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
40 /* At least one required character in the target string is expressible only in
42 static const char* const non_utf8_target_but_utf8_required
43 = "Can't match, because target string needs to be in UTF-8\n";
45 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
46 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
51 * pregcomp and pregexec -- regsub and regerror are not used in perl
53 * Copyright (c) 1986 by University of Toronto.
54 * Written by Henry Spencer. Not derived from licensed software.
56 * Permission is granted to anyone to use this software for any
57 * purpose on any computer system, and to redistribute it freely,
58 * subject to the following restrictions:
60 * 1. The author is not responsible for the consequences of use of
61 * this software, no matter how awful, even if they arise
64 * 2. The origin of this software must not be misrepresented, either
65 * by explicit claim or by omission.
67 * 3. Altered versions must be plainly marked as such, and must not
68 * be misrepresented as being the original software.
70 **** Alterations to Henry's code are...
72 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
73 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
74 **** by Larry Wall and others
76 **** You may distribute under the terms of either the GNU General Public
77 **** License or the Artistic License, as specified in the README file.
79 * Beware that some of this code is subtly aware of the way operator
80 * precedence is structured in regular expressions. Serious changes in
81 * regular-expression syntax might require a total rethink.
84 #define PERL_IN_REGEXEC_C
88 #ifdef PERL_IN_XSUB_RE
94 #include "inline_invlist.c"
95 #include "unicode_constants.h"
97 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
100 #define STATIC static
103 /* Valid for non-utf8 strings: avoids the reginclass
104 * call if there are no complications: i.e., if everything matchable is
105 * straight forward in the bitmap */
106 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \
107 : ANYOF_BITMAP_TEST(p,*(c)))
113 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
114 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
116 #define HOPc(pos,off) \
117 (char *)(PL_reg_match_utf8 \
118 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
120 #define HOPBACKc(pos, off) \
121 (char*)(PL_reg_match_utf8\
122 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
123 : (pos - off >= PL_bostr) \
127 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
128 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
131 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
132 #define NEXTCHR_IS_EOS (nextchr < 0)
134 #define SET_nextchr \
135 nextchr = ((locinput < PL_regeol) ? UCHARAT(locinput) : NEXTCHR_EOS)
137 #define SET_locinput(p) \
142 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name) STMT_START { \
144 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
145 ENTER; save_re_context(); \
146 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
147 1, 0, NULL, &flags); \
153 /* If in debug mode, we test that a known character properly matches */
155 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
157 utf8_char_in_property) \
158 LOAD_UTF8_CHARCLASS(swash_ptr, property_name); \
159 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
161 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
163 utf8_char_in_property) \
164 LOAD_UTF8_CHARCLASS(swash_ptr, property_name)
167 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
168 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
169 swash_property_names[_CC_WORDCHAR], \
170 GREEK_SMALL_LETTER_IOTA_UTF8)
172 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
174 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
175 "_X_regular_begin", \
176 GREEK_SMALL_LETTER_IOTA_UTF8); \
177 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
179 COMBINING_GRAVE_ACCENT_UTF8); \
182 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
183 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
185 /* for use after a quantifier and before an EXACT-like node -- japhy */
186 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
188 * NOTE that *nothing* that affects backtracking should be in here, specifically
189 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
190 * node that is in between two EXACT like nodes when ascertaining what the required
191 * "follow" character is. This should probably be moved to regex compile time
192 * although it may be done at run time beause of the REF possibility - more
193 * investigation required. -- demerphq
195 #define JUMPABLE(rn) ( \
197 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
199 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
200 OP(rn) == PLUS || OP(rn) == MINMOD || \
202 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
204 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
206 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
209 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
210 we don't need this definition. */
211 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
212 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
213 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
216 /* ... so we use this as its faster. */
217 #define IS_TEXT(rn) ( OP(rn)==EXACT )
218 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
219 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
220 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
225 Search for mandatory following text node; for lookahead, the text must
226 follow but for lookbehind (rn->flags != 0) we skip to the next step.
228 #define FIND_NEXT_IMPT(rn) STMT_START { \
229 while (JUMPABLE(rn)) { \
230 const OPCODE type = OP(rn); \
231 if (type == SUSPEND || PL_regkind[type] == CURLY) \
232 rn = NEXTOPER(NEXTOPER(rn)); \
233 else if (type == PLUS) \
235 else if (type == IFMATCH) \
236 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
237 else rn += NEXT_OFF(rn); \
241 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
242 * These are for the pre-composed Hangul syllables, which are all in a
243 * contiguous block and arranged there in such a way so as to facilitate
244 * alorithmic determination of their characteristics. As such, they don't need
245 * a swash, but can be determined by simple arithmetic. Almost all are
246 * GCB=LVT, but every 28th one is a GCB=LV */
247 #define SBASE 0xAC00 /* Start of block */
248 #define SCount 11172 /* Length of block */
251 static void restore_pos(pTHX_ void *arg);
253 #define REGCP_PAREN_ELEMS 3
254 #define REGCP_OTHER_ELEMS 3
255 #define REGCP_FRAME_ELEMS 1
256 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
257 * are needed for the regexp context stack bookkeeping. */
260 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
263 const int retval = PL_savestack_ix;
264 const int paren_elems_to_push =
265 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
266 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
267 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
269 GET_RE_DEBUG_FLAGS_DECL;
271 PERL_ARGS_ASSERT_REGCPPUSH;
273 if (paren_elems_to_push < 0)
274 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
275 paren_elems_to_push);
277 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
278 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
279 " out of range (%lu-%ld)",
281 (unsigned long)maxopenparen,
284 SSGROW(total_elems + REGCP_FRAME_ELEMS);
287 if ((int)maxopenparen > (int)parenfloor)
288 PerlIO_printf(Perl_debug_log,
289 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
294 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
295 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
296 SSPUSHINT(rex->offs[p].end);
297 SSPUSHINT(rex->offs[p].start);
298 SSPUSHINT(rex->offs[p].start_tmp);
299 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
300 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
302 (IV)rex->offs[p].start,
303 (IV)rex->offs[p].start_tmp,
307 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
308 SSPUSHINT(maxopenparen);
309 SSPUSHINT(rex->lastparen);
310 SSPUSHINT(rex->lastcloseparen);
311 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
316 /* These are needed since we do not localize EVAL nodes: */
317 #define REGCP_SET(cp) \
319 PerlIO_printf(Perl_debug_log, \
320 " Setting an EVAL scope, savestack=%"IVdf"\n", \
321 (IV)PL_savestack_ix)); \
324 #define REGCP_UNWIND(cp) \
326 if (cp != PL_savestack_ix) \
327 PerlIO_printf(Perl_debug_log, \
328 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
329 (IV)(cp), (IV)PL_savestack_ix)); \
332 #define UNWIND_PAREN(lp, lcp) \
333 for (n = rex->lastparen; n > lp; n--) \
334 rex->offs[n].end = -1; \
335 rex->lastparen = n; \
336 rex->lastcloseparen = lcp;
340 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
345 GET_RE_DEBUG_FLAGS_DECL;
347 PERL_ARGS_ASSERT_REGCPPOP;
349 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
351 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
352 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
353 rex->lastcloseparen = SSPOPINT;
354 rex->lastparen = SSPOPINT;
355 *maxopenparen_p = SSPOPINT;
357 i -= REGCP_OTHER_ELEMS;
358 /* Now restore the parentheses context. */
360 if (i || rex->lastparen + 1 <= rex->nparens)
361 PerlIO_printf(Perl_debug_log,
362 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
367 paren = *maxopenparen_p;
368 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
370 rex->offs[paren].start_tmp = SSPOPINT;
371 rex->offs[paren].start = SSPOPINT;
373 if (paren <= rex->lastparen)
374 rex->offs[paren].end = tmps;
375 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
376 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
378 (IV)rex->offs[paren].start,
379 (IV)rex->offs[paren].start_tmp,
380 (IV)rex->offs[paren].end,
381 (paren > rex->lastparen ? "(skipped)" : ""));
386 /* It would seem that the similar code in regtry()
387 * already takes care of this, and in fact it is in
388 * a better location to since this code can #if 0-ed out
389 * but the code in regtry() is needed or otherwise tests
390 * requiring null fields (pat.t#187 and split.t#{13,14}
391 * (as of patchlevel 7877) will fail. Then again,
392 * this code seems to be necessary or otherwise
393 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
394 * --jhi updated by dapm */
395 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
396 if (i > *maxopenparen_p)
397 rex->offs[i].start = -1;
398 rex->offs[i].end = -1;
399 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
400 " \\%"UVuf": %s ..-1 undeffing\n",
402 (i > *maxopenparen_p) ? "-1" : " "
408 /* restore the parens and associated vars at savestack position ix,
409 * but without popping the stack */
412 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
414 I32 tmpix = PL_savestack_ix;
415 PL_savestack_ix = ix;
416 regcppop(rex, maxopenparen_p);
417 PL_savestack_ix = tmpix;
420 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
423 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
425 /* Returns a boolean as to whether or not 'character' is a member of the
426 * Posix character class given by 'classnum' that should be equivalent to a
427 * value in the typedef '_char_class_number'.
429 * Ideally this could be replaced by a just an array of function pointers
430 * to the C library functions that implement the macros this calls.
431 * However, to compile, the precise function signatures are required, and
432 * these may vary from platform to to platform. To avoid having to figure
433 * out what those all are on each platform, I (khw) am using this method,
434 * which adds an extra layer of function call overhead (unless the C
435 * optimizer strips it away). But we don't particularly care about
436 * performance with locales anyway. */
438 switch ((_char_class_number) classnum) {
439 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
440 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
441 case _CC_ENUM_ASCII: return isASCII_LC(character);
442 case _CC_ENUM_BLANK: return isBLANK_LC(character);
443 case _CC_ENUM_CASED: return isLOWER_LC(character)
444 || isUPPER_LC(character);
445 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
446 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
447 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
448 case _CC_ENUM_LOWER: return isLOWER_LC(character);
449 case _CC_ENUM_PRINT: return isPRINT_LC(character);
450 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
451 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
452 case _CC_ENUM_SPACE: return isSPACE_LC(character);
453 case _CC_ENUM_UPPER: return isUPPER_LC(character);
454 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
455 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
456 default: /* VERTSPACE should never occur in locales */
457 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
460 assert(0); /* NOTREACHED */
465 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
467 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
468 * 'character' is a member of the Posix character class given by 'classnum'
469 * that should be equivalent to a value in the typedef
470 * '_char_class_number'.
472 * This just calls isFOO_lc on the code point for the character if it is in
473 * the range 0-255. Outside that range, all characters avoid Unicode
474 * rules, ignoring any locale. So use the Unicode function if this class
475 * requires a swash, and use the Unicode macro otherwise. */
477 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
479 if (UTF8_IS_INVARIANT(*character)) {
480 return isFOO_lc(classnum, *character);
482 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
483 return isFOO_lc(classnum,
484 TWO_BYTE_UTF8_TO_UNI(*character, *(character + 1)));
487 if (classnum < _FIRST_NON_SWASH_CC) {
489 /* Initialize the swash unless done already */
490 if (! PL_utf8_swash_ptrs[classnum]) {
491 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
492 PL_utf8_swash_ptrs[classnum] = _core_swash_init("utf8",
493 swash_property_names[classnum], &PL_sv_undef, 1, 0, NULL, &flags);
496 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
498 TRUE /* is UTF */ ));
501 switch ((_char_class_number) classnum) {
503 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
505 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
506 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
507 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
508 default: return 0; /* Things like CNTRL are always
512 assert(0); /* NOTREACHED */
517 * pregexec and friends
520 #ifndef PERL_IN_XSUB_RE
522 - pregexec - match a regexp against a string
525 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
526 char *strbeg, I32 minend, SV *screamer, U32 nosave)
527 /* stringarg: the point in the string at which to begin matching */
528 /* strend: pointer to null at end of string */
529 /* strbeg: real beginning of string */
530 /* minend: end of match must be >= minend bytes after stringarg. */
531 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
532 * itself is accessed via the pointers above */
533 /* nosave: For optimizations. */
535 PERL_ARGS_ASSERT_PREGEXEC;
538 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
539 nosave ? 0 : REXEC_COPY_STR);
544 * Need to implement the following flags for reg_anch:
546 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
548 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
549 * INTUIT_AUTORITATIVE_ML
550 * INTUIT_ONCE_NOML - Intuit can match in one location only.
553 * Another flag for this function: SECOND_TIME (so that float substrs
554 * with giant delta may be not rechecked).
557 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
559 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
560 Otherwise, only SvCUR(sv) is used to get strbeg. */
562 /* XXXX We assume that strpos is strbeg unless sv. */
564 /* XXXX Some places assume that there is a fixed substring.
565 An update may be needed if optimizer marks as "INTUITable"
566 RExen without fixed substrings. Similarly, it is assumed that
567 lengths of all the strings are no more than minlen, thus they
568 cannot come from lookahead.
569 (Or minlen should take into account lookahead.)
570 NOTE: Some of this comment is not correct. minlen does now take account
571 of lookahead/behind. Further research is required. -- demerphq
575 /* A failure to find a constant substring means that there is no need to make
576 an expensive call to REx engine, thus we celebrate a failure. Similarly,
577 finding a substring too deep into the string means that fewer calls to
578 regtry() should be needed.
580 REx compiler's optimizer found 4 possible hints:
581 a) Anchored substring;
583 c) Whether we are anchored (beginning-of-line or \G);
584 d) First node (of those at offset 0) which may distinguish positions;
585 We use a)b)d) and multiline-part of c), and try to find a position in the
586 string which does not contradict any of them.
589 /* Most of decisions we do here should have been done at compile time.
590 The nodes of the REx which we used for the search should have been
591 deleted from the finite automaton. */
594 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
595 char *strend, const U32 flags, re_scream_pos_data *data)
598 struct regexp *const prog = ReANY(rx);
600 /* Should be nonnegative! */
606 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
608 char *other_last = NULL; /* other substr checked before this */
609 char *check_at = NULL; /* check substr found at this pos */
610 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
611 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
612 RXi_GET_DECL(prog,progi);
615 const char * const i_strpos = strpos;
617 GET_RE_DEBUG_FLAGS_DECL;
619 PERL_ARGS_ASSERT_RE_INTUIT_START;
620 PERL_UNUSED_ARG(flags);
621 PERL_UNUSED_ARG(data);
623 RX_MATCH_UTF8_set(rx,utf8_target);
625 is_utf8_pat = cBOOL(RX_UTF8(rx));
628 debug_start_match(rx, utf8_target, strpos, strend,
629 sv ? "Guessing start of match in sv for"
630 : "Guessing start of match in string for");
633 /* CHR_DIST() would be more correct here but it makes things slow. */
634 if (prog->minlen > strend - strpos) {
635 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
636 "String too short... [re_intuit_start]\n"));
640 /* XXX we need to pass strbeg as a separate arg: the following is
641 * guesswork and can be wrong... */
642 if (sv && SvPOK(sv)) {
643 char * p = SvPVX(sv);
644 STRLEN cur = SvCUR(sv);
645 if (p <= strpos && strpos < p + cur) {
647 assert(p <= strend && strend <= p + cur);
650 strbeg = strend - cur;
657 if (!prog->check_utf8 && prog->check_substr)
658 to_utf8_substr(prog);
659 check = prog->check_utf8;
661 if (!prog->check_substr && prog->check_utf8) {
662 if (! to_byte_substr(prog)) {
663 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
666 check = prog->check_substr;
668 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
669 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
670 || ( (prog->extflags & RXf_ANCH_BOL)
671 && !multiline ) ); /* Check after \n? */
674 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
675 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
676 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
678 && (strpos != strbeg)) {
679 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
682 if (prog->check_offset_min == prog->check_offset_max
683 && !(prog->extflags & RXf_CANY_SEEN)
684 && ! multiline) /* /m can cause \n's to match that aren't
685 accounted for in the string max length.
686 See [perl #115242] */
688 /* Substring at constant offset from beg-of-str... */
691 s = HOP3c(strpos, prog->check_offset_min, strend);
694 slen = SvCUR(check); /* >= 1 */
696 if ( strend - s > slen || strend - s < slen - 1
697 || (strend - s == slen && strend[-1] != '\n')) {
698 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
701 /* Now should match s[0..slen-2] */
703 if (slen && (*SvPVX_const(check) != *s
705 && memNE(SvPVX_const(check), s, slen)))) {
707 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
711 else if (*SvPVX_const(check) != *s
712 || ((slen = SvCUR(check)) > 1
713 && memNE(SvPVX_const(check), s, slen)))
716 goto success_at_start;
719 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
721 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
722 end_shift = prog->check_end_shift;
725 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
726 - (SvTAIL(check) != 0);
727 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
729 if (end_shift < eshift)
733 else { /* Can match at random position */
736 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
737 end_shift = prog->check_end_shift;
739 /* end shift should be non negative here */
742 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
744 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
745 (IV)end_shift, RX_PRECOMP(prog));
749 /* Find a possible match in the region s..strend by looking for
750 the "check" substring in the region corrected by start/end_shift. */
753 I32 srch_start_shift = start_shift;
754 I32 srch_end_shift = end_shift;
757 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
758 srch_end_shift -= ((strbeg - s) - srch_start_shift);
759 srch_start_shift = strbeg - s;
761 DEBUG_OPTIMISE_MORE_r({
762 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
763 (IV)prog->check_offset_min,
764 (IV)srch_start_shift,
766 (IV)prog->check_end_shift);
769 if (prog->extflags & RXf_CANY_SEEN) {
770 start_point= (U8*)(s + srch_start_shift);
771 end_point= (U8*)(strend - srch_end_shift);
773 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
774 end_point= HOP3(strend, -srch_end_shift, strbeg);
776 DEBUG_OPTIMISE_MORE_r({
777 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
778 (int)(end_point - start_point),
779 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
783 s = fbm_instr( start_point, end_point,
784 check, multiline ? FBMrf_MULTILINE : 0);
786 /* Update the count-of-usability, remove useless subpatterns,
790 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
791 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
792 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
793 (s ? "Found" : "Did not find"),
794 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
795 ? "anchored" : "floating"),
798 (s ? " at offset " : "...\n") );
803 /* Finish the diagnostic message */
804 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
806 /* XXX dmq: first branch is for positive lookbehind...
807 Our check string is offset from the beginning of the pattern.
808 So we need to do any stclass tests offset forward from that
817 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
818 Start with the other substr.
819 XXXX no SCREAM optimization yet - and a very coarse implementation
820 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
821 *always* match. Probably should be marked during compile...
822 Probably it is right to do no SCREAM here...
825 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
826 : (prog->float_substr && prog->anchored_substr))
828 /* Take into account the "other" substring. */
829 /* XXXX May be hopelessly wrong for UTF... */
832 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
835 char * const last = HOP3c(s, -start_shift, strbeg);
837 char * const saved_s = s;
840 t = s - prog->check_offset_max;
841 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
843 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
848 t = HOP3c(t, prog->anchored_offset, strend);
849 if (t < other_last) /* These positions already checked */
851 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
854 /* XXXX It is not documented what units *_offsets are in.
855 We assume bytes, but this is clearly wrong.
856 Meaning this code needs to be carefully reviewed for errors.
860 /* On end-of-str: see comment below. */
861 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
862 if (must == &PL_sv_undef) {
864 DEBUG_r(must = prog->anchored_utf8); /* for debug */
869 HOP3(HOP3(last1, prog->anchored_offset, strend)
870 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
872 multiline ? FBMrf_MULTILINE : 0
875 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
876 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
877 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
878 (s ? "Found" : "Contradicts"),
879 quoted, RE_SV_TAIL(must));
884 if (last1 >= last2) {
885 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
886 ", giving up...\n"));
889 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
890 ", trying floating at offset %ld...\n",
891 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
892 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
893 s = HOP3c(last, 1, strend);
897 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
898 (long)(s - i_strpos)));
899 t = HOP3c(s, -prog->anchored_offset, strbeg);
900 other_last = HOP3c(s, 1, strend);
908 else { /* Take into account the floating substring. */
910 char * const saved_s = s;
913 t = HOP3c(s, -start_shift, strbeg);
915 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
916 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
917 last = HOP3c(t, prog->float_max_offset, strend);
918 s = HOP3c(t, prog->float_min_offset, strend);
921 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
922 must = utf8_target ? prog->float_utf8 : prog->float_substr;
923 /* fbm_instr() takes into account exact value of end-of-str
924 if the check is SvTAIL(ed). Since false positives are OK,
925 and end-of-str is not later than strend we are OK. */
926 if (must == &PL_sv_undef) {
928 DEBUG_r(must = prog->float_utf8); /* for debug message */
931 s = fbm_instr((unsigned char*)s,
932 (unsigned char*)last + SvCUR(must)
934 must, multiline ? FBMrf_MULTILINE : 0);
936 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
937 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
938 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
939 (s ? "Found" : "Contradicts"),
940 quoted, RE_SV_TAIL(must));
944 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
945 ", giving up...\n"));
948 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
949 ", trying anchored starting at offset %ld...\n",
950 (long)(saved_s + 1 - i_strpos)));
952 s = HOP3c(t, 1, strend);
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
957 (long)(s - i_strpos)));
958 other_last = s; /* Fix this later. --Hugo */
968 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
970 DEBUG_OPTIMISE_MORE_r(
971 PerlIO_printf(Perl_debug_log,
972 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
973 (IV)prog->check_offset_min,
974 (IV)prog->check_offset_max,
982 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
984 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
987 /* Fixed substring is found far enough so that the match
988 cannot start at strpos. */
990 if (ml_anch && t[-1] != '\n') {
991 /* Eventually fbm_*() should handle this, but often
992 anchored_offset is not 0, so this check will not be wasted. */
993 /* XXXX In the code below we prefer to look for "^" even in
994 presence of anchored substrings. And we search even
995 beyond the found float position. These pessimizations
996 are historical artefacts only. */
998 while (t < strend - prog->minlen) {
1000 if (t < check_at - prog->check_offset_min) {
1001 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1002 /* Since we moved from the found position,
1003 we definitely contradict the found anchored
1004 substr. Due to the above check we do not
1005 contradict "check" substr.
1006 Thus we can arrive here only if check substr
1007 is float. Redo checking for "other"=="fixed".
1010 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1011 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
1012 goto do_other_anchored;
1014 /* We don't contradict the found floating substring. */
1015 /* XXXX Why not check for STCLASS? */
1017 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1018 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1021 /* Position contradicts check-string */
1022 /* XXXX probably better to look for check-string
1023 than for "\n", so one should lower the limit for t? */
1024 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1025 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1026 other_last = strpos = s = t + 1;
1031 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1032 PL_colors[0], PL_colors[1]));
1036 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1037 PL_colors[0], PL_colors[1]));
1041 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1044 /* The found string does not prohibit matching at strpos,
1045 - no optimization of calling REx engine can be performed,
1046 unless it was an MBOL and we are not after MBOL,
1047 or a future STCLASS check will fail this. */
1049 /* Even in this situation we may use MBOL flag if strpos is offset
1050 wrt the start of the string. */
1051 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1052 && (strpos != strbeg) && strpos[-1] != '\n'
1053 /* May be due to an implicit anchor of m{.*foo} */
1054 && !(prog->intflags & PREGf_IMPLICIT))
1059 DEBUG_EXECUTE_r( if (ml_anch)
1060 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1061 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1064 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1066 prog->check_utf8 /* Could be deleted already */
1067 && --BmUSEFUL(prog->check_utf8) < 0
1068 && (prog->check_utf8 == prog->float_utf8)
1070 prog->check_substr /* Could be deleted already */
1071 && --BmUSEFUL(prog->check_substr) < 0
1072 && (prog->check_substr == prog->float_substr)
1075 /* If flags & SOMETHING - do not do it many times on the same match */
1076 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1077 /* XXX Does the destruction order has to change with utf8_target? */
1078 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1079 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1080 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1081 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1082 check = NULL; /* abort */
1084 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1085 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1086 if (prog->intflags & PREGf_IMPLICIT)
1087 prog->extflags &= ~RXf_ANCH_MBOL;
1088 /* XXXX This is a remnant of the old implementation. It
1089 looks wasteful, since now INTUIT can use many
1090 other heuristics. */
1091 prog->extflags &= ~RXf_USE_INTUIT;
1092 /* XXXX What other flags might need to be cleared in this branch? */
1098 /* Last resort... */
1099 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1100 /* trie stclasses are too expensive to use here, we are better off to
1101 leave it to regmatch itself */
1102 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1103 /* minlen == 0 is possible if regstclass is \b or \B,
1104 and the fixed substr is ''$.
1105 Since minlen is already taken into account, s+1 is before strend;
1106 accidentally, minlen >= 1 guaranties no false positives at s + 1
1107 even for \b or \B. But (minlen? 1 : 0) below assumes that
1108 regstclass does not come from lookahead... */
1109 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1110 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1111 const U8* const str = (U8*)STRING(progi->regstclass);
1112 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1113 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1116 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1117 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1118 else if (prog->float_substr || prog->float_utf8)
1119 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1123 if (checked_upto < s)
1125 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1126 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1129 s = find_byclass(prog, progi->regstclass, checked_upto, endpos,
1135 const char *what = NULL;
1137 if (endpos == strend) {
1138 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1139 "Could not match STCLASS...\n") );
1142 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1143 "This position contradicts STCLASS...\n") );
1144 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1146 checked_upto = HOPBACKc(endpos, start_shift);
1147 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1148 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1149 /* Contradict one of substrings */
1150 if (prog->anchored_substr || prog->anchored_utf8) {
1151 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1152 DEBUG_EXECUTE_r( what = "anchored" );
1154 s = HOP3c(t, 1, strend);
1155 if (s + start_shift + end_shift > strend) {
1156 /* XXXX Should be taken into account earlier? */
1157 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1158 "Could not match STCLASS...\n") );
1163 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1164 "Looking for %s substr starting at offset %ld...\n",
1165 what, (long)(s + start_shift - i_strpos)) );
1168 /* Have both, check_string is floating */
1169 if (t + start_shift >= check_at) /* Contradicts floating=check */
1170 goto retry_floating_check;
1171 /* Recheck anchored substring, but not floating... */
1175 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1176 "Looking for anchored substr starting at offset %ld...\n",
1177 (long)(other_last - i_strpos)) );
1178 goto do_other_anchored;
1180 /* Another way we could have checked stclass at the
1181 current position only: */
1186 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1187 "Looking for /%s^%s/m starting at offset %ld...\n",
1188 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1191 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1193 /* Check is floating substring. */
1194 retry_floating_check:
1195 t = check_at - start_shift;
1196 DEBUG_EXECUTE_r( what = "floating" );
1197 goto hop_and_restart;
1200 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1201 "By STCLASS: moving %ld --> %ld\n",
1202 (long)(t - i_strpos), (long)(s - i_strpos))
1206 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1207 "Does not contradict STCLASS...\n");
1212 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1213 PL_colors[4], (check ? "Guessed" : "Giving up"),
1214 PL_colors[5], (long)(s - i_strpos)) );
1217 fail_finish: /* Substring not found */
1218 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1219 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1221 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1222 PL_colors[4], PL_colors[5]));
1226 #define DECL_TRIE_TYPE(scan) \
1227 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1228 trie_type = ((scan->flags == EXACT) \
1229 ? (utf8_target ? trie_utf8 : trie_plain) \
1230 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1232 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1233 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1235 switch (trie_type) { \
1236 case trie_utf8_fold: \
1237 if ( foldlen>0 ) { \
1238 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1243 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1244 len = UTF8SKIP(uc); \
1245 skiplen = UNISKIP( uvc ); \
1246 foldlen -= skiplen; \
1247 uscan = foldbuf + skiplen; \
1250 case trie_latin_utf8_fold: \
1251 if ( foldlen>0 ) { \
1252 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1258 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1259 skiplen = UNISKIP( uvc ); \
1260 foldlen -= skiplen; \
1261 uscan = foldbuf + skiplen; \
1265 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1272 charid = trie->charmap[ uvc ]; \
1276 if (widecharmap) { \
1277 SV** const svpp = hv_fetch(widecharmap, \
1278 (char*)&uvc, sizeof(UV), 0); \
1280 charid = (U16)SvIV(*svpp); \
1285 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1289 && (ln == 1 || folder(s, pat_string, ln)) \
1290 && (!reginfo || regtry(reginfo, &s)) ) \
1296 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1298 while (s < strend) { \
1304 #define REXEC_FBC_SCAN(CoDe) \
1306 while (s < strend) { \
1312 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1313 REXEC_FBC_UTF8_SCAN( \
1315 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1324 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1327 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1336 #define REXEC_FBC_TRYIT \
1337 if ((!reginfo || regtry(reginfo, &s))) \
1340 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1341 if (utf8_target) { \
1342 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1345 REXEC_FBC_CLASS_SCAN(CoNd); \
1348 #define DUMP_EXEC_POS(li,s,doutf8) \
1349 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1352 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1353 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1354 tmp = TEST_NON_UTF8(tmp); \
1355 REXEC_FBC_UTF8_SCAN( \
1356 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1365 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1366 if (s == PL_bostr) { \
1370 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1371 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1374 LOAD_UTF8_CHARCLASS_ALNUM(); \
1375 REXEC_FBC_UTF8_SCAN( \
1376 if (tmp == ! (TeSt2_UtF8)) { \
1385 /* The only difference between the BOUND and NBOUND cases is that
1386 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1387 * NBOUND. This is accomplished by passing it in either the if or else clause,
1388 * with the other one being empty */
1389 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1390 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1392 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1393 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1395 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1396 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1398 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1399 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1402 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1403 * be passed in completely with the variable name being tested, which isn't
1404 * such a clean interface, but this is easier to read than it was before. We
1405 * are looking for the boundary (or non-boundary between a word and non-word
1406 * character. The utf8 and non-utf8 cases have the same logic, but the details
1407 * must be different. Find the "wordness" of the character just prior to this
1408 * one, and compare it with the wordness of this one. If they differ, we have
1409 * a boundary. At the beginning of the string, pretend that the previous
1410 * character was a new-line */
1411 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1412 if (utf8_target) { \
1415 else { /* Not utf8 */ \
1416 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1417 tmp = TEST_NON_UTF8(tmp); \
1419 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1428 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1431 /* We know what class REx starts with. Try to find this position... */
1432 /* if reginfo is NULL, its a dryrun */
1433 /* annoyingly all the vars in this routine have different names from their counterparts
1434 in regmatch. /grrr */
1437 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1438 const char *strend, regmatch_info *reginfo, bool is_utf8_pat)
1441 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1442 char *pat_string; /* The pattern's exactish string */
1443 char *pat_end; /* ptr to end char of pat_string */
1444 re_fold_t folder; /* Function for computing non-utf8 folds */
1445 const U8 *fold_array; /* array for folding ords < 256 */
1451 I32 tmp = 1; /* Scratch variable? */
1452 const bool utf8_target = PL_reg_match_utf8;
1453 UV utf8_fold_flags = 0;
1454 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1455 with a result inverts that result, as 0^1 =
1457 _char_class_number classnum;
1459 RXi_GET_DECL(prog,progi);
1461 PERL_ARGS_ASSERT_FIND_BYCLASS;
1463 /* We know what class it must start with. */
1466 case ANYOF_SYNTHETIC:
1467 case ANYOF_WARN_SUPER:
1469 REXEC_FBC_UTF8_CLASS_SCAN(
1470 reginclass(prog, c, (U8*)s, utf8_target));
1473 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1478 if (tmp && (!reginfo || regtry(reginfo, &s)))
1486 if (is_utf8_pat || utf8_target) {
1487 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1488 goto do_exactf_utf8;
1490 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1491 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1492 goto do_exactf_non_utf8; /* isn't dealt with by these */
1497 /* regcomp.c already folded this if pattern is in UTF-8 */
1498 utf8_fold_flags = 0;
1499 goto do_exactf_utf8;
1501 fold_array = PL_fold;
1503 goto do_exactf_non_utf8;
1506 if (is_utf8_pat || utf8_target) {
1507 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1508 goto do_exactf_utf8;
1510 fold_array = PL_fold_locale;
1511 folder = foldEQ_locale;
1512 goto do_exactf_non_utf8;
1516 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1518 goto do_exactf_utf8;
1520 case EXACTFU_TRICKYFOLD:
1522 if (is_utf8_pat || utf8_target) {
1523 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1524 goto do_exactf_utf8;
1527 /* Any 'ss' in the pattern should have been replaced by regcomp,
1528 * so we don't have to worry here about this single special case
1529 * in the Latin1 range */
1530 fold_array = PL_fold_latin1;
1531 folder = foldEQ_latin1;
1535 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1536 are no glitches with fold-length differences
1537 between the target string and pattern */
1539 /* The idea in the non-utf8 EXACTF* cases is to first find the
1540 * first character of the EXACTF* node and then, if necessary,
1541 * case-insensitively compare the full text of the node. c1 is the
1542 * first character. c2 is its fold. This logic will not work for
1543 * Unicode semantics and the german sharp ss, which hence should
1544 * not be compiled into a node that gets here. */
1545 pat_string = STRING(c);
1546 ln = STR_LEN(c); /* length to match in octets/bytes */
1548 /* We know that we have to match at least 'ln' bytes (which is the
1549 * same as characters, since not utf8). If we have to match 3
1550 * characters, and there are only 2 availabe, we know without
1551 * trying that it will fail; so don't start a match past the
1552 * required minimum number from the far end */
1553 e = HOP3c(strend, -((I32)ln), s);
1555 if (!reginfo && e < s) {
1556 e = s; /* Due to minlen logic of intuit() */
1560 c2 = fold_array[c1];
1561 if (c1 == c2) { /* If char and fold are the same */
1562 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1565 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1573 /* If one of the operands is in utf8, we can't use the simpler folding
1574 * above, due to the fact that many different characters can have the
1575 * same fold, or portion of a fold, or different- length fold */
1576 pat_string = STRING(c);
1577 ln = STR_LEN(c); /* length to match in octets/bytes */
1578 pat_end = pat_string + ln;
1579 lnc = is_utf8_pat /* length to match in characters */
1580 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1583 /* We have 'lnc' characters to match in the pattern, but because of
1584 * multi-character folding, each character in the target can match
1585 * up to 3 characters (Unicode guarantees it will never exceed
1586 * this) if it is utf8-encoded; and up to 2 if not (based on the
1587 * fact that the Latin 1 folds are already determined, and the
1588 * only multi-char fold in that range is the sharp-s folding to
1589 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1590 * string character. Adjust lnc accordingly, rounding up, so that
1591 * if we need to match at least 4+1/3 chars, that really is 5. */
1592 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1593 lnc = (lnc + expansion - 1) / expansion;
1595 /* As in the non-UTF8 case, if we have to match 3 characters, and
1596 * only 2 are left, it's guaranteed to fail, so don't start a
1597 * match that would require us to go beyond the end of the string
1599 e = HOP3c(strend, -((I32)lnc), s);
1601 if (!reginfo && e < s) {
1602 e = s; /* Due to minlen logic of intuit() */
1605 /* XXX Note that we could recalculate e to stop the loop earlier,
1606 * as the worst case expansion above will rarely be met, and as we
1607 * go along we would usually find that e moves further to the left.
1608 * This would happen only after we reached the point in the loop
1609 * where if there were no expansion we should fail. Unclear if
1610 * worth the expense */
1613 char *my_strend= (char *)strend;
1614 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1615 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1616 && (!reginfo || regtry(reginfo, &s)) )
1620 s += (utf8_target) ? UTF8SKIP(s) : 1;
1625 RXp_MATCH_TAINTED_on(prog);
1626 FBC_BOUND(isWORDCHAR_LC,
1627 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1628 isWORDCHAR_LC_utf8((U8*)s));
1631 RXp_MATCH_TAINTED_on(prog);
1632 FBC_NBOUND(isWORDCHAR_LC,
1633 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1634 isWORDCHAR_LC_utf8((U8*)s));
1637 FBC_BOUND(isWORDCHAR,
1638 isWORDCHAR_uni(tmp),
1639 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1642 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1644 isWORDCHAR_A((U8*)s));
1647 FBC_NBOUND(isWORDCHAR,
1648 isWORDCHAR_uni(tmp),
1649 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1652 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1654 isWORDCHAR_A((U8*)s));
1657 FBC_BOUND(isWORDCHAR_L1,
1658 isWORDCHAR_uni(tmp),
1659 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1662 FBC_NBOUND(isWORDCHAR_L1,
1663 isWORDCHAR_uni(tmp),
1664 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1667 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1668 is_LNBREAK_latin1_safe(s, strend)
1672 /* The argument to all the POSIX node types is the class number to pass to
1673 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1680 RXp_MATCH_TAINTED_on(prog);
1681 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1682 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1697 /* The complement of something that matches only ASCII matches all
1698 * UTF-8 variant code points, plus everything in ASCII that isn't
1700 REXEC_FBC_UTF8_CLASS_SCAN(! UTF8_IS_INVARIANT(*s)
1701 || ! _generic_isCC_A(*s, FLAGS(c)));
1710 /* Don't need to worry about utf8, as it can match only a single
1711 * byte invariant character. */
1712 REXEC_FBC_CLASS_SCAN(
1713 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1721 if (! utf8_target) {
1722 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1728 classnum = (_char_class_number) FLAGS(c);
1729 if (classnum < _FIRST_NON_SWASH_CC) {
1730 while (s < strend) {
1732 /* We avoid loading in the swash as long as possible, but
1733 * should we have to, we jump to a separate loop. This
1734 * extra 'if' statement is what keeps this code from being
1735 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1736 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1737 goto found_above_latin1;
1739 if ((UTF8_IS_INVARIANT(*s)
1740 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1742 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1743 && to_complement ^ cBOOL(
1744 _generic_isCC(TWO_BYTE_UTF8_TO_UNI(*s, *(s + 1)),
1747 if (tmp && (!reginfo || regtry(reginfo, &s)))
1759 else switch (classnum) { /* These classes are implemented as
1761 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1762 revert the change of \v matching this */
1765 case _CC_ENUM_PSXSPC:
1766 REXEC_FBC_UTF8_CLASS_SCAN(
1767 to_complement ^ cBOOL(isSPACE_utf8(s)));
1770 case _CC_ENUM_BLANK:
1771 REXEC_FBC_UTF8_CLASS_SCAN(
1772 to_complement ^ cBOOL(isBLANK_utf8(s)));
1775 case _CC_ENUM_XDIGIT:
1776 REXEC_FBC_UTF8_CLASS_SCAN(
1777 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1780 case _CC_ENUM_VERTSPACE:
1781 REXEC_FBC_UTF8_CLASS_SCAN(
1782 to_complement ^ cBOOL(isVERTWS_utf8(s)));
1785 case _CC_ENUM_CNTRL:
1786 REXEC_FBC_UTF8_CLASS_SCAN(
1787 to_complement ^ cBOOL(isCNTRL_utf8(s)));
1791 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
1792 assert(0); /* NOTREACHED */
1797 found_above_latin1: /* Here we have to load a swash to get the result
1798 for the current code point */
1799 if (! PL_utf8_swash_ptrs[classnum]) {
1800 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1801 PL_utf8_swash_ptrs[classnum] =
1802 _core_swash_init("utf8", swash_property_names[classnum],
1803 &PL_sv_undef, 1, 0, NULL, &flags);
1806 /* This is a copy of the loop above for swash classes, though using the
1807 * FBC macro instead of being expanded out. Since we've loaded the
1808 * swash, we don't have to check for that each time through the loop */
1809 REXEC_FBC_UTF8_CLASS_SCAN(
1810 to_complement ^ cBOOL(_generic_utf8(
1813 swash_fetch(PL_utf8_swash_ptrs[classnum],
1821 /* what trie are we using right now */
1822 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1823 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
1824 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1826 const char *last_start = strend - trie->minlen;
1828 const char *real_start = s;
1830 STRLEN maxlen = trie->maxlen;
1832 U8 **points; /* map of where we were in the input string
1833 when reading a given char. For ASCII this
1834 is unnecessary overhead as the relationship
1835 is always 1:1, but for Unicode, especially
1836 case folded Unicode this is not true. */
1837 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1841 GET_RE_DEBUG_FLAGS_DECL;
1843 /* We can't just allocate points here. We need to wrap it in
1844 * an SV so it gets freed properly if there is a croak while
1845 * running the match */
1848 sv_points=newSV(maxlen * sizeof(U8 *));
1849 SvCUR_set(sv_points,
1850 maxlen * sizeof(U8 *));
1851 SvPOK_on(sv_points);
1852 sv_2mortal(sv_points);
1853 points=(U8**)SvPV_nolen(sv_points );
1854 if ( trie_type != trie_utf8_fold
1855 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1858 bitmap=(U8*)trie->bitmap;
1860 bitmap=(U8*)ANYOF_BITMAP(c);
1862 /* this is the Aho-Corasick algorithm modified a touch
1863 to include special handling for long "unknown char" sequences.
1864 The basic idea being that we use AC as long as we are dealing
1865 with a possible matching char, when we encounter an unknown char
1866 (and we have not encountered an accepting state) we scan forward
1867 until we find a legal starting char.
1868 AC matching is basically that of trie matching, except that when
1869 we encounter a failing transition, we fall back to the current
1870 states "fail state", and try the current char again, a process
1871 we repeat until we reach the root state, state 1, or a legal
1872 transition. If we fail on the root state then we can either
1873 terminate if we have reached an accepting state previously, or
1874 restart the entire process from the beginning if we have not.
1877 while (s <= last_start) {
1878 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1886 U8 *uscan = (U8*)NULL;
1887 U8 *leftmost = NULL;
1889 U32 accepted_word= 0;
1893 while ( state && uc <= (U8*)strend ) {
1895 U32 word = aho->states[ state ].wordnum;
1899 DEBUG_TRIE_EXECUTE_r(
1900 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1901 dump_exec_pos( (char *)uc, c, strend, real_start,
1902 (char *)uc, utf8_target );
1903 PerlIO_printf( Perl_debug_log,
1904 " Scanning for legal start char...\n");
1908 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1912 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1918 if (uc >(U8*)last_start) break;
1922 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1923 if (!leftmost || lpos < leftmost) {
1924 DEBUG_r(accepted_word=word);
1930 points[pointpos++ % maxlen]= uc;
1931 if (foldlen || uc < (U8*)strend) {
1932 REXEC_TRIE_READ_CHAR(trie_type, trie,
1934 uscan, len, uvc, charid, foldlen,
1936 DEBUG_TRIE_EXECUTE_r({
1937 dump_exec_pos( (char *)uc, c, strend,
1938 real_start, s, utf8_target);
1939 PerlIO_printf(Perl_debug_log,
1940 " Charid:%3u CP:%4"UVxf" ",
1952 word = aho->states[ state ].wordnum;
1954 base = aho->states[ state ].trans.base;
1956 DEBUG_TRIE_EXECUTE_r({
1958 dump_exec_pos( (char *)uc, c, strend, real_start,
1960 PerlIO_printf( Perl_debug_log,
1961 "%sState: %4"UVxf", word=%"UVxf,
1962 failed ? " Fail transition to " : "",
1963 (UV)state, (UV)word);
1969 ( ((offset = base + charid
1970 - 1 - trie->uniquecharcount)) >= 0)
1971 && ((U32)offset < trie->lasttrans)
1972 && trie->trans[offset].check == state
1973 && (tmp=trie->trans[offset].next))
1975 DEBUG_TRIE_EXECUTE_r(
1976 PerlIO_printf( Perl_debug_log," - legal\n"));
1981 DEBUG_TRIE_EXECUTE_r(
1982 PerlIO_printf( Perl_debug_log," - fail\n"));
1984 state = aho->fail[state];
1988 /* we must be accepting here */
1989 DEBUG_TRIE_EXECUTE_r(
1990 PerlIO_printf( Perl_debug_log," - accepting\n"));
1999 if (!state) state = 1;
2002 if ( aho->states[ state ].wordnum ) {
2003 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2004 if (!leftmost || lpos < leftmost) {
2005 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2010 s = (char*)leftmost;
2011 DEBUG_TRIE_EXECUTE_r({
2013 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2014 (UV)accepted_word, (IV)(s - real_start)
2017 if (!reginfo || regtry(reginfo, &s)) {
2023 DEBUG_TRIE_EXECUTE_r({
2024 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2027 DEBUG_TRIE_EXECUTE_r(
2028 PerlIO_printf( Perl_debug_log,"No match.\n"));
2037 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2047 - regexec_flags - match a regexp against a string
2050 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2051 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2052 /* stringarg: the point in the string at which to begin matching */
2053 /* strend: pointer to null at end of string */
2054 /* strbeg: real beginning of string */
2055 /* minend: end of match must be >= minend bytes after stringarg. */
2056 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2057 * itself is accessed via the pointers above */
2058 /* data: May be used for some additional optimizations.
2059 Currently its only used, with a U32 cast, for transmitting
2060 the ganch offset when doing a /g match. This will change */
2061 /* nosave: For optimizations. */
2065 struct regexp *const prog = ReANY(rx);
2068 char *startpos = stringarg;
2069 I32 minlen; /* must match at least this many chars */
2070 I32 dontbother = 0; /* how many characters not to try at end */
2071 I32 end_shift = 0; /* Same for the end. */ /* CC */
2072 I32 scream_pos = -1; /* Internal iterator of scream. */
2073 char *scream_olds = NULL;
2074 const bool utf8_target = cBOOL(DO_UTF8(sv));
2076 RXi_GET_DECL(prog,progi);
2077 regmatch_info reginfo; /* create some info to pass to regtry etc */
2078 regexp_paren_pair *swap = NULL;
2079 GET_RE_DEBUG_FLAGS_DECL;
2081 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2082 PERL_UNUSED_ARG(data);
2084 /* Be paranoid... */
2085 if (prog == NULL || startpos == NULL) {
2086 Perl_croak(aTHX_ "NULL regexp parameter");
2090 multiline = prog->extflags & RXf_PMf_MULTILINE;
2091 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2093 RX_MATCH_UTF8_set(rx, utf8_target);
2095 debug_start_match(rx, utf8_target, startpos, strend,
2099 minlen = prog->minlen;
2101 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2102 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2103 "String too short [regexec_flags]...\n"));
2108 /* Check validity of program. */
2109 if (UCHARAT(progi->program) != REG_MAGIC) {
2110 Perl_croak(aTHX_ "corrupted regexp program");
2113 RX_MATCH_TAINTED_off(rx);
2114 PL_reg_state.re_state_eval_setup_done = FALSE;
2117 reginfo.is_utf8_pat = cBOOL(RX_UTF8(rx));
2118 reginfo.warned = FALSE;
2119 /* Mark beginning of line for ^ and lookbehind. */
2120 reginfo.bol = startpos; /* XXX not used ??? */
2124 /* Mark end of line for $ (and such) */
2127 /* see how far we have to get to not match where we matched before */
2128 reginfo.till = startpos+minend;
2130 /* If there is a "must appear" string, look for it. */
2133 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2135 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2136 reginfo.ganch = startpos + prog->gofs;
2137 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2138 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2139 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2141 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2142 && mg->mg_len >= 0) {
2143 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2144 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2145 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2147 if (prog->extflags & RXf_ANCH_GPOS) {
2148 if (s > reginfo.ganch)
2150 s = reginfo.ganch - prog->gofs;
2151 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2152 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2158 reginfo.ganch = strbeg + PTR2UV(data);
2159 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2160 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2162 } else { /* pos() not defined */
2163 reginfo.ganch = strbeg;
2164 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2165 "GPOS: reginfo.ganch = strbeg\n"));
2168 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2169 /* We have to be careful. If the previous successful match
2170 was from this regex we don't want a subsequent partially
2171 successful match to clobber the old results.
2172 So when we detect this possibility we add a swap buffer
2173 to the re, and switch the buffer each match. If we fail,
2174 we switch it back; otherwise we leave it swapped.
2177 /* do we need a save destructor here for eval dies? */
2178 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2179 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2180 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2186 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2187 re_scream_pos_data d;
2189 d.scream_olds = &scream_olds;
2190 d.scream_pos = &scream_pos;
2191 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2193 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2194 goto phooey; /* not present */
2200 /* Simplest case: anchored match need be tried only once. */
2201 /* [unless only anchor is BOL and multiline is set] */
2202 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2203 if (s == startpos && regtry(®info, &startpos))
2205 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2206 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2211 dontbother = minlen - 1;
2212 end = HOP3c(strend, -dontbother, strbeg) - 1;
2213 /* for multiline we only have to try after newlines */
2214 if (prog->check_substr || prog->check_utf8) {
2215 /* because of the goto we can not easily reuse the macros for bifurcating the
2216 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2219 goto after_try_utf8;
2221 if (regtry(®info, &s)) {
2228 if (prog->extflags & RXf_USE_INTUIT) {
2229 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2238 } /* end search for check string in unicode */
2240 if (s == startpos) {
2241 goto after_try_latin;
2244 if (regtry(®info, &s)) {
2251 if (prog->extflags & RXf_USE_INTUIT) {
2252 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2261 } /* end search for check string in latin*/
2262 } /* end search for check string */
2263 else { /* search for newline */
2265 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2268 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2269 while (s <= end) { /* note it could be possible to match at the end of the string */
2270 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2271 if (regtry(®info, &s))
2275 } /* end search for newline */
2276 } /* end anchored/multiline check string search */
2278 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2280 /* the warning about reginfo.ganch being used without initialization
2281 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2282 and we only enter this block when the same bit is set. */
2283 char *tmp_s = reginfo.ganch - prog->gofs;
2285 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2290 /* Messy cases: unanchored match. */
2291 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2292 /* we have /x+whatever/ */
2293 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2299 if (! prog->anchored_utf8) {
2300 to_utf8_substr(prog);
2302 ch = SvPVX_const(prog->anchored_utf8)[0];
2305 DEBUG_EXECUTE_r( did_match = 1 );
2306 if (regtry(®info, &s)) goto got_it;
2308 while (s < strend && *s == ch)
2315 if (! prog->anchored_substr) {
2316 if (! to_byte_substr(prog)) {
2317 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2320 ch = SvPVX_const(prog->anchored_substr)[0];
2323 DEBUG_EXECUTE_r( did_match = 1 );
2324 if (regtry(®info, &s)) goto got_it;
2326 while (s < strend && *s == ch)
2331 DEBUG_EXECUTE_r(if (!did_match)
2332 PerlIO_printf(Perl_debug_log,
2333 "Did not find anchored character...\n")
2336 else if (prog->anchored_substr != NULL
2337 || prog->anchored_utf8 != NULL
2338 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2339 && prog->float_max_offset < strend - s)) {
2344 char *last1; /* Last position checked before */
2348 if (prog->anchored_substr || prog->anchored_utf8) {
2350 if (! prog->anchored_utf8) {
2351 to_utf8_substr(prog);
2353 must = prog->anchored_utf8;
2356 if (! prog->anchored_substr) {
2357 if (! to_byte_substr(prog)) {
2358 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2361 must = prog->anchored_substr;
2363 back_max = back_min = prog->anchored_offset;
2366 if (! prog->float_utf8) {
2367 to_utf8_substr(prog);
2369 must = prog->float_utf8;
2372 if (! prog->float_substr) {
2373 if (! to_byte_substr(prog)) {
2374 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2377 must = prog->float_substr;
2379 back_max = prog->float_max_offset;
2380 back_min = prog->float_min_offset;
2386 last = HOP3c(strend, /* Cannot start after this */
2387 -(I32)(CHR_SVLEN(must)
2388 - (SvTAIL(must) != 0) + back_min), strbeg);
2391 last1 = HOPc(s, -1);
2393 last1 = s - 1; /* bogus */
2395 /* XXXX check_substr already used to find "s", can optimize if
2396 check_substr==must. */
2398 dontbother = end_shift;
2399 strend = HOPc(strend, -dontbother);
2400 while ( (s <= last) &&
2401 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2402 (unsigned char*)strend, must,
2403 multiline ? FBMrf_MULTILINE : 0)) ) {
2404 DEBUG_EXECUTE_r( did_match = 1 );
2405 if (HOPc(s, -back_max) > last1) {
2406 last1 = HOPc(s, -back_min);
2407 s = HOPc(s, -back_max);
2410 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2412 last1 = HOPc(s, -back_min);
2416 while (s <= last1) {
2417 if (regtry(®info, &s))
2420 s++; /* to break out of outer loop */
2427 while (s <= last1) {
2428 if (regtry(®info, &s))
2434 DEBUG_EXECUTE_r(if (!did_match) {
2435 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2436 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2437 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2438 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2439 ? "anchored" : "floating"),
2440 quoted, RE_SV_TAIL(must));
2444 else if ( (c = progi->regstclass) ) {
2446 const OPCODE op = OP(progi->regstclass);
2447 /* don't bother with what can't match */
2448 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2449 strend = HOPc(strend, -(minlen - 1));
2452 SV * const prop = sv_newmortal();
2453 regprop(prog, prop, c);
2455 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2457 PerlIO_printf(Perl_debug_log,
2458 "Matching stclass %.*s against %s (%d bytes)\n",
2459 (int)SvCUR(prop), SvPVX_const(prop),
2460 quoted, (int)(strend - s));
2463 if (find_byclass(prog, c, s, strend, ®info, reginfo.is_utf8_pat))
2465 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2469 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2477 if (! prog->float_utf8) {
2478 to_utf8_substr(prog);
2480 float_real = prog->float_utf8;
2483 if (! prog->float_substr) {
2484 if (! to_byte_substr(prog)) {
2485 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2488 float_real = prog->float_substr;
2491 little = SvPV_const(float_real, len);
2492 if (SvTAIL(float_real)) {
2493 /* This means that float_real contains an artificial \n on
2494 * the end due to the presence of something like this:
2495 * /foo$/ where we can match both "foo" and "foo\n" at the
2496 * end of the string. So we have to compare the end of the
2497 * string first against the float_real without the \n and
2498 * then against the full float_real with the string. We
2499 * have to watch out for cases where the string might be
2500 * smaller than the float_real or the float_real without
2502 char *checkpos= strend - len;
2504 PerlIO_printf(Perl_debug_log,
2505 "%sChecking for float_real.%s\n",
2506 PL_colors[4], PL_colors[5]));
2507 if (checkpos + 1 < strbeg) {
2508 /* can't match, even if we remove the trailing \n
2509 * string is too short to match */
2511 PerlIO_printf(Perl_debug_log,
2512 "%sString shorter than required trailing substring, cannot match.%s\n",
2513 PL_colors[4], PL_colors[5]));
2515 } else if (memEQ(checkpos + 1, little, len - 1)) {
2516 /* can match, the end of the string matches without the
2518 last = checkpos + 1;
2519 } else if (checkpos < strbeg) {
2520 /* cant match, string is too short when the "\n" is
2523 PerlIO_printf(Perl_debug_log,
2524 "%sString does not contain required trailing substring, cannot match.%s\n",
2525 PL_colors[4], PL_colors[5]));
2527 } else if (!multiline) {
2528 /* non multiline match, so compare with the "\n" at the
2529 * end of the string */
2530 if (memEQ(checkpos, little, len)) {
2534 PerlIO_printf(Perl_debug_log,
2535 "%sString does not contain required trailing substring, cannot match.%s\n",
2536 PL_colors[4], PL_colors[5]));
2540 /* multiline match, so we have to search for a place
2541 * where the full string is located */
2547 last = rninstr(s, strend, little, little + len);
2549 last = strend; /* matching "$" */
2552 /* at one point this block contained a comment which was
2553 * probably incorrect, which said that this was a "should not
2554 * happen" case. Even if it was true when it was written I am
2555 * pretty sure it is not anymore, so I have removed the comment
2556 * and replaced it with this one. Yves */
2558 PerlIO_printf(Perl_debug_log,
2559 "String does not contain required substring, cannot match.\n"
2563 dontbother = strend - last + prog->float_min_offset;
2565 if (minlen && (dontbother < minlen))
2566 dontbother = minlen - 1;
2567 strend -= dontbother; /* this one's always in bytes! */
2568 /* We don't know much -- general case. */
2571 if (regtry(®info, &s))
2580 if (regtry(®info, &s))
2582 } while (s++ < strend);
2592 PerlIO_printf(Perl_debug_log,
2593 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2600 if (PL_reg_state.re_state_eval_setup_done)
2601 restore_pos(aTHX_ prog);
2602 if (RXp_PAREN_NAMES(prog))
2603 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2605 /* make sure $`, $&, $', and $digit will work later */
2606 if ( !(flags & REXEC_NOT_FIRST) ) {
2607 if (flags & REXEC_COPY_STR) {
2611 PerlIO_printf(Perl_debug_log,
2612 "Copy on write: regexp capture, type %d\n",
2615 RX_MATCH_COPY_FREE(rx);
2616 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2617 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2618 assert (SvPOKp(prog->saved_copy));
2619 prog->sublen = PL_regeol - strbeg;
2620 prog->suboffset = 0;
2621 prog->subcoffset = 0;
2626 I32 max = PL_regeol - strbeg;
2629 if ( (flags & REXEC_COPY_SKIP_POST)
2630 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2631 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2632 ) { /* don't copy $' part of string */
2635 /* calculate the right-most part of the string covered
2636 * by a capture. Due to look-ahead, this may be to
2637 * the right of $&, so we have to scan all captures */
2638 while (n <= prog->lastparen) {
2639 if (prog->offs[n].end > max)
2640 max = prog->offs[n].end;
2644 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2645 ? prog->offs[0].start
2647 assert(max >= 0 && max <= PL_regeol - strbeg);
2650 if ( (flags & REXEC_COPY_SKIP_PRE)
2651 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2652 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2653 ) { /* don't copy $` part of string */
2656 /* calculate the left-most part of the string covered
2657 * by a capture. Due to look-behind, this may be to
2658 * the left of $&, so we have to scan all captures */
2659 while (min && n <= prog->lastparen) {
2660 if ( prog->offs[n].start != -1
2661 && prog->offs[n].start < min)
2663 min = prog->offs[n].start;
2667 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2668 && min > prog->offs[0].end
2670 min = prog->offs[0].end;
2674 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2677 if (RX_MATCH_COPIED(rx)) {
2678 if (sublen > prog->sublen)
2680 (char*)saferealloc(prog->subbeg, sublen+1);
2683 prog->subbeg = (char*)safemalloc(sublen+1);
2684 Copy(strbeg + min, prog->subbeg, sublen, char);
2685 prog->subbeg[sublen] = '\0';
2686 prog->suboffset = min;
2687 prog->sublen = sublen;
2688 RX_MATCH_COPIED_on(rx);
2690 prog->subcoffset = prog->suboffset;
2691 if (prog->suboffset && utf8_target) {
2692 /* Convert byte offset to chars.
2693 * XXX ideally should only compute this if @-/@+
2694 * has been seen, a la PL_sawampersand ??? */
2696 /* If there's a direct correspondence between the
2697 * string which we're matching and the original SV,
2698 * then we can use the utf8 len cache associated with
2699 * the SV. In particular, it means that under //g,
2700 * sv_pos_b2u() will use the previously cached
2701 * position to speed up working out the new length of
2702 * subcoffset, rather than counting from the start of
2703 * the string each time. This stops
2704 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2705 * from going quadratic */
2706 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2707 sv_pos_b2u(sv, &(prog->subcoffset));
2709 prog->subcoffset = utf8_length((U8*)strbeg,
2710 (U8*)(strbeg+prog->suboffset));
2714 RX_MATCH_COPY_FREE(rx);
2715 prog->subbeg = strbeg;
2716 prog->suboffset = 0;
2717 prog->subcoffset = 0;
2718 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2725 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2726 PL_colors[4], PL_colors[5]));
2727 if (PL_reg_state.re_state_eval_setup_done)
2728 restore_pos(aTHX_ prog);
2730 /* we failed :-( roll it back */
2731 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2732 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2737 Safefree(prog->offs);
2744 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2745 * Do inc before dec, in case old and new rex are the same */
2746 #define SET_reg_curpm(Re2) \
2747 if (PL_reg_state.re_state_eval_setup_done) { \
2748 (void)ReREFCNT_inc(Re2); \
2749 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2750 PM_SETRE((PL_reg_curpm), (Re2)); \
2755 - regtry - try match at specific point
2757 STATIC I32 /* 0 failure, 1 success */
2758 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2762 REGEXP *const rx = reginfo->prog;
2763 regexp *const prog = ReANY(rx);
2765 RXi_GET_DECL(prog,progi);
2766 GET_RE_DEBUG_FLAGS_DECL;
2768 PERL_ARGS_ASSERT_REGTRY;
2770 reginfo->cutpoint=NULL;
2772 if ((prog->extflags & RXf_EVAL_SEEN)
2773 && !PL_reg_state.re_state_eval_setup_done)
2777 PL_reg_state.re_state_eval_setup_done = TRUE;
2779 /* Make $_ available to executed code. */
2780 if (reginfo->sv != DEFSV) {
2782 DEFSV_set(reginfo->sv);
2785 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2786 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2787 /* prepare for quick setting of pos */
2788 #ifdef PERL_OLD_COPY_ON_WRITE
2789 if (SvIsCOW(reginfo->sv))
2790 sv_force_normal_flags(reginfo->sv, 0);
2792 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2793 &PL_vtbl_mglob, NULL, 0);
2797 PL_reg_oldpos = mg->mg_len;
2798 SAVEDESTRUCTOR_X(restore_pos, prog);
2800 if (!PL_reg_curpm) {
2801 Newxz(PL_reg_curpm, 1, PMOP);
2804 SV* const repointer = &PL_sv_undef;
2805 /* this regexp is also owned by the new PL_reg_curpm, which
2806 will try to free it. */
2807 av_push(PL_regex_padav, repointer);
2808 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2809 PL_regex_pad = AvARRAY(PL_regex_padav);
2814 PL_reg_oldcurpm = PL_curpm;
2815 PL_curpm = PL_reg_curpm;
2816 if (RXp_MATCH_COPIED(prog)) {
2817 /* Here is a serious problem: we cannot rewrite subbeg,
2818 since it may be needed if this match fails. Thus
2819 $` inside (?{}) could fail... */
2820 PL_reg_oldsaved = prog->subbeg;
2821 PL_reg_oldsavedlen = prog->sublen;
2822 PL_reg_oldsavedoffset = prog->suboffset;
2823 PL_reg_oldsavedcoffset = prog->suboffset;
2825 PL_nrs = prog->saved_copy;
2827 RXp_MATCH_COPIED_off(prog);
2830 PL_reg_oldsaved = NULL;
2831 prog->subbeg = PL_bostr;
2832 prog->suboffset = 0;
2833 prog->subcoffset = 0;
2834 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2837 PL_reg_starttry = *startposp;
2839 prog->offs[0].start = *startposp - PL_bostr;
2840 prog->lastparen = 0;
2841 prog->lastcloseparen = 0;
2843 /* XXXX What this code is doing here?!!! There should be no need
2844 to do this again and again, prog->lastparen should take care of
2847 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2848 * Actually, the code in regcppop() (which Ilya may be meaning by
2849 * prog->lastparen), is not needed at all by the test suite
2850 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2851 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2852 * Meanwhile, this code *is* needed for the
2853 * above-mentioned test suite tests to succeed. The common theme
2854 * on those tests seems to be returning null fields from matches.
2855 * --jhi updated by dapm */
2857 if (prog->nparens) {
2858 regexp_paren_pair *pp = prog->offs;
2860 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2868 result = regmatch(reginfo, *startposp, progi->program + 1);
2870 prog->offs[0].end = result;
2873 if (reginfo->cutpoint)
2874 *startposp= reginfo->cutpoint;
2875 REGCP_UNWIND(lastcp);
2880 #define sayYES goto yes
2881 #define sayNO goto no
2882 #define sayNO_SILENT goto no_silent
2884 /* we dont use STMT_START/END here because it leads to
2885 "unreachable code" warnings, which are bogus, but distracting. */
2886 #define CACHEsayNO \
2887 if (ST.cache_mask) \
2888 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2891 /* this is used to determine how far from the left messages like
2892 'failed...' are printed. It should be set such that messages
2893 are inline with the regop output that created them.
2895 #define REPORT_CODE_OFF 32
2898 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2899 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2900 #define CHRTEST_NOT_A_CP_1 -999
2901 #define CHRTEST_NOT_A_CP_2 -998
2903 #define SLAB_FIRST(s) (&(s)->states[0])
2904 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2906 /* grab a new slab and return the first slot in it */
2908 STATIC regmatch_state *
2911 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2914 regmatch_slab *s = PL_regmatch_slab->next;
2916 Newx(s, 1, regmatch_slab);
2917 s->prev = PL_regmatch_slab;
2919 PL_regmatch_slab->next = s;
2921 PL_regmatch_slab = s;
2922 return SLAB_FIRST(s);
2926 /* push a new state then goto it */
2928 #define PUSH_STATE_GOTO(state, node, input) \
2929 pushinput = input; \
2931 st->resume_state = state; \
2934 /* push a new state with success backtracking, then goto it */
2936 #define PUSH_YES_STATE_GOTO(state, node, input) \
2937 pushinput = input; \
2939 st->resume_state = state; \
2940 goto push_yes_state;
2947 regmatch() - main matching routine
2949 This is basically one big switch statement in a loop. We execute an op,
2950 set 'next' to point the next op, and continue. If we come to a point which
2951 we may need to backtrack to on failure such as (A|B|C), we push a
2952 backtrack state onto the backtrack stack. On failure, we pop the top
2953 state, and re-enter the loop at the state indicated. If there are no more
2954 states to pop, we return failure.
2956 Sometimes we also need to backtrack on success; for example /A+/, where
2957 after successfully matching one A, we need to go back and try to
2958 match another one; similarly for lookahead assertions: if the assertion
2959 completes successfully, we backtrack to the state just before the assertion
2960 and then carry on. In these cases, the pushed state is marked as
2961 'backtrack on success too'. This marking is in fact done by a chain of
2962 pointers, each pointing to the previous 'yes' state. On success, we pop to
2963 the nearest yes state, discarding any intermediate failure-only states.
2964 Sometimes a yes state is pushed just to force some cleanup code to be
2965 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2966 it to free the inner regex.
2968 Note that failure backtracking rewinds the cursor position, while
2969 success backtracking leaves it alone.
2971 A pattern is complete when the END op is executed, while a subpattern
2972 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2973 ops trigger the "pop to last yes state if any, otherwise return true"
2976 A common convention in this function is to use A and B to refer to the two
2977 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2978 the subpattern to be matched possibly multiple times, while B is the entire
2979 rest of the pattern. Variable and state names reflect this convention.
2981 The states in the main switch are the union of ops and failure/success of
2982 substates associated with with that op. For example, IFMATCH is the op
2983 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2984 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2985 successfully matched A and IFMATCH_A_fail is a state saying that we have
2986 just failed to match A. Resume states always come in pairs. The backtrack
2987 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2988 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2989 on success or failure.
2991 The struct that holds a backtracking state is actually a big union, with
2992 one variant for each major type of op. The variable st points to the
2993 top-most backtrack struct. To make the code clearer, within each
2994 block of code we #define ST to alias the relevant union.
2996 Here's a concrete example of a (vastly oversimplified) IFMATCH
3002 #define ST st->u.ifmatch
3004 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3005 ST.foo = ...; // some state we wish to save
3007 // push a yes backtrack state with a resume value of
3008 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3010 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3013 case IFMATCH_A: // we have successfully executed A; now continue with B
3015 bar = ST.foo; // do something with the preserved value
3018 case IFMATCH_A_fail: // A failed, so the assertion failed
3019 ...; // do some housekeeping, then ...
3020 sayNO; // propagate the failure
3027 For any old-timers reading this who are familiar with the old recursive
3028 approach, the code above is equivalent to:
3030 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3039 ...; // do some housekeeping, then ...
3040 sayNO; // propagate the failure
3043 The topmost backtrack state, pointed to by st, is usually free. If you
3044 want to claim it, populate any ST.foo fields in it with values you wish to
3045 save, then do one of
3047 PUSH_STATE_GOTO(resume_state, node, newinput);
3048 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3050 which sets that backtrack state's resume value to 'resume_state', pushes a
3051 new free entry to the top of the backtrack stack, then goes to 'node'.
3052 On backtracking, the free slot is popped, and the saved state becomes the
3053 new free state. An ST.foo field in this new top state can be temporarily
3054 accessed to retrieve values, but once the main loop is re-entered, it
3055 becomes available for reuse.
3057 Note that the depth of the backtrack stack constantly increases during the
3058 left-to-right execution of the pattern, rather than going up and down with
3059 the pattern nesting. For example the stack is at its maximum at Z at the
3060 end of the pattern, rather than at X in the following:
3062 /(((X)+)+)+....(Y)+....Z/
3064 The only exceptions to this are lookahead/behind assertions and the cut,
3065 (?>A), which pop all the backtrack states associated with A before
3068 Backtrack state structs are allocated in slabs of about 4K in size.
3069 PL_regmatch_state and st always point to the currently active state,
3070 and PL_regmatch_slab points to the slab currently containing
3071 PL_regmatch_state. The first time regmatch() is called, the first slab is
3072 allocated, and is never freed until interpreter destruction. When the slab
3073 is full, a new one is allocated and chained to the end. At exit from
3074 regmatch(), slabs allocated since entry are freed.
3079 #define DEBUG_STATE_pp(pp) \
3081 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3082 PerlIO_printf(Perl_debug_log, \
3083 " %*s"pp" %s%s%s%s%s\n", \
3085 PL_reg_name[st->resume_state], \
3086 ((st==yes_state||st==mark_state) ? "[" : ""), \
3087 ((st==yes_state) ? "Y" : ""), \
3088 ((st==mark_state) ? "M" : ""), \
3089 ((st==yes_state||st==mark_state) ? "]" : "") \
3094 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3099 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3100 const char *start, const char *end, const char *blurb)
3102 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3104 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3109 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3110 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3112 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3113 start, end - start, 60);
3115 PerlIO_printf(Perl_debug_log,
3116 "%s%s REx%s %s against %s\n",
3117 PL_colors[4], blurb, PL_colors[5], s0, s1);
3119 if (utf8_target||utf8_pat)
3120 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3121 utf8_pat ? "pattern" : "",
3122 utf8_pat && utf8_target ? " and " : "",
3123 utf8_target ? "string" : ""
3129 S_dump_exec_pos(pTHX_ const char *locinput,
3130 const regnode *scan,
3131 const char *loc_regeol,
3132 const char *loc_bostr,
3133 const char *loc_reg_starttry,
3134 const bool utf8_target)
3136 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3137 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3138 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3139 /* The part of the string before starttry has one color
3140 (pref0_len chars), between starttry and current
3141 position another one (pref_len - pref0_len chars),
3142 after the current position the third one.
3143 We assume that pref0_len <= pref_len, otherwise we
3144 decrease pref0_len. */
3145 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3146 ? (5 + taill) - l : locinput - loc_bostr;
3149 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3151 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3153 pref0_len = pref_len - (locinput - loc_reg_starttry);
3154 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3155 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3156 ? (5 + taill) - pref_len : loc_regeol - locinput);
3157 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3161 if (pref0_len > pref_len)
3162 pref0_len = pref_len;
3164 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3166 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3167 (locinput - pref_len),pref0_len, 60, 4, 5);
3169 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3170 (locinput - pref_len + pref0_len),
3171 pref_len - pref0_len, 60, 2, 3);
3173 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3174 locinput, loc_regeol - locinput, 10, 0, 1);
3176 const STRLEN tlen=len0+len1+len2;
3177 PerlIO_printf(Perl_debug_log,
3178 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3179 (IV)(locinput - loc_bostr),
3182 (docolor ? "" : "> <"),
3184 (int)(tlen > 19 ? 0 : 19 - tlen),
3191 /* reg_check_named_buff_matched()
3192 * Checks to see if a named buffer has matched. The data array of
3193 * buffer numbers corresponding to the buffer is expected to reside
3194 * in the regexp->data->data array in the slot stored in the ARG() of
3195 * node involved. Note that this routine doesn't actually care about the
3196 * name, that information is not preserved from compilation to execution.
3197 * Returns the index of the leftmost defined buffer with the given name
3198 * or 0 if non of the buffers matched.
3201 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3204 RXi_GET_DECL(rex,rexi);
3205 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3206 I32 *nums=(I32*)SvPVX(sv_dat);
3208 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3210 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3211 if ((I32)rex->lastparen >= nums[n] &&
3212 rex->offs[nums[n]].end != -1)
3221 /* free all slabs above current one - called during LEAVE_SCOPE */
3224 S_clear_backtrack_stack(pTHX_ void *p)
3226 regmatch_slab *s = PL_regmatch_slab->next;
3231 PL_regmatch_slab->next = NULL;
3233 regmatch_slab * const osl = s;
3239 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3240 U8* c1_utf8, int *c2p, U8* c2_utf8, bool is_utf8_pat)
3242 /* This function determines if there are one or two characters that match
3243 * the first character of the passed-in EXACTish node <text_node>, and if
3244 * so, returns them in the passed-in pointers.
3246 * If it determines that no possible character in the target string can
3247 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3248 * the first character in <text_node> requires UTF-8 to represent, and the
3249 * target string isn't in UTF-8.)
3251 * If there are more than two characters that could match the beginning of
3252 * <text_node>, or if more context is required to determine a match or not,
3253 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3255 * The motiviation behind this function is to allow the caller to set up
3256 * tight loops for matching. If <text_node> is of type EXACT, there is
3257 * only one possible character that can match its first character, and so
3258 * the situation is quite simple. But things get much more complicated if
3259 * folding is involved. It may be that the first character of an EXACTFish
3260 * node doesn't participate in any possible fold, e.g., punctuation, so it
3261 * can be matched only by itself. The vast majority of characters that are
3262 * in folds match just two things, their lower and upper-case equivalents.
3263 * But not all are like that; some have multiple possible matches, or match
3264 * sequences of more than one character. This function sorts all that out.
3266 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3267 * loop of trying to match A*, we know we can't exit where the thing
3268 * following it isn't a B. And something can't be a B unless it is the
3269 * beginning of B. By putting a quick test for that beginning in a tight
3270 * loop, we can rule out things that can't possibly be B without having to
3271 * break out of the loop, thus avoiding work. Similarly, if A is a single
3272 * character, we can make a tight loop matching A*, using the outputs of
3275 * If the target string to match isn't in UTF-8, and there aren't
3276 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3277 * the one or two possible octets (which are characters in this situation)
3278 * that can match. In all cases, if there is only one character that can
3279 * match, *<c1p> and *<c2p> will be identical.
3281 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3282 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3283 * can match the beginning of <text_node>. They should be declared with at
3284 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3285 * undefined what these contain.) If one or both of the buffers are
3286 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3287 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3288 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3289 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3290 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3292 const bool utf8_target = PL_reg_match_utf8;
3294 UV c1 = CHRTEST_NOT_A_CP_1;
3295 UV c2 = CHRTEST_NOT_A_CP_2;
3296 bool use_chrtest_void = FALSE;
3298 /* Used when we have both utf8 input and utf8 output, to avoid converting
3299 * to/from code points */
3300 bool utf8_has_been_setup = FALSE;
3304 U8 *pat = (U8*)STRING(text_node);
3306 if (OP(text_node) == EXACT) {
3308 /* In an exact node, only one thing can be matched, that first
3309 * character. If both the pat and the target are UTF-8, we can just
3310 * copy the input to the output, avoiding finding the code point of
3315 else if (utf8_target) {
3316 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3317 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3318 utf8_has_been_setup = TRUE;
3321 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3324 else /* an EXACTFish node */
3326 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3327 pat + STR_LEN(text_node)))
3329 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3330 pat + STR_LEN(text_node))))
3332 /* Multi-character folds require more context to sort out. Also
3333 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3334 * handled outside this routine */
3335 use_chrtest_void = TRUE;
3337 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3338 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3340 /* Load the folds hash, if not already done */
3342 if (! PL_utf8_foldclosures) {
3343 if (! PL_utf8_tofold) {
3344 U8 dummy[UTF8_MAXBYTES+1];
3346 /* Force loading this by folding an above-Latin1 char */
3347 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3348 assert(PL_utf8_tofold); /* Verify that worked */
3350 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3353 /* The fold closures data structure is a hash with the keys being
3354 * the UTF-8 of every character that is folded to, like 'k', and
3355 * the values each an array of all code points that fold to its
3356 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3358 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3363 /* Not found in the hash, therefore there are no folds
3364 * containing it, so there is only a single character that
3368 else { /* Does participate in folds */
3369 AV* list = (AV*) *listp;
3370 if (av_len(list) != 1) {
3372 /* If there aren't exactly two folds to this, it is outside
3373 * the scope of this function */
3374 use_chrtest_void = TRUE;
3376 else { /* There are two. Get them */
3377 SV** c_p = av_fetch(list, 0, FALSE);
3379 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3383 c_p = av_fetch(list, 1, FALSE);
3385 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3389 /* Folds that cross the 255/256 boundary are forbidden if
3390 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3391 * pattern character is above 256, and its only other match
3392 * is below 256, the only legal match will be to itself.
3393 * We have thrown away the original, so have to compute
3394 * which is the one above 255 */
3395 if ((c1 < 256) != (c2 < 256)) {
3396 if (OP(text_node) == EXACTFL
3397 || (OP(text_node) == EXACTFA
3398 && (isASCII(c1) || isASCII(c2))))
3411 else /* Here, c1 is < 255 */
3413 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3414 && OP(text_node) != EXACTFL
3415 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3417 /* Here, there could be something above Latin1 in the target which
3418 * folds to this character in the pattern. All such cases except
3419 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3420 * involved in their folds, so are outside the scope of this
3422 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3423 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3426 use_chrtest_void = TRUE;
3429 else { /* Here nothing above Latin1 can fold to the pattern character */
3430 switch (OP(text_node)) {
3432 case EXACTFL: /* /l rules */
3433 c2 = PL_fold_locale[c1];
3437 if (! utf8_target) { /* /d rules */
3442 /* /u rules for all these. This happens to work for
3443 * EXACTFA as nothing in Latin1 folds to ASCII */
3445 case EXACTFU_TRICKYFOLD:
3448 c2 = PL_fold_latin1[c1];
3452 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3453 assert(0); /* NOTREACHED */
3458 /* Here have figured things out. Set up the returns */
3459 if (use_chrtest_void) {
3460 *c2p = *c1p = CHRTEST_VOID;
3462 else if (utf8_target) {
3463 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3464 uvchr_to_utf8(c1_utf8, c1);
3465 uvchr_to_utf8(c2_utf8, c2);
3468 /* Invariants are stored in both the utf8 and byte outputs; Use
3469 * negative numbers otherwise for the byte ones. Make sure that the
3470 * byte ones are the same iff the utf8 ones are the same */
3471 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3472 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3475 ? CHRTEST_NOT_A_CP_1
3476 : CHRTEST_NOT_A_CP_2;
3478 else if (c1 > 255) {
3479 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3484 *c1p = *c2p = c2; /* c2 is the only representable value */
3486 else { /* c1 is representable; see about c2 */
3488 *c2p = (c2 < 256) ? c2 : c1;
3494 /* returns -1 on failure, $+[0] on success */
3496 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3498 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3502 const bool utf8_target = PL_reg_match_utf8;
3503 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3504 REGEXP *rex_sv = reginfo->prog;
3505 regexp *rex = ReANY(rex_sv);
3506 RXi_GET_DECL(rex,rexi);
3508 /* the current state. This is a cached copy of PL_regmatch_state */
3510 /* cache heavy used fields of st in registers */
3513 U32 n = 0; /* general value; init to avoid compiler warning */
3514 I32 ln = 0; /* len or last; init to avoid compiler warning */
3515 char *locinput = startpos;
3516 char *pushinput; /* where to continue after a PUSH */
3517 I32 nextchr; /* is always set to UCHARAT(locinput) */
3519 bool result = 0; /* return value of S_regmatch */
3520 int depth = 0; /* depth of backtrack stack */
3521 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3522 const U32 max_nochange_depth =
3523 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3524 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3525 regmatch_state *yes_state = NULL; /* state to pop to on success of
3527 /* mark_state piggy backs on the yes_state logic so that when we unwind
3528 the stack on success we can update the mark_state as we go */
3529 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3530 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3531 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3533 bool no_final = 0; /* prevent failure from backtracking? */
3534 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3535 char *startpoint = locinput;
3536 SV *popmark = NULL; /* are we looking for a mark? */
3537 SV *sv_commit = NULL; /* last mark name seen in failure */
3538 SV *sv_yes_mark = NULL; /* last mark name we have seen
3539 during a successful match */
3540 U32 lastopen = 0; /* last open we saw */
3541 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3542 SV* const oreplsv = GvSV(PL_replgv);
3543 /* these three flags are set by various ops to signal information to
3544 * the very next op. They have a useful lifetime of exactly one loop
3545 * iteration, and are not preserved or restored by state pushes/pops
3547 bool sw = 0; /* the condition value in (?(cond)a|b) */
3548 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3549 int logical = 0; /* the following EVAL is:
3553 or the following IFMATCH/UNLESSM is:
3554 false: plain (?=foo)
3555 true: used as a condition: (?(?=foo))
3557 PAD* last_pad = NULL;
3559 I32 gimme = G_SCALAR;
3560 CV *caller_cv = NULL; /* who called us */
3561 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3562 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3563 U32 maxopenparen = 0; /* max '(' index seen so far */
3564 int to_complement; /* Invert the result? */
3565 _char_class_number classnum;
3566 bool is_utf8_pat = reginfo->is_utf8_pat;
3569 GET_RE_DEBUG_FLAGS_DECL;
3572 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3573 multicall_oldcatch = 0;
3574 multicall_cv = NULL;
3576 PERL_UNUSED_VAR(multicall_cop);
3577 PERL_UNUSED_VAR(newsp);
3580 PERL_ARGS_ASSERT_REGMATCH;
3582 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3583 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3585 /* on first ever call to regmatch, allocate first slab */
3586 if (!PL_regmatch_slab) {
3587 Newx(PL_regmatch_slab, 1, regmatch_slab);
3588 PL_regmatch_slab->prev = NULL;
3589 PL_regmatch_slab->next = NULL;
3590 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3593 oldsave = PL_savestack_ix;
3594 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3595 SAVEVPTR(PL_regmatch_slab);
3596 SAVEVPTR(PL_regmatch_state);
3598 /* grab next free state slot */
3599 st = ++PL_regmatch_state;
3600 if (st > SLAB_LAST(PL_regmatch_slab))
3601 st = PL_regmatch_state = S_push_slab(aTHX);
3603 /* Note that nextchr is a byte even in UTF */
3606 while (scan != NULL) {
3609 SV * const prop = sv_newmortal();
3610 regnode *rnext=regnext(scan);
3611 DUMP_EXEC_POS( locinput, scan, utf8_target );
3612 regprop(rex, prop, scan);
3614 PerlIO_printf(Perl_debug_log,
3615 "%3"IVdf":%*s%s(%"IVdf")\n",
3616 (IV)(scan - rexi->program), depth*2, "",
3618 (PL_regkind[OP(scan)] == END || !rnext) ?
3619 0 : (IV)(rnext - rexi->program));
3622 next = scan + NEXT_OFF(scan);
3625 state_num = OP(scan);
3627 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3632 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3634 switch (state_num) {
3635 case BOL: /* /^../ */
3636 if (locinput == PL_bostr)
3638 /* reginfo->till = reginfo->bol; */
3643 case MBOL: /* /^../m */
3644 if (locinput == PL_bostr ||
3645 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3651 case SBOL: /* /^../s */
3652 if (locinput == PL_bostr)
3657 if (locinput == reginfo->ganch)
3661 case KEEPS: /* \K */
3662 /* update the startpoint */
3663 st->u.keeper.val = rex->offs[0].start;
3664 rex->offs[0].start = locinput - PL_bostr;
3665 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3666 assert(0); /*NOTREACHED*/
3667 case KEEPS_next_fail:
3668 /* rollback the start point change */
3669 rex->offs[0].start = st->u.keeper.val;
3671 assert(0); /*NOTREACHED*/
3673 case EOL: /* /..$/ */
3676 case MEOL: /* /..$/m */
3677 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3681 case SEOL: /* /..$/s */
3683 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3685 if (PL_regeol - locinput > 1)
3690 if (!NEXTCHR_IS_EOS)
3694 case SANY: /* /./s */
3697 goto increment_locinput;
3705 case REG_ANY: /* /./ */
3706 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3708 goto increment_locinput;
3712 #define ST st->u.trie
3713 case TRIEC: /* (ab|cd) with known charclass */
3714 /* In this case the charclass data is available inline so
3715 we can fail fast without a lot of extra overhead.
3717 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3719 PerlIO_printf(Perl_debug_log,
3720 "%*s %sfailed to match trie start class...%s\n",
3721 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3724 assert(0); /* NOTREACHED */
3727 case TRIE: /* (ab|cd) */
3728 /* the basic plan of execution of the trie is:
3729 * At the beginning, run though all the states, and
3730 * find the longest-matching word. Also remember the position
3731 * of the shortest matching word. For example, this pattern:
3734 * when matched against the string "abcde", will generate
3735 * accept states for all words except 3, with the longest
3736 * matching word being 4, and the shortest being 2 (with
3737 * the position being after char 1 of the string).
3739 * Then for each matching word, in word order (i.e. 1,2,4,5),
3740 * we run the remainder of the pattern; on each try setting
3741 * the current position to the character following the word,
3742 * returning to try the next word on failure.
3744 * We avoid having to build a list of words at runtime by
3745 * using a compile-time structure, wordinfo[].prev, which
3746 * gives, for each word, the previous accepting word (if any).
3747 * In the case above it would contain the mappings 1->2, 2->0,
3748 * 3->0, 4->5, 5->1. We can use this table to generate, from
3749 * the longest word (4 above), a list of all words, by
3750 * following the list of prev pointers; this gives us the
3751 * unordered list 4,5,1,2. Then given the current word we have
3752 * just tried, we can go through the list and find the
3753 * next-biggest word to try (so if we just failed on word 2,
3754 * the next in the list is 4).
3756 * Since at runtime we don't record the matching position in
3757 * the string for each word, we have to work that out for
3758 * each word we're about to process. The wordinfo table holds
3759 * the character length of each word; given that we recorded
3760 * at the start: the position of the shortest word and its
3761 * length in chars, we just need to move the pointer the
3762 * difference between the two char lengths. Depending on
3763 * Unicode status and folding, that's cheap or expensive.
3765 * This algorithm is optimised for the case where are only a
3766 * small number of accept states, i.e. 0,1, or maybe 2.
3767 * With lots of accepts states, and having to try all of them,
3768 * it becomes quadratic on number of accept states to find all
3773 /* what type of TRIE am I? (utf8 makes this contextual) */
3774 DECL_TRIE_TYPE(scan);
3776 /* what trie are we using right now */
3777 reg_trie_data * const trie
3778 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3779 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3780 U32 state = trie->startstate;
3783 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3785 if (trie->states[ state ].wordnum) {
3787 PerlIO_printf(Perl_debug_log,
3788 "%*s %smatched empty string...%s\n",
3789 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3795 PerlIO_printf(Perl_debug_log,
3796 "%*s %sfailed to match trie start class...%s\n",
3797 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3804 U8 *uc = ( U8* )locinput;
3808 U8 *uscan = (U8*)NULL;
3809 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3810 U32 charcount = 0; /* how many input chars we have matched */
3811 U32 accepted = 0; /* have we seen any accepting states? */
3813 ST.jump = trie->jump;
3816 ST.longfold = FALSE; /* char longer if folded => it's harder */
3819 /* fully traverse the TRIE; note the position of the
3820 shortest accept state and the wordnum of the longest
3823 while ( state && uc <= (U8*)PL_regeol ) {
3824 U32 base = trie->states[ state ].trans.base;
3828 wordnum = trie->states[ state ].wordnum;
3830 if (wordnum) { /* it's an accept state */
3833 /* record first match position */
3835 ST.firstpos = (U8*)locinput;
3840 ST.firstchars = charcount;
3843 if (!ST.nextword || wordnum < ST.nextword)
3844 ST.nextword = wordnum;
3845 ST.topword = wordnum;
3848 DEBUG_TRIE_EXECUTE_r({
3849 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3850 PerlIO_printf( Perl_debug_log,
3851 "%*s %sState: %4"UVxf" Accepted: %c ",
3852 2+depth * 2, "", PL_colors[4],
3853 (UV)state, (accepted ? 'Y' : 'N'));
3856 /* read a char and goto next state */
3857 if ( base && (foldlen || uc < (U8*)PL_regeol)) {
3859 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3860 uscan, len, uvc, charid, foldlen,
3867 base + charid - 1 - trie->uniquecharcount)) >= 0)
3869 && ((U32)offset < trie->lasttrans)
3870 && trie->trans[offset].check == state)
3872 state = trie->trans[offset].next;
3883 DEBUG_TRIE_EXECUTE_r(
3884 PerlIO_printf( Perl_debug_log,
3885 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3886 charid, uvc, (UV)state, PL_colors[5] );
3892 /* calculate total number of accept states */
3897 w = trie->wordinfo[w].prev;
3900 ST.accepted = accepted;
3904 PerlIO_printf( Perl_debug_log,
3905 "%*s %sgot %"IVdf" possible matches%s\n",
3906 REPORT_CODE_OFF + depth * 2, "",
3907 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3909 goto trie_first_try; /* jump into the fail handler */
3911 assert(0); /* NOTREACHED */
3913 case TRIE_next_fail: /* we failed - try next alternative */
3917 REGCP_UNWIND(ST.cp);
3918 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3920 if (!--ST.accepted) {
3922 PerlIO_printf( Perl_debug_log,
3923 "%*s %sTRIE failed...%s\n",
3924 REPORT_CODE_OFF+depth*2, "",
3931 /* Find next-highest word to process. Note that this code
3932 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3935 U16 const nextword = ST.nextword;
3936 reg_trie_wordinfo * const wordinfo
3937 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3938 for (word=ST.topword; word; word=wordinfo[word].prev) {
3939 if (word > nextword && (!min || word < min))
3952 ST.lastparen = rex->lastparen;
3953 ST.lastcloseparen = rex->lastcloseparen;
3957 /* find start char of end of current word */
3959 U32 chars; /* how many chars to skip */
3960 reg_trie_data * const trie
3961 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3963 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3965 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3970 /* the hard option - fold each char in turn and find
3971 * its folded length (which may be different */
3972 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3980 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3988 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3993 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
4009 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4010 ? ST.jump[ST.nextword]
4014 PerlIO_printf( Perl_debug_log,
4015 "%*s %sTRIE matched word #%d, continuing%s\n",
4016 REPORT_CODE_OFF+depth*2, "",
4023 if (ST.accepted > 1 || has_cutgroup) {
4024 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4025 assert(0); /* NOTREACHED */
4027 /* only one choice left - just continue */
4029 AV *const trie_words
4030 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4031 SV ** const tmp = av_fetch( trie_words,
4033 SV *sv= tmp ? sv_newmortal() : NULL;
4035 PerlIO_printf( Perl_debug_log,
4036 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4037 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4039 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4040 PL_colors[0], PL_colors[1],
4041 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4043 : "not compiled under -Dr",
4047 locinput = (char*)uc;
4048 continue; /* execute rest of RE */
4049 assert(0); /* NOTREACHED */
4053 case EXACT: { /* /abc/ */
4054 char *s = STRING(scan);
4056 if (utf8_target != is_utf8_pat) {
4057 /* The target and the pattern have differing utf8ness. */
4059 const char * const e = s + ln;
4062 /* The target is utf8, the pattern is not utf8.
4063 * Above-Latin1 code points can't match the pattern;
4064 * invariants match exactly, and the other Latin1 ones need
4065 * to be downgraded to a single byte in order to do the
4066 * comparison. (If we could be confident that the target
4067 * is not malformed, this could be refactored to have fewer
4068 * tests by just assuming that if the first bytes match, it
4069 * is an invariant, but there are tests in the test suite
4070 * dealing with (??{...}) which violate this) */
4072 if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) l)) {
4075 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4082 if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
4091 /* The target is not utf8, the pattern is utf8. */
4093 if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4097 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4104 if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
4115 /* The target and the pattern have the same utf8ness. */
4116 /* Inline the first character, for speed. */
4117 if (PL_regeol - locinput < ln
4118 || UCHARAT(s) != nextchr
4119 || (ln > 1 && memNE(s, locinput, ln)))
4128 case EXACTFL: { /* /abc/il */
4130 const U8 * fold_array;
4132 U32 fold_utf8_flags;
4134 RX_MATCH_TAINTED_on(reginfo->prog);
4135 folder = foldEQ_locale;
4136 fold_array = PL_fold_locale;
4137 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
4140 case EXACTFU_SS: /* /\x{df}/iu */
4141 case EXACTFU_TRICKYFOLD: /* /\x{390}/iu */
4142 case EXACTFU: /* /abc/iu */
4143 folder = foldEQ_latin1;
4144 fold_array = PL_fold_latin1;
4145 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4148 case EXACTFA: /* /abc/iaa */
4149 folder = foldEQ_latin1;
4150 fold_array = PL_fold_latin1;
4151 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4154 case EXACTF: /* /abc/i */
4156 fold_array = PL_fold;
4157 fold_utf8_flags = 0;
4163 if (utf8_target || is_utf8_pat || state_num == EXACTFU_SS) {
4164 /* Either target or the pattern are utf8, or has the issue where
4165 * the fold lengths may differ. */
4166 const char * const l = locinput;
4167 char *e = PL_regeol;
4169 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4170 l, &e, 0, utf8_target, fold_utf8_flags))
4178 /* Neither the target nor the pattern are utf8 */
4179 if (UCHARAT(s) != nextchr
4181 && UCHARAT(s) != fold_array[nextchr])
4185 if (PL_regeol - locinput < ln)
4187 if (ln > 1 && ! folder(s, locinput, ln))
4193 /* XXX Could improve efficiency by separating these all out using a
4194 * macro or in-line function. At that point regcomp.c would no longer
4195 * have to set the FLAGS fields of these */
4196 case BOUNDL: /* /\b/l */
4197 case NBOUNDL: /* /\B/l */
4198 RX_MATCH_TAINTED_on(reginfo->prog);
4200 case BOUND: /* /\b/ */
4201 case BOUNDU: /* /\b/u */
4202 case BOUNDA: /* /\b/a */
4203 case NBOUND: /* /\B/ */
4204 case NBOUNDU: /* /\B/u */
4205 case NBOUNDA: /* /\B/a */
4206 /* was last char in word? */
4208 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4209 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4211 if (locinput == PL_bostr)
4214 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
4216 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
4218 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4219 ln = isWORDCHAR_uni(ln);
4223 LOAD_UTF8_CHARCLASS_ALNUM();
4224 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4229 ln = isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(ln));
4230 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4235 /* Here the string isn't utf8, or is utf8 and only ascii
4236 * characters are to match \w. In the latter case looking at
4237 * the byte just prior to the current one may be just the final
4238 * byte of a multi-byte character. This is ok. There are two
4240 * 1) it is a single byte character, and then the test is doing
4241 * just what it's supposed to.
4242 * 2) it is a multi-byte character, in which case the final
4243 * byte is never mistakable for ASCII, and so the test
4244 * will say it is not a word character, which is the
4245 * correct answer. */
4246 ln = (locinput != PL_bostr) ?
4247 UCHARAT(locinput - 1) : '\n';
4248 switch (FLAGS(scan)) {
4249 case REGEX_UNICODE_CHARSET:
4250 ln = isWORDCHAR_L1(ln);
4251 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4253 case REGEX_LOCALE_CHARSET:
4254 ln = isWORDCHAR_LC(ln);
4255 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4257 case REGEX_DEPENDS_CHARSET:
4258 ln = isWORDCHAR(ln);
4259 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4261 case REGEX_ASCII_RESTRICTED_CHARSET:
4262 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4263 ln = isWORDCHAR_A(ln);
4264 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4267 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4271 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4273 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4277 case ANYOF: /* /[abc]/ */
4278 case ANYOF_WARN_SUPER:
4282 if (!reginclass(rex, scan, (U8*)locinput, utf8_target))
4284 locinput += UTF8SKIP(locinput);
4287 if (!REGINCLASS(rex, scan, (U8*)locinput))
4293 /* The argument (FLAGS) to all the POSIX node types is the class number
4296 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4300 case POSIXL: /* \w or [:punct:] etc. under /l */
4304 /* The locale hasn't influenced the outcome before this, so defer
4305 * tainting until now */
4306 RX_MATCH_TAINTED_on(reginfo->prog);
4308 /* Use isFOO_lc() for characters within Latin1. (Note that
4309 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4310 * wouldn't be invariant) */
4311 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4312 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4316 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4317 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4318 (U8) TWO_BYTE_UTF8_TO_UNI(nextchr,
4319 *(locinput + 1))))))
4324 else { /* Here, must be an above Latin-1 code point */
4325 goto utf8_posix_not_eos;
4328 /* Here, must be utf8 */
4329 locinput += UTF8SKIP(locinput);
4332 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4336 case POSIXD: /* \w or [:punct:] etc. under /d */
4342 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4344 if (NEXTCHR_IS_EOS) {
4348 /* All UTF-8 variants match */
4349 if (! UTF8_IS_INVARIANT(nextchr)) {
4350 goto increment_locinput;
4356 case POSIXA: /* \w or [:punct:] etc. under /a */
4359 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4360 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4361 * character is a single byte */
4364 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4370 /* Here we are either not in utf8, or we matched a utf8-invariant,
4371 * so the next char is the next byte */
4375 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4379 case POSIXU: /* \w or [:punct:] etc. under /u */
4381 if (NEXTCHR_IS_EOS) {
4386 /* Use _generic_isCC() for characters within Latin1. (Note that
4387 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4388 * wouldn't be invariant) */
4389 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4390 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4397 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4398 if (! (to_complement
4399 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(nextchr,
4407 else { /* Handle above Latin-1 code points */
4408 classnum = (_char_class_number) FLAGS(scan);
4409 if (classnum < _FIRST_NON_SWASH_CC) {
4411 /* Here, uses a swash to find such code points. Load if if
4412 * not done already */
4413 if (! PL_utf8_swash_ptrs[classnum]) {
4414 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4415 PL_utf8_swash_ptrs[classnum]
4416 = _core_swash_init("utf8",
4417 swash_property_names[classnum],
4418 &PL_sv_undef, 1, 0, NULL, &flags);
4420 if (! (to_complement
4421 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4422 (U8 *) locinput, TRUE))))
4427 else { /* Here, uses macros to find above Latin-1 code points */
4429 case _CC_ENUM_SPACE: /* XXX would require separate
4430 code if we revert the change
4431 of \v matching this */
4432 case _CC_ENUM_PSXSPC:
4433 if (! (to_complement
4434 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4439 case _CC_ENUM_BLANK:
4440 if (! (to_complement
4441 ^ cBOOL(is_HORIZWS_high(locinput))))
4446 case _CC_ENUM_XDIGIT:
4447 if (! (to_complement
4448 ^ cBOOL(is_XDIGIT_high(locinput))))
4453 case _CC_ENUM_VERTSPACE:
4454 if (! (to_complement
4455 ^ cBOOL(is_VERTWS_high(locinput))))
4460 default: /* The rest, e.g. [:cntrl:], can't match
4462 if (! to_complement) {
4468 locinput += UTF8SKIP(locinput);
4472 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4473 a Unicode extended Grapheme Cluster */
4474 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4475 extended Grapheme Cluster is:
4478 | Prepend* Begin Extend*
4481 Begin is: ( Special_Begin | ! Control )
4482 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4483 Extend is: ( Grapheme_Extend | Spacing_Mark )
4484 Control is: [ GCB_Control | CR | LF ]
4485 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4487 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4490 Begin is ( Regular_Begin + Special Begin )
4492 It turns out that 98.4% of all Unicode code points match
4493 Regular_Begin. Doing it this way eliminates a table match in
4494 the previous implementation for almost all Unicode code points.
4496 There is a subtlety with Prepend* which showed up in testing.
4497 Note that the Begin, and only the Begin is required in:
4498 | Prepend* Begin Extend*
4499 Also, Begin contains '! Control'. A Prepend must be a
4500 '! Control', which means it must also be a Begin. What it
4501 comes down to is that if we match Prepend* and then find no
4502 suitable Begin afterwards, that if we backtrack the last
4503 Prepend, that one will be a suitable Begin.
4508 if (! utf8_target) {
4510 /* Match either CR LF or '.', as all the other possibilities
4512 locinput++; /* Match the . or CR */
4513 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4515 && locinput < PL_regeol
4516 && UCHARAT(locinput) == '\n')
4523 /* Utf8: See if is ( CR LF ); already know that locinput <
4524 * PL_regeol, so locinput+1 is in bounds */
4525 if ( nextchr == '\r' && locinput+1 < PL_regeol
4526 && UCHARAT(locinput + 1) == '\n')
4533 /* In case have to backtrack to beginning, then match '.' */
4534 char *starting = locinput;
4536 /* In case have to backtrack the last prepend */
4537 char *previous_prepend = NULL;
4539 LOAD_UTF8_CHARCLASS_GCB();
4541 /* Match (prepend)* */
4542 while (locinput < PL_regeol
4543 && (len = is_GCB_Prepend_utf8(locinput)))
4545 previous_prepend = locinput;
4549 /* As noted above, if we matched a prepend character, but
4550 * the next thing won't match, back off the last prepend we
4551 * matched, as it is guaranteed to match the begin */
4552 if (previous_prepend
4553 && (locinput >= PL_regeol
4554 || (! swash_fetch(PL_utf8_X_regular_begin,
4555 (U8*)locinput, utf8_target)
4556 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4559 locinput = previous_prepend;
4562 /* Note that here we know PL_regeol > locinput, as we
4563 * tested that upon input to this switch case, and if we
4564 * moved locinput forward, we tested the result just above
4565 * and it either passed, or we backed off so that it will
4567 if (swash_fetch(PL_utf8_X_regular_begin,
4568 (U8*)locinput, utf8_target)) {
4569 locinput += UTF8SKIP(locinput);
4571 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4573 /* Here did not match the required 'Begin' in the
4574 * second term. So just match the very first
4575 * character, the '.' of the final term of the regex */
4576 locinput = starting + UTF8SKIP(starting);
4580 /* Here is a special begin. It can be composed of
4581 * several individual characters. One possibility is
4583 if ((len = is_GCB_RI_utf8(locinput))) {
4585 while (locinput < PL_regeol
4586 && (len = is_GCB_RI_utf8(locinput)))
4590 } else if ((len = is_GCB_T_utf8(locinput))) {
4591 /* Another possibility is T+ */
4593 while (locinput < PL_regeol
4594 && (len = is_GCB_T_utf8(locinput)))
4600 /* Here, neither RI+ nor T+; must be some other
4601 * Hangul. That means it is one of the others: L,
4602 * LV, LVT or V, and matches:
4603 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4606 while (locinput < PL_regeol
4607 && (len = is_GCB_L_utf8(locinput)))
4612 /* Here, have exhausted L*. If the next character
4613 * is not an LV, LVT nor V, it means we had to have
4614 * at least one L, so matches L+ in the original
4615 * equation, we have a complete hangul syllable.
4618 if (locinput < PL_regeol
4619 && is_GCB_LV_LVT_V_utf8(locinput))
4621 /* Otherwise keep going. Must be LV, LVT or V.
4622 * See if LVT, by first ruling out V, then LV */
4623 if (! is_GCB_V_utf8(locinput)
4624 /* All but every TCount one is LV */
4625 && (valid_utf8_to_uvchr((U8 *) locinput,
4630 locinput += UTF8SKIP(locinput);
4633 /* Must be V or LV. Take it, then match
4635 locinput += UTF8SKIP(locinput);
4636 while (locinput < PL_regeol
4637 && (len = is_GCB_V_utf8(locinput)))
4643 /* And any of LV, LVT, or V can be followed
4645 while (locinput < PL_regeol
4646 && (len = is_GCB_T_utf8(locinput)))
4654 /* Match any extender */
4655 while (locinput < PL_regeol
4656 && swash_fetch(PL_utf8_X_extend,
4657 (U8*)locinput, utf8_target))
4659 locinput += UTF8SKIP(locinput);
4663 if (locinput > PL_regeol) sayNO;
4667 case NREFFL: /* /\g{name}/il */
4668 { /* The capture buffer cases. The ones beginning with N for the
4669 named buffers just convert to the equivalent numbered and
4670 pretend they were called as the corresponding numbered buffer
4672 /* don't initialize these in the declaration, it makes C++
4677 const U8 *fold_array;
4680 RX_MATCH_TAINTED_on(reginfo->prog);
4681 folder = foldEQ_locale;
4682 fold_array = PL_fold_locale;
4684 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4687 case NREFFA: /* /\g{name}/iaa */
4688 folder = foldEQ_latin1;
4689 fold_array = PL_fold_latin1;
4691 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4694 case NREFFU: /* /\g{name}/iu */
4695 folder = foldEQ_latin1;
4696 fold_array = PL_fold_latin1;
4698 utf8_fold_flags = 0;
4701 case NREFF: /* /\g{name}/i */
4703 fold_array = PL_fold;
4705 utf8_fold_flags = 0;
4708 case NREF: /* /\g{name}/ */
4712 utf8_fold_flags = 0;
4715 /* For the named back references, find the corresponding buffer
4717 n = reg_check_named_buff_matched(rex,scan);
4722 goto do_nref_ref_common;
4724 case REFFL: /* /\1/il */
4725 RX_MATCH_TAINTED_on(reginfo->prog);
4726 folder = foldEQ_locale;
4727 fold_array = PL_fold_locale;
4728 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4731 case REFFA: /* /\1/iaa */
4732 folder = foldEQ_latin1;
4733 fold_array = PL_fold_latin1;
4734 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4737 case REFFU: /* /\1/iu */
4738 folder = foldEQ_latin1;
4739 fold_array = PL_fold_latin1;
4740 utf8_fold_flags = 0;
4743 case REFF: /* /\1/i */
4745 fold_array = PL_fold;
4746 utf8_fold_flags = 0;
4749 case REF: /* /\1/ */
4752 utf8_fold_flags = 0;
4756 n = ARG(scan); /* which paren pair */
4759 ln = rex->offs[n].start;
4760 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4761 if (rex->lastparen < n || ln == -1)
4762 sayNO; /* Do not match unless seen CLOSEn. */
4763 if (ln == rex->offs[n].end)
4767 if (type != REF /* REF can do byte comparison */
4768 && (utf8_target || type == REFFU))
4769 { /* XXX handle REFFL better */
4770 char * limit = PL_regeol;
4772 /* This call case insensitively compares the entire buffer
4773 * at s, with the current input starting at locinput, but
4774 * not going off the end given by PL_regeol, and returns in
4775 * <limit> upon success, how much of the current input was
4777 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4778 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4786 /* Not utf8: Inline the first character, for speed. */
4787 if (!NEXTCHR_IS_EOS &&
4788 UCHARAT(s) != nextchr &&
4790 UCHARAT(s) != fold_array[nextchr]))
4792 ln = rex->offs[n].end - ln;
4793 if (locinput + ln > PL_regeol)
4795 if (ln > 1 && (type == REF
4796 ? memNE(s, locinput, ln)
4797 : ! folder(s, locinput, ln)))
4803 case NOTHING: /* null op; e.g. the 'nothing' following
4804 * the '*' in m{(a+|b)*}' */
4806 case TAIL: /* placeholder while compiling (A|B|C) */
4809 case BACK: /* ??? doesn't appear to be used ??? */
4813 #define ST st->u.eval
4818 regexp_internal *rei;
4819 regnode *startpoint;
4821 case GOSTART: /* (?R) */
4822 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4823 if (cur_eval && cur_eval->locinput==locinput) {
4824 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4825 Perl_croak(aTHX_ "Infinite recursion in regex");
4826 if ( ++nochange_depth > max_nochange_depth )
4828 "Pattern subroutine nesting without pos change"
4829 " exceeded limit in regex");
4836 if (OP(scan)==GOSUB) {
4837 startpoint = scan + ARG2L(scan);
4838 ST.close_paren = ARG(scan);
4840 startpoint = rei->program+1;
4843 goto eval_recurse_doit;
4844 assert(0); /* NOTREACHED */
4846 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4847 if (cur_eval && cur_eval->locinput==locinput) {
4848 if ( ++nochange_depth > max_nochange_depth )
4849 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4854 /* execute the code in the {...} */
4858 OP * const oop = PL_op;
4859 COP * const ocurcop = PL_curcop;
4861 char *saved_regeol = PL_regeol;
4862 struct re_save_state saved_state;
4865 /* save *all* paren positions */
4866 regcppush(rex, 0, maxopenparen);
4867 REGCP_SET(runops_cp);
4869 /* To not corrupt the existing regex state while executing the
4870 * eval we would normally put it on the save stack, like with
4871 * save_re_context. However, re-evals have a weird scoping so we
4872 * can't just add ENTER/LEAVE here. With that, things like
4874 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4876 * would break, as they expect the localisation to be unwound
4877 * only when the re-engine backtracks through the bit that
4880 * What we do instead is just saving the state in a local c
4883 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4885 PL_reg_state.re_reparsing = FALSE;
4888 caller_cv = find_runcv(NULL);
4892 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4894 (REGEXP*)(rexi->data->data[n])
4897 nop = (OP*)rexi->data->data[n+1];
4899 else if (rexi->data->what[n] == 'l') { /* literal code */
4901 nop = (OP*)rexi->data->data[n];
4902 assert(CvDEPTH(newcv));
4905 /* literal with own CV */
4906 assert(rexi->data->what[n] == 'L');
4907 newcv = rex->qr_anoncv;
4908 nop = (OP*)rexi->data->data[n];
4911 /* normally if we're about to execute code from the same
4912 * CV that we used previously, we just use the existing
4913 * CX stack entry. However, its possible that in the
4914 * meantime we may have backtracked, popped from the save
4915 * stack, and undone the SAVECOMPPAD(s) associated with
4916 * PUSH_MULTICALL; in which case PL_comppad no longer
4917 * points to newcv's pad. */
4918 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4920 I32 depth = (newcv == caller_cv) ? 0 : 1;
4921 if (last_pushed_cv) {
4922 CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
4925 PUSH_MULTICALL_WITHDEPTH(newcv, depth);
4927 last_pushed_cv = newcv;
4930 /* these assignments are just to silence compiler
4932 multicall_cop = NULL;
4935 last_pad = PL_comppad;
4937 /* the initial nextstate you would normally execute
4938 * at the start of an eval (which would cause error
4939 * messages to come from the eval), may be optimised
4940 * away from the execution path in the regex code blocks;
4941 * so manually set PL_curcop to it initially */
4943 OP *o = cUNOPx(nop)->op_first;
4944 assert(o->op_type == OP_NULL);
4945 if (o->op_targ == OP_SCOPE) {
4946 o = cUNOPo->op_first;
4949 assert(o->op_targ == OP_LEAVE);
4950 o = cUNOPo->op_first;
4951 assert(o->op_type == OP_ENTER);
4955 if (o->op_type != OP_STUB) {
4956 assert( o->op_type == OP_NEXTSTATE
4957 || o->op_type == OP_DBSTATE
4958 || (o->op_type == OP_NULL
4959 && ( o->op_targ == OP_NEXTSTATE
4960 || o->op_targ == OP_DBSTATE
4964 PL_curcop = (COP*)o;
4969 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4970 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4972 rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4975 SV *sv_mrk = get_sv("REGMARK", 1);
4976 sv_setsv(sv_mrk, sv_yes_mark);
4979 /* we don't use MULTICALL here as we want to call the
4980 * first op of the block of interest, rather than the
4981 * first op of the sub */
4982 before = (IV)(SP-PL_stack_base);
4984 CALLRUNOPS(aTHX); /* Scalar context. */
4986 if ((IV)(SP-PL_stack_base) == before)
4987 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4993 /* before restoring everything, evaluate the returned
4994 * value, so that 'uninit' warnings don't use the wrong
4995 * PL_op or pad. Also need to process any magic vars
4996 * (e.g. $1) *before* parentheses are restored */
5001 if (logical == 0) /* (?{})/ */
5002 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5003 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5004 sw = cBOOL(SvTRUE(ret));
5007 else { /* /(??{}) */
5008 /* if its overloaded, let the regex compiler handle
5009 * it; otherwise extract regex, or stringify */
5010 if (!SvAMAGIC(ret)) {
5014 if (SvTYPE(sv) == SVt_REGEXP)
5015 re_sv = (REGEXP*) sv;
5016 else if (SvSMAGICAL(sv)) {
5017 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
5019 re_sv = (REGEXP *) mg->mg_obj;
5022 /* force any magic, undef warnings here */
5024 ret = sv_mortalcopy(ret);
5025 (void) SvPV_force_nolen(ret);
5031 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
5033 /* *** Note that at this point we don't restore
5034 * PL_comppad, (or pop the CxSUB) on the assumption it may
5035 * be used again soon. This is safe as long as nothing
5036 * in the regexp code uses the pad ! */
5038 PL_curcop = ocurcop;
5039 PL_regeol = saved_regeol;
5040 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5046 /* only /(??{})/ from now on */
5049 /* extract RE object from returned value; compiling if
5053 re_sv = reg_temp_copy(NULL, re_sv);
5058 if (SvUTF8(ret) && IN_BYTES) {
5059 /* In use 'bytes': make a copy of the octet
5060 * sequence, but without the flag on */
5062 const char *const p = SvPV(ret, len);
5063 ret = newSVpvn_flags(p, len, SVs_TEMP);
5065 if (rex->intflags & PREGf_USE_RE_EVAL)
5066 pm_flags |= PMf_USE_RE_EVAL;
5068 /* if we got here, it should be an engine which
5069 * supports compiling code blocks and stuff */
5070 assert(rex->engine && rex->engine->op_comp);
5071 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5072 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5073 rex->engine, NULL, NULL,
5074 /* copy /msix etc to inner pattern */
5079 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
5081 /* This isn't a first class regexp. Instead, it's
5082 caching a regexp onto an existing, Perl visible
5084 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5086 /* safe to do now that any $1 etc has been
5087 * interpolated into the new pattern string and
5089 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5094 RXp_MATCH_COPIED_off(re);
5095 re->subbeg = rex->subbeg;
5096 re->sublen = rex->sublen;
5097 re->suboffset = rex->suboffset;
5098 re->subcoffset = rex->subcoffset;
5101 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
5102 "Matching embedded");
5104 startpoint = rei->program + 1;
5105 ST.close_paren = 0; /* only used for GOSUB */
5107 eval_recurse_doit: /* Share code with GOSUB below this line */
5108 /* run the pattern returned from (??{...}) */
5110 /* Save *all* the positions. */
5111 ST.cp = regcppush(rex, 0, maxopenparen);
5112 REGCP_SET(ST.lastcp);
5115 re->lastcloseparen = 0;
5119 /* XXXX This is too dramatic a measure... */
5122 ST.saved_utf8_pat = is_utf8_pat;
5123 is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5125 ST.prev_rex = rex_sv;
5126 ST.prev_curlyx = cur_curlyx;
5128 SET_reg_curpm(rex_sv);
5133 ST.prev_eval = cur_eval;
5135 /* now continue from first node in postoned RE */
5136 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5137 assert(0); /* NOTREACHED */
5140 case EVAL_AB: /* cleanup after a successful (??{A})B */
5141 /* note: this is called twice; first after popping B, then A */
5142 is_utf8_pat = ST.saved_utf8_pat;
5143 rex_sv = ST.prev_rex;
5144 SET_reg_curpm(rex_sv);
5145 rex = ReANY(rex_sv);
5146 rexi = RXi_GET(rex);
5148 cur_eval = ST.prev_eval;
5149 cur_curlyx = ST.prev_curlyx;
5151 /* XXXX This is too dramatic a measure... */
5153 if ( nochange_depth )
5158 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5159 /* note: this is called twice; first after popping B, then A */
5160 is_utf8_pat = ST.saved_utf8_pat;
5161 rex_sv = ST.prev_rex;
5162 SET_reg_curpm(rex_sv);
5163 rex = ReANY(rex_sv);
5164 rexi = RXi_GET(rex);
5166 REGCP_UNWIND(ST.lastcp);
5167 regcppop(rex, &maxopenparen);
5168 cur_eval = ST.prev_eval;
5169 cur_curlyx = ST.prev_curlyx;
5170 /* XXXX This is too dramatic a measure... */
5172 if ( nochange_depth )
5178 n = ARG(scan); /* which paren pair */
5179 rex->offs[n].start_tmp = locinput - PL_bostr;
5180 if (n > maxopenparen)
5182 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5183 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5187 (IV)rex->offs[n].start_tmp,
5193 /* XXX really need to log other places start/end are set too */
5194 #define CLOSE_CAPTURE \
5195 rex->offs[n].start = rex->offs[n].start_tmp; \
5196 rex->offs[n].end = locinput - PL_bostr; \
5197 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5198 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5200 PTR2UV(rex->offs), \
5202 (IV)rex->offs[n].start, \
5203 (IV)rex->offs[n].end \
5207 n = ARG(scan); /* which paren pair */
5209 if (n > rex->lastparen)
5211 rex->lastcloseparen = n;
5212 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5217 case ACCEPT: /* (*ACCEPT) */
5221 cursor && OP(cursor)!=END;
5222 cursor=regnext(cursor))
5224 if ( OP(cursor)==CLOSE ){
5226 if ( n <= lastopen ) {
5228 if (n > rex->lastparen)
5230 rex->lastcloseparen = n;
5231 if ( n == ARG(scan) || (cur_eval &&
5232 cur_eval->u.eval.close_paren == n))
5241 case GROUPP: /* (?(1)) */
5242 n = ARG(scan); /* which paren pair */
5243 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5246 case NGROUPP: /* (?(<name>)) */
5247 /* reg_check_named_buff_matched returns 0 for no match */
5248 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5251 case INSUBP: /* (?(R)) */
5253 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5256 case DEFINEP: /* (?(DEFINE)) */
5260 case IFTHEN: /* (?(cond)A|B) */
5261 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
5263 next = NEXTOPER(NEXTOPER(scan));
5265 next = scan + ARG(scan);
5266 if (OP(next) == IFTHEN) /* Fake one. */
5267 next = NEXTOPER(NEXTOPER(next));
5271 case LOGICAL: /* modifier for EVAL and IFMATCH */
5272 logical = scan->flags;
5275 /*******************************************************************
5277 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5278 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5279 STAR/PLUS/CURLY/CURLYN are used instead.)
5281 A*B is compiled as <CURLYX><A><WHILEM><B>
5283 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5284 state, which contains the current count, initialised to -1. It also sets
5285 cur_curlyx to point to this state, with any previous value saved in the
5288 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5289 since the pattern may possibly match zero times (i.e. it's a while {} loop
5290 rather than a do {} while loop).
5292 Each entry to WHILEM represents a successful match of A. The count in the
5293 CURLYX block is incremented, another WHILEM state is pushed, and execution
5294 passes to A or B depending on greediness and the current count.
5296 For example, if matching against the string a1a2a3b (where the aN are
5297 substrings that match /A/), then the match progresses as follows: (the
5298 pushed states are interspersed with the bits of strings matched so far):
5301 <CURLYX cnt=0><WHILEM>
5302 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5303 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5304 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5305 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5307 (Contrast this with something like CURLYM, which maintains only a single
5311 a1 <CURLYM cnt=1> a2
5312 a1 a2 <CURLYM cnt=2> a3
5313 a1 a2 a3 <CURLYM cnt=3> b
5316 Each WHILEM state block marks a point to backtrack to upon partial failure
5317 of A or B, and also contains some minor state data related to that
5318 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5319 overall state, such as the count, and pointers to the A and B ops.
5321 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5322 must always point to the *current* CURLYX block, the rules are:
5324 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5325 and set cur_curlyx to point the new block.
5327 When popping the CURLYX block after a successful or unsuccessful match,
5328 restore the previous cur_curlyx.
5330 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5331 to the outer one saved in the CURLYX block.
5333 When popping the WHILEM block after a successful or unsuccessful B match,
5334 restore the previous cur_curlyx.
5336 Here's an example for the pattern (AI* BI)*BO
5337 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5340 curlyx backtrack stack
5341 ------ ---------------
5343 CO <CO prev=NULL> <WO>
5344 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5345 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5346 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5348 At this point the pattern succeeds, and we work back down the stack to
5349 clean up, restoring as we go:
5351 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5352 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5353 CO <CO prev=NULL> <WO>
5356 *******************************************************************/
5358 #define ST st->u.curlyx
5360 case CURLYX: /* start of /A*B/ (for complex A) */
5362 /* No need to save/restore up to this paren */
5363 I32 parenfloor = scan->flags;
5365 assert(next); /* keep Coverity happy */
5366 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5369 /* XXXX Probably it is better to teach regpush to support
5370 parenfloor > maxopenparen ... */
5371 if (parenfloor > (I32)rex->lastparen)
5372 parenfloor = rex->lastparen; /* Pessimization... */
5374 ST.prev_curlyx= cur_curlyx;
5376 ST.cp = PL_savestack_ix;
5378 /* these fields contain the state of the current curly.
5379 * they are accessed by subsequent WHILEMs */
5380 ST.parenfloor = parenfloor;
5385 ST.count = -1; /* this will be updated by WHILEM */
5386 ST.lastloc = NULL; /* this will be updated by WHILEM */
5388 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5389 assert(0); /* NOTREACHED */
5392 case CURLYX_end: /* just finished matching all of A*B */
5393 cur_curlyx = ST.prev_curlyx;
5395 assert(0); /* NOTREACHED */
5397 case CURLYX_end_fail: /* just failed to match all of A*B */
5399 cur_curlyx = ST.prev_curlyx;
5401 assert(0); /* NOTREACHED */
5405 #define ST st->u.whilem
5407 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5409 /* see the discussion above about CURLYX/WHILEM */
5411 int min = ARG1(cur_curlyx->u.curlyx.me);
5412 int max = ARG2(cur_curlyx->u.curlyx.me);
5413 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5415 assert(cur_curlyx); /* keep Coverity happy */
5416 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5417 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5418 ST.cache_offset = 0;
5422 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5423 "%*s whilem: matched %ld out of %d..%d\n",
5424 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5427 /* First just match a string of min A's. */
5430 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5432 cur_curlyx->u.curlyx.lastloc = locinput;
5433 REGCP_SET(ST.lastcp);
5435 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5436 assert(0); /* NOTREACHED */
5439 /* If degenerate A matches "", assume A done. */
5441 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5442 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5443 "%*s whilem: empty match detected, trying continuation...\n",
5444 REPORT_CODE_OFF+depth*2, "")
5446 goto do_whilem_B_max;
5449 /* super-linear cache processing */
5453 if (!PL_reg_maxiter) {
5454 /* start the countdown: Postpone detection until we
5455 * know the match is not *that* much linear. */
5456 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
5457 /* possible overflow for long strings and many CURLYX's */
5458 if (PL_reg_maxiter < 0)
5459 PL_reg_maxiter = I32_MAX;
5460 PL_reg_leftiter = PL_reg_maxiter;
5463 if (PL_reg_leftiter-- == 0) {
5464 /* initialise cache */
5465 const I32 size = (PL_reg_maxiter + 7)/8;
5466 if (PL_reg_poscache) {
5467 if ((I32)PL_reg_poscache_size < size) {
5468 Renew(PL_reg_poscache, size, char);
5469 PL_reg_poscache_size = size;
5471 Zero(PL_reg_poscache, size, char);
5474 PL_reg_poscache_size = size;
5475 Newxz(PL_reg_poscache, size, char);
5477 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5478 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5479 PL_colors[4], PL_colors[5])
5483 if (PL_reg_leftiter < 0) {
5484 /* have we already failed at this position? */
5486 offset = (scan->flags & 0xf) - 1
5487 + (locinput - PL_bostr) * (scan->flags>>4);
5488 mask = 1 << (offset % 8);
5490 if (PL_reg_poscache[offset] & mask) {
5491 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5492 "%*s whilem: (cache) already tried at this position...\n",
5493 REPORT_CODE_OFF+depth*2, "")
5495 sayNO; /* cache records failure */
5497 ST.cache_offset = offset;
5498 ST.cache_mask = mask;
5502 /* Prefer B over A for minimal matching. */
5504 if (cur_curlyx->u.curlyx.minmod) {
5505 ST.save_curlyx = cur_curlyx;
5506 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5507 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5509 REGCP_SET(ST.lastcp);
5510 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5512 assert(0); /* NOTREACHED */
5515 /* Prefer A over B for maximal matching. */
5517 if (n < max) { /* More greed allowed? */
5518 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5520 cur_curlyx->u.curlyx.lastloc = locinput;
5521 REGCP_SET(ST.lastcp);
5522 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5523 assert(0); /* NOTREACHED */
5525 goto do_whilem_B_max;
5527 assert(0); /* NOTREACHED */
5529 case WHILEM_B_min: /* just matched B in a minimal match */
5530 case WHILEM_B_max: /* just matched B in a maximal match */
5531 cur_curlyx = ST.save_curlyx;
5533 assert(0); /* NOTREACHED */
5535 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5536 cur_curlyx = ST.save_curlyx;
5537 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5538 cur_curlyx->u.curlyx.count--;
5540 assert(0); /* NOTREACHED */
5542 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5544 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5545 REGCP_UNWIND(ST.lastcp);
5546 regcppop(rex, &maxopenparen);
5547 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5548 cur_curlyx->u.curlyx.count--;
5550 assert(0); /* NOTREACHED */
5552 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5553 REGCP_UNWIND(ST.lastcp);
5554 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5555 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5556 "%*s whilem: failed, trying continuation...\n",
5557 REPORT_CODE_OFF+depth*2, "")
5560 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5561 && ckWARN(WARN_REGEXP)
5562 && !reginfo->warned)
5564 reginfo->warned = TRUE;
5565 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5566 "Complex regular subexpression recursion limit (%d) "
5572 ST.save_curlyx = cur_curlyx;
5573 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5574 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5576 assert(0); /* NOTREACHED */
5578 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5579 cur_curlyx = ST.save_curlyx;
5580 REGCP_UNWIND(ST.lastcp);
5581 regcppop(rex, &maxopenparen);
5583 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5584 /* Maximum greed exceeded */
5585 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5586 && ckWARN(WARN_REGEXP)
5587 && !reginfo->warned)
5589 reginfo->warned = TRUE;
5590 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5591 "Complex regular subexpression recursion "
5592 "limit (%d) exceeded",
5595 cur_curlyx->u.curlyx.count--;
5599 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5600 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5602 /* Try grabbing another A and see if it helps. */
5603 cur_curlyx->u.curlyx.lastloc = locinput;
5604 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5606 REGCP_SET(ST.lastcp);
5607 PUSH_STATE_GOTO(WHILEM_A_min,
5608 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5610 assert(0); /* NOTREACHED */
5613 #define ST st->u.branch
5615 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5616 next = scan + ARG(scan);
5619 scan = NEXTOPER(scan);
5622 case BRANCH: /* /(...|A|...)/ */
5623 scan = NEXTOPER(scan); /* scan now points to inner node */
5624 ST.lastparen = rex->lastparen;
5625 ST.lastcloseparen = rex->lastcloseparen;
5626 ST.next_branch = next;
5629 /* Now go into the branch */
5631 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5633 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5635 assert(0); /* NOTREACHED */
5637 case CUTGROUP: /* /(*THEN)/ */
5638 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5639 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5640 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5641 assert(0); /* NOTREACHED */
5643 case CUTGROUP_next_fail:
5646 if (st->u.mark.mark_name)
5647 sv_commit = st->u.mark.mark_name;
5649 assert(0); /* NOTREACHED */
5653 assert(0); /* NOTREACHED */
5655 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5660 REGCP_UNWIND(ST.cp);
5661 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5662 scan = ST.next_branch;
5663 /* no more branches? */
5664 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5666 PerlIO_printf( Perl_debug_log,
5667 "%*s %sBRANCH failed...%s\n",
5668 REPORT_CODE_OFF+depth*2, "",
5674 continue; /* execute next BRANCH[J] op */
5675 assert(0); /* NOTREACHED */
5677 case MINMOD: /* next op will be non-greedy, e.g. A*? */
5682 #define ST st->u.curlym
5684 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5686 /* This is an optimisation of CURLYX that enables us to push
5687 * only a single backtracking state, no matter how many matches
5688 * there are in {m,n}. It relies on the pattern being constant
5689 * length, with no parens to influence future backrefs
5693 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5695 ST.lastparen = rex->lastparen;
5696 ST.lastcloseparen = rex->lastcloseparen;
5698 /* if paren positive, emulate an OPEN/CLOSE around A */
5700 U32 paren = ST.me->flags;
5701 if (paren > maxopenparen)
5702 maxopenparen = paren;
5703 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5711 ST.c1 = CHRTEST_UNINIT;
5714 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5717 curlym_do_A: /* execute the A in /A{m,n}B/ */
5718 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5719 assert(0); /* NOTREACHED */
5721 case CURLYM_A: /* we've just matched an A */
5723 /* after first match, determine A's length: u.curlym.alen */
5724 if (ST.count == 1) {
5725 if (PL_reg_match_utf8) {
5726 char *s = st->locinput;
5727 while (s < locinput) {
5733 ST.alen = locinput - st->locinput;
5736 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5739 PerlIO_printf(Perl_debug_log,
5740 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5741 (int)(REPORT_CODE_OFF+(depth*2)), "",
5742 (IV) ST.count, (IV)ST.alen)
5745 if (cur_eval && cur_eval->u.eval.close_paren &&
5746 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5750 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5751 if ( max == REG_INFTY || ST.count < max )
5752 goto curlym_do_A; /* try to match another A */
5754 goto curlym_do_B; /* try to match B */
5756 case CURLYM_A_fail: /* just failed to match an A */
5757 REGCP_UNWIND(ST.cp);
5759 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5760 || (cur_eval && cur_eval->u.eval.close_paren &&
5761 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5764 curlym_do_B: /* execute the B in /A{m,n}B/ */
5765 if (ST.c1 == CHRTEST_UNINIT) {
5766 /* calculate c1 and c2 for possible match of 1st char
5767 * following curly */
5768 ST.c1 = ST.c2 = CHRTEST_VOID;
5769 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5770 regnode *text_node = ST.B;
5771 if (! HAS_TEXT(text_node))
5772 FIND_NEXT_IMPT(text_node);
5775 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5777 But the former is redundant in light of the latter.
5779 if this changes back then the macro for
5780 IS_TEXT and friends need to change.
5782 if (PL_regkind[OP(text_node)] == EXACT) {
5783 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5784 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5794 PerlIO_printf(Perl_debug_log,
5795 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5796 (int)(REPORT_CODE_OFF+(depth*2)),
5799 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
5800 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
5801 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
5802 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
5804 /* simulate B failing */
5806 PerlIO_printf(Perl_debug_log,
5807 "%*s CURLYM Fast bail next target=U+%"UVXf" c1=U+%"UVXf" c2=U+%"UVXf"\n",
5808 (int)(REPORT_CODE_OFF+(depth*2)),"",
5809 valid_utf8_to_uvchr((U8 *) locinput, NULL),
5810 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
5811 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
5813 state_num = CURLYM_B_fail;
5814 goto reenter_switch;
5817 else if (nextchr != ST.c1 && nextchr != ST.c2) {
5818 /* simulate B failing */
5820 PerlIO_printf(Perl_debug_log,
5821 "%*s CURLYM Fast bail next target=U+%X c1=U+%X c2=U+%X\n",
5822 (int)(REPORT_CODE_OFF+(depth*2)),"",
5823 (int) nextchr, ST.c1, ST.c2)
5825 state_num = CURLYM_B_fail;
5826 goto reenter_switch;
5831 /* emulate CLOSE: mark current A as captured */
5832 I32 paren = ST.me->flags;
5834 rex->offs[paren].start
5835 = HOPc(locinput, -ST.alen) - PL_bostr;
5836 rex->offs[paren].end = locinput - PL_bostr;
5837 if ((U32)paren > rex->lastparen)
5838 rex->lastparen = paren;
5839 rex->lastcloseparen = paren;
5842 rex->offs[paren].end = -1;
5843 if (cur_eval && cur_eval->u.eval.close_paren &&
5844 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5853 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5854 assert(0); /* NOTREACHED */
5856 case CURLYM_B_fail: /* just failed to match a B */
5857 REGCP_UNWIND(ST.cp);
5858 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5860 I32 max = ARG2(ST.me);
5861 if (max != REG_INFTY && ST.count == max)
5863 goto curlym_do_A; /* try to match a further A */
5865 /* backtrack one A */
5866 if (ST.count == ARG1(ST.me) /* min */)
5869 SET_locinput(HOPc(locinput, -ST.alen));
5870 goto curlym_do_B; /* try to match B */
5873 #define ST st->u.curly
5875 #define CURLY_SETPAREN(paren, success) \
5878 rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5879 rex->offs[paren].end = locinput - PL_bostr; \
5880 if (paren > rex->lastparen) \
5881 rex->lastparen = paren; \
5882 rex->lastcloseparen = paren; \
5885 rex->offs[paren].end = -1; \
5886 rex->lastparen = ST.lastparen; \
5887 rex->lastcloseparen = ST.lastcloseparen; \
5891 case STAR: /* /A*B/ where A is width 1 char */
5895 scan = NEXTOPER(scan);
5898 case PLUS: /* /A+B/ where A is width 1 char */
5902 scan = NEXTOPER(scan);
5905 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
5906 ST.paren = scan->flags; /* Which paren to set */
5907 ST.lastparen = rex->lastparen;
5908 ST.lastcloseparen = rex->lastcloseparen;
5909 if (ST.paren > maxopenparen)
5910 maxopenparen = ST.paren;
5911 ST.min = ARG1(scan); /* min to match */
5912 ST.max = ARG2(scan); /* max to match */
5913 if (cur_eval && cur_eval->u.eval.close_paren &&
5914 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5918 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5921 case CURLY: /* /A{m,n}B/ where A is width 1 char */
5923 ST.min = ARG1(scan); /* min to match */
5924 ST.max = ARG2(scan); /* max to match */
5925 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5928 * Lookahead to avoid useless match attempts
5929 * when we know what character comes next.
5931 * Used to only do .*x and .*?x, but now it allows
5932 * for )'s, ('s and (?{ ... })'s to be in the way
5933 * of the quantifier and the EXACT-like node. -- japhy
5936 assert(ST.min <= ST.max);
5937 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
5938 ST.c1 = ST.c2 = CHRTEST_VOID;
5941 regnode *text_node = next;
5943 if (! HAS_TEXT(text_node))
5944 FIND_NEXT_IMPT(text_node);
5946 if (! HAS_TEXT(text_node))
5947 ST.c1 = ST.c2 = CHRTEST_VOID;
5949 if ( PL_regkind[OP(text_node)] != EXACT ) {
5950 ST.c1 = ST.c2 = CHRTEST_VOID;
5954 /* Currently we only get here when
5956 PL_rekind[OP(text_node)] == EXACT
5958 if this changes back then the macro for IS_TEXT and
5959 friends need to change. */
5960 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5961 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5973 char *li = locinput;
5976 regrepeat(rex, &li, ST.A, ST.min, depth, is_utf8_pat)
5982 if (ST.c1 == CHRTEST_VOID)
5983 goto curly_try_B_min;
5985 ST.oldloc = locinput;
5987 /* set ST.maxpos to the furthest point along the
5988 * string that could possibly match */
5989 if (ST.max == REG_INFTY) {
5990 ST.maxpos = PL_regeol - 1;
5992 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5995 else if (utf8_target) {
5996 int m = ST.max - ST.min;
5997 for (ST.maxpos = locinput;
5998 m >0 && ST.maxpos < PL_regeol; m--)
5999 ST.maxpos += UTF8SKIP(ST.maxpos);
6002 ST.maxpos = locinput + ST.max - ST.min;
6003 if (ST.maxpos >= PL_regeol)
6004 ST.maxpos = PL_regeol - 1;
6006 goto curly_try_B_min_known;
6010 /* avoid taking address of locinput, so it can remain
6012 char *li = locinput;
6013 ST.count = regrepeat(rex, &li, ST.A, ST.max, depth,
6015 if (ST.count < ST.min)
6018 if ((ST.count > ST.min)
6019 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6021 /* A{m,n} must come at the end of the string, there's
6022 * no point in backing off ... */
6024 /* ...except that $ and \Z can match before *and* after
6025 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6026 We may back off by one in this case. */
6027 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6031 goto curly_try_B_max;
6033 assert(0); /* NOTREACHED */
6036 case CURLY_B_min_known_fail:
6037 /* failed to find B in a non-greedy match where c1,c2 valid */
6039 REGCP_UNWIND(ST.cp);
6041 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6043 /* Couldn't or didn't -- move forward. */
6044 ST.oldloc = locinput;
6046 locinput += UTF8SKIP(locinput);
6050 curly_try_B_min_known:
6051 /* find the next place where 'B' could work, then call B */
6055 n = (ST.oldloc == locinput) ? 0 : 1;
6056 if (ST.c1 == ST.c2) {
6057 /* set n to utf8_distance(oldloc, locinput) */
6058 while (locinput <= ST.maxpos
6059 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6061 locinput += UTF8SKIP(locinput);
6066 /* set n to utf8_distance(oldloc, locinput) */
6067 while (locinput <= ST.maxpos
6068 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6069 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6071 locinput += UTF8SKIP(locinput);
6076 else { /* Not utf8_target */
6077 if (ST.c1 == ST.c2) {
6078 while (locinput <= ST.maxpos &&
6079 UCHARAT(locinput) != ST.c1)
6083 while (locinput <= ST.maxpos
6084 && UCHARAT(locinput) != ST.c1
6085 && UCHARAT(locinput) != ST.c2)
6088 n = locinput - ST.oldloc;
6090 if (locinput > ST.maxpos)
6093 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6094 * at b; check that everything between oldloc and
6095 * locinput matches */
6096 char *li = ST.oldloc;
6098 if (regrepeat(rex, &li, ST.A, n, depth, is_utf8_pat) < n)
6100 assert(n == REG_INFTY || locinput == li);
6102 CURLY_SETPAREN(ST.paren, ST.count);
6103 if (cur_eval && cur_eval->u.eval.close_paren &&
6104 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6107 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6109 assert(0); /* NOTREACHED */
6112 case CURLY_B_min_fail:
6113 /* failed to find B in a non-greedy match where c1,c2 invalid */
6115 REGCP_UNWIND(ST.cp);
6117 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6119 /* failed -- move forward one */
6121 char *li = locinput;
6122 if (!regrepeat(rex, &li, ST.A, 1, depth, is_utf8_pat)) {
6129 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6130 ST.count > 0)) /* count overflow ? */
6133 CURLY_SETPAREN(ST.paren, ST.count);
6134 if (cur_eval && cur_eval->u.eval.close_paren &&
6135 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6138 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6142 assert(0); /* NOTREACHED */
6146 /* a successful greedy match: now try to match B */
6147 if (cur_eval && cur_eval->u.eval.close_paren &&
6148 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6152 bool could_match = locinput < PL_regeol;
6154 /* If it could work, try it. */
6155 if (ST.c1 != CHRTEST_VOID && could_match) {
6156 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6158 could_match = memEQ(locinput,
6163 UTF8SKIP(locinput));
6166 could_match = UCHARAT(locinput) == ST.c1
6167 || UCHARAT(locinput) == ST.c2;
6170 if (ST.c1 == CHRTEST_VOID || could_match) {
6171 CURLY_SETPAREN(ST.paren, ST.count);
6172 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6173 assert(0); /* NOTREACHED */
6178 case CURLY_B_max_fail:
6179 /* failed to find B in a greedy match */
6181 REGCP_UNWIND(ST.cp);
6183 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6186 if (--ST.count < ST.min)
6188 locinput = HOPc(locinput, -1);
6189 goto curly_try_B_max;
6193 case END: /* last op of main pattern */
6196 /* we've just finished A in /(??{A})B/; now continue with B */
6197 st->u.eval.saved_utf8_pat = is_utf8_pat;
6198 is_utf8_pat = cur_eval->u.eval.saved_utf8_pat;
6200 st->u.eval.prev_rex = rex_sv; /* inner */
6202 /* Save *all* the positions. */
6203 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6204 rex_sv = cur_eval->u.eval.prev_rex;
6205 SET_reg_curpm(rex_sv);
6206 rex = ReANY(rex_sv);
6207 rexi = RXi_GET(rex);
6208 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6210 REGCP_SET(st->u.eval.lastcp);
6212 /* Restore parens of the outer rex without popping the
6214 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6217 st->u.eval.prev_eval = cur_eval;
6218 cur_eval = cur_eval->u.eval.prev_eval;
6220 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6221 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6222 if ( nochange_depth )
6225 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6226 locinput); /* match B */
6229 if (locinput < reginfo->till) {
6230 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6231 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6233 (long)(locinput - PL_reg_starttry),
6234 (long)(reginfo->till - PL_reg_starttry),
6237 sayNO_SILENT; /* Cannot match: too short. */
6239 sayYES; /* Success! */
6241 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6243 PerlIO_printf(Perl_debug_log,
6244 "%*s %ssubpattern success...%s\n",
6245 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6246 sayYES; /* Success! */
6249 #define ST st->u.ifmatch
6254 case SUSPEND: /* (?>A) */
6256 newstart = locinput;
6259 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6261 goto ifmatch_trivial_fail_test;
6263 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6265 ifmatch_trivial_fail_test:
6267 char * const s = HOPBACKc(locinput, scan->flags);
6272 sw = 1 - cBOOL(ST.wanted);
6276 next = scan + ARG(scan);
6284 newstart = locinput;
6288 ST.logical = logical;
6289 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6291 /* execute body of (?...A) */
6292 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6293 assert(0); /* NOTREACHED */
6296 case IFMATCH_A_fail: /* body of (?...A) failed */
6297 ST.wanted = !ST.wanted;
6300 case IFMATCH_A: /* body of (?...A) succeeded */
6302 sw = cBOOL(ST.wanted);
6304 else if (!ST.wanted)
6307 if (OP(ST.me) != SUSPEND) {
6308 /* restore old position except for (?>...) */
6309 locinput = st->locinput;
6311 scan = ST.me + ARG(ST.me);
6314 continue; /* execute B */
6318 case LONGJMP: /* alternative with many branches compiles to
6319 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6320 next = scan + ARG(scan);
6325 case COMMIT: /* (*COMMIT) */
6326 reginfo->cutpoint = PL_regeol;
6329 case PRUNE: /* (*PRUNE) */
6331 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6332 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6333 assert(0); /* NOTREACHED */
6335 case COMMIT_next_fail:
6339 case OPFAIL: /* (*FAIL) */
6341 assert(0); /* NOTREACHED */
6343 #define ST st->u.mark
6344 case MARKPOINT: /* (*MARK:foo) */
6345 ST.prev_mark = mark_state;
6346 ST.mark_name = sv_commit = sv_yes_mark
6347 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6349 ST.mark_loc = locinput;
6350 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6351 assert(0); /* NOTREACHED */
6353 case MARKPOINT_next:
6354 mark_state = ST.prev_mark;
6356 assert(0); /* NOTREACHED */
6358 case MARKPOINT_next_fail:
6359 if (popmark && sv_eq(ST.mark_name,popmark))
6361 if (ST.mark_loc > startpoint)
6362 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6363 popmark = NULL; /* we found our mark */
6364 sv_commit = ST.mark_name;
6367 PerlIO_printf(Perl_debug_log,
6368 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6369 REPORT_CODE_OFF+depth*2, "",
6370 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6373 mark_state = ST.prev_mark;
6374 sv_yes_mark = mark_state ?
6375 mark_state->u.mark.mark_name : NULL;
6377 assert(0); /* NOTREACHED */
6379 case SKIP: /* (*SKIP) */
6381 /* (*SKIP) : if we fail we cut here*/
6382 ST.mark_name = NULL;
6383 ST.mark_loc = locinput;
6384 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6386 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6387 otherwise do nothing. Meaning we need to scan
6389 regmatch_state *cur = mark_state;
6390 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6393 if ( sv_eq( cur->u.mark.mark_name,
6396 ST.mark_name = find;
6397 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6399 cur = cur->u.mark.prev_mark;
6402 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6405 case SKIP_next_fail:
6407 /* (*CUT:NAME) - Set up to search for the name as we
6408 collapse the stack*/
6409 popmark = ST.mark_name;
6411 /* (*CUT) - No name, we cut here.*/
6412 if (ST.mark_loc > startpoint)
6413 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6414 /* but we set sv_commit to latest mark_name if there
6415 is one so they can test to see how things lead to this
6418 sv_commit=mark_state->u.mark.mark_name;
6422 assert(0); /* NOTREACHED */
6425 case LNBREAK: /* \R */
6426 if ((n=is_LNBREAK_safe(locinput, PL_regeol, utf8_target))) {
6433 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6434 PTR2UV(scan), OP(scan));
6435 Perl_croak(aTHX_ "regexp memory corruption");
6437 /* this is a point to jump to in order to increment
6438 * locinput by one character */
6440 assert(!NEXTCHR_IS_EOS);
6442 locinput += PL_utf8skip[nextchr];
6443 /* locinput is allowed to go 1 char off the end, but not 2+ */
6444 if (locinput > PL_regeol)
6453 /* switch break jumps here */
6454 scan = next; /* prepare to execute the next op and ... */
6455 continue; /* ... jump back to the top, reusing st */
6456 assert(0); /* NOTREACHED */
6459 /* push a state that backtracks on success */
6460 st->u.yes.prev_yes_state = yes_state;
6464 /* push a new regex state, then continue at scan */
6466 regmatch_state *newst;
6469 regmatch_state *cur = st;
6470 regmatch_state *curyes = yes_state;
6472 regmatch_slab *slab = PL_regmatch_slab;
6473 for (;curd > -1;cur--,curd--) {
6474 if (cur < SLAB_FIRST(slab)) {
6476 cur = SLAB_LAST(slab);
6478 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6479 REPORT_CODE_OFF + 2 + depth * 2,"",
6480 curd, PL_reg_name[cur->resume_state],
6481 (curyes == cur) ? "yes" : ""
6484 curyes = cur->u.yes.prev_yes_state;
6487 DEBUG_STATE_pp("push")
6490 st->locinput = locinput;
6492 if (newst > SLAB_LAST(PL_regmatch_slab))
6493 newst = S_push_slab(aTHX);
6494 PL_regmatch_state = newst;
6496 locinput = pushinput;
6499 assert(0); /* NOTREACHED */
6504 * We get here only if there's trouble -- normally "case END" is
6505 * the terminating point.
6507 Perl_croak(aTHX_ "corrupted regexp pointers");
6513 /* we have successfully completed a subexpression, but we must now
6514 * pop to the state marked by yes_state and continue from there */
6515 assert(st != yes_state);
6517 while (st != yes_state) {
6519 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6520 PL_regmatch_slab = PL_regmatch_slab->prev;
6521 st = SLAB_LAST(PL_regmatch_slab);
6525 DEBUG_STATE_pp("pop (no final)");
6527 DEBUG_STATE_pp("pop (yes)");
6533 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6534 || yes_state > SLAB_LAST(PL_regmatch_slab))
6536 /* not in this slab, pop slab */
6537 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6538 PL_regmatch_slab = PL_regmatch_slab->prev;
6539 st = SLAB_LAST(PL_regmatch_slab);
6541 depth -= (st - yes_state);
6544 yes_state = st->u.yes.prev_yes_state;
6545 PL_regmatch_state = st;
6548 locinput= st->locinput;
6549 state_num = st->resume_state + no_final;
6550 goto reenter_switch;
6553 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6554 PL_colors[4], PL_colors[5]));
6556 if (PL_reg_state.re_state_eval_setup_done) {
6557 /* each successfully executed (?{...}) block does the equivalent of
6558 * local $^R = do {...}
6559 * When popping the save stack, all these locals would be undone;
6560 * bypass this by setting the outermost saved $^R to the latest
6562 if (oreplsv != GvSV(PL_replgv))
6563 sv_setsv(oreplsv, GvSV(PL_replgv));
6570 PerlIO_printf(Perl_debug_log,
6571 "%*s %sfailed...%s\n",
6572 REPORT_CODE_OFF+depth*2, "",
6573 PL_colors[4], PL_colors[5])
6585 /* there's a previous state to backtrack to */
6587 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6588 PL_regmatch_slab = PL_regmatch_slab->prev;
6589 st = SLAB_LAST(PL_regmatch_slab);
6591 PL_regmatch_state = st;
6592 locinput= st->locinput;
6594 DEBUG_STATE_pp("pop");
6596 if (yes_state == st)
6597 yes_state = st->u.yes.prev_yes_state;
6599 state_num = st->resume_state + 1; /* failure = success + 1 */
6600 goto reenter_switch;
6605 if (rex->intflags & PREGf_VERBARG_SEEN) {
6606 SV *sv_err = get_sv("REGERROR", 1);
6607 SV *sv_mrk = get_sv("REGMARK", 1);
6609 sv_commit = &PL_sv_no;
6611 sv_yes_mark = &PL_sv_yes;
6614 sv_commit = &PL_sv_yes;
6615 sv_yes_mark = &PL_sv_no;
6617 sv_setsv(sv_err, sv_commit);
6618 sv_setsv(sv_mrk, sv_yes_mark);
6622 if (last_pushed_cv) {
6625 PERL_UNUSED_VAR(SP);
6628 /* clean up; in particular, free all slabs above current one */
6629 LEAVE_SCOPE(oldsave);
6631 assert(!result || locinput - PL_bostr >= 0);
6632 return result ? locinput - PL_bostr : -1;
6636 - regrepeat - repeatedly match something simple, report how many
6638 * What 'simple' means is a node which can be the operand of a quantifier like
6641 * startposp - pointer a pointer to the start position. This is updated
6642 * to point to the byte following the highest successful
6644 * p - the regnode to be repeatedly matched against.
6645 * max - maximum number of things to match.
6646 * depth - (for debugging) backtracking depth.
6649 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
6650 I32 max, int depth, bool is_utf8_pat)
6653 char *scan; /* Pointer to current position in target string */
6655 char *loceol = PL_regeol; /* local version */
6656 I32 hardcount = 0; /* How many matches so far */
6657 bool utf8_target = PL_reg_match_utf8;
6658 int to_complement = 0; /* Invert the result? */
6660 _char_class_number classnum;
6662 PERL_UNUSED_ARG(depth);
6665 PERL_ARGS_ASSERT_REGREPEAT;
6668 if (max == REG_INFTY)
6670 else if (! utf8_target && scan + max < loceol)
6671 loceol = scan + max;
6673 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
6674 * to the maximum of how far we should go in it (leaving it set to the real
6675 * end, if the maximum permissible would take us beyond that). This allows
6676 * us to make the loop exit condition that we haven't gone past <loceol> to
6677 * also mean that we haven't exceeded the max permissible count, saving a
6678 * test each time through the loop. But it assumes that the OP matches a
6679 * single byte, which is true for most of the OPs below when applied to a
6680 * non-UTF-8 target. Those relatively few OPs that don't have this
6681 * characteristic will have to compensate.
6683 * There is no adjustment for UTF-8 targets, as the number of bytes per
6684 * character varies. OPs will have to test both that the count is less
6685 * than the max permissible (using <hardcount> to keep track), and that we
6686 * are still within the bounds of the string (using <loceol>. A few OPs
6687 * match a single byte no matter what the encoding. They can omit the max
6688 * test if, for the UTF-8 case, they do the adjustment that was skipped
6691 * Thus, the code above sets things up for the common case; and exceptional
6692 * cases need extra work; the common case is to make sure <scan> doesn't
6693 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
6694 * count doesn't exceed the maximum permissible */
6699 while (scan < loceol && hardcount < max && *scan != '\n') {
6700 scan += UTF8SKIP(scan);
6704 while (scan < loceol && *scan != '\n')
6710 while (scan < loceol && hardcount < max) {
6711 scan += UTF8SKIP(scan);
6718 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
6719 if (utf8_target && scan + max < loceol) {
6721 /* <loceol> hadn't been adjusted in the UTF-8 case */
6729 assert(STR_LEN(p) == is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6733 /* Can use a simple loop if the pattern char to match on is invariant
6734 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
6735 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
6736 * true iff it doesn't matter if the argument is in UTF-8 or not */
6737 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! is_utf8_pat)) {
6738 if (utf8_target && scan + max < loceol) {
6739 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6740 * since here, to match at all, 1 char == 1 byte */
6741 loceol = scan + max;
6743 while (scan < loceol && UCHARAT(scan) == c) {
6747 else if (is_utf8_pat) {
6749 STRLEN scan_char_len;
6751 /* When both target and pattern are UTF-8, we have to do
6753 while (hardcount < max
6755 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
6756 && memEQ(scan, STRING(p), scan_char_len))
6758 scan += scan_char_len;
6762 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
6764 /* Target isn't utf8; convert the character in the UTF-8
6765 * pattern to non-UTF8, and do a simple loop */
6766 c = TWO_BYTE_UTF8_TO_UNI(c, *(STRING(p) + 1));
6767 while (scan < loceol && UCHARAT(scan) == c) {
6770 } /* else pattern char is above Latin1, can't possibly match the
6775 /* Here, the string must be utf8; pattern isn't, and <c> is
6776 * different in utf8 than not, so can't compare them directly.
6777 * Outside the loop, find the two utf8 bytes that represent c, and
6778 * then look for those in sequence in the utf8 string */
6779 U8 high = UTF8_TWO_BYTE_HI(c);
6780 U8 low = UTF8_TWO_BYTE_LO(c);
6782 while (hardcount < max
6783 && scan + 1 < loceol
6784 && UCHARAT(scan) == high
6785 && UCHARAT(scan + 1) == low)
6794 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6798 RXp_MATCH_TAINTED_on(prog);
6799 utf8_flags = FOLDEQ_UTF8_LOCALE;
6807 case EXACTFU_TRICKYFOLD:
6809 utf8_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6813 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
6815 assert(STR_LEN(p) == is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6817 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
6820 if (c1 == CHRTEST_VOID) {
6821 /* Use full Unicode fold matching */
6822 char *tmpeol = PL_regeol;
6823 STRLEN pat_len = is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
6824 while (hardcount < max
6825 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6826 STRING(p), NULL, pat_len,
6827 is_utf8_pat, utf8_flags))
6834 else if (utf8_target) {
6836 while (scan < loceol
6838 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
6840 scan += UTF8SKIP(scan);
6845 while (scan < loceol
6847 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
6848 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
6850 scan += UTF8SKIP(scan);
6855 else if (c1 == c2) {
6856 while (scan < loceol && UCHARAT(scan) == c1) {
6861 while (scan < loceol &&
6862 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
6871 case ANYOF_WARN_SUPER:
6873 while (hardcount < max
6875 && reginclass(prog, p, (U8*)scan, utf8_target))
6877 scan += UTF8SKIP(scan);
6881 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6886 /* The argument (FLAGS) to all the POSIX node types is the class number */
6893 RXp_MATCH_TAINTED_on(prog);
6894 if (! utf8_target) {
6895 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
6901 while (hardcount < max && scan < loceol
6902 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
6905 scan += UTF8SKIP(scan);
6918 if (utf8_target && scan + max < loceol) {
6920 /* We didn't adjust <loceol> at the beginning of this routine
6921 * because is UTF-8, but it is actually ok to do so, since here, to
6922 * match, 1 char == 1 byte. */
6923 loceol = scan + max;
6925 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6938 if (! utf8_target) {
6939 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6945 /* The complement of something that matches only ASCII matches all
6946 * UTF-8 variant code points, plus everything in ASCII that isn't
6948 while (hardcount < max && scan < loceol
6949 && (! UTF8_IS_INVARIANT(*scan)
6950 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
6952 scan += UTF8SKIP(scan);
6963 if (! utf8_target) {
6964 while (scan < loceol && to_complement
6965 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
6972 classnum = (_char_class_number) FLAGS(p);
6973 if (classnum < _FIRST_NON_SWASH_CC) {
6975 /* Here, a swash is needed for above-Latin1 code points.
6976 * Process as many Latin1 code points using the built-in rules.
6977 * Go to another loop to finish processing upon encountering
6978 * the first Latin1 code point. We could do that in this loop
6979 * as well, but the other way saves having to test if the swash
6980 * has been loaded every time through the loop: extra space to
6982 while (hardcount < max && scan < loceol) {
6983 if (UTF8_IS_INVARIANT(*scan)) {
6984 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
6991 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
6992 if (! (to_complement
6993 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(*scan,
7002 goto found_above_latin1;
7009 /* For these character classes, the knowledge of how to handle
7010 * every code point is compiled in to Perl via a macro. This
7011 * code is written for making the loops as tight as possible.
7012 * It could be refactored to save space instead */
7014 case _CC_ENUM_SPACE: /* XXX would require separate code
7015 if we revert the change of \v
7018 case _CC_ENUM_PSXSPC:
7019 while (hardcount < max
7021 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7023 scan += UTF8SKIP(scan);
7027 case _CC_ENUM_BLANK:
7028 while (hardcount < max
7030 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7032 scan += UTF8SKIP(scan);
7036 case _CC_ENUM_XDIGIT:
7037 while (hardcount < max
7039 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7041 scan += UTF8SKIP(scan);
7045 case _CC_ENUM_VERTSPACE:
7046 while (hardcount < max
7048 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7050 scan += UTF8SKIP(scan);
7054 case _CC_ENUM_CNTRL:
7055 while (hardcount < max
7057 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7059 scan += UTF8SKIP(scan);
7064 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7070 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7072 /* Load the swash if not already present */
7073 if (! PL_utf8_swash_ptrs[classnum]) {
7074 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7075 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7076 "utf8", swash_property_names[classnum],
7077 &PL_sv_undef, 1, 0, NULL, &flags);
7080 while (hardcount < max && scan < loceol
7081 && to_complement ^ cBOOL(_generic_utf8(
7084 swash_fetch(PL_utf8_swash_ptrs[classnum],
7088 scan += UTF8SKIP(scan);
7095 while (hardcount < max && scan < loceol &&
7096 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7101 /* LNBREAK can match one or two latin chars, which is ok, but we
7102 * have to use hardcount in this situation, and throw away the
7103 * adjustment to <loceol> done before the switch statement */
7105 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7126 /* These are all 0 width, so match right here or not at all. */
7130 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7131 assert(0); /* NOTREACHED */
7138 c = scan - *startposp;
7142 GET_RE_DEBUG_FLAGS_DECL;
7144 SV * const prop = sv_newmortal();
7145 regprop(prog, prop, p);
7146 PerlIO_printf(Perl_debug_log,
7147 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7148 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7156 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7158 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7159 create a copy so that changes the caller makes won't change the shared one.
7160 If <altsvp> is non-null, will return NULL in it, for back-compat.
7163 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7165 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7171 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp));
7176 S_core_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp)
7178 /* Returns the swash for the input 'node' in the regex 'prog'.
7179 * If <doinit> is true, will attempt to create the swash if not already
7181 * If <listsvp> is non-null, will return the swash initialization string in
7183 * Tied intimately to how regcomp.c sets up the data structure */
7190 RXi_GET_DECL(prog,progi);
7191 const struct reg_data * const data = prog ? progi->data : NULL;
7193 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
7195 assert(ANYOF_NONBITMAP(node));
7197 if (data && data->count) {
7198 const U32 n = ARG(node);
7200 if (data->what[n] == 's') {
7201 SV * const rv = MUTABLE_SV(data->data[n]);
7202 AV * const av = MUTABLE_AV(SvRV(rv));
7203 SV **const ary = AvARRAY(av);
7204 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7206 si = *ary; /* ary[0] = the string to initialize the swash with */
7208 /* Elements 2 and 3 are either both present or both absent. [2] is
7209 * any inversion list generated at compile time; [3] indicates if
7210 * that inversion list has any user-defined properties in it. */
7211 if (av_len(av) >= 2) {
7214 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7221 /* Element [1] is reserved for the set-up swash. If already there,
7222 * return it; if not, create it and store it there */
7223 if (SvROK(ary[1])) {
7226 else if (si && doinit) {
7228 sw = _core_swash_init("utf8", /* the utf8 package */
7232 0, /* not from tr/// */
7235 (void)av_store(av, 1, sw);
7241 SV* matches_string = newSVpvn("", 0);
7243 /* Use the swash, if any, which has to have incorporated into it all
7245 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7246 && (si && si != &PL_sv_undef))
7249 /* If no swash, use the input initialization string, if available */
7250 sv_catsv(matches_string, si);
7253 /* Add the inversion list to whatever we have. This may have come from
7254 * the swash, or from an input parameter */
7256 sv_catsv(matches_string, _invlist_contents(invlist));
7258 *listsvp = matches_string;
7265 - reginclass - determine if a character falls into a character class
7267 n is the ANYOF regnode
7268 p is the target string
7269 utf8_target tells whether p is in UTF-8.
7271 Returns true if matched; false otherwise.
7273 Note that this can be a synthetic start class, a combination of various
7274 nodes, so things you think might be mutually exclusive, such as locale,
7275 aren't. It can match both locale and non-locale
7280 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const bool utf8_target)
7283 const char flags = ANYOF_FLAGS(n);
7287 PERL_ARGS_ASSERT_REGINCLASS;
7289 /* If c is not already the code point, get it. Note that
7290 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7291 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7293 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
7294 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7295 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7296 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7297 * UTF8_ALLOW_FFFF */
7298 if (c_len == (STRLEN)-1)
7299 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7302 /* If this character is potentially in the bitmap, check it */
7304 if (ANYOF_BITMAP_TEST(n, c))
7306 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
7312 else if (flags & ANYOF_LOCALE) {
7313 RXp_MATCH_TAINTED_on(prog);
7315 if ((flags & ANYOF_LOC_FOLD)
7316 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7320 else if (ANYOF_CLASS_TEST_ANY_SET(n)) {
7322 /* The data structure is arranged so bits 0, 2, 4, ... are set
7323 * if the class includes the Posix character class given by
7324 * bit/2; and 1, 3, 5, ... are set if the class includes the
7325 * complemented Posix class given by int(bit/2). So we loop
7326 * through the bits, each time changing whether we complement
7327 * the result or not. Suppose for the sake of illustration
7328 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7329 * is set, it means there is a match for this ANYOF node if the
7330 * character is in the class given by the expression (0 / 2 = 0
7331 * = \w). If it is in that class, isFOO_lc() will return 1,
7332 * and since 'to_complement' is 0, the result will stay TRUE,
7333 * and we exit the loop. Suppose instead that bit 0 is 0, but
7334 * bit 1 is 1. That means there is a match if the character
7335 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7336 * but will on bit 1. On the second iteration 'to_complement'
7337 * will be 1, so the exclusive or will reverse things, so we
7338 * are testing for \W. On the third iteration, 'to_complement'
7339 * will be 0, and we would be testing for \s; the fourth
7340 * iteration would test for \S, etc.
7342 * Note that this code assumes that all the classes are closed
7343 * under folding. For example, if a character matches \w, then
7344 * its fold does too; and vice versa. This should be true for
7345 * any well-behaved locale for all the currently defined Posix
7346 * classes, except for :lower: and :upper:, which are handled
7347 * by the pseudo-class :cased: which matches if either of the
7348 * other two does. To get rid of this assumption, an outer
7349 * loop could be used below to iterate over both the source
7350 * character, and its fold (if different) */
7353 int to_complement = 0;
7354 while (count < ANYOF_MAX) {
7355 if (ANYOF_CLASS_TEST(n, count)
7356 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7368 /* If the bitmap didn't (or couldn't) match, and something outside the
7369 * bitmap could match, try that. Locale nodes specify completely the
7370 * behavior of code points in the bit map (otherwise, a utf8 target would
7371 * cause them to be treated as Unicode and not locale), except in
7372 * the very unlikely event when this node is a synthetic start class, which
7373 * could be a combination of locale and non-locale nodes. So allow locale
7374 * to match for the synthetic start class, which will give a false
7375 * positive that will be resolved when the match is done again as not part
7376 * of the synthetic start class */
7378 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
7379 match = TRUE; /* Everything above 255 matches */
7381 else if (ANYOF_NONBITMAP(n)
7382 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
7385 || (! (flags & ANYOF_LOCALE))
7386 || OP(n) == ANYOF_SYNTHETIC))))
7388 SV * const sw = core_regclass_swash(prog, n, TRUE, 0);
7393 } else { /* Convert to utf8 */
7395 utf8_p = bytes_to_utf8(p, &len);
7398 if (swash_fetch(sw, utf8_p, TRUE)) {
7402 /* If we allocated a string above, free it */
7403 if (! utf8_target) Safefree(utf8_p);
7407 if (UNICODE_IS_SUPER(c)
7408 && OP(n) == ANYOF_WARN_SUPER
7409 && ckWARN_d(WARN_NON_UNICODE))
7411 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7412 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7416 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7417 return cBOOL(flags & ANYOF_INVERT) ^ match;
7421 S_reghop3(U8 *s, I32 off, const U8* lim)
7423 /* return the position 'off' UTF-8 characters away from 's', forward if
7424 * 'off' >= 0, backwards if negative. But don't go outside of position
7425 * 'lim', which better be < s if off < 0 */
7429 PERL_ARGS_ASSERT_REGHOP3;
7432 while (off-- && s < lim) {
7433 /* XXX could check well-formedness here */
7438 while (off++ && s > lim) {
7440 if (UTF8_IS_CONTINUED(*s)) {
7441 while (s > lim && UTF8_IS_CONTINUATION(*s))
7444 /* XXX could check well-formedness here */
7451 /* there are a bunch of places where we use two reghop3's that should
7452 be replaced with this routine. but since thats not done yet
7453 we ifdef it out - dmq
7456 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7460 PERL_ARGS_ASSERT_REGHOP4;
7463 while (off-- && s < rlim) {
7464 /* XXX could check well-formedness here */
7469 while (off++ && s > llim) {
7471 if (UTF8_IS_CONTINUED(*s)) {
7472 while (s > llim && UTF8_IS_CONTINUATION(*s))
7475 /* XXX could check well-formedness here */
7483 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7487 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7490 while (off-- && s < lim) {
7491 /* XXX could check well-formedness here */
7498 while (off++ && s > lim) {
7500 if (UTF8_IS_CONTINUED(*s)) {
7501 while (s > lim && UTF8_IS_CONTINUATION(*s))
7504 /* XXX could check well-formedness here */
7513 restore_pos(pTHX_ void *arg)
7516 regexp * const rex = (regexp *)arg;
7517 if (PL_reg_state.re_state_eval_setup_done) {
7518 if (PL_reg_oldsaved) {
7519 rex->subbeg = PL_reg_oldsaved;
7520 rex->sublen = PL_reg_oldsavedlen;
7521 rex->suboffset = PL_reg_oldsavedoffset;
7522 rex->subcoffset = PL_reg_oldsavedcoffset;
7524 rex->saved_copy = PL_nrs;
7526 RXp_MATCH_COPIED_on(rex);
7528 PL_reg_magic->mg_len = PL_reg_oldpos;
7529 PL_reg_state.re_state_eval_setup_done = FALSE;
7530 PL_curpm = PL_reg_oldcurpm;
7535 S_to_utf8_substr(pTHX_ regexp *prog)
7537 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
7538 * on the converted value */
7542 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7545 if (prog->substrs->data[i].substr
7546 && !prog->substrs->data[i].utf8_substr) {
7547 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7548 prog->substrs->data[i].utf8_substr = sv;
7549 sv_utf8_upgrade(sv);
7550 if (SvVALID(prog->substrs->data[i].substr)) {
7551 if (SvTAIL(prog->substrs->data[i].substr)) {
7552 /* Trim the trailing \n that fbm_compile added last
7554 SvCUR_set(sv, SvCUR(sv) - 1);
7555 /* Whilst this makes the SV technically "invalid" (as its
7556 buffer is no longer followed by "\0") when fbm_compile()
7557 adds the "\n" back, a "\0" is restored. */
7558 fbm_compile(sv, FBMcf_TAIL);
7562 if (prog->substrs->data[i].substr == prog->check_substr)
7563 prog->check_utf8 = sv;
7569 S_to_byte_substr(pTHX_ regexp *prog)
7571 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
7572 * on the converted value; returns FALSE if can't be converted. */
7577 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7580 if (prog->substrs->data[i].utf8_substr
7581 && !prog->substrs->data[i].substr) {
7582 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7583 if (! sv_utf8_downgrade(sv, TRUE)) {
7586 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7587 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7588 /* Trim the trailing \n that fbm_compile added last
7590 SvCUR_set(sv, SvCUR(sv) - 1);
7591 fbm_compile(sv, FBMcf_TAIL);
7595 prog->substrs->data[i].substr = sv;
7596 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7597 prog->check_substr = sv;
7606 * c-indentation-style: bsd
7608 * indent-tabs-mode: nil
7611 * ex: set ts=8 sts=4 sw=4 et: