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 PERL_UNUSED_DECL; \
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 = to_utf8_fold( (U8 *) uc, foldbuf, &foldlen ); \
1205 len = UTF8SKIP(uc); \
1206 foldlen -= UNISKIP( uvc ); \
1207 uscan = foldbuf + UNISKIP( uvc ); \
1210 case trie_latin_utf8_fold: \
1211 if ( foldlen>0 ) { \
1212 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1218 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1219 foldlen -= UNISKIP( uvc ); \
1220 uscan = foldbuf + UNISKIP( uvc ); \
1224 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1231 charid = trie->charmap[ uvc ]; \
1235 if (widecharmap) { \
1236 SV** const svpp = hv_fetch(widecharmap, \
1237 (char*)&uvc, sizeof(UV), 0); \
1239 charid = (U16)SvIV(*svpp); \
1244 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1248 && (ln == 1 || folder(s, pat_string, ln)) \
1249 && (!reginfo || regtry(reginfo, &s)) ) \
1255 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1257 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1263 #define REXEC_FBC_SCAN(CoDe) \
1265 while (s < strend) { \
1271 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1272 REXEC_FBC_UTF8_SCAN( \
1274 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1283 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1286 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1295 #define REXEC_FBC_TRYIT \
1296 if ((!reginfo || regtry(reginfo, &s))) \
1299 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1300 if (utf8_target) { \
1301 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1304 REXEC_FBC_CLASS_SCAN(CoNd); \
1307 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1308 if (utf8_target) { \
1310 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1313 REXEC_FBC_CLASS_SCAN(CoNd); \
1316 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1317 PL_reg_flags |= RF_tainted; \
1318 if (utf8_target) { \
1319 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1322 REXEC_FBC_CLASS_SCAN(CoNd); \
1325 #define DUMP_EXEC_POS(li,s,doutf8) \
1326 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1329 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1330 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1331 tmp = TEST_NON_UTF8(tmp); \
1332 REXEC_FBC_UTF8_SCAN( \
1333 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1342 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1343 if (s == PL_bostr) { \
1347 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1348 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1351 LOAD_UTF8_CHARCLASS_ALNUM(); \
1352 REXEC_FBC_UTF8_SCAN( \
1353 if (tmp == ! (TeSt2_UtF8)) { \
1362 /* The only difference between the BOUND and NBOUND cases is that
1363 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1364 * NBOUND. This is accomplished by passing it in either the if or else clause,
1365 * with the other one being empty */
1366 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1367 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1369 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1370 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1372 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1373 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1375 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1376 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1379 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1380 * be passed in completely with the variable name being tested, which isn't
1381 * such a clean interface, but this is easier to read than it was before. We
1382 * are looking for the boundary (or non-boundary between a word and non-word
1383 * character. The utf8 and non-utf8 cases have the same logic, but the details
1384 * must be different. Find the "wordness" of the character just prior to this
1385 * one, and compare it with the wordness of this one. If they differ, we have
1386 * a boundary. At the beginning of the string, pretend that the previous
1387 * character was a new-line */
1388 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1389 if (utf8_target) { \
1392 else { /* Not utf8 */ \
1393 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1394 tmp = TEST_NON_UTF8(tmp); \
1396 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1405 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1408 /* We know what class REx starts with. Try to find this position... */
1409 /* if reginfo is NULL, its a dryrun */
1410 /* annoyingly all the vars in this routine have different names from their counterparts
1411 in regmatch. /grrr */
1414 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1415 const char *strend, regmatch_info *reginfo)
1418 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1419 char *pat_string; /* The pattern's exactish string */
1420 char *pat_end; /* ptr to end char of pat_string */
1421 re_fold_t folder; /* Function for computing non-utf8 folds */
1422 const U8 *fold_array; /* array for folding ords < 256 */
1425 register STRLEN uskip;
1429 register I32 tmp = 1; /* Scratch variable? */
1430 register const bool utf8_target = PL_reg_match_utf8;
1431 UV utf8_fold_flags = 0;
1432 RXi_GET_DECL(prog,progi);
1434 PERL_ARGS_ASSERT_FIND_BYCLASS;
1436 /* We know what class it must start with. */
1440 if (utf8_target || OP(c) == ANYOFV) {
1441 STRLEN inclasslen = strend - s;
1442 REXEC_FBC_UTF8_CLASS_SCAN(
1443 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1446 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1451 if (tmp && (!reginfo || regtry(reginfo, &s)))
1459 if (UTF_PATTERN || utf8_target) {
1460 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1461 goto do_exactf_utf8;
1463 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1464 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1465 goto do_exactf_non_utf8; /* isn't dealt with by these */
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, rounding up, so that
1557 * if we need to match at least 4+1/3 chars, that really is 5. */
1558 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1559 lnc = (lnc + expansion - 1) / 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 to stop the loop earlier,
1572 * as the worst case expansion above will rarely be met, and as we
1573 * go along we would usually find that e moves further to the left.
1574 * This would happen only after we reached the point in the loop
1575 * where if there were no expansion we should fail. Unclear if
1576 * worth the expense */
1579 char *my_strend= (char *)strend;
1580 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1581 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1582 && (!reginfo || regtry(reginfo, &s)) )
1586 s += (utf8_target) ? UTF8SKIP(s) : 1;
1591 PL_reg_flags |= RF_tainted;
1592 FBC_BOUND(isALNUM_LC,
1593 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1594 isALNUM_LC_utf8((U8*)s));
1597 PL_reg_flags |= RF_tainted;
1598 FBC_NBOUND(isALNUM_LC,
1599 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1600 isALNUM_LC_utf8((U8*)s));
1603 FBC_BOUND(isWORDCHAR,
1605 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1608 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1610 isWORDCHAR_A((U8*)s));
1613 FBC_NBOUND(isWORDCHAR,
1615 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1618 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1620 isWORDCHAR_A((U8*)s));
1623 FBC_BOUND(isWORDCHAR_L1,
1625 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1628 FBC_NBOUND(isWORDCHAR_L1,
1630 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1633 REXEC_FBC_CSCAN_TAINT(
1634 isALNUM_LC_utf8((U8*)s),
1639 REXEC_FBC_CSCAN_PRELOAD(
1640 LOAD_UTF8_CHARCLASS_ALNUM(),
1641 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1642 isWORDCHAR_L1((U8) *s)
1646 REXEC_FBC_CSCAN_PRELOAD(
1647 LOAD_UTF8_CHARCLASS_ALNUM(),
1648 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1653 /* Don't need to worry about utf8, as it can match only a single
1654 * byte invariant character */
1655 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1658 REXEC_FBC_CSCAN_PRELOAD(
1659 LOAD_UTF8_CHARCLASS_ALNUM(),
1660 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1661 ! isWORDCHAR_L1((U8) *s)
1665 REXEC_FBC_CSCAN_PRELOAD(
1666 LOAD_UTF8_CHARCLASS_ALNUM(),
1667 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1678 REXEC_FBC_CSCAN_TAINT(
1679 !isALNUM_LC_utf8((U8*)s),
1684 REXEC_FBC_CSCAN_PRELOAD(
1685 LOAD_UTF8_CHARCLASS_SPACE(),
1686 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1691 REXEC_FBC_CSCAN_PRELOAD(
1692 LOAD_UTF8_CHARCLASS_SPACE(),
1693 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1698 /* Don't need to worry about utf8, as it can match only a single
1699 * byte invariant character */
1700 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1703 REXEC_FBC_CSCAN_TAINT(
1704 isSPACE_LC_utf8((U8*)s),
1709 REXEC_FBC_CSCAN_PRELOAD(
1710 LOAD_UTF8_CHARCLASS_SPACE(),
1711 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1712 ! isSPACE_L1((U8) *s)
1716 REXEC_FBC_CSCAN_PRELOAD(
1717 LOAD_UTF8_CHARCLASS_SPACE(),
1718 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1729 REXEC_FBC_CSCAN_TAINT(
1730 !isSPACE_LC_utf8((U8*)s),
1735 REXEC_FBC_CSCAN_PRELOAD(
1736 LOAD_UTF8_CHARCLASS_DIGIT(),
1737 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1742 /* Don't need to worry about utf8, as it can match only a single
1743 * byte invariant character */
1744 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1747 REXEC_FBC_CSCAN_TAINT(
1748 isDIGIT_LC_utf8((U8*)s),
1753 REXEC_FBC_CSCAN_PRELOAD(
1754 LOAD_UTF8_CHARCLASS_DIGIT(),
1755 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1766 REXEC_FBC_CSCAN_TAINT(
1767 !isDIGIT_LC_utf8((U8*)s),
1774 is_LNBREAK_latin1(s)
1786 !is_VERTWS_latin1(s)
1792 is_HORIZWS_latin1(s)
1797 !is_HORIZWS_utf8(s),
1798 !is_HORIZWS_latin1(s)
1805 /* what trie are we using right now */
1807 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1809 = (reg_trie_data*)progi->data->data[ aho->trie ];
1810 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1812 const char *last_start = strend - trie->minlen;
1814 const char *real_start = s;
1816 STRLEN maxlen = trie->maxlen;
1818 U8 **points; /* map of where we were in the input string
1819 when reading a given char. For ASCII this
1820 is unnecessary overhead as the relationship
1821 is always 1:1, but for Unicode, especially
1822 case folded Unicode this is not true. */
1823 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1827 GET_RE_DEBUG_FLAGS_DECL;
1829 /* We can't just allocate points here. We need to wrap it in
1830 * an SV so it gets freed properly if there is a croak while
1831 * running the match */
1834 sv_points=newSV(maxlen * sizeof(U8 *));
1835 SvCUR_set(sv_points,
1836 maxlen * sizeof(U8 *));
1837 SvPOK_on(sv_points);
1838 sv_2mortal(sv_points);
1839 points=(U8**)SvPV_nolen(sv_points );
1840 if ( trie_type != trie_utf8_fold
1841 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1844 bitmap=(U8*)trie->bitmap;
1846 bitmap=(U8*)ANYOF_BITMAP(c);
1848 /* this is the Aho-Corasick algorithm modified a touch
1849 to include special handling for long "unknown char"
1850 sequences. The basic idea being that we use AC as long
1851 as we are dealing with a possible matching char, when
1852 we encounter an unknown char (and we have not encountered
1853 an accepting state) we scan forward until we find a legal
1855 AC matching is basically that of trie matching, except
1856 that when we encounter a failing transition, we fall back
1857 to the current states "fail state", and try the current char
1858 again, a process we repeat until we reach the root state,
1859 state 1, or a legal transition. If we fail on the root state
1860 then we can either terminate if we have reached an accepting
1861 state previously, or restart the entire process from the beginning
1865 while (s <= last_start) {
1866 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1874 U8 *uscan = (U8*)NULL;
1875 U8 *leftmost = NULL;
1877 U32 accepted_word= 0;
1881 while ( state && uc <= (U8*)strend ) {
1883 U32 word = aho->states[ state ].wordnum;
1887 DEBUG_TRIE_EXECUTE_r(
1888 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1889 dump_exec_pos( (char *)uc, c, strend, real_start,
1890 (char *)uc, utf8_target );
1891 PerlIO_printf( Perl_debug_log,
1892 " Scanning for legal start char...\n");
1896 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1900 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1906 if (uc >(U8*)last_start) break;
1910 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1911 if (!leftmost || lpos < leftmost) {
1912 DEBUG_r(accepted_word=word);
1918 points[pointpos++ % maxlen]= uc;
1919 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1920 uscan, len, uvc, charid, foldlen,
1922 DEBUG_TRIE_EXECUTE_r({
1923 dump_exec_pos( (char *)uc, c, strend, real_start,
1925 PerlIO_printf(Perl_debug_log,
1926 " Charid:%3u CP:%4"UVxf" ",
1932 word = aho->states[ state ].wordnum;
1934 base = aho->states[ state ].trans.base;
1936 DEBUG_TRIE_EXECUTE_r({
1938 dump_exec_pos( (char *)uc, c, strend, real_start,
1940 PerlIO_printf( Perl_debug_log,
1941 "%sState: %4"UVxf", word=%"UVxf,
1942 failed ? " Fail transition to " : "",
1943 (UV)state, (UV)word);
1949 ( ((offset = base + charid
1950 - 1 - trie->uniquecharcount)) >= 0)
1951 && ((U32)offset < trie->lasttrans)
1952 && trie->trans[offset].check == state
1953 && (tmp=trie->trans[offset].next))
1955 DEBUG_TRIE_EXECUTE_r(
1956 PerlIO_printf( Perl_debug_log," - legal\n"));
1961 DEBUG_TRIE_EXECUTE_r(
1962 PerlIO_printf( Perl_debug_log," - fail\n"));
1964 state = aho->fail[state];
1968 /* we must be accepting here */
1969 DEBUG_TRIE_EXECUTE_r(
1970 PerlIO_printf( Perl_debug_log," - accepting\n"));
1979 if (!state) state = 1;
1982 if ( aho->states[ state ].wordnum ) {
1983 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1984 if (!leftmost || lpos < leftmost) {
1985 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1990 s = (char*)leftmost;
1991 DEBUG_TRIE_EXECUTE_r({
1993 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1994 (UV)accepted_word, (IV)(s - real_start)
1997 if (!reginfo || regtry(reginfo, &s)) {
2003 DEBUG_TRIE_EXECUTE_r({
2004 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2007 DEBUG_TRIE_EXECUTE_r(
2008 PerlIO_printf( Perl_debug_log,"No match.\n"));
2017 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2027 - regexec_flags - match a regexp against a string
2030 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2031 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2032 /* strend: pointer to null at end of string */
2033 /* strbeg: real beginning of string */
2034 /* minend: end of match must be >=minend after stringarg. */
2035 /* data: May be used for some additional optimizations.
2036 Currently its only used, with a U32 cast, for transmitting
2037 the ganch offset when doing a /g match. This will change */
2038 /* nosave: For optimizations. */
2041 struct regexp *const prog = (struct regexp *)SvANY(rx);
2042 /*register*/ char *s;
2043 register regnode *c;
2044 /*register*/ char *startpos = stringarg;
2045 I32 minlen; /* must match at least this many chars */
2046 I32 dontbother = 0; /* how many characters not to try at end */
2047 I32 end_shift = 0; /* Same for the end. */ /* CC */
2048 I32 scream_pos = -1; /* Internal iterator of scream. */
2049 char *scream_olds = NULL;
2050 const bool utf8_target = cBOOL(DO_UTF8(sv));
2052 RXi_GET_DECL(prog,progi);
2053 regmatch_info reginfo; /* create some info to pass to regtry etc */
2054 regexp_paren_pair *swap = NULL;
2055 GET_RE_DEBUG_FLAGS_DECL;
2057 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2058 PERL_UNUSED_ARG(data);
2060 /* Be paranoid... */
2061 if (prog == NULL || startpos == NULL) {
2062 Perl_croak(aTHX_ "NULL regexp parameter");
2066 multiline = prog->extflags & RXf_PMf_MULTILINE;
2067 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2069 RX_MATCH_UTF8_set(rx, utf8_target);
2071 debug_start_match(rx, utf8_target, startpos, strend,
2075 minlen = prog->minlen;
2077 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2078 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2079 "String too short [regexec_flags]...\n"));
2084 /* Check validity of program. */
2085 if (UCHARAT(progi->program) != REG_MAGIC) {
2086 Perl_croak(aTHX_ "corrupted regexp program");
2090 PL_reg_eval_set = 0;
2094 PL_reg_flags |= RF_utf8;
2096 /* Mark beginning of line for ^ and lookbehind. */
2097 reginfo.bol = startpos; /* XXX not used ??? */
2101 /* Mark end of line for $ (and such) */
2104 /* see how far we have to get to not match where we matched before */
2105 reginfo.till = startpos+minend;
2107 /* If there is a "must appear" string, look for it. */
2110 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2112 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2113 reginfo.ganch = startpos + prog->gofs;
2114 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2115 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2116 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2118 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2119 && mg->mg_len >= 0) {
2120 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2121 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2122 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2124 if (prog->extflags & RXf_ANCH_GPOS) {
2125 if (s > reginfo.ganch)
2127 s = reginfo.ganch - prog->gofs;
2128 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2129 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2135 reginfo.ganch = strbeg + PTR2UV(data);
2136 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2137 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2139 } else { /* pos() not defined */
2140 reginfo.ganch = strbeg;
2141 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2142 "GPOS: reginfo.ganch = strbeg\n"));
2145 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2146 /* We have to be careful. If the previous successful match
2147 was from this regex we don't want a subsequent partially
2148 successful match to clobber the old results.
2149 So when we detect this possibility we add a swap buffer
2150 to the re, and switch the buffer each match. If we fail
2151 we switch it back, otherwise we leave it swapped.
2154 /* do we need a save destructor here for eval dies? */
2155 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2157 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2158 re_scream_pos_data d;
2160 d.scream_olds = &scream_olds;
2161 d.scream_pos = &scream_pos;
2162 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2164 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2165 goto phooey; /* not present */
2171 /* Simplest case: anchored match need be tried only once. */
2172 /* [unless only anchor is BOL and multiline is set] */
2173 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2174 if (s == startpos && regtry(®info, &startpos))
2176 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2177 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2182 dontbother = minlen - 1;
2183 end = HOP3c(strend, -dontbother, strbeg) - 1;
2184 /* for multiline we only have to try after newlines */
2185 if (prog->check_substr || prog->check_utf8) {
2186 /* because of the goto we can not easily reuse the macros for bifurcating the
2187 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2190 goto after_try_utf8;
2192 if (regtry(®info, &s)) {
2199 if (prog->extflags & RXf_USE_INTUIT) {
2200 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2209 } /* end search for check string in unicode */
2211 if (s == startpos) {
2212 goto after_try_latin;
2215 if (regtry(®info, &s)) {
2222 if (prog->extflags & RXf_USE_INTUIT) {
2223 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2232 } /* end search for check string in latin*/
2233 } /* end search for check string */
2234 else { /* search for newline */
2236 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2239 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2241 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2242 if (regtry(®info, &s))
2246 } /* end search for newline */
2247 } /* end anchored/multiline check string search */
2249 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2251 /* the warning about reginfo.ganch being used without initialization
2252 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2253 and we only enter this block when the same bit is set. */
2254 char *tmp_s = reginfo.ganch - prog->gofs;
2256 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2261 /* Messy cases: unanchored match. */
2262 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2263 /* we have /x+whatever/ */
2264 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2269 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2270 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2271 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2276 DEBUG_EXECUTE_r( did_match = 1 );
2277 if (regtry(®info, &s)) goto got_it;
2279 while (s < strend && *s == ch)
2287 DEBUG_EXECUTE_r( did_match = 1 );
2288 if (regtry(®info, &s)) goto got_it;
2290 while (s < strend && *s == ch)
2295 DEBUG_EXECUTE_r(if (!did_match)
2296 PerlIO_printf(Perl_debug_log,
2297 "Did not find anchored character...\n")
2300 else if (prog->anchored_substr != NULL
2301 || prog->anchored_utf8 != NULL
2302 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2303 && prog->float_max_offset < strend - s)) {
2308 char *last1; /* Last position checked before */
2312 if (prog->anchored_substr || prog->anchored_utf8) {
2313 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2314 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2315 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2316 back_max = back_min = prog->anchored_offset;
2318 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2319 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2320 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2321 back_max = prog->float_max_offset;
2322 back_min = prog->float_min_offset;
2326 if (must == &PL_sv_undef)
2327 /* could not downgrade utf8 check substring, so must fail */
2333 last = HOP3c(strend, /* Cannot start after this */
2334 -(I32)(CHR_SVLEN(must)
2335 - (SvTAIL(must) != 0) + back_min), strbeg);
2338 last1 = HOPc(s, -1);
2340 last1 = s - 1; /* bogus */
2342 /* XXXX check_substr already used to find "s", can optimize if
2343 check_substr==must. */
2345 dontbother = end_shift;
2346 strend = HOPc(strend, -dontbother);
2347 while ( (s <= last) &&
2348 ((flags & REXEC_SCREAM) && SvSCREAM(sv)
2349 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2350 end_shift, &scream_pos, 0))
2351 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2352 (unsigned char*)strend, must,
2353 multiline ? FBMrf_MULTILINE : 0))) ) {
2354 /* we may be pointing at the wrong string */
2355 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2356 s = strbeg + (s - SvPVX_const(sv));
2357 DEBUG_EXECUTE_r( did_match = 1 );
2358 if (HOPc(s, -back_max) > last1) {
2359 last1 = HOPc(s, -back_min);
2360 s = HOPc(s, -back_max);
2363 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2365 last1 = HOPc(s, -back_min);
2369 while (s <= last1) {
2370 if (regtry(®info, &s))
2376 while (s <= last1) {
2377 if (regtry(®info, &s))
2383 DEBUG_EXECUTE_r(if (!did_match) {
2384 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2385 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2386 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2387 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2388 ? "anchored" : "floating"),
2389 quoted, RE_SV_TAIL(must));
2393 else if ( (c = progi->regstclass) ) {
2395 const OPCODE op = OP(progi->regstclass);
2396 /* don't bother with what can't match */
2397 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2398 strend = HOPc(strend, -(minlen - 1));
2401 SV * const prop = sv_newmortal();
2402 regprop(prog, prop, c);
2404 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2406 PerlIO_printf(Perl_debug_log,
2407 "Matching stclass %.*s against %s (%d bytes)\n",
2408 (int)SvCUR(prop), SvPVX_const(prop),
2409 quoted, (int)(strend - s));
2412 if (find_byclass(prog, c, s, strend, ®info))
2414 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2418 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2423 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2424 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2425 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2427 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
2428 last = screaminstr(sv, float_real, s - strbeg,
2429 end_shift, &scream_pos, 1); /* last one */
2431 last = scream_olds; /* Only one occurrence. */
2432 /* we may be pointing at the wrong string */
2433 else if (RXp_MATCH_COPIED(prog))
2434 s = strbeg + (s - SvPVX_const(sv));
2438 const char * const little = SvPV_const(float_real, len);
2440 if (SvTAIL(float_real)) {
2441 if (memEQ(strend - len + 1, little, len - 1))
2442 last = strend - len + 1;
2443 else if (!multiline)
2444 last = memEQ(strend - len, little, len)
2445 ? strend - len : NULL;
2451 last = rninstr(s, strend, little, little + len);
2453 last = strend; /* matching "$" */
2458 PerlIO_printf(Perl_debug_log,
2459 "%sCan't trim the tail, match fails (should not happen)%s\n",
2460 PL_colors[4], PL_colors[5]));
2461 goto phooey; /* Should not happen! */
2463 dontbother = strend - last + prog->float_min_offset;
2465 if (minlen && (dontbother < minlen))
2466 dontbother = minlen - 1;
2467 strend -= dontbother; /* this one's always in bytes! */
2468 /* We don't know much -- general case. */
2471 if (regtry(®info, &s))
2480 if (regtry(®info, &s))
2482 } while (s++ < strend);
2491 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2493 if (PL_reg_eval_set)
2494 restore_pos(aTHX_ prog);
2495 if (RXp_PAREN_NAMES(prog))
2496 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2498 /* make sure $`, $&, $', and $digit will work later */
2499 if ( !(flags & REXEC_NOT_FIRST) ) {
2500 RX_MATCH_COPY_FREE(rx);
2501 if (flags & REXEC_COPY_STR) {
2502 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2503 #ifdef PERL_OLD_COPY_ON_WRITE
2505 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2507 PerlIO_printf(Perl_debug_log,
2508 "Copy on write: regexp capture, type %d\n",
2511 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2512 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2513 assert (SvPOKp(prog->saved_copy));
2517 RX_MATCH_COPIED_on(rx);
2518 s = savepvn(strbeg, i);
2524 prog->subbeg = strbeg;
2525 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2532 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2533 PL_colors[4], PL_colors[5]));
2534 if (PL_reg_eval_set)
2535 restore_pos(aTHX_ prog);
2537 /* we failed :-( roll it back */
2538 Safefree(prog->offs);
2547 - regtry - try match at specific point
2549 STATIC I32 /* 0 failure, 1 success */
2550 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2554 REGEXP *const rx = reginfo->prog;
2555 regexp *const prog = (struct regexp *)SvANY(rx);
2556 RXi_GET_DECL(prog,progi);
2557 GET_RE_DEBUG_FLAGS_DECL;
2559 PERL_ARGS_ASSERT_REGTRY;
2561 reginfo->cutpoint=NULL;
2563 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2566 PL_reg_eval_set = RS_init;
2567 DEBUG_EXECUTE_r(DEBUG_s(
2568 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2569 (IV)(PL_stack_sp - PL_stack_base));
2572 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2573 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2575 /* Apparently this is not needed, judging by wantarray. */
2576 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2577 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2580 /* Make $_ available to executed code. */
2581 if (reginfo->sv != DEFSV) {
2583 DEFSV_set(reginfo->sv);
2586 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2587 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2588 /* prepare for quick setting of pos */
2589 #ifdef PERL_OLD_COPY_ON_WRITE
2590 if (SvIsCOW(reginfo->sv))
2591 sv_force_normal_flags(reginfo->sv, 0);
2593 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2594 &PL_vtbl_mglob, NULL, 0);
2598 PL_reg_oldpos = mg->mg_len;
2599 SAVEDESTRUCTOR_X(restore_pos, prog);
2601 if (!PL_reg_curpm) {
2602 Newxz(PL_reg_curpm, 1, PMOP);
2605 SV* const repointer = &PL_sv_undef;
2606 /* this regexp is also owned by the new PL_reg_curpm, which
2607 will try to free it. */
2608 av_push(PL_regex_padav, repointer);
2609 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2610 PL_regex_pad = AvARRAY(PL_regex_padav);
2615 /* It seems that non-ithreads works both with and without this code.
2616 So for efficiency reasons it seems best not to have the code
2617 compiled when it is not needed. */
2618 /* This is safe against NULLs: */
2619 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2620 /* PM_reg_curpm owns a reference to this regexp. */
2621 (void)ReREFCNT_inc(rx);
2623 PM_SETRE(PL_reg_curpm, rx);
2624 PL_reg_oldcurpm = PL_curpm;
2625 PL_curpm = PL_reg_curpm;
2626 if (RXp_MATCH_COPIED(prog)) {
2627 /* Here is a serious problem: we cannot rewrite subbeg,
2628 since it may be needed if this match fails. Thus
2629 $` inside (?{}) could fail... */
2630 PL_reg_oldsaved = prog->subbeg;
2631 PL_reg_oldsavedlen = prog->sublen;
2632 #ifdef PERL_OLD_COPY_ON_WRITE
2633 PL_nrs = prog->saved_copy;
2635 RXp_MATCH_COPIED_off(prog);
2638 PL_reg_oldsaved = NULL;
2639 prog->subbeg = PL_bostr;
2640 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2642 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2643 prog->offs[0].start = *startpos - PL_bostr;
2644 PL_reginput = *startpos;
2645 PL_reglastparen = &prog->lastparen;
2646 PL_reglastcloseparen = &prog->lastcloseparen;
2647 prog->lastparen = 0;
2648 prog->lastcloseparen = 0;
2650 PL_regoffs = prog->offs;
2651 if (PL_reg_start_tmpl <= prog->nparens) {
2652 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2653 if(PL_reg_start_tmp)
2654 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2656 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2659 /* XXXX What this code is doing here?!!! There should be no need
2660 to do this again and again, PL_reglastparen should take care of
2663 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2664 * Actually, the code in regcppop() (which Ilya may be meaning by
2665 * PL_reglastparen), is not needed at all by the test suite
2666 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2667 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2668 * Meanwhile, this code *is* needed for the
2669 * above-mentioned test suite tests to succeed. The common theme
2670 * on those tests seems to be returning null fields from matches.
2671 * --jhi updated by dapm */
2673 if (prog->nparens) {
2674 regexp_paren_pair *pp = PL_regoffs;
2676 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2684 if (regmatch(reginfo, progi->program + 1)) {
2685 PL_regoffs[0].end = PL_reginput - PL_bostr;
2688 if (reginfo->cutpoint)
2689 *startpos= reginfo->cutpoint;
2690 REGCP_UNWIND(lastcp);
2695 #define sayYES goto yes
2696 #define sayNO goto no
2697 #define sayNO_SILENT goto no_silent
2699 /* we dont use STMT_START/END here because it leads to
2700 "unreachable code" warnings, which are bogus, but distracting. */
2701 #define CACHEsayNO \
2702 if (ST.cache_mask) \
2703 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2706 /* this is used to determine how far from the left messages like
2707 'failed...' are printed. It should be set such that messages
2708 are inline with the regop output that created them.
2710 #define REPORT_CODE_OFF 32
2713 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2714 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2716 #define SLAB_FIRST(s) (&(s)->states[0])
2717 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2719 /* grab a new slab and return the first slot in it */
2721 STATIC regmatch_state *
2724 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2727 regmatch_slab *s = PL_regmatch_slab->next;
2729 Newx(s, 1, regmatch_slab);
2730 s->prev = PL_regmatch_slab;
2732 PL_regmatch_slab->next = s;
2734 PL_regmatch_slab = s;
2735 return SLAB_FIRST(s);
2739 /* push a new state then goto it */
2741 #define PUSH_STATE_GOTO(state, node) \
2743 st->resume_state = state; \
2746 /* push a new state with success backtracking, then goto it */
2748 #define PUSH_YES_STATE_GOTO(state, node) \
2750 st->resume_state = state; \
2751 goto push_yes_state;
2757 regmatch() - main matching routine
2759 This is basically one big switch statement in a loop. We execute an op,
2760 set 'next' to point the next op, and continue. If we come to a point which
2761 we may need to backtrack to on failure such as (A|B|C), we push a
2762 backtrack state onto the backtrack stack. On failure, we pop the top
2763 state, and re-enter the loop at the state indicated. If there are no more
2764 states to pop, we return failure.
2766 Sometimes we also need to backtrack on success; for example /A+/, where
2767 after successfully matching one A, we need to go back and try to
2768 match another one; similarly for lookahead assertions: if the assertion
2769 completes successfully, we backtrack to the state just before the assertion
2770 and then carry on. In these cases, the pushed state is marked as
2771 'backtrack on success too'. This marking is in fact done by a chain of
2772 pointers, each pointing to the previous 'yes' state. On success, we pop to
2773 the nearest yes state, discarding any intermediate failure-only states.
2774 Sometimes a yes state is pushed just to force some cleanup code to be
2775 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2776 it to free the inner regex.
2778 Note that failure backtracking rewinds the cursor position, while
2779 success backtracking leaves it alone.
2781 A pattern is complete when the END op is executed, while a subpattern
2782 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2783 ops trigger the "pop to last yes state if any, otherwise return true"
2786 A common convention in this function is to use A and B to refer to the two
2787 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2788 the subpattern to be matched possibly multiple times, while B is the entire
2789 rest of the pattern. Variable and state names reflect this convention.
2791 The states in the main switch are the union of ops and failure/success of
2792 substates associated with with that op. For example, IFMATCH is the op
2793 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2794 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2795 successfully matched A and IFMATCH_A_fail is a state saying that we have
2796 just failed to match A. Resume states always come in pairs. The backtrack
2797 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2798 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2799 on success or failure.
2801 The struct that holds a backtracking state is actually a big union, with
2802 one variant for each major type of op. The variable st points to the
2803 top-most backtrack struct. To make the code clearer, within each
2804 block of code we #define ST to alias the relevant union.
2806 Here's a concrete example of a (vastly oversimplified) IFMATCH
2812 #define ST st->u.ifmatch
2814 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2815 ST.foo = ...; // some state we wish to save
2817 // push a yes backtrack state with a resume value of
2818 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2820 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2823 case IFMATCH_A: // we have successfully executed A; now continue with B
2825 bar = ST.foo; // do something with the preserved value
2828 case IFMATCH_A_fail: // A failed, so the assertion failed
2829 ...; // do some housekeeping, then ...
2830 sayNO; // propagate the failure
2837 For any old-timers reading this who are familiar with the old recursive
2838 approach, the code above is equivalent to:
2840 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2849 ...; // do some housekeeping, then ...
2850 sayNO; // propagate the failure
2853 The topmost backtrack state, pointed to by st, is usually free. If you
2854 want to claim it, populate any ST.foo fields in it with values you wish to
2855 save, then do one of
2857 PUSH_STATE_GOTO(resume_state, node);
2858 PUSH_YES_STATE_GOTO(resume_state, node);
2860 which sets that backtrack state's resume value to 'resume_state', pushes a
2861 new free entry to the top of the backtrack stack, then goes to 'node'.
2862 On backtracking, the free slot is popped, and the saved state becomes the
2863 new free state. An ST.foo field in this new top state can be temporarily
2864 accessed to retrieve values, but once the main loop is re-entered, it
2865 becomes available for reuse.
2867 Note that the depth of the backtrack stack constantly increases during the
2868 left-to-right execution of the pattern, rather than going up and down with
2869 the pattern nesting. For example the stack is at its maximum at Z at the
2870 end of the pattern, rather than at X in the following:
2872 /(((X)+)+)+....(Y)+....Z/
2874 The only exceptions to this are lookahead/behind assertions and the cut,
2875 (?>A), which pop all the backtrack states associated with A before
2878 Backtrack state structs are allocated in slabs of about 4K in size.
2879 PL_regmatch_state and st always point to the currently active state,
2880 and PL_regmatch_slab points to the slab currently containing
2881 PL_regmatch_state. The first time regmatch() is called, the first slab is
2882 allocated, and is never freed until interpreter destruction. When the slab
2883 is full, a new one is allocated and chained to the end. At exit from
2884 regmatch(), slabs allocated since entry are freed.
2889 #define DEBUG_STATE_pp(pp) \
2891 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2892 PerlIO_printf(Perl_debug_log, \
2893 " %*s"pp" %s%s%s%s%s\n", \
2895 PL_reg_name[st->resume_state], \
2896 ((st==yes_state||st==mark_state) ? "[" : ""), \
2897 ((st==yes_state) ? "Y" : ""), \
2898 ((st==mark_state) ? "M" : ""), \
2899 ((st==yes_state||st==mark_state) ? "]" : "") \
2904 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2909 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2910 const char *start, const char *end, const char *blurb)
2912 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2914 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2919 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2920 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2922 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2923 start, end - start, 60);
2925 PerlIO_printf(Perl_debug_log,
2926 "%s%s REx%s %s against %s\n",
2927 PL_colors[4], blurb, PL_colors[5], s0, s1);
2929 if (utf8_target||utf8_pat)
2930 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2931 utf8_pat ? "pattern" : "",
2932 utf8_pat && utf8_target ? " and " : "",
2933 utf8_target ? "string" : ""
2939 S_dump_exec_pos(pTHX_ const char *locinput,
2940 const regnode *scan,
2941 const char *loc_regeol,
2942 const char *loc_bostr,
2943 const char *loc_reg_starttry,
2944 const bool utf8_target)
2946 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2947 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2948 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2949 /* The part of the string before starttry has one color
2950 (pref0_len chars), between starttry and current
2951 position another one (pref_len - pref0_len chars),
2952 after the current position the third one.
2953 We assume that pref0_len <= pref_len, otherwise we
2954 decrease pref0_len. */
2955 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2956 ? (5 + taill) - l : locinput - loc_bostr;
2959 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2961 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2963 pref0_len = pref_len - (locinput - loc_reg_starttry);
2964 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2965 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2966 ? (5 + taill) - pref_len : loc_regeol - locinput);
2967 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2971 if (pref0_len > pref_len)
2972 pref0_len = pref_len;
2974 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2976 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2977 (locinput - pref_len),pref0_len, 60, 4, 5);
2979 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2980 (locinput - pref_len + pref0_len),
2981 pref_len - pref0_len, 60, 2, 3);
2983 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2984 locinput, loc_regeol - locinput, 10, 0, 1);
2986 const STRLEN tlen=len0+len1+len2;
2987 PerlIO_printf(Perl_debug_log,
2988 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2989 (IV)(locinput - loc_bostr),
2992 (docolor ? "" : "> <"),
2994 (int)(tlen > 19 ? 0 : 19 - tlen),
3001 /* reg_check_named_buff_matched()
3002 * Checks to see if a named buffer has matched. The data array of
3003 * buffer numbers corresponding to the buffer is expected to reside
3004 * in the regexp->data->data array in the slot stored in the ARG() of
3005 * node involved. Note that this routine doesn't actually care about the
3006 * name, that information is not preserved from compilation to execution.
3007 * Returns the index of the leftmost defined buffer with the given name
3008 * or 0 if non of the buffers matched.
3011 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3014 RXi_GET_DECL(rex,rexi);
3015 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3016 I32 *nums=(I32*)SvPVX(sv_dat);
3018 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3020 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3021 if ((I32)*PL_reglastparen >= nums[n] &&
3022 PL_regoffs[nums[n]].end != -1)
3031 /* free all slabs above current one - called during LEAVE_SCOPE */
3034 S_clear_backtrack_stack(pTHX_ void *p)
3036 regmatch_slab *s = PL_regmatch_slab->next;
3041 PL_regmatch_slab->next = NULL;
3043 regmatch_slab * const osl = s;
3050 #define SETREX(Re1,Re2) \
3051 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3054 STATIC I32 /* 0 failure, 1 success */
3055 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3057 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3061 register const bool utf8_target = PL_reg_match_utf8;
3062 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3063 REGEXP *rex_sv = reginfo->prog;
3064 regexp *rex = (struct regexp *)SvANY(rex_sv);
3065 RXi_GET_DECL(rex,rexi);
3067 /* the current state. This is a cached copy of PL_regmatch_state */
3068 register regmatch_state *st;
3069 /* cache heavy used fields of st in registers */
3070 register regnode *scan;
3071 register regnode *next;
3072 register U32 n = 0; /* general value; init to avoid compiler warning */
3073 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3074 register char *locinput = PL_reginput;
3075 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3077 bool result = 0; /* return value of S_regmatch */
3078 int depth = 0; /* depth of backtrack stack */
3079 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3080 const U32 max_nochange_depth =
3081 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3082 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3083 regmatch_state *yes_state = NULL; /* state to pop to on success of
3085 /* mark_state piggy backs on the yes_state logic so that when we unwind
3086 the stack on success we can update the mark_state as we go */
3087 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3088 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3089 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3091 bool no_final = 0; /* prevent failure from backtracking? */
3092 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3093 char *startpoint = PL_reginput;
3094 SV *popmark = NULL; /* are we looking for a mark? */
3095 SV *sv_commit = NULL; /* last mark name seen in failure */
3096 SV *sv_yes_mark = NULL; /* last mark name we have seen
3097 during a successful match */
3098 U32 lastopen = 0; /* last open we saw */
3099 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3100 SV* const oreplsv = GvSV(PL_replgv);
3101 /* these three flags are set by various ops to signal information to
3102 * the very next op. They have a useful lifetime of exactly one loop
3103 * iteration, and are not preserved or restored by state pushes/pops
3105 bool sw = 0; /* the condition value in (?(cond)a|b) */
3106 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3107 int logical = 0; /* the following EVAL is:
3111 or the following IFMATCH/UNLESSM is:
3112 false: plain (?=foo)
3113 true: used as a condition: (?(?=foo))
3116 GET_RE_DEBUG_FLAGS_DECL;
3119 PERL_ARGS_ASSERT_REGMATCH;
3121 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3122 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3124 /* on first ever call to regmatch, allocate first slab */
3125 if (!PL_regmatch_slab) {
3126 Newx(PL_regmatch_slab, 1, regmatch_slab);
3127 PL_regmatch_slab->prev = NULL;
3128 PL_regmatch_slab->next = NULL;
3129 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3132 oldsave = PL_savestack_ix;
3133 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3134 SAVEVPTR(PL_regmatch_slab);
3135 SAVEVPTR(PL_regmatch_state);
3137 /* grab next free state slot */
3138 st = ++PL_regmatch_state;
3139 if (st > SLAB_LAST(PL_regmatch_slab))
3140 st = PL_regmatch_state = S_push_slab(aTHX);
3142 /* Note that nextchr is a byte even in UTF */
3143 nextchr = UCHARAT(locinput);
3145 while (scan != NULL) {
3148 SV * const prop = sv_newmortal();
3149 regnode *rnext=regnext(scan);
3150 DUMP_EXEC_POS( locinput, scan, utf8_target );
3151 regprop(rex, prop, scan);
3153 PerlIO_printf(Perl_debug_log,
3154 "%3"IVdf":%*s%s(%"IVdf")\n",
3155 (IV)(scan - rexi->program), depth*2, "",
3157 (PL_regkind[OP(scan)] == END || !rnext) ?
3158 0 : (IV)(rnext - rexi->program));
3161 next = scan + NEXT_OFF(scan);
3164 state_num = OP(scan);
3166 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3169 assert(PL_reglastparen == &rex->lastparen);
3170 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3171 assert(PL_regoffs == rex->offs);
3173 switch (state_num) {
3175 if (locinput == PL_bostr)
3177 /* reginfo->till = reginfo->bol; */
3182 if (locinput == PL_bostr ||
3183 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3189 if (locinput == PL_bostr)
3193 if (locinput == reginfo->ganch)
3198 /* update the startpoint */
3199 st->u.keeper.val = PL_regoffs[0].start;
3200 PL_reginput = locinput;
3201 PL_regoffs[0].start = locinput - PL_bostr;
3202 PUSH_STATE_GOTO(KEEPS_next, next);
3204 case KEEPS_next_fail:
3205 /* rollback the start point change */
3206 PL_regoffs[0].start = st->u.keeper.val;
3212 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3217 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3219 if (PL_regeol - locinput > 1)
3223 if (PL_regeol != locinput)
3227 if (!nextchr && locinput >= PL_regeol)
3230 locinput += PL_utf8skip[nextchr];
3231 if (locinput > PL_regeol)
3233 nextchr = UCHARAT(locinput);
3236 nextchr = UCHARAT(++locinput);
3239 if (!nextchr && locinput >= PL_regeol)
3241 nextchr = UCHARAT(++locinput);
3244 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3247 locinput += PL_utf8skip[nextchr];
3248 if (locinput > PL_regeol)
3250 nextchr = UCHARAT(locinput);
3253 nextchr = UCHARAT(++locinput);
3257 #define ST st->u.trie
3259 /* In this case the charclass data is available inline so
3260 we can fail fast without a lot of extra overhead.
3262 if (scan->flags == EXACT || !utf8_target) {
3263 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3265 PerlIO_printf(Perl_debug_log,
3266 "%*s %sfailed to match trie start class...%s\n",
3267 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3275 /* the basic plan of execution of the trie is:
3276 * At the beginning, run though all the states, and
3277 * find the longest-matching word. Also remember the position
3278 * of the shortest matching word. For example, this pattern:
3281 * when matched against the string "abcde", will generate
3282 * accept states for all words except 3, with the longest
3283 * matching word being 4, and the shortest being 1 (with
3284 * the position being after char 1 of the string).
3286 * Then for each matching word, in word order (i.e. 1,2,4,5),
3287 * we run the remainder of the pattern; on each try setting
3288 * the current position to the character following the word,
3289 * returning to try the next word on failure.
3291 * We avoid having to build a list of words at runtime by
3292 * using a compile-time structure, wordinfo[].prev, which
3293 * gives, for each word, the previous accepting word (if any).
3294 * In the case above it would contain the mappings 1->2, 2->0,
3295 * 3->0, 4->5, 5->1. We can use this table to generate, from
3296 * the longest word (4 above), a list of all words, by
3297 * following the list of prev pointers; this gives us the
3298 * unordered list 4,5,1,2. Then given the current word we have
3299 * just tried, we can go through the list and find the
3300 * next-biggest word to try (so if we just failed on word 2,
3301 * the next in the list is 4).
3303 * Since at runtime we don't record the matching position in
3304 * the string for each word, we have to work that out for
3305 * each word we're about to process. The wordinfo table holds
3306 * the character length of each word; given that we recorded
3307 * at the start: the position of the shortest word and its
3308 * length in chars, we just need to move the pointer the
3309 * difference between the two char lengths. Depending on
3310 * Unicode status and folding, that's cheap or expensive.
3312 * This algorithm is optimised for the case where are only a
3313 * small number of accept states, i.e. 0,1, or maybe 2.
3314 * With lots of accepts states, and having to try all of them,
3315 * it becomes quadratic on number of accept states to find all
3320 /* what type of TRIE am I? (utf8 makes this contextual) */
3321 DECL_TRIE_TYPE(scan);
3323 /* what trie are we using right now */
3324 reg_trie_data * const trie
3325 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3326 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3327 U32 state = trie->startstate;
3329 if (trie->bitmap && trie_type != trie_utf8_fold &&
3330 !TRIE_BITMAP_TEST(trie,*locinput)
3332 if (trie->states[ state ].wordnum) {
3334 PerlIO_printf(Perl_debug_log,
3335 "%*s %smatched empty string...%s\n",
3336 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3342 PerlIO_printf(Perl_debug_log,
3343 "%*s %sfailed to match trie start class...%s\n",
3344 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3351 U8 *uc = ( U8* )locinput;
3355 U8 *uscan = (U8*)NULL;
3356 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3357 U32 charcount = 0; /* how many input chars we have matched */
3358 U32 accepted = 0; /* have we seen any accepting states? */
3361 ST.jump = trie->jump;
3364 ST.longfold = FALSE; /* char longer if folded => it's harder */
3367 /* fully traverse the TRIE; note the position of the
3368 shortest accept state and the wordnum of the longest
3371 while ( state && uc <= (U8*)PL_regeol ) {
3372 U32 base = trie->states[ state ].trans.base;
3376 wordnum = trie->states[ state ].wordnum;
3378 if (wordnum) { /* it's an accept state */
3381 /* record first match position */
3383 ST.firstpos = (U8*)locinput;
3388 ST.firstchars = charcount;
3391 if (!ST.nextword || wordnum < ST.nextword)
3392 ST.nextword = wordnum;
3393 ST.topword = wordnum;
3396 DEBUG_TRIE_EXECUTE_r({
3397 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3398 PerlIO_printf( Perl_debug_log,
3399 "%*s %sState: %4"UVxf" Accepted: %c ",
3400 2+depth * 2, "", PL_colors[4],
3401 (UV)state, (accepted ? 'Y' : 'N'));
3404 /* read a char and goto next state */
3407 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3408 uscan, len, uvc, charid, foldlen,
3415 base + charid - 1 - trie->uniquecharcount)) >= 0)
3417 && ((U32)offset < trie->lasttrans)
3418 && trie->trans[offset].check == state)
3420 state = trie->trans[offset].next;
3431 DEBUG_TRIE_EXECUTE_r(
3432 PerlIO_printf( Perl_debug_log,
3433 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3434 charid, uvc, (UV)state, PL_colors[5] );
3440 /* calculate total number of accept states */
3445 w = trie->wordinfo[w].prev;
3448 ST.accepted = accepted;
3452 PerlIO_printf( Perl_debug_log,
3453 "%*s %sgot %"IVdf" possible matches%s\n",
3454 REPORT_CODE_OFF + depth * 2, "",
3455 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3457 goto trie_first_try; /* jump into the fail handler */
3461 case TRIE_next_fail: /* we failed - try next alternative */
3463 REGCP_UNWIND(ST.cp);
3464 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3465 PL_regoffs[n].end = -1;
3466 *PL_reglastparen = n;
3468 if (!--ST.accepted) {
3470 PerlIO_printf( Perl_debug_log,
3471 "%*s %sTRIE failed...%s\n",
3472 REPORT_CODE_OFF+depth*2, "",
3479 /* Find next-highest word to process. Note that this code
3480 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3481 register U16 min = 0;
3483 register U16 const nextword = ST.nextword;
3484 register reg_trie_wordinfo * const wordinfo
3485 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3486 for (word=ST.topword; word; word=wordinfo[word].prev) {
3487 if (word > nextword && (!min || word < min))
3500 ST.lastparen = *PL_reglastparen;
3504 /* find start char of end of current word */
3506 U32 chars; /* how many chars to skip */
3507 U8 *uc = ST.firstpos;
3508 reg_trie_data * const trie
3509 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3511 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3513 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3517 /* the hard option - fold each char in turn and find
3518 * its folded length (which may be different */
3519 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3527 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3535 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3540 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3554 PL_reginput = (char *)uc;
3557 scan = (ST.jump && ST.jump[ST.nextword])
3558 ? ST.me + ST.jump[ST.nextword]
3562 PerlIO_printf( Perl_debug_log,
3563 "%*s %sTRIE matched word #%d, continuing%s\n",
3564 REPORT_CODE_OFF+depth*2, "",
3571 if (ST.accepted > 1 || has_cutgroup) {
3572 PUSH_STATE_GOTO(TRIE_next, scan);
3575 /* only one choice left - just continue */
3577 AV *const trie_words
3578 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3579 SV ** const tmp = av_fetch( trie_words,
3581 SV *sv= tmp ? sv_newmortal() : NULL;
3583 PerlIO_printf( Perl_debug_log,
3584 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3585 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3587 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3588 PL_colors[0], PL_colors[1],
3589 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3591 : "not compiled under -Dr",
3595 locinput = PL_reginput;
3596 nextchr = UCHARAT(locinput);
3597 continue; /* execute rest of RE */
3602 char *s = STRING(scan);
3604 if (utf8_target != UTF_PATTERN) {
3605 /* The target and the pattern have differing utf8ness. */
3607 const char * const e = s + ln;
3610 /* The target is utf8, the pattern is not utf8. */
3615 if (NATIVE_TO_UNI(*(U8*)s) !=
3616 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3624 /* The target is not utf8, the pattern is utf8. */
3629 if (NATIVE_TO_UNI(*((U8*)l)) !=
3630 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3638 nextchr = UCHARAT(locinput);
3641 /* The target and the pattern have the same utf8ness. */
3642 /* Inline the first character, for speed. */
3643 if (UCHARAT(s) != nextchr)
3645 if (PL_regeol - locinput < ln)
3647 if (ln > 1 && memNE(s, locinput, ln))
3650 nextchr = UCHARAT(locinput);
3655 const U8 * fold_array;
3657 U32 fold_utf8_flags;
3659 PL_reg_flags |= RF_tainted;
3660 folder = foldEQ_locale;
3661 fold_array = PL_fold_locale;
3662 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3666 folder = foldEQ_latin1;
3667 fold_array = PL_fold_latin1;
3668 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3672 folder = foldEQ_latin1;
3673 fold_array = PL_fold_latin1;
3674 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3679 fold_array = PL_fold;
3680 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3686 if (utf8_target || UTF_PATTERN) {
3687 /* Either target or the pattern are utf8. */
3688 const char * const l = locinput;
3689 char *e = PL_regeol;
3691 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3692 l, &e, 0, utf8_target, fold_utf8_flags))
3697 nextchr = UCHARAT(locinput);
3701 /* Neither the target nor the pattern are utf8 */
3702 if (UCHARAT(s) != nextchr &&
3703 UCHARAT(s) != fold_array[nextchr])
3707 if (PL_regeol - locinput < ln)
3709 if (ln > 1 && ! folder(s, locinput, ln))
3712 nextchr = UCHARAT(locinput);
3716 /* XXX Could improve efficiency by separating these all out using a
3717 * macro or in-line function. At that point regcomp.c would no longer
3718 * have to set the FLAGS fields of these */
3721 PL_reg_flags |= RF_tainted;
3729 /* was last char in word? */
3731 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3732 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3734 if (locinput == PL_bostr)
3737 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3739 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3741 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3742 ln = isALNUM_uni(ln);
3743 LOAD_UTF8_CHARCLASS_ALNUM();
3744 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3747 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3748 n = isALNUM_LC_utf8((U8*)locinput);
3753 /* Here the string isn't utf8, or is utf8 and only ascii
3754 * characters are to match \w. In the latter case looking at
3755 * the byte just prior to the current one may be just the final
3756 * byte of a multi-byte character. This is ok. There are two
3758 * 1) it is a single byte character, and then the test is doing
3759 * just what it's supposed to.
3760 * 2) it is a multi-byte character, in which case the final
3761 * byte is never mistakable for ASCII, and so the test
3762 * will say it is not a word character, which is the
3763 * correct answer. */
3764 ln = (locinput != PL_bostr) ?
3765 UCHARAT(locinput - 1) : '\n';
3766 switch (FLAGS(scan)) {
3767 case REGEX_UNICODE_CHARSET:
3768 ln = isWORDCHAR_L1(ln);
3769 n = isWORDCHAR_L1(nextchr);
3771 case REGEX_LOCALE_CHARSET:
3772 ln = isALNUM_LC(ln);
3773 n = isALNUM_LC(nextchr);
3775 case REGEX_DEPENDS_CHARSET:
3777 n = isALNUM(nextchr);
3779 case REGEX_ASCII_RESTRICTED_CHARSET:
3780 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3781 ln = isWORDCHAR_A(ln);
3782 n = isWORDCHAR_A(nextchr);
3785 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3789 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3791 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3796 if (utf8_target || state_num == ANYOFV) {
3797 STRLEN inclasslen = PL_regeol - locinput;
3798 if (locinput >= PL_regeol)
3801 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3803 locinput += inclasslen;
3804 nextchr = UCHARAT(locinput);
3809 nextchr = UCHARAT(locinput);
3810 if (!nextchr && locinput >= PL_regeol)
3812 if (!REGINCLASS(rex, scan, (U8*)locinput))
3814 nextchr = UCHARAT(++locinput);
3818 /* Special char classes - The defines start on line 129 or so */
3819 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3820 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3821 ALNUMU, NALNUMU, isWORDCHAR_L1,
3822 ALNUMA, NALNUMA, isWORDCHAR_A,
3825 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3826 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3827 SPACEU, NSPACEU, isSPACE_L1,
3828 SPACEA, NSPACEA, isSPACE_A,
3831 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3832 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3833 DIGITA, NDIGITA, isDIGIT_A,
3836 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3837 a Unicode extended Grapheme Cluster */
3838 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3839 extended Grapheme Cluster is:
3842 | Prepend* Begin Extend*
3845 Begin is (Hangul-syllable | ! Control)
3846 Extend is (Grapheme_Extend | Spacing_Mark)
3847 Control is [ GCB_Control CR LF ]
3849 The discussion below shows how the code for CLUMP is derived
3850 from this regex. Note that most of these concepts are from
3851 property values of the Grapheme Cluster Boundary (GCB) property.
3852 No code point can have multiple property values for a given
3853 property. Thus a code point in Prepend can't be in Control, but
3854 it must be in !Control. This is why Control above includes
3855 GCB_Control plus CR plus LF. The latter two are used in the GCB
3856 property separately, and so can't be in GCB_Control, even though
3857 they logically are controls. Control is not the same as gc=cc,
3858 but includes format and other characters as well.
3860 The Unicode definition of Hangul-syllable is:
3862 | (L* ( ( V | LV ) V* | LVT ) T*)
3865 Each of these is a value for the GCB property, and hence must be
3866 disjoint, so the order they are tested is immaterial, so the
3867 above can safely be changed to
3870 | (L* ( LVT | ( V | LV ) V*) T*)
3872 The last two terms can be combined like this:
3874 | (( LVT | ( V | LV ) V*) T*))
3876 And refactored into this:
3877 L* (L | LVT T* | V V* T* | LV V* T*)
3879 That means that if we have seen any L's at all we can quit
3880 there, but if the next character is an LVT, a V, or an LV we
3883 There is a subtlety with Prepend* which showed up in testing.
3884 Note that the Begin, and only the Begin is required in:
3885 | Prepend* Begin Extend*
3886 Also, Begin contains '! Control'. A Prepend must be a
3887 '! Control', which means it must also be a Begin. What it
3888 comes down to is that if we match Prepend* and then find no
3889 suitable Begin afterwards, that if we backtrack the last
3890 Prepend, that one will be a suitable Begin.
3893 if (locinput >= PL_regeol)
3895 if (! utf8_target) {
3897 /* Match either CR LF or '.', as all the other possibilities
3899 locinput++; /* Match the . or CR */
3900 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3902 && locinput < PL_regeol
3903 && UCHARAT(locinput) == '\n') locinput++;
3907 /* Utf8: See if is ( CR LF ); already know that locinput <
3908 * PL_regeol, so locinput+1 is in bounds */
3909 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3913 /* In case have to backtrack to beginning, then match '.' */
3914 char *starting = locinput;
3916 /* In case have to backtrack the last prepend */
3917 char *previous_prepend = 0;
3919 LOAD_UTF8_CHARCLASS_GCB();
3921 /* Match (prepend)* */
3922 while (locinput < PL_regeol
3923 && swash_fetch(PL_utf8_X_prepend,
3924 (U8*)locinput, utf8_target))
3926 previous_prepend = locinput;
3927 locinput += UTF8SKIP(locinput);
3930 /* As noted above, if we matched a prepend character, but
3931 * the next thing won't match, back off the last prepend we
3932 * matched, as it is guaranteed to match the begin */
3933 if (previous_prepend
3934 && (locinput >= PL_regeol
3935 || ! swash_fetch(PL_utf8_X_begin,
3936 (U8*)locinput, utf8_target)))
3938 locinput = previous_prepend;
3941 /* Note that here we know PL_regeol > locinput, as we
3942 * tested that upon input to this switch case, and if we
3943 * moved locinput forward, we tested the result just above
3944 * and it either passed, or we backed off so that it will
3946 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3948 /* Here did not match the required 'Begin' in the
3949 * second term. So just match the very first
3950 * character, the '.' of the final term of the regex */
3951 locinput = starting + UTF8SKIP(starting);
3954 /* Here is the beginning of a character that can have
3955 * an extender. It is either a hangul syllable, or a
3957 if (swash_fetch(PL_utf8_X_non_hangul,
3958 (U8*)locinput, utf8_target))
3961 /* Here not a Hangul syllable, must be a
3962 * ('! * Control') */
3963 locinput += UTF8SKIP(locinput);
3966 /* Here is a Hangul syllable. It can be composed
3967 * of several individual characters. One
3968 * possibility is T+ */
3969 if (swash_fetch(PL_utf8_X_T,
3970 (U8*)locinput, utf8_target))
3972 while (locinput < PL_regeol
3973 && swash_fetch(PL_utf8_X_T,
3974 (U8*)locinput, utf8_target))
3976 locinput += UTF8SKIP(locinput);
3980 /* Here, not T+, but is a Hangul. That means
3981 * it is one of the others: L, LV, LVT or V,
3983 * L* (L | LVT T* | V V* T* | LV V* T*) */
3986 while (locinput < PL_regeol
3987 && swash_fetch(PL_utf8_X_L,
3988 (U8*)locinput, utf8_target))
3990 locinput += UTF8SKIP(locinput);
3993 /* Here, have exhausted L*. If the next
3994 * character is not an LV, LVT nor V, it means
3995 * we had to have at least one L, so matches L+
3996 * in the original equation, we have a complete
3997 * hangul syllable. Are done. */
3999 if (locinput < PL_regeol
4000 && swash_fetch(PL_utf8_X_LV_LVT_V,
4001 (U8*)locinput, utf8_target))
4004 /* Otherwise keep going. Must be LV, LVT
4005 * or V. See if LVT */
4006 if (swash_fetch(PL_utf8_X_LVT,
4007 (U8*)locinput, utf8_target))
4009 locinput += UTF8SKIP(locinput);
4012 /* Must be V or LV. Take it, then
4014 locinput += UTF8SKIP(locinput);
4015 while (locinput < PL_regeol
4016 && swash_fetch(PL_utf8_X_V,
4017 (U8*)locinput, utf8_target))
4019 locinput += UTF8SKIP(locinput);
4023 /* And any of LV, LVT, or V can be followed
4025 while (locinput < PL_regeol
4026 && swash_fetch(PL_utf8_X_T,
4030 locinput += UTF8SKIP(locinput);
4036 /* Match any extender */
4037 while (locinput < PL_regeol
4038 && swash_fetch(PL_utf8_X_extend,
4039 (U8*)locinput, utf8_target))
4041 locinput += UTF8SKIP(locinput);
4045 if (locinput > PL_regeol) sayNO;
4047 nextchr = UCHARAT(locinput);
4051 { /* The capture buffer cases. The ones beginning with N for the
4052 named buffers just convert to the equivalent numbered and
4053 pretend they were called as the corresponding numbered buffer
4055 /* don't initialize these in the declaration, it makes C++
4060 const U8 *fold_array;
4063 PL_reg_flags |= RF_tainted;
4064 folder = foldEQ_locale;
4065 fold_array = PL_fold_locale;
4067 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4071 folder = foldEQ_latin1;
4072 fold_array = PL_fold_latin1;
4074 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4078 folder = foldEQ_latin1;
4079 fold_array = PL_fold_latin1;
4081 utf8_fold_flags = 0;
4086 fold_array = PL_fold;
4088 utf8_fold_flags = 0;
4095 utf8_fold_flags = 0;
4098 /* For the named back references, find the corresponding buffer
4100 n = reg_check_named_buff_matched(rex,scan);
4105 goto do_nref_ref_common;
4108 PL_reg_flags |= RF_tainted;
4109 folder = foldEQ_locale;
4110 fold_array = PL_fold_locale;
4111 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4115 folder = foldEQ_latin1;
4116 fold_array = PL_fold_latin1;
4117 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4121 folder = foldEQ_latin1;
4122 fold_array = PL_fold_latin1;
4123 utf8_fold_flags = 0;
4128 fold_array = PL_fold;
4129 utf8_fold_flags = 0;
4135 utf8_fold_flags = 0;
4139 n = ARG(scan); /* which paren pair */
4142 ln = PL_regoffs[n].start;
4143 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4144 if (*PL_reglastparen < n || ln == -1)
4145 sayNO; /* Do not match unless seen CLOSEn. */
4146 if (ln == PL_regoffs[n].end)
4150 if (type != REF /* REF can do byte comparison */
4151 && (utf8_target || type == REFFU))
4152 { /* XXX handle REFFL better */
4153 char * limit = PL_regeol;
4155 /* This call case insensitively compares the entire buffer
4156 * at s, with the current input starting at locinput, but
4157 * not going off the end given by PL_regeol, and returns in
4158 * limit upon success, how much of the current input was
4160 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4161 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4166 nextchr = UCHARAT(locinput);
4170 /* Not utf8: Inline the first character, for speed. */
4171 if (UCHARAT(s) != nextchr &&
4173 UCHARAT(s) != fold_array[nextchr]))
4175 ln = PL_regoffs[n].end - ln;
4176 if (locinput + ln > PL_regeol)
4178 if (ln > 1 && (type == REF
4179 ? memNE(s, locinput, ln)
4180 : ! folder(s, locinput, ln)))
4183 nextchr = UCHARAT(locinput);
4193 #define ST st->u.eval
4198 regexp_internal *rei;
4199 regnode *startpoint;
4202 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4203 if (cur_eval && cur_eval->locinput==locinput) {
4204 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4205 Perl_croak(aTHX_ "Infinite recursion in regex");
4206 if ( ++nochange_depth > max_nochange_depth )
4208 "Pattern subroutine nesting without pos change"
4209 " exceeded limit in regex");
4216 (void)ReREFCNT_inc(rex_sv);
4217 if (OP(scan)==GOSUB) {
4218 startpoint = scan + ARG2L(scan);
4219 ST.close_paren = ARG(scan);
4221 startpoint = rei->program+1;
4224 goto eval_recurse_doit;
4226 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4227 if (cur_eval && cur_eval->locinput==locinput) {
4228 if ( ++nochange_depth > max_nochange_depth )
4229 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4234 /* execute the code in the {...} */
4236 SV ** const before = SP;
4237 OP_4tree * const oop = PL_op;
4238 COP * const ocurcop = PL_curcop;
4240 char *saved_regeol = PL_regeol;
4241 struct re_save_state saved_state;
4243 /* To not corrupt the existing regex state while executing the
4244 * eval we would normally put it on the save stack, like with
4245 * save_re_context. However, re-evals have a weird scoping so we
4246 * can't just add ENTER/LEAVE here. With that, things like
4248 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4250 * would break, as they expect the localisation to be unwound
4251 * only when the re-engine backtracks through the bit that
4254 * What we do instead is just saving the state in a local c
4257 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4260 PL_op = (OP_4tree*)rexi->data->data[n];
4261 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4262 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4263 /* wrap the call in two SAVECOMPPADs. This ensures that
4264 * when the save stack is eventually unwound, all the
4265 * accumulated SAVEt_CLEARSV's will be processed with
4266 * interspersed SAVEt_COMPPAD's to ensure that lexicals
4267 * are cleared in the right pad */
4269 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4270 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4273 SV *sv_mrk = get_sv("REGMARK", 1);
4274 sv_setsv(sv_mrk, sv_yes_mark);
4277 CALLRUNOPS(aTHX); /* Scalar context. */
4280 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4286 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4290 PAD_RESTORE_LOCAL(old_comppad);
4291 PL_curcop = ocurcop;
4292 PL_regeol = saved_regeol;
4295 sv_setsv(save_scalar(PL_replgv), ret);
4299 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4302 /* extract RE object from returned value; compiling if
4308 SV *const sv = SvRV(ret);
4310 if (SvTYPE(sv) == SVt_REGEXP) {
4312 } else if (SvSMAGICAL(sv)) {
4313 mg = mg_find(sv, PERL_MAGIC_qr);
4316 } else if (SvTYPE(ret) == SVt_REGEXP) {
4318 } else if (SvSMAGICAL(ret)) {
4319 if (SvGMAGICAL(ret)) {
4320 /* I don't believe that there is ever qr magic
4322 assert(!mg_find(ret, PERL_MAGIC_qr));
4323 sv_unmagic(ret, PERL_MAGIC_qr);
4326 mg = mg_find(ret, PERL_MAGIC_qr);
4327 /* testing suggests mg only ends up non-NULL for
4328 scalars who were upgraded and compiled in the
4329 else block below. In turn, this is only
4330 triggered in the "postponed utf8 string" tests
4336 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4340 rx = reg_temp_copy(NULL, rx);
4344 const I32 osize = PL_regsize;
4347 assert (SvUTF8(ret));
4348 } else if (SvUTF8(ret)) {
4349 /* Not doing UTF-8, despite what the SV says. Is
4350 this only if we're trapped in use 'bytes'? */
4351 /* Make a copy of the octet sequence, but without
4352 the flag on, as the compiler now honours the
4353 SvUTF8 flag on ret. */
4355 const char *const p = SvPV(ret, len);
4356 ret = newSVpvn_flags(p, len, SVs_TEMP);
4358 rx = CALLREGCOMP(ret, pm_flags);
4360 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4362 /* This isn't a first class regexp. Instead, it's
4363 caching a regexp onto an existing, Perl visible
4365 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4370 re = (struct regexp *)SvANY(rx);
4372 RXp_MATCH_COPIED_off(re);
4373 re->subbeg = rex->subbeg;
4374 re->sublen = rex->sublen;
4377 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4378 "Matching embedded");
4380 startpoint = rei->program + 1;
4381 ST.close_paren = 0; /* only used for GOSUB */
4382 /* borrowed from regtry */
4383 if (PL_reg_start_tmpl <= re->nparens) {
4384 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4385 if(PL_reg_start_tmp)
4386 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4388 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4391 eval_recurse_doit: /* Share code with GOSUB below this line */
4392 /* run the pattern returned from (??{...}) */
4393 ST.cp = regcppush(0); /* Save *all* the positions. */
4394 REGCP_SET(ST.lastcp);
4396 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4398 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4399 PL_reglastparen = &re->lastparen;
4400 PL_reglastcloseparen = &re->lastcloseparen;
4402 re->lastcloseparen = 0;
4404 PL_reginput = locinput;
4407 /* XXXX This is too dramatic a measure... */
4410 ST.toggle_reg_flags = PL_reg_flags;
4412 PL_reg_flags |= RF_utf8;
4414 PL_reg_flags &= ~RF_utf8;
4415 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4417 ST.prev_rex = rex_sv;
4418 ST.prev_curlyx = cur_curlyx;
4419 SETREX(rex_sv,re_sv);
4424 ST.prev_eval = cur_eval;
4426 /* now continue from first node in postoned RE */
4427 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4430 /* logical is 1, /(?(?{...})X|Y)/ */
4431 sw = cBOOL(SvTRUE(ret));
4436 case EVAL_AB: /* cleanup after a successful (??{A})B */
4437 /* note: this is called twice; first after popping B, then A */
4438 PL_reg_flags ^= ST.toggle_reg_flags;
4439 ReREFCNT_dec(rex_sv);
4440 SETREX(rex_sv,ST.prev_rex);
4441 rex = (struct regexp *)SvANY(rex_sv);
4442 rexi = RXi_GET(rex);
4444 cur_eval = ST.prev_eval;
4445 cur_curlyx = ST.prev_curlyx;
4447 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4448 PL_reglastparen = &rex->lastparen;
4449 PL_reglastcloseparen = &rex->lastcloseparen;
4450 /* also update PL_regoffs */
4451 PL_regoffs = rex->offs;
4453 /* XXXX This is too dramatic a measure... */
4455 if ( nochange_depth )
4460 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4461 /* note: this is called twice; first after popping B, then A */
4462 PL_reg_flags ^= ST.toggle_reg_flags;
4463 ReREFCNT_dec(rex_sv);
4464 SETREX(rex_sv,ST.prev_rex);
4465 rex = (struct regexp *)SvANY(rex_sv);
4466 rexi = RXi_GET(rex);
4467 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4468 PL_reglastparen = &rex->lastparen;
4469 PL_reglastcloseparen = &rex->lastcloseparen;
4471 PL_reginput = locinput;
4472 REGCP_UNWIND(ST.lastcp);
4474 cur_eval = ST.prev_eval;
4475 cur_curlyx = ST.prev_curlyx;
4476 /* XXXX This is too dramatic a measure... */
4478 if ( nochange_depth )
4484 n = ARG(scan); /* which paren pair */
4485 PL_reg_start_tmp[n] = locinput;
4491 n = ARG(scan); /* which paren pair */
4492 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4493 PL_regoffs[n].end = locinput - PL_bostr;
4494 /*if (n > PL_regsize)
4496 if (n > *PL_reglastparen)
4497 *PL_reglastparen = n;
4498 *PL_reglastcloseparen = n;
4499 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4507 cursor && OP(cursor)!=END;
4508 cursor=regnext(cursor))
4510 if ( OP(cursor)==CLOSE ){
4512 if ( n <= lastopen ) {
4514 = PL_reg_start_tmp[n] - PL_bostr;
4515 PL_regoffs[n].end = locinput - PL_bostr;
4516 /*if (n > PL_regsize)
4518 if (n > *PL_reglastparen)
4519 *PL_reglastparen = n;
4520 *PL_reglastcloseparen = n;
4521 if ( n == ARG(scan) || (cur_eval &&
4522 cur_eval->u.eval.close_paren == n))
4531 n = ARG(scan); /* which paren pair */
4532 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4535 /* reg_check_named_buff_matched returns 0 for no match */
4536 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4540 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4546 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4548 next = NEXTOPER(NEXTOPER(scan));
4550 next = scan + ARG(scan);
4551 if (OP(next) == IFTHEN) /* Fake one. */
4552 next = NEXTOPER(NEXTOPER(next));
4556 logical = scan->flags;
4559 /*******************************************************************
4561 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4562 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4563 STAR/PLUS/CURLY/CURLYN are used instead.)
4565 A*B is compiled as <CURLYX><A><WHILEM><B>
4567 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4568 state, which contains the current count, initialised to -1. It also sets
4569 cur_curlyx to point to this state, with any previous value saved in the
4572 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4573 since the pattern may possibly match zero times (i.e. it's a while {} loop
4574 rather than a do {} while loop).
4576 Each entry to WHILEM represents a successful match of A. The count in the
4577 CURLYX block is incremented, another WHILEM state is pushed, and execution
4578 passes to A or B depending on greediness and the current count.
4580 For example, if matching against the string a1a2a3b (where the aN are
4581 substrings that match /A/), then the match progresses as follows: (the
4582 pushed states are interspersed with the bits of strings matched so far):
4585 <CURLYX cnt=0><WHILEM>
4586 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4587 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4588 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4589 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4591 (Contrast this with something like CURLYM, which maintains only a single
4595 a1 <CURLYM cnt=1> a2
4596 a1 a2 <CURLYM cnt=2> a3
4597 a1 a2 a3 <CURLYM cnt=3> b
4600 Each WHILEM state block marks a point to backtrack to upon partial failure
4601 of A or B, and also contains some minor state data related to that
4602 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4603 overall state, such as the count, and pointers to the A and B ops.
4605 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4606 must always point to the *current* CURLYX block, the rules are:
4608 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4609 and set cur_curlyx to point the new block.
4611 When popping the CURLYX block after a successful or unsuccessful match,
4612 restore the previous cur_curlyx.
4614 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4615 to the outer one saved in the CURLYX block.
4617 When popping the WHILEM block after a successful or unsuccessful B match,
4618 restore the previous cur_curlyx.
4620 Here's an example for the pattern (AI* BI)*BO
4621 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4624 curlyx backtrack stack
4625 ------ ---------------
4627 CO <CO prev=NULL> <WO>
4628 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4629 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4630 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4632 At this point the pattern succeeds, and we work back down the stack to
4633 clean up, restoring as we go:
4635 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4636 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4637 CO <CO prev=NULL> <WO>
4640 *******************************************************************/
4642 #define ST st->u.curlyx
4644 case CURLYX: /* start of /A*B/ (for complex A) */
4646 /* No need to save/restore up to this paren */
4647 I32 parenfloor = scan->flags;
4649 assert(next); /* keep Coverity happy */
4650 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4653 /* XXXX Probably it is better to teach regpush to support
4654 parenfloor > PL_regsize... */
4655 if (parenfloor > (I32)*PL_reglastparen)
4656 parenfloor = *PL_reglastparen; /* Pessimization... */
4658 ST.prev_curlyx= cur_curlyx;
4660 ST.cp = PL_savestack_ix;
4662 /* these fields contain the state of the current curly.
4663 * they are accessed by subsequent WHILEMs */
4664 ST.parenfloor = parenfloor;
4669 ST.count = -1; /* this will be updated by WHILEM */
4670 ST.lastloc = NULL; /* this will be updated by WHILEM */
4672 PL_reginput = locinput;
4673 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4677 case CURLYX_end: /* just finished matching all of A*B */
4678 cur_curlyx = ST.prev_curlyx;
4682 case CURLYX_end_fail: /* just failed to match all of A*B */
4684 cur_curlyx = ST.prev_curlyx;
4690 #define ST st->u.whilem
4692 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4694 /* see the discussion above about CURLYX/WHILEM */
4696 int min = ARG1(cur_curlyx->u.curlyx.me);
4697 int max = ARG2(cur_curlyx->u.curlyx.me);
4698 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4700 assert(cur_curlyx); /* keep Coverity happy */
4701 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4702 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4703 ST.cache_offset = 0;
4706 PL_reginput = locinput;
4708 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4709 "%*s whilem: matched %ld out of %d..%d\n",
4710 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4713 /* First just match a string of min A's. */
4716 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4717 cur_curlyx->u.curlyx.lastloc = locinput;
4718 REGCP_SET(ST.lastcp);
4720 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4724 /* If degenerate A matches "", assume A done. */
4726 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4727 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4728 "%*s whilem: empty match detected, trying continuation...\n",
4729 REPORT_CODE_OFF+depth*2, "")
4731 goto do_whilem_B_max;
4734 /* super-linear cache processing */
4738 if (!PL_reg_maxiter) {
4739 /* start the countdown: Postpone detection until we
4740 * know the match is not *that* much linear. */
4741 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4742 /* possible overflow for long strings and many CURLYX's */
4743 if (PL_reg_maxiter < 0)
4744 PL_reg_maxiter = I32_MAX;
4745 PL_reg_leftiter = PL_reg_maxiter;
4748 if (PL_reg_leftiter-- == 0) {
4749 /* initialise cache */
4750 const I32 size = (PL_reg_maxiter + 7)/8;
4751 if (PL_reg_poscache) {
4752 if ((I32)PL_reg_poscache_size < size) {
4753 Renew(PL_reg_poscache, size, char);
4754 PL_reg_poscache_size = size;
4756 Zero(PL_reg_poscache, size, char);
4759 PL_reg_poscache_size = size;
4760 Newxz(PL_reg_poscache, size, char);
4762 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4763 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4764 PL_colors[4], PL_colors[5])
4768 if (PL_reg_leftiter < 0) {
4769 /* have we already failed at this position? */
4771 offset = (scan->flags & 0xf) - 1
4772 + (locinput - PL_bostr) * (scan->flags>>4);
4773 mask = 1 << (offset % 8);
4775 if (PL_reg_poscache[offset] & mask) {
4776 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4777 "%*s whilem: (cache) already tried at this position...\n",
4778 REPORT_CODE_OFF+depth*2, "")
4780 sayNO; /* cache records failure */
4782 ST.cache_offset = offset;
4783 ST.cache_mask = mask;
4787 /* Prefer B over A for minimal matching. */
4789 if (cur_curlyx->u.curlyx.minmod) {
4790 ST.save_curlyx = cur_curlyx;
4791 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4792 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4793 REGCP_SET(ST.lastcp);
4794 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4798 /* Prefer A over B for maximal matching. */
4800 if (n < max) { /* More greed allowed? */
4801 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4802 cur_curlyx->u.curlyx.lastloc = locinput;
4803 REGCP_SET(ST.lastcp);
4804 PUSH_STATE_GOTO(WHILEM_A_max, A);
4807 goto do_whilem_B_max;
4811 case WHILEM_B_min: /* just matched B in a minimal match */
4812 case WHILEM_B_max: /* just matched B in a maximal match */
4813 cur_curlyx = ST.save_curlyx;
4817 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4818 cur_curlyx = ST.save_curlyx;
4819 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4820 cur_curlyx->u.curlyx.count--;
4824 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4826 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4827 REGCP_UNWIND(ST.lastcp);
4829 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4830 cur_curlyx->u.curlyx.count--;
4834 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4835 REGCP_UNWIND(ST.lastcp);
4836 regcppop(rex); /* Restore some previous $<digit>s? */
4837 PL_reginput = locinput;
4838 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4839 "%*s whilem: failed, trying continuation...\n",
4840 REPORT_CODE_OFF+depth*2, "")
4843 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4844 && ckWARN(WARN_REGEXP)
4845 && !(PL_reg_flags & RF_warned))
4847 PL_reg_flags |= RF_warned;
4848 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4849 "Complex regular subexpression recursion",
4854 ST.save_curlyx = cur_curlyx;
4855 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4856 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4859 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4860 cur_curlyx = ST.save_curlyx;
4861 REGCP_UNWIND(ST.lastcp);
4864 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4865 /* Maximum greed exceeded */
4866 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4867 && ckWARN(WARN_REGEXP)
4868 && !(PL_reg_flags & RF_warned))
4870 PL_reg_flags |= RF_warned;
4871 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4872 "%s limit (%d) exceeded",
4873 "Complex regular subexpression recursion",
4876 cur_curlyx->u.curlyx.count--;
4880 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4881 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4883 /* Try grabbing another A and see if it helps. */
4884 PL_reginput = locinput;
4885 cur_curlyx->u.curlyx.lastloc = locinput;
4886 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4887 REGCP_SET(ST.lastcp);
4888 PUSH_STATE_GOTO(WHILEM_A_min,
4889 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4893 #define ST st->u.branch
4895 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4896 next = scan + ARG(scan);
4899 scan = NEXTOPER(scan);
4902 case BRANCH: /* /(...|A|...)/ */
4903 scan = NEXTOPER(scan); /* scan now points to inner node */
4904 ST.lastparen = *PL_reglastparen;
4905 ST.next_branch = next;
4907 PL_reginput = locinput;
4909 /* Now go into the branch */
4911 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4913 PUSH_STATE_GOTO(BRANCH_next, scan);
4917 PL_reginput = locinput;
4918 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4919 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4920 PUSH_STATE_GOTO(CUTGROUP_next,next);
4922 case CUTGROUP_next_fail:
4925 if (st->u.mark.mark_name)
4926 sv_commit = st->u.mark.mark_name;
4932 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4937 REGCP_UNWIND(ST.cp);
4938 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4939 PL_regoffs[n].end = -1;
4940 *PL_reglastparen = n;
4941 /*dmq: *PL_reglastcloseparen = n; */
4942 scan = ST.next_branch;
4943 /* no more branches? */
4944 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4946 PerlIO_printf( Perl_debug_log,
4947 "%*s %sBRANCH failed...%s\n",
4948 REPORT_CODE_OFF+depth*2, "",
4954 continue; /* execute next BRANCH[J] op */
4962 #define ST st->u.curlym
4964 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4966 /* This is an optimisation of CURLYX that enables us to push
4967 * only a single backtracking state, no matter how many matches
4968 * there are in {m,n}. It relies on the pattern being constant
4969 * length, with no parens to influence future backrefs
4973 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4975 /* if paren positive, emulate an OPEN/CLOSE around A */
4977 U32 paren = ST.me->flags;
4978 if (paren > PL_regsize)
4980 if (paren > *PL_reglastparen)
4981 *PL_reglastparen = paren;
4982 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4990 ST.c1 = CHRTEST_UNINIT;
4993 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4996 curlym_do_A: /* execute the A in /A{m,n}B/ */
4997 PL_reginput = locinput;
4998 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
5001 case CURLYM_A: /* we've just matched an A */
5002 locinput = st->locinput;
5003 nextchr = UCHARAT(locinput);
5006 /* after first match, determine A's length: u.curlym.alen */
5007 if (ST.count == 1) {
5008 if (PL_reg_match_utf8) {
5010 while (s < PL_reginput) {
5016 ST.alen = PL_reginput - locinput;
5019 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5022 PerlIO_printf(Perl_debug_log,
5023 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5024 (int)(REPORT_CODE_OFF+(depth*2)), "",
5025 (IV) ST.count, (IV)ST.alen)
5028 locinput = PL_reginput;
5030 if (cur_eval && cur_eval->u.eval.close_paren &&
5031 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5035 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5036 if ( max == REG_INFTY || ST.count < max )
5037 goto curlym_do_A; /* try to match another A */
5039 goto curlym_do_B; /* try to match B */
5041 case CURLYM_A_fail: /* just failed to match an A */
5042 REGCP_UNWIND(ST.cp);
5044 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5045 || (cur_eval && cur_eval->u.eval.close_paren &&
5046 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5049 curlym_do_B: /* execute the B in /A{m,n}B/ */
5050 PL_reginput = locinput;
5051 if (ST.c1 == CHRTEST_UNINIT) {
5052 /* calculate c1 and c2 for possible match of 1st char
5053 * following curly */
5054 ST.c1 = ST.c2 = CHRTEST_VOID;
5055 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5056 regnode *text_node = ST.B;
5057 if (! HAS_TEXT(text_node))
5058 FIND_NEXT_IMPT(text_node);
5061 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5063 But the former is redundant in light of the latter.
5065 if this changes back then the macro for
5066 IS_TEXT and friends need to change.
5068 if (PL_regkind[OP(text_node)] == EXACT)
5071 ST.c1 = (U8)*STRING(text_node);
5072 switch (OP(text_node)) {
5073 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5075 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5076 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5077 default: ST.c2 = ST.c1;
5084 PerlIO_printf(Perl_debug_log,
5085 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5086 (int)(REPORT_CODE_OFF+(depth*2)),
5089 if (ST.c1 != CHRTEST_VOID
5090 && UCHARAT(PL_reginput) != ST.c1
5091 && UCHARAT(PL_reginput) != ST.c2)
5093 /* simulate B failing */
5095 PerlIO_printf(Perl_debug_log,
5096 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5097 (int)(REPORT_CODE_OFF+(depth*2)),"",
5100 state_num = CURLYM_B_fail;
5101 goto reenter_switch;
5105 /* mark current A as captured */
5106 I32 paren = ST.me->flags;
5108 PL_regoffs[paren].start
5109 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5110 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5111 /*dmq: *PL_reglastcloseparen = paren; */
5114 PL_regoffs[paren].end = -1;
5115 if (cur_eval && cur_eval->u.eval.close_paren &&
5116 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5125 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5128 case CURLYM_B_fail: /* just failed to match a B */
5129 REGCP_UNWIND(ST.cp);
5131 I32 max = ARG2(ST.me);
5132 if (max != REG_INFTY && ST.count == max)
5134 goto curlym_do_A; /* try to match a further A */
5136 /* backtrack one A */
5137 if (ST.count == ARG1(ST.me) /* min */)
5140 locinput = HOPc(locinput, -ST.alen);
5141 goto curlym_do_B; /* try to match B */
5144 #define ST st->u.curly
5146 #define CURLY_SETPAREN(paren, success) \
5149 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5150 PL_regoffs[paren].end = locinput - PL_bostr; \
5151 *PL_reglastcloseparen = paren; \
5154 PL_regoffs[paren].end = -1; \
5157 case STAR: /* /A*B/ where A is width 1 */
5161 scan = NEXTOPER(scan);
5163 case PLUS: /* /A+B/ where A is width 1 */
5167 scan = NEXTOPER(scan);
5169 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5170 ST.paren = scan->flags; /* Which paren to set */
5171 if (ST.paren > PL_regsize)
5172 PL_regsize = ST.paren;
5173 if (ST.paren > *PL_reglastparen)
5174 *PL_reglastparen = ST.paren;
5175 ST.min = ARG1(scan); /* min to match */
5176 ST.max = ARG2(scan); /* max to match */
5177 if (cur_eval && cur_eval->u.eval.close_paren &&
5178 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5182 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5184 case CURLY: /* /A{m,n}B/ where A is width 1 */
5186 ST.min = ARG1(scan); /* min to match */
5187 ST.max = ARG2(scan); /* max to match */
5188 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5191 * Lookahead to avoid useless match attempts
5192 * when we know what character comes next.
5194 * Used to only do .*x and .*?x, but now it allows
5195 * for )'s, ('s and (?{ ... })'s to be in the way
5196 * of the quantifier and the EXACT-like node. -- japhy
5199 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5201 if (HAS_TEXT(next) || JUMPABLE(next)) {
5203 regnode *text_node = next;
5205 if (! HAS_TEXT(text_node))
5206 FIND_NEXT_IMPT(text_node);
5208 if (! HAS_TEXT(text_node))
5209 ST.c1 = ST.c2 = CHRTEST_VOID;
5211 if ( PL_regkind[OP(text_node)] != EXACT ) {
5212 ST.c1 = ST.c2 = CHRTEST_VOID;
5213 goto assume_ok_easy;
5216 s = (U8*)STRING(text_node);
5218 /* Currently we only get here when
5220 PL_rekind[OP(text_node)] == EXACT
5222 if this changes back then the macro for IS_TEXT and
5223 friends need to change. */
5226 switch (OP(text_node)) {
5227 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5229 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5230 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5231 default: ST.c2 = ST.c1; break;
5234 else { /* UTF_PATTERN */
5235 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5236 STRLEN ulen1, ulen2;
5237 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5238 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5240 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5241 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5243 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5245 0 : UTF8_ALLOW_ANY);
5246 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5248 0 : UTF8_ALLOW_ANY);
5250 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5252 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5257 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5264 ST.c1 = ST.c2 = CHRTEST_VOID;
5269 PL_reginput = locinput;
5272 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5275 locinput = PL_reginput;
5277 if (ST.c1 == CHRTEST_VOID)
5278 goto curly_try_B_min;
5280 ST.oldloc = locinput;
5282 /* set ST.maxpos to the furthest point along the
5283 * string that could possibly match */
5284 if (ST.max == REG_INFTY) {
5285 ST.maxpos = PL_regeol - 1;
5287 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5290 else if (utf8_target) {
5291 int m = ST.max - ST.min;
5292 for (ST.maxpos = locinput;
5293 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5294 ST.maxpos += UTF8SKIP(ST.maxpos);
5297 ST.maxpos = locinput + ST.max - ST.min;
5298 if (ST.maxpos >= PL_regeol)
5299 ST.maxpos = PL_regeol - 1;
5301 goto curly_try_B_min_known;
5305 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5306 locinput = PL_reginput;
5307 if (ST.count < ST.min)
5309 if ((ST.count > ST.min)
5310 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5312 /* A{m,n} must come at the end of the string, there's
5313 * no point in backing off ... */
5315 /* ...except that $ and \Z can match before *and* after
5316 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5317 We may back off by one in this case. */
5318 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5322 goto curly_try_B_max;
5327 case CURLY_B_min_known_fail:
5328 /* failed to find B in a non-greedy match where c1,c2 valid */
5329 if (ST.paren && ST.count)
5330 PL_regoffs[ST.paren].end = -1;
5332 PL_reginput = locinput; /* Could be reset... */
5333 REGCP_UNWIND(ST.cp);
5334 /* Couldn't or didn't -- move forward. */
5335 ST.oldloc = locinput;
5337 locinput += UTF8SKIP(locinput);
5341 curly_try_B_min_known:
5342 /* find the next place where 'B' could work, then call B */
5346 n = (ST.oldloc == locinput) ? 0 : 1;
5347 if (ST.c1 == ST.c2) {
5349 /* set n to utf8_distance(oldloc, locinput) */
5350 while (locinput <= ST.maxpos &&
5351 utf8n_to_uvchr((U8*)locinput,
5352 UTF8_MAXBYTES, &len,
5353 uniflags) != (UV)ST.c1) {
5359 /* set n to utf8_distance(oldloc, locinput) */
5360 while (locinput <= ST.maxpos) {
5362 const UV c = utf8n_to_uvchr((U8*)locinput,
5363 UTF8_MAXBYTES, &len,
5365 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5373 if (ST.c1 == ST.c2) {
5374 while (locinput <= ST.maxpos &&
5375 UCHARAT(locinput) != ST.c1)
5379 while (locinput <= ST.maxpos
5380 && UCHARAT(locinput) != ST.c1
5381 && UCHARAT(locinput) != ST.c2)
5384 n = locinput - ST.oldloc;
5386 if (locinput > ST.maxpos)
5388 /* PL_reginput == oldloc now */
5391 if (regrepeat(rex, ST.A, n, depth) < n)
5394 PL_reginput = locinput;
5395 CURLY_SETPAREN(ST.paren, ST.count);
5396 if (cur_eval && cur_eval->u.eval.close_paren &&
5397 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5400 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5405 case CURLY_B_min_fail:
5406 /* failed to find B in a non-greedy match where c1,c2 invalid */
5407 if (ST.paren && ST.count)
5408 PL_regoffs[ST.paren].end = -1;
5410 REGCP_UNWIND(ST.cp);
5411 /* failed -- move forward one */
5412 PL_reginput = locinput;
5413 if (regrepeat(rex, ST.A, 1, depth)) {
5415 locinput = PL_reginput;
5416 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5417 ST.count > 0)) /* count overflow ? */
5420 CURLY_SETPAREN(ST.paren, ST.count);
5421 if (cur_eval && cur_eval->u.eval.close_paren &&
5422 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5425 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5433 /* a successful greedy match: now try to match B */
5434 if (cur_eval && cur_eval->u.eval.close_paren &&
5435 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5440 if (ST.c1 != CHRTEST_VOID)
5441 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5442 UTF8_MAXBYTES, 0, uniflags)
5443 : (UV) UCHARAT(PL_reginput);
5444 /* If it could work, try it. */
5445 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5446 CURLY_SETPAREN(ST.paren, ST.count);
5447 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5452 case CURLY_B_max_fail:
5453 /* failed to find B in a greedy match */
5454 if (ST.paren && ST.count)
5455 PL_regoffs[ST.paren].end = -1;
5457 REGCP_UNWIND(ST.cp);
5459 if (--ST.count < ST.min)
5461 PL_reginput = locinput = HOPc(locinput, -1);
5462 goto curly_try_B_max;
5469 /* we've just finished A in /(??{A})B/; now continue with B */
5471 st->u.eval.toggle_reg_flags
5472 = cur_eval->u.eval.toggle_reg_flags;
5473 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5475 st->u.eval.prev_rex = rex_sv; /* inner */
5476 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5477 rex = (struct regexp *)SvANY(rex_sv);
5478 rexi = RXi_GET(rex);
5479 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5480 (void)ReREFCNT_inc(rex_sv);
5481 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5483 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5484 PL_reglastparen = &rex->lastparen;
5485 PL_reglastcloseparen = &rex->lastcloseparen;
5487 REGCP_SET(st->u.eval.lastcp);
5488 PL_reginput = locinput;
5490 /* Restore parens of the outer rex without popping the
5492 tmpix = PL_savestack_ix;
5493 PL_savestack_ix = cur_eval->u.eval.lastcp;
5495 PL_savestack_ix = tmpix;
5497 st->u.eval.prev_eval = cur_eval;
5498 cur_eval = cur_eval->u.eval.prev_eval;
5500 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5501 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5502 if ( nochange_depth )
5505 PUSH_YES_STATE_GOTO(EVAL_AB,
5506 st->u.eval.prev_eval->u.eval.B); /* match B */
5509 if (locinput < reginfo->till) {
5510 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5511 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5513 (long)(locinput - PL_reg_starttry),
5514 (long)(reginfo->till - PL_reg_starttry),
5517 sayNO_SILENT; /* Cannot match: too short. */
5519 PL_reginput = locinput; /* put where regtry can find it */
5520 sayYES; /* Success! */
5522 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5524 PerlIO_printf(Perl_debug_log,
5525 "%*s %ssubpattern success...%s\n",
5526 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5527 PL_reginput = locinput; /* put where regtry can find it */
5528 sayYES; /* Success! */
5531 #define ST st->u.ifmatch
5533 case SUSPEND: /* (?>A) */
5535 PL_reginput = locinput;
5538 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5540 goto ifmatch_trivial_fail_test;
5542 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5544 ifmatch_trivial_fail_test:
5546 char * const s = HOPBACKc(locinput, scan->flags);
5551 sw = 1 - cBOOL(ST.wanted);
5555 next = scan + ARG(scan);
5563 PL_reginput = locinput;
5567 ST.logical = logical;
5568 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5570 /* execute body of (?...A) */
5571 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5574 case IFMATCH_A_fail: /* body of (?...A) failed */
5575 ST.wanted = !ST.wanted;
5578 case IFMATCH_A: /* body of (?...A) succeeded */
5580 sw = cBOOL(ST.wanted);
5582 else if (!ST.wanted)
5585 if (OP(ST.me) == SUSPEND)
5586 locinput = PL_reginput;
5588 locinput = PL_reginput = st->locinput;
5589 nextchr = UCHARAT(locinput);
5591 scan = ST.me + ARG(ST.me);
5594 continue; /* execute B */
5599 next = scan + ARG(scan);
5604 reginfo->cutpoint = PL_regeol;
5607 PL_reginput = locinput;
5609 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5610 PUSH_STATE_GOTO(COMMIT_next,next);
5612 case COMMIT_next_fail:
5619 #define ST st->u.mark
5621 ST.prev_mark = mark_state;
5622 ST.mark_name = sv_commit = sv_yes_mark
5623 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5625 ST.mark_loc = PL_reginput = locinput;
5626 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5628 case MARKPOINT_next:
5629 mark_state = ST.prev_mark;
5632 case MARKPOINT_next_fail:
5633 if (popmark && sv_eq(ST.mark_name,popmark))
5635 if (ST.mark_loc > startpoint)
5636 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5637 popmark = NULL; /* we found our mark */
5638 sv_commit = ST.mark_name;
5641 PerlIO_printf(Perl_debug_log,
5642 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5643 REPORT_CODE_OFF+depth*2, "",
5644 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5647 mark_state = ST.prev_mark;
5648 sv_yes_mark = mark_state ?
5649 mark_state->u.mark.mark_name : NULL;
5653 PL_reginput = locinput;
5655 /* (*SKIP) : if we fail we cut here*/
5656 ST.mark_name = NULL;
5657 ST.mark_loc = locinput;
5658 PUSH_STATE_GOTO(SKIP_next,next);
5660 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5661 otherwise do nothing. Meaning we need to scan
5663 regmatch_state *cur = mark_state;
5664 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5667 if ( sv_eq( cur->u.mark.mark_name,
5670 ST.mark_name = find;
5671 PUSH_STATE_GOTO( SKIP_next, next );
5673 cur = cur->u.mark.prev_mark;
5676 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5678 case SKIP_next_fail:
5680 /* (*CUT:NAME) - Set up to search for the name as we
5681 collapse the stack*/
5682 popmark = ST.mark_name;
5684 /* (*CUT) - No name, we cut here.*/
5685 if (ST.mark_loc > startpoint)
5686 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5687 /* but we set sv_commit to latest mark_name if there
5688 is one so they can test to see how things lead to this
5691 sv_commit=mark_state->u.mark.mark_name;
5699 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5701 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5704 U8 folded[UTF8_MAXBYTES_CASE+1];
5706 const char * const l = locinput;
5707 char *e = PL_regeol;
5708 to_uni_fold(n, folded, &foldlen);
5710 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5711 l, &e, 0, utf8_target)) {
5716 nextchr = UCHARAT(locinput);
5719 if ((n=is_LNBREAK(locinput,utf8_target))) {
5721 nextchr = UCHARAT(locinput);
5726 #define CASE_CLASS(nAmE) \
5728 if (locinput >= PL_regeol) \
5730 if ((n=is_##nAmE(locinput,utf8_target))) { \
5732 nextchr = UCHARAT(locinput); \
5737 if (locinput >= PL_regeol) \
5739 if ((n=is_##nAmE(locinput,utf8_target))) { \
5742 locinput += UTF8SKIP(locinput); \
5743 nextchr = UCHARAT(locinput); \
5748 CASE_CLASS(HORIZWS);
5752 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5753 PTR2UV(scan), OP(scan));
5754 Perl_croak(aTHX_ "regexp memory corruption");
5758 /* switch break jumps here */
5759 scan = next; /* prepare to execute the next op and ... */
5760 continue; /* ... jump back to the top, reusing st */
5764 /* push a state that backtracks on success */
5765 st->u.yes.prev_yes_state = yes_state;
5769 /* push a new regex state, then continue at scan */
5771 regmatch_state *newst;
5774 regmatch_state *cur = st;
5775 regmatch_state *curyes = yes_state;
5777 regmatch_slab *slab = PL_regmatch_slab;
5778 for (;curd > -1;cur--,curd--) {
5779 if (cur < SLAB_FIRST(slab)) {
5781 cur = SLAB_LAST(slab);
5783 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5784 REPORT_CODE_OFF + 2 + depth * 2,"",
5785 curd, PL_reg_name[cur->resume_state],
5786 (curyes == cur) ? "yes" : ""
5789 curyes = cur->u.yes.prev_yes_state;
5792 DEBUG_STATE_pp("push")
5795 st->locinput = locinput;
5797 if (newst > SLAB_LAST(PL_regmatch_slab))
5798 newst = S_push_slab(aTHX);
5799 PL_regmatch_state = newst;
5801 locinput = PL_reginput;
5802 nextchr = UCHARAT(locinput);
5810 * We get here only if there's trouble -- normally "case END" is
5811 * the terminating point.
5813 Perl_croak(aTHX_ "corrupted regexp pointers");
5819 /* we have successfully completed a subexpression, but we must now
5820 * pop to the state marked by yes_state and continue from there */
5821 assert(st != yes_state);
5823 while (st != yes_state) {
5825 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5826 PL_regmatch_slab = PL_regmatch_slab->prev;
5827 st = SLAB_LAST(PL_regmatch_slab);
5831 DEBUG_STATE_pp("pop (no final)");
5833 DEBUG_STATE_pp("pop (yes)");
5839 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5840 || yes_state > SLAB_LAST(PL_regmatch_slab))
5842 /* not in this slab, pop slab */
5843 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5844 PL_regmatch_slab = PL_regmatch_slab->prev;
5845 st = SLAB_LAST(PL_regmatch_slab);
5847 depth -= (st - yes_state);
5850 yes_state = st->u.yes.prev_yes_state;
5851 PL_regmatch_state = st;
5854 locinput= st->locinput;
5855 nextchr = UCHARAT(locinput);
5857 state_num = st->resume_state + no_final;
5858 goto reenter_switch;
5861 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5862 PL_colors[4], PL_colors[5]));
5864 if (PL_reg_eval_set) {
5865 /* each successfully executed (?{...}) block does the equivalent of
5866 * local $^R = do {...}
5867 * When popping the save stack, all these locals would be undone;
5868 * bypass this by setting the outermost saved $^R to the latest
5870 if (oreplsv != GvSV(PL_replgv))
5871 sv_setsv(oreplsv, GvSV(PL_replgv));
5878 PerlIO_printf(Perl_debug_log,
5879 "%*s %sfailed...%s\n",
5880 REPORT_CODE_OFF+depth*2, "",
5881 PL_colors[4], PL_colors[5])
5893 /* there's a previous state to backtrack to */
5895 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5896 PL_regmatch_slab = PL_regmatch_slab->prev;
5897 st = SLAB_LAST(PL_regmatch_slab);
5899 PL_regmatch_state = st;
5900 locinput= st->locinput;
5901 nextchr = UCHARAT(locinput);
5903 DEBUG_STATE_pp("pop");
5905 if (yes_state == st)
5906 yes_state = st->u.yes.prev_yes_state;
5908 state_num = st->resume_state + 1; /* failure = success + 1 */
5909 goto reenter_switch;
5914 if (rex->intflags & PREGf_VERBARG_SEEN) {
5915 SV *sv_err = get_sv("REGERROR", 1);
5916 SV *sv_mrk = get_sv("REGMARK", 1);
5918 sv_commit = &PL_sv_no;
5920 sv_yes_mark = &PL_sv_yes;
5923 sv_commit = &PL_sv_yes;
5924 sv_yes_mark = &PL_sv_no;
5926 sv_setsv(sv_err, sv_commit);
5927 sv_setsv(sv_mrk, sv_yes_mark);
5930 /* clean up; in particular, free all slabs above current one */
5931 LEAVE_SCOPE(oldsave);
5937 - regrepeat - repeatedly match something simple, report how many
5940 * [This routine now assumes that it will only match on things of length 1.
5941 * That was true before, but now we assume scan - reginput is the count,
5942 * rather than incrementing count on every character. [Er, except utf8.]]
5945 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5948 register char *scan;
5950 register char *loceol = PL_regeol;
5951 register I32 hardcount = 0;
5952 register bool utf8_target = PL_reg_match_utf8;
5955 PERL_UNUSED_ARG(depth);
5958 PERL_ARGS_ASSERT_REGREPEAT;
5961 if (max == REG_INFTY)
5963 else if (max < loceol - scan)
5964 loceol = scan + max;
5969 while (scan < loceol && hardcount < max && *scan != '\n') {
5970 scan += UTF8SKIP(scan);
5974 while (scan < loceol && *scan != '\n')
5981 while (scan < loceol && hardcount < max) {
5982 scan += UTF8SKIP(scan);
5993 /* To get here, EXACTish nodes must have *byte* length == 1. That
5994 * means they match only characters in the string that can be expressed
5995 * as a single byte. For non-utf8 strings, that means a simple match.
5996 * For utf8 strings, the character matched must be an invariant, or
5997 * downgradable to a single byte. The pattern's utf8ness is
5998 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5999 * it is, it's an invariant */
6002 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6004 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6005 while (scan < loceol && UCHARAT(scan) == c) {
6011 /* Here, the string is utf8, and the pattern char is different
6012 * in utf8 than not, so can't compare them directly. Outside the
6013 * loop, find the two utf8 bytes that represent c, and then
6014 * look for those in sequence in the utf8 string */
6015 U8 high = UTF8_TWO_BYTE_HI(c);
6016 U8 low = UTF8_TWO_BYTE_LO(c);
6019 while (hardcount < max
6020 && scan + 1 < loceol
6021 && UCHARAT(scan) == high
6022 && UCHARAT(scan + 1) == low)
6030 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6034 PL_reg_flags |= RF_tainted;
6035 utf8_flags = FOLDEQ_UTF8_LOCALE;
6040 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6042 /* The comments for the EXACT case above apply as well to these fold
6047 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6049 if (utf8_target) { /* Use full Unicode fold matching */
6050 char *tmpeol = loceol;
6051 while (hardcount < max
6052 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6053 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6060 /* XXX Note that the above handles properly the German sharp s in
6061 * the pattern matching ss in the string. But it doesn't handle
6062 * properly cases where the string contains say 'LIGATURE ff' and
6063 * the pattern is 'f+'. This would require, say, a new function or
6064 * revised interface to foldEQ_utf8(), in which the maximum number
6065 * of characters to match could be passed and it would return how
6066 * many actually did. This is just one of many cases where
6067 * multi-char folds don't work properly, and so the fix is being
6073 /* Here, the string isn't utf8 and c is a single byte; and either
6074 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6075 * doesn't affect c. Can just do simple comparisons for exact or
6078 case EXACTF: folded = PL_fold[c]; break;
6080 case EXACTFU: folded = PL_fold_latin1[c]; break;
6081 case EXACTFL: folded = PL_fold_locale[c]; break;
6082 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6084 while (scan < loceol &&
6085 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6093 if (utf8_target || OP(p) == ANYOFV) {
6096 inclasslen = loceol - scan;
6097 while (hardcount < max
6098 && ((inclasslen = loceol - scan) > 0)
6099 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6105 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6113 LOAD_UTF8_CHARCLASS_ALNUM();
6114 while (hardcount < max && scan < loceol &&
6115 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6117 scan += UTF8SKIP(scan);
6121 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6129 while (scan < loceol && isALNUM((U8) *scan)) {
6134 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6139 PL_reg_flags |= RF_tainted;
6142 while (hardcount < max && scan < loceol &&
6143 isALNUM_LC_utf8((U8*)scan)) {
6144 scan += UTF8SKIP(scan);
6148 while (scan < loceol && isALNUM_LC(*scan))
6158 LOAD_UTF8_CHARCLASS_ALNUM();
6159 while (hardcount < max && scan < loceol &&
6160 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6162 scan += UTF8SKIP(scan);
6166 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6173 goto utf8_Nwordchar;
6174 while (scan < loceol && ! isALNUM((U8) *scan)) {
6180 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6181 scan += UTF8SKIP(scan);
6185 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6191 PL_reg_flags |= RF_tainted;
6194 while (hardcount < max && scan < loceol &&
6195 !isALNUM_LC_utf8((U8*)scan)) {
6196 scan += UTF8SKIP(scan);
6200 while (scan < loceol && !isALNUM_LC(*scan))
6210 LOAD_UTF8_CHARCLASS_SPACE();
6211 while (hardcount < max && scan < loceol &&
6213 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6215 scan += UTF8SKIP(scan);
6221 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6230 while (scan < loceol && isSPACE((U8) *scan)) {
6235 while (scan < loceol && isSPACE_A((U8) *scan)) {
6240 PL_reg_flags |= RF_tainted;
6243 while (hardcount < max && scan < loceol &&
6244 isSPACE_LC_utf8((U8*)scan)) {
6245 scan += UTF8SKIP(scan);
6249 while (scan < loceol && isSPACE_LC(*scan))
6259 LOAD_UTF8_CHARCLASS_SPACE();
6260 while (hardcount < max && scan < loceol &&
6262 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6264 scan += UTF8SKIP(scan);
6270 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6279 while (scan < loceol && ! isSPACE((U8) *scan)) {
6285 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6286 scan += UTF8SKIP(scan);
6290 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6296 PL_reg_flags |= RF_tainted;
6299 while (hardcount < max && scan < loceol &&
6300 !isSPACE_LC_utf8((U8*)scan)) {
6301 scan += UTF8SKIP(scan);
6305 while (scan < loceol && !isSPACE_LC(*scan))
6312 LOAD_UTF8_CHARCLASS_DIGIT();
6313 while (hardcount < max && scan < loceol &&
6314 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6315 scan += UTF8SKIP(scan);
6319 while (scan < loceol && isDIGIT(*scan))
6324 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6329 PL_reg_flags |= RF_tainted;
6332 while (hardcount < max && scan < loceol &&
6333 isDIGIT_LC_utf8((U8*)scan)) {
6334 scan += UTF8SKIP(scan);
6338 while (scan < loceol && isDIGIT_LC(*scan))
6345 LOAD_UTF8_CHARCLASS_DIGIT();
6346 while (hardcount < max && scan < loceol &&
6347 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6348 scan += UTF8SKIP(scan);
6352 while (scan < loceol && !isDIGIT(*scan))
6358 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6359 scan += UTF8SKIP(scan);
6363 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6369 PL_reg_flags |= RF_tainted;
6372 while (hardcount < max && scan < loceol &&
6373 !isDIGIT_LC_utf8((U8*)scan)) {
6374 scan += UTF8SKIP(scan);
6378 while (scan < loceol && !isDIGIT_LC(*scan))
6385 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6391 LNBREAK can match two latin chars, which is ok,
6392 because we have a null terminated string, but we
6393 have to use hardcount in this situation
6395 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6404 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6409 while (scan < loceol && is_HORIZWS_latin1(scan))
6416 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6417 scan += UTF8SKIP(scan);
6421 while (scan < loceol && !is_HORIZWS_latin1(scan))
6429 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6434 while (scan < loceol && is_VERTWS_latin1(scan))
6442 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6443 scan += UTF8SKIP(scan);
6447 while (scan < loceol && !is_VERTWS_latin1(scan))
6453 default: /* Called on something of 0 width. */
6454 break; /* So match right here or not at all. */
6460 c = scan - PL_reginput;
6464 GET_RE_DEBUG_FLAGS_DECL;
6466 SV * const prop = sv_newmortal();
6467 regprop(prog, prop, p);
6468 PerlIO_printf(Perl_debug_log,
6469 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6470 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6478 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6480 - regclass_swash - prepare the utf8 swash
6484 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6490 RXi_GET_DECL(prog,progi);
6491 const struct reg_data * const data = prog ? progi->data : NULL;
6493 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6495 assert(ANYOF_NONBITMAP(node));
6497 if (data && data->count) {
6498 const U32 n = ARG(node);
6500 if (data->what[n] == 's') {
6501 SV * const rv = MUTABLE_SV(data->data[n]);
6502 AV * const av = MUTABLE_AV(SvRV(rv));
6503 SV **const ary = AvARRAY(av);
6506 /* See the end of regcomp.c:S_regclass() for
6507 * documentation of these array elements. */
6510 a = SvROK(ary[1]) ? &ary[1] : NULL;
6511 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6515 else if (si && doinit) {
6516 sw = swash_init("utf8", "", si, 1, 0);
6517 (void)av_store(av, 1, sw);
6534 - reginclass - determine if a character falls into a character class
6536 n is the ANYOF regnode
6537 p is the target string
6538 lenp is pointer to the maximum number of bytes of how far to go in p
6539 (This is assumed wthout checking to always be at least the current
6541 utf8_target tells whether p is in UTF-8.
6543 Returns true if matched; false otherwise. If lenp is not NULL, on return
6544 from a successful match, the value it points to will be updated to how many
6545 bytes in p were matched. If there was no match, the value is undefined,
6546 possibly changed from the input.
6548 Note that this can be a synthetic start class, a combination of various
6549 nodes, so things you think might be mutually exclusive, such as locale,
6550 aren't. It can match both locale and non-locale
6555 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6558 const char flags = ANYOF_FLAGS(n);
6564 PERL_ARGS_ASSERT_REGINCLASS;
6566 /* If c is not already the code point, get it */
6567 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6568 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6569 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6570 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6571 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6572 * UTF8_ALLOW_FFFF */
6573 if (c_len == (STRLEN)-1)
6574 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6580 /* Use passed in max length, or one character if none passed in or less
6581 * than one character. And assume will match just one character. This is
6582 * overwritten later if matched more. */
6584 maxlen = (*lenp > c_len) ? *lenp : c_len;
6592 /* If this character is potentially in the bitmap, check it */
6594 if (ANYOF_BITMAP_TEST(n, c))
6596 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6603 else if (flags & ANYOF_LOCALE) {
6604 PL_reg_flags |= RF_tainted;
6606 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6607 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6611 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6612 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6613 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6614 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6615 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6616 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6617 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6618 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6619 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6620 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6621 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6622 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6623 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6624 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6625 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6626 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6627 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6628 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6629 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6630 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6631 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6632 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6633 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6634 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6635 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6636 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6637 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6638 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6639 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6640 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6641 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6642 ) /* How's that for a conditional? */
6649 /* If the bitmap didn't (or couldn't) match, and something outside the
6650 * bitmap could match, try that. Locale nodes specifiy completely the
6651 * behavior of code points in the bit map (otherwise, a utf8 target would
6652 * cause them to be treated as Unicode and not locale), except in
6653 * the very unlikely event when this node is a synthetic start class, which
6654 * could be a combination of locale and non-locale nodes. So allow locale
6655 * to match for the synthetic start class, which will give a false
6656 * positive that will be resolved when the match is done again as not part
6657 * of the synthetic start class */
6659 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6660 match = TRUE; /* Everything above 255 matches */
6662 else if (ANYOF_NONBITMAP(n)
6663 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6666 || (! (flags & ANYOF_LOCALE))
6667 || (flags & ANYOF_IS_SYNTHETIC)))))
6670 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6678 /* Not utf8. Convert as much of the string as available up
6679 * to the limit of how far the (single) character in the
6680 * pattern can possibly match (no need to go further). If
6681 * the node is a straight ANYOF or not folding, it can't
6682 * match more than one. Otherwise, It can match up to how
6683 * far a single char can fold to. Since not utf8, each
6684 * character is a single byte, so the max it can be in
6685 * bytes is the same as the max it can be in characters */
6686 STRLEN len = (OP(n) == ANYOF
6687 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6689 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6691 : UTF8_MAX_FOLD_CHAR_EXPAND;
6692 utf8_p = bytes_to_utf8(p, &len);
6695 if (swash_fetch(sw, utf8_p, TRUE))
6697 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6699 /* Here, we need to test if the fold of the target string
6700 * matches. The non-multi char folds have all been moved to
6701 * the compilation phase, and the multi-char folds have
6702 * been stored by regcomp into 'av'; we linearly check to
6703 * see if any match the target string (folded). We know
6704 * that the originals were each one character, but we don't
6705 * currently know how many characters/bytes each folded to,
6706 * except we do know that there are small limits imposed by
6707 * Unicode. XXX A performance enhancement would be to have
6708 * regcomp.c store the max number of chars/bytes that are
6709 * in an av entry, as, say the 0th element. Even better
6710 * would be to have a hash of the few characters that can
6711 * start a multi-char fold to the max number of chars of
6714 * If there is a match, we will need to advance (if lenp is
6715 * specified) the match pointer in the target string. But
6716 * what we are comparing here isn't that string directly,
6717 * but its fold, whose length may differ from the original.
6718 * As we go along in constructing the fold, therefore, we
6719 * create a map so that we know how many bytes in the
6720 * source to advance given that we have matched a certain
6721 * number of bytes in the fold. This map is stored in
6722 * 'map_fold_len_back'. Let n mean the number of bytes in
6723 * the fold of the first character that we are folding.
6724 * Then map_fold_len_back[n] is set to the number of bytes
6725 * in that first character. Similarly let m be the
6726 * corresponding number for the second character to be
6727 * folded. Then map_fold_len_back[n+m] is set to the
6728 * number of bytes occupied by the first two source
6729 * characters. ... */
6730 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6731 U8 folded[UTF8_MAXBYTES_CASE+1];
6732 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6733 STRLEN total_foldlen = 0; /* num bytes in fold of all
6736 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6738 /* Here, only need to fold the first char of the target
6739 * string. It the source wasn't utf8, is 1 byte long */
6740 to_utf8_fold(utf8_p, folded, &foldlen);
6741 total_foldlen = foldlen;
6742 map_fold_len_back[foldlen] = (utf8_target)
6748 /* Here, need to fold more than the first char. Do so
6749 * up to the limits */
6750 U8* source_ptr = utf8_p; /* The source for the fold
6753 U8* folded_ptr = folded;
6754 U8* e = utf8_p + maxlen; /* Can't go beyond last
6755 available byte in the
6759 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6763 /* Fold the next character */
6764 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6765 STRLEN this_char_foldlen;
6766 to_utf8_fold(source_ptr,
6768 &this_char_foldlen);
6770 /* Bail if it would exceed the byte limit for
6771 * folding a single char. */
6772 if (this_char_foldlen + folded_ptr - folded >
6778 /* Add the fold of this character */
6779 Copy(this_char_folded,
6783 source_ptr += UTF8SKIP(source_ptr);
6784 folded_ptr += this_char_foldlen;
6785 total_foldlen = folded_ptr - folded;
6787 /* Create map from the number of bytes in the fold
6788 * back to the number of bytes in the source. If
6789 * the source isn't utf8, the byte count is just
6790 * the number of characters so far */
6791 map_fold_len_back[total_foldlen]
6793 ? source_ptr - utf8_p
6800 /* Do the linear search to see if the fold is in the list
6801 * of multi-char folds. */
6804 for (i = 0; i <= av_len(av); i++) {
6805 SV* const sv = *av_fetch(av, i, FALSE);
6807 const char * const s = SvPV_const(sv, len);
6809 if (len <= total_foldlen
6810 && memEQ(s, (char*)folded, len)
6812 /* If 0, means matched a partial char. See
6814 && map_fold_len_back[len])
6817 /* Advance the target string ptr to account for
6818 * this fold, but have to translate from the
6819 * folded length to the corresponding source
6822 *lenp = map_fold_len_back[len];
6831 /* If we allocated a string above, free it */
6832 if (! utf8_target) Safefree(utf8_p);
6837 return (flags & ANYOF_INVERT) ? !match : match;
6841 S_reghop3(U8 *s, I32 off, const U8* lim)
6843 /* return the position 'off' UTF-8 characters away from 's', forward if
6844 * 'off' >= 0, backwards if negative. But don't go outside of position
6845 * 'lim', which better be < s if off < 0 */
6849 PERL_ARGS_ASSERT_REGHOP3;
6852 while (off-- && s < lim) {
6853 /* XXX could check well-formedness here */
6858 while (off++ && s > lim) {
6860 if (UTF8_IS_CONTINUED(*s)) {
6861 while (s > lim && UTF8_IS_CONTINUATION(*s))
6864 /* XXX could check well-formedness here */
6871 /* there are a bunch of places where we use two reghop3's that should
6872 be replaced with this routine. but since thats not done yet
6873 we ifdef it out - dmq
6876 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6880 PERL_ARGS_ASSERT_REGHOP4;
6883 while (off-- && s < rlim) {
6884 /* XXX could check well-formedness here */
6889 while (off++ && s > llim) {
6891 if (UTF8_IS_CONTINUED(*s)) {
6892 while (s > llim && UTF8_IS_CONTINUATION(*s))
6895 /* XXX could check well-formedness here */
6903 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6907 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6910 while (off-- && s < lim) {
6911 /* XXX could check well-formedness here */
6918 while (off++ && s > lim) {
6920 if (UTF8_IS_CONTINUED(*s)) {
6921 while (s > lim && UTF8_IS_CONTINUATION(*s))
6924 /* XXX could check well-formedness here */
6933 restore_pos(pTHX_ void *arg)
6936 regexp * const rex = (regexp *)arg;
6937 if (PL_reg_eval_set) {
6938 if (PL_reg_oldsaved) {
6939 rex->subbeg = PL_reg_oldsaved;
6940 rex->sublen = PL_reg_oldsavedlen;
6941 #ifdef PERL_OLD_COPY_ON_WRITE
6942 rex->saved_copy = PL_nrs;
6944 RXp_MATCH_COPIED_on(rex);
6946 PL_reg_magic->mg_len = PL_reg_oldpos;
6947 PL_reg_eval_set = 0;
6948 PL_curpm = PL_reg_oldcurpm;
6953 S_to_utf8_substr(pTHX_ register regexp *prog)
6957 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6960 if (prog->substrs->data[i].substr
6961 && !prog->substrs->data[i].utf8_substr) {
6962 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6963 prog->substrs->data[i].utf8_substr = sv;
6964 sv_utf8_upgrade(sv);
6965 if (SvVALID(prog->substrs->data[i].substr)) {
6966 if (SvTAIL(prog->substrs->data[i].substr)) {
6967 /* Trim the trailing \n that fbm_compile added last
6969 SvCUR_set(sv, SvCUR(sv) - 1);
6970 /* Whilst this makes the SV technically "invalid" (as its
6971 buffer is no longer followed by "\0") when fbm_compile()
6972 adds the "\n" back, a "\0" is restored. */
6973 fbm_compile(sv, FBMcf_TAIL);
6977 if (prog->substrs->data[i].substr == prog->check_substr)
6978 prog->check_utf8 = sv;
6984 S_to_byte_substr(pTHX_ register regexp *prog)
6989 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6992 if (prog->substrs->data[i].utf8_substr
6993 && !prog->substrs->data[i].substr) {
6994 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6995 if (sv_utf8_downgrade(sv, TRUE)) {
6996 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6997 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
6998 /* Trim the trailing \n that fbm_compile added last
7000 SvCUR_set(sv, SvCUR(sv) - 1);
7001 fbm_compile(sv, FBMcf_TAIL);
7009 prog->substrs->data[i].substr = sv;
7010 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7011 prog->check_substr = sv;
7018 * c-indentation-style: bsd
7020 * indent-tabs-mode: t
7023 * ex: set ts=8 sts=4 sw=4 noet: