5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
78 #ifdef PERL_IN_XSUB_RE
84 #define RF_tainted 1 /* tainted information used? */
85 #define RF_warned 2 /* warned about big count? */
87 #define RF_utf8 8 /* Pattern contains multibyte chars? */
89 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
91 #define RS_init 1 /* eval environment created */
92 #define RS_set 2 /* replsv value is set */
98 /* Valid for non-utf8 strings only: avoids the reginclass call if there are no
99 * complications: i.e., if everything matchable is straight forward in the
101 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
102 : ANYOF_BITMAP_TEST(p,*(c)))
108 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
109 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
111 #define HOPc(pos,off) \
112 (char *)(PL_reg_match_utf8 \
113 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
115 #define HOPBACKc(pos, off) \
116 (char*)(PL_reg_match_utf8\
117 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
118 : (pos - off >= PL_bostr) \
122 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
123 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
125 /* these are unrolled below in the CCC_TRY_XXX defined */
126 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
127 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
129 /* Doesn't do an assert to verify that is correct */
130 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
131 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END
133 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
134 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
135 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
137 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
138 LOAD_UTF8_CHARCLASS(X_begin, " "); \
139 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
140 /* These are utf8 constants, and not utf-ebcdic constants, so the \
141 * assert should likely and hopefully fail on an EBCDIC machine */ \
142 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
144 /* No asserts are done for these, in case called on an early \
145 * Unicode version in which they map to nothing */ \
146 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
147 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
148 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
149 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
150 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
151 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
152 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
155 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
156 so that it is possible to override the option here without having to
157 rebuild the entire core. as we are required to do if we change regcomp.h
158 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
160 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
161 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
164 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
165 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS_ALNUM()
166 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS_SPACE()
167 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS_DIGIT()
168 #define RE_utf8_perl_word PL_utf8_alnum
169 #define RE_utf8_perl_space PL_utf8_space
170 #define RE_utf8_posix_digit PL_utf8_digit
171 #define perl_word alnum
172 #define perl_space space
173 #define posix_digit digit
175 #define LOAD_UTF8_CHARCLASS_PERL_WORD() LOAD_UTF8_CHARCLASS(perl_word,"a")
176 #define LOAD_UTF8_CHARCLASS_PERL_SPACE() LOAD_UTF8_CHARCLASS(perl_space," ")
177 #define LOAD_UTF8_CHARCLASS_POSIX_DIGIT() LOAD_UTF8_CHARCLASS(posix_digit,"0")
178 #define RE_utf8_perl_word PL_utf8_perl_word
179 #define RE_utf8_perl_space PL_utf8_perl_space
180 #define RE_utf8_posix_digit PL_utf8_posix_digit
184 #define _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
186 PL_reg_flags |= RF_tainted; \
191 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
192 if (!CAT2(PL_utf8_,CLASS)) { \
196 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
200 if (!(OP(scan) == NAME \
201 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
202 : LCFUNC_utf8((U8*)locinput))) \
206 locinput += PL_utf8skip[nextchr]; \
207 nextchr = UCHARAT(locinput); \
210 /* Drops through to the macro that calls this one */
212 #define CCC_TRY_AFF(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
213 _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
214 if (!(OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
216 nextchr = UCHARAT(++locinput); \
219 /* Almost identical to the above, but has a case for a node that matches chars
220 * between 128 and 255 using Unicode (latin1) semantics. */
221 #define CCC_TRY_AFF_U(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU,LCFUNC) \
222 _CCC_TRY_AFF_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
223 if (!(OP(scan) == NAMEL ? LCFUNC(nextchr) : (FUNCU(nextchr) && (isASCII(nextchr) || (FLAGS(scan) & USE_UNI))))) \
225 nextchr = UCHARAT(++locinput); \
228 #define _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
230 PL_reg_flags |= RF_tainted; \
233 if (!nextchr && locinput >= PL_regeol) \
235 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
236 if (!CAT2(PL_utf8_,CLASS)) { \
240 ok=CAT2(is_utf8_,CLASS)((const U8*)STR); \
244 if ((OP(scan) == NAME \
245 ? cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), (U8*)locinput, utf8_target)) \
246 : LCFUNC_utf8((U8*)locinput))) \
250 locinput += PL_utf8skip[nextchr]; \
251 nextchr = UCHARAT(locinput); \
255 #define CCC_TRY_NEG(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC,LCFUNC) \
256 _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNC) \
257 if ((OP(scan) == NAME ? FUNC(nextchr) : LCFUNC(nextchr))) \
259 nextchr = UCHARAT(++locinput); \
263 #define CCC_TRY_NEG_U(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU,LCFUNC) \
264 _CCC_TRY_NEG_COMMON(NAME,NAMEL,CLASS,STR,LCFUNC_utf8,FUNCU) \
265 if ((OP(scan) == NAMEL ? LCFUNC(nextchr) : (FUNCU(nextchr) && (isASCII(nextchr) || (FLAGS(scan) & USE_UNI))))) \
267 nextchr = UCHARAT(++locinput); \
272 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
274 /* for use after a quantifier and before an EXACT-like node -- japhy */
275 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
277 * NOTE that *nothing* that affects backtracking should be in here, specifically
278 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
279 * node that is in between two EXACT like nodes when ascertaining what the required
280 * "follow" character is. This should probably be moved to regex compile time
281 * although it may be done at run time beause of the REF possibility - more
282 * investigation required. -- demerphq
284 #define JUMPABLE(rn) ( \
286 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
288 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
289 OP(rn) == PLUS || OP(rn) == MINMOD || \
291 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
293 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
295 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
298 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
299 we don't need this definition. */
300 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
301 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
302 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
305 /* ... so we use this as its faster. */
306 #define IS_TEXT(rn) ( OP(rn)==EXACT )
307 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU )
308 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
309 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
314 Search for mandatory following text node; for lookahead, the text must
315 follow but for lookbehind (rn->flags != 0) we skip to the next step.
317 #define FIND_NEXT_IMPT(rn) STMT_START { \
318 while (JUMPABLE(rn)) { \
319 const OPCODE type = OP(rn); \
320 if (type == SUSPEND || PL_regkind[type] == CURLY) \
321 rn = NEXTOPER(NEXTOPER(rn)); \
322 else if (type == PLUS) \
324 else if (type == IFMATCH) \
325 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
326 else rn += NEXT_OFF(rn); \
331 static void restore_pos(pTHX_ void *arg);
333 #define REGCP_PAREN_ELEMS 4
334 #define REGCP_OTHER_ELEMS 5
335 #define REGCP_FRAME_ELEMS 1
336 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
337 * are needed for the regexp context stack bookkeeping. */
340 S_regcppush(pTHX_ I32 parenfloor)
343 const int retval = PL_savestack_ix;
344 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
345 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
346 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
348 GET_RE_DEBUG_FLAGS_DECL;
350 if (paren_elems_to_push < 0)
351 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
353 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
354 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
355 " out of range (%lu-%ld)",
356 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
358 SSGROW(total_elems + REGCP_FRAME_ELEMS);
360 for (p = PL_regsize; p > parenfloor; p--) {
361 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
362 SSPUSHINT(PL_regoffs[p].end);
363 SSPUSHINT(PL_regoffs[p].start);
364 SSPUSHPTR(PL_reg_start_tmp[p]);
366 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
367 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
368 (UV)p, (IV)PL_regoffs[p].start,
369 (IV)(PL_reg_start_tmp[p] - PL_bostr),
370 (IV)PL_regoffs[p].end
373 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
374 SSPUSHPTR(PL_regoffs);
375 SSPUSHINT(PL_regsize);
376 SSPUSHINT(*PL_reglastparen);
377 SSPUSHINT(*PL_reglastcloseparen);
378 SSPUSHPTR(PL_reginput);
379 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
384 /* These are needed since we do not localize EVAL nodes: */
385 #define REGCP_SET(cp) \
387 PerlIO_printf(Perl_debug_log, \
388 " Setting an EVAL scope, savestack=%"IVdf"\n", \
389 (IV)PL_savestack_ix)); \
392 #define REGCP_UNWIND(cp) \
394 if (cp != PL_savestack_ix) \
395 PerlIO_printf(Perl_debug_log, \
396 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
397 (IV)(cp), (IV)PL_savestack_ix)); \
401 S_regcppop(pTHX_ const regexp *rex)
406 GET_RE_DEBUG_FLAGS_DECL;
408 PERL_ARGS_ASSERT_REGCPPOP;
410 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
412 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
413 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
414 input = (char *) SSPOPPTR;
415 *PL_reglastcloseparen = SSPOPINT;
416 *PL_reglastparen = SSPOPINT;
417 PL_regsize = SSPOPINT;
418 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
420 i -= REGCP_OTHER_ELEMS;
421 /* Now restore the parentheses context. */
422 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
424 U32 paren = (U32)SSPOPINT;
425 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
426 PL_regoffs[paren].start = SSPOPINT;
428 if (paren <= *PL_reglastparen)
429 PL_regoffs[paren].end = tmps;
431 PerlIO_printf(Perl_debug_log,
432 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
433 (UV)paren, (IV)PL_regoffs[paren].start,
434 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
435 (IV)PL_regoffs[paren].end,
436 (paren > *PL_reglastparen ? "(no)" : ""));
440 if (*PL_reglastparen + 1 <= rex->nparens) {
441 PerlIO_printf(Perl_debug_log,
442 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
443 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
447 /* It would seem that the similar code in regtry()
448 * already takes care of this, and in fact it is in
449 * a better location to since this code can #if 0-ed out
450 * but the code in regtry() is needed or otherwise tests
451 * requiring null fields (pat.t#187 and split.t#{13,14}
452 * (as of patchlevel 7877) will fail. Then again,
453 * this code seems to be necessary or otherwise
454 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
455 * --jhi updated by dapm */
456 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
458 PL_regoffs[i].start = -1;
459 PL_regoffs[i].end = -1;
465 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
468 * pregexec and friends
471 #ifndef PERL_IN_XSUB_RE
473 - pregexec - match a regexp against a string
476 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
477 char *strbeg, I32 minend, SV *screamer, U32 nosave)
478 /* strend: pointer to null at end of string */
479 /* strbeg: real beginning of string */
480 /* minend: end of match must be >=minend after stringarg. */
481 /* nosave: For optimizations. */
483 PERL_ARGS_ASSERT_PREGEXEC;
486 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
487 nosave ? 0 : REXEC_COPY_STR);
492 * Need to implement the following flags for reg_anch:
494 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
496 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
497 * INTUIT_AUTORITATIVE_ML
498 * INTUIT_ONCE_NOML - Intuit can match in one location only.
501 * Another flag for this function: SECOND_TIME (so that float substrs
502 * with giant delta may be not rechecked).
505 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
507 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
508 Otherwise, only SvCUR(sv) is used to get strbeg. */
510 /* XXXX We assume that strpos is strbeg unless sv. */
512 /* XXXX Some places assume that there is a fixed substring.
513 An update may be needed if optimizer marks as "INTUITable"
514 RExen without fixed substrings. Similarly, it is assumed that
515 lengths of all the strings are no more than minlen, thus they
516 cannot come from lookahead.
517 (Or minlen should take into account lookahead.)
518 NOTE: Some of this comment is not correct. minlen does now take account
519 of lookahead/behind. Further research is required. -- demerphq
523 /* A failure to find a constant substring means that there is no need to make
524 an expensive call to REx engine, thus we celebrate a failure. Similarly,
525 finding a substring too deep into the string means that less calls to
526 regtry() should be needed.
528 REx compiler's optimizer found 4 possible hints:
529 a) Anchored substring;
531 c) Whether we are anchored (beginning-of-line or \G);
532 d) First node (of those at offset 0) which may distingush positions;
533 We use a)b)d) and multiline-part of c), and try to find a position in the
534 string which does not contradict any of them.
537 /* Most of decisions we do here should have been done at compile time.
538 The nodes of the REx which we used for the search should have been
539 deleted from the finite automaton. */
542 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
543 char *strend, const U32 flags, re_scream_pos_data *data)
546 struct regexp *const prog = (struct regexp *)SvANY(rx);
547 register I32 start_shift = 0;
548 /* Should be nonnegative! */
549 register I32 end_shift = 0;
554 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
556 register char *other_last = NULL; /* other substr checked before this */
557 char *check_at = NULL; /* check substr found at this pos */
558 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
559 RXi_GET_DECL(prog,progi);
561 const char * const i_strpos = strpos;
563 GET_RE_DEBUG_FLAGS_DECL;
565 PERL_ARGS_ASSERT_RE_INTUIT_START;
567 RX_MATCH_UTF8_set(rx,utf8_target);
570 PL_reg_flags |= RF_utf8;
573 debug_start_match(rx, utf8_target, strpos, strend,
574 sv ? "Guessing start of match in sv for"
575 : "Guessing start of match in string for");
578 /* CHR_DIST() would be more correct here but it makes things slow. */
579 if (prog->minlen > strend - strpos) {
580 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
581 "String too short... [re_intuit_start]\n"));
585 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
588 if (!prog->check_utf8 && prog->check_substr)
589 to_utf8_substr(prog);
590 check = prog->check_utf8;
592 if (!prog->check_substr && prog->check_utf8)
593 to_byte_substr(prog);
594 check = prog->check_substr;
596 if (check == &PL_sv_undef) {
597 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
598 "Non-utf8 string cannot match utf8 check string\n"));
601 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
602 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
603 || ( (prog->extflags & RXf_ANCH_BOL)
604 && !multiline ) ); /* Check after \n? */
607 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
608 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
609 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
611 && (strpos != strbeg)) {
612 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
615 if (prog->check_offset_min == prog->check_offset_max &&
616 !(prog->extflags & RXf_CANY_SEEN)) {
617 /* Substring at constant offset from beg-of-str... */
620 s = HOP3c(strpos, prog->check_offset_min, strend);
623 slen = SvCUR(check); /* >= 1 */
625 if ( strend - s > slen || strend - s < slen - 1
626 || (strend - s == slen && strend[-1] != '\n')) {
627 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
630 /* Now should match s[0..slen-2] */
632 if (slen && (*SvPVX_const(check) != *s
634 && memNE(SvPVX_const(check), s, slen)))) {
636 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
640 else if (*SvPVX_const(check) != *s
641 || ((slen = SvCUR(check)) > 1
642 && memNE(SvPVX_const(check), s, slen)))
645 goto success_at_start;
648 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
650 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
651 end_shift = prog->check_end_shift;
654 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
655 - (SvTAIL(check) != 0);
656 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
658 if (end_shift < eshift)
662 else { /* Can match at random position */
665 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
666 end_shift = prog->check_end_shift;
668 /* end shift should be non negative here */
671 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
673 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
674 (IV)end_shift, RX_PRECOMP(prog));
678 /* Find a possible match in the region s..strend by looking for
679 the "check" substring in the region corrected by start/end_shift. */
682 I32 srch_start_shift = start_shift;
683 I32 srch_end_shift = end_shift;
684 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
685 srch_end_shift -= ((strbeg - s) - srch_start_shift);
686 srch_start_shift = strbeg - s;
688 DEBUG_OPTIMISE_MORE_r({
689 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
690 (IV)prog->check_offset_min,
691 (IV)srch_start_shift,
693 (IV)prog->check_end_shift);
696 if (flags & REXEC_SCREAM) {
697 I32 p = -1; /* Internal iterator of scream. */
698 I32 * const pp = data ? data->scream_pos : &p;
700 if (PL_screamfirst[BmRARE(check)] >= 0
701 || ( BmRARE(check) == '\n'
702 && (BmPREVIOUS(check) == SvCUR(check) - 1)
704 s = screaminstr(sv, check,
705 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
708 /* we may be pointing at the wrong string */
709 if (s && RXp_MATCH_COPIED(prog))
710 s = strbeg + (s - SvPVX_const(sv));
712 *data->scream_olds = s;
717 if (prog->extflags & RXf_CANY_SEEN) {
718 start_point= (U8*)(s + srch_start_shift);
719 end_point= (U8*)(strend - srch_end_shift);
721 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
722 end_point= HOP3(strend, -srch_end_shift, strbeg);
724 DEBUG_OPTIMISE_MORE_r({
725 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
726 (int)(end_point - start_point),
727 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
731 s = fbm_instr( start_point, end_point,
732 check, multiline ? FBMrf_MULTILINE : 0);
735 /* Update the count-of-usability, remove useless subpatterns,
739 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
740 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
741 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
742 (s ? "Found" : "Did not find"),
743 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
744 ? "anchored" : "floating"),
747 (s ? " at offset " : "...\n") );
752 /* Finish the diagnostic message */
753 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
755 /* XXX dmq: first branch is for positive lookbehind...
756 Our check string is offset from the beginning of the pattern.
757 So we need to do any stclass tests offset forward from that
766 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
767 Start with the other substr.
768 XXXX no SCREAM optimization yet - and a very coarse implementation
769 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
770 *always* match. Probably should be marked during compile...
771 Probably it is right to do no SCREAM here...
774 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
775 : (prog->float_substr && prog->anchored_substr))
777 /* Take into account the "other" substring. */
778 /* XXXX May be hopelessly wrong for UTF... */
781 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
784 char * const last = HOP3c(s, -start_shift, strbeg);
786 char * const saved_s = s;
789 t = s - prog->check_offset_max;
790 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
792 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
797 t = HOP3c(t, prog->anchored_offset, strend);
798 if (t < other_last) /* These positions already checked */
800 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
803 /* XXXX It is not documented what units *_offsets are in.
804 We assume bytes, but this is clearly wrong.
805 Meaning this code needs to be carefully reviewed for errors.
809 /* On end-of-str: see comment below. */
810 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
811 if (must == &PL_sv_undef) {
813 DEBUG_r(must = prog->anchored_utf8); /* for debug */
818 HOP3(HOP3(last1, prog->anchored_offset, strend)
819 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
821 multiline ? FBMrf_MULTILINE : 0
824 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
825 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
826 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
827 (s ? "Found" : "Contradicts"),
828 quoted, RE_SV_TAIL(must));
833 if (last1 >= last2) {
834 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
835 ", giving up...\n"));
838 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
839 ", trying floating at offset %ld...\n",
840 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
841 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
842 s = HOP3c(last, 1, strend);
846 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
847 (long)(s - i_strpos)));
848 t = HOP3c(s, -prog->anchored_offset, strbeg);
849 other_last = HOP3c(s, 1, strend);
857 else { /* Take into account the floating substring. */
859 char * const saved_s = s;
862 t = HOP3c(s, -start_shift, strbeg);
864 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
865 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
866 last = HOP3c(t, prog->float_max_offset, strend);
867 s = HOP3c(t, prog->float_min_offset, strend);
870 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
871 must = utf8_target ? prog->float_utf8 : prog->float_substr;
872 /* fbm_instr() takes into account exact value of end-of-str
873 if the check is SvTAIL(ed). Since false positives are OK,
874 and end-of-str is not later than strend we are OK. */
875 if (must == &PL_sv_undef) {
877 DEBUG_r(must = prog->float_utf8); /* for debug message */
880 s = fbm_instr((unsigned char*)s,
881 (unsigned char*)last + SvCUR(must)
883 must, multiline ? FBMrf_MULTILINE : 0);
885 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
886 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
887 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
888 (s ? "Found" : "Contradicts"),
889 quoted, RE_SV_TAIL(must));
893 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
894 ", giving up...\n"));
897 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
898 ", trying anchored starting at offset %ld...\n",
899 (long)(saved_s + 1 - i_strpos)));
901 s = HOP3c(t, 1, strend);
905 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
906 (long)(s - i_strpos)));
907 other_last = s; /* Fix this later. --Hugo */
917 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
919 DEBUG_OPTIMISE_MORE_r(
920 PerlIO_printf(Perl_debug_log,
921 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
922 (IV)prog->check_offset_min,
923 (IV)prog->check_offset_max,
931 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
933 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
936 /* Fixed substring is found far enough so that the match
937 cannot start at strpos. */
939 if (ml_anch && t[-1] != '\n') {
940 /* Eventually fbm_*() should handle this, but often
941 anchored_offset is not 0, so this check will not be wasted. */
942 /* XXXX In the code below we prefer to look for "^" even in
943 presence of anchored substrings. And we search even
944 beyond the found float position. These pessimizations
945 are historical artefacts only. */
947 while (t < strend - prog->minlen) {
949 if (t < check_at - prog->check_offset_min) {
950 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
951 /* Since we moved from the found position,
952 we definitely contradict the found anchored
953 substr. Due to the above check we do not
954 contradict "check" substr.
955 Thus we can arrive here only if check substr
956 is float. Redo checking for "other"=="fixed".
959 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
960 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
961 goto do_other_anchored;
963 /* We don't contradict the found floating substring. */
964 /* XXXX Why not check for STCLASS? */
966 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
967 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
970 /* Position contradicts check-string */
971 /* XXXX probably better to look for check-string
972 than for "\n", so one should lower the limit for t? */
973 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
974 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
975 other_last = strpos = s = t + 1;
980 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
981 PL_colors[0], PL_colors[1]));
985 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
986 PL_colors[0], PL_colors[1]));
990 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
993 /* The found string does not prohibit matching at strpos,
994 - no optimization of calling REx engine can be performed,
995 unless it was an MBOL and we are not after MBOL,
996 or a future STCLASS check will fail this. */
998 /* Even in this situation we may use MBOL flag if strpos is offset
999 wrt the start of the string. */
1000 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1001 && (strpos != strbeg) && strpos[-1] != '\n'
1002 /* May be due to an implicit anchor of m{.*foo} */
1003 && !(prog->intflags & PREGf_IMPLICIT))
1008 DEBUG_EXECUTE_r( if (ml_anch)
1009 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1010 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1013 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1015 prog->check_utf8 /* Could be deleted already */
1016 && --BmUSEFUL(prog->check_utf8) < 0
1017 && (prog->check_utf8 == prog->float_utf8)
1019 prog->check_substr /* Could be deleted already */
1020 && --BmUSEFUL(prog->check_substr) < 0
1021 && (prog->check_substr == prog->float_substr)
1024 /* If flags & SOMETHING - do not do it many times on the same match */
1025 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1026 /* XXX Does the destruction order has to change with utf8_target? */
1027 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1028 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1029 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1030 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1031 check = NULL; /* abort */
1033 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevent flag
1034 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1035 if (prog->intflags & PREGf_IMPLICIT)
1036 prog->extflags &= ~RXf_ANCH_MBOL;
1037 /* XXXX This is a remnant of the old implementation. It
1038 looks wasteful, since now INTUIT can use many
1039 other heuristics. */
1040 prog->extflags &= ~RXf_USE_INTUIT;
1041 /* XXXX What other flags might need to be cleared in this branch? */
1047 /* Last resort... */
1048 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1049 /* trie stclasses are too expensive to use here, we are better off to
1050 leave it to regmatch itself */
1051 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1052 /* minlen == 0 is possible if regstclass is \b or \B,
1053 and the fixed substr is ''$.
1054 Since minlen is already taken into account, s+1 is before strend;
1055 accidentally, minlen >= 1 guaranties no false positives at s + 1
1056 even for \b or \B. But (minlen? 1 : 0) below assumes that
1057 regstclass does not come from lookahead... */
1058 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1059 This leaves EXACTF, EXACTFU only, which are dealt with in find_byclass(). */
1060 const U8* const str = (U8*)STRING(progi->regstclass);
1061 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1062 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1065 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1066 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1067 else if (prog->float_substr || prog->float_utf8)
1068 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1072 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1073 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1076 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1079 const char *what = NULL;
1081 if (endpos == strend) {
1082 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1083 "Could not match STCLASS...\n") );
1086 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1087 "This position contradicts STCLASS...\n") );
1088 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1090 /* Contradict one of substrings */
1091 if (prog->anchored_substr || prog->anchored_utf8) {
1092 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1093 DEBUG_EXECUTE_r( what = "anchored" );
1095 s = HOP3c(t, 1, strend);
1096 if (s + start_shift + end_shift > strend) {
1097 /* XXXX Should be taken into account earlier? */
1098 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1099 "Could not match STCLASS...\n") );
1104 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1105 "Looking for %s substr starting at offset %ld...\n",
1106 what, (long)(s + start_shift - i_strpos)) );
1109 /* Have both, check_string is floating */
1110 if (t + start_shift >= check_at) /* Contradicts floating=check */
1111 goto retry_floating_check;
1112 /* Recheck anchored substring, but not floating... */
1116 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1117 "Looking for anchored substr starting at offset %ld...\n",
1118 (long)(other_last - i_strpos)) );
1119 goto do_other_anchored;
1121 /* Another way we could have checked stclass at the
1122 current position only: */
1127 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1128 "Looking for /%s^%s/m starting at offset %ld...\n",
1129 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1132 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1134 /* Check is floating subtring. */
1135 retry_floating_check:
1136 t = check_at - start_shift;
1137 DEBUG_EXECUTE_r( what = "floating" );
1138 goto hop_and_restart;
1141 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1142 "By STCLASS: moving %ld --> %ld\n",
1143 (long)(t - i_strpos), (long)(s - i_strpos))
1147 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1148 "Does not contradict STCLASS...\n");
1153 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1154 PL_colors[4], (check ? "Guessed" : "Giving up"),
1155 PL_colors[5], (long)(s - i_strpos)) );
1158 fail_finish: /* Substring not found */
1159 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1160 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1162 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1163 PL_colors[4], PL_colors[5]));
1167 #define DECL_TRIE_TYPE(scan) \
1168 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1169 trie_type = (scan->flags != EXACT) \
1170 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1171 : (utf8_target ? trie_utf8 : trie_plain)
1173 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1174 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1175 switch (trie_type) { \
1176 case trie_utf8_fold: \
1177 if ( foldlen>0 ) { \
1178 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1183 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1184 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1185 foldlen -= UNISKIP( uvc ); \
1186 uscan = foldbuf + UNISKIP( uvc ); \
1189 case trie_latin_utf8_fold: \
1190 if ( foldlen>0 ) { \
1191 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1197 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1198 foldlen -= UNISKIP( uvc ); \
1199 uscan = foldbuf + UNISKIP( uvc ); \
1203 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1210 charid = trie->charmap[ uvc ]; \
1214 if (widecharmap) { \
1215 SV** const svpp = hv_fetch(widecharmap, \
1216 (char*)&uvc, sizeof(UV), 0); \
1218 charid = (U16)SvIV(*svpp); \
1223 #define REXEC_FBC_EXACTISH_CHECK(CoNd) \
1225 char *my_strend= (char *)strend; \
1228 foldEQ_utf8(s, &my_strend, 0, utf8_target, \
1229 m, NULL, ln, cBOOL(UTF_PATTERN))) \
1230 && (!reginfo || regtry(reginfo, &s)) ) \
1233 U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
1234 uvchr_to_utf8(tmpbuf, c); \
1235 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
1237 && (f == c1 || f == c2) \
1239 foldEQ_utf8(s, &my_strend, 0, utf8_target,\
1240 m, NULL, ln, cBOOL(UTF_PATTERN)))\
1241 && (!reginfo || regtry(reginfo, &s)) ) \
1247 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1251 case EXACTFU: folder = foldEQ_latin1; break; \
1252 case EXACTFL: folder = foldEQ_locale; break; \
1253 case EXACTF: folder = foldEQ; break; \
1255 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(c)); \
1259 && (ln == 1 || folder(s, m, ln)) \
1260 && (!reginfo || regtry(reginfo, &s)) ) \
1266 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1268 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1274 #define REXEC_FBC_SCAN(CoDe) \
1276 while (s < strend) { \
1282 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1283 REXEC_FBC_UTF8_SCAN( \
1285 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1294 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1297 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1306 #define REXEC_FBC_TRYIT \
1307 if ((!reginfo || regtry(reginfo, &s))) \
1310 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1311 if (utf8_target) { \
1312 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1315 REXEC_FBC_CLASS_SCAN(CoNd); \
1319 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1320 if (utf8_target) { \
1322 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1325 REXEC_FBC_CLASS_SCAN(CoNd); \
1329 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1330 PL_reg_flags |= RF_tainted; \
1331 if (utf8_target) { \
1332 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1335 REXEC_FBC_CLASS_SCAN(CoNd); \
1339 #define DUMP_EXEC_POS(li,s,doutf8) \
1340 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1342 /* We know what class REx starts with. Try to find this position... */
1343 /* if reginfo is NULL, its a dryrun */
1344 /* annoyingly all the vars in this routine have different names from their counterparts
1345 in regmatch. /grrr */
1348 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1349 const char *strend, regmatch_info *reginfo)
1352 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1356 register STRLEN uskip;
1360 register I32 tmp = 1; /* Scratch variable? */
1361 register const bool utf8_target = PL_reg_match_utf8;
1362 RXi_GET_DECL(prog,progi);
1364 PERL_ARGS_ASSERT_FIND_BYCLASS;
1366 /* We know what class it must start with. */
1370 REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_NONBITMAP) ||
1371 !UTF8_IS_INVARIANT((U8)s[0]) ?
1372 reginclass(prog, c, (U8*)s, 0, utf8_target) :
1373 REGINCLASS(prog, c, (U8*)s));
1376 while (s < strend) {
1379 if (REGINCLASS(prog, c, (U8*)s) ||
1380 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1381 /* The assignment of 2 is intentional:
1382 * for the folded sharp s, the skip is 2. */
1383 (skip = SHARP_S_SKIP))) {
1384 if (tmp && (!reginfo || regtry(reginfo, &s)))
1397 if (tmp && (!reginfo || regtry(reginfo, &s)))
1406 ln = STR_LEN(c); /* length to match in octets/bytes */
1407 lnc = (I32) ln; /* length to match in characters */
1409 STRLEN ulen1, ulen2;
1411 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1412 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1413 /* used by commented-out code below */
1414 /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
1416 /* XXX: Since the node will be case folded at compile
1417 time this logic is a little odd, although im not
1418 sure that its actually wrong. --dmq */
1420 c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1421 c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1423 /* XXX: This is kinda strange. to_utf8_XYZ returns the
1424 codepoint of the first character in the converted
1425 form, yet originally we did the extra step.
1426 No tests fail by commenting this code out however
1427 so Ive left it out. -- dmq.
1429 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1431 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1436 while (sm < ((U8 *) m + ln)) {
1443 if (utf8_target || OP(c) == EXACTFU) {
1445 /* Micro sign folds to GREEK SMALL LETTER MU;
1446 LATIN_SMALL_LETTER_SHARP_S folds to 'ss', and this sets
1447 c2 to the first 's' of the pair, and the code below will
1449 c2 = (c1 == MICRO_SIGN)
1450 ? GREEK_SMALL_LETTER_MU
1451 : (c1 == LATIN_SMALL_LETTER_SHARP_S)
1453 : PL_fold_latin1[c1];
1454 } else c2 = PL_fold[c1];
1462 c2 = PL_fold_locale[c1];
1464 e = HOP3c(strend, -((I32)lnc), s);
1466 if (!reginfo && e < s)
1467 e = s; /* Due to minlen logic of intuit() */
1469 /* The idea in the EXACTF* cases is to first find the
1470 * first character of the EXACTF* node and then, if
1471 * necessary, case-insensitively compare the full
1472 * text of the node. The c1 and c2 are the first
1473 * characters (though in Unicode it gets a bit
1474 * more complicated because there are more cases
1475 * than just upper and lower: one needs to use
1476 * the so-called folding case for case-insensitive
1477 * matching (called "loose matching" in Unicode).
1478 * foldEQ_utf8() will do just that. */
1480 if (utf8_target || UTF_PATTERN) {
1482 U8 tmpbuf [UTF8_MAXBYTES+1];
1485 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1487 /* Upper and lower of 1st char are equal -
1488 * probably not a "letter". */
1491 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1496 REXEC_FBC_EXACTISH_CHECK(c == c1);
1502 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1508 /* Handle some of the three Greek sigmas cases.
1509 * Note that not all the possible combinations
1510 * are handled here: some of them are handled
1511 * by the standard folding rules, and some of
1512 * them (the character class or ANYOF cases)
1513 * are handled during compiletime in
1514 * regexec.c:S_regclass(). */
1515 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1516 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1517 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1519 REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2);
1524 /* Neither pattern nor string are UTF8 */
1526 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1528 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1532 PL_reg_flags |= RF_tainted;
1539 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1540 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1542 tmp = ((OP(c) == BOUND ?
1543 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1544 LOAD_UTF8_CHARCLASS_ALNUM();
1545 REXEC_FBC_UTF8_SCAN(
1546 if (tmp == !(OP(c) == BOUND ?
1547 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1548 isALNUM_LC_utf8((U8*)s)))
1555 else { /* Not utf8 */
1556 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1557 tmp = cBOOL((OP(c) == BOUNDL)
1559 : (isWORDCHAR_L1(tmp)
1560 && (isASCII(tmp) || (FLAGS(c) & USE_UNI))));
1565 : (isWORDCHAR_L1((U8) *s)
1566 && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))))
1573 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s)))
1577 PL_reg_flags |= RF_tainted;
1584 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1585 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT);
1587 tmp = ((OP(c) == NBOUND ?
1588 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1589 LOAD_UTF8_CHARCLASS_ALNUM();
1590 REXEC_FBC_UTF8_SCAN(
1591 if (tmp == !(OP(c) == NBOUND ?
1592 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)) :
1593 isALNUM_LC_utf8((U8*)s)))
1595 else REXEC_FBC_TRYIT;
1599 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1600 tmp = cBOOL((OP(c) == NBOUNDL)
1602 : (isWORDCHAR_L1(tmp)
1603 && (isASCII(tmp) || (FLAGS(c) & USE_UNI))));
1608 : (isWORDCHAR_L1((U8) *s)
1609 && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))))
1613 else REXEC_FBC_TRYIT;
1616 if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, &s)))
1620 REXEC_FBC_CSCAN_PRELOAD(
1621 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1622 swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1623 (FLAGS(c) & USE_UNI) ? isWORDCHAR_L1((U8) *s) : isALNUM(*s)
1626 REXEC_FBC_CSCAN_TAINT(
1627 isALNUM_LC_utf8((U8*)s),
1631 REXEC_FBC_CSCAN_PRELOAD(
1632 LOAD_UTF8_CHARCLASS_PERL_WORD(),
1633 !swash_fetch(RE_utf8_perl_word, (U8*)s, utf8_target),
1634 ! ((FLAGS(c) & USE_UNI) ? isWORDCHAR_L1((U8) *s) : isALNUM(*s))
1637 REXEC_FBC_CSCAN_TAINT(
1638 !isALNUM_LC_utf8((U8*)s),
1642 REXEC_FBC_CSCAN_PRELOAD(
1643 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1644 *s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target),
1645 isSPACE_L1((U8) *s) && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI))
1648 REXEC_FBC_CSCAN_TAINT(
1649 isSPACE_LC_utf8((U8*)s),
1653 REXEC_FBC_CSCAN_PRELOAD(
1654 LOAD_UTF8_CHARCLASS_PERL_SPACE(),
1655 !(*s == ' ' || swash_fetch(RE_utf8_perl_space,(U8*)s, utf8_target)),
1656 !(isSPACE_L1((U8) *s) && (isASCII((U8) *s) || (FLAGS(c) & USE_UNI)))
1659 REXEC_FBC_CSCAN_TAINT(
1660 !isSPACE_LC_utf8((U8*)s),
1664 REXEC_FBC_CSCAN_PRELOAD(
1665 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1666 swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1670 REXEC_FBC_CSCAN_TAINT(
1671 isDIGIT_LC_utf8((U8*)s),
1675 REXEC_FBC_CSCAN_PRELOAD(
1676 LOAD_UTF8_CHARCLASS_POSIX_DIGIT(),
1677 !swash_fetch(RE_utf8_posix_digit,(U8*)s, utf8_target),
1681 REXEC_FBC_CSCAN_TAINT(
1682 !isDIGIT_LC_utf8((U8*)s),
1688 is_LNBREAK_latin1(s)
1698 !is_VERTWS_latin1(s)
1703 is_HORIZWS_latin1(s)
1707 !is_HORIZWS_utf8(s),
1708 !is_HORIZWS_latin1(s)
1714 /* what trie are we using right now */
1716 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1718 = (reg_trie_data*)progi->data->data[ aho->trie ];
1719 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1721 const char *last_start = strend - trie->minlen;
1723 const char *real_start = s;
1725 STRLEN maxlen = trie->maxlen;
1727 U8 **points; /* map of where we were in the input string
1728 when reading a given char. For ASCII this
1729 is unnecessary overhead as the relationship
1730 is always 1:1, but for Unicode, especially
1731 case folded Unicode this is not true. */
1732 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1736 GET_RE_DEBUG_FLAGS_DECL;
1738 /* We can't just allocate points here. We need to wrap it in
1739 * an SV so it gets freed properly if there is a croak while
1740 * running the match */
1743 sv_points=newSV(maxlen * sizeof(U8 *));
1744 SvCUR_set(sv_points,
1745 maxlen * sizeof(U8 *));
1746 SvPOK_on(sv_points);
1747 sv_2mortal(sv_points);
1748 points=(U8**)SvPV_nolen(sv_points );
1749 if ( trie_type != trie_utf8_fold
1750 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1753 bitmap=(U8*)trie->bitmap;
1755 bitmap=(U8*)ANYOF_BITMAP(c);
1757 /* this is the Aho-Corasick algorithm modified a touch
1758 to include special handling for long "unknown char"
1759 sequences. The basic idea being that we use AC as long
1760 as we are dealing with a possible matching char, when
1761 we encounter an unknown char (and we have not encountered
1762 an accepting state) we scan forward until we find a legal
1764 AC matching is basically that of trie matching, except
1765 that when we encounter a failing transition, we fall back
1766 to the current states "fail state", and try the current char
1767 again, a process we repeat until we reach the root state,
1768 state 1, or a legal transition. If we fail on the root state
1769 then we can either terminate if we have reached an accepting
1770 state previously, or restart the entire process from the beginning
1774 while (s <= last_start) {
1775 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1783 U8 *uscan = (U8*)NULL;
1784 U8 *leftmost = NULL;
1786 U32 accepted_word= 0;
1790 while ( state && uc <= (U8*)strend ) {
1792 U32 word = aho->states[ state ].wordnum;
1796 DEBUG_TRIE_EXECUTE_r(
1797 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1798 dump_exec_pos( (char *)uc, c, strend, real_start,
1799 (char *)uc, utf8_target );
1800 PerlIO_printf( Perl_debug_log,
1801 " Scanning for legal start char...\n");
1805 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1809 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1815 if (uc >(U8*)last_start) break;
1819 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1820 if (!leftmost || lpos < leftmost) {
1821 DEBUG_r(accepted_word=word);
1827 points[pointpos++ % maxlen]= uc;
1828 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1829 uscan, len, uvc, charid, foldlen,
1831 DEBUG_TRIE_EXECUTE_r({
1832 dump_exec_pos( (char *)uc, c, strend, real_start,
1834 PerlIO_printf(Perl_debug_log,
1835 " Charid:%3u CP:%4"UVxf" ",
1841 word = aho->states[ state ].wordnum;
1843 base = aho->states[ state ].trans.base;
1845 DEBUG_TRIE_EXECUTE_r({
1847 dump_exec_pos( (char *)uc, c, strend, real_start,
1849 PerlIO_printf( Perl_debug_log,
1850 "%sState: %4"UVxf", word=%"UVxf,
1851 failed ? " Fail transition to " : "",
1852 (UV)state, (UV)word);
1858 ( ((offset = base + charid
1859 - 1 - trie->uniquecharcount)) >= 0)
1860 && ((U32)offset < trie->lasttrans)
1861 && trie->trans[offset].check == state
1862 && (tmp=trie->trans[offset].next))
1864 DEBUG_TRIE_EXECUTE_r(
1865 PerlIO_printf( Perl_debug_log," - legal\n"));
1870 DEBUG_TRIE_EXECUTE_r(
1871 PerlIO_printf( Perl_debug_log," - fail\n"));
1873 state = aho->fail[state];
1877 /* we must be accepting here */
1878 DEBUG_TRIE_EXECUTE_r(
1879 PerlIO_printf( Perl_debug_log," - accepting\n"));
1888 if (!state) state = 1;
1891 if ( aho->states[ state ].wordnum ) {
1892 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1893 if (!leftmost || lpos < leftmost) {
1894 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1899 s = (char*)leftmost;
1900 DEBUG_TRIE_EXECUTE_r({
1902 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1903 (UV)accepted_word, (IV)(s - real_start)
1906 if (!reginfo || regtry(reginfo, &s)) {
1912 DEBUG_TRIE_EXECUTE_r({
1913 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1916 DEBUG_TRIE_EXECUTE_r(
1917 PerlIO_printf( Perl_debug_log,"No match.\n"));
1926 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1936 - regexec_flags - match a regexp against a string
1939 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1940 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1941 /* strend: pointer to null at end of string */
1942 /* strbeg: real beginning of string */
1943 /* minend: end of match must be >=minend after stringarg. */
1944 /* data: May be used for some additional optimizations.
1945 Currently its only used, with a U32 cast, for transmitting
1946 the ganch offset when doing a /g match. This will change */
1947 /* nosave: For optimizations. */
1950 struct regexp *const prog = (struct regexp *)SvANY(rx);
1951 /*register*/ char *s;
1952 register regnode *c;
1953 /*register*/ char *startpos = stringarg;
1954 I32 minlen; /* must match at least this many chars */
1955 I32 dontbother = 0; /* how many characters not to try at end */
1956 I32 end_shift = 0; /* Same for the end. */ /* CC */
1957 I32 scream_pos = -1; /* Internal iterator of scream. */
1958 char *scream_olds = NULL;
1959 const bool utf8_target = cBOOL(DO_UTF8(sv));
1961 RXi_GET_DECL(prog,progi);
1962 regmatch_info reginfo; /* create some info to pass to regtry etc */
1963 regexp_paren_pair *swap = NULL;
1964 GET_RE_DEBUG_FLAGS_DECL;
1966 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1967 PERL_UNUSED_ARG(data);
1969 /* Be paranoid... */
1970 if (prog == NULL || startpos == NULL) {
1971 Perl_croak(aTHX_ "NULL regexp parameter");
1975 multiline = prog->extflags & RXf_PMf_MULTILINE;
1976 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
1978 RX_MATCH_UTF8_set(rx, utf8_target);
1980 debug_start_match(rx, utf8_target, startpos, strend,
1984 minlen = prog->minlen;
1986 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
1987 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1988 "String too short [regexec_flags]...\n"));
1993 /* Check validity of program. */
1994 if (UCHARAT(progi->program) != REG_MAGIC) {
1995 Perl_croak(aTHX_ "corrupted regexp program");
1999 PL_reg_eval_set = 0;
2003 PL_reg_flags |= RF_utf8;
2005 /* Mark beginning of line for ^ and lookbehind. */
2006 reginfo.bol = startpos; /* XXX not used ??? */
2010 /* Mark end of line for $ (and such) */
2013 /* see how far we have to get to not match where we matched before */
2014 reginfo.till = startpos+minend;
2016 /* If there is a "must appear" string, look for it. */
2019 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2021 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2022 reginfo.ganch = startpos + prog->gofs;
2023 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2024 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2025 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2027 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2028 && mg->mg_len >= 0) {
2029 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2030 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2031 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2033 if (prog->extflags & RXf_ANCH_GPOS) {
2034 if (s > reginfo.ganch)
2036 s = reginfo.ganch - prog->gofs;
2037 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2038 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2044 reginfo.ganch = strbeg + PTR2UV(data);
2045 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2046 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2048 } else { /* pos() not defined */
2049 reginfo.ganch = strbeg;
2050 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2051 "GPOS: reginfo.ganch = strbeg\n"));
2054 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2055 /* We have to be careful. If the previous successful match
2056 was from this regex we don't want a subsequent partially
2057 successful match to clobber the old results.
2058 So when we detect this possibility we add a swap buffer
2059 to the re, and switch the buffer each match. If we fail
2060 we switch it back, otherwise we leave it swapped.
2063 /* do we need a save destructor here for eval dies? */
2064 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2066 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2067 re_scream_pos_data d;
2069 d.scream_olds = &scream_olds;
2070 d.scream_pos = &scream_pos;
2071 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2073 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2074 goto phooey; /* not present */
2080 /* Simplest case: anchored match need be tried only once. */
2081 /* [unless only anchor is BOL and multiline is set] */
2082 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2083 if (s == startpos && regtry(®info, &startpos))
2085 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2086 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2091 dontbother = minlen - 1;
2092 end = HOP3c(strend, -dontbother, strbeg) - 1;
2093 /* for multiline we only have to try after newlines */
2094 if (prog->check_substr || prog->check_utf8) {
2095 /* because of the goto we can not easily reuse the macros for bifurcating the
2096 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2099 goto after_try_utf8;
2101 if (regtry(®info, &s)) {
2108 if (prog->extflags & RXf_USE_INTUIT) {
2109 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2118 } /* end search for check string in unicode */
2120 if (s == startpos) {
2121 goto after_try_latin;
2124 if (regtry(®info, &s)) {
2131 if (prog->extflags & RXf_USE_INTUIT) {
2132 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2141 } /* end search for check string in latin*/
2142 } /* end search for check string */
2143 else { /* search for newline */
2145 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2148 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2150 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2151 if (regtry(®info, &s))
2155 } /* end search for newline */
2156 } /* end anchored/multiline check string search */
2158 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2160 /* the warning about reginfo.ganch being used without intialization
2161 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2162 and we only enter this block when the same bit is set. */
2163 char *tmp_s = reginfo.ganch - prog->gofs;
2165 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2170 /* Messy cases: unanchored match. */
2171 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2172 /* we have /x+whatever/ */
2173 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2178 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2179 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2180 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2185 DEBUG_EXECUTE_r( did_match = 1 );
2186 if (regtry(®info, &s)) goto got_it;
2188 while (s < strend && *s == ch)
2196 DEBUG_EXECUTE_r( did_match = 1 );
2197 if (regtry(®info, &s)) goto got_it;
2199 while (s < strend && *s == ch)
2204 DEBUG_EXECUTE_r(if (!did_match)
2205 PerlIO_printf(Perl_debug_log,
2206 "Did not find anchored character...\n")
2209 else if (prog->anchored_substr != NULL
2210 || prog->anchored_utf8 != NULL
2211 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2212 && prog->float_max_offset < strend - s)) {
2217 char *last1; /* Last position checked before */
2221 if (prog->anchored_substr || prog->anchored_utf8) {
2222 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2223 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2224 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2225 back_max = back_min = prog->anchored_offset;
2227 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2228 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2229 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2230 back_max = prog->float_max_offset;
2231 back_min = prog->float_min_offset;
2235 if (must == &PL_sv_undef)
2236 /* could not downgrade utf8 check substring, so must fail */
2242 last = HOP3c(strend, /* Cannot start after this */
2243 -(I32)(CHR_SVLEN(must)
2244 - (SvTAIL(must) != 0) + back_min), strbeg);
2247 last1 = HOPc(s, -1);
2249 last1 = s - 1; /* bogus */
2251 /* XXXX check_substr already used to find "s", can optimize if
2252 check_substr==must. */
2254 dontbother = end_shift;
2255 strend = HOPc(strend, -dontbother);
2256 while ( (s <= last) &&
2257 ((flags & REXEC_SCREAM)
2258 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2259 end_shift, &scream_pos, 0))
2260 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2261 (unsigned char*)strend, must,
2262 multiline ? FBMrf_MULTILINE : 0))) ) {
2263 /* we may be pointing at the wrong string */
2264 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2265 s = strbeg + (s - SvPVX_const(sv));
2266 DEBUG_EXECUTE_r( did_match = 1 );
2267 if (HOPc(s, -back_max) > last1) {
2268 last1 = HOPc(s, -back_min);
2269 s = HOPc(s, -back_max);
2272 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2274 last1 = HOPc(s, -back_min);
2278 while (s <= last1) {
2279 if (regtry(®info, &s))
2285 while (s <= last1) {
2286 if (regtry(®info, &s))
2292 DEBUG_EXECUTE_r(if (!did_match) {
2293 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2294 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2295 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2296 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2297 ? "anchored" : "floating"),
2298 quoted, RE_SV_TAIL(must));
2302 else if ( (c = progi->regstclass) ) {
2304 const OPCODE op = OP(progi->regstclass);
2305 /* don't bother with what can't match */
2306 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2307 strend = HOPc(strend, -(minlen - 1));
2310 SV * const prop = sv_newmortal();
2311 regprop(prog, prop, c);
2313 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2315 PerlIO_printf(Perl_debug_log,
2316 "Matching stclass %.*s against %s (%d bytes)\n",
2317 (int)SvCUR(prop), SvPVX_const(prop),
2318 quoted, (int)(strend - s));
2321 if (find_byclass(prog, c, s, strend, ®info))
2323 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2327 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2332 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2333 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2334 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2336 if (flags & REXEC_SCREAM) {
2337 last = screaminstr(sv, float_real, s - strbeg,
2338 end_shift, &scream_pos, 1); /* last one */
2340 last = scream_olds; /* Only one occurrence. */
2341 /* we may be pointing at the wrong string */
2342 else if (RXp_MATCH_COPIED(prog))
2343 s = strbeg + (s - SvPVX_const(sv));
2347 const char * const little = SvPV_const(float_real, len);
2349 if (SvTAIL(float_real)) {
2350 if (memEQ(strend - len + 1, little, len - 1))
2351 last = strend - len + 1;
2352 else if (!multiline)
2353 last = memEQ(strend - len, little, len)
2354 ? strend - len : NULL;
2360 last = rninstr(s, strend, little, little + len);
2362 last = strend; /* matching "$" */
2367 PerlIO_printf(Perl_debug_log,
2368 "%sCan't trim the tail, match fails (should not happen)%s\n",
2369 PL_colors[4], PL_colors[5]));
2370 goto phooey; /* Should not happen! */
2372 dontbother = strend - last + prog->float_min_offset;
2374 if (minlen && (dontbother < minlen))
2375 dontbother = minlen - 1;
2376 strend -= dontbother; /* this one's always in bytes! */
2377 /* We don't know much -- general case. */
2380 if (regtry(®info, &s))
2389 if (regtry(®info, &s))
2391 } while (s++ < strend);
2400 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2402 if (PL_reg_eval_set)
2403 restore_pos(aTHX_ prog);
2404 if (RXp_PAREN_NAMES(prog))
2405 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2407 /* make sure $`, $&, $', and $digit will work later */
2408 if ( !(flags & REXEC_NOT_FIRST) ) {
2409 RX_MATCH_COPY_FREE(rx);
2410 if (flags & REXEC_COPY_STR) {
2411 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2412 #ifdef PERL_OLD_COPY_ON_WRITE
2414 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2416 PerlIO_printf(Perl_debug_log,
2417 "Copy on write: regexp capture, type %d\n",
2420 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2421 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2422 assert (SvPOKp(prog->saved_copy));
2426 RX_MATCH_COPIED_on(rx);
2427 s = savepvn(strbeg, i);
2433 prog->subbeg = strbeg;
2434 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2441 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2442 PL_colors[4], PL_colors[5]));
2443 if (PL_reg_eval_set)
2444 restore_pos(aTHX_ prog);
2446 /* we failed :-( roll it back */
2447 Safefree(prog->offs);
2456 - regtry - try match at specific point
2458 STATIC I32 /* 0 failure, 1 success */
2459 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2463 REGEXP *const rx = reginfo->prog;
2464 regexp *const prog = (struct regexp *)SvANY(rx);
2465 RXi_GET_DECL(prog,progi);
2466 GET_RE_DEBUG_FLAGS_DECL;
2468 PERL_ARGS_ASSERT_REGTRY;
2470 reginfo->cutpoint=NULL;
2472 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2475 PL_reg_eval_set = RS_init;
2476 DEBUG_EXECUTE_r(DEBUG_s(
2477 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2478 (IV)(PL_stack_sp - PL_stack_base));
2481 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2482 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2484 /* Apparently this is not needed, judging by wantarray. */
2485 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2486 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2489 /* Make $_ available to executed code. */
2490 if (reginfo->sv != DEFSV) {
2492 DEFSV_set(reginfo->sv);
2495 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2496 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2497 /* prepare for quick setting of pos */
2498 #ifdef PERL_OLD_COPY_ON_WRITE
2499 if (SvIsCOW(reginfo->sv))
2500 sv_force_normal_flags(reginfo->sv, 0);
2502 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2503 &PL_vtbl_mglob, NULL, 0);
2507 PL_reg_oldpos = mg->mg_len;
2508 SAVEDESTRUCTOR_X(restore_pos, prog);
2510 if (!PL_reg_curpm) {
2511 Newxz(PL_reg_curpm, 1, PMOP);
2514 SV* const repointer = &PL_sv_undef;
2515 /* this regexp is also owned by the new PL_reg_curpm, which
2516 will try to free it. */
2517 av_push(PL_regex_padav, repointer);
2518 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2519 PL_regex_pad = AvARRAY(PL_regex_padav);
2524 /* It seems that non-ithreads works both with and without this code.
2525 So for efficiency reasons it seems best not to have the code
2526 compiled when it is not needed. */
2527 /* This is safe against NULLs: */
2528 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2529 /* PM_reg_curpm owns a reference to this regexp. */
2532 PM_SETRE(PL_reg_curpm, rx);
2533 PL_reg_oldcurpm = PL_curpm;
2534 PL_curpm = PL_reg_curpm;
2535 if (RXp_MATCH_COPIED(prog)) {
2536 /* Here is a serious problem: we cannot rewrite subbeg,
2537 since it may be needed if this match fails. Thus
2538 $` inside (?{}) could fail... */
2539 PL_reg_oldsaved = prog->subbeg;
2540 PL_reg_oldsavedlen = prog->sublen;
2541 #ifdef PERL_OLD_COPY_ON_WRITE
2542 PL_nrs = prog->saved_copy;
2544 RXp_MATCH_COPIED_off(prog);
2547 PL_reg_oldsaved = NULL;
2548 prog->subbeg = PL_bostr;
2549 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2551 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2552 prog->offs[0].start = *startpos - PL_bostr;
2553 PL_reginput = *startpos;
2554 PL_reglastparen = &prog->lastparen;
2555 PL_reglastcloseparen = &prog->lastcloseparen;
2556 prog->lastparen = 0;
2557 prog->lastcloseparen = 0;
2559 PL_regoffs = prog->offs;
2560 if (PL_reg_start_tmpl <= prog->nparens) {
2561 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2562 if(PL_reg_start_tmp)
2563 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2565 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2568 /* XXXX What this code is doing here?!!! There should be no need
2569 to do this again and again, PL_reglastparen should take care of
2572 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2573 * Actually, the code in regcppop() (which Ilya may be meaning by
2574 * PL_reglastparen), is not needed at all by the test suite
2575 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2576 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2577 * Meanwhile, this code *is* needed for the
2578 * above-mentioned test suite tests to succeed. The common theme
2579 * on those tests seems to be returning null fields from matches.
2580 * --jhi updated by dapm */
2582 if (prog->nparens) {
2583 regexp_paren_pair *pp = PL_regoffs;
2585 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2593 if (regmatch(reginfo, progi->program + 1)) {
2594 PL_regoffs[0].end = PL_reginput - PL_bostr;
2597 if (reginfo->cutpoint)
2598 *startpos= reginfo->cutpoint;
2599 REGCP_UNWIND(lastcp);
2604 #define sayYES goto yes
2605 #define sayNO goto no
2606 #define sayNO_SILENT goto no_silent
2608 /* we dont use STMT_START/END here because it leads to
2609 "unreachable code" warnings, which are bogus, but distracting. */
2610 #define CACHEsayNO \
2611 if (ST.cache_mask) \
2612 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2615 /* this is used to determine how far from the left messages like
2616 'failed...' are printed. It should be set such that messages
2617 are inline with the regop output that created them.
2619 #define REPORT_CODE_OFF 32
2622 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2623 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2625 #define SLAB_FIRST(s) (&(s)->states[0])
2626 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2628 /* grab a new slab and return the first slot in it */
2630 STATIC regmatch_state *
2633 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2636 regmatch_slab *s = PL_regmatch_slab->next;
2638 Newx(s, 1, regmatch_slab);
2639 s->prev = PL_regmatch_slab;
2641 PL_regmatch_slab->next = s;
2643 PL_regmatch_slab = s;
2644 return SLAB_FIRST(s);
2648 /* push a new state then goto it */
2650 #define PUSH_STATE_GOTO(state, node) \
2652 st->resume_state = state; \
2655 /* push a new state with success backtracking, then goto it */
2657 #define PUSH_YES_STATE_GOTO(state, node) \
2659 st->resume_state = state; \
2660 goto push_yes_state;
2666 regmatch() - main matching routine
2668 This is basically one big switch statement in a loop. We execute an op,
2669 set 'next' to point the next op, and continue. If we come to a point which
2670 we may need to backtrack to on failure such as (A|B|C), we push a
2671 backtrack state onto the backtrack stack. On failure, we pop the top
2672 state, and re-enter the loop at the state indicated. If there are no more
2673 states to pop, we return failure.
2675 Sometimes we also need to backtrack on success; for example /A+/, where
2676 after successfully matching one A, we need to go back and try to
2677 match another one; similarly for lookahead assertions: if the assertion
2678 completes successfully, we backtrack to the state just before the assertion
2679 and then carry on. In these cases, the pushed state is marked as
2680 'backtrack on success too'. This marking is in fact done by a chain of
2681 pointers, each pointing to the previous 'yes' state. On success, we pop to
2682 the nearest yes state, discarding any intermediate failure-only states.
2683 Sometimes a yes state is pushed just to force some cleanup code to be
2684 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2685 it to free the inner regex.
2687 Note that failure backtracking rewinds the cursor position, while
2688 success backtracking leaves it alone.
2690 A pattern is complete when the END op is executed, while a subpattern
2691 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2692 ops trigger the "pop to last yes state if any, otherwise return true"
2695 A common convention in this function is to use A and B to refer to the two
2696 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2697 the subpattern to be matched possibly multiple times, while B is the entire
2698 rest of the pattern. Variable and state names reflect this convention.
2700 The states in the main switch are the union of ops and failure/success of
2701 substates associated with with that op. For example, IFMATCH is the op
2702 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2703 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2704 successfully matched A and IFMATCH_A_fail is a state saying that we have
2705 just failed to match A. Resume states always come in pairs. The backtrack
2706 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2707 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2708 on success or failure.
2710 The struct that holds a backtracking state is actually a big union, with
2711 one variant for each major type of op. The variable st points to the
2712 top-most backtrack struct. To make the code clearer, within each
2713 block of code we #define ST to alias the relevant union.
2715 Here's a concrete example of a (vastly oversimplified) IFMATCH
2721 #define ST st->u.ifmatch
2723 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2724 ST.foo = ...; // some state we wish to save
2726 // push a yes backtrack state with a resume value of
2727 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2729 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2732 case IFMATCH_A: // we have successfully executed A; now continue with B
2734 bar = ST.foo; // do something with the preserved value
2737 case IFMATCH_A_fail: // A failed, so the assertion failed
2738 ...; // do some housekeeping, then ...
2739 sayNO; // propagate the failure
2746 For any old-timers reading this who are familiar with the old recursive
2747 approach, the code above is equivalent to:
2749 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2758 ...; // do some housekeeping, then ...
2759 sayNO; // propagate the failure
2762 The topmost backtrack state, pointed to by st, is usually free. If you
2763 want to claim it, populate any ST.foo fields in it with values you wish to
2764 save, then do one of
2766 PUSH_STATE_GOTO(resume_state, node);
2767 PUSH_YES_STATE_GOTO(resume_state, node);
2769 which sets that backtrack state's resume value to 'resume_state', pushes a
2770 new free entry to the top of the backtrack stack, then goes to 'node'.
2771 On backtracking, the free slot is popped, and the saved state becomes the
2772 new free state. An ST.foo field in this new top state can be temporarily
2773 accessed to retrieve values, but once the main loop is re-entered, it
2774 becomes available for reuse.
2776 Note that the depth of the backtrack stack constantly increases during the
2777 left-to-right execution of the pattern, rather than going up and down with
2778 the pattern nesting. For example the stack is at its maximum at Z at the
2779 end of the pattern, rather than at X in the following:
2781 /(((X)+)+)+....(Y)+....Z/
2783 The only exceptions to this are lookahead/behind assertions and the cut,
2784 (?>A), which pop all the backtrack states associated with A before
2787 Bascktrack state structs are allocated in slabs of about 4K in size.
2788 PL_regmatch_state and st always point to the currently active state,
2789 and PL_regmatch_slab points to the slab currently containing
2790 PL_regmatch_state. The first time regmatch() is called, the first slab is
2791 allocated, and is never freed until interpreter destruction. When the slab
2792 is full, a new one is allocated and chained to the end. At exit from
2793 regmatch(), slabs allocated since entry are freed.
2798 #define DEBUG_STATE_pp(pp) \
2800 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2801 PerlIO_printf(Perl_debug_log, \
2802 " %*s"pp" %s%s%s%s%s\n", \
2804 PL_reg_name[st->resume_state], \
2805 ((st==yes_state||st==mark_state) ? "[" : ""), \
2806 ((st==yes_state) ? "Y" : ""), \
2807 ((st==mark_state) ? "M" : ""), \
2808 ((st==yes_state||st==mark_state) ? "]" : "") \
2813 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2818 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2819 const char *start, const char *end, const char *blurb)
2821 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2823 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2828 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2829 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2831 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2832 start, end - start, 60);
2834 PerlIO_printf(Perl_debug_log,
2835 "%s%s REx%s %s against %s\n",
2836 PL_colors[4], blurb, PL_colors[5], s0, s1);
2838 if (utf8_target||utf8_pat)
2839 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2840 utf8_pat ? "pattern" : "",
2841 utf8_pat && utf8_target ? " and " : "",
2842 utf8_target ? "string" : ""
2848 S_dump_exec_pos(pTHX_ const char *locinput,
2849 const regnode *scan,
2850 const char *loc_regeol,
2851 const char *loc_bostr,
2852 const char *loc_reg_starttry,
2853 const bool utf8_target)
2855 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2856 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2857 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2858 /* The part of the string before starttry has one color
2859 (pref0_len chars), between starttry and current
2860 position another one (pref_len - pref0_len chars),
2861 after the current position the third one.
2862 We assume that pref0_len <= pref_len, otherwise we
2863 decrease pref0_len. */
2864 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2865 ? (5 + taill) - l : locinput - loc_bostr;
2868 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2870 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2872 pref0_len = pref_len - (locinput - loc_reg_starttry);
2873 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2874 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2875 ? (5 + taill) - pref_len : loc_regeol - locinput);
2876 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2880 if (pref0_len > pref_len)
2881 pref0_len = pref_len;
2883 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2885 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2886 (locinput - pref_len),pref0_len, 60, 4, 5);
2888 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2889 (locinput - pref_len + pref0_len),
2890 pref_len - pref0_len, 60, 2, 3);
2892 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2893 locinput, loc_regeol - locinput, 10, 0, 1);
2895 const STRLEN tlen=len0+len1+len2;
2896 PerlIO_printf(Perl_debug_log,
2897 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2898 (IV)(locinput - loc_bostr),
2901 (docolor ? "" : "> <"),
2903 (int)(tlen > 19 ? 0 : 19 - tlen),
2910 /* reg_check_named_buff_matched()
2911 * Checks to see if a named buffer has matched. The data array of
2912 * buffer numbers corresponding to the buffer is expected to reside
2913 * in the regexp->data->data array in the slot stored in the ARG() of
2914 * node involved. Note that this routine doesn't actually care about the
2915 * name, that information is not preserved from compilation to execution.
2916 * Returns the index of the leftmost defined buffer with the given name
2917 * or 0 if non of the buffers matched.
2920 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2923 RXi_GET_DECL(rex,rexi);
2924 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2925 I32 *nums=(I32*)SvPVX(sv_dat);
2927 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2929 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2930 if ((I32)*PL_reglastparen >= nums[n] &&
2931 PL_regoffs[nums[n]].end != -1)
2940 /* free all slabs above current one - called during LEAVE_SCOPE */
2943 S_clear_backtrack_stack(pTHX_ void *p)
2945 regmatch_slab *s = PL_regmatch_slab->next;
2950 PL_regmatch_slab->next = NULL;
2952 regmatch_slab * const osl = s;
2959 #define SETREX(Re1,Re2) \
2960 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2963 STATIC I32 /* 0 failure, 1 success */
2964 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2966 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2970 register const bool utf8_target = PL_reg_match_utf8;
2971 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2972 REGEXP *rex_sv = reginfo->prog;
2973 regexp *rex = (struct regexp *)SvANY(rex_sv);
2974 RXi_GET_DECL(rex,rexi);
2976 /* the current state. This is a cached copy of PL_regmatch_state */
2977 register regmatch_state *st;
2978 /* cache heavy used fields of st in registers */
2979 register regnode *scan;
2980 register regnode *next;
2981 register U32 n = 0; /* general value; init to avoid compiler warning */
2982 register I32 ln = 0; /* len or last; init to avoid compiler warning */
2983 register char *locinput = PL_reginput;
2984 register I32 nextchr; /* is always set to UCHARAT(locinput) */
2986 bool result = 0; /* return value of S_regmatch */
2987 int depth = 0; /* depth of backtrack stack */
2988 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
2989 const U32 max_nochange_depth =
2990 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
2991 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
2992 regmatch_state *yes_state = NULL; /* state to pop to on success of
2994 /* mark_state piggy backs on the yes_state logic so that when we unwind
2995 the stack on success we can update the mark_state as we go */
2996 regmatch_state *mark_state = NULL; /* last mark state we have seen */
2997 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
2998 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3000 bool no_final = 0; /* prevent failure from backtracking? */
3001 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3002 char *startpoint = PL_reginput;
3003 SV *popmark = NULL; /* are we looking for a mark? */
3004 SV *sv_commit = NULL; /* last mark name seen in failure */
3005 SV *sv_yes_mark = NULL; /* last mark name we have seen
3006 during a successfull match */
3007 U32 lastopen = 0; /* last open we saw */
3008 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3009 SV* const oreplsv = GvSV(PL_replgv);
3010 /* these three flags are set by various ops to signal information to
3011 * the very next op. They have a useful lifetime of exactly one loop
3012 * iteration, and are not preserved or restored by state pushes/pops
3014 bool sw = 0; /* the condition value in (?(cond)a|b) */
3015 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3016 int logical = 0; /* the following EVAL is:
3020 or the following IFMATCH/UNLESSM is:
3021 false: plain (?=foo)
3022 true: used as a condition: (?(?=foo))
3025 GET_RE_DEBUG_FLAGS_DECL;
3028 PERL_ARGS_ASSERT_REGMATCH;
3030 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3031 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3033 /* on first ever call to regmatch, allocate first slab */
3034 if (!PL_regmatch_slab) {
3035 Newx(PL_regmatch_slab, 1, regmatch_slab);
3036 PL_regmatch_slab->prev = NULL;
3037 PL_regmatch_slab->next = NULL;
3038 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3041 oldsave = PL_savestack_ix;
3042 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3043 SAVEVPTR(PL_regmatch_slab);
3044 SAVEVPTR(PL_regmatch_state);
3046 /* grab next free state slot */
3047 st = ++PL_regmatch_state;
3048 if (st > SLAB_LAST(PL_regmatch_slab))
3049 st = PL_regmatch_state = S_push_slab(aTHX);
3051 /* Note that nextchr is a byte even in UTF */
3052 nextchr = UCHARAT(locinput);
3054 while (scan != NULL) {
3057 SV * const prop = sv_newmortal();
3058 regnode *rnext=regnext(scan);
3059 DUMP_EXEC_POS( locinput, scan, utf8_target );
3060 regprop(rex, prop, scan);
3062 PerlIO_printf(Perl_debug_log,
3063 "%3"IVdf":%*s%s(%"IVdf")\n",
3064 (IV)(scan - rexi->program), depth*2, "",
3066 (PL_regkind[OP(scan)] == END || !rnext) ?
3067 0 : (IV)(rnext - rexi->program));
3070 next = scan + NEXT_OFF(scan);
3073 state_num = OP(scan);
3075 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3078 assert(PL_reglastparen == &rex->lastparen);
3079 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3080 assert(PL_regoffs == rex->offs);
3082 switch (state_num) {
3084 if (locinput == PL_bostr)
3086 /* reginfo->till = reginfo->bol; */
3091 if (locinput == PL_bostr ||
3092 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3098 if (locinput == PL_bostr)
3102 if (locinput == reginfo->ganch)
3107 /* update the startpoint */
3108 st->u.keeper.val = PL_regoffs[0].start;
3109 PL_reginput = locinput;
3110 PL_regoffs[0].start = locinput - PL_bostr;
3111 PUSH_STATE_GOTO(KEEPS_next, next);
3113 case KEEPS_next_fail:
3114 /* rollback the start point change */
3115 PL_regoffs[0].start = st->u.keeper.val;
3121 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3126 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3128 if (PL_regeol - locinput > 1)
3132 if (PL_regeol != locinput)
3136 if (!nextchr && locinput >= PL_regeol)
3139 locinput += PL_utf8skip[nextchr];
3140 if (locinput > PL_regeol)
3142 nextchr = UCHARAT(locinput);
3145 nextchr = UCHARAT(++locinput);
3148 if (!nextchr && locinput >= PL_regeol)
3150 nextchr = UCHARAT(++locinput);
3153 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3156 locinput += PL_utf8skip[nextchr];
3157 if (locinput > PL_regeol)
3159 nextchr = UCHARAT(locinput);
3162 nextchr = UCHARAT(++locinput);
3166 #define ST st->u.trie
3168 /* In this case the charclass data is available inline so
3169 we can fail fast without a lot of extra overhead.
3171 if (scan->flags == EXACT || !utf8_target) {
3172 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3174 PerlIO_printf(Perl_debug_log,
3175 "%*s %sfailed to match trie start class...%s\n",
3176 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3184 /* the basic plan of execution of the trie is:
3185 * At the beginning, run though all the states, and
3186 * find the longest-matching word. Also remember the position
3187 * of the shortest matching word. For example, this pattern:
3190 * when matched against the string "abcde", will generate
3191 * accept states for all words except 3, with the longest
3192 * matching word being 4, and the shortest being 1 (with
3193 * the position being after char 1 of the string).
3195 * Then for each matching word, in word order (i.e. 1,2,4,5),
3196 * we run the remainder of the pattern; on each try setting
3197 * the current position to the character following the word,
3198 * returning to try the next word on failure.
3200 * We avoid having to build a list of words at runtime by
3201 * using a compile-time structure, wordinfo[].prev, which
3202 * gives, for each word, the previous accepting word (if any).
3203 * In the case above it would contain the mappings 1->2, 2->0,
3204 * 3->0, 4->5, 5->1. We can use this table to generate, from
3205 * the longest word (4 above), a list of all words, by
3206 * following the list of prev pointers; this gives us the
3207 * unordered list 4,5,1,2. Then given the current word we have
3208 * just tried, we can go through the list and find the
3209 * next-biggest word to try (so if we just failed on word 2,
3210 * the next in the list is 4).
3212 * Since at runtime we don't record the matching position in
3213 * the string for each word, we have to work that out for
3214 * each word we're about to process. The wordinfo table holds
3215 * the character length of each word; given that we recorded
3216 * at the start: the position of the shortest word and its
3217 * length in chars, we just need to move the pointer the
3218 * difference between the two char lengths. Depending on
3219 * Unicode status and folding, that's cheap or expensive.
3221 * This algorithm is optimised for the case where are only a
3222 * small number of accept states, i.e. 0,1, or maybe 2.
3223 * With lots of accepts states, and having to try all of them,
3224 * it becomes quadratic on number of accept states to find all
3229 /* what type of TRIE am I? (utf8 makes this contextual) */
3230 DECL_TRIE_TYPE(scan);
3232 /* what trie are we using right now */
3233 reg_trie_data * const trie
3234 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3235 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3236 U32 state = trie->startstate;
3238 if (trie->bitmap && trie_type != trie_utf8_fold &&
3239 !TRIE_BITMAP_TEST(trie,*locinput)
3241 if (trie->states[ state ].wordnum) {
3243 PerlIO_printf(Perl_debug_log,
3244 "%*s %smatched empty string...%s\n",
3245 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3251 PerlIO_printf(Perl_debug_log,
3252 "%*s %sfailed to match trie start class...%s\n",
3253 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3260 U8 *uc = ( U8* )locinput;
3264 U8 *uscan = (U8*)NULL;
3265 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3266 U32 charcount = 0; /* how many input chars we have matched */
3267 U32 accepted = 0; /* have we seen any accepting states? */
3270 ST.jump = trie->jump;
3273 ST.longfold = FALSE; /* char longer if folded => it's harder */
3276 /* fully traverse the TRIE; note the position of the
3277 shortest accept state and the wordnum of the longest
3280 while ( state && uc <= (U8*)PL_regeol ) {
3281 U32 base = trie->states[ state ].trans.base;
3285 wordnum = trie->states[ state ].wordnum;
3287 if (wordnum) { /* it's an accept state */
3290 /* record first match position */
3292 ST.firstpos = (U8*)locinput;
3297 ST.firstchars = charcount;
3300 if (!ST.nextword || wordnum < ST.nextword)
3301 ST.nextword = wordnum;
3302 ST.topword = wordnum;
3305 DEBUG_TRIE_EXECUTE_r({
3306 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3307 PerlIO_printf( Perl_debug_log,
3308 "%*s %sState: %4"UVxf" Accepted: %c ",
3309 2+depth * 2, "", PL_colors[4],
3310 (UV)state, (accepted ? 'Y' : 'N'));
3313 /* read a char and goto next state */
3316 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3317 uscan, len, uvc, charid, foldlen,
3324 base + charid - 1 - trie->uniquecharcount)) >= 0)
3326 && ((U32)offset < trie->lasttrans)
3327 && trie->trans[offset].check == state)
3329 state = trie->trans[offset].next;
3340 DEBUG_TRIE_EXECUTE_r(
3341 PerlIO_printf( Perl_debug_log,
3342 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3343 charid, uvc, (UV)state, PL_colors[5] );
3349 /* calculate total number of accept states */
3354 w = trie->wordinfo[w].prev;
3357 ST.accepted = accepted;
3361 PerlIO_printf( Perl_debug_log,
3362 "%*s %sgot %"IVdf" possible matches%s\n",
3363 REPORT_CODE_OFF + depth * 2, "",
3364 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3366 goto trie_first_try; /* jump into the fail handler */
3370 case TRIE_next_fail: /* we failed - try next alternative */
3372 REGCP_UNWIND(ST.cp);
3373 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3374 PL_regoffs[n].end = -1;
3375 *PL_reglastparen = n;
3377 if (!--ST.accepted) {
3379 PerlIO_printf( Perl_debug_log,
3380 "%*s %sTRIE failed...%s\n",
3381 REPORT_CODE_OFF+depth*2, "",
3388 /* Find next-highest word to process. Note that this code
3389 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3390 register U16 min = 0;
3392 register U16 const nextword = ST.nextword;
3393 register reg_trie_wordinfo * const wordinfo
3394 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3395 for (word=ST.topword; word; word=wordinfo[word].prev) {
3396 if (word > nextword && (!min || word < min))
3409 ST.lastparen = *PL_reglastparen;
3413 /* find start char of end of current word */
3415 U32 chars; /* how many chars to skip */
3416 U8 *uc = ST.firstpos;
3417 reg_trie_data * const trie
3418 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3420 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3422 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3426 /* the hard option - fold each char in turn and find
3427 * its folded length (which may be different */
3428 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3436 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3444 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3449 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3463 PL_reginput = (char *)uc;
3466 scan = (ST.jump && ST.jump[ST.nextword])
3467 ? ST.me + ST.jump[ST.nextword]
3471 PerlIO_printf( Perl_debug_log,
3472 "%*s %sTRIE matched word #%d, continuing%s\n",
3473 REPORT_CODE_OFF+depth*2, "",
3480 if (ST.accepted > 1 || has_cutgroup) {
3481 PUSH_STATE_GOTO(TRIE_next, scan);
3484 /* only one choice left - just continue */
3486 AV *const trie_words
3487 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3488 SV ** const tmp = av_fetch( trie_words,
3490 SV *sv= tmp ? sv_newmortal() : NULL;
3492 PerlIO_printf( Perl_debug_log,
3493 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3494 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3496 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3497 PL_colors[0], PL_colors[1],
3498 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
3500 : "not compiled under -Dr",
3504 locinput = PL_reginput;
3505 nextchr = UCHARAT(locinput);
3506 continue; /* execute rest of RE */
3511 char *s = STRING(scan);
3513 if (utf8_target != UTF_PATTERN) {
3514 /* The target and the pattern have differing utf8ness. */
3516 const char * const e = s + ln;
3519 /* The target is utf8, the pattern is not utf8. */
3524 if (NATIVE_TO_UNI(*(U8*)s) !=
3525 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3533 /* The target is not utf8, the pattern is utf8. */
3538 if (NATIVE_TO_UNI(*((U8*)l)) !=
3539 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3547 nextchr = UCHARAT(locinput);
3550 /* The target and the pattern have the same utf8ness. */
3551 /* Inline the first character, for speed. */
3552 if (UCHARAT(s) != nextchr)
3554 if (PL_regeol - locinput < ln)
3556 if (ln > 1 && memNE(s, locinput, ln))
3559 nextchr = UCHARAT(locinput);
3564 const U8 * fold_array;
3567 PL_reg_flags |= RF_tainted;
3568 folder = foldEQ_locale;
3569 fold_array = PL_fold_locale;
3573 folder = foldEQ_latin1;
3574 fold_array = PL_fold_latin1;
3579 fold_array = PL_fold;
3585 if (utf8_target || UTF_PATTERN) {
3586 /* Either target or the pattern are utf8. */
3587 const char * const l = locinput;
3588 char *e = PL_regeol;
3590 if (! foldEQ_utf8(s, 0, ln, cBOOL(UTF_PATTERN),
3591 l, &e, 0, utf8_target)) {
3592 /* One more case for the sharp s:
3593 * pack("U0U*", 0xDF) =~ /ss/i,
3594 * the 0xC3 0x9F are the UTF-8
3595 * byte sequence for the U+00DF. */
3597 if (!(utf8_target &&
3598 toLOWER(s[0]) == 's' &&
3600 toLOWER(s[1]) == 's' &&
3607 nextchr = UCHARAT(locinput);
3611 /* Neither the target and the pattern are utf8. */
3613 /* Inline the first character, for speed. */
3614 if (UCHARAT(s) != nextchr &&
3615 UCHARAT(s) != fold_array[nextchr])
3619 if (PL_regeol - locinput < ln)
3621 if (ln > 1 && ! folder(s, locinput, ln))
3624 nextchr = UCHARAT(locinput);
3629 PL_reg_flags |= RF_tainted;
3633 /* was last char in word? */
3635 if (locinput == PL_bostr)
3638 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3640 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3642 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3643 ln = isALNUM_uni(ln);
3644 LOAD_UTF8_CHARCLASS_ALNUM();
3645 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3648 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3649 n = isALNUM_LC_utf8((U8*)locinput);
3653 ln = (locinput != PL_bostr) ?
3654 UCHARAT(locinput - 1) : '\n';
3655 if (FLAGS(scan) & USE_UNI) {
3657 /* Here, can't be BOUNDL or NBOUNDL because they never set
3658 * the flags to USE_UNI */
3659 ln = isWORDCHAR_L1(ln);
3660 n = isWORDCHAR_L1(nextchr);
3662 else if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3664 n = isALNUM(nextchr);
3667 ln = isALNUM_LC(ln);
3668 n = isALNUM_LC(nextchr);
3671 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3672 OP(scan) == BOUNDL))
3677 STRLEN inclasslen = PL_regeol - locinput;
3678 if (locinput >= PL_regeol)
3681 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3683 locinput += inclasslen;
3684 nextchr = UCHARAT(locinput);
3689 nextchr = UCHARAT(locinput);
3690 if (!nextchr && locinput >= PL_regeol)
3692 if (!REGINCLASS(rex, scan, (U8*)locinput))
3694 nextchr = UCHARAT(++locinput);
3698 /* If we might have the case of the German sharp s
3699 * in a casefolding Unicode character class. */
3701 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
3702 locinput += SHARP_S_SKIP;
3703 nextchr = UCHARAT(locinput);
3708 /* Special char classes - The defines start on line 129 or so */
3709 CCC_TRY_AFF_U( ALNUM, ALNUML, perl_word, "a", isALNUM_LC_utf8, isWORDCHAR_L1, isALNUM_LC);
3710 CCC_TRY_NEG_U(NALNUM, NALNUML, perl_word, "a", isALNUM_LC_utf8, isWORDCHAR_L1, isALNUM_LC);
3712 CCC_TRY_AFF_U( SPACE, SPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE_L1, isSPACE_LC);
3713 CCC_TRY_NEG_U(NSPACE, NSPACEL, perl_space, " ", isSPACE_LC_utf8, isSPACE_L1, isSPACE_LC);
3715 CCC_TRY_AFF( DIGIT, DIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3716 CCC_TRY_NEG(NDIGIT, NDIGITL, posix_digit, "0", isDIGIT_LC_utf8, isDIGIT, isDIGIT_LC);
3718 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3719 a Unicode extended Grapheme Cluster */
3720 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3721 extended Grapheme Cluster is:
3724 | Prepend* Begin Extend*
3727 Begin is (Hangul-syllable | ! Control)
3728 Extend is (Grapheme_Extend | Spacing_Mark)
3729 Control is [ GCB_Control CR LF ]
3731 The discussion below shows how the code for CLUMP is derived
3732 from this regex. Note that most of these concepts are from
3733 property values of the Grapheme Cluster Boundary (GCB) property.
3734 No code point can have multiple property values for a given
3735 property. Thus a code point in Prepend can't be in Control, but
3736 it must be in !Control. This is why Control above includes
3737 GCB_Control plus CR plus LF. The latter two are used in the GCB
3738 property separately, and so can't be in GCB_Control, even though
3739 they logically are controls. Control is not the same as gc=cc,
3740 but includes format and other characters as well.
3742 The Unicode definition of Hangul-syllable is:
3744 | (L* ( ( V | LV ) V* | LVT ) T*)
3747 Each of these is a value for the GCB property, and hence must be
3748 disjoint, so the order they are tested is immaterial, so the
3749 above can safely be changed to
3752 | (L* ( LVT | ( V | LV ) V*) T*)
3754 The last two terms can be combined like this:
3756 | (( LVT | ( V | LV ) V*) T*))
3758 And refactored into this:
3759 L* (L | LVT T* | V V* T* | LV V* T*)
3761 That means that if we have seen any L's at all we can quit
3762 there, but if the next character is a LVT, a V or and LV we
3765 There is a subtlety with Prepend* which showed up in testing.
3766 Note that the Begin, and only the Begin is required in:
3767 | Prepend* Begin Extend*
3768 Also, Begin contains '! Control'. A Prepend must be a '!
3769 Control', which means it must be a Begin. What it comes down to
3770 is that if we match Prepend* and then find no suitable Begin
3771 afterwards, that if we backtrack the last Prepend, that one will
3772 be a suitable Begin.
3775 if (locinput >= PL_regeol)
3777 if (! utf8_target) {
3779 /* Match either CR LF or '.', as all the other possibilities
3781 locinput++; /* Match the . or CR */
3783 && locinput < PL_regeol
3784 && UCHARAT(locinput) == '\n') locinput++;
3788 /* Utf8: See if is ( CR LF ); already know that locinput <
3789 * PL_regeol, so locinput+1 is in bounds */
3790 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3794 /* In case have to backtrack to beginning, then match '.' */
3795 char *starting = locinput;
3797 /* In case have to backtrack the last prepend */
3798 char *previous_prepend = 0;
3800 LOAD_UTF8_CHARCLASS_GCB();
3802 /* Match (prepend)* */
3803 while (locinput < PL_regeol
3804 && swash_fetch(PL_utf8_X_prepend,
3805 (U8*)locinput, utf8_target))
3807 previous_prepend = locinput;
3808 locinput += UTF8SKIP(locinput);
3811 /* As noted above, if we matched a prepend character, but
3812 * the next thing won't match, back off the last prepend we
3813 * matched, as it is guaranteed to match the begin */
3814 if (previous_prepend
3815 && (locinput >= PL_regeol
3816 || ! swash_fetch(PL_utf8_X_begin,
3817 (U8*)locinput, utf8_target)))
3819 locinput = previous_prepend;
3822 /* Note that here we know PL_regeol > locinput, as we
3823 * tested that upon input to this switch case, and if we
3824 * moved locinput forward, we tested the result just above
3825 * and it either passed, or we backed off so that it will
3827 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3829 /* Here did not match the required 'Begin' in the
3830 * second term. So just match the very first
3831 * character, the '.' of the final term of the regex */
3832 locinput = starting + UTF8SKIP(starting);
3835 /* Here is the beginning of a character that can have
3836 * an extender. It is either a hangul syllable, or a
3838 if (swash_fetch(PL_utf8_X_non_hangul,
3839 (U8*)locinput, utf8_target))
3842 /* Here not a Hangul syllable, must be a
3843 * ('! * Control') */
3844 locinput += UTF8SKIP(locinput);
3847 /* Here is a Hangul syllable. It can be composed
3848 * of several individual characters. One
3849 * possibility is T+ */
3850 if (swash_fetch(PL_utf8_X_T,
3851 (U8*)locinput, utf8_target))
3853 while (locinput < PL_regeol
3854 && swash_fetch(PL_utf8_X_T,
3855 (U8*)locinput, utf8_target))
3857 locinput += UTF8SKIP(locinput);
3861 /* Here, not T+, but is a Hangul. That means
3862 * it is one of the others: L, LV, LVT or V,
3864 * L* (L | LVT T* | V V* T* | LV V* T*) */
3867 while (locinput < PL_regeol
3868 && swash_fetch(PL_utf8_X_L,
3869 (U8*)locinput, utf8_target))
3871 locinput += UTF8SKIP(locinput);
3874 /* Here, have exhausted L*. If the next
3875 * character is not an LV, LVT nor V, it means
3876 * we had to have at least one L, so matches L+
3877 * in the original equation, we have a complete
3878 * hangul syllable. Are done. */
3880 if (locinput < PL_regeol
3881 && swash_fetch(PL_utf8_X_LV_LVT_V,
3882 (U8*)locinput, utf8_target))
3885 /* Otherwise keep going. Must be LV, LVT
3886 * or V. See if LVT */
3887 if (swash_fetch(PL_utf8_X_LVT,
3888 (U8*)locinput, utf8_target))
3890 locinput += UTF8SKIP(locinput);
3893 /* Must be V or LV. Take it, then
3895 locinput += UTF8SKIP(locinput);
3896 while (locinput < PL_regeol
3897 && swash_fetch(PL_utf8_X_V,
3898 (U8*)locinput, utf8_target))
3900 locinput += UTF8SKIP(locinput);
3904 /* And any of LV, LVT, or V can be followed
3906 while (locinput < PL_regeol
3907 && swash_fetch(PL_utf8_X_T,
3911 locinput += UTF8SKIP(locinput);
3917 /* Match any extender */
3918 while (locinput < PL_regeol
3919 && swash_fetch(PL_utf8_X_extend,
3920 (U8*)locinput, utf8_target))
3922 locinput += UTF8SKIP(locinput);
3926 if (locinput > PL_regeol) sayNO;
3928 nextchr = UCHARAT(locinput);
3932 { /* The capture buffer cases. The ones beginning with N for the
3933 named buffers just convert to the equivalent numbered and
3934 pretend they were called as the corresponding numbered buffer
3936 /* don't initialize these, it makes C++ unhappy */
3940 const U8 *fold_array;
3942 folder = NULL; /* NULL assumes will be NREF, REF: no
3946 PL_reg_flags |= RF_tainted;
3947 folder = foldEQ_locale;
3948 fold_array = PL_fold_locale;
3953 folder = foldEQ_latin1;
3954 fold_array = PL_fold_latin1;
3960 fold_array = PL_fold;
3970 /* For the named back references, find the corresponding buffer
3972 n = reg_check_named_buff_matched(rex,scan);
3977 goto do_nref_ref_common;
3980 PL_reg_flags |= RF_tainted;
3981 folder = foldEQ_locale;
3982 fold_array = PL_fold_locale;
3986 folder = foldEQ_latin1;
3987 fold_array = PL_fold_latin1;
3992 fold_array = PL_fold;
4001 n = ARG(scan); /* which paren pair */
4004 ln = PL_regoffs[n].start;
4005 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4006 if (*PL_reglastparen < n || ln == -1)
4007 sayNO; /* Do not match unless seen CLOSEn. */
4008 if (ln == PL_regoffs[n].end)
4012 if (type != REF /* REF can do byte comparison */
4015 && (*s == (char) LATIN_SMALL_LETTER_SHARP_S
4016 || *locinput == (char) LATIN_SMALL_LETTER_SHARP_S))))
4017 { /* XXX handle REFFL better */
4018 char * limit = PL_regeol;
4020 /* This call case insensitively compares the entire buffer
4021 * at s, with the current input starting at locinput, but
4022 * not going off the end given by PL_regeol, and returns in
4023 * limit upon success, how much of the current input was
4025 if (! foldEQ_utf8(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4026 locinput, &limit, 0, utf8_target))
4031 nextchr = UCHARAT(locinput);
4035 /* Not utf8: Inline the first character, for speed. */
4036 if (UCHARAT(s) != nextchr &&
4038 UCHARAT(s) != fold_array[nextchr]))
4040 ln = PL_regoffs[n].end - ln;
4041 if (locinput + ln > PL_regeol)
4043 if (ln > 1 && (type == REF
4044 ? memNE(s, locinput, ln)
4045 : ! folder(s, locinput, ln)))
4048 nextchr = UCHARAT(locinput);
4058 #define ST st->u.eval
4063 regexp_internal *rei;
4064 regnode *startpoint;
4067 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4068 if (cur_eval && cur_eval->locinput==locinput) {
4069 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4070 Perl_croak(aTHX_ "Infinite recursion in regex");
4071 if ( ++nochange_depth > max_nochange_depth )
4073 "Pattern subroutine nesting without pos change"
4074 " exceeded limit in regex");
4081 (void)ReREFCNT_inc(rex_sv);
4082 if (OP(scan)==GOSUB) {
4083 startpoint = scan + ARG2L(scan);
4084 ST.close_paren = ARG(scan);
4086 startpoint = rei->program+1;
4089 goto eval_recurse_doit;
4091 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4092 if (cur_eval && cur_eval->locinput==locinput) {
4093 if ( ++nochange_depth > max_nochange_depth )
4094 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4099 /* execute the code in the {...} */
4101 SV ** const before = SP;
4102 OP_4tree * const oop = PL_op;
4103 COP * const ocurcop = PL_curcop;
4105 char *saved_regeol = PL_regeol;
4106 struct re_save_state saved_state;
4108 /* To not corrupt the existing regex state while executing the
4109 * eval we would normally put it on the save stack, like with
4110 * save_re_context. However, re-evals have a weird scoping so we
4111 * can't just add ENTER/LEAVE here. With that, things like
4113 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4115 * would break, as they expect the localisation to be unwound
4116 * only when the re-engine backtracks through the bit that
4119 * What we do instead is just saving the state in a local c
4122 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4125 PL_op = (OP_4tree*)rexi->data->data[n];
4126 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4127 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4128 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4129 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4132 SV *sv_mrk = get_sv("REGMARK", 1);
4133 sv_setsv(sv_mrk, sv_yes_mark);
4136 CALLRUNOPS(aTHX); /* Scalar context. */
4139 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4145 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4148 PAD_RESTORE_LOCAL(old_comppad);
4149 PL_curcop = ocurcop;
4150 PL_regeol = saved_regeol;
4153 sv_setsv(save_scalar(PL_replgv), ret);
4157 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4160 /* extract RE object from returned value; compiling if
4166 SV *const sv = SvRV(ret);
4168 if (SvTYPE(sv) == SVt_REGEXP) {
4170 } else if (SvSMAGICAL(sv)) {
4171 mg = mg_find(sv, PERL_MAGIC_qr);
4174 } else if (SvTYPE(ret) == SVt_REGEXP) {
4176 } else if (SvSMAGICAL(ret)) {
4177 if (SvGMAGICAL(ret)) {
4178 /* I don't believe that there is ever qr magic
4180 assert(!mg_find(ret, PERL_MAGIC_qr));
4181 sv_unmagic(ret, PERL_MAGIC_qr);
4184 mg = mg_find(ret, PERL_MAGIC_qr);
4185 /* testing suggests mg only ends up non-NULL for
4186 scalars who were upgraded and compiled in the
4187 else block below. In turn, this is only
4188 triggered in the "postponed utf8 string" tests
4194 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4198 rx = reg_temp_copy(NULL, rx);
4202 const I32 osize = PL_regsize;
4205 assert (SvUTF8(ret));
4206 } else if (SvUTF8(ret)) {
4207 /* Not doing UTF-8, despite what the SV says. Is
4208 this only if we're trapped in use 'bytes'? */
4209 /* Make a copy of the octet sequence, but without
4210 the flag on, as the compiler now honours the
4211 SvUTF8 flag on ret. */
4213 const char *const p = SvPV(ret, len);
4214 ret = newSVpvn_flags(p, len, SVs_TEMP);
4216 rx = CALLREGCOMP(ret, pm_flags);
4218 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4220 /* This isn't a first class regexp. Instead, it's
4221 caching a regexp onto an existing, Perl visible
4223 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4228 re = (struct regexp *)SvANY(rx);
4230 RXp_MATCH_COPIED_off(re);
4231 re->subbeg = rex->subbeg;
4232 re->sublen = rex->sublen;
4235 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4236 "Matching embedded");
4238 startpoint = rei->program + 1;
4239 ST.close_paren = 0; /* only used for GOSUB */
4240 /* borrowed from regtry */
4241 if (PL_reg_start_tmpl <= re->nparens) {
4242 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4243 if(PL_reg_start_tmp)
4244 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4246 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4249 eval_recurse_doit: /* Share code with GOSUB below this line */
4250 /* run the pattern returned from (??{...}) */
4251 ST.cp = regcppush(0); /* Save *all* the positions. */
4252 REGCP_SET(ST.lastcp);
4254 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4256 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4257 PL_reglastparen = &re->lastparen;
4258 PL_reglastcloseparen = &re->lastcloseparen;
4260 re->lastcloseparen = 0;
4262 PL_reginput = locinput;
4265 /* XXXX This is too dramatic a measure... */
4268 ST.toggle_reg_flags = PL_reg_flags;
4270 PL_reg_flags |= RF_utf8;
4272 PL_reg_flags &= ~RF_utf8;
4273 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4275 ST.prev_rex = rex_sv;
4276 ST.prev_curlyx = cur_curlyx;
4277 SETREX(rex_sv,re_sv);
4282 ST.prev_eval = cur_eval;
4284 /* now continue from first node in postoned RE */
4285 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4288 /* logical is 1, /(?(?{...})X|Y)/ */
4289 sw = cBOOL(SvTRUE(ret));
4294 case EVAL_AB: /* cleanup after a successful (??{A})B */
4295 /* note: this is called twice; first after popping B, then A */
4296 PL_reg_flags ^= ST.toggle_reg_flags;
4297 ReREFCNT_dec(rex_sv);
4298 SETREX(rex_sv,ST.prev_rex);
4299 rex = (struct regexp *)SvANY(rex_sv);
4300 rexi = RXi_GET(rex);
4302 cur_eval = ST.prev_eval;
4303 cur_curlyx = ST.prev_curlyx;
4305 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4306 PL_reglastparen = &rex->lastparen;
4307 PL_reglastcloseparen = &rex->lastcloseparen;
4308 /* also update PL_regoffs */
4309 PL_regoffs = rex->offs;
4311 /* XXXX This is too dramatic a measure... */
4313 if ( nochange_depth )
4318 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4319 /* note: this is called twice; first after popping B, then A */
4320 PL_reg_flags ^= ST.toggle_reg_flags;
4321 ReREFCNT_dec(rex_sv);
4322 SETREX(rex_sv,ST.prev_rex);
4323 rex = (struct regexp *)SvANY(rex_sv);
4324 rexi = RXi_GET(rex);
4325 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4326 PL_reglastparen = &rex->lastparen;
4327 PL_reglastcloseparen = &rex->lastcloseparen;
4329 PL_reginput = locinput;
4330 REGCP_UNWIND(ST.lastcp);
4332 cur_eval = ST.prev_eval;
4333 cur_curlyx = ST.prev_curlyx;
4334 /* XXXX This is too dramatic a measure... */
4336 if ( nochange_depth )
4342 n = ARG(scan); /* which paren pair */
4343 PL_reg_start_tmp[n] = locinput;
4349 n = ARG(scan); /* which paren pair */
4350 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4351 PL_regoffs[n].end = locinput - PL_bostr;
4352 /*if (n > PL_regsize)
4354 if (n > *PL_reglastparen)
4355 *PL_reglastparen = n;
4356 *PL_reglastcloseparen = n;
4357 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4365 cursor && OP(cursor)!=END;
4366 cursor=regnext(cursor))
4368 if ( OP(cursor)==CLOSE ){
4370 if ( n <= lastopen ) {
4372 = PL_reg_start_tmp[n] - PL_bostr;
4373 PL_regoffs[n].end = locinput - PL_bostr;
4374 /*if (n > PL_regsize)
4376 if (n > *PL_reglastparen)
4377 *PL_reglastparen = n;
4378 *PL_reglastcloseparen = n;
4379 if ( n == ARG(scan) || (cur_eval &&
4380 cur_eval->u.eval.close_paren == n))
4389 n = ARG(scan); /* which paren pair */
4390 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4393 /* reg_check_named_buff_matched returns 0 for no match */
4394 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4398 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4404 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4406 next = NEXTOPER(NEXTOPER(scan));
4408 next = scan + ARG(scan);
4409 if (OP(next) == IFTHEN) /* Fake one. */
4410 next = NEXTOPER(NEXTOPER(next));
4414 logical = scan->flags;
4417 /*******************************************************************
4419 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4420 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4421 STAR/PLUS/CURLY/CURLYN are used instead.)
4423 A*B is compiled as <CURLYX><A><WHILEM><B>
4425 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4426 state, which contains the current count, initialised to -1. It also sets
4427 cur_curlyx to point to this state, with any previous value saved in the
4430 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4431 since the pattern may possibly match zero times (i.e. it's a while {} loop
4432 rather than a do {} while loop).
4434 Each entry to WHILEM represents a successful match of A. The count in the
4435 CURLYX block is incremented, another WHILEM state is pushed, and execution
4436 passes to A or B depending on greediness and the current count.
4438 For example, if matching against the string a1a2a3b (where the aN are
4439 substrings that match /A/), then the match progresses as follows: (the
4440 pushed states are interspersed with the bits of strings matched so far):
4443 <CURLYX cnt=0><WHILEM>
4444 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4445 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4446 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4447 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4449 (Contrast this with something like CURLYM, which maintains only a single
4453 a1 <CURLYM cnt=1> a2
4454 a1 a2 <CURLYM cnt=2> a3
4455 a1 a2 a3 <CURLYM cnt=3> b
4458 Each WHILEM state block marks a point to backtrack to upon partial failure
4459 of A or B, and also contains some minor state data related to that
4460 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4461 overall state, such as the count, and pointers to the A and B ops.
4463 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4464 must always point to the *current* CURLYX block, the rules are:
4466 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4467 and set cur_curlyx to point the new block.
4469 When popping the CURLYX block after a successful or unsuccessful match,
4470 restore the previous cur_curlyx.
4472 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4473 to the outer one saved in the CURLYX block.
4475 When popping the WHILEM block after a successful or unsuccessful B match,
4476 restore the previous cur_curlyx.
4478 Here's an example for the pattern (AI* BI)*BO
4479 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4482 curlyx backtrack stack
4483 ------ ---------------
4485 CO <CO prev=NULL> <WO>
4486 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4487 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4488 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4490 At this point the pattern succeeds, and we work back down the stack to
4491 clean up, restoring as we go:
4493 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4494 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4495 CO <CO prev=NULL> <WO>
4498 *******************************************************************/
4500 #define ST st->u.curlyx
4502 case CURLYX: /* start of /A*B/ (for complex A) */
4504 /* No need to save/restore up to this paren */
4505 I32 parenfloor = scan->flags;
4507 assert(next); /* keep Coverity happy */
4508 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4511 /* XXXX Probably it is better to teach regpush to support
4512 parenfloor > PL_regsize... */
4513 if (parenfloor > (I32)*PL_reglastparen)
4514 parenfloor = *PL_reglastparen; /* Pessimization... */
4516 ST.prev_curlyx= cur_curlyx;
4518 ST.cp = PL_savestack_ix;
4520 /* these fields contain the state of the current curly.
4521 * they are accessed by subsequent WHILEMs */
4522 ST.parenfloor = parenfloor;
4527 ST.count = -1; /* this will be updated by WHILEM */
4528 ST.lastloc = NULL; /* this will be updated by WHILEM */
4530 PL_reginput = locinput;
4531 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4535 case CURLYX_end: /* just finished matching all of A*B */
4536 cur_curlyx = ST.prev_curlyx;
4540 case CURLYX_end_fail: /* just failed to match all of A*B */
4542 cur_curlyx = ST.prev_curlyx;
4548 #define ST st->u.whilem
4550 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4552 /* see the discussion above about CURLYX/WHILEM */
4554 int min = ARG1(cur_curlyx->u.curlyx.me);
4555 int max = ARG2(cur_curlyx->u.curlyx.me);
4556 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4558 assert(cur_curlyx); /* keep Coverity happy */
4559 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4560 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4561 ST.cache_offset = 0;
4564 PL_reginput = locinput;
4566 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4567 "%*s whilem: matched %ld out of %d..%d\n",
4568 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4571 /* First just match a string of min A's. */
4574 cur_curlyx->u.curlyx.lastloc = locinput;
4575 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4579 /* If degenerate A matches "", assume A done. */
4581 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4582 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4583 "%*s whilem: empty match detected, trying continuation...\n",
4584 REPORT_CODE_OFF+depth*2, "")
4586 goto do_whilem_B_max;
4589 /* super-linear cache processing */
4593 if (!PL_reg_maxiter) {
4594 /* start the countdown: Postpone detection until we
4595 * know the match is not *that* much linear. */
4596 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4597 /* possible overflow for long strings and many CURLYX's */
4598 if (PL_reg_maxiter < 0)
4599 PL_reg_maxiter = I32_MAX;
4600 PL_reg_leftiter = PL_reg_maxiter;
4603 if (PL_reg_leftiter-- == 0) {
4604 /* initialise cache */
4605 const I32 size = (PL_reg_maxiter + 7)/8;
4606 if (PL_reg_poscache) {
4607 if ((I32)PL_reg_poscache_size < size) {
4608 Renew(PL_reg_poscache, size, char);
4609 PL_reg_poscache_size = size;
4611 Zero(PL_reg_poscache, size, char);
4614 PL_reg_poscache_size = size;
4615 Newxz(PL_reg_poscache, size, char);
4617 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4618 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4619 PL_colors[4], PL_colors[5])
4623 if (PL_reg_leftiter < 0) {
4624 /* have we already failed at this position? */
4626 offset = (scan->flags & 0xf) - 1
4627 + (locinput - PL_bostr) * (scan->flags>>4);
4628 mask = 1 << (offset % 8);
4630 if (PL_reg_poscache[offset] & mask) {
4631 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4632 "%*s whilem: (cache) already tried at this position...\n",
4633 REPORT_CODE_OFF+depth*2, "")
4635 sayNO; /* cache records failure */
4637 ST.cache_offset = offset;
4638 ST.cache_mask = mask;
4642 /* Prefer B over A for minimal matching. */
4644 if (cur_curlyx->u.curlyx.minmod) {
4645 ST.save_curlyx = cur_curlyx;
4646 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4647 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4648 REGCP_SET(ST.lastcp);
4649 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4653 /* Prefer A over B for maximal matching. */
4655 if (n < max) { /* More greed allowed? */
4656 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4657 cur_curlyx->u.curlyx.lastloc = locinput;
4658 REGCP_SET(ST.lastcp);
4659 PUSH_STATE_GOTO(WHILEM_A_max, A);
4662 goto do_whilem_B_max;
4666 case WHILEM_B_min: /* just matched B in a minimal match */
4667 case WHILEM_B_max: /* just matched B in a maximal match */
4668 cur_curlyx = ST.save_curlyx;
4672 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4673 cur_curlyx = ST.save_curlyx;
4674 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4675 cur_curlyx->u.curlyx.count--;
4679 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4680 REGCP_UNWIND(ST.lastcp);
4683 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4684 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4685 cur_curlyx->u.curlyx.count--;
4689 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4690 REGCP_UNWIND(ST.lastcp);
4691 regcppop(rex); /* Restore some previous $<digit>s? */
4692 PL_reginput = locinput;
4693 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4694 "%*s whilem: failed, trying continuation...\n",
4695 REPORT_CODE_OFF+depth*2, "")
4698 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4699 && ckWARN(WARN_REGEXP)
4700 && !(PL_reg_flags & RF_warned))
4702 PL_reg_flags |= RF_warned;
4703 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4704 "Complex regular subexpression recursion",
4709 ST.save_curlyx = cur_curlyx;
4710 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4711 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4714 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4715 cur_curlyx = ST.save_curlyx;
4716 REGCP_UNWIND(ST.lastcp);
4719 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4720 /* Maximum greed exceeded */
4721 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4722 && ckWARN(WARN_REGEXP)
4723 && !(PL_reg_flags & RF_warned))
4725 PL_reg_flags |= RF_warned;
4726 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4727 "%s limit (%d) exceeded",
4728 "Complex regular subexpression recursion",
4731 cur_curlyx->u.curlyx.count--;
4735 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4736 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4738 /* Try grabbing another A and see if it helps. */
4739 PL_reginput = locinput;
4740 cur_curlyx->u.curlyx.lastloc = locinput;
4741 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4742 REGCP_SET(ST.lastcp);
4743 PUSH_STATE_GOTO(WHILEM_A_min,
4744 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4748 #define ST st->u.branch
4750 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4751 next = scan + ARG(scan);
4754 scan = NEXTOPER(scan);
4757 case BRANCH: /* /(...|A|...)/ */
4758 scan = NEXTOPER(scan); /* scan now points to inner node */
4759 ST.lastparen = *PL_reglastparen;
4760 ST.next_branch = next;
4762 PL_reginput = locinput;
4764 /* Now go into the branch */
4766 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4768 PUSH_STATE_GOTO(BRANCH_next, scan);
4772 PL_reginput = locinput;
4773 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4774 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4775 PUSH_STATE_GOTO(CUTGROUP_next,next);
4777 case CUTGROUP_next_fail:
4780 if (st->u.mark.mark_name)
4781 sv_commit = st->u.mark.mark_name;
4787 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4792 REGCP_UNWIND(ST.cp);
4793 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4794 PL_regoffs[n].end = -1;
4795 *PL_reglastparen = n;
4796 /*dmq: *PL_reglastcloseparen = n; */
4797 scan = ST.next_branch;
4798 /* no more branches? */
4799 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4801 PerlIO_printf( Perl_debug_log,
4802 "%*s %sBRANCH failed...%s\n",
4803 REPORT_CODE_OFF+depth*2, "",
4809 continue; /* execute next BRANCH[J] op */
4817 #define ST st->u.curlym
4819 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4821 /* This is an optimisation of CURLYX that enables us to push
4822 * only a single backtracking state, no matter how many matches
4823 * there are in {m,n}. It relies on the pattern being constant
4824 * length, with no parens to influence future backrefs
4828 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4830 /* if paren positive, emulate an OPEN/CLOSE around A */
4832 U32 paren = ST.me->flags;
4833 if (paren > PL_regsize)
4835 if (paren > *PL_reglastparen)
4836 *PL_reglastparen = paren;
4837 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4845 ST.c1 = CHRTEST_UNINIT;
4848 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4851 curlym_do_A: /* execute the A in /A{m,n}B/ */
4852 PL_reginput = locinput;
4853 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4856 case CURLYM_A: /* we've just matched an A */
4857 locinput = st->locinput;
4858 nextchr = UCHARAT(locinput);
4861 /* after first match, determine A's length: u.curlym.alen */
4862 if (ST.count == 1) {
4863 if (PL_reg_match_utf8) {
4865 while (s < PL_reginput) {
4871 ST.alen = PL_reginput - locinput;
4874 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4877 PerlIO_printf(Perl_debug_log,
4878 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4879 (int)(REPORT_CODE_OFF+(depth*2)), "",
4880 (IV) ST.count, (IV)ST.alen)
4883 locinput = PL_reginput;
4885 if (cur_eval && cur_eval->u.eval.close_paren &&
4886 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4890 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4891 if ( max == REG_INFTY || ST.count < max )
4892 goto curlym_do_A; /* try to match another A */
4894 goto curlym_do_B; /* try to match B */
4896 case CURLYM_A_fail: /* just failed to match an A */
4897 REGCP_UNWIND(ST.cp);
4899 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4900 || (cur_eval && cur_eval->u.eval.close_paren &&
4901 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4904 curlym_do_B: /* execute the B in /A{m,n}B/ */
4905 PL_reginput = locinput;
4906 if (ST.c1 == CHRTEST_UNINIT) {
4907 /* calculate c1 and c2 for possible match of 1st char
4908 * following curly */
4909 ST.c1 = ST.c2 = CHRTEST_VOID;
4910 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4911 regnode *text_node = ST.B;
4912 if (! HAS_TEXT(text_node))
4913 FIND_NEXT_IMPT(text_node);
4916 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4918 But the former is redundant in light of the latter.
4920 if this changes back then the macro for
4921 IS_TEXT and friends need to change.
4923 if (PL_regkind[OP(text_node)] == EXACT)
4926 ST.c1 = (U8)*STRING(text_node);
4927 switch (OP(text_node)) {
4928 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
4929 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
4930 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
4931 default: ST.c2 = ST.c1;
4938 PerlIO_printf(Perl_debug_log,
4939 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
4940 (int)(REPORT_CODE_OFF+(depth*2)),
4943 if (ST.c1 != CHRTEST_VOID
4944 && UCHARAT(PL_reginput) != ST.c1
4945 && UCHARAT(PL_reginput) != ST.c2)
4947 /* simulate B failing */
4949 PerlIO_printf(Perl_debug_log,
4950 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
4951 (int)(REPORT_CODE_OFF+(depth*2)),"",
4954 state_num = CURLYM_B_fail;
4955 goto reenter_switch;
4959 /* mark current A as captured */
4960 I32 paren = ST.me->flags;
4962 PL_regoffs[paren].start
4963 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
4964 PL_regoffs[paren].end = PL_reginput - PL_bostr;
4965 /*dmq: *PL_reglastcloseparen = paren; */
4968 PL_regoffs[paren].end = -1;
4969 if (cur_eval && cur_eval->u.eval.close_paren &&
4970 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4979 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
4982 case CURLYM_B_fail: /* just failed to match a B */
4983 REGCP_UNWIND(ST.cp);
4985 I32 max = ARG2(ST.me);
4986 if (max != REG_INFTY && ST.count == max)
4988 goto curlym_do_A; /* try to match a further A */
4990 /* backtrack one A */
4991 if (ST.count == ARG1(ST.me) /* min */)
4994 locinput = HOPc(locinput, -ST.alen);
4995 goto curlym_do_B; /* try to match B */
4998 #define ST st->u.curly
5000 #define CURLY_SETPAREN(paren, success) \
5003 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5004 PL_regoffs[paren].end = locinput - PL_bostr; \
5005 *PL_reglastcloseparen = paren; \
5008 PL_regoffs[paren].end = -1; \
5011 case STAR: /* /A*B/ where A is width 1 */
5015 scan = NEXTOPER(scan);
5017 case PLUS: /* /A+B/ where A is width 1 */
5021 scan = NEXTOPER(scan);
5023 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5024 ST.paren = scan->flags; /* Which paren to set */
5025 if (ST.paren > PL_regsize)
5026 PL_regsize = ST.paren;
5027 if (ST.paren > *PL_reglastparen)
5028 *PL_reglastparen = ST.paren;
5029 ST.min = ARG1(scan); /* min to match */
5030 ST.max = ARG2(scan); /* max to match */
5031 if (cur_eval && cur_eval->u.eval.close_paren &&
5032 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5036 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5038 case CURLY: /* /A{m,n}B/ where A is width 1 */
5040 ST.min = ARG1(scan); /* min to match */
5041 ST.max = ARG2(scan); /* max to match */
5042 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5045 * Lookahead to avoid useless match attempts
5046 * when we know what character comes next.
5048 * Used to only do .*x and .*?x, but now it allows
5049 * for )'s, ('s and (?{ ... })'s to be in the way
5050 * of the quantifier and the EXACT-like node. -- japhy
5053 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5055 if (HAS_TEXT(next) || JUMPABLE(next)) {
5057 regnode *text_node = next;
5059 if (! HAS_TEXT(text_node))
5060 FIND_NEXT_IMPT(text_node);
5062 if (! HAS_TEXT(text_node))
5063 ST.c1 = ST.c2 = CHRTEST_VOID;
5065 if ( PL_regkind[OP(text_node)] != EXACT ) {
5066 ST.c1 = ST.c2 = CHRTEST_VOID;
5067 goto assume_ok_easy;
5070 s = (U8*)STRING(text_node);
5072 /* Currently we only get here when
5074 PL_rekind[OP(text_node)] == EXACT
5076 if this changes back then the macro for IS_TEXT and
5077 friends need to change. */
5080 switch (OP(text_node)) {
5081 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5082 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5083 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5084 default: ST.c2 = ST.c1; break;
5087 else { /* UTF_PATTERN */
5088 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5089 STRLEN ulen1, ulen2;
5090 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5091 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5093 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5094 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5096 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5098 0 : UTF8_ALLOW_ANY);
5099 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5101 0 : UTF8_ALLOW_ANY);
5103 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5105 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5110 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5117 ST.c1 = ST.c2 = CHRTEST_VOID;
5122 PL_reginput = locinput;
5125 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5128 locinput = PL_reginput;
5130 if (ST.c1 == CHRTEST_VOID)
5131 goto curly_try_B_min;
5133 ST.oldloc = locinput;
5135 /* set ST.maxpos to the furthest point along the
5136 * string that could possibly match */
5137 if (ST.max == REG_INFTY) {
5138 ST.maxpos = PL_regeol - 1;
5140 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5143 else if (utf8_target) {
5144 int m = ST.max - ST.min;
5145 for (ST.maxpos = locinput;
5146 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5147 ST.maxpos += UTF8SKIP(ST.maxpos);
5150 ST.maxpos = locinput + ST.max - ST.min;
5151 if (ST.maxpos >= PL_regeol)
5152 ST.maxpos = PL_regeol - 1;
5154 goto curly_try_B_min_known;
5158 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5159 locinput = PL_reginput;
5160 if (ST.count < ST.min)
5162 if ((ST.count > ST.min)
5163 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5165 /* A{m,n} must come at the end of the string, there's
5166 * no point in backing off ... */
5168 /* ...except that $ and \Z can match before *and* after
5169 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5170 We may back off by one in this case. */
5171 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5175 goto curly_try_B_max;
5180 case CURLY_B_min_known_fail:
5181 /* failed to find B in a non-greedy match where c1,c2 valid */
5182 if (ST.paren && ST.count)
5183 PL_regoffs[ST.paren].end = -1;
5185 PL_reginput = locinput; /* Could be reset... */
5186 REGCP_UNWIND(ST.cp);
5187 /* Couldn't or didn't -- move forward. */
5188 ST.oldloc = locinput;
5190 locinput += UTF8SKIP(locinput);
5194 curly_try_B_min_known:
5195 /* find the next place where 'B' could work, then call B */
5199 n = (ST.oldloc == locinput) ? 0 : 1;
5200 if (ST.c1 == ST.c2) {
5202 /* set n to utf8_distance(oldloc, locinput) */
5203 while (locinput <= ST.maxpos &&
5204 utf8n_to_uvchr((U8*)locinput,
5205 UTF8_MAXBYTES, &len,
5206 uniflags) != (UV)ST.c1) {
5212 /* set n to utf8_distance(oldloc, locinput) */
5213 while (locinput <= ST.maxpos) {
5215 const UV c = utf8n_to_uvchr((U8*)locinput,
5216 UTF8_MAXBYTES, &len,
5218 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5226 if (ST.c1 == ST.c2) {
5227 while (locinput <= ST.maxpos &&
5228 UCHARAT(locinput) != ST.c1)
5232 while (locinput <= ST.maxpos
5233 && UCHARAT(locinput) != ST.c1
5234 && UCHARAT(locinput) != ST.c2)
5237 n = locinput - ST.oldloc;
5239 if (locinput > ST.maxpos)
5241 /* PL_reginput == oldloc now */
5244 if (regrepeat(rex, ST.A, n, depth) < n)
5247 PL_reginput = locinput;
5248 CURLY_SETPAREN(ST.paren, ST.count);
5249 if (cur_eval && cur_eval->u.eval.close_paren &&
5250 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5253 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5258 case CURLY_B_min_fail:
5259 /* failed to find B in a non-greedy match where c1,c2 invalid */
5260 if (ST.paren && ST.count)
5261 PL_regoffs[ST.paren].end = -1;
5263 REGCP_UNWIND(ST.cp);
5264 /* failed -- move forward one */
5265 PL_reginput = locinput;
5266 if (regrepeat(rex, ST.A, 1, depth)) {
5268 locinput = PL_reginput;
5269 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5270 ST.count > 0)) /* count overflow ? */
5273 CURLY_SETPAREN(ST.paren, ST.count);
5274 if (cur_eval && cur_eval->u.eval.close_paren &&
5275 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5278 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5286 /* a successful greedy match: now try to match B */
5287 if (cur_eval && cur_eval->u.eval.close_paren &&
5288 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5293 if (ST.c1 != CHRTEST_VOID)
5294 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5295 UTF8_MAXBYTES, 0, uniflags)
5296 : (UV) UCHARAT(PL_reginput);
5297 /* If it could work, try it. */
5298 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5299 CURLY_SETPAREN(ST.paren, ST.count);
5300 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5305 case CURLY_B_max_fail:
5306 /* failed to find B in a greedy match */
5307 if (ST.paren && ST.count)
5308 PL_regoffs[ST.paren].end = -1;
5310 REGCP_UNWIND(ST.cp);
5312 if (--ST.count < ST.min)
5314 PL_reginput = locinput = HOPc(locinput, -1);
5315 goto curly_try_B_max;
5322 /* we've just finished A in /(??{A})B/; now continue with B */
5324 st->u.eval.toggle_reg_flags
5325 = cur_eval->u.eval.toggle_reg_flags;
5326 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5328 st->u.eval.prev_rex = rex_sv; /* inner */
5329 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5330 rex = (struct regexp *)SvANY(rex_sv);
5331 rexi = RXi_GET(rex);
5332 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5333 ReREFCNT_inc(rex_sv);
5334 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5336 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5337 PL_reglastparen = &rex->lastparen;
5338 PL_reglastcloseparen = &rex->lastcloseparen;
5340 REGCP_SET(st->u.eval.lastcp);
5341 PL_reginput = locinput;
5343 /* Restore parens of the outer rex without popping the
5345 tmpix = PL_savestack_ix;
5346 PL_savestack_ix = cur_eval->u.eval.lastcp;
5348 PL_savestack_ix = tmpix;
5350 st->u.eval.prev_eval = cur_eval;
5351 cur_eval = cur_eval->u.eval.prev_eval;
5353 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5354 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5355 if ( nochange_depth )
5358 PUSH_YES_STATE_GOTO(EVAL_AB,
5359 st->u.eval.prev_eval->u.eval.B); /* match B */
5362 if (locinput < reginfo->till) {
5363 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5364 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5366 (long)(locinput - PL_reg_starttry),
5367 (long)(reginfo->till - PL_reg_starttry),
5370 sayNO_SILENT; /* Cannot match: too short. */
5372 PL_reginput = locinput; /* put where regtry can find it */
5373 sayYES; /* Success! */
5375 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5377 PerlIO_printf(Perl_debug_log,
5378 "%*s %ssubpattern success...%s\n",
5379 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5380 PL_reginput = locinput; /* put where regtry can find it */
5381 sayYES; /* Success! */
5384 #define ST st->u.ifmatch
5386 case SUSPEND: /* (?>A) */
5388 PL_reginput = locinput;
5391 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5393 goto ifmatch_trivial_fail_test;
5395 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5397 ifmatch_trivial_fail_test:
5399 char * const s = HOPBACKc(locinput, scan->flags);
5404 sw = 1 - cBOOL(ST.wanted);
5408 next = scan + ARG(scan);
5416 PL_reginput = locinput;
5420 ST.logical = logical;
5421 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5423 /* execute body of (?...A) */
5424 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5427 case IFMATCH_A_fail: /* body of (?...A) failed */
5428 ST.wanted = !ST.wanted;
5431 case IFMATCH_A: /* body of (?...A) succeeded */
5433 sw = cBOOL(ST.wanted);
5435 else if (!ST.wanted)
5438 if (OP(ST.me) == SUSPEND)
5439 locinput = PL_reginput;
5441 locinput = PL_reginput = st->locinput;
5442 nextchr = UCHARAT(locinput);
5444 scan = ST.me + ARG(ST.me);
5447 continue; /* execute B */
5452 next = scan + ARG(scan);
5457 reginfo->cutpoint = PL_regeol;
5460 PL_reginput = locinput;
5462 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5463 PUSH_STATE_GOTO(COMMIT_next,next);
5465 case COMMIT_next_fail:
5472 #define ST st->u.mark
5474 ST.prev_mark = mark_state;
5475 ST.mark_name = sv_commit = sv_yes_mark
5476 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5478 ST.mark_loc = PL_reginput = locinput;
5479 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5481 case MARKPOINT_next:
5482 mark_state = ST.prev_mark;
5485 case MARKPOINT_next_fail:
5486 if (popmark && sv_eq(ST.mark_name,popmark))
5488 if (ST.mark_loc > startpoint)
5489 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5490 popmark = NULL; /* we found our mark */
5491 sv_commit = ST.mark_name;
5494 PerlIO_printf(Perl_debug_log,
5495 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5496 REPORT_CODE_OFF+depth*2, "",
5497 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5500 mark_state = ST.prev_mark;
5501 sv_yes_mark = mark_state ?
5502 mark_state->u.mark.mark_name : NULL;
5506 PL_reginput = locinput;
5508 /* (*SKIP) : if we fail we cut here*/
5509 ST.mark_name = NULL;
5510 ST.mark_loc = locinput;
5511 PUSH_STATE_GOTO(SKIP_next,next);
5513 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5514 otherwise do nothing. Meaning we need to scan
5516 regmatch_state *cur = mark_state;
5517 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5520 if ( sv_eq( cur->u.mark.mark_name,
5523 ST.mark_name = find;
5524 PUSH_STATE_GOTO( SKIP_next, next );
5526 cur = cur->u.mark.prev_mark;
5529 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5531 case SKIP_next_fail:
5533 /* (*CUT:NAME) - Set up to search for the name as we
5534 collapse the stack*/
5535 popmark = ST.mark_name;
5537 /* (*CUT) - No name, we cut here.*/
5538 if (ST.mark_loc > startpoint)
5539 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5540 /* but we set sv_commit to latest mark_name if there
5541 is one so they can test to see how things lead to this
5544 sv_commit=mark_state->u.mark.mark_name;
5552 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5554 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5557 U8 folded[UTF8_MAXBYTES_CASE+1];
5559 const char * const l = locinput;
5560 char *e = PL_regeol;
5561 to_uni_fold(n, folded, &foldlen);
5563 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5564 l, &e, 0, utf8_target)) {
5569 nextchr = UCHARAT(locinput);
5572 if ((n=is_LNBREAK(locinput,utf8_target))) {
5574 nextchr = UCHARAT(locinput);
5579 #define CASE_CLASS(nAmE) \
5581 if ((n=is_##nAmE(locinput,utf8_target))) { \
5583 nextchr = UCHARAT(locinput); \
5588 if ((n=is_##nAmE(locinput,utf8_target))) { \
5591 locinput += UTF8SKIP(locinput); \
5592 nextchr = UCHARAT(locinput); \
5597 CASE_CLASS(HORIZWS);
5601 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5602 PTR2UV(scan), OP(scan));
5603 Perl_croak(aTHX_ "regexp memory corruption");
5607 /* switch break jumps here */
5608 scan = next; /* prepare to execute the next op and ... */
5609 continue; /* ... jump back to the top, reusing st */
5613 /* push a state that backtracks on success */
5614 st->u.yes.prev_yes_state = yes_state;
5618 /* push a new regex state, then continue at scan */
5620 regmatch_state *newst;
5623 regmatch_state *cur = st;
5624 regmatch_state *curyes = yes_state;
5626 regmatch_slab *slab = PL_regmatch_slab;
5627 for (;curd > -1;cur--,curd--) {
5628 if (cur < SLAB_FIRST(slab)) {
5630 cur = SLAB_LAST(slab);
5632 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5633 REPORT_CODE_OFF + 2 + depth * 2,"",
5634 curd, PL_reg_name[cur->resume_state],
5635 (curyes == cur) ? "yes" : ""
5638 curyes = cur->u.yes.prev_yes_state;
5641 DEBUG_STATE_pp("push")
5644 st->locinput = locinput;
5646 if (newst > SLAB_LAST(PL_regmatch_slab))
5647 newst = S_push_slab(aTHX);
5648 PL_regmatch_state = newst;
5650 locinput = PL_reginput;
5651 nextchr = UCHARAT(locinput);
5659 * We get here only if there's trouble -- normally "case END" is
5660 * the terminating point.
5662 Perl_croak(aTHX_ "corrupted regexp pointers");
5668 /* we have successfully completed a subexpression, but we must now
5669 * pop to the state marked by yes_state and continue from there */
5670 assert(st != yes_state);
5672 while (st != yes_state) {
5674 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5675 PL_regmatch_slab = PL_regmatch_slab->prev;
5676 st = SLAB_LAST(PL_regmatch_slab);
5680 DEBUG_STATE_pp("pop (no final)");
5682 DEBUG_STATE_pp("pop (yes)");
5688 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5689 || yes_state > SLAB_LAST(PL_regmatch_slab))
5691 /* not in this slab, pop slab */
5692 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5693 PL_regmatch_slab = PL_regmatch_slab->prev;
5694 st = SLAB_LAST(PL_regmatch_slab);
5696 depth -= (st - yes_state);
5699 yes_state = st->u.yes.prev_yes_state;
5700 PL_regmatch_state = st;
5703 locinput= st->locinput;
5704 nextchr = UCHARAT(locinput);
5706 state_num = st->resume_state + no_final;
5707 goto reenter_switch;
5710 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5711 PL_colors[4], PL_colors[5]));
5713 if (PL_reg_eval_set) {
5714 /* each successfully executed (?{...}) block does the equivalent of
5715 * local $^R = do {...}
5716 * When popping the save stack, all these locals would be undone;
5717 * bypass this by setting the outermost saved $^R to the latest
5719 if (oreplsv != GvSV(PL_replgv))
5720 sv_setsv(oreplsv, GvSV(PL_replgv));
5727 PerlIO_printf(Perl_debug_log,
5728 "%*s %sfailed...%s\n",
5729 REPORT_CODE_OFF+depth*2, "",
5730 PL_colors[4], PL_colors[5])
5742 /* there's a previous state to backtrack to */
5744 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5745 PL_regmatch_slab = PL_regmatch_slab->prev;
5746 st = SLAB_LAST(PL_regmatch_slab);
5748 PL_regmatch_state = st;
5749 locinput= st->locinput;
5750 nextchr = UCHARAT(locinput);
5752 DEBUG_STATE_pp("pop");
5754 if (yes_state == st)
5755 yes_state = st->u.yes.prev_yes_state;
5757 state_num = st->resume_state + 1; /* failure = success + 1 */
5758 goto reenter_switch;
5763 if (rex->intflags & PREGf_VERBARG_SEEN) {
5764 SV *sv_err = get_sv("REGERROR", 1);
5765 SV *sv_mrk = get_sv("REGMARK", 1);
5767 sv_commit = &PL_sv_no;
5769 sv_yes_mark = &PL_sv_yes;
5772 sv_commit = &PL_sv_yes;
5773 sv_yes_mark = &PL_sv_no;
5775 sv_setsv(sv_err, sv_commit);
5776 sv_setsv(sv_mrk, sv_yes_mark);
5779 /* clean up; in particular, free all slabs above current one */
5780 LEAVE_SCOPE(oldsave);
5786 - regrepeat - repeatedly match something simple, report how many
5789 * [This routine now assumes that it will only match on things of length 1.
5790 * That was true before, but now we assume scan - reginput is the count,
5791 * rather than incrementing count on every character. [Er, except utf8.]]
5794 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5797 register char *scan;
5799 register char *loceol = PL_regeol;
5800 register I32 hardcount = 0;
5801 register bool utf8_target = PL_reg_match_utf8;
5803 PERL_UNUSED_ARG(depth);
5806 PERL_ARGS_ASSERT_REGREPEAT;
5809 if (max == REG_INFTY)
5811 else if (max < loceol - scan)
5812 loceol = scan + max;
5817 while (scan < loceol && hardcount < max && *scan != '\n') {
5818 scan += UTF8SKIP(scan);
5822 while (scan < loceol && *scan != '\n')
5829 while (scan < loceol && hardcount < max) {
5830 scan += UTF8SKIP(scan);
5841 /* To get here, EXACTish nodes must have *byte* length == 1. That
5842 * means they match only characters in the string that can be expressed
5843 * as a single byte. For non-utf8 strings, that means a simple match.
5844 * For utf8 strings, the character matched must be an invariant, or
5845 * downgradable to a single byte. The pattern's utf8ness is
5846 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5847 * it is, it's an invariant */
5850 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5852 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5853 while (scan < loceol && UCHARAT(scan) == c) {
5859 /* Here, the string is utf8, and the pattern char is different
5860 * in utf8 than not, so can't compare them directly. Outside the
5861 * loop, find find the two utf8 bytes that represent c, and then
5862 * look for those in sequence in the utf8 string */
5863 U8 high = UTF8_TWO_BYTE_HI(c);
5864 U8 low = UTF8_TWO_BYTE_LO(c);
5867 while (hardcount < max
5868 && scan + 1 < loceol
5869 && UCHARAT(scan) == high
5870 && UCHARAT(scan + 1) == low)
5878 PL_reg_flags |= RF_tainted;
5883 /* The comments for the EXACT case above apply as well to these fold
5887 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5889 if (utf8_target) { /* Use full Unicode fold matching */
5891 /* For the EXACTFL case, It doesn't really make sense to compare
5892 * locale and utf8, but it is best we can do. The documents warn
5893 * against mixing them */
5895 char *tmpeol = loceol;
5896 while (hardcount < max
5897 && foldEQ_utf8(scan, &tmpeol, 0, utf8_target,
5898 STRING(p), NULL, 1, cBOOL(UTF_PATTERN)))
5905 /* XXX Note that the above handles properly the German sharp s in
5906 * the pattern matching ss in the string. But it doesn't handle
5907 * properly cases where the string contains say 'LIGATURE ff' and
5908 * the pattern is 'f+'. This would require, say, a new function or
5909 * revised interface to foldEQ_utf8(), in which the maximum number
5910 * of characters to match could be passed and it would return how
5911 * many actually did. This is just one of many cases where
5912 * multi-char folds don't work properly, and so the fix is being
5918 /* Here, the string isn't utf8 and c is a single byte; and either
5919 * the pattern isn't utf8 or c is an invariant, so its utf8ness
5920 * doesn't affect c. Can just do simple comparisons for exact or
5923 case EXACTF: folded = PL_fold[c]; break;
5924 case EXACTFU: folded = PL_fold_latin1[c]; break;
5925 case EXACTFL: folded = PL_fold_locale[c]; break;
5926 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
5928 while (scan < loceol &&
5929 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
5938 while (hardcount < max && scan < loceol &&
5939 reginclass(prog, p, (U8*)scan, 0, utf8_target)) {
5940 scan += UTF8SKIP(scan);
5944 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
5951 LOAD_UTF8_CHARCLASS_ALNUM();
5952 while (hardcount < max && scan < loceol &&
5953 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
5955 scan += UTF8SKIP(scan);
5958 } else if (FLAGS(p) & USE_UNI) {
5959 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
5963 while (scan < loceol && isALNUM((U8) *scan)) {
5969 PL_reg_flags |= RF_tainted;
5972 while (hardcount < max && scan < loceol &&
5973 isALNUM_LC_utf8((U8*)scan)) {
5974 scan += UTF8SKIP(scan);
5978 while (scan < loceol && isALNUM_LC(*scan))
5985 LOAD_UTF8_CHARCLASS_ALNUM();
5986 while (hardcount < max && scan < loceol &&
5987 !swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
5989 scan += UTF8SKIP(scan);
5992 } else if (FLAGS(p) & USE_UNI) {
5993 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
5997 while (scan < loceol && ! isALNUM((U8) *scan)) {
6003 PL_reg_flags |= RF_tainted;
6006 while (hardcount < max && scan < loceol &&
6007 !isALNUM_LC_utf8((U8*)scan)) {
6008 scan += UTF8SKIP(scan);
6012 while (scan < loceol && !isALNUM_LC(*scan))
6019 LOAD_UTF8_CHARCLASS_SPACE();
6020 while (hardcount < max && scan < loceol &&
6022 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6024 scan += UTF8SKIP(scan);
6027 } else if (FLAGS(p) & USE_UNI) {
6028 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6032 while (scan < loceol && isSPACE((U8) *scan))
6037 PL_reg_flags |= RF_tainted;
6040 while (hardcount < max && scan < loceol &&
6041 isSPACE_LC_utf8((U8*)scan)) {
6042 scan += UTF8SKIP(scan);
6046 while (scan < loceol && isSPACE_LC(*scan))
6053 LOAD_UTF8_CHARCLASS_SPACE();
6054 while (hardcount < max && scan < loceol &&
6056 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6058 scan += UTF8SKIP(scan);
6061 } else if (FLAGS(p) & USE_UNI) {
6062 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6066 while (scan < loceol && ! isSPACE((U8) *scan)) {
6072 PL_reg_flags |= RF_tainted;
6075 while (hardcount < max && scan < loceol &&
6076 !isSPACE_LC_utf8((U8*)scan)) {
6077 scan += UTF8SKIP(scan);
6081 while (scan < loceol && !isSPACE_LC(*scan))
6088 LOAD_UTF8_CHARCLASS_DIGIT();
6089 while (hardcount < max && scan < loceol &&
6090 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6091 scan += UTF8SKIP(scan);
6095 while (scan < loceol && isDIGIT(*scan))
6100 PL_reg_flags |= RF_tainted;
6103 while (hardcount < max && scan < loceol &&
6104 isDIGIT_LC_utf8((U8*)scan)) {
6105 scan += UTF8SKIP(scan);
6109 while (scan < loceol && isDIGIT_LC(*scan))
6116 LOAD_UTF8_CHARCLASS_DIGIT();
6117 while (hardcount < max && scan < loceol &&
6118 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6119 scan += UTF8SKIP(scan);
6123 while (scan < loceol && !isDIGIT(*scan))
6127 PL_reg_flags |= RF_tainted;
6130 while (hardcount < max && scan < loceol &&
6131 !isDIGIT_LC_utf8((U8*)scan)) {
6132 scan += UTF8SKIP(scan);
6136 while (scan < loceol && !isDIGIT_LC(*scan))
6143 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6149 LNBREAK can match two latin chars, which is ok,
6150 because we have a null terminated string, but we
6151 have to use hardcount in this situation
6153 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6162 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6167 while (scan < loceol && is_HORIZWS_latin1(scan))
6174 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6175 scan += UTF8SKIP(scan);
6179 while (scan < loceol && !is_HORIZWS_latin1(scan))
6187 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6192 while (scan < loceol && is_VERTWS_latin1(scan))
6200 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6201 scan += UTF8SKIP(scan);
6205 while (scan < loceol && !is_VERTWS_latin1(scan))
6211 default: /* Called on something of 0 width. */
6212 break; /* So match right here or not at all. */
6218 c = scan - PL_reginput;
6222 GET_RE_DEBUG_FLAGS_DECL;
6224 SV * const prop = sv_newmortal();
6225 regprop(prog, prop, p);
6226 PerlIO_printf(Perl_debug_log,
6227 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6228 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6236 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6238 - regclass_swash - prepare the utf8 swash
6242 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6248 RXi_GET_DECL(prog,progi);
6249 const struct reg_data * const data = prog ? progi->data : NULL;
6251 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6253 if (data && data->count) {
6254 const U32 n = ARG(node);
6256 if (data->what[n] == 's') {
6257 SV * const rv = MUTABLE_SV(data->data[n]);
6258 AV * const av = MUTABLE_AV(SvRV(rv));
6259 SV **const ary = AvARRAY(av);
6262 /* See the end of regcomp.c:S_regclass() for
6263 * documentation of these array elements. */
6266 a = SvROK(ary[1]) ? &ary[1] : NULL;
6267 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6271 else if (si && doinit) {
6272 sw = swash_init("utf8", "", si, 1, 0);
6273 (void)av_store(av, 1, sw);
6290 - reginclass - determine if a character falls into a character class
6292 n is the ANYOF regnode
6293 p is the target string
6294 lenp is pointer to the maximum number of bytes of how far to go in p
6295 (This is assumed wthout checking to always be at least the current
6297 utf8_target tells whether p is in UTF-8.
6299 Returns true if matched; false otherwise. If lenp is not NULL, on return
6300 from a successful match, the value it points to will be updated to how many
6301 bytes in p were matched. If there was no match, the value is undefined,
6302 possibly changed from the input.
6307 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6310 const char flags = ANYOF_FLAGS(n);
6316 PERL_ARGS_ASSERT_REGINCLASS;
6318 /* If c is not already the code point, get it */
6319 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6320 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6321 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6322 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6323 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6324 * UTF8_ALLOW_FFFF */
6325 if (c_len == (STRLEN)-1)
6326 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6332 /* Use passed in max length, or one character if none passed in or less
6333 * than one character. And assume will match just one character. This is
6334 * overwritten later if matched more. */
6336 maxlen = (*lenp > c_len) ? *lenp : c_len;
6344 /* If this character is potentially in the bitmap, check it */
6346 if (ANYOF_BITMAP_TEST(n, c))
6348 else if (flags & ANYOF_FOLD) {
6351 if (flags & ANYOF_LOCALE) {
6352 PL_reg_flags |= RF_tainted;
6353 f = PL_fold_locale[c];
6357 if (f != c && ANYOF_BITMAP_TEST(n, f))
6361 if (!match && ANYOF_CLASS_TEST_ANY_SET(n)) {
6362 PL_reg_flags |= RF_tainted; /* CLASS implies LOCALE */
6364 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6365 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6366 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6367 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6368 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6369 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6370 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6371 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6372 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6373 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6374 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6375 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6376 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6377 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6378 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6379 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6380 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6381 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6382 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6383 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6384 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6385 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6386 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6387 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6388 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6389 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6390 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6391 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6392 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6393 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6394 ) /* How's that for a conditional? */
6401 /* If the bitmap didn't (or couldn't) match, and something outside the
6402 * bitmap could match, try that */
6404 if (utf8_target && (flags & ANYOF_UNICODE_ALL)) {
6406 || ((flags & ANYOF_FOLD) /* Latin1 1 that has a non-Latin1 fold
6408 && _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(c)))
6413 if (!match && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6414 || (utf8_target && flags & ANYOF_UTF8)))
6417 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6425 utf8_p = bytes_to_utf8(p, &len);
6427 if (swash_fetch(sw, utf8_p, 1))
6429 else if (flags & ANYOF_FOLD) {
6430 if (!match && lenp && av) {
6432 for (i = 0; i <= av_len(av); i++) {
6433 SV* const sv = *av_fetch(av, i, FALSE);
6435 const char * const s = SvPV_const(sv, len);
6436 if (len <= maxlen && memEQ(s, (char*)utf8_p, len)) {
6443 if (!match) { /* See if the folded version matches */
6444 U8 folded[UTF8_MAXBYTES_CASE+1];
6448 to_utf8_fold(utf8_p, folded, &foldlen);
6450 /* Consider "k" =~ /[K]/i. The line above would have
6451 * just folded the 'k' to itself, and that isn't going
6452 * to match 'K'. So we look through the closure of
6453 * everything that folds to 'k'. That will find the
6454 * 'K'. Initialize the list, if necessary */
6455 if (! PL_utf8_foldclosures) {
6457 /* If the folds haven't been read in, call a fold
6458 * function to force that */
6459 if (! PL_utf8_tofold) {
6460 U8 dummy[UTF8_MAXBYTES+1];
6462 to_utf8_fold((U8*) "A", dummy, &dummy_len);
6464 PL_utf8_foldclosures =
6465 _swash_inversion_hash(PL_utf8_tofold);
6468 /* The data structure is a hash with the keys every
6469 * character that is folded to, like 'k', and the
6470 * values each an array of everything that folds to its
6471 * key. e.g. [ 'k', 'K', KELVIN_SIGN ] */
6472 if ((listp = hv_fetch(PL_utf8_foldclosures,
6473 (char *) folded, foldlen, FALSE)))
6475 AV* list = (AV*) *listp;
6477 for (i = 0; i <= av_len(list); i++) {
6478 SV** try_p = av_fetch(list, i, FALSE);
6480 if (try_p == NULL) {
6481 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
6483 /* Don't have to worry about embeded nulls
6484 * since NULL isn't folded or foldable */
6485 try_c = SvPVX(*try_p);
6487 /* The fold in a few cases of an above Latin1
6488 * char is in the Latin1 range, and hence may
6489 * be in the bitmap */
6490 if (UTF8_IS_INVARIANT(*try_c)
6491 && ANYOF_BITMAP_TEST(n,
6492 UNI_TO_NATIVE(*try_c)))
6498 (UTF8_IS_DOWNGRADEABLE_START(*try_c)
6499 && ANYOF_BITMAP_TEST(n, UNI_TO_NATIVE(
6500 TWO_BYTE_UTF8_TO_UNI(try_c[0],
6503 /* Since the fold comes from internally
6504 * generated data, we can safely assume it
6505 * is valid utf8 in the test above */
6508 } else if (swash_fetch(sw, (U8*) try_c, 1)) {
6517 /* If we allocated a string above, free it */
6518 if (! utf8_target) Safefree(utf8_p);
6523 return (flags & ANYOF_INVERT) ? !match : match;
6527 S_reghop3(U8 *s, I32 off, const U8* lim)
6531 PERL_ARGS_ASSERT_REGHOP3;
6534 while (off-- && s < lim) {
6535 /* XXX could check well-formedness here */
6540 while (off++ && s > lim) {
6542 if (UTF8_IS_CONTINUED(*s)) {
6543 while (s > lim && UTF8_IS_CONTINUATION(*s))
6546 /* XXX could check well-formedness here */
6553 /* there are a bunch of places where we use two reghop3's that should
6554 be replaced with this routine. but since thats not done yet
6555 we ifdef it out - dmq
6558 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6562 PERL_ARGS_ASSERT_REGHOP4;
6565 while (off-- && s < rlim) {
6566 /* XXX could check well-formedness here */
6571 while (off++ && s > llim) {
6573 if (UTF8_IS_CONTINUED(*s)) {
6574 while (s > llim && UTF8_IS_CONTINUATION(*s))
6577 /* XXX could check well-formedness here */
6585 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6589 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6592 while (off-- && s < lim) {
6593 /* XXX could check well-formedness here */
6600 while (off++ && s > lim) {
6602 if (UTF8_IS_CONTINUED(*s)) {
6603 while (s > lim && UTF8_IS_CONTINUATION(*s))
6606 /* XXX could check well-formedness here */
6615 restore_pos(pTHX_ void *arg)
6618 regexp * const rex = (regexp *)arg;
6619 if (PL_reg_eval_set) {
6620 if (PL_reg_oldsaved) {
6621 rex->subbeg = PL_reg_oldsaved;
6622 rex->sublen = PL_reg_oldsavedlen;
6623 #ifdef PERL_OLD_COPY_ON_WRITE
6624 rex->saved_copy = PL_nrs;
6626 RXp_MATCH_COPIED_on(rex);
6628 PL_reg_magic->mg_len = PL_reg_oldpos;
6629 PL_reg_eval_set = 0;
6630 PL_curpm = PL_reg_oldcurpm;
6635 S_to_utf8_substr(pTHX_ register regexp *prog)
6639 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6642 if (prog->substrs->data[i].substr
6643 && !prog->substrs->data[i].utf8_substr) {
6644 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6645 prog->substrs->data[i].utf8_substr = sv;
6646 sv_utf8_upgrade(sv);
6647 if (SvVALID(prog->substrs->data[i].substr)) {
6648 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6649 if (flags & FBMcf_TAIL) {
6650 /* Trim the trailing \n that fbm_compile added last
6652 SvCUR_set(sv, SvCUR(sv) - 1);
6653 /* Whilst this makes the SV technically "invalid" (as its
6654 buffer is no longer followed by "\0") when fbm_compile()
6655 adds the "\n" back, a "\0" is restored. */
6657 fbm_compile(sv, flags);
6659 if (prog->substrs->data[i].substr == prog->check_substr)
6660 prog->check_utf8 = sv;
6666 S_to_byte_substr(pTHX_ register regexp *prog)
6671 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6674 if (prog->substrs->data[i].utf8_substr
6675 && !prog->substrs->data[i].substr) {
6676 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6677 if (sv_utf8_downgrade(sv, TRUE)) {
6678 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6680 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6681 if (flags & FBMcf_TAIL) {
6682 /* Trim the trailing \n that fbm_compile added last
6684 SvCUR_set(sv, SvCUR(sv) - 1);
6686 fbm_compile(sv, flags);
6692 prog->substrs->data[i].substr = sv;
6693 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6694 prog->check_substr = sv;
6701 * c-indentation-style: bsd
6703 * indent-tabs-mode: t
6706 * ex: set ts=8 sts=4 sw=4 noet: