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
78 #ifdef PERL_IN_XSUB_RE
84 #define RF_tainted 1 /* tainted information used? e.g. locale */
85 #define RF_warned 2 /* warned about big count? */
87 #define RF_utf8 8 /* Pattern contains multibyte chars? */
89 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
91 #define RS_init 1 /* eval environment created */
92 #define RS_set 2 /* replsv value is set */
98 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
99 * call if there are no complications: i.e., if everything matchable is
100 * straight forward in the bitmap */
101 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
102 : ANYOF_BITMAP_TEST(p,*(c)))
108 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
109 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
111 #define HOPc(pos,off) \
112 (char *)(PL_reg_match_utf8 \
113 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
115 #define HOPBACKc(pos, off) \
116 (char*)(PL_reg_match_utf8\
117 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
118 : (pos - off >= PL_bostr) \
122 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
123 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
125 /* these are unrolled below in the CCC_TRY_XXX defined */
127 /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
128 * skip the check on EBCDIC platforms */
129 # define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
131 # define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
132 if (!CAT2(PL_utf8_,class)) { \
134 ENTER; save_re_context(); \
135 ok=CAT2(is_utf8_,class)((const U8*)str); \
136 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
139 /* Doesn't do an assert to verify that is correct */
140 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
141 if (!CAT2(PL_utf8_,class)) { \
142 bool throw_away __attribute__unused__; \
143 ENTER; save_re_context(); \
144 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
147 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
148 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
149 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
151 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
152 LOAD_UTF8_CHARCLASS(X_begin, " "); \
153 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
154 /* These are utf8 constants, and not utf-ebcdic constants, so the \
155 * assert should likely and hopefully fail on an EBCDIC machine */ \
156 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
158 /* No asserts are done for these, in case called on an early \
159 * Unicode version in which they map to nothing */ \
160 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
161 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
162 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
163 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
164 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
165 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
166 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
168 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
170 /* The actual code for CCC_TRY, which uses several variables from the routine
171 * it's callable from. It is designed to be the bulk of a case statement.
172 * FUNC is the macro or function to call on non-utf8 targets that indicate if
173 * nextchr matches the class.
174 * UTF8_TEST is the whole test string to use for utf8 targets
175 * LOAD is what to use to test, and if not present to load in the swash for the
177 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
179 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
180 * utf8 and a variant, load the swash if necessary and test using the utf8
181 * test. Advance to the next character if test is ok, otherwise fail; If not
182 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
183 * fails, or advance to the next character */
185 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
186 if (locinput >= PL_regeol) { \
189 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
190 LOAD_UTF8_CHARCLASS(CLASS, STR); \
191 if (POS_OR_NEG (UTF8_TEST)) { \
194 locinput += PL_utf8skip[nextchr]; \
195 nextchr = UCHARAT(locinput); \
198 if (POS_OR_NEG (FUNC(nextchr))) { \
201 nextchr = UCHARAT(++locinput); \
204 /* Handle the non-locale cases for a character class and its complement. It
205 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
206 * This is because that code fails when the test succeeds, so we want to have
207 * the test fail so that the code succeeds. The swash is stored in a
208 * predictable PL_ place */
209 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
212 _CCC_TRY_CODE( !, FUNC, \
213 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
214 (U8*)locinput, TRUE)), \
217 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
218 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
219 (U8*)locinput, TRUE)), \
222 /* Generate the case statements for both locale and non-locale character
223 * classes in regmatch for classes that don't have special unicode semantics.
224 * Locales don't use an immediate swash, but an intermediary special locale
225 * function that is called on the pointer to the current place in the input
226 * string. That function will resolve to needing the same swash. One might
227 * think that because we don't know what the locale will match, we shouldn't
228 * check with the swash loading function that it loaded properly; ie, that we
229 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
230 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
232 #define CCC_TRY(NAME, NNAME, FUNC, \
233 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
234 NAMEA, NNAMEA, FUNCA, \
237 PL_reg_flags |= RF_tainted; \
238 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
240 PL_reg_flags |= RF_tainted; \
241 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
244 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
247 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
248 nextchr = UCHARAT(++locinput); \
251 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
255 locinput += PL_utf8skip[nextchr]; \
256 nextchr = UCHARAT(locinput); \
259 nextchr = UCHARAT(++locinput); \
262 /* Generate the non-locale cases */ \
263 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
265 /* This is like CCC_TRY, but has an extra set of parameters for generating case
266 * statements to handle separate Unicode semantics nodes */
267 #define CCC_TRY_U(NAME, NNAME, FUNC, \
268 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
269 NAMEU, NNAMEU, FUNCU, \
270 NAMEA, NNAMEA, FUNCA, \
272 CCC_TRY(NAME, NNAME, FUNC, \
273 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
274 NAMEA, NNAMEA, FUNCA, \
276 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
278 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
280 /* for use after a quantifier and before an EXACT-like node -- japhy */
281 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
283 * NOTE that *nothing* that affects backtracking should be in here, specifically
284 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
285 * node that is in between two EXACT like nodes when ascertaining what the required
286 * "follow" character is. This should probably be moved to regex compile time
287 * although it may be done at run time beause of the REF possibility - more
288 * investigation required. -- demerphq
290 #define JUMPABLE(rn) ( \
292 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
294 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
295 OP(rn) == PLUS || OP(rn) == MINMOD || \
297 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
299 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
301 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
304 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
305 we don't need this definition. */
306 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
307 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
308 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
311 /* ... so we use this as its faster. */
312 #define IS_TEXT(rn) ( OP(rn)==EXACT )
313 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
314 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
315 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
320 Search for mandatory following text node; for lookahead, the text must
321 follow but for lookbehind (rn->flags != 0) we skip to the next step.
323 #define FIND_NEXT_IMPT(rn) STMT_START { \
324 while (JUMPABLE(rn)) { \
325 const OPCODE type = OP(rn); \
326 if (type == SUSPEND || PL_regkind[type] == CURLY) \
327 rn = NEXTOPER(NEXTOPER(rn)); \
328 else if (type == PLUS) \
330 else if (type == IFMATCH) \
331 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
332 else rn += NEXT_OFF(rn); \
337 static void restore_pos(pTHX_ void *arg);
339 #define REGCP_PAREN_ELEMS 4
340 #define REGCP_OTHER_ELEMS 5
341 #define REGCP_FRAME_ELEMS 1
342 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
343 * are needed for the regexp context stack bookkeeping. */
346 S_regcppush(pTHX_ I32 parenfloor)
349 const int retval = PL_savestack_ix;
350 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
351 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
352 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
354 GET_RE_DEBUG_FLAGS_DECL;
356 if (paren_elems_to_push < 0)
357 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
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 = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1205 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
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 */
1468 if (UTF_PATTERN || utf8_target) {
1470 /* regcomp.c already folded this if pattern is in UTF-8 */
1471 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 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 */
1503 /* The idea in the non-utf8 EXACTF* cases is to first find the
1504 * first character of the EXACTF* node and then, if necessary,
1505 * case-insensitively compare the full text of the node. c1 is the
1506 * first character. c2 is its fold. This logic will not work for
1507 * Unicode semantics and the german sharp ss, which hence should
1508 * not be compiled into a node that gets here. */
1509 pat_string = STRING(c);
1510 ln = STR_LEN(c); /* length to match in octets/bytes */
1512 /* We know that we have to match at least 'ln' bytes (which is the
1513 * same as characters, since not utf8). If we have to match 3
1514 * characters, and there are only 2 availabe, we know without
1515 * trying that it will fail; so don't start a match past the
1516 * required minimum number from the far end */
1517 e = HOP3c(strend, -((I32)ln), s);
1519 if (!reginfo && e < s) {
1520 e = s; /* Due to minlen logic of intuit() */
1524 c2 = fold_array[c1];
1525 if (c1 == c2) { /* If char and fold are the same */
1526 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1529 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1538 /* If one of the operands is in utf8, we can't use the simpler
1539 * folding above, due to the fact that many different characters
1540 * can have the same fold, or portion of a fold, or different-
1542 pat_string = STRING(c);
1543 ln = STR_LEN(c); /* length to match in octets/bytes */
1544 pat_end = pat_string + ln;
1545 lnc = (UTF_PATTERN) /* length to match in characters */
1546 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1549 /* We have 'lnc' characters to match in the pattern, but because of
1550 * multi-character folding, each character in the target can match
1551 * up to 3 characters (Unicode guarantees it will never exceed
1552 * this) if it is utf8-encoded; and up to 2 if not (based on the
1553 * fact that the Latin 1 folds are already determined, and the
1554 * only multi-char fold in that range is the sharp-s folding to
1555 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1556 * string character. Adjust lnc accordingly, always matching at
1558 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1559 lnc = (lnc < expansion) ? 1 : lnc / expansion;
1561 /* As in the non-UTF8 case, if we have to match 3 characters, and
1562 * only 2 are left, it's guaranteed to fail, so don't start a
1563 * match that would require us to go beyond the end of the string
1565 e = HOP3c(strend, -((I32)lnc), s);
1567 if (!reginfo && e < s) {
1568 e = s; /* Due to minlen logic of intuit() */
1571 /* XXX Note that we could recalculate e every so-often through the
1572 * loop to stop earlier, as the worst case expansion above will
1573 * rarely be met, and as we go along we would usually find that e
1574 * moves further to the left. Unclear if worth the expense */
1577 char *my_strend= (char *)strend;
1578 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1579 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1580 && (!reginfo || regtry(reginfo, &s)) )
1589 PL_reg_flags |= RF_tainted;
1590 FBC_BOUND(isALNUM_LC,
1591 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1592 isALNUM_LC_utf8((U8*)s));
1595 PL_reg_flags |= RF_tainted;
1596 FBC_NBOUND(isALNUM_LC,
1597 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1598 isALNUM_LC_utf8((U8*)s));
1601 FBC_BOUND(isWORDCHAR,
1603 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1606 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1608 isWORDCHAR_A((U8*)s));
1611 FBC_NBOUND(isWORDCHAR,
1613 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1616 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1618 isWORDCHAR_A((U8*)s));
1621 FBC_BOUND(isWORDCHAR_L1,
1623 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1626 FBC_NBOUND(isWORDCHAR_L1,
1628 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1631 REXEC_FBC_CSCAN_TAINT(
1632 isALNUM_LC_utf8((U8*)s),
1637 REXEC_FBC_CSCAN_PRELOAD(
1638 LOAD_UTF8_CHARCLASS_ALNUM(),
1639 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1640 isWORDCHAR_L1((U8) *s)
1644 REXEC_FBC_CSCAN_PRELOAD(
1645 LOAD_UTF8_CHARCLASS_ALNUM(),
1646 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1651 /* Don't need to worry about utf8, as it can match only a single
1652 * byte invariant character */
1653 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1656 REXEC_FBC_CSCAN_PRELOAD(
1657 LOAD_UTF8_CHARCLASS_ALNUM(),
1658 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1659 ! isWORDCHAR_L1((U8) *s)
1663 REXEC_FBC_CSCAN_PRELOAD(
1664 LOAD_UTF8_CHARCLASS_ALNUM(),
1665 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1676 REXEC_FBC_CSCAN_TAINT(
1677 !isALNUM_LC_utf8((U8*)s),
1682 REXEC_FBC_CSCAN_PRELOAD(
1683 LOAD_UTF8_CHARCLASS_SPACE(),
1684 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1689 REXEC_FBC_CSCAN_PRELOAD(
1690 LOAD_UTF8_CHARCLASS_SPACE(),
1691 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1696 /* Don't need to worry about utf8, as it can match only a single
1697 * byte invariant character */
1698 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1701 REXEC_FBC_CSCAN_TAINT(
1702 isSPACE_LC_utf8((U8*)s),
1707 REXEC_FBC_CSCAN_PRELOAD(
1708 LOAD_UTF8_CHARCLASS_SPACE(),
1709 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1710 ! isSPACE_L1((U8) *s)
1714 REXEC_FBC_CSCAN_PRELOAD(
1715 LOAD_UTF8_CHARCLASS_SPACE(),
1716 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1727 REXEC_FBC_CSCAN_TAINT(
1728 !isSPACE_LC_utf8((U8*)s),
1733 REXEC_FBC_CSCAN_PRELOAD(
1734 LOAD_UTF8_CHARCLASS_DIGIT(),
1735 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1740 /* Don't need to worry about utf8, as it can match only a single
1741 * byte invariant character */
1742 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1745 REXEC_FBC_CSCAN_TAINT(
1746 isDIGIT_LC_utf8((U8*)s),
1751 REXEC_FBC_CSCAN_PRELOAD(
1752 LOAD_UTF8_CHARCLASS_DIGIT(),
1753 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1764 REXEC_FBC_CSCAN_TAINT(
1765 !isDIGIT_LC_utf8((U8*)s),
1772 is_LNBREAK_latin1(s)
1784 !is_VERTWS_latin1(s)
1790 is_HORIZWS_latin1(s)
1795 !is_HORIZWS_utf8(s),
1796 !is_HORIZWS_latin1(s)
1803 /* what trie are we using right now */
1805 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1807 = (reg_trie_data*)progi->data->data[ aho->trie ];
1808 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1810 const char *last_start = strend - trie->minlen;
1812 const char *real_start = s;
1814 STRLEN maxlen = trie->maxlen;
1816 U8 **points; /* map of where we were in the input string
1817 when reading a given char. For ASCII this
1818 is unnecessary overhead as the relationship
1819 is always 1:1, but for Unicode, especially
1820 case folded Unicode this is not true. */
1821 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1825 GET_RE_DEBUG_FLAGS_DECL;
1827 /* We can't just allocate points here. We need to wrap it in
1828 * an SV so it gets freed properly if there is a croak while
1829 * running the match */
1832 sv_points=newSV(maxlen * sizeof(U8 *));
1833 SvCUR_set(sv_points,
1834 maxlen * sizeof(U8 *));
1835 SvPOK_on(sv_points);
1836 sv_2mortal(sv_points);
1837 points=(U8**)SvPV_nolen(sv_points );
1838 if ( trie_type != trie_utf8_fold
1839 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1842 bitmap=(U8*)trie->bitmap;
1844 bitmap=(U8*)ANYOF_BITMAP(c);
1846 /* this is the Aho-Corasick algorithm modified a touch
1847 to include special handling for long "unknown char"
1848 sequences. The basic idea being that we use AC as long
1849 as we are dealing with a possible matching char, when
1850 we encounter an unknown char (and we have not encountered
1851 an accepting state) we scan forward until we find a legal
1853 AC matching is basically that of trie matching, except
1854 that when we encounter a failing transition, we fall back
1855 to the current states "fail state", and try the current char
1856 again, a process we repeat until we reach the root state,
1857 state 1, or a legal transition. If we fail on the root state
1858 then we can either terminate if we have reached an accepting
1859 state previously, or restart the entire process from the beginning
1863 while (s <= last_start) {
1864 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1872 U8 *uscan = (U8*)NULL;
1873 U8 *leftmost = NULL;
1875 U32 accepted_word= 0;
1879 while ( state && uc <= (U8*)strend ) {
1881 U32 word = aho->states[ state ].wordnum;
1885 DEBUG_TRIE_EXECUTE_r(
1886 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1887 dump_exec_pos( (char *)uc, c, strend, real_start,
1888 (char *)uc, utf8_target );
1889 PerlIO_printf( Perl_debug_log,
1890 " Scanning for legal start char...\n");
1894 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1898 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1904 if (uc >(U8*)last_start) break;
1908 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1909 if (!leftmost || lpos < leftmost) {
1910 DEBUG_r(accepted_word=word);
1916 points[pointpos++ % maxlen]= uc;
1917 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1918 uscan, len, uvc, charid, foldlen,
1920 DEBUG_TRIE_EXECUTE_r({
1921 dump_exec_pos( (char *)uc, c, strend, real_start,
1923 PerlIO_printf(Perl_debug_log,
1924 " Charid:%3u CP:%4"UVxf" ",
1930 word = aho->states[ state ].wordnum;
1932 base = aho->states[ state ].trans.base;
1934 DEBUG_TRIE_EXECUTE_r({
1936 dump_exec_pos( (char *)uc, c, strend, real_start,
1938 PerlIO_printf( Perl_debug_log,
1939 "%sState: %4"UVxf", word=%"UVxf,
1940 failed ? " Fail transition to " : "",
1941 (UV)state, (UV)word);
1947 ( ((offset = base + charid
1948 - 1 - trie->uniquecharcount)) >= 0)
1949 && ((U32)offset < trie->lasttrans)
1950 && trie->trans[offset].check == state
1951 && (tmp=trie->trans[offset].next))
1953 DEBUG_TRIE_EXECUTE_r(
1954 PerlIO_printf( Perl_debug_log," - legal\n"));
1959 DEBUG_TRIE_EXECUTE_r(
1960 PerlIO_printf( Perl_debug_log," - fail\n"));
1962 state = aho->fail[state];
1966 /* we must be accepting here */
1967 DEBUG_TRIE_EXECUTE_r(
1968 PerlIO_printf( Perl_debug_log," - accepting\n"));
1977 if (!state) state = 1;
1980 if ( aho->states[ state ].wordnum ) {
1981 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1982 if (!leftmost || lpos < leftmost) {
1983 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1988 s = (char*)leftmost;
1989 DEBUG_TRIE_EXECUTE_r({
1991 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1992 (UV)accepted_word, (IV)(s - real_start)
1995 if (!reginfo || regtry(reginfo, &s)) {
2001 DEBUG_TRIE_EXECUTE_r({
2002 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2005 DEBUG_TRIE_EXECUTE_r(
2006 PerlIO_printf( Perl_debug_log,"No match.\n"));
2015 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2025 - regexec_flags - match a regexp against a string
2028 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2029 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2030 /* strend: pointer to null at end of string */
2031 /* strbeg: real beginning of string */
2032 /* minend: end of match must be >=minend after stringarg. */
2033 /* data: May be used for some additional optimizations.
2034 Currently its only used, with a U32 cast, for transmitting
2035 the ganch offset when doing a /g match. This will change */
2036 /* nosave: For optimizations. */
2039 struct regexp *const prog = (struct regexp *)SvANY(rx);
2040 /*register*/ char *s;
2041 register regnode *c;
2042 /*register*/ char *startpos = stringarg;
2043 I32 minlen; /* must match at least this many chars */
2044 I32 dontbother = 0; /* how many characters not to try at end */
2045 I32 end_shift = 0; /* Same for the end. */ /* CC */
2046 I32 scream_pos = -1; /* Internal iterator of scream. */
2047 char *scream_olds = NULL;
2048 const bool utf8_target = cBOOL(DO_UTF8(sv));
2050 RXi_GET_DECL(prog,progi);
2051 regmatch_info reginfo; /* create some info to pass to regtry etc */
2052 regexp_paren_pair *swap = NULL;
2053 GET_RE_DEBUG_FLAGS_DECL;
2055 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2056 PERL_UNUSED_ARG(data);
2058 /* Be paranoid... */
2059 if (prog == NULL || startpos == NULL) {
2060 Perl_croak(aTHX_ "NULL regexp parameter");
2064 multiline = prog->extflags & RXf_PMf_MULTILINE;
2065 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2067 RX_MATCH_UTF8_set(rx, utf8_target);
2069 debug_start_match(rx, utf8_target, startpos, strend,
2073 minlen = prog->minlen;
2075 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2076 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2077 "String too short [regexec_flags]...\n"));
2082 /* Check validity of program. */
2083 if (UCHARAT(progi->program) != REG_MAGIC) {
2084 Perl_croak(aTHX_ "corrupted regexp program");
2088 PL_reg_eval_set = 0;
2092 PL_reg_flags |= RF_utf8;
2094 /* Mark beginning of line for ^ and lookbehind. */
2095 reginfo.bol = startpos; /* XXX not used ??? */
2099 /* Mark end of line for $ (and such) */
2102 /* see how far we have to get to not match where we matched before */
2103 reginfo.till = startpos+minend;
2105 /* If there is a "must appear" string, look for it. */
2108 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2110 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2111 reginfo.ganch = startpos + prog->gofs;
2112 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2113 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2114 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2116 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2117 && mg->mg_len >= 0) {
2118 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2119 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2120 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2122 if (prog->extflags & RXf_ANCH_GPOS) {
2123 if (s > reginfo.ganch)
2125 s = reginfo.ganch - prog->gofs;
2126 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2127 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2133 reginfo.ganch = strbeg + PTR2UV(data);
2134 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2135 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2137 } else { /* pos() not defined */
2138 reginfo.ganch = strbeg;
2139 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2140 "GPOS: reginfo.ganch = strbeg\n"));
2143 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2144 /* We have to be careful. If the previous successful match
2145 was from this regex we don't want a subsequent partially
2146 successful match to clobber the old results.
2147 So when we detect this possibility we add a swap buffer
2148 to the re, and switch the buffer each match. If we fail
2149 we switch it back, otherwise we leave it swapped.
2152 /* do we need a save destructor here for eval dies? */
2153 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2155 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2156 re_scream_pos_data d;
2158 d.scream_olds = &scream_olds;
2159 d.scream_pos = &scream_pos;
2160 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2162 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2163 goto phooey; /* not present */
2169 /* Simplest case: anchored match need be tried only once. */
2170 /* [unless only anchor is BOL and multiline is set] */
2171 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2172 if (s == startpos && regtry(®info, &startpos))
2174 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2175 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2180 dontbother = minlen - 1;
2181 end = HOP3c(strend, -dontbother, strbeg) - 1;
2182 /* for multiline we only have to try after newlines */
2183 if (prog->check_substr || prog->check_utf8) {
2184 /* because of the goto we can not easily reuse the macros for bifurcating the
2185 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2188 goto after_try_utf8;
2190 if (regtry(®info, &s)) {
2197 if (prog->extflags & RXf_USE_INTUIT) {
2198 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2207 } /* end search for check string in unicode */
2209 if (s == startpos) {
2210 goto after_try_latin;
2213 if (regtry(®info, &s)) {
2220 if (prog->extflags & RXf_USE_INTUIT) {
2221 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2230 } /* end search for check string in latin*/
2231 } /* end search for check string */
2232 else { /* search for newline */
2234 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2237 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2239 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2240 if (regtry(®info, &s))
2244 } /* end search for newline */
2245 } /* end anchored/multiline check string search */
2247 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2249 /* the warning about reginfo.ganch being used without initialization
2250 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2251 and we only enter this block when the same bit is set. */
2252 char *tmp_s = reginfo.ganch - prog->gofs;
2254 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2259 /* Messy cases: unanchored match. */
2260 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2261 /* we have /x+whatever/ */
2262 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2267 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2268 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2269 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2274 DEBUG_EXECUTE_r( did_match = 1 );
2275 if (regtry(®info, &s)) goto got_it;
2277 while (s < strend && *s == ch)
2285 DEBUG_EXECUTE_r( did_match = 1 );
2286 if (regtry(®info, &s)) goto got_it;
2288 while (s < strend && *s == ch)
2293 DEBUG_EXECUTE_r(if (!did_match)
2294 PerlIO_printf(Perl_debug_log,
2295 "Did not find anchored character...\n")
2298 else if (prog->anchored_substr != NULL
2299 || prog->anchored_utf8 != NULL
2300 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2301 && prog->float_max_offset < strend - s)) {
2306 char *last1; /* Last position checked before */
2310 if (prog->anchored_substr || prog->anchored_utf8) {
2311 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2312 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2313 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2314 back_max = back_min = prog->anchored_offset;
2316 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2317 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2318 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2319 back_max = prog->float_max_offset;
2320 back_min = prog->float_min_offset;
2324 if (must == &PL_sv_undef)
2325 /* could not downgrade utf8 check substring, so must fail */
2331 last = HOP3c(strend, /* Cannot start after this */
2332 -(I32)(CHR_SVLEN(must)
2333 - (SvTAIL(must) != 0) + back_min), strbeg);
2336 last1 = HOPc(s, -1);
2338 last1 = s - 1; /* bogus */
2340 /* XXXX check_substr already used to find "s", can optimize if
2341 check_substr==must. */
2343 dontbother = end_shift;
2344 strend = HOPc(strend, -dontbother);
2345 while ( (s <= last) &&
2346 ((flags & REXEC_SCREAM) && SvSCREAM(sv)
2347 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2348 end_shift, &scream_pos, 0))
2349 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2350 (unsigned char*)strend, must,
2351 multiline ? FBMrf_MULTILINE : 0))) ) {
2352 /* we may be pointing at the wrong string */
2353 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2354 s = strbeg + (s - SvPVX_const(sv));
2355 DEBUG_EXECUTE_r( did_match = 1 );
2356 if (HOPc(s, -back_max) > last1) {
2357 last1 = HOPc(s, -back_min);
2358 s = HOPc(s, -back_max);
2361 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2363 last1 = HOPc(s, -back_min);
2367 while (s <= last1) {
2368 if (regtry(®info, &s))
2374 while (s <= last1) {
2375 if (regtry(®info, &s))
2381 DEBUG_EXECUTE_r(if (!did_match) {
2382 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2383 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2384 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2385 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2386 ? "anchored" : "floating"),
2387 quoted, RE_SV_TAIL(must));
2391 else if ( (c = progi->regstclass) ) {
2393 const OPCODE op = OP(progi->regstclass);
2394 /* don't bother with what can't match */
2395 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2396 strend = HOPc(strend, -(minlen - 1));
2399 SV * const prop = sv_newmortal();
2400 regprop(prog, prop, c);
2402 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2404 PerlIO_printf(Perl_debug_log,
2405 "Matching stclass %.*s against %s (%d bytes)\n",
2406 (int)SvCUR(prop), SvPVX_const(prop),
2407 quoted, (int)(strend - s));
2410 if (find_byclass(prog, c, s, strend, ®info))
2412 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2416 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2421 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2422 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2423 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2425 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
2426 last = screaminstr(sv, float_real, s - strbeg,
2427 end_shift, &scream_pos, 1); /* last one */
2429 last = scream_olds; /* Only one occurrence. */
2430 /* we may be pointing at the wrong string */
2431 else if (RXp_MATCH_COPIED(prog))
2432 s = strbeg + (s - SvPVX_const(sv));
2436 const char * const little = SvPV_const(float_real, len);
2438 if (SvTAIL(float_real)) {
2439 if (memEQ(strend - len + 1, little, len - 1))
2440 last = strend - len + 1;
2441 else if (!multiline)
2442 last = memEQ(strend - len, little, len)
2443 ? strend - len : NULL;
2449 last = rninstr(s, strend, little, little + len);
2451 last = strend; /* matching "$" */
2456 PerlIO_printf(Perl_debug_log,
2457 "%sCan't trim the tail, match fails (should not happen)%s\n",
2458 PL_colors[4], PL_colors[5]));
2459 goto phooey; /* Should not happen! */
2461 dontbother = strend - last + prog->float_min_offset;
2463 if (minlen && (dontbother < minlen))
2464 dontbother = minlen - 1;
2465 strend -= dontbother; /* this one's always in bytes! */
2466 /* We don't know much -- general case. */
2469 if (regtry(®info, &s))
2478 if (regtry(®info, &s))
2480 } while (s++ < strend);
2489 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2491 if (PL_reg_eval_set)
2492 restore_pos(aTHX_ prog);
2493 if (RXp_PAREN_NAMES(prog))
2494 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2496 /* make sure $`, $&, $', and $digit will work later */
2497 if ( !(flags & REXEC_NOT_FIRST) ) {
2498 RX_MATCH_COPY_FREE(rx);
2499 if (flags & REXEC_COPY_STR) {
2500 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2501 #ifdef PERL_OLD_COPY_ON_WRITE
2503 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2505 PerlIO_printf(Perl_debug_log,
2506 "Copy on write: regexp capture, type %d\n",
2509 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2510 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2511 assert (SvPOKp(prog->saved_copy));
2515 RX_MATCH_COPIED_on(rx);
2516 s = savepvn(strbeg, i);
2522 prog->subbeg = strbeg;
2523 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2530 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2531 PL_colors[4], PL_colors[5]));
2532 if (PL_reg_eval_set)
2533 restore_pos(aTHX_ prog);
2535 /* we failed :-( roll it back */
2536 Safefree(prog->offs);
2545 - regtry - try match at specific point
2547 STATIC I32 /* 0 failure, 1 success */
2548 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2552 REGEXP *const rx = reginfo->prog;
2553 regexp *const prog = (struct regexp *)SvANY(rx);
2554 RXi_GET_DECL(prog,progi);
2555 GET_RE_DEBUG_FLAGS_DECL;
2557 PERL_ARGS_ASSERT_REGTRY;
2559 reginfo->cutpoint=NULL;
2561 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2564 PL_reg_eval_set = RS_init;
2565 DEBUG_EXECUTE_r(DEBUG_s(
2566 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2567 (IV)(PL_stack_sp - PL_stack_base));
2570 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2571 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2573 /* Apparently this is not needed, judging by wantarray. */
2574 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2575 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2578 /* Make $_ available to executed code. */
2579 if (reginfo->sv != DEFSV) {
2581 DEFSV_set(reginfo->sv);
2584 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2585 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2586 /* prepare for quick setting of pos */
2587 #ifdef PERL_OLD_COPY_ON_WRITE
2588 if (SvIsCOW(reginfo->sv))
2589 sv_force_normal_flags(reginfo->sv, 0);
2591 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2592 &PL_vtbl_mglob, NULL, 0);
2596 PL_reg_oldpos = mg->mg_len;
2597 SAVEDESTRUCTOR_X(restore_pos, prog);
2599 if (!PL_reg_curpm) {
2600 Newxz(PL_reg_curpm, 1, PMOP);
2603 SV* const repointer = &PL_sv_undef;
2604 /* this regexp is also owned by the new PL_reg_curpm, which
2605 will try to free it. */
2606 av_push(PL_regex_padav, repointer);
2607 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2608 PL_regex_pad = AvARRAY(PL_regex_padav);
2613 /* It seems that non-ithreads works both with and without this code.
2614 So for efficiency reasons it seems best not to have the code
2615 compiled when it is not needed. */
2616 /* This is safe against NULLs: */
2617 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2618 /* PM_reg_curpm owns a reference to this regexp. */
2619 (void)ReREFCNT_inc(rx);
2621 PM_SETRE(PL_reg_curpm, rx);
2622 PL_reg_oldcurpm = PL_curpm;
2623 PL_curpm = PL_reg_curpm;
2624 if (RXp_MATCH_COPIED(prog)) {
2625 /* Here is a serious problem: we cannot rewrite subbeg,
2626 since it may be needed if this match fails. Thus
2627 $` inside (?{}) could fail... */
2628 PL_reg_oldsaved = prog->subbeg;
2629 PL_reg_oldsavedlen = prog->sublen;
2630 #ifdef PERL_OLD_COPY_ON_WRITE
2631 PL_nrs = prog->saved_copy;
2633 RXp_MATCH_COPIED_off(prog);
2636 PL_reg_oldsaved = NULL;
2637 prog->subbeg = PL_bostr;
2638 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2640 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2641 prog->offs[0].start = *startpos - PL_bostr;
2642 PL_reginput = *startpos;
2643 PL_reglastparen = &prog->lastparen;
2644 PL_reglastcloseparen = &prog->lastcloseparen;
2645 prog->lastparen = 0;
2646 prog->lastcloseparen = 0;
2648 PL_regoffs = prog->offs;
2649 if (PL_reg_start_tmpl <= prog->nparens) {
2650 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2651 if(PL_reg_start_tmp)
2652 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2654 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2657 /* XXXX What this code is doing here?!!! There should be no need
2658 to do this again and again, PL_reglastparen should take care of
2661 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2662 * Actually, the code in regcppop() (which Ilya may be meaning by
2663 * PL_reglastparen), is not needed at all by the test suite
2664 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2665 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2666 * Meanwhile, this code *is* needed for the
2667 * above-mentioned test suite tests to succeed. The common theme
2668 * on those tests seems to be returning null fields from matches.
2669 * --jhi updated by dapm */
2671 if (prog->nparens) {
2672 regexp_paren_pair *pp = PL_regoffs;
2674 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2682 if (regmatch(reginfo, progi->program + 1)) {
2683 PL_regoffs[0].end = PL_reginput - PL_bostr;
2686 if (reginfo->cutpoint)
2687 *startpos= reginfo->cutpoint;
2688 REGCP_UNWIND(lastcp);
2693 #define sayYES goto yes
2694 #define sayNO goto no
2695 #define sayNO_SILENT goto no_silent
2697 /* we dont use STMT_START/END here because it leads to
2698 "unreachable code" warnings, which are bogus, but distracting. */
2699 #define CACHEsayNO \
2700 if (ST.cache_mask) \
2701 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2704 /* this is used to determine how far from the left messages like
2705 'failed...' are printed. It should be set such that messages
2706 are inline with the regop output that created them.
2708 #define REPORT_CODE_OFF 32
2711 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2712 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2714 #define SLAB_FIRST(s) (&(s)->states[0])
2715 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2717 /* grab a new slab and return the first slot in it */
2719 STATIC regmatch_state *
2722 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2725 regmatch_slab *s = PL_regmatch_slab->next;
2727 Newx(s, 1, regmatch_slab);
2728 s->prev = PL_regmatch_slab;
2730 PL_regmatch_slab->next = s;
2732 PL_regmatch_slab = s;
2733 return SLAB_FIRST(s);
2737 /* push a new state then goto it */
2739 #define PUSH_STATE_GOTO(state, node) \
2741 st->resume_state = state; \
2744 /* push a new state with success backtracking, then goto it */
2746 #define PUSH_YES_STATE_GOTO(state, node) \
2748 st->resume_state = state; \
2749 goto push_yes_state;
2755 regmatch() - main matching routine
2757 This is basically one big switch statement in a loop. We execute an op,
2758 set 'next' to point the next op, and continue. If we come to a point which
2759 we may need to backtrack to on failure such as (A|B|C), we push a
2760 backtrack state onto the backtrack stack. On failure, we pop the top
2761 state, and re-enter the loop at the state indicated. If there are no more
2762 states to pop, we return failure.
2764 Sometimes we also need to backtrack on success; for example /A+/, where
2765 after successfully matching one A, we need to go back and try to
2766 match another one; similarly for lookahead assertions: if the assertion
2767 completes successfully, we backtrack to the state just before the assertion
2768 and then carry on. In these cases, the pushed state is marked as
2769 'backtrack on success too'. This marking is in fact done by a chain of
2770 pointers, each pointing to the previous 'yes' state. On success, we pop to
2771 the nearest yes state, discarding any intermediate failure-only states.
2772 Sometimes a yes state is pushed just to force some cleanup code to be
2773 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2774 it to free the inner regex.
2776 Note that failure backtracking rewinds the cursor position, while
2777 success backtracking leaves it alone.
2779 A pattern is complete when the END op is executed, while a subpattern
2780 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2781 ops trigger the "pop to last yes state if any, otherwise return true"
2784 A common convention in this function is to use A and B to refer to the two
2785 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2786 the subpattern to be matched possibly multiple times, while B is the entire
2787 rest of the pattern. Variable and state names reflect this convention.
2789 The states in the main switch are the union of ops and failure/success of
2790 substates associated with with that op. For example, IFMATCH is the op
2791 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2792 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2793 successfully matched A and IFMATCH_A_fail is a state saying that we have
2794 just failed to match A. Resume states always come in pairs. The backtrack
2795 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2796 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2797 on success or failure.
2799 The struct that holds a backtracking state is actually a big union, with
2800 one variant for each major type of op. The variable st points to the
2801 top-most backtrack struct. To make the code clearer, within each
2802 block of code we #define ST to alias the relevant union.
2804 Here's a concrete example of a (vastly oversimplified) IFMATCH
2810 #define ST st->u.ifmatch
2812 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2813 ST.foo = ...; // some state we wish to save
2815 // push a yes backtrack state with a resume value of
2816 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2818 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2821 case IFMATCH_A: // we have successfully executed A; now continue with B
2823 bar = ST.foo; // do something with the preserved value
2826 case IFMATCH_A_fail: // A failed, so the assertion failed
2827 ...; // do some housekeeping, then ...
2828 sayNO; // propagate the failure
2835 For any old-timers reading this who are familiar with the old recursive
2836 approach, the code above is equivalent to:
2838 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2847 ...; // do some housekeeping, then ...
2848 sayNO; // propagate the failure
2851 The topmost backtrack state, pointed to by st, is usually free. If you
2852 want to claim it, populate any ST.foo fields in it with values you wish to
2853 save, then do one of
2855 PUSH_STATE_GOTO(resume_state, node);
2856 PUSH_YES_STATE_GOTO(resume_state, node);
2858 which sets that backtrack state's resume value to 'resume_state', pushes a
2859 new free entry to the top of the backtrack stack, then goes to 'node'.
2860 On backtracking, the free slot is popped, and the saved state becomes the
2861 new free state. An ST.foo field in this new top state can be temporarily
2862 accessed to retrieve values, but once the main loop is re-entered, it
2863 becomes available for reuse.
2865 Note that the depth of the backtrack stack constantly increases during the
2866 left-to-right execution of the pattern, rather than going up and down with
2867 the pattern nesting. For example the stack is at its maximum at Z at the
2868 end of the pattern, rather than at X in the following:
2870 /(((X)+)+)+....(Y)+....Z/
2872 The only exceptions to this are lookahead/behind assertions and the cut,
2873 (?>A), which pop all the backtrack states associated with A before
2876 Backtrack state structs are allocated in slabs of about 4K in size.
2877 PL_regmatch_state and st always point to the currently active state,
2878 and PL_regmatch_slab points to the slab currently containing
2879 PL_regmatch_state. The first time regmatch() is called, the first slab is
2880 allocated, and is never freed until interpreter destruction. When the slab
2881 is full, a new one is allocated and chained to the end. At exit from
2882 regmatch(), slabs allocated since entry are freed.
2887 #define DEBUG_STATE_pp(pp) \
2889 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2890 PerlIO_printf(Perl_debug_log, \
2891 " %*s"pp" %s%s%s%s%s\n", \
2893 PL_reg_name[st->resume_state], \
2894 ((st==yes_state||st==mark_state) ? "[" : ""), \
2895 ((st==yes_state) ? "Y" : ""), \
2896 ((st==mark_state) ? "M" : ""), \
2897 ((st==yes_state||st==mark_state) ? "]" : "") \
2902 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2907 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2908 const char *start, const char *end, const char *blurb)
2910 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2912 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2917 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2918 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2920 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2921 start, end - start, 60);
2923 PerlIO_printf(Perl_debug_log,
2924 "%s%s REx%s %s against %s\n",
2925 PL_colors[4], blurb, PL_colors[5], s0, s1);
2927 if (utf8_target||utf8_pat)
2928 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2929 utf8_pat ? "pattern" : "",
2930 utf8_pat && utf8_target ? " and " : "",
2931 utf8_target ? "string" : ""
2937 S_dump_exec_pos(pTHX_ const char *locinput,
2938 const regnode *scan,
2939 const char *loc_regeol,
2940 const char *loc_bostr,
2941 const char *loc_reg_starttry,
2942 const bool utf8_target)
2944 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2945 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2946 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2947 /* The part of the string before starttry has one color
2948 (pref0_len chars), between starttry and current
2949 position another one (pref_len - pref0_len chars),
2950 after the current position the third one.
2951 We assume that pref0_len <= pref_len, otherwise we
2952 decrease pref0_len. */
2953 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2954 ? (5 + taill) - l : locinput - loc_bostr;
2957 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2959 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2961 pref0_len = pref_len - (locinput - loc_reg_starttry);
2962 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2963 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2964 ? (5 + taill) - pref_len : loc_regeol - locinput);
2965 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2969 if (pref0_len > pref_len)
2970 pref0_len = pref_len;
2972 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2974 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2975 (locinput - pref_len),pref0_len, 60, 4, 5);
2977 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2978 (locinput - pref_len + pref0_len),
2979 pref_len - pref0_len, 60, 2, 3);
2981 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2982 locinput, loc_regeol - locinput, 10, 0, 1);
2984 const STRLEN tlen=len0+len1+len2;
2985 PerlIO_printf(Perl_debug_log,
2986 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2987 (IV)(locinput - loc_bostr),
2990 (docolor ? "" : "> <"),
2992 (int)(tlen > 19 ? 0 : 19 - tlen),
2999 /* reg_check_named_buff_matched()
3000 * Checks to see if a named buffer has matched. The data array of
3001 * buffer numbers corresponding to the buffer is expected to reside
3002 * in the regexp->data->data array in the slot stored in the ARG() of
3003 * node involved. Note that this routine doesn't actually care about the
3004 * name, that information is not preserved from compilation to execution.
3005 * Returns the index of the leftmost defined buffer with the given name
3006 * or 0 if non of the buffers matched.
3009 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3012 RXi_GET_DECL(rex,rexi);
3013 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3014 I32 *nums=(I32*)SvPVX(sv_dat);
3016 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3018 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3019 if ((I32)*PL_reglastparen >= nums[n] &&
3020 PL_regoffs[nums[n]].end != -1)
3029 /* free all slabs above current one - called during LEAVE_SCOPE */
3032 S_clear_backtrack_stack(pTHX_ void *p)
3034 regmatch_slab *s = PL_regmatch_slab->next;
3039 PL_regmatch_slab->next = NULL;
3041 regmatch_slab * const osl = s;
3048 #define SETREX(Re1,Re2) \
3049 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3052 STATIC I32 /* 0 failure, 1 success */
3053 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3055 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3059 register const bool utf8_target = PL_reg_match_utf8;
3060 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3061 REGEXP *rex_sv = reginfo->prog;
3062 regexp *rex = (struct regexp *)SvANY(rex_sv);
3063 RXi_GET_DECL(rex,rexi);
3065 /* the current state. This is a cached copy of PL_regmatch_state */
3066 register regmatch_state *st;
3067 /* cache heavy used fields of st in registers */
3068 register regnode *scan;
3069 register regnode *next;
3070 register U32 n = 0; /* general value; init to avoid compiler warning */
3071 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3072 register char *locinput = PL_reginput;
3073 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3075 bool result = 0; /* return value of S_regmatch */
3076 int depth = 0; /* depth of backtrack stack */
3077 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3078 const U32 max_nochange_depth =
3079 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3080 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3081 regmatch_state *yes_state = NULL; /* state to pop to on success of
3083 /* mark_state piggy backs on the yes_state logic so that when we unwind
3084 the stack on success we can update the mark_state as we go */
3085 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3086 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3087 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3089 bool no_final = 0; /* prevent failure from backtracking? */
3090 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3091 char *startpoint = PL_reginput;
3092 SV *popmark = NULL; /* are we looking for a mark? */
3093 SV *sv_commit = NULL; /* last mark name seen in failure */
3094 SV *sv_yes_mark = NULL; /* last mark name we have seen
3095 during a successful match */
3096 U32 lastopen = 0; /* last open we saw */
3097 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3098 SV* const oreplsv = GvSV(PL_replgv);
3099 /* these three flags are set by various ops to signal information to
3100 * the very next op. They have a useful lifetime of exactly one loop
3101 * iteration, and are not preserved or restored by state pushes/pops
3103 bool sw = 0; /* the condition value in (?(cond)a|b) */
3104 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3105 int logical = 0; /* the following EVAL is:
3109 or the following IFMATCH/UNLESSM is:
3110 false: plain (?=foo)
3111 true: used as a condition: (?(?=foo))
3114 GET_RE_DEBUG_FLAGS_DECL;
3117 PERL_ARGS_ASSERT_REGMATCH;
3119 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3120 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3122 /* on first ever call to regmatch, allocate first slab */
3123 if (!PL_regmatch_slab) {
3124 Newx(PL_regmatch_slab, 1, regmatch_slab);
3125 PL_regmatch_slab->prev = NULL;
3126 PL_regmatch_slab->next = NULL;
3127 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3130 oldsave = PL_savestack_ix;
3131 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3132 SAVEVPTR(PL_regmatch_slab);
3133 SAVEVPTR(PL_regmatch_state);
3135 /* grab next free state slot */
3136 st = ++PL_regmatch_state;
3137 if (st > SLAB_LAST(PL_regmatch_slab))
3138 st = PL_regmatch_state = S_push_slab(aTHX);
3140 /* Note that nextchr is a byte even in UTF */
3141 nextchr = UCHARAT(locinput);
3143 while (scan != NULL) {
3146 SV * const prop = sv_newmortal();
3147 regnode *rnext=regnext(scan);
3148 DUMP_EXEC_POS( locinput, scan, utf8_target );
3149 regprop(rex, prop, scan);
3151 PerlIO_printf(Perl_debug_log,
3152 "%3"IVdf":%*s%s(%"IVdf")\n",
3153 (IV)(scan - rexi->program), depth*2, "",
3155 (PL_regkind[OP(scan)] == END || !rnext) ?
3156 0 : (IV)(rnext - rexi->program));
3159 next = scan + NEXT_OFF(scan);
3162 state_num = OP(scan);
3164 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3167 assert(PL_reglastparen == &rex->lastparen);
3168 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3169 assert(PL_regoffs == rex->offs);
3171 switch (state_num) {
3173 if (locinput == PL_bostr)
3175 /* reginfo->till = reginfo->bol; */
3180 if (locinput == PL_bostr ||
3181 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3187 if (locinput == PL_bostr)
3191 if (locinput == reginfo->ganch)
3196 /* update the startpoint */
3197 st->u.keeper.val = PL_regoffs[0].start;
3198 PL_reginput = locinput;
3199 PL_regoffs[0].start = locinput - PL_bostr;
3200 PUSH_STATE_GOTO(KEEPS_next, next);
3202 case KEEPS_next_fail:
3203 /* rollback the start point change */
3204 PL_regoffs[0].start = st->u.keeper.val;
3210 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3215 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3217 if (PL_regeol - locinput > 1)
3221 if (PL_regeol != locinput)
3225 if (!nextchr && locinput >= PL_regeol)
3228 locinput += PL_utf8skip[nextchr];
3229 if (locinput > PL_regeol)
3231 nextchr = UCHARAT(locinput);
3234 nextchr = UCHARAT(++locinput);
3237 if (!nextchr && locinput >= PL_regeol)
3239 nextchr = UCHARAT(++locinput);
3242 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3245 locinput += PL_utf8skip[nextchr];
3246 if (locinput > PL_regeol)
3248 nextchr = UCHARAT(locinput);
3251 nextchr = UCHARAT(++locinput);
3255 #define ST st->u.trie
3257 /* In this case the charclass data is available inline so
3258 we can fail fast without a lot of extra overhead.
3260 if (scan->flags == EXACT || !utf8_target) {
3261 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3263 PerlIO_printf(Perl_debug_log,
3264 "%*s %sfailed to match trie start class...%s\n",
3265 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3273 /* the basic plan of execution of the trie is:
3274 * At the beginning, run though all the states, and
3275 * find the longest-matching word. Also remember the position
3276 * of the shortest matching word. For example, this pattern:
3279 * when matched against the string "abcde", will generate
3280 * accept states for all words except 3, with the longest
3281 * matching word being 4, and the shortest being 1 (with
3282 * the position being after char 1 of the string).
3284 * Then for each matching word, in word order (i.e. 1,2,4,5),
3285 * we run the remainder of the pattern; on each try setting
3286 * the current position to the character following the word,
3287 * returning to try the next word on failure.
3289 * We avoid having to build a list of words at runtime by
3290 * using a compile-time structure, wordinfo[].prev, which
3291 * gives, for each word, the previous accepting word (if any).
3292 * In the case above it would contain the mappings 1->2, 2->0,
3293 * 3->0, 4->5, 5->1. We can use this table to generate, from
3294 * the longest word (4 above), a list of all words, by
3295 * following the list of prev pointers; this gives us the
3296 * unordered list 4,5,1,2. Then given the current word we have
3297 * just tried, we can go through the list and find the
3298 * next-biggest word to try (so if we just failed on word 2,
3299 * the next in the list is 4).
3301 * Since at runtime we don't record the matching position in
3302 * the string for each word, we have to work that out for
3303 * each word we're about to process. The wordinfo table holds
3304 * the character length of each word; given that we recorded
3305 * at the start: the position of the shortest word and its
3306 * length in chars, we just need to move the pointer the
3307 * difference between the two char lengths. Depending on
3308 * Unicode status and folding, that's cheap or expensive.
3310 * This algorithm is optimised for the case where are only a
3311 * small number of accept states, i.e. 0,1, or maybe 2.
3312 * With lots of accepts states, and having to try all of them,
3313 * it becomes quadratic on number of accept states to find all
3318 /* what type of TRIE am I? (utf8 makes this contextual) */
3319 DECL_TRIE_TYPE(scan);
3321 /* what trie are we using right now */
3322 reg_trie_data * const trie
3323 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3324 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3325 U32 state = trie->startstate;
3327 if (trie->bitmap && trie_type != trie_utf8_fold &&
3328 !TRIE_BITMAP_TEST(trie,*locinput)
3330 if (trie->states[ state ].wordnum) {
3332 PerlIO_printf(Perl_debug_log,
3333 "%*s %smatched empty string...%s\n",
3334 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3340 PerlIO_printf(Perl_debug_log,
3341 "%*s %sfailed to match trie start class...%s\n",
3342 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3349 U8 *uc = ( U8* )locinput;
3353 U8 *uscan = (U8*)NULL;
3354 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3355 U32 charcount = 0; /* how many input chars we have matched */
3356 U32 accepted = 0; /* have we seen any accepting states? */
3359 ST.jump = trie->jump;
3362 ST.longfold = FALSE; /* char longer if folded => it's harder */
3365 /* fully traverse the TRIE; note the position of the
3366 shortest accept state and the wordnum of the longest
3369 while ( state && uc <= (U8*)PL_regeol ) {
3370 U32 base = trie->states[ state ].trans.base;
3374 wordnum = trie->states[ state ].wordnum;
3376 if (wordnum) { /* it's an accept state */
3379 /* record first match position */
3381 ST.firstpos = (U8*)locinput;
3386 ST.firstchars = charcount;
3389 if (!ST.nextword || wordnum < ST.nextword)
3390 ST.nextword = wordnum;
3391 ST.topword = wordnum;
3394 DEBUG_TRIE_EXECUTE_r({
3395 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3396 PerlIO_printf( Perl_debug_log,
3397 "%*s %sState: %4"UVxf" Accepted: %c ",
3398 2+depth * 2, "", PL_colors[4],
3399 (UV)state, (accepted ? 'Y' : 'N'));
3402 /* read a char and goto next state */
3405 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3406 uscan, len, uvc, charid, foldlen,
3413 base + charid - 1 - trie->uniquecharcount)) >= 0)
3415 && ((U32)offset < trie->lasttrans)
3416 && trie->trans[offset].check == state)
3418 state = trie->trans[offset].next;
3429 DEBUG_TRIE_EXECUTE_r(
3430 PerlIO_printf( Perl_debug_log,
3431 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3432 charid, uvc, (UV)state, PL_colors[5] );
3438 /* calculate total number of accept states */
3443 w = trie->wordinfo[w].prev;
3446 ST.accepted = accepted;
3450 PerlIO_printf( Perl_debug_log,
3451 "%*s %sgot %"IVdf" possible matches%s\n",
3452 REPORT_CODE_OFF + depth * 2, "",
3453 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3455 goto trie_first_try; /* jump into the fail handler */
3459 case TRIE_next_fail: /* we failed - try next alternative */
3461 REGCP_UNWIND(ST.cp);
3462 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3463 PL_regoffs[n].end = -1;
3464 *PL_reglastparen = n;
3466 if (!--ST.accepted) {
3468 PerlIO_printf( Perl_debug_log,
3469 "%*s %sTRIE failed...%s\n",
3470 REPORT_CODE_OFF+depth*2, "",
3477 /* Find next-highest word to process. Note that this code
3478 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3479 register U16 min = 0;
3481 register U16 const nextword = ST.nextword;
3482 register reg_trie_wordinfo * const wordinfo
3483 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3484 for (word=ST.topword; word; word=wordinfo[word].prev) {
3485 if (word > nextword && (!min || word < min))
3498 ST.lastparen = *PL_reglastparen;
3502 /* find start char of end of current word */
3504 U32 chars; /* how many chars to skip */
3505 U8 *uc = ST.firstpos;
3506 reg_trie_data * const trie
3507 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3509 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3511 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3515 /* the hard option - fold each char in turn and find
3516 * its folded length (which may be different */
3517 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3525 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3533 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3538 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3552 PL_reginput = (char *)uc;
3555 scan = (ST.jump && ST.jump[ST.nextword])
3556 ? ST.me + ST.jump[ST.nextword]
3560 PerlIO_printf( Perl_debug_log,
3561 "%*s %sTRIE matched word #%d, continuing%s\n",
3562 REPORT_CODE_OFF+depth*2, "",
3569 if (ST.accepted > 1 || has_cutgroup) {
3570 PUSH_STATE_GOTO(TRIE_next, scan);
3573 /* only one choice left - just continue */
3575 AV *const trie_words
3576 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3577 SV ** const tmp = av_fetch( trie_words,
3579 SV *sv= tmp ? sv_newmortal() : NULL;
3581 PerlIO_printf( Perl_debug_log,
3582 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3583 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3585 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3586 PL_colors[0], PL_colors[1],
3587 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3589 : "not compiled under -Dr",
3593 locinput = PL_reginput;
3594 nextchr = UCHARAT(locinput);
3595 continue; /* execute rest of RE */
3600 char *s = STRING(scan);
3602 if (utf8_target != UTF_PATTERN) {
3603 /* The target and the pattern have differing utf8ness. */
3605 const char * const e = s + ln;
3608 /* The target is utf8, the pattern is not utf8. */
3613 if (NATIVE_TO_UNI(*(U8*)s) !=
3614 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3622 /* The target is not utf8, the pattern is utf8. */
3627 if (NATIVE_TO_UNI(*((U8*)l)) !=
3628 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3636 nextchr = UCHARAT(locinput);
3639 /* The target and the pattern have the same utf8ness. */
3640 /* Inline the first character, for speed. */
3641 if (UCHARAT(s) != nextchr)
3643 if (PL_regeol - locinput < ln)
3645 if (ln > 1 && memNE(s, locinput, ln))
3648 nextchr = UCHARAT(locinput);
3653 const U8 * fold_array;
3655 U32 fold_utf8_flags;
3657 PL_reg_flags |= RF_tainted;
3658 folder = foldEQ_locale;
3659 fold_array = PL_fold_locale;
3660 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3664 folder = foldEQ_latin1;
3665 fold_array = PL_fold_latin1;
3666 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3670 folder = foldEQ_latin1;
3671 fold_array = PL_fold_latin1;
3672 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3677 fold_array = PL_fold;
3678 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3684 if (utf8_target || UTF_PATTERN) {
3685 /* Either target or the pattern are utf8. */
3686 const char * const l = locinput;
3687 char *e = PL_regeol;
3689 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3690 l, &e, 0, utf8_target, fold_utf8_flags))
3695 nextchr = UCHARAT(locinput);
3699 /* Neither the target nor the pattern are utf8 */
3700 if (UCHARAT(s) != nextchr &&
3701 UCHARAT(s) != fold_array[nextchr])
3705 if (PL_regeol - locinput < ln)
3707 if (ln > 1 && ! folder(s, locinput, ln))
3710 nextchr = UCHARAT(locinput);
3714 /* XXX Could improve efficiency by separating these all out using a
3715 * macro or in-line function. At that point regcomp.c would no longer
3716 * have to set the FLAGS fields of these */
3719 PL_reg_flags |= RF_tainted;
3727 /* was last char in word? */
3729 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3730 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3732 if (locinput == PL_bostr)
3735 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3737 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3739 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3740 ln = isALNUM_uni(ln);
3741 LOAD_UTF8_CHARCLASS_ALNUM();
3742 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3745 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3746 n = isALNUM_LC_utf8((U8*)locinput);
3751 /* Here the string isn't utf8, or is utf8 and only ascii
3752 * characters are to match \w. In the latter case looking at
3753 * the byte just prior to the current one may be just the final
3754 * byte of a multi-byte character. This is ok. There are two
3756 * 1) it is a single byte character, and then the test is doing
3757 * just what it's supposed to.
3758 * 2) it is a multi-byte character, in which case the final
3759 * byte is never mistakable for ASCII, and so the test
3760 * will say it is not a word character, which is the
3761 * correct answer. */
3762 ln = (locinput != PL_bostr) ?
3763 UCHARAT(locinput - 1) : '\n';
3764 switch (FLAGS(scan)) {
3765 case REGEX_UNICODE_CHARSET:
3766 ln = isWORDCHAR_L1(ln);
3767 n = isWORDCHAR_L1(nextchr);
3769 case REGEX_LOCALE_CHARSET:
3770 ln = isALNUM_LC(ln);
3771 n = isALNUM_LC(nextchr);
3773 case REGEX_DEPENDS_CHARSET:
3775 n = isALNUM(nextchr);
3777 case REGEX_ASCII_RESTRICTED_CHARSET:
3778 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3779 ln = isWORDCHAR_A(ln);
3780 n = isWORDCHAR_A(nextchr);
3783 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3787 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3789 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3794 if (utf8_target || state_num == ANYOFV) {
3795 STRLEN inclasslen = PL_regeol - locinput;
3796 if (locinput >= PL_regeol)
3799 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3801 locinput += inclasslen;
3802 nextchr = UCHARAT(locinput);
3807 nextchr = UCHARAT(locinput);
3808 if (!nextchr && locinput >= PL_regeol)
3810 if (!REGINCLASS(rex, scan, (U8*)locinput))
3812 nextchr = UCHARAT(++locinput);
3816 /* Special char classes - The defines start on line 129 or so */
3817 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3818 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3819 ALNUMU, NALNUMU, isWORDCHAR_L1,
3820 ALNUMA, NALNUMA, isWORDCHAR_A,
3823 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3824 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3825 SPACEU, NSPACEU, isSPACE_L1,
3826 SPACEA, NSPACEA, isSPACE_A,
3829 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3830 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3831 DIGITA, NDIGITA, isDIGIT_A,
3834 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3835 a Unicode extended Grapheme Cluster */
3836 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3837 extended Grapheme Cluster is:
3840 | Prepend* Begin Extend*
3843 Begin is (Hangul-syllable | ! Control)
3844 Extend is (Grapheme_Extend | Spacing_Mark)
3845 Control is [ GCB_Control CR LF ]
3847 The discussion below shows how the code for CLUMP is derived
3848 from this regex. Note that most of these concepts are from
3849 property values of the Grapheme Cluster Boundary (GCB) property.
3850 No code point can have multiple property values for a given
3851 property. Thus a code point in Prepend can't be in Control, but
3852 it must be in !Control. This is why Control above includes
3853 GCB_Control plus CR plus LF. The latter two are used in the GCB
3854 property separately, and so can't be in GCB_Control, even though
3855 they logically are controls. Control is not the same as gc=cc,
3856 but includes format and other characters as well.
3858 The Unicode definition of Hangul-syllable is:
3860 | (L* ( ( V | LV ) V* | LVT ) T*)
3863 Each of these is a value for the GCB property, and hence must be
3864 disjoint, so the order they are tested is immaterial, so the
3865 above can safely be changed to
3868 | (L* ( LVT | ( V | LV ) V*) T*)
3870 The last two terms can be combined like this:
3872 | (( LVT | ( V | LV ) V*) T*))
3874 And refactored into this:
3875 L* (L | LVT T* | V V* T* | LV V* T*)
3877 That means that if we have seen any L's at all we can quit
3878 there, but if the next character is an LVT, a V, or an LV we
3881 There is a subtlety with Prepend* which showed up in testing.
3882 Note that the Begin, and only the Begin is required in:
3883 | Prepend* Begin Extend*
3884 Also, Begin contains '! Control'. A Prepend must be a
3885 '! Control', which means it must also be a Begin. What it
3886 comes down to is that if we match Prepend* and then find no
3887 suitable Begin afterwards, that if we backtrack the last
3888 Prepend, that one will be a suitable Begin.
3891 if (locinput >= PL_regeol)
3893 if (! utf8_target) {
3895 /* Match either CR LF or '.', as all the other possibilities
3897 locinput++; /* Match the . or CR */
3898 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3900 && locinput < PL_regeol
3901 && UCHARAT(locinput) == '\n') locinput++;
3905 /* Utf8: See if is ( CR LF ); already know that locinput <
3906 * PL_regeol, so locinput+1 is in bounds */
3907 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3911 /* In case have to backtrack to beginning, then match '.' */
3912 char *starting = locinput;
3914 /* In case have to backtrack the last prepend */
3915 char *previous_prepend = 0;
3917 LOAD_UTF8_CHARCLASS_GCB();
3919 /* Match (prepend)* */
3920 while (locinput < PL_regeol
3921 && swash_fetch(PL_utf8_X_prepend,
3922 (U8*)locinput, utf8_target))
3924 previous_prepend = locinput;
3925 locinput += UTF8SKIP(locinput);
3928 /* As noted above, if we matched a prepend character, but
3929 * the next thing won't match, back off the last prepend we
3930 * matched, as it is guaranteed to match the begin */
3931 if (previous_prepend
3932 && (locinput >= PL_regeol
3933 || ! swash_fetch(PL_utf8_X_begin,
3934 (U8*)locinput, utf8_target)))
3936 locinput = previous_prepend;
3939 /* Note that here we know PL_regeol > locinput, as we
3940 * tested that upon input to this switch case, and if we
3941 * moved locinput forward, we tested the result just above
3942 * and it either passed, or we backed off so that it will
3944 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3946 /* Here did not match the required 'Begin' in the
3947 * second term. So just match the very first
3948 * character, the '.' of the final term of the regex */
3949 locinput = starting + UTF8SKIP(starting);
3952 /* Here is the beginning of a character that can have
3953 * an extender. It is either a hangul syllable, or a
3955 if (swash_fetch(PL_utf8_X_non_hangul,
3956 (U8*)locinput, utf8_target))
3959 /* Here not a Hangul syllable, must be a
3960 * ('! * Control') */
3961 locinput += UTF8SKIP(locinput);
3964 /* Here is a Hangul syllable. It can be composed
3965 * of several individual characters. One
3966 * possibility is T+ */
3967 if (swash_fetch(PL_utf8_X_T,
3968 (U8*)locinput, utf8_target))
3970 while (locinput < PL_regeol
3971 && swash_fetch(PL_utf8_X_T,
3972 (U8*)locinput, utf8_target))
3974 locinput += UTF8SKIP(locinput);
3978 /* Here, not T+, but is a Hangul. That means
3979 * it is one of the others: L, LV, LVT or V,
3981 * L* (L | LVT T* | V V* T* | LV V* T*) */
3984 while (locinput < PL_regeol
3985 && swash_fetch(PL_utf8_X_L,
3986 (U8*)locinput, utf8_target))
3988 locinput += UTF8SKIP(locinput);
3991 /* Here, have exhausted L*. If the next
3992 * character is not an LV, LVT nor V, it means
3993 * we had to have at least one L, so matches L+
3994 * in the original equation, we have a complete
3995 * hangul syllable. Are done. */
3997 if (locinput < PL_regeol
3998 && swash_fetch(PL_utf8_X_LV_LVT_V,
3999 (U8*)locinput, utf8_target))
4002 /* Otherwise keep going. Must be LV, LVT
4003 * or V. See if LVT */
4004 if (swash_fetch(PL_utf8_X_LVT,
4005 (U8*)locinput, utf8_target))
4007 locinput += UTF8SKIP(locinput);
4010 /* Must be V or LV. Take it, then
4012 locinput += UTF8SKIP(locinput);
4013 while (locinput < PL_regeol
4014 && swash_fetch(PL_utf8_X_V,
4015 (U8*)locinput, utf8_target))
4017 locinput += UTF8SKIP(locinput);
4021 /* And any of LV, LVT, or V can be followed
4023 while (locinput < PL_regeol
4024 && swash_fetch(PL_utf8_X_T,
4028 locinput += UTF8SKIP(locinput);
4034 /* Match any extender */
4035 while (locinput < PL_regeol
4036 && swash_fetch(PL_utf8_X_extend,
4037 (U8*)locinput, utf8_target))
4039 locinput += UTF8SKIP(locinput);
4043 if (locinput > PL_regeol) sayNO;
4045 nextchr = UCHARAT(locinput);
4049 { /* The capture buffer cases. The ones beginning with N for the
4050 named buffers just convert to the equivalent numbered and
4051 pretend they were called as the corresponding numbered buffer
4053 /* don't initialize these in the declaration, it makes C++
4058 const U8 *fold_array;
4061 PL_reg_flags |= RF_tainted;
4062 folder = foldEQ_locale;
4063 fold_array = PL_fold_locale;
4065 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4069 folder = foldEQ_latin1;
4070 fold_array = PL_fold_latin1;
4072 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4076 folder = foldEQ_latin1;
4077 fold_array = PL_fold_latin1;
4079 utf8_fold_flags = 0;
4084 fold_array = PL_fold;
4086 utf8_fold_flags = 0;
4093 utf8_fold_flags = 0;
4096 /* For the named back references, find the corresponding buffer
4098 n = reg_check_named_buff_matched(rex,scan);
4103 goto do_nref_ref_common;
4106 PL_reg_flags |= RF_tainted;
4107 folder = foldEQ_locale;
4108 fold_array = PL_fold_locale;
4109 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4113 folder = foldEQ_latin1;
4114 fold_array = PL_fold_latin1;
4115 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4119 folder = foldEQ_latin1;
4120 fold_array = PL_fold_latin1;
4121 utf8_fold_flags = 0;
4126 fold_array = PL_fold;
4127 utf8_fold_flags = 0;
4133 utf8_fold_flags = 0;
4137 n = ARG(scan); /* which paren pair */
4140 ln = PL_regoffs[n].start;
4141 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4142 if (*PL_reglastparen < n || ln == -1)
4143 sayNO; /* Do not match unless seen CLOSEn. */
4144 if (ln == PL_regoffs[n].end)
4148 if (type != REF /* REF can do byte comparison */
4149 && (utf8_target || type == REFFU))
4150 { /* XXX handle REFFL better */
4151 char * limit = PL_regeol;
4153 /* This call case insensitively compares the entire buffer
4154 * at s, with the current input starting at locinput, but
4155 * not going off the end given by PL_regeol, and returns in
4156 * limit upon success, how much of the current input was
4158 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4159 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4164 nextchr = UCHARAT(locinput);
4168 /* Not utf8: Inline the first character, for speed. */
4169 if (UCHARAT(s) != nextchr &&
4171 UCHARAT(s) != fold_array[nextchr]))
4173 ln = PL_regoffs[n].end - ln;
4174 if (locinput + ln > PL_regeol)
4176 if (ln > 1 && (type == REF
4177 ? memNE(s, locinput, ln)
4178 : ! folder(s, locinput, ln)))
4181 nextchr = UCHARAT(locinput);
4191 #define ST st->u.eval
4196 regexp_internal *rei;
4197 regnode *startpoint;
4200 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4201 if (cur_eval && cur_eval->locinput==locinput) {
4202 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4203 Perl_croak(aTHX_ "Infinite recursion in regex");
4204 if ( ++nochange_depth > max_nochange_depth )
4206 "Pattern subroutine nesting without pos change"
4207 " exceeded limit in regex");
4214 (void)ReREFCNT_inc(rex_sv);
4215 if (OP(scan)==GOSUB) {
4216 startpoint = scan + ARG2L(scan);
4217 ST.close_paren = ARG(scan);
4219 startpoint = rei->program+1;
4222 goto eval_recurse_doit;
4224 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4225 if (cur_eval && cur_eval->locinput==locinput) {
4226 if ( ++nochange_depth > max_nochange_depth )
4227 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4232 /* execute the code in the {...} */
4234 SV ** const before = SP;
4235 OP_4tree * const oop = PL_op;
4236 COP * const ocurcop = PL_curcop;
4238 char *saved_regeol = PL_regeol;
4239 struct re_save_state saved_state;
4241 /* To not corrupt the existing regex state while executing the
4242 * eval we would normally put it on the save stack, like with
4243 * save_re_context. However, re-evals have a weird scoping so we
4244 * can't just add ENTER/LEAVE here. With that, things like
4246 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4248 * would break, as they expect the localisation to be unwound
4249 * only when the re-engine backtracks through the bit that
4252 * What we do instead is just saving the state in a local c
4255 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4258 PL_op = (OP_4tree*)rexi->data->data[n];
4259 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4260 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4261 /* wrap the call in two SAVECOMPPADs. This ensures that
4262 * when the save stack is eventually unwound, all the
4263 * accumulated SAVEt_CLEARSV's will be processed with
4264 * interspersed SAVEt_COMPPAD's to ensure that lexicals
4265 * are cleared in the right pad */
4267 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4268 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4271 SV *sv_mrk = get_sv("REGMARK", 1);
4272 sv_setsv(sv_mrk, sv_yes_mark);
4275 CALLRUNOPS(aTHX); /* Scalar context. */
4278 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4284 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4288 PAD_RESTORE_LOCAL(old_comppad);
4289 PL_curcop = ocurcop;
4290 PL_regeol = saved_regeol;
4293 sv_setsv(save_scalar(PL_replgv), ret);
4297 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4300 /* extract RE object from returned value; compiling if
4306 SV *const sv = SvRV(ret);
4308 if (SvTYPE(sv) == SVt_REGEXP) {
4310 } else if (SvSMAGICAL(sv)) {
4311 mg = mg_find(sv, PERL_MAGIC_qr);
4314 } else if (SvTYPE(ret) == SVt_REGEXP) {
4316 } else if (SvSMAGICAL(ret)) {
4317 if (SvGMAGICAL(ret)) {
4318 /* I don't believe that there is ever qr magic
4320 assert(!mg_find(ret, PERL_MAGIC_qr));
4321 sv_unmagic(ret, PERL_MAGIC_qr);
4324 mg = mg_find(ret, PERL_MAGIC_qr);
4325 /* testing suggests mg only ends up non-NULL for
4326 scalars who were upgraded and compiled in the
4327 else block below. In turn, this is only
4328 triggered in the "postponed utf8 string" tests
4334 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4338 rx = reg_temp_copy(NULL, rx);
4342 const I32 osize = PL_regsize;
4345 assert (SvUTF8(ret));
4346 } else if (SvUTF8(ret)) {
4347 /* Not doing UTF-8, despite what the SV says. Is
4348 this only if we're trapped in use 'bytes'? */
4349 /* Make a copy of the octet sequence, but without
4350 the flag on, as the compiler now honours the
4351 SvUTF8 flag on ret. */
4353 const char *const p = SvPV(ret, len);
4354 ret = newSVpvn_flags(p, len, SVs_TEMP);
4356 rx = CALLREGCOMP(ret, pm_flags);
4358 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4360 /* This isn't a first class regexp. Instead, it's
4361 caching a regexp onto an existing, Perl visible
4363 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4368 re = (struct regexp *)SvANY(rx);
4370 RXp_MATCH_COPIED_off(re);
4371 re->subbeg = rex->subbeg;
4372 re->sublen = rex->sublen;
4375 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4376 "Matching embedded");
4378 startpoint = rei->program + 1;
4379 ST.close_paren = 0; /* only used for GOSUB */
4380 /* borrowed from regtry */
4381 if (PL_reg_start_tmpl <= re->nparens) {
4382 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4383 if(PL_reg_start_tmp)
4384 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4386 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4389 eval_recurse_doit: /* Share code with GOSUB below this line */
4390 /* run the pattern returned from (??{...}) */
4391 ST.cp = regcppush(0); /* Save *all* the positions. */
4392 REGCP_SET(ST.lastcp);
4394 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4396 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4397 PL_reglastparen = &re->lastparen;
4398 PL_reglastcloseparen = &re->lastcloseparen;
4400 re->lastcloseparen = 0;
4402 PL_reginput = locinput;
4405 /* XXXX This is too dramatic a measure... */
4408 ST.toggle_reg_flags = PL_reg_flags;
4410 PL_reg_flags |= RF_utf8;
4412 PL_reg_flags &= ~RF_utf8;
4413 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4415 ST.prev_rex = rex_sv;
4416 ST.prev_curlyx = cur_curlyx;
4417 SETREX(rex_sv,re_sv);
4422 ST.prev_eval = cur_eval;
4424 /* now continue from first node in postoned RE */
4425 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4428 /* logical is 1, /(?(?{...})X|Y)/ */
4429 sw = cBOOL(SvTRUE(ret));
4434 case EVAL_AB: /* cleanup after a successful (??{A})B */
4435 /* note: this is called twice; first after popping B, then A */
4436 PL_reg_flags ^= ST.toggle_reg_flags;
4437 ReREFCNT_dec(rex_sv);
4438 SETREX(rex_sv,ST.prev_rex);
4439 rex = (struct regexp *)SvANY(rex_sv);
4440 rexi = RXi_GET(rex);
4442 cur_eval = ST.prev_eval;
4443 cur_curlyx = ST.prev_curlyx;
4445 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4446 PL_reglastparen = &rex->lastparen;
4447 PL_reglastcloseparen = &rex->lastcloseparen;
4448 /* also update PL_regoffs */
4449 PL_regoffs = rex->offs;
4451 /* XXXX This is too dramatic a measure... */
4453 if ( nochange_depth )
4458 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4459 /* note: this is called twice; first after popping B, then A */
4460 PL_reg_flags ^= ST.toggle_reg_flags;
4461 ReREFCNT_dec(rex_sv);
4462 SETREX(rex_sv,ST.prev_rex);
4463 rex = (struct regexp *)SvANY(rex_sv);
4464 rexi = RXi_GET(rex);
4465 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4466 PL_reglastparen = &rex->lastparen;
4467 PL_reglastcloseparen = &rex->lastcloseparen;
4469 PL_reginput = locinput;
4470 REGCP_UNWIND(ST.lastcp);
4472 cur_eval = ST.prev_eval;
4473 cur_curlyx = ST.prev_curlyx;
4474 /* XXXX This is too dramatic a measure... */
4476 if ( nochange_depth )
4482 n = ARG(scan); /* which paren pair */
4483 PL_reg_start_tmp[n] = locinput;
4489 n = ARG(scan); /* which paren pair */
4490 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4491 PL_regoffs[n].end = locinput - PL_bostr;
4492 /*if (n > PL_regsize)
4494 if (n > *PL_reglastparen)
4495 *PL_reglastparen = n;
4496 *PL_reglastcloseparen = n;
4497 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4505 cursor && OP(cursor)!=END;
4506 cursor=regnext(cursor))
4508 if ( OP(cursor)==CLOSE ){
4510 if ( n <= lastopen ) {
4512 = PL_reg_start_tmp[n] - PL_bostr;
4513 PL_regoffs[n].end = locinput - PL_bostr;
4514 /*if (n > PL_regsize)
4516 if (n > *PL_reglastparen)
4517 *PL_reglastparen = n;
4518 *PL_reglastcloseparen = n;
4519 if ( n == ARG(scan) || (cur_eval &&
4520 cur_eval->u.eval.close_paren == n))
4529 n = ARG(scan); /* which paren pair */
4530 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4533 /* reg_check_named_buff_matched returns 0 for no match */
4534 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4538 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4544 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4546 next = NEXTOPER(NEXTOPER(scan));
4548 next = scan + ARG(scan);
4549 if (OP(next) == IFTHEN) /* Fake one. */
4550 next = NEXTOPER(NEXTOPER(next));
4554 logical = scan->flags;
4557 /*******************************************************************
4559 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4560 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4561 STAR/PLUS/CURLY/CURLYN are used instead.)
4563 A*B is compiled as <CURLYX><A><WHILEM><B>
4565 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4566 state, which contains the current count, initialised to -1. It also sets
4567 cur_curlyx to point to this state, with any previous value saved in the
4570 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4571 since the pattern may possibly match zero times (i.e. it's a while {} loop
4572 rather than a do {} while loop).
4574 Each entry to WHILEM represents a successful match of A. The count in the
4575 CURLYX block is incremented, another WHILEM state is pushed, and execution
4576 passes to A or B depending on greediness and the current count.
4578 For example, if matching against the string a1a2a3b (where the aN are
4579 substrings that match /A/), then the match progresses as follows: (the
4580 pushed states are interspersed with the bits of strings matched so far):
4583 <CURLYX cnt=0><WHILEM>
4584 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4585 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4586 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4587 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4589 (Contrast this with something like CURLYM, which maintains only a single
4593 a1 <CURLYM cnt=1> a2
4594 a1 a2 <CURLYM cnt=2> a3
4595 a1 a2 a3 <CURLYM cnt=3> b
4598 Each WHILEM state block marks a point to backtrack to upon partial failure
4599 of A or B, and also contains some minor state data related to that
4600 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4601 overall state, such as the count, and pointers to the A and B ops.
4603 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4604 must always point to the *current* CURLYX block, the rules are:
4606 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4607 and set cur_curlyx to point the new block.
4609 When popping the CURLYX block after a successful or unsuccessful match,
4610 restore the previous cur_curlyx.
4612 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4613 to the outer one saved in the CURLYX block.
4615 When popping the WHILEM block after a successful or unsuccessful B match,
4616 restore the previous cur_curlyx.
4618 Here's an example for the pattern (AI* BI)*BO
4619 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4622 curlyx backtrack stack
4623 ------ ---------------
4625 CO <CO prev=NULL> <WO>
4626 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4627 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4628 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4630 At this point the pattern succeeds, and we work back down the stack to
4631 clean up, restoring as we go:
4633 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4634 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4635 CO <CO prev=NULL> <WO>
4638 *******************************************************************/
4640 #define ST st->u.curlyx
4642 case CURLYX: /* start of /A*B/ (for complex A) */
4644 /* No need to save/restore up to this paren */
4645 I32 parenfloor = scan->flags;
4647 assert(next); /* keep Coverity happy */
4648 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4651 /* XXXX Probably it is better to teach regpush to support
4652 parenfloor > PL_regsize... */
4653 if (parenfloor > (I32)*PL_reglastparen)
4654 parenfloor = *PL_reglastparen; /* Pessimization... */
4656 ST.prev_curlyx= cur_curlyx;
4658 ST.cp = PL_savestack_ix;
4660 /* these fields contain the state of the current curly.
4661 * they are accessed by subsequent WHILEMs */
4662 ST.parenfloor = parenfloor;
4667 ST.count = -1; /* this will be updated by WHILEM */
4668 ST.lastloc = NULL; /* this will be updated by WHILEM */
4670 PL_reginput = locinput;
4671 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4675 case CURLYX_end: /* just finished matching all of A*B */
4676 cur_curlyx = ST.prev_curlyx;
4680 case CURLYX_end_fail: /* just failed to match all of A*B */
4682 cur_curlyx = ST.prev_curlyx;
4688 #define ST st->u.whilem
4690 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4692 /* see the discussion above about CURLYX/WHILEM */
4694 int min = ARG1(cur_curlyx->u.curlyx.me);
4695 int max = ARG2(cur_curlyx->u.curlyx.me);
4696 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4698 assert(cur_curlyx); /* keep Coverity happy */
4699 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4700 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4701 ST.cache_offset = 0;
4704 PL_reginput = locinput;
4706 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4707 "%*s whilem: matched %ld out of %d..%d\n",
4708 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4711 /* First just match a string of min A's. */
4714 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4715 cur_curlyx->u.curlyx.lastloc = locinput;
4716 REGCP_SET(ST.lastcp);
4718 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4722 /* If degenerate A matches "", assume A done. */
4724 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4725 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4726 "%*s whilem: empty match detected, trying continuation...\n",
4727 REPORT_CODE_OFF+depth*2, "")
4729 goto do_whilem_B_max;
4732 /* super-linear cache processing */
4736 if (!PL_reg_maxiter) {
4737 /* start the countdown: Postpone detection until we
4738 * know the match is not *that* much linear. */
4739 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4740 /* possible overflow for long strings and many CURLYX's */
4741 if (PL_reg_maxiter < 0)
4742 PL_reg_maxiter = I32_MAX;
4743 PL_reg_leftiter = PL_reg_maxiter;
4746 if (PL_reg_leftiter-- == 0) {
4747 /* initialise cache */
4748 const I32 size = (PL_reg_maxiter + 7)/8;
4749 if (PL_reg_poscache) {
4750 if ((I32)PL_reg_poscache_size < size) {
4751 Renew(PL_reg_poscache, size, char);
4752 PL_reg_poscache_size = size;
4754 Zero(PL_reg_poscache, size, char);
4757 PL_reg_poscache_size = size;
4758 Newxz(PL_reg_poscache, size, char);
4760 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4761 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4762 PL_colors[4], PL_colors[5])
4766 if (PL_reg_leftiter < 0) {
4767 /* have we already failed at this position? */
4769 offset = (scan->flags & 0xf) - 1
4770 + (locinput - PL_bostr) * (scan->flags>>4);
4771 mask = 1 << (offset % 8);
4773 if (PL_reg_poscache[offset] & mask) {
4774 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4775 "%*s whilem: (cache) already tried at this position...\n",
4776 REPORT_CODE_OFF+depth*2, "")
4778 sayNO; /* cache records failure */
4780 ST.cache_offset = offset;
4781 ST.cache_mask = mask;
4785 /* Prefer B over A for minimal matching. */
4787 if (cur_curlyx->u.curlyx.minmod) {
4788 ST.save_curlyx = cur_curlyx;
4789 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4790 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4791 REGCP_SET(ST.lastcp);
4792 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4796 /* Prefer A over B for maximal matching. */
4798 if (n < max) { /* More greed allowed? */
4799 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4800 cur_curlyx->u.curlyx.lastloc = locinput;
4801 REGCP_SET(ST.lastcp);
4802 PUSH_STATE_GOTO(WHILEM_A_max, A);
4805 goto do_whilem_B_max;
4809 case WHILEM_B_min: /* just matched B in a minimal match */
4810 case WHILEM_B_max: /* just matched B in a maximal match */
4811 cur_curlyx = ST.save_curlyx;
4815 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4816 cur_curlyx = ST.save_curlyx;
4817 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4818 cur_curlyx->u.curlyx.count--;
4822 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4824 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4825 REGCP_UNWIND(ST.lastcp);
4827 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4828 cur_curlyx->u.curlyx.count--;
4832 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4833 REGCP_UNWIND(ST.lastcp);
4834 regcppop(rex); /* Restore some previous $<digit>s? */
4835 PL_reginput = locinput;
4836 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4837 "%*s whilem: failed, trying continuation...\n",
4838 REPORT_CODE_OFF+depth*2, "")
4841 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4842 && ckWARN(WARN_REGEXP)
4843 && !(PL_reg_flags & RF_warned))
4845 PL_reg_flags |= RF_warned;
4846 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4847 "Complex regular subexpression recursion",
4852 ST.save_curlyx = cur_curlyx;
4853 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4854 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4857 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4858 cur_curlyx = ST.save_curlyx;
4859 REGCP_UNWIND(ST.lastcp);
4862 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4863 /* Maximum greed exceeded */
4864 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4865 && ckWARN(WARN_REGEXP)
4866 && !(PL_reg_flags & RF_warned))
4868 PL_reg_flags |= RF_warned;
4869 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4870 "%s limit (%d) exceeded",
4871 "Complex regular subexpression recursion",
4874 cur_curlyx->u.curlyx.count--;
4878 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4879 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4881 /* Try grabbing another A and see if it helps. */
4882 PL_reginput = locinput;
4883 cur_curlyx->u.curlyx.lastloc = locinput;
4884 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4885 REGCP_SET(ST.lastcp);
4886 PUSH_STATE_GOTO(WHILEM_A_min,
4887 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4891 #define ST st->u.branch
4893 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4894 next = scan + ARG(scan);
4897 scan = NEXTOPER(scan);
4900 case BRANCH: /* /(...|A|...)/ */
4901 scan = NEXTOPER(scan); /* scan now points to inner node */
4902 ST.lastparen = *PL_reglastparen;
4903 ST.next_branch = next;
4905 PL_reginput = locinput;
4907 /* Now go into the branch */
4909 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4911 PUSH_STATE_GOTO(BRANCH_next, scan);
4915 PL_reginput = locinput;
4916 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4917 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4918 PUSH_STATE_GOTO(CUTGROUP_next,next);
4920 case CUTGROUP_next_fail:
4923 if (st->u.mark.mark_name)
4924 sv_commit = st->u.mark.mark_name;
4930 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4935 REGCP_UNWIND(ST.cp);
4936 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4937 PL_regoffs[n].end = -1;
4938 *PL_reglastparen = n;
4939 /*dmq: *PL_reglastcloseparen = n; */
4940 scan = ST.next_branch;
4941 /* no more branches? */
4942 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4944 PerlIO_printf( Perl_debug_log,
4945 "%*s %sBRANCH failed...%s\n",
4946 REPORT_CODE_OFF+depth*2, "",
4952 continue; /* execute next BRANCH[J] op */
4960 #define ST st->u.curlym
4962 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4964 /* This is an optimisation of CURLYX that enables us to push
4965 * only a single backtracking state, no matter how many matches
4966 * there are in {m,n}. It relies on the pattern being constant
4967 * length, with no parens to influence future backrefs
4971 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4973 /* if paren positive, emulate an OPEN/CLOSE around A */
4975 U32 paren = ST.me->flags;
4976 if (paren > PL_regsize)
4978 if (paren > *PL_reglastparen)
4979 *PL_reglastparen = paren;
4980 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4988 ST.c1 = CHRTEST_UNINIT;
4991 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4994 curlym_do_A: /* execute the A in /A{m,n}B/ */
4995 PL_reginput = locinput;
4996 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4999 case CURLYM_A: /* we've just matched an A */
5000 locinput = st->locinput;
5001 nextchr = UCHARAT(locinput);
5004 /* after first match, determine A's length: u.curlym.alen */
5005 if (ST.count == 1) {
5006 if (PL_reg_match_utf8) {
5008 while (s < PL_reginput) {
5014 ST.alen = PL_reginput - locinput;
5017 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5020 PerlIO_printf(Perl_debug_log,
5021 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5022 (int)(REPORT_CODE_OFF+(depth*2)), "",
5023 (IV) ST.count, (IV)ST.alen)
5026 locinput = PL_reginput;
5028 if (cur_eval && cur_eval->u.eval.close_paren &&
5029 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5033 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5034 if ( max == REG_INFTY || ST.count < max )
5035 goto curlym_do_A; /* try to match another A */
5037 goto curlym_do_B; /* try to match B */
5039 case CURLYM_A_fail: /* just failed to match an A */
5040 REGCP_UNWIND(ST.cp);
5042 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5043 || (cur_eval && cur_eval->u.eval.close_paren &&
5044 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5047 curlym_do_B: /* execute the B in /A{m,n}B/ */
5048 PL_reginput = locinput;
5049 if (ST.c1 == CHRTEST_UNINIT) {
5050 /* calculate c1 and c2 for possible match of 1st char
5051 * following curly */
5052 ST.c1 = ST.c2 = CHRTEST_VOID;
5053 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5054 regnode *text_node = ST.B;
5055 if (! HAS_TEXT(text_node))
5056 FIND_NEXT_IMPT(text_node);
5059 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5061 But the former is redundant in light of the latter.
5063 if this changes back then the macro for
5064 IS_TEXT and friends need to change.
5066 if (PL_regkind[OP(text_node)] == EXACT)
5069 ST.c1 = (U8)*STRING(text_node);
5070 switch (OP(text_node)) {
5071 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5073 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5074 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5075 default: ST.c2 = ST.c1;
5082 PerlIO_printf(Perl_debug_log,
5083 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5084 (int)(REPORT_CODE_OFF+(depth*2)),
5087 if (ST.c1 != CHRTEST_VOID
5088 && UCHARAT(PL_reginput) != ST.c1
5089 && UCHARAT(PL_reginput) != ST.c2)
5091 /* simulate B failing */
5093 PerlIO_printf(Perl_debug_log,
5094 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5095 (int)(REPORT_CODE_OFF+(depth*2)),"",
5098 state_num = CURLYM_B_fail;
5099 goto reenter_switch;
5103 /* mark current A as captured */
5104 I32 paren = ST.me->flags;
5106 PL_regoffs[paren].start
5107 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5108 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5109 /*dmq: *PL_reglastcloseparen = paren; */
5112 PL_regoffs[paren].end = -1;
5113 if (cur_eval && cur_eval->u.eval.close_paren &&
5114 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5123 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5126 case CURLYM_B_fail: /* just failed to match a B */
5127 REGCP_UNWIND(ST.cp);
5129 I32 max = ARG2(ST.me);
5130 if (max != REG_INFTY && ST.count == max)
5132 goto curlym_do_A; /* try to match a further A */
5134 /* backtrack one A */
5135 if (ST.count == ARG1(ST.me) /* min */)
5138 locinput = HOPc(locinput, -ST.alen);
5139 goto curlym_do_B; /* try to match B */
5142 #define ST st->u.curly
5144 #define CURLY_SETPAREN(paren, success) \
5147 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5148 PL_regoffs[paren].end = locinput - PL_bostr; \
5149 *PL_reglastcloseparen = paren; \
5152 PL_regoffs[paren].end = -1; \
5155 case STAR: /* /A*B/ where A is width 1 */
5159 scan = NEXTOPER(scan);
5161 case PLUS: /* /A+B/ where A is width 1 */
5165 scan = NEXTOPER(scan);
5167 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5168 ST.paren = scan->flags; /* Which paren to set */
5169 if (ST.paren > PL_regsize)
5170 PL_regsize = ST.paren;
5171 if (ST.paren > *PL_reglastparen)
5172 *PL_reglastparen = ST.paren;
5173 ST.min = ARG1(scan); /* min to match */
5174 ST.max = ARG2(scan); /* max to match */
5175 if (cur_eval && cur_eval->u.eval.close_paren &&
5176 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5180 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5182 case CURLY: /* /A{m,n}B/ where A is width 1 */
5184 ST.min = ARG1(scan); /* min to match */
5185 ST.max = ARG2(scan); /* max to match */
5186 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5189 * Lookahead to avoid useless match attempts
5190 * when we know what character comes next.
5192 * Used to only do .*x and .*?x, but now it allows
5193 * for )'s, ('s and (?{ ... })'s to be in the way
5194 * of the quantifier and the EXACT-like node. -- japhy
5197 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5199 if (HAS_TEXT(next) || JUMPABLE(next)) {
5201 regnode *text_node = next;
5203 if (! HAS_TEXT(text_node))
5204 FIND_NEXT_IMPT(text_node);
5206 if (! HAS_TEXT(text_node))
5207 ST.c1 = ST.c2 = CHRTEST_VOID;
5209 if ( PL_regkind[OP(text_node)] != EXACT ) {
5210 ST.c1 = ST.c2 = CHRTEST_VOID;
5211 goto assume_ok_easy;
5214 s = (U8*)STRING(text_node);
5216 /* Currently we only get here when
5218 PL_rekind[OP(text_node)] == EXACT
5220 if this changes back then the macro for IS_TEXT and
5221 friends need to change. */
5224 switch (OP(text_node)) {
5225 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5227 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5228 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5229 default: ST.c2 = ST.c1; break;
5232 else { /* UTF_PATTERN */
5233 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5234 STRLEN ulen1, ulen2;
5235 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5236 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5238 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5239 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5241 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5243 0 : UTF8_ALLOW_ANY);
5244 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5246 0 : UTF8_ALLOW_ANY);
5248 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5250 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5255 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5262 ST.c1 = ST.c2 = CHRTEST_VOID;
5267 PL_reginput = locinput;
5270 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5273 locinput = PL_reginput;
5275 if (ST.c1 == CHRTEST_VOID)
5276 goto curly_try_B_min;
5278 ST.oldloc = locinput;
5280 /* set ST.maxpos to the furthest point along the
5281 * string that could possibly match */
5282 if (ST.max == REG_INFTY) {
5283 ST.maxpos = PL_regeol - 1;
5285 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5288 else if (utf8_target) {
5289 int m = ST.max - ST.min;
5290 for (ST.maxpos = locinput;
5291 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5292 ST.maxpos += UTF8SKIP(ST.maxpos);
5295 ST.maxpos = locinput + ST.max - ST.min;
5296 if (ST.maxpos >= PL_regeol)
5297 ST.maxpos = PL_regeol - 1;
5299 goto curly_try_B_min_known;
5303 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5304 locinput = PL_reginput;
5305 if (ST.count < ST.min)
5307 if ((ST.count > ST.min)
5308 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5310 /* A{m,n} must come at the end of the string, there's
5311 * no point in backing off ... */
5313 /* ...except that $ and \Z can match before *and* after
5314 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5315 We may back off by one in this case. */
5316 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5320 goto curly_try_B_max;
5325 case CURLY_B_min_known_fail:
5326 /* failed to find B in a non-greedy match where c1,c2 valid */
5327 if (ST.paren && ST.count)
5328 PL_regoffs[ST.paren].end = -1;
5330 PL_reginput = locinput; /* Could be reset... */
5331 REGCP_UNWIND(ST.cp);
5332 /* Couldn't or didn't -- move forward. */
5333 ST.oldloc = locinput;
5335 locinput += UTF8SKIP(locinput);
5339 curly_try_B_min_known:
5340 /* find the next place where 'B' could work, then call B */
5344 n = (ST.oldloc == locinput) ? 0 : 1;
5345 if (ST.c1 == ST.c2) {
5347 /* set n to utf8_distance(oldloc, locinput) */
5348 while (locinput <= ST.maxpos &&
5349 utf8n_to_uvchr((U8*)locinput,
5350 UTF8_MAXBYTES, &len,
5351 uniflags) != (UV)ST.c1) {
5357 /* set n to utf8_distance(oldloc, locinput) */
5358 while (locinput <= ST.maxpos) {
5360 const UV c = utf8n_to_uvchr((U8*)locinput,
5361 UTF8_MAXBYTES, &len,
5363 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5371 if (ST.c1 == ST.c2) {
5372 while (locinput <= ST.maxpos &&
5373 UCHARAT(locinput) != ST.c1)
5377 while (locinput <= ST.maxpos
5378 && UCHARAT(locinput) != ST.c1
5379 && UCHARAT(locinput) != ST.c2)
5382 n = locinput - ST.oldloc;
5384 if (locinput > ST.maxpos)
5386 /* PL_reginput == oldloc now */
5389 if (regrepeat(rex, ST.A, n, depth) < n)
5392 PL_reginput = locinput;
5393 CURLY_SETPAREN(ST.paren, ST.count);
5394 if (cur_eval && cur_eval->u.eval.close_paren &&
5395 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5398 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5403 case CURLY_B_min_fail:
5404 /* failed to find B in a non-greedy match where c1,c2 invalid */
5405 if (ST.paren && ST.count)
5406 PL_regoffs[ST.paren].end = -1;
5408 REGCP_UNWIND(ST.cp);
5409 /* failed -- move forward one */
5410 PL_reginput = locinput;
5411 if (regrepeat(rex, ST.A, 1, depth)) {
5413 locinput = PL_reginput;
5414 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5415 ST.count > 0)) /* count overflow ? */
5418 CURLY_SETPAREN(ST.paren, ST.count);
5419 if (cur_eval && cur_eval->u.eval.close_paren &&
5420 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5423 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5431 /* a successful greedy match: now try to match B */
5432 if (cur_eval && cur_eval->u.eval.close_paren &&
5433 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5438 if (ST.c1 != CHRTEST_VOID)
5439 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5440 UTF8_MAXBYTES, 0, uniflags)
5441 : (UV) UCHARAT(PL_reginput);
5442 /* If it could work, try it. */
5443 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5444 CURLY_SETPAREN(ST.paren, ST.count);
5445 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5450 case CURLY_B_max_fail:
5451 /* failed to find B in a greedy match */
5452 if (ST.paren && ST.count)
5453 PL_regoffs[ST.paren].end = -1;
5455 REGCP_UNWIND(ST.cp);
5457 if (--ST.count < ST.min)
5459 PL_reginput = locinput = HOPc(locinput, -1);
5460 goto curly_try_B_max;
5467 /* we've just finished A in /(??{A})B/; now continue with B */
5469 st->u.eval.toggle_reg_flags
5470 = cur_eval->u.eval.toggle_reg_flags;
5471 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5473 st->u.eval.prev_rex = rex_sv; /* inner */
5474 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5475 rex = (struct regexp *)SvANY(rex_sv);
5476 rexi = RXi_GET(rex);
5477 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5478 (void)ReREFCNT_inc(rex_sv);
5479 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5481 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5482 PL_reglastparen = &rex->lastparen;
5483 PL_reglastcloseparen = &rex->lastcloseparen;
5485 REGCP_SET(st->u.eval.lastcp);
5486 PL_reginput = locinput;
5488 /* Restore parens of the outer rex without popping the
5490 tmpix = PL_savestack_ix;
5491 PL_savestack_ix = cur_eval->u.eval.lastcp;
5493 PL_savestack_ix = tmpix;
5495 st->u.eval.prev_eval = cur_eval;
5496 cur_eval = cur_eval->u.eval.prev_eval;
5498 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5499 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5500 if ( nochange_depth )
5503 PUSH_YES_STATE_GOTO(EVAL_AB,
5504 st->u.eval.prev_eval->u.eval.B); /* match B */
5507 if (locinput < reginfo->till) {
5508 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5509 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5511 (long)(locinput - PL_reg_starttry),
5512 (long)(reginfo->till - PL_reg_starttry),
5515 sayNO_SILENT; /* Cannot match: too short. */
5517 PL_reginput = locinput; /* put where regtry can find it */
5518 sayYES; /* Success! */
5520 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5522 PerlIO_printf(Perl_debug_log,
5523 "%*s %ssubpattern success...%s\n",
5524 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5525 PL_reginput = locinput; /* put where regtry can find it */
5526 sayYES; /* Success! */
5529 #define ST st->u.ifmatch
5531 case SUSPEND: /* (?>A) */
5533 PL_reginput = locinput;
5536 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5538 goto ifmatch_trivial_fail_test;
5540 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5542 ifmatch_trivial_fail_test:
5544 char * const s = HOPBACKc(locinput, scan->flags);
5549 sw = 1 - cBOOL(ST.wanted);
5553 next = scan + ARG(scan);
5561 PL_reginput = locinput;
5565 ST.logical = logical;
5566 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5568 /* execute body of (?...A) */
5569 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5572 case IFMATCH_A_fail: /* body of (?...A) failed */
5573 ST.wanted = !ST.wanted;
5576 case IFMATCH_A: /* body of (?...A) succeeded */
5578 sw = cBOOL(ST.wanted);
5580 else if (!ST.wanted)
5583 if (OP(ST.me) == SUSPEND)
5584 locinput = PL_reginput;
5586 locinput = PL_reginput = st->locinput;
5587 nextchr = UCHARAT(locinput);
5589 scan = ST.me + ARG(ST.me);
5592 continue; /* execute B */
5597 next = scan + ARG(scan);
5602 reginfo->cutpoint = PL_regeol;
5605 PL_reginput = locinput;
5607 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5608 PUSH_STATE_GOTO(COMMIT_next,next);
5610 case COMMIT_next_fail:
5617 #define ST st->u.mark
5619 ST.prev_mark = mark_state;
5620 ST.mark_name = sv_commit = sv_yes_mark
5621 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5623 ST.mark_loc = PL_reginput = locinput;
5624 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5626 case MARKPOINT_next:
5627 mark_state = ST.prev_mark;
5630 case MARKPOINT_next_fail:
5631 if (popmark && sv_eq(ST.mark_name,popmark))
5633 if (ST.mark_loc > startpoint)
5634 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5635 popmark = NULL; /* we found our mark */
5636 sv_commit = ST.mark_name;
5639 PerlIO_printf(Perl_debug_log,
5640 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5641 REPORT_CODE_OFF+depth*2, "",
5642 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5645 mark_state = ST.prev_mark;
5646 sv_yes_mark = mark_state ?
5647 mark_state->u.mark.mark_name : NULL;
5651 PL_reginput = locinput;
5653 /* (*SKIP) : if we fail we cut here*/
5654 ST.mark_name = NULL;
5655 ST.mark_loc = locinput;
5656 PUSH_STATE_GOTO(SKIP_next,next);
5658 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5659 otherwise do nothing. Meaning we need to scan
5661 regmatch_state *cur = mark_state;
5662 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5665 if ( sv_eq( cur->u.mark.mark_name,
5668 ST.mark_name = find;
5669 PUSH_STATE_GOTO( SKIP_next, next );
5671 cur = cur->u.mark.prev_mark;
5674 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5676 case SKIP_next_fail:
5678 /* (*CUT:NAME) - Set up to search for the name as we
5679 collapse the stack*/
5680 popmark = ST.mark_name;
5682 /* (*CUT) - No name, we cut here.*/
5683 if (ST.mark_loc > startpoint)
5684 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5685 /* but we set sv_commit to latest mark_name if there
5686 is one so they can test to see how things lead to this
5689 sv_commit=mark_state->u.mark.mark_name;
5697 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5699 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5702 U8 folded[UTF8_MAXBYTES_CASE+1];
5704 const char * const l = locinput;
5705 char *e = PL_regeol;
5706 to_uni_fold(n, folded, &foldlen);
5708 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5709 l, &e, 0, utf8_target)) {
5714 nextchr = UCHARAT(locinput);
5717 if ((n=is_LNBREAK(locinput,utf8_target))) {
5719 nextchr = UCHARAT(locinput);
5724 #define CASE_CLASS(nAmE) \
5726 if (locinput >= PL_regeol) \
5728 if ((n=is_##nAmE(locinput,utf8_target))) { \
5730 nextchr = UCHARAT(locinput); \
5735 if (locinput >= PL_regeol) \
5737 if ((n=is_##nAmE(locinput,utf8_target))) { \
5740 locinput += UTF8SKIP(locinput); \
5741 nextchr = UCHARAT(locinput); \
5746 CASE_CLASS(HORIZWS);
5750 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5751 PTR2UV(scan), OP(scan));
5752 Perl_croak(aTHX_ "regexp memory corruption");
5756 /* switch break jumps here */
5757 scan = next; /* prepare to execute the next op and ... */
5758 continue; /* ... jump back to the top, reusing st */
5762 /* push a state that backtracks on success */
5763 st->u.yes.prev_yes_state = yes_state;
5767 /* push a new regex state, then continue at scan */
5769 regmatch_state *newst;
5772 regmatch_state *cur = st;
5773 regmatch_state *curyes = yes_state;
5775 regmatch_slab *slab = PL_regmatch_slab;
5776 for (;curd > -1;cur--,curd--) {
5777 if (cur < SLAB_FIRST(slab)) {
5779 cur = SLAB_LAST(slab);
5781 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5782 REPORT_CODE_OFF + 2 + depth * 2,"",
5783 curd, PL_reg_name[cur->resume_state],
5784 (curyes == cur) ? "yes" : ""
5787 curyes = cur->u.yes.prev_yes_state;
5790 DEBUG_STATE_pp("push")
5793 st->locinput = locinput;
5795 if (newst > SLAB_LAST(PL_regmatch_slab))
5796 newst = S_push_slab(aTHX);
5797 PL_regmatch_state = newst;
5799 locinput = PL_reginput;
5800 nextchr = UCHARAT(locinput);
5808 * We get here only if there's trouble -- normally "case END" is
5809 * the terminating point.
5811 Perl_croak(aTHX_ "corrupted regexp pointers");
5817 /* we have successfully completed a subexpression, but we must now
5818 * pop to the state marked by yes_state and continue from there */
5819 assert(st != yes_state);
5821 while (st != yes_state) {
5823 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5824 PL_regmatch_slab = PL_regmatch_slab->prev;
5825 st = SLAB_LAST(PL_regmatch_slab);
5829 DEBUG_STATE_pp("pop (no final)");
5831 DEBUG_STATE_pp("pop (yes)");
5837 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5838 || yes_state > SLAB_LAST(PL_regmatch_slab))
5840 /* not in this slab, pop slab */
5841 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5842 PL_regmatch_slab = PL_regmatch_slab->prev;
5843 st = SLAB_LAST(PL_regmatch_slab);
5845 depth -= (st - yes_state);
5848 yes_state = st->u.yes.prev_yes_state;
5849 PL_regmatch_state = st;
5852 locinput= st->locinput;
5853 nextchr = UCHARAT(locinput);
5855 state_num = st->resume_state + no_final;
5856 goto reenter_switch;
5859 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5860 PL_colors[4], PL_colors[5]));
5862 if (PL_reg_eval_set) {
5863 /* each successfully executed (?{...}) block does the equivalent of
5864 * local $^R = do {...}
5865 * When popping the save stack, all these locals would be undone;
5866 * bypass this by setting the outermost saved $^R to the latest
5868 if (oreplsv != GvSV(PL_replgv))
5869 sv_setsv(oreplsv, GvSV(PL_replgv));
5876 PerlIO_printf(Perl_debug_log,
5877 "%*s %sfailed...%s\n",
5878 REPORT_CODE_OFF+depth*2, "",
5879 PL_colors[4], PL_colors[5])
5891 /* there's a previous state to backtrack to */
5893 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5894 PL_regmatch_slab = PL_regmatch_slab->prev;
5895 st = SLAB_LAST(PL_regmatch_slab);
5897 PL_regmatch_state = st;
5898 locinput= st->locinput;
5899 nextchr = UCHARAT(locinput);
5901 DEBUG_STATE_pp("pop");
5903 if (yes_state == st)
5904 yes_state = st->u.yes.prev_yes_state;
5906 state_num = st->resume_state + 1; /* failure = success + 1 */
5907 goto reenter_switch;
5912 if (rex->intflags & PREGf_VERBARG_SEEN) {
5913 SV *sv_err = get_sv("REGERROR", 1);
5914 SV *sv_mrk = get_sv("REGMARK", 1);
5916 sv_commit = &PL_sv_no;
5918 sv_yes_mark = &PL_sv_yes;
5921 sv_commit = &PL_sv_yes;
5922 sv_yes_mark = &PL_sv_no;
5924 sv_setsv(sv_err, sv_commit);
5925 sv_setsv(sv_mrk, sv_yes_mark);
5928 /* clean up; in particular, free all slabs above current one */
5929 LEAVE_SCOPE(oldsave);
5935 - regrepeat - repeatedly match something simple, report how many
5938 * [This routine now assumes that it will only match on things of length 1.
5939 * That was true before, but now we assume scan - reginput is the count,
5940 * rather than incrementing count on every character. [Er, except utf8.]]
5943 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5946 register char *scan;
5948 register char *loceol = PL_regeol;
5949 register I32 hardcount = 0;
5950 register bool utf8_target = PL_reg_match_utf8;
5953 PERL_UNUSED_ARG(depth);
5956 PERL_ARGS_ASSERT_REGREPEAT;
5959 if (max == REG_INFTY)
5961 else if (max < loceol - scan)
5962 loceol = scan + max;
5967 while (scan < loceol && hardcount < max && *scan != '\n') {
5968 scan += UTF8SKIP(scan);
5972 while (scan < loceol && *scan != '\n')
5979 while (scan < loceol && hardcount < max) {
5980 scan += UTF8SKIP(scan);
5991 /* To get here, EXACTish nodes must have *byte* length == 1. That
5992 * means they match only characters in the string that can be expressed
5993 * as a single byte. For non-utf8 strings, that means a simple match.
5994 * For utf8 strings, the character matched must be an invariant, or
5995 * downgradable to a single byte. The pattern's utf8ness is
5996 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5997 * it is, it's an invariant */
6000 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6002 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6003 while (scan < loceol && UCHARAT(scan) == c) {
6009 /* Here, the string is utf8, and the pattern char is different
6010 * in utf8 than not, so can't compare them directly. Outside the
6011 * loop, find find the two utf8 bytes that represent c, and then
6012 * look for those in sequence in the utf8 string */
6013 U8 high = UTF8_TWO_BYTE_HI(c);
6014 U8 low = UTF8_TWO_BYTE_LO(c);
6017 while (hardcount < max
6018 && scan + 1 < loceol
6019 && UCHARAT(scan) == high
6020 && UCHARAT(scan + 1) == low)
6028 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6032 PL_reg_flags |= RF_tainted;
6033 utf8_flags = FOLDEQ_UTF8_LOCALE;
6038 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6040 /* The comments for the EXACT case above apply as well to these fold
6045 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6047 if (utf8_target) { /* Use full Unicode fold matching */
6048 char *tmpeol = loceol;
6049 while (hardcount < max
6050 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6051 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6058 /* XXX Note that the above handles properly the German sharp s in
6059 * the pattern matching ss in the string. But it doesn't handle
6060 * properly cases where the string contains say 'LIGATURE ff' and
6061 * the pattern is 'f+'. This would require, say, a new function or
6062 * revised interface to foldEQ_utf8(), in which the maximum number
6063 * of characters to match could be passed and it would return how
6064 * many actually did. This is just one of many cases where
6065 * multi-char folds don't work properly, and so the fix is being
6071 /* Here, the string isn't utf8 and c is a single byte; and either
6072 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6073 * doesn't affect c. Can just do simple comparisons for exact or
6076 case EXACTF: folded = PL_fold[c]; break;
6078 case EXACTFU: folded = PL_fold_latin1[c]; break;
6079 case EXACTFL: folded = PL_fold_locale[c]; break;
6080 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6082 while (scan < loceol &&
6083 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6091 if (utf8_target || OP(p) == ANYOFV) {
6094 inclasslen = loceol - scan;
6095 while (hardcount < max
6096 && ((inclasslen = loceol - scan) > 0)
6097 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6103 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6111 LOAD_UTF8_CHARCLASS_ALNUM();
6112 while (hardcount < max && scan < loceol &&
6113 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6115 scan += UTF8SKIP(scan);
6119 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6127 while (scan < loceol && isALNUM((U8) *scan)) {
6132 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6137 PL_reg_flags |= RF_tainted;
6140 while (hardcount < max && scan < loceol &&
6141 isALNUM_LC_utf8((U8*)scan)) {
6142 scan += UTF8SKIP(scan);
6146 while (scan < loceol && isALNUM_LC(*scan))
6156 LOAD_UTF8_CHARCLASS_ALNUM();
6157 while (hardcount < max && scan < loceol &&
6158 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6160 scan += UTF8SKIP(scan);
6164 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6171 goto utf8_Nwordchar;
6172 while (scan < loceol && ! isALNUM((U8) *scan)) {
6178 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6179 scan += UTF8SKIP(scan);
6183 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6189 PL_reg_flags |= RF_tainted;
6192 while (hardcount < max && scan < loceol &&
6193 !isALNUM_LC_utf8((U8*)scan)) {
6194 scan += UTF8SKIP(scan);
6198 while (scan < loceol && !isALNUM_LC(*scan))
6208 LOAD_UTF8_CHARCLASS_SPACE();
6209 while (hardcount < max && scan < loceol &&
6211 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6213 scan += UTF8SKIP(scan);
6219 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6228 while (scan < loceol && isSPACE((U8) *scan)) {
6233 while (scan < loceol && isSPACE_A((U8) *scan)) {
6238 PL_reg_flags |= RF_tainted;
6241 while (hardcount < max && scan < loceol &&
6242 isSPACE_LC_utf8((U8*)scan)) {
6243 scan += UTF8SKIP(scan);
6247 while (scan < loceol && isSPACE_LC(*scan))
6257 LOAD_UTF8_CHARCLASS_SPACE();
6258 while (hardcount < max && scan < loceol &&
6260 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6262 scan += UTF8SKIP(scan);
6268 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6277 while (scan < loceol && ! isSPACE((U8) *scan)) {
6283 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6284 scan += UTF8SKIP(scan);
6288 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6294 PL_reg_flags |= RF_tainted;
6297 while (hardcount < max && scan < loceol &&
6298 !isSPACE_LC_utf8((U8*)scan)) {
6299 scan += UTF8SKIP(scan);
6303 while (scan < loceol && !isSPACE_LC(*scan))
6310 LOAD_UTF8_CHARCLASS_DIGIT();
6311 while (hardcount < max && scan < loceol &&
6312 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6313 scan += UTF8SKIP(scan);
6317 while (scan < loceol && isDIGIT(*scan))
6322 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6327 PL_reg_flags |= RF_tainted;
6330 while (hardcount < max && scan < loceol &&
6331 isDIGIT_LC_utf8((U8*)scan)) {
6332 scan += UTF8SKIP(scan);
6336 while (scan < loceol && isDIGIT_LC(*scan))
6343 LOAD_UTF8_CHARCLASS_DIGIT();
6344 while (hardcount < max && scan < loceol &&
6345 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6346 scan += UTF8SKIP(scan);
6350 while (scan < loceol && !isDIGIT(*scan))
6356 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6357 scan += UTF8SKIP(scan);
6361 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6367 PL_reg_flags |= RF_tainted;
6370 while (hardcount < max && scan < loceol &&
6371 !isDIGIT_LC_utf8((U8*)scan)) {
6372 scan += UTF8SKIP(scan);
6376 while (scan < loceol && !isDIGIT_LC(*scan))
6383 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6389 LNBREAK can match two latin chars, which is ok,
6390 because we have a null terminated string, but we
6391 have to use hardcount in this situation
6393 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6402 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6407 while (scan < loceol && is_HORIZWS_latin1(scan))
6414 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6415 scan += UTF8SKIP(scan);
6419 while (scan < loceol && !is_HORIZWS_latin1(scan))
6427 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6432 while (scan < loceol && is_VERTWS_latin1(scan))
6440 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6441 scan += UTF8SKIP(scan);
6445 while (scan < loceol && !is_VERTWS_latin1(scan))
6451 default: /* Called on something of 0 width. */
6452 break; /* So match right here or not at all. */
6458 c = scan - PL_reginput;
6462 GET_RE_DEBUG_FLAGS_DECL;
6464 SV * const prop = sv_newmortal();
6465 regprop(prog, prop, p);
6466 PerlIO_printf(Perl_debug_log,
6467 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6468 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6476 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6478 - regclass_swash - prepare the utf8 swash
6482 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6488 RXi_GET_DECL(prog,progi);
6489 const struct reg_data * const data = prog ? progi->data : NULL;
6491 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6493 assert(ANYOF_NONBITMAP(node));
6495 if (data && data->count) {
6496 const U32 n = ARG(node);
6498 if (data->what[n] == 's') {
6499 SV * const rv = MUTABLE_SV(data->data[n]);
6500 AV * const av = MUTABLE_AV(SvRV(rv));
6501 SV **const ary = AvARRAY(av);
6504 /* See the end of regcomp.c:S_regclass() for
6505 * documentation of these array elements. */
6508 a = SvROK(ary[1]) ? &ary[1] : NULL;
6509 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6513 else if (si && doinit) {
6514 sw = swash_init("utf8", "", si, 1, 0);
6515 (void)av_store(av, 1, sw);
6532 - reginclass - determine if a character falls into a character class
6534 n is the ANYOF regnode
6535 p is the target string
6536 lenp is pointer to the maximum number of bytes of how far to go in p
6537 (This is assumed wthout checking to always be at least the current
6539 utf8_target tells whether p is in UTF-8.
6541 Returns true if matched; false otherwise. If lenp is not NULL, on return
6542 from a successful match, the value it points to will be updated to how many
6543 bytes in p were matched. If there was no match, the value is undefined,
6544 possibly changed from the input.
6546 Note that this can be a synthetic start class, a combination of various
6547 nodes, so things you think might be mutually exclusive, such as locale,
6548 aren't. It can match both locale and non-locale
6553 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6556 const char flags = ANYOF_FLAGS(n);
6562 PERL_ARGS_ASSERT_REGINCLASS;
6564 /* If c is not already the code point, get it */
6565 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6566 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6567 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6568 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6569 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6570 * UTF8_ALLOW_FFFF */
6571 if (c_len == (STRLEN)-1)
6572 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6578 /* Use passed in max length, or one character if none passed in or less
6579 * than one character. And assume will match just one character. This is
6580 * overwritten later if matched more. */
6582 maxlen = (*lenp > c_len) ? *lenp : c_len;
6590 /* If this character is potentially in the bitmap, check it */
6592 if (ANYOF_BITMAP_TEST(n, c))
6594 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6601 else if (flags & ANYOF_LOCALE) {
6602 PL_reg_flags |= RF_tainted;
6604 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6605 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6609 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6610 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6611 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6612 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6613 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6614 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6615 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6616 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6617 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6618 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6619 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6620 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6621 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6622 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6623 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6624 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6625 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6626 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6627 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6628 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6629 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6630 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6631 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6632 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6633 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6634 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6635 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6636 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6637 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6638 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6639 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6640 ) /* How's that for a conditional? */
6647 /* If the bitmap didn't (or couldn't) match, and something outside the
6648 * bitmap could match, try that. Locale nodes specifiy completely the
6649 * behavior of code points in the bit map (otherwise, a utf8 target would
6650 * cause them to be treated as Unicode and not locale), except in
6651 * the very unlikely event when this node is a synthetic start class, which
6652 * could be a combination of locale and non-locale nodes. So allow locale
6653 * to match for the synthetic start class, which will give a false
6654 * positive that will be resolved when the match is done again as not part
6655 * of the synthetic start class */
6657 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6658 match = TRUE; /* Everything above 255 matches */
6660 else if (ANYOF_NONBITMAP(n)
6661 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6664 || (! (flags & ANYOF_LOCALE))
6665 || (flags & ANYOF_IS_SYNTHETIC)))))
6668 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6676 /* Not utf8. Convert as much of the string as available up
6677 * to the limit of how far the (single) character in the
6678 * pattern can possibly match (no need to go further). If
6679 * the node is a straight ANYOF or not folding, it can't
6680 * match more than one. Otherwise, It can match up to how
6681 * far a single char can fold to. Since not utf8, each
6682 * character is a single byte, so the max it can be in
6683 * bytes is the same as the max it can be in characters */
6684 STRLEN len = (OP(n) == ANYOF
6685 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6687 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6689 : UTF8_MAX_FOLD_CHAR_EXPAND;
6690 utf8_p = bytes_to_utf8(p, &len);
6693 if (swash_fetch(sw, utf8_p, TRUE))
6695 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6697 /* Here, we need to test if the fold of the target string
6698 * matches. The non-multi char folds have all been moved to
6699 * the compilation phase, and the multi-char folds have
6700 * been stored by regcomp into 'av'; we linearly check to
6701 * see if any match the target string (folded). We know
6702 * that the originals were each one character, but we don't
6703 * currently know how many characters/bytes each folded to,
6704 * except we do know that there are small limits imposed by
6705 * Unicode. XXX A performance enhancement would be to have
6706 * regcomp.c store the max number of chars/bytes that are
6707 * in an av entry, as, say the 0th element. Even better
6708 * would be to have a hash of the few characters that can
6709 * start a multi-char fold to the max number of chars of
6712 * If there is a match, we will need to advance (if lenp is
6713 * specified) the match pointer in the target string. But
6714 * what we are comparing here isn't that string directly,
6715 * but its fold, whose length may differ from the original.
6716 * As we go along in constructing the fold, therefore, we
6717 * create a map so that we know how many bytes in the
6718 * source to advance given that we have matched a certain
6719 * number of bytes in the fold. This map is stored in
6720 * 'map_fold_len_back'. Let n mean the number of bytes in
6721 * the fold of the first character that we are folding.
6722 * Then map_fold_len_back[n] is set to the number of bytes
6723 * in that first character. Similarly let m be the
6724 * corresponding number for the second character to be
6725 * folded. Then map_fold_len_back[n+m] is set to the
6726 * number of bytes occupied by the first two source
6727 * characters. ... */
6728 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6729 U8 folded[UTF8_MAXBYTES_CASE+1];
6730 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6731 STRLEN total_foldlen = 0; /* num bytes in fold of all
6734 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6736 /* Here, only need to fold the first char of the target
6737 * string. It the source wasn't utf8, is 1 byte long */
6738 to_utf8_fold(utf8_p, folded, &foldlen);
6739 total_foldlen = foldlen;
6740 map_fold_len_back[foldlen] = (utf8_target)
6746 /* Here, need to fold more than the first char. Do so
6747 * up to the limits */
6748 U8* source_ptr = utf8_p; /* The source for the fold
6751 U8* folded_ptr = folded;
6752 U8* e = utf8_p + maxlen; /* Can't go beyond last
6753 available byte in the
6757 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6761 /* Fold the next character */
6762 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6763 STRLEN this_char_foldlen;
6764 to_utf8_fold(source_ptr,
6766 &this_char_foldlen);
6768 /* Bail if it would exceed the byte limit for
6769 * folding a single char. */
6770 if (this_char_foldlen + folded_ptr - folded >
6776 /* Add the fold of this character */
6777 Copy(this_char_folded,
6781 source_ptr += UTF8SKIP(source_ptr);
6782 folded_ptr += this_char_foldlen;
6783 total_foldlen = folded_ptr - folded;
6785 /* Create map from the number of bytes in the fold
6786 * back to the number of bytes in the source. If
6787 * the source isn't utf8, the byte count is just
6788 * the number of characters so far */
6789 map_fold_len_back[total_foldlen]
6791 ? source_ptr - utf8_p
6798 /* Do the linear search to see if the fold is in the list
6799 * of multi-char folds. */
6802 for (i = 0; i <= av_len(av); i++) {
6803 SV* const sv = *av_fetch(av, i, FALSE);
6805 const char * const s = SvPV_const(sv, len);
6807 if (len <= total_foldlen
6808 && memEQ(s, (char*)folded, len)
6810 /* If 0, means matched a partial char. See
6812 && map_fold_len_back[len])
6815 /* Advance the target string ptr to account for
6816 * this fold, but have to translate from the
6817 * folded length to the corresponding source
6820 *lenp = map_fold_len_back[len];
6829 /* If we allocated a string above, free it */
6830 if (! utf8_target) Safefree(utf8_p);
6835 return (flags & ANYOF_INVERT) ? !match : match;
6839 S_reghop3(U8 *s, I32 off, const U8* lim)
6841 /* return the position 'off' UTF-8 characters away from 's', forward if
6842 * 'off' >= 0, backwards if negative. But don't go outside of position
6843 * 'lim', which better be < s if off < 0 */
6847 PERL_ARGS_ASSERT_REGHOP3;
6850 while (off-- && s < lim) {
6851 /* XXX could check well-formedness here */
6856 while (off++ && s > lim) {
6858 if (UTF8_IS_CONTINUED(*s)) {
6859 while (s > lim && UTF8_IS_CONTINUATION(*s))
6862 /* XXX could check well-formedness here */
6869 /* there are a bunch of places where we use two reghop3's that should
6870 be replaced with this routine. but since thats not done yet
6871 we ifdef it out - dmq
6874 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6878 PERL_ARGS_ASSERT_REGHOP4;
6881 while (off-- && s < rlim) {
6882 /* XXX could check well-formedness here */
6887 while (off++ && s > llim) {
6889 if (UTF8_IS_CONTINUED(*s)) {
6890 while (s > llim && UTF8_IS_CONTINUATION(*s))
6893 /* XXX could check well-formedness here */
6901 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6905 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6908 while (off-- && s < lim) {
6909 /* XXX could check well-formedness here */
6916 while (off++ && s > lim) {
6918 if (UTF8_IS_CONTINUED(*s)) {
6919 while (s > lim && UTF8_IS_CONTINUATION(*s))
6922 /* XXX could check well-formedness here */
6931 restore_pos(pTHX_ void *arg)
6934 regexp * const rex = (regexp *)arg;
6935 if (PL_reg_eval_set) {
6936 if (PL_reg_oldsaved) {
6937 rex->subbeg = PL_reg_oldsaved;
6938 rex->sublen = PL_reg_oldsavedlen;
6939 #ifdef PERL_OLD_COPY_ON_WRITE
6940 rex->saved_copy = PL_nrs;
6942 RXp_MATCH_COPIED_on(rex);
6944 PL_reg_magic->mg_len = PL_reg_oldpos;
6945 PL_reg_eval_set = 0;
6946 PL_curpm = PL_reg_oldcurpm;
6951 S_to_utf8_substr(pTHX_ register regexp *prog)
6955 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6958 if (prog->substrs->data[i].substr
6959 && !prog->substrs->data[i].utf8_substr) {
6960 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6961 prog->substrs->data[i].utf8_substr = sv;
6962 sv_utf8_upgrade(sv);
6963 if (SvVALID(prog->substrs->data[i].substr)) {
6964 if (SvTAIL(prog->substrs->data[i].substr)) {
6965 /* Trim the trailing \n that fbm_compile added last
6967 SvCUR_set(sv, SvCUR(sv) - 1);
6968 /* Whilst this makes the SV technically "invalid" (as its
6969 buffer is no longer followed by "\0") when fbm_compile()
6970 adds the "\n" back, a "\0" is restored. */
6971 fbm_compile(sv, FBMcf_TAIL);
6975 if (prog->substrs->data[i].substr == prog->check_substr)
6976 prog->check_utf8 = sv;
6982 S_to_byte_substr(pTHX_ register regexp *prog)
6987 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6990 if (prog->substrs->data[i].utf8_substr
6991 && !prog->substrs->data[i].substr) {
6992 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6993 if (sv_utf8_downgrade(sv, TRUE)) {
6994 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6995 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
6996 /* Trim the trailing \n that fbm_compile added last
6998 SvCUR_set(sv, SvCUR(sv) - 1);
6999 fbm_compile(sv, FBMcf_TAIL);
7007 prog->substrs->data[i].substr = sv;
7008 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7009 prog->check_substr = sv;
7016 * c-indentation-style: bsd
7018 * indent-tabs-mode: t
7021 * ex: set ts=8 sts=4 sw=4 noet: