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 */
126 /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
127 * skip the check on EBCDIC platforms */
128 # define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
130 # define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
131 if (!CAT2(PL_utf8_,class)) { \
133 ENTER; save_re_context(); \
134 ok=CAT2(is_utf8_,class)((const U8*)str); \
135 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
138 /* Doesn't do an assert to verify that is correct */
139 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
140 if (!CAT2(PL_utf8_,class)) { \
141 bool throw_away PERL_UNUSED_DECL; \
142 ENTER; save_re_context(); \
143 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
146 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
147 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
148 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
150 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
151 LOAD_UTF8_CHARCLASS(X_begin, " "); \
152 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
153 /* These are utf8 constants, and not utf-ebcdic constants, so the \
154 * assert should likely and hopefully fail on an EBCDIC machine */ \
155 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
157 /* No asserts are done for these, in case called on an early \
158 * Unicode version in which they map to nothing */ \
159 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
160 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
161 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
162 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
163 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
164 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
165 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
167 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
169 /* The actual code for CCC_TRY, which uses several variables from the routine
170 * it's callable from. It is designed to be the bulk of a case statement.
171 * FUNC is the macro or function to call on non-utf8 targets that indicate if
172 * nextchr matches the class.
173 * UTF8_TEST is the whole test string to use for utf8 targets
174 * LOAD is what to use to test, and if not present to load in the swash for the
176 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
178 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
179 * utf8 and a variant, load the swash if necessary and test using the utf8
180 * test. Advance to the next character if test is ok, otherwise fail; If not
181 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
182 * fails, or advance to the next character */
184 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
185 if (locinput >= PL_regeol) { \
188 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
189 LOAD_UTF8_CHARCLASS(CLASS, STR); \
190 if (POS_OR_NEG (UTF8_TEST)) { \
193 locinput += PL_utf8skip[nextchr]; \
194 nextchr = UCHARAT(locinput); \
197 if (POS_OR_NEG (FUNC(nextchr))) { \
200 nextchr = UCHARAT(++locinput); \
203 /* Handle the non-locale cases for a character class and its complement. It
204 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
205 * This is because that code fails when the test succeeds, so we want to have
206 * the test fail so that the code succeeds. The swash is stored in a
207 * predictable PL_ place */
208 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
211 _CCC_TRY_CODE( !, FUNC, \
212 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
213 (U8*)locinput, TRUE)), \
216 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
217 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
218 (U8*)locinput, TRUE)), \
221 /* Generate the case statements for both locale and non-locale character
222 * classes in regmatch for classes that don't have special unicode semantics.
223 * Locales don't use an immediate swash, but an intermediary special locale
224 * function that is called on the pointer to the current place in the input
225 * string. That function will resolve to needing the same swash. One might
226 * think that because we don't know what the locale will match, we shouldn't
227 * check with the swash loading function that it loaded properly; ie, that we
228 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
229 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
231 #define CCC_TRY(NAME, NNAME, FUNC, \
232 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
233 NAMEA, NNAMEA, FUNCA, \
236 PL_reg_flags |= RF_tainted; \
237 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
239 PL_reg_flags |= RF_tainted; \
240 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
243 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
246 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
247 nextchr = UCHARAT(++locinput); \
250 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
254 locinput += PL_utf8skip[nextchr]; \
255 nextchr = UCHARAT(locinput); \
258 nextchr = UCHARAT(++locinput); \
261 /* Generate the non-locale cases */ \
262 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
264 /* This is like CCC_TRY, but has an extra set of parameters for generating case
265 * statements to handle separate Unicode semantics nodes */
266 #define CCC_TRY_U(NAME, NNAME, FUNC, \
267 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
268 NAMEU, NNAMEU, FUNCU, \
269 NAMEA, NNAMEA, FUNCA, \
271 CCC_TRY(NAME, NNAME, FUNC, \
272 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
273 NAMEA, NNAMEA, FUNCA, \
275 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
277 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
279 /* for use after a quantifier and before an EXACT-like node -- japhy */
280 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
282 * NOTE that *nothing* that affects backtracking should be in here, specifically
283 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
284 * node that is in between two EXACT like nodes when ascertaining what the required
285 * "follow" character is. This should probably be moved to regex compile time
286 * although it may be done at run time beause of the REF possibility - more
287 * investigation required. -- demerphq
289 #define JUMPABLE(rn) ( \
291 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
293 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
294 OP(rn) == PLUS || OP(rn) == MINMOD || \
296 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
298 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
300 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
303 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
304 we don't need this definition. */
305 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
306 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
307 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
310 /* ... so we use this as its faster. */
311 #define IS_TEXT(rn) ( OP(rn)==EXACT )
312 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
313 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
314 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
319 Search for mandatory following text node; for lookahead, the text must
320 follow but for lookbehind (rn->flags != 0) we skip to the next step.
322 #define FIND_NEXT_IMPT(rn) STMT_START { \
323 while (JUMPABLE(rn)) { \
324 const OPCODE type = OP(rn); \
325 if (type == SUSPEND || PL_regkind[type] == CURLY) \
326 rn = NEXTOPER(NEXTOPER(rn)); \
327 else if (type == PLUS) \
329 else if (type == IFMATCH) \
330 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
331 else rn += NEXT_OFF(rn); \
336 static void restore_pos(pTHX_ void *arg);
338 #define REGCP_PAREN_ELEMS 4
339 #define REGCP_OTHER_ELEMS 5
340 #define REGCP_FRAME_ELEMS 1
341 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
342 * are needed for the regexp context stack bookkeeping. */
345 S_regcppush(pTHX_ I32 parenfloor)
348 const int retval = PL_savestack_ix;
349 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
350 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
351 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
353 GET_RE_DEBUG_FLAGS_DECL;
355 if (paren_elems_to_push < 0)
356 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
357 paren_elems_to_push);
359 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
360 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
361 " out of range (%lu-%ld)",
362 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
364 SSGROW(total_elems + REGCP_FRAME_ELEMS);
366 for (p = PL_regsize; p > parenfloor; p--) {
367 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
368 SSPUSHINT(PL_regoffs[p].end);
369 SSPUSHINT(PL_regoffs[p].start);
370 SSPUSHPTR(PL_reg_start_tmp[p]);
372 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
373 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
374 (UV)p, (IV)PL_regoffs[p].start,
375 (IV)(PL_reg_start_tmp[p] - PL_bostr),
376 (IV)PL_regoffs[p].end
379 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
380 SSPUSHPTR(PL_regoffs);
381 SSPUSHINT(PL_regsize);
382 SSPUSHINT(*PL_reglastparen);
383 SSPUSHINT(*PL_reglastcloseparen);
384 SSPUSHPTR(PL_reginput);
385 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
390 /* These are needed since we do not localize EVAL nodes: */
391 #define REGCP_SET(cp) \
393 PerlIO_printf(Perl_debug_log, \
394 " Setting an EVAL scope, savestack=%"IVdf"\n", \
395 (IV)PL_savestack_ix)); \
398 #define REGCP_UNWIND(cp) \
400 if (cp != PL_savestack_ix) \
401 PerlIO_printf(Perl_debug_log, \
402 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
403 (IV)(cp), (IV)PL_savestack_ix)); \
407 S_regcppop(pTHX_ const regexp *rex)
412 GET_RE_DEBUG_FLAGS_DECL;
414 PERL_ARGS_ASSERT_REGCPPOP;
416 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
418 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
419 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
420 input = (char *) SSPOPPTR;
421 *PL_reglastcloseparen = SSPOPINT;
422 *PL_reglastparen = SSPOPINT;
423 PL_regsize = SSPOPINT;
424 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
426 i -= REGCP_OTHER_ELEMS;
427 /* Now restore the parentheses context. */
428 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
430 U32 paren = (U32)SSPOPINT;
431 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
432 PL_regoffs[paren].start = SSPOPINT;
434 if (paren <= *PL_reglastparen)
435 PL_regoffs[paren].end = tmps;
437 PerlIO_printf(Perl_debug_log,
438 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
439 (UV)paren, (IV)PL_regoffs[paren].start,
440 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
441 (IV)PL_regoffs[paren].end,
442 (paren > *PL_reglastparen ? "(no)" : ""));
446 if (*PL_reglastparen + 1 <= rex->nparens) {
447 PerlIO_printf(Perl_debug_log,
448 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
449 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
453 /* It would seem that the similar code in regtry()
454 * already takes care of this, and in fact it is in
455 * a better location to since this code can #if 0-ed out
456 * but the code in regtry() is needed or otherwise tests
457 * requiring null fields (pat.t#187 and split.t#{13,14}
458 * (as of patchlevel 7877) will fail. Then again,
459 * this code seems to be necessary or otherwise
460 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
461 * --jhi updated by dapm */
462 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
464 PL_regoffs[i].start = -1;
465 PL_regoffs[i].end = -1;
471 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
474 * pregexec and friends
477 #ifndef PERL_IN_XSUB_RE
479 - pregexec - match a regexp against a string
482 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
483 char *strbeg, I32 minend, SV *screamer, U32 nosave)
484 /* strend: pointer to null at end of string */
485 /* strbeg: real beginning of string */
486 /* minend: end of match must be >=minend after stringarg. */
487 /* nosave: For optimizations. */
489 PERL_ARGS_ASSERT_PREGEXEC;
492 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
493 nosave ? 0 : REXEC_COPY_STR);
498 * Need to implement the following flags for reg_anch:
500 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
502 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
503 * INTUIT_AUTORITATIVE_ML
504 * INTUIT_ONCE_NOML - Intuit can match in one location only.
507 * Another flag for this function: SECOND_TIME (so that float substrs
508 * with giant delta may be not rechecked).
511 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
513 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
514 Otherwise, only SvCUR(sv) is used to get strbeg. */
516 /* XXXX We assume that strpos is strbeg unless sv. */
518 /* XXXX Some places assume that there is a fixed substring.
519 An update may be needed if optimizer marks as "INTUITable"
520 RExen without fixed substrings. Similarly, it is assumed that
521 lengths of all the strings are no more than minlen, thus they
522 cannot come from lookahead.
523 (Or minlen should take into account lookahead.)
524 NOTE: Some of this comment is not correct. minlen does now take account
525 of lookahead/behind. Further research is required. -- demerphq
529 /* A failure to find a constant substring means that there is no need to make
530 an expensive call to REx engine, thus we celebrate a failure. Similarly,
531 finding a substring too deep into the string means that less calls to
532 regtry() should be needed.
534 REx compiler's optimizer found 4 possible hints:
535 a) Anchored substring;
537 c) Whether we are anchored (beginning-of-line or \G);
538 d) First node (of those at offset 0) which may distinguish positions;
539 We use a)b)d) and multiline-part of c), and try to find a position in the
540 string which does not contradict any of them.
543 /* Most of decisions we do here should have been done at compile time.
544 The nodes of the REx which we used for the search should have been
545 deleted from the finite automaton. */
548 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
549 char *strend, const U32 flags, re_scream_pos_data *data)
552 struct regexp *const prog = (struct regexp *)SvANY(rx);
553 register I32 start_shift = 0;
554 /* Should be nonnegative! */
555 register I32 end_shift = 0;
560 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
562 register char *other_last = NULL; /* other substr checked before this */
563 char *check_at = NULL; /* check substr found at this pos */
564 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
565 RXi_GET_DECL(prog,progi);
567 const char * const i_strpos = strpos;
569 GET_RE_DEBUG_FLAGS_DECL;
571 PERL_ARGS_ASSERT_RE_INTUIT_START;
573 RX_MATCH_UTF8_set(rx,utf8_target);
576 PL_reg_flags |= RF_utf8;
579 debug_start_match(rx, utf8_target, strpos, strend,
580 sv ? "Guessing start of match in sv for"
581 : "Guessing start of match in string for");
584 /* CHR_DIST() would be more correct here but it makes things slow. */
585 if (prog->minlen > strend - strpos) {
586 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
587 "String too short... [re_intuit_start]\n"));
591 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
594 if (!prog->check_utf8 && prog->check_substr)
595 to_utf8_substr(prog);
596 check = prog->check_utf8;
598 if (!prog->check_substr && prog->check_utf8)
599 to_byte_substr(prog);
600 check = prog->check_substr;
602 if (check == &PL_sv_undef) {
603 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
604 "Non-utf8 string cannot match utf8 check string\n"));
607 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
608 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
609 || ( (prog->extflags & RXf_ANCH_BOL)
610 && !multiline ) ); /* Check after \n? */
613 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
614 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
615 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
617 && (strpos != strbeg)) {
618 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
621 if (prog->check_offset_min == prog->check_offset_max &&
622 !(prog->extflags & RXf_CANY_SEEN)) {
623 /* Substring at constant offset from beg-of-str... */
626 s = HOP3c(strpos, prog->check_offset_min, strend);
629 slen = SvCUR(check); /* >= 1 */
631 if ( strend - s > slen || strend - s < slen - 1
632 || (strend - s == slen && strend[-1] != '\n')) {
633 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
636 /* Now should match s[0..slen-2] */
638 if (slen && (*SvPVX_const(check) != *s
640 && memNE(SvPVX_const(check), s, slen)))) {
642 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
646 else if (*SvPVX_const(check) != *s
647 || ((slen = SvCUR(check)) > 1
648 && memNE(SvPVX_const(check), s, slen)))
651 goto success_at_start;
654 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
656 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
657 end_shift = prog->check_end_shift;
660 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
661 - (SvTAIL(check) != 0);
662 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
664 if (end_shift < eshift)
668 else { /* Can match at random position */
671 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
672 end_shift = prog->check_end_shift;
674 /* end shift should be non negative here */
677 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
679 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
680 (IV)end_shift, RX_PRECOMP(prog));
684 /* Find a possible match in the region s..strend by looking for
685 the "check" substring in the region corrected by start/end_shift. */
688 I32 srch_start_shift = start_shift;
689 I32 srch_end_shift = end_shift;
690 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
691 srch_end_shift -= ((strbeg - s) - srch_start_shift);
692 srch_start_shift = strbeg - s;
694 DEBUG_OPTIMISE_MORE_r({
695 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
696 (IV)prog->check_offset_min,
697 (IV)srch_start_shift,
699 (IV)prog->check_end_shift);
702 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
703 I32 p = -1; /* Internal iterator of scream. */
704 I32 * const pp = data ? data->scream_pos : &p;
708 assert(SvMAGICAL(sv));
709 mg = mg_find(sv, PERL_MAGIC_study);
712 if (mg->mg_private == 1) {
713 found = ((U8 *)mg->mg_ptr)[BmRARE(check)] != (U8)~0;
714 } else if (mg->mg_private == 2) {
715 found = ((U16 *)mg->mg_ptr)[BmRARE(check)] != (U16)~0;
717 assert (mg->mg_private == 4);
718 found = ((U32 *)mg->mg_ptr)[BmRARE(check)] != (U32)~0;
722 || ( BmRARE(check) == '\n'
723 && (BmPREVIOUS(check) == SvCUR(check) - 1)
725 s = screaminstr(sv, check,
726 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
729 /* we may be pointing at the wrong string */
730 if (s && RXp_MATCH_COPIED(prog))
731 s = strbeg + (s - SvPVX_const(sv));
733 *data->scream_olds = s;
738 if (prog->extflags & RXf_CANY_SEEN) {
739 start_point= (U8*)(s + srch_start_shift);
740 end_point= (U8*)(strend - srch_end_shift);
742 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
743 end_point= HOP3(strend, -srch_end_shift, strbeg);
745 DEBUG_OPTIMISE_MORE_r({
746 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
747 (int)(end_point - start_point),
748 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
752 s = fbm_instr( start_point, end_point,
753 check, multiline ? FBMrf_MULTILINE : 0);
756 /* Update the count-of-usability, remove useless subpatterns,
760 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
761 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
762 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
763 (s ? "Found" : "Did not find"),
764 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
765 ? "anchored" : "floating"),
768 (s ? " at offset " : "...\n") );
773 /* Finish the diagnostic message */
774 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
776 /* XXX dmq: first branch is for positive lookbehind...
777 Our check string is offset from the beginning of the pattern.
778 So we need to do any stclass tests offset forward from that
787 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
788 Start with the other substr.
789 XXXX no SCREAM optimization yet - and a very coarse implementation
790 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
791 *always* match. Probably should be marked during compile...
792 Probably it is right to do no SCREAM here...
795 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
796 : (prog->float_substr && prog->anchored_substr))
798 /* Take into account the "other" substring. */
799 /* XXXX May be hopelessly wrong for UTF... */
802 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
805 char * const last = HOP3c(s, -start_shift, strbeg);
807 char * const saved_s = s;
810 t = s - prog->check_offset_max;
811 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
813 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
818 t = HOP3c(t, prog->anchored_offset, strend);
819 if (t < other_last) /* These positions already checked */
821 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
824 /* XXXX It is not documented what units *_offsets are in.
825 We assume bytes, but this is clearly wrong.
826 Meaning this code needs to be carefully reviewed for errors.
830 /* On end-of-str: see comment below. */
831 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
832 if (must == &PL_sv_undef) {
834 DEBUG_r(must = prog->anchored_utf8); /* for debug */
839 HOP3(HOP3(last1, prog->anchored_offset, strend)
840 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
842 multiline ? FBMrf_MULTILINE : 0
845 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
846 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
847 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
848 (s ? "Found" : "Contradicts"),
849 quoted, RE_SV_TAIL(must));
854 if (last1 >= last2) {
855 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
856 ", giving up...\n"));
859 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
860 ", trying floating at offset %ld...\n",
861 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
862 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
863 s = HOP3c(last, 1, strend);
867 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
868 (long)(s - i_strpos)));
869 t = HOP3c(s, -prog->anchored_offset, strbeg);
870 other_last = HOP3c(s, 1, strend);
878 else { /* Take into account the floating substring. */
880 char * const saved_s = s;
883 t = HOP3c(s, -start_shift, strbeg);
885 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
886 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
887 last = HOP3c(t, prog->float_max_offset, strend);
888 s = HOP3c(t, prog->float_min_offset, strend);
891 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
892 must = utf8_target ? prog->float_utf8 : prog->float_substr;
893 /* fbm_instr() takes into account exact value of end-of-str
894 if the check is SvTAIL(ed). Since false positives are OK,
895 and end-of-str is not later than strend we are OK. */
896 if (must == &PL_sv_undef) {
898 DEBUG_r(must = prog->float_utf8); /* for debug message */
901 s = fbm_instr((unsigned char*)s,
902 (unsigned char*)last + SvCUR(must)
904 must, multiline ? FBMrf_MULTILINE : 0);
906 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
907 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
908 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
909 (s ? "Found" : "Contradicts"),
910 quoted, RE_SV_TAIL(must));
914 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
915 ", giving up...\n"));
918 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
919 ", trying anchored starting at offset %ld...\n",
920 (long)(saved_s + 1 - i_strpos)));
922 s = HOP3c(t, 1, strend);
926 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
927 (long)(s - i_strpos)));
928 other_last = s; /* Fix this later. --Hugo */
938 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
940 DEBUG_OPTIMISE_MORE_r(
941 PerlIO_printf(Perl_debug_log,
942 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
943 (IV)prog->check_offset_min,
944 (IV)prog->check_offset_max,
952 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
954 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
957 /* Fixed substring is found far enough so that the match
958 cannot start at strpos. */
960 if (ml_anch && t[-1] != '\n') {
961 /* Eventually fbm_*() should handle this, but often
962 anchored_offset is not 0, so this check will not be wasted. */
963 /* XXXX In the code below we prefer to look for "^" even in
964 presence of anchored substrings. And we search even
965 beyond the found float position. These pessimizations
966 are historical artefacts only. */
968 while (t < strend - prog->minlen) {
970 if (t < check_at - prog->check_offset_min) {
971 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
972 /* Since we moved from the found position,
973 we definitely contradict the found anchored
974 substr. Due to the above check we do not
975 contradict "check" substr.
976 Thus we can arrive here only if check substr
977 is float. Redo checking for "other"=="fixed".
980 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
981 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
982 goto do_other_anchored;
984 /* We don't contradict the found floating substring. */
985 /* XXXX Why not check for STCLASS? */
987 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
988 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
991 /* Position contradicts check-string */
992 /* XXXX probably better to look for check-string
993 than for "\n", so one should lower the limit for t? */
994 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
995 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
996 other_last = strpos = s = t + 1;
1001 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1002 PL_colors[0], PL_colors[1]));
1006 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1007 PL_colors[0], PL_colors[1]));
1011 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1014 /* The found string does not prohibit matching at strpos,
1015 - no optimization of calling REx engine can be performed,
1016 unless it was an MBOL and we are not after MBOL,
1017 or a future STCLASS check will fail this. */
1019 /* Even in this situation we may use MBOL flag if strpos is offset
1020 wrt the start of the string. */
1021 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1022 && (strpos != strbeg) && strpos[-1] != '\n'
1023 /* May be due to an implicit anchor of m{.*foo} */
1024 && !(prog->intflags & PREGf_IMPLICIT))
1029 DEBUG_EXECUTE_r( if (ml_anch)
1030 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1031 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1034 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1036 prog->check_utf8 /* Could be deleted already */
1037 && --BmUSEFUL(prog->check_utf8) < 0
1038 && (prog->check_utf8 == prog->float_utf8)
1040 prog->check_substr /* Could be deleted already */
1041 && --BmUSEFUL(prog->check_substr) < 0
1042 && (prog->check_substr == prog->float_substr)
1045 /* If flags & SOMETHING - do not do it many times on the same match */
1046 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1047 /* XXX Does the destruction order has to change with utf8_target? */
1048 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1049 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1050 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1051 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1052 check = NULL; /* abort */
1054 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1055 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1056 if (prog->intflags & PREGf_IMPLICIT)
1057 prog->extflags &= ~RXf_ANCH_MBOL;
1058 /* XXXX This is a remnant of the old implementation. It
1059 looks wasteful, since now INTUIT can use many
1060 other heuristics. */
1061 prog->extflags &= ~RXf_USE_INTUIT;
1062 /* XXXX What other flags might need to be cleared in this branch? */
1068 /* Last resort... */
1069 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1070 /* trie stclasses are too expensive to use here, we are better off to
1071 leave it to regmatch itself */
1072 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1073 /* minlen == 0 is possible if regstclass is \b or \B,
1074 and the fixed substr is ''$.
1075 Since minlen is already taken into account, s+1 is before strend;
1076 accidentally, minlen >= 1 guaranties no false positives at s + 1
1077 even for \b or \B. But (minlen? 1 : 0) below assumes that
1078 regstclass does not come from lookahead... */
1079 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1080 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1081 const U8* const str = (U8*)STRING(progi->regstclass);
1082 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1083 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1086 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1087 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1088 else if (prog->float_substr || prog->float_utf8)
1089 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1093 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1094 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1097 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1100 const char *what = NULL;
1102 if (endpos == strend) {
1103 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1104 "Could not match STCLASS...\n") );
1107 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1108 "This position contradicts STCLASS...\n") );
1109 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1111 /* Contradict one of substrings */
1112 if (prog->anchored_substr || prog->anchored_utf8) {
1113 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1114 DEBUG_EXECUTE_r( what = "anchored" );
1116 s = HOP3c(t, 1, strend);
1117 if (s + start_shift + end_shift > strend) {
1118 /* XXXX Should be taken into account earlier? */
1119 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1120 "Could not match STCLASS...\n") );
1125 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1126 "Looking for %s substr starting at offset %ld...\n",
1127 what, (long)(s + start_shift - i_strpos)) );
1130 /* Have both, check_string is floating */
1131 if (t + start_shift >= check_at) /* Contradicts floating=check */
1132 goto retry_floating_check;
1133 /* Recheck anchored substring, but not floating... */
1137 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1138 "Looking for anchored substr starting at offset %ld...\n",
1139 (long)(other_last - i_strpos)) );
1140 goto do_other_anchored;
1142 /* Another way we could have checked stclass at the
1143 current position only: */
1148 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1149 "Looking for /%s^%s/m starting at offset %ld...\n",
1150 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1153 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1155 /* Check is floating substring. */
1156 retry_floating_check:
1157 t = check_at - start_shift;
1158 DEBUG_EXECUTE_r( what = "floating" );
1159 goto hop_and_restart;
1162 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1163 "By STCLASS: moving %ld --> %ld\n",
1164 (long)(t - i_strpos), (long)(s - i_strpos))
1168 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1169 "Does not contradict STCLASS...\n");
1174 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1175 PL_colors[4], (check ? "Guessed" : "Giving up"),
1176 PL_colors[5], (long)(s - i_strpos)) );
1179 fail_finish: /* Substring not found */
1180 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1181 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1183 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1184 PL_colors[4], PL_colors[5]));
1188 #define DECL_TRIE_TYPE(scan) \
1189 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1190 trie_type = (scan->flags != EXACT) \
1191 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1192 : (utf8_target ? trie_utf8 : trie_plain)
1194 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1195 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1196 switch (trie_type) { \
1197 case trie_utf8_fold: \
1198 if ( foldlen>0 ) { \
1199 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1204 uvc = to_utf8_fold( (U8 *) uc, foldbuf, &foldlen ); \
1205 len = UTF8SKIP(uc); \
1206 foldlen -= UNISKIP( uvc ); \
1207 uscan = foldbuf + UNISKIP( uvc ); \
1210 case trie_latin_utf8_fold: \
1211 if ( foldlen>0 ) { \
1212 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1218 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1219 foldlen -= UNISKIP( uvc ); \
1220 uscan = foldbuf + UNISKIP( uvc ); \
1224 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1231 charid = trie->charmap[ uvc ]; \
1235 if (widecharmap) { \
1236 SV** const svpp = hv_fetch(widecharmap, \
1237 (char*)&uvc, sizeof(UV), 0); \
1239 charid = (U16)SvIV(*svpp); \
1244 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1248 && (ln == 1 || folder(s, pat_string, ln)) \
1249 && (!reginfo || regtry(reginfo, &s)) ) \
1255 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1257 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1263 #define REXEC_FBC_SCAN(CoDe) \
1265 while (s < strend) { \
1271 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1272 REXEC_FBC_UTF8_SCAN( \
1274 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1283 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1286 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1295 #define REXEC_FBC_TRYIT \
1296 if ((!reginfo || regtry(reginfo, &s))) \
1299 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1300 if (utf8_target) { \
1301 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1304 REXEC_FBC_CLASS_SCAN(CoNd); \
1307 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1308 if (utf8_target) { \
1310 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1313 REXEC_FBC_CLASS_SCAN(CoNd); \
1316 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1317 PL_reg_flags |= RF_tainted; \
1318 if (utf8_target) { \
1319 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1322 REXEC_FBC_CLASS_SCAN(CoNd); \
1325 #define DUMP_EXEC_POS(li,s,doutf8) \
1326 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1329 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1330 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1331 tmp = TEST_NON_UTF8(tmp); \
1332 REXEC_FBC_UTF8_SCAN( \
1333 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1342 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1343 if (s == PL_bostr) { \
1347 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1348 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1351 LOAD_UTF8_CHARCLASS_ALNUM(); \
1352 REXEC_FBC_UTF8_SCAN( \
1353 if (tmp == ! (TeSt2_UtF8)) { \
1362 /* The only difference between the BOUND and NBOUND cases is that
1363 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1364 * NBOUND. This is accomplished by passing it in either the if or else clause,
1365 * with the other one being empty */
1366 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1367 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1369 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1370 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1372 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1373 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1375 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1376 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1379 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1380 * be passed in completely with the variable name being tested, which isn't
1381 * such a clean interface, but this is easier to read than it was before. We
1382 * are looking for the boundary (or non-boundary between a word and non-word
1383 * character. The utf8 and non-utf8 cases have the same logic, but the details
1384 * must be different. Find the "wordness" of the character just prior to this
1385 * one, and compare it with the wordness of this one. If they differ, we have
1386 * a boundary. At the beginning of the string, pretend that the previous
1387 * character was a new-line */
1388 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1389 if (utf8_target) { \
1392 else { /* Not utf8 */ \
1393 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1394 tmp = TEST_NON_UTF8(tmp); \
1396 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1405 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1408 /* We know what class REx starts with. Try to find this position... */
1409 /* if reginfo is NULL, its a dryrun */
1410 /* annoyingly all the vars in this routine have different names from their counterparts
1411 in regmatch. /grrr */
1414 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1415 const char *strend, regmatch_info *reginfo)
1418 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1419 char *pat_string; /* The pattern's exactish string */
1420 char *pat_end; /* ptr to end char of pat_string */
1421 re_fold_t folder; /* Function for computing non-utf8 folds */
1422 const U8 *fold_array; /* array for folding ords < 256 */
1425 register STRLEN uskip;
1429 register I32 tmp = 1; /* Scratch variable? */
1430 register const bool utf8_target = PL_reg_match_utf8;
1431 UV utf8_fold_flags = 0;
1432 RXi_GET_DECL(prog,progi);
1434 PERL_ARGS_ASSERT_FIND_BYCLASS;
1436 /* We know what class it must start with. */
1440 if (utf8_target || OP(c) == ANYOFV) {
1441 STRLEN inclasslen = strend - s;
1442 REXEC_FBC_UTF8_CLASS_SCAN(
1443 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1446 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1451 if (tmp && (!reginfo || regtry(reginfo, &s)))
1459 if (UTF_PATTERN || utf8_target) {
1460 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1461 goto do_exactf_utf8;
1463 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1464 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1465 goto do_exactf_non_utf8; /* isn't dealt with by these */
1470 /* regcomp.c already folded this if pattern is in UTF-8 */
1471 utf8_fold_flags = 0;
1472 goto do_exactf_utf8;
1474 fold_array = PL_fold;
1476 goto do_exactf_non_utf8;
1479 if (UTF_PATTERN || utf8_target) {
1480 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1481 goto do_exactf_utf8;
1483 fold_array = PL_fold_locale;
1484 folder = foldEQ_locale;
1485 goto do_exactf_non_utf8;
1488 if (UTF_PATTERN || utf8_target) {
1489 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1490 goto do_exactf_utf8;
1493 /* Any 'ss' in the pattern should have been replaced by regcomp,
1494 * so we don't have to worry here about this single special case
1495 * in the Latin1 range */
1496 fold_array = PL_fold_latin1;
1497 folder = foldEQ_latin1;
1501 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1502 are no glitches with fold-length differences
1503 between the target string and pattern */
1505 /* The idea in the non-utf8 EXACTF* cases is to first find the
1506 * first character of the EXACTF* node and then, if necessary,
1507 * case-insensitively compare the full text of the node. c1 is the
1508 * first character. c2 is its fold. This logic will not work for
1509 * Unicode semantics and the german sharp ss, which hence should
1510 * not be compiled into a node that gets here. */
1511 pat_string = STRING(c);
1512 ln = STR_LEN(c); /* length to match in octets/bytes */
1514 /* We know that we have to match at least 'ln' bytes (which is the
1515 * same as characters, since not utf8). If we have to match 3
1516 * characters, and there are only 2 availabe, we know without
1517 * trying that it will fail; so don't start a match past the
1518 * required minimum number from the far end */
1519 e = HOP3c(strend, -((I32)ln), s);
1521 if (!reginfo && e < s) {
1522 e = s; /* Due to minlen logic of intuit() */
1526 c2 = fold_array[c1];
1527 if (c1 == c2) { /* If char and fold are the same */
1528 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1531 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1540 /* If one of the operands is in utf8, we can't use the simpler
1541 * folding above, due to the fact that many different characters
1542 * can have the same fold, or portion of a fold, or different-
1544 pat_string = STRING(c);
1545 ln = STR_LEN(c); /* length to match in octets/bytes */
1546 pat_end = pat_string + ln;
1547 lnc = (UTF_PATTERN) /* length to match in characters */
1548 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1551 /* We have 'lnc' characters to match in the pattern, but because of
1552 * multi-character folding, each character in the target can match
1553 * up to 3 characters (Unicode guarantees it will never exceed
1554 * this) if it is utf8-encoded; and up to 2 if not (based on the
1555 * fact that the Latin 1 folds are already determined, and the
1556 * only multi-char fold in that range is the sharp-s folding to
1557 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1558 * string character. Adjust lnc accordingly, rounding up, so that
1559 * if we need to match at least 4+1/3 chars, that really is 5. */
1560 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1561 lnc = (lnc + expansion - 1) / expansion;
1563 /* As in the non-UTF8 case, if we have to match 3 characters, and
1564 * only 2 are left, it's guaranteed to fail, so don't start a
1565 * match that would require us to go beyond the end of the string
1567 e = HOP3c(strend, -((I32)lnc), s);
1569 if (!reginfo && e < s) {
1570 e = s; /* Due to minlen logic of intuit() */
1573 /* XXX Note that we could recalculate e to stop the loop earlier,
1574 * as the worst case expansion above will rarely be met, and as we
1575 * go along we would usually find that e moves further to the left.
1576 * This would happen only after we reached the point in the loop
1577 * where if there were no expansion we should fail. Unclear if
1578 * worth the expense */
1581 char *my_strend= (char *)strend;
1582 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1583 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1584 && (!reginfo || regtry(reginfo, &s)) )
1588 s += (utf8_target) ? UTF8SKIP(s) : 1;
1593 PL_reg_flags |= RF_tainted;
1594 FBC_BOUND(isALNUM_LC,
1595 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1596 isALNUM_LC_utf8((U8*)s));
1599 PL_reg_flags |= RF_tainted;
1600 FBC_NBOUND(isALNUM_LC,
1601 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1602 isALNUM_LC_utf8((U8*)s));
1605 FBC_BOUND(isWORDCHAR,
1607 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1610 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1612 isWORDCHAR_A((U8*)s));
1615 FBC_NBOUND(isWORDCHAR,
1617 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1620 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1622 isWORDCHAR_A((U8*)s));
1625 FBC_BOUND(isWORDCHAR_L1,
1627 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1630 FBC_NBOUND(isWORDCHAR_L1,
1632 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1635 REXEC_FBC_CSCAN_TAINT(
1636 isALNUM_LC_utf8((U8*)s),
1641 REXEC_FBC_CSCAN_PRELOAD(
1642 LOAD_UTF8_CHARCLASS_ALNUM(),
1643 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1644 isWORDCHAR_L1((U8) *s)
1648 REXEC_FBC_CSCAN_PRELOAD(
1649 LOAD_UTF8_CHARCLASS_ALNUM(),
1650 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1655 /* Don't need to worry about utf8, as it can match only a single
1656 * byte invariant character */
1657 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1660 REXEC_FBC_CSCAN_PRELOAD(
1661 LOAD_UTF8_CHARCLASS_ALNUM(),
1662 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1663 ! isWORDCHAR_L1((U8) *s)
1667 REXEC_FBC_CSCAN_PRELOAD(
1668 LOAD_UTF8_CHARCLASS_ALNUM(),
1669 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1680 REXEC_FBC_CSCAN_TAINT(
1681 !isALNUM_LC_utf8((U8*)s),
1686 REXEC_FBC_CSCAN_PRELOAD(
1687 LOAD_UTF8_CHARCLASS_SPACE(),
1688 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1693 REXEC_FBC_CSCAN_PRELOAD(
1694 LOAD_UTF8_CHARCLASS_SPACE(),
1695 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1700 /* Don't need to worry about utf8, as it can match only a single
1701 * byte invariant character */
1702 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1705 REXEC_FBC_CSCAN_TAINT(
1706 isSPACE_LC_utf8((U8*)s),
1711 REXEC_FBC_CSCAN_PRELOAD(
1712 LOAD_UTF8_CHARCLASS_SPACE(),
1713 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1714 ! isSPACE_L1((U8) *s)
1718 REXEC_FBC_CSCAN_PRELOAD(
1719 LOAD_UTF8_CHARCLASS_SPACE(),
1720 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1731 REXEC_FBC_CSCAN_TAINT(
1732 !isSPACE_LC_utf8((U8*)s),
1737 REXEC_FBC_CSCAN_PRELOAD(
1738 LOAD_UTF8_CHARCLASS_DIGIT(),
1739 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1744 /* Don't need to worry about utf8, as it can match only a single
1745 * byte invariant character */
1746 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1749 REXEC_FBC_CSCAN_TAINT(
1750 isDIGIT_LC_utf8((U8*)s),
1755 REXEC_FBC_CSCAN_PRELOAD(
1756 LOAD_UTF8_CHARCLASS_DIGIT(),
1757 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1768 REXEC_FBC_CSCAN_TAINT(
1769 !isDIGIT_LC_utf8((U8*)s),
1776 is_LNBREAK_latin1(s)
1788 !is_VERTWS_latin1(s)
1794 is_HORIZWS_latin1(s)
1799 !is_HORIZWS_utf8(s),
1800 !is_HORIZWS_latin1(s)
1807 /* what trie are we using right now */
1809 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1811 = (reg_trie_data*)progi->data->data[ aho->trie ];
1812 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1814 const char *last_start = strend - trie->minlen;
1816 const char *real_start = s;
1818 STRLEN maxlen = trie->maxlen;
1820 U8 **points; /* map of where we were in the input string
1821 when reading a given char. For ASCII this
1822 is unnecessary overhead as the relationship
1823 is always 1:1, but for Unicode, especially
1824 case folded Unicode this is not true. */
1825 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1829 GET_RE_DEBUG_FLAGS_DECL;
1831 /* We can't just allocate points here. We need to wrap it in
1832 * an SV so it gets freed properly if there is a croak while
1833 * running the match */
1836 sv_points=newSV(maxlen * sizeof(U8 *));
1837 SvCUR_set(sv_points,
1838 maxlen * sizeof(U8 *));
1839 SvPOK_on(sv_points);
1840 sv_2mortal(sv_points);
1841 points=(U8**)SvPV_nolen(sv_points );
1842 if ( trie_type != trie_utf8_fold
1843 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1846 bitmap=(U8*)trie->bitmap;
1848 bitmap=(U8*)ANYOF_BITMAP(c);
1850 /* this is the Aho-Corasick algorithm modified a touch
1851 to include special handling for long "unknown char"
1852 sequences. The basic idea being that we use AC as long
1853 as we are dealing with a possible matching char, when
1854 we encounter an unknown char (and we have not encountered
1855 an accepting state) we scan forward until we find a legal
1857 AC matching is basically that of trie matching, except
1858 that when we encounter a failing transition, we fall back
1859 to the current states "fail state", and try the current char
1860 again, a process we repeat until we reach the root state,
1861 state 1, or a legal transition. If we fail on the root state
1862 then we can either terminate if we have reached an accepting
1863 state previously, or restart the entire process from the beginning
1867 while (s <= last_start) {
1868 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1876 U8 *uscan = (U8*)NULL;
1877 U8 *leftmost = NULL;
1879 U32 accepted_word= 0;
1883 while ( state && uc <= (U8*)strend ) {
1885 U32 word = aho->states[ state ].wordnum;
1889 DEBUG_TRIE_EXECUTE_r(
1890 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1891 dump_exec_pos( (char *)uc, c, strend, real_start,
1892 (char *)uc, utf8_target );
1893 PerlIO_printf( Perl_debug_log,
1894 " Scanning for legal start char...\n");
1898 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1902 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1908 if (uc >(U8*)last_start) break;
1912 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1913 if (!leftmost || lpos < leftmost) {
1914 DEBUG_r(accepted_word=word);
1920 points[pointpos++ % maxlen]= uc;
1921 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1922 uscan, len, uvc, charid, foldlen,
1924 DEBUG_TRIE_EXECUTE_r({
1925 dump_exec_pos( (char *)uc, c, strend, real_start,
1927 PerlIO_printf(Perl_debug_log,
1928 " Charid:%3u CP:%4"UVxf" ",
1934 word = aho->states[ state ].wordnum;
1936 base = aho->states[ state ].trans.base;
1938 DEBUG_TRIE_EXECUTE_r({
1940 dump_exec_pos( (char *)uc, c, strend, real_start,
1942 PerlIO_printf( Perl_debug_log,
1943 "%sState: %4"UVxf", word=%"UVxf,
1944 failed ? " Fail transition to " : "",
1945 (UV)state, (UV)word);
1951 ( ((offset = base + charid
1952 - 1 - trie->uniquecharcount)) >= 0)
1953 && ((U32)offset < trie->lasttrans)
1954 && trie->trans[offset].check == state
1955 && (tmp=trie->trans[offset].next))
1957 DEBUG_TRIE_EXECUTE_r(
1958 PerlIO_printf( Perl_debug_log," - legal\n"));
1963 DEBUG_TRIE_EXECUTE_r(
1964 PerlIO_printf( Perl_debug_log," - fail\n"));
1966 state = aho->fail[state];
1970 /* we must be accepting here */
1971 DEBUG_TRIE_EXECUTE_r(
1972 PerlIO_printf( Perl_debug_log," - accepting\n"));
1981 if (!state) state = 1;
1984 if ( aho->states[ state ].wordnum ) {
1985 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1986 if (!leftmost || lpos < leftmost) {
1987 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1992 s = (char*)leftmost;
1993 DEBUG_TRIE_EXECUTE_r({
1995 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1996 (UV)accepted_word, (IV)(s - real_start)
1999 if (!reginfo || regtry(reginfo, &s)) {
2005 DEBUG_TRIE_EXECUTE_r({
2006 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2009 DEBUG_TRIE_EXECUTE_r(
2010 PerlIO_printf( Perl_debug_log,"No match.\n"));
2019 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2029 - regexec_flags - match a regexp against a string
2032 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2033 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2034 /* strend: pointer to null at end of string */
2035 /* strbeg: real beginning of string */
2036 /* minend: end of match must be >=minend after stringarg. */
2037 /* data: May be used for some additional optimizations.
2038 Currently its only used, with a U32 cast, for transmitting
2039 the ganch offset when doing a /g match. This will change */
2040 /* nosave: For optimizations. */
2043 struct regexp *const prog = (struct regexp *)SvANY(rx);
2044 /*register*/ char *s;
2045 register regnode *c;
2046 /*register*/ char *startpos = stringarg;
2047 I32 minlen; /* must match at least this many chars */
2048 I32 dontbother = 0; /* how many characters not to try at end */
2049 I32 end_shift = 0; /* Same for the end. */ /* CC */
2050 I32 scream_pos = -1; /* Internal iterator of scream. */
2051 char *scream_olds = NULL;
2052 const bool utf8_target = cBOOL(DO_UTF8(sv));
2054 RXi_GET_DECL(prog,progi);
2055 regmatch_info reginfo; /* create some info to pass to regtry etc */
2056 regexp_paren_pair *swap = NULL;
2057 GET_RE_DEBUG_FLAGS_DECL;
2059 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2060 PERL_UNUSED_ARG(data);
2062 /* Be paranoid... */
2063 if (prog == NULL || startpos == NULL) {
2064 Perl_croak(aTHX_ "NULL regexp parameter");
2068 multiline = prog->extflags & RXf_PMf_MULTILINE;
2069 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2071 RX_MATCH_UTF8_set(rx, utf8_target);
2073 debug_start_match(rx, utf8_target, startpos, strend,
2077 minlen = prog->minlen;
2079 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2080 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2081 "String too short [regexec_flags]...\n"));
2086 /* Check validity of program. */
2087 if (UCHARAT(progi->program) != REG_MAGIC) {
2088 Perl_croak(aTHX_ "corrupted regexp program");
2092 PL_reg_eval_set = 0;
2096 PL_reg_flags |= RF_utf8;
2098 /* Mark beginning of line for ^ and lookbehind. */
2099 reginfo.bol = startpos; /* XXX not used ??? */
2103 /* Mark end of line for $ (and such) */
2106 /* see how far we have to get to not match where we matched before */
2107 reginfo.till = startpos+minend;
2109 /* If there is a "must appear" string, look for it. */
2112 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2114 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2115 reginfo.ganch = startpos + prog->gofs;
2116 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2117 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2118 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2120 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2121 && mg->mg_len >= 0) {
2122 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2123 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2124 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2126 if (prog->extflags & RXf_ANCH_GPOS) {
2127 if (s > reginfo.ganch)
2129 s = reginfo.ganch - prog->gofs;
2130 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2131 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2137 reginfo.ganch = strbeg + PTR2UV(data);
2138 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2139 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2141 } else { /* pos() not defined */
2142 reginfo.ganch = strbeg;
2143 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2144 "GPOS: reginfo.ganch = strbeg\n"));
2147 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2148 /* We have to be careful. If the previous successful match
2149 was from this regex we don't want a subsequent partially
2150 successful match to clobber the old results.
2151 So when we detect this possibility we add a swap buffer
2152 to the re, and switch the buffer each match. If we fail
2153 we switch it back, otherwise we leave it swapped.
2156 /* do we need a save destructor here for eval dies? */
2157 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2159 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2160 re_scream_pos_data d;
2162 d.scream_olds = &scream_olds;
2163 d.scream_pos = &scream_pos;
2164 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2166 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2167 goto phooey; /* not present */
2173 /* Simplest case: anchored match need be tried only once. */
2174 /* [unless only anchor is BOL and multiline is set] */
2175 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2176 if (s == startpos && regtry(®info, &startpos))
2178 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2179 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2184 dontbother = minlen - 1;
2185 end = HOP3c(strend, -dontbother, strbeg) - 1;
2186 /* for multiline we only have to try after newlines */
2187 if (prog->check_substr || prog->check_utf8) {
2188 /* because of the goto we can not easily reuse the macros for bifurcating the
2189 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2192 goto after_try_utf8;
2194 if (regtry(®info, &s)) {
2201 if (prog->extflags & RXf_USE_INTUIT) {
2202 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2211 } /* end search for check string in unicode */
2213 if (s == startpos) {
2214 goto after_try_latin;
2217 if (regtry(®info, &s)) {
2224 if (prog->extflags & RXf_USE_INTUIT) {
2225 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2234 } /* end search for check string in latin*/
2235 } /* end search for check string */
2236 else { /* search for newline */
2238 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2241 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2243 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2244 if (regtry(®info, &s))
2248 } /* end search for newline */
2249 } /* end anchored/multiline check string search */
2251 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2253 /* the warning about reginfo.ganch being used without initialization
2254 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2255 and we only enter this block when the same bit is set. */
2256 char *tmp_s = reginfo.ganch - prog->gofs;
2258 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2263 /* Messy cases: unanchored match. */
2264 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2265 /* we have /x+whatever/ */
2266 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2271 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2272 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2273 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2278 DEBUG_EXECUTE_r( did_match = 1 );
2279 if (regtry(®info, &s)) goto got_it;
2281 while (s < strend && *s == ch)
2289 DEBUG_EXECUTE_r( did_match = 1 );
2290 if (regtry(®info, &s)) goto got_it;
2292 while (s < strend && *s == ch)
2297 DEBUG_EXECUTE_r(if (!did_match)
2298 PerlIO_printf(Perl_debug_log,
2299 "Did not find anchored character...\n")
2302 else if (prog->anchored_substr != NULL
2303 || prog->anchored_utf8 != NULL
2304 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2305 && prog->float_max_offset < strend - s)) {
2310 char *last1; /* Last position checked before */
2314 if (prog->anchored_substr || prog->anchored_utf8) {
2315 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2316 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2317 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2318 back_max = back_min = prog->anchored_offset;
2320 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2321 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2322 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2323 back_max = prog->float_max_offset;
2324 back_min = prog->float_min_offset;
2328 if (must == &PL_sv_undef)
2329 /* could not downgrade utf8 check substring, so must fail */
2335 last = HOP3c(strend, /* Cannot start after this */
2336 -(I32)(CHR_SVLEN(must)
2337 - (SvTAIL(must) != 0) + back_min), strbeg);
2340 last1 = HOPc(s, -1);
2342 last1 = s - 1; /* bogus */
2344 /* XXXX check_substr already used to find "s", can optimize if
2345 check_substr==must. */
2347 dontbother = end_shift;
2348 strend = HOPc(strend, -dontbother);
2349 while ( (s <= last) &&
2350 ((flags & REXEC_SCREAM) && SvSCREAM(sv)
2351 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2352 end_shift, &scream_pos, 0))
2353 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2354 (unsigned char*)strend, must,
2355 multiline ? FBMrf_MULTILINE : 0))) ) {
2356 /* we may be pointing at the wrong string */
2357 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2358 s = strbeg + (s - SvPVX_const(sv));
2359 DEBUG_EXECUTE_r( did_match = 1 );
2360 if (HOPc(s, -back_max) > last1) {
2361 last1 = HOPc(s, -back_min);
2362 s = HOPc(s, -back_max);
2365 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2367 last1 = HOPc(s, -back_min);
2371 while (s <= last1) {
2372 if (regtry(®info, &s))
2378 while (s <= last1) {
2379 if (regtry(®info, &s))
2385 DEBUG_EXECUTE_r(if (!did_match) {
2386 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2387 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2388 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2389 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2390 ? "anchored" : "floating"),
2391 quoted, RE_SV_TAIL(must));
2395 else if ( (c = progi->regstclass) ) {
2397 const OPCODE op = OP(progi->regstclass);
2398 /* don't bother with what can't match */
2399 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2400 strend = HOPc(strend, -(minlen - 1));
2403 SV * const prop = sv_newmortal();
2404 regprop(prog, prop, c);
2406 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2408 PerlIO_printf(Perl_debug_log,
2409 "Matching stclass %.*s against %s (%d bytes)\n",
2410 (int)SvCUR(prop), SvPVX_const(prop),
2411 quoted, (int)(strend - s));
2414 if (find_byclass(prog, c, s, strend, ®info))
2416 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2420 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2425 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2426 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2427 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2429 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
2430 last = screaminstr(sv, float_real, s - strbeg,
2431 end_shift, &scream_pos, 1); /* last one */
2433 last = scream_olds; /* Only one occurrence. */
2434 /* we may be pointing at the wrong string */
2435 else if (RXp_MATCH_COPIED(prog))
2436 s = strbeg + (s - SvPVX_const(sv));
2440 const char * const little = SvPV_const(float_real, len);
2442 if (SvTAIL(float_real)) {
2443 if (memEQ(strend - len + 1, little, len - 1))
2444 last = strend - len + 1;
2445 else if (!multiline)
2446 last = memEQ(strend - len, little, len)
2447 ? strend - len : NULL;
2453 last = rninstr(s, strend, little, little + len);
2455 last = strend; /* matching "$" */
2460 PerlIO_printf(Perl_debug_log,
2461 "%sCan't trim the tail, match fails (should not happen)%s\n",
2462 PL_colors[4], PL_colors[5]));
2463 goto phooey; /* Should not happen! */
2465 dontbother = strend - last + prog->float_min_offset;
2467 if (minlen && (dontbother < minlen))
2468 dontbother = minlen - 1;
2469 strend -= dontbother; /* this one's always in bytes! */
2470 /* We don't know much -- general case. */
2473 if (regtry(®info, &s))
2482 if (regtry(®info, &s))
2484 } while (s++ < strend);
2493 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2495 if (PL_reg_eval_set)
2496 restore_pos(aTHX_ prog);
2497 if (RXp_PAREN_NAMES(prog))
2498 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2500 /* make sure $`, $&, $', and $digit will work later */
2501 if ( !(flags & REXEC_NOT_FIRST) ) {
2502 RX_MATCH_COPY_FREE(rx);
2503 if (flags & REXEC_COPY_STR) {
2504 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2505 #ifdef PERL_OLD_COPY_ON_WRITE
2507 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2509 PerlIO_printf(Perl_debug_log,
2510 "Copy on write: regexp capture, type %d\n",
2513 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2514 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2515 assert (SvPOKp(prog->saved_copy));
2519 RX_MATCH_COPIED_on(rx);
2520 s = savepvn(strbeg, i);
2526 prog->subbeg = strbeg;
2527 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2534 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2535 PL_colors[4], PL_colors[5]));
2536 if (PL_reg_eval_set)
2537 restore_pos(aTHX_ prog);
2539 /* we failed :-( roll it back */
2540 Safefree(prog->offs);
2549 - regtry - try match at specific point
2551 STATIC I32 /* 0 failure, 1 success */
2552 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2556 REGEXP *const rx = reginfo->prog;
2557 regexp *const prog = (struct regexp *)SvANY(rx);
2558 RXi_GET_DECL(prog,progi);
2559 GET_RE_DEBUG_FLAGS_DECL;
2561 PERL_ARGS_ASSERT_REGTRY;
2563 reginfo->cutpoint=NULL;
2565 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2568 PL_reg_eval_set = RS_init;
2569 DEBUG_EXECUTE_r(DEBUG_s(
2570 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2571 (IV)(PL_stack_sp - PL_stack_base));
2574 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2575 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2577 /* Apparently this is not needed, judging by wantarray. */
2578 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2579 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2582 /* Make $_ available to executed code. */
2583 if (reginfo->sv != DEFSV) {
2585 DEFSV_set(reginfo->sv);
2588 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2589 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2590 /* prepare for quick setting of pos */
2591 #ifdef PERL_OLD_COPY_ON_WRITE
2592 if (SvIsCOW(reginfo->sv))
2593 sv_force_normal_flags(reginfo->sv, 0);
2595 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2596 &PL_vtbl_mglob, NULL, 0);
2600 PL_reg_oldpos = mg->mg_len;
2601 SAVEDESTRUCTOR_X(restore_pos, prog);
2603 if (!PL_reg_curpm) {
2604 Newxz(PL_reg_curpm, 1, PMOP);
2607 SV* const repointer = &PL_sv_undef;
2608 /* this regexp is also owned by the new PL_reg_curpm, which
2609 will try to free it. */
2610 av_push(PL_regex_padav, repointer);
2611 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2612 PL_regex_pad = AvARRAY(PL_regex_padav);
2617 /* It seems that non-ithreads works both with and without this code.
2618 So for efficiency reasons it seems best not to have the code
2619 compiled when it is not needed. */
2620 /* This is safe against NULLs: */
2621 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2622 /* PM_reg_curpm owns a reference to this regexp. */
2623 (void)ReREFCNT_inc(rx);
2625 PM_SETRE(PL_reg_curpm, rx);
2626 PL_reg_oldcurpm = PL_curpm;
2627 PL_curpm = PL_reg_curpm;
2628 if (RXp_MATCH_COPIED(prog)) {
2629 /* Here is a serious problem: we cannot rewrite subbeg,
2630 since it may be needed if this match fails. Thus
2631 $` inside (?{}) could fail... */
2632 PL_reg_oldsaved = prog->subbeg;
2633 PL_reg_oldsavedlen = prog->sublen;
2634 #ifdef PERL_OLD_COPY_ON_WRITE
2635 PL_nrs = prog->saved_copy;
2637 RXp_MATCH_COPIED_off(prog);
2640 PL_reg_oldsaved = NULL;
2641 prog->subbeg = PL_bostr;
2642 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2644 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2645 prog->offs[0].start = *startpos - PL_bostr;
2646 PL_reginput = *startpos;
2647 PL_reglastparen = &prog->lastparen;
2648 PL_reglastcloseparen = &prog->lastcloseparen;
2649 prog->lastparen = 0;
2650 prog->lastcloseparen = 0;
2652 PL_regoffs = prog->offs;
2653 if (PL_reg_start_tmpl <= prog->nparens) {
2654 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2655 if(PL_reg_start_tmp)
2656 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2658 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2661 /* XXXX What this code is doing here?!!! There should be no need
2662 to do this again and again, PL_reglastparen should take care of
2665 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2666 * Actually, the code in regcppop() (which Ilya may be meaning by
2667 * PL_reglastparen), is not needed at all by the test suite
2668 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2669 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2670 * Meanwhile, this code *is* needed for the
2671 * above-mentioned test suite tests to succeed. The common theme
2672 * on those tests seems to be returning null fields from matches.
2673 * --jhi updated by dapm */
2675 if (prog->nparens) {
2676 regexp_paren_pair *pp = PL_regoffs;
2678 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2686 if (regmatch(reginfo, progi->program + 1)) {
2687 PL_regoffs[0].end = PL_reginput - PL_bostr;
2690 if (reginfo->cutpoint)
2691 *startpos= reginfo->cutpoint;
2692 REGCP_UNWIND(lastcp);
2697 #define sayYES goto yes
2698 #define sayNO goto no
2699 #define sayNO_SILENT goto no_silent
2701 /* we dont use STMT_START/END here because it leads to
2702 "unreachable code" warnings, which are bogus, but distracting. */
2703 #define CACHEsayNO \
2704 if (ST.cache_mask) \
2705 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2708 /* this is used to determine how far from the left messages like
2709 'failed...' are printed. It should be set such that messages
2710 are inline with the regop output that created them.
2712 #define REPORT_CODE_OFF 32
2715 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2716 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2718 #define SLAB_FIRST(s) (&(s)->states[0])
2719 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2721 /* grab a new slab and return the first slot in it */
2723 STATIC regmatch_state *
2726 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2729 regmatch_slab *s = PL_regmatch_slab->next;
2731 Newx(s, 1, regmatch_slab);
2732 s->prev = PL_regmatch_slab;
2734 PL_regmatch_slab->next = s;
2736 PL_regmatch_slab = s;
2737 return SLAB_FIRST(s);
2741 /* push a new state then goto it */
2743 #define PUSH_STATE_GOTO(state, node) \
2745 st->resume_state = state; \
2748 /* push a new state with success backtracking, then goto it */
2750 #define PUSH_YES_STATE_GOTO(state, node) \
2752 st->resume_state = state; \
2753 goto push_yes_state;
2759 regmatch() - main matching routine
2761 This is basically one big switch statement in a loop. We execute an op,
2762 set 'next' to point the next op, and continue. If we come to a point which
2763 we may need to backtrack to on failure such as (A|B|C), we push a
2764 backtrack state onto the backtrack stack. On failure, we pop the top
2765 state, and re-enter the loop at the state indicated. If there are no more
2766 states to pop, we return failure.
2768 Sometimes we also need to backtrack on success; for example /A+/, where
2769 after successfully matching one A, we need to go back and try to
2770 match another one; similarly for lookahead assertions: if the assertion
2771 completes successfully, we backtrack to the state just before the assertion
2772 and then carry on. In these cases, the pushed state is marked as
2773 'backtrack on success too'. This marking is in fact done by a chain of
2774 pointers, each pointing to the previous 'yes' state. On success, we pop to
2775 the nearest yes state, discarding any intermediate failure-only states.
2776 Sometimes a yes state is pushed just to force some cleanup code to be
2777 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2778 it to free the inner regex.
2780 Note that failure backtracking rewinds the cursor position, while
2781 success backtracking leaves it alone.
2783 A pattern is complete when the END op is executed, while a subpattern
2784 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2785 ops trigger the "pop to last yes state if any, otherwise return true"
2788 A common convention in this function is to use A and B to refer to the two
2789 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2790 the subpattern to be matched possibly multiple times, while B is the entire
2791 rest of the pattern. Variable and state names reflect this convention.
2793 The states in the main switch are the union of ops and failure/success of
2794 substates associated with with that op. For example, IFMATCH is the op
2795 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2796 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2797 successfully matched A and IFMATCH_A_fail is a state saying that we have
2798 just failed to match A. Resume states always come in pairs. The backtrack
2799 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2800 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2801 on success or failure.
2803 The struct that holds a backtracking state is actually a big union, with
2804 one variant for each major type of op. The variable st points to the
2805 top-most backtrack struct. To make the code clearer, within each
2806 block of code we #define ST to alias the relevant union.
2808 Here's a concrete example of a (vastly oversimplified) IFMATCH
2814 #define ST st->u.ifmatch
2816 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2817 ST.foo = ...; // some state we wish to save
2819 // push a yes backtrack state with a resume value of
2820 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2822 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2825 case IFMATCH_A: // we have successfully executed A; now continue with B
2827 bar = ST.foo; // do something with the preserved value
2830 case IFMATCH_A_fail: // A failed, so the assertion failed
2831 ...; // do some housekeeping, then ...
2832 sayNO; // propagate the failure
2839 For any old-timers reading this who are familiar with the old recursive
2840 approach, the code above is equivalent to:
2842 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2851 ...; // do some housekeeping, then ...
2852 sayNO; // propagate the failure
2855 The topmost backtrack state, pointed to by st, is usually free. If you
2856 want to claim it, populate any ST.foo fields in it with values you wish to
2857 save, then do one of
2859 PUSH_STATE_GOTO(resume_state, node);
2860 PUSH_YES_STATE_GOTO(resume_state, node);
2862 which sets that backtrack state's resume value to 'resume_state', pushes a
2863 new free entry to the top of the backtrack stack, then goes to 'node'.
2864 On backtracking, the free slot is popped, and the saved state becomes the
2865 new free state. An ST.foo field in this new top state can be temporarily
2866 accessed to retrieve values, but once the main loop is re-entered, it
2867 becomes available for reuse.
2869 Note that the depth of the backtrack stack constantly increases during the
2870 left-to-right execution of the pattern, rather than going up and down with
2871 the pattern nesting. For example the stack is at its maximum at Z at the
2872 end of the pattern, rather than at X in the following:
2874 /(((X)+)+)+....(Y)+....Z/
2876 The only exceptions to this are lookahead/behind assertions and the cut,
2877 (?>A), which pop all the backtrack states associated with A before
2880 Backtrack state structs are allocated in slabs of about 4K in size.
2881 PL_regmatch_state and st always point to the currently active state,
2882 and PL_regmatch_slab points to the slab currently containing
2883 PL_regmatch_state. The first time regmatch() is called, the first slab is
2884 allocated, and is never freed until interpreter destruction. When the slab
2885 is full, a new one is allocated and chained to the end. At exit from
2886 regmatch(), slabs allocated since entry are freed.
2891 #define DEBUG_STATE_pp(pp) \
2893 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2894 PerlIO_printf(Perl_debug_log, \
2895 " %*s"pp" %s%s%s%s%s\n", \
2897 PL_reg_name[st->resume_state], \
2898 ((st==yes_state||st==mark_state) ? "[" : ""), \
2899 ((st==yes_state) ? "Y" : ""), \
2900 ((st==mark_state) ? "M" : ""), \
2901 ((st==yes_state||st==mark_state) ? "]" : "") \
2906 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2911 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2912 const char *start, const char *end, const char *blurb)
2914 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2916 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2921 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2922 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2924 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2925 start, end - start, 60);
2927 PerlIO_printf(Perl_debug_log,
2928 "%s%s REx%s %s against %s\n",
2929 PL_colors[4], blurb, PL_colors[5], s0, s1);
2931 if (utf8_target||utf8_pat)
2932 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2933 utf8_pat ? "pattern" : "",
2934 utf8_pat && utf8_target ? " and " : "",
2935 utf8_target ? "string" : ""
2941 S_dump_exec_pos(pTHX_ const char *locinput,
2942 const regnode *scan,
2943 const char *loc_regeol,
2944 const char *loc_bostr,
2945 const char *loc_reg_starttry,
2946 const bool utf8_target)
2948 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2949 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2950 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2951 /* The part of the string before starttry has one color
2952 (pref0_len chars), between starttry and current
2953 position another one (pref_len - pref0_len chars),
2954 after the current position the third one.
2955 We assume that pref0_len <= pref_len, otherwise we
2956 decrease pref0_len. */
2957 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2958 ? (5 + taill) - l : locinput - loc_bostr;
2961 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2963 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2965 pref0_len = pref_len - (locinput - loc_reg_starttry);
2966 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2967 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2968 ? (5 + taill) - pref_len : loc_regeol - locinput);
2969 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2973 if (pref0_len > pref_len)
2974 pref0_len = pref_len;
2976 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2978 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2979 (locinput - pref_len),pref0_len, 60, 4, 5);
2981 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2982 (locinput - pref_len + pref0_len),
2983 pref_len - pref0_len, 60, 2, 3);
2985 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2986 locinput, loc_regeol - locinput, 10, 0, 1);
2988 const STRLEN tlen=len0+len1+len2;
2989 PerlIO_printf(Perl_debug_log,
2990 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2991 (IV)(locinput - loc_bostr),
2994 (docolor ? "" : "> <"),
2996 (int)(tlen > 19 ? 0 : 19 - tlen),
3003 /* reg_check_named_buff_matched()
3004 * Checks to see if a named buffer has matched. The data array of
3005 * buffer numbers corresponding to the buffer is expected to reside
3006 * in the regexp->data->data array in the slot stored in the ARG() of
3007 * node involved. Note that this routine doesn't actually care about the
3008 * name, that information is not preserved from compilation to execution.
3009 * Returns the index of the leftmost defined buffer with the given name
3010 * or 0 if non of the buffers matched.
3013 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3016 RXi_GET_DECL(rex,rexi);
3017 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3018 I32 *nums=(I32*)SvPVX(sv_dat);
3020 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3022 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3023 if ((I32)*PL_reglastparen >= nums[n] &&
3024 PL_regoffs[nums[n]].end != -1)
3033 /* free all slabs above current one - called during LEAVE_SCOPE */
3036 S_clear_backtrack_stack(pTHX_ void *p)
3038 regmatch_slab *s = PL_regmatch_slab->next;
3043 PL_regmatch_slab->next = NULL;
3045 regmatch_slab * const osl = s;
3052 #define SETREX(Re1,Re2) \
3053 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3056 STATIC I32 /* 0 failure, 1 success */
3057 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3059 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3063 register const bool utf8_target = PL_reg_match_utf8;
3064 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3065 REGEXP *rex_sv = reginfo->prog;
3066 regexp *rex = (struct regexp *)SvANY(rex_sv);
3067 RXi_GET_DECL(rex,rexi);
3069 /* the current state. This is a cached copy of PL_regmatch_state */
3070 register regmatch_state *st;
3071 /* cache heavy used fields of st in registers */
3072 register regnode *scan;
3073 register regnode *next;
3074 register U32 n = 0; /* general value; init to avoid compiler warning */
3075 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3076 register char *locinput = PL_reginput;
3077 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3079 bool result = 0; /* return value of S_regmatch */
3080 int depth = 0; /* depth of backtrack stack */
3081 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3082 const U32 max_nochange_depth =
3083 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3084 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3085 regmatch_state *yes_state = NULL; /* state to pop to on success of
3087 /* mark_state piggy backs on the yes_state logic so that when we unwind
3088 the stack on success we can update the mark_state as we go */
3089 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3090 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3091 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3093 bool no_final = 0; /* prevent failure from backtracking? */
3094 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3095 char *startpoint = PL_reginput;
3096 SV *popmark = NULL; /* are we looking for a mark? */
3097 SV *sv_commit = NULL; /* last mark name seen in failure */
3098 SV *sv_yes_mark = NULL; /* last mark name we have seen
3099 during a successful match */
3100 U32 lastopen = 0; /* last open we saw */
3101 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3102 SV* const oreplsv = GvSV(PL_replgv);
3103 /* these three flags are set by various ops to signal information to
3104 * the very next op. They have a useful lifetime of exactly one loop
3105 * iteration, and are not preserved or restored by state pushes/pops
3107 bool sw = 0; /* the condition value in (?(cond)a|b) */
3108 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3109 int logical = 0; /* the following EVAL is:
3113 or the following IFMATCH/UNLESSM is:
3114 false: plain (?=foo)
3115 true: used as a condition: (?(?=foo))
3118 GET_RE_DEBUG_FLAGS_DECL;
3121 PERL_ARGS_ASSERT_REGMATCH;
3123 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3124 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3126 /* on first ever call to regmatch, allocate first slab */
3127 if (!PL_regmatch_slab) {
3128 Newx(PL_regmatch_slab, 1, regmatch_slab);
3129 PL_regmatch_slab->prev = NULL;
3130 PL_regmatch_slab->next = NULL;
3131 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3134 oldsave = PL_savestack_ix;
3135 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3136 SAVEVPTR(PL_regmatch_slab);
3137 SAVEVPTR(PL_regmatch_state);
3139 /* grab next free state slot */
3140 st = ++PL_regmatch_state;
3141 if (st > SLAB_LAST(PL_regmatch_slab))
3142 st = PL_regmatch_state = S_push_slab(aTHX);
3144 /* Note that nextchr is a byte even in UTF */
3145 nextchr = UCHARAT(locinput);
3147 while (scan != NULL) {
3150 SV * const prop = sv_newmortal();
3151 regnode *rnext=regnext(scan);
3152 DUMP_EXEC_POS( locinput, scan, utf8_target );
3153 regprop(rex, prop, scan);
3155 PerlIO_printf(Perl_debug_log,
3156 "%3"IVdf":%*s%s(%"IVdf")\n",
3157 (IV)(scan - rexi->program), depth*2, "",
3159 (PL_regkind[OP(scan)] == END || !rnext) ?
3160 0 : (IV)(rnext - rexi->program));
3163 next = scan + NEXT_OFF(scan);
3166 state_num = OP(scan);
3170 assert(PL_reglastparen == &rex->lastparen);
3171 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3172 assert(PL_regoffs == rex->offs);
3174 switch (state_num) {
3176 if (locinput == PL_bostr)
3178 /* reginfo->till = reginfo->bol; */
3183 if (locinput == PL_bostr ||
3184 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3190 if (locinput == PL_bostr)
3194 if (locinput == reginfo->ganch)
3199 /* update the startpoint */
3200 st->u.keeper.val = PL_regoffs[0].start;
3201 PL_reginput = locinput;
3202 PL_regoffs[0].start = locinput - PL_bostr;
3203 PUSH_STATE_GOTO(KEEPS_next, next);
3205 case KEEPS_next_fail:
3206 /* rollback the start point change */
3207 PL_regoffs[0].start = st->u.keeper.val;
3213 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3218 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3220 if (PL_regeol - locinput > 1)
3224 if (PL_regeol != locinput)
3228 if (!nextchr && locinput >= PL_regeol)
3231 locinput += PL_utf8skip[nextchr];
3232 if (locinput > PL_regeol)
3234 nextchr = UCHARAT(locinput);
3237 nextchr = UCHARAT(++locinput);
3240 if (!nextchr && locinput >= PL_regeol)
3242 nextchr = UCHARAT(++locinput);
3245 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3248 locinput += PL_utf8skip[nextchr];
3249 if (locinput > PL_regeol)
3251 nextchr = UCHARAT(locinput);
3254 nextchr = UCHARAT(++locinput);
3258 #define ST st->u.trie
3260 /* In this case the charclass data is available inline so
3261 we can fail fast without a lot of extra overhead.
3263 if (scan->flags == EXACT || !utf8_target) {
3264 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3266 PerlIO_printf(Perl_debug_log,
3267 "%*s %sfailed to match trie start class...%s\n",
3268 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3276 /* the basic plan of execution of the trie is:
3277 * At the beginning, run though all the states, and
3278 * find the longest-matching word. Also remember the position
3279 * of the shortest matching word. For example, this pattern:
3282 * when matched against the string "abcde", will generate
3283 * accept states for all words except 3, with the longest
3284 * matching word being 4, and the shortest being 1 (with
3285 * the position being after char 1 of the string).
3287 * Then for each matching word, in word order (i.e. 1,2,4,5),
3288 * we run the remainder of the pattern; on each try setting
3289 * the current position to the character following the word,
3290 * returning to try the next word on failure.
3292 * We avoid having to build a list of words at runtime by
3293 * using a compile-time structure, wordinfo[].prev, which
3294 * gives, for each word, the previous accepting word (if any).
3295 * In the case above it would contain the mappings 1->2, 2->0,
3296 * 3->0, 4->5, 5->1. We can use this table to generate, from
3297 * the longest word (4 above), a list of all words, by
3298 * following the list of prev pointers; this gives us the
3299 * unordered list 4,5,1,2. Then given the current word we have
3300 * just tried, we can go through the list and find the
3301 * next-biggest word to try (so if we just failed on word 2,
3302 * the next in the list is 4).
3304 * Since at runtime we don't record the matching position in
3305 * the string for each word, we have to work that out for
3306 * each word we're about to process. The wordinfo table holds
3307 * the character length of each word; given that we recorded
3308 * at the start: the position of the shortest word and its
3309 * length in chars, we just need to move the pointer the
3310 * difference between the two char lengths. Depending on
3311 * Unicode status and folding, that's cheap or expensive.
3313 * This algorithm is optimised for the case where are only a
3314 * small number of accept states, i.e. 0,1, or maybe 2.
3315 * With lots of accepts states, and having to try all of them,
3316 * it becomes quadratic on number of accept states to find all
3321 /* what type of TRIE am I? (utf8 makes this contextual) */
3322 DECL_TRIE_TYPE(scan);
3324 /* what trie are we using right now */
3325 reg_trie_data * const trie
3326 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3327 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3328 U32 state = trie->startstate;
3330 if (trie->bitmap && trie_type != trie_utf8_fold &&
3331 !TRIE_BITMAP_TEST(trie,*locinput)
3333 if (trie->states[ state ].wordnum) {
3335 PerlIO_printf(Perl_debug_log,
3336 "%*s %smatched empty string...%s\n",
3337 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3343 PerlIO_printf(Perl_debug_log,
3344 "%*s %sfailed to match trie start class...%s\n",
3345 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3352 U8 *uc = ( U8* )locinput;
3356 U8 *uscan = (U8*)NULL;
3357 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3358 U32 charcount = 0; /* how many input chars we have matched */
3359 U32 accepted = 0; /* have we seen any accepting states? */
3362 ST.jump = trie->jump;
3365 ST.longfold = FALSE; /* char longer if folded => it's harder */
3368 /* fully traverse the TRIE; note the position of the
3369 shortest accept state and the wordnum of the longest
3372 while ( state && uc <= (U8*)PL_regeol ) {
3373 U32 base = trie->states[ state ].trans.base;
3377 wordnum = trie->states[ state ].wordnum;
3379 if (wordnum) { /* it's an accept state */
3382 /* record first match position */
3384 ST.firstpos = (U8*)locinput;
3389 ST.firstchars = charcount;
3392 if (!ST.nextword || wordnum < ST.nextword)
3393 ST.nextword = wordnum;
3394 ST.topword = wordnum;
3397 DEBUG_TRIE_EXECUTE_r({
3398 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3399 PerlIO_printf( Perl_debug_log,
3400 "%*s %sState: %4"UVxf" Accepted: %c ",
3401 2+depth * 2, "", PL_colors[4],
3402 (UV)state, (accepted ? 'Y' : 'N'));
3405 /* read a char and goto next state */
3408 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3409 uscan, len, uvc, charid, foldlen,
3416 base + charid - 1 - trie->uniquecharcount)) >= 0)
3418 && ((U32)offset < trie->lasttrans)
3419 && trie->trans[offset].check == state)
3421 state = trie->trans[offset].next;
3432 DEBUG_TRIE_EXECUTE_r(
3433 PerlIO_printf( Perl_debug_log,
3434 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3435 charid, uvc, (UV)state, PL_colors[5] );
3441 /* calculate total number of accept states */
3446 w = trie->wordinfo[w].prev;
3449 ST.accepted = accepted;
3453 PerlIO_printf( Perl_debug_log,
3454 "%*s %sgot %"IVdf" possible matches%s\n",
3455 REPORT_CODE_OFF + depth * 2, "",
3456 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3458 goto trie_first_try; /* jump into the fail handler */
3462 case TRIE_next_fail: /* we failed - try next alternative */
3464 REGCP_UNWIND(ST.cp);
3465 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3466 PL_regoffs[n].end = -1;
3467 *PL_reglastparen = n;
3469 if (!--ST.accepted) {
3471 PerlIO_printf( Perl_debug_log,
3472 "%*s %sTRIE failed...%s\n",
3473 REPORT_CODE_OFF+depth*2, "",
3480 /* Find next-highest word to process. Note that this code
3481 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3482 register U16 min = 0;
3484 register U16 const nextword = ST.nextword;
3485 register reg_trie_wordinfo * const wordinfo
3486 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3487 for (word=ST.topword; word; word=wordinfo[word].prev) {
3488 if (word > nextword && (!min || word < min))
3501 ST.lastparen = *PL_reglastparen;
3505 /* find start char of end of current word */
3507 U32 chars; /* how many chars to skip */
3508 U8 *uc = ST.firstpos;
3509 reg_trie_data * const trie
3510 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3512 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3514 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3518 /* the hard option - fold each char in turn and find
3519 * its folded length (which may be different */
3520 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3528 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3536 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3541 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3555 PL_reginput = (char *)uc;
3558 scan = (ST.jump && ST.jump[ST.nextword])
3559 ? ST.me + ST.jump[ST.nextword]
3563 PerlIO_printf( Perl_debug_log,
3564 "%*s %sTRIE matched word #%d, continuing%s\n",
3565 REPORT_CODE_OFF+depth*2, "",
3572 if (ST.accepted > 1 || has_cutgroup) {
3573 PUSH_STATE_GOTO(TRIE_next, scan);
3576 /* only one choice left - just continue */
3578 AV *const trie_words
3579 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3580 SV ** const tmp = av_fetch( trie_words,
3582 SV *sv= tmp ? sv_newmortal() : NULL;
3584 PerlIO_printf( Perl_debug_log,
3585 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3586 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3588 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3589 PL_colors[0], PL_colors[1],
3590 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3592 : "not compiled under -Dr",
3596 locinput = PL_reginput;
3597 nextchr = UCHARAT(locinput);
3598 continue; /* execute rest of RE */
3603 char *s = STRING(scan);
3605 if (utf8_target != UTF_PATTERN) {
3606 /* The target and the pattern have differing utf8ness. */
3608 const char * const e = s + ln;
3611 /* The target is utf8, the pattern is not utf8. */
3616 if (NATIVE_TO_UNI(*(U8*)s) !=
3617 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3625 /* The target is not utf8, the pattern is utf8. */
3630 if (NATIVE_TO_UNI(*((U8*)l)) !=
3631 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3639 nextchr = UCHARAT(locinput);
3642 /* The target and the pattern have the same utf8ness. */
3643 /* Inline the first character, for speed. */
3644 if (UCHARAT(s) != nextchr)
3646 if (PL_regeol - locinput < ln)
3648 if (ln > 1 && memNE(s, locinput, ln))
3651 nextchr = UCHARAT(locinput);
3656 const U8 * fold_array;
3658 U32 fold_utf8_flags;
3660 PL_reg_flags |= RF_tainted;
3661 folder = foldEQ_locale;
3662 fold_array = PL_fold_locale;
3663 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3667 folder = foldEQ_latin1;
3668 fold_array = PL_fold_latin1;
3669 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3673 folder = foldEQ_latin1;
3674 fold_array = PL_fold_latin1;
3675 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3680 fold_array = PL_fold;
3681 fold_utf8_flags = 0;
3687 if (utf8_target || UTF_PATTERN) {
3688 /* Either target or the pattern are utf8. */
3689 const char * const l = locinput;
3690 char *e = PL_regeol;
3692 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3693 l, &e, 0, utf8_target, fold_utf8_flags))
3698 nextchr = UCHARAT(locinput);
3702 /* Neither the target nor the pattern are utf8 */
3703 if (UCHARAT(s) != nextchr &&
3704 UCHARAT(s) != fold_array[nextchr])
3708 if (PL_regeol - locinput < ln)
3710 if (ln > 1 && ! folder(s, locinput, ln))
3713 nextchr = UCHARAT(locinput);
3717 /* XXX Could improve efficiency by separating these all out using a
3718 * macro or in-line function. At that point regcomp.c would no longer
3719 * have to set the FLAGS fields of these */
3722 PL_reg_flags |= RF_tainted;
3730 /* was last char in word? */
3732 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3733 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3735 if (locinput == PL_bostr)
3738 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3740 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3742 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3743 ln = isALNUM_uni(ln);
3744 LOAD_UTF8_CHARCLASS_ALNUM();
3745 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3748 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3749 n = isALNUM_LC_utf8((U8*)locinput);
3754 /* Here the string isn't utf8, or is utf8 and only ascii
3755 * characters are to match \w. In the latter case looking at
3756 * the byte just prior to the current one may be just the final
3757 * byte of a multi-byte character. This is ok. There are two
3759 * 1) it is a single byte character, and then the test is doing
3760 * just what it's supposed to.
3761 * 2) it is a multi-byte character, in which case the final
3762 * byte is never mistakable for ASCII, and so the test
3763 * will say it is not a word character, which is the
3764 * correct answer. */
3765 ln = (locinput != PL_bostr) ?
3766 UCHARAT(locinput - 1) : '\n';
3767 switch (FLAGS(scan)) {
3768 case REGEX_UNICODE_CHARSET:
3769 ln = isWORDCHAR_L1(ln);
3770 n = isWORDCHAR_L1(nextchr);
3772 case REGEX_LOCALE_CHARSET:
3773 ln = isALNUM_LC(ln);
3774 n = isALNUM_LC(nextchr);
3776 case REGEX_DEPENDS_CHARSET:
3778 n = isALNUM(nextchr);
3780 case REGEX_ASCII_RESTRICTED_CHARSET:
3781 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3782 ln = isWORDCHAR_A(ln);
3783 n = isWORDCHAR_A(nextchr);
3786 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3790 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3792 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3797 if (utf8_target || state_num == ANYOFV) {
3798 STRLEN inclasslen = PL_regeol - locinput;
3799 if (locinput >= PL_regeol)
3802 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3804 locinput += inclasslen;
3805 nextchr = UCHARAT(locinput);
3810 nextchr = UCHARAT(locinput);
3811 if (!nextchr && locinput >= PL_regeol)
3813 if (!REGINCLASS(rex, scan, (U8*)locinput))
3815 nextchr = UCHARAT(++locinput);
3819 /* Special char classes - The defines start on line 129 or so */
3820 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3821 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3822 ALNUMU, NALNUMU, isWORDCHAR_L1,
3823 ALNUMA, NALNUMA, isWORDCHAR_A,
3826 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3827 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3828 SPACEU, NSPACEU, isSPACE_L1,
3829 SPACEA, NSPACEA, isSPACE_A,
3832 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3833 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3834 DIGITA, NDIGITA, isDIGIT_A,
3837 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3838 a Unicode extended Grapheme Cluster */
3839 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3840 extended Grapheme Cluster is:
3843 | Prepend* Begin Extend*
3846 Begin is (Hangul-syllable | ! Control)
3847 Extend is (Grapheme_Extend | Spacing_Mark)
3848 Control is [ GCB_Control CR LF ]
3850 The discussion below shows how the code for CLUMP is derived
3851 from this regex. Note that most of these concepts are from
3852 property values of the Grapheme Cluster Boundary (GCB) property.
3853 No code point can have multiple property values for a given
3854 property. Thus a code point in Prepend can't be in Control, but
3855 it must be in !Control. This is why Control above includes
3856 GCB_Control plus CR plus LF. The latter two are used in the GCB
3857 property separately, and so can't be in GCB_Control, even though
3858 they logically are controls. Control is not the same as gc=cc,
3859 but includes format and other characters as well.
3861 The Unicode definition of Hangul-syllable is:
3863 | (L* ( ( V | LV ) V* | LVT ) T*)
3866 Each of these is a value for the GCB property, and hence must be
3867 disjoint, so the order they are tested is immaterial, so the
3868 above can safely be changed to
3871 | (L* ( LVT | ( V | LV ) V*) T*)
3873 The last two terms can be combined like this:
3875 | (( LVT | ( V | LV ) V*) T*))
3877 And refactored into this:
3878 L* (L | LVT T* | V V* T* | LV V* T*)
3880 That means that if we have seen any L's at all we can quit
3881 there, but if the next character is an LVT, a V, or an LV we
3884 There is a subtlety with Prepend* which showed up in testing.
3885 Note that the Begin, and only the Begin is required in:
3886 | Prepend* Begin Extend*
3887 Also, Begin contains '! Control'. A Prepend must be a
3888 '! Control', which means it must also be a Begin. What it
3889 comes down to is that if we match Prepend* and then find no
3890 suitable Begin afterwards, that if we backtrack the last
3891 Prepend, that one will be a suitable Begin.
3894 if (locinput >= PL_regeol)
3896 if (! utf8_target) {
3898 /* Match either CR LF or '.', as all the other possibilities
3900 locinput++; /* Match the . or CR */
3901 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3903 && locinput < PL_regeol
3904 && UCHARAT(locinput) == '\n') locinput++;
3908 /* Utf8: See if is ( CR LF ); already know that locinput <
3909 * PL_regeol, so locinput+1 is in bounds */
3910 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3914 /* In case have to backtrack to beginning, then match '.' */
3915 char *starting = locinput;
3917 /* In case have to backtrack the last prepend */
3918 char *previous_prepend = 0;
3920 LOAD_UTF8_CHARCLASS_GCB();
3922 /* Match (prepend)* */
3923 while (locinput < PL_regeol
3924 && swash_fetch(PL_utf8_X_prepend,
3925 (U8*)locinput, utf8_target))
3927 previous_prepend = locinput;
3928 locinput += UTF8SKIP(locinput);
3931 /* As noted above, if we matched a prepend character, but
3932 * the next thing won't match, back off the last prepend we
3933 * matched, as it is guaranteed to match the begin */
3934 if (previous_prepend
3935 && (locinput >= PL_regeol
3936 || ! swash_fetch(PL_utf8_X_begin,
3937 (U8*)locinput, utf8_target)))
3939 locinput = previous_prepend;
3942 /* Note that here we know PL_regeol > locinput, as we
3943 * tested that upon input to this switch case, and if we
3944 * moved locinput forward, we tested the result just above
3945 * and it either passed, or we backed off so that it will
3947 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3949 /* Here did not match the required 'Begin' in the
3950 * second term. So just match the very first
3951 * character, the '.' of the final term of the regex */
3952 locinput = starting + UTF8SKIP(starting);
3955 /* Here is the beginning of a character that can have
3956 * an extender. It is either a hangul syllable, or a
3958 if (swash_fetch(PL_utf8_X_non_hangul,
3959 (U8*)locinput, utf8_target))
3962 /* Here not a Hangul syllable, must be a
3963 * ('! * Control') */
3964 locinput += UTF8SKIP(locinput);
3967 /* Here is a Hangul syllable. It can be composed
3968 * of several individual characters. One
3969 * possibility is T+ */
3970 if (swash_fetch(PL_utf8_X_T,
3971 (U8*)locinput, utf8_target))
3973 while (locinput < PL_regeol
3974 && swash_fetch(PL_utf8_X_T,
3975 (U8*)locinput, utf8_target))
3977 locinput += UTF8SKIP(locinput);
3981 /* Here, not T+, but is a Hangul. That means
3982 * it is one of the others: L, LV, LVT or V,
3984 * L* (L | LVT T* | V V* T* | LV V* T*) */
3987 while (locinput < PL_regeol
3988 && swash_fetch(PL_utf8_X_L,
3989 (U8*)locinput, utf8_target))
3991 locinput += UTF8SKIP(locinput);
3994 /* Here, have exhausted L*. If the next
3995 * character is not an LV, LVT nor V, it means
3996 * we had to have at least one L, so matches L+
3997 * in the original equation, we have a complete
3998 * hangul syllable. Are done. */
4000 if (locinput < PL_regeol
4001 && swash_fetch(PL_utf8_X_LV_LVT_V,
4002 (U8*)locinput, utf8_target))
4005 /* Otherwise keep going. Must be LV, LVT
4006 * or V. See if LVT */
4007 if (swash_fetch(PL_utf8_X_LVT,
4008 (U8*)locinput, utf8_target))
4010 locinput += UTF8SKIP(locinput);
4013 /* Must be V or LV. Take it, then
4015 locinput += UTF8SKIP(locinput);
4016 while (locinput < PL_regeol
4017 && swash_fetch(PL_utf8_X_V,
4018 (U8*)locinput, utf8_target))
4020 locinput += UTF8SKIP(locinput);
4024 /* And any of LV, LVT, or V can be followed
4026 while (locinput < PL_regeol
4027 && swash_fetch(PL_utf8_X_T,
4031 locinput += UTF8SKIP(locinput);
4037 /* Match any extender */
4038 while (locinput < PL_regeol
4039 && swash_fetch(PL_utf8_X_extend,
4040 (U8*)locinput, utf8_target))
4042 locinput += UTF8SKIP(locinput);
4046 if (locinput > PL_regeol) sayNO;
4048 nextchr = UCHARAT(locinput);
4052 { /* The capture buffer cases. The ones beginning with N for the
4053 named buffers just convert to the equivalent numbered and
4054 pretend they were called as the corresponding numbered buffer
4056 /* don't initialize these in the declaration, it makes C++
4061 const U8 *fold_array;
4064 PL_reg_flags |= RF_tainted;
4065 folder = foldEQ_locale;
4066 fold_array = PL_fold_locale;
4068 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4072 folder = foldEQ_latin1;
4073 fold_array = PL_fold_latin1;
4075 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4079 folder = foldEQ_latin1;
4080 fold_array = PL_fold_latin1;
4082 utf8_fold_flags = 0;
4087 fold_array = PL_fold;
4089 utf8_fold_flags = 0;
4096 utf8_fold_flags = 0;
4099 /* For the named back references, find the corresponding buffer
4101 n = reg_check_named_buff_matched(rex,scan);
4106 goto do_nref_ref_common;
4109 PL_reg_flags |= RF_tainted;
4110 folder = foldEQ_locale;
4111 fold_array = PL_fold_locale;
4112 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4116 folder = foldEQ_latin1;
4117 fold_array = PL_fold_latin1;
4118 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4122 folder = foldEQ_latin1;
4123 fold_array = PL_fold_latin1;
4124 utf8_fold_flags = 0;
4129 fold_array = PL_fold;
4130 utf8_fold_flags = 0;
4136 utf8_fold_flags = 0;
4140 n = ARG(scan); /* which paren pair */
4143 ln = PL_regoffs[n].start;
4144 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4145 if (*PL_reglastparen < n || ln == -1)
4146 sayNO; /* Do not match unless seen CLOSEn. */
4147 if (ln == PL_regoffs[n].end)
4151 if (type != REF /* REF can do byte comparison */
4152 && (utf8_target || type == REFFU))
4153 { /* XXX handle REFFL better */
4154 char * limit = PL_regeol;
4156 /* This call case insensitively compares the entire buffer
4157 * at s, with the current input starting at locinput, but
4158 * not going off the end given by PL_regeol, and returns in
4159 * limit upon success, how much of the current input was
4161 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4162 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4167 nextchr = UCHARAT(locinput);
4171 /* Not utf8: Inline the first character, for speed. */
4172 if (UCHARAT(s) != nextchr &&
4174 UCHARAT(s) != fold_array[nextchr]))
4176 ln = PL_regoffs[n].end - ln;
4177 if (locinput + ln > PL_regeol)
4179 if (ln > 1 && (type == REF
4180 ? memNE(s, locinput, ln)
4181 : ! folder(s, locinput, ln)))
4184 nextchr = UCHARAT(locinput);
4194 #define ST st->u.eval
4199 regexp_internal *rei;
4200 regnode *startpoint;
4203 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4204 if (cur_eval && cur_eval->locinput==locinput) {
4205 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4206 Perl_croak(aTHX_ "Infinite recursion in regex");
4207 if ( ++nochange_depth > max_nochange_depth )
4209 "Pattern subroutine nesting without pos change"
4210 " exceeded limit in regex");
4217 (void)ReREFCNT_inc(rex_sv);
4218 if (OP(scan)==GOSUB) {
4219 startpoint = scan + ARG2L(scan);
4220 ST.close_paren = ARG(scan);
4222 startpoint = rei->program+1;
4225 goto eval_recurse_doit;
4227 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4228 if (cur_eval && cur_eval->locinput==locinput) {
4229 if ( ++nochange_depth > max_nochange_depth )
4230 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4235 /* execute the code in the {...} */
4237 SV ** const before = SP;
4238 OP_4tree * const oop = PL_op;
4239 COP * const ocurcop = PL_curcop;
4241 char *saved_regeol = PL_regeol;
4242 struct re_save_state saved_state;
4244 /* To not corrupt the existing regex state while executing the
4245 * eval we would normally put it on the save stack, like with
4246 * save_re_context. However, re-evals have a weird scoping so we
4247 * can't just add ENTER/LEAVE here. With that, things like
4249 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4251 * would break, as they expect the localisation to be unwound
4252 * only when the re-engine backtracks through the bit that
4255 * What we do instead is just saving the state in a local c
4258 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4261 PL_op = (OP_4tree*)rexi->data->data[n];
4262 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4263 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4264 /* wrap the call in two SAVECOMPPADs. This ensures that
4265 * when the save stack is eventually unwound, all the
4266 * accumulated SAVEt_CLEARSV's will be processed with
4267 * interspersed SAVEt_COMPPAD's to ensure that lexicals
4268 * are cleared in the right pad */
4270 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4271 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4274 SV *sv_mrk = get_sv("REGMARK", 1);
4275 sv_setsv(sv_mrk, sv_yes_mark);
4278 CALLRUNOPS(aTHX); /* Scalar context. */
4281 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4287 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4291 PAD_RESTORE_LOCAL(old_comppad);
4292 PL_curcop = ocurcop;
4293 PL_regeol = saved_regeol;
4296 sv_setsv(save_scalar(PL_replgv), ret);
4300 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4303 /* extract RE object from returned value; compiling if
4309 SV *const sv = SvRV(ret);
4311 if (SvTYPE(sv) == SVt_REGEXP) {
4313 } else if (SvSMAGICAL(sv)) {
4314 mg = mg_find(sv, PERL_MAGIC_qr);
4317 } else if (SvTYPE(ret) == SVt_REGEXP) {
4319 } else if (SvSMAGICAL(ret)) {
4320 if (SvGMAGICAL(ret)) {
4321 /* I don't believe that there is ever qr magic
4323 assert(!mg_find(ret, PERL_MAGIC_qr));
4324 sv_unmagic(ret, PERL_MAGIC_qr);
4327 mg = mg_find(ret, PERL_MAGIC_qr);
4328 /* testing suggests mg only ends up non-NULL for
4329 scalars who were upgraded and compiled in the
4330 else block below. In turn, this is only
4331 triggered in the "postponed utf8 string" tests
4337 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4341 rx = reg_temp_copy(NULL, rx);
4345 const I32 osize = PL_regsize;
4348 assert (SvUTF8(ret));
4349 } else if (SvUTF8(ret)) {
4350 /* Not doing UTF-8, despite what the SV says. Is
4351 this only if we're trapped in use 'bytes'? */
4352 /* Make a copy of the octet sequence, but without
4353 the flag on, as the compiler now honours the
4354 SvUTF8 flag on ret. */
4356 const char *const p = SvPV(ret, len);
4357 ret = newSVpvn_flags(p, len, SVs_TEMP);
4359 rx = CALLREGCOMP(ret, pm_flags);
4361 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4363 /* This isn't a first class regexp. Instead, it's
4364 caching a regexp onto an existing, Perl visible
4366 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4371 re = (struct regexp *)SvANY(rx);
4373 RXp_MATCH_COPIED_off(re);
4374 re->subbeg = rex->subbeg;
4375 re->sublen = rex->sublen;
4378 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4379 "Matching embedded");
4381 startpoint = rei->program + 1;
4382 ST.close_paren = 0; /* only used for GOSUB */
4383 /* borrowed from regtry */
4384 if (PL_reg_start_tmpl <= re->nparens) {
4385 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4386 if(PL_reg_start_tmp)
4387 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4389 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4392 eval_recurse_doit: /* Share code with GOSUB below this line */
4393 /* run the pattern returned from (??{...}) */
4394 ST.cp = regcppush(0); /* Save *all* the positions. */
4395 REGCP_SET(ST.lastcp);
4397 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4399 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4400 PL_reglastparen = &re->lastparen;
4401 PL_reglastcloseparen = &re->lastcloseparen;
4403 re->lastcloseparen = 0;
4405 PL_reginput = locinput;
4408 /* XXXX This is too dramatic a measure... */
4411 ST.toggle_reg_flags = PL_reg_flags;
4413 PL_reg_flags |= RF_utf8;
4415 PL_reg_flags &= ~RF_utf8;
4416 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4418 ST.prev_rex = rex_sv;
4419 ST.prev_curlyx = cur_curlyx;
4420 SETREX(rex_sv,re_sv);
4425 ST.prev_eval = cur_eval;
4427 /* now continue from first node in postoned RE */
4428 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4431 /* logical is 1, /(?(?{...})X|Y)/ */
4432 sw = cBOOL(SvTRUE(ret));
4437 case EVAL_AB: /* cleanup after a successful (??{A})B */
4438 /* note: this is called twice; first after popping B, then A */
4439 PL_reg_flags ^= ST.toggle_reg_flags;
4440 ReREFCNT_dec(rex_sv);
4441 SETREX(rex_sv,ST.prev_rex);
4442 rex = (struct regexp *)SvANY(rex_sv);
4443 rexi = RXi_GET(rex);
4445 cur_eval = ST.prev_eval;
4446 cur_curlyx = ST.prev_curlyx;
4448 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4449 PL_reglastparen = &rex->lastparen;
4450 PL_reglastcloseparen = &rex->lastcloseparen;
4451 /* also update PL_regoffs */
4452 PL_regoffs = rex->offs;
4454 /* XXXX This is too dramatic a measure... */
4456 if ( nochange_depth )
4461 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4462 /* note: this is called twice; first after popping B, then A */
4463 PL_reg_flags ^= ST.toggle_reg_flags;
4464 ReREFCNT_dec(rex_sv);
4465 SETREX(rex_sv,ST.prev_rex);
4466 rex = (struct regexp *)SvANY(rex_sv);
4467 rexi = RXi_GET(rex);
4468 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4469 PL_reglastparen = &rex->lastparen;
4470 PL_reglastcloseparen = &rex->lastcloseparen;
4472 PL_reginput = locinput;
4473 REGCP_UNWIND(ST.lastcp);
4475 cur_eval = ST.prev_eval;
4476 cur_curlyx = ST.prev_curlyx;
4477 /* XXXX This is too dramatic a measure... */
4479 if ( nochange_depth )
4485 n = ARG(scan); /* which paren pair */
4486 PL_reg_start_tmp[n] = locinput;
4492 n = ARG(scan); /* which paren pair */
4493 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4494 PL_regoffs[n].end = locinput - PL_bostr;
4495 /*if (n > PL_regsize)
4497 if (n > *PL_reglastparen)
4498 *PL_reglastparen = n;
4499 *PL_reglastcloseparen = n;
4500 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4508 cursor && OP(cursor)!=END;
4509 cursor=regnext(cursor))
4511 if ( OP(cursor)==CLOSE ){
4513 if ( n <= lastopen ) {
4515 = PL_reg_start_tmp[n] - PL_bostr;
4516 PL_regoffs[n].end = locinput - PL_bostr;
4517 /*if (n > PL_regsize)
4519 if (n > *PL_reglastparen)
4520 *PL_reglastparen = n;
4521 *PL_reglastcloseparen = n;
4522 if ( n == ARG(scan) || (cur_eval &&
4523 cur_eval->u.eval.close_paren == n))
4532 n = ARG(scan); /* which paren pair */
4533 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4536 /* reg_check_named_buff_matched returns 0 for no match */
4537 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4541 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4547 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4549 next = NEXTOPER(NEXTOPER(scan));
4551 next = scan + ARG(scan);
4552 if (OP(next) == IFTHEN) /* Fake one. */
4553 next = NEXTOPER(NEXTOPER(next));
4557 logical = scan->flags;
4560 /*******************************************************************
4562 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4563 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4564 STAR/PLUS/CURLY/CURLYN are used instead.)
4566 A*B is compiled as <CURLYX><A><WHILEM><B>
4568 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4569 state, which contains the current count, initialised to -1. It also sets
4570 cur_curlyx to point to this state, with any previous value saved in the
4573 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4574 since the pattern may possibly match zero times (i.e. it's a while {} loop
4575 rather than a do {} while loop).
4577 Each entry to WHILEM represents a successful match of A. The count in the
4578 CURLYX block is incremented, another WHILEM state is pushed, and execution
4579 passes to A or B depending on greediness and the current count.
4581 For example, if matching against the string a1a2a3b (where the aN are
4582 substrings that match /A/), then the match progresses as follows: (the
4583 pushed states are interspersed with the bits of strings matched so far):
4586 <CURLYX cnt=0><WHILEM>
4587 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4588 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4589 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4590 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4592 (Contrast this with something like CURLYM, which maintains only a single
4596 a1 <CURLYM cnt=1> a2
4597 a1 a2 <CURLYM cnt=2> a3
4598 a1 a2 a3 <CURLYM cnt=3> b
4601 Each WHILEM state block marks a point to backtrack to upon partial failure
4602 of A or B, and also contains some minor state data related to that
4603 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4604 overall state, such as the count, and pointers to the A and B ops.
4606 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4607 must always point to the *current* CURLYX block, the rules are:
4609 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4610 and set cur_curlyx to point the new block.
4612 When popping the CURLYX block after a successful or unsuccessful match,
4613 restore the previous cur_curlyx.
4615 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4616 to the outer one saved in the CURLYX block.
4618 When popping the WHILEM block after a successful or unsuccessful B match,
4619 restore the previous cur_curlyx.
4621 Here's an example for the pattern (AI* BI)*BO
4622 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4625 curlyx backtrack stack
4626 ------ ---------------
4628 CO <CO prev=NULL> <WO>
4629 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4630 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4631 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4633 At this point the pattern succeeds, and we work back down the stack to
4634 clean up, restoring as we go:
4636 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4637 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4638 CO <CO prev=NULL> <WO>
4641 *******************************************************************/
4643 #define ST st->u.curlyx
4645 case CURLYX: /* start of /A*B/ (for complex A) */
4647 /* No need to save/restore up to this paren */
4648 I32 parenfloor = scan->flags;
4650 assert(next); /* keep Coverity happy */
4651 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4654 /* XXXX Probably it is better to teach regpush to support
4655 parenfloor > PL_regsize... */
4656 if (parenfloor > (I32)*PL_reglastparen)
4657 parenfloor = *PL_reglastparen; /* Pessimization... */
4659 ST.prev_curlyx= cur_curlyx;
4661 ST.cp = PL_savestack_ix;
4663 /* these fields contain the state of the current curly.
4664 * they are accessed by subsequent WHILEMs */
4665 ST.parenfloor = parenfloor;
4670 ST.count = -1; /* this will be updated by WHILEM */
4671 ST.lastloc = NULL; /* this will be updated by WHILEM */
4673 PL_reginput = locinput;
4674 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4678 case CURLYX_end: /* just finished matching all of A*B */
4679 cur_curlyx = ST.prev_curlyx;
4683 case CURLYX_end_fail: /* just failed to match all of A*B */
4685 cur_curlyx = ST.prev_curlyx;
4691 #define ST st->u.whilem
4693 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4695 /* see the discussion above about CURLYX/WHILEM */
4697 int min = ARG1(cur_curlyx->u.curlyx.me);
4698 int max = ARG2(cur_curlyx->u.curlyx.me);
4699 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4701 assert(cur_curlyx); /* keep Coverity happy */
4702 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4703 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4704 ST.cache_offset = 0;
4707 PL_reginput = locinput;
4709 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4710 "%*s whilem: matched %ld out of %d..%d\n",
4711 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4714 /* First just match a string of min A's. */
4717 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4718 cur_curlyx->u.curlyx.lastloc = locinput;
4719 REGCP_SET(ST.lastcp);
4721 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4725 /* If degenerate A matches "", assume A done. */
4727 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4728 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4729 "%*s whilem: empty match detected, trying continuation...\n",
4730 REPORT_CODE_OFF+depth*2, "")
4732 goto do_whilem_B_max;
4735 /* super-linear cache processing */
4739 if (!PL_reg_maxiter) {
4740 /* start the countdown: Postpone detection until we
4741 * know the match is not *that* much linear. */
4742 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4743 /* possible overflow for long strings and many CURLYX's */
4744 if (PL_reg_maxiter < 0)
4745 PL_reg_maxiter = I32_MAX;
4746 PL_reg_leftiter = PL_reg_maxiter;
4749 if (PL_reg_leftiter-- == 0) {
4750 /* initialise cache */
4751 const I32 size = (PL_reg_maxiter + 7)/8;
4752 if (PL_reg_poscache) {
4753 if ((I32)PL_reg_poscache_size < size) {
4754 Renew(PL_reg_poscache, size, char);
4755 PL_reg_poscache_size = size;
4757 Zero(PL_reg_poscache, size, char);
4760 PL_reg_poscache_size = size;
4761 Newxz(PL_reg_poscache, size, char);
4763 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4764 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4765 PL_colors[4], PL_colors[5])
4769 if (PL_reg_leftiter < 0) {
4770 /* have we already failed at this position? */
4772 offset = (scan->flags & 0xf) - 1
4773 + (locinput - PL_bostr) * (scan->flags>>4);
4774 mask = 1 << (offset % 8);
4776 if (PL_reg_poscache[offset] & mask) {
4777 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4778 "%*s whilem: (cache) already tried at this position...\n",
4779 REPORT_CODE_OFF+depth*2, "")
4781 sayNO; /* cache records failure */
4783 ST.cache_offset = offset;
4784 ST.cache_mask = mask;
4788 /* Prefer B over A for minimal matching. */
4790 if (cur_curlyx->u.curlyx.minmod) {
4791 ST.save_curlyx = cur_curlyx;
4792 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4793 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4794 REGCP_SET(ST.lastcp);
4795 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4799 /* Prefer A over B for maximal matching. */
4801 if (n < max) { /* More greed allowed? */
4802 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4803 cur_curlyx->u.curlyx.lastloc = locinput;
4804 REGCP_SET(ST.lastcp);
4805 PUSH_STATE_GOTO(WHILEM_A_max, A);
4808 goto do_whilem_B_max;
4812 case WHILEM_B_min: /* just matched B in a minimal match */
4813 case WHILEM_B_max: /* just matched B in a maximal match */
4814 cur_curlyx = ST.save_curlyx;
4818 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4819 cur_curlyx = ST.save_curlyx;
4820 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4821 cur_curlyx->u.curlyx.count--;
4825 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4827 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4828 REGCP_UNWIND(ST.lastcp);
4830 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4831 cur_curlyx->u.curlyx.count--;
4835 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4836 REGCP_UNWIND(ST.lastcp);
4837 regcppop(rex); /* Restore some previous $<digit>s? */
4838 PL_reginput = locinput;
4839 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4840 "%*s whilem: failed, trying continuation...\n",
4841 REPORT_CODE_OFF+depth*2, "")
4844 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4845 && ckWARN(WARN_REGEXP)
4846 && !(PL_reg_flags & RF_warned))
4848 PL_reg_flags |= RF_warned;
4849 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4850 "Complex regular subexpression recursion limit (%d) "
4856 ST.save_curlyx = cur_curlyx;
4857 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4858 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4861 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4862 cur_curlyx = ST.save_curlyx;
4863 REGCP_UNWIND(ST.lastcp);
4866 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4867 /* Maximum greed exceeded */
4868 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4869 && ckWARN(WARN_REGEXP)
4870 && !(PL_reg_flags & RF_warned))
4872 PL_reg_flags |= RF_warned;
4873 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4874 "Complex regular subexpression recursion "
4875 "limit (%d) exceeded",
4878 cur_curlyx->u.curlyx.count--;
4882 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4883 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4885 /* Try grabbing another A and see if it helps. */
4886 PL_reginput = locinput;
4887 cur_curlyx->u.curlyx.lastloc = locinput;
4888 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4889 REGCP_SET(ST.lastcp);
4890 PUSH_STATE_GOTO(WHILEM_A_min,
4891 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4895 #define ST st->u.branch
4897 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4898 next = scan + ARG(scan);
4901 scan = NEXTOPER(scan);
4904 case BRANCH: /* /(...|A|...)/ */
4905 scan = NEXTOPER(scan); /* scan now points to inner node */
4906 ST.lastparen = *PL_reglastparen;
4907 ST.next_branch = next;
4909 PL_reginput = locinput;
4911 /* Now go into the branch */
4913 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4915 PUSH_STATE_GOTO(BRANCH_next, scan);
4919 PL_reginput = locinput;
4920 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4921 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4922 PUSH_STATE_GOTO(CUTGROUP_next,next);
4924 case CUTGROUP_next_fail:
4927 if (st->u.mark.mark_name)
4928 sv_commit = st->u.mark.mark_name;
4934 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4939 REGCP_UNWIND(ST.cp);
4940 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4941 PL_regoffs[n].end = -1;
4942 *PL_reglastparen = n;
4943 /*dmq: *PL_reglastcloseparen = n; */
4944 scan = ST.next_branch;
4945 /* no more branches? */
4946 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4948 PerlIO_printf( Perl_debug_log,
4949 "%*s %sBRANCH failed...%s\n",
4950 REPORT_CODE_OFF+depth*2, "",
4956 continue; /* execute next BRANCH[J] op */
4964 #define ST st->u.curlym
4966 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4968 /* This is an optimisation of CURLYX that enables us to push
4969 * only a single backtracking state, no matter how many matches
4970 * there are in {m,n}. It relies on the pattern being constant
4971 * length, with no parens to influence future backrefs
4975 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4977 /* if paren positive, emulate an OPEN/CLOSE around A */
4979 U32 paren = ST.me->flags;
4980 if (paren > PL_regsize)
4982 if (paren > *PL_reglastparen)
4983 *PL_reglastparen = paren;
4984 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4992 ST.c1 = CHRTEST_UNINIT;
4995 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4998 curlym_do_A: /* execute the A in /A{m,n}B/ */
4999 PL_reginput = locinput;
5000 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
5003 case CURLYM_A: /* we've just matched an A */
5004 locinput = st->locinput;
5005 nextchr = UCHARAT(locinput);
5008 /* after first match, determine A's length: u.curlym.alen */
5009 if (ST.count == 1) {
5010 if (PL_reg_match_utf8) {
5012 while (s < PL_reginput) {
5018 ST.alen = PL_reginput - locinput;
5021 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5024 PerlIO_printf(Perl_debug_log,
5025 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5026 (int)(REPORT_CODE_OFF+(depth*2)), "",
5027 (IV) ST.count, (IV)ST.alen)
5030 locinput = PL_reginput;
5032 if (cur_eval && cur_eval->u.eval.close_paren &&
5033 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5037 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5038 if ( max == REG_INFTY || ST.count < max )
5039 goto curlym_do_A; /* try to match another A */
5041 goto curlym_do_B; /* try to match B */
5043 case CURLYM_A_fail: /* just failed to match an A */
5044 REGCP_UNWIND(ST.cp);
5046 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5047 || (cur_eval && cur_eval->u.eval.close_paren &&
5048 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5051 curlym_do_B: /* execute the B in /A{m,n}B/ */
5052 PL_reginput = locinput;
5053 if (ST.c1 == CHRTEST_UNINIT) {
5054 /* calculate c1 and c2 for possible match of 1st char
5055 * following curly */
5056 ST.c1 = ST.c2 = CHRTEST_VOID;
5057 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5058 regnode *text_node = ST.B;
5059 if (! HAS_TEXT(text_node))
5060 FIND_NEXT_IMPT(text_node);
5063 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5065 But the former is redundant in light of the latter.
5067 if this changes back then the macro for
5068 IS_TEXT and friends need to change.
5070 if (PL_regkind[OP(text_node)] == EXACT)
5073 ST.c1 = (U8)*STRING(text_node);
5074 switch (OP(text_node)) {
5075 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5077 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5078 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5079 default: ST.c2 = ST.c1;
5086 PerlIO_printf(Perl_debug_log,
5087 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5088 (int)(REPORT_CODE_OFF+(depth*2)),
5091 if (ST.c1 != CHRTEST_VOID
5092 && UCHARAT(PL_reginput) != ST.c1
5093 && UCHARAT(PL_reginput) != ST.c2)
5095 /* simulate B failing */
5097 PerlIO_printf(Perl_debug_log,
5098 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5099 (int)(REPORT_CODE_OFF+(depth*2)),"",
5102 state_num = CURLYM_B_fail;
5103 goto reenter_switch;
5107 /* mark current A as captured */
5108 I32 paren = ST.me->flags;
5110 PL_regoffs[paren].start
5111 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5112 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5113 /*dmq: *PL_reglastcloseparen = paren; */
5116 PL_regoffs[paren].end = -1;
5117 if (cur_eval && cur_eval->u.eval.close_paren &&
5118 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5127 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5130 case CURLYM_B_fail: /* just failed to match a B */
5131 REGCP_UNWIND(ST.cp);
5133 I32 max = ARG2(ST.me);
5134 if (max != REG_INFTY && ST.count == max)
5136 goto curlym_do_A; /* try to match a further A */
5138 /* backtrack one A */
5139 if (ST.count == ARG1(ST.me) /* min */)
5142 locinput = HOPc(locinput, -ST.alen);
5143 goto curlym_do_B; /* try to match B */
5146 #define ST st->u.curly
5148 #define CURLY_SETPAREN(paren, success) \
5151 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5152 PL_regoffs[paren].end = locinput - PL_bostr; \
5153 *PL_reglastcloseparen = paren; \
5156 PL_regoffs[paren].end = -1; \
5159 case STAR: /* /A*B/ where A is width 1 */
5163 scan = NEXTOPER(scan);
5165 case PLUS: /* /A+B/ where A is width 1 */
5169 scan = NEXTOPER(scan);
5171 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5172 ST.paren = scan->flags; /* Which paren to set */
5173 if (ST.paren > PL_regsize)
5174 PL_regsize = ST.paren;
5175 if (ST.paren > *PL_reglastparen)
5176 *PL_reglastparen = ST.paren;
5177 ST.min = ARG1(scan); /* min to match */
5178 ST.max = ARG2(scan); /* max to match */
5179 if (cur_eval && cur_eval->u.eval.close_paren &&
5180 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5184 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5186 case CURLY: /* /A{m,n}B/ where A is width 1 */
5188 ST.min = ARG1(scan); /* min to match */
5189 ST.max = ARG2(scan); /* max to match */
5190 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5193 * Lookahead to avoid useless match attempts
5194 * when we know what character comes next.
5196 * Used to only do .*x and .*?x, but now it allows
5197 * for )'s, ('s and (?{ ... })'s to be in the way
5198 * of the quantifier and the EXACT-like node. -- japhy
5201 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5203 if (HAS_TEXT(next) || JUMPABLE(next)) {
5205 regnode *text_node = next;
5207 if (! HAS_TEXT(text_node))
5208 FIND_NEXT_IMPT(text_node);
5210 if (! HAS_TEXT(text_node))
5211 ST.c1 = ST.c2 = CHRTEST_VOID;
5213 if ( PL_regkind[OP(text_node)] != EXACT ) {
5214 ST.c1 = ST.c2 = CHRTEST_VOID;
5215 goto assume_ok_easy;
5218 s = (U8*)STRING(text_node);
5220 /* Currently we only get here when
5222 PL_rekind[OP(text_node)] == EXACT
5224 if this changes back then the macro for IS_TEXT and
5225 friends need to change. */
5228 switch (OP(text_node)) {
5229 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5231 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5232 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5233 default: ST.c2 = ST.c1; break;
5236 else { /* UTF_PATTERN */
5237 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5238 STRLEN ulen1, ulen2;
5239 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5240 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5242 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5243 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5245 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5247 0 : UTF8_ALLOW_ANY);
5248 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5250 0 : UTF8_ALLOW_ANY);
5252 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5254 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5259 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5266 ST.c1 = ST.c2 = CHRTEST_VOID;
5271 PL_reginput = locinput;
5274 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5277 locinput = PL_reginput;
5279 if (ST.c1 == CHRTEST_VOID)
5280 goto curly_try_B_min;
5282 ST.oldloc = locinput;
5284 /* set ST.maxpos to the furthest point along the
5285 * string that could possibly match */
5286 if (ST.max == REG_INFTY) {
5287 ST.maxpos = PL_regeol - 1;
5289 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5292 else if (utf8_target) {
5293 int m = ST.max - ST.min;
5294 for (ST.maxpos = locinput;
5295 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5296 ST.maxpos += UTF8SKIP(ST.maxpos);
5299 ST.maxpos = locinput + ST.max - ST.min;
5300 if (ST.maxpos >= PL_regeol)
5301 ST.maxpos = PL_regeol - 1;
5303 goto curly_try_B_min_known;
5307 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5308 locinput = PL_reginput;
5309 if (ST.count < ST.min)
5311 if ((ST.count > ST.min)
5312 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5314 /* A{m,n} must come at the end of the string, there's
5315 * no point in backing off ... */
5317 /* ...except that $ and \Z can match before *and* after
5318 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5319 We may back off by one in this case. */
5320 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5324 goto curly_try_B_max;
5329 case CURLY_B_min_known_fail:
5330 /* failed to find B in a non-greedy match where c1,c2 valid */
5331 if (ST.paren && ST.count)
5332 PL_regoffs[ST.paren].end = -1;
5334 PL_reginput = locinput; /* Could be reset... */
5335 REGCP_UNWIND(ST.cp);
5336 /* Couldn't or didn't -- move forward. */
5337 ST.oldloc = locinput;
5339 locinput += UTF8SKIP(locinput);
5343 curly_try_B_min_known:
5344 /* find the next place where 'B' could work, then call B */
5348 n = (ST.oldloc == locinput) ? 0 : 1;
5349 if (ST.c1 == ST.c2) {
5351 /* set n to utf8_distance(oldloc, locinput) */
5352 while (locinput <= ST.maxpos &&
5353 utf8n_to_uvchr((U8*)locinput,
5354 UTF8_MAXBYTES, &len,
5355 uniflags) != (UV)ST.c1) {
5361 /* set n to utf8_distance(oldloc, locinput) */
5362 while (locinput <= ST.maxpos) {
5364 const UV c = utf8n_to_uvchr((U8*)locinput,
5365 UTF8_MAXBYTES, &len,
5367 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5375 if (ST.c1 == ST.c2) {
5376 while (locinput <= ST.maxpos &&
5377 UCHARAT(locinput) != ST.c1)
5381 while (locinput <= ST.maxpos
5382 && UCHARAT(locinput) != ST.c1
5383 && UCHARAT(locinput) != ST.c2)
5386 n = locinput - ST.oldloc;
5388 if (locinput > ST.maxpos)
5390 /* PL_reginput == oldloc now */
5393 if (regrepeat(rex, ST.A, n, depth) < n)
5396 PL_reginput = locinput;
5397 CURLY_SETPAREN(ST.paren, ST.count);
5398 if (cur_eval && cur_eval->u.eval.close_paren &&
5399 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5402 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5407 case CURLY_B_min_fail:
5408 /* failed to find B in a non-greedy match where c1,c2 invalid */
5409 if (ST.paren && ST.count)
5410 PL_regoffs[ST.paren].end = -1;
5412 REGCP_UNWIND(ST.cp);
5413 /* failed -- move forward one */
5414 PL_reginput = locinput;
5415 if (regrepeat(rex, ST.A, 1, depth)) {
5417 locinput = PL_reginput;
5418 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5419 ST.count > 0)) /* count overflow ? */
5422 CURLY_SETPAREN(ST.paren, ST.count);
5423 if (cur_eval && cur_eval->u.eval.close_paren &&
5424 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5427 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5435 /* a successful greedy match: now try to match B */
5436 if (cur_eval && cur_eval->u.eval.close_paren &&
5437 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5442 if (ST.c1 != CHRTEST_VOID)
5443 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5444 UTF8_MAXBYTES, 0, uniflags)
5445 : (UV) UCHARAT(PL_reginput);
5446 /* If it could work, try it. */
5447 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5448 CURLY_SETPAREN(ST.paren, ST.count);
5449 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5454 case CURLY_B_max_fail:
5455 /* failed to find B in a greedy match */
5456 if (ST.paren && ST.count)
5457 PL_regoffs[ST.paren].end = -1;
5459 REGCP_UNWIND(ST.cp);
5461 if (--ST.count < ST.min)
5463 PL_reginput = locinput = HOPc(locinput, -1);
5464 goto curly_try_B_max;
5471 /* we've just finished A in /(??{A})B/; now continue with B */
5473 st->u.eval.toggle_reg_flags
5474 = cur_eval->u.eval.toggle_reg_flags;
5475 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5477 st->u.eval.prev_rex = rex_sv; /* inner */
5478 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5479 rex = (struct regexp *)SvANY(rex_sv);
5480 rexi = RXi_GET(rex);
5481 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5482 (void)ReREFCNT_inc(rex_sv);
5483 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5485 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5486 PL_reglastparen = &rex->lastparen;
5487 PL_reglastcloseparen = &rex->lastcloseparen;
5489 REGCP_SET(st->u.eval.lastcp);
5490 PL_reginput = locinput;
5492 /* Restore parens of the outer rex without popping the
5494 tmpix = PL_savestack_ix;
5495 PL_savestack_ix = cur_eval->u.eval.lastcp;
5497 PL_savestack_ix = tmpix;
5499 st->u.eval.prev_eval = cur_eval;
5500 cur_eval = cur_eval->u.eval.prev_eval;
5502 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5503 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5504 if ( nochange_depth )
5507 PUSH_YES_STATE_GOTO(EVAL_AB,
5508 st->u.eval.prev_eval->u.eval.B); /* match B */
5511 if (locinput < reginfo->till) {
5512 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5513 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5515 (long)(locinput - PL_reg_starttry),
5516 (long)(reginfo->till - PL_reg_starttry),
5519 sayNO_SILENT; /* Cannot match: too short. */
5521 PL_reginput = locinput; /* put where regtry can find it */
5522 sayYES; /* Success! */
5524 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5526 PerlIO_printf(Perl_debug_log,
5527 "%*s %ssubpattern success...%s\n",
5528 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5529 PL_reginput = locinput; /* put where regtry can find it */
5530 sayYES; /* Success! */
5533 #define ST st->u.ifmatch
5535 case SUSPEND: /* (?>A) */
5537 PL_reginput = locinput;
5540 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5542 goto ifmatch_trivial_fail_test;
5544 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5546 ifmatch_trivial_fail_test:
5548 char * const s = HOPBACKc(locinput, scan->flags);
5553 sw = 1 - cBOOL(ST.wanted);
5557 next = scan + ARG(scan);
5565 PL_reginput = locinput;
5569 ST.logical = logical;
5570 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5572 /* execute body of (?...A) */
5573 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5576 case IFMATCH_A_fail: /* body of (?...A) failed */
5577 ST.wanted = !ST.wanted;
5580 case IFMATCH_A: /* body of (?...A) succeeded */
5582 sw = cBOOL(ST.wanted);
5584 else if (!ST.wanted)
5587 if (OP(ST.me) == SUSPEND)
5588 locinput = PL_reginput;
5590 locinput = PL_reginput = st->locinput;
5591 nextchr = UCHARAT(locinput);
5593 scan = ST.me + ARG(ST.me);
5596 continue; /* execute B */
5601 next = scan + ARG(scan);
5606 reginfo->cutpoint = PL_regeol;
5609 PL_reginput = locinput;
5611 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5612 PUSH_STATE_GOTO(COMMIT_next,next);
5614 case COMMIT_next_fail:
5621 #define ST st->u.mark
5623 ST.prev_mark = mark_state;
5624 ST.mark_name = sv_commit = sv_yes_mark
5625 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5627 ST.mark_loc = PL_reginput = locinput;
5628 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5630 case MARKPOINT_next:
5631 mark_state = ST.prev_mark;
5634 case MARKPOINT_next_fail:
5635 if (popmark && sv_eq(ST.mark_name,popmark))
5637 if (ST.mark_loc > startpoint)
5638 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5639 popmark = NULL; /* we found our mark */
5640 sv_commit = ST.mark_name;
5643 PerlIO_printf(Perl_debug_log,
5644 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5645 REPORT_CODE_OFF+depth*2, "",
5646 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5649 mark_state = ST.prev_mark;
5650 sv_yes_mark = mark_state ?
5651 mark_state->u.mark.mark_name : NULL;
5655 PL_reginput = locinput;
5657 /* (*SKIP) : if we fail we cut here*/
5658 ST.mark_name = NULL;
5659 ST.mark_loc = locinput;
5660 PUSH_STATE_GOTO(SKIP_next,next);
5662 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5663 otherwise do nothing. Meaning we need to scan
5665 regmatch_state *cur = mark_state;
5666 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5669 if ( sv_eq( cur->u.mark.mark_name,
5672 ST.mark_name = find;
5673 PUSH_STATE_GOTO( SKIP_next, next );
5675 cur = cur->u.mark.prev_mark;
5678 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5680 case SKIP_next_fail:
5682 /* (*CUT:NAME) - Set up to search for the name as we
5683 collapse the stack*/
5684 popmark = ST.mark_name;
5686 /* (*CUT) - No name, we cut here.*/
5687 if (ST.mark_loc > startpoint)
5688 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5689 /* but we set sv_commit to latest mark_name if there
5690 is one so they can test to see how things lead to this
5693 sv_commit=mark_state->u.mark.mark_name;
5701 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5703 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5706 U8 folded[UTF8_MAXBYTES_CASE+1];
5708 const char * const l = locinput;
5709 char *e = PL_regeol;
5710 to_uni_fold(n, folded, &foldlen);
5712 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5713 l, &e, 0, utf8_target)) {
5718 nextchr = UCHARAT(locinput);
5721 if ((n=is_LNBREAK(locinput,utf8_target))) {
5723 nextchr = UCHARAT(locinput);
5728 #define CASE_CLASS(nAmE) \
5730 if (locinput >= PL_regeol) \
5732 if ((n=is_##nAmE(locinput,utf8_target))) { \
5734 nextchr = UCHARAT(locinput); \
5739 if (locinput >= PL_regeol) \
5741 if ((n=is_##nAmE(locinput,utf8_target))) { \
5744 locinput += UTF8SKIP(locinput); \
5745 nextchr = UCHARAT(locinput); \
5750 CASE_CLASS(HORIZWS);
5754 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5755 PTR2UV(scan), OP(scan));
5756 Perl_croak(aTHX_ "regexp memory corruption");
5760 /* switch break jumps here */
5761 scan = next; /* prepare to execute the next op and ... */
5762 continue; /* ... jump back to the top, reusing st */
5766 /* push a state that backtracks on success */
5767 st->u.yes.prev_yes_state = yes_state;
5771 /* push a new regex state, then continue at scan */
5773 regmatch_state *newst;
5776 regmatch_state *cur = st;
5777 regmatch_state *curyes = yes_state;
5779 regmatch_slab *slab = PL_regmatch_slab;
5780 for (;curd > -1;cur--,curd--) {
5781 if (cur < SLAB_FIRST(slab)) {
5783 cur = SLAB_LAST(slab);
5785 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5786 REPORT_CODE_OFF + 2 + depth * 2,"",
5787 curd, PL_reg_name[cur->resume_state],
5788 (curyes == cur) ? "yes" : ""
5791 curyes = cur->u.yes.prev_yes_state;
5794 DEBUG_STATE_pp("push")
5797 st->locinput = locinput;
5799 if (newst > SLAB_LAST(PL_regmatch_slab))
5800 newst = S_push_slab(aTHX);
5801 PL_regmatch_state = newst;
5803 locinput = PL_reginput;
5804 nextchr = UCHARAT(locinput);
5812 * We get here only if there's trouble -- normally "case END" is
5813 * the terminating point.
5815 Perl_croak(aTHX_ "corrupted regexp pointers");
5821 /* we have successfully completed a subexpression, but we must now
5822 * pop to the state marked by yes_state and continue from there */
5823 assert(st != yes_state);
5825 while (st != yes_state) {
5827 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5828 PL_regmatch_slab = PL_regmatch_slab->prev;
5829 st = SLAB_LAST(PL_regmatch_slab);
5833 DEBUG_STATE_pp("pop (no final)");
5835 DEBUG_STATE_pp("pop (yes)");
5841 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5842 || yes_state > SLAB_LAST(PL_regmatch_slab))
5844 /* not in this slab, pop slab */
5845 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5846 PL_regmatch_slab = PL_regmatch_slab->prev;
5847 st = SLAB_LAST(PL_regmatch_slab);
5849 depth -= (st - yes_state);
5852 yes_state = st->u.yes.prev_yes_state;
5853 PL_regmatch_state = st;
5856 locinput= st->locinput;
5857 nextchr = UCHARAT(locinput);
5859 state_num = st->resume_state + no_final;
5860 goto reenter_switch;
5863 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5864 PL_colors[4], PL_colors[5]));
5866 if (PL_reg_eval_set) {
5867 /* each successfully executed (?{...}) block does the equivalent of
5868 * local $^R = do {...}
5869 * When popping the save stack, all these locals would be undone;
5870 * bypass this by setting the outermost saved $^R to the latest
5872 if (oreplsv != GvSV(PL_replgv))
5873 sv_setsv(oreplsv, GvSV(PL_replgv));
5880 PerlIO_printf(Perl_debug_log,
5881 "%*s %sfailed...%s\n",
5882 REPORT_CODE_OFF+depth*2, "",
5883 PL_colors[4], PL_colors[5])
5895 /* there's a previous state to backtrack to */
5897 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5898 PL_regmatch_slab = PL_regmatch_slab->prev;
5899 st = SLAB_LAST(PL_regmatch_slab);
5901 PL_regmatch_state = st;
5902 locinput= st->locinput;
5903 nextchr = UCHARAT(locinput);
5905 DEBUG_STATE_pp("pop");
5907 if (yes_state == st)
5908 yes_state = st->u.yes.prev_yes_state;
5910 state_num = st->resume_state + 1; /* failure = success + 1 */
5911 goto reenter_switch;
5916 if (rex->intflags & PREGf_VERBARG_SEEN) {
5917 SV *sv_err = get_sv("REGERROR", 1);
5918 SV *sv_mrk = get_sv("REGMARK", 1);
5920 sv_commit = &PL_sv_no;
5922 sv_yes_mark = &PL_sv_yes;
5925 sv_commit = &PL_sv_yes;
5926 sv_yes_mark = &PL_sv_no;
5928 sv_setsv(sv_err, sv_commit);
5929 sv_setsv(sv_mrk, sv_yes_mark);
5932 /* clean up; in particular, free all slabs above current one */
5933 LEAVE_SCOPE(oldsave);
5939 - regrepeat - repeatedly match something simple, report how many
5942 * [This routine now assumes that it will only match on things of length 1.
5943 * That was true before, but now we assume scan - reginput is the count,
5944 * rather than incrementing count on every character. [Er, except utf8.]]
5947 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5950 register char *scan;
5952 register char *loceol = PL_regeol;
5953 register I32 hardcount = 0;
5954 register bool utf8_target = PL_reg_match_utf8;
5957 PERL_UNUSED_ARG(depth);
5960 PERL_ARGS_ASSERT_REGREPEAT;
5963 if (max == REG_INFTY)
5965 else if (max < loceol - scan)
5966 loceol = scan + max;
5971 while (scan < loceol && hardcount < max && *scan != '\n') {
5972 scan += UTF8SKIP(scan);
5976 while (scan < loceol && *scan != '\n')
5983 while (scan < loceol && hardcount < max) {
5984 scan += UTF8SKIP(scan);
5995 /* To get here, EXACTish nodes must have *byte* length == 1. That
5996 * means they match only characters in the string that can be expressed
5997 * as a single byte. For non-utf8 strings, that means a simple match.
5998 * For utf8 strings, the character matched must be an invariant, or
5999 * downgradable to a single byte. The pattern's utf8ness is
6000 * irrelevant, as since it's a single byte, it either isn't utf8, or if
6001 * it is, it's an invariant */
6004 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6006 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6007 while (scan < loceol && UCHARAT(scan) == c) {
6013 /* Here, the string is utf8, and the pattern char is different
6014 * in utf8 than not, so can't compare them directly. Outside the
6015 * loop, find the two utf8 bytes that represent c, and then
6016 * look for those in sequence in the utf8 string */
6017 U8 high = UTF8_TWO_BYTE_HI(c);
6018 U8 low = UTF8_TWO_BYTE_LO(c);
6021 while (hardcount < max
6022 && scan + 1 < loceol
6023 && UCHARAT(scan) == high
6024 && UCHARAT(scan + 1) == low)
6032 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6036 PL_reg_flags |= RF_tainted;
6037 utf8_flags = FOLDEQ_UTF8_LOCALE;
6045 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6047 /* The comments for the EXACT case above apply as well to these fold
6052 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6054 if (utf8_target) { /* Use full Unicode fold matching */
6055 char *tmpeol = loceol;
6056 while (hardcount < max
6057 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6058 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6065 /* XXX Note that the above handles properly the German sharp s in
6066 * the pattern matching ss in the string. But it doesn't handle
6067 * properly cases where the string contains say 'LIGATURE ff' and
6068 * the pattern is 'f+'. This would require, say, a new function or
6069 * revised interface to foldEQ_utf8(), in which the maximum number
6070 * of characters to match could be passed and it would return how
6071 * many actually did. This is just one of many cases where
6072 * multi-char folds don't work properly, and so the fix is being
6078 /* Here, the string isn't utf8 and c is a single byte; and either
6079 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6080 * doesn't affect c. Can just do simple comparisons for exact or
6083 case EXACTF: folded = PL_fold[c]; break;
6085 case EXACTFU: folded = PL_fold_latin1[c]; break;
6086 case EXACTFL: folded = PL_fold_locale[c]; break;
6087 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6089 while (scan < loceol &&
6090 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6098 if (utf8_target || OP(p) == ANYOFV) {
6101 inclasslen = loceol - scan;
6102 while (hardcount < max
6103 && ((inclasslen = loceol - scan) > 0)
6104 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6110 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6118 LOAD_UTF8_CHARCLASS_ALNUM();
6119 while (hardcount < max && scan < loceol &&
6120 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6122 scan += UTF8SKIP(scan);
6126 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6134 while (scan < loceol && isALNUM((U8) *scan)) {
6139 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6144 PL_reg_flags |= RF_tainted;
6147 while (hardcount < max && scan < loceol &&
6148 isALNUM_LC_utf8((U8*)scan)) {
6149 scan += UTF8SKIP(scan);
6153 while (scan < loceol && isALNUM_LC(*scan))
6163 LOAD_UTF8_CHARCLASS_ALNUM();
6164 while (hardcount < max && scan < loceol &&
6165 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6167 scan += UTF8SKIP(scan);
6171 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6178 goto utf8_Nwordchar;
6179 while (scan < loceol && ! isALNUM((U8) *scan)) {
6185 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6186 scan += UTF8SKIP(scan);
6190 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6196 PL_reg_flags |= RF_tainted;
6199 while (hardcount < max && scan < loceol &&
6200 !isALNUM_LC_utf8((U8*)scan)) {
6201 scan += UTF8SKIP(scan);
6205 while (scan < loceol && !isALNUM_LC(*scan))
6215 LOAD_UTF8_CHARCLASS_SPACE();
6216 while (hardcount < max && scan < loceol &&
6218 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6220 scan += UTF8SKIP(scan);
6226 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6235 while (scan < loceol && isSPACE((U8) *scan)) {
6240 while (scan < loceol && isSPACE_A((U8) *scan)) {
6245 PL_reg_flags |= RF_tainted;
6248 while (hardcount < max && scan < loceol &&
6249 isSPACE_LC_utf8((U8*)scan)) {
6250 scan += UTF8SKIP(scan);
6254 while (scan < loceol && isSPACE_LC(*scan))
6264 LOAD_UTF8_CHARCLASS_SPACE();
6265 while (hardcount < max && scan < loceol &&
6267 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6269 scan += UTF8SKIP(scan);
6275 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6284 while (scan < loceol && ! isSPACE((U8) *scan)) {
6290 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6291 scan += UTF8SKIP(scan);
6295 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6301 PL_reg_flags |= RF_tainted;
6304 while (hardcount < max && scan < loceol &&
6305 !isSPACE_LC_utf8((U8*)scan)) {
6306 scan += UTF8SKIP(scan);
6310 while (scan < loceol && !isSPACE_LC(*scan))
6317 LOAD_UTF8_CHARCLASS_DIGIT();
6318 while (hardcount < max && scan < loceol &&
6319 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6320 scan += UTF8SKIP(scan);
6324 while (scan < loceol && isDIGIT(*scan))
6329 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6334 PL_reg_flags |= RF_tainted;
6337 while (hardcount < max && scan < loceol &&
6338 isDIGIT_LC_utf8((U8*)scan)) {
6339 scan += UTF8SKIP(scan);
6343 while (scan < loceol && isDIGIT_LC(*scan))
6350 LOAD_UTF8_CHARCLASS_DIGIT();
6351 while (hardcount < max && scan < loceol &&
6352 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6353 scan += UTF8SKIP(scan);
6357 while (scan < loceol && !isDIGIT(*scan))
6363 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6364 scan += UTF8SKIP(scan);
6368 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6374 PL_reg_flags |= RF_tainted;
6377 while (hardcount < max && scan < loceol &&
6378 !isDIGIT_LC_utf8((U8*)scan)) {
6379 scan += UTF8SKIP(scan);
6383 while (scan < loceol && !isDIGIT_LC(*scan))
6390 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6396 LNBREAK can match two latin chars, which is ok,
6397 because we have a null terminated string, but we
6398 have to use hardcount in this situation
6400 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6409 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6414 while (scan < loceol && is_HORIZWS_latin1(scan))
6421 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6422 scan += UTF8SKIP(scan);
6426 while (scan < loceol && !is_HORIZWS_latin1(scan))
6434 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6439 while (scan < loceol && is_VERTWS_latin1(scan))
6447 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6448 scan += UTF8SKIP(scan);
6452 while (scan < loceol && !is_VERTWS_latin1(scan))
6458 default: /* Called on something of 0 width. */
6459 break; /* So match right here or not at all. */
6465 c = scan - PL_reginput;
6469 GET_RE_DEBUG_FLAGS_DECL;
6471 SV * const prop = sv_newmortal();
6472 regprop(prog, prop, p);
6473 PerlIO_printf(Perl_debug_log,
6474 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6475 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6483 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6485 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
6486 create a copy so that changes the caller makes won't change the shared one
6489 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6491 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6492 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp, altsvp));
6497 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6499 /* Returns the swash for the input 'node' in the regex 'prog'.
6500 * If <doinit> is true, will attempt to create the swash if not already
6502 * If <listsvp> is non-null, will return the swash initialization string in
6504 * If <altsvp> is non-null, will return the alternates to the regular swash
6506 * Tied intimately to how regcomp.c sets up the data structure */
6514 RXi_GET_DECL(prog,progi);
6515 const struct reg_data * const data = prog ? progi->data : NULL;
6517 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
6519 assert(ANYOF_NONBITMAP(node));
6521 if (data && data->count) {
6522 const U32 n = ARG(node);
6524 if (data->what[n] == 's') {
6525 SV * const rv = MUTABLE_SV(data->data[n]);
6526 AV * const av = MUTABLE_AV(SvRV(rv));
6527 SV **const ary = AvARRAY(av);
6528 bool invlist_has_user_defined_property;
6530 si = *ary; /* ary[0] = the string to initialize the swash with */
6532 /* Elements 3 and 4 are either both present or both absent. [3] is
6533 * any inversion list generated at compile time; [4] indicates if
6534 * that inversion list has any user-defined properties in it. */
6535 if (av_len(av) >= 3) {
6537 invlist_has_user_defined_property = cBOOL(SvUV(ary[4]));
6541 invlist_has_user_defined_property = FALSE;
6544 /* Element [1] is reserved for the set-up swash. If already there,
6545 * return it; if not, create it and store it there */
6546 if (SvROK(ary[1])) {
6549 else if (si && doinit) {
6551 sw = _core_swash_init("utf8", /* the utf8 package */
6555 0, /* not from tr/// */
6556 FALSE, /* is error if can't find
6559 invlist_has_user_defined_property);
6560 (void)av_store(av, 1, sw);
6563 /* Element [2] is for any multi-char folds. Note that is a
6564 * fundamentally flawed design, because can't backtrack and try
6565 * again. See [perl #89774] */
6566 if (SvTYPE(ary[2]) == SVt_PVAV) {
6573 SV* matches_string = newSVpvn("", 0);
6576 /* Use the swash, if any, which has to have incorporated into it all
6580 && SvTYPE(SvRV(sw)) == SVt_PVHV
6581 && (invlistsvp = hv_fetchs(MUTABLE_HV(SvRV(sw)), "INVLIST", FALSE)))
6583 invlist = *invlistsvp;
6585 else if (si && si != &PL_sv_undef) {
6587 /* If no swash, use the input nitialization string, if available */
6588 sv_catsv(matches_string, si);
6591 /* Add the inversion list to whatever we have. This may have come from
6592 * the swash, or from an input parameter */
6594 sv_catsv(matches_string, _invlist_contents(invlist));
6596 *listsvp = matches_string;
6606 - reginclass - determine if a character falls into a character class
6608 n is the ANYOF regnode
6609 p is the target string
6610 lenp is pointer to the maximum number of bytes of how far to go in p
6611 (This is assumed wthout checking to always be at least the current
6613 utf8_target tells whether p is in UTF-8.
6615 Returns true if matched; false otherwise. If lenp is not NULL, on return
6616 from a successful match, the value it points to will be updated to how many
6617 bytes in p were matched. If there was no match, the value is undefined,
6618 possibly changed from the input.
6620 Note that this can be a synthetic start class, a combination of various
6621 nodes, so things you think might be mutually exclusive, such as locale,
6622 aren't. It can match both locale and non-locale
6627 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6630 const char flags = ANYOF_FLAGS(n);
6636 PERL_ARGS_ASSERT_REGINCLASS;
6638 /* If c is not already the code point, get it */
6639 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6640 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6641 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6642 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6643 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6644 * UTF8_ALLOW_FFFF */
6645 if (c_len == (STRLEN)-1)
6646 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6652 /* Use passed in max length, or one character if none passed in or less
6653 * than one character. And assume will match just one character. This is
6654 * overwritten later if matched more. */
6656 maxlen = (*lenp > c_len) ? *lenp : c_len;
6664 /* If this character is potentially in the bitmap, check it */
6666 if (ANYOF_BITMAP_TEST(n, c))
6668 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6675 else if (flags & ANYOF_LOCALE) {
6676 PL_reg_flags |= RF_tainted;
6678 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6679 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6683 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6684 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6685 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6686 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6687 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6688 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6689 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6690 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6691 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6692 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6693 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6694 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6695 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6696 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6697 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6698 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6699 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6700 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6701 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6702 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6703 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6704 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6705 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6706 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6707 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6708 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6709 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6710 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6711 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6712 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6713 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6714 ) /* How's that for a conditional? */
6721 /* If the bitmap didn't (or couldn't) match, and something outside the
6722 * bitmap could match, try that. Locale nodes specifiy completely the
6723 * behavior of code points in the bit map (otherwise, a utf8 target would
6724 * cause them to be treated as Unicode and not locale), except in
6725 * the very unlikely event when this node is a synthetic start class, which
6726 * could be a combination of locale and non-locale nodes. So allow locale
6727 * to match for the synthetic start class, which will give a false
6728 * positive that will be resolved when the match is done again as not part
6729 * of the synthetic start class */
6731 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6732 match = TRUE; /* Everything above 255 matches */
6734 else if (ANYOF_NONBITMAP(n)
6735 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6738 || (! (flags & ANYOF_LOCALE))
6739 || (flags & ANYOF_IS_SYNTHETIC)))))
6742 SV * const sw = core_regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6750 /* Not utf8. Convert as much of the string as available up
6751 * to the limit of how far the (single) character in the
6752 * pattern can possibly match (no need to go further). If
6753 * the node is a straight ANYOF or not folding, it can't
6754 * match more than one. Otherwise, It can match up to how
6755 * far a single char can fold to. Since not utf8, each
6756 * character is a single byte, so the max it can be in
6757 * bytes is the same as the max it can be in characters */
6758 STRLEN len = (OP(n) == ANYOF
6759 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6761 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6763 : UTF8_MAX_FOLD_CHAR_EXPAND;
6764 utf8_p = bytes_to_utf8(p, &len);
6767 if (swash_fetch(sw, utf8_p, TRUE))
6769 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6771 /* Here, we need to test if the fold of the target string
6772 * matches. The non-multi char folds have all been moved to
6773 * the compilation phase, and the multi-char folds have
6774 * been stored by regcomp into 'av'; we linearly check to
6775 * see if any match the target string (folded). We know
6776 * that the originals were each one character, but we don't
6777 * currently know how many characters/bytes each folded to,
6778 * except we do know that there are small limits imposed by
6779 * Unicode. XXX A performance enhancement would be to have
6780 * regcomp.c store the max number of chars/bytes that are
6781 * in an av entry, as, say the 0th element. Even better
6782 * would be to have a hash of the few characters that can
6783 * start a multi-char fold to the max number of chars of
6786 * If there is a match, we will need to advance (if lenp is
6787 * specified) the match pointer in the target string. But
6788 * what we are comparing here isn't that string directly,
6789 * but its fold, whose length may differ from the original.
6790 * As we go along in constructing the fold, therefore, we
6791 * create a map so that we know how many bytes in the
6792 * source to advance given that we have matched a certain
6793 * number of bytes in the fold. This map is stored in
6794 * 'map_fold_len_back'. Let n mean the number of bytes in
6795 * the fold of the first character that we are folding.
6796 * Then map_fold_len_back[n] is set to the number of bytes
6797 * in that first character. Similarly let m be the
6798 * corresponding number for the second character to be
6799 * folded. Then map_fold_len_back[n+m] is set to the
6800 * number of bytes occupied by the first two source
6801 * characters. ... */
6802 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6803 U8 folded[UTF8_MAXBYTES_CASE+1];
6804 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6805 STRLEN total_foldlen = 0; /* num bytes in fold of all
6808 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6810 /* Here, only need to fold the first char of the target
6811 * string. It the source wasn't utf8, is 1 byte long */
6812 to_utf8_fold(utf8_p, folded, &foldlen);
6813 total_foldlen = foldlen;
6814 map_fold_len_back[foldlen] = (utf8_target)
6820 /* Here, need to fold more than the first char. Do so
6821 * up to the limits */
6822 U8* source_ptr = utf8_p; /* The source for the fold
6825 U8* folded_ptr = folded;
6826 U8* e = utf8_p + maxlen; /* Can't go beyond last
6827 available byte in the
6831 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6835 /* Fold the next character */
6836 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6837 STRLEN this_char_foldlen;
6838 to_utf8_fold(source_ptr,
6840 &this_char_foldlen);
6842 /* Bail if it would exceed the byte limit for
6843 * folding a single char. */
6844 if (this_char_foldlen + folded_ptr - folded >
6850 /* Add the fold of this character */
6851 Copy(this_char_folded,
6855 source_ptr += UTF8SKIP(source_ptr);
6856 folded_ptr += this_char_foldlen;
6857 total_foldlen = folded_ptr - folded;
6859 /* Create map from the number of bytes in the fold
6860 * back to the number of bytes in the source. If
6861 * the source isn't utf8, the byte count is just
6862 * the number of characters so far */
6863 map_fold_len_back[total_foldlen]
6865 ? source_ptr - utf8_p
6872 /* Do the linear search to see if the fold is in the list
6873 * of multi-char folds. */
6876 for (i = 0; i <= av_len(av); i++) {
6877 SV* const sv = *av_fetch(av, i, FALSE);
6879 const char * const s = SvPV_const(sv, len);
6881 if (len <= total_foldlen
6882 && memEQ(s, (char*)folded, len)
6884 /* If 0, means matched a partial char. See
6886 && map_fold_len_back[len])
6889 /* Advance the target string ptr to account for
6890 * this fold, but have to translate from the
6891 * folded length to the corresponding source
6894 *lenp = map_fold_len_back[len];
6903 /* If we allocated a string above, free it */
6904 if (! utf8_target) Safefree(utf8_p);
6909 return (flags & ANYOF_INVERT) ? !match : match;
6913 S_reghop3(U8 *s, I32 off, const U8* lim)
6915 /* return the position 'off' UTF-8 characters away from 's', forward if
6916 * 'off' >= 0, backwards if negative. But don't go outside of position
6917 * 'lim', which better be < s if off < 0 */
6921 PERL_ARGS_ASSERT_REGHOP3;
6924 while (off-- && s < lim) {
6925 /* XXX could check well-formedness here */
6930 while (off++ && s > lim) {
6932 if (UTF8_IS_CONTINUED(*s)) {
6933 while (s > lim && UTF8_IS_CONTINUATION(*s))
6936 /* XXX could check well-formedness here */
6943 /* there are a bunch of places where we use two reghop3's that should
6944 be replaced with this routine. but since thats not done yet
6945 we ifdef it out - dmq
6948 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6952 PERL_ARGS_ASSERT_REGHOP4;
6955 while (off-- && s < rlim) {
6956 /* XXX could check well-formedness here */
6961 while (off++ && s > llim) {
6963 if (UTF8_IS_CONTINUED(*s)) {
6964 while (s > llim && UTF8_IS_CONTINUATION(*s))
6967 /* XXX could check well-formedness here */
6975 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6979 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6982 while (off-- && s < lim) {
6983 /* XXX could check well-formedness here */
6990 while (off++ && s > lim) {
6992 if (UTF8_IS_CONTINUED(*s)) {
6993 while (s > lim && UTF8_IS_CONTINUATION(*s))
6996 /* XXX could check well-formedness here */
7005 restore_pos(pTHX_ void *arg)
7008 regexp * const rex = (regexp *)arg;
7009 if (PL_reg_eval_set) {
7010 if (PL_reg_oldsaved) {
7011 rex->subbeg = PL_reg_oldsaved;
7012 rex->sublen = PL_reg_oldsavedlen;
7013 #ifdef PERL_OLD_COPY_ON_WRITE
7014 rex->saved_copy = PL_nrs;
7016 RXp_MATCH_COPIED_on(rex);
7018 PL_reg_magic->mg_len = PL_reg_oldpos;
7019 PL_reg_eval_set = 0;
7020 PL_curpm = PL_reg_oldcurpm;
7025 S_to_utf8_substr(pTHX_ register regexp *prog)
7029 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7032 if (prog->substrs->data[i].substr
7033 && !prog->substrs->data[i].utf8_substr) {
7034 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7035 prog->substrs->data[i].utf8_substr = sv;
7036 sv_utf8_upgrade(sv);
7037 if (SvVALID(prog->substrs->data[i].substr)) {
7038 if (SvTAIL(prog->substrs->data[i].substr)) {
7039 /* Trim the trailing \n that fbm_compile added last
7041 SvCUR_set(sv, SvCUR(sv) - 1);
7042 /* Whilst this makes the SV technically "invalid" (as its
7043 buffer is no longer followed by "\0") when fbm_compile()
7044 adds the "\n" back, a "\0" is restored. */
7045 fbm_compile(sv, FBMcf_TAIL);
7049 if (prog->substrs->data[i].substr == prog->check_substr)
7050 prog->check_utf8 = sv;
7056 S_to_byte_substr(pTHX_ register regexp *prog)
7061 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7064 if (prog->substrs->data[i].utf8_substr
7065 && !prog->substrs->data[i].substr) {
7066 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7067 if (sv_utf8_downgrade(sv, TRUE)) {
7068 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7069 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7070 /* Trim the trailing \n that fbm_compile added last
7072 SvCUR_set(sv, SvCUR(sv) - 1);
7073 fbm_compile(sv, FBMcf_TAIL);
7081 prog->substrs->data[i].substr = sv;
7082 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7083 prog->check_substr = sv;
7090 * c-indentation-style: bsd
7092 * indent-tabs-mode: t
7095 * ex: set ts=8 sts=4 sw=4 noet: