvc, (UV)state, PL_colors[5] );
3388 /* calculate total number of accept states */
3393 w = trie->wordinfo[w].prev;
3396 ST.accepted = accepted;
3400 PerlIO_printf( Perl_debug_log,
3401 "%*s %sgot %"IVdf" possible matches%s\n",
3402 REPORT_CODE_OFF + depth * 2, "",
3403 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3405 goto trie_first_try; /* jump into the fail handler */
3409 case TRIE_next_fail: /* we failed - try next alternative */
3411 REGCP_UNWIND(ST.cp);
3412 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3413 PL_regoffs[n].end = -1;
3414 *PL_reglastparen = n;
3416 if (!--ST.accepted) {
3418 PerlIO_printf( Perl_debug_log,
3419 "%*s %sTRIE failed...%s\n",
3420 REPORT_CODE_OFF+depth*2, "",
3427 /* Find next-highest word to process. Note that this code
3428 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3429 register U16 min = 0;
3431 register U16 const nextword = ST.nextword;
3432 register reg_trie_wordinfo * const wordinfo
3433 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3434 for (word=ST.topword; word; word=wordinfo[word].prev) {
3435 if (word > nextword && (!min || word < min))
3448 ST.lastparen = *PL_reglastparen;
3452 /* find start char of end of current word */
3454 U32 chars; /* how many chars to skip */
3455 U8 *uc = ST.firstpos;
3456 reg_trie_data * const trie
3457 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3459 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3461 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3465 /* the hard option - fold each char in turn and find
3466 * its folded length (which may be different */
3467 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3475 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3483 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3488 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3502 PL_reginput = (char *)uc;
3505 scan = (ST.jump && ST.jump[ST.nextword])
3506 ? ST.me + ST.jump[ST.nextword]
3510 PerlIO_printf( Perl_debug_log,
3511 "%*s %sTRIE matched word #%d, continuing%s\n",
3512 REPORT_CODE_OFF+depth*2, "",
3519 if (ST.accepted > 1 || has_cutgroup) {
3520 PUSH_STATE_GOTO(TRIE_next, scan);
3523 /* only one choice left - just continue */
3525 AV *const trie_words
3526 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3527 SV ** const tmp = av_fetch( trie_words,
3529 SV *sv= tmp ? sv_newmortal() : NULL;
3531 PerlIO_printf( Perl_debug_log,
3532 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3533 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3535 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3536 PL_colors[0], PL_colors[1],
3537 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3539 : "not compiled under -Dr",
3543 locinput = PL_reginput;
3544 nextchr = UCHARAT(locinput);
3545 continue; /* execute rest of RE */
3550 char *s = STRING(scan);
3552 if (utf8_target != UTF_PATTERN) {
3553 /* The target and the pattern have differing utf8ness. */
3555 const char * const e = s + ln;
3558 /* The target is utf8, the pattern is not utf8. */
3563 if (NATIVE_TO_UNI(*(U8*)s) !=
3564 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3572 /* The target is not utf8, the pattern is utf8. */
3577 if (NATIVE_TO_UNI(*((U8*)l)) !=
3578 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3586 nextchr = UCHARAT(locinput);
3589 /* The target and the pattern have the same utf8ness. */
3590 /* Inline the first character, for speed. */
3591 if (UCHARAT(s) != nextchr)
3593 if (PL_regeol - locinput < ln)
3595 if (ln > 1 && memNE(s, locinput, ln))
3598 nextchr = UCHARAT(locinput);
3603 const U8 * fold_array;
3605 U32 fold_utf8_flags;
3607 PL_reg_flags |= RF_tainted;
3608 folder = foldEQ_locale;
3609 fold_array = PL_fold_locale;
3610 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3614 folder = foldEQ_latin1;
3615 fold_array = PL_fold_latin1;
3616 fold_utf8_flags = 0;
3620 folder = foldEQ_latin1;
3621 fold_array = PL_fold_latin1;
3622 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3627 fold_array = PL_fold;
3628 fold_utf8_flags = 0;
3634 if (utf8_target || UTF_PATTERN) {
3635 /* Either target or the pattern are utf8. */
3636 const char * const l = locinput;
3637 char *e = PL_regeol;
3639 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3640 l, &e, 0, utf8_target, fold_utf8_flags)) {
3641 /* One more case for the sharp s:
3642 * pack("U0U*", 0xDF) =~ /ss/i,
3643 * the 0xC3 0x9F are the UTF-8
3644 * byte sequence for the U+00DF. */
3646 if (!(utf8_target &&
3647 toLOWER(s[0]) == 's' &&
3649 toLOWER(s[1]) == 's' &&
3656 nextchr = UCHARAT(locinput);
3660 /* Neither the target nor the pattern are utf8 */
3661 if (UCHARAT(s) != nextchr &&
3662 UCHARAT(s) != fold_array[nextchr])
3666 if (PL_regeol - locinput < ln)
3668 if (ln > 1 && ! folder(s, locinput, ln))
3671 nextchr = UCHARAT(locinput);
3675 /* XXX Could improve efficiency by separating these all out using a
3676 * macro or in-line function. At that point regcomp.c would no longer
3677 * have to set the FLAGS fields of these */
3680 PL_reg_flags |= RF_tainted;
3688 /* was last char in word? */
3689 if (utf8_target && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET) {
3690 if (locinput == PL_bostr)
3693 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3695 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3697 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3698 ln = isALNUM_uni(ln);
3699 LOAD_UTF8_CHARCLASS_ALNUM();
3700 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3703 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3704 n = isALNUM_LC_utf8((U8*)locinput);
3709 /* Here the string isn't utf8, or is utf8 and only ascii
3710 * characters are to match \w. In the latter case looking at
3711 * the byte just prior to the current one may be just the final
3712 * byte of a multi-byte character. This is ok. There are two
3714 * 1) it is a single byte character, and then the test is doing
3715 * just what it's supposed to.
3716 * 2) it is a multi-byte character, in which case the final
3717 * byte is never mistakable for ASCII, and so the test
3718 * will say it is not a word character, which is the
3719 * correct answer. */
3720 ln = (locinput != PL_bostr) ?
3721 UCHARAT(locinput - 1) : '\n';
3722 switch (FLAGS(scan)) {
3723 case REGEX_UNICODE_CHARSET:
3724 ln = isWORDCHAR_L1(ln);
3725 n = isWORDCHAR_L1(nextchr);
3727 case REGEX_LOCALE_CHARSET:
3728 ln = isALNUM_LC(ln);
3729 n = isALNUM_LC(nextchr);
3731 case REGEX_DEPENDS_CHARSET:
3733 n = isALNUM(nextchr);
3735 case REGEX_ASCII_RESTRICTED_CHARSET:
3736 ln = isWORDCHAR_A(ln);
3737 n = isWORDCHAR_A(nextchr);
3740 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3744 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3746 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3751 if (utf8_target || state_num == ANYOFV) {
3752 STRLEN inclasslen = PL_regeol - locinput;
3753 if (locinput >= PL_regeol)
3756 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3758 locinput += inclasslen;
3759 nextchr = UCHARAT(locinput);
3764 nextchr = UCHARAT(locinput);
3765 if (!nextchr && locinput >= PL_regeol)
3767 if (!REGINCLASS(rex, scan, (U8*)locinput))
3769 nextchr = UCHARAT(++locinput);
3773 /* Special char classes - The defines start on line 129 or so */
3774 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3775 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3776 ALNUMU, NALNUMU, isWORDCHAR_L1,
3777 ALNUMA, NALNUMA, isWORDCHAR_A,
3780 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3781 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3782 SPACEU, NSPACEU, isSPACE_L1,
3783 SPACEA, NSPACEA, isSPACE_A,
3786 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3787 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3788 DIGITA, NDIGITA, isDIGIT_A,
3791 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3792 a Unicode extended Grapheme Cluster */
3793 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3794 extended Grapheme Cluster is:
3797 | Prepend* Begin Extend*
3800 Begin is (Hangul-syllable | ! Control)
3801 Extend is (Grapheme_Extend | Spacing_Mark)
3802 Control is [ GCB_Control CR LF ]
3804 The discussion below shows how the code for CLUMP is derived
3805 from this regex. Note that most of these concepts are from
3806 property values of the Grapheme Cluster Boundary (GCB) property.
3807 No code point can have multiple property values for a given
3808 property. Thus a code point in Prepend can't be in Control, but
3809 it must be in !Control. This is why Control above includes
3810 GCB_Control plus CR plus LF. The latter two are used in the GCB
3811 property separately, and so can't be in GCB_Control, even though
3812 they logically are controls. Control is not the same as gc=cc,
3813 but includes format and other characters as well.
3815 The Unicode definition of Hangul-syllable is:
3817 | (L* ( ( V | LV ) V* | LVT ) T*)
3820 Each of these is a value for the GCB property, and hence must be
3821 disjoint, so the order they are tested is immaterial, so the
3822 above can safely be changed to
3825 | (L* ( LVT | ( V | LV ) V*) T*)
3827 The last two terms can be combined like this:
3829 | (( LVT | ( V | LV ) V*) T*))
3831 And refactored into this:
3832 L* (L | LVT T* | V V* T* | LV V* T*)
3834 That means that if we have seen any L's at all we can quit
3835 there, but if the next character is a LVT, a V or and LV we
3838 There is a subtlety with Prepend* which showed up in testing.
3839 Note that the Begin, and only the Begin is required in:
3840 | Prepend* Begin Extend*
3841 Also, Begin contains '! Control'. A Prepend must be a '!
3842 Control', which means it must be a Begin. What it comes down to
3843 is that if we match Prepend* and then find no suitable Begin
3844 afterwards, that if we backtrack the last Prepend, that one will
3845 be a suitable Begin.
3848 if (locinput >= PL_regeol)
3850 if (! utf8_target) {
3852 /* Match either CR LF or '.', as all the other possibilities
3854 locinput++; /* Match the . or CR */
3856 && locinput < PL_regeol
3857 && UCHARAT(locinput) == '\n') locinput++;
3861 /* Utf8: See if is ( CR LF ); already know that locinput <
3862 * PL_regeol, so locinput+1 is in bounds */
3863 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3867 /* In case have to backtrack to beginning, then match '.' */
3868 char *starting = locinput;
3870 /* In case have to backtrack the last prepend */
3871 char *previous_prepend = 0;
3873 LOAD_UTF8_CHARCLASS_GCB();
3875 /* Match (prepend)* */
3876 while (locinput < PL_regeol
3877 && swash_fetch(PL_utf8_X_prepend,
3878 (U8*)locinput, utf8_target))
3880 previous_prepend = locinput;
3881 locinput += UTF8SKIP(locinput);
3884 /* As noted above, if we matched a prepend character, but
3885 * the next thing won't match, back off the last prepend we
3886 * matched, as it is guaranteed to match the begin */
3887 if (previous_prepend
3888 && (locinput >= PL_regeol
3889 || ! swash_fetch(PL_utf8_X_begin,
3890 (U8*)locinput, utf8_target)))
3892 locinput = previous_prepend;
3895 /* Note that here we know PL_regeol > locinput, as we
3896 * tested that upon input to this switch case, and if we
3897 * moved locinput forward, we tested the result just above
3898 * and it either passed, or we backed off so that it will
3900 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3902 /* Here did not match the required 'Begin' in the
3903 * second term. So just match the very first
3904 * character, the '.' of the final term of the regex */
3905 locinput = starting + UTF8SKIP(starting);
3908 /* Here is the beginning of a character that can have
3909 * an extender. It is either a hangul syllable, or a
3911 if (swash_fetch(PL_utf8_X_non_hangul,
3912 (U8*)locinput, utf8_target))
3915 /* Here not a Hangul syllable, must be a
3916 * ('! * Control') */
3917 locinput += UTF8SKIP(locinput);
3920 /* Here is a Hangul syllable. It can be composed
3921 * of several individual characters. One
3922 * possibility is T+ */
3923 if (swash_fetch(PL_utf8_X_T,
3924 (U8*)locinput, utf8_target))
3926 while (locinput < PL_regeol
3927 && swash_fetch(PL_utf8_X_T,
3928 (U8*)locinput, utf8_target))
3930 locinput += UTF8SKIP(locinput);
3934 /* Here, not T+, but is a Hangul. That means
3935 * it is one of the others: L, LV, LVT or V,
3937 * L* (L | LVT T* | V V* T* | LV V* T*) */
3940 while (locinput < PL_regeol
3941 && swash_fetch(PL_utf8_X_L,
3942 (U8*)locinput, utf8_target))
3944 locinput += UTF8SKIP(locinput);
3947 /* Here, have exhausted L*. If the next
3948 * character is not an LV, LVT nor V, it means
3949 * we had to have at least one L, so matches L+
3950 * in the original equation, we have a complete
3951 * hangul syllable. Are done. */
3953 if (locinput < PL_regeol
3954 && swash_fetch(PL_utf8_X_LV_LVT_V,
3955 (U8*)locinput, utf8_target))
3958 /* Otherwise keep going. Must be LV, LVT
3959 * or V. See if LVT */
3960 if (swash_fetch(PL_utf8_X_LVT,
3961 (U8*)locinput, utf8_target))
3963 locinput += UTF8SKIP(locinput);
3966 /* Must be V or LV. Take it, then
3968 locinput += UTF8SKIP(locinput);
3969 while (locinput < PL_regeol
3970 && swash_fetch(PL_utf8_X_V,
3971 (U8*)locinput, utf8_target))
3973 locinput += UTF8SKIP(locinput);
3977 /* And any of LV, LVT, or V can be followed
3979 while (locinput < PL_regeol
3980 && swash_fetch(PL_utf8_X_T,
3984 locinput += UTF8SKIP(locinput);
3990 /* Match any extender */
3991 while (locinput < PL_regeol
3992 && swash_fetch(PL_utf8_X_extend,
3993 (U8*)locinput, utf8_target))
3995 locinput += UTF8SKIP(locinput);
3999 if (locinput > PL_regeol) sayNO;
4001 nextchr = UCHARAT(locinput);
4005 { /* The capture buffer cases. The ones beginning with N for the
4006 named buffers just convert to the equivalent numbered and
4007 pretend they were called as the corresponding numbered buffer
4009 /* don't initialize these in the declaration, it makes C++
4014 const U8 *fold_array;
4017 PL_reg_flags |= RF_tainted;
4018 folder = foldEQ_locale;
4019 fold_array = PL_fold_locale;
4021 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4025 folder = foldEQ_latin1;
4026 fold_array = PL_fold_latin1;
4028 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4032 folder = foldEQ_latin1;
4033 fold_array = PL_fold_latin1;
4035 utf8_fold_flags = 0;
4040 fold_array = PL_fold;
4042 utf8_fold_flags = 0;
4049 utf8_fold_flags = 0;
4052 /* For the named back references, find the corresponding buffer
4054 n = reg_check_named_buff_matched(rex,scan);
4059 goto do_nref_ref_common;
4062 PL_reg_flags |= RF_tainted;
4063 folder = foldEQ_locale;
4064 fold_array = PL_fold_locale;
4065 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4069 folder = foldEQ_latin1;
4070 fold_array = PL_fold_latin1;
4071 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4075 folder = foldEQ_latin1;
4076 fold_array = PL_fold_latin1;
4077 utf8_fold_flags = 0;
4082 fold_array = PL_fold;
4083 utf8_fold_flags = 0;
4089 utf8_fold_flags = 0;
4093 n = ARG(scan); /* which paren pair */
4096 ln = PL_regoffs[n].start;
4097 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4098 if (*PL_reglastparen < n || ln == -1)
4099 sayNO; /* Do not match unless seen CLOSEn. */
4100 if (ln == PL_regoffs[n].end)
4104 if (type != REF /* REF can do byte comparison */
4105 && (utf8_target || type == REFFU))
4106 { /* XXX handle REFFL better */
4107 char * limit = PL_regeol;
4109 /* This call case insensitively compares the entire buffer
4110 * at s, with the current input starting at locinput, but
4111 * not going off the end given by PL_regeol, and returns in
4112 * limit upon success, how much of the current input was
4114 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4115 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4120 nextchr = UCHARAT(locinput);
4124 /* Not utf8: Inline the first character, for speed. */
4125 if (UCHARAT(s) != nextchr &&
4127 UCHARAT(s) != fold_array[nextchr]))
4129 ln = PL_regoffs[n].end - ln;
4130 if (locinput + ln > PL_regeol)
4132 if (ln > 1 && (type == REF
4133 ? memNE(s, locinput, ln)
4134 : ! folder(s, locinput, ln)))
4137 nextchr = UCHARAT(locinput);
4147 #define ST st->u.eval
4152 regexp_internal *rei;
4153 regnode *startpoint;
4156 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4157 if (cur_eval && cur_eval->locinput==locinput) {
4158 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4159 Perl_croak(aTHX_ "Infinite recursion in regex");
4160 if ( ++nochange_depth > max_nochange_depth )
4162 "Pattern subroutine nesting without pos change"
4163 " exceeded limit in regex");
4170 (void)ReREFCNT_inc(rex_sv);
4171 if (OP(scan)==GOSUB) {
4172 startpoint = scan + ARG2L(scan);
4173 ST.close_paren = ARG(scan);
4175 startpoint = rei->program+1;
4178 goto eval_recurse_doit;
4180 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4181 if (cur_eval && cur_eval->locinput==locinput) {
4182 if ( ++nochange_depth > max_nochange_depth )
4183 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4188 /* execute the code in the {...} */
4190 SV ** const before = SP;
4191 OP_4tree * const oop = PL_op;
4192 COP * const ocurcop = PL_curcop;
4194 char *saved_regeol = PL_regeol;
4195 struct re_save_state saved_state;
4197 /* To not corrupt the existing regex state while executing the
4198 * eval we would normally put it on the save stack, like with
4199 * save_re_context. However, re-evals have a weird scoping so we
4200 * can't just add ENTER/LEAVE here. With that, things like
4202 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4204 * would break, as they expect the localisation to be unwound
4205 * only when the re-engine backtracks through the bit that
4208 * What we do instead is just saving the state in a local c
4211 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4214 PL_op = (OP_4tree*)rexi->data->data[n];
4215 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4216 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4217 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4218 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4221 SV *sv_mrk = get_sv("REGMARK", 1);
4222 sv_setsv(sv_mrk, sv_yes_mark);
4225 CALLRUNOPS(aTHX); /* Scalar context. */
4228 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4234 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4237 PAD_RESTORE_LOCAL(old_comppad);
4238 PL_curcop = ocurcop;
4239 PL_regeol = saved_regeol;
4242 sv_setsv(save_scalar(PL_replgv), ret);
4246 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4249 /* extract RE object from returned value; compiling if
4255 SV *const sv = SvRV(ret);
4257 if (SvTYPE(sv) == SVt_REGEXP) {
4259 } else if (SvSMAGICAL(sv)) {
4260 mg = mg_find(sv, PERL_MAGIC_qr);
4263 } else if (SvTYPE(ret) == SVt_REGEXP) {
4265 } else if (SvSMAGICAL(ret)) {
4266 if (SvGMAGICAL(ret)) {
4267 /* I don't believe that there is ever qr magic
4269 assert(!mg_find(ret, PERL_MAGIC_qr));
4270 sv_unmagic(ret, PERL_MAGIC_qr);
4273 mg = mg_find(ret, PERL_MAGIC_qr);
4274 /* testing suggests mg only ends up non-NULL for
4275 scalars who were upgraded and compiled in the
4276 else block below. In turn, this is only
4277 triggered in the "postponed utf8 string" tests
4283 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4287 rx = reg_temp_copy(NULL, rx);
4291 const I32 osize = PL_regsize;
4294 assert (SvUTF8(ret));
4295 } else if (SvUTF8(ret)) {
4296 /* Not doing UTF-8, despite what the SV says. Is
4297 this only if we're trapped in use 'bytes'? */
4298 /* Make a copy of the octet sequence, but without
4299 the flag on, as the compiler now honours the
4300 SvUTF8 flag on ret. */
4302 const char *const p = SvPV(ret, len);
4303 ret = newSVpvn_flags(p, len, SVs_TEMP);
4305 rx = CALLREGCOMP(ret, pm_flags);
4307 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4309 /* This isn't a first class regexp. Instead, it's
4310 caching a regexp onto an existing, Perl visible
4312 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4317 re = (struct regexp *)SvANY(rx);
4319 RXp_MATCH_COPIED_off(re);
4320 re->subbeg = rex->subbeg;
4321 re->sublen = rex->sublen;
4324 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4325 "Matching embedded");
4327 startpoint = rei->program + 1;
4328 ST.close_paren = 0; /* only used for GOSUB */
4329 /* borrowed from regtry */
4330 if (PL_reg_start_tmpl <= re->nparens) {
4331 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4332 if(PL_reg_start_tmp)
4333 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4335 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4338 eval_recurse_doit: /* Share code with GOSUB below this line */
4339 /* run the pattern returned from (??{...}) */
4340 ST.cp = regcppush(0); /* Save *all* the positions. */
4341 REGCP_SET(ST.lastcp);
4343 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4345 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4346 PL_reglastparen = &re->lastparen;
4347 PL_reglastcloseparen = &re->lastcloseparen;
4349 re->lastcloseparen = 0;
4351 PL_reginput = locinput;
4354 /* XXXX This is too dramatic a measure... */
4357 ST.toggle_reg_flags = PL_reg_flags;
4359 PL_reg_flags |= RF_utf8;
4361 PL_reg_flags &= ~RF_utf8;
4362 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4364 ST.prev_rex = rex_sv;
4365 ST.prev_curlyx = cur_curlyx;
4366 SETREX(rex_sv,re_sv);
4371 ST.prev_eval = cur_eval;
4373 /* now continue from first node in postoned RE */
4374 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4377 /* logical is 1, /(?(?{...})X|Y)/ */
4378 sw = cBOOL(SvTRUE(ret));
4383 case EVAL_AB: /* cleanup after a successful (??{A})B */
4384 /* note: this is called twice; first after popping B, then A */
4385 PL_reg_flags ^= ST.toggle_reg_flags;
4386 ReREFCNT_dec(rex_sv);
4387 SETREX(rex_sv,ST.prev_rex);
4388 rex = (struct regexp *)SvANY(rex_sv);
4389 rexi = RXi_GET(rex);
4391 cur_eval = ST.prev_eval;
4392 cur_curlyx = ST.prev_curlyx;
4394 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4395 PL_reglastparen = &rex->lastparen;
4396 PL_reglastcloseparen = &rex->lastcloseparen;
4397 /* also update PL_regoffs */
4398 PL_regoffs = rex->offs;
4400 /* XXXX This is too dramatic a measure... */
4402 if ( nochange_depth )
4407 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4408 /* note: this is called twice; first after popping B, then A */
4409 PL_reg_flags ^= ST.toggle_reg_flags;
4410 ReREFCNT_dec(rex_sv);
4411 SETREX(rex_sv,ST.prev_rex);
4412 rex = (struct regexp *)SvANY(rex_sv);
4413 rexi = RXi_GET(rex);
4414 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4415 PL_reglastparen = &rex->lastparen;
4416 PL_reglastcloseparen = &rex->lastcloseparen;
4418 PL_reginput = locinput;
4419 REGCP_UNWIND(ST.lastcp);
4421 cur_eval = ST.prev_eval;
4422 cur_curlyx = ST.prev_curlyx;
4423 /* XXXX This is too dramatic a measure... */
4425 if ( nochange_depth )
4431 n = ARG(scan); /* which paren pair */
4432 PL_reg_start_tmp[n] = locinput;
4438 n = ARG(scan); /* which paren pair */
4439 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4440 PL_regoffs[n].end = locinput - PL_bostr;
4441 /*if (n > PL_regsize)
4443 if (n > *PL_reglastparen)
4444 *PL_reglastparen = n;
4445 *PL_reglastcloseparen = n;
4446 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4454 cursor && OP(cursor)!=END;
4455 cursor=regnext(cursor))
4457 if ( OP(cursor)==CLOSE ){
4459 if ( n <= lastopen ) {
4461 = PL_reg_start_tmp[n] - PL_bostr;
4462 PL_regoffs[n].end = locinput - PL_bostr;
4463 /*if (n > PL_regsize)
4465 if (n > *PL_reglastparen)
4466 *PL_reglastparen = n;
4467 *PL_reglastcloseparen = n;
4468 if ( n == ARG(scan) || (cur_eval &&
4469 cur_eval->u.eval.close_paren == n))
4478 n = ARG(scan); /* which paren pair */
4479 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4482 /* reg_check_named_buff_matched returns 0 for no match */
4483 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4487 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4493 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4495 next = NEXTOPER(NEXTOPER(scan));
4497 next = scan + ARG(scan);
4498 if (OP(next) == IFTHEN) /* Fake one. */
4499 next = NEXTOPER(NEXTOPER(next));
4503 logical = scan->flags;
4506 /*******************************************************************
4508 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4509 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4510 STAR/PLUS/CURLY/CURLYN are used instead.)
4512 A*B is compiled as <CURLYX><A><WHILEM><B>
4514 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4515 state, which contains the current count, initialised to -1. It also sets
4516 cur_curlyx to point to this state, with any previous value saved in the
4519 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4520 since the pattern may possibly match zero times (i.e. it's a while {} loop
4521 rather than a do {} while loop).
4523 Each entry to WHILEM represents a successful match of A. The count in the
4524 CURLYX block is incremented, another WHILEM state is pushed, and execution
4525 passes to A or B depending on greediness and the current count.
4527 For example, if matching against the string a1a2a3b (where the aN are
4528 substrings that match /A/), then the match progresses as follows: (the
4529 pushed states are interspersed with the bits of strings matched so far):
4532 <CURLYX cnt=0><WHILEM>
4533 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4534 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4535 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4536 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4538 (Contrast this with something like CURLYM, which maintains only a single
4542 a1 <CURLYM cnt=1> a2
4543 a1 a2 <CURLYM cnt=2> a3
4544 a1 a2 a3 <CURLYM cnt=3> b
4547 Each WHILEM state block marks a point to backtrack to upon partial failure
4548 of A or B, and also contains some minor state data related to that
4549 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4550 overall state, such as the count, and pointers to the A and B ops.
4552 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4553 must always point to the *current* CURLYX block, the rules are:
4555 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4556 and set cur_curlyx to point the new block.
4558 When popping the CURLYX block after a successful or unsuccessful match,
4559 restore the previous cur_curlyx.
4561 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4562 to the outer one saved in the CURLYX block.
4564 When popping the WHILEM block after a successful or unsuccessful B match,
4565 restore the previous cur_curlyx.
4567 Here's an example for the pattern (AI* BI)*BO
4568 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4571 curlyx backtrack stack
4572 ------ ---------------
4574 CO <CO prev=NULL> <WO>
4575 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4576 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4577 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4579 At this point the pattern succeeds, and we work back down the stack to
4580 clean up, restoring as we go:
4582 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4583 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4584 CO <CO prev=NULL> <WO>
4587 *******************************************************************/
4589 #define ST st->u.curlyx
4591 case CURLYX: /* start of /A*B/ (for complex A) */
4593 /* No need to save/restore up to this paren */
4594 I32 parenfloor = scan->flags;
4596 assert(next); /* keep Coverity happy */
4597 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4600 /* XXXX Probably it is better to teach regpush to support
4601 parenfloor > PL_regsize... */
4602 if (parenfloor > (I32)*PL_reglastparen)
4603 parenfloor = *PL_reglastparen; /* Pessimization... */
4605 ST.prev_curlyx= cur_curlyx;
4607 ST.cp = PL_savestack_ix;
4609 /* these fields contain the state of the current curly.
4610 * they are accessed by subsequent WHILEMs */
4611 ST.parenfloor = parenfloor;
4616 ST.count = -1; /* this will be updated by WHILEM */
4617 ST.lastloc = NULL; /* this will be updated by WHILEM */
4619 PL_reginput = locinput;
4620 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4624 case CURLYX_end: /* just finished matching all of A*B */
4625 cur_curlyx = ST.prev_curlyx;
4629 case CURLYX_end_fail: /* just failed to match all of A*B */
4631 cur_curlyx = ST.prev_curlyx;
4637 #define ST st->u.whilem
4639 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4641 /* see the discussion above about CURLYX/WHILEM */
4643 int min = ARG1(cur_curlyx->u.curlyx.me);
4644 int max = ARG2(cur_curlyx->u.curlyx.me);
4645 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4647 assert(cur_curlyx); /* keep Coverity happy */
4648 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4649 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4650 ST.cache_offset = 0;
4653 PL_reginput = locinput;
4655 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4656 "%*s whilem: matched %ld out of %d..%d\n",
4657 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4660 /* First just match a string of min A's. */
4663 cur_curlyx->u.curlyx.lastloc = locinput;
4664 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4668 /* If degenerate A matches "", assume A done. */
4670 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4671 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4672 "%*s whilem: empty match detected, trying continuation...\n",
4673 REPORT_CODE_OFF+depth*2, "")
4675 goto do_whilem_B_max;
4678 /* super-linear cache processing */
4682 if (!PL_reg_maxiter) {
4683 /* start the countdown: Postpone detection until we
4684 * know the match is not *that* much linear. */
4685 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4686 /* possible overflow for long strings and many CURLYX's */
4687 if (PL_reg_maxiter < 0)
4688 PL_reg_maxiter = I32_MAX;
4689 PL_reg_leftiter = PL_reg_maxiter;
4692 if (PL_reg_leftiter-- == 0) {
4693 /* initialise cache */
4694 const I32 size = (PL_reg_maxiter + 7)/8;
4695 if (PL_reg_poscache) {
4696 if ((I32)PL_reg_poscache_size < size) {
4697 Renew(PL_reg_poscache, size, char);
4698 PL_reg_poscache_size = size;
4700 Zero(PL_reg_poscache, size, char);
4703 PL_reg_poscache_size = size;
4704 Newxz(PL_reg_poscache, size, char);
4706 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4707 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4708 PL_colors[4], PL_colors[5])
4712 if (PL_reg_leftiter < 0) {
4713 /* have we already failed at this position? */
4715 offset = (scan->flags & 0xf) - 1
4716 + (locinput - PL_bostr) * (scan->flags>>4);
4717 mask = 1 << (offset % 8);
4719 if (PL_reg_poscache[offset] & mask) {
4720 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4721 "%*s whilem: (cache) already tried at this position...\n",
4722 REPORT_CODE_OFF+depth*2, "")
4724 sayNO; /* cache records failure */
4726 ST.cache_offset = offset;
4727 ST.cache_mask = mask;
4731 /* Prefer B over A for minimal matching. */
4733 if (cur_curlyx->u.curlyx.minmod) {
4734 ST.save_curlyx = cur_curlyx;
4735 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4736 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4737 REGCP_SET(ST.lastcp);
4738 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4742 /* Prefer A over B for maximal matching. */
4744 if (n < max) { /* More greed allowed? */
4745 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4746 cur_curlyx->u.curlyx.lastloc = locinput;
4747 REGCP_SET(ST.lastcp);
4748 PUSH_STATE_GOTO(WHILEM_A_max, A);
4751 goto do_whilem_B_max;
4755 case WHILEM_B_min: /* just matched B in a minimal match */
4756 case WHILEM_B_max: /* just matched B in a maximal match */
4757 cur_curlyx = ST.save_curlyx;
4761 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4762 cur_curlyx = ST.save_curlyx;
4763 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4764 cur_curlyx->u.curlyx.count--;
4768 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4769 REGCP_UNWIND(ST.lastcp);
4772 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4773 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4774 cur_curlyx->u.curlyx.count--;
4778 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4779 REGCP_UNWIND(ST.lastcp);
4780 regcppop(rex); /* Restore some previous $<digit>s? */
4781 PL_reginput = locinput;
4782 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4783 "%*s whilem: failed, trying continuation...\n",
4784 REPORT_CODE_OFF+depth*2, "")
4787 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4788 && ckWARN(WARN_REGEXP)
4789 && !(PL_reg_flags & RF_warned))
4791 PL_reg_flags |= RF_warned;
4792 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4793 "Complex regular subexpression recursion",
4798 ST.save_curlyx = cur_curlyx;
4799 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4800 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4803 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4804 cur_curlyx = ST.save_curlyx;
4805 REGCP_UNWIND(ST.lastcp);
4808 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4809 /* Maximum greed exceeded */
4810 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4811 && ckWARN(WARN_REGEXP)
4812 && !(PL_reg_flags & RF_warned))
4814 PL_reg_flags |= RF_warned;
4815 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4816 "%s limit (%d) exceeded",
4817 "Complex regular subexpression recursion",
4820 cur_curlyx->u.curlyx.count--;
4824 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4825 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4827 /* Try grabbing another A and see if it helps. */
4828 PL_reginput = locinput;
4829 cur_curlyx->u.curlyx.lastloc = locinput;
4830 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4831 REGCP_SET(ST.lastcp);
4832 PUSH_STATE_GOTO(WHILEM_A_min,
4833 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4837 #define ST st->u.branch
4839 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4840 next = scan + ARG(scan);
4843 scan = NEXTOPER(scan);
4846 case BRANCH: /* /(...|A|...)/ */
4847 scan = NEXTOPER(scan); /* scan now points to inner node */
4848 ST.lastparen = *PL_reglastparen;
4849 ST.next_branch = next;
4851 PL_reginput = locinput;
4853 /* Now go into the branch */
4855 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4857 PUSH_STATE_GOTO(BRANCH_next, scan);
4861 PL_reginput = locinput;
4862 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4863 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4864 PUSH_STATE_GOTO(CUTGROUP_next,next);
4866 case CUTGROUP_next_fail:
4869 if (st->u.mark.mark_name)
4870 sv_commit = st->u.mark.mark_name;
4876 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4881 REGCP_UNWIND(ST.cp);
4882 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4883 PL_regoffs[n].end = -1;
4884 *PL_reglastparen = n;
4885 /*dmq: *PL_reglastcloseparen = n; */
4886 scan = ST.next_branch;
4887 /* no more branches? */
4888 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4890 PerlIO_printf( Perl_debug_log,
4891 "%*s %sBRANCH failed...%s\n",
4892 REPORT_CODE_OFF+depth*2, "",
4898 continue; /* execute next BRANCH[J] op */
4906 #define ST st->u.curlym
4908 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4910 /* This is an optimisation of CURLYX that enables us to push
4911 * only a single backtracking state, no matter how many matches
4912 * there are in {m,n}. It relies on the pattern being constant
4913 * length, with no parens to influence future backrefs
4917 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4919 /* if paren positive, emulate an OPEN/CLOSE around A */
4921 U32 paren = ST.me->flags;
4922 if (paren > PL_regsize)
4924 if (paren > *PL_reglastparen)
4925 *PL_reglastparen = paren;
4926 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4934 ST.c1 = CHRTEST_UNINIT;
4937 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4940 curlym_do_A: /* execute the A in /A{m,n}B/ */
4941 PL_reginput = locinput;
4942 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4945 case CURLYM_A: /* we've just matched an A */
4946 locinput = st->locinput;
4947 nextchr = UCHARAT(locinput);
4950 /* after first match, determine A's length: u.curlym.alen */
4951 if (ST.count == 1) {
4952 if (PL_reg_match_utf8) {
4954 while (s < PL_reginput) {
4960 ST.alen = PL_reginput - locinput;
4963 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4966 PerlIO_printf(Perl_debug_log,
4967 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4968 (int)(REPORT_CODE_OFF+(depth*2)), "",
4969 (IV) ST.count, (IV)ST.alen)
4972 locinput = PL_reginput;
4974 if (cur_eval && cur_eval->u.eval.close_paren &&
4975 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4979 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4980 if ( max == REG_INFTY || ST.count < max )
4981 goto curlym_do_A; /* try to match another A */
4983 goto curlym_do_B; /* try to match B */
4985 case CURLYM_A_fail: /* just failed to match an A */
4986 REGCP_UNWIND(ST.cp);
4988 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4989 || (cur_eval && cur_eval->u.eval.close_paren &&
4990 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4993 curlym_do_B: /* execute the B in /A{m,n}B/ */
4994 PL_reginput = locinput;
4995 if (ST.c1 == CHRTEST_UNINIT) {
4996 /* calculate c1 and c2 for possible match of 1st char
4997 * following curly */
4998 ST.c1 = ST.c2 = CHRTEST_VOID;
4999 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5000 regnode *text_node = ST.B;
5001 if (! HAS_TEXT(text_node))
5002 FIND_NEXT_IMPT(text_node);
5005 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5007 But the former is redundant in light of the latter.
5009 if this changes back then the macro for
5010 IS_TEXT and friends need to change.
5012 if (PL_regkind[OP(text_node)] == EXACT)
5015 ST.c1 = (U8)*STRING(text_node);
5016 switch (OP(text_node)) {
5017 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5019 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5020 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5021 default: ST.c2 = ST.c1;
5028 PerlIO_printf(Perl_debug_log,
5029 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5030 (int)(REPORT_CODE_OFF+(depth*2)),
5033 if (ST.c1 != CHRTEST_VOID
5034 && UCHARAT(PL_reginput) != ST.c1
5035 && UCHARAT(PL_reginput) != ST.c2)
5037 /* simulate B failing */
5039 PerlIO_printf(Perl_debug_log,
5040 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5041 (int)(REPORT_CODE_OFF+(depth*2)),"",
5044 state_num = CURLYM_B_fail;
5045 goto reenter_switch;
5049 /* mark current A as captured */
5050 I32 paren = ST.me->flags;
5052 PL_regoffs[paren].start
5053 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5054 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5055 /*dmq: *PL_reglastcloseparen = paren; */
5058 PL_regoffs[paren].end = -1;
5059 if (cur_eval && cur_eval->u.eval.close_paren &&
5060 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5069 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5072 case CURLYM_B_fail: /* just failed to match a B */
5073 REGCP_UNWIND(ST.cp);
5075 I32 max = ARG2(ST.me);
5076 if (max != REG_INFTY && ST.count == max)
5078 goto curlym_do_A; /* try to match a further A */
5080 /* backtrack one A */
5081 if (ST.count == ARG1(ST.me) /* min */)
5084 locinput = HOPc(locinput, -ST.alen);
5085 goto curlym_do_B; /* try to match B */
5088 #define ST st->u.curly
5090 #define CURLY_SETPAREN(paren, success) \
5093 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5094 PL_regoffs[paren].end = locinput - PL_bostr; \
5095 *PL_reglastcloseparen = paren; \
5098 PL_regoffs[paren].end = -1; \
5101 case STAR: /* /A*B/ where A is width 1 */
5105 scan = NEXTOPER(scan);
5107 case PLUS: /* /A+B/ where A is width 1 */
5111 scan = NEXTOPER(scan);
5113 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5114 ST.paren = scan->flags; /* Which paren to set */
5115 if (ST.paren > PL_regsize)
5116 PL_regsize = ST.paren;
5117 if (ST.paren > *PL_reglastparen)
5118 *PL_reglastparen = ST.paren;
5119 ST.min = ARG1(scan); /* min to match */
5120 ST.max = ARG2(scan); /* max to match */
5121 if (cur_eval && cur_eval->u.eval.close_paren &&
5122 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5126 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5128 case CURLY: /* /A{m,n}B/ where A is width 1 */
5130 ST.min = ARG1(scan); /* min to match */
5131 ST.max = ARG2(scan); /* max to match */
5132 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5135 * Lookahead to avoid useless match attempts
5136 * when we know what character comes next.
5138 * Used to only do .*x and .*?x, but now it allows
5139 * for )'s, ('s and (?{ ... })'s to be in the way
5140 * of the quantifier and the EXACT-like node. -- japhy
5143 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5145 if (HAS_TEXT(next) || JUMPABLE(next)) {
5147 regnode *text_node = next;
5149 if (! HAS_TEXT(text_node))
5150 FIND_NEXT_IMPT(text_node);
5152 if (! HAS_TEXT(text_node))
5153 ST.c1 = ST.c2 = CHRTEST_VOID;
5155 if ( PL_regkind[OP(text_node)] != EXACT ) {
5156 ST.c1 = ST.c2 = CHRTEST_VOID;
5157 goto assume_ok_easy;
5160 s = (U8*)STRING(text_node);
5162 /* Currently we only get here when
5164 PL_rekind[OP(text_node)] == EXACT
5166 if this changes back then the macro for IS_TEXT and
5167 friends need to change. */
5170 switch (OP(text_node)) {
5171 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5173 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5174 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5175 default: ST.c2 = ST.c1; break;
5178 else { /* UTF_PATTERN */
5179 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5180 STRLEN ulen1, ulen2;
5181 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5182 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5184 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5185 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5187 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5189 0 : UTF8_ALLOW_ANY);
5190 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5192 0 : UTF8_ALLOW_ANY);
5194 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5196 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5201 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5208 ST.c1 = ST.c2 = CHRTEST_VOID;
5213 PL_reginput = locinput;
5216 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5219 locinput = PL_reginput;
5221 if (ST.c1 == CHRTEST_VOID)
5222 goto curly_try_B_min;
5224 ST.oldloc = locinput;
5226 /* set ST.maxpos to the furthest point along the
5227 * string that could possibly match */
5228 if (ST.max == REG_INFTY) {
5229 ST.maxpos = PL_regeol - 1;
5231 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5234 else if (utf8_target) {
5235 int m = ST.max - ST.min;
5236 for (ST.maxpos = locinput;
5237 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5238 ST.maxpos += UTF8SKIP(ST.maxpos);
5241 ST.maxpos = locinput + ST.max - ST.min;
5242 if (ST.maxpos >= PL_regeol)
5243 ST.maxpos = PL_regeol - 1;
5245 goto curly_try_B_min_known;
5249 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5250 locinput = PL_reginput;
5251 if (ST.count < ST.min)
5253 if ((ST.count > ST.min)
5254 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5256 /* A{m,n} must come at the end of the string, there's
5257 * no point in backing off ... */
5259 /* ...except that $ and \Z can match before *and* after
5260 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5261 We may back off by one in this case. */
5262 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5266 goto curly_try_B_max;
5271 case CURLY_B_min_known_fail:
5272 /* failed to find B in a non-greedy match where c1,c2 valid */
5273 if (ST.paren && ST.count)
5274 PL_regoffs[ST.paren].end = -1;
5276 PL_reginput = locinput; /* Could be reset... */
5277 REGCP_UNWIND(ST.cp);
5278 /* Couldn't or didn't -- move forward. */
5279 ST.oldloc = locinput;
5281 locinput += UTF8SKIP(locinput);
5285 curly_try_B_min_known:
5286 /* find the next place where 'B' could work, then call B */
5290 n = (ST.oldloc == locinput) ? 0 : 1;
5291 if (ST.c1 == ST.c2) {
5293 /* set n to utf8_distance(oldloc, locinput) */
5294 while (locinput <= ST.maxpos &&
5295 utf8n_to_uvchr((U8*)locinput,
5296 UTF8_MAXBYTES, &len,
5297 uniflags) != (UV)ST.c1) {
5303 /* set n to utf8_distance(oldloc, locinput) */
5304 while (locinput <= ST.maxpos) {
5306 const UV c = utf8n_to_uvchr((U8*)locinput,
5307 UTF8_MAXBYTES, &len,
5309 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5317 if (ST.c1 == ST.c2) {
5318 while (locinput <= ST.maxpos &&
5319 UCHARAT(locinput) != ST.c1)
5323 while (locinput <= ST.maxpos
5324 && UCHARAT(locinput) != ST.c1
5325 && UCHARAT(locinput) != ST.c2)
5328 n = locinput - ST.oldloc;
5330 if (locinput > ST.maxpos)
5332 /* PL_reginput == oldloc now */
5335 if (regrepeat(rex, ST.A, n, depth) < n)
5338 PL_reginput = locinput;
5339 CURLY_SETPAREN(ST.paren, ST.count);
5340 if (cur_eval && cur_eval->u.eval.close_paren &&
5341 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5344 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5349 case CURLY_B_min_fail:
5350 /* failed to find B in a non-greedy match where c1,c2 invalid */
5351 if (ST.paren && ST.count)
5352 PL_regoffs[ST.paren].end = -1;
5354 REGCP_UNWIND(ST.cp);
5355 /* failed -- move forward one */
5356 PL_reginput = locinput;
5357 if (regrepeat(rex, ST.A, 1, depth)) {
5359 locinput = PL_reginput;
5360 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5361 ST.count > 0)) /* count overflow ? */
5364 CURLY_SETPAREN(ST.paren, ST.count);
5365 if (cur_eval && cur_eval->u.eval.close_paren &&
5366 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5369 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5377 /* a successful greedy match: now try to match B */
5378 if (cur_eval && cur_eval->u.eval.close_paren &&
5379 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5384 if (ST.c1 != CHRTEST_VOID)
5385 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5386 UTF8_MAXBYTES, 0, uniflags)
5387 : (UV) UCHARAT(PL_reginput);
5388 /* If it could work, try it. */
5389 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5390 CURLY_SETPAREN(ST.paren, ST.count);
5391 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5396 case CURLY_B_max_fail:
5397 /* failed to find B in a greedy match */
5398 if (ST.paren && ST.count)
5399 PL_regoffs[ST.paren].end = -1;
5401 REGCP_UNWIND(ST.cp);
5403 if (--ST.count < ST.min)
5405 PL_reginput = locinput = HOPc(locinput, -1);
5406 goto curly_try_B_max;
5413 /* we've just finished A in /(??{A})B/; now continue with B */
5415 st->u.eval.toggle_reg_flags
5416 = cur_eval->u.eval.toggle_reg_flags;
5417 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5419 st->u.eval.prev_rex = rex_sv; /* inner */
5420 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5421 rex = (struct regexp *)SvANY(rex_sv);
5422 rexi = RXi_GET(rex);
5423 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5424 ReREFCNT_inc(rex_sv);
5425 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5427 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5428 PL_reglastparen = &rex->lastparen;
5429 PL_reglastcloseparen = &rex->lastcloseparen;
5431 REGCP_SET(st->u.eval.lastcp);
5432 PL_reginput = locinput;
5434 /* Restore parens of the outer rex without popping the
5436 tmpix = PL_savestack_ix;
5437 PL_savestack_ix = cur_eval->u.eval.lastcp;
5439 PL_savestack_ix = tmpix;
5441 st->u.eval.prev_eval = cur_eval;
5442 cur_eval = cur_eval->u.eval.prev_eval;
5444 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5445 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5446 if ( nochange_depth )
5449 PUSH_YES_STATE_GOTO(EVAL_AB,
5450 st->u.eval.prev_eval->u.eval.B); /* match B */
5453 if (locinput < reginfo->till) {
5454 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5455 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5457 (long)(locinput - PL_reg_starttry),
5458 (long)(reginfo->till - PL_reg_starttry),
5461 sayNO_SILENT; /* Cannot match: too short. */
5463 PL_reginput = locinput; /* put where regtry can find it */
5464 sayYES; /* Success! */
5466 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5468 PerlIO_printf(Perl_debug_log,
5469 "%*s %ssubpattern success...%s\n",
5470 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5471 PL_reginput = locinput; /* put where regtry can find it */
5472 sayYES; /* Success! */
5475 #define ST st->u.ifmatch
5477 case SUSPEND: /* (?>A) */
5479 PL_reginput = locinput;
5482 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5484 goto ifmatch_trivial_fail_test;
5486 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5488 ifmatch_trivial_fail_test:
5490 char * const s = HOPBACKc(locinput, scan->flags);
5495 sw = 1 - cBOOL(ST.wanted);
5499 next = scan + ARG(scan);
5507 PL_reginput = locinput;
5511 ST.logical = logical;
5512 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5514 /* execute body of (?...A) */
5515 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5518 case IFMATCH_A_fail: /* body of (?...A) failed */
5519 ST.wanted = !ST.wanted;
5522 case IFMATCH_A: /* body of (?...A) succeeded */
5524 sw = cBOOL(ST.wanted);
5526 else if (!ST.wanted)
5529 if (OP(ST.me) == SUSPEND)
5530 locinput = PL_reginput;
5532 locinput = PL_reginput = st->locinput;
5533 nextchr = UCHARAT(locinput);
5535 scan = ST.me + ARG(ST.me);
5538 continue; /* execute B */
5543 next = scan + ARG(scan);
5548 reginfo->cutpoint = PL_regeol;
5551 PL_reginput = locinput;
5553 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5554 PUSH_STATE_GOTO(COMMIT_next,next);
5556 case COMMIT_next_fail:
5563 #define ST st->u.mark
5565 ST.prev_mark = mark_state;
5566 ST.mark_name = sv_commit = sv_yes_mark
5567 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5569 ST.mark_loc = PL_reginput = locinput;
5570 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5572 case MARKPOINT_next:
5573 mark_state = ST.prev_mark;
5576 case MARKPOINT_next_fail:
5577 if (popmark && sv_eq(ST.mark_name,popmark))
5579 if (ST.mark_loc > startpoint)
5580 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5581 popmark = NULL; /* we found our mark */
5582 sv_commit = ST.mark_name;
5585 PerlIO_printf(Perl_debug_log,
5586 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5587 REPORT_CODE_OFF+depth*2, "",
5588 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5591 mark_state = ST.prev_mark;
5592 sv_yes_mark = mark_state ?
5593 mark_state->u.mark.mark_name : NULL;
5597 PL_reginput = locinput;
5599 /* (*SKIP) : if we fail we cut here*/
5600 ST.mark_name = NULL;
5601 ST.mark_loc = locinput;
5602 PUSH_STATE_GOTO(SKIP_next,next);
5604 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5605 otherwise do nothing. Meaning we need to scan
5607 regmatch_state *cur = mark_state;
5608 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5611 if ( sv_eq( cur->u.mark.mark_name,
5614 ST.mark_name = find;
5615 PUSH_STATE_GOTO( SKIP_next, next );
5617 cur = cur->u.mark.prev_mark;
5620 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5622 case SKIP_next_fail:
5624 /* (*CUT:NAME) - Set up to search for the name as we
5625 collapse the stack*/
5626 popmark = ST.mark_name;
5628 /* (*CUT) - No name, we cut here.*/
5629 if (ST.mark_loc > startpoint)
5630 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5631 /* but we set sv_commit to latest mark_name if there
5632 is one so they can test to see how things lead to this
5635 sv_commit=mark_state->u.mark.mark_name;
5643 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5645 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5648 U8 folded[UTF8_MAXBYTES_CASE+1];
5650 const char * const l = locinput;
5651 char *e = PL_regeol;
5652 to_uni_fold(n, folded, &foldlen);
5654 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5655 l, &e, 0, utf8_target)) {
5660 nextchr = UCHARAT(locinput);
5663 if ((n=is_LNBREAK(locinput,utf8_target))) {
5665 nextchr = UCHARAT(locinput);
5670 #define CASE_CLASS(nAmE) \
5672 if ((n=is_##nAmE(locinput,utf8_target))) { \
5674 nextchr = UCHARAT(locinput); \
5679 if ((n=is_##nAmE(locinput,utf8_target))) { \
5682 locinput += UTF8SKIP(locinput); \
5683 nextchr = UCHARAT(locinput); \
5688 CASE_CLASS(HORIZWS);
5692 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5693 PTR2UV(scan), OP(scan));
5694 Perl_croak(aTHX_ "regexp memory corruption");
5698 /* switch break jumps here */
5699 scan = next; /* prepare to execute the next op and ... */
5700 continue; /* ... jump back to the top, reusing st */
5704 /* push a state that backtracks on success */
5705 st->u.yes.prev_yes_state = yes_state;
5709 /* push a new regex state, then continue at scan */
5711 regmatch_state *newst;
5714 regmatch_state *cur = st;
5715 regmatch_state *curyes = yes_state;
5717 regmatch_slab *slab = PL_regmatch_slab;
5718 for (;curd > -1;cur--,curd--) {
5719 if (cur < SLAB_FIRST(slab)) {
5721 cur = SLAB_LAST(slab);
5723 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5724 REPORT_CODE_OFF + 2 + depth * 2,"",
5725 curd, PL_reg_name[cur->resume_state],
5726 (curyes == cur) ? "yes" : ""
5729 curyes = cur->u.yes.prev_yes_state;
5732 DEBUG_STATE_pp("push")
5735 st->locinput = locinput;
5737 if (newst > SLAB_LAST(PL_regmatch_slab))
5738 newst = S_push_slab(aTHX);
5739 PL_regmatch_state = newst;
5741 locinput = PL_reginput;
5742 nextchr = UCHARAT(locinput);
5750 * We get here only if there's trouble -- normally "case END" is
5751 * the terminating point.
5753 Perl_croak(aTHX_ "corrupted regexp pointers");
5759 /* we have successfully completed a subexpression, but we must now
5760 * pop to the state marked by yes_state and continue from there */
5761 assert(st != yes_state);
5763 while (st != yes_state) {
5765 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5766 PL_regmatch_slab = PL_regmatch_slab->prev;
5767 st = SLAB_LAST(PL_regmatch_slab);
5771 DEBUG_STATE_pp("pop (no final)");
5773 DEBUG_STATE_pp("pop (yes)");
5779 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5780 || yes_state > SLAB_LAST(PL_regmatch_slab))
5782 /* not in this slab, pop slab */
5783 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5784 PL_regmatch_slab = PL_regmatch_slab->prev;
5785 st = SLAB_LAST(PL_regmatch_slab);
5787 depth -= (st - yes_state);
5790 yes_state = st->u.yes.prev_yes_state;
5791 PL_regmatch_state = st;
5794 locinput= st->locinput;
5795 nextchr = UCHARAT(locinput);
5797 state_num = st->resume_state + no_final;
5798 goto reenter_switch;
5801 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5802 PL_colors[4], PL_colors[5]));
5804 if (PL_reg_eval_set) {
5805 /* each successfully executed (?{...}) block does the equivalent of
5806 * local $^R = do {...}
5807 * When popping the save stack, all these locals would be undone;
5808 * bypass this by setting the outermost saved $^R to the latest
5810 if (oreplsv != GvSV(PL_replgv))
5811 sv_setsv(oreplsv, GvSV(PL_replgv));
5818 PerlIO_printf(Perl_debug_log,
5819 "%*s %sfailed...%s\n",
5820 REPORT_CODE_OFF+depth*2, "",
5821 PL_colors[4], PL_colors[5])
5833 /* there's a previous state to backtrack to */
5835 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5836 PL_regmatch_slab = PL_regmatch_slab->prev;
5837 st = SLAB_LAST(PL_regmatch_slab);
5839 PL_regmatch_state = st;
5840 locinput= st->locinput;
5841 nextchr = UCHARAT(locinput);
5843 DEBUG_STATE_pp("pop");
5845 if (yes_state == st)
5846 yes_state = st->u.yes.prev_yes_state;
5848 state_num = st->resume_state + 1; /* failure = success + 1 */
5849 goto reenter_switch;
5854 if (rex->intflags & PREGf_VERBARG_SEEN) {
5855 SV *sv_err = get_sv("REGERROR", 1);
5856 SV *sv_mrk = get_sv("REGMARK", 1);
5858 sv_commit = &PL_sv_no;
5860 sv_yes_mark = &PL_sv_yes;
5863 sv_commit = &PL_sv_yes;
5864 sv_yes_mark = &PL_sv_no;
5866 sv_setsv(sv_err, sv_commit);
5867 sv_setsv(sv_mrk, sv_yes_mark);
5870 /* clean up; in particular, free all slabs above current one */
5871 LEAVE_SCOPE(oldsave);
5877 - regrepeat - repeatedly match something simple, report how many
5880 * [This routine now assumes that it will only match on things of length 1.
5881 * That was true before, but now we assume scan - reginput is the count,
5882 * rather than incrementing count on every character. [Er, except utf8.]]
5885 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5888 register char *scan;
5890 register char *loceol = PL_regeol;
5891 register I32 hardcount = 0;
5892 register bool utf8_target = PL_reg_match_utf8;
5895 PERL_UNUSED_ARG(depth);
5898 PERL_ARGS_ASSERT_REGREPEAT;
5901 if (max == REG_INFTY)
5903 else if (max < loceol - scan)
5904 loceol = scan + max;
5909 while (scan < loceol && hardcount < max && *scan != '\n') {
5910 scan += UTF8SKIP(scan);
5914 while (scan < loceol && *scan != '\n')
5921 while (scan < loceol && hardcount < max) {
5922 scan += UTF8SKIP(scan);
5933 /* To get here, EXACTish nodes must have *byte* length == 1. That
5934 * means they match only characters in the string that can be expressed
5935 * as a single byte. For non-utf8 strings, that means a simple match.
5936 * For utf8 strings, the character matched must be an invariant, or
5937 * downgradable to a single byte. The pattern's utf8ness is
5938 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5939 * it is, it's an invariant */
5942 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5944 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5945 while (scan < loceol && UCHARAT(scan) == c) {
5951 /* Here, the string is utf8, and the pattern char is different
5952 * in utf8 than not, so can't compare them directly. Outside the
5953 * loop, find find the two utf8 bytes that represent c, and then
5954 * look for those in sequence in the utf8 string */
5955 U8 high = UTF8_TWO_BYTE_HI(c);
5956 U8 low = UTF8_TWO_BYTE_LO(c);
5959 while (hardcount < max
5960 && scan + 1 < loceol
5961 && UCHARAT(scan) == high
5962 && UCHARAT(scan + 1) == low)
5970 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5974 PL_reg_flags |= RF_tainted;
5975 utf8_flags = FOLDEQ_UTF8_LOCALE;
5982 /* The comments for the EXACT case above apply as well to these fold
5987 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5989 if (utf8_target) { /* Use full Unicode fold matching */
5990 char *tmpeol = loceol;
5991 while (hardcount < max
5992 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
5993 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6000 /* XXX Note that the above handles properly the German sharp s in
6001 * the pattern matching ss in the string. But it doesn't handle
6002 * properly cases where the string contains say 'LIGATURE ff' and
6003 * the pattern is 'f+'. This would require, say, a new function or
6004 * revised interface to foldEQ_utf8(), in which the maximum number
6005 * of characters to match could be passed and it would return how
6006 * many actually did. This is just one of many cases where
6007 * multi-char folds don't work properly, and so the fix is being
6013 /* Here, the string isn't utf8 and c is a single byte; and either
6014 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6015 * doesn't affect c. Can just do simple comparisons for exact or
6018 case EXACTF: folded = PL_fold[c]; break;
6020 case EXACTFU: folded = PL_fold_latin1[c]; break;
6021 case EXACTFL: folded = PL_fold_locale[c]; break;
6022 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6024 while (scan < loceol &&
6025 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6033 if (utf8_target || OP(p) == ANYOFV) {
6036 inclasslen = loceol - scan;
6037 while (hardcount < max
6038 && ((inclasslen = loceol - scan) > 0)
6039 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6045 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6053 LOAD_UTF8_CHARCLASS_ALNUM();
6054 while (hardcount < max && scan < loceol &&
6055 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6057 scan += UTF8SKIP(scan);
6061 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6069 while (scan < loceol && isALNUM((U8) *scan)) {
6074 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6079 PL_reg_flags |= RF_tainted;
6082 while (hardcount < max && scan < loceol &&
6083 isALNUM_LC_utf8((U8*)scan)) {
6084 scan += UTF8SKIP(scan);
6088 while (scan < loceol && isALNUM_LC(*scan))
6098 LOAD_UTF8_CHARCLASS_ALNUM();
6099 while (hardcount < max && scan < loceol &&
6100 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6102 scan += UTF8SKIP(scan);
6106 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6113 goto utf8_Nwordchar;
6114 while (scan < loceol && ! isALNUM((U8) *scan)) {
6120 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6121 scan += UTF8SKIP(scan);
6125 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6131 PL_reg_flags |= RF_tainted;
6134 while (hardcount < max && scan < loceol &&
6135 !isALNUM_LC_utf8((U8*)scan)) {
6136 scan += UTF8SKIP(scan);
6140 while (scan < loceol && !isALNUM_LC(*scan))
6150 LOAD_UTF8_CHARCLASS_SPACE();
6151 while (hardcount < max && scan < loceol &&
6153 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6155 scan += UTF8SKIP(scan);
6161 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6170 while (scan < loceol && isSPACE((U8) *scan)) {
6175 while (scan < loceol && isSPACE_A((U8) *scan)) {
6180 PL_reg_flags |= RF_tainted;
6183 while (hardcount < max && scan < loceol &&
6184 isSPACE_LC_utf8((U8*)scan)) {
6185 scan += UTF8SKIP(scan);
6189 while (scan < loceol && isSPACE_LC(*scan))
6199 LOAD_UTF8_CHARCLASS_SPACE();
6200 while (hardcount < max && scan < loceol &&
6202 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6204 scan += UTF8SKIP(scan);
6210 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6219 while (scan < loceol && ! isSPACE((U8) *scan)) {
6225 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6226 scan += UTF8SKIP(scan);
6230 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6236 PL_reg_flags |= RF_tainted;
6239 while (hardcount < max && scan < loceol &&
6240 !isSPACE_LC_utf8((U8*)scan)) {
6241 scan += UTF8SKIP(scan);
6245 while (scan < loceol && !isSPACE_LC(*scan))
6252 LOAD_UTF8_CHARCLASS_DIGIT();
6253 while (hardcount < max && scan < loceol &&
6254 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6255 scan += UTF8SKIP(scan);
6259 while (scan < loceol && isDIGIT(*scan))
6264 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6269 PL_reg_flags |= RF_tainted;
6272 while (hardcount < max && scan < loceol &&
6273 isDIGIT_LC_utf8((U8*)scan)) {
6274 scan += UTF8SKIP(scan);
6278 while (scan < loceol && isDIGIT_LC(*scan))
6285 LOAD_UTF8_CHARCLASS_DIGIT();
6286 while (hardcount < max && scan < loceol &&
6287 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6288 scan += UTF8SKIP(scan);
6292 while (scan < loceol && !isDIGIT(*scan))
6298 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6299 scan += UTF8SKIP(scan);
6303 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6309 PL_reg_flags |= RF_tainted;
6312 while (hardcount < max && scan < loceol &&
6313 !isDIGIT_LC_utf8((U8*)scan)) {
6314 scan += UTF8SKIP(scan);
6318 while (scan < loceol && !isDIGIT_LC(*scan))
6325 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6331 LNBREAK can match two latin chars, which is ok,
6332 because we have a null terminated string, but we
6333 have to use hardcount in this situation
6335 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6344 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6349 while (scan < loceol && is_HORIZWS_latin1(scan))
6356 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6357 scan += UTF8SKIP(scan);
6361 while (scan < loceol && !is_HORIZWS_latin1(scan))
6369 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6374 while (scan < loceol && is_VERTWS_latin1(scan))
6382 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6383 scan += UTF8SKIP(scan);
6387 while (scan < loceol && !is_VERTWS_latin1(scan))
6393 default: /* Called on something of 0 width. */
6394 break; /* So match right here or not at all. */
6400 c = scan - PL_reginput;
6404 GET_RE_DEBUG_FLAGS_DECL;
6406 SV * const prop = sv_newmortal();
6407 regprop(prog, prop, p);
6408 PerlIO_printf(Perl_debug_log,
6409 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6410 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6418 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6420 - regclass_swash - prepare the utf8 swash
6424 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6430 RXi_GET_DECL(prog,progi);
6431 const struct reg_data * const data = prog ? progi->data : NULL;
6433 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6435 if (data && data->count) {
6436 const U32 n = ARG(node);
6438 if (data->what[n] == 's') {
6439 SV * const rv = MUTABLE_SV(data->data[n]);
6440 AV * const av = MUTABLE_AV(SvRV(rv));
6441 SV **const ary = AvARRAY(av);
6444 /* See the end of regcomp.c:S_regclass() for
6445 * documentation of these array elements. */
6448 a = SvROK(ary[1]) ? &ary[1] : NULL;
6449 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6453 else if (si && doinit) {
6454 sw = swash_init("utf8", "", si, 1, 0);
6455 (void)av_store(av, 1, sw);
6472 - reginclass - determine if a character falls into a character class
6474 n is the ANYOF regnode
6475 p is the target string
6476 lenp is pointer to the maximum number of bytes of how far to go in p
6477 (This is assumed wthout checking to always be at least the current
6479 utf8_target tells whether p is in UTF-8.
6481 Returns true if matched; false otherwise. If lenp is not NULL, on return
6482 from a successful match, the value it points to will be updated to how many
6483 bytes in p were matched. If there was no match, the value is undefined,
6484 possibly changed from the input.
6486 Note that this can be a synthetic start class, a combination of various
6487 nodes, so things you think might be mutually exclusive, such as locale,
6488 aren't. It can match both locale and non-locale
6493 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6496 const char flags = ANYOF_FLAGS(n);
6502 PERL_ARGS_ASSERT_REGINCLASS;
6504 /* If c is not already the code point, get it */
6505 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6506 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6507 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6508 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6509 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6510 * UTF8_ALLOW_FFFF */
6511 if (c_len == (STRLEN)-1)
6512 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6518 /* Use passed in max length, or one character if none passed in or less
6519 * than one character. And assume will match just one character. This is
6520 * overwritten later if matched more. */
6522 maxlen = (*lenp > c_len) ? *lenp : c_len;
6530 /* If this character is potentially in the bitmap, check it */
6532 if (ANYOF_BITMAP_TEST(n, c))
6534 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6541 else if (flags & ANYOF_LOCALE) {
6542 PL_reg_flags |= RF_tainted;
6544 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6545 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6549 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6550 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6551 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6552 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6553 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6554 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6555 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6556 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6557 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6558 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6559 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6560 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6561 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6562 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6563 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6564 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6565 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6566 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6567 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6568 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6569 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6570 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6571 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6572 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6573 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6574 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6575 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6576 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6577 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6578 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6579 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6580 ) /* How's that for a conditional? */
6587 /* If the bitmap didn't (or couldn't) match, and something outside the
6588 * bitmap could match, try that. Locale nodes specifiy completely the
6589 * behavior of code points in the bit map (otherwise, a utf8 target would
6590 * cause them to be treated as Unicode and not locale), except in
6591 * the very unlikely event when this node is a synthetic start class, which
6592 * could be a combination of locale and non-locale nodes. So allow locale
6593 * to match for the synthetic start class, which will give a false
6594 * positive that will be resolved when the match is done again as not part
6595 * of the synthetic start class */
6597 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6598 match = TRUE; /* Everything above 255 matches */
6600 else if (ANYOF_NONBITMAP(n)
6601 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6604 || (! (flags & ANYOF_LOCALE))
6605 || (flags & ANYOF_IS_SYNTHETIC)))))
6608 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6616 /* Not utf8. Convert as much of the string as available up
6617 * to the limit of how far the (single) character in the
6618 * pattern can possibly match (no need to go further). If
6619 * the node is a straight ANYOF or not folding, it can't
6620 * match more than one. Otherwise, It can match up to how
6621 * far a single char can fold to. Since not utf8, each
6622 * character is a single byte, so the max it can be in
6623 * bytes is the same as the max it can be in characters */
6624 STRLEN len = (OP(n) == ANYOF
6625 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6627 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6629 : UTF8_MAX_FOLD_CHAR_EXPAND;
6630 utf8_p = bytes_to_utf8(p, &len);
6633 if (swash_fetch(sw, utf8_p, TRUE))
6635 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6637 /* Here, we need to test if the fold of the target string
6638 * matches. In the case of a multi-char fold that is
6639 * caught by regcomp.c, it has stored all such folds into
6640 * 'av'; we linearly check to see if any match the target
6641 * string (folded). We know that the originals were each
6642 * one character, but we don't currently know how many
6643 * characters/bytes each folded to, except we do know that
6644 * there are small limits imposed by Unicode. XXX A
6645 * performance enhancement would be to have regcomp.c store
6646 * the max number of chars/bytes that are in an av entry,
6647 * as, say the 0th element. Even better would be to have a
6648 * hash of the few characters that can start a multi-char
6649 * fold to the max number of chars of those folds.
6651 * Further down, if there isn't a
6652 * match in the av, we will check if there is another
6653 * fold-type match. For that, we also need the fold, but
6654 * only the first character. No sense in folding it twice,
6655 * so we do it here, even if there isn't any multi-char
6656 * fold, so we always fold at least the first character.
6657 * If the node is a straight ANYOF node, or there is only
6658 * one character available in the string, or if there isn't
6659 * any av, that's all we have to fold. In the case of a
6660 * multi-char fold, we do have guarantees in Unicode that
6661 * it can only expand up to so many characters and so many
6662 * bytes. We keep track so don't exceed either.
6664 * If there is a match, we will need to advance (if lenp is
6665 * specified) the match pointer in the target string. But
6666 * what we are comparing here isn't that string directly,
6667 * but its fold, whose length may differ from the original.
6668 * As we go along in constructing the fold, therefore, we
6669 * create a map so that we know how many bytes in the
6670 * source to advance given that we have matched a certain
6671 * number of bytes in the fold. This map is stored in
6672 * 'map_fold_len_back'. The first character in the fold
6673 * has array element 1 contain the number of bytes in the
6674 * source that folded to it; the 2nd is the cumulative
6675 * number to match it; ... */
6676 U8 map_fold_len_back[UTF8_MAX_FOLD_CHAR_EXPAND+1] = { 0 };
6677 U8 folded[UTF8_MAXBYTES_CASE+1];
6678 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6679 STRLEN foldlen_for_av; /* num bytes in fold of all chars */
6681 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6683 /* Here, only need to fold the first char of the target
6685 to_utf8_fold(utf8_p, folded, &foldlen);
6686 foldlen_for_av = foldlen;
6687 map_fold_len_back[1] = UTF8SKIP(utf8_p);
6691 /* Here, need to fold more than the first char. Do so
6692 * up to the limits */
6694 U8* source_ptr = utf8_p; /* The source for the fold
6697 U8* folded_ptr = folded;
6698 U8* e = utf8_p + maxlen; /* Can't go beyond last
6699 available byte in the
6701 while (which_char < UTF8_MAX_FOLD_CHAR_EXPAND
6705 /* Fold the next character */
6706 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6707 STRLEN this_char_foldlen;
6708 to_utf8_fold(source_ptr,
6710 &this_char_foldlen);
6712 /* Bail if it would exceed the byte limit for
6713 * folding a single char. */
6714 if (this_char_foldlen + folded_ptr - folded >
6720 /* Save the first character's folded length, in
6721 * case we have to use it later */
6723 foldlen = this_char_foldlen;
6726 /* Here, add the fold of this character */
6727 Copy(this_char_folded,
6732 map_fold_len_back[which_char] =
6733 map_fold_len_back[which_char - 1]
6734 + UTF8SKIP(source_ptr);
6735 folded_ptr += this_char_foldlen;
6736 source_ptr += UTF8SKIP(source_ptr);
6739 foldlen_for_av = folded_ptr - folded;
6743 /* Do the linear search to see if the fold is in the list
6744 * of multi-char folds. (Useless to look if won't be able
6745 * to store that it is a multi-char fold in *lenp) */
6748 for (i = 0; i <= av_len(av); i++) {
6749 SV* const sv = *av_fetch(av, i, FALSE);
6751 const char * const s = SvPV_const(sv, len);
6752 if (len <= foldlen_for_av && memEQ(s,
6757 /* Advance the target string ptr to account for
6758 * this fold, but have to translate from the
6759 * folded length to the corresponding source
6760 * length. The array is indexed by how many
6761 * characters in the match */
6762 *lenp = map_fold_len_back[
6763 utf8_length(folded, folded + len)];
6771 /* If we allocated a string above, free it */
6772 if (! utf8_target) Safefree(utf8_p);
6777 return (flags & ANYOF_INVERT) ? !match : match;
6781 S_reghop3(U8 *s, I32 off, const U8* lim)
6785 PERL_ARGS_ASSERT_REGHOP3;
6788 while (off-- && s < lim) {
6789 /* XXX could check well-formedness here */
6794 while (off++ && s > lim) {
6796 if (UTF8_IS_CONTINUED(*s)) {
6797 while (s > lim && UTF8_IS_CONTINUATION(*s))
6800 /* XXX could check well-formedness here */
6807 /* there are a bunch of places where we use two reghop3's that should
6808 be replaced with this routine. but since thats not done yet
6809 we ifdef it out - dmq
6812 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6816 PERL_ARGS_ASSERT_REGHOP4;
6819 while (off-- && s < rlim) {
6820 /* XXX could check well-formedness here */
6825 while (off++ && s > llim) {
6827 if (UTF8_IS_CONTINUED(*s)) {
6828 while (s > llim && UTF8_IS_CONTINUATION(*s))
6831 /* XXX could check well-formedness here */
6839 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6843 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6846 while (off-- && s < lim) {
6847 /* XXX could check well-formedness here */
6854 while (off++ && s > lim) {
6856 if (UTF8_IS_CONTINUED(*s)) {
6857 while (s > lim && UTF8_IS_CONTINUATION(*s))
6860 /* XXX could check well-formedness here */
6869 restore_pos(pTHX_ void *arg)
6872 regexp * const rex = (regexp *)arg;
6873 if (PL_reg_eval_set) {
6874 if (PL_reg_oldsaved) {
6875 rex->subbeg = PL_reg_oldsaved;
6876 rex->sublen = PL_reg_oldsavedlen;
6877 #ifdef PERL_OLD_COPY_ON_WRITE
6878 rex->saved_copy = PL_nrs;
6880 RXp_MATCH_COPIED_on(rex);
6882 PL_reg_magic->mg_len = PL_reg_oldpos;
6883 PL_reg_eval_set = 0;
6884 PL_curpm = PL_reg_oldcurpm;
6889 S_to_utf8_substr(pTHX_ register regexp *prog)
6893 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6896 if (prog->substrs->data[i].substr
6897 && !prog->substrs->data[i].utf8_substr) {
6898 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6899 prog->substrs->data[i].utf8_substr = sv;
6900 sv_utf8_upgrade(sv);
6901 if (SvVALID(prog->substrs->data[i].substr)) {
6902 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6903 if (flags & FBMcf_TAIL) {
6904 /* Trim the trailing \n that fbm_compile added last
6906 SvCUR_set(sv, SvCUR(sv) - 1);
6907 /* Whilst this makes the SV technically "invalid" (as its
6908 buffer is no longer followed by "\0") when fbm_compile()
6909 adds the "\n" back, a "\0" is restored. */
6911 fbm_compile(sv, flags);
6913 if (prog->substrs->data[i].substr == prog->check_substr)
6914 prog->check_utf8 = sv;
6920 S_to_byte_substr(pTHX_ register regexp *prog)
6925 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6928 if (prog->substrs->data[i].utf8_substr
6929 && !prog->substrs->data[i].substr) {
6930 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6931 if (sv_utf8_downgrade(sv, TRUE)) {
6932 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6934 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6935 if (flags & FBMcf_TAIL) {
6936 /* Trim the trailing \n that fbm_compile added last
6938 SvCUR_set(sv, SvCUR(sv) - 1);
6940 fbm_compile(sv, flags);
6946 prog->substrs->data[i].substr = sv;
6947 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6948 prog->check_substr = sv;
6955 * c-indentation-style: bsd
6957 * indent-tabs-mode: t
6960 * ex: set ts=8 sts=4 sw=4 noet: