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
87 #ifdef PERL_IN_XSUB_RE
93 #include "inline_invlist.c"
94 #include "unicode_constants.h"
96 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
102 /* Valid for non-utf8 strings: avoids the reginclass
103 * call if there are no complications: i.e., if everything matchable is
104 * straight forward in the bitmap */
105 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \
106 : ANYOF_BITMAP_TEST(p,*(c)))
112 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
113 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
115 #define HOPc(pos,off) \
116 (char *)(PL_reg_match_utf8 \
117 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
119 #define HOPBACKc(pos, off) \
120 (char*)(PL_reg_match_utf8\
121 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
122 : (pos - off >= PL_bostr) \
126 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
127 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
130 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
131 #define NEXTCHR_IS_EOS (nextchr < 0)
133 #define SET_nextchr \
134 nextchr = ((locinput < PL_regeol) ? UCHARAT(locinput) : NEXTCHR_EOS)
136 #define SET_locinput(p) \
141 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name) STMT_START { \
143 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
144 ENTER; save_re_context(); \
145 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
146 1, 0, NULL, &flags); \
151 /* If in debug mode, we test that a known character properly matches */
153 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
155 utf8_char_in_property) \
156 LOAD_UTF8_CHARCLASS(swash_ptr, property_name); \
157 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
159 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
161 utf8_char_in_property) \
162 LOAD_UTF8_CHARCLASS(swash_ptr, property_name)
165 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
166 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
167 swash_property_names[_CC_WORDCHAR], \
168 GREEK_SMALL_LETTER_IOTA_UTF8)
170 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
172 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
173 "_X_regular_begin", \
174 GREEK_SMALL_LETTER_IOTA_UTF8); \
175 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
177 COMBINING_GRAVE_ACCENT_UTF8); \
180 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
181 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
183 /* for use after a quantifier and before an EXACT-like node -- japhy */
184 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
186 * NOTE that *nothing* that affects backtracking should be in here, specifically
187 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
188 * node that is in between two EXACT like nodes when ascertaining what the required
189 * "follow" character is. This should probably be moved to regex compile time
190 * although it may be done at run time beause of the REF possibility - more
191 * investigation required. -- demerphq
193 #define JUMPABLE(rn) ( \
195 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
197 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
198 OP(rn) == PLUS || OP(rn) == MINMOD || \
200 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
202 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
204 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
207 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
208 we don't need this definition. */
209 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
210 #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 )
211 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
214 /* ... so we use this as its faster. */
215 #define IS_TEXT(rn) ( OP(rn)==EXACT )
216 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
217 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
218 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
223 Search for mandatory following text node; for lookahead, the text must
224 follow but for lookbehind (rn->flags != 0) we skip to the next step.
226 #define FIND_NEXT_IMPT(rn) STMT_START { \
227 while (JUMPABLE(rn)) { \
228 const OPCODE type = OP(rn); \
229 if (type == SUSPEND || PL_regkind[type] == CURLY) \
230 rn = NEXTOPER(NEXTOPER(rn)); \
231 else if (type == PLUS) \
233 else if (type == IFMATCH) \
234 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
235 else rn += NEXT_OFF(rn); \
239 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
240 * These are for the pre-composed Hangul syllables, which are all in a
241 * contiguous block and arranged there in such a way so as to facilitate
242 * alorithmic determination of their characteristics. As such, they don't need
243 * a swash, but can be determined by simple arithmetic. Almost all are
244 * GCB=LVT, but every 28th one is a GCB=LV */
245 #define SBASE 0xAC00 /* Start of block */
246 #define SCount 11172 /* Length of block */
249 static void restore_pos(pTHX_ void *arg);
251 #define REGCP_PAREN_ELEMS 3
252 #define REGCP_OTHER_ELEMS 3
253 #define REGCP_FRAME_ELEMS 1
254 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
255 * are needed for the regexp context stack bookkeeping. */
258 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
261 const int retval = PL_savestack_ix;
262 const int paren_elems_to_push =
263 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
264 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
265 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
267 GET_RE_DEBUG_FLAGS_DECL;
269 PERL_ARGS_ASSERT_REGCPPUSH;
271 if (paren_elems_to_push < 0)
272 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
273 paren_elems_to_push);
275 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
276 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
277 " out of range (%lu-%ld)",
279 (unsigned long)maxopenparen,
282 SSGROW(total_elems + REGCP_FRAME_ELEMS);
285 if ((int)maxopenparen > (int)parenfloor)
286 PerlIO_printf(Perl_debug_log,
287 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
292 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
293 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
294 SSPUSHINT(rex->offs[p].end);
295 SSPUSHINT(rex->offs[p].start);
296 SSPUSHINT(rex->offs[p].start_tmp);
297 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
298 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
300 (IV)rex->offs[p].start,
301 (IV)rex->offs[p].start_tmp,
305 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
306 SSPUSHINT(maxopenparen);
307 SSPUSHINT(rex->lastparen);
308 SSPUSHINT(rex->lastcloseparen);
309 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
314 /* These are needed since we do not localize EVAL nodes: */
315 #define REGCP_SET(cp) \
317 PerlIO_printf(Perl_debug_log, \
318 " Setting an EVAL scope, savestack=%"IVdf"\n", \
319 (IV)PL_savestack_ix)); \
322 #define REGCP_UNWIND(cp) \
324 if (cp != PL_savestack_ix) \
325 PerlIO_printf(Perl_debug_log, \
326 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
327 (IV)(cp), (IV)PL_savestack_ix)); \
330 #define UNWIND_PAREN(lp, lcp) \
331 for (n = rex->lastparen; n > lp; n--) \
332 rex->offs[n].end = -1; \
333 rex->lastparen = n; \
334 rex->lastcloseparen = lcp;
338 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
343 GET_RE_DEBUG_FLAGS_DECL;
345 PERL_ARGS_ASSERT_REGCPPOP;
347 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
349 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
350 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
351 rex->lastcloseparen = SSPOPINT;
352 rex->lastparen = SSPOPINT;
353 *maxopenparen_p = SSPOPINT;
355 i -= REGCP_OTHER_ELEMS;
356 /* Now restore the parentheses context. */
358 if (i || rex->lastparen + 1 <= rex->nparens)
359 PerlIO_printf(Perl_debug_log,
360 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
365 paren = *maxopenparen_p;
366 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
368 rex->offs[paren].start_tmp = SSPOPINT;
369 rex->offs[paren].start = SSPOPINT;
371 if (paren <= rex->lastparen)
372 rex->offs[paren].end = tmps;
373 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
374 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
376 (IV)rex->offs[paren].start,
377 (IV)rex->offs[paren].start_tmp,
378 (IV)rex->offs[paren].end,
379 (paren > rex->lastparen ? "(skipped)" : ""));
384 /* It would seem that the similar code in regtry()
385 * already takes care of this, and in fact it is in
386 * a better location to since this code can #if 0-ed out
387 * but the code in regtry() is needed or otherwise tests
388 * requiring null fields (pat.t#187 and split.t#{13,14}
389 * (as of patchlevel 7877) will fail. Then again,
390 * this code seems to be necessary or otherwise
391 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
392 * --jhi updated by dapm */
393 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
394 if (i > *maxopenparen_p)
395 rex->offs[i].start = -1;
396 rex->offs[i].end = -1;
397 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
398 " \\%"UVuf": %s ..-1 undeffing\n",
400 (i > *maxopenparen_p) ? "-1" : " "
406 /* restore the parens and associated vars at savestack position ix,
407 * but without popping the stack */
410 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
412 I32 tmpix = PL_savestack_ix;
413 PL_savestack_ix = ix;
414 regcppop(rex, maxopenparen_p);
415 PL_savestack_ix = tmpix;
418 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
421 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
423 /* Returns a boolean as to whether or not 'character' is a member of the
424 * Posix character class given by 'classnum' that should be equivalent to a
425 * value in the typedef '_char_class_number'.
427 * Ideally this could be replaced by a just an array of function pointers
428 * to the C library functions that implement the macros this calls.
429 * However, to compile, the precise function signatures are required, and
430 * these may vary from platform to to platform. To avoid having to figure
431 * out what those all are on each platform, I (khw) am using this method,
432 * which adds an extra layer of function call overhead (unless the C
433 * optimizer strips it away). But we don't particularly care about
434 * performance with locales anyway. */
436 switch ((_char_class_number) classnum) {
437 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
438 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
439 case _CC_ENUM_ASCII: return isASCII_LC(character);
440 case _CC_ENUM_BLANK: return isBLANK_LC(character);
441 case _CC_ENUM_CASED: return isLOWER_LC(character)
442 || isUPPER_LC(character);
443 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
444 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
445 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
446 case _CC_ENUM_LOWER: return isLOWER_LC(character);
447 case _CC_ENUM_PRINT: return isPRINT_LC(character);
448 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
449 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
450 case _CC_ENUM_SPACE: return isSPACE_LC(character);
451 case _CC_ENUM_UPPER: return isUPPER_LC(character);
452 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
453 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
454 default: /* VERTSPACE should never occur in locales */
455 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
458 assert(0); /* NOTREACHED */
463 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
465 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
466 * 'character' is a member of the Posix character class given by 'classnum'
467 * that should be equivalent to a value in the typedef
468 * '_char_class_number'.
470 * This just calls isFOO_lc on the code point for the character if it is in
471 * the range 0-255. Outside that range, all characters avoid Unicode
472 * rules, ignoring any locale. So use the Unicode function if this class
473 * requires a swash, and use the Unicode macro otherwise. */
475 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
477 if (UTF8_IS_INVARIANT(*character)) {
478 return isFOO_lc(classnum, *character);
480 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
481 return isFOO_lc(classnum,
482 TWO_BYTE_UTF8_TO_UNI(*character, *(character + 1)));
485 if (classnum < _FIRST_NON_SWASH_CC) {
487 /* Initialize the swash unless done already */
488 if (! PL_utf8_swash_ptrs[classnum]) {
489 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
490 PL_utf8_swash_ptrs[classnum] = _core_swash_init("utf8",
491 swash_property_names[classnum], &PL_sv_undef, 1, 0, NULL, &flags);
494 return swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *) character, TRUE);
497 switch ((_char_class_number) classnum) {
499 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
501 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
502 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
503 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
504 default: return 0; /* Things like CNTRL are always
508 assert(0); /* NOTREACHED */
513 * pregexec and friends
516 #ifndef PERL_IN_XSUB_RE
518 - pregexec - match a regexp against a string
521 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
522 char *strbeg, I32 minend, SV *screamer, U32 nosave)
523 /* stringarg: the point in the string at which to begin matching */
524 /* strend: pointer to null at end of string */
525 /* strbeg: real beginning of string */
526 /* minend: end of match must be >= minend bytes after stringarg. */
527 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
528 * itself is accessed via the pointers above */
529 /* nosave: For optimizations. */
531 PERL_ARGS_ASSERT_PREGEXEC;
534 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
535 nosave ? 0 : REXEC_COPY_STR);
540 * Need to implement the following flags for reg_anch:
542 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
544 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
545 * INTUIT_AUTORITATIVE_ML
546 * INTUIT_ONCE_NOML - Intuit can match in one location only.
549 * Another flag for this function: SECOND_TIME (so that float substrs
550 * with giant delta may be not rechecked).
553 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
555 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
556 Otherwise, only SvCUR(sv) is used to get strbeg. */
558 /* XXXX We assume that strpos is strbeg unless sv. */
560 /* XXXX Some places assume that there is a fixed substring.
561 An update may be needed if optimizer marks as "INTUITable"
562 RExen without fixed substrings. Similarly, it is assumed that
563 lengths of all the strings are no more than minlen, thus they
564 cannot come from lookahead.
565 (Or minlen should take into account lookahead.)
566 NOTE: Some of this comment is not correct. minlen does now take account
567 of lookahead/behind. Further research is required. -- demerphq
571 /* A failure to find a constant substring means that there is no need to make
572 an expensive call to REx engine, thus we celebrate a failure. Similarly,
573 finding a substring too deep into the string means that fewer calls to
574 regtry() should be needed.
576 REx compiler's optimizer found 4 possible hints:
577 a) Anchored substring;
579 c) Whether we are anchored (beginning-of-line or \G);
580 d) First node (of those at offset 0) which may distinguish positions;
581 We use a)b)d) and multiline-part of c), and try to find a position in the
582 string which does not contradict any of them.
585 /* Most of decisions we do here should have been done at compile time.
586 The nodes of the REx which we used for the search should have been
587 deleted from the finite automaton. */
590 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
591 char *strend, const U32 flags, re_scream_pos_data *data)
594 struct regexp *const prog = ReANY(rx);
596 /* Should be nonnegative! */
602 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
604 char *other_last = NULL; /* other substr checked before this */
605 char *check_at = NULL; /* check substr found at this pos */
606 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
607 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
608 RXi_GET_DECL(prog,progi);
611 const char * const i_strpos = strpos;
613 GET_RE_DEBUG_FLAGS_DECL;
615 PERL_ARGS_ASSERT_RE_INTUIT_START;
616 PERL_UNUSED_ARG(flags);
617 PERL_UNUSED_ARG(data);
619 RX_MATCH_UTF8_set(rx,utf8_target);
621 is_utf8_pat = cBOOL(RX_UTF8(rx));
624 debug_start_match(rx, utf8_target, strpos, strend,
625 sv ? "Guessing start of match in sv for"
626 : "Guessing start of match in string for");
629 /* CHR_DIST() would be more correct here but it makes things slow. */
630 if (prog->minlen > strend - strpos) {
631 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
632 "String too short... [re_intuit_start]\n"));
636 /* XXX we need to pass strbeg as a separate arg: the following is
637 * guesswork and can be wrong... */
638 if (sv && SvPOK(sv)) {
639 char * p = SvPVX(sv);
640 STRLEN cur = SvCUR(sv);
641 if (p <= strpos && strpos < p + cur) {
643 assert(p <= strend && strend <= p + cur);
646 strbeg = strend - cur;
653 if (!prog->check_utf8 && prog->check_substr)
654 to_utf8_substr(prog);
655 check = prog->check_utf8;
657 if (!prog->check_substr && prog->check_utf8) {
658 if (! to_byte_substr(prog)) {
659 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
662 check = prog->check_substr;
664 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
665 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
666 || ( (prog->extflags & RXf_ANCH_BOL)
667 && !multiline ) ); /* Check after \n? */
670 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
671 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
672 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
674 && (strpos != strbeg)) {
675 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
678 if (prog->check_offset_min == prog->check_offset_max
679 && !(prog->extflags & RXf_CANY_SEEN)
680 && ! multiline) /* /m can cause \n's to match that aren't
681 accounted for in the string max length.
682 See [perl #115242] */
684 /* Substring at constant offset from beg-of-str... */
687 s = HOP3c(strpos, prog->check_offset_min, strend);
690 slen = SvCUR(check); /* >= 1 */
692 if ( strend - s > slen || strend - s < slen - 1
693 || (strend - s == slen && strend[-1] != '\n')) {
694 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
697 /* Now should match s[0..slen-2] */
699 if (slen && (*SvPVX_const(check) != *s
701 && memNE(SvPVX_const(check), s, slen)))) {
703 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
707 else if (*SvPVX_const(check) != *s
708 || ((slen = SvCUR(check)) > 1
709 && memNE(SvPVX_const(check), s, slen)))
712 goto success_at_start;
715 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
717 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
718 end_shift = prog->check_end_shift;
721 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
722 - (SvTAIL(check) != 0);
723 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
725 if (end_shift < eshift)
729 else { /* Can match at random position */
732 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
733 end_shift = prog->check_end_shift;
735 /* end shift should be non negative here */
738 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
740 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
741 (IV)end_shift, RX_PRECOMP(prog));
745 /* Find a possible match in the region s..strend by looking for
746 the "check" substring in the region corrected by start/end_shift. */
749 I32 srch_start_shift = start_shift;
750 I32 srch_end_shift = end_shift;
753 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
754 srch_end_shift -= ((strbeg - s) - srch_start_shift);
755 srch_start_shift = strbeg - s;
757 DEBUG_OPTIMISE_MORE_r({
758 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
759 (IV)prog->check_offset_min,
760 (IV)srch_start_shift,
762 (IV)prog->check_end_shift);
765 if (prog->extflags & RXf_CANY_SEEN) {
766 start_point= (U8*)(s + srch_start_shift);
767 end_point= (U8*)(strend - srch_end_shift);
769 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
770 end_point= HOP3(strend, -srch_end_shift, strbeg);
772 DEBUG_OPTIMISE_MORE_r({
773 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
774 (int)(end_point - start_point),
775 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
779 s = fbm_instr( start_point, end_point,
780 check, multiline ? FBMrf_MULTILINE : 0);
782 /* Update the count-of-usability, remove useless subpatterns,
786 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
787 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
788 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
789 (s ? "Found" : "Did not find"),
790 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
791 ? "anchored" : "floating"),
794 (s ? " at offset " : "...\n") );
799 /* Finish the diagnostic message */
800 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
802 /* XXX dmq: first branch is for positive lookbehind...
803 Our check string is offset from the beginning of the pattern.
804 So we need to do any stclass tests offset forward from that
813 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
814 Start with the other substr.
815 XXXX no SCREAM optimization yet - and a very coarse implementation
816 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
817 *always* match. Probably should be marked during compile...
818 Probably it is right to do no SCREAM here...
821 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
822 : (prog->float_substr && prog->anchored_substr))
824 /* Take into account the "other" substring. */
825 /* XXXX May be hopelessly wrong for UTF... */
828 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
831 char * const last = HOP3c(s, -start_shift, strbeg);
833 char * const saved_s = s;
836 t = s - prog->check_offset_max;
837 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
839 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
844 t = HOP3c(t, prog->anchored_offset, strend);
845 if (t < other_last) /* These positions already checked */
847 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
850 /* XXXX It is not documented what units *_offsets are in.
851 We assume bytes, but this is clearly wrong.
852 Meaning this code needs to be carefully reviewed for errors.
856 /* On end-of-str: see comment below. */
857 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
858 if (must == &PL_sv_undef) {
860 DEBUG_r(must = prog->anchored_utf8); /* for debug */
865 HOP3(HOP3(last1, prog->anchored_offset, strend)
866 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
868 multiline ? FBMrf_MULTILINE : 0
871 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
872 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
873 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
874 (s ? "Found" : "Contradicts"),
875 quoted, RE_SV_TAIL(must));
880 if (last1 >= last2) {
881 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
882 ", giving up...\n"));
885 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
886 ", trying floating at offset %ld...\n",
887 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
888 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
889 s = HOP3c(last, 1, strend);
893 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
894 (long)(s - i_strpos)));
895 t = HOP3c(s, -prog->anchored_offset, strbeg);
896 other_last = HOP3c(s, 1, strend);
904 else { /* Take into account the floating substring. */
906 char * const saved_s = s;
909 t = HOP3c(s, -start_shift, strbeg);
911 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
912 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
913 last = HOP3c(t, prog->float_max_offset, strend);
914 s = HOP3c(t, prog->float_min_offset, strend);
917 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
918 must = utf8_target ? prog->float_utf8 : prog->float_substr;
919 /* fbm_instr() takes into account exact value of end-of-str
920 if the check is SvTAIL(ed). Since false positives are OK,
921 and end-of-str is not later than strend we are OK. */
922 if (must == &PL_sv_undef) {
924 DEBUG_r(must = prog->float_utf8); /* for debug message */
927 s = fbm_instr((unsigned char*)s,
928 (unsigned char*)last + SvCUR(must)
930 must, multiline ? FBMrf_MULTILINE : 0);
932 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
933 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
934 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
935 (s ? "Found" : "Contradicts"),
936 quoted, RE_SV_TAIL(must));
940 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
941 ", giving up...\n"));
944 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
945 ", trying anchored starting at offset %ld...\n",
946 (long)(saved_s + 1 - i_strpos)));
948 s = HOP3c(t, 1, strend);
952 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
953 (long)(s - i_strpos)));
954 other_last = s; /* Fix this later. --Hugo */
964 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
966 DEBUG_OPTIMISE_MORE_r(
967 PerlIO_printf(Perl_debug_log,
968 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
969 (IV)prog->check_offset_min,
970 (IV)prog->check_offset_max,
978 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
980 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
983 /* Fixed substring is found far enough so that the match
984 cannot start at strpos. */
986 if (ml_anch && t[-1] != '\n') {
987 /* Eventually fbm_*() should handle this, but often
988 anchored_offset is not 0, so this check will not be wasted. */
989 /* XXXX In the code below we prefer to look for "^" even in
990 presence of anchored substrings. And we search even
991 beyond the found float position. These pessimizations
992 are historical artefacts only. */
994 while (t < strend - prog->minlen) {
996 if (t < check_at - prog->check_offset_min) {
997 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
998 /* Since we moved from the found position,
999 we definitely contradict the found anchored
1000 substr. Due to the above check we do not
1001 contradict "check" substr.
1002 Thus we can arrive here only if check substr
1003 is float. Redo checking for "other"=="fixed".
1006 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1007 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
1008 goto do_other_anchored;
1010 /* We don't contradict the found floating substring. */
1011 /* XXXX Why not check for STCLASS? */
1013 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1014 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1017 /* Position contradicts check-string */
1018 /* XXXX probably better to look for check-string
1019 than for "\n", so one should lower the limit for t? */
1020 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1021 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1022 other_last = strpos = s = t + 1;
1027 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1028 PL_colors[0], PL_colors[1]));
1032 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1033 PL_colors[0], PL_colors[1]));
1037 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1040 /* The found string does not prohibit matching at strpos,
1041 - no optimization of calling REx engine can be performed,
1042 unless it was an MBOL and we are not after MBOL,
1043 or a future STCLASS check will fail this. */
1045 /* Even in this situation we may use MBOL flag if strpos is offset
1046 wrt the start of the string. */
1047 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1048 && (strpos != strbeg) && strpos[-1] != '\n'
1049 /* May be due to an implicit anchor of m{.*foo} */
1050 && !(prog->intflags & PREGf_IMPLICIT))
1055 DEBUG_EXECUTE_r( if (ml_anch)
1056 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1057 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1060 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1062 prog->check_utf8 /* Could be deleted already */
1063 && --BmUSEFUL(prog->check_utf8) < 0
1064 && (prog->check_utf8 == prog->float_utf8)
1066 prog->check_substr /* Could be deleted already */
1067 && --BmUSEFUL(prog->check_substr) < 0
1068 && (prog->check_substr == prog->float_substr)
1071 /* If flags & SOMETHING - do not do it many times on the same match */
1072 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1073 /* XXX Does the destruction order has to change with utf8_target? */
1074 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1075 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1076 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1077 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1078 check = NULL; /* abort */
1080 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1081 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1082 if (prog->intflags & PREGf_IMPLICIT)
1083 prog->extflags &= ~RXf_ANCH_MBOL;
1084 /* XXXX This is a remnant of the old implementation. It
1085 looks wasteful, since now INTUIT can use many
1086 other heuristics. */
1087 prog->extflags &= ~RXf_USE_INTUIT;
1088 /* XXXX What other flags might need to be cleared in this branch? */
1094 /* Last resort... */
1095 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1096 /* trie stclasses are too expensive to use here, we are better off to
1097 leave it to regmatch itself */
1098 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1099 /* minlen == 0 is possible if regstclass is \b or \B,
1100 and the fixed substr is ''$.
1101 Since minlen is already taken into account, s+1 is before strend;
1102 accidentally, minlen >= 1 guaranties no false positives at s + 1
1103 even for \b or \B. But (minlen? 1 : 0) below assumes that
1104 regstclass does not come from lookahead... */
1105 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1106 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1107 const U8* const str = (U8*)STRING(progi->regstclass);
1108 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1109 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1112 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1113 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1114 else if (prog->float_substr || prog->float_utf8)
1115 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1119 if (checked_upto < s)
1121 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1122 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1125 s = find_byclass(prog, progi->regstclass, checked_upto, endpos,
1131 const char *what = NULL;
1133 if (endpos == strend) {
1134 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1135 "Could not match STCLASS...\n") );
1138 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1139 "This position contradicts STCLASS...\n") );
1140 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1142 checked_upto = HOPBACKc(endpos, start_shift);
1143 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1144 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1145 /* Contradict one of substrings */
1146 if (prog->anchored_substr || prog->anchored_utf8) {
1147 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1148 DEBUG_EXECUTE_r( what = "anchored" );
1150 s = HOP3c(t, 1, strend);
1151 if (s + start_shift + end_shift > strend) {
1152 /* XXXX Should be taken into account earlier? */
1153 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1154 "Could not match STCLASS...\n") );
1159 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1160 "Looking for %s substr starting at offset %ld...\n",
1161 what, (long)(s + start_shift - i_strpos)) );
1164 /* Have both, check_string is floating */
1165 if (t + start_shift >= check_at) /* Contradicts floating=check */
1166 goto retry_floating_check;
1167 /* Recheck anchored substring, but not floating... */
1171 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1172 "Looking for anchored substr starting at offset %ld...\n",
1173 (long)(other_last - i_strpos)) );
1174 goto do_other_anchored;
1176 /* Another way we could have checked stclass at the
1177 current position only: */
1182 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1183 "Looking for /%s^%s/m starting at offset %ld...\n",
1184 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1187 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1189 /* Check is floating substring. */
1190 retry_floating_check:
1191 t = check_at - start_shift;
1192 DEBUG_EXECUTE_r( what = "floating" );
1193 goto hop_and_restart;
1196 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1197 "By STCLASS: moving %ld --> %ld\n",
1198 (long)(t - i_strpos), (long)(s - i_strpos))
1202 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1203 "Does not contradict STCLASS...\n");
1208 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1209 PL_colors[4], (check ? "Guessed" : "Giving up"),
1210 PL_colors[5], (long)(s - i_strpos)) );
1213 fail_finish: /* Substring not found */
1214 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1215 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1217 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1218 PL_colors[4], PL_colors[5]));
1222 #define DECL_TRIE_TYPE(scan) \
1223 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1224 trie_type = ((scan->flags == EXACT) \
1225 ? (utf8_target ? trie_utf8 : trie_plain) \
1226 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1228 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1229 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1231 switch (trie_type) { \
1232 case trie_utf8_fold: \
1233 if ( foldlen>0 ) { \
1234 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1239 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1240 len = UTF8SKIP(uc); \
1241 skiplen = UNISKIP( uvc ); \
1242 foldlen -= skiplen; \
1243 uscan = foldbuf + skiplen; \
1246 case trie_latin_utf8_fold: \
1247 if ( foldlen>0 ) { \
1248 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1254 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1255 skiplen = UNISKIP( uvc ); \
1256 foldlen -= skiplen; \
1257 uscan = foldbuf + skiplen; \
1261 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1268 charid = trie->charmap[ uvc ]; \
1272 if (widecharmap) { \
1273 SV** const svpp = hv_fetch(widecharmap, \
1274 (char*)&uvc, sizeof(UV), 0); \
1276 charid = (U16)SvIV(*svpp); \
1281 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1285 && (ln == 1 || folder(s, pat_string, ln)) \
1286 && (!reginfo || regtry(reginfo, &s)) ) \
1292 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1294 while (s < strend) { \
1300 #define REXEC_FBC_SCAN(CoDe) \
1302 while (s < strend) { \
1308 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1309 REXEC_FBC_UTF8_SCAN( \
1311 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1320 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1323 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1332 #define REXEC_FBC_TRYIT \
1333 if ((!reginfo || regtry(reginfo, &s))) \
1336 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1337 if (utf8_target) { \
1338 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1341 REXEC_FBC_CLASS_SCAN(CoNd); \
1344 #define DUMP_EXEC_POS(li,s,doutf8) \
1345 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1348 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1349 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1350 tmp = TEST_NON_UTF8(tmp); \
1351 REXEC_FBC_UTF8_SCAN( \
1352 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1361 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1362 if (s == PL_bostr) { \
1366 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1367 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1370 LOAD_UTF8_CHARCLASS_ALNUM(); \
1371 REXEC_FBC_UTF8_SCAN( \
1372 if (tmp == ! (TeSt2_UtF8)) { \
1381 /* The only difference between the BOUND and NBOUND cases is that
1382 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1383 * NBOUND. This is accomplished by passing it in either the if or else clause,
1384 * with the other one being empty */
1385 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1386 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1388 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1389 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1391 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1392 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1394 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1395 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1398 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1399 * be passed in completely with the variable name being tested, which isn't
1400 * such a clean interface, but this is easier to read than it was before. We
1401 * are looking for the boundary (or non-boundary between a word and non-word
1402 * character. The utf8 and non-utf8 cases have the same logic, but the details
1403 * must be different. Find the "wordness" of the character just prior to this
1404 * one, and compare it with the wordness of this one. If they differ, we have
1405 * a boundary. At the beginning of the string, pretend that the previous
1406 * character was a new-line */
1407 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1408 if (utf8_target) { \
1411 else { /* Not utf8 */ \
1412 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1413 tmp = TEST_NON_UTF8(tmp); \
1415 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1424 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1427 /* We know what class REx starts with. Try to find this position... */
1428 /* if reginfo is NULL, its a dryrun */
1429 /* annoyingly all the vars in this routine have different names from their counterparts
1430 in regmatch. /grrr */
1433 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1434 const char *strend, regmatch_info *reginfo, bool is_utf8_pat)
1437 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1438 char *pat_string; /* The pattern's exactish string */
1439 char *pat_end; /* ptr to end char of pat_string */
1440 re_fold_t folder; /* Function for computing non-utf8 folds */
1441 const U8 *fold_array; /* array for folding ords < 256 */
1447 I32 tmp = 1; /* Scratch variable? */
1448 const bool utf8_target = PL_reg_match_utf8;
1449 UV utf8_fold_flags = 0;
1450 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1451 with a result inverts that result, as 0^1 =
1453 _char_class_number classnum;
1455 RXi_GET_DECL(prog,progi);
1457 PERL_ARGS_ASSERT_FIND_BYCLASS;
1459 /* We know what class it must start with. */
1462 case ANYOF_SYNTHETIC:
1463 case ANYOF_WARN_SUPER:
1465 REXEC_FBC_UTF8_CLASS_SCAN(
1466 reginclass(prog, c, (U8*)s, utf8_target));
1469 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1474 if (tmp && (!reginfo || regtry(reginfo, &s)))
1482 if (is_utf8_pat || utf8_target) {
1483 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1484 goto do_exactf_utf8;
1486 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1487 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1488 goto do_exactf_non_utf8; /* isn't dealt with by these */
1493 /* regcomp.c already folded this if pattern is in UTF-8 */
1494 utf8_fold_flags = 0;
1495 goto do_exactf_utf8;
1497 fold_array = PL_fold;
1499 goto do_exactf_non_utf8;
1502 if (is_utf8_pat || utf8_target) {
1503 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1504 goto do_exactf_utf8;
1506 fold_array = PL_fold_locale;
1507 folder = foldEQ_locale;
1508 goto do_exactf_non_utf8;
1512 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1514 goto do_exactf_utf8;
1516 case EXACTFU_TRICKYFOLD:
1518 if (is_utf8_pat || utf8_target) {
1519 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1520 goto do_exactf_utf8;
1523 /* Any 'ss' in the pattern should have been replaced by regcomp,
1524 * so we don't have to worry here about this single special case
1525 * in the Latin1 range */
1526 fold_array = PL_fold_latin1;
1527 folder = foldEQ_latin1;
1531 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1532 are no glitches with fold-length differences
1533 between the target string and pattern */
1535 /* The idea in the non-utf8 EXACTF* cases is to first find the
1536 * first character of the EXACTF* node and then, if necessary,
1537 * case-insensitively compare the full text of the node. c1 is the
1538 * first character. c2 is its fold. This logic will not work for
1539 * Unicode semantics and the german sharp ss, which hence should
1540 * not be compiled into a node that gets here. */
1541 pat_string = STRING(c);
1542 ln = STR_LEN(c); /* length to match in octets/bytes */
1544 /* We know that we have to match at least 'ln' bytes (which is the
1545 * same as characters, since not utf8). If we have to match 3
1546 * characters, and there are only 2 availabe, we know without
1547 * trying that it will fail; so don't start a match past the
1548 * required minimum number from the far end */
1549 e = HOP3c(strend, -((I32)ln), s);
1551 if (!reginfo && e < s) {
1552 e = s; /* Due to minlen logic of intuit() */
1556 c2 = fold_array[c1];
1557 if (c1 == c2) { /* If char and fold are the same */
1558 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1561 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1569 /* If one of the operands is in utf8, we can't use the simpler folding
1570 * above, due to the fact that many different characters can have the
1571 * same fold, or portion of a fold, or different- length fold */
1572 pat_string = STRING(c);
1573 ln = STR_LEN(c); /* length to match in octets/bytes */
1574 pat_end = pat_string + ln;
1575 lnc = is_utf8_pat /* length to match in characters */
1576 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1579 /* We have 'lnc' characters to match in the pattern, but because of
1580 * multi-character folding, each character in the target can match
1581 * up to 3 characters (Unicode guarantees it will never exceed
1582 * this) if it is utf8-encoded; and up to 2 if not (based on the
1583 * fact that the Latin 1 folds are already determined, and the
1584 * only multi-char fold in that range is the sharp-s folding to
1585 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1586 * string character. Adjust lnc accordingly, rounding up, so that
1587 * if we need to match at least 4+1/3 chars, that really is 5. */
1588 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1589 lnc = (lnc + expansion - 1) / expansion;
1591 /* As in the non-UTF8 case, if we have to match 3 characters, and
1592 * only 2 are left, it's guaranteed to fail, so don't start a
1593 * match that would require us to go beyond the end of the string
1595 e = HOP3c(strend, -((I32)lnc), s);
1597 if (!reginfo && e < s) {
1598 e = s; /* Due to minlen logic of intuit() */
1601 /* XXX Note that we could recalculate e to stop the loop earlier,
1602 * as the worst case expansion above will rarely be met, and as we
1603 * go along we would usually find that e moves further to the left.
1604 * This would happen only after we reached the point in the loop
1605 * where if there were no expansion we should fail. Unclear if
1606 * worth the expense */
1609 char *my_strend= (char *)strend;
1610 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1611 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1612 && (!reginfo || regtry(reginfo, &s)) )
1616 s += (utf8_target) ? UTF8SKIP(s) : 1;
1621 RXp_MATCH_TAINTED_on(prog);
1622 FBC_BOUND(isWORDCHAR_LC,
1623 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1624 isWORDCHAR_LC_utf8((U8*)s));
1627 RXp_MATCH_TAINTED_on(prog);
1628 FBC_NBOUND(isWORDCHAR_LC,
1629 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1630 isWORDCHAR_LC_utf8((U8*)s));
1633 FBC_BOUND(isWORDCHAR,
1634 isWORDCHAR_uni(tmp),
1635 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1638 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1640 isWORDCHAR_A((U8*)s));
1643 FBC_NBOUND(isWORDCHAR,
1644 isWORDCHAR_uni(tmp),
1645 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1648 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1650 isWORDCHAR_A((U8*)s));
1653 FBC_BOUND(isWORDCHAR_L1,
1654 isWORDCHAR_uni(tmp),
1655 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1658 FBC_NBOUND(isWORDCHAR_L1,
1659 isWORDCHAR_uni(tmp),
1660 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
1663 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1664 is_LNBREAK_latin1_safe(s, strend)
1668 /* The argument to all the POSIX node types is the class number to pass to
1669 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1676 RXp_MATCH_TAINTED_on(prog);
1677 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1678 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1693 /* The complement of something that matches only ASCII matches all
1694 * UTF-8 variant code points, plus everything in ASCII that isn't
1696 REXEC_FBC_UTF8_CLASS_SCAN(! UTF8_IS_INVARIANT(*s)
1697 || ! _generic_isCC_A(*s, FLAGS(c)));
1706 /* Don't need to worry about utf8, as it can match only a single
1707 * byte invariant character. */
1708 REXEC_FBC_CLASS_SCAN(
1709 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1717 if (! utf8_target) {
1718 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1724 classnum = (_char_class_number) FLAGS(c);
1725 if (classnum < _FIRST_NON_SWASH_CC) {
1726 while (s < strend) {
1728 /* We avoid loading in the swash as long as possible, but
1729 * should we have to, we jump to a separate loop. This
1730 * extra 'if' statement is what keeps this code from being
1731 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1732 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1733 goto found_above_latin1;
1735 if ((UTF8_IS_INVARIANT(*s)
1736 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1738 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1739 && to_complement ^ cBOOL(
1740 _generic_isCC(TWO_BYTE_UTF8_TO_UNI(*s, *(s + 1)),
1743 if (tmp && (!reginfo || regtry(reginfo, &s)))
1755 else switch (classnum) { /* These classes are implemented as
1757 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1758 revert the change of \v matching this */
1761 case _CC_ENUM_PSXSPC:
1762 REXEC_FBC_UTF8_CLASS_SCAN(
1763 to_complement ^ cBOOL(isSPACE_utf8(s)));
1766 case _CC_ENUM_BLANK:
1767 REXEC_FBC_UTF8_CLASS_SCAN(
1768 to_complement ^ cBOOL(isBLANK_utf8(s)));
1771 case _CC_ENUM_XDIGIT:
1772 REXEC_FBC_UTF8_CLASS_SCAN(
1773 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1776 case _CC_ENUM_VERTSPACE:
1777 REXEC_FBC_UTF8_CLASS_SCAN(
1778 to_complement ^ cBOOL(isVERTWS_utf8(s)));
1781 case _CC_ENUM_CNTRL:
1782 REXEC_FBC_UTF8_CLASS_SCAN(
1783 to_complement ^ cBOOL(isCNTRL_utf8(s)));
1787 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
1788 assert(0); /* NOTREACHED */
1793 found_above_latin1: /* Here we have to load a swash to get the result
1794 for the current code point */
1795 if (! PL_utf8_swash_ptrs[classnum]) {
1796 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1797 PL_utf8_swash_ptrs[classnum] =
1798 _core_swash_init("utf8", swash_property_names[classnum],
1799 &PL_sv_undef, 1, 0, NULL, &flags);
1802 /* This is a copy of the loop above for swash classes, though using the
1803 * FBC macro instead of being expanded out. Since we've loaded the
1804 * swash, we don't have to check for that each time through the loop */
1805 REXEC_FBC_UTF8_CLASS_SCAN(
1806 to_complement ^ cBOOL(_generic_utf8(
1809 swash_fetch(PL_utf8_swash_ptrs[classnum],
1817 /* what trie are we using right now */
1818 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1819 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
1820 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1822 const char *last_start = strend - trie->minlen;
1824 const char *real_start = s;
1826 STRLEN maxlen = trie->maxlen;
1828 U8 **points; /* map of where we were in the input string
1829 when reading a given char. For ASCII this
1830 is unnecessary overhead as the relationship
1831 is always 1:1, but for Unicode, especially
1832 case folded Unicode this is not true. */
1833 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1837 GET_RE_DEBUG_FLAGS_DECL;
1839 /* We can't just allocate points here. We need to wrap it in
1840 * an SV so it gets freed properly if there is a croak while
1841 * running the match */
1844 sv_points=newSV(maxlen * sizeof(U8 *));
1845 SvCUR_set(sv_points,
1846 maxlen * sizeof(U8 *));
1847 SvPOK_on(sv_points);
1848 sv_2mortal(sv_points);
1849 points=(U8**)SvPV_nolen(sv_points );
1850 if ( trie_type != trie_utf8_fold
1851 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1854 bitmap=(U8*)trie->bitmap;
1856 bitmap=(U8*)ANYOF_BITMAP(c);
1858 /* this is the Aho-Corasick algorithm modified a touch
1859 to include special handling for long "unknown char" sequences.
1860 The basic idea being that we use AC as long as we are dealing
1861 with a possible matching char, when we encounter an unknown char
1862 (and we have not encountered an accepting state) we scan forward
1863 until we find a legal starting char.
1864 AC matching is basically that of trie matching, except that when
1865 we encounter a failing transition, we fall back to the current
1866 states "fail state", and try the current char again, a process
1867 we repeat until we reach the root state, state 1, or a legal
1868 transition. If we fail on the root state then we can either
1869 terminate if we have reached an accepting state previously, or
1870 restart the entire process from the beginning if we have not.
1873 while (s <= last_start) {
1874 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1882 U8 *uscan = (U8*)NULL;
1883 U8 *leftmost = NULL;
1885 U32 accepted_word= 0;
1889 while ( state && uc <= (U8*)strend ) {
1891 U32 word = aho->states[ state ].wordnum;
1895 DEBUG_TRIE_EXECUTE_r(
1896 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1897 dump_exec_pos( (char *)uc, c, strend, real_start,
1898 (char *)uc, utf8_target );
1899 PerlIO_printf( Perl_debug_log,
1900 " Scanning for legal start char...\n");
1904 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1908 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1914 if (uc >(U8*)last_start) break;
1918 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1919 if (!leftmost || lpos < leftmost) {
1920 DEBUG_r(accepted_word=word);
1926 points[pointpos++ % maxlen]= uc;
1927 if (foldlen || uc < (U8*)strend) {
1928 REXEC_TRIE_READ_CHAR(trie_type, trie,
1930 uscan, len, uvc, charid, foldlen,
1932 DEBUG_TRIE_EXECUTE_r({
1933 dump_exec_pos( (char *)uc, c, strend,
1934 real_start, s, utf8_target);
1935 PerlIO_printf(Perl_debug_log,
1936 " Charid:%3u CP:%4"UVxf" ",
1948 word = aho->states[ state ].wordnum;
1950 base = aho->states[ state ].trans.base;
1952 DEBUG_TRIE_EXECUTE_r({
1954 dump_exec_pos( (char *)uc, c, strend, real_start,
1956 PerlIO_printf( Perl_debug_log,
1957 "%sState: %4"UVxf", word=%"UVxf,
1958 failed ? " Fail transition to " : "",
1959 (UV)state, (UV)word);
1965 ( ((offset = base + charid
1966 - 1 - trie->uniquecharcount)) >= 0)
1967 && ((U32)offset < trie->lasttrans)
1968 && trie->trans[offset].check == state
1969 && (tmp=trie->trans[offset].next))
1971 DEBUG_TRIE_EXECUTE_r(
1972 PerlIO_printf( Perl_debug_log," - legal\n"));
1977 DEBUG_TRIE_EXECUTE_r(
1978 PerlIO_printf( Perl_debug_log," - fail\n"));
1980 state = aho->fail[state];
1984 /* we must be accepting here */
1985 DEBUG_TRIE_EXECUTE_r(
1986 PerlIO_printf( Perl_debug_log," - accepting\n"));
1995 if (!state) state = 1;
1998 if ( aho->states[ state ].wordnum ) {
1999 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2000 if (!leftmost || lpos < leftmost) {
2001 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2006 s = (char*)leftmost;
2007 DEBUG_TRIE_EXECUTE_r({
2009 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2010 (UV)accepted_word, (IV)(s - real_start)
2013 if (!reginfo || regtry(reginfo, &s)) {
2019 DEBUG_TRIE_EXECUTE_r({
2020 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2023 DEBUG_TRIE_EXECUTE_r(
2024 PerlIO_printf( Perl_debug_log,"No match.\n"));
2033 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2043 - regexec_flags - match a regexp against a string
2046 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2047 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2048 /* stringarg: the point in the string at which to begin matching */
2049 /* strend: pointer to null at end of string */
2050 /* strbeg: real beginning of string */
2051 /* minend: end of match must be >= minend bytes after stringarg. */
2052 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2053 * itself is accessed via the pointers above */
2054 /* data: May be used for some additional optimizations.
2055 Currently its only used, with a U32 cast, for transmitting
2056 the ganch offset when doing a /g match. This will change */
2057 /* nosave: For optimizations. */
2061 struct regexp *const prog = ReANY(rx);
2064 char *startpos = stringarg;
2065 I32 minlen; /* must match at least this many chars */
2066 I32 dontbother = 0; /* how many characters not to try at end */
2067 I32 end_shift = 0; /* Same for the end. */ /* CC */
2068 I32 scream_pos = -1; /* Internal iterator of scream. */
2069 char *scream_olds = NULL;
2070 const bool utf8_target = cBOOL(DO_UTF8(sv));
2072 RXi_GET_DECL(prog,progi);
2073 regmatch_info reginfo; /* create some info to pass to regtry etc */
2074 regexp_paren_pair *swap = NULL;
2075 GET_RE_DEBUG_FLAGS_DECL;
2077 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2078 PERL_UNUSED_ARG(data);
2080 /* Be paranoid... */
2081 if (prog == NULL || startpos == NULL) {
2082 Perl_croak(aTHX_ "NULL regexp parameter");
2086 multiline = prog->extflags & RXf_PMf_MULTILINE;
2087 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2089 RX_MATCH_UTF8_set(rx, utf8_target);
2091 debug_start_match(rx, utf8_target, startpos, strend,
2095 minlen = prog->minlen;
2097 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2098 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2099 "String too short [regexec_flags]...\n"));
2104 /* Check validity of program. */
2105 if (UCHARAT(progi->program) != REG_MAGIC) {
2106 Perl_croak(aTHX_ "corrupted regexp program");
2109 RX_MATCH_TAINTED_off(rx);
2110 PL_reg_state.re_state_eval_setup_done = FALSE;
2113 reginfo.is_utf8_pat = cBOOL(RX_UTF8(rx));
2114 reginfo.warned = FALSE;
2115 /* Mark beginning of line for ^ and lookbehind. */
2116 reginfo.bol = startpos; /* XXX not used ??? */
2120 /* Mark end of line for $ (and such) */
2123 /* see how far we have to get to not match where we matched before */
2124 reginfo.till = startpos+minend;
2126 /* If there is a "must appear" string, look for it. */
2129 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2131 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2132 reginfo.ganch = startpos + prog->gofs;
2133 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2134 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2135 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2137 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2138 && mg->mg_len >= 0) {
2139 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2140 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2141 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2143 if (prog->extflags & RXf_ANCH_GPOS) {
2144 if (s > reginfo.ganch)
2146 s = reginfo.ganch - prog->gofs;
2147 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2148 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2154 reginfo.ganch = strbeg + PTR2UV(data);
2155 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2156 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2158 } else { /* pos() not defined */
2159 reginfo.ganch = strbeg;
2160 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2161 "GPOS: reginfo.ganch = strbeg\n"));
2164 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2165 /* We have to be careful. If the previous successful match
2166 was from this regex we don't want a subsequent partially
2167 successful match to clobber the old results.
2168 So when we detect this possibility we add a swap buffer
2169 to the re, and switch the buffer each match. If we fail,
2170 we switch it back; otherwise we leave it swapped.
2173 /* do we need a save destructor here for eval dies? */
2174 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2175 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2176 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2182 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2183 re_scream_pos_data d;
2185 d.scream_olds = &scream_olds;
2186 d.scream_pos = &scream_pos;
2187 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2189 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2190 goto phooey; /* not present */
2196 /* Simplest case: anchored match need be tried only once. */
2197 /* [unless only anchor is BOL and multiline is set] */
2198 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2199 if (s == startpos && regtry(®info, &startpos))
2201 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2202 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2207 dontbother = minlen - 1;
2208 end = HOP3c(strend, -dontbother, strbeg) - 1;
2209 /* for multiline we only have to try after newlines */
2210 if (prog->check_substr || prog->check_utf8) {
2211 /* because of the goto we can not easily reuse the macros for bifurcating the
2212 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2215 goto after_try_utf8;
2217 if (regtry(®info, &s)) {
2224 if (prog->extflags & RXf_USE_INTUIT) {
2225 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2234 } /* end search for check string in unicode */
2236 if (s == startpos) {
2237 goto after_try_latin;
2240 if (regtry(®info, &s)) {
2247 if (prog->extflags & RXf_USE_INTUIT) {
2248 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2257 } /* end search for check string in latin*/
2258 } /* end search for check string */
2259 else { /* search for newline */
2261 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2264 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2265 while (s <= end) { /* note it could be possible to match at the end of the string */
2266 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2267 if (regtry(®info, &s))
2271 } /* end search for newline */
2272 } /* end anchored/multiline check string search */
2274 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2276 /* the warning about reginfo.ganch being used without initialization
2277 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2278 and we only enter this block when the same bit is set. */
2279 char *tmp_s = reginfo.ganch - prog->gofs;
2281 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2286 /* Messy cases: unanchored match. */
2287 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2288 /* we have /x+whatever/ */
2289 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2295 if (! prog->anchored_utf8) {
2296 to_utf8_substr(prog);
2298 ch = SvPVX_const(prog->anchored_utf8)[0];
2301 DEBUG_EXECUTE_r( did_match = 1 );
2302 if (regtry(®info, &s)) goto got_it;
2304 while (s < strend && *s == ch)
2311 if (! prog->anchored_substr) {
2312 if (! to_byte_substr(prog)) {
2313 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2316 ch = SvPVX_const(prog->anchored_substr)[0];
2319 DEBUG_EXECUTE_r( did_match = 1 );
2320 if (regtry(®info, &s)) goto got_it;
2322 while (s < strend && *s == ch)
2327 DEBUG_EXECUTE_r(if (!did_match)
2328 PerlIO_printf(Perl_debug_log,
2329 "Did not find anchored character...\n")
2332 else if (prog->anchored_substr != NULL
2333 || prog->anchored_utf8 != NULL
2334 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2335 && prog->float_max_offset < strend - s)) {
2340 char *last1; /* Last position checked before */
2344 if (prog->anchored_substr || prog->anchored_utf8) {
2346 if (! prog->anchored_utf8) {
2347 to_utf8_substr(prog);
2349 must = prog->anchored_utf8;
2352 if (! prog->anchored_substr) {
2353 if (! to_byte_substr(prog)) {
2354 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2357 must = prog->anchored_substr;
2359 back_max = back_min = prog->anchored_offset;
2362 if (! prog->float_utf8) {
2363 to_utf8_substr(prog);
2365 must = prog->float_utf8;
2368 if (! prog->float_substr) {
2369 if (! to_byte_substr(prog)) {
2370 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2373 must = prog->float_substr;
2375 back_max = prog->float_max_offset;
2376 back_min = prog->float_min_offset;
2382 last = HOP3c(strend, /* Cannot start after this */
2383 -(I32)(CHR_SVLEN(must)
2384 - (SvTAIL(must) != 0) + back_min), strbeg);
2387 last1 = HOPc(s, -1);
2389 last1 = s - 1; /* bogus */
2391 /* XXXX check_substr already used to find "s", can optimize if
2392 check_substr==must. */
2394 dontbother = end_shift;
2395 strend = HOPc(strend, -dontbother);
2396 while ( (s <= last) &&
2397 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2398 (unsigned char*)strend, must,
2399 multiline ? FBMrf_MULTILINE : 0)) ) {
2400 DEBUG_EXECUTE_r( did_match = 1 );
2401 if (HOPc(s, -back_max) > last1) {
2402 last1 = HOPc(s, -back_min);
2403 s = HOPc(s, -back_max);
2406 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2408 last1 = HOPc(s, -back_min);
2412 while (s <= last1) {
2413 if (regtry(®info, &s))
2416 s++; /* to break out of outer loop */
2423 while (s <= last1) {
2424 if (regtry(®info, &s))
2430 DEBUG_EXECUTE_r(if (!did_match) {
2431 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2432 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2433 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2434 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2435 ? "anchored" : "floating"),
2436 quoted, RE_SV_TAIL(must));
2440 else if ( (c = progi->regstclass) ) {
2442 const OPCODE op = OP(progi->regstclass);
2443 /* don't bother with what can't match */
2444 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2445 strend = HOPc(strend, -(minlen - 1));
2448 SV * const prop = sv_newmortal();
2449 regprop(prog, prop, c);
2451 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2453 PerlIO_printf(Perl_debug_log,
2454 "Matching stclass %.*s against %s (%d bytes)\n",
2455 (int)SvCUR(prop), SvPVX_const(prop),
2456 quoted, (int)(strend - s));
2459 if (find_byclass(prog, c, s, strend, ®info, reginfo.is_utf8_pat))
2461 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2465 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2473 if (! prog->float_utf8) {
2474 to_utf8_substr(prog);
2476 float_real = prog->float_utf8;
2479 if (! prog->float_substr) {
2480 if (! to_byte_substr(prog)) {
2481 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2484 float_real = prog->float_substr;
2487 little = SvPV_const(float_real, len);
2488 if (SvTAIL(float_real)) {
2489 /* This means that float_real contains an artificial \n on
2490 * the end due to the presence of something like this:
2491 * /foo$/ where we can match both "foo" and "foo\n" at the
2492 * end of the string. So we have to compare the end of the
2493 * string first against the float_real without the \n and
2494 * then against the full float_real with the string. We
2495 * have to watch out for cases where the string might be
2496 * smaller than the float_real or the float_real without
2498 char *checkpos= strend - len;
2500 PerlIO_printf(Perl_debug_log,
2501 "%sChecking for float_real.%s\n",
2502 PL_colors[4], PL_colors[5]));
2503 if (checkpos + 1 < strbeg) {
2504 /* can't match, even if we remove the trailing \n
2505 * string is too short to match */
2507 PerlIO_printf(Perl_debug_log,
2508 "%sString shorter than required trailing substring, cannot match.%s\n",
2509 PL_colors[4], PL_colors[5]));
2511 } else if (memEQ(checkpos + 1, little, len - 1)) {
2512 /* can match, the end of the string matches without the
2514 last = checkpos + 1;
2515 } else if (checkpos < strbeg) {
2516 /* cant match, string is too short when the "\n" is
2519 PerlIO_printf(Perl_debug_log,
2520 "%sString does not contain required trailing substring, cannot match.%s\n",
2521 PL_colors[4], PL_colors[5]));
2523 } else if (!multiline) {
2524 /* non multiline match, so compare with the "\n" at the
2525 * end of the string */
2526 if (memEQ(checkpos, little, len)) {
2530 PerlIO_printf(Perl_debug_log,
2531 "%sString does not contain required trailing substring, cannot match.%s\n",
2532 PL_colors[4], PL_colors[5]));
2536 /* multiline match, so we have to search for a place
2537 * where the full string is located */
2543 last = rninstr(s, strend, little, little + len);
2545 last = strend; /* matching "$" */
2548 /* at one point this block contained a comment which was
2549 * probably incorrect, which said that this was a "should not
2550 * happen" case. Even if it was true when it was written I am
2551 * pretty sure it is not anymore, so I have removed the comment
2552 * and replaced it with this one. Yves */
2554 PerlIO_printf(Perl_debug_log,
2555 "String does not contain required substring, cannot match.\n"
2559 dontbother = strend - last + prog->float_min_offset;
2561 if (minlen && (dontbother < minlen))
2562 dontbother = minlen - 1;
2563 strend -= dontbother; /* this one's always in bytes! */
2564 /* We don't know much -- general case. */
2567 if (regtry(®info, &s))
2576 if (regtry(®info, &s))
2578 } while (s++ < strend);
2588 PerlIO_printf(Perl_debug_log,
2589 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2596 if (PL_reg_state.re_state_eval_setup_done)
2597 restore_pos(aTHX_ prog);
2598 if (RXp_PAREN_NAMES(prog))
2599 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2601 /* make sure $`, $&, $', and $digit will work later */
2602 if ( !(flags & REXEC_NOT_FIRST) ) {
2603 if (flags & REXEC_COPY_STR) {
2607 PerlIO_printf(Perl_debug_log,
2608 "Copy on write: regexp capture, type %d\n",
2611 RX_MATCH_COPY_FREE(rx);
2612 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2613 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2614 assert (SvPOKp(prog->saved_copy));
2615 prog->sublen = PL_regeol - strbeg;
2616 prog->suboffset = 0;
2617 prog->subcoffset = 0;
2622 I32 max = PL_regeol - strbeg;
2625 if ( (flags & REXEC_COPY_SKIP_POST)
2626 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2627 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2628 ) { /* don't copy $' part of string */
2631 /* calculate the right-most part of the string covered
2632 * by a capture. Due to look-ahead, this may be to
2633 * the right of $&, so we have to scan all captures */
2634 while (n <= prog->lastparen) {
2635 if (prog->offs[n].end > max)
2636 max = prog->offs[n].end;
2640 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2641 ? prog->offs[0].start
2643 assert(max >= 0 && max <= PL_regeol - strbeg);
2646 if ( (flags & REXEC_COPY_SKIP_PRE)
2647 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2648 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2649 ) { /* don't copy $` part of string */
2652 /* calculate the left-most part of the string covered
2653 * by a capture. Due to look-behind, this may be to
2654 * the left of $&, so we have to scan all captures */
2655 while (min && n <= prog->lastparen) {
2656 if ( prog->offs[n].start != -1
2657 && prog->offs[n].start < min)
2659 min = prog->offs[n].start;
2663 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2664 && min > prog->offs[0].end
2666 min = prog->offs[0].end;
2670 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2673 if (RX_MATCH_COPIED(rx)) {
2674 if (sublen > prog->sublen)
2676 (char*)saferealloc(prog->subbeg, sublen+1);
2679 prog->subbeg = (char*)safemalloc(sublen+1);
2680 Copy(strbeg + min, prog->subbeg, sublen, char);
2681 prog->subbeg[sublen] = '\0';
2682 prog->suboffset = min;
2683 prog->sublen = sublen;
2684 RX_MATCH_COPIED_on(rx);
2686 prog->subcoffset = prog->suboffset;
2687 if (prog->suboffset && utf8_target) {
2688 /* Convert byte offset to chars.
2689 * XXX ideally should only compute this if @-/@+
2690 * has been seen, a la PL_sawampersand ??? */
2692 /* If there's a direct correspondence between the
2693 * string which we're matching and the original SV,
2694 * then we can use the utf8 len cache associated with
2695 * the SV. In particular, it means that under //g,
2696 * sv_pos_b2u() will use the previously cached
2697 * position to speed up working out the new length of
2698 * subcoffset, rather than counting from the start of
2699 * the string each time. This stops
2700 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2701 * from going quadratic */
2702 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2703 sv_pos_b2u(sv, &(prog->subcoffset));
2705 prog->subcoffset = utf8_length((U8*)strbeg,
2706 (U8*)(strbeg+prog->suboffset));
2710 RX_MATCH_COPY_FREE(rx);
2711 prog->subbeg = strbeg;
2712 prog->suboffset = 0;
2713 prog->subcoffset = 0;
2714 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2721 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2722 PL_colors[4], PL_colors[5]));
2723 if (PL_reg_state.re_state_eval_setup_done)
2724 restore_pos(aTHX_ prog);
2726 /* we failed :-( roll it back */
2727 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2728 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2733 Safefree(prog->offs);
2740 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2741 * Do inc before dec, in case old and new rex are the same */
2742 #define SET_reg_curpm(Re2) \
2743 if (PL_reg_state.re_state_eval_setup_done) { \
2744 (void)ReREFCNT_inc(Re2); \
2745 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2746 PM_SETRE((PL_reg_curpm), (Re2)); \
2751 - regtry - try match at specific point
2753 STATIC I32 /* 0 failure, 1 success */
2754 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2758 REGEXP *const rx = reginfo->prog;
2759 regexp *const prog = ReANY(rx);
2761 RXi_GET_DECL(prog,progi);
2762 GET_RE_DEBUG_FLAGS_DECL;
2764 PERL_ARGS_ASSERT_REGTRY;
2766 reginfo->cutpoint=NULL;
2768 if ((prog->extflags & RXf_EVAL_SEEN)
2769 && !PL_reg_state.re_state_eval_setup_done)
2773 PL_reg_state.re_state_eval_setup_done = TRUE;
2775 /* Make $_ available to executed code. */
2776 if (reginfo->sv != DEFSV) {
2778 DEFSV_set(reginfo->sv);
2781 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2782 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2783 /* prepare for quick setting of pos */
2784 #ifdef PERL_OLD_COPY_ON_WRITE
2785 if (SvIsCOW(reginfo->sv))
2786 sv_force_normal_flags(reginfo->sv, 0);
2788 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2789 &PL_vtbl_mglob, NULL, 0);
2793 PL_reg_oldpos = mg->mg_len;
2794 SAVEDESTRUCTOR_X(restore_pos, prog);
2796 if (!PL_reg_curpm) {
2797 Newxz(PL_reg_curpm, 1, PMOP);
2800 SV* const repointer = &PL_sv_undef;
2801 /* this regexp is also owned by the new PL_reg_curpm, which
2802 will try to free it. */
2803 av_push(PL_regex_padav, repointer);
2804 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2805 PL_regex_pad = AvARRAY(PL_regex_padav);
2810 PL_reg_oldcurpm = PL_curpm;
2811 PL_curpm = PL_reg_curpm;
2812 if (RXp_MATCH_COPIED(prog)) {
2813 /* Here is a serious problem: we cannot rewrite subbeg,
2814 since it may be needed if this match fails. Thus
2815 $` inside (?{}) could fail... */
2816 PL_reg_oldsaved = prog->subbeg;
2817 PL_reg_oldsavedlen = prog->sublen;
2818 PL_reg_oldsavedoffset = prog->suboffset;
2819 PL_reg_oldsavedcoffset = prog->suboffset;
2821 PL_nrs = prog->saved_copy;
2823 RXp_MATCH_COPIED_off(prog);
2826 PL_reg_oldsaved = NULL;
2827 prog->subbeg = PL_bostr;
2828 prog->suboffset = 0;
2829 prog->subcoffset = 0;
2830 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2833 PL_reg_starttry = *startposp;
2835 prog->offs[0].start = *startposp - PL_bostr;
2836 prog->lastparen = 0;
2837 prog->lastcloseparen = 0;
2839 /* XXXX What this code is doing here?!!! There should be no need
2840 to do this again and again, prog->lastparen should take care of
2843 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2844 * Actually, the code in regcppop() (which Ilya may be meaning by
2845 * prog->lastparen), is not needed at all by the test suite
2846 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2847 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2848 * Meanwhile, this code *is* needed for the
2849 * above-mentioned test suite tests to succeed. The common theme
2850 * on those tests seems to be returning null fields from matches.
2851 * --jhi updated by dapm */
2853 if (prog->nparens) {
2854 regexp_paren_pair *pp = prog->offs;
2856 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2864 result = regmatch(reginfo, *startposp, progi->program + 1);
2866 prog->offs[0].end = result;
2869 if (reginfo->cutpoint)
2870 *startposp= reginfo->cutpoint;
2871 REGCP_UNWIND(lastcp);
2876 #define sayYES goto yes
2877 #define sayNO goto no
2878 #define sayNO_SILENT goto no_silent
2880 /* we dont use STMT_START/END here because it leads to
2881 "unreachable code" warnings, which are bogus, but distracting. */
2882 #define CACHEsayNO \
2883 if (ST.cache_mask) \
2884 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2887 /* this is used to determine how far from the left messages like
2888 'failed...' are printed. It should be set such that messages
2889 are inline with the regop output that created them.
2891 #define REPORT_CODE_OFF 32
2894 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2895 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2896 #define CHRTEST_NOT_A_CP_1 -999
2897 #define CHRTEST_NOT_A_CP_2 -998
2899 #define SLAB_FIRST(s) (&(s)->states[0])
2900 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2902 /* grab a new slab and return the first slot in it */
2904 STATIC regmatch_state *
2907 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2910 regmatch_slab *s = PL_regmatch_slab->next;
2912 Newx(s, 1, regmatch_slab);
2913 s->prev = PL_regmatch_slab;
2915 PL_regmatch_slab->next = s;
2917 PL_regmatch_slab = s;
2918 return SLAB_FIRST(s);
2922 /* push a new state then goto it */
2924 #define PUSH_STATE_GOTO(state, node, input) \
2925 pushinput = input; \
2927 st->resume_state = state; \
2930 /* push a new state with success backtracking, then goto it */
2932 #define PUSH_YES_STATE_GOTO(state, node, input) \
2933 pushinput = input; \
2935 st->resume_state = state; \
2936 goto push_yes_state;
2943 regmatch() - main matching routine
2945 This is basically one big switch statement in a loop. We execute an op,
2946 set 'next' to point the next op, and continue. If we come to a point which
2947 we may need to backtrack to on failure such as (A|B|C), we push a
2948 backtrack state onto the backtrack stack. On failure, we pop the top
2949 state, and re-enter the loop at the state indicated. If there are no more
2950 states to pop, we return failure.
2952 Sometimes we also need to backtrack on success; for example /A+/, where
2953 after successfully matching one A, we need to go back and try to
2954 match another one; similarly for lookahead assertions: if the assertion
2955 completes successfully, we backtrack to the state just before the assertion
2956 and then carry on. In these cases, the pushed state is marked as
2957 'backtrack on success too'. This marking is in fact done by a chain of
2958 pointers, each pointing to the previous 'yes' state. On success, we pop to
2959 the nearest yes state, discarding any intermediate failure-only states.
2960 Sometimes a yes state is pushed just to force some cleanup code to be
2961 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2962 it to free the inner regex.
2964 Note that failure backtracking rewinds the cursor position, while
2965 success backtracking leaves it alone.
2967 A pattern is complete when the END op is executed, while a subpattern
2968 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2969 ops trigger the "pop to last yes state if any, otherwise return true"
2972 A common convention in this function is to use A and B to refer to the two
2973 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2974 the subpattern to be matched possibly multiple times, while B is the entire
2975 rest of the pattern. Variable and state names reflect this convention.
2977 The states in the main switch are the union of ops and failure/success of
2978 substates associated with with that op. For example, IFMATCH is the op
2979 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2980 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2981 successfully matched A and IFMATCH_A_fail is a state saying that we have
2982 just failed to match A. Resume states always come in pairs. The backtrack
2983 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2984 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2985 on success or failure.
2987 The struct that holds a backtracking state is actually a big union, with
2988 one variant for each major type of op. The variable st points to the
2989 top-most backtrack struct. To make the code clearer, within each
2990 block of code we #define ST to alias the relevant union.
2992 Here's a concrete example of a (vastly oversimplified) IFMATCH
2998 #define ST st->u.ifmatch
3000 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3001 ST.foo = ...; // some state we wish to save
3003 // push a yes backtrack state with a resume value of
3004 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3006 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3009 case IFMATCH_A: // we have successfully executed A; now continue with B
3011 bar = ST.foo; // do something with the preserved value
3014 case IFMATCH_A_fail: // A failed, so the assertion failed
3015 ...; // do some housekeeping, then ...
3016 sayNO; // propagate the failure
3023 For any old-timers reading this who are familiar with the old recursive
3024 approach, the code above is equivalent to:
3026 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3035 ...; // do some housekeeping, then ...
3036 sayNO; // propagate the failure
3039 The topmost backtrack state, pointed to by st, is usually free. If you
3040 want to claim it, populate any ST.foo fields in it with values you wish to
3041 save, then do one of
3043 PUSH_STATE_GOTO(resume_state, node, newinput);
3044 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3046 which sets that backtrack state's resume value to 'resume_state', pushes a
3047 new free entry to the top of the backtrack stack, then goes to 'node'.
3048 On backtracking, the free slot is popped, and the saved state becomes the
3049 new free state. An ST.foo field in this new top state can be temporarily
3050 accessed to retrieve values, but once the main loop is re-entered, it
3051 becomes available for reuse.
3053 Note that the depth of the backtrack stack constantly increases during the
3054 left-to-right execution of the pattern, rather than going up and down with
3055 the pattern nesting. For example the stack is at its maximum at Z at the
3056 end of the pattern, rather than at X in the following:
3058 /(((X)+)+)+....(Y)+....Z/
3060 The only exceptions to this are lookahead/behind assertions and the cut,
3061 (?>A), which pop all the backtrack states associated with A before
3064 Backtrack state structs are allocated in slabs of about 4K in size.
3065 PL_regmatch_state and st always point to the currently active state,
3066 and PL_regmatch_slab points to the slab currently containing
3067 PL_regmatch_state. The first time regmatch() is called, the first slab is
3068 allocated, and is never freed until interpreter destruction. When the slab
3069 is full, a new one is allocated and chained to the end. At exit from
3070 regmatch(), slabs allocated since entry are freed.
3075 #define DEBUG_STATE_pp(pp) \
3077 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3078 PerlIO_printf(Perl_debug_log, \
3079 " %*s"pp" %s%s%s%s%s\n", \
3081 PL_reg_name[st->resume_state], \
3082 ((st==yes_state||st==mark_state) ? "[" : ""), \
3083 ((st==yes_state) ? "Y" : ""), \
3084 ((st==mark_state) ? "M" : ""), \
3085 ((st==yes_state||st==mark_state) ? "]" : "") \
3090 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3095 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3096 const char *start, const char *end, const char *blurb)
3098 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3100 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3105 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3106 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3108 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3109 start, end - start, 60);
3111 PerlIO_printf(Perl_debug_log,
3112 "%s%s REx%s %s against %s\n",
3113 PL_colors[4], blurb, PL_colors[5], s0, s1);
3115 if (utf8_target||utf8_pat)
3116 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3117 utf8_pat ? "pattern" : "",
3118 utf8_pat && utf8_target ? " and " : "",
3119 utf8_target ? "string" : ""
3125 S_dump_exec_pos(pTHX_ const char *locinput,
3126 const regnode *scan,
3127 const char *loc_regeol,
3128 const char *loc_bostr,
3129 const char *loc_reg_starttry,
3130 const bool utf8_target)
3132 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3133 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3134 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3135 /* The part of the string before starttry has one color
3136 (pref0_len chars), between starttry and current
3137 position another one (pref_len - pref0_len chars),
3138 after the current position the third one.
3139 We assume that pref0_len <= pref_len, otherwise we
3140 decrease pref0_len. */
3141 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3142 ? (5 + taill) - l : locinput - loc_bostr;
3145 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3147 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3149 pref0_len = pref_len - (locinput - loc_reg_starttry);
3150 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3151 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3152 ? (5 + taill) - pref_len : loc_regeol - locinput);
3153 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3157 if (pref0_len > pref_len)
3158 pref0_len = pref_len;
3160 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3162 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3163 (locinput - pref_len),pref0_len, 60, 4, 5);
3165 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3166 (locinput - pref_len + pref0_len),
3167 pref_len - pref0_len, 60, 2, 3);
3169 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3170 locinput, loc_regeol - locinput, 10, 0, 1);
3172 const STRLEN tlen=len0+len1+len2;
3173 PerlIO_printf(Perl_debug_log,
3174 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3175 (IV)(locinput - loc_bostr),
3178 (docolor ? "" : "> <"),
3180 (int)(tlen > 19 ? 0 : 19 - tlen),
3187 /* reg_check_named_buff_matched()
3188 * Checks to see if a named buffer has matched. The data array of
3189 * buffer numbers corresponding to the buffer is expected to reside
3190 * in the regexp->data->data array in the slot stored in the ARG() of
3191 * node involved. Note that this routine doesn't actually care about the
3192 * name, that information is not preserved from compilation to execution.
3193 * Returns the index of the leftmost defined buffer with the given name
3194 * or 0 if non of the buffers matched.
3197 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3200 RXi_GET_DECL(rex,rexi);
3201 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3202 I32 *nums=(I32*)SvPVX(sv_dat);
3204 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3206 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3207 if ((I32)rex->lastparen >= nums[n] &&
3208 rex->offs[nums[n]].end != -1)
3217 /* free all slabs above current one - called during LEAVE_SCOPE */
3220 S_clear_backtrack_stack(pTHX_ void *p)
3222 regmatch_slab *s = PL_regmatch_slab->next;
3227 PL_regmatch_slab->next = NULL;
3229 regmatch_slab * const osl = s;
3235 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3236 U8* c1_utf8, int *c2p, U8* c2_utf8, bool is_utf8_pat)
3238 /* This function determines if there are one or two characters that match
3239 * the first character of the passed-in EXACTish node <text_node>, and if
3240 * so, returns them in the passed-in pointers.
3242 * If it determines that no possible character in the target string can
3243 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3244 * the first character in <text_node> requires UTF-8 to represent, and the
3245 * target string isn't in UTF-8.)
3247 * If there are more than two characters that could match the beginning of
3248 * <text_node>, or if more context is required to determine a match or not,
3249 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3251 * The motiviation behind this function is to allow the caller to set up
3252 * tight loops for matching. If <text_node> is of type EXACT, there is
3253 * only one possible character that can match its first character, and so
3254 * the situation is quite simple. But things get much more complicated if
3255 * folding is involved. It may be that the first character of an EXACTFish
3256 * node doesn't participate in any possible fold, e.g., punctuation, so it
3257 * can be matched only by itself. The vast majority of characters that are
3258 * in folds match just two things, their lower and upper-case equivalents.
3259 * But not all are like that; some have multiple possible matches, or match
3260 * sequences of more than one character. This function sorts all that out.
3262 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3263 * loop of trying to match A*, we know we can't exit where the thing
3264 * following it isn't a B. And something can't be a B unless it is the
3265 * beginning of B. By putting a quick test for that beginning in a tight
3266 * loop, we can rule out things that can't possibly be B without having to
3267 * break out of the loop, thus avoiding work. Similarly, if A is a single
3268 * character, we can make a tight loop matching A*, using the outputs of
3271 * If the target string to match isn't in UTF-8, and there aren't
3272 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3273 * the one or two possible octets (which are characters in this situation)
3274 * that can match. In all cases, if there is only one character that can
3275 * match, *<c1p> and *<c2p> will be identical.
3277 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3278 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3279 * can match the beginning of <text_node>. They should be declared with at
3280 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3281 * undefined what these contain.) If one or both of the buffers are
3282 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3283 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3284 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3285 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3286 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3288 const bool utf8_target = PL_reg_match_utf8;
3290 UV c1 = CHRTEST_NOT_A_CP_1;
3291 UV c2 = CHRTEST_NOT_A_CP_2;
3292 bool use_chrtest_void = FALSE;
3294 /* Used when we have both utf8 input and utf8 output, to avoid converting
3295 * to/from code points */
3296 bool utf8_has_been_setup = FALSE;
3300 U8 *pat = (U8*)STRING(text_node);
3302 if (OP(text_node) == EXACT) {
3304 /* In an exact node, only one thing can be matched, that first
3305 * character. If both the pat and the target are UTF-8, we can just
3306 * copy the input to the output, avoiding finding the code point of
3311 else if (utf8_target) {
3312 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3313 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3314 utf8_has_been_setup = TRUE;
3317 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3320 else /* an EXACTFish node */
3322 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3323 pat + STR_LEN(text_node)))
3325 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3326 pat + STR_LEN(text_node))))
3328 /* Multi-character folds require more context to sort out. Also
3329 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3330 * handled outside this routine */
3331 use_chrtest_void = TRUE;
3333 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3334 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3336 /* Load the folds hash, if not already done */
3338 if (! PL_utf8_foldclosures) {
3339 if (! PL_utf8_tofold) {
3340 U8 dummy[UTF8_MAXBYTES+1];
3342 /* Force loading this by folding an above-Latin1 char */
3343 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3344 assert(PL_utf8_tofold); /* Verify that worked */
3346 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3349 /* The fold closures data structure is a hash with the keys being
3350 * the UTF-8 of every character that is folded to, like 'k', and
3351 * the values each an array of all code points that fold to its
3352 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3354 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3359 /* Not found in the hash, therefore there are no folds
3360 * containing it, so there is only a single character that
3364 else { /* Does participate in folds */
3365 AV* list = (AV*) *listp;
3366 if (av_len(list) != 1) {
3368 /* If there aren't exactly two folds to this, it is outside
3369 * the scope of this function */
3370 use_chrtest_void = TRUE;
3372 else { /* There are two. Get them */
3373 SV** c_p = av_fetch(list, 0, FALSE);
3375 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3379 c_p = av_fetch(list, 1, FALSE);
3381 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3385 /* Folds that cross the 255/256 boundary are forbidden if
3386 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3387 * pattern character is above 256, and its only other match
3388 * is below 256, the only legal match will be to itself.
3389 * We have thrown away the original, so have to compute
3390 * which is the one above 255 */
3391 if ((c1 < 256) != (c2 < 256)) {
3392 if (OP(text_node) == EXACTFL
3393 || (OP(text_node) == EXACTFA
3394 && (isASCII(c1) || isASCII(c2))))
3407 else /* Here, c1 is < 255 */
3409 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3410 && OP(text_node) != EXACTFL
3411 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3413 /* Here, there could be something above Latin1 in the target which
3414 * folds to this character in the pattern. All such cases except
3415 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3416 * involved in their folds, so are outside the scope of this
3418 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3419 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3422 use_chrtest_void = TRUE;
3425 else { /* Here nothing above Latin1 can fold to the pattern character */
3426 switch (OP(text_node)) {
3428 case EXACTFL: /* /l rules */
3429 c2 = PL_fold_locale[c1];
3433 if (! utf8_target) { /* /d rules */
3438 /* /u rules for all these. This happens to work for
3439 * EXACTFA as nothing in Latin1 folds to ASCII */
3441 case EXACTFU_TRICKYFOLD:
3444 c2 = PL_fold_latin1[c1];
3448 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3449 assert(0); /* NOTREACHED */
3454 /* Here have figured things out. Set up the returns */
3455 if (use_chrtest_void) {
3456 *c2p = *c1p = CHRTEST_VOID;
3458 else if (utf8_target) {
3459 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3460 uvchr_to_utf8(c1_utf8, c1);
3461 uvchr_to_utf8(c2_utf8, c2);
3464 /* Invariants are stored in both the utf8 and byte outputs; Use
3465 * negative numbers otherwise for the byte ones. Make sure that the
3466 * byte ones are the same iff the utf8 ones are the same */
3467 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3468 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3471 ? CHRTEST_NOT_A_CP_1
3472 : CHRTEST_NOT_A_CP_2;
3474 else if (c1 > 255) {
3475 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3480 *c1p = *c2p = c2; /* c2 is the only representable value */
3482 else { /* c1 is representable; see about c2 */
3484 *c2p = (c2 < 256) ? c2 : c1;
3490 /* returns -1 on failure, $+[0] on success */
3492 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3494 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3498 const bool utf8_target = PL_reg_match_utf8;
3499 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3500 REGEXP *rex_sv = reginfo->prog;
3501 regexp *rex = ReANY(rex_sv);
3502 RXi_GET_DECL(rex,rexi);
3504 /* the current state. This is a cached copy of PL_regmatch_state */
3506 /* cache heavy used fields of st in registers */
3509 U32 n = 0; /* general value; init to avoid compiler warning */
3510 I32 ln = 0; /* len or last; init to avoid compiler warning */
3511 char *locinput = startpos;
3512 char *pushinput; /* where to continue after a PUSH */
3513 I32 nextchr; /* is always set to UCHARAT(locinput) */
3515 bool result = 0; /* return value of S_regmatch */
3516 int depth = 0; /* depth of backtrack stack */
3517 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3518 const U32 max_nochange_depth =
3519 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3520 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3521 regmatch_state *yes_state = NULL; /* state to pop to on success of
3523 /* mark_state piggy backs on the yes_state logic so that when we unwind
3524 the stack on success we can update the mark_state as we go */
3525 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3526 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3527 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3529 bool no_final = 0; /* prevent failure from backtracking? */
3530 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3531 char *startpoint = locinput;
3532 SV *popmark = NULL; /* are we looking for a mark? */
3533 SV *sv_commit = NULL; /* last mark name seen in failure */
3534 SV *sv_yes_mark = NULL; /* last mark name we have seen
3535 during a successful match */
3536 U32 lastopen = 0; /* last open we saw */
3537 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3538 SV* const oreplsv = GvSV(PL_replgv);
3539 /* these three flags are set by various ops to signal information to
3540 * the very next op. They have a useful lifetime of exactly one loop
3541 * iteration, and are not preserved or restored by state pushes/pops
3543 bool sw = 0; /* the condition value in (?(cond)a|b) */
3544 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3545 int logical = 0; /* the following EVAL is:
3549 or the following IFMATCH/UNLESSM is:
3550 false: plain (?=foo)
3551 true: used as a condition: (?(?=foo))
3553 PAD* last_pad = NULL;
3555 I32 gimme = G_SCALAR;
3556 CV *caller_cv = NULL; /* who called us */
3557 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3558 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3559 U32 maxopenparen = 0; /* max '(' index seen so far */
3560 int to_complement; /* Invert the result? */
3561 _char_class_number classnum;
3562 bool is_utf8_pat = reginfo->is_utf8_pat;
3565 GET_RE_DEBUG_FLAGS_DECL;
3568 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3569 multicall_oldcatch = 0;
3570 multicall_cv = NULL;
3572 PERL_UNUSED_VAR(multicall_cop);
3573 PERL_UNUSED_VAR(newsp);
3576 PERL_ARGS_ASSERT_REGMATCH;
3578 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3579 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3581 /* on first ever call to regmatch, allocate first slab */
3582 if (!PL_regmatch_slab) {
3583 Newx(PL_regmatch_slab, 1, regmatch_slab);
3584 PL_regmatch_slab->prev = NULL;
3585 PL_regmatch_slab->next = NULL;
3586 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3589 oldsave = PL_savestack_ix;
3590 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3591 SAVEVPTR(PL_regmatch_slab);
3592 SAVEVPTR(PL_regmatch_state);
3594 /* grab next free state slot */
3595 st = ++PL_regmatch_state;
3596 if (st > SLAB_LAST(PL_regmatch_slab))
3597 st = PL_regmatch_state = S_push_slab(aTHX);
3599 /* Note that nextchr is a byte even in UTF */
3602 while (scan != NULL) {
3605 SV * const prop = sv_newmortal();
3606 regnode *rnext=regnext(scan);
3607 DUMP_EXEC_POS( locinput, scan, utf8_target );
3608 regprop(rex, prop, scan);
3610 PerlIO_printf(Perl_debug_log,
3611 "%3"IVdf":%*s%s(%"IVdf")\n",
3612 (IV)(scan - rexi->program), depth*2, "",
3614 (PL_regkind[OP(scan)] == END || !rnext) ?
3615 0 : (IV)(rnext - rexi->program));
3618 next = scan + NEXT_OFF(scan);
3621 state_num = OP(scan);
3627 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3629 switch (state_num) {
3630 case BOL: /* /^../ */
3631 if (locinput == PL_bostr)
3633 /* reginfo->till = reginfo->bol; */
3638 case MBOL: /* /^../m */
3639 if (locinput == PL_bostr ||
3640 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3646 case SBOL: /* /^../s */
3647 if (locinput == PL_bostr)
3652 if (locinput == reginfo->ganch)
3656 case KEEPS: /* \K */
3657 /* update the startpoint */
3658 st->u.keeper.val = rex->offs[0].start;
3659 rex->offs[0].start = locinput - PL_bostr;
3660 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3661 assert(0); /*NOTREACHED*/
3662 case KEEPS_next_fail:
3663 /* rollback the start point change */
3664 rex->offs[0].start = st->u.keeper.val;
3666 assert(0); /*NOTREACHED*/
3668 case EOL: /* /..$/ */
3671 case MEOL: /* /..$/m */
3672 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3676 case SEOL: /* /..$/s */
3678 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3680 if (PL_regeol - locinput > 1)
3685 if (!NEXTCHR_IS_EOS)
3689 case SANY: /* /./s */
3692 goto increment_locinput;
3700 case REG_ANY: /* /./ */
3701 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3703 goto increment_locinput;
3707 #define ST st->u.trie
3708 case TRIEC: /* (ab|cd) with known charclass */
3709 /* In this case the charclass data is available inline so
3710 we can fail fast without a lot of extra overhead.
3712 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3714 PerlIO_printf(Perl_debug_log,
3715 "%*s %sfailed to match trie start class...%s\n",
3716 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3719 assert(0); /* NOTREACHED */
3722 case TRIE: /* (ab|cd) */
3723 /* the basic plan of execution of the trie is:
3724 * At the beginning, run though all the states, and
3725 * find the longest-matching word. Also remember the position
3726 * of the shortest matching word. For example, this pattern:
3729 * when matched against the string "abcde", will generate
3730 * accept states for all words except 3, with the longest
3731 * matching word being 4, and the shortest being 2 (with
3732 * the position being after char 1 of the string).
3734 * Then for each matching word, in word order (i.e. 1,2,4,5),
3735 * we run the remainder of the pattern; on each try setting
3736 * the current position to the character following the word,
3737 * returning to try the next word on failure.
3739 * We avoid having to build a list of words at runtime by
3740 * using a compile-time structure, wordinfo[].prev, which
3741 * gives, for each word, the previous accepting word (if any).
3742 * In the case above it would contain the mappings 1->2, 2->0,
3743 * 3->0, 4->5, 5->1. We can use this table to generate, from
3744 * the longest word (4 above), a list of all words, by
3745 * following the list of prev pointers; this gives us the
3746 * unordered list 4,5,1,2. Then given the current word we have
3747 * just tried, we can go through the list and find the
3748 * next-biggest word to try (so if we just failed on word 2,
3749 * the next in the list is 4).
3751 * Since at runtime we don't record the matching position in
3752 * the string for each word, we have to work that out for
3753 * each word we're about to process. The wordinfo table holds
3754 * the character length of each word; given that we recorded
3755 * at the start: the position of the shortest word and its
3756 * length in chars, we just need to move the pointer the
3757 * difference between the two char lengths. Depending on
3758 * Unicode status and folding, that's cheap or expensive.
3760 * This algorithm is optimised for the case where are only a
3761 * small number of accept states, i.e. 0,1, or maybe 2.
3762 * With lots of accepts states, and having to try all of them,
3763 * it becomes quadratic on number of accept states to find all
3768 /* what type of TRIE am I? (utf8 makes this contextual) */
3769 DECL_TRIE_TYPE(scan);
3771 /* what trie are we using right now */
3772 reg_trie_data * const trie
3773 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3774 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3775 U32 state = trie->startstate;
3778 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3780 if (trie->states[ state ].wordnum) {
3782 PerlIO_printf(Perl_debug_log,
3783 "%*s %smatched empty string...%s\n",
3784 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3790 PerlIO_printf(Perl_debug_log,
3791 "%*s %sfailed to match trie start class...%s\n",
3792 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3799 U8 *uc = ( U8* )locinput;
3803 U8 *uscan = (U8*)NULL;
3804 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3805 U32 charcount = 0; /* how many input chars we have matched */
3806 U32 accepted = 0; /* have we seen any accepting states? */
3808 ST.jump = trie->jump;
3811 ST.longfold = FALSE; /* char longer if folded => it's harder */
3814 /* fully traverse the TRIE; note the position of the
3815 shortest accept state and the wordnum of the longest
3818 while ( state && uc <= (U8*)PL_regeol ) {
3819 U32 base = trie->states[ state ].trans.base;
3823 wordnum = trie->states[ state ].wordnum;
3825 if (wordnum) { /* it's an accept state */
3828 /* record first match position */
3830 ST.firstpos = (U8*)locinput;
3835 ST.firstchars = charcount;
3838 if (!ST.nextword || wordnum < ST.nextword)
3839 ST.nextword = wordnum;
3840 ST.topword = wordnum;
3843 DEBUG_TRIE_EXECUTE_r({
3844 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3845 PerlIO_printf( Perl_debug_log,
3846 "%*s %sState: %4"UVxf" Accepted: %c ",
3847 2+depth * 2, "", PL_colors[4],
3848 (UV)state, (accepted ? 'Y' : 'N'));
3851 /* read a char and goto next state */
3852 if ( base && (foldlen || uc < (U8*)PL_regeol)) {
3854 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3855 uscan, len, uvc, charid, foldlen,
3862 base + charid - 1 - trie->uniquecharcount)) >= 0)
3864 && ((U32)offset < trie->lasttrans)
3865 && trie->trans[offset].check == state)
3867 state = trie->trans[offset].next;
3878 DEBUG_TRIE_EXECUTE_r(
3879 PerlIO_printf( Perl_debug_log,
3880 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3881 charid, uvc, (UV)state, PL_colors[5] );
3887 /* calculate total number of accept states */
3892 w = trie->wordinfo[w].prev;
3895 ST.accepted = accepted;
3899 PerlIO_printf( Perl_debug_log,
3900 "%*s %sgot %"IVdf" possible matches%s\n",
3901 REPORT_CODE_OFF + depth * 2, "",
3902 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3904 goto trie_first_try; /* jump into the fail handler */
3906 assert(0); /* NOTREACHED */
3908 case TRIE_next_fail: /* we failed - try next alternative */
3912 REGCP_UNWIND(ST.cp);
3913 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3915 if (!--ST.accepted) {
3917 PerlIO_printf( Perl_debug_log,
3918 "%*s %sTRIE failed...%s\n",
3919 REPORT_CODE_OFF+depth*2, "",
3926 /* Find next-highest word to process. Note that this code
3927 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3930 U16 const nextword = ST.nextword;
3931 reg_trie_wordinfo * const wordinfo
3932 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3933 for (word=ST.topword; word; word=wordinfo[word].prev) {
3934 if (word > nextword && (!min || word < min))
3947 ST.lastparen = rex->lastparen;
3948 ST.lastcloseparen = rex->lastcloseparen;
3952 /* find start char of end of current word */
3954 U32 chars; /* how many chars to skip */
3955 reg_trie_data * const trie
3956 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3958 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3960 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3965 /* the hard option - fold each char in turn and find
3966 * its folded length (which may be different */
3967 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3975 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3983 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3988 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
4004 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4005 ? ST.jump[ST.nextword]
4009 PerlIO_printf( Perl_debug_log,
4010 "%*s %sTRIE matched word #%d, continuing%s\n",
4011 REPORT_CODE_OFF+depth*2, "",
4018 if (ST.accepted > 1 || has_cutgroup) {
4019 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4020 assert(0); /* NOTREACHED */
4022 /* only one choice left - just continue */
4024 AV *const trie_words
4025 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4026 SV ** const tmp = av_fetch( trie_words,
4028 SV *sv= tmp ? sv_newmortal() : NULL;
4030 PerlIO_printf( Perl_debug_log,
4031 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4032 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4034 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4035 PL_colors[0], PL_colors[1],
4036 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4038 : "not compiled under -Dr",
4042 locinput = (char*)uc;
4043 continue; /* execute rest of RE */
4044 assert(0); /* NOTREACHED */
4048 case EXACT: { /* /abc/ */
4049 char *s = STRING(scan);
4051 if (utf8_target != is_utf8_pat) {
4052 /* The target and the pattern have differing utf8ness. */
4054 const char * const e = s + ln;
4057 /* The target is utf8, the pattern is not utf8.
4058 * Above-Latin1 code points can't match the pattern;
4059 * invariants match exactly, and the other Latin1 ones need
4060 * to be downgraded to a single byte in order to do the
4061 * comparison. (If we could be confident that the target
4062 * is not malformed, this could be refactored to have fewer
4063 * tests by just assuming that if the first bytes match, it
4064 * is an invariant, but there are tests in the test suite
4065 * dealing with (??{...}) which violate this) */
4067 if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) l)) {
4070 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4077 if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
4086 /* The target is not utf8, the pattern is utf8. */
4088 if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4092 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4099 if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
4110 /* The target and the pattern have the same utf8ness. */
4111 /* Inline the first character, for speed. */
4112 if (PL_regeol - locinput < ln
4113 || UCHARAT(s) != nextchr
4114 || (ln > 1 && memNE(s, locinput, ln)))
4123 case EXACTFL: { /* /abc/il */
4125 const U8 * fold_array;
4127 U32 fold_utf8_flags;
4129 RX_MATCH_TAINTED_on(reginfo->prog);
4130 folder = foldEQ_locale;
4131 fold_array = PL_fold_locale;
4132 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
4135 case EXACTFU_SS: /* /\x{df}/iu */
4136 case EXACTFU_TRICKYFOLD: /* /\x{390}/iu */
4137 case EXACTFU: /* /abc/iu */
4138 folder = foldEQ_latin1;
4139 fold_array = PL_fold_latin1;
4140 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4143 case EXACTFA: /* /abc/iaa */
4144 folder = foldEQ_latin1;
4145 fold_array = PL_fold_latin1;
4146 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4149 case EXACTF: /* /abc/i */
4151 fold_array = PL_fold;
4152 fold_utf8_flags = 0;
4158 if (utf8_target || is_utf8_pat || state_num == EXACTFU_SS) {
4159 /* Either target or the pattern are utf8, or has the issue where
4160 * the fold lengths may differ. */
4161 const char * const l = locinput;
4162 char *e = PL_regeol;
4164 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4165 l, &e, 0, utf8_target, fold_utf8_flags))
4173 /* Neither the target nor the pattern are utf8 */
4174 if (UCHARAT(s) != nextchr
4176 && UCHARAT(s) != fold_array[nextchr])
4180 if (PL_regeol - locinput < ln)
4182 if (ln > 1 && ! folder(s, locinput, ln))
4188 /* XXX Could improve efficiency by separating these all out using a
4189 * macro or in-line function. At that point regcomp.c would no longer
4190 * have to set the FLAGS fields of these */
4191 case BOUNDL: /* /\b/l */
4192 case NBOUNDL: /* /\B/l */
4193 RX_MATCH_TAINTED_on(reginfo->prog);
4195 case BOUND: /* /\b/ */
4196 case BOUNDU: /* /\b/u */
4197 case BOUNDA: /* /\b/a */
4198 case NBOUND: /* /\B/ */
4199 case NBOUNDU: /* /\B/u */
4200 case NBOUNDA: /* /\B/a */
4201 /* was last char in word? */
4203 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4204 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4206 if (locinput == PL_bostr)
4209 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
4211 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
4213 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4214 ln = isWORDCHAR_uni(ln);
4218 LOAD_UTF8_CHARCLASS_ALNUM();
4219 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4224 ln = isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(ln));
4225 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4230 /* Here the string isn't utf8, or is utf8 and only ascii
4231 * characters are to match \w. In the latter case looking at
4232 * the byte just prior to the current one may be just the final
4233 * byte of a multi-byte character. This is ok. There are two
4235 * 1) it is a single byte character, and then the test is doing
4236 * just what it's supposed to.
4237 * 2) it is a multi-byte character, in which case the final
4238 * byte is never mistakable for ASCII, and so the test
4239 * will say it is not a word character, which is the
4240 * correct answer. */
4241 ln = (locinput != PL_bostr) ?
4242 UCHARAT(locinput - 1) : '\n';
4243 switch (FLAGS(scan)) {
4244 case REGEX_UNICODE_CHARSET:
4245 ln = isWORDCHAR_L1(ln);
4246 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4248 case REGEX_LOCALE_CHARSET:
4249 ln = isWORDCHAR_LC(ln);
4250 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4252 case REGEX_DEPENDS_CHARSET:
4253 ln = isWORDCHAR(ln);
4254 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4256 case REGEX_ASCII_RESTRICTED_CHARSET:
4257 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4258 ln = isWORDCHAR_A(ln);
4259 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4262 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4266 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4268 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4272 case ANYOF: /* /[abc]/ */
4273 case ANYOF_WARN_SUPER:
4277 if (!reginclass(rex, scan, (U8*)locinput, utf8_target))
4279 locinput += UTF8SKIP(locinput);
4282 if (!REGINCLASS(rex, scan, (U8*)locinput))
4288 /* The argument (FLAGS) to all the POSIX node types is the class number
4291 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4295 case POSIXL: /* \w or [:punct:] etc. under /l */
4299 /* The locale hasn't influenced the outcome before this, so defer
4300 * tainting until now */
4301 RX_MATCH_TAINTED_on(reginfo->prog);
4303 /* Use isFOO_lc() for characters within Latin1. (Note that
4304 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4305 * wouldn't be invariant) */
4306 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4307 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), nextchr)))) {
4311 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4312 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4313 TWO_BYTE_UTF8_TO_UNI(nextchr,
4314 *(locinput + 1))))))
4319 else { /* Here, must be an above Latin-1 code point */
4320 goto utf8_posix_not_eos;
4323 /* Here, must be utf8 */
4324 locinput += UTF8SKIP(locinput);
4327 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4331 case POSIXD: /* \w or [:punct:] etc. under /d */
4337 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4339 if (NEXTCHR_IS_EOS) {
4343 /* All UTF-8 variants match */
4344 if (! UTF8_IS_INVARIANT(nextchr)) {
4345 goto increment_locinput;
4351 case POSIXA: /* \w or [:punct:] etc. under /a */
4354 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4355 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4356 * character is a single byte */
4359 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4365 /* Here we are either not in utf8, or we matched a utf8-invariant,
4366 * so the next char is the next byte */
4370 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4374 case POSIXU: /* \w or [:punct:] etc. under /u */
4376 if (NEXTCHR_IS_EOS) {
4381 /* Use _generic_isCC() for characters within Latin1. (Note that
4382 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4383 * wouldn't be invariant) */
4384 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4385 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4392 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4393 if (! (to_complement
4394 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(nextchr,
4402 else { /* Handle above Latin-1 code points */
4403 classnum = (_char_class_number) FLAGS(scan);
4404 if (classnum < _FIRST_NON_SWASH_CC) {
4406 /* Here, uses a swash to find such code points. Load if if
4407 * not done already */
4408 if (! PL_utf8_swash_ptrs[classnum]) {
4409 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4410 PL_utf8_swash_ptrs[classnum]
4411 = _core_swash_init("utf8",
4412 swash_property_names[classnum],
4413 &PL_sv_undef, 1, 0, NULL, &flags);
4415 if (! (to_complement
4416 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4417 (U8 *) locinput, TRUE))))
4422 else { /* Here, uses macros to find above Latin-1 code points */
4424 case _CC_ENUM_SPACE: /* XXX would require separate
4425 code if we revert the change
4426 of \v matching this */
4427 case _CC_ENUM_PSXSPC:
4428 if (! (to_complement
4429 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4434 case _CC_ENUM_BLANK:
4435 if (! (to_complement
4436 ^ cBOOL(is_HORIZWS_high(locinput))))
4441 case _CC_ENUM_XDIGIT:
4442 if (! (to_complement
4443 ^ cBOOL(is_XDIGIT_high(locinput))))
4448 case _CC_ENUM_VERTSPACE:
4449 if (! (to_complement
4450 ^ cBOOL(is_VERTWS_high(locinput))))
4455 default: /* The rest, e.g. [:cntrl:], can't match
4457 if (! to_complement) {
4463 locinput += UTF8SKIP(locinput);
4467 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4468 a Unicode extended Grapheme Cluster */
4469 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4470 extended Grapheme Cluster is:
4473 | Prepend* Begin Extend*
4476 Begin is: ( Special_Begin | ! Control )
4477 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4478 Extend is: ( Grapheme_Extend | Spacing_Mark )
4479 Control is: [ GCB_Control | CR | LF ]
4480 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4482 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4485 Begin is ( Regular_Begin + Special Begin )
4487 It turns out that 98.4% of all Unicode code points match
4488 Regular_Begin. Doing it this way eliminates a table match in
4489 the previous implementation for almost all Unicode code points.
4491 There is a subtlety with Prepend* which showed up in testing.
4492 Note that the Begin, and only the Begin is required in:
4493 | Prepend* Begin Extend*
4494 Also, Begin contains '! Control'. A Prepend must be a
4495 '! Control', which means it must also be a Begin. What it
4496 comes down to is that if we match Prepend* and then find no
4497 suitable Begin afterwards, that if we backtrack the last
4498 Prepend, that one will be a suitable Begin.
4503 if (! utf8_target) {
4505 /* Match either CR LF or '.', as all the other possibilities
4507 locinput++; /* Match the . or CR */
4508 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4510 && locinput < PL_regeol
4511 && UCHARAT(locinput) == '\n')
4518 /* Utf8: See if is ( CR LF ); already know that locinput <
4519 * PL_regeol, so locinput+1 is in bounds */
4520 if ( nextchr == '\r' && locinput+1 < PL_regeol
4521 && UCHARAT(locinput + 1) == '\n')
4528 /* In case have to backtrack to beginning, then match '.' */
4529 char *starting = locinput;
4531 /* In case have to backtrack the last prepend */
4532 char *previous_prepend = NULL;
4534 LOAD_UTF8_CHARCLASS_GCB();
4536 /* Match (prepend)* */
4537 while (locinput < PL_regeol
4538 && (len = is_GCB_Prepend_utf8(locinput)))
4540 previous_prepend = locinput;
4544 /* As noted above, if we matched a prepend character, but
4545 * the next thing won't match, back off the last prepend we
4546 * matched, as it is guaranteed to match the begin */
4547 if (previous_prepend
4548 && (locinput >= PL_regeol
4549 || (! swash_fetch(PL_utf8_X_regular_begin,
4550 (U8*)locinput, utf8_target)
4551 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4554 locinput = previous_prepend;
4557 /* Note that here we know PL_regeol > locinput, as we
4558 * tested that upon input to this switch case, and if we
4559 * moved locinput forward, we tested the result just above
4560 * and it either passed, or we backed off so that it will
4562 if (swash_fetch(PL_utf8_X_regular_begin,
4563 (U8*)locinput, utf8_target)) {
4564 locinput += UTF8SKIP(locinput);
4566 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4568 /* Here did not match the required 'Begin' in the
4569 * second term. So just match the very first
4570 * character, the '.' of the final term of the regex */
4571 locinput = starting + UTF8SKIP(starting);
4575 /* Here is a special begin. It can be composed of
4576 * several individual characters. One possibility is
4578 if ((len = is_GCB_RI_utf8(locinput))) {
4580 while (locinput < PL_regeol
4581 && (len = is_GCB_RI_utf8(locinput)))
4585 } else if ((len = is_GCB_T_utf8(locinput))) {
4586 /* Another possibility is T+ */
4588 while (locinput < PL_regeol
4589 && (len = is_GCB_T_utf8(locinput)))
4595 /* Here, neither RI+ nor T+; must be some other
4596 * Hangul. That means it is one of the others: L,
4597 * LV, LVT or V, and matches:
4598 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4601 while (locinput < PL_regeol
4602 && (len = is_GCB_L_utf8(locinput)))
4607 /* Here, have exhausted L*. If the next character
4608 * is not an LV, LVT nor V, it means we had to have
4609 * at least one L, so matches L+ in the original
4610 * equation, we have a complete hangul syllable.
4613 if (locinput < PL_regeol
4614 && is_GCB_LV_LVT_V_utf8(locinput))
4616 /* Otherwise keep going. Must be LV, LVT or V.
4617 * See if LVT, by first ruling out V, then LV */
4618 if (! is_GCB_V_utf8(locinput)
4619 /* All but every TCount one is LV */
4620 && (valid_utf8_to_uvchr((U8 *) locinput,
4625 locinput += UTF8SKIP(locinput);
4628 /* Must be V or LV. Take it, then match
4630 locinput += UTF8SKIP(locinput);
4631 while (locinput < PL_regeol
4632 && (len = is_GCB_V_utf8(locinput)))
4638 /* And any of LV, LVT, or V can be followed
4640 while (locinput < PL_regeol
4641 && (len = is_GCB_T_utf8(locinput)))
4649 /* Match any extender */
4650 while (locinput < PL_regeol
4651 && swash_fetch(PL_utf8_X_extend,
4652 (U8*)locinput, utf8_target))
4654 locinput += UTF8SKIP(locinput);
4658 if (locinput > PL_regeol) sayNO;
4662 case NREFFL: /* /\g{name}/il */
4663 { /* The capture buffer cases. The ones beginning with N for the
4664 named buffers just convert to the equivalent numbered and
4665 pretend they were called as the corresponding numbered buffer
4667 /* don't initialize these in the declaration, it makes C++
4672 const U8 *fold_array;
4675 RX_MATCH_TAINTED_on(reginfo->prog);
4676 folder = foldEQ_locale;
4677 fold_array = PL_fold_locale;
4679 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4682 case NREFFA: /* /\g{name}/iaa */
4683 folder = foldEQ_latin1;
4684 fold_array = PL_fold_latin1;
4686 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4689 case NREFFU: /* /\g{name}/iu */
4690 folder = foldEQ_latin1;
4691 fold_array = PL_fold_latin1;
4693 utf8_fold_flags = 0;
4696 case NREFF: /* /\g{name}/i */
4698 fold_array = PL_fold;
4700 utf8_fold_flags = 0;
4703 case NREF: /* /\g{name}/ */
4707 utf8_fold_flags = 0;
4710 /* For the named back references, find the corresponding buffer
4712 n = reg_check_named_buff_matched(rex,scan);
4717 goto do_nref_ref_common;
4719 case REFFL: /* /\1/il */
4720 RX_MATCH_TAINTED_on(reginfo->prog);
4721 folder = foldEQ_locale;
4722 fold_array = PL_fold_locale;
4723 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4726 case REFFA: /* /\1/iaa */
4727 folder = foldEQ_latin1;
4728 fold_array = PL_fold_latin1;
4729 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4732 case REFFU: /* /\1/iu */
4733 folder = foldEQ_latin1;
4734 fold_array = PL_fold_latin1;
4735 utf8_fold_flags = 0;
4738 case REFF: /* /\1/i */
4740 fold_array = PL_fold;
4741 utf8_fold_flags = 0;
4744 case REF: /* /\1/ */
4747 utf8_fold_flags = 0;
4751 n = ARG(scan); /* which paren pair */
4754 ln = rex->offs[n].start;
4755 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4756 if (rex->lastparen < n || ln == -1)
4757 sayNO; /* Do not match unless seen CLOSEn. */
4758 if (ln == rex->offs[n].end)
4762 if (type != REF /* REF can do byte comparison */
4763 && (utf8_target || type == REFFU))
4764 { /* XXX handle REFFL better */
4765 char * limit = PL_regeol;
4767 /* This call case insensitively compares the entire buffer
4768 * at s, with the current input starting at locinput, but
4769 * not going off the end given by PL_regeol, and returns in
4770 * <limit> upon success, how much of the current input was
4772 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4773 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4781 /* Not utf8: Inline the first character, for speed. */
4782 if (!NEXTCHR_IS_EOS &&
4783 UCHARAT(s) != nextchr &&
4785 UCHARAT(s) != fold_array[nextchr]))
4787 ln = rex->offs[n].end - ln;
4788 if (locinput + ln > PL_regeol)
4790 if (ln > 1 && (type == REF
4791 ? memNE(s, locinput, ln)
4792 : ! folder(s, locinput, ln)))
4798 case NOTHING: /* null op; e.g. the 'nothing' following
4799 * the '*' in m{(a+|b)*}' */
4801 case TAIL: /* placeholder while compiling (A|B|C) */
4804 case BACK: /* ??? doesn't appear to be used ??? */
4808 #define ST st->u.eval
4813 regexp_internal *rei;
4814 regnode *startpoint;
4816 case GOSTART: /* (?R) */
4817 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4818 if (cur_eval && cur_eval->locinput==locinput) {
4819 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4820 Perl_croak(aTHX_ "Infinite recursion in regex");
4821 if ( ++nochange_depth > max_nochange_depth )
4823 "Pattern subroutine nesting without pos change"
4824 " exceeded limit in regex");
4831 if (OP(scan)==GOSUB) {
4832 startpoint = scan + ARG2L(scan);
4833 ST.close_paren = ARG(scan);
4835 startpoint = rei->program+1;
4838 goto eval_recurse_doit;
4839 assert(0); /* NOTREACHED */
4841 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4842 if (cur_eval && cur_eval->locinput==locinput) {
4843 if ( ++nochange_depth > max_nochange_depth )
4844 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4849 /* execute the code in the {...} */
4853 OP * const oop = PL_op;
4854 COP * const ocurcop = PL_curcop;
4856 char *saved_regeol = PL_regeol;
4857 struct re_save_state saved_state;
4860 /* save *all* paren positions */
4861 regcppush(rex, 0, maxopenparen);
4862 REGCP_SET(runops_cp);
4864 /* To not corrupt the existing regex state while executing the
4865 * eval we would normally put it on the save stack, like with
4866 * save_re_context. However, re-evals have a weird scoping so we
4867 * can't just add ENTER/LEAVE here. With that, things like
4869 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4871 * would break, as they expect the localisation to be unwound
4872 * only when the re-engine backtracks through the bit that
4875 * What we do instead is just saving the state in a local c
4878 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4880 PL_reg_state.re_reparsing = FALSE;
4883 caller_cv = find_runcv(NULL);
4887 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4889 (REGEXP*)(rexi->data->data[n])
4892 nop = (OP*)rexi->data->data[n+1];
4894 else if (rexi->data->what[n] == 'l') { /* literal code */
4896 nop = (OP*)rexi->data->data[n];
4897 assert(CvDEPTH(newcv));
4900 /* literal with own CV */
4901 assert(rexi->data->what[n] == 'L');
4902 newcv = rex->qr_anoncv;
4903 nop = (OP*)rexi->data->data[n];
4906 /* normally if we're about to execute code from the same
4907 * CV that we used previously, we just use the existing
4908 * CX stack entry. However, its possible that in the
4909 * meantime we may have backtracked, popped from the save
4910 * stack, and undone the SAVECOMPPAD(s) associated with
4911 * PUSH_MULTICALL; in which case PL_comppad no longer
4912 * points to newcv's pad. */
4913 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4915 I32 depth = (newcv == caller_cv) ? 0 : 1;
4916 if (last_pushed_cv) {
4917 CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
4920 PUSH_MULTICALL_WITHDEPTH(newcv, depth);
4922 last_pushed_cv = newcv;
4925 /* these assignments are just to silence compiler
4927 multicall_cop = NULL;
4930 last_pad = PL_comppad;
4932 /* the initial nextstate you would normally execute
4933 * at the start of an eval (which would cause error
4934 * messages to come from the eval), may be optimised
4935 * away from the execution path in the regex code blocks;
4936 * so manually set PL_curcop to it initially */
4938 OP *o = cUNOPx(nop)->op_first;
4939 assert(o->op_type == OP_NULL);
4940 if (o->op_targ == OP_SCOPE) {
4941 o = cUNOPo->op_first;
4944 assert(o->op_targ == OP_LEAVE);
4945 o = cUNOPo->op_first;
4946 assert(o->op_type == OP_ENTER);
4950 if (o->op_type != OP_STUB) {
4951 assert( o->op_type == OP_NEXTSTATE
4952 || o->op_type == OP_DBSTATE
4953 || (o->op_type == OP_NULL
4954 && ( o->op_targ == OP_NEXTSTATE
4955 || o->op_targ == OP_DBSTATE
4959 PL_curcop = (COP*)o;
4964 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4965 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4967 rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4970 SV *sv_mrk = get_sv("REGMARK", 1);
4971 sv_setsv(sv_mrk, sv_yes_mark);
4974 /* we don't use MULTICALL here as we want to call the
4975 * first op of the block of interest, rather than the
4976 * first op of the sub */
4977 before = (IV)(SP-PL_stack_base);
4979 CALLRUNOPS(aTHX); /* Scalar context. */
4981 if ((IV)(SP-PL_stack_base) == before)
4982 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4988 /* before restoring everything, evaluate the returned
4989 * value, so that 'uninit' warnings don't use the wrong
4990 * PL_op or pad. Also need to process any magic vars
4991 * (e.g. $1) *before* parentheses are restored */
4996 if (logical == 0) /* (?{})/ */
4997 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4998 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4999 sw = cBOOL(SvTRUE(ret));
5002 else { /* /(??{}) */
5003 /* if its overloaded, let the regex compiler handle
5004 * it; otherwise extract regex, or stringify */
5005 if (!SvAMAGIC(ret)) {
5009 if (SvTYPE(sv) == SVt_REGEXP)
5010 re_sv = (REGEXP*) sv;
5011 else if (SvSMAGICAL(sv)) {
5012 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
5014 re_sv = (REGEXP *) mg->mg_obj;
5017 /* force any magic, undef warnings here */
5019 ret = sv_mortalcopy(ret);
5020 (void) SvPV_force_nolen(ret);
5026 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
5028 /* *** Note that at this point we don't restore
5029 * PL_comppad, (or pop the CxSUB) on the assumption it may
5030 * be used again soon. This is safe as long as nothing
5031 * in the regexp code uses the pad ! */
5033 PL_curcop = ocurcop;
5034 PL_regeol = saved_regeol;
5035 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5041 /* only /(??{})/ from now on */
5044 /* extract RE object from returned value; compiling if
5048 re_sv = reg_temp_copy(NULL, re_sv);
5053 if (SvUTF8(ret) && IN_BYTES) {
5054 /* In use 'bytes': make a copy of the octet
5055 * sequence, but without the flag on */
5057 const char *const p = SvPV(ret, len);
5058 ret = newSVpvn_flags(p, len, SVs_TEMP);
5060 if (rex->intflags & PREGf_USE_RE_EVAL)
5061 pm_flags |= PMf_USE_RE_EVAL;
5063 /* if we got here, it should be an engine which
5064 * supports compiling code blocks and stuff */
5065 assert(rex->engine && rex->engine->op_comp);
5066 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5067 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5068 rex->engine, NULL, NULL,
5069 /* copy /msix etc to inner pattern */
5074 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
5076 /* This isn't a first class regexp. Instead, it's
5077 caching a regexp onto an existing, Perl visible
5079 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5081 /* safe to do now that any $1 etc has been
5082 * interpolated into the new pattern string and
5084 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5089 RXp_MATCH_COPIED_off(re);
5090 re->subbeg = rex->subbeg;
5091 re->sublen = rex->sublen;
5092 re->suboffset = rex->suboffset;
5093 re->subcoffset = rex->subcoffset;
5096 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
5097 "Matching embedded");
5099 startpoint = rei->program + 1;
5100 ST.close_paren = 0; /* only used for GOSUB */
5102 eval_recurse_doit: /* Share code with GOSUB below this line */
5103 /* run the pattern returned from (??{...}) */
5105 /* Save *all* the positions. */
5106 ST.cp = regcppush(rex, 0, maxopenparen);
5107 REGCP_SET(ST.lastcp);
5110 re->lastcloseparen = 0;
5114 /* XXXX This is too dramatic a measure... */
5117 ST.saved_utf8_pat = is_utf8_pat;
5118 is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5120 ST.prev_rex = rex_sv;
5121 ST.prev_curlyx = cur_curlyx;
5123 SET_reg_curpm(rex_sv);
5128 ST.prev_eval = cur_eval;
5130 /* now continue from first node in postoned RE */
5131 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5132 assert(0); /* NOTREACHED */
5135 case EVAL_AB: /* cleanup after a successful (??{A})B */
5136 /* note: this is called twice; first after popping B, then A */
5137 is_utf8_pat = ST.saved_utf8_pat;
5138 rex_sv = ST.prev_rex;
5139 SET_reg_curpm(rex_sv);
5140 rex = ReANY(rex_sv);
5141 rexi = RXi_GET(rex);
5143 cur_eval = ST.prev_eval;
5144 cur_curlyx = ST.prev_curlyx;
5146 /* XXXX This is too dramatic a measure... */
5148 if ( nochange_depth )
5153 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5154 /* note: this is called twice; first after popping B, then A */
5155 is_utf8_pat = ST.saved_utf8_pat;
5156 rex_sv = ST.prev_rex;
5157 SET_reg_curpm(rex_sv);
5158 rex = ReANY(rex_sv);
5159 rexi = RXi_GET(rex);
5161 REGCP_UNWIND(ST.lastcp);
5162 regcppop(rex, &maxopenparen);
5163 cur_eval = ST.prev_eval;
5164 cur_curlyx = ST.prev_curlyx;
5165 /* XXXX This is too dramatic a measure... */
5167 if ( nochange_depth )
5173 n = ARG(scan); /* which paren pair */
5174 rex->offs[n].start_tmp = locinput - PL_bostr;
5175 if (n > maxopenparen)
5177 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5178 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5182 (IV)rex->offs[n].start_tmp,
5188 /* XXX really need to log other places start/end are set too */
5189 #define CLOSE_CAPTURE \
5190 rex->offs[n].start = rex->offs[n].start_tmp; \
5191 rex->offs[n].end = locinput - PL_bostr; \
5192 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5193 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5195 PTR2UV(rex->offs), \
5197 (IV)rex->offs[n].start, \
5198 (IV)rex->offs[n].end \
5202 n = ARG(scan); /* which paren pair */
5204 if (n > rex->lastparen)
5206 rex->lastcloseparen = n;
5207 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5212 case ACCEPT: /* (*ACCEPT) */
5216 cursor && OP(cursor)!=END;
5217 cursor=regnext(cursor))
5219 if ( OP(cursor)==CLOSE ){
5221 if ( n <= lastopen ) {
5223 if (n > rex->lastparen)
5225 rex->lastcloseparen = n;
5226 if ( n == ARG(scan) || (cur_eval &&
5227 cur_eval->u.eval.close_paren == n))
5236 case GROUPP: /* (?(1)) */
5237 n = ARG(scan); /* which paren pair */
5238 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5241 case NGROUPP: /* (?(<name>)) */
5242 /* reg_check_named_buff_matched returns 0 for no match */
5243 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5246 case INSUBP: /* (?(R)) */
5248 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5251 case DEFINEP: /* (?(DEFINE)) */
5255 case IFTHEN: /* (?(cond)A|B) */
5256 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
5258 next = NEXTOPER(NEXTOPER(scan));
5260 next = scan + ARG(scan);
5261 if (OP(next) == IFTHEN) /* Fake one. */
5262 next = NEXTOPER(NEXTOPER(next));
5266 case LOGICAL: /* modifier for EVAL and IFMATCH */
5267 logical = scan->flags;
5270 /*******************************************************************
5272 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5273 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5274 STAR/PLUS/CURLY/CURLYN are used instead.)
5276 A*B is compiled as <CURLYX><A><WHILEM><B>
5278 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5279 state, which contains the current count, initialised to -1. It also sets
5280 cur_curlyx to point to this state, with any previous value saved in the
5283 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5284 since the pattern may possibly match zero times (i.e. it's a while {} loop
5285 rather than a do {} while loop).
5287 Each entry to WHILEM represents a successful match of A. The count in the
5288 CURLYX block is incremented, another WHILEM state is pushed, and execution
5289 passes to A or B depending on greediness and the current count.
5291 For example, if matching against the string a1a2a3b (where the aN are
5292 substrings that match /A/), then the match progresses as follows: (the
5293 pushed states are interspersed with the bits of strings matched so far):
5296 <CURLYX cnt=0><WHILEM>
5297 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5298 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5299 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5300 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5302 (Contrast this with something like CURLYM, which maintains only a single
5306 a1 <CURLYM cnt=1> a2
5307 a1 a2 <CURLYM cnt=2> a3
5308 a1 a2 a3 <CURLYM cnt=3> b
5311 Each WHILEM state block marks a point to backtrack to upon partial failure
5312 of A or B, and also contains some minor state data related to that
5313 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5314 overall state, such as the count, and pointers to the A and B ops.
5316 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5317 must always point to the *current* CURLYX block, the rules are:
5319 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5320 and set cur_curlyx to point the new block.
5322 When popping the CURLYX block after a successful or unsuccessful match,
5323 restore the previous cur_curlyx.
5325 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5326 to the outer one saved in the CURLYX block.
5328 When popping the WHILEM block after a successful or unsuccessful B match,
5329 restore the previous cur_curlyx.
5331 Here's an example for the pattern (AI* BI)*BO
5332 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5335 curlyx backtrack stack
5336 ------ ---------------
5338 CO <CO prev=NULL> <WO>
5339 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5340 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5341 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5343 At this point the pattern succeeds, and we work back down the stack to
5344 clean up, restoring as we go:
5346 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5347 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5348 CO <CO prev=NULL> <WO>
5351 *******************************************************************/
5353 #define ST st->u.curlyx
5355 case CURLYX: /* start of /A*B/ (for complex A) */
5357 /* No need to save/restore up to this paren */
5358 I32 parenfloor = scan->flags;
5360 assert(next); /* keep Coverity happy */
5361 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5364 /* XXXX Probably it is better to teach regpush to support
5365 parenfloor > maxopenparen ... */
5366 if (parenfloor > (I32)rex->lastparen)
5367 parenfloor = rex->lastparen; /* Pessimization... */
5369 ST.prev_curlyx= cur_curlyx;
5371 ST.cp = PL_savestack_ix;
5373 /* these fields contain the state of the current curly.
5374 * they are accessed by subsequent WHILEMs */
5375 ST.parenfloor = parenfloor;
5380 ST.count = -1; /* this will be updated by WHILEM */
5381 ST.lastloc = NULL; /* this will be updated by WHILEM */
5383 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5384 assert(0); /* NOTREACHED */
5387 case CURLYX_end: /* just finished matching all of A*B */
5388 cur_curlyx = ST.prev_curlyx;
5390 assert(0); /* NOTREACHED */
5392 case CURLYX_end_fail: /* just failed to match all of A*B */
5394 cur_curlyx = ST.prev_curlyx;
5396 assert(0); /* NOTREACHED */
5400 #define ST st->u.whilem
5402 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5404 /* see the discussion above about CURLYX/WHILEM */
5406 int min = ARG1(cur_curlyx->u.curlyx.me);
5407 int max = ARG2(cur_curlyx->u.curlyx.me);
5408 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5410 assert(cur_curlyx); /* keep Coverity happy */
5411 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5412 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5413 ST.cache_offset = 0;
5417 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5418 "%*s whilem: matched %ld out of %d..%d\n",
5419 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5422 /* First just match a string of min A's. */
5425 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5427 cur_curlyx->u.curlyx.lastloc = locinput;
5428 REGCP_SET(ST.lastcp);
5430 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5431 assert(0); /* NOTREACHED */
5434 /* If degenerate A matches "", assume A done. */
5436 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5437 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5438 "%*s whilem: empty match detected, trying continuation...\n",
5439 REPORT_CODE_OFF+depth*2, "")
5441 goto do_whilem_B_max;
5444 /* super-linear cache processing */
5448 if (!PL_reg_maxiter) {
5449 /* start the countdown: Postpone detection until we
5450 * know the match is not *that* much linear. */
5451 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
5452 /* possible overflow for long strings and many CURLYX's */
5453 if (PL_reg_maxiter < 0)
5454 PL_reg_maxiter = I32_MAX;
5455 PL_reg_leftiter = PL_reg_maxiter;
5458 if (PL_reg_leftiter-- == 0) {
5459 /* initialise cache */
5460 const I32 size = (PL_reg_maxiter + 7)/8;
5461 if (PL_reg_poscache) {
5462 if ((I32)PL_reg_poscache_size < size) {
5463 Renew(PL_reg_poscache, size, char);
5464 PL_reg_poscache_size = size;
5466 Zero(PL_reg_poscache, size, char);
5469 PL_reg_poscache_size = size;
5470 Newxz(PL_reg_poscache, size, char);
5472 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5473 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5474 PL_colors[4], PL_colors[5])
5478 if (PL_reg_leftiter < 0) {
5479 /* have we already failed at this position? */
5481 offset = (scan->flags & 0xf) - 1
5482 + (locinput - PL_bostr) * (scan->flags>>4);
5483 mask = 1 << (offset % 8);
5485 if (PL_reg_poscache[offset] & mask) {
5486 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5487 "%*s whilem: (cache) already tried at this position...\n",
5488 REPORT_CODE_OFF+depth*2, "")
5490 sayNO; /* cache records failure */
5492 ST.cache_offset = offset;
5493 ST.cache_mask = mask;
5497 /* Prefer B over A for minimal matching. */
5499 if (cur_curlyx->u.curlyx.minmod) {
5500 ST.save_curlyx = cur_curlyx;
5501 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5502 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5504 REGCP_SET(ST.lastcp);
5505 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5507 assert(0); /* NOTREACHED */
5510 /* Prefer A over B for maximal matching. */
5512 if (n < max) { /* More greed allowed? */
5513 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5515 cur_curlyx->u.curlyx.lastloc = locinput;
5516 REGCP_SET(ST.lastcp);
5517 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5518 assert(0); /* NOTREACHED */
5520 goto do_whilem_B_max;
5522 assert(0); /* NOTREACHED */
5524 case WHILEM_B_min: /* just matched B in a minimal match */
5525 case WHILEM_B_max: /* just matched B in a maximal match */
5526 cur_curlyx = ST.save_curlyx;
5528 assert(0); /* NOTREACHED */
5530 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5531 cur_curlyx = ST.save_curlyx;
5532 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5533 cur_curlyx->u.curlyx.count--;
5535 assert(0); /* NOTREACHED */
5537 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5539 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5540 REGCP_UNWIND(ST.lastcp);
5541 regcppop(rex, &maxopenparen);
5542 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5543 cur_curlyx->u.curlyx.count--;
5545 assert(0); /* NOTREACHED */
5547 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5548 REGCP_UNWIND(ST.lastcp);
5549 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5550 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5551 "%*s whilem: failed, trying continuation...\n",
5552 REPORT_CODE_OFF+depth*2, "")
5555 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5556 && ckWARN(WARN_REGEXP)
5557 && !reginfo->warned)
5559 reginfo->warned = TRUE;
5560 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5561 "Complex regular subexpression recursion limit (%d) "
5567 ST.save_curlyx = cur_curlyx;
5568 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5569 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5571 assert(0); /* NOTREACHED */
5573 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5574 cur_curlyx = ST.save_curlyx;
5575 REGCP_UNWIND(ST.lastcp);
5576 regcppop(rex, &maxopenparen);
5578 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5579 /* Maximum greed exceeded */
5580 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5581 && ckWARN(WARN_REGEXP)
5582 && !reginfo->warned)
5584 reginfo->warned = TRUE;
5585 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5586 "Complex regular subexpression recursion "
5587 "limit (%d) exceeded",
5590 cur_curlyx->u.curlyx.count--;
5594 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5595 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5597 /* Try grabbing another A and see if it helps. */
5598 cur_curlyx->u.curlyx.lastloc = locinput;
5599 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5601 REGCP_SET(ST.lastcp);
5602 PUSH_STATE_GOTO(WHILEM_A_min,
5603 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5605 assert(0); /* NOTREACHED */
5608 #define ST st->u.branch
5610 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5611 next = scan + ARG(scan);
5614 scan = NEXTOPER(scan);
5617 case BRANCH: /* /(...|A|...)/ */
5618 scan = NEXTOPER(scan); /* scan now points to inner node */
5619 ST.lastparen = rex->lastparen;
5620 ST.lastcloseparen = rex->lastcloseparen;
5621 ST.next_branch = next;
5624 /* Now go into the branch */
5626 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5628 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5630 assert(0); /* NOTREACHED */
5632 case CUTGROUP: /* /(*THEN)/ */
5633 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5634 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5635 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5636 assert(0); /* NOTREACHED */
5638 case CUTGROUP_next_fail:
5641 if (st->u.mark.mark_name)
5642 sv_commit = st->u.mark.mark_name;
5644 assert(0); /* NOTREACHED */
5648 assert(0); /* NOTREACHED */
5650 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5655 REGCP_UNWIND(ST.cp);
5656 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5657 scan = ST.next_branch;
5658 /* no more branches? */
5659 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5661 PerlIO_printf( Perl_debug_log,
5662 "%*s %sBRANCH failed...%s\n",
5663 REPORT_CODE_OFF+depth*2, "",
5669 continue; /* execute next BRANCH[J] op */
5670 assert(0); /* NOTREACHED */
5672 case MINMOD: /* next op will be non-greedy, e.g. A*? */
5677 #define ST st->u.curlym
5679 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5681 /* This is an optimisation of CURLYX that enables us to push
5682 * only a single backtracking state, no matter how many matches
5683 * there are in {m,n}. It relies on the pattern being constant
5684 * length, with no parens to influence future backrefs
5688 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5690 ST.lastparen = rex->lastparen;
5691 ST.lastcloseparen = rex->lastcloseparen;
5693 /* if paren positive, emulate an OPEN/CLOSE around A */
5695 U32 paren = ST.me->flags;
5696 if (paren > maxopenparen)
5697 maxopenparen = paren;
5698 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5706 ST.c1 = CHRTEST_UNINIT;
5709 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5712 curlym_do_A: /* execute the A in /A{m,n}B/ */
5713 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5714 assert(0); /* NOTREACHED */
5716 case CURLYM_A: /* we've just matched an A */
5718 /* after first match, determine A's length: u.curlym.alen */
5719 if (ST.count == 1) {
5720 if (PL_reg_match_utf8) {
5721 char *s = st->locinput;
5722 while (s < locinput) {
5728 ST.alen = locinput - st->locinput;
5731 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5734 PerlIO_printf(Perl_debug_log,
5735 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5736 (int)(REPORT_CODE_OFF+(depth*2)), "",
5737 (IV) ST.count, (IV)ST.alen)
5740 if (cur_eval && cur_eval->u.eval.close_paren &&
5741 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5745 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5746 if ( max == REG_INFTY || ST.count < max )
5747 goto curlym_do_A; /* try to match another A */
5749 goto curlym_do_B; /* try to match B */
5751 case CURLYM_A_fail: /* just failed to match an A */
5752 REGCP_UNWIND(ST.cp);
5754 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5755 || (cur_eval && cur_eval->u.eval.close_paren &&
5756 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5759 curlym_do_B: /* execute the B in /A{m,n}B/ */
5760 if (ST.c1 == CHRTEST_UNINIT) {
5761 /* calculate c1 and c2 for possible match of 1st char
5762 * following curly */
5763 ST.c1 = ST.c2 = CHRTEST_VOID;
5764 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5765 regnode *text_node = ST.B;
5766 if (! HAS_TEXT(text_node))
5767 FIND_NEXT_IMPT(text_node);
5770 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5772 But the former is redundant in light of the latter.
5774 if this changes back then the macro for
5775 IS_TEXT and friends need to change.
5777 if (PL_regkind[OP(text_node)] == EXACT) {
5778 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5779 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5789 PerlIO_printf(Perl_debug_log,
5790 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5791 (int)(REPORT_CODE_OFF+(depth*2)),
5794 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
5795 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
5796 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
5797 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
5799 /* simulate B failing */
5801 PerlIO_printf(Perl_debug_log,
5802 "%*s CURLYM Fast bail next target=U+%"UVXf" c1=U+%"UVXf" c2=U+%"UVXf"\n",
5803 (int)(REPORT_CODE_OFF+(depth*2)),"",
5804 valid_utf8_to_uvchr((U8 *) locinput, NULL),
5805 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
5806 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
5808 state_num = CURLYM_B_fail;
5809 goto reenter_switch;
5812 else if (nextchr != ST.c1 && nextchr != ST.c2) {
5813 /* simulate B failing */
5815 PerlIO_printf(Perl_debug_log,
5816 "%*s CURLYM Fast bail next target=U+%X c1=U+%X c2=U+%X\n",
5817 (int)(REPORT_CODE_OFF+(depth*2)),"",
5818 (int) nextchr, ST.c1, ST.c2)
5820 state_num = CURLYM_B_fail;
5821 goto reenter_switch;
5826 /* emulate CLOSE: mark current A as captured */
5827 I32 paren = ST.me->flags;
5829 rex->offs[paren].start
5830 = HOPc(locinput, -ST.alen) - PL_bostr;
5831 rex->offs[paren].end = locinput - PL_bostr;
5832 if ((U32)paren > rex->lastparen)
5833 rex->lastparen = paren;
5834 rex->lastcloseparen = paren;
5837 rex->offs[paren].end = -1;
5838 if (cur_eval && cur_eval->u.eval.close_paren &&
5839 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5848 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5849 assert(0); /* NOTREACHED */
5851 case CURLYM_B_fail: /* just failed to match a B */
5852 REGCP_UNWIND(ST.cp);
5853 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5855 I32 max = ARG2(ST.me);
5856 if (max != REG_INFTY && ST.count == max)
5858 goto curlym_do_A; /* try to match a further A */
5860 /* backtrack one A */
5861 if (ST.count == ARG1(ST.me) /* min */)
5864 SET_locinput(HOPc(locinput, -ST.alen));
5865 goto curlym_do_B; /* try to match B */
5868 #define ST st->u.curly
5870 #define CURLY_SETPAREN(paren, success) \
5873 rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5874 rex->offs[paren].end = locinput - PL_bostr; \
5875 if (paren > rex->lastparen) \
5876 rex->lastparen = paren; \
5877 rex->lastcloseparen = paren; \
5880 rex->offs[paren].end = -1; \
5881 rex->lastparen = ST.lastparen; \
5882 rex->lastcloseparen = ST.lastcloseparen; \
5886 case STAR: /* /A*B/ where A is width 1 char */
5890 scan = NEXTOPER(scan);
5893 case PLUS: /* /A+B/ where A is width 1 char */
5897 scan = NEXTOPER(scan);
5900 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
5901 ST.paren = scan->flags; /* Which paren to set */
5902 ST.lastparen = rex->lastparen;
5903 ST.lastcloseparen = rex->lastcloseparen;
5904 if (ST.paren > maxopenparen)
5905 maxopenparen = ST.paren;
5906 ST.min = ARG1(scan); /* min to match */
5907 ST.max = ARG2(scan); /* max to match */
5908 if (cur_eval && cur_eval->u.eval.close_paren &&
5909 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5913 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5916 case CURLY: /* /A{m,n}B/ where A is width 1 char */
5918 ST.min = ARG1(scan); /* min to match */
5919 ST.max = ARG2(scan); /* max to match */
5920 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5923 * Lookahead to avoid useless match attempts
5924 * when we know what character comes next.
5926 * Used to only do .*x and .*?x, but now it allows
5927 * for )'s, ('s and (?{ ... })'s to be in the way
5928 * of the quantifier and the EXACT-like node. -- japhy
5931 assert(ST.min <= ST.max);
5932 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
5933 ST.c1 = ST.c2 = CHRTEST_VOID;
5936 regnode *text_node = next;
5938 if (! HAS_TEXT(text_node))
5939 FIND_NEXT_IMPT(text_node);
5941 if (! HAS_TEXT(text_node))
5942 ST.c1 = ST.c2 = CHRTEST_VOID;
5944 if ( PL_regkind[OP(text_node)] != EXACT ) {
5945 ST.c1 = ST.c2 = CHRTEST_VOID;
5949 /* Currently we only get here when
5951 PL_rekind[OP(text_node)] == EXACT
5953 if this changes back then the macro for IS_TEXT and
5954 friends need to change. */
5955 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5956 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
5968 char *li = locinput;
5971 regrepeat(rex, &li, ST.A, ST.min, depth, is_utf8_pat)
5977 if (ST.c1 == CHRTEST_VOID)
5978 goto curly_try_B_min;
5980 ST.oldloc = locinput;
5982 /* set ST.maxpos to the furthest point along the
5983 * string that could possibly match */
5984 if (ST.max == REG_INFTY) {
5985 ST.maxpos = PL_regeol - 1;
5987 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5990 else if (utf8_target) {
5991 int m = ST.max - ST.min;
5992 for (ST.maxpos = locinput;
5993 m >0 && ST.maxpos < PL_regeol; m--)
5994 ST.maxpos += UTF8SKIP(ST.maxpos);
5997 ST.maxpos = locinput + ST.max - ST.min;
5998 if (ST.maxpos >= PL_regeol)
5999 ST.maxpos = PL_regeol - 1;
6001 goto curly_try_B_min_known;
6005 /* avoid taking address of locinput, so it can remain
6007 char *li = locinput;
6008 ST.count = regrepeat(rex, &li, ST.A, ST.max, depth,
6010 if (ST.count < ST.min)
6013 if ((ST.count > ST.min)
6014 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6016 /* A{m,n} must come at the end of the string, there's
6017 * no point in backing off ... */
6019 /* ...except that $ and \Z can match before *and* after
6020 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6021 We may back off by one in this case. */
6022 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6026 goto curly_try_B_max;
6028 assert(0); /* NOTREACHED */
6031 case CURLY_B_min_known_fail:
6032 /* failed to find B in a non-greedy match where c1,c2 valid */
6034 REGCP_UNWIND(ST.cp);
6036 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6038 /* Couldn't or didn't -- move forward. */
6039 ST.oldloc = locinput;
6041 locinput += UTF8SKIP(locinput);
6045 curly_try_B_min_known:
6046 /* find the next place where 'B' could work, then call B */
6050 n = (ST.oldloc == locinput) ? 0 : 1;
6051 if (ST.c1 == ST.c2) {
6052 /* set n to utf8_distance(oldloc, locinput) */
6053 while (locinput <= ST.maxpos
6054 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6056 locinput += UTF8SKIP(locinput);
6061 /* set n to utf8_distance(oldloc, locinput) */
6062 while (locinput <= ST.maxpos
6063 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6064 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6066 locinput += UTF8SKIP(locinput);
6071 else { /* Not utf8_target */
6072 if (ST.c1 == ST.c2) {
6073 while (locinput <= ST.maxpos &&
6074 UCHARAT(locinput) != ST.c1)
6078 while (locinput <= ST.maxpos
6079 && UCHARAT(locinput) != ST.c1
6080 && UCHARAT(locinput) != ST.c2)
6083 n = locinput - ST.oldloc;
6085 if (locinput > ST.maxpos)
6088 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6089 * at b; check that everything between oldloc and
6090 * locinput matches */
6091 char *li = ST.oldloc;
6093 if (regrepeat(rex, &li, ST.A, n, depth, is_utf8_pat) < n)
6095 assert(n == REG_INFTY || locinput == li);
6097 CURLY_SETPAREN(ST.paren, ST.count);
6098 if (cur_eval && cur_eval->u.eval.close_paren &&
6099 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6102 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6104 assert(0); /* NOTREACHED */
6107 case CURLY_B_min_fail:
6108 /* failed to find B in a non-greedy match where c1,c2 invalid */
6110 REGCP_UNWIND(ST.cp);
6112 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6114 /* failed -- move forward one */
6116 char *li = locinput;
6117 if (!regrepeat(rex, &li, ST.A, 1, depth, is_utf8_pat)) {
6124 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6125 ST.count > 0)) /* count overflow ? */
6128 CURLY_SETPAREN(ST.paren, ST.count);
6129 if (cur_eval && cur_eval->u.eval.close_paren &&
6130 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6133 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6137 assert(0); /* NOTREACHED */
6141 /* a successful greedy match: now try to match B */
6142 if (cur_eval && cur_eval->u.eval.close_paren &&
6143 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6147 bool could_match = locinput < PL_regeol;
6149 /* If it could work, try it. */
6150 if (ST.c1 != CHRTEST_VOID && could_match) {
6151 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6153 could_match = memEQ(locinput,
6158 UTF8SKIP(locinput));
6161 could_match = UCHARAT(locinput) == ST.c1
6162 || UCHARAT(locinput) == ST.c2;
6165 if (ST.c1 == CHRTEST_VOID || could_match) {
6166 CURLY_SETPAREN(ST.paren, ST.count);
6167 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6168 assert(0); /* NOTREACHED */
6173 case CURLY_B_max_fail:
6174 /* failed to find B in a greedy match */
6176 REGCP_UNWIND(ST.cp);
6178 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6181 if (--ST.count < ST.min)
6183 locinput = HOPc(locinput, -1);
6184 goto curly_try_B_max;
6188 case END: /* last op of main pattern */
6191 /* we've just finished A in /(??{A})B/; now continue with B */
6192 st->u.eval.saved_utf8_pat = is_utf8_pat;
6193 is_utf8_pat = cur_eval->u.eval.saved_utf8_pat;
6195 st->u.eval.prev_rex = rex_sv; /* inner */
6197 /* Save *all* the positions. */
6198 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6199 rex_sv = cur_eval->u.eval.prev_rex;
6200 SET_reg_curpm(rex_sv);
6201 rex = ReANY(rex_sv);
6202 rexi = RXi_GET(rex);
6203 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6205 REGCP_SET(st->u.eval.lastcp);
6207 /* Restore parens of the outer rex without popping the
6209 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6212 st->u.eval.prev_eval = cur_eval;
6213 cur_eval = cur_eval->u.eval.prev_eval;
6215 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6216 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6217 if ( nochange_depth )
6220 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6221 locinput); /* match B */
6224 if (locinput < reginfo->till) {
6225 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6226 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6228 (long)(locinput - PL_reg_starttry),
6229 (long)(reginfo->till - PL_reg_starttry),
6232 sayNO_SILENT; /* Cannot match: too short. */
6234 sayYES; /* Success! */
6236 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6238 PerlIO_printf(Perl_debug_log,
6239 "%*s %ssubpattern success...%s\n",
6240 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6241 sayYES; /* Success! */
6244 #define ST st->u.ifmatch
6249 case SUSPEND: /* (?>A) */
6251 newstart = locinput;
6254 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6256 goto ifmatch_trivial_fail_test;
6258 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6260 ifmatch_trivial_fail_test:
6262 char * const s = HOPBACKc(locinput, scan->flags);
6267 sw = 1 - cBOOL(ST.wanted);
6271 next = scan + ARG(scan);
6279 newstart = locinput;
6283 ST.logical = logical;
6284 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6286 /* execute body of (?...A) */
6287 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6288 assert(0); /* NOTREACHED */
6291 case IFMATCH_A_fail: /* body of (?...A) failed */
6292 ST.wanted = !ST.wanted;
6295 case IFMATCH_A: /* body of (?...A) succeeded */
6297 sw = cBOOL(ST.wanted);
6299 else if (!ST.wanted)
6302 if (OP(ST.me) != SUSPEND) {
6303 /* restore old position except for (?>...) */
6304 locinput = st->locinput;
6306 scan = ST.me + ARG(ST.me);
6309 continue; /* execute B */
6313 case LONGJMP: /* alternative with many branches compiles to
6314 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6315 next = scan + ARG(scan);
6320 case COMMIT: /* (*COMMIT) */
6321 reginfo->cutpoint = PL_regeol;
6324 case PRUNE: /* (*PRUNE) */
6326 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6327 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6328 assert(0); /* NOTREACHED */
6330 case COMMIT_next_fail:
6334 case OPFAIL: /* (*FAIL) */
6336 assert(0); /* NOTREACHED */
6338 #define ST st->u.mark
6339 case MARKPOINT: /* (*MARK:foo) */
6340 ST.prev_mark = mark_state;
6341 ST.mark_name = sv_commit = sv_yes_mark
6342 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6344 ST.mark_loc = locinput;
6345 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6346 assert(0); /* NOTREACHED */
6348 case MARKPOINT_next:
6349 mark_state = ST.prev_mark;
6351 assert(0); /* NOTREACHED */
6353 case MARKPOINT_next_fail:
6354 if (popmark && sv_eq(ST.mark_name,popmark))
6356 if (ST.mark_loc > startpoint)
6357 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6358 popmark = NULL; /* we found our mark */
6359 sv_commit = ST.mark_name;
6362 PerlIO_printf(Perl_debug_log,
6363 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6364 REPORT_CODE_OFF+depth*2, "",
6365 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6368 mark_state = ST.prev_mark;
6369 sv_yes_mark = mark_state ?
6370 mark_state->u.mark.mark_name : NULL;
6372 assert(0); /* NOTREACHED */
6374 case SKIP: /* (*SKIP) */
6376 /* (*SKIP) : if we fail we cut here*/
6377 ST.mark_name = NULL;
6378 ST.mark_loc = locinput;
6379 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6381 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6382 otherwise do nothing. Meaning we need to scan
6384 regmatch_state *cur = mark_state;
6385 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6388 if ( sv_eq( cur->u.mark.mark_name,
6391 ST.mark_name = find;
6392 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6394 cur = cur->u.mark.prev_mark;
6397 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6400 case SKIP_next_fail:
6402 /* (*CUT:NAME) - Set up to search for the name as we
6403 collapse the stack*/
6404 popmark = ST.mark_name;
6406 /* (*CUT) - No name, we cut here.*/
6407 if (ST.mark_loc > startpoint)
6408 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6409 /* but we set sv_commit to latest mark_name if there
6410 is one so they can test to see how things lead to this
6413 sv_commit=mark_state->u.mark.mark_name;
6417 assert(0); /* NOTREACHED */
6420 case LNBREAK: /* \R */
6421 if ((n=is_LNBREAK_safe(locinput, PL_regeol, utf8_target))) {
6428 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6429 PTR2UV(scan), OP(scan));
6430 Perl_croak(aTHX_ "regexp memory corruption");
6432 /* this is a point to jump to in order to increment
6433 * locinput by one character */
6435 assert(!NEXTCHR_IS_EOS);
6437 locinput += PL_utf8skip[nextchr];
6438 /* locinput is allowed to go 1 char off the end, but not 2+ */
6439 if (locinput > PL_regeol)
6448 /* switch break jumps here */
6449 scan = next; /* prepare to execute the next op and ... */
6450 continue; /* ... jump back to the top, reusing st */
6451 assert(0); /* NOTREACHED */
6454 /* push a state that backtracks on success */
6455 st->u.yes.prev_yes_state = yes_state;
6459 /* push a new regex state, then continue at scan */
6461 regmatch_state *newst;
6464 regmatch_state *cur = st;
6465 regmatch_state *curyes = yes_state;
6467 regmatch_slab *slab = PL_regmatch_slab;
6468 for (;curd > -1;cur--,curd--) {
6469 if (cur < SLAB_FIRST(slab)) {
6471 cur = SLAB_LAST(slab);
6473 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6474 REPORT_CODE_OFF + 2 + depth * 2,"",
6475 curd, PL_reg_name[cur->resume_state],
6476 (curyes == cur) ? "yes" : ""
6479 curyes = cur->u.yes.prev_yes_state;
6482 DEBUG_STATE_pp("push")
6485 st->locinput = locinput;
6487 if (newst > SLAB_LAST(PL_regmatch_slab))
6488 newst = S_push_slab(aTHX);
6489 PL_regmatch_state = newst;
6491 locinput = pushinput;
6494 assert(0); /* NOTREACHED */
6499 * We get here only if there's trouble -- normally "case END" is
6500 * the terminating point.
6502 Perl_croak(aTHX_ "corrupted regexp pointers");
6508 /* we have successfully completed a subexpression, but we must now
6509 * pop to the state marked by yes_state and continue from there */
6510 assert(st != yes_state);
6512 while (st != yes_state) {
6514 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6515 PL_regmatch_slab = PL_regmatch_slab->prev;
6516 st = SLAB_LAST(PL_regmatch_slab);
6520 DEBUG_STATE_pp("pop (no final)");
6522 DEBUG_STATE_pp("pop (yes)");
6528 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6529 || yes_state > SLAB_LAST(PL_regmatch_slab))
6531 /* not in this slab, pop slab */
6532 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6533 PL_regmatch_slab = PL_regmatch_slab->prev;
6534 st = SLAB_LAST(PL_regmatch_slab);
6536 depth -= (st - yes_state);
6539 yes_state = st->u.yes.prev_yes_state;
6540 PL_regmatch_state = st;
6543 locinput= st->locinput;
6544 state_num = st->resume_state + no_final;
6545 goto reenter_switch;
6548 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6549 PL_colors[4], PL_colors[5]));
6551 if (PL_reg_state.re_state_eval_setup_done) {
6552 /* each successfully executed (?{...}) block does the equivalent of
6553 * local $^R = do {...}
6554 * When popping the save stack, all these locals would be undone;
6555 * bypass this by setting the outermost saved $^R to the latest
6557 if (oreplsv != GvSV(PL_replgv))
6558 sv_setsv(oreplsv, GvSV(PL_replgv));
6565 PerlIO_printf(Perl_debug_log,
6566 "%*s %sfailed...%s\n",
6567 REPORT_CODE_OFF+depth*2, "",
6568 PL_colors[4], PL_colors[5])
6580 /* there's a previous state to backtrack to */
6582 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6583 PL_regmatch_slab = PL_regmatch_slab->prev;
6584 st = SLAB_LAST(PL_regmatch_slab);
6586 PL_regmatch_state = st;
6587 locinput= st->locinput;
6589 DEBUG_STATE_pp("pop");
6591 if (yes_state == st)
6592 yes_state = st->u.yes.prev_yes_state;
6594 state_num = st->resume_state + 1; /* failure = success + 1 */
6595 goto reenter_switch;
6600 if (rex->intflags & PREGf_VERBARG_SEEN) {
6601 SV *sv_err = get_sv("REGERROR", 1);
6602 SV *sv_mrk = get_sv("REGMARK", 1);
6604 sv_commit = &PL_sv_no;
6606 sv_yes_mark = &PL_sv_yes;
6609 sv_commit = &PL_sv_yes;
6610 sv_yes_mark = &PL_sv_no;
6612 sv_setsv(sv_err, sv_commit);
6613 sv_setsv(sv_mrk, sv_yes_mark);
6617 if (last_pushed_cv) {
6620 PERL_UNUSED_VAR(SP);
6623 /* clean up; in particular, free all slabs above current one */
6624 LEAVE_SCOPE(oldsave);
6626 assert(!result || locinput - PL_bostr >= 0);
6627 return result ? locinput - PL_bostr : -1;
6631 - regrepeat - repeatedly match something simple, report how many
6633 * What 'simple' means is a node which can be the operand of a quantifier like
6636 * startposp - pointer a pointer to the start position. This is updated
6637 * to point to the byte following the highest successful
6639 * p - the regnode to be repeatedly matched against.
6640 * max - maximum number of things to match.
6641 * depth - (for debugging) backtracking depth.
6644 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
6645 I32 max, int depth, bool is_utf8_pat)
6648 char *scan; /* Pointer to current position in target string */
6650 char *loceol = PL_regeol; /* local version */
6651 I32 hardcount = 0; /* How many matches so far */
6652 bool utf8_target = PL_reg_match_utf8;
6653 int to_complement = 0; /* Invert the result? */
6655 _char_class_number classnum;
6657 PERL_UNUSED_ARG(depth);
6660 PERL_ARGS_ASSERT_REGREPEAT;
6663 if (max == REG_INFTY)
6665 else if (! utf8_target && scan + max < loceol)
6666 loceol = scan + max;
6668 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
6669 * to the maximum of how far we should go in it (leaving it set to the real
6670 * end, if the maximum permissible would take us beyond that). This allows
6671 * us to make the loop exit condition that we haven't gone past <loceol> to
6672 * also mean that we haven't exceeded the max permissible count, saving a
6673 * test each time through the loop. But it assumes that the OP matches a
6674 * single byte, which is true for most of the OPs below when applied to a
6675 * non-UTF-8 target. Those relatively few OPs that don't have this
6676 * characteristic will have to compensate.
6678 * There is no adjustment for UTF-8 targets, as the number of bytes per
6679 * character varies. OPs will have to test both that the count is less
6680 * than the max permissible (using <hardcount> to keep track), and that we
6681 * are still within the bounds of the string (using <loceol>. A few OPs
6682 * match a single byte no matter what the encoding. They can omit the max
6683 * test if, for the UTF-8 case, they do the adjustment that was skipped
6686 * Thus, the code above sets things up for the common case; and exceptional
6687 * cases need extra work; the common case is to make sure <scan> doesn't
6688 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
6689 * count doesn't exceed the maximum permissible */
6694 while (scan < loceol && hardcount < max && *scan != '\n') {
6695 scan += UTF8SKIP(scan);
6699 while (scan < loceol && *scan != '\n')
6705 while (scan < loceol && hardcount < max) {
6706 scan += UTF8SKIP(scan);
6713 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
6714 if (utf8_target && scan + max < loceol) {
6716 /* <loceol> hadn't been adjusted in the UTF-8 case */
6724 assert(STR_LEN(p) == is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6728 /* Can use a simple loop if the pattern char to match on is invariant
6729 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
6730 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
6731 * true iff it doesn't matter if the argument is in UTF-8 or not */
6732 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! is_utf8_pat)) {
6733 if (utf8_target && scan + max < loceol) {
6734 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6735 * since here, to match at all, 1 char == 1 byte */
6736 loceol = scan + max;
6738 while (scan < loceol && UCHARAT(scan) == c) {
6742 else if (is_utf8_pat) {
6744 STRLEN scan_char_len;
6746 /* When both target and pattern are UTF-8, we have to do
6748 while (hardcount < max
6750 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
6751 && memEQ(scan, STRING(p), scan_char_len))
6753 scan += scan_char_len;
6757 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
6759 /* Target isn't utf8; convert the character in the UTF-8
6760 * pattern to non-UTF8, and do a simple loop */
6761 c = TWO_BYTE_UTF8_TO_UNI(c, *(STRING(p) + 1));
6762 while (scan < loceol && UCHARAT(scan) == c) {
6765 } /* else pattern char is above Latin1, can't possibly match the
6770 /* Here, the string must be utf8; pattern isn't, and <c> is
6771 * different in utf8 than not, so can't compare them directly.
6772 * Outside the loop, find the two utf8 bytes that represent c, and
6773 * then look for those in sequence in the utf8 string */
6774 U8 high = UTF8_TWO_BYTE_HI(c);
6775 U8 low = UTF8_TWO_BYTE_LO(c);
6777 while (hardcount < max
6778 && scan + 1 < loceol
6779 && UCHARAT(scan) == high
6780 && UCHARAT(scan + 1) == low)
6789 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6793 RXp_MATCH_TAINTED_on(prog);
6794 utf8_flags = FOLDEQ_UTF8_LOCALE;
6802 case EXACTFU_TRICKYFOLD:
6804 utf8_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6808 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
6810 assert(STR_LEN(p) == is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
6812 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
6815 if (c1 == CHRTEST_VOID) {
6816 /* Use full Unicode fold matching */
6817 char *tmpeol = PL_regeol;
6818 STRLEN pat_len = is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
6819 while (hardcount < max
6820 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6821 STRING(p), NULL, pat_len,
6822 is_utf8_pat, utf8_flags))
6829 else if (utf8_target) {
6831 while (scan < loceol
6833 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
6835 scan += UTF8SKIP(scan);
6840 while (scan < loceol
6842 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
6843 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
6845 scan += UTF8SKIP(scan);
6850 else if (c1 == c2) {
6851 while (scan < loceol && UCHARAT(scan) == c1) {
6856 while (scan < loceol &&
6857 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
6866 case ANYOF_WARN_SUPER:
6868 while (hardcount < max
6870 && reginclass(prog, p, (U8*)scan, utf8_target))
6872 scan += UTF8SKIP(scan);
6876 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6881 /* The argument (FLAGS) to all the POSIX node types is the class number */
6888 RXp_MATCH_TAINTED_on(prog);
6889 if (! utf8_target) {
6890 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
6896 while (hardcount < max && scan < loceol
6897 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
6900 scan += UTF8SKIP(scan);
6913 if (utf8_target && scan + max < loceol) {
6915 /* We didn't adjust <loceol> at the beginning of this routine
6916 * because is UTF-8, but it is actually ok to do so, since here, to
6917 * match, 1 char == 1 byte. */
6918 loceol = scan + max;
6920 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6933 if (! utf8_target) {
6934 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6940 /* The complement of something that matches only ASCII matches all
6941 * UTF-8 variant code points, plus everything in ASCII that isn't
6943 while (hardcount < max && scan < loceol
6944 && (! UTF8_IS_INVARIANT(*scan)
6945 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
6947 scan += UTF8SKIP(scan);
6958 if (! utf8_target) {
6959 while (scan < loceol && to_complement
6960 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
6967 classnum = (_char_class_number) FLAGS(p);
6968 if (classnum < _FIRST_NON_SWASH_CC) {
6970 /* Here, a swash is needed for above-Latin1 code points.
6971 * Process as many Latin1 code points using the built-in rules.
6972 * Go to another loop to finish processing upon encountering
6973 * the first Latin1 code point. We could do that in this loop
6974 * as well, but the other way saves having to test if the swash
6975 * has been loaded every time through the loop: extra space to
6977 while (hardcount < max && scan < loceol) {
6978 if (UTF8_IS_INVARIANT(*scan)) {
6979 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
6986 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
6987 if (! (to_complement
6988 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(*scan,
6997 goto found_above_latin1;
7004 /* For these character classes, the knowledge of how to handle
7005 * every code point is compiled in to Perl via a macro. This
7006 * code is written for making the loops as tight as possible.
7007 * It could be refactored to save space instead */
7009 case _CC_ENUM_SPACE: /* XXX would require separate code
7010 if we revert the change of \v
7013 case _CC_ENUM_PSXSPC:
7014 while (hardcount < max
7016 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7018 scan += UTF8SKIP(scan);
7022 case _CC_ENUM_BLANK:
7023 while (hardcount < max
7025 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7027 scan += UTF8SKIP(scan);
7031 case _CC_ENUM_XDIGIT:
7032 while (hardcount < max
7034 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7036 scan += UTF8SKIP(scan);
7040 case _CC_ENUM_VERTSPACE:
7041 while (hardcount < max
7043 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7045 scan += UTF8SKIP(scan);
7049 case _CC_ENUM_CNTRL:
7050 while (hardcount < max
7052 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7054 scan += UTF8SKIP(scan);
7059 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7065 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7067 /* Load the swash if not already present */
7068 if (! PL_utf8_swash_ptrs[classnum]) {
7069 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7070 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7071 "utf8", swash_property_names[classnum],
7072 &PL_sv_undef, 1, 0, NULL, &flags);
7075 while (hardcount < max && scan < loceol
7076 && to_complement ^ cBOOL(_generic_utf8(
7079 swash_fetch(PL_utf8_swash_ptrs[classnum],
7083 scan += UTF8SKIP(scan);
7090 while (hardcount < max && scan < loceol &&
7091 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7096 /* LNBREAK can match one or two latin chars, which is ok, but we
7097 * have to use hardcount in this situation, and throw away the
7098 * adjustment to <loceol> done before the switch statement */
7100 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7121 /* These are all 0 width, so match right here or not at all. */
7125 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7126 assert(0); /* NOTREACHED */
7133 c = scan - *startposp;
7137 GET_RE_DEBUG_FLAGS_DECL;
7139 SV * const prop = sv_newmortal();
7140 regprop(prog, prop, p);
7141 PerlIO_printf(Perl_debug_log,
7142 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7143 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7151 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7153 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7154 create a copy so that changes the caller makes won't change the shared one.
7155 If <altsvp> is non-null, will return NULL in it, for back-compat.
7158 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7160 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7166 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp));
7171 S_core_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp)
7173 /* Returns the swash for the input 'node' in the regex 'prog'.
7174 * If <doinit> is true, will attempt to create the swash if not already
7176 * If <listsvp> is non-null, will return the swash initialization string in
7178 * Tied intimately to how regcomp.c sets up the data structure */
7185 RXi_GET_DECL(prog,progi);
7186 const struct reg_data * const data = prog ? progi->data : NULL;
7188 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
7190 assert(ANYOF_NONBITMAP(node));
7192 if (data && data->count) {
7193 const U32 n = ARG(node);
7195 if (data->what[n] == 's') {
7196 SV * const rv = MUTABLE_SV(data->data[n]);
7197 AV * const av = MUTABLE_AV(SvRV(rv));
7198 SV **const ary = AvARRAY(av);
7199 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7201 si = *ary; /* ary[0] = the string to initialize the swash with */
7203 /* Elements 2 and 3 are either both present or both absent. [2] is
7204 * any inversion list generated at compile time; [3] indicates if
7205 * that inversion list has any user-defined properties in it. */
7206 if (av_len(av) >= 2) {
7209 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7216 /* Element [1] is reserved for the set-up swash. If already there,
7217 * return it; if not, create it and store it there */
7218 if (SvROK(ary[1])) {
7221 else if (si && doinit) {
7223 sw = _core_swash_init("utf8", /* the utf8 package */
7227 0, /* not from tr/// */
7230 (void)av_store(av, 1, sw);
7236 SV* matches_string = newSVpvn("", 0);
7238 /* Use the swash, if any, which has to have incorporated into it all
7240 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7241 && (si && si != &PL_sv_undef))
7244 /* If no swash, use the input initialization string, if available */
7245 sv_catsv(matches_string, si);
7248 /* Add the inversion list to whatever we have. This may have come from
7249 * the swash, or from an input parameter */
7251 sv_catsv(matches_string, _invlist_contents(invlist));
7253 *listsvp = matches_string;
7260 - reginclass - determine if a character falls into a character class
7262 n is the ANYOF regnode
7263 p is the target string
7264 utf8_target tells whether p is in UTF-8.
7266 Returns true if matched; false otherwise.
7268 Note that this can be a synthetic start class, a combination of various
7269 nodes, so things you think might be mutually exclusive, such as locale,
7270 aren't. It can match both locale and non-locale
7275 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const bool utf8_target)
7278 const char flags = ANYOF_FLAGS(n);
7282 PERL_ARGS_ASSERT_REGINCLASS;
7284 /* If c is not already the code point, get it. Note that
7285 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7286 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7288 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
7289 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7290 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7291 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7292 * UTF8_ALLOW_FFFF */
7293 if (c_len == (STRLEN)-1)
7294 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7297 /* If this character is potentially in the bitmap, check it */
7299 if (ANYOF_BITMAP_TEST(n, c))
7301 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
7307 else if (flags & ANYOF_LOCALE) {
7308 RXp_MATCH_TAINTED_on(prog);
7310 if ((flags & ANYOF_LOC_FOLD)
7311 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7315 else if (ANYOF_CLASS_TEST_ANY_SET(n)) {
7317 /* The data structure is arranged so bits 0, 2, 4, ... are set
7318 * if the class includes the Posix character class given by
7319 * bit/2; and 1, 3, 5, ... are set if the class includes the
7320 * complemented Posix class given by int(bit/2). So we loop
7321 * through the bits, each time changing whether we complement
7322 * the result or not. Suppose for the sake of illustration
7323 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7324 * is set, it means there is a match for this ANYOF node if the
7325 * character is in the class given by the expression (0 / 2 = 0
7326 * = \w). If it is in that class, isFOO_lc() will return 1,
7327 * and since 'to_complement' is 0, the result will stay TRUE,
7328 * and we exit the loop. Suppose instead that bit 0 is 0, but
7329 * bit 1 is 1. That means there is a match if the character
7330 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7331 * but will on bit 1. On the second iteration 'to_complement'
7332 * will be 1, so the exclusive or will reverse things, so we
7333 * are testing for \W. On the third iteration, 'to_complement'
7334 * will be 0, and we would be testing for \s; the fourth
7335 * iteration would test for \S, etc.
7337 * Note that this code assumes that all the classes are closed
7338 * under folding. For example, if a character matches \w, then
7339 * its fold does too; and vice versa. This should be true for
7340 * any well-behaved locale for all the currently defined Posix
7341 * classes, except for :lower: and :upper:, which are handled
7342 * by the pseudo-class :cased: which matches if either of the
7343 * other two does. To get rid of this assumption, an outer
7344 * loop could be used below to iterate over both the source
7345 * character, and its fold (if different) */
7348 int to_complement = 0;
7349 while (count < ANYOF_MAX) {
7350 if (ANYOF_CLASS_TEST(n, count)
7351 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7363 /* If the bitmap didn't (or couldn't) match, and something outside the
7364 * bitmap could match, try that. Locale nodes specify completely the
7365 * behavior of code points in the bit map (otherwise, a utf8 target would
7366 * cause them to be treated as Unicode and not locale), except in
7367 * the very unlikely event when this node is a synthetic start class, which
7368 * could be a combination of locale and non-locale nodes. So allow locale
7369 * to match for the synthetic start class, which will give a false
7370 * positive that will be resolved when the match is done again as not part
7371 * of the synthetic start class */
7373 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
7374 match = TRUE; /* Everything above 255 matches */
7376 else if (ANYOF_NONBITMAP(n)
7377 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
7380 || (! (flags & ANYOF_LOCALE))
7381 || OP(n) == ANYOF_SYNTHETIC))))
7383 SV * const sw = core_regclass_swash(prog, n, TRUE, 0);
7388 } else { /* Convert to utf8 */
7390 utf8_p = bytes_to_utf8(p, &len);
7393 if (swash_fetch(sw, utf8_p, TRUE)) {
7397 /* If we allocated a string above, free it */
7398 if (! utf8_target) Safefree(utf8_p);
7402 if (UNICODE_IS_SUPER(c)
7403 && OP(n) == ANYOF_WARN_SUPER
7404 && ckWARN_d(WARN_NON_UNICODE))
7406 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7407 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7411 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7412 return cBOOL(flags & ANYOF_INVERT) ^ match;
7416 S_reghop3(U8 *s, I32 off, const U8* lim)
7418 /* return the position 'off' UTF-8 characters away from 's', forward if
7419 * 'off' >= 0, backwards if negative. But don't go outside of position
7420 * 'lim', which better be < s if off < 0 */
7424 PERL_ARGS_ASSERT_REGHOP3;
7427 while (off-- && s < lim) {
7428 /* XXX could check well-formedness here */
7433 while (off++ && s > lim) {
7435 if (UTF8_IS_CONTINUED(*s)) {
7436 while (s > lim && UTF8_IS_CONTINUATION(*s))
7439 /* XXX could check well-formedness here */
7446 /* there are a bunch of places where we use two reghop3's that should
7447 be replaced with this routine. but since thats not done yet
7448 we ifdef it out - dmq
7451 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7455 PERL_ARGS_ASSERT_REGHOP4;
7458 while (off-- && s < rlim) {
7459 /* XXX could check well-formedness here */
7464 while (off++ && s > llim) {
7466 if (UTF8_IS_CONTINUED(*s)) {
7467 while (s > llim && UTF8_IS_CONTINUATION(*s))
7470 /* XXX could check well-formedness here */
7478 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7482 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7485 while (off-- && s < lim) {
7486 /* XXX could check well-formedness here */
7493 while (off++ && s > lim) {
7495 if (UTF8_IS_CONTINUED(*s)) {
7496 while (s > lim && UTF8_IS_CONTINUATION(*s))
7499 /* XXX could check well-formedness here */
7508 restore_pos(pTHX_ void *arg)
7511 regexp * const rex = (regexp *)arg;
7512 if (PL_reg_state.re_state_eval_setup_done) {
7513 if (PL_reg_oldsaved) {
7514 rex->subbeg = PL_reg_oldsaved;
7515 rex->sublen = PL_reg_oldsavedlen;
7516 rex->suboffset = PL_reg_oldsavedoffset;
7517 rex->subcoffset = PL_reg_oldsavedcoffset;
7519 rex->saved_copy = PL_nrs;
7521 RXp_MATCH_COPIED_on(rex);
7523 PL_reg_magic->mg_len = PL_reg_oldpos;
7524 PL_reg_state.re_state_eval_setup_done = FALSE;
7525 PL_curpm = PL_reg_oldcurpm;
7530 S_to_utf8_substr(pTHX_ regexp *prog)
7532 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
7533 * on the converted value */
7537 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7540 if (prog->substrs->data[i].substr
7541 && !prog->substrs->data[i].utf8_substr) {
7542 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7543 prog->substrs->data[i].utf8_substr = sv;
7544 sv_utf8_upgrade(sv);
7545 if (SvVALID(prog->substrs->data[i].substr)) {
7546 if (SvTAIL(prog->substrs->data[i].substr)) {
7547 /* Trim the trailing \n that fbm_compile added last
7549 SvCUR_set(sv, SvCUR(sv) - 1);
7550 /* Whilst this makes the SV technically "invalid" (as its
7551 buffer is no longer followed by "\0") when fbm_compile()
7552 adds the "\n" back, a "\0" is restored. */
7553 fbm_compile(sv, FBMcf_TAIL);
7557 if (prog->substrs->data[i].substr == prog->check_substr)
7558 prog->check_utf8 = sv;
7564 S_to_byte_substr(pTHX_ regexp *prog)
7566 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
7567 * on the converted value; returns FALSE if can't be converted. */
7572 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7575 if (prog->substrs->data[i].utf8_substr
7576 && !prog->substrs->data[i].substr) {
7577 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7578 if (! sv_utf8_downgrade(sv, TRUE)) {
7581 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7582 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7583 /* Trim the trailing \n that fbm_compile added last
7585 SvCUR_set(sv, SvCUR(sv) - 1);
7586 fbm_compile(sv, FBMcf_TAIL);
7590 prog->substrs->data[i].substr = sv;
7591 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7592 prog->check_substr = sv;
7601 * c-indentation-style: bsd
7603 * indent-tabs-mode: nil
7606 * ex: set ts=8 sts=4 sw=4 et: