5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? e.g. locale */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
126 /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
127 * skip the check on EBCDIC platforms */
128 # define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
130 # define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
131 if (!CAT2(PL_utf8_,class)) { \
133 ENTER; save_re_context(); \
134 ok=CAT2(is_utf8_,class)((const U8*)str); \
135 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
138 /* Doesn't do an assert to verify that is correct */
139 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
140 if (!CAT2(PL_utf8_,class)) { \
141 bool throw_away __attribute__unused__; \
142 ENTER; save_re_context(); \
143 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
146 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
147 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
148 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
150 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
151 LOAD_UTF8_CHARCLASS(X_begin, " "); \
152 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
153 /* These are utf8 constants, and not utf-ebcdic constants, so the \
154 * assert should likely and hopefully fail on an EBCDIC machine */ \
155 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
157 /* No asserts are done for these, in case called on an early \
158 * Unicode version in which they map to nothing */ \
159 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
160 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
161 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
162 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
163 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
164 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
165 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
167 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
169 /* The actual code for CCC_TRY, which uses several variables from the routine
170 * it's callable from. It is designed to be the bulk of a case statement.
171 * FUNC is the macro or function to call on non-utf8 targets that indicate if
172 * nextchr matches the class.
173 * UTF8_TEST is the whole test string to use for utf8 targets
174 * LOAD is what to use to test, and if not present to load in the swash for the
176 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
178 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
179 * utf8 and a variant, load the swash if necessary and test using the utf8
180 * test. Advance to the next character if test is ok, otherwise fail; If not
181 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
182 * fails, or advance to the next character */
184 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
185 if (locinput >= PL_regeol) { \
188 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
189 LOAD_UTF8_CHARCLASS(CLASS, STR); \
190 if (POS_OR_NEG (UTF8_TEST)) { \
193 locinput += PL_utf8skip[nextchr]; \
194 nextchr = UCHARAT(locinput); \
197 if (POS_OR_NEG (FUNC(nextchr))) { \
200 nextchr = UCHARAT(++locinput); \
203 /* Handle the non-locale cases for a character class and its complement. It
204 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
205 * This is because that code fails when the test succeeds, so we want to have
206 * the test fail so that the code succeeds. The swash is stored in a
207 * predictable PL_ place */
208 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
211 _CCC_TRY_CODE( !, FUNC, \
212 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
213 (U8*)locinput, TRUE)), \
216 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
217 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
218 (U8*)locinput, TRUE)), \
221 /* Generate the case statements for both locale and non-locale character
222 * classes in regmatch for classes that don't have special unicode semantics.
223 * Locales don't use an immediate swash, but an intermediary special locale
224 * function that is called on the pointer to the current place in the input
225 * string. That function will resolve to needing the same swash. One might
226 * think that because we don't know what the locale will match, we shouldn't
227 * check with the swash loading function that it loaded properly; ie, that we
228 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
229 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
231 #define CCC_TRY(NAME, NNAME, FUNC, \
232 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
233 NAMEA, NNAMEA, FUNCA, \
236 PL_reg_flags |= RF_tainted; \
237 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
239 PL_reg_flags |= RF_tainted; \
240 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
243 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
246 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
247 nextchr = UCHARAT(++locinput); \
250 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
254 locinput += PL_utf8skip[nextchr]; \
255 nextchr = UCHARAT(locinput); \
258 nextchr = UCHARAT(++locinput); \
261 /* Generate the non-locale cases */ \
262 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
264 /* This is like CCC_TRY, but has an extra set of parameters for generating case
265 * statements to handle separate Unicode semantics nodes */
266 #define CCC_TRY_U(NAME, NNAME, FUNC, \
267 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
268 NAMEU, NNAMEU, FUNCU, \
269 NAMEA, NNAMEA, FUNCA, \
271 CCC_TRY(NAME, NNAME, FUNC, \
272 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
273 NAMEA, NNAMEA, FUNCA, \
275 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
277 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
279 /* for use after a quantifier and before an EXACT-like node -- japhy */
280 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
282 * NOTE that *nothing* that affects backtracking should be in here, specifically
283 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
284 * node that is in between two EXACT like nodes when ascertaining what the required
285 * "follow" character is. This should probably be moved to regex compile time
286 * although it may be done at run time beause of the REF possibility - more
287 * investigation required. -- demerphq
289 #define JUMPABLE(rn) ( \
291 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
293 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
294 OP(rn) == PLUS || OP(rn) == MINMOD || \
296 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
298 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
300 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
303 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
304 we don't need this definition. */
305 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
306 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
307 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
310 /* ... so we use this as its faster. */
311 #define IS_TEXT(rn) ( OP(rn)==EXACT )
312 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
313 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
314 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
319 Search for mandatory following text node; for lookahead, the text must
320 follow but for lookbehind (rn->flags != 0) we skip to the next step.
322 #define FIND_NEXT_IMPT(rn) STMT_START { \
323 while (JUMPABLE(rn)) { \
324 const OPCODE type = OP(rn); \
325 if (type == SUSPEND || PL_regkind[type] == CURLY) \
326 rn = NEXTOPER(NEXTOPER(rn)); \
327 else if (type == PLUS) \
329 else if (type == IFMATCH) \
330 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
331 else rn += NEXT_OFF(rn); \
336 static void restore_pos(pTHX_ void *arg);
338 #define REGCP_PAREN_ELEMS 4
339 #define REGCP_OTHER_ELEMS 5
340 #define REGCP_FRAME_ELEMS 1
341 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
342 * are needed for the regexp context stack bookkeeping. */
345 S_regcppush(pTHX_ I32 parenfloor)
348 const int retval = PL_savestack_ix;
349 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
350 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
351 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
353 GET_RE_DEBUG_FLAGS_DECL;
355 if (paren_elems_to_push < 0)
356 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
358 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
359 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
360 " out of range (%lu-%ld)",
361 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
363 SSGROW(total_elems + REGCP_FRAME_ELEMS);
365 for (p = PL_regsize; p > parenfloor; p--) {
366 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
367 SSPUSHINT(PL_regoffs[p].end);
368 SSPUSHINT(PL_regoffs[p].start);
369 SSPUSHPTR(PL_reg_start_tmp[p]);
371 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
372 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
373 (UV)p, (IV)PL_regoffs[p].start,
374 (IV)(PL_reg_start_tmp[p] - PL_bostr),
375 (IV)PL_regoffs[p].end
378 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
379 SSPUSHPTR(PL_regoffs);
380 SSPUSHINT(PL_regsize);
381 SSPUSHINT(*PL_reglastparen);
382 SSPUSHINT(*PL_reglastcloseparen);
383 SSPUSHPTR(PL_reginput);
384 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
389 /* These are needed since we do not localize EVAL nodes: */
390 #define REGCP_SET(cp) \
392 PerlIO_printf(Perl_debug_log, \
393 " Setting an EVAL scope, savestack=%"IVdf"\n", \
394 (IV)PL_savestack_ix)); \
397 #define REGCP_UNWIND(cp) \
399 if (cp != PL_savestack_ix) \
400 PerlIO_printf(Perl_debug_log, \
401 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
402 (IV)(cp), (IV)PL_savestack_ix)); \
406 S_regcppop(pTHX_ const regexp *rex)
411 GET_RE_DEBUG_FLAGS_DECL;
413 PERL_ARGS_ASSERT_REGCPPOP;
415 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
417 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
418 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
419 input = (char *) SSPOPPTR;
420 *PL_reglastcloseparen = SSPOPINT;
421 *PL_reglastparen = SSPOPINT;
422 PL_regsize = SSPOPINT;
423 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
425 i -= REGCP_OTHER_ELEMS;
426 /* Now restore the parentheses context. */
427 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
429 U32 paren = (U32)SSPOPINT;
430 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
431 PL_regoffs[paren].start = SSPOPINT;
433 if (paren <= *PL_reglastparen)
434 PL_regoffs[paren].end = tmps;
436 PerlIO_printf(Perl_debug_log,
437 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
438 (UV)paren, (IV)PL_regoffs[paren].start,
439 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
440 (IV)PL_regoffs[paren].end,
441 (paren > *PL_reglastparen ? "(no)" : ""));
445 if (*PL_reglastparen + 1 <= rex->nparens) {
446 PerlIO_printf(Perl_debug_log,
447 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
448 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
452 /* It would seem that the similar code in regtry()
453 * already takes care of this, and in fact it is in
454 * a better location to since this code can #if 0-ed out
455 * but the code in regtry() is needed or otherwise tests
456 * requiring null fields (pat.t#187 and split.t#{13,14}
457 * (as of patchlevel 7877) will fail. Then again,
458 * this code seems to be necessary or otherwise
459 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
460 * --jhi updated by dapm */
461 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
463 PL_regoffs[i].start = -1;
464 PL_regoffs[i].end = -1;
470 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
473 * pregexec and friends
476 #ifndef PERL_IN_XSUB_RE
478 - pregexec - match a regexp against a string
481 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
482 char *strbeg, I32 minend, SV *screamer, U32 nosave)
483 /* strend: pointer to null at end of string */
484 /* strbeg: real beginning of string */
485 /* minend: end of match must be >=minend after stringarg. */
486 /* nosave: For optimizations. */
488 PERL_ARGS_ASSERT_PREGEXEC;
491 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
492 nosave ? 0 : REXEC_COPY_STR);
497 * Need to implement the following flags for reg_anch:
499 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
501 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
502 * INTUIT_AUTORITATIVE_ML
503 * INTUIT_ONCE_NOML - Intuit can match in one location only.
506 * Another flag for this function: SECOND_TIME (so that float substrs
507 * with giant delta may be not rechecked).
510 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
512 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
513 Otherwise, only SvCUR(sv) is used to get strbeg. */
515 /* XXXX We assume that strpos is strbeg unless sv. */
517 /* XXXX Some places assume that there is a fixed substring.
518 An update may be needed if optimizer marks as "INTUITable"
519 RExen without fixed substrings. Similarly, it is assumed that
520 lengths of all the strings are no more than minlen, thus they
521 cannot come from lookahead.
522 (Or minlen should take into account lookahead.)
523 NOTE: Some of this comment is not correct. minlen does now take account
524 of lookahead/behind. Further research is required. -- demerphq
528 /* A failure to find a constant substring means that there is no need to make
529 an expensive call to REx engine, thus we celebrate a failure. Similarly,
530 finding a substring too deep into the string means that less calls to
531 regtry() should be needed.
533 REx compiler's optimizer found 4 possible hints:
534 a) Anchored substring;
536 c) Whether we are anchored (beginning-of-line or \G);
537 d) First node (of those at offset 0) which may distinguish positions;
538 We use a)b)d) and multiline-part of c), and try to find a position in the
539 string which does not contradict any of them.
542 /* Most of decisions we do here should have been done at compile time.
543 The nodes of the REx which we used for the search should have been
544 deleted from the finite automaton. */
547 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
548 char *strend, const U32 flags, re_scream_pos_data *data)
551 struct regexp *const prog = (struct regexp *)SvANY(rx);
552 register I32 start_shift = 0;
553 /* Should be nonnegative! */
554 register I32 end_shift = 0;
559 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
561 register char *other_last = NULL; /* other substr checked before this */
562 char *check_at = NULL; /* check substr found at this pos */
563 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
564 RXi_GET_DECL(prog,progi);
566 const char * const i_strpos = strpos;
568 GET_RE_DEBUG_FLAGS_DECL;
570 PERL_ARGS_ASSERT_RE_INTUIT_START;
572 RX_MATCH_UTF8_set(rx,utf8_target);
575 PL_reg_flags |= RF_utf8;
578 debug_start_match(rx, utf8_target, strpos, strend,
579 sv ? "Guessing start of match in sv for"
580 : "Guessing start of match in string for");
583 /* CHR_DIST() would be more correct here but it makes things slow. */
584 if (prog->minlen > strend - strpos) {
585 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
586 "String too short... [re_intuit_start]\n"));
590 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
593 if (!prog->check_utf8 && prog->check_substr)
594 to_utf8_substr(prog);
595 check = prog->check_utf8;
597 if (!prog->check_substr && prog->check_utf8)
598 to_byte_substr(prog);
599 check = prog->check_substr;
601 if (check == &PL_sv_undef) {
602 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
603 "Non-utf8 string cannot match utf8 check string\n"));
606 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
607 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
608 || ( (prog->extflags & RXf_ANCH_BOL)
609 && !multiline ) ); /* Check after \n? */
612 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
613 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
614 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
616 && (strpos != strbeg)) {
617 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
620 if (prog->check_offset_min == prog->check_offset_max &&
621 !(prog->extflags & RXf_CANY_SEEN)) {
622 /* Substring at constant offset from beg-of-str... */
625 s = HOP3c(strpos, prog->check_offset_min, strend);
628 slen = SvCUR(check); /* >= 1 */
630 if ( strend - s > slen || strend - s < slen - 1
631 || (strend - s == slen && strend[-1] != '\n')) {
632 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
635 /* Now should match s[0..slen-2] */
637 if (slen && (*SvPVX_const(check) != *s
639 && memNE(SvPVX_const(check), s, slen)))) {
641 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
645 else if (*SvPVX_const(check) != *s
646 || ((slen = SvCUR(check)) > 1
647 && memNE(SvPVX_const(check), s, slen)))
650 goto success_at_start;
653 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
655 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
656 end_shift = prog->check_end_shift;
659 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
660 - (SvTAIL(check) != 0);
661 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
663 if (end_shift < eshift)
667 else { /* Can match at random position */
670 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
671 end_shift = prog->check_end_shift;
673 /* end shift should be non negative here */
676 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
678 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
679 (IV)end_shift, RX_PRECOMP(prog));
683 /* Find a possible match in the region s..strend by looking for
684 the "check" substring in the region corrected by start/end_shift. */
687 I32 srch_start_shift = start_shift;
688 I32 srch_end_shift = end_shift;
689 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
690 srch_end_shift -= ((strbeg - s) - srch_start_shift);
691 srch_start_shift = strbeg - s;
693 DEBUG_OPTIMISE_MORE_r({
694 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
695 (IV)prog->check_offset_min,
696 (IV)srch_start_shift,
698 (IV)prog->check_end_shift);
701 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
702 I32 p = -1; /* Internal iterator of scream. */
703 I32 * const pp = data ? data->scream_pos : &p;
707 assert(SvMAGICAL(sv));
708 mg = mg_find(sv, PERL_MAGIC_study);
711 if (mg->mg_private == 1) {
712 found = ((U8 *)mg->mg_ptr)[BmRARE(check)] != (U8)~0;
713 } else if (mg->mg_private == 2) {
714 found = ((U16 *)mg->mg_ptr)[BmRARE(check)] != (U16)~0;
716 assert (mg->mg_private == 4);
717 found = ((U32 *)mg->mg_ptr)[BmRARE(check)] != (U32)~0;
721 || ( BmRARE(check) == '\n'
722 && (BmPREVIOUS(check) == SvCUR(check) - 1)
724 s = screaminstr(sv, check,
725 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
728 /* we may be pointing at the wrong string */
729 if (s && RXp_MATCH_COPIED(prog))
730 s = strbeg + (s - SvPVX_const(sv));
732 *data->scream_olds = s;
737 if (prog->extflags & RXf_CANY_SEEN) {
738 start_point= (U8*)(s + srch_start_shift);
739 end_point= (U8*)(strend - srch_end_shift);
741 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
742 end_point= HOP3(strend, -srch_end_shift, strbeg);
744 DEBUG_OPTIMISE_MORE_r({
745 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
746 (int)(end_point - start_point),
747 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
751 s = fbm_instr( start_point, end_point,
752 check, multiline ? FBMrf_MULTILINE : 0);
755 /* Update the count-of-usability, remove useless subpatterns,
759 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
760 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
761 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
762 (s ? "Found" : "Did not find"),
763 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
764 ? "anchored" : "floating"),
767 (s ? " at offset " : "...\n") );
772 /* Finish the diagnostic message */
773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
775 /* XXX dmq: first branch is for positive lookbehind...
776 Our check string is offset from the beginning of the pattern.
777 So we need to do any stclass tests offset forward from that
786 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
787 Start with the other substr.
788 XXXX no SCREAM optimization yet - and a very coarse implementation
789 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
790 *always* match. Probably should be marked during compile...
791 Probably it is right to do no SCREAM here...
794 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
795 : (prog->float_substr && prog->anchored_substr))
797 /* Take into account the "other" substring. */
798 /* XXXX May be hopelessly wrong for UTF... */
801 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
804 char * const last = HOP3c(s, -start_shift, strbeg);
806 char * const saved_s = s;
809 t = s - prog->check_offset_max;
810 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
812 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
817 t = HOP3c(t, prog->anchored_offset, strend);
818 if (t < other_last) /* These positions already checked */
820 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
823 /* XXXX It is not documented what units *_offsets are in.
824 We assume bytes, but this is clearly wrong.
825 Meaning this code needs to be carefully reviewed for errors.
829 /* On end-of-str: see comment below. */
830 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
831 if (must == &PL_sv_undef) {
833 DEBUG_r(must = prog->anchored_utf8); /* for debug */
838 HOP3(HOP3(last1, prog->anchored_offset, strend)
839 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
841 multiline ? FBMrf_MULTILINE : 0
844 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
845 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
846 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
847 (s ? "Found" : "Contradicts"),
848 quoted, RE_SV_TAIL(must));
853 if (last1 >= last2) {
854 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
855 ", giving up...\n"));
858 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
859 ", trying floating at offset %ld...\n",
860 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
861 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
862 s = HOP3c(last, 1, strend);
866 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
867 (long)(s - i_strpos)));
868 t = HOP3c(s, -prog->anchored_offset, strbeg);
869 other_last = HOP3c(s, 1, strend);
877 else { /* Take into account the floating substring. */
879 char * const saved_s = s;
882 t = HOP3c(s, -start_shift, strbeg);
884 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
885 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
886 last = HOP3c(t, prog->float_max_offset, strend);
887 s = HOP3c(t, prog->float_min_offset, strend);
890 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
891 must = utf8_target ? prog->float_utf8 : prog->float_substr;
892 /* fbm_instr() takes into account exact value of end-of-str
893 if the check is SvTAIL(ed). Since false positives are OK,
894 and end-of-str is not later than strend we are OK. */
895 if (must == &PL_sv_undef) {
897 DEBUG_r(must = prog->float_utf8); /* for debug message */
900 s = fbm_instr((unsigned char*)s,
901 (unsigned char*)last + SvCUR(must)
903 must, multiline ? FBMrf_MULTILINE : 0);
905 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
906 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
907 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
908 (s ? "Found" : "Contradicts"),
909 quoted, RE_SV_TAIL(must));
913 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
914 ", giving up...\n"));
917 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
918 ", trying anchored starting at offset %ld...\n",
919 (long)(saved_s + 1 - i_strpos)));
921 s = HOP3c(t, 1, strend);
925 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
926 (long)(s - i_strpos)));
927 other_last = s; /* Fix this later. --Hugo */
937 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
939 DEBUG_OPTIMISE_MORE_r(
940 PerlIO_printf(Perl_debug_log,
941 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
942 (IV)prog->check_offset_min,
943 (IV)prog->check_offset_max,
951 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
953 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
956 /* Fixed substring is found far enough so that the match
957 cannot start at strpos. */
959 if (ml_anch && t[-1] != '\n') {
960 /* Eventually fbm_*() should handle this, but often
961 anchored_offset is not 0, so this check will not be wasted. */
962 /* XXXX In the code below we prefer to look for "^" even in
963 presence of anchored substrings. And we search even
964 beyond the found float position. These pessimizations
965 are historical artefacts only. */
967 while (t < strend - prog->minlen) {
969 if (t < check_at - prog->check_offset_min) {
970 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
971 /* Since we moved from the found position,
972 we definitely contradict the found anchored
973 substr. Due to the above check we do not
974 contradict "check" substr.
975 Thus we can arrive here only if check substr
976 is float. Redo checking for "other"=="fixed".
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
980 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
981 goto do_other_anchored;
983 /* We don't contradict the found floating substring. */
984 /* XXXX Why not check for STCLASS? */
986 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
987 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
990 /* Position contradicts check-string */
991 /* XXXX probably better to look for check-string
992 than for "\n", so one should lower the limit for t? */
993 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
994 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
995 other_last = strpos = s = t + 1;
1000 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1001 PL_colors[0], PL_colors[1]));
1005 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1006 PL_colors[0], PL_colors[1]));
1010 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1013 /* The found string does not prohibit matching at strpos,
1014 - no optimization of calling REx engine can be performed,
1015 unless it was an MBOL and we are not after MBOL,
1016 or a future STCLASS check will fail this. */
1018 /* Even in this situation we may use MBOL flag if strpos is offset
1019 wrt the start of the string. */
1020 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1021 && (strpos != strbeg) && strpos[-1] != '\n'
1022 /* May be due to an implicit anchor of m{.*foo} */
1023 && !(prog->intflags & PREGf_IMPLICIT))
1028 DEBUG_EXECUTE_r( if (ml_anch)
1029 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1030 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1033 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1035 prog->check_utf8 /* Could be deleted already */
1036 && --BmUSEFUL(prog->check_utf8) < 0
1037 && (prog->check_utf8 == prog->float_utf8)
1039 prog->check_substr /* Could be deleted already */
1040 && --BmUSEFUL(prog->check_substr) < 0
1041 && (prog->check_substr == prog->float_substr)
1044 /* If flags & SOMETHING - do not do it many times on the same match */
1045 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1046 /* XXX Does the destruction order has to change with utf8_target? */
1047 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1048 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1049 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1050 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1051 check = NULL; /* abort */
1053 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1054 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1055 if (prog->intflags & PREGf_IMPLICIT)
1056 prog->extflags &= ~RXf_ANCH_MBOL;
1057 /* XXXX This is a remnant of the old implementation. It
1058 looks wasteful, since now INTUIT can use many
1059 other heuristics. */
1060 prog->extflags &= ~RXf_USE_INTUIT;
1061 /* XXXX What other flags might need to be cleared in this branch? */
1067 /* Last resort... */
1068 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1069 /* trie stclasses are too expensive to use here, we are better off to
1070 leave it to regmatch itself */
1071 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1072 /* minlen == 0 is possible if regstclass is \b or \B,
1073 and the fixed substr is ''$.
1074 Since minlen is already taken into account, s+1 is before strend;
1075 accidentally, minlen >= 1 guaranties no false positives at s + 1
1076 even for \b or \B. But (minlen? 1 : 0) below assumes that
1077 regstclass does not come from lookahead... */
1078 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1079 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1080 const U8* const str = (U8*)STRING(progi->regstclass);
1081 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1082 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1085 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1086 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1087 else if (prog->float_substr || prog->float_utf8)
1088 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1092 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1093 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1096 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1099 const char *what = NULL;
1101 if (endpos == strend) {
1102 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1103 "Could not match STCLASS...\n") );
1106 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1107 "This position contradicts STCLASS...\n") );
1108 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1110 /* Contradict one of substrings */
1111 if (prog->anchored_substr || prog->anchored_utf8) {
1112 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1113 DEBUG_EXECUTE_r( what = "anchored" );
1115 s = HOP3c(t, 1, strend);
1116 if (s + start_shift + end_shift > strend) {
1117 /* XXXX Should be taken into account earlier? */
1118 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1119 "Could not match STCLASS...\n") );
1124 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1125 "Looking for %s substr starting at offset %ld...\n",
1126 what, (long)(s + start_shift - i_strpos)) );
1129 /* Have both, check_string is floating */
1130 if (t + start_shift >= check_at) /* Contradicts floating=check */
1131 goto retry_floating_check;
1132 /* Recheck anchored substring, but not floating... */
1136 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1137 "Looking for anchored substr starting at offset %ld...\n",
1138 (long)(other_last - i_strpos)) );
1139 goto do_other_anchored;
1141 /* Another way we could have checked stclass at the
1142 current position only: */
1147 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1148 "Looking for /%s^%s/m starting at offset %ld...\n",
1149 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1152 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1154 /* Check is floating substring. */
1155 retry_floating_check:
1156 t = check_at - start_shift;
1157 DEBUG_EXECUTE_r( what = "floating" );
1158 goto hop_and_restart;
1161 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1162 "By STCLASS: moving %ld --> %ld\n",
1163 (long)(t - i_strpos), (long)(s - i_strpos))
1167 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1168 "Does not contradict STCLASS...\n");
1173 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1174 PL_colors[4], (check ? "Guessed" : "Giving up"),
1175 PL_colors[5], (long)(s - i_strpos)) );
1178 fail_finish: /* Substring not found */
1179 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1180 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1182 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1183 PL_colors[4], PL_colors[5]));
1187 #define DECL_TRIE_TYPE(scan) \
1188 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1189 trie_type = (scan->flags != EXACT) \
1190 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1191 : (utf8_target ? trie_utf8 : trie_plain)
1193 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1194 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1195 switch (trie_type) { \
1196 case trie_utf8_fold: \
1197 if ( foldlen>0 ) { \
1198 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1203 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1204 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1205 foldlen -= UNISKIP( uvc ); \
1206 uscan = foldbuf + UNISKIP( uvc ); \
1209 case trie_latin_utf8_fold: \
1210 if ( foldlen>0 ) { \
1211 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1217 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1218 foldlen -= UNISKIP( uvc ); \
1219 uscan = foldbuf + UNISKIP( uvc ); \
1223 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1230 charid = trie->charmap[ uvc ]; \
1234 if (widecharmap) { \
1235 SV** const svpp = hv_fetch(widecharmap, \
1236 (char*)&uvc, sizeof(UV), 0); \
1238 charid = (U16)SvIV(*svpp); \
1243 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1247 && (ln == 1 || folder(s, pat_string, ln)) \
1248 && (!reginfo || regtry(reginfo, &s)) ) \
1254 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1256 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1262 #define REXEC_FBC_SCAN(CoDe) \
1264 while (s < strend) { \
1270 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1271 REXEC_FBC_UTF8_SCAN( \
1273 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1282 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1285 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1294 #define REXEC_FBC_TRYIT \
1295 if ((!reginfo || regtry(reginfo, &s))) \
1298 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1299 if (utf8_target) { \
1300 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1303 REXEC_FBC_CLASS_SCAN(CoNd); \
1306 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1307 if (utf8_target) { \
1309 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1312 REXEC_FBC_CLASS_SCAN(CoNd); \
1315 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1316 PL_reg_flags |= RF_tainted; \
1317 if (utf8_target) { \
1318 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1321 REXEC_FBC_CLASS_SCAN(CoNd); \
1324 #define DUMP_EXEC_POS(li,s,doutf8) \
1325 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1328 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1329 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1330 tmp = TEST_NON_UTF8(tmp); \
1331 REXEC_FBC_UTF8_SCAN( \
1332 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1341 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1342 if (s == PL_bostr) { \
1346 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1347 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1350 LOAD_UTF8_CHARCLASS_ALNUM(); \
1351 REXEC_FBC_UTF8_SCAN( \
1352 if (tmp == ! (TeSt2_UtF8)) { \
1361 /* The only difference between the BOUND and NBOUND cases is that
1362 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1363 * NBOUND. This is accomplished by passing it in either the if or else clause,
1364 * with the other one being empty */
1365 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1366 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1368 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1369 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1371 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1372 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1374 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1375 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1378 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1379 * be passed in completely with the variable name being tested, which isn't
1380 * such a clean interface, but this is easier to read than it was before. We
1381 * are looking for the boundary (or non-boundary between a word and non-word
1382 * character. The utf8 and non-utf8 cases have the same logic, but the details
1383 * must be different. Find the "wordness" of the character just prior to this
1384 * one, and compare it with the wordness of this one. If they differ, we have
1385 * a boundary. At the beginning of the string, pretend that the previous
1386 * character was a new-line */
1387 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1388 if (utf8_target) { \
1391 else { /* Not utf8 */ \
1392 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1393 tmp = TEST_NON_UTF8(tmp); \
1395 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1404 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1407 /* We know what class REx starts with. Try to find this position... */
1408 /* if reginfo is NULL, its a dryrun */
1409 /* annoyingly all the vars in this routine have different names from their counterparts
1410 in regmatch. /grrr */
1413 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1414 const char *strend, regmatch_info *reginfo)
1417 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1418 char *pat_string; /* The pattern's exactish string */
1419 char *pat_end; /* ptr to end char of pat_string */
1420 re_fold_t folder; /* Function for computing non-utf8 folds */
1421 const U8 *fold_array; /* array for folding ords < 256 */
1424 register STRLEN uskip;
1428 register I32 tmp = 1; /* Scratch variable? */
1429 register const bool utf8_target = PL_reg_match_utf8;
1430 UV utf8_fold_flags = 0;
1431 RXi_GET_DECL(prog,progi);
1433 PERL_ARGS_ASSERT_FIND_BYCLASS;
1435 /* We know what class it must start with. */
1439 if (utf8_target || OP(c) == ANYOFV) {
1440 STRLEN inclasslen = strend - s;
1441 REXEC_FBC_UTF8_CLASS_SCAN(
1442 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1445 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1450 if (tmp && (!reginfo || regtry(reginfo, &s)))
1458 if (UTF_PATTERN || utf8_target) {
1459 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1460 goto do_exactf_utf8;
1462 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1463 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1464 goto do_exactf_non_utf8; /* isn't dealt with by these */
1467 if (UTF_PATTERN || utf8_target) {
1469 /* regcomp.c already folded this if pattern is in UTF-8 */
1470 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1471 goto do_exactf_utf8;
1473 fold_array = PL_fold;
1475 goto do_exactf_non_utf8;
1478 if (UTF_PATTERN || utf8_target) {
1479 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1480 goto do_exactf_utf8;
1482 fold_array = PL_fold_locale;
1483 folder = foldEQ_locale;
1484 goto do_exactf_non_utf8;
1487 if (UTF_PATTERN || utf8_target) {
1488 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1489 goto do_exactf_utf8;
1492 /* Any 'ss' in the pattern should have been replaced by regcomp,
1493 * so we don't have to worry here about this single special case
1494 * in the Latin1 range */
1495 fold_array = PL_fold_latin1;
1496 folder = foldEQ_latin1;
1500 do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */
1502 /* The idea in the non-utf8 EXACTF* cases is to first find the
1503 * first character of the EXACTF* node and then, if necessary,
1504 * case-insensitively compare the full text of the node. c1 is the
1505 * first character. c2 is its fold. This logic will not work for
1506 * Unicode semantics and the german sharp ss, which hence should
1507 * not be compiled into a node that gets here. */
1508 pat_string = STRING(c);
1509 ln = STR_LEN(c); /* length to match in octets/bytes */
1511 /* We know that we have to match at least 'ln' bytes (which is the
1512 * same as characters, since not utf8). If we have to match 3
1513 * characters, and there are only 2 availabe, we know without
1514 * trying that it will fail; so don't start a match past the
1515 * required minimum number from the far end */
1516 e = HOP3c(strend, -((I32)ln), s);
1518 if (!reginfo && e < s) {
1519 e = s; /* Due to minlen logic of intuit() */
1523 c2 = fold_array[c1];
1524 if (c1 == c2) { /* If char and fold are the same */
1525 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1528 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1537 /* If one of the operands is in utf8, we can't use the simpler
1538 * folding above, due to the fact that many different characters
1539 * can have the same fold, or portion of a fold, or different-
1541 pat_string = STRING(c);
1542 ln = STR_LEN(c); /* length to match in octets/bytes */
1543 pat_end = pat_string + ln;
1544 lnc = (UTF_PATTERN) /* length to match in characters */
1545 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1548 /* We have 'lnc' characters to match in the pattern, but because of
1549 * multi-character folding, each character in the target can match
1550 * up to 3 characters (Unicode guarantees it will never exceed
1551 * this) if it is utf8-encoded; and up to 2 if not (based on the
1552 * fact that the Latin 1 folds are already determined, and the
1553 * only multi-char fold in that range is the sharp-s folding to
1554 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1555 * string character. Adjust lnc accordingly, always matching at
1557 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1558 lnc = (lnc < expansion) ? 1 : lnc / expansion;
1560 /* As in the non-UTF8 case, if we have to match 3 characters, and
1561 * only 2 are left, it's guaranteed to fail, so don't start a
1562 * match that would require us to go beyond the end of the string
1564 e = HOP3c(strend, -((I32)lnc), s);
1566 if (!reginfo && e < s) {
1567 e = s; /* Due to minlen logic of intuit() */
1570 /* XXX Note that we could recalculate e every so-often through the
1571 * loop to stop earlier, as the worst case expansion above will
1572 * rarely be met, and as we go along we would usually find that e
1573 * moves further to the left. Unclear if worth the expense */
1576 char *my_strend= (char *)strend;
1577 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1578 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1579 && (!reginfo || regtry(reginfo, &s)) )
1588 PL_reg_flags |= RF_tainted;
1589 FBC_BOUND(isALNUM_LC,
1590 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1591 isALNUM_LC_utf8((U8*)s));
1594 PL_reg_flags |= RF_tainted;
1595 FBC_NBOUND(isALNUM_LC,
1596 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1597 isALNUM_LC_utf8((U8*)s));
1600 FBC_BOUND(isWORDCHAR,
1602 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1605 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1607 isWORDCHAR_A((U8*)s));
1610 FBC_NBOUND(isWORDCHAR,
1612 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1615 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1617 isWORDCHAR_A((U8*)s));
1620 FBC_BOUND(isWORDCHAR_L1,
1622 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1625 FBC_NBOUND(isWORDCHAR_L1,
1627 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1630 REXEC_FBC_CSCAN_TAINT(
1631 isALNUM_LC_utf8((U8*)s),
1636 REXEC_FBC_CSCAN_PRELOAD(
1637 LOAD_UTF8_CHARCLASS_ALNUM(),
1638 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1639 isWORDCHAR_L1((U8) *s)
1643 REXEC_FBC_CSCAN_PRELOAD(
1644 LOAD_UTF8_CHARCLASS_ALNUM(),
1645 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1650 /* Don't need to worry about utf8, as it can match only a single
1651 * byte invariant character */
1652 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1655 REXEC_FBC_CSCAN_PRELOAD(
1656 LOAD_UTF8_CHARCLASS_ALNUM(),
1657 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1658 ! isWORDCHAR_L1((U8) *s)
1662 REXEC_FBC_CSCAN_PRELOAD(
1663 LOAD_UTF8_CHARCLASS_ALNUM(),
1664 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1675 REXEC_FBC_CSCAN_TAINT(
1676 !isALNUM_LC_utf8((U8*)s),
1681 REXEC_FBC_CSCAN_PRELOAD(
1682 LOAD_UTF8_CHARCLASS_SPACE(),
1683 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1688 REXEC_FBC_CSCAN_PRELOAD(
1689 LOAD_UTF8_CHARCLASS_SPACE(),
1690 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1695 /* Don't need to worry about utf8, as it can match only a single
1696 * byte invariant character */
1697 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1700 REXEC_FBC_CSCAN_TAINT(
1701 isSPACE_LC_utf8((U8*)s),
1706 REXEC_FBC_CSCAN_PRELOAD(
1707 LOAD_UTF8_CHARCLASS_SPACE(),
1708 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1709 ! isSPACE_L1((U8) *s)
1713 REXEC_FBC_CSCAN_PRELOAD(
1714 LOAD_UTF8_CHARCLASS_SPACE(),
1715 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1726 REXEC_FBC_CSCAN_TAINT(
1727 !isSPACE_LC_utf8((U8*)s),
1732 REXEC_FBC_CSCAN_PRELOAD(
1733 LOAD_UTF8_CHARCLASS_DIGIT(),
1734 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1739 /* Don't need to worry about utf8, as it can match only a single
1740 * byte invariant character */
1741 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1744 REXEC_FBC_CSCAN_TAINT(
1745 isDIGIT_LC_utf8((U8*)s),
1750 REXEC_FBC_CSCAN_PRELOAD(
1751 LOAD_UTF8_CHARCLASS_DIGIT(),
1752 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1763 REXEC_FBC_CSCAN_TAINT(
1764 !isDIGIT_LC_utf8((U8*)s),
1771 is_LNBREAK_latin1(s)
1783 !is_VERTWS_latin1(s)
1789 is_HORIZWS_latin1(s)
1794 !is_HORIZWS_utf8(s),
1795 !is_HORIZWS_latin1(s)
1802 /* what trie are we using right now */
1804 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1806 = (reg_trie_data*)progi->data->data[ aho->trie ];
1807 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1809 const char *last_start = strend - trie->minlen;
1811 const char *real_start = s;
1813 STRLEN maxlen = trie->maxlen;
1815 U8 **points; /* map of where we were in the input string
1816 when reading a given char. For ASCII this
1817 is unnecessary overhead as the relationship
1818 is always 1:1, but for Unicode, especially
1819 case folded Unicode this is not true. */
1820 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1824 GET_RE_DEBUG_FLAGS_DECL;
1826 /* We can't just allocate points here. We need to wrap it in
1827 * an SV so it gets freed properly if there is a croak while
1828 * running the match */
1831 sv_points=newSV(maxlen * sizeof(U8 *));
1832 SvCUR_set(sv_points,
1833 maxlen * sizeof(U8 *));
1834 SvPOK_on(sv_points);
1835 sv_2mortal(sv_points);
1836 points=(U8**)SvPV_nolen(sv_points );
1837 if ( trie_type != trie_utf8_fold
1838 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1841 bitmap=(U8*)trie->bitmap;
1843 bitmap=(U8*)ANYOF_BITMAP(c);
1845 /* this is the Aho-Corasick algorithm modified a touch
1846 to include special handling for long "unknown char"
1847 sequences. The basic idea being that we use AC as long
1848 as we are dealing with a possible matching char, when
1849 we encounter an unknown char (and we have not encountered
1850 an accepting state) we scan forward until we find a legal
1852 AC matching is basically that of trie matching, except
1853 that when we encounter a failing transition, we fall back
1854 to the current states "fail state", and try the current char
1855 again, a process we repeat until we reach the root state,
1856 state 1, or a legal transition. If we fail on the root state
1857 then we can either terminate if we have reached an accepting
1858 state previously, or restart the entire process from the beginning
1862 while (s <= last_start) {
1863 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1871 U8 *uscan = (U8*)NULL;
1872 U8 *leftmost = NULL;
1874 U32 accepted_word= 0;
1878 while ( state && uc <= (U8*)strend ) {
1880 U32 word = aho->states[ state ].wordnum;
1884 DEBUG_TRIE_EXECUTE_r(
1885 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1886 dump_exec_pos( (char *)uc, c, strend, real_start,
1887 (char *)uc, utf8_target );
1888 PerlIO_printf( Perl_debug_log,
1889 " Scanning for legal start char...\n");
1893 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1897 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1903 if (uc >(U8*)last_start) break;
1907 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1908 if (!leftmost || lpos < leftmost) {
1909 DEBUG_r(accepted_word=word);
1915 points[pointpos++ % maxlen]= uc;
1916 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1917 uscan, len, uvc, charid, foldlen,
1919 DEBUG_TRIE_EXECUTE_r({
1920 dump_exec_pos( (char *)uc, c, strend, real_start,
1922 PerlIO_printf(Perl_debug_log,
1923 " Charid:%3u CP:%4"UVxf" ",
1929 word = aho->states[ state ].wordnum;
1931 base = aho->states[ state ].trans.base;
1933 DEBUG_TRIE_EXECUTE_r({
1935 dump_exec_pos( (char *)uc, c, strend, real_start,
1937 PerlIO_printf( Perl_debug_log,
1938 "%sState: %4"UVxf", word=%"UVxf,
1939 failed ? " Fail transition to " : "",
1940 (UV)state, (UV)word);
1946 ( ((offset = base + charid
1947 - 1 - trie->uniquecharcount)) >= 0)
1948 && ((U32)offset < trie->lasttrans)
1949 && trie->trans[offset].check == state
1950 && (tmp=trie->trans[offset].next))
1952 DEBUG_TRIE_EXECUTE_r(
1953 PerlIO_printf( Perl_debug_log," - legal\n"));
1958 DEBUG_TRIE_EXECUTE_r(
1959 PerlIO_printf( Perl_debug_log," - fail\n"));
1961 state = aho->fail[state];
1965 /* we must be accepting here */
1966 DEBUG_TRIE_EXECUTE_r(
1967 PerlIO_printf( Perl_debug_log," - accepting\n"));
1976 if (!state) state = 1;
1979 if ( aho->states[ state ].wordnum ) {
1980 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1981 if (!leftmost || lpos < leftmost) {
1982 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1987 s = (char*)leftmost;
1988 DEBUG_TRIE_EXECUTE_r({
1990 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1991 (UV)accepted_word, (IV)(s - real_start)
1994 if (!reginfo || regtry(reginfo, &s)) {
2000 DEBUG_TRIE_EXECUTE_r({
2001 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2004 DEBUG_TRIE_EXECUTE_r(
2005 PerlIO_printf( Perl_debug_log,"No match.\n"));
2014 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2024 - regexec_flags - match a regexp against a string
2027 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2028 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2029 /* strend: pointer to null at end of string */
2030 /* strbeg: real beginning of string */
2031 /* minend: end of match must be >=minend after stringarg. */
2032 /* data: May be used for some additional optimizations.
2033 Currently its only used, with a U32 cast, for transmitting
2034 the ganch offset when doing a /g match. This will change */
2035 /* nosave: For optimizations. */
2038 struct regexp *const prog = (struct regexp *)SvANY(rx);
2039 /*register*/ char *s;
2040 register regnode *c;
2041 /*register*/ char *startpos = stringarg;
2042 I32 minlen; /* must match at least this many chars */
2043 I32 dontbother = 0; /* how many characters not to try at end */
2044 I32 end_shift = 0; /* Same for the end. */ /* CC */
2045 I32 scream_pos = -1; /* Internal iterator of scream. */
2046 char *scream_olds = NULL;
2047 const bool utf8_target = cBOOL(DO_UTF8(sv));
2049 RXi_GET_DECL(prog,progi);
2050 regmatch_info reginfo; /* create some info to pass to regtry etc */
2051 regexp_paren_pair *swap = NULL;
2052 GET_RE_DEBUG_FLAGS_DECL;
2054 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2055 PERL_UNUSED_ARG(data);
2057 /* Be paranoid... */
2058 if (prog == NULL || startpos == NULL) {
2059 Perl_croak(aTHX_ "NULL regexp parameter");
2063 multiline = prog->extflags & RXf_PMf_MULTILINE;
2064 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2066 RX_MATCH_UTF8_set(rx, utf8_target);
2068 debug_start_match(rx, utf8_target, startpos, strend,
2072 minlen = prog->minlen;
2074 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2075 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2076 "String too short [regexec_flags]...\n"));
2081 /* Check validity of program. */
2082 if (UCHARAT(progi->program) != REG_MAGIC) {
2083 Perl_croak(aTHX_ "corrupted regexp program");
2087 PL_reg_eval_set = 0;
2091 PL_reg_flags |= RF_utf8;
2093 /* Mark beginning of line for ^ and lookbehind. */
2094 reginfo.bol = startpos; /* XXX not used ??? */
2098 /* Mark end of line for $ (and such) */
2101 /* see how far we have to get to not match where we matched before */
2102 reginfo.till = startpos+minend;
2104 /* If there is a "must appear" string, look for it. */
2107 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2109 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2110 reginfo.ganch = startpos + prog->gofs;
2111 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2112 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2113 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2115 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2116 && mg->mg_len >= 0) {
2117 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2118 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2119 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2121 if (prog->extflags & RXf_ANCH_GPOS) {
2122 if (s > reginfo.ganch)
2124 s = reginfo.ganch - prog->gofs;
2125 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2126 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2132 reginfo.ganch = strbeg + PTR2UV(data);
2133 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2134 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2136 } else { /* pos() not defined */
2137 reginfo.ganch = strbeg;
2138 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2139 "GPOS: reginfo.ganch = strbeg\n"));
2142 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2143 /* We have to be careful. If the previous successful match
2144 was from this regex we don't want a subsequent partially
2145 successful match to clobber the old results.
2146 So when we detect this possibility we add a swap buffer
2147 to the re, and switch the buffer each match. If we fail
2148 we switch it back, otherwise we leave it swapped.
2151 /* do we need a save destructor here for eval dies? */
2152 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2154 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2155 re_scream_pos_data d;
2157 d.scream_olds = &scream_olds;
2158 d.scream_pos = &scream_pos;
2159 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2161 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2162 goto phooey; /* not present */
2168 /* Simplest case: anchored match need be tried only once. */
2169 /* [unless only anchor is BOL and multiline is set] */
2170 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2171 if (s == startpos && regtry(®info, &startpos))
2173 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2174 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2179 dontbother = minlen - 1;
2180 end = HOP3c(strend, -dontbother, strbeg) - 1;
2181 /* for multiline we only have to try after newlines */
2182 if (prog->check_substr || prog->check_utf8) {
2183 /* because of the goto we can not easily reuse the macros for bifurcating the
2184 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2187 goto after_try_utf8;
2189 if (regtry(®info, &s)) {
2196 if (prog->extflags & RXf_USE_INTUIT) {
2197 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2206 } /* end search for check string in unicode */
2208 if (s == startpos) {
2209 goto after_try_latin;
2212 if (regtry(®info, &s)) {
2219 if (prog->extflags & RXf_USE_INTUIT) {
2220 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2229 } /* end search for check string in latin*/
2230 } /* end search for check string */
2231 else { /* search for newline */
2233 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2236 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2238 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2239 if (regtry(®info, &s))
2243 } /* end search for newline */
2244 } /* end anchored/multiline check string search */
2246 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2248 /* the warning about reginfo.ganch being used without initialization
2249 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2250 and we only enter this block when the same bit is set. */
2251 char *tmp_s = reginfo.ganch - prog->gofs;
2253 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2258 /* Messy cases: unanchored match. */
2259 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2260 /* we have /x+whatever/ */
2261 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2266 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2267 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2268 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2273 DEBUG_EXECUTE_r( did_match = 1 );
2274 if (regtry(®info, &s)) goto got_it;
2276 while (s < strend && *s == ch)
2284 DEBUG_EXECUTE_r( did_match = 1 );
2285 if (regtry(®info, &s)) goto got_it;
2287 while (s < strend && *s == ch)
2292 DEBUG_EXECUTE_r(if (!did_match)
2293 PerlIO_printf(Perl_debug_log,
2294 "Did not find anchored character...\n")
2297 else if (prog->anchored_substr != NULL
2298 || prog->anchored_utf8 != NULL
2299 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2300 && prog->float_max_offset < strend - s)) {
2305 char *last1; /* Last position checked before */
2309 if (prog->anchored_substr || prog->anchored_utf8) {
2310 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2311 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2312 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2313 back_max = back_min = prog->anchored_offset;
2315 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2316 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2317 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2318 back_max = prog->float_max_offset;
2319 back_min = prog->float_min_offset;
2323 if (must == &PL_sv_undef)
2324 /* could not downgrade utf8 check substring, so must fail */
2330 last = HOP3c(strend, /* Cannot start after this */
2331 -(I32)(CHR_SVLEN(must)
2332 - (SvTAIL(must) != 0) + back_min), strbeg);
2335 last1 = HOPc(s, -1);
2337 last1 = s - 1; /* bogus */
2339 /* XXXX check_substr already used to find "s", can optimize if
2340 check_substr==must. */
2342 dontbother = end_shift;
2343 strend = HOPc(strend, -dontbother);
2344 while ( (s <= last) &&
2345 ((flags & REXEC_SCREAM) && SvSCREAM(sv)
2346 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2347 end_shift, &scream_pos, 0))
2348 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2349 (unsigned char*)strend, must,
2350 multiline ? FBMrf_MULTILINE : 0))) ) {
2351 /* we may be pointing at the wrong string */
2352 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2353 s = strbeg + (s - SvPVX_const(sv));
2354 DEBUG_EXECUTE_r( did_match = 1 );
2355 if (HOPc(s, -back_max) > last1) {
2356 last1 = HOPc(s, -back_min);
2357 s = HOPc(s, -back_max);
2360 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2362 last1 = HOPc(s, -back_min);
2366 while (s <= last1) {
2367 if (regtry(®info, &s))
2373 while (s <= last1) {
2374 if (regtry(®info, &s))
2380 DEBUG_EXECUTE_r(if (!did_match) {
2381 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2382 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2383 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2384 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2385 ? "anchored" : "floating"),
2386 quoted, RE_SV_TAIL(must));
2390 else if ( (c = progi->regstclass) ) {
2392 const OPCODE op = OP(progi->regstclass);
2393 /* don't bother with what can't match */
2394 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2395 strend = HOPc(strend, -(minlen - 1));
2398 SV * const prop = sv_newmortal();
2399 regprop(prog, prop, c);
2401 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2403 PerlIO_printf(Perl_debug_log,
2404 "Matching stclass %.*s against %s (%d bytes)\n",
2405 (int)SvCUR(prop), SvPVX_const(prop),
2406 quoted, (int)(strend - s));
2409 if (find_byclass(prog, c, s, strend, ®info))
2411 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2415 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2420 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2421 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2422 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2424 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
2425 last = screaminstr(sv, float_real, s - strbeg,
2426 end_shift, &scream_pos, 1); /* last one */
2428 last = scream_olds; /* Only one occurrence. */
2429 /* we may be pointing at the wrong string */
2430 else if (RXp_MATCH_COPIED(prog))
2431 s = strbeg + (s - SvPVX_const(sv));
2435 const char * const little = SvPV_const(float_real, len);
2437 if (SvTAIL(float_real)) {
2438 if (memEQ(strend - len + 1, little, len - 1))
2439 last = strend - len + 1;
2440 else if (!multiline)
2441 last = memEQ(strend - len, little, len)
2442 ? strend - len : NULL;
2448 last = rninstr(s, strend, little, little + len);
2450 last = strend; /* matching "$" */
2455 PerlIO_printf(Perl_debug_log,
2456 "%sCan't trim the tail, match fails (should not happen)%s\n",
2457 PL_colors[4], PL_colors[5]));
2458 goto phooey; /* Should not happen! */
2460 dontbother = strend - last + prog->float_min_offset;
2462 if (minlen && (dontbother < minlen))
2463 dontbother = minlen - 1;
2464 strend -= dontbother; /* this one's always in bytes! */
2465 /* We don't know much -- general case. */
2468 if (regtry(®info, &s))
2477 if (regtry(®info, &s))
2479 } while (s++ < strend);
2488 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2490 if (PL_reg_eval_set)
2491 restore_pos(aTHX_ prog);
2492 if (RXp_PAREN_NAMES(prog))
2493 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2495 /* make sure $`, $&, $', and $digit will work later */
2496 if ( !(flags & REXEC_NOT_FIRST) ) {
2497 RX_MATCH_COPY_FREE(rx);
2498 if (flags & REXEC_COPY_STR) {
2499 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2500 #ifdef PERL_OLD_COPY_ON_WRITE
2502 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2504 PerlIO_printf(Perl_debug_log,
2505 "Copy on write: regexp capture, type %d\n",
2508 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2509 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2510 assert (SvPOKp(prog->saved_copy));
2514 RX_MATCH_COPIED_on(rx);
2515 s = savepvn(strbeg, i);
2521 prog->subbeg = strbeg;
2522 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2529 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2530 PL_colors[4], PL_colors[5]));
2531 if (PL_reg_eval_set)
2532 restore_pos(aTHX_ prog);
2534 /* we failed :-( roll it back */
2535 Safefree(prog->offs);
2544 - regtry - try match at specific point
2546 STATIC I32 /* 0 failure, 1 success */
2547 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2551 REGEXP *const rx = reginfo->prog;
2552 regexp *const prog = (struct regexp *)SvANY(rx);
2553 RXi_GET_DECL(prog,progi);
2554 GET_RE_DEBUG_FLAGS_DECL;
2556 PERL_ARGS_ASSERT_REGTRY;
2558 reginfo->cutpoint=NULL;
2560 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2563 PL_reg_eval_set = RS_init;
2564 DEBUG_EXECUTE_r(DEBUG_s(
2565 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2566 (IV)(PL_stack_sp - PL_stack_base));
2569 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2570 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2572 /* Apparently this is not needed, judging by wantarray. */
2573 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2574 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2577 /* Make $_ available to executed code. */
2578 if (reginfo->sv != DEFSV) {
2580 DEFSV_set(reginfo->sv);
2583 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2584 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2585 /* prepare for quick setting of pos */
2586 #ifdef PERL_OLD_COPY_ON_WRITE
2587 if (SvIsCOW(reginfo->sv))
2588 sv_force_normal_flags(reginfo->sv, 0);
2590 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2591 &PL_vtbl_mglob, NULL, 0);
2595 PL_reg_oldpos = mg->mg_len;
2596 SAVEDESTRUCTOR_X(restore_pos, prog);
2598 if (!PL_reg_curpm) {
2599 Newxz(PL_reg_curpm, 1, PMOP);
2602 SV* const repointer = &PL_sv_undef;
2603 /* this regexp is also owned by the new PL_reg_curpm, which
2604 will try to free it. */
2605 av_push(PL_regex_padav, repointer);
2606 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2607 PL_regex_pad = AvARRAY(PL_regex_padav);
2612 /* It seems that non-ithreads works both with and without this code.
2613 So for efficiency reasons it seems best not to have the code
2614 compiled when it is not needed. */
2615 /* This is safe against NULLs: */
2616 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2617 /* PM_reg_curpm owns a reference to this regexp. */
2618 (void)ReREFCNT_inc(rx);
2620 PM_SETRE(PL_reg_curpm, rx);
2621 PL_reg_oldcurpm = PL_curpm;
2622 PL_curpm = PL_reg_curpm;
2623 if (RXp_MATCH_COPIED(prog)) {
2624 /* Here is a serious problem: we cannot rewrite subbeg,
2625 since it may be needed if this match fails. Thus
2626 $` inside (?{}) could fail... */
2627 PL_reg_oldsaved = prog->subbeg;
2628 PL_reg_oldsavedlen = prog->sublen;
2629 #ifdef PERL_OLD_COPY_ON_WRITE
2630 PL_nrs = prog->saved_copy;
2632 RXp_MATCH_COPIED_off(prog);
2635 PL_reg_oldsaved = NULL;
2636 prog->subbeg = PL_bostr;
2637 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2639 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2640 prog->offs[0].start = *startpos - PL_bostr;
2641 PL_reginput = *startpos;
2642 PL_reglastparen = &prog->lastparen;
2643 PL_reglastcloseparen = &prog->lastcloseparen;
2644 prog->lastparen = 0;
2645 prog->lastcloseparen = 0;
2647 PL_regoffs = prog->offs;
2648 if (PL_reg_start_tmpl <= prog->nparens) {
2649 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2650 if(PL_reg_start_tmp)
2651 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2653 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2656 /* XXXX What this code is doing here?!!! There should be no need
2657 to do this again and again, PL_reglastparen should take care of
2660 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2661 * Actually, the code in regcppop() (which Ilya may be meaning by
2662 * PL_reglastparen), is not needed at all by the test suite
2663 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2664 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2665 * Meanwhile, this code *is* needed for the
2666 * above-mentioned test suite tests to succeed. The common theme
2667 * on those tests seems to be returning null fields from matches.
2668 * --jhi updated by dapm */
2670 if (prog->nparens) {
2671 regexp_paren_pair *pp = PL_regoffs;
2673 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2681 if (regmatch(reginfo, progi->program + 1)) {
2682 PL_regoffs[0].end = PL_reginput - PL_bostr;
2685 if (reginfo->cutpoint)
2686 *startpos= reginfo->cutpoint;
2687 REGCP_UNWIND(lastcp);
2692 #define sayYES goto yes
2693 #define sayNO goto no
2694 #define sayNO_SILENT goto no_silent
2696 /* we dont use STMT_START/END here because it leads to
2697 "unreachable code" warnings, which are bogus, but distracting. */
2698 #define CACHEsayNO \
2699 if (ST.cache_mask) \
2700 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2703 /* this is used to determine how far from the left messages like
2704 'failed...' are printed. It should be set such that messages
2705 are inline with the regop output that created them.
2707 #define REPORT_CODE_OFF 32
2710 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2711 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2713 #define SLAB_FIRST(s) (&(s)->states[0])
2714 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2716 /* grab a new slab and return the first slot in it */
2718 STATIC regmatch_state *
2721 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2724 regmatch_slab *s = PL_regmatch_slab->next;
2726 Newx(s, 1, regmatch_slab);
2727 s->prev = PL_regmatch_slab;
2729 PL_regmatch_slab->next = s;
2731 PL_regmatch_slab = s;
2732 return SLAB_FIRST(s);
2736 /* push a new state then goto it */
2738 #define PUSH_STATE_GOTO(state, node) \
2740 st->resume_state = state; \
2743 /* push a new state with success backtracking, then goto it */
2745 #define PUSH_YES_STATE_GOTO(state, node) \
2747 st->resume_state = state; \
2748 goto push_yes_state;
2754 regmatch() - main matching routine
2756 This is basically one big switch statement in a loop. We execute an op,
2757 set 'next' to point the next op, and continue. If we come to a point which
2758 we may need to backtrack to on failure such as (A|B|C), we push a
2759 backtrack state onto the backtrack stack. On failure, we pop the top
2760 state, and re-enter the loop at the state indicated. If there are no more
2761 states to pop, we return failure.
2763 Sometimes we also need to backtrack on success; for example /A+/, where
2764 after successfully matching one A, we need to go back and try to
2765 match another one; similarly for lookahead assertions: if the assertion
2766 completes successfully, we backtrack to the state just before the assertion
2767 and then carry on. In these cases, the pushed state is marked as
2768 'backtrack on success too'. This marking is in fact done by a chain of
2769 pointers, each pointing to the previous 'yes' state. On success, we pop to
2770 the nearest yes state, discarding any intermediate failure-only states.
2771 Sometimes a yes state is pushed just to force some cleanup code to be
2772 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2773 it to free the inner regex.
2775 Note that failure backtracking rewinds the cursor position, while
2776 success backtracking leaves it alone.
2778 A pattern is complete when the END op is executed, while a subpattern
2779 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2780 ops trigger the "pop to last yes state if any, otherwise return true"
2783 A common convention in this function is to use A and B to refer to the two
2784 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2785 the subpattern to be matched possibly multiple times, while B is the entire
2786 rest of the pattern. Variable and state names reflect this convention.
2788 The states in the main switch are the union of ops and failure/success of
2789 substates associated with with that op. For example, IFMATCH is the op
2790 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2791 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2792 successfully matched A and IFMATCH_A_fail is a state saying that we have
2793 just failed to match A. Resume states always come in pairs. The backtrack
2794 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2795 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2796 on success or failure.
2798 The struct that holds a backtracking state is actually a big union, with
2799 one variant for each major type of op. The variable st points to the
2800 top-most backtrack struct. To make the code clearer, within each
2801 block of code we #define ST to alias the relevant union.
2803 Here's a concrete example of a (vastly oversimplified) IFMATCH
2809 #define ST st->u.ifmatch
2811 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2812 ST.foo = ...; // some state we wish to save
2814 // push a yes backtrack state with a resume value of
2815 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2817 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2820 case IFMATCH_A: // we have successfully executed A; now continue with B
2822 bar = ST.foo; // do something with the preserved value
2825 case IFMATCH_A_fail: // A failed, so the assertion failed
2826 ...; // do some housekeeping, then ...
2827 sayNO; // propagate the failure
2834 For any old-timers reading this who are familiar with the old recursive
2835 approach, the code above is equivalent to:
2837 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2846 ...; // do some housekeeping, then ...
2847 sayNO; // propagate the failure
2850 The topmost backtrack state, pointed to by st, is usually free. If you
2851 want to claim it, populate any ST.foo fields in it with values you wish to
2852 save, then do one of
2854 PUSH_STATE_GOTO(resume_state, node);
2855 PUSH_YES_STATE_GOTO(resume_state, node);
2857 which sets that backtrack state's resume value to 'resume_state', pushes a
2858 new free entry to the top of the backtrack stack, then goes to 'node'.
2859 On backtracking, the free slot is popped, and the saved state becomes the
2860 new free state. An ST.foo field in this new top state can be temporarily
2861 accessed to retrieve values, but once the main loop is re-entered, it
2862 becomes available for reuse.
2864 Note that the depth of the backtrack stack constantly increases during the
2865 left-to-right execution of the pattern, rather than going up and down with
2866 the pattern nesting. For example the stack is at its maximum at Z at the
2867 end of the pattern, rather than at X in the following:
2869 /(((X)+)+)+....(Y)+....Z/
2871 The only exceptions to this are lookahead/behind assertions and the cut,
2872 (?>A), which pop all the backtrack states associated with A before
2875 Backtrack state structs are allocated in slabs of about 4K in size.
2876 PL_regmatch_state and st always point to the currently active state,
2877 and PL_regmatch_slab points to the slab currently containing
2878 PL_regmatch_state. The first time regmatch() is called, the first slab is
2879 allocated, and is never freed until interpreter destruction. When the slab
2880 is full, a new one is allocated and chained to the end. At exit from
2881 regmatch(), slabs allocated since entry are freed.
2886 #define DEBUG_STATE_pp(pp) \
2888 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2889 PerlIO_printf(Perl_debug_log, \
2890 " %*s"pp" %s%s%s%s%s\n", \
2892 PL_reg_name[st->resume_state], \
2893 ((st==yes_state||st==mark_state) ? "[" : ""), \
2894 ((st==yes_state) ? "Y" : ""), \
2895 ((st==mark_state) ? "M" : ""), \
2896 ((st==yes_state||st==mark_state) ? "]" : "") \
2901 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2906 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2907 const char *start, const char *end, const char *blurb)
2909 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2911 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2916 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2917 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2919 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2920 start, end - start, 60);
2922 PerlIO_printf(Perl_debug_log,
2923 "%s%s REx%s %s against %s\n",
2924 PL_colors[4], blurb, PL_colors[5], s0, s1);
2926 if (utf8_target||utf8_pat)
2927 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2928 utf8_pat ? "pattern" : "",
2929 utf8_pat && utf8_target ? " and " : "",
2930 utf8_target ? "string" : ""
2936 S_dump_exec_pos(pTHX_ const char *locinput,
2937 const regnode *scan,
2938 const char *loc_regeol,
2939 const char *loc_bostr,
2940 const char *loc_reg_starttry,
2941 const bool utf8_target)
2943 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2944 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2945 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2946 /* The part of the string before starttry has one color
2947 (pref0_len chars), between starttry and current
2948 position another one (pref_len - pref0_len chars),
2949 after the current position the third one.
2950 We assume that pref0_len <= pref_len, otherwise we
2951 decrease pref0_len. */
2952 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2953 ? (5 + taill) - l : locinput - loc_bostr;
2956 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2958 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2960 pref0_len = pref_len - (locinput - loc_reg_starttry);
2961 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2962 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2963 ? (5 + taill) - pref_len : loc_regeol - locinput);
2964 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2968 if (pref0_len > pref_len)
2969 pref0_len = pref_len;
2971 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2973 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2974 (locinput - pref_len),pref0_len, 60, 4, 5);
2976 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2977 (locinput - pref_len + pref0_len),
2978 pref_len - pref0_len, 60, 2, 3);
2980 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2981 locinput, loc_regeol - locinput, 10, 0, 1);
2983 const STRLEN tlen=len0+len1+len2;
2984 PerlIO_printf(Perl_debug_log,
2985 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2986 (IV)(locinput - loc_bostr),
2989 (docolor ? "" : "> <"),
2991 (int)(tlen > 19 ? 0 : 19 - tlen),
2998 /* reg_check_named_buff_matched()
2999 * Checks to see if a named buffer has matched. The data array of
3000 * buffer numbers corresponding to the buffer is expected to reside
3001 * in the regexp->data->data array in the slot stored in the ARG() of
3002 * node involved. Note that this routine doesn't actually care about the
3003 * name, that information is not preserved from compilation to execution.
3004 * Returns the index of the leftmost defined buffer with the given name
3005 * or 0 if non of the buffers matched.
3008 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3011 RXi_GET_DECL(rex,rexi);
3012 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3013 I32 *nums=(I32*)SvPVX(sv_dat);
3015 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3017 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3018 if ((I32)*PL_reglastparen >= nums[n] &&
3019 PL_regoffs[nums[n]].end != -1)
3028 /* free all slabs above current one - called during LEAVE_SCOPE */
3031 S_clear_backtrack_stack(pTHX_ void *p)
3033 regmatch_slab *s = PL_regmatch_slab->next;
3038 PL_regmatch_slab->next = NULL;
3040 regmatch_slab * const osl = s;
3047 #define SETREX(Re1,Re2) \
3048 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3051 STATIC I32 /* 0 failure, 1 success */
3052 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3054 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3058 register const bool utf8_target = PL_reg_match_utf8;
3059 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3060 REGEXP *rex_sv = reginfo->prog;
3061 regexp *rex = (struct regexp *)SvANY(rex_sv);
3062 RXi_GET_DECL(rex,rexi);
3064 /* the current state. This is a cached copy of PL_regmatch_state */
3065 register regmatch_state *st;
3066 /* cache heavy used fields of st in registers */
3067 register regnode *scan;
3068 register regnode *next;
3069 register U32 n = 0; /* general value; init to avoid compiler warning */
3070 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3071 register char *locinput = PL_reginput;
3072 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3074 bool result = 0; /* return value of S_regmatch */
3075 int depth = 0; /* depth of backtrack stack */
3076 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3077 const U32 max_nochange_depth =
3078 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3079 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3080 regmatch_state *yes_state = NULL; /* state to pop to on success of
3082 /* mark_state piggy backs on the yes_state logic so that when we unwind
3083 the stack on success we can update the mark_state as we go */
3084 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3085 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3086 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3088 bool no_final = 0; /* prevent failure from backtracking? */
3089 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3090 char *startpoint = PL_reginput;
3091 SV *popmark = NULL; /* are we looking for a mark? */
3092 SV *sv_commit = NULL; /* last mark name seen in failure */
3093 SV *sv_yes_mark = NULL; /* last mark name we have seen
3094 during a successful match */
3095 U32 lastopen = 0; /* last open we saw */
3096 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3097 SV* const oreplsv = GvSV(PL_replgv);
3098 /* these three flags are set by various ops to signal information to
3099 * the very next op. They have a useful lifetime of exactly one loop
3100 * iteration, and are not preserved or restored by state pushes/pops
3102 bool sw = 0; /* the condition value in (?(cond)a|b) */
3103 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3104 int logical = 0; /* the following EVAL is:
3108 or the following IFMATCH/UNLESSM is:
3109 false: plain (?=foo)
3110 true: used as a condition: (?(?=foo))
3113 GET_RE_DEBUG_FLAGS_DECL;
3116 PERL_ARGS_ASSERT_REGMATCH;
3118 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3119 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3121 /* on first ever call to regmatch, allocate first slab */
3122 if (!PL_regmatch_slab) {
3123 Newx(PL_regmatch_slab, 1, regmatch_slab);
3124 PL_regmatch_slab->prev = NULL;
3125 PL_regmatch_slab->next = NULL;
3126 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3129 oldsave = PL_savestack_ix;
3130 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3131 SAVEVPTR(PL_regmatch_slab);
3132 SAVEVPTR(PL_regmatch_state);
3134 /* grab next free state slot */
3135 st = ++PL_regmatch_state;
3136 if (st > SLAB_LAST(PL_regmatch_slab))
3137 st = PL_regmatch_state = S_push_slab(aTHX);
3139 /* Note that nextchr is a byte even in UTF */
3140 nextchr = UCHARAT(locinput);
3142 while (scan != NULL) {
3145 SV * const prop = sv_newmortal();
3146 regnode *rnext=regnext(scan);
3147 DUMP_EXEC_POS( locinput, scan, utf8_target );
3148 regprop(rex, prop, scan);
3150 PerlIO_printf(Perl_debug_log,
3151 "%3"IVdf":%*s%s(%"IVdf")\n",
3152 (IV)(scan - rexi->program), depth*2, "",
3154 (PL_regkind[OP(scan)] == END || !rnext) ?
3155 0 : (IV)(rnext - rexi->program));
3158 next = scan + NEXT_OFF(scan);
3161 state_num = OP(scan);
3165 assert(PL_reglastparen == &rex->lastparen);
3166 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3167 assert(PL_regoffs == rex->offs);
3169 switch (state_num) {
3171 if (locinput == PL_bostr)
3173 /* reginfo->till = reginfo->bol; */
3178 if (locinput == PL_bostr ||
3179 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3185 if (locinput == PL_bostr)
3189 if (locinput == reginfo->ganch)
3194 /* update the startpoint */
3195 st->u.keeper.val = PL_regoffs[0].start;
3196 PL_reginput = locinput;
3197 PL_regoffs[0].start = locinput - PL_bostr;
3198 PUSH_STATE_GOTO(KEEPS_next, next);
3200 case KEEPS_next_fail:
3201 /* rollback the start point change */
3202 PL_regoffs[0].start = st->u.keeper.val;
3208 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3213 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3215 if (PL_regeol - locinput > 1)
3219 if (PL_regeol != locinput)
3223 if (!nextchr && locinput >= PL_regeol)
3226 locinput += PL_utf8skip[nextchr];
3227 if (locinput > PL_regeol)
3229 nextchr = UCHARAT(locinput);
3232 nextchr = UCHARAT(++locinput);
3235 if (!nextchr && locinput >= PL_regeol)
3237 nextchr = UCHARAT(++locinput);
3240 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3243 locinput += PL_utf8skip[nextchr];
3244 if (locinput > PL_regeol)
3246 nextchr = UCHARAT(locinput);
3249 nextchr = UCHARAT(++locinput);
3253 #define ST st->u.trie
3255 /* In this case the charclass data is available inline so
3256 we can fail fast without a lot of extra overhead.
3258 if (scan->flags == EXACT || !utf8_target) {
3259 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3261 PerlIO_printf(Perl_debug_log,
3262 "%*s %sfailed to match trie start class...%s\n",
3263 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3271 /* the basic plan of execution of the trie is:
3272 * At the beginning, run though all the states, and
3273 * find the longest-matching word. Also remember the position
3274 * of the shortest matching word. For example, this pattern:
3277 * when matched against the string "abcde", will generate
3278 * accept states for all words except 3, with the longest
3279 * matching word being 4, and the shortest being 1 (with
3280 * the position being after char 1 of the string).
3282 * Then for each matching word, in word order (i.e. 1,2,4,5),
3283 * we run the remainder of the pattern; on each try setting
3284 * the current position to the character following the word,
3285 * returning to try the next word on failure.
3287 * We avoid having to build a list of words at runtime by
3288 * using a compile-time structure, wordinfo[].prev, which
3289 * gives, for each word, the previous accepting word (if any).
3290 * In the case above it would contain the mappings 1->2, 2->0,
3291 * 3->0, 4->5, 5->1. We can use this table to generate, from
3292 * the longest word (4 above), a list of all words, by
3293 * following the list of prev pointers; this gives us the
3294 * unordered list 4,5,1,2. Then given the current word we have
3295 * just tried, we can go through the list and find the
3296 * next-biggest word to try (so if we just failed on word 2,
3297 * the next in the list is 4).
3299 * Since at runtime we don't record the matching position in
3300 * the string for each word, we have to work that out for
3301 * each word we're about to process. The wordinfo table holds
3302 * the character length of each word; given that we recorded
3303 * at the start: the position of the shortest word and its
3304 * length in chars, we just need to move the pointer the
3305 * difference between the two char lengths. Depending on
3306 * Unicode status and folding, that's cheap or expensive.
3308 * This algorithm is optimised for the case where are only a
3309 * small number of accept states, i.e. 0,1, or maybe 2.
3310 * With lots of accepts states, and having to try all of them,
3311 * it becomes quadratic on number of accept states to find all
3316 /* what type of TRIE am I? (utf8 makes this contextual) */
3317 DECL_TRIE_TYPE(scan);
3319 /* what trie are we using right now */
3320 reg_trie_data * const trie
3321 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3322 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3323 U32 state = trie->startstate;
3325 if (trie->bitmap && trie_type != trie_utf8_fold &&
3326 !TRIE_BITMAP_TEST(trie,*locinput)
3328 if (trie->states[ state ].wordnum) {
3330 PerlIO_printf(Perl_debug_log,
3331 "%*s %smatched empty string...%s\n",
3332 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3338 PerlIO_printf(Perl_debug_log,
3339 "%*s %sfailed to match trie start class...%s\n",
3340 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3347 U8 *uc = ( U8* )locinput;
3351 U8 *uscan = (U8*)NULL;
3352 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3353 U32 charcount = 0; /* how many input chars we have matched */
3354 U32 accepted = 0; /* have we seen any accepting states? */
3357 ST.jump = trie->jump;
3360 ST.longfold = FALSE; /* char longer if folded => it's harder */
3363 /* fully traverse the TRIE; note the position of the
3364 shortest accept state and the wordnum of the longest
3367 while ( state && uc <= (U8*)PL_regeol ) {
3368 U32 base = trie->states[ state ].trans.base;
3372 wordnum = trie->states[ state ].wordnum;
3374 if (wordnum) { /* it's an accept state */
3377 /* record first match position */
3379 ST.firstpos = (U8*)locinput;
3384 ST.firstchars = charcount;
3387 if (!ST.nextword || wordnum < ST.nextword)
3388 ST.nextword = wordnum;
3389 ST.topword = wordnum;
3392 DEBUG_TRIE_EXECUTE_r({
3393 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3394 PerlIO_printf( Perl_debug_log,
3395 "%*s %sState: %4"UVxf" Accepted: %c ",
3396 2+depth * 2, "", PL_colors[4],
3397 (UV)state, (accepted ? 'Y' : 'N'));
3400 /* read a char and goto next state */
3403 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3404 uscan, len, uvc, charid, foldlen,
3411 base + charid - 1 - trie->uniquecharcount)) >= 0)
3413 && ((U32)offset < trie->lasttrans)
3414 && trie->trans[offset].check == state)
3416 state = trie->trans[offset].next;
3427 DEBUG_TRIE_EXECUTE_r(
3428 PerlIO_printf( Perl_debug_log,
3429 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3430 charid, uvc, (UV)state, PL_colors[5] );
3436 /* calculate total number of accept states */
3441 w = trie->wordinfo[w].prev;
3444 ST.accepted = accepted;
3448 PerlIO_printf( Perl_debug_log,
3449 "%*s %sgot %"IVdf" possible matches%s\n",
3450 REPORT_CODE_OFF + depth * 2, "",
3451 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3453 goto trie_first_try; /* jump into the fail handler */
3457 case TRIE_next_fail: /* we failed - try next alternative */
3459 REGCP_UNWIND(ST.cp);
3460 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3461 PL_regoffs[n].end = -1;
3462 *PL_reglastparen = n;
3464 if (!--ST.accepted) {
3466 PerlIO_printf( Perl_debug_log,
3467 "%*s %sTRIE failed...%s\n",
3468 REPORT_CODE_OFF+depth*2, "",
3475 /* Find next-highest word to process. Note that this code
3476 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3477 register U16 min = 0;
3479 register U16 const nextword = ST.nextword;
3480 register reg_trie_wordinfo * const wordinfo
3481 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3482 for (word=ST.topword; word; word=wordinfo[word].prev) {
3483 if (word > nextword && (!min || word < min))
3496 ST.lastparen = *PL_reglastparen;
3500 /* find start char of end of current word */
3502 U32 chars; /* how many chars to skip */
3503 U8 *uc = ST.firstpos;
3504 reg_trie_data * const trie
3505 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3507 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3509 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3513 /* the hard option - fold each char in turn and find
3514 * its folded length (which may be different */
3515 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3523 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3531 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3536 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3550 PL_reginput = (char *)uc;
3553 scan = (ST.jump && ST.jump[ST.nextword])
3554 ? ST.me + ST.jump[ST.nextword]
3558 PerlIO_printf( Perl_debug_log,
3559 "%*s %sTRIE matched word #%d, continuing%s\n",
3560 REPORT_CODE_OFF+depth*2, "",
3567 if (ST.accepted > 1 || has_cutgroup) {
3568 PUSH_STATE_GOTO(TRIE_next, scan);
3571 /* only one choice left - just continue */
3573 AV *const trie_words
3574 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3575 SV ** const tmp = av_fetch( trie_words,
3577 SV *sv= tmp ? sv_newmortal() : NULL;
3579 PerlIO_printf( Perl_debug_log,
3580 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3581 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3583 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3584 PL_colors[0], PL_colors[1],
3585 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3587 : "not compiled under -Dr",
3591 locinput = PL_reginput;
3592 nextchr = UCHARAT(locinput);
3593 continue; /* execute rest of RE */
3598 char *s = STRING(scan);
3600 if (utf8_target != UTF_PATTERN) {
3601 /* The target and the pattern have differing utf8ness. */
3603 const char * const e = s + ln;
3606 /* The target is utf8, the pattern is not utf8. */
3611 if (NATIVE_TO_UNI(*(U8*)s) !=
3612 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3620 /* The target is not utf8, the pattern is utf8. */
3625 if (NATIVE_TO_UNI(*((U8*)l)) !=
3626 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3634 nextchr = UCHARAT(locinput);
3637 /* The target and the pattern have the same utf8ness. */
3638 /* Inline the first character, for speed. */
3639 if (UCHARAT(s) != nextchr)
3641 if (PL_regeol - locinput < ln)
3643 if (ln > 1 && memNE(s, locinput, ln))
3646 nextchr = UCHARAT(locinput);
3651 const U8 * fold_array;
3653 U32 fold_utf8_flags;
3655 PL_reg_flags |= RF_tainted;
3656 folder = foldEQ_locale;
3657 fold_array = PL_fold_locale;
3658 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3662 folder = foldEQ_latin1;
3663 fold_array = PL_fold_latin1;
3664 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3668 folder = foldEQ_latin1;
3669 fold_array = PL_fold_latin1;
3670 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3675 fold_array = PL_fold;
3676 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3682 if (utf8_target || UTF_PATTERN) {
3683 /* Either target or the pattern are utf8. */
3684 const char * const l = locinput;
3685 char *e = PL_regeol;
3687 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3688 l, &e, 0, utf8_target, fold_utf8_flags))
3693 nextchr = UCHARAT(locinput);
3697 /* Neither the target nor the pattern are utf8 */
3698 if (UCHARAT(s) != nextchr &&
3699 UCHARAT(s) != fold_array[nextchr])
3703 if (PL_regeol - locinput < ln)
3705 if (ln > 1 && ! folder(s, locinput, ln))
3708 nextchr = UCHARAT(locinput);
3712 /* XXX Could improve efficiency by separating these all out using a
3713 * macro or in-line function. At that point regcomp.c would no longer
3714 * have to set the FLAGS fields of these */
3717 PL_reg_flags |= RF_tainted;
3725 /* was last char in word? */
3727 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3728 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3730 if (locinput == PL_bostr)
3733 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3735 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3737 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3738 ln = isALNUM_uni(ln);
3739 LOAD_UTF8_CHARCLASS_ALNUM();
3740 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3743 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3744 n = isALNUM_LC_utf8((U8*)locinput);
3749 /* Here the string isn't utf8, or is utf8 and only ascii
3750 * characters are to match \w. In the latter case looking at
3751 * the byte just prior to the current one may be just the final
3752 * byte of a multi-byte character. This is ok. There are two
3754 * 1) it is a single byte character, and then the test is doing
3755 * just what it's supposed to.
3756 * 2) it is a multi-byte character, in which case the final
3757 * byte is never mistakable for ASCII, and so the test
3758 * will say it is not a word character, which is the
3759 * correct answer. */
3760 ln = (locinput != PL_bostr) ?
3761 UCHARAT(locinput - 1) : '\n';
3762 switch (FLAGS(scan)) {
3763 case REGEX_UNICODE_CHARSET:
3764 ln = isWORDCHAR_L1(ln);
3765 n = isWORDCHAR_L1(nextchr);
3767 case REGEX_LOCALE_CHARSET:
3768 ln = isALNUM_LC(ln);
3769 n = isALNUM_LC(nextchr);
3771 case REGEX_DEPENDS_CHARSET:
3773 n = isALNUM(nextchr);
3775 case REGEX_ASCII_RESTRICTED_CHARSET:
3776 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3777 ln = isWORDCHAR_A(ln);
3778 n = isWORDCHAR_A(nextchr);
3781 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3785 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3787 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3792 if (utf8_target || state_num == ANYOFV) {
3793 STRLEN inclasslen = PL_regeol - locinput;
3794 if (locinput >= PL_regeol)
3797 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3799 locinput += inclasslen;
3800 nextchr = UCHARAT(locinput);
3805 nextchr = UCHARAT(locinput);
3806 if (!nextchr && locinput >= PL_regeol)
3808 if (!REGINCLASS(rex, scan, (U8*)locinput))
3810 nextchr = UCHARAT(++locinput);
3814 /* Special char classes - The defines start on line 129 or so */
3815 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3816 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3817 ALNUMU, NALNUMU, isWORDCHAR_L1,
3818 ALNUMA, NALNUMA, isWORDCHAR_A,
3821 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3822 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3823 SPACEU, NSPACEU, isSPACE_L1,
3824 SPACEA, NSPACEA, isSPACE_A,
3827 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3828 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3829 DIGITA, NDIGITA, isDIGIT_A,
3832 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3833 a Unicode extended Grapheme Cluster */
3834 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3835 extended Grapheme Cluster is:
3838 | Prepend* Begin Extend*
3841 Begin is (Hangul-syllable | ! Control)
3842 Extend is (Grapheme_Extend | Spacing_Mark)
3843 Control is [ GCB_Control CR LF ]
3845 The discussion below shows how the code for CLUMP is derived
3846 from this regex. Note that most of these concepts are from
3847 property values of the Grapheme Cluster Boundary (GCB) property.
3848 No code point can have multiple property values for a given
3849 property. Thus a code point in Prepend can't be in Control, but
3850 it must be in !Control. This is why Control above includes
3851 GCB_Control plus CR plus LF. The latter two are used in the GCB
3852 property separately, and so can't be in GCB_Control, even though
3853 they logically are controls. Control is not the same as gc=cc,
3854 but includes format and other characters as well.
3856 The Unicode definition of Hangul-syllable is:
3858 | (L* ( ( V | LV ) V* | LVT ) T*)
3861 Each of these is a value for the GCB property, and hence must be
3862 disjoint, so the order they are tested is immaterial, so the
3863 above can safely be changed to
3866 | (L* ( LVT | ( V | LV ) V*) T*)
3868 The last two terms can be combined like this:
3870 | (( LVT | ( V | LV ) V*) T*))
3872 And refactored into this:
3873 L* (L | LVT T* | V V* T* | LV V* T*)
3875 That means that if we have seen any L's at all we can quit
3876 there, but if the next character is an LVT, a V, or an LV we
3879 There is a subtlety with Prepend* which showed up in testing.
3880 Note that the Begin, and only the Begin is required in:
3881 | Prepend* Begin Extend*
3882 Also, Begin contains '! Control'. A Prepend must be a
3883 '! Control', which means it must also be a Begin. What it
3884 comes down to is that if we match Prepend* and then find no
3885 suitable Begin afterwards, that if we backtrack the last
3886 Prepend, that one will be a suitable Begin.
3889 if (locinput >= PL_regeol)
3891 if (! utf8_target) {
3893 /* Match either CR LF or '.', as all the other possibilities
3895 locinput++; /* Match the . or CR */
3896 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3898 && locinput < PL_regeol
3899 && UCHARAT(locinput) == '\n') locinput++;
3903 /* Utf8: See if is ( CR LF ); already know that locinput <
3904 * PL_regeol, so locinput+1 is in bounds */
3905 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3909 /* In case have to backtrack to beginning, then match '.' */
3910 char *starting = locinput;
3912 /* In case have to backtrack the last prepend */
3913 char *previous_prepend = 0;
3915 LOAD_UTF8_CHARCLASS_GCB();
3917 /* Match (prepend)* */
3918 while (locinput < PL_regeol
3919 && swash_fetch(PL_utf8_X_prepend,
3920 (U8*)locinput, utf8_target))
3922 previous_prepend = locinput;
3923 locinput += UTF8SKIP(locinput);
3926 /* As noted above, if we matched a prepend character, but
3927 * the next thing won't match, back off the last prepend we
3928 * matched, as it is guaranteed to match the begin */
3929 if (previous_prepend
3930 && (locinput >= PL_regeol
3931 || ! swash_fetch(PL_utf8_X_begin,
3932 (U8*)locinput, utf8_target)))
3934 locinput = previous_prepend;
3937 /* Note that here we know PL_regeol > locinput, as we
3938 * tested that upon input to this switch case, and if we
3939 * moved locinput forward, we tested the result just above
3940 * and it either passed, or we backed off so that it will
3942 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3944 /* Here did not match the required 'Begin' in the
3945 * second term. So just match the very first
3946 * character, the '.' of the final term of the regex */
3947 locinput = starting + UTF8SKIP(starting);
3950 /* Here is the beginning of a character that can have
3951 * an extender. It is either a hangul syllable, or a
3953 if (swash_fetch(PL_utf8_X_non_hangul,
3954 (U8*)locinput, utf8_target))
3957 /* Here not a Hangul syllable, must be a
3958 * ('! * Control') */
3959 locinput += UTF8SKIP(locinput);
3962 /* Here is a Hangul syllable. It can be composed
3963 * of several individual characters. One
3964 * possibility is T+ */
3965 if (swash_fetch(PL_utf8_X_T,
3966 (U8*)locinput, utf8_target))
3968 while (locinput < PL_regeol
3969 && swash_fetch(PL_utf8_X_T,
3970 (U8*)locinput, utf8_target))
3972 locinput += UTF8SKIP(locinput);
3976 /* Here, not T+, but is a Hangul. That means
3977 * it is one of the others: L, LV, LVT or V,
3979 * L* (L | LVT T* | V V* T* | LV V* T*) */
3982 while (locinput < PL_regeol
3983 && swash_fetch(PL_utf8_X_L,
3984 (U8*)locinput, utf8_target))
3986 locinput += UTF8SKIP(locinput);
3989 /* Here, have exhausted L*. If the next
3990 * character is not an LV, LVT nor V, it means
3991 * we had to have at least one L, so matches L+
3992 * in the original equation, we have a complete
3993 * hangul syllable. Are done. */
3995 if (locinput < PL_regeol
3996 && swash_fetch(PL_utf8_X_LV_LVT_V,
3997 (U8*)locinput, utf8_target))
4000 /* Otherwise keep going. Must be LV, LVT
4001 * or V. See if LVT */
4002 if (swash_fetch(PL_utf8_X_LVT,
4003 (U8*)locinput, utf8_target))
4005 locinput += UTF8SKIP(locinput);
4008 /* Must be V or LV. Take it, then
4010 locinput += UTF8SKIP(locinput);
4011 while (locinput < PL_regeol
4012 && swash_fetch(PL_utf8_X_V,
4013 (U8*)locinput, utf8_target))
4015 locinput += UTF8SKIP(locinput);
4019 /* And any of LV, LVT, or V can be followed
4021 while (locinput < PL_regeol
4022 && swash_fetch(PL_utf8_X_T,
4026 locinput += UTF8SKIP(locinput);
4032 /* Match any extender */
4033 while (locinput < PL_regeol
4034 && swash_fetch(PL_utf8_X_extend,
4035 (U8*)locinput, utf8_target))
4037 locinput += UTF8SKIP(locinput);
4041 if (locinput > PL_regeol) sayNO;
4043 nextchr = UCHARAT(locinput);
4047 { /* The capture buffer cases. The ones beginning with N for the
4048 named buffers just convert to the equivalent numbered and
4049 pretend they were called as the corresponding numbered buffer
4051 /* don't initialize these in the declaration, it makes C++
4056 const U8 *fold_array;
4059 PL_reg_flags |= RF_tainted;
4060 folder = foldEQ_locale;
4061 fold_array = PL_fold_locale;
4063 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4067 folder = foldEQ_latin1;
4068 fold_array = PL_fold_latin1;
4070 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4074 folder = foldEQ_latin1;
4075 fold_array = PL_fold_latin1;
4077 utf8_fold_flags = 0;
4082 fold_array = PL_fold;
4084 utf8_fold_flags = 0;
4091 utf8_fold_flags = 0;
4094 /* For the named back references, find the corresponding buffer
4096 n = reg_check_named_buff_matched(rex,scan);
4101 goto do_nref_ref_common;
4104 PL_reg_flags |= RF_tainted;
4105 folder = foldEQ_locale;
4106 fold_array = PL_fold_locale;
4107 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4111 folder = foldEQ_latin1;
4112 fold_array = PL_fold_latin1;
4113 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4117 folder = foldEQ_latin1;
4118 fold_array = PL_fold_latin1;
4119 utf8_fold_flags = 0;
4124 fold_array = PL_fold;
4125 utf8_fold_flags = 0;
4131 utf8_fold_flags = 0;
4135 n = ARG(scan); /* which paren pair */
4138 ln = PL_regoffs[n].start;
4139 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4140 if (*PL_reglastparen < n || ln == -1)
4141 sayNO; /* Do not match unless seen CLOSEn. */
4142 if (ln == PL_regoffs[n].end)
4146 if (type != REF /* REF can do byte comparison */
4147 && (utf8_target || type == REFFU))
4148 { /* XXX handle REFFL better */
4149 char * limit = PL_regeol;
4151 /* This call case insensitively compares the entire buffer
4152 * at s, with the current input starting at locinput, but
4153 * not going off the end given by PL_regeol, and returns in
4154 * limit upon success, how much of the current input was
4156 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4157 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4162 nextchr = UCHARAT(locinput);
4166 /* Not utf8: Inline the first character, for speed. */
4167 if (UCHARAT(s) != nextchr &&
4169 UCHARAT(s) != fold_array[nextchr]))
4171 ln = PL_regoffs[n].end - ln;
4172 if (locinput + ln > PL_regeol)
4174 if (ln > 1 && (type == REF
4175 ? memNE(s, locinput, ln)
4176 : ! folder(s, locinput, ln)))
4179 nextchr = UCHARAT(locinput);
4189 #define ST st->u.eval
4194 regexp_internal *rei;
4195 regnode *startpoint;
4198 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4199 if (cur_eval && cur_eval->locinput==locinput) {
4200 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4201 Perl_croak(aTHX_ "Infinite recursion in regex");
4202 if ( ++nochange_depth > max_nochange_depth )
4204 "Pattern subroutine nesting without pos change"
4205 " exceeded limit in regex");
4212 (void)ReREFCNT_inc(rex_sv);
4213 if (OP(scan)==GOSUB) {
4214 startpoint = scan + ARG2L(scan);
4215 ST.close_paren = ARG(scan);
4217 startpoint = rei->program+1;
4220 goto eval_recurse_doit;
4222 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4223 if (cur_eval && cur_eval->locinput==locinput) {
4224 if ( ++nochange_depth > max_nochange_depth )
4225 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4230 /* execute the code in the {...} */
4232 SV ** const before = SP;
4233 OP_4tree * const oop = PL_op;
4234 COP * const ocurcop = PL_curcop;
4236 char *saved_regeol = PL_regeol;
4237 struct re_save_state saved_state;
4239 /* To not corrupt the existing regex state while executing the
4240 * eval we would normally put it on the save stack, like with
4241 * save_re_context. However, re-evals have a weird scoping so we
4242 * can't just add ENTER/LEAVE here. With that, things like
4244 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4246 * would break, as they expect the localisation to be unwound
4247 * only when the re-engine backtracks through the bit that
4250 * What we do instead is just saving the state in a local c
4253 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4256 PL_op = (OP_4tree*)rexi->data->data[n];
4257 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4258 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4259 /* wrap the call in two SAVECOMPPADs. This ensures that
4260 * when the save stack is eventually unwound, all the
4261 * accumulated SAVEt_CLEARSV's will be processed with
4262 * interspersed SAVEt_COMPPAD's to ensure that lexicals
4263 * are cleared in the right pad */
4265 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4266 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4269 SV *sv_mrk = get_sv("REGMARK", 1);
4270 sv_setsv(sv_mrk, sv_yes_mark);
4273 CALLRUNOPS(aTHX); /* Scalar context. */
4276 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4282 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4286 PAD_RESTORE_LOCAL(old_comppad);
4287 PL_curcop = ocurcop;
4288 PL_regeol = saved_regeol;
4291 sv_setsv(save_scalar(PL_replgv), ret);
4295 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4298 /* extract RE object from returned value; compiling if
4304 SV *const sv = SvRV(ret);
4306 if (SvTYPE(sv) == SVt_REGEXP) {
4308 } else if (SvSMAGICAL(sv)) {
4309 mg = mg_find(sv, PERL_MAGIC_qr);
4312 } else if (SvTYPE(ret) == SVt_REGEXP) {
4314 } else if (SvSMAGICAL(ret)) {
4315 if (SvGMAGICAL(ret)) {
4316 /* I don't believe that there is ever qr magic
4318 assert(!mg_find(ret, PERL_MAGIC_qr));
4319 sv_unmagic(ret, PERL_MAGIC_qr);
4322 mg = mg_find(ret, PERL_MAGIC_qr);
4323 /* testing suggests mg only ends up non-NULL for
4324 scalars who were upgraded and compiled in the
4325 else block below. In turn, this is only
4326 triggered in the "postponed utf8 string" tests
4332 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4336 rx = reg_temp_copy(NULL, rx);
4340 const I32 osize = PL_regsize;
4343 assert (SvUTF8(ret));
4344 } else if (SvUTF8(ret)) {
4345 /* Not doing UTF-8, despite what the SV says. Is
4346 this only if we're trapped in use 'bytes'? */
4347 /* Make a copy of the octet sequence, but without
4348 the flag on, as the compiler now honours the
4349 SvUTF8 flag on ret. */
4351 const char *const p = SvPV(ret, len);
4352 ret = newSVpvn_flags(p, len, SVs_TEMP);
4354 rx = CALLREGCOMP(ret, pm_flags);
4356 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4358 /* This isn't a first class regexp. Instead, it's
4359 caching a regexp onto an existing, Perl visible
4361 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4366 re = (struct regexp *)SvANY(rx);
4368 RXp_MATCH_COPIED_off(re);
4369 re->subbeg = rex->subbeg;
4370 re->sublen = rex->sublen;
4373 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4374 "Matching embedded");
4376 startpoint = rei->program + 1;
4377 ST.close_paren = 0; /* only used for GOSUB */
4378 /* borrowed from regtry */
4379 if (PL_reg_start_tmpl <= re->nparens) {
4380 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4381 if(PL_reg_start_tmp)
4382 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4384 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4387 eval_recurse_doit: /* Share code with GOSUB below this line */
4388 /* run the pattern returned from (??{...}) */
4389 ST.cp = regcppush(0); /* Save *all* the positions. */
4390 REGCP_SET(ST.lastcp);
4392 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4394 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4395 PL_reglastparen = &re->lastparen;
4396 PL_reglastcloseparen = &re->lastcloseparen;
4398 re->lastcloseparen = 0;
4400 PL_reginput = locinput;
4403 /* XXXX This is too dramatic a measure... */
4406 ST.toggle_reg_flags = PL_reg_flags;
4408 PL_reg_flags |= RF_utf8;
4410 PL_reg_flags &= ~RF_utf8;
4411 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4413 ST.prev_rex = rex_sv;
4414 ST.prev_curlyx = cur_curlyx;
4415 SETREX(rex_sv,re_sv);
4420 ST.prev_eval = cur_eval;
4422 /* now continue from first node in postoned RE */
4423 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4426 /* logical is 1, /(?(?{...})X|Y)/ */
4427 sw = cBOOL(SvTRUE(ret));
4432 case EVAL_AB: /* cleanup after a successful (??{A})B */
4433 /* note: this is called twice; first after popping B, then A */
4434 PL_reg_flags ^= ST.toggle_reg_flags;
4435 ReREFCNT_dec(rex_sv);
4436 SETREX(rex_sv,ST.prev_rex);
4437 rex = (struct regexp *)SvANY(rex_sv);
4438 rexi = RXi_GET(rex);
4440 cur_eval = ST.prev_eval;
4441 cur_curlyx = ST.prev_curlyx;
4443 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4444 PL_reglastparen = &rex->lastparen;
4445 PL_reglastcloseparen = &rex->lastcloseparen;
4446 /* also update PL_regoffs */
4447 PL_regoffs = rex->offs;
4449 /* XXXX This is too dramatic a measure... */
4451 if ( nochange_depth )
4456 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4457 /* note: this is called twice; first after popping B, then A */
4458 PL_reg_flags ^= ST.toggle_reg_flags;
4459 ReREFCNT_dec(rex_sv);
4460 SETREX(rex_sv,ST.prev_rex);
4461 rex = (struct regexp *)SvANY(rex_sv);
4462 rexi = RXi_GET(rex);
4463 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4464 PL_reglastparen = &rex->lastparen;
4465 PL_reglastcloseparen = &rex->lastcloseparen;
4467 PL_reginput = locinput;
4468 REGCP_UNWIND(ST.lastcp);
4470 cur_eval = ST.prev_eval;
4471 cur_curlyx = ST.prev_curlyx;
4472 /* XXXX This is too dramatic a measure... */
4474 if ( nochange_depth )
4480 n = ARG(scan); /* which paren pair */
4481 PL_reg_start_tmp[n] = locinput;
4487 n = ARG(scan); /* which paren pair */
4488 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4489 PL_regoffs[n].end = locinput - PL_bostr;
4490 /*if (n > PL_regsize)
4492 if (n > *PL_reglastparen)
4493 *PL_reglastparen = n;
4494 *PL_reglastcloseparen = n;
4495 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4503 cursor && OP(cursor)!=END;
4504 cursor=regnext(cursor))
4506 if ( OP(cursor)==CLOSE ){
4508 if ( n <= lastopen ) {
4510 = PL_reg_start_tmp[n] - PL_bostr;
4511 PL_regoffs[n].end = locinput - PL_bostr;
4512 /*if (n > PL_regsize)
4514 if (n > *PL_reglastparen)
4515 *PL_reglastparen = n;
4516 *PL_reglastcloseparen = n;
4517 if ( n == ARG(scan) || (cur_eval &&
4518 cur_eval->u.eval.close_paren == n))
4527 n = ARG(scan); /* which paren pair */
4528 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4531 /* reg_check_named_buff_matched returns 0 for no match */
4532 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4536 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4542 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4544 next = NEXTOPER(NEXTOPER(scan));
4546 next = scan + ARG(scan);
4547 if (OP(next) == IFTHEN) /* Fake one. */
4548 next = NEXTOPER(NEXTOPER(next));
4552 logical = scan->flags;
4555 /*******************************************************************
4557 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4558 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4559 STAR/PLUS/CURLY/CURLYN are used instead.)
4561 A*B is compiled as <CURLYX><A><WHILEM><B>
4563 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4564 state, which contains the current count, initialised to -1. It also sets
4565 cur_curlyx to point to this state, with any previous value saved in the
4568 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4569 since the pattern may possibly match zero times (i.e. it's a while {} loop
4570 rather than a do {} while loop).
4572 Each entry to WHILEM represents a successful match of A. The count in the
4573 CURLYX block is incremented, another WHILEM state is pushed, and execution
4574 passes to A or B depending on greediness and the current count.
4576 For example, if matching against the string a1a2a3b (where the aN are
4577 substrings that match /A/), then the match progresses as follows: (the
4578 pushed states are interspersed with the bits of strings matched so far):
4581 <CURLYX cnt=0><WHILEM>
4582 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4583 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4584 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4585 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4587 (Contrast this with something like CURLYM, which maintains only a single
4591 a1 <CURLYM cnt=1> a2
4592 a1 a2 <CURLYM cnt=2> a3
4593 a1 a2 a3 <CURLYM cnt=3> b
4596 Each WHILEM state block marks a point to backtrack to upon partial failure
4597 of A or B, and also contains some minor state data related to that
4598 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4599 overall state, such as the count, and pointers to the A and B ops.
4601 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4602 must always point to the *current* CURLYX block, the rules are:
4604 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4605 and set cur_curlyx to point the new block.
4607 When popping the CURLYX block after a successful or unsuccessful match,
4608 restore the previous cur_curlyx.
4610 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4611 to the outer one saved in the CURLYX block.
4613 When popping the WHILEM block after a successful or unsuccessful B match,
4614 restore the previous cur_curlyx.
4616 Here's an example for the pattern (AI* BI)*BO
4617 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4620 curlyx backtrack stack
4621 ------ ---------------
4623 CO <CO prev=NULL> <WO>
4624 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4625 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4626 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4628 At this point the pattern succeeds, and we work back down the stack to
4629 clean up, restoring as we go:
4631 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4632 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4633 CO <CO prev=NULL> <WO>
4636 *******************************************************************/
4638 #define ST st->u.curlyx
4640 case CURLYX: /* start of /A*B/ (for complex A) */
4642 /* No need to save/restore up to this paren */
4643 I32 parenfloor = scan->flags;
4645 assert(next); /* keep Coverity happy */
4646 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4649 /* XXXX Probably it is better to teach regpush to support
4650 parenfloor > PL_regsize... */
4651 if (parenfloor > (I32)*PL_reglastparen)
4652 parenfloor = *PL_reglastparen; /* Pessimization... */
4654 ST.prev_curlyx= cur_curlyx;
4656 ST.cp = PL_savestack_ix;
4658 /* these fields contain the state of the current curly.
4659 * they are accessed by subsequent WHILEMs */
4660 ST.parenfloor = parenfloor;
4665 ST.count = -1; /* this will be updated by WHILEM */
4666 ST.lastloc = NULL; /* this will be updated by WHILEM */
4668 PL_reginput = locinput;
4669 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4673 case CURLYX_end: /* just finished matching all of A*B */
4674 cur_curlyx = ST.prev_curlyx;
4678 case CURLYX_end_fail: /* just failed to match all of A*B */
4680 cur_curlyx = ST.prev_curlyx;
4686 #define ST st->u.whilem
4688 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4690 /* see the discussion above about CURLYX/WHILEM */
4692 int min = ARG1(cur_curlyx->u.curlyx.me);
4693 int max = ARG2(cur_curlyx->u.curlyx.me);
4694 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4696 assert(cur_curlyx); /* keep Coverity happy */
4697 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4698 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4699 ST.cache_offset = 0;
4702 PL_reginput = locinput;
4704 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4705 "%*s whilem: matched %ld out of %d..%d\n",
4706 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4709 /* First just match a string of min A's. */
4712 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4713 cur_curlyx->u.curlyx.lastloc = locinput;
4714 REGCP_SET(ST.lastcp);
4716 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4720 /* If degenerate A matches "", assume A done. */
4722 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4723 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4724 "%*s whilem: empty match detected, trying continuation...\n",
4725 REPORT_CODE_OFF+depth*2, "")
4727 goto do_whilem_B_max;
4730 /* super-linear cache processing */
4734 if (!PL_reg_maxiter) {
4735 /* start the countdown: Postpone detection until we
4736 * know the match is not *that* much linear. */
4737 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4738 /* possible overflow for long strings and many CURLYX's */
4739 if (PL_reg_maxiter < 0)
4740 PL_reg_maxiter = I32_MAX;
4741 PL_reg_leftiter = PL_reg_maxiter;
4744 if (PL_reg_leftiter-- == 0) {
4745 /* initialise cache */
4746 const I32 size = (PL_reg_maxiter + 7)/8;
4747 if (PL_reg_poscache) {
4748 if ((I32)PL_reg_poscache_size < size) {
4749 Renew(PL_reg_poscache, size, char);
4750 PL_reg_poscache_size = size;
4752 Zero(PL_reg_poscache, size, char);
4755 PL_reg_poscache_size = size;
4756 Newxz(PL_reg_poscache, size, char);
4758 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4759 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4760 PL_colors[4], PL_colors[5])
4764 if (PL_reg_leftiter < 0) {
4765 /* have we already failed at this position? */
4767 offset = (scan->flags & 0xf) - 1
4768 + (locinput - PL_bostr) * (scan->flags>>4);
4769 mask = 1 << (offset % 8);
4771 if (PL_reg_poscache[offset] & mask) {
4772 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4773 "%*s whilem: (cache) already tried at this position...\n",
4774 REPORT_CODE_OFF+depth*2, "")
4776 sayNO; /* cache records failure */
4778 ST.cache_offset = offset;
4779 ST.cache_mask = mask;
4783 /* Prefer B over A for minimal matching. */
4785 if (cur_curlyx->u.curlyx.minmod) {
4786 ST.save_curlyx = cur_curlyx;
4787 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4788 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4789 REGCP_SET(ST.lastcp);
4790 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4794 /* Prefer A over B for maximal matching. */
4796 if (n < max) { /* More greed allowed? */
4797 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4798 cur_curlyx->u.curlyx.lastloc = locinput;
4799 REGCP_SET(ST.lastcp);
4800 PUSH_STATE_GOTO(WHILEM_A_max, A);
4803 goto do_whilem_B_max;
4807 case WHILEM_B_min: /* just matched B in a minimal match */
4808 case WHILEM_B_max: /* just matched B in a maximal match */
4809 cur_curlyx = ST.save_curlyx;
4813 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4814 cur_curlyx = ST.save_curlyx;
4815 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4816 cur_curlyx->u.curlyx.count--;
4820 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4822 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4823 REGCP_UNWIND(ST.lastcp);
4825 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4826 cur_curlyx->u.curlyx.count--;
4830 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4831 REGCP_UNWIND(ST.lastcp);
4832 regcppop(rex); /* Restore some previous $<digit>s? */
4833 PL_reginput = locinput;
4834 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4835 "%*s whilem: failed, trying continuation...\n",
4836 REPORT_CODE_OFF+depth*2, "")
4839 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4840 && ckWARN(WARN_REGEXP)
4841 && !(PL_reg_flags & RF_warned))
4843 PL_reg_flags |= RF_warned;
4844 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4845 "Complex regular subexpression recursion",
4850 ST.save_curlyx = cur_curlyx;
4851 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4852 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4855 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4856 cur_curlyx = ST.save_curlyx;
4857 REGCP_UNWIND(ST.lastcp);
4860 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4861 /* Maximum greed exceeded */
4862 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4863 && ckWARN(WARN_REGEXP)
4864 && !(PL_reg_flags & RF_warned))
4866 PL_reg_flags |= RF_warned;
4867 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4868 "%s limit (%d) exceeded",
4869 "Complex regular subexpression recursion",
4872 cur_curlyx->u.curlyx.count--;
4876 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4877 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4879 /* Try grabbing another A and see if it helps. */
4880 PL_reginput = locinput;
4881 cur_curlyx->u.curlyx.lastloc = locinput;
4882 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4883 REGCP_SET(ST.lastcp);
4884 PUSH_STATE_GOTO(WHILEM_A_min,
4885 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4889 #define ST st->u.branch
4891 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4892 next = scan + ARG(scan);
4895 scan = NEXTOPER(scan);
4898 case BRANCH: /* /(...|A|...)/ */
4899 scan = NEXTOPER(scan); /* scan now points to inner node */
4900 ST.lastparen = *PL_reglastparen;
4901 ST.next_branch = next;
4903 PL_reginput = locinput;
4905 /* Now go into the branch */
4907 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4909 PUSH_STATE_GOTO(BRANCH_next, scan);
4913 PL_reginput = locinput;
4914 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4915 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4916 PUSH_STATE_GOTO(CUTGROUP_next,next);
4918 case CUTGROUP_next_fail:
4921 if (st->u.mark.mark_name)
4922 sv_commit = st->u.mark.mark_name;
4928 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4933 REGCP_UNWIND(ST.cp);
4934 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4935 PL_regoffs[n].end = -1;
4936 *PL_reglastparen = n;
4937 /*dmq: *PL_reglastcloseparen = n; */
4938 scan = ST.next_branch;
4939 /* no more branches? */
4940 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4942 PerlIO_printf( Perl_debug_log,
4943 "%*s %sBRANCH failed...%s\n",
4944 REPORT_CODE_OFF+depth*2, "",
4950 continue; /* execute next BRANCH[J] op */
4958 #define ST st->u.curlym
4960 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4962 /* This is an optimisation of CURLYX that enables us to push
4963 * only a single backtracking state, no matter how many matches
4964 * there are in {m,n}. It relies on the pattern being constant
4965 * length, with no parens to influence future backrefs
4969 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4971 /* if paren positive, emulate an OPEN/CLOSE around A */
4973 U32 paren = ST.me->flags;
4974 if (paren > PL_regsize)
4976 if (paren > *PL_reglastparen)
4977 *PL_reglastparen = paren;
4978 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4986 ST.c1 = CHRTEST_UNINIT;
4989 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4992 curlym_do_A: /* execute the A in /A{m,n}B/ */
4993 PL_reginput = locinput;
4994 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4997 case CURLYM_A: /* we've just matched an A */
4998 locinput = st->locinput;
4999 nextchr = UCHARAT(locinput);
5002 /* after first match, determine A's length: u.curlym.alen */
5003 if (ST.count == 1) {
5004 if (PL_reg_match_utf8) {
5006 while (s < PL_reginput) {
5012 ST.alen = PL_reginput - locinput;
5015 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5018 PerlIO_printf(Perl_debug_log,
5019 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5020 (int)(REPORT_CODE_OFF+(depth*2)), "",
5021 (IV) ST.count, (IV)ST.alen)
5024 locinput = PL_reginput;
5026 if (cur_eval && cur_eval->u.eval.close_paren &&
5027 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5031 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5032 if ( max == REG_INFTY || ST.count < max )
5033 goto curlym_do_A; /* try to match another A */
5035 goto curlym_do_B; /* try to match B */
5037 case CURLYM_A_fail: /* just failed to match an A */
5038 REGCP_UNWIND(ST.cp);
5040 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5041 || (cur_eval && cur_eval->u.eval.close_paren &&
5042 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5045 curlym_do_B: /* execute the B in /A{m,n}B/ */
5046 PL_reginput = locinput;
5047 if (ST.c1 == CHRTEST_UNINIT) {
5048 /* calculate c1 and c2 for possible match of 1st char
5049 * following curly */
5050 ST.c1 = ST.c2 = CHRTEST_VOID;
5051 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5052 regnode *text_node = ST.B;
5053 if (! HAS_TEXT(text_node))
5054 FIND_NEXT_IMPT(text_node);
5057 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5059 But the former is redundant in light of the latter.
5061 if this changes back then the macro for
5062 IS_TEXT and friends need to change.
5064 if (PL_regkind[OP(text_node)] == EXACT)
5067 ST.c1 = (U8)*STRING(text_node);
5068 switch (OP(text_node)) {
5069 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5071 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5072 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5073 default: ST.c2 = ST.c1;
5080 PerlIO_printf(Perl_debug_log,
5081 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5082 (int)(REPORT_CODE_OFF+(depth*2)),
5085 if (ST.c1 != CHRTEST_VOID
5086 && UCHARAT(PL_reginput) != ST.c1
5087 && UCHARAT(PL_reginput) != ST.c2)
5089 /* simulate B failing */
5091 PerlIO_printf(Perl_debug_log,
5092 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5093 (int)(REPORT_CODE_OFF+(depth*2)),"",
5096 state_num = CURLYM_B_fail;
5097 goto reenter_switch;
5101 /* mark current A as captured */
5102 I32 paren = ST.me->flags;
5104 PL_regoffs[paren].start
5105 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5106 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5107 /*dmq: *PL_reglastcloseparen = paren; */
5110 PL_regoffs[paren].end = -1;
5111 if (cur_eval && cur_eval->u.eval.close_paren &&
5112 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5121 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5124 case CURLYM_B_fail: /* just failed to match a B */
5125 REGCP_UNWIND(ST.cp);
5127 I32 max = ARG2(ST.me);
5128 if (max != REG_INFTY && ST.count == max)
5130 goto curlym_do_A; /* try to match a further A */
5132 /* backtrack one A */
5133 if (ST.count == ARG1(ST.me) /* min */)
5136 locinput = HOPc(locinput, -ST.alen);
5137 goto curlym_do_B; /* try to match B */
5140 #define ST st->u.curly
5142 #define CURLY_SETPAREN(paren, success) \
5145 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5146 PL_regoffs[paren].end = locinput - PL_bostr; \
5147 *PL_reglastcloseparen = paren; \
5150 PL_regoffs[paren].end = -1; \
5153 case STAR: /* /A*B/ where A is width 1 */
5157 scan = NEXTOPER(scan);
5159 case PLUS: /* /A+B/ where A is width 1 */
5163 scan = NEXTOPER(scan);
5165 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5166 ST.paren = scan->flags; /* Which paren to set */
5167 if (ST.paren > PL_regsize)
5168 PL_regsize = ST.paren;
5169 if (ST.paren > *PL_reglastparen)
5170 *PL_reglastparen = ST.paren;
5171 ST.min = ARG1(scan); /* min to match */
5172 ST.max = ARG2(scan); /* max to match */
5173 if (cur_eval && cur_eval->u.eval.close_paren &&
5174 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5178 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5180 case CURLY: /* /A{m,n}B/ where A is width 1 */
5182 ST.min = ARG1(scan); /* min to match */
5183 ST.max = ARG2(scan); /* max to match */
5184 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5187 * Lookahead to avoid useless match attempts
5188 * when we know what character comes next.
5190 * Used to only do .*x and .*?x, but now it allows
5191 * for )'s, ('s and (?{ ... })'s to be in the way
5192 * of the quantifier and the EXACT-like node. -- japhy
5195 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5197 if (HAS_TEXT(next) || JUMPABLE(next)) {
5199 regnode *text_node = next;
5201 if (! HAS_TEXT(text_node))
5202 FIND_NEXT_IMPT(text_node);
5204 if (! HAS_TEXT(text_node))
5205 ST.c1 = ST.c2 = CHRTEST_VOID;
5207 if ( PL_regkind[OP(text_node)] != EXACT ) {
5208 ST.c1 = ST.c2 = CHRTEST_VOID;
5209 goto assume_ok_easy;
5212 s = (U8*)STRING(text_node);
5214 /* Currently we only get here when
5216 PL_rekind[OP(text_node)] == EXACT
5218 if this changes back then the macro for IS_TEXT and
5219 friends need to change. */
5222 switch (OP(text_node)) {
5223 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5225 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5226 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5227 default: ST.c2 = ST.c1; break;
5230 else { /* UTF_PATTERN */
5231 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5232 STRLEN ulen1, ulen2;
5233 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5234 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5236 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5237 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5239 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5241 0 : UTF8_ALLOW_ANY);
5242 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5244 0 : UTF8_ALLOW_ANY);
5246 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5248 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5253 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5260 ST.c1 = ST.c2 = CHRTEST_VOID;
5265 PL_reginput = locinput;
5268 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5271 locinput = PL_reginput;
5273 if (ST.c1 == CHRTEST_VOID)
5274 goto curly_try_B_min;
5276 ST.oldloc = locinput;
5278 /* set ST.maxpos to the furthest point along the
5279 * string that could possibly match */
5280 if (ST.max == REG_INFTY) {
5281 ST.maxpos = PL_regeol - 1;
5283 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5286 else if (utf8_target) {
5287 int m = ST.max - ST.min;
5288 for (ST.maxpos = locinput;
5289 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5290 ST.maxpos += UTF8SKIP(ST.maxpos);
5293 ST.maxpos = locinput + ST.max - ST.min;
5294 if (ST.maxpos >= PL_regeol)
5295 ST.maxpos = PL_regeol - 1;
5297 goto curly_try_B_min_known;
5301 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5302 locinput = PL_reginput;
5303 if (ST.count < ST.min)
5305 if ((ST.count > ST.min)
5306 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5308 /* A{m,n} must come at the end of the string, there's
5309 * no point in backing off ... */
5311 /* ...except that $ and \Z can match before *and* after
5312 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5313 We may back off by one in this case. */
5314 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5318 goto curly_try_B_max;
5323 case CURLY_B_min_known_fail:
5324 /* failed to find B in a non-greedy match where c1,c2 valid */
5325 if (ST.paren && ST.count)
5326 PL_regoffs[ST.paren].end = -1;
5328 PL_reginput = locinput; /* Could be reset... */
5329 REGCP_UNWIND(ST.cp);
5330 /* Couldn't or didn't -- move forward. */
5331 ST.oldloc = locinput;
5333 locinput += UTF8SKIP(locinput);
5337 curly_try_B_min_known:
5338 /* find the next place where 'B' could work, then call B */
5342 n = (ST.oldloc == locinput) ? 0 : 1;
5343 if (ST.c1 == ST.c2) {
5345 /* set n to utf8_distance(oldloc, locinput) */
5346 while (locinput <= ST.maxpos &&
5347 utf8n_to_uvchr((U8*)locinput,
5348 UTF8_MAXBYTES, &len,
5349 uniflags) != (UV)ST.c1) {
5355 /* set n to utf8_distance(oldloc, locinput) */
5356 while (locinput <= ST.maxpos) {
5358 const UV c = utf8n_to_uvchr((U8*)locinput,
5359 UTF8_MAXBYTES, &len,
5361 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5369 if (ST.c1 == ST.c2) {
5370 while (locinput <= ST.maxpos &&
5371 UCHARAT(locinput) != ST.c1)
5375 while (locinput <= ST.maxpos
5376 && UCHARAT(locinput) != ST.c1
5377 && UCHARAT(locinput) != ST.c2)
5380 n = locinput - ST.oldloc;
5382 if (locinput > ST.maxpos)
5384 /* PL_reginput == oldloc now */
5387 if (regrepeat(rex, ST.A, n, depth) < n)
5390 PL_reginput = locinput;
5391 CURLY_SETPAREN(ST.paren, ST.count);
5392 if (cur_eval && cur_eval->u.eval.close_paren &&
5393 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5396 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5401 case CURLY_B_min_fail:
5402 /* failed to find B in a non-greedy match where c1,c2 invalid */
5403 if (ST.paren && ST.count)
5404 PL_regoffs[ST.paren].end = -1;
5406 REGCP_UNWIND(ST.cp);
5407 /* failed -- move forward one */
5408 PL_reginput = locinput;
5409 if (regrepeat(rex, ST.A, 1, depth)) {
5411 locinput = PL_reginput;
5412 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5413 ST.count > 0)) /* count overflow ? */
5416 CURLY_SETPAREN(ST.paren, ST.count);
5417 if (cur_eval && cur_eval->u.eval.close_paren &&
5418 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5421 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5429 /* a successful greedy match: now try to match B */
5430 if (cur_eval && cur_eval->u.eval.close_paren &&
5431 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5436 if (ST.c1 != CHRTEST_VOID)
5437 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5438 UTF8_MAXBYTES, 0, uniflags)
5439 : (UV) UCHARAT(PL_reginput);
5440 /* If it could work, try it. */
5441 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5442 CURLY_SETPAREN(ST.paren, ST.count);
5443 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5448 case CURLY_B_max_fail:
5449 /* failed to find B in a greedy match */
5450 if (ST.paren && ST.count)
5451 PL_regoffs[ST.paren].end = -1;
5453 REGCP_UNWIND(ST.cp);
5455 if (--ST.count < ST.min)
5457 PL_reginput = locinput = HOPc(locinput, -1);
5458 goto curly_try_B_max;
5465 /* we've just finished A in /(??{A})B/; now continue with B */
5467 st->u.eval.toggle_reg_flags
5468 = cur_eval->u.eval.toggle_reg_flags;
5469 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5471 st->u.eval.prev_rex = rex_sv; /* inner */
5472 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5473 rex = (struct regexp *)SvANY(rex_sv);
5474 rexi = RXi_GET(rex);
5475 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5476 (void)ReREFCNT_inc(rex_sv);
5477 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5479 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5480 PL_reglastparen = &rex->lastparen;
5481 PL_reglastcloseparen = &rex->lastcloseparen;
5483 REGCP_SET(st->u.eval.lastcp);
5484 PL_reginput = locinput;
5486 /* Restore parens of the outer rex without popping the
5488 tmpix = PL_savestack_ix;
5489 PL_savestack_ix = cur_eval->u.eval.lastcp;
5491 PL_savestack_ix = tmpix;
5493 st->u.eval.prev_eval = cur_eval;
5494 cur_eval = cur_eval->u.eval.prev_eval;
5496 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5497 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5498 if ( nochange_depth )
5501 PUSH_YES_STATE_GOTO(EVAL_AB,
5502 st->u.eval.prev_eval->u.eval.B); /* match B */
5505 if (locinput < reginfo->till) {
5506 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5507 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5509 (long)(locinput - PL_reg_starttry),
5510 (long)(reginfo->till - PL_reg_starttry),
5513 sayNO_SILENT; /* Cannot match: too short. */
5515 PL_reginput = locinput; /* put where regtry can find it */
5516 sayYES; /* Success! */
5518 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5520 PerlIO_printf(Perl_debug_log,
5521 "%*s %ssubpattern success...%s\n",
5522 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5523 PL_reginput = locinput; /* put where regtry can find it */
5524 sayYES; /* Success! */
5527 #define ST st->u.ifmatch
5529 case SUSPEND: /* (?>A) */
5531 PL_reginput = locinput;
5534 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5536 goto ifmatch_trivial_fail_test;
5538 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5540 ifmatch_trivial_fail_test:
5542 char * const s = HOPBACKc(locinput, scan->flags);
5547 sw = 1 - cBOOL(ST.wanted);
5551 next = scan + ARG(scan);
5559 PL_reginput = locinput;
5563 ST.logical = logical;
5564 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5566 /* execute body of (?...A) */
5567 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5570 case IFMATCH_A_fail: /* body of (?...A) failed */
5571 ST.wanted = !ST.wanted;
5574 case IFMATCH_A: /* body of (?...A) succeeded */
5576 sw = cBOOL(ST.wanted);
5578 else if (!ST.wanted)
5581 if (OP(ST.me) == SUSPEND)
5582 locinput = PL_reginput;
5584 locinput = PL_reginput = st->locinput;
5585 nextchr = UCHARAT(locinput);
5587 scan = ST.me + ARG(ST.me);
5590 continue; /* execute B */
5595 next = scan + ARG(scan);
5600 reginfo->cutpoint = PL_regeol;
5603 PL_reginput = locinput;
5605 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5606 PUSH_STATE_GOTO(COMMIT_next,next);
5608 case COMMIT_next_fail:
5615 #define ST st->u.mark
5617 ST.prev_mark = mark_state;
5618 ST.mark_name = sv_commit = sv_yes_mark
5619 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5621 ST.mark_loc = PL_reginput = locinput;
5622 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5624 case MARKPOINT_next:
5625 mark_state = ST.prev_mark;
5628 case MARKPOINT_next_fail:
5629 if (popmark && sv_eq(ST.mark_name,popmark))
5631 if (ST.mark_loc > startpoint)
5632 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5633 popmark = NULL; /* we found our mark */
5634 sv_commit = ST.mark_name;
5637 PerlIO_printf(Perl_debug_log,
5638 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5639 REPORT_CODE_OFF+depth*2, "",
5640 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5643 mark_state = ST.prev_mark;
5644 sv_yes_mark = mark_state ?
5645 mark_state->u.mark.mark_name : NULL;
5649 PL_reginput = locinput;
5651 /* (*SKIP) : if we fail we cut here*/
5652 ST.mark_name = NULL;
5653 ST.mark_loc = locinput;
5654 PUSH_STATE_GOTO(SKIP_next,next);
5656 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5657 otherwise do nothing. Meaning we need to scan
5659 regmatch_state *cur = mark_state;
5660 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5663 if ( sv_eq( cur->u.mark.mark_name,
5666 ST.mark_name = find;
5667 PUSH_STATE_GOTO( SKIP_next, next );
5669 cur = cur->u.mark.prev_mark;
5672 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5674 case SKIP_next_fail:
5676 /* (*CUT:NAME) - Set up to search for the name as we
5677 collapse the stack*/
5678 popmark = ST.mark_name;
5680 /* (*CUT) - No name, we cut here.*/
5681 if (ST.mark_loc > startpoint)
5682 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5683 /* but we set sv_commit to latest mark_name if there
5684 is one so they can test to see how things lead to this
5687 sv_commit=mark_state->u.mark.mark_name;
5695 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5697 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5700 U8 folded[UTF8_MAXBYTES_CASE+1];
5702 const char * const l = locinput;
5703 char *e = PL_regeol;
5704 to_uni_fold(n, folded, &foldlen);
5706 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5707 l, &e, 0, utf8_target)) {
5712 nextchr = UCHARAT(locinput);
5715 if ((n=is_LNBREAK(locinput,utf8_target))) {
5717 nextchr = UCHARAT(locinput);
5722 #define CASE_CLASS(nAmE) \
5724 if (locinput >= PL_regeol) \
5726 if ((n=is_##nAmE(locinput,utf8_target))) { \
5728 nextchr = UCHARAT(locinput); \
5733 if (locinput >= PL_regeol) \
5735 if ((n=is_##nAmE(locinput,utf8_target))) { \
5738 locinput += UTF8SKIP(locinput); \
5739 nextchr = UCHARAT(locinput); \
5744 CASE_CLASS(HORIZWS);
5748 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5749 PTR2UV(scan), OP(scan));
5750 Perl_croak(aTHX_ "regexp memory corruption");
5754 /* switch break jumps here */
5755 scan = next; /* prepare to execute the next op and ... */
5756 continue; /* ... jump back to the top, reusing st */
5760 /* push a state that backtracks on success */
5761 st->u.yes.prev_yes_state = yes_state;
5765 /* push a new regex state, then continue at scan */
5767 regmatch_state *newst;
5770 regmatch_state *cur = st;
5771 regmatch_state *curyes = yes_state;
5773 regmatch_slab *slab = PL_regmatch_slab;
5774 for (;curd > -1;cur--,curd--) {
5775 if (cur < SLAB_FIRST(slab)) {
5777 cur = SLAB_LAST(slab);
5779 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5780 REPORT_CODE_OFF + 2 + depth * 2,"",
5781 curd, PL_reg_name[cur->resume_state],
5782 (curyes == cur) ? "yes" : ""
5785 curyes = cur->u.yes.prev_yes_state;
5788 DEBUG_STATE_pp("push")
5791 st->locinput = locinput;
5793 if (newst > SLAB_LAST(PL_regmatch_slab))
5794 newst = S_push_slab(aTHX);
5795 PL_regmatch_state = newst;
5797 locinput = PL_reginput;
5798 nextchr = UCHARAT(locinput);
5806 * We get here only if there's trouble -- normally "case END" is
5807 * the terminating point.
5809 Perl_croak(aTHX_ "corrupted regexp pointers");
5815 /* we have successfully completed a subexpression, but we must now
5816 * pop to the state marked by yes_state and continue from there */
5817 assert(st != yes_state);
5819 while (st != yes_state) {
5821 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5822 PL_regmatch_slab = PL_regmatch_slab->prev;
5823 st = SLAB_LAST(PL_regmatch_slab);
5827 DEBUG_STATE_pp("pop (no final)");
5829 DEBUG_STATE_pp("pop (yes)");
5835 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5836 || yes_state > SLAB_LAST(PL_regmatch_slab))
5838 /* not in this slab, pop slab */
5839 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5840 PL_regmatch_slab = PL_regmatch_slab->prev;
5841 st = SLAB_LAST(PL_regmatch_slab);
5843 depth -= (st - yes_state);
5846 yes_state = st->u.yes.prev_yes_state;
5847 PL_regmatch_state = st;
5850 locinput= st->locinput;
5851 nextchr = UCHARAT(locinput);
5853 state_num = st->resume_state + no_final;
5854 goto reenter_switch;
5857 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5858 PL_colors[4], PL_colors[5]));
5860 if (PL_reg_eval_set) {
5861 /* each successfully executed (?{...}) block does the equivalent of
5862 * local $^R = do {...}
5863 * When popping the save stack, all these locals would be undone;
5864 * bypass this by setting the outermost saved $^R to the latest
5866 if (oreplsv != GvSV(PL_replgv))
5867 sv_setsv(oreplsv, GvSV(PL_replgv));
5874 PerlIO_printf(Perl_debug_log,
5875 "%*s %sfailed...%s\n",
5876 REPORT_CODE_OFF+depth*2, "",
5877 PL_colors[4], PL_colors[5])
5889 /* there's a previous state to backtrack to */
5891 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5892 PL_regmatch_slab = PL_regmatch_slab->prev;
5893 st = SLAB_LAST(PL_regmatch_slab);
5895 PL_regmatch_state = st;
5896 locinput= st->locinput;
5897 nextchr = UCHARAT(locinput);
5899 DEBUG_STATE_pp("pop");
5901 if (yes_state == st)
5902 yes_state = st->u.yes.prev_yes_state;
5904 state_num = st->resume_state + 1; /* failure = success + 1 */
5905 goto reenter_switch;
5910 if (rex->intflags & PREGf_VERBARG_SEEN) {
5911 SV *sv_err = get_sv("REGERROR", 1);
5912 SV *sv_mrk = get_sv("REGMARK", 1);
5914 sv_commit = &PL_sv_no;
5916 sv_yes_mark = &PL_sv_yes;
5919 sv_commit = &PL_sv_yes;
5920 sv_yes_mark = &PL_sv_no;
5922 sv_setsv(sv_err, sv_commit);
5923 sv_setsv(sv_mrk, sv_yes_mark);
5926 /* clean up; in particular, free all slabs above current one */
5927 LEAVE_SCOPE(oldsave);
5933 - regrepeat - repeatedly match something simple, report how many
5936 * [This routine now assumes that it will only match on things of length 1.
5937 * That was true before, but now we assume scan - reginput is the count,
5938 * rather than incrementing count on every character. [Er, except utf8.]]
5941 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5944 register char *scan;
5946 register char *loceol = PL_regeol;
5947 register I32 hardcount = 0;
5948 register bool utf8_target = PL_reg_match_utf8;
5951 PERL_UNUSED_ARG(depth);
5954 PERL_ARGS_ASSERT_REGREPEAT;
5957 if (max == REG_INFTY)
5959 else if (max < loceol - scan)
5960 loceol = scan + max;
5965 while (scan < loceol && hardcount < max && *scan != '\n') {
5966 scan += UTF8SKIP(scan);
5970 while (scan < loceol && *scan != '\n')
5977 while (scan < loceol && hardcount < max) {
5978 scan += UTF8SKIP(scan);
5989 /* To get here, EXACTish nodes must have *byte* length == 1. That
5990 * means they match only characters in the string that can be expressed
5991 * as a single byte. For non-utf8 strings, that means a simple match.
5992 * For utf8 strings, the character matched must be an invariant, or
5993 * downgradable to a single byte. The pattern's utf8ness is
5994 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5995 * it is, it's an invariant */
5998 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6000 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6001 while (scan < loceol && UCHARAT(scan) == c) {
6007 /* Here, the string is utf8, and the pattern char is different
6008 * in utf8 than not, so can't compare them directly. Outside the
6009 * loop, find find the two utf8 bytes that represent c, and then
6010 * look for those in sequence in the utf8 string */
6011 U8 high = UTF8_TWO_BYTE_HI(c);
6012 U8 low = UTF8_TWO_BYTE_LO(c);
6015 while (hardcount < max
6016 && scan + 1 < loceol
6017 && UCHARAT(scan) == high
6018 && UCHARAT(scan + 1) == low)
6026 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6030 PL_reg_flags |= RF_tainted;
6031 utf8_flags = FOLDEQ_UTF8_LOCALE;
6036 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6038 /* The comments for the EXACT case above apply as well to these fold
6043 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6045 if (utf8_target) { /* Use full Unicode fold matching */
6046 char *tmpeol = loceol;
6047 while (hardcount < max
6048 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6049 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6056 /* XXX Note that the above handles properly the German sharp s in
6057 * the pattern matching ss in the string. But it doesn't handle
6058 * properly cases where the string contains say 'LIGATURE ff' and
6059 * the pattern is 'f+'. This would require, say, a new function or
6060 * revised interface to foldEQ_utf8(), in which the maximum number
6061 * of characters to match could be passed and it would return how
6062 * many actually did. This is just one of many cases where
6063 * multi-char folds don't work properly, and so the fix is being
6069 /* Here, the string isn't utf8 and c is a single byte; and either
6070 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6071 * doesn't affect c. Can just do simple comparisons for exact or
6074 case EXACTF: folded = PL_fold[c]; break;
6076 case EXACTFU: folded = PL_fold_latin1[c]; break;
6077 case EXACTFL: folded = PL_fold_locale[c]; break;
6078 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6080 while (scan < loceol &&
6081 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6089 if (utf8_target || OP(p) == ANYOFV) {
6092 inclasslen = loceol - scan;
6093 while (hardcount < max
6094 && ((inclasslen = loceol - scan) > 0)
6095 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6101 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6109 LOAD_UTF8_CHARCLASS_ALNUM();
6110 while (hardcount < max && scan < loceol &&
6111 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6113 scan += UTF8SKIP(scan);
6117 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6125 while (scan < loceol && isALNUM((U8) *scan)) {
6130 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6135 PL_reg_flags |= RF_tainted;
6138 while (hardcount < max && scan < loceol &&
6139 isALNUM_LC_utf8((U8*)scan)) {
6140 scan += UTF8SKIP(scan);
6144 while (scan < loceol && isALNUM_LC(*scan))
6154 LOAD_UTF8_CHARCLASS_ALNUM();
6155 while (hardcount < max && scan < loceol &&
6156 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6158 scan += UTF8SKIP(scan);
6162 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6169 goto utf8_Nwordchar;
6170 while (scan < loceol && ! isALNUM((U8) *scan)) {
6176 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6177 scan += UTF8SKIP(scan);
6181 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6187 PL_reg_flags |= RF_tainted;
6190 while (hardcount < max && scan < loceol &&
6191 !isALNUM_LC_utf8((U8*)scan)) {
6192 scan += UTF8SKIP(scan);
6196 while (scan < loceol && !isALNUM_LC(*scan))
6206 LOAD_UTF8_CHARCLASS_SPACE();
6207 while (hardcount < max && scan < loceol &&
6209 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6211 scan += UTF8SKIP(scan);
6217 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6226 while (scan < loceol && isSPACE((U8) *scan)) {
6231 while (scan < loceol && isSPACE_A((U8) *scan)) {
6236 PL_reg_flags |= RF_tainted;
6239 while (hardcount < max && scan < loceol &&
6240 isSPACE_LC_utf8((U8*)scan)) {
6241 scan += UTF8SKIP(scan);
6245 while (scan < loceol && isSPACE_LC(*scan))
6255 LOAD_UTF8_CHARCLASS_SPACE();
6256 while (hardcount < max && scan < loceol &&
6258 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6260 scan += UTF8SKIP(scan);
6266 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6275 while (scan < loceol && ! isSPACE((U8) *scan)) {
6281 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6282 scan += UTF8SKIP(scan);
6286 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6292 PL_reg_flags |= RF_tainted;
6295 while (hardcount < max && scan < loceol &&
6296 !isSPACE_LC_utf8((U8*)scan)) {
6297 scan += UTF8SKIP(scan);
6301 while (scan < loceol && !isSPACE_LC(*scan))
6308 LOAD_UTF8_CHARCLASS_DIGIT();
6309 while (hardcount < max && scan < loceol &&
6310 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6311 scan += UTF8SKIP(scan);
6315 while (scan < loceol && isDIGIT(*scan))
6320 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6325 PL_reg_flags |= RF_tainted;
6328 while (hardcount < max && scan < loceol &&
6329 isDIGIT_LC_utf8((U8*)scan)) {
6330 scan += UTF8SKIP(scan);
6334 while (scan < loceol && isDIGIT_LC(*scan))
6341 LOAD_UTF8_CHARCLASS_DIGIT();
6342 while (hardcount < max && scan < loceol &&
6343 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6344 scan += UTF8SKIP(scan);
6348 while (scan < loceol && !isDIGIT(*scan))
6354 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6355 scan += UTF8SKIP(scan);
6359 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6365 PL_reg_flags |= RF_tainted;
6368 while (hardcount < max && scan < loceol &&
6369 !isDIGIT_LC_utf8((U8*)scan)) {
6370 scan += UTF8SKIP(scan);
6374 while (scan < loceol && !isDIGIT_LC(*scan))
6381 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6387 LNBREAK can match two latin chars, which is ok,
6388 because we have a null terminated string, but we
6389 have to use hardcount in this situation
6391 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6400 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6405 while (scan < loceol && is_HORIZWS_latin1(scan))
6412 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6413 scan += UTF8SKIP(scan);
6417 while (scan < loceol && !is_HORIZWS_latin1(scan))
6425 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6430 while (scan < loceol && is_VERTWS_latin1(scan))
6438 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6439 scan += UTF8SKIP(scan);
6443 while (scan < loceol && !is_VERTWS_latin1(scan))
6449 default: /* Called on something of 0 width. */
6450 break; /* So match right here or not at all. */
6456 c = scan - PL_reginput;
6460 GET_RE_DEBUG_FLAGS_DECL;
6462 SV * const prop = sv_newmortal();
6463 regprop(prog, prop, p);
6464 PerlIO_printf(Perl_debug_log,
6465 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6466 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6474 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6476 - regclass_swash - prepare the utf8 swash
6480 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6486 RXi_GET_DECL(prog,progi);
6487 const struct reg_data * const data = prog ? progi->data : NULL;
6489 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6491 assert(ANYOF_NONBITMAP(node));
6493 if (data && data->count) {
6494 const U32 n = ARG(node);
6496 if (data->what[n] == 's') {
6497 SV * const rv = MUTABLE_SV(data->data[n]);
6498 AV * const av = MUTABLE_AV(SvRV(rv));
6499 SV **const ary = AvARRAY(av);
6502 /* See the end of regcomp.c:S_regclass() for
6503 * documentation of these array elements. */
6506 a = SvROK(ary[1]) ? &ary[1] : NULL;
6507 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6511 else if (si && doinit) {
6512 sw = swash_init("utf8", "", si, 1, 0);
6513 (void)av_store(av, 1, sw);
6530 - reginclass - determine if a character falls into a character class
6532 n is the ANYOF regnode
6533 p is the target string
6534 lenp is pointer to the maximum number of bytes of how far to go in p
6535 (This is assumed wthout checking to always be at least the current
6537 utf8_target tells whether p is in UTF-8.
6539 Returns true if matched; false otherwise. If lenp is not NULL, on return
6540 from a successful match, the value it points to will be updated to how many
6541 bytes in p were matched. If there was no match, the value is undefined,
6542 possibly changed from the input.
6544 Note that this can be a synthetic start class, a combination of various
6545 nodes, so things you think might be mutually exclusive, such as locale,
6546 aren't. It can match both locale and non-locale
6551 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6554 const char flags = ANYOF_FLAGS(n);
6560 PERL_ARGS_ASSERT_REGINCLASS;
6562 /* If c is not already the code point, get it */
6563 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6564 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6565 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6566 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6567 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6568 * UTF8_ALLOW_FFFF */
6569 if (c_len == (STRLEN)-1)
6570 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6576 /* Use passed in max length, or one character if none passed in or less
6577 * than one character. And assume will match just one character. This is
6578 * overwritten later if matched more. */
6580 maxlen = (*lenp > c_len) ? *lenp : c_len;
6588 /* If this character is potentially in the bitmap, check it */
6590 if (ANYOF_BITMAP_TEST(n, c))
6592 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6599 else if (flags & ANYOF_LOCALE) {
6600 PL_reg_flags |= RF_tainted;
6602 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6603 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6607 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6608 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6609 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6610 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6611 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6612 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6613 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6614 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6615 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6616 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6617 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6618 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6619 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6620 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6621 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6622 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6623 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6624 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6625 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6626 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6627 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6628 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6629 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6630 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6631 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6632 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6633 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6634 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6635 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6636 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6637 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6638 ) /* How's that for a conditional? */
6645 /* If the bitmap didn't (or couldn't) match, and something outside the
6646 * bitmap could match, try that. Locale nodes specifiy completely the
6647 * behavior of code points in the bit map (otherwise, a utf8 target would
6648 * cause them to be treated as Unicode and not locale), except in
6649 * the very unlikely event when this node is a synthetic start class, which
6650 * could be a combination of locale and non-locale nodes. So allow locale
6651 * to match for the synthetic start class, which will give a false
6652 * positive that will be resolved when the match is done again as not part
6653 * of the synthetic start class */
6655 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6656 match = TRUE; /* Everything above 255 matches */
6658 else if (ANYOF_NONBITMAP(n)
6659 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6662 || (! (flags & ANYOF_LOCALE))
6663 || (flags & ANYOF_IS_SYNTHETIC)))))
6666 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6674 /* Not utf8. Convert as much of the string as available up
6675 * to the limit of how far the (single) character in the
6676 * pattern can possibly match (no need to go further). If
6677 * the node is a straight ANYOF or not folding, it can't
6678 * match more than one. Otherwise, It can match up to how
6679 * far a single char can fold to. Since not utf8, each
6680 * character is a single byte, so the max it can be in
6681 * bytes is the same as the max it can be in characters */
6682 STRLEN len = (OP(n) == ANYOF
6683 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6685 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6687 : UTF8_MAX_FOLD_CHAR_EXPAND;
6688 utf8_p = bytes_to_utf8(p, &len);
6691 if (swash_fetch(sw, utf8_p, TRUE))
6693 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6695 /* Here, we need to test if the fold of the target string
6696 * matches. The non-multi char folds have all been moved to
6697 * the compilation phase, and the multi-char folds have
6698 * been stored by regcomp into 'av'; we linearly check to
6699 * see if any match the target string (folded). We know
6700 * that the originals were each one character, but we don't
6701 * currently know how many characters/bytes each folded to,
6702 * except we do know that there are small limits imposed by
6703 * Unicode. XXX A performance enhancement would be to have
6704 * regcomp.c store the max number of chars/bytes that are
6705 * in an av entry, as, say the 0th element. Even better
6706 * would be to have a hash of the few characters that can
6707 * start a multi-char fold to the max number of chars of
6710 * If there is a match, we will need to advance (if lenp is
6711 * specified) the match pointer in the target string. But
6712 * what we are comparing here isn't that string directly,
6713 * but its fold, whose length may differ from the original.
6714 * As we go along in constructing the fold, therefore, we
6715 * create a map so that we know how many bytes in the
6716 * source to advance given that we have matched a certain
6717 * number of bytes in the fold. This map is stored in
6718 * 'map_fold_len_back'. Let n mean the number of bytes in
6719 * the fold of the first character that we are folding.
6720 * Then map_fold_len_back[n] is set to the number of bytes
6721 * in that first character. Similarly let m be the
6722 * corresponding number for the second character to be
6723 * folded. Then map_fold_len_back[n+m] is set to the
6724 * number of bytes occupied by the first two source
6725 * characters. ... */
6726 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6727 U8 folded[UTF8_MAXBYTES_CASE+1];
6728 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6729 STRLEN total_foldlen = 0; /* num bytes in fold of all
6732 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6734 /* Here, only need to fold the first char of the target
6735 * string. It the source wasn't utf8, is 1 byte long */
6736 to_utf8_fold(utf8_p, folded, &foldlen);
6737 total_foldlen = foldlen;
6738 map_fold_len_back[foldlen] = (utf8_target)
6744 /* Here, need to fold more than the first char. Do so
6745 * up to the limits */
6746 U8* source_ptr = utf8_p; /* The source for the fold
6749 U8* folded_ptr = folded;
6750 U8* e = utf8_p + maxlen; /* Can't go beyond last
6751 available byte in the
6755 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6759 /* Fold the next character */
6760 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6761 STRLEN this_char_foldlen;
6762 to_utf8_fold(source_ptr,
6764 &this_char_foldlen);
6766 /* Bail if it would exceed the byte limit for
6767 * folding a single char. */
6768 if (this_char_foldlen + folded_ptr - folded >
6774 /* Add the fold of this character */
6775 Copy(this_char_folded,
6779 source_ptr += UTF8SKIP(source_ptr);
6780 folded_ptr += this_char_foldlen;
6781 total_foldlen = folded_ptr - folded;
6783 /* Create map from the number of bytes in the fold
6784 * back to the number of bytes in the source. If
6785 * the source isn't utf8, the byte count is just
6786 * the number of characters so far */
6787 map_fold_len_back[total_foldlen]
6789 ? source_ptr - utf8_p
6796 /* Do the linear search to see if the fold is in the list
6797 * of multi-char folds. */
6800 for (i = 0; i <= av_len(av); i++) {
6801 SV* const sv = *av_fetch(av, i, FALSE);
6803 const char * const s = SvPV_const(sv, len);
6805 if (len <= total_foldlen
6806 && memEQ(s, (char*)folded, len)
6808 /* If 0, means matched a partial char. See
6810 && map_fold_len_back[len])
6813 /* Advance the target string ptr to account for
6814 * this fold, but have to translate from the
6815 * folded length to the corresponding source
6818 *lenp = map_fold_len_back[len];
6827 /* If we allocated a string above, free it */
6828 if (! utf8_target) Safefree(utf8_p);
6833 return (flags & ANYOF_INVERT) ? !match : match;
6837 S_reghop3(U8 *s, I32 off, const U8* lim)
6839 /* return the position 'off' UTF-8 characters away from 's', forward if
6840 * 'off' >= 0, backwards if negative. But don't go outside of position
6841 * 'lim', which better be < s if off < 0 */
6845 PERL_ARGS_ASSERT_REGHOP3;
6848 while (off-- && s < lim) {
6849 /* XXX could check well-formedness here */
6854 while (off++ && s > lim) {
6856 if (UTF8_IS_CONTINUED(*s)) {
6857 while (s > lim && UTF8_IS_CONTINUATION(*s))
6860 /* XXX could check well-formedness here */
6867 /* there are a bunch of places where we use two reghop3's that should
6868 be replaced with this routine. but since thats not done yet
6869 we ifdef it out - dmq
6872 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6876 PERL_ARGS_ASSERT_REGHOP4;
6879 while (off-- && s < rlim) {
6880 /* XXX could check well-formedness here */
6885 while (off++ && s > llim) {
6887 if (UTF8_IS_CONTINUED(*s)) {
6888 while (s > llim && UTF8_IS_CONTINUATION(*s))
6891 /* XXX could check well-formedness here */
6899 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6903 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6906 while (off-- && s < lim) {
6907 /* XXX could check well-formedness here */
6914 while (off++ && s > lim) {
6916 if (UTF8_IS_CONTINUED(*s)) {
6917 while (s > lim && UTF8_IS_CONTINUATION(*s))
6920 /* XXX could check well-formedness here */
6929 restore_pos(pTHX_ void *arg)
6932 regexp * const rex = (regexp *)arg;
6933 if (PL_reg_eval_set) {
6934 if (PL_reg_oldsaved) {
6935 rex->subbeg = PL_reg_oldsaved;
6936 rex->sublen = PL_reg_oldsavedlen;
6937 #ifdef PERL_OLD_COPY_ON_WRITE
6938 rex->saved_copy = PL_nrs;
6940 RXp_MATCH_COPIED_on(rex);
6942 PL_reg_magic->mg_len = PL_reg_oldpos;
6943 PL_reg_eval_set = 0;
6944 PL_curpm = PL_reg_oldcurpm;
6949 S_to_utf8_substr(pTHX_ register regexp *prog)
6953 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6956 if (prog->substrs->data[i].substr
6957 && !prog->substrs->data[i].utf8_substr) {
6958 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6959 prog->substrs->data[i].utf8_substr = sv;
6960 sv_utf8_upgrade(sv);
6961 if (SvVALID(prog->substrs->data[i].substr)) {
6962 if (SvTAIL(prog->substrs->data[i].substr)) {
6963 /* Trim the trailing \n that fbm_compile added last
6965 SvCUR_set(sv, SvCUR(sv) - 1);
6966 /* Whilst this makes the SV technically "invalid" (as its
6967 buffer is no longer followed by "\0") when fbm_compile()
6968 adds the "\n" back, a "\0" is restored. */
6969 fbm_compile(sv, FBMcf_TAIL);
6973 if (prog->substrs->data[i].substr == prog->check_substr)
6974 prog->check_utf8 = sv;
6980 S_to_byte_substr(pTHX_ register regexp *prog)
6985 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6988 if (prog->substrs->data[i].utf8_substr
6989 && !prog->substrs->data[i].substr) {
6990 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6991 if (sv_utf8_downgrade(sv, TRUE)) {
6992 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6993 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
6994 /* Trim the trailing \n that fbm_compile added last
6996 SvCUR_set(sv, SvCUR(sv) - 1);
6997 fbm_compile(sv, FBMcf_TAIL);
7005 prog->substrs->data[i].substr = sv;
7006 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7007 prog->check_substr = sv;
7014 * c-indentation-style: bsd
7016 * indent-tabs-mode: t
7019 * ex: set ts=8 sts=4 sw=4 noet: