5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "invlist_inline.h"
84 #include "unicode_constants.h"
86 #define B_ON_NON_UTF8_LOCALE_IS_WRONG \
87 "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale"
89 static const char utf8_locale_required[] =
90 "Use of (?[ ]) for non-UTF-8 locale is wrong. Assuming a UTF-8 locale";
93 /* At least one required character in the target string is expressible only in
95 static const char* const non_utf8_target_but_utf8_required
96 = "Can't match, because target string needs to be in UTF-8\n";
99 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
100 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%s", non_utf8_target_but_utf8_required));\
104 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
107 #define STATIC static
110 /* Valid only if 'c', the character being looke-up, is an invariant under
111 * UTF-8: it avoids the reginclass call if there are no complications: i.e., if
112 * everything matchable is straight forward in the bitmap */
113 #define REGINCLASS(prog,p,c,u) (ANYOF_FLAGS(p) \
114 ? reginclass(prog,p,c,c+1,u) \
115 : ANYOF_BITMAP_TEST(p,*(c)))
121 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
122 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
124 #define HOPc(pos,off) \
125 (char *)(reginfo->is_utf8_target \
126 ? reghop3((U8*)pos, off, \
127 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
130 #define HOPBACKc(pos, off) \
131 (char*)(reginfo->is_utf8_target \
132 ? reghopmaybe3((U8*)pos, (SSize_t)0-off, (U8*)(reginfo->strbeg)) \
133 : (pos - off >= reginfo->strbeg) \
137 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
138 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
140 /* lim must be +ve. Returns NULL on overshoot */
141 #define HOPMAYBE3(pos,off,lim) \
142 (reginfo->is_utf8_target \
143 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
144 : ((U8*)pos + off <= lim) \
148 /* like HOP3, but limits the result to <= lim even for the non-utf8 case.
149 * off must be >=0; args should be vars rather than expressions */
150 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
151 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
152 : (U8*)((pos + off) > lim ? lim : (pos + off)))
154 #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
155 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
157 #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
159 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
160 #define NEXTCHR_IS_EOS (nextchr < 0)
162 #define SET_nextchr \
163 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
165 #define SET_locinput(p) \
170 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
172 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
173 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
174 1, 0, invlist, &flags); \
179 /* If in debug mode, we test that a known character properly matches */
181 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
184 utf8_char_in_property) \
185 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \
186 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
188 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
191 utf8_char_in_property) \
192 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
195 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
196 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
198 PL_XPosix_ptrs[_CC_WORDCHAR], \
199 LATIN_SMALL_LIGATURE_LONG_S_T_UTF8);
201 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
202 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
204 /* for use after a quantifier and before an EXACT-like node -- japhy */
205 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
207 * NOTE that *nothing* that affects backtracking should be in here, specifically
208 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
209 * node that is in between two EXACT like nodes when ascertaining what the required
210 * "follow" character is. This should probably be moved to regex compile time
211 * although it may be done at run time beause of the REF possibility - more
212 * investigation required. -- demerphq
214 #define JUMPABLE(rn) ( \
216 (OP(rn) == CLOSE && \
217 !EVAL_CLOSE_PAREN_IS(cur_eval,ARG(rn)) ) || \
219 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
220 OP(rn) == PLUS || OP(rn) == MINMOD || \
222 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
224 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
226 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
229 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
230 we don't need this definition. XXX These are now out-of-sync*/
231 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
232 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
233 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
236 /* ... so we use this as its faster. */
237 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==EXACTL )
238 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFLU8 || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
239 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
240 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
245 Search for mandatory following text node; for lookahead, the text must
246 follow but for lookbehind (rn->flags != 0) we skip to the next step.
248 #define FIND_NEXT_IMPT(rn) STMT_START { \
249 while (JUMPABLE(rn)) { \
250 const OPCODE type = OP(rn); \
251 if (type == SUSPEND || PL_regkind[type] == CURLY) \
252 rn = NEXTOPER(NEXTOPER(rn)); \
253 else if (type == PLUS) \
255 else if (type == IFMATCH) \
256 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
257 else rn += NEXT_OFF(rn); \
261 #define SLAB_FIRST(s) (&(s)->states[0])
262 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
264 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
265 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
266 static regmatch_state * S_push_slab(pTHX);
268 #define REGCP_PAREN_ELEMS 3
269 #define REGCP_OTHER_ELEMS 3
270 #define REGCP_FRAME_ELEMS 1
271 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
272 * are needed for the regexp context stack bookkeeping. */
275 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen _pDEPTH)
277 const int retval = PL_savestack_ix;
278 const int paren_elems_to_push =
279 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
280 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
281 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
283 GET_RE_DEBUG_FLAGS_DECL;
285 PERL_ARGS_ASSERT_REGCPPUSH;
287 if (paren_elems_to_push < 0)
288 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
289 (int)paren_elems_to_push, (int)maxopenparen,
290 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
292 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
293 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %" UVuf
294 " out of range (%lu-%ld)",
296 (unsigned long)maxopenparen,
299 SSGROW(total_elems + REGCP_FRAME_ELEMS);
302 if ((int)maxopenparen > (int)parenfloor)
303 Perl_re_exec_indentf( aTHX_
304 "rex=0x%" UVxf " offs=0x%" UVxf ": saving capture indices:\n",
310 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
311 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
312 SSPUSHIV(rex->offs[p].end);
313 SSPUSHIV(rex->offs[p].start);
314 SSPUSHINT(rex->offs[p].start_tmp);
315 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
316 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "\n",
319 (IV)rex->offs[p].start,
320 (IV)rex->offs[p].start_tmp,
324 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
325 SSPUSHINT(maxopenparen);
326 SSPUSHINT(rex->lastparen);
327 SSPUSHINT(rex->lastcloseparen);
328 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
333 /* These are needed since we do not localize EVAL nodes: */
334 #define REGCP_SET(cp) \
336 Perl_re_exec_indentf( aTHX_ \
337 "Setting an EVAL scope, savestack=%" IVdf ",\n", \
338 depth, (IV)PL_savestack_ix \
343 #define REGCP_UNWIND(cp) \
345 if (cp != PL_savestack_ix) \
346 Perl_re_exec_indentf( aTHX_ \
347 "Clearing an EVAL scope, savestack=%" \
348 IVdf "..%" IVdf "\n", \
349 depth, (IV)(cp), (IV)PL_savestack_ix \
354 #define UNWIND_PAREN(lp, lcp) \
355 for (n = rex->lastparen; n > lp; n--) \
356 rex->offs[n].end = -1; \
357 rex->lastparen = n; \
358 rex->lastcloseparen = lcp;
362 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p _pDEPTH)
366 GET_RE_DEBUG_FLAGS_DECL;
368 PERL_ARGS_ASSERT_REGCPPOP;
370 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
372 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
373 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
374 rex->lastcloseparen = SSPOPINT;
375 rex->lastparen = SSPOPINT;
376 *maxopenparen_p = SSPOPINT;
378 i -= REGCP_OTHER_ELEMS;
379 /* Now restore the parentheses context. */
381 if (i || rex->lastparen + 1 <= rex->nparens)
382 Perl_re_exec_indentf( aTHX_
383 "rex=0x%" UVxf " offs=0x%" UVxf ": restoring capture indices to:\n",
389 paren = *maxopenparen_p;
390 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
392 rex->offs[paren].start_tmp = SSPOPINT;
393 rex->offs[paren].start = SSPOPIV;
395 if (paren <= rex->lastparen)
396 rex->offs[paren].end = tmps;
397 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
398 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "%s\n",
401 (IV)rex->offs[paren].start,
402 (IV)rex->offs[paren].start_tmp,
403 (IV)rex->offs[paren].end,
404 (paren > rex->lastparen ? "(skipped)" : ""));
409 /* It would seem that the similar code in regtry()
410 * already takes care of this, and in fact it is in
411 * a better location to since this code can #if 0-ed out
412 * but the code in regtry() is needed or otherwise tests
413 * requiring null fields (pat.t#187 and split.t#{13,14}
414 * (as of patchlevel 7877) will fail. Then again,
415 * this code seems to be necessary or otherwise
416 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
417 * --jhi updated by dapm */
418 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
419 if (i > *maxopenparen_p)
420 rex->offs[i].start = -1;
421 rex->offs[i].end = -1;
422 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
423 " \\%" UVuf ": %s ..-1 undeffing\n",
426 (i > *maxopenparen_p) ? "-1" : " "
432 /* restore the parens and associated vars at savestack position ix,
433 * but without popping the stack */
436 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p _pDEPTH)
438 I32 tmpix = PL_savestack_ix;
439 PERL_ARGS_ASSERT_REGCP_RESTORE;
441 PL_savestack_ix = ix;
442 regcppop(rex, maxopenparen_p);
443 PL_savestack_ix = tmpix;
446 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
449 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
451 /* Returns a boolean as to whether or not 'character' is a member of the
452 * Posix character class given by 'classnum' that should be equivalent to a
453 * value in the typedef '_char_class_number'.
455 * Ideally this could be replaced by a just an array of function pointers
456 * to the C library functions that implement the macros this calls.
457 * However, to compile, the precise function signatures are required, and
458 * these may vary from platform to to platform. To avoid having to figure
459 * out what those all are on each platform, I (khw) am using this method,
460 * which adds an extra layer of function call overhead (unless the C
461 * optimizer strips it away). But we don't particularly care about
462 * performance with locales anyway. */
464 switch ((_char_class_number) classnum) {
465 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
466 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
467 case _CC_ENUM_ASCII: return isASCII_LC(character);
468 case _CC_ENUM_BLANK: return isBLANK_LC(character);
469 case _CC_ENUM_CASED: return isLOWER_LC(character)
470 || isUPPER_LC(character);
471 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
472 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
473 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
474 case _CC_ENUM_LOWER: return isLOWER_LC(character);
475 case _CC_ENUM_PRINT: return isPRINT_LC(character);
476 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
477 case _CC_ENUM_SPACE: return isSPACE_LC(character);
478 case _CC_ENUM_UPPER: return isUPPER_LC(character);
479 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
480 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
481 default: /* VERTSPACE should never occur in locales */
482 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
485 NOT_REACHED; /* NOTREACHED */
490 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
492 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
493 * 'character' is a member of the Posix character class given by 'classnum'
494 * that should be equivalent to a value in the typedef
495 * '_char_class_number'.
497 * This just calls isFOO_lc on the code point for the character if it is in
498 * the range 0-255. Outside that range, all characters use Unicode
499 * rules, ignoring any locale. So use the Unicode function if this class
500 * requires a swash, and use the Unicode macro otherwise. */
502 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
504 if (UTF8_IS_INVARIANT(*character)) {
505 return isFOO_lc(classnum, *character);
507 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
508 return isFOO_lc(classnum,
509 EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1)));
512 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, character + UTF8SKIP(character));
514 if (classnum < _FIRST_NON_SWASH_CC) {
516 /* Initialize the swash unless done already */
517 if (! PL_utf8_swash_ptrs[classnum]) {
518 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
519 PL_utf8_swash_ptrs[classnum] =
520 _core_swash_init("utf8",
523 PL_XPosix_ptrs[classnum], &flags);
526 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
528 TRUE /* is UTF */ ));
531 switch ((_char_class_number) classnum) {
532 case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character);
533 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
534 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
535 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
539 return FALSE; /* Things like CNTRL are always below 256 */
543 * pregexec and friends
546 #ifndef PERL_IN_XSUB_RE
548 - pregexec - match a regexp against a string
551 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
552 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
553 /* stringarg: the point in the string at which to begin matching */
554 /* strend: pointer to null at end of string */
555 /* strbeg: real beginning of string */
556 /* minend: end of match must be >= minend bytes after stringarg. */
557 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
558 * itself is accessed via the pointers above */
559 /* nosave: For optimizations. */
561 PERL_ARGS_ASSERT_PREGEXEC;
564 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
565 nosave ? 0 : REXEC_COPY_STR);
571 /* re_intuit_start():
573 * Based on some optimiser hints, try to find the earliest position in the
574 * string where the regex could match.
576 * rx: the regex to match against
577 * sv: the SV being matched: only used for utf8 flag; the string
578 * itself is accessed via the pointers below. Note that on
579 * something like an overloaded SV, SvPOK(sv) may be false
580 * and the string pointers may point to something unrelated to
582 * strbeg: real beginning of string
583 * strpos: the point in the string at which to begin matching
584 * strend: pointer to the byte following the last char of the string
585 * flags currently unused; set to 0
586 * data: currently unused; set to NULL
588 * The basic idea of re_intuit_start() is to use some known information
589 * about the pattern, namely:
591 * a) the longest known anchored substring (i.e. one that's at a
592 * constant offset from the beginning of the pattern; but not
593 * necessarily at a fixed offset from the beginning of the
595 * b) the longest floating substring (i.e. one that's not at a constant
596 * offset from the beginning of the pattern);
597 * c) Whether the pattern is anchored to the string; either
598 * an absolute anchor: /^../, or anchored to \n: /^.../m,
599 * or anchored to pos(): /\G/;
600 * d) A start class: a real or synthetic character class which
601 * represents which characters are legal at the start of the pattern;
603 * to either quickly reject the match, or to find the earliest position
604 * within the string at which the pattern might match, thus avoiding
605 * running the full NFA engine at those earlier locations, only to
606 * eventually fail and retry further along.
608 * Returns NULL if the pattern can't match, or returns the address within
609 * the string which is the earliest place the match could occur.
611 * The longest of the anchored and floating substrings is called 'check'
612 * and is checked first. The other is called 'other' and is checked
613 * second. The 'other' substring may not be present. For example,
615 * /(abc|xyz)ABC\d{0,3}DEFG/
619 * check substr (float) = "DEFG", offset 6..9 chars
620 * other substr (anchored) = "ABC", offset 3..3 chars
623 * Be aware that during the course of this function, sometimes 'anchored'
624 * refers to a substring being anchored relative to the start of the
625 * pattern, and sometimes to the pattern itself being anchored relative to
626 * the string. For example:
628 * /\dabc/: "abc" is anchored to the pattern;
629 * /^\dabc/: "abc" is anchored to the pattern and the string;
630 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
631 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
632 * but the pattern is anchored to the string.
636 Perl_re_intuit_start(pTHX_
639 const char * const strbeg,
643 re_scream_pos_data *data)
645 struct regexp *const prog = ReANY(rx);
646 SSize_t start_shift = prog->check_offset_min;
647 /* Should be nonnegative! */
648 SSize_t end_shift = 0;
649 /* current lowest pos in string where the regex can start matching */
650 char *rx_origin = strpos;
652 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
653 U8 other_ix = 1 - prog->substrs->check_ix;
655 char *other_last = strpos;/* latest pos 'other' substr already checked to */
656 char *check_at = NULL; /* check substr found at this pos */
657 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
658 RXi_GET_DECL(prog,progi);
659 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
660 regmatch_info *const reginfo = ®info_buf;
661 GET_RE_DEBUG_FLAGS_DECL;
663 PERL_ARGS_ASSERT_RE_INTUIT_START;
664 PERL_UNUSED_ARG(flags);
665 PERL_UNUSED_ARG(data);
667 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
668 "Intuit: trying to determine minimum start position...\n"));
670 /* for now, assume that all substr offsets are positive. If at some point
671 * in the future someone wants to do clever things with lookbehind and
672 * -ve offsets, they'll need to fix up any code in this function
673 * which uses these offsets. See the thread beginning
674 * <20140113145929.GF27210@iabyn.com>
676 assert(prog->substrs->data[0].min_offset >= 0);
677 assert(prog->substrs->data[0].max_offset >= 0);
678 assert(prog->substrs->data[1].min_offset >= 0);
679 assert(prog->substrs->data[1].max_offset >= 0);
680 assert(prog->substrs->data[2].min_offset >= 0);
681 assert(prog->substrs->data[2].max_offset >= 0);
683 /* for now, assume that if both present, that the floating substring
684 * doesn't start before the anchored substring.
685 * If you break this assumption (e.g. doing better optimisations
686 * with lookahead/behind), then you'll need to audit the code in this
687 * function carefully first
690 ! ( (prog->anchored_utf8 || prog->anchored_substr)
691 && (prog->float_utf8 || prog->float_substr))
692 || (prog->float_min_offset >= prog->anchored_offset));
694 /* byte rather than char calculation for efficiency. It fails
695 * to quickly reject some cases that can't match, but will reject
696 * them later after doing full char arithmetic */
697 if (prog->minlen > strend - strpos) {
698 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
699 " String too short...\n"));
703 RX_MATCH_UTF8_set(rx,utf8_target);
704 reginfo->is_utf8_target = cBOOL(utf8_target);
705 reginfo->info_aux = NULL;
706 reginfo->strbeg = strbeg;
707 reginfo->strend = strend;
708 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
710 /* not actually used within intuit, but zero for safety anyway */
711 reginfo->poscache_maxiter = 0;
714 if ((!prog->anchored_utf8 && prog->anchored_substr)
715 || (!prog->float_utf8 && prog->float_substr))
716 to_utf8_substr(prog);
717 check = prog->check_utf8;
719 if (!prog->check_substr && prog->check_utf8) {
720 if (! to_byte_substr(prog)) {
721 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
724 check = prog->check_substr;
727 /* dump the various substring data */
728 DEBUG_OPTIMISE_MORE_r({
730 for (i=0; i<=2; i++) {
731 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
732 : prog->substrs->data[i].substr);
736 Perl_re_printf( aTHX_
737 " substrs[%d]: min=%" IVdf " max=%" IVdf " end shift=%" IVdf
738 " useful=%" IVdf " utf8=%d [%s]\n",
740 (IV)prog->substrs->data[i].min_offset,
741 (IV)prog->substrs->data[i].max_offset,
742 (IV)prog->substrs->data[i].end_shift,
749 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
751 /* ml_anch: check after \n?
753 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
754 * with /.*.../, these flags will have been added by the
756 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
757 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
759 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
760 && !(prog->intflags & PREGf_IMPLICIT);
762 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
763 /* we are only allowed to match at BOS or \G */
765 /* trivially reject if there's a BOS anchor and we're not at BOS.
767 * Note that we don't try to do a similar quick reject for
768 * \G, since generally the caller will have calculated strpos
769 * based on pos() and gofs, so the string is already correctly
770 * anchored by definition; and handling the exceptions would
771 * be too fiddly (e.g. REXEC_IGNOREPOS).
773 if ( strpos != strbeg
774 && (prog->intflags & PREGf_ANCH_SBOL))
776 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
777 " Not at start...\n"));
781 /* in the presence of an anchor, the anchored (relative to the
782 * start of the regex) substr must also be anchored relative
783 * to strpos. So quickly reject if substr isn't found there.
784 * This works for \G too, because the caller will already have
785 * subtracted gofs from pos, and gofs is the offset from the
786 * \G to the start of the regex. For example, in /.abc\Gdef/,
787 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
788 * caller will have set strpos=pos()-4; we look for the substr
789 * at position pos()-4+1, which lines up with the "a" */
791 if (prog->check_offset_min == prog->check_offset_max) {
792 /* Substring at constant offset from beg-of-str... */
793 SSize_t slen = SvCUR(check);
794 char *s = HOP3c(strpos, prog->check_offset_min, strend);
796 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
797 " Looking for check substr at fixed offset %" IVdf "...\n",
798 (IV)prog->check_offset_min));
801 /* In this case, the regex is anchored at the end too.
802 * Unless it's a multiline match, the lengths must match
803 * exactly, give or take a \n. NB: slen >= 1 since
804 * the last char of check is \n */
806 && ( strend - s > slen
807 || strend - s < slen - 1
808 || (strend - s == slen && strend[-1] != '\n')))
810 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
811 " String too long...\n"));
814 /* Now should match s[0..slen-2] */
817 if (slen && (strend - s < slen
818 || *SvPVX_const(check) != *s
819 || (slen > 1 && (memNE(SvPVX_const(check), s, slen)))))
821 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
822 " String not equal...\n"));
827 goto success_at_start;
832 end_shift = prog->check_end_shift;
834 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
836 Perl_croak(aTHX_ "panic: end_shift: %" IVdf " pattern:\n%s\n ",
837 (IV)end_shift, RX_PRECOMP(prog));
842 /* This is the (re)entry point of the main loop in this function.
843 * The goal of this loop is to:
844 * 1) find the "check" substring in the region rx_origin..strend
845 * (adjusted by start_shift / end_shift). If not found, reject
847 * 2) If it exists, look for the "other" substr too if defined; for
848 * example, if the check substr maps to the anchored substr, then
849 * check the floating substr, and vice-versa. If not found, go
850 * back to (1) with rx_origin suitably incremented.
851 * 3) If we find an rx_origin position that doesn't contradict
852 * either of the substrings, then check the possible additional
853 * constraints on rx_origin of /^.../m or a known start class.
854 * If these fail, then depending on which constraints fail, jump
855 * back to here, or to various other re-entry points further along
856 * that skip some of the first steps.
857 * 4) If we pass all those tests, update the BmUSEFUL() count on the
858 * substring. If the start position was determined to be at the
859 * beginning of the string - so, not rejected, but not optimised,
860 * since we have to run regmatch from position 0 - decrement the
861 * BmUSEFUL() count. Otherwise increment it.
865 /* first, look for the 'check' substring */
871 DEBUG_OPTIMISE_MORE_r({
872 Perl_re_printf( aTHX_
873 " At restart: rx_origin=%" IVdf " Check offset min: %" IVdf
874 " Start shift: %" IVdf " End shift %" IVdf
875 " Real end Shift: %" IVdf "\n",
876 (IV)(rx_origin - strbeg),
877 (IV)prog->check_offset_min,
880 (IV)prog->check_end_shift);
883 end_point = HOP3(strend, -end_shift, strbeg);
884 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
889 /* If the regex is absolutely anchored to either the start of the
890 * string (SBOL) or to pos() (ANCH_GPOS), then
891 * check_offset_max represents an upper bound on the string where
892 * the substr could start. For the ANCH_GPOS case, we assume that
893 * the caller of intuit will have already set strpos to
894 * pos()-gofs, so in this case strpos + offset_max will still be
895 * an upper bound on the substr.
898 && prog->intflags & PREGf_ANCH
899 && prog->check_offset_max != SSize_t_MAX)
901 SSize_t len = SvCUR(check) - !!SvTAIL(check);
902 const char * const anchor =
903 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
905 /* do a bytes rather than chars comparison. It's conservative;
906 * so it skips doing the HOP if the result can't possibly end
907 * up earlier than the old value of end_point.
909 if ((char*)end_point - anchor > prog->check_offset_max) {
910 end_point = HOP3lim((U8*)anchor,
911 prog->check_offset_max,
917 check_at = fbm_instr( start_point, end_point,
918 check, multiline ? FBMrf_MULTILINE : 0);
920 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
921 " doing 'check' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
922 (IV)((char*)start_point - strbeg),
923 (IV)((char*)end_point - strbeg),
924 (IV)(check_at ? check_at - strbeg : -1)
927 /* Update the count-of-usability, remove useless subpatterns,
931 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
932 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
933 Perl_re_printf( aTHX_ " %s %s substr %s%s%s",
934 (check_at ? "Found" : "Did not find"),
935 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
936 ? "anchored" : "floating"),
939 (check_at ? " at offset " : "...\n") );
944 /* set rx_origin to the minimum position where the regex could start
945 * matching, given the constraint of the just-matched check substring.
946 * But don't set it lower than previously.
949 if (check_at - rx_origin > prog->check_offset_max)
950 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
951 /* Finish the diagnostic message */
952 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
953 "%ld (rx_origin now %" IVdf ")...\n",
954 (long)(check_at - strbeg),
955 (IV)(rx_origin - strbeg)
960 /* now look for the 'other' substring if defined */
962 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
963 : prog->substrs->data[other_ix].substr)
965 /* Take into account the "other" substring. */
969 struct reg_substr_datum *other;
972 other = &prog->substrs->data[other_ix];
974 /* if "other" is anchored:
975 * we've previously found a floating substr starting at check_at.
976 * This means that the regex origin must lie somewhere
977 * between min (rx_origin): HOP3(check_at, -check_offset_max)
978 * and max: HOP3(check_at, -check_offset_min)
979 * (except that min will be >= strpos)
980 * So the fixed substr must lie somewhere between
981 * HOP3(min, anchored_offset)
982 * HOP3(max, anchored_offset) + SvCUR(substr)
985 /* if "other" is floating
986 * Calculate last1, the absolute latest point where the
987 * floating substr could start in the string, ignoring any
988 * constraints from the earlier fixed match. It is calculated
991 * strend - prog->minlen (in chars) is the absolute latest
992 * position within the string where the origin of the regex
993 * could appear. The latest start point for the floating
994 * substr is float_min_offset(*) on from the start of the
995 * regex. last1 simply combines thee two offsets.
997 * (*) You might think the latest start point should be
998 * float_max_offset from the regex origin, and technically
999 * you'd be correct. However, consider
1001 * Here, float min, max are 3,5 and minlen is 7.
1002 * This can match either
1006 * In the first case, the regex matches minlen chars; in the
1007 * second, minlen+1, in the third, minlen+2.
1008 * In the first case, the floating offset is 3 (which equals
1009 * float_min), in the second, 4, and in the third, 5 (which
1010 * equals float_max). In all cases, the floating string bcd
1011 * can never start more than 4 chars from the end of the
1012 * string, which equals minlen - float_min. As the substring
1013 * starts to match more than float_min from the start of the
1014 * regex, it makes the regex match more than minlen chars,
1015 * and the two cancel each other out. So we can always use
1016 * float_min - minlen, rather than float_max - minlen for the
1017 * latest position in the string.
1019 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1020 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1023 assert(prog->minlen >= other->min_offset);
1024 last1 = HOP3c(strend,
1025 other->min_offset - prog->minlen, strbeg);
1027 if (other_ix) {/* i.e. if (other-is-float) */
1028 /* last is the latest point where the floating substr could
1029 * start, *given* any constraints from the earlier fixed
1030 * match. This constraint is that the floating string starts
1031 * <= float_max_offset chars from the regex origin (rx_origin).
1032 * If this value is less than last1, use it instead.
1034 assert(rx_origin <= last1);
1036 /* this condition handles the offset==infinity case, and
1037 * is a short-cut otherwise. Although it's comparing a
1038 * byte offset to a char length, it does so in a safe way,
1039 * since 1 char always occupies 1 or more bytes,
1040 * so if a string range is (last1 - rx_origin) bytes,
1041 * it will be less than or equal to (last1 - rx_origin)
1042 * chars; meaning it errs towards doing the accurate HOP3
1043 * rather than just using last1 as a short-cut */
1044 (last1 - rx_origin) < other->max_offset
1046 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1049 assert(strpos + start_shift <= check_at);
1050 last = HOP4c(check_at, other->min_offset - start_shift,
1054 s = HOP3c(rx_origin, other->min_offset, strend);
1055 if (s < other_last) /* These positions already checked */
1058 must = utf8_target ? other->utf8_substr : other->substr;
1059 assert(SvPOK(must));
1062 char *to = last + SvCUR(must) - (SvTAIL(must)!=0);
1068 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1069 " skipping 'other' fbm scan: %" IVdf " > %" IVdf "\n",
1070 (IV)(from - strbeg),
1076 (unsigned char*)from,
1079 multiline ? FBMrf_MULTILINE : 0
1081 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1082 " doing 'other' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
1083 (IV)(from - strbeg),
1085 (IV)(s ? s - strbeg : -1)
1091 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1092 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1093 Perl_re_printf( aTHX_ " %s %s substr %s%s",
1094 s ? "Found" : "Contradicts",
1095 other_ix ? "floating" : "anchored",
1096 quoted, RE_SV_TAIL(must));
1101 /* last1 is latest possible substr location. If we didn't
1102 * find it before there, we never will */
1103 if (last >= last1) {
1104 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1105 "; giving up...\n"));
1109 /* try to find the check substr again at a later
1110 * position. Maybe next time we'll find the "other" substr
1112 other_last = HOP3c(last, 1, strend) /* highest failure */;
1114 other_ix /* i.e. if other-is-float */
1115 ? HOP3c(rx_origin, 1, strend)
1116 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1117 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1118 "; about to retry %s at offset %ld (rx_origin now %" IVdf ")...\n",
1119 (other_ix ? "floating" : "anchored"),
1120 (long)(HOP3c(check_at, 1, strend) - strbeg),
1121 (IV)(rx_origin - strbeg)
1126 if (other_ix) { /* if (other-is-float) */
1127 /* other_last is set to s, not s+1, since its possible for
1128 * a floating substr to fail first time, then succeed
1129 * second time at the same floating position; e.g.:
1130 * "-AB--AABZ" =~ /\wAB\d*Z/
1131 * The first time round, anchored and float match at
1132 * "-(AB)--AAB(Z)" then fail on the initial \w character
1133 * class. Second time round, they match at "-AB--A(AB)(Z)".
1138 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1139 other_last = HOP3c(s, 1, strend);
1141 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1142 " at offset %ld (rx_origin now %" IVdf ")...\n",
1144 (IV)(rx_origin - strbeg)
1150 DEBUG_OPTIMISE_MORE_r(
1151 Perl_re_printf( aTHX_
1152 " Check-only match: offset min:%" IVdf " max:%" IVdf
1153 " check_at:%" IVdf " rx_origin:%" IVdf " rx_origin-check_at:%" IVdf
1154 " strend:%" IVdf "\n",
1155 (IV)prog->check_offset_min,
1156 (IV)prog->check_offset_max,
1157 (IV)(check_at-strbeg),
1158 (IV)(rx_origin-strbeg),
1159 (IV)(rx_origin-check_at),
1165 postprocess_substr_matches:
1167 /* handle the extra constraint of /^.../m if present */
1169 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1172 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1173 " looking for /^/m anchor"));
1175 /* we have failed the constraint of a \n before rx_origin.
1176 * Find the next \n, if any, even if it's beyond the current
1177 * anchored and/or floating substrings. Whether we should be
1178 * scanning ahead for the next \n or the next substr is debatable.
1179 * On the one hand you'd expect rare substrings to appear less
1180 * often than \n's. On the other hand, searching for \n means
1181 * we're effectively flipping between check_substr and "\n" on each
1182 * iteration as the current "rarest" string candidate, which
1183 * means for example that we'll quickly reject the whole string if
1184 * hasn't got a \n, rather than trying every substr position
1188 s = HOP3c(strend, - prog->minlen, strpos);
1189 if (s <= rx_origin ||
1190 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1192 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1193 " Did not find /%s^%s/m...\n",
1194 PL_colors[0], PL_colors[1]));
1198 /* earliest possible origin is 1 char after the \n.
1199 * (since *rx_origin == '\n', it's safe to ++ here rather than
1200 * HOP(rx_origin, 1)) */
1203 if (prog->substrs->check_ix == 0 /* check is anchored */
1204 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1206 /* Position contradicts check-string; either because
1207 * check was anchored (and thus has no wiggle room),
1208 * or check was float and rx_origin is above the float range */
1209 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1210 " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
1211 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1215 /* if we get here, the check substr must have been float,
1216 * is in range, and we may or may not have had an anchored
1217 * "other" substr which still contradicts */
1218 assert(prog->substrs->check_ix); /* check is float */
1220 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1221 /* whoops, the anchored "other" substr exists, so we still
1222 * contradict. On the other hand, the float "check" substr
1223 * didn't contradict, so just retry the anchored "other"
1225 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1226 " Found /%s^%s/m, rescanning for anchored from offset %" IVdf " (rx_origin now %" IVdf ")...\n",
1227 PL_colors[0], PL_colors[1],
1228 (IV)(rx_origin - strbeg + prog->anchored_offset),
1229 (IV)(rx_origin - strbeg)
1231 goto do_other_substr;
1234 /* success: we don't contradict the found floating substring
1235 * (and there's no anchored substr). */
1236 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1237 " Found /%s^%s/m with rx_origin %ld...\n",
1238 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1241 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1242 " (multiline anchor test skipped)\n"));
1248 /* if we have a starting character class, then test that extra constraint.
1249 * (trie stclasses are too expensive to use here, we are better off to
1250 * leave it to regmatch itself) */
1252 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1253 const U8* const str = (U8*)STRING(progi->regstclass);
1255 /* XXX this value could be pre-computed */
1256 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1257 ? (reginfo->is_utf8_pat
1258 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1259 : STR_LEN(progi->regstclass))
1263 /* latest pos that a matching float substr constrains rx start to */
1264 char *rx_max_float = NULL;
1266 /* if the current rx_origin is anchored, either by satisfying an
1267 * anchored substring constraint, or a /^.../m constraint, then we
1268 * can reject the current origin if the start class isn't found
1269 * at the current position. If we have a float-only match, then
1270 * rx_origin is constrained to a range; so look for the start class
1271 * in that range. if neither, then look for the start class in the
1272 * whole rest of the string */
1274 /* XXX DAPM it's not clear what the minlen test is for, and why
1275 * it's not used in the floating case. Nothing in the test suite
1276 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1277 * Here are some old comments, which may or may not be correct:
1279 * minlen == 0 is possible if regstclass is \b or \B,
1280 * and the fixed substr is ''$.
1281 * Since minlen is already taken into account, rx_origin+1 is
1282 * before strend; accidentally, minlen >= 1 guaranties no false
1283 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1284 * 0) below assumes that regstclass does not come from lookahead...
1285 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1286 * This leaves EXACTF-ish only, which are dealt with in
1290 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1291 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1292 else if (prog->float_substr || prog->float_utf8) {
1293 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1294 endpos= HOP3c(rx_max_float, cl_l, strend);
1299 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1300 " looking for class: start_shift: %" IVdf " check_at: %" IVdf
1301 " rx_origin: %" IVdf " endpos: %" IVdf "\n",
1302 (IV)start_shift, (IV)(check_at - strbeg),
1303 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1305 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1308 if (endpos == strend) {
1309 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1310 " Could not match STCLASS...\n") );
1313 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1314 " This position contradicts STCLASS...\n") );
1315 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1316 && !(prog->intflags & PREGf_IMPLICIT))
1319 /* Contradict one of substrings */
1320 if (prog->anchored_substr || prog->anchored_utf8) {
1321 if (prog->substrs->check_ix == 1) { /* check is float */
1322 /* Have both, check_string is floating */
1323 assert(rx_origin + start_shift <= check_at);
1324 if (rx_origin + start_shift != check_at) {
1325 /* not at latest position float substr could match:
1326 * Recheck anchored substring, but not floating.
1327 * The condition above is in bytes rather than
1328 * chars for efficiency. It's conservative, in
1329 * that it errs on the side of doing 'goto
1330 * do_other_substr'. In this case, at worst,
1331 * an extra anchored search may get done, but in
1332 * practice the extra fbm_instr() is likely to
1333 * get skipped anyway. */
1334 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1335 " about to retry anchored at offset %ld (rx_origin now %" IVdf ")...\n",
1336 (long)(other_last - strbeg),
1337 (IV)(rx_origin - strbeg)
1339 goto do_other_substr;
1347 /* In the presence of ml_anch, we might be able to
1348 * find another \n without breaking the current float
1351 /* strictly speaking this should be HOP3c(..., 1, ...),
1352 * but since we goto a block of code that's going to
1353 * search for the next \n if any, its safe here */
1355 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1356 " about to look for /%s^%s/m starting at rx_origin %ld...\n",
1357 PL_colors[0], PL_colors[1],
1358 (long)(rx_origin - strbeg)) );
1359 goto postprocess_substr_matches;
1362 /* strictly speaking this can never be true; but might
1363 * be if we ever allow intuit without substrings */
1364 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1367 rx_origin = rx_max_float;
1370 /* at this point, any matching substrings have been
1371 * contradicted. Start again... */
1373 rx_origin = HOP3c(rx_origin, 1, strend);
1375 /* uses bytes rather than char calculations for efficiency.
1376 * It's conservative: it errs on the side of doing 'goto restart',
1377 * where there is code that does a proper char-based test */
1378 if (rx_origin + start_shift + end_shift > strend) {
1379 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1380 " Could not match STCLASS...\n") );
1383 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1384 " about to look for %s substr starting at offset %ld (rx_origin now %" IVdf ")...\n",
1385 (prog->substrs->check_ix ? "floating" : "anchored"),
1386 (long)(rx_origin + start_shift - strbeg),
1387 (IV)(rx_origin - strbeg)
1394 if (rx_origin != s) {
1395 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1396 " By STCLASS: moving %ld --> %ld\n",
1397 (long)(rx_origin - strbeg), (long)(s - strbeg))
1401 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1402 " Does not contradict STCLASS...\n");
1407 /* Decide whether using the substrings helped */
1409 if (rx_origin != strpos) {
1410 /* Fixed substring is found far enough so that the match
1411 cannot start at strpos. */
1413 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " try at offset...\n"));
1414 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1417 /* The found rx_origin position does not prohibit matching at
1418 * strpos, so calling intuit didn't gain us anything. Decrement
1419 * the BmUSEFUL() count on the check substring, and if we reach
1421 if (!(prog->intflags & PREGf_NAUGHTY)
1423 prog->check_utf8 /* Could be deleted already */
1424 && --BmUSEFUL(prog->check_utf8) < 0
1425 && (prog->check_utf8 == prog->float_utf8)
1427 prog->check_substr /* Could be deleted already */
1428 && --BmUSEFUL(prog->check_substr) < 0
1429 && (prog->check_substr == prog->float_substr)
1432 /* If flags & SOMETHING - do not do it many times on the same match */
1433 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " ... Disabling check substring...\n"));
1434 /* XXX Does the destruction order has to change with utf8_target? */
1435 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1436 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1437 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1438 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1439 check = NULL; /* abort */
1440 /* XXXX This is a remnant of the old implementation. It
1441 looks wasteful, since now INTUIT can use many
1442 other heuristics. */
1443 prog->extflags &= ~RXf_USE_INTUIT;
1447 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1448 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1449 PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );
1453 fail_finish: /* Substring not found */
1454 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1455 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1457 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch rejected by optimizer%s\n",
1458 PL_colors[4], PL_colors[5]));
1463 #define DECL_TRIE_TYPE(scan) \
1464 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1465 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1466 trie_utf8l, trie_flu8 } \
1467 trie_type = ((scan->flags == EXACT) \
1468 ? (utf8_target ? trie_utf8 : trie_plain) \
1469 : (scan->flags == EXACTL) \
1470 ? (utf8_target ? trie_utf8l : trie_plain) \
1471 : (scan->flags == EXACTFA) \
1473 ? trie_utf8_exactfa_fold \
1474 : trie_latin_utf8_exactfa_fold) \
1475 : (scan->flags == EXACTFLU8 \
1479 : trie_latin_utf8_fold)))
1481 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1484 U8 flags = FOLD_FLAGS_FULL; \
1485 switch (trie_type) { \
1487 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1488 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1489 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1491 goto do_trie_utf8_fold; \
1492 case trie_utf8_exactfa_fold: \
1493 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1495 case trie_utf8_fold: \
1496 do_trie_utf8_fold: \
1497 if ( foldlen>0 ) { \
1498 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1503 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1504 len = UTF8SKIP(uc); \
1505 skiplen = UVCHR_SKIP( uvc ); \
1506 foldlen -= skiplen; \
1507 uscan = foldbuf + skiplen; \
1510 case trie_latin_utf8_exactfa_fold: \
1511 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1513 case trie_latin_utf8_fold: \
1514 if ( foldlen>0 ) { \
1515 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1521 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1522 skiplen = UVCHR_SKIP( uvc ); \
1523 foldlen -= skiplen; \
1524 uscan = foldbuf + skiplen; \
1528 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1529 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1530 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1534 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1541 charid = trie->charmap[ uvc ]; \
1545 if (widecharmap) { \
1546 SV** const svpp = hv_fetch(widecharmap, \
1547 (char*)&uvc, sizeof(UV), 0); \
1549 charid = (U16)SvIV(*svpp); \
1554 #define DUMP_EXEC_POS(li,s,doutf8,depth) \
1555 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1556 startpos, doutf8, depth)
1558 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1562 && (ln == 1 || folder(s, pat_string, ln)) \
1563 && (reginfo->intuit || regtry(reginfo, &s)) )\
1569 #define REXEC_FBC_UTF8_SCAN(CODE) \
1571 while (s < strend) { \
1577 #define REXEC_FBC_SCAN(CODE) \
1579 while (s < strend) { \
1585 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1586 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1588 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1597 #define REXEC_FBC_CLASS_SCAN(COND) \
1598 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1600 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1609 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1610 if (utf8_target) { \
1611 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1614 REXEC_FBC_CLASS_SCAN(COND); \
1617 /* The three macros below are slightly different versions of the same logic.
1619 * The first is for /a and /aa when the target string is UTF-8. This can only
1620 * match ascii, but it must advance based on UTF-8. The other two handle the
1621 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1622 * for the boundary (or non-boundary) between a word and non-word character.
1623 * The utf8 and non-utf8 cases have the same logic, but the details must be
1624 * different. Find the "wordness" of the character just prior to this one, and
1625 * compare it with the wordness of this one. If they differ, we have a
1626 * boundary. At the beginning of the string, pretend that the previous
1627 * character was a new-line.
1629 * All these macros uncleanly have side-effects with each other and outside
1630 * variables. So far it's been too much trouble to clean-up
1632 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1633 * a word character or not.
1634 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1636 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1638 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1639 * are looking for a boundary or for a non-boundary. If we are looking for a
1640 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1641 * see if this tentative match actually works, and if so, to quit the loop
1642 * here. And vice-versa if we are looking for a non-boundary.
1644 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1645 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1646 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1647 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1648 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1649 * complement. But in that branch we complement tmp, meaning that at the
1650 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1651 * which means at the top of the loop in the next iteration, it is
1652 * TEST_NON_UTF8(s-1) */
1653 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1654 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1655 tmp = TEST_NON_UTF8(tmp); \
1656 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1657 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1659 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1666 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1667 * TEST_UTF8 is a macro that for the same input code points returns identically
1668 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1669 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1670 if (s == reginfo->strbeg) { \
1673 else { /* Back-up to the start of the previous character */ \
1674 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1675 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1676 0, UTF8_ALLOW_DEFAULT); \
1678 tmp = TEST_UV(tmp); \
1679 LOAD_UTF8_CHARCLASS_ALNUM(); \
1680 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1681 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1690 /* Like the above two macros. UTF8_CODE is the complete code for handling
1691 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1693 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1694 if (utf8_target) { \
1697 else { /* Not utf8 */ \
1698 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1699 tmp = TEST_NON_UTF8(tmp); \
1700 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1701 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1710 /* Here, things have been set up by the previous code so that tmp is the \
1711 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1712 * utf8ness of the target). We also have to check if this matches against \
1713 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1714 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1716 if (tmp == ! TEST_NON_UTF8('\n')) { \
1723 /* This is the macro to use when we want to see if something that looks like it
1724 * could match, actually does, and if so exits the loop */
1725 #define REXEC_FBC_TRYIT \
1726 if ((reginfo->intuit || regtry(reginfo, &s))) \
1729 /* The only difference between the BOUND and NBOUND cases is that
1730 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1731 * NBOUND. This is accomplished by passing it as either the if or else clause,
1732 * with the other one being empty (PLACEHOLDER is defined as empty).
1734 * The TEST_FOO parameters are for operating on different forms of input, but
1735 * all should be ones that return identically for the same underlying code
1737 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1739 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1740 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1742 #define FBC_BOUND_A(TEST_NON_UTF8) \
1744 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1745 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1747 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1749 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1750 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1752 #define FBC_NBOUND_A(TEST_NON_UTF8) \
1754 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1755 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1759 S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) {
1760 IV cp_out = Perl__invlist_search(invlist, cp_in);
1761 assert(cp_out >= 0);
1764 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
1765 invmap[S_get_break_val_cp_checked(invlist, cp)]
1767 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
1768 invmap[_invlist_search(invlist, cp)]
1771 /* Takes a pointer to an inversion list, a pointer to its corresponding
1772 * inversion map, and a code point, and returns the code point's value
1773 * according to the two arrays. It assumes that all code points have a value.
1774 * This is used as the base macro for macros for particular properties */
1775 #define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \
1776 _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp)
1778 /* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
1779 * of a code point, returning the value for the first code point in the string.
1780 * And it takes the particular macro name that finds the desired value given a
1781 * code point. Merely convert the UTF-8 to code point and call the cp macro */
1782 #define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \
1783 (__ASSERT_(pos < strend) \
1784 /* Note assumes is valid UTF-8 */ \
1785 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))
1787 /* Returns the GCB value for the input code point */
1788 #define getGCB_VAL_CP(cp) \
1789 _generic_GET_BREAK_VAL_CP( \
1794 /* Returns the GCB value for the first code point in the UTF-8 encoded string
1795 * bounded by pos and strend */
1796 #define getGCB_VAL_UTF8(pos, strend) \
1797 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)
1799 /* Returns the LB value for the input code point */
1800 #define getLB_VAL_CP(cp) \
1801 _generic_GET_BREAK_VAL_CP( \
1806 /* Returns the LB value for the first code point in the UTF-8 encoded string
1807 * bounded by pos and strend */
1808 #define getLB_VAL_UTF8(pos, strend) \
1809 _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend)
1812 /* Returns the SB value for the input code point */
1813 #define getSB_VAL_CP(cp) \
1814 _generic_GET_BREAK_VAL_CP( \
1819 /* Returns the SB value for the first code point in the UTF-8 encoded string
1820 * bounded by pos and strend */
1821 #define getSB_VAL_UTF8(pos, strend) \
1822 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)
1824 /* Returns the WB value for the input code point */
1825 #define getWB_VAL_CP(cp) \
1826 _generic_GET_BREAK_VAL_CP( \
1831 /* Returns the WB value for the first code point in the UTF-8 encoded string
1832 * bounded by pos and strend */
1833 #define getWB_VAL_UTF8(pos, strend) \
1834 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)
1836 /* We know what class REx starts with. Try to find this position... */
1837 /* if reginfo->intuit, its a dryrun */
1838 /* annoyingly all the vars in this routine have different names from their counterparts
1839 in regmatch. /grrr */
1841 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1842 const char *strend, regmatch_info *reginfo)
1845 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1846 char *pat_string; /* The pattern's exactish string */
1847 char *pat_end; /* ptr to end char of pat_string */
1848 re_fold_t folder; /* Function for computing non-utf8 folds */
1849 const U8 *fold_array; /* array for folding ords < 256 */
1855 I32 tmp = 1; /* Scratch variable? */
1856 const bool utf8_target = reginfo->is_utf8_target;
1857 UV utf8_fold_flags = 0;
1858 const bool is_utf8_pat = reginfo->is_utf8_pat;
1859 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1860 with a result inverts that result, as 0^1 =
1862 _char_class_number classnum;
1864 RXi_GET_DECL(prog,progi);
1866 PERL_ARGS_ASSERT_FIND_BYCLASS;
1868 /* We know what class it must start with. */
1871 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1873 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) {
1874 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
1881 REXEC_FBC_UTF8_CLASS_SCAN(
1882 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1884 else if (ANYOF_FLAGS(c)) {
1885 REXEC_FBC_CLASS_SCAN(reginclass(prog,c, (U8*)s, (U8*)s+1, 0));
1888 REXEC_FBC_CLASS_SCAN(ANYOF_BITMAP_TEST(c, *((U8*)s)));
1892 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1893 assert(! is_utf8_pat);
1896 if (is_utf8_pat || utf8_target) {
1897 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1898 goto do_exactf_utf8;
1900 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1901 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1902 goto do_exactf_non_utf8; /* isn't dealt with by these */
1904 case EXACTF: /* This node only generated for non-utf8 patterns */
1905 assert(! is_utf8_pat);
1907 utf8_fold_flags = 0;
1908 goto do_exactf_utf8;
1910 fold_array = PL_fold;
1912 goto do_exactf_non_utf8;
1915 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1916 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1917 utf8_fold_flags = FOLDEQ_LOCALE;
1918 goto do_exactf_utf8;
1920 fold_array = PL_fold_locale;
1921 folder = foldEQ_locale;
1922 goto do_exactf_non_utf8;
1926 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1928 goto do_exactf_utf8;
1931 if (! utf8_target) { /* All code points in this node require
1932 UTF-8 to express. */
1935 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
1936 | FOLDEQ_S2_FOLDS_SANE;
1937 goto do_exactf_utf8;
1940 if (is_utf8_pat || utf8_target) {
1941 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1942 goto do_exactf_utf8;
1945 /* Any 'ss' in the pattern should have been replaced by regcomp,
1946 * so we don't have to worry here about this single special case
1947 * in the Latin1 range */
1948 fold_array = PL_fold_latin1;
1949 folder = foldEQ_latin1;
1953 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1954 are no glitches with fold-length differences
1955 between the target string and pattern */
1957 /* The idea in the non-utf8 EXACTF* cases is to first find the
1958 * first character of the EXACTF* node and then, if necessary,
1959 * case-insensitively compare the full text of the node. c1 is the
1960 * first character. c2 is its fold. This logic will not work for
1961 * Unicode semantics and the german sharp ss, which hence should
1962 * not be compiled into a node that gets here. */
1963 pat_string = STRING(c);
1964 ln = STR_LEN(c); /* length to match in octets/bytes */
1966 /* We know that we have to match at least 'ln' bytes (which is the
1967 * same as characters, since not utf8). If we have to match 3
1968 * characters, and there are only 2 availabe, we know without
1969 * trying that it will fail; so don't start a match past the
1970 * required minimum number from the far end */
1971 e = HOP3c(strend, -((SSize_t)ln), s);
1973 if (reginfo->intuit && e < s) {
1974 e = s; /* Due to minlen logic of intuit() */
1978 c2 = fold_array[c1];
1979 if (c1 == c2) { /* If char and fold are the same */
1980 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1983 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1991 /* If one of the operands is in utf8, we can't use the simpler folding
1992 * above, due to the fact that many different characters can have the
1993 * same fold, or portion of a fold, or different- length fold */
1994 pat_string = STRING(c);
1995 ln = STR_LEN(c); /* length to match in octets/bytes */
1996 pat_end = pat_string + ln;
1997 lnc = is_utf8_pat /* length to match in characters */
1998 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
2001 /* We have 'lnc' characters to match in the pattern, but because of
2002 * multi-character folding, each character in the target can match
2003 * up to 3 characters (Unicode guarantees it will never exceed
2004 * this) if it is utf8-encoded; and up to 2 if not (based on the
2005 * fact that the Latin 1 folds are already determined, and the
2006 * only multi-char fold in that range is the sharp-s folding to
2007 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
2008 * string character. Adjust lnc accordingly, rounding up, so that
2009 * if we need to match at least 4+1/3 chars, that really is 5. */
2010 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
2011 lnc = (lnc + expansion - 1) / expansion;
2013 /* As in the non-UTF8 case, if we have to match 3 characters, and
2014 * only 2 are left, it's guaranteed to fail, so don't start a
2015 * match that would require us to go beyond the end of the string
2017 e = HOP3c(strend, -((SSize_t)lnc), s);
2019 if (reginfo->intuit && e < s) {
2020 e = s; /* Due to minlen logic of intuit() */
2023 /* XXX Note that we could recalculate e to stop the loop earlier,
2024 * as the worst case expansion above will rarely be met, and as we
2025 * go along we would usually find that e moves further to the left.
2026 * This would happen only after we reached the point in the loop
2027 * where if there were no expansion we should fail. Unclear if
2028 * worth the expense */
2031 char *my_strend= (char *)strend;
2032 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
2033 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
2034 && (reginfo->intuit || regtry(reginfo, &s)) )
2038 s += (utf8_target) ? UTF8SKIP(s) : 1;
2044 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2045 if (FLAGS(c) != TRADITIONAL_BOUND) {
2046 if (! IN_UTF8_CTYPE_LOCALE) {
2047 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2048 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2053 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
2057 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2058 if (FLAGS(c) != TRADITIONAL_BOUND) {
2059 if (! IN_UTF8_CTYPE_LOCALE) {
2060 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2061 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2066 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
2069 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2071 assert(FLAGS(c) == TRADITIONAL_BOUND);
2073 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
2076 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2078 assert(FLAGS(c) == TRADITIONAL_BOUND);
2080 FBC_BOUND_A(isWORDCHAR_A);
2083 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2085 assert(FLAGS(c) == TRADITIONAL_BOUND);
2087 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
2090 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2092 assert(FLAGS(c) == TRADITIONAL_BOUND);
2094 FBC_NBOUND_A(isWORDCHAR_A);
2098 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
2099 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
2110 switch((bound_type) FLAGS(c)) {
2111 case TRADITIONAL_BOUND:
2112 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
2115 if (s == reginfo->strbeg) {
2116 if (reginfo->intuit || regtry(reginfo, &s))
2121 /* Didn't match. Try at the next position (if there is one) */
2122 s += (utf8_target) ? UTF8SKIP(s) : 1;
2123 if (UNLIKELY(s >= reginfo->strend)) {
2129 GCB_enum before = getGCB_VAL_UTF8(
2131 (U8*)(reginfo->strbeg)),
2132 (U8*) reginfo->strend);
2133 while (s < strend) {
2134 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
2135 (U8*) reginfo->strend);
2136 if ( (to_complement ^ isGCB(before,
2138 (U8*) reginfo->strbeg,
2141 && (reginfo->intuit || regtry(reginfo, &s)))
2149 else { /* Not utf8. Everything is a GCB except between CR and
2151 while (s < strend) {
2152 if ((to_complement ^ ( UCHARAT(s - 1) != '\r'
2153 || UCHARAT(s) != '\n'))
2154 && (reginfo->intuit || regtry(reginfo, &s)))
2162 /* And, since this is a bound, it can match after the final
2163 * character in the string */
2164 if ((reginfo->intuit || regtry(reginfo, &s))) {
2170 if (s == reginfo->strbeg) {
2171 if (reginfo->intuit || regtry(reginfo, &s)) {
2174 s += (utf8_target) ? UTF8SKIP(s) : 1;
2175 if (UNLIKELY(s >= reginfo->strend)) {
2181 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
2183 (U8*)(reginfo->strbeg)),
2184 (U8*) reginfo->strend);
2185 while (s < strend) {
2186 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
2187 if (to_complement ^ isLB(before,
2189 (U8*) reginfo->strbeg,
2191 (U8*) reginfo->strend,
2193 && (reginfo->intuit || regtry(reginfo, &s)))
2201 else { /* Not utf8. */
2202 LB_enum before = getLB_VAL_CP((U8) *(s -1));
2203 while (s < strend) {
2204 LB_enum after = getLB_VAL_CP((U8) *s);
2205 if (to_complement ^ isLB(before,
2207 (U8*) reginfo->strbeg,
2209 (U8*) reginfo->strend,
2211 && (reginfo->intuit || regtry(reginfo, &s)))
2220 if (reginfo->intuit || regtry(reginfo, &s)) {
2227 if (s == reginfo->strbeg) {
2228 if (reginfo->intuit || regtry(reginfo, &s)) {
2231 s += (utf8_target) ? UTF8SKIP(s) : 1;
2232 if (UNLIKELY(s >= reginfo->strend)) {
2238 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
2240 (U8*)(reginfo->strbeg)),
2241 (U8*) reginfo->strend);
2242 while (s < strend) {
2243 SB_enum after = getSB_VAL_UTF8((U8*) s,
2244 (U8*) reginfo->strend);
2245 if ((to_complement ^ isSB(before,
2247 (U8*) reginfo->strbeg,
2249 (U8*) reginfo->strend,
2251 && (reginfo->intuit || regtry(reginfo, &s)))
2259 else { /* Not utf8. */
2260 SB_enum before = getSB_VAL_CP((U8) *(s -1));
2261 while (s < strend) {
2262 SB_enum after = getSB_VAL_CP((U8) *s);
2263 if ((to_complement ^ isSB(before,
2265 (U8*) reginfo->strbeg,
2267 (U8*) reginfo->strend,
2269 && (reginfo->intuit || regtry(reginfo, &s)))
2278 /* Here are at the final position in the target string. The SB
2279 * value is always true here, so matches, depending on other
2281 if (reginfo->intuit || regtry(reginfo, &s)) {
2288 if (s == reginfo->strbeg) {
2289 if (reginfo->intuit || regtry(reginfo, &s)) {
2292 s += (utf8_target) ? UTF8SKIP(s) : 1;
2293 if (UNLIKELY(s >= reginfo->strend)) {
2299 /* We are at a boundary between char_sub_0 and char_sub_1.
2300 * We also keep track of the value for char_sub_-1 as we
2301 * loop through the line. Context may be needed to make a
2302 * determination, and if so, this can save having to
2304 WB_enum previous = WB_UNKNOWN;
2305 WB_enum before = getWB_VAL_UTF8(
2308 (U8*)(reginfo->strbeg)),
2309 (U8*) reginfo->strend);
2310 while (s < strend) {
2311 WB_enum after = getWB_VAL_UTF8((U8*) s,
2312 (U8*) reginfo->strend);
2313 if ((to_complement ^ isWB(previous,
2316 (U8*) reginfo->strbeg,
2318 (U8*) reginfo->strend,
2320 && (reginfo->intuit || regtry(reginfo, &s)))
2329 else { /* Not utf8. */
2330 WB_enum previous = WB_UNKNOWN;
2331 WB_enum before = getWB_VAL_CP((U8) *(s -1));
2332 while (s < strend) {
2333 WB_enum after = getWB_VAL_CP((U8) *s);
2334 if ((to_complement ^ isWB(previous,
2337 (U8*) reginfo->strbeg,
2339 (U8*) reginfo->strend,
2341 && (reginfo->intuit || regtry(reginfo, &s)))
2351 if (reginfo->intuit || regtry(reginfo, &s)) {
2358 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2359 is_LNBREAK_latin1_safe(s, strend)
2363 /* The argument to all the POSIX node types is the class number to pass to
2364 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2371 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2372 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
2373 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
2388 /* The complement of something that matches only ASCII matches all
2389 * non-ASCII, plus everything in ASCII that isn't in the class. */
2390 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
2391 || ! _generic_isCC_A(*s, FLAGS(c)));
2400 /* Don't need to worry about utf8, as it can match only a single
2401 * byte invariant character. */
2402 REXEC_FBC_CLASS_SCAN(
2403 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
2411 if (! utf8_target) {
2412 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
2418 classnum = (_char_class_number) FLAGS(c);
2419 if (classnum < _FIRST_NON_SWASH_CC) {
2420 while (s < strend) {
2422 /* We avoid loading in the swash as long as possible, but
2423 * should we have to, we jump to a separate loop. This
2424 * extra 'if' statement is what keeps this code from being
2425 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
2426 if (UTF8_IS_ABOVE_LATIN1(*s)) {
2427 goto found_above_latin1;
2429 if ((UTF8_IS_INVARIANT(*s)
2430 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2432 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2433 && to_complement ^ cBOOL(
2434 _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*s,
2438 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2450 else switch (classnum) { /* These classes are implemented as
2452 case _CC_ENUM_SPACE:
2453 REXEC_FBC_UTF8_CLASS_SCAN(
2454 to_complement ^ cBOOL(isSPACE_utf8(s)));
2457 case _CC_ENUM_BLANK:
2458 REXEC_FBC_UTF8_CLASS_SCAN(
2459 to_complement ^ cBOOL(isBLANK_utf8(s)));
2462 case _CC_ENUM_XDIGIT:
2463 REXEC_FBC_UTF8_CLASS_SCAN(
2464 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2467 case _CC_ENUM_VERTSPACE:
2468 REXEC_FBC_UTF8_CLASS_SCAN(
2469 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2472 case _CC_ENUM_CNTRL:
2473 REXEC_FBC_UTF8_CLASS_SCAN(
2474 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2478 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2479 NOT_REACHED; /* NOTREACHED */
2484 found_above_latin1: /* Here we have to load a swash to get the result
2485 for the current code point */
2486 if (! PL_utf8_swash_ptrs[classnum]) {
2487 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2488 PL_utf8_swash_ptrs[classnum] =
2489 _core_swash_init("utf8",
2492 PL_XPosix_ptrs[classnum], &flags);
2495 /* This is a copy of the loop above for swash classes, though using the
2496 * FBC macro instead of being expanded out. Since we've loaded the
2497 * swash, we don't have to check for that each time through the loop */
2498 REXEC_FBC_UTF8_CLASS_SCAN(
2499 to_complement ^ cBOOL(_generic_utf8(
2502 swash_fetch(PL_utf8_swash_ptrs[classnum],
2510 /* what trie are we using right now */
2511 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2512 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2513 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2515 const char *last_start = strend - trie->minlen;
2517 const char *real_start = s;
2519 STRLEN maxlen = trie->maxlen;
2521 U8 **points; /* map of where we were in the input string
2522 when reading a given char. For ASCII this
2523 is unnecessary overhead as the relationship
2524 is always 1:1, but for Unicode, especially
2525 case folded Unicode this is not true. */
2526 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2530 GET_RE_DEBUG_FLAGS_DECL;
2532 /* We can't just allocate points here. We need to wrap it in
2533 * an SV so it gets freed properly if there is a croak while
2534 * running the match */
2537 sv_points=newSV(maxlen * sizeof(U8 *));
2538 SvCUR_set(sv_points,
2539 maxlen * sizeof(U8 *));
2540 SvPOK_on(sv_points);
2541 sv_2mortal(sv_points);
2542 points=(U8**)SvPV_nolen(sv_points );
2543 if ( trie_type != trie_utf8_fold
2544 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2547 bitmap=(U8*)trie->bitmap;
2549 bitmap=(U8*)ANYOF_BITMAP(c);
2551 /* this is the Aho-Corasick algorithm modified a touch
2552 to include special handling for long "unknown char" sequences.
2553 The basic idea being that we use AC as long as we are dealing
2554 with a possible matching char, when we encounter an unknown char
2555 (and we have not encountered an accepting state) we scan forward
2556 until we find a legal starting char.
2557 AC matching is basically that of trie matching, except that when
2558 we encounter a failing transition, we fall back to the current
2559 states "fail state", and try the current char again, a process
2560 we repeat until we reach the root state, state 1, or a legal
2561 transition. If we fail on the root state then we can either
2562 terminate if we have reached an accepting state previously, or
2563 restart the entire process from the beginning if we have not.
2566 while (s <= last_start) {
2567 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2575 U8 *uscan = (U8*)NULL;
2576 U8 *leftmost = NULL;
2578 U32 accepted_word= 0;
2582 while ( state && uc <= (U8*)strend ) {
2584 U32 word = aho->states[ state ].wordnum;
2588 DEBUG_TRIE_EXECUTE_r(
2589 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2590 dump_exec_pos( (char *)uc, c, strend, real_start,
2591 (char *)uc, utf8_target, 0 );
2592 Perl_re_printf( aTHX_
2593 " Scanning for legal start char...\n");
2597 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2601 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2607 if (uc >(U8*)last_start) break;
2611 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2612 if (!leftmost || lpos < leftmost) {
2613 DEBUG_r(accepted_word=word);
2619 points[pointpos++ % maxlen]= uc;
2620 if (foldlen || uc < (U8*)strend) {
2621 REXEC_TRIE_READ_CHAR(trie_type, trie,
2623 uscan, len, uvc, charid, foldlen,
2625 DEBUG_TRIE_EXECUTE_r({
2626 dump_exec_pos( (char *)uc, c, strend,
2627 real_start, s, utf8_target, 0);
2628 Perl_re_printf( aTHX_
2629 " Charid:%3u CP:%4" UVxf " ",
2641 word = aho->states[ state ].wordnum;
2643 base = aho->states[ state ].trans.base;
2645 DEBUG_TRIE_EXECUTE_r({
2647 dump_exec_pos( (char *)uc, c, strend, real_start,
2648 s, utf8_target, 0 );
2649 Perl_re_printf( aTHX_
2650 "%sState: %4" UVxf ", word=%" UVxf,
2651 failed ? " Fail transition to " : "",
2652 (UV)state, (UV)word);
2658 ( ((offset = base + charid
2659 - 1 - trie->uniquecharcount)) >= 0)
2660 && ((U32)offset < trie->lasttrans)
2661 && trie->trans[offset].check == state
2662 && (tmp=trie->trans[offset].next))
2664 DEBUG_TRIE_EXECUTE_r(
2665 Perl_re_printf( aTHX_ " - legal\n"));
2670 DEBUG_TRIE_EXECUTE_r(
2671 Perl_re_printf( aTHX_ " - fail\n"));
2673 state = aho->fail[state];
2677 /* we must be accepting here */
2678 DEBUG_TRIE_EXECUTE_r(
2679 Perl_re_printf( aTHX_ " - accepting\n"));
2688 if (!state) state = 1;
2691 if ( aho->states[ state ].wordnum ) {
2692 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2693 if (!leftmost || lpos < leftmost) {
2694 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2699 s = (char*)leftmost;
2700 DEBUG_TRIE_EXECUTE_r({
2701 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
2702 (UV)accepted_word, (IV)(s - real_start)
2705 if (reginfo->intuit || regtry(reginfo, &s)) {
2711 DEBUG_TRIE_EXECUTE_r({
2712 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
2715 DEBUG_TRIE_EXECUTE_r(
2716 Perl_re_printf( aTHX_ "No match.\n"));
2725 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2732 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2733 * flags have same meanings as with regexec_flags() */
2736 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2743 struct regexp *const prog = ReANY(rx);
2745 if (flags & REXEC_COPY_STR) {
2748 DEBUG_C(Perl_re_printf( aTHX_
2749 "Copy on write: regexp capture, type %d\n",
2751 /* Create a new COW SV to share the match string and store
2752 * in saved_copy, unless the current COW SV in saved_copy
2753 * is valid and suitable for our purpose */
2754 if (( prog->saved_copy
2755 && SvIsCOW(prog->saved_copy)
2756 && SvPOKp(prog->saved_copy)
2759 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2761 /* just reuse saved_copy SV */
2762 if (RXp_MATCH_COPIED(prog)) {
2763 Safefree(prog->subbeg);
2764 RXp_MATCH_COPIED_off(prog);
2768 /* create new COW SV to share string */
2769 RX_MATCH_COPY_FREE(rx);
2770 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2772 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2773 assert (SvPOKp(prog->saved_copy));
2774 prog->sublen = strend - strbeg;
2775 prog->suboffset = 0;
2776 prog->subcoffset = 0;
2781 SSize_t max = strend - strbeg;
2784 if ( (flags & REXEC_COPY_SKIP_POST)
2785 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2786 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2787 ) { /* don't copy $' part of string */
2790 /* calculate the right-most part of the string covered
2791 * by a capture. Due to lookahead, this may be to
2792 * the right of $&, so we have to scan all captures */
2793 while (n <= prog->lastparen) {
2794 if (prog->offs[n].end > max)
2795 max = prog->offs[n].end;
2799 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2800 ? prog->offs[0].start
2802 assert(max >= 0 && max <= strend - strbeg);
2805 if ( (flags & REXEC_COPY_SKIP_PRE)
2806 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2807 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2808 ) { /* don't copy $` part of string */
2811 /* calculate the left-most part of the string covered
2812 * by a capture. Due to lookbehind, this may be to
2813 * the left of $&, so we have to scan all captures */
2814 while (min && n <= prog->lastparen) {
2815 if ( prog->offs[n].start != -1
2816 && prog->offs[n].start < min)
2818 min = prog->offs[n].start;
2822 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2823 && min > prog->offs[0].end
2825 min = prog->offs[0].end;
2829 assert(min >= 0 && min <= max && min <= strend - strbeg);
2832 if (RX_MATCH_COPIED(rx)) {
2833 if (sublen > prog->sublen)
2835 (char*)saferealloc(prog->subbeg, sublen+1);
2838 prog->subbeg = (char*)safemalloc(sublen+1);
2839 Copy(strbeg + min, prog->subbeg, sublen, char);
2840 prog->subbeg[sublen] = '\0';
2841 prog->suboffset = min;
2842 prog->sublen = sublen;
2843 RX_MATCH_COPIED_on(rx);
2845 prog->subcoffset = prog->suboffset;
2846 if (prog->suboffset && utf8_target) {
2847 /* Convert byte offset to chars.
2848 * XXX ideally should only compute this if @-/@+
2849 * has been seen, a la PL_sawampersand ??? */
2851 /* If there's a direct correspondence between the
2852 * string which we're matching and the original SV,
2853 * then we can use the utf8 len cache associated with
2854 * the SV. In particular, it means that under //g,
2855 * sv_pos_b2u() will use the previously cached
2856 * position to speed up working out the new length of
2857 * subcoffset, rather than counting from the start of
2858 * the string each time. This stops
2859 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2860 * from going quadratic */
2861 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2862 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2863 SV_GMAGIC|SV_CONST_RETURN);
2865 prog->subcoffset = utf8_length((U8*)strbeg,
2866 (U8*)(strbeg+prog->suboffset));
2870 RX_MATCH_COPY_FREE(rx);
2871 prog->subbeg = strbeg;
2872 prog->suboffset = 0;
2873 prog->subcoffset = 0;
2874 prog->sublen = strend - strbeg;
2882 - regexec_flags - match a regexp against a string
2885 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2886 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2887 /* stringarg: the point in the string at which to begin matching */
2888 /* strend: pointer to null at end of string */
2889 /* strbeg: real beginning of string */
2890 /* minend: end of match must be >= minend bytes after stringarg. */
2891 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2892 * itself is accessed via the pointers above */
2893 /* data: May be used for some additional optimizations.
2894 Currently unused. */
2895 /* flags: For optimizations. See REXEC_* in regexp.h */
2898 struct regexp *const prog = ReANY(rx);
2902 SSize_t minlen; /* must match at least this many chars */
2903 SSize_t dontbother = 0; /* how many characters not to try at end */
2904 const bool utf8_target = cBOOL(DO_UTF8(sv));
2906 RXi_GET_DECL(prog,progi);
2907 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2908 regmatch_info *const reginfo = ®info_buf;
2909 regexp_paren_pair *swap = NULL;
2911 GET_RE_DEBUG_FLAGS_DECL;
2913 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2914 PERL_UNUSED_ARG(data);
2916 /* Be paranoid... */
2918 Perl_croak(aTHX_ "NULL regexp parameter");
2922 debug_start_match(rx, utf8_target, stringarg, strend,
2926 startpos = stringarg;
2928 /* set these early as they may be used by the HOP macros below */
2929 reginfo->strbeg = strbeg;
2930 reginfo->strend = strend;
2931 reginfo->is_utf8_target = cBOOL(utf8_target);
2933 if (prog->intflags & PREGf_GPOS_SEEN) {
2936 /* set reginfo->ganch, the position where \G can match */
2939 (flags & REXEC_IGNOREPOS)
2940 ? stringarg /* use start pos rather than pos() */
2941 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2942 /* Defined pos(): */
2943 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2944 : strbeg; /* pos() not defined; use start of string */
2946 DEBUG_GPOS_r(Perl_re_printf( aTHX_
2947 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
2949 /* in the presence of \G, we may need to start looking earlier in
2950 * the string than the suggested start point of stringarg:
2951 * if prog->gofs is set, then that's a known, fixed minimum
2954 * /ab|c\G/: gofs = 1
2955 * or if the minimum offset isn't known, then we have to go back
2956 * to the start of the string, e.g. /w+\G/
2959 if (prog->intflags & PREGf_ANCH_GPOS) {
2961 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
2963 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
2965 DEBUG_r(Perl_re_printf( aTHX_
2966 "fail: ganch-gofs before earliest possible start\n"));
2971 startpos = reginfo->ganch;
2973 else if (prog->gofs) {
2974 startpos = HOPBACKc(startpos, prog->gofs);
2978 else if (prog->intflags & PREGf_GPOS_FLOAT)
2982 minlen = prog->minlen;
2983 if ((startpos + minlen) > strend || startpos < strbeg) {
2984 DEBUG_r(Perl_re_printf( aTHX_
2985 "Regex match can't succeed, so not even tried\n"));
2989 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2990 * which will call destuctors to reset PL_regmatch_state, free higher
2991 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2992 * regmatch_info_aux_eval */
2994 oldsave = PL_savestack_ix;
2998 if ((prog->extflags & RXf_USE_INTUIT)
2999 && !(flags & REXEC_CHECKED))
3001 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
3006 if (prog->extflags & RXf_CHECK_ALL) {
3007 /* we can match based purely on the result of INTUIT.
3008 * Set up captures etc just for $& and $-[0]
3009 * (an intuit-only match wont have $1,$2,..) */
3010 assert(!prog->nparens);
3012 /* s/// doesn't like it if $& is earlier than where we asked it to
3013 * start searching (which can happen on something like /.\G/) */
3014 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3017 /* this should only be possible under \G */
3018 assert(prog->intflags & PREGf_GPOS_SEEN);
3019 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3020 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3024 /* match via INTUIT shouldn't have any captures.
3025 * Let @-, @+, $^N know */
3026 prog->lastparen = prog->lastcloseparen = 0;
3027 RX_MATCH_UTF8_set(rx, utf8_target);
3028 prog->offs[0].start = s - strbeg;
3029 prog->offs[0].end = utf8_target
3030 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
3031 : s - strbeg + prog->minlenret;
3032 if ( !(flags & REXEC_NOT_FIRST) )
3033 S_reg_set_capture_string(aTHX_ rx,
3035 sv, flags, utf8_target);
3041 multiline = prog->extflags & RXf_PMf_MULTILINE;
3043 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
3044 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3045 "String too short [regexec_flags]...\n"));
3049 /* Check validity of program. */
3050 if (UCHARAT(progi->program) != REG_MAGIC) {
3051 Perl_croak(aTHX_ "corrupted regexp program");
3054 RX_MATCH_TAINTED_off(rx);
3055 RX_MATCH_UTF8_set(rx, utf8_target);
3057 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3058 reginfo->intuit = 0;
3059 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3060 reginfo->warned = FALSE;
3062 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
3063 /* see how far we have to get to not match where we matched before */
3064 reginfo->till = stringarg + minend;
3066 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
3067 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3068 S_cleanup_regmatch_info_aux has executed (registered by
3069 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3070 magic belonging to this SV.
3071 Not newSVsv, either, as it does not COW.
3073 reginfo->sv = newSV(0);
3074 SvSetSV_nosteal(reginfo->sv, sv);
3075 SAVEFREESV(reginfo->sv);
3078 /* reserve next 2 or 3 slots in PL_regmatch_state:
3079 * slot N+0: may currently be in use: skip it
3080 * slot N+1: use for regmatch_info_aux struct
3081 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3082 * slot N+3: ready for use by regmatch()
3086 regmatch_state *old_regmatch_state;
3087 regmatch_slab *old_regmatch_slab;
3088 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3090 /* on first ever match, allocate first slab */
3091 if (!PL_regmatch_slab) {
3092 Newx(PL_regmatch_slab, 1, regmatch_slab);
3093 PL_regmatch_slab->prev = NULL;
3094 PL_regmatch_slab->next = NULL;
3095 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3098 old_regmatch_state = PL_regmatch_state;
3099 old_regmatch_slab = PL_regmatch_slab;
3101 for (i=0; i <= max; i++) {
3103 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3105 reginfo->info_aux_eval =
3106 reginfo->info_aux->info_aux_eval =
3107 &(PL_regmatch_state->u.info_aux_eval);
3109 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3110 PL_regmatch_state = S_push_slab(aTHX);
3113 /* note initial PL_regmatch_state position; at end of match we'll
3114 * pop back to there and free any higher slabs */
3116 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3117 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
3118 reginfo->info_aux->poscache = NULL;
3120 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
3122 if ((prog->extflags & RXf_EVAL_SEEN))
3123 S_setup_eval_state(aTHX_ reginfo);
3125 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
3128 /* If there is a "must appear" string, look for it. */
3130 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
3131 /* We have to be careful. If the previous successful match
3132 was from this regex we don't want a subsequent partially
3133 successful match to clobber the old results.
3134 So when we detect this possibility we add a swap buffer
3135 to the re, and switch the buffer each match. If we fail,
3136 we switch it back; otherwise we leave it swapped.
3139 /* do we need a save destructor here for eval dies? */
3140 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
3141 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3142 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
3150 if (prog->recurse_locinput)
3151 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3153 /* Simplest case: anchored match need be tried only once, or with
3154 * MBOL, only at the beginning of each line.
3156 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3157 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3158 * match at the start of the string then it won't match anywhere else
3159 * either; while with /.*.../, if it doesn't match at the beginning,
3160 * the earliest it could match is at the start of the next line */
3162 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3165 if (regtry(reginfo, &s))
3168 if (!(prog->intflags & PREGf_ANCH_MBOL))
3171 /* didn't match at start, try at other newline positions */
3174 dontbother = minlen - 1;
3175 end = HOP3c(strend, -dontbother, strbeg) - 1;
3177 /* skip to next newline */
3179 while (s <= end) { /* note it could be possible to match at the end of the string */
3180 /* NB: newlines are the same in unicode as they are in latin */
3183 if (prog->check_substr || prog->check_utf8) {
3184 /* note that with PREGf_IMPLICIT, intuit can only fail
3185 * or return the start position, so it's of limited utility.
3186 * Nevertheless, I made the decision that the potential for
3187 * quick fail was still worth it - DAPM */
3188 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3192 if (regtry(reginfo, &s))
3196 } /* end anchored search */
3198 if (prog->intflags & PREGf_ANCH_GPOS)
3200 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3201 assert(prog->intflags & PREGf_GPOS_SEEN);
3202 /* For anchored \G, the only position it can match from is
3203 * (ganch-gofs); we already set startpos to this above; if intuit
3204 * moved us on from there, we can't possibly succeed */
3205 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
3206 if (s == startpos && regtry(reginfo, &s))
3211 /* Messy cases: unanchored match. */
3212 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
3213 /* we have /x+whatever/ */
3214 /* it must be a one character string (XXXX Except is_utf8_pat?) */
3220 if (! prog->anchored_utf8) {
3221 to_utf8_substr(prog);
3223 ch = SvPVX_const(prog->anchored_utf8)[0];
3226 DEBUG_EXECUTE_r( did_match = 1 );
3227 if (regtry(reginfo, &s)) goto got_it;
3229 while (s < strend && *s == ch)
3236 if (! prog->anchored_substr) {
3237 if (! to_byte_substr(prog)) {
3238 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3241 ch = SvPVX_const(prog->anchored_substr)[0];
3244 DEBUG_EXECUTE_r( did_match = 1 );
3245 if (regtry(reginfo, &s)) goto got_it;
3247 while (s < strend && *s == ch)
3252 DEBUG_EXECUTE_r(if (!did_match)
3253 Perl_re_printf( aTHX_
3254 "Did not find anchored character...\n")
3257 else if (prog->anchored_substr != NULL
3258 || prog->anchored_utf8 != NULL
3259 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
3260 && prog->float_max_offset < strend - s)) {
3265 char *last1; /* Last position checked before */
3269 if (prog->anchored_substr || prog->anchored_utf8) {
3271 if (! prog->anchored_utf8) {
3272 to_utf8_substr(prog);
3274 must = prog->anchored_utf8;
3277 if (! prog->anchored_substr) {
3278 if (! to_byte_substr(prog)) {
3279 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3282 must = prog->anchored_substr;
3284 back_max = back_min = prog->anchored_offset;
3287 if (! prog->float_utf8) {
3288 to_utf8_substr(prog);
3290 must = prog->float_utf8;
3293 if (! prog->float_substr) {
3294 if (! to_byte_substr(prog)) {
3295 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3298 must = prog->float_substr;
3300 back_max = prog->float_max_offset;
3301 back_min = prog->float_min_offset;
3307 last = HOP3c(strend, /* Cannot start after this */
3308 -(SSize_t)(CHR_SVLEN(must)
3309 - (SvTAIL(must) != 0) + back_min), strbeg);
3311 if (s > reginfo->strbeg)
3312 last1 = HOPc(s, -1);
3314 last1 = s - 1; /* bogus */
3316 /* XXXX check_substr already used to find "s", can optimize if
3317 check_substr==must. */
3319 strend = HOPc(strend, -dontbother);
3320 while ( (s <= last) &&
3321 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
3322 (unsigned char*)strend, must,
3323 multiline ? FBMrf_MULTILINE : 0)) ) {
3324 DEBUG_EXECUTE_r( did_match = 1 );
3325 if (HOPc(s, -back_max) > last1) {
3326 last1 = HOPc(s, -back_min);
3327 s = HOPc(s, -back_max);
3330 char * const t = (last1 >= reginfo->strbeg)
3331 ? HOPc(last1, 1) : last1 + 1;
3333 last1 = HOPc(s, -back_min);
3337 while (s <= last1) {
3338 if (regtry(reginfo, &s))
3341 s++; /* to break out of outer loop */
3348 while (s <= last1) {
3349 if (regtry(reginfo, &s))
3355 DEBUG_EXECUTE_r(if (!did_match) {
3356 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
3357 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
3358 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n",
3359 ((must == prog->anchored_substr || must == prog->anchored_utf8)
3360 ? "anchored" : "floating"),
3361 quoted, RE_SV_TAIL(must));
3365 else if ( (c = progi->regstclass) ) {
3367 const OPCODE op = OP(progi->regstclass);
3368 /* don't bother with what can't match */
3369 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
3370 strend = HOPc(strend, -(minlen - 1));
3373 SV * const prop = sv_newmortal();
3374 regprop(prog, prop, c, reginfo, NULL);
3376 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
3378 Perl_re_printf( aTHX_
3379 "Matching stclass %.*s against %s (%d bytes)\n",
3380 (int)SvCUR(prop), SvPVX_const(prop),
3381 quoted, (int)(strend - s));
3384 if (find_byclass(prog, c, s, strend, reginfo))
3386 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n"));
3390 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3398 if (! prog->float_utf8) {
3399 to_utf8_substr(prog);
3401 float_real = prog->float_utf8;
3404 if (! prog->float_substr) {
3405 if (! to_byte_substr(prog)) {
3406 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3409 float_real = prog->float_substr;
3412 little = SvPV_const(float_real, len);
3413 if (SvTAIL(float_real)) {
3414 /* This means that float_real contains an artificial \n on
3415 * the end due to the presence of something like this:
3416 * /foo$/ where we can match both "foo" and "foo\n" at the
3417 * end of the string. So we have to compare the end of the
3418 * string first against the float_real without the \n and
3419 * then against the full float_real with the string. We
3420 * have to watch out for cases where the string might be
3421 * smaller than the float_real or the float_real without
3423 char *checkpos= strend - len;
3425 Perl_re_printf( aTHX_
3426 "%sChecking for float_real.%s\n",
3427 PL_colors[4], PL_colors[5]));
3428 if (checkpos + 1 < strbeg) {
3429 /* can't match, even if we remove the trailing \n
3430 * string is too short to match */
3432 Perl_re_printf( aTHX_
3433 "%sString shorter than required trailing substring, cannot match.%s\n",
3434 PL_colors[4], PL_colors[5]));
3436 } else if (memEQ(checkpos + 1, little, len - 1)) {
3437 /* can match, the end of the string matches without the
3439 last = checkpos + 1;
3440 } else if (checkpos < strbeg) {
3441 /* cant match, string is too short when the "\n" is
3444 Perl_re_printf( aTHX_
3445 "%sString does not contain required trailing substring, cannot match.%s\n",
3446 PL_colors[4], PL_colors[5]));
3448 } else if (!multiline) {
3449 /* non multiline match, so compare with the "\n" at the
3450 * end of the string */
3451 if (memEQ(checkpos, little, len)) {
3455 Perl_re_printf( aTHX_
3456 "%sString does not contain required trailing substring, cannot match.%s\n",
3457 PL_colors[4], PL_colors[5]));
3461 /* multiline match, so we have to search for a place
3462 * where the full string is located */
3468 last = rninstr(s, strend, little, little + len);
3470 last = strend; /* matching "$" */
3473 /* at one point this block contained a comment which was
3474 * probably incorrect, which said that this was a "should not
3475 * happen" case. Even if it was true when it was written I am
3476 * pretty sure it is not anymore, so I have removed the comment
3477 * and replaced it with this one. Yves */
3479 Perl_re_printf( aTHX_
3480 "%sString does not contain required substring, cannot match.%s\n",
3481 PL_colors[4], PL_colors[5]
3485 dontbother = strend - last + prog->float_min_offset;
3487 if (minlen && (dontbother < minlen))
3488 dontbother = minlen - 1;
3489 strend -= dontbother; /* this one's always in bytes! */
3490 /* We don't know much -- general case. */
3493 if (regtry(reginfo, &s))
3502 if (regtry(reginfo, &s))
3504 } while (s++ < strend);
3512 /* s/// doesn't like it if $& is earlier than where we asked it to
3513 * start searching (which can happen on something like /.\G/) */
3514 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3515 && (prog->offs[0].start < stringarg - strbeg))
3517 /* this should only be possible under \G */
3518 assert(prog->intflags & PREGf_GPOS_SEEN);
3519 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3520 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3526 Perl_re_exec_indentf( aTHX_
3527 "rex=0x%" UVxf " freeing offs: 0x%" UVxf "\n",
3535 /* clean up; this will trigger destructors that will free all slabs
3536 * above the current one, and cleanup the regmatch_info_aux
3537 * and regmatch_info_aux_eval sructs */
3539 LEAVE_SCOPE(oldsave);
3541 if (RXp_PAREN_NAMES(prog))
3542 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3544 /* make sure $`, $&, $', and $digit will work later */
3545 if ( !(flags & REXEC_NOT_FIRST) )
3546 S_reg_set_capture_string(aTHX_ rx,
3547 strbeg, reginfo->strend,
3548 sv, flags, utf8_target);
3553 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n",
3554 PL_colors[4], PL_colors[5]));
3556 /* clean up; this will trigger destructors that will free all slabs
3557 * above the current one, and cleanup the regmatch_info_aux
3558 * and regmatch_info_aux_eval sructs */
3560 LEAVE_SCOPE(oldsave);
3563 /* we failed :-( roll it back */
3564 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3565 "rex=0x%" UVxf " rolling back offs: freeing=0x%" UVxf " restoring=0x%" UVxf "\n",
3571 Safefree(prog->offs);
3578 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3579 * Do inc before dec, in case old and new rex are the same */
3580 #define SET_reg_curpm(Re2) \
3581 if (reginfo->info_aux_eval) { \
3582 (void)ReREFCNT_inc(Re2); \
3583 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3584 PM_SETRE((PL_reg_curpm), (Re2)); \
3589 - regtry - try match at specific point
3591 STATIC bool /* 0 failure, 1 success */
3592 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3595 REGEXP *const rx = reginfo->prog;
3596 regexp *const prog = ReANY(rx);
3599 U32 depth = 0; /* used by REGCP_SET */
3601 RXi_GET_DECL(prog,progi);
3602 GET_RE_DEBUG_FLAGS_DECL;
3604 PERL_ARGS_ASSERT_REGTRY;
3606 reginfo->cutpoint=NULL;
3608 prog->offs[0].start = *startposp - reginfo->strbeg;
3609 prog->lastparen = 0;
3610 prog->lastcloseparen = 0;
3612 /* XXXX What this code is doing here?!!! There should be no need
3613 to do this again and again, prog->lastparen should take care of
3616 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3617 * Actually, the code in regcppop() (which Ilya may be meaning by
3618 * prog->lastparen), is not needed at all by the test suite
3619 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3620 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3621 * Meanwhile, this code *is* needed for the
3622 * above-mentioned test suite tests to succeed. The common theme
3623 * on those tests seems to be returning null fields from matches.
3624 * --jhi updated by dapm */
3626 /* After encountering a variant of the issue mentioned above I think
3627 * the point Ilya was making is that if we properly unwind whenever
3628 * we set lastparen to a smaller value then we should not need to do
3629 * this every time, only when needed. So if we have tests that fail if
3630 * we remove this, then it suggests somewhere else we are improperly
3631 * unwinding the lastparen/paren buffers. See UNWIND_PARENS() and
3632 * places it is called, and related regcp() routines. - Yves */
3634 if (prog->nparens) {
3635 regexp_paren_pair *pp = prog->offs;
3637 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3645 result = regmatch(reginfo, *startposp, progi->program + 1);
3647 prog->offs[0].end = result;
3650 if (reginfo->cutpoint)
3651 *startposp= reginfo->cutpoint;
3652 REGCP_UNWIND(lastcp);
3657 #define sayYES goto yes
3658 #define sayNO goto no
3659 #define sayNO_SILENT goto no_silent
3661 /* we dont use STMT_START/END here because it leads to
3662 "unreachable code" warnings, which are bogus, but distracting. */
3663 #define CACHEsayNO \
3664 if (ST.cache_mask) \
3665 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3668 /* this is used to determine how far from the left messages like
3669 'failed...' are printed in regexec.c. It should be set such that
3670 messages are inline with the regop output that created them.
3672 #define REPORT_CODE_OFF 29
3673 #define INDENT_CHARS(depth) ((int)(depth) % 20)
3676 Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...)
3680 PerlIO *f= Perl_debug_log;
3681 PERL_ARGS_ASSERT_RE_EXEC_INDENTF;
3682 va_start(ap, depth);
3683 PerlIO_printf(f, "%*s|%4" UVuf "| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" );
3684 result = PerlIO_vprintf(f, fmt, ap);
3688 #endif /* DEBUGGING */
3691 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3692 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3693 #define CHRTEST_NOT_A_CP_1 -999
3694 #define CHRTEST_NOT_A_CP_2 -998
3696 /* grab a new slab and return the first slot in it */
3698 STATIC regmatch_state *
3701 regmatch_slab *s = PL_regmatch_slab->next;
3703 Newx(s, 1, regmatch_slab);
3704 s->prev = PL_regmatch_slab;
3706 PL_regmatch_slab->next = s;
3708 PL_regmatch_slab = s;
3709 return SLAB_FIRST(s);
3713 /* push a new state then goto it */
3715 #define PUSH_STATE_GOTO(state, node, input) \
3716 pushinput = input; \
3718 st->resume_state = state; \
3721 /* push a new state with success backtracking, then goto it */
3723 #define PUSH_YES_STATE_GOTO(state, node, input) \
3724 pushinput = input; \
3726 st->resume_state = state; \
3727 goto push_yes_state;
3734 regmatch() - main matching routine
3736 This is basically one big switch statement in a loop. We execute an op,
3737 set 'next' to point the next op, and continue. If we come to a point which
3738 we may need to backtrack to on failure such as (A|B|C), we push a
3739 backtrack state onto the backtrack stack. On failure, we pop the top
3740 state, and re-enter the loop at the state indicated. If there are no more
3741 states to pop, we return failure.
3743 Sometimes we also need to backtrack on success; for example /A+/, where
3744 after successfully matching one A, we need to go back and try to
3745 match another one; similarly for lookahead assertions: if the assertion
3746 completes successfully, we backtrack to the state just before the assertion
3747 and then carry on. In these cases, the pushed state is marked as
3748 'backtrack on success too'. This marking is in fact done by a chain of
3749 pointers, each pointing to the previous 'yes' state. On success, we pop to
3750 the nearest yes state, discarding any intermediate failure-only states.
3751 Sometimes a yes state is pushed just to force some cleanup code to be
3752 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3753 it to free the inner regex.
3755 Note that failure backtracking rewinds the cursor position, while
3756 success backtracking leaves it alone.
3758 A pattern is complete when the END op is executed, while a subpattern
3759 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3760 ops trigger the "pop to last yes state if any, otherwise return true"
3763 A common convention in this function is to use A and B to refer to the two
3764 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3765 the subpattern to be matched possibly multiple times, while B is the entire
3766 rest of the pattern. Variable and state names reflect this convention.
3768 The states in the main switch are the union of ops and failure/success of
3769 substates associated with with that op. For example, IFMATCH is the op
3770 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3771 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3772 successfully matched A and IFMATCH_A_fail is a state saying that we have
3773 just failed to match A. Resume states always come in pairs. The backtrack
3774 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3775 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3776 on success or failure.
3778 The struct that holds a backtracking state is actually a big union, with
3779 one variant for each major type of op. The variable st points to the
3780 top-most backtrack struct. To make the code clearer, within each
3781 block of code we #define ST to alias the relevant union.
3783 Here's a concrete example of a (vastly oversimplified) IFMATCH
3789 #define ST st->u.ifmatch
3791 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3792 ST.foo = ...; // some state we wish to save
3794 // push a yes backtrack state with a resume value of
3795 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3797 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3800 case IFMATCH_A: // we have successfully executed A; now continue with B
3802 bar = ST.foo; // do something with the preserved value
3805 case IFMATCH_A_fail: // A failed, so the assertion failed
3806 ...; // do some housekeeping, then ...
3807 sayNO; // propagate the failure
3814 For any old-timers reading this who are familiar with the old recursive
3815 approach, the code above is equivalent to:
3817 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3826 ...; // do some housekeeping, then ...
3827 sayNO; // propagate the failure
3830 The topmost backtrack state, pointed to by st, is usually free. If you
3831 want to claim it, populate any ST.foo fields in it with values you wish to
3832 save, then do one of
3834 PUSH_STATE_GOTO(resume_state, node, newinput);
3835 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3837 which sets that backtrack state's resume value to 'resume_state', pushes a
3838 new free entry to the top of the backtrack stack, then goes to 'node'.
3839 On backtracking, the free slot is popped, and the saved state becomes the
3840 new free state. An ST.foo field in this new top state can be temporarily
3841 accessed to retrieve values, but once the main loop is re-entered, it
3842 becomes available for reuse.
3844 Note that the depth of the backtrack stack constantly increases during the
3845 left-to-right execution of the pattern, rather than going up and down with
3846 the pattern nesting. For example the stack is at its maximum at Z at the
3847 end of the pattern, rather than at X in the following:
3849 /(((X)+)+)+....(Y)+....Z/
3851 The only exceptions to this are lookahead/behind assertions and the cut,
3852 (?>A), which pop all the backtrack states associated with A before
3855 Backtrack state structs are allocated in slabs of about 4K in size.
3856 PL_regmatch_state and st always point to the currently active state,
3857 and PL_regmatch_slab points to the slab currently containing
3858 PL_regmatch_state. The first time regmatch() is called, the first slab is
3859 allocated, and is never freed until interpreter destruction. When the slab
3860 is full, a new one is allocated and chained to the end. At exit from
3861 regmatch(), slabs allocated since entry are freed.
3866 #define DEBUG_STATE_pp(pp) \
3868 DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \
3869 Perl_re_printf( aTHX_ \
3870 "%*s" pp " %s%s%s%s%s\n", \
3871 INDENT_CHARS(depth), "", \
3872 PL_reg_name[st->resume_state], \
3873 ((st==yes_state||st==mark_state) ? "[" : ""), \
3874 ((st==yes_state) ? "Y" : ""), \
3875 ((st==mark_state) ? "M" : ""), \
3876 ((st==yes_state||st==mark_state) ? "]" : "") \
3881 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3886 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3887 const char *start, const char *end, const char *blurb)
3889 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3891 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3896 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3897 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3899 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3900 start, end - start, 60);
3902 Perl_re_printf( aTHX_
3903 "%s%s REx%s %s against %s\n",
3904 PL_colors[4], blurb, PL_colors[5], s0, s1);
3906 if (utf8_target||utf8_pat)
3907 Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n",
3908 utf8_pat ? "pattern" : "",
3909 utf8_pat && utf8_target ? " and " : "",
3910 utf8_target ? "string" : ""
3916 S_dump_exec_pos(pTHX_ const char *locinput,
3917 const regnode *scan,
3918 const char *loc_regeol,
3919 const char *loc_bostr,
3920 const char *loc_reg_starttry,
3921 const bool utf8_target,
3925 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3926 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3927 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3928 /* The part of the string before starttry has one color
3929 (pref0_len chars), between starttry and current
3930 position another one (pref_len - pref0_len chars),
3931 after the current position the third one.
3932 We assume that pref0_len <= pref_len, otherwise we
3933 decrease pref0_len. */
3934 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3935 ? (5 + taill) - l : locinput - loc_bostr;
3938 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3940 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3942 pref0_len = pref_len - (locinput - loc_reg_starttry);
3943 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3944 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3945 ? (5 + taill) - pref_len : loc_regeol - locinput);
3946 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3950 if (pref0_len > pref_len)
3951 pref0_len = pref_len;
3953 const int is_uni = utf8_target ? 1 : 0;
3955 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3956 (locinput - pref_len),pref0_len, 60, 4, 5);
3958 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3959 (locinput - pref_len + pref0_len),
3960 pref_len - pref0_len, 60, 2, 3);
3962 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3963 locinput, loc_regeol - locinput, 10, 0, 1);
3965 const STRLEN tlen=len0+len1+len2;
3966 Perl_re_printf( aTHX_
3967 "%4" IVdf " <%.*s%.*s%s%.*s>%*s|%4u| ",
3968 (IV)(locinput - loc_bostr),
3971 (docolor ? "" : "> <"),
3973 (int)(tlen > 19 ? 0 : 19 - tlen),
3981 /* reg_check_named_buff_matched()
3982 * Checks to see if a named buffer has matched. The data array of
3983 * buffer numbers corresponding to the buffer is expected to reside
3984 * in the regexp->data->data array in the slot stored in the ARG() of
3985 * node involved. Note that this routine doesn't actually care about the
3986 * name, that information is not preserved from compilation to execution.
3987 * Returns the index of the leftmost defined buffer with the given name
3988 * or 0 if non of the buffers matched.
3991 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3994 RXi_GET_DECL(rex,rexi);
3995 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3996 I32 *nums=(I32*)SvPVX(sv_dat);
3998 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
4000 for ( n=0; n<SvIVX(sv_dat); n++ ) {
4001 if ((I32)rex->lastparen >= nums[n] &&
4002 rex->offs[nums[n]].end != -1)
4012 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
4013 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
4015 /* This function determines if there are one or two characters that match
4016 * the first character of the passed-in EXACTish node <text_node>, and if
4017 * so, returns them in the passed-in pointers.
4019 * If it determines that no possible character in the target string can
4020 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
4021 * the first character in <text_node> requires UTF-8 to represent, and the
4022 * target string isn't in UTF-8.)
4024 * If there are more than two characters that could match the beginning of
4025 * <text_node>, or if more context is required to determine a match or not,
4026 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
4028 * The motiviation behind this function is to allow the caller to set up
4029 * tight loops for matching. If <text_node> is of type EXACT, there is
4030 * only one possible character that can match its first character, and so
4031 * the situation is quite simple. But things get much more complicated if
4032 * folding is involved. It may be that the first character of an EXACTFish
4033 * node doesn't participate in any possible fold, e.g., punctuation, so it
4034 * can be matched only by itself. The vast majority of characters that are
4035 * in folds match just two things, their lower and upper-case equivalents.
4036 * But not all are like that; some have multiple possible matches, or match
4037 * sequences of more than one character. This function sorts all that out.
4039 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
4040 * loop of trying to match A*, we know we can't exit where the thing
4041 * following it isn't a B. And something can't be a B unless it is the
4042 * beginning of B. By putting a quick test for that beginning in a tight
4043 * loop, we can rule out things that can't possibly be B without having to
4044 * break out of the loop, thus avoiding work. Similarly, if A is a single
4045 * character, we can make a tight loop matching A*, using the outputs of
4048 * If the target string to match isn't in UTF-8, and there aren't
4049 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
4050 * the one or two possible octets (which are characters in this situation)
4051 * that can match. In all cases, if there is only one character that can
4052 * match, *<c1p> and *<c2p> will be identical.
4054 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
4055 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
4056 * can match the beginning of <text_node>. They should be declared with at
4057 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
4058 * undefined what these contain.) If one or both of the buffers are
4059 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
4060 * corresponding invariant. If variant, the corresponding *<c1p> and/or
4061 * *<c2p> will be set to a negative number(s) that shouldn't match any code
4062 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
4063 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
4065 const bool utf8_target = reginfo->is_utf8_target;
4067 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
4068 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
4069 bool use_chrtest_void = FALSE;
4070 const bool is_utf8_pat = reginfo->is_utf8_pat;
4072 /* Used when we have both utf8 input and utf8 output, to avoid converting
4073 * to/from code points */
4074 bool utf8_has_been_setup = FALSE;
4078 U8 *pat = (U8*)STRING(text_node);
4079 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
4081 if (OP(text_node) == EXACT || OP(text_node) == EXACTL) {
4083 /* In an exact node, only one thing can be matched, that first
4084 * character. If both the pat and the target are UTF-8, we can just
4085 * copy the input to the output, avoiding finding the code point of
4090 else if (utf8_target) {
4091 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
4092 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
4093 utf8_has_been_setup = TRUE;
4096 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
4099 else { /* an EXACTFish node */
4100 U8 *pat_end = pat + STR_LEN(text_node);
4102 /* An EXACTFL node has at least some characters unfolded, because what
4103 * they match is not known until now. So, now is the time to fold
4104 * the first few of them, as many as are needed to determine 'c1' and
4105 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
4106 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
4107 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
4108 * need to fold as many characters as a single character can fold to,
4109 * so that later we can check if the first ones are such a multi-char
4110 * fold. But, in such a pattern only locale-problematic characters
4111 * aren't folded, so we can skip this completely if the first character
4112 * in the node isn't one of the tricky ones */
4113 if (OP(text_node) == EXACTFL) {
4115 if (! is_utf8_pat) {
4116 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
4118 folded[0] = folded[1] = 's';
4120 pat_end = folded + 2;
4123 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
4128 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
4130 *(d++) = (U8) toFOLD_LC(*s);
4135 _to_utf8_fold_flags(s,
4138 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
4149 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
4150 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
4152 /* Multi-character folds require more context to sort out. Also
4153 * PL_utf8_foldclosures used below doesn't handle them, so have to
4154 * be handled outside this routine */
4155 use_chrtest_void = TRUE;
4157 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
4158 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
4160 /* Load the folds hash, if not already done */
4162 if (! PL_utf8_foldclosures) {
4163 _load_PL_utf8_foldclosures();
4166 /* The fold closures data structure is a hash with the keys
4167 * being the UTF-8 of every character that is folded to, like
4168 * 'k', and the values each an array of all code points that
4169 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
4170 * Multi-character folds are not included */
4171 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
4176 /* Not found in the hash, therefore there are no folds
4177 * containing it, so there is only a single character that
4181 else { /* Does participate in folds */
4182 AV* list = (AV*) *listp;
4183 if (av_tindex_nomg(list) != 1) {
4185 /* If there aren't exactly two folds to this, it is
4186 * outside the scope of this function */
4187 use_chrtest_void = TRUE;
4189 else { /* There are two. Get them */
4190 SV** c_p = av_fetch(list, 0, FALSE);
4192 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
4196 c_p = av_fetch(list, 1, FALSE);
4198 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
4202 /* Folds that cross the 255/256 boundary are forbidden
4203 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
4204 * one is ASCIII. Since the pattern character is above
4205 * 255, and its only other match is below 256, the only
4206 * legal match will be to itself. We have thrown away
4207 * the original, so have to compute which is the one
4209 if ((c1 < 256) != (c2 < 256)) {
4210 if ((OP(text_node) == EXACTFL
4211 && ! IN_UTF8_CTYPE_LOCALE)
4212 || ((OP(text_node) == EXACTFA
4213 || OP(text_node) == EXACTFA_NO_TRIE)
4214 && (isASCII(c1) || isASCII(c2))))
4227 else /* Here, c1 is <= 255 */
4229 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
4230 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
4231 && ((OP(text_node) != EXACTFA
4232 && OP(text_node) != EXACTFA_NO_TRIE)
4235 /* Here, there could be something above Latin1 in the target
4236 * which folds to this character in the pattern. All such
4237 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
4238 * than two characters involved in their folds, so are outside
4239 * the scope of this function */
4240 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4241 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
4244 use_chrtest_void = TRUE;
4247 else { /* Here nothing above Latin1 can fold to the pattern
4249 switch (OP(text_node)) {
4251 case EXACTFL: /* /l rules */
4252 c2 = PL_fold_locale[c1];
4255 case EXACTF: /* This node only generated for non-utf8
4257 assert(! is_utf8_pat);
4258 if (! utf8_target) { /* /d rules */
4263 /* /u rules for all these. This happens to work for
4264 * EXACTFA as nothing in Latin1 folds to ASCII */
4265 case EXACTFA_NO_TRIE: /* This node only generated for
4266 non-utf8 patterns */
4267 assert(! is_utf8_pat);
4272 c2 = PL_fold_latin1[c1];
4276 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
4277 NOT_REACHED; /* NOTREACHED */
4283 /* Here have figured things out. Set up the returns */
4284 if (use_chrtest_void) {
4285 *c2p = *c1p = CHRTEST_VOID;
4287 else if (utf8_target) {
4288 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
4289 uvchr_to_utf8(c1_utf8, c1);
4290 uvchr_to_utf8(c2_utf8, c2);
4293 /* Invariants are stored in both the utf8 and byte outputs; Use
4294 * negative numbers otherwise for the byte ones. Make sure that the
4295 * byte ones are the same iff the utf8 ones are the same */
4296 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
4297 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
4300 ? CHRTEST_NOT_A_CP_1
4301 : CHRTEST_NOT_A_CP_2;
4303 else if (c1 > 255) {
4304 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
4309 *c1p = *c2p = c2; /* c2 is the only representable value */
4311 else { /* c1 is representable; see about c2 */
4313 *c2p = (c2 < 256) ? c2 : c1;
4320 S_isGCB(pTHX_ const GCB_enum before, const GCB_enum after, const U8 * const strbeg, const U8 * const curpos, const bool utf8_target)
4322 /* returns a boolean indicating if there is a Grapheme Cluster Boundary
4323 * between the inputs. See http://www.unicode.org/reports/tr29/. */
4325 PERL_ARGS_ASSERT_ISGCB;
4327 switch (GCB_table[before][after]) {
4334 case GCB_RI_then_RI:
4337 U8 * temp_pos = (U8 *) curpos;
4339 /* Do not break within emoji flag sequences. That is, do not
4340 * break between regional indicator (RI) symbols if there is an
4341 * odd number of RI characters before the break point.
4342 * GB12 ^ (RI RI)* RI × RI
4343 * GB13 [^RI] (RI RI)* RI × RI */
4345 while (backup_one_GCB(strbeg,
4347 utf8_target) == GCB_Regional_Indicator)
4352 return RI_count % 2 != 1;
4355 case GCB_EX_then_EM:
4357 /* GB10 ( E_Base | E_Base_GAZ ) Extend* × E_Modifier */
4359 U8 * temp_pos = (U8 *) curpos;
4363 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4365 while (prev == GCB_Extend);
4367 return prev != GCB_E_Base && prev != GCB_E_Base_GAZ;
4375 Perl_re_printf( aTHX_ "Unhandled GCB pair: GCB_table[%d, %d] = %d\n",
4376 before, after, GCB_table[before][after]);
4383 S_backup_one_GCB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4387 PERL_ARGS_ASSERT_BACKUP_ONE_GCB;
4389 if (*curpos < strbeg) {
4394 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4395 U8 * prev_prev_char_pos;
4397 if (! prev_char_pos) {
4401 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4402 gcb = getGCB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4403 *curpos = prev_char_pos;
4404 prev_char_pos = prev_prev_char_pos;
4407 *curpos = (U8 *) strbeg;
4412 if (*curpos - 2 < strbeg) {
4413 *curpos = (U8 *) strbeg;
4417 gcb = getGCB_VAL_CP(*(*curpos - 1));
4423 /* Combining marks attach to most classes that precede them, but this defines
4424 * the exceptions (from TR14) */
4425 #define LB_CM_ATTACHES_TO(prev) ( ! ( prev == LB_EDGE \
4426 || prev == LB_Mandatory_Break \
4427 || prev == LB_Carriage_Return \
4428 || prev == LB_Line_Feed \
4429 || prev == LB_Next_Line \
4430 || prev == LB_Space \
4431 || prev == LB_ZWSpace))
4434 S_isLB(pTHX_ LB_enum before,
4436 const U8 * const strbeg,
4437 const U8 * const curpos,
4438 const U8 * const strend,
4439 const bool utf8_target)
4441 U8 * temp_pos = (U8 *) curpos;
4442 LB_enum prev = before;
4444 /* Is the boundary between 'before' and 'after' line-breakable?
4445 * Most of this is just a table lookup of a generated table from Unicode
4446 * rules. But some rules require context to decide, and so have to be
4447 * implemented in code */
4449 PERL_ARGS_ASSERT_ISLB;
4451 /* Rule numbers in the comments below are as of Unicode 9.0 */
4455 switch (LB_table[before][after]) {
4460 case LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4463 case LB_SP_foo + LB_BREAKABLE:
4464 case LB_SP_foo + LB_NOBREAK:
4465 case LB_SP_foo + LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4467 /* When we have something following a SP, we have to look at the
4468 * context in order to know what to do.
4470 * SP SP should not reach here because LB7: Do not break before
4471 * spaces. (For two spaces in a row there is nothing that
4472 * overrides that) */
4473 assert(after != LB_Space);
4475 /* Here we have a space followed by a non-space. Mostly this is a
4476 * case of LB18: "Break after spaces". But there are complications
4477 * as the handling of spaces is somewhat tricky. They are in a
4478 * number of rules, which have to be applied in priority order, but
4479 * something earlier in the string can cause a rule to be skipped
4480 * and a lower priority rule invoked. A prime example is LB7 which
4481 * says don't break before a space. But rule LB8 (lower priority)
4482 * says that the first break opportunity after a ZW is after any
4483 * span of spaces immediately after it. If a ZW comes before a SP
4484 * in the input, rule LB8 applies, and not LB7. Other such rules
4485 * involve combining marks which are rules 9 and 10, but they may
4486 * override higher priority rules if they come earlier in the
4487 * string. Since we're doing random access into the middle of the
4488 * string, we have to look for rules that should get applied based
4489 * on both string position and priority. Combining marks do not
4490 * attach to either ZW nor SP, so we don't have to consider them
4493 * To check for LB8, we have to find the first non-space character
4494 * before this span of spaces */
4496 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4498 while (prev == LB_Space);
4500 /* LB8 Break before any character following a zero-width space,
4501 * even if one or more spaces intervene.
4503 * So if we have a ZW just before this span, and to get here this
4504 * is the final space in the span. */
4505 if (prev == LB_ZWSpace) {
4509 /* Here, not ZW SP+. There are several rules that have higher
4510 * priority than LB18 and can be resolved now, as they don't depend
4511 * on anything earlier in the string (except ZW, which we have
4512 * already handled). One of these rules is LB11 Do not break
4513 * before Word joiner, but we have specially encoded that in the
4514 * lookup table so it is caught by the single test below which
4515 * catches the other ones. */
4516 if (LB_table[LB_Space][after] - LB_SP_foo
4517 == LB_NOBREAK_EVEN_WITH_SP_BETWEEN)
4522 /* If we get here, we have to XXX consider combining marks. */
4523 if (prev == LB_Combining_Mark) {
4525 /* What happens with these depends on the character they
4528 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4530 while (prev == LB_Combining_Mark);
4532 /* Most times these attach to and inherit the characteristics
4533 * of that character, but not always, and when not, they are to
4534 * be treated as AL by rule LB10. */
4535 if (! LB_CM_ATTACHES_TO(prev)) {
4536 prev = LB_Alphabetic;
4540 /* Here, we have the character preceding the span of spaces all set
4541 * up. We follow LB18: "Break after spaces" unless the table shows
4542 * that is overriden */
4543 return LB_table[prev][after] != LB_NOBREAK_EVEN_WITH_SP_BETWEEN;
4547 /* We don't know how to treat the CM except by looking at the first
4548 * non-CM character preceding it. ZWJ is treated as CM */
4550 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4552 while (prev == LB_Combining_Mark || prev == LB_ZWJ);
4554 /* Here, 'prev' is that first earlier non-CM character. If the CM
4555 * attatches to it, then it inherits the behavior of 'prev'. If it
4556 * doesn't attach, it is to be treated as an AL */
4557 if (! LB_CM_ATTACHES_TO(prev)) {
4558 prev = LB_Alphabetic;
4563 case LB_HY_or_BA_then_foo + LB_BREAKABLE:
4564 case LB_HY_or_BA_then_foo + LB_NOBREAK:
4566 /* LB21a Don't break after Hebrew + Hyphen.
4567 * HL (HY | BA) × */
4569 if (backup_one_LB(strbeg, &temp_pos, utf8_target)
4570 == LB_Hebrew_Letter)
4575 return LB_table[prev][after] - LB_HY_or_BA_then_foo == LB_BREAKABLE;
4577 case LB_PR_or_PO_then_OP_or_HY + LB_BREAKABLE:
4578 case LB_PR_or_PO_then_OP_or_HY + LB_NOBREAK:
4580 /* LB25a (PR | PO) × ( OP | HY )? NU */
4581 if (advance_one_LB(&temp_pos, strend, utf8_target) == LB_Numeric) {
4585 return LB_table[prev][after] - LB_PR_or_PO_then_OP_or_HY
4588 case LB_SY_or_IS_then_various + LB_BREAKABLE:
4589 case LB_SY_or_IS_then_various + LB_NOBREAK:
4591 /* LB25d NU (SY | IS)* × (NU | SY | IS | CL | CP ) */
4593 LB_enum temp = prev;
4595 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4597 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric);
4598 if (temp == LB_Numeric) {
4602 return LB_table[prev][after] - LB_SY_or_IS_then_various
4606 case LB_various_then_PO_or_PR + LB_BREAKABLE:
4607 case LB_various_then_PO_or_PR + LB_NOBREAK:
4609 /* LB25e NU (SY | IS)* (CL | CP)? × (PO | PR) */
4611 LB_enum temp = prev;
4612 if (temp == LB_Close_Punctuation || temp == LB_Close_Parenthesis)
4614 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4616 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric) {
4617 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4619 if (temp == LB_Numeric) {
4622 return LB_various_then_PO_or_PR;
4625 case LB_RI_then_RI + LB_NOBREAK:
4626 case LB_RI_then_RI + LB_BREAKABLE:
4630 /* LB30a Break between two regional indicator symbols if and
4631 * only if there are an even number of regional indicators
4632 * preceding the position of the break.
4634 * sot (RI RI)* RI × RI
4635 * [^RI] (RI RI)* RI × RI */
4637 while (backup_one_LB(strbeg,
4639 utf8_target) == LB_Regional_Indicator)
4644 return RI_count % 2 == 0;
4652 Perl_re_printf( aTHX_ "Unhandled LB pair: LB_table[%d, %d] = %d\n",
4653 before, after, LB_table[before][after]);
4660 S_advance_one_LB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4664 PERL_ARGS_ASSERT_ADVANCE_ONE_LB;
4666 if (*curpos >= strend) {
4671 *curpos += UTF8SKIP(*curpos);
4672 if (*curpos >= strend) {
4675 lb = getLB_VAL_UTF8(*curpos, strend);
4679 if (*curpos >= strend) {
4682 lb = getLB_VAL_CP(**curpos);
4689 S_backup_one_LB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4693 PERL_ARGS_ASSERT_BACKUP_ONE_LB;
4695 if (*curpos < strbeg) {
4700 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4701 U8 * prev_prev_char_pos;
4703 if (! prev_char_pos) {
4707 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4708 lb = getLB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4709 *curpos = prev_char_pos;
4710 prev_char_pos = prev_prev_char_pos;
4713 *curpos = (U8 *) strbeg;
4718 if (*curpos - 2 < strbeg) {
4719 *curpos = (U8 *) strbeg;
4723 lb = getLB_VAL_CP(*(*curpos - 1));
4730 S_isSB(pTHX_ SB_enum before,
4732 const U8 * const strbeg,
4733 const U8 * const curpos,
4734 const U8 * const strend,
4735 const bool utf8_target)
4737 /* returns a boolean indicating if there is a Sentence Boundary Break
4738 * between the inputs. See http://www.unicode.org/reports/tr29/ */
4740 U8 * lpos = (U8 *) curpos;
4741 bool has_para_sep = FALSE;
4742 bool has_sp = FALSE;
4744 PERL_ARGS_ASSERT_ISSB;
4746 /* Break at the start and end of text.
4749 But unstated in Unicode is don't break if the text is empty */
4750 if (before == SB_EDGE || after == SB_EDGE) {
4751 return before != after;
4754 /* SB 3: Do not break within CRLF. */
4755 if (before == SB_CR && after == SB_LF) {
4759 /* Break after paragraph separators. CR and LF are considered
4760 * so because Unicode views text as like word processing text where there
4761 * are no newlines except between paragraphs, and the word processor takes
4762 * care of wrapping without there being hard line-breaks in the text *./
4763 SB4. Sep | CR | LF ÷ */
4764 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4768 /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF.
4769 * (See Section 6.2, Replacing Ignore Rules.)
4770 SB5. X (Extend | Format)* → X */
4771 if (after == SB_Extend || after == SB_Format) {
4773 /* Implied is that the these characters attach to everything
4774 * immediately prior to them except for those separator-type
4775 * characters. And the rules earlier have already handled the case
4776 * when one of those immediately precedes the extend char */
4780 if (before == SB_Extend || before == SB_Format) {
4781 U8 * temp_pos = lpos;
4782 const SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4783 if ( backup != SB_EDGE
4792 /* Here, both 'before' and 'backup' are these types; implied is that we
4793 * don't break between them */
4794 if (backup == SB_Extend || backup == SB_Format) {
4799 /* Do not break after ambiguous terminators like period, if they are
4800 * immediately followed by a number or lowercase letter, if they are
4801 * between uppercase letters, if the first following letter (optionally
4802 * after certain punctuation) is lowercase, or if they are followed by
4803 * "continuation" punctuation such as comma, colon, or semicolon. For
4804 * example, a period may be an abbreviation or numeric period, and thus may
4805 * not mark the end of a sentence.
4807 * SB6. ATerm × Numeric */
4808 if (before == SB_ATerm && after == SB_Numeric) {
4812 /* SB7. (Upper | Lower) ATerm × Upper */
4813 if (before == SB_ATerm && after == SB_Upper) {
4814 U8 * temp_pos = lpos;
4815 SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4816 if (backup == SB_Upper || backup == SB_Lower) {
4821 /* The remaining rules that aren't the final one, all require an STerm or
4822 * an ATerm after having backed up over some Close* Sp*, and in one case an
4823 * optional Paragraph separator, although one rule doesn't have any Sp's in it.
4824 * So do that backup now, setting flags if either Sp or a paragraph
4825 * separator are found */
4827 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4828 has_para_sep = TRUE;
4829 before = backup_one_SB(strbeg, &lpos, utf8_target);
4832 if (before == SB_Sp) {
4835 before = backup_one_SB(strbeg, &lpos, utf8_target);
4837 while (before == SB_Sp);
4840 while (before == SB_Close) {
4841 before = backup_one_SB(strbeg, &lpos, utf8_target);
4844 /* The next few rules apply only when the backed-up-to is an ATerm, and in
4845 * most cases an STerm */
4846 if (before == SB_STerm || before == SB_ATerm) {
4848 /* So, here the lhs matches
4849 * (STerm | ATerm) Close* Sp* (Sep | CR | LF)?
4850 * and we have set flags if we found an Sp, or the optional Sep,CR,LF.
4851 * The rules that apply here are:
4853 * SB8 ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR
4854 | LF | STerm | ATerm) )* Lower
4855 SB8a (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm)
4856 SB9 (STerm | ATerm) Close* × (Close | Sp | Sep | CR | LF)
4857 SB10 (STerm | ATerm) Close* Sp* × (Sp | Sep | CR | LF)
4858 SB11 (STerm | ATerm) Close* Sp* (Sep | CR | LF)? ÷
4861 /* And all but SB11 forbid having seen a paragraph separator */
4862 if (! has_para_sep) {
4863 if (before == SB_ATerm) { /* SB8 */
4864 U8 * rpos = (U8 *) curpos;
4865 SB_enum later = after;
4867 while ( later != SB_OLetter
4868 && later != SB_Upper
4869 && later != SB_Lower
4873 && later != SB_STerm
4874 && later != SB_ATerm
4875 && later != SB_EDGE)
4877 later = advance_one_SB(&rpos, strend, utf8_target);
4879 if (later == SB_Lower) {
4884 if ( after == SB_SContinue /* SB8a */
4885 || after == SB_STerm
4886 || after == SB_ATerm)
4891 if (! has_sp) { /* SB9 applies only if there was no Sp* */
4892 if ( after == SB_Close
4902 /* SB10. This and SB9 could probably be combined some way, but khw
4903 * has decided to follow the Unicode rule book precisely for
4904 * simplified maintenance */
4918 /* Otherwise, do not break.
4925 S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4929 PERL_ARGS_ASSERT_ADVANCE_ONE_SB;
4931 if (*curpos >= strend) {
4937 *curpos += UTF8SKIP(*curpos);
4938 if (*curpos >= strend) {
4941 sb = getSB_VAL_UTF8(*curpos, strend);
4942 } while (sb == SB_Extend || sb == SB_Format);
4947 if (*curpos >= strend) {
4950 sb = getSB_VAL_CP(**curpos);
4951 } while (sb == SB_Extend || sb == SB_Format);
4958 S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4962 PERL_ARGS_ASSERT_BACKUP_ONE_SB;
4964 if (*curpos < strbeg) {
4969 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4970 if (! prev_char_pos) {
4974 /* Back up over Extend and Format. curpos is always just to the right
4975 * of the characater whose value we are getting */
4977 U8 * prev_prev_char_pos;
4978 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1,
4981 sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4982 *curpos = prev_char_pos;
4983 prev_char_pos = prev_prev_char_pos;
4986 *curpos = (U8 *) strbeg;
4989 } while (sb == SB_Extend || sb == SB_Format);
4993 if (*curpos - 2 < strbeg) {
4994 *curpos = (U8 *) strbeg;
4998 sb = getSB_VAL_CP(*(*curpos - 1));
4999 } while (sb == SB_Extend || sb == SB_Format);
5006 S_isWB(pTHX_ WB_enum previous,
5009 const U8 * const strbeg,
5010 const U8 * const curpos,
5011 const U8 * const strend,
5012 const bool utf8_target)
5014 /* Return a boolean as to if the boundary between 'before' and 'after' is
5015 * a Unicode word break, using their published algorithm, but tailored for
5016 * Perl by treating spans of white space as one unit. Context may be
5017 * needed to make this determination. If the value for the character
5018 * before 'before' is known, it is passed as 'previous'; otherwise that
5019 * should be set to WB_UNKNOWN. The other input parameters give the
5020 * boundaries and current position in the matching of the string. That
5021 * is, 'curpos' marks the position where the character whose wb value is
5022 * 'after' begins. See http://www.unicode.org/reports/tr29/ */
5024 U8 * before_pos = (U8 *) curpos;
5025 U8 * after_pos = (U8 *) curpos;
5026 WB_enum prev = before;
5029 PERL_ARGS_ASSERT_ISWB;
5031 /* Rule numbers in the comments below are as of Unicode 9.0 */
5035 switch (WB_table[before][after]) {
5042 case WB_hs_then_hs: /* 2 horizontal spaces in a row */
5043 next = advance_one_WB(&after_pos, strend, utf8_target,
5044 FALSE /* Don't skip Extend nor Format */ );
5045 /* A space immediately preceeding an Extend or Format is attached
5046 * to by them, and hence gets separated from previous spaces.
5047 * Otherwise don't break between horizontal white space */
5048 return next == WB_Extend || next == WB_Format;
5050 /* WB4 Ignore Format and Extend characters, except when they appear at
5051 * the beginning of a region of text. This code currently isn't
5052 * general purpose, but it works as the rules are currently and likely
5053 * to be laid out. The reason it works is that when 'they appear at
5054 * the beginning of a region of text', the rule is to break before
5055 * them, just like any other character. Therefore, the default rule
5056 * applies and we don't have to look in more depth. Should this ever
5057 * change, we would have to have 2 'case' statements, like in the rules
5058 * below, and backup a single character (not spacing over the extend
5059 * ones) and then see if that is one of the region-end characters and
5061 case WB_Ex_or_FO_or_ZWJ_then_foo:
5062 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5065 case WB_DQ_then_HL + WB_BREAKABLE:
5066 case WB_DQ_then_HL + WB_NOBREAK:
5068 /* WB7c Hebrew_Letter Double_Quote × Hebrew_Letter */
5070 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5071 == WB_Hebrew_Letter)
5076 return WB_table[before][after] - WB_DQ_then_HL == WB_BREAKABLE;
5078 case WB_HL_then_DQ + WB_BREAKABLE:
5079 case WB_HL_then_DQ + WB_NOBREAK:
5081 /* WB7b Hebrew_Letter × Double_Quote Hebrew_Letter */
5083 if (advance_one_WB(&after_pos, strend, utf8_target,
5084 TRUE /* Do skip Extend and Format */ )
5085 == WB_Hebrew_Letter)
5090 return WB_table[before][after] - WB_HL_then_DQ == WB_BREAKABLE;
5092 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_NOBREAK:
5093 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_BREAKABLE:
5095 /* WB6 (ALetter | Hebrew_Letter) × (MidLetter | MidNumLet
5096 * | Single_Quote) (ALetter | Hebrew_Letter) */
5098 next = advance_one_WB(&after_pos, strend, utf8_target,
5099 TRUE /* Do skip Extend and Format */ );
5101 if (next == WB_ALetter || next == WB_Hebrew_Letter)
5106 return WB_table[before][after]
5107 - WB_LE_or_HL_then_MB_or_ML_or_SQ == WB_BREAKABLE;
5109 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_NOBREAK:
5110 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_BREAKABLE:
5112 /* WB7 (ALetter | Hebrew_Letter) (MidLetter | MidNumLet
5113 * | Single_Quote) × (ALetter | Hebrew_Letter) */
5115 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5116 if (prev == WB_ALetter || prev == WB_Hebrew_Letter)
5121 return WB_table[before][after]
5122 - WB_MB_or_ML_or_SQ_then_LE_or_HL == WB_BREAKABLE;
5124 case WB_MB_or_MN_or_SQ_then_NU + WB_NOBREAK:
5125 case WB_MB_or_MN_or_SQ_then_NU + WB_BREAKABLE:
5127 /* WB11 Numeric (MidNum | (MidNumLet | Single_Quote)) × Numeric
5130 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5136 return WB_table[before][after]
5137 - WB_MB_or_MN_or_SQ_then_NU == WB_BREAKABLE;
5139 case WB_NU_then_MB_or_MN_or_SQ + WB_NOBREAK:
5140 case WB_NU_then_MB_or_MN_or_SQ + WB_BREAKABLE:
5142 /* WB12 Numeric × (MidNum | MidNumLet | Single_Quote) Numeric */
5144 if (advance_one_WB(&after_pos, strend, utf8_target,
5145 TRUE /* Do skip Extend and Format */ )
5151 return WB_table[before][after]
5152 - WB_NU_then_MB_or_MN_or_SQ == WB_BREAKABLE;
5154 case WB_RI_then_RI + WB_NOBREAK:
5155 case WB_RI_then_RI + WB_BREAKABLE:
5159 /* Do not break within emoji flag sequences. That is, do not
5160 * break between regional indicator (RI) symbols if there is an
5161 * odd number of RI characters before the potential break
5164 * WB15 ^ (RI RI)* RI × RI
5165 * WB16 [^RI] (RI RI)* RI × RI */
5167 while (backup_one_WB(&previous,
5170 utf8_target) == WB_Regional_Indicator)
5175 return RI_count % 2 != 1;
5183 Perl_re_printf( aTHX_ "Unhandled WB pair: WB_table[%d, %d] = %d\n",
5184 before, after, WB_table[before][after]);
5191 S_advance_one_WB(pTHX_ U8 ** curpos,
5192 const U8 * const strend,
5193 const bool utf8_target,
5194 const bool skip_Extend_Format)
5198 PERL_ARGS_ASSERT_ADVANCE_ONE_WB;
5200 if (*curpos >= strend) {
5206 /* Advance over Extend and Format */
5208 *curpos += UTF8SKIP(*curpos);
5209 if (*curpos >= strend) {
5212 wb = getWB_VAL_UTF8(*curpos, strend);
5213 } while ( skip_Extend_Format
5214 && (wb == WB_Extend || wb == WB_Format));
5219 if (*curpos >= strend) {
5222 wb = getWB_VAL_CP(**curpos);
5223 } while ( skip_Extend_Format
5224 && (wb == WB_Extend || wb == WB_Format));
5231 S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5235 PERL_ARGS_ASSERT_BACKUP_ONE_WB;
5237 /* If we know what the previous character's break value is, don't have
5239 if (*previous != WB_UNKNOWN) {
5242 /* But we need to move backwards by one */
5244 *curpos = reghopmaybe3(*curpos, -1, strbeg);
5246 *previous = WB_EDGE;
5247 *curpos = (U8 *) strbeg;
5250 *previous = WB_UNKNOWN;
5255 *previous = (*curpos <= strbeg) ? WB_EDGE : WB_UNKNOWN;
5258 /* And we always back up over these three types */
5259 if (wb != WB_Extend && wb != WB_Format && wb != WB_ZWJ) {
5264 if (*curpos < strbeg) {
5269 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5270 if (! prev_char_pos) {
5274 /* Back up over Extend and Format. curpos is always just to the right
5275 * of the characater whose value we are getting */
5277 U8 * prev_prev_char_pos;
5278 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos,
5282 wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5283 *curpos = prev_char_pos;
5284 prev_char_pos = prev_prev_char_pos;
5287 *curpos = (U8 *) strbeg;
5290 } while (wb == WB_Extend || wb == WB_Format || wb == WB_ZWJ);
5294 if (*curpos - 2 < strbeg) {
5295 *curpos = (U8 *) strbeg;
5299 wb = getWB_VAL_CP(*(*curpos - 1));
5300 } while (wb == WB_Extend || wb == WB_Format);
5306 #define EVAL_CLOSE_PAREN_IS(st,expr) \
5309 ( ( st )->u.eval.close_paren ) && \
5310 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5313 #define EVAL_CLOSE_PAREN_IS_TRUE(st,expr) \
5316 ( ( st )->u.eval.close_paren ) && \
5318 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5322 #define EVAL_CLOSE_PAREN_SET(st,expr) \
5323 (st)->u.eval.close_paren = ( (expr) + 1 )
5325 #define EVAL_CLOSE_PAREN_CLEAR(st) \
5326 (st)->u.eval.close_paren = 0
5328 /* returns -1 on failure, $+[0] on success */
5330 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
5333 const bool utf8_target = reginfo->is_utf8_target;
5334 const U32 uniflags = UTF8_ALLOW_DEFAULT;
5335 REGEXP *rex_sv = reginfo->prog;
5336 regexp *rex = ReANY(rex_sv);
5337 RXi_GET_DECL(rex,rexi);
5338 /* the current state. This is a cached copy of PL_regmatch_state */
5340 /* cache heavy used fields of st in registers */
5343 U32 n = 0; /* general value; init to avoid compiler warning */
5344 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
5345 char *locinput = startpos;
5346 char *pushinput; /* where to continue after a PUSH */
5347 I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */
5349 bool result = 0; /* return value of S_regmatch */
5350 U32 depth = 0; /* depth of backtrack stack */
5351 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
5352 const U32 max_nochange_depth =
5353 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
5354 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
5355 regmatch_state *yes_state = NULL; /* state to pop to on success of
5357 /* mark_state piggy backs on the yes_state logic so that when we unwind
5358 the stack on success we can update the mark_state as we go */
5359 regmatch_state *mark_state = NULL; /* last mark state we have seen */
5360 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
5361 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
5363 bool no_final = 0; /* prevent failure from backtracking? */
5364 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
5365 char *startpoint = locinput;
5366 SV *popmark = NULL; /* are we looking for a mark? */
5367 SV *sv_commit = NULL; /* last mark name seen in failure */
5368 SV *sv_yes_mark = NULL; /* last mark name we have seen
5369 during a successful match */
5370 U32 lastopen = 0; /* last open we saw */
5371 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
5372 SV* const oreplsv = GvSVn(PL_replgv);
5373 /* these three flags are set by various ops to signal information to
5374 * the very next op. They have a useful lifetime of exactly one loop
5375 * iteration, and are not preserved or restored by state pushes/pops
5377 bool sw = 0; /* the condition value in (?(cond)a|b) */
5378 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
5379 int logical = 0; /* the following EVAL is:
5383 or the following IFMATCH/UNLESSM is:
5384 false: plain (?=foo)
5385 true: used as a condition: (?(?=foo))
5387 PAD* last_pad = NULL;
5389 U8 gimme = G_SCALAR;
5390 CV *caller_cv = NULL; /* who called us */
5391 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
5392 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
5393 U32 maxopenparen = 0; /* max '(' index seen so far */
5394 int to_complement; /* Invert the result? */
5395 _char_class_number classnum;
5396 bool is_utf8_pat = reginfo->is_utf8_pat;
5399 /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */
5400 #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL))
5401 # define SOLARIS_BAD_OPTIMIZER
5402 const U32 *pl_charclass_dup = PL_charclass;
5403 # define PL_charclass pl_charclass_dup
5407 GET_RE_DEBUG_FLAGS_DECL;
5410 /* protect against undef(*^R) */
5411 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
5413 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
5414 multicall_oldcatch = 0;
5415 PERL_UNUSED_VAR(multicall_cop);
5417 PERL_ARGS_ASSERT_REGMATCH;
5419 st = PL_regmatch_state;
5421 /* Note that nextchr is a byte even in UTF */
5425 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
5426 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5427 Perl_re_printf( aTHX_ "regmatch start\n" );
5430 while (scan != NULL) {
5433 next = scan + NEXT_OFF(scan);
5436 state_num = OP(scan);
5440 if (state_num <= REGNODE_MAX) {
5441 SV * const prop = sv_newmortal();
5442 regnode *rnext = regnext(scan);
5444 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5445 regprop(rex, prop, scan, reginfo, NULL);
5446 Perl_re_printf( aTHX_
5447 "%*s%" IVdf ":%s(%" IVdf ")\n",
5448 INDENT_CHARS(depth), "",
5449 (IV)(scan - rexi->program),
5451 (PL_regkind[OP(scan)] == END || !rnext) ?
5452 0 : (IV)(rnext - rexi->program));
5459 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
5461 switch (state_num) {
5462 case SBOL: /* /^../ and /\A../ */
5463 if (locinput == reginfo->strbeg)
5467 case MBOL: /* /^../m */
5468 if (locinput == reginfo->strbeg ||
5469 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
5476 if (locinput == reginfo->ganch)
5480 case KEEPS: /* \K */
5481 /* update the startpoint */
5482 st->u.keeper.val = rex->offs[0].start;
5483 rex->offs[0].start = locinput - reginfo->strbeg;
5484 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
5485 NOT_REACHED; /* NOTREACHED */
5487 case KEEPS_next_fail:
5488 /* rollback the start point change */
5489 rex->offs[0].start = st->u.keeper.val;
5491 NOT_REACHED; /* NOTREACHED */
5493 case MEOL: /* /..$/m */
5494 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5498 case SEOL: /* /..$/ */
5499 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5501 if (reginfo->strend - locinput > 1)
5506 if (!NEXTCHR_IS_EOS)
5510 case SANY: /* /./s */
5513 goto increment_locinput;
5515 case REG_ANY: /* /./ */
5516 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
5518 goto increment_locinput;
5522 #define ST st->u.trie
5523 case TRIEC: /* (ab|cd) with known charclass */
5524 /* In this case the charclass data is available inline so
5525 we can fail fast without a lot of extra overhead.
5527 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
5529 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n",
5530 depth, PL_colors[4], PL_colors[5])
5533 NOT_REACHED; /* NOTREACHED */
5536 case TRIE: /* (ab|cd) */
5537 /* the basic plan of execution of the trie is:
5538 * At the beginning, run though all the states, and
5539 * find the longest-matching word. Also remember the position
5540 * of the shortest matching word. For example, this pattern:
5543 * when matched against the string "abcde", will generate
5544 * accept states for all words except 3, with the longest
5545 * matching word being 4, and the shortest being 2 (with
5546 * the position being after char 1 of the string).
5548 * Then for each matching word, in word order (i.e. 1,2,4,5),
5549 * we run the remainder of the pattern; on each try setting
5550 * the current position to the character following the word,
5551 * returning to try the next word on failure.
5553 * We avoid having to build a list of words at runtime by
5554 * using a compile-time structure, wordinfo[].prev, which
5555 * gives, for each word, the previous accepting word (if any).
5556 * In the case above it would contain the mappings 1->2, 2->0,
5557 * 3->0, 4->5, 5->1. We can use this table to generate, from
5558 * the longest word (4 above), a list of all words, by
5559 * following the list of prev pointers; this gives us the
5560 * unordered list 4,5,1,2. Then given the current word we have
5561 * just tried, we can go through the list and find the
5562 * next-biggest word to try (so if we just failed on word 2,
5563 * the next in the list is 4).
5565 * Since at runtime we don't record the matching position in
5566 * the string for each word, we have to work that out for
5567 * each word we're about to process. The wordinfo table holds
5568 * the character length of each word; given that we recorded
5569 * at the start: the position of the shortest word and its
5570 * length in chars, we just need to move the pointer the
5571 * difference between the two char lengths. Depending on
5572 * Unicode status and folding, that's cheap or expensive.
5574 * This algorithm is optimised for the case where are only a
5575 * small number of accept states, i.e. 0,1, or maybe 2.
5576 * With lots of accepts states, and having to try all of them,
5577 * it becomes quadratic on number of accept states to find all
5582 /* what type of TRIE am I? (utf8 makes this contextual) */
5583 DECL_TRIE_TYPE(scan);
5585 /* what trie are we using right now */
5586 reg_trie_data * const trie
5587 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
5588 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
5589 U32 state = trie->startstate;
5591 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
5592 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5594 && UTF8_IS_ABOVE_LATIN1(nextchr)
5595 && scan->flags == EXACTL)
5597 /* We only output for EXACTL, as we let the folder
5598 * output this message for EXACTFLU8 to avoid
5600 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
5605 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
5607 if (trie->states[ state ].wordnum) {
5609 Perl_re_exec_indentf( aTHX_ "%smatched empty string...%s\n",
5610 depth, PL_colors[4], PL_colors[5])
5616 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n",
5617 depth, PL_colors[4], PL_colors[5])
5624 U8 *uc = ( U8* )locinput;
5628 U8 *uscan = (U8*)NULL;
5629 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
5630 U32 charcount = 0; /* how many input chars we have matched */
5631 U32 accepted = 0; /* have we seen any accepting states? */
5633 ST.jump = trie->jump;
5636 ST.longfold = FALSE; /* char longer if folded => it's harder */
5639 /* fully traverse the TRIE; note the position of the
5640 shortest accept state and the wordnum of the longest
5643 while ( state && uc <= (U8*)(reginfo->strend) ) {
5644 U32 base = trie->states[ state ].trans.base;
5648 wordnum = trie->states[ state ].wordnum;
5650 if (wordnum) { /* it's an accept state */
5653 /* record first match position */
5655 ST.firstpos = (U8*)locinput;
5660 ST.firstchars = charcount;
5663 if (!ST.nextword || wordnum < ST.nextword)
5664 ST.nextword = wordnum;
5665 ST.topword = wordnum;
5668 DEBUG_TRIE_EXECUTE_r({
5669 DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth );
5671 PerlIO_printf( Perl_debug_log,
5672 "%*s%sState: %4" UVxf " Accepted: %c ",
5673 INDENT_CHARS(depth), "", PL_colors[4],
5674 (UV)state, (accepted ? 'Y' : 'N'));
5677 /* read a char and goto next state */
5678 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
5680 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
5681 uscan, len, uvc, charid, foldlen,
5688 base + charid - 1 - trie->uniquecharcount)) >= 0)
5690 && ((U32)offset < trie->lasttrans)
5691 && trie->trans[offset].check == state)
5693 state = trie->trans[offset].next;
5704 DEBUG_TRIE_EXECUTE_r(
5705 Perl_re_printf( aTHX_
5706 "Charid:%3x CP:%4" UVxf " After State: %4" UVxf "%s\n",
5707 charid, uvc, (UV)state, PL_colors[5] );
5713 /* calculate total number of accept states */
5718 w = trie->wordinfo[w].prev;
5721 ST.accepted = accepted;
5725 Perl_re_exec_indentf( aTHX_ "%sgot %" IVdf " possible matches%s\n",
5727 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
5729 goto trie_first_try; /* jump into the fail handler */
5731 NOT_REACHED; /* NOTREACHED */
5733 case TRIE_next_fail: /* we failed - try next alternative */
5737 REGCP_UNWIND(ST.cp);
5738 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5740 if (!--ST.accepted) {
5742 Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n",
5750 /* Find next-highest word to process. Note that this code
5751 * is O(N^2) per trie run (O(N) per branch), so keep tight */
5754 U16 const nextword = ST.nextword;
5755 reg_trie_wordinfo * const wordinfo
5756 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
5757 for (word=ST.topword; word; word=wordinfo[word].prev) {
5758 if (word > nextword && (!min || word < min))
5771 ST.lastparen = rex->lastparen;
5772 ST.lastcloseparen = rex->lastcloseparen;
5776 /* find start char of end of current word */
5778 U32 chars; /* how many chars to skip */
5779 reg_trie_data * const trie
5780 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
5782 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
5784 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
5789 /* the hard option - fold each char in turn and find
5790 * its folded length (which may be different */
5791 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
5799 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
5807 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
5812 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
5828 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
5829 ? ST.jump[ST.nextword]
5833 Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n",
5841 if ( ST.accepted > 1 || has_cutgroup || ST.jump ) {
5842 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
5843 NOT_REACHED; /* NOTREACHED */
5845 /* only one choice left - just continue */
5847 AV *const trie_words
5848 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
5849 SV ** const tmp = trie_words
5850 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL;
5851 SV *sv= tmp ? sv_newmortal() : NULL;
5853 Perl_re_exec_indentf( aTHX_ "%sonly one match left, short-circuiting: #%d <%s>%s\n",
5854 depth, PL_colors[4],
5856 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
5857 PL_colors[0], PL_colors[1],
5858 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
5860 : "not compiled under -Dr",
5864 locinput = (char*)uc;
5865 continue; /* execute rest of RE */
5870 case EXACTL: /* /abc/l */
5871 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5873 /* Complete checking would involve going through every character
5874 * matched by the string to see if any is above latin1. But the
5875 * comparision otherwise might very well be a fast assembly
5876 * language routine, and I (khw) don't think slowing things down
5877 * just to check for this warning is worth it. So this just checks
5878 * the first character */
5879 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
5880 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
5883 case EXACT: { /* /abc/ */
5884 char *s = STRING(scan);
5886 if (utf8_target != is_utf8_pat) {
5887 /* The target and the pattern have differing utf8ness. */
5889 const char * const e = s + ln;
5892 /* The target is utf8, the pattern is not utf8.
5893 * Above-Latin1 code points can't match the pattern;
5894 * invariants match exactly, and the other Latin1 ones need
5895 * to be downgraded to a single byte in order to do the
5896 * comparison. (If we could be confident that the target
5897 * is not malformed, this could be refactored to have fewer
5898 * tests by just assuming that if the first bytes match, it
5899 * is an invariant, but there are tests in the test suite
5900 * dealing with (??{...}) which violate this) */
5902 if (l >= reginfo->strend
5903 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
5907 if (UTF8_IS_INVARIANT(*(U8*)l)) {
5914 if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
5924 /* The target is not utf8, the pattern is utf8. */
5926 if (l >= reginfo->strend
5927 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
5931 if (UTF8_IS_INVARIANT(*(U8*)s)) {
5938 if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
5950 /* The target and the pattern have the same utf8ness. */
5951 /* Inline the first character, for speed. */
5952 if (reginfo->strend - locinput < ln
5953 || UCHARAT(s) != nextchr
5954 || (ln > 1 && memNE(s, locinput, ln)))
5963 case EXACTFL: { /* /abc/il */
5965 const U8 * fold_array;
5967 U32 fold_utf8_flags;
5969 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5970 folder = foldEQ_locale;
5971 fold_array = PL_fold_locale;
5972 fold_utf8_flags = FOLDEQ_LOCALE;
5975 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
5976 is effectively /u; hence to match, target
5978 if (! utf8_target) {
5981 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S1_ALREADY_FOLDED
5982 | FOLDEQ_S1_FOLDS_SANE;
5983 folder = foldEQ_latin1;
5984 fold_array = PL_fold_latin1;
5987 case EXACTFU_SS: /* /\x{df}/iu */
5988 case EXACTFU: /* /abc/iu */
5989 folder = foldEQ_latin1;
5990 fold_array = PL_fold_latin1;
5991 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
5994 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
5996 assert(! is_utf8_pat);
5998 case EXACTFA: /* /abc/iaa */
5999 folder = foldEQ_latin1;
6000 fold_array = PL_fold_latin1;
6001 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6004 case EXACTF: /* /abc/i This node only generated for
6005 non-utf8 patterns */
6006 assert(! is_utf8_pat);
6008 fold_array = PL_fold;
6009 fold_utf8_flags = 0;
6017 || state_num == EXACTFU_SS
6018 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
6020 /* Either target or the pattern are utf8, or has the issue where
6021 * the fold lengths may differ. */
6022 const char * const l = locinput;
6023 char *e = reginfo->strend;
6025 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
6026 l, &e, 0, utf8_target, fold_utf8_flags))
6034 /* Neither the target nor the pattern are utf8 */
6035 if (UCHARAT(s) != nextchr
6037 && UCHARAT(s) != fold_array[nextchr])
6041 if (reginfo->strend - locinput < ln)
6043 if (ln > 1 && ! folder(s, locinput, ln))
6049 case NBOUNDL: /* /\B/l */
6053 case BOUNDL: /* /\b/l */
6056 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6058 if (FLAGS(scan) != TRADITIONAL_BOUND) {
6059 if (! IN_UTF8_CTYPE_LOCALE) {
6060 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
6061 B_ON_NON_UTF8_LOCALE_IS_WRONG);
6067 if (locinput == reginfo->strbeg)
6068 b1 = isWORDCHAR_LC('\n');
6070 b1 = isWORDCHAR_LC_utf8(reghop3((U8*)locinput, -1,
6071 (U8*)(reginfo->strbeg)));
6073 b2 = (NEXTCHR_IS_EOS)
6074 ? isWORDCHAR_LC('\n')
6075 : isWORDCHAR_LC_utf8((U8*)locinput);
6077 else { /* Here the string isn't utf8 */
6078 b1 = (locinput == reginfo->strbeg)
6079 ? isWORDCHAR_LC('\n')
6080 : isWORDCHAR_LC(UCHARAT(locinput - 1));
6081 b2 = (NEXTCHR_IS_EOS)
6082 ? isWORDCHAR_LC('\n')
6083 : isWORDCHAR_LC(nextchr);
6085 if (to_complement ^ (b1 == b2)) {
6091 case NBOUND: /* /\B/ */
6095 case BOUND: /* /\b/ */
6099 goto bound_ascii_match_only;
6101 case NBOUNDA: /* /\B/a */
6105 case BOUNDA: /* /\b/a */
6109 bound_ascii_match_only:
6110 /* Here the string isn't utf8, or is utf8 and only ascii characters
6111 * are to match \w. In the latter case looking at the byte just
6112 * prior to the current one may be just the final byte of a
6113 * multi-byte character. This is ok. There are two cases:
6114 * 1) it is a single byte character, and then the test is doing
6115 * just what it's supposed to.
6116 * 2) it is a multi-byte character, in which case the final byte is
6117 * never mistakable for ASCII, and so the test will say it is
6118 * not a word character, which is the correct answer. */
6119 b1 = (locinput == reginfo->strbeg)
6120 ? isWORDCHAR_A('\n')
6121 : isWORDCHAR_A(UCHARAT(locinput - 1));
6122 b2 = (NEXTCHR_IS_EOS)
6123 ? isWORDCHAR_A('\n')
6124 : isWORDCHAR_A(nextchr);
6125 if (to_complement ^ (b1 == b2)) {
6131 case NBOUNDU: /* /\B/u */
6135 case BOUNDU: /* /\b/u */
6138 if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) {
6141 else if (utf8_target) {
6143 switch((bound_type) FLAGS(scan)) {
6144 case TRADITIONAL_BOUND:
6147 b1 = (locinput == reginfo->strbeg)
6148 ? 0 /* isWORDCHAR_L1('\n') */
6149 : isWORDCHAR_utf8(reghop3((U8*)locinput, -1,
6150 (U8*)(reginfo->strbeg)));
6151 b2 = (NEXTCHR_IS_EOS)
6152 ? 0 /* isWORDCHAR_L1('\n') */
6153 : isWORDCHAR_utf8((U8*)locinput);
6154 match = cBOOL(b1 != b2);
6158 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6159 match = TRUE; /* GCB always matches at begin and
6163 /* Find the gcb values of previous and current
6164 * chars, then see if is a break point */
6165 match = isGCB(getGCB_VAL_UTF8(
6166 reghop3((U8*)locinput,
6168 (U8*)(reginfo->strbeg)),
6169 (U8*) reginfo->strend),
6170 getGCB_VAL_UTF8((U8*) locinput,
6171 (U8*) reginfo->strend),
6172 (U8*) reginfo->strbeg,
6179 if (locinput == reginfo->strbeg) {
6182 else if (NEXTCHR_IS_EOS) {
6186 match = isLB(getLB_VAL_UTF8(
6187 reghop3((U8*)locinput,
6189 (U8*)(reginfo->strbeg)),
6190 (U8*) reginfo->strend),
6191 getLB_VAL_UTF8((U8*) locinput,
6192 (U8*) reginfo->strend),
6193 (U8*) reginfo->strbeg,
6195 (U8*) reginfo->strend,
6200 case SB_BOUND: /* Always matches at begin and end */
6201 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6205 match = isSB(getSB_VAL_UTF8(
6206 reghop3((U8*)locinput,
6208 (U8*)(reginfo->strbeg)),
6209 (U8*) reginfo->strend),
6210 getSB_VAL_UTF8((U8*) locinput,
6211 (U8*) reginfo->strend),
6212 (U8*) reginfo->strbeg,
6214 (U8*) reginfo->strend,
6220 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6224 match = isWB(WB_UNKNOWN,
6226 reghop3((U8*)locinput,
6228 (U8*)(reginfo->strbeg)),
6229 (U8*) reginfo->strend),
6230 getWB_VAL_UTF8((U8*) locinput,
6231 (U8*) reginfo->strend),
6232 (U8*) reginfo->strbeg,
6234 (U8*) reginfo->strend,
6240 else { /* Not utf8 target */
6241 switch((bound_type) FLAGS(scan)) {
6242 case TRADITIONAL_BOUND:
6245 b1 = (locinput == reginfo->strbeg)
6246 ? 0 /* isWORDCHAR_L1('\n') */
6247 : isWORDCHAR_L1(UCHARAT(locinput - 1));
6248 b2 = (NEXTCHR_IS_EOS)
6249 ? 0 /* isWORDCHAR_L1('\n') */
6250 : isWORDCHAR_L1(nextchr);
6251 match = cBOOL(b1 != b2);
6256 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6257 match = TRUE; /* GCB always matches at begin and
6260 else { /* Only CR-LF combo isn't a GCB in 0-255
6262 match = UCHARAT(locinput - 1) != '\r'
6263 || UCHARAT(locinput) != '\n';
6268 if (locinput == reginfo->strbeg) {
6271 else if (NEXTCHR_IS_EOS) {
6275 match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)),
6276 getLB_VAL_CP(UCHARAT(locinput)),
6277 (U8*) reginfo->strbeg,
6279 (U8*) reginfo->strend,
6284 case SB_BOUND: /* Always matches at begin and end */
6285 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6289 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)),
6290 getSB_VAL_CP(UCHARAT(locinput)),
6291 (U8*) reginfo->strbeg,
6293 (U8*) reginfo->strend,
6299 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6303 match = isWB(WB_UNKNOWN,
6304 getWB_VAL_CP(UCHARAT(locinput -1)),
6305 getWB_VAL_CP(UCHARAT(locinput)),
6306 (U8*) reginfo->strbeg,
6308 (U8*) reginfo->strend,
6315 if (to_complement ^ ! match) {
6320 case ANYOFL: /* /[abc]/l */
6321 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6323 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE)
6325 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
6328 case ANYOFD: /* /[abc]/d */
6329 case ANYOF: /* /[abc]/ */
6332 if (utf8_target && ! UTF8_IS_INVARIANT(*locinput)) {
6333 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
6336 locinput += UTF8SKIP(locinput);
6339 if (!REGINCLASS(rex, scan, (U8*)locinput, utf8_target))
6345 /* The argument (FLAGS) to all the POSIX node types is the class number
6348 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
6352 case POSIXL: /* \w or [:punct:] etc. under /l */
6353 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6357 /* Use isFOO_lc() for characters within Latin1. (Note that
6358 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6359 * wouldn't be invariant) */
6360 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6361 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
6369 if (! UTF8_IS_DOWNGRADEABLE_START(nextchr)) { /* An above Latin-1 code point */
6370 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
6371 goto utf8_posix_above_latin1;
6374 /* Here is a UTF-8 variant code point below 256 and the target is
6376 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
6377 EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6378 *(locinput + 1))))))
6383 goto increment_locinput;
6385 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
6389 case POSIXD: /* \w or [:punct:] etc. under /d */
6395 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
6397 if (NEXTCHR_IS_EOS) {
6401 /* All UTF-8 variants match */
6402 if (! UTF8_IS_INVARIANT(nextchr)) {
6403 goto increment_locinput;
6409 case POSIXA: /* \w or [:punct:] etc. under /a */
6412 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
6413 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
6414 * character is a single byte */
6416 if (NEXTCHR_IS_EOS) {
6422 if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
6428 /* Here we are either not in utf8, or we matched a utf8-invariant,
6429 * so the next char is the next byte */
6433 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
6437 case POSIXU: /* \w or [:punct:] etc. under /u */
6439 if (NEXTCHR_IS_EOS) {
6443 /* Use _generic_isCC() for characters within Latin1. (Note that
6444 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6445 * wouldn't be invariant) */
6446 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6447 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
6454 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
6455 if (! (to_complement
6456 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6464 else { /* Handle above Latin-1 code points */
6465 utf8_posix_above_latin1:
6466 classnum = (_char_class_number) FLAGS(scan);
6467 if (classnum < _FIRST_NON_SWASH_CC) {
6469 /* Here, uses a swash to find such code points. Load if if
6470 * not done already */
6471 if (! PL_utf8_swash_ptrs[classnum]) {
6472 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
6473 PL_utf8_swash_ptrs[classnum]
6474 = _core_swash_init("utf8",
6477 PL_XPosix_ptrs[classnum], &flags);
6479 if (! (to_complement
6480 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
6481 (U8 *) locinput, TRUE))))
6486 else { /* Here, uses macros to find above Latin-1 code points */
6488 case _CC_ENUM_SPACE:
6489 if (! (to_complement
6490 ^ cBOOL(is_XPERLSPACE_high(locinput))))
6495 case _CC_ENUM_BLANK:
6496 if (! (to_complement
6497 ^ cBOOL(is_HORIZWS_high(locinput))))
6502 case _CC_ENUM_XDIGIT:
6503 if (! (to_complement
6504 ^ cBOOL(is_XDIGIT_high(locinput))))
6509 case _CC_ENUM_VERTSPACE:
6510 if (! (to_complement
6511 ^ cBOOL(is_VERTWS_high(locinput))))
6516 default: /* The rest, e.g. [:cntrl:], can't match
6518 if (! to_complement) {
6524 locinput += UTF8SKIP(locinput);
6528 case CLUMP: /* Match \X: logical Unicode character. This is defined as
6529 a Unicode extended Grapheme Cluster */
6532 if (! utf8_target) {
6534 /* Match either CR LF or '.', as all the other possibilities
6536 locinput++; /* Match the . or CR */
6537 if (nextchr == '\r' /* And if it was CR, and the next is LF,
6539 && locinput < reginfo->strend
6540 && UCHARAT(locinput) == '\n')
6547 /* Get the gcb type for the current character */
6548 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput,
6549 (U8*) reginfo->strend);
6551 /* Then scan through the input until we get to the first
6552 * character whose type is supposed to be a gcb with the
6553 * current character. (There is always a break at the
6555 locinput += UTF8SKIP(locinput);
6556 while (locinput < reginfo->strend) {
6557 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput,
6558 (U8*) reginfo->strend);
6559 if (isGCB(prev_gcb, cur_gcb,
6560 (U8*) reginfo->strbeg, (U8*) locinput,
6567 locinput += UTF8SKIP(locinput);
6574 case NREFFL: /* /\g{name}/il */
6575 { /* The capture buffer cases. The ones beginning with N for the
6576 named buffers just convert to the equivalent numbered and
6577 pretend they were called as the corresponding numbered buffer
6579 /* don't initialize these in the declaration, it makes C++
6584 const U8 *fold_array;
6587 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6588 folder = foldEQ_locale;
6589 fold_array = PL_fold_locale;
6591 utf8_fold_flags = FOLDEQ_LOCALE;
6594 case NREFFA: /* /\g{name}/iaa */
6595 folder = foldEQ_latin1;
6596 fold_array = PL_fold_latin1;
6598 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6601 case NREFFU: /* /\g{name}/iu */
6602 folder = foldEQ_latin1;
6603 fold_array = PL_fold_latin1;
6605 utf8_fold_flags = 0;
6608 case NREFF: /* /\g{name}/i */
6610 fold_array = PL_fold;
6612 utf8_fold_flags = 0;
6615 case NREF: /* /\g{name}/ */
6619 utf8_fold_flags = 0;
6622 /* For the named back references, find the corresponding buffer
6624 n = reg_check_named_buff_matched(rex,scan);
6629 goto do_nref_ref_common;
6631 case REFFL: /* /\1/il */
6632 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6633 folder = foldEQ_locale;
6634 fold_array = PL_fold_locale;
6635 utf8_fold_flags = FOLDEQ_LOCALE;
6638 case REFFA: /* /\1/iaa */
6639 folder = foldEQ_latin1;
6640 fold_array = PL_fold_latin1;
6641 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6644 case REFFU: /* /\1/iu */
6645 folder = foldEQ_latin1;
6646 fold_array = PL_fold_latin1;
6647 utf8_fold_flags = 0;
6650 case REFF: /* /\1/i */
6652 fold_array = PL_fold;
6653 utf8_fold_flags = 0;
6656 case REF: /* /\1/ */
6659 utf8_fold_flags = 0;
6663 n = ARG(scan); /* which paren pair */
6666 ln = rex->offs[n].start;
6667 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
6668 if (rex->lastparen < n || ln == -1)
6669 sayNO; /* Do not match unless seen CLOSEn. */
6670 if (ln == rex->offs[n].end)
6673 s = reginfo->strbeg + ln;
6674 if (type != REF /* REF can do byte comparison */
6675 && (utf8_target || type == REFFU || type == REFFL))
6677 char * limit = reginfo->strend;
6679 /* This call case insensitively compares the entire buffer
6680 * at s, with the current input starting at locinput, but
6681 * not going off the end given by reginfo->strend, and
6682 * returns in <limit> upon success, how much of the
6683 * current input was matched */
6684 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
6685 locinput, &limit, 0, utf8_target, utf8_fold_flags))
6693 /* Not utf8: Inline the first character, for speed. */
6694 if (!NEXTCHR_IS_EOS &&
6695 UCHARAT(s) != nextchr &&
6697 UCHARAT(s) != fold_array[nextchr]))
6699 ln = rex->offs[n].end - ln;
6700 if (locinput + ln > reginfo->strend)
6702 if (ln > 1 && (type == REF
6703 ? memNE(s, locinput, ln)
6704 : ! folder(s, locinput, ln)))
6710 case NOTHING: /* null op; e.g. the 'nothing' following
6711 * the '*' in m{(a+|b)*}' */
6713 case TAIL: /* placeholder while compiling (A|B|C) */
6717 #define ST st->u.eval
6718 #define CUR_EVAL cur_eval->u.eval
6724 regexp_internal *rei;
6725 regnode *startpoint;
6728 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
6729 arg= (U32)ARG(scan);
6730 if (cur_eval && cur_eval->locinput == locinput) {
6731 if ( ++nochange_depth > max_nochange_depth )
6733 "Pattern subroutine nesting without pos change"
6734 " exceeded limit in regex");
6741 startpoint = scan + ARG2L(scan);
6742 EVAL_CLOSE_PAREN_SET( st, arg );
6743 /* Detect infinite recursion
6745 * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/
6746 * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever.
6747 * So we track the position in the string we are at each time
6748 * we recurse and if we try to enter the same routine twice from
6749 * the same position we throw an error.
6751 if ( rex->recurse_locinput[arg] == locinput ) {
6752 /* FIXME: we should show the regop that is failing as part
6753 * of the error message. */
6754 Perl_croak(aTHX_ "Infinite recursion in regex");
6756 ST.prev_recurse_locinput= rex->recurse_locinput[arg];
6757 rex->recurse_locinput[arg]= locinput;
6760 GET_RE_DEBUG_FLAGS_DECL;
6762 Perl_re_exec_indentf( aTHX_
6763 "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n",
6764 depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg]
6770 /* Save all the positions seen so far. */
6771 ST.cp = regcppush(rex, 0, maxopenparen);
6772 REGCP_SET(ST.lastcp);
6774 /* and then jump to the code we share with EVAL */
6775 goto eval_recurse_doit;
6778 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
6779 if (cur_eval && cur_eval->locinput==locinput) {
6780 if ( ++nochange_depth > max_nochange_depth )
6781 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
6786 /* execute the code in the {...} */
6790 OP * const oop = PL_op;
6791 COP * const ocurcop = PL_curcop;
6795 /* save *all* paren positions */
6796 regcppush(rex, 0, maxopenparen);
6797 REGCP_SET(runops_cp);
6800 caller_cv = find_runcv(NULL);
6804 if (rexi->data->what[n] == 'r') { /* code from an external qr */
6806 (REGEXP*)(rexi->data->data[n])
6808 nop = (OP*)rexi->data->data[n+1];
6810 else if (rexi->data->what[n] == 'l') { /* literal code */
6812 nop = (OP*)rexi->data->data[n];
6813 assert(CvDEPTH(newcv));
6816 /* literal with own CV */
6817 assert(rexi->data->what[n] == 'L');
6818 newcv = rex->qr_anoncv;
6819 nop = (OP*)rexi->data->data[n];
6822 /* normally if we're about to execute code from the same
6823 * CV that we used previously, we just use the existing
6824 * CX stack entry. However, its possible that in the
6825 * meantime we may have backtracked, popped from the save
6826 * stack, and undone the SAVECOMPPAD(s) associated with
6827 * PUSH_MULTICALL; in which case PL_comppad no longer
6828 * points to newcv's pad. */
6829 if (newcv != last_pushed_cv || PL_comppad != last_pad)
6831 U8 flags = (CXp_SUB_RE |
6832 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
6833 if (last_pushed_cv) {
6834 /* PUSH/POP_MULTICALL save and restore the
6835 * caller's PL_comppad; if we call multiple subs
6836 * using the same CX block, we have to save and
6837 * unwind the varying PL_comppad's ourselves,
6838 * especially restoring the right PL_comppad on
6839 * backtrack - so save it on the save stack */
6841 CHANGE_MULTICALL_FLAGS(newcv, flags);
6844 PUSH_MULTICALL_FLAGS(newcv, flags);
6846 last_pushed_cv = newcv;
6849 /* these assignments are just to silence compiler
6851 multicall_cop = NULL;
6853 last_pad = PL_comppad;
6855 /* the initial nextstate you would normally execute
6856 * at the start of an eval (which would cause error
6857 * messages to come from the eval), may be optimised
6858 * away from the execution path in the regex code blocks;
6859 * so manually set PL_curcop to it initially */
6861 OP *o = cUNOPx(nop)->op_first;
6862 assert(o->op_type == OP_NULL);
6863 if (o->op_targ == OP_SCOPE) {
6864 o = cUNOPo->op_first;
6867 assert(o->op_targ == OP_LEAVE);
6868 o = cUNOPo->op_first;
6869 assert(o->op_type == OP_ENTER);
6873 if (o->op_type != OP_STUB) {
6874 assert( o->op_type == OP_NEXTSTATE
6875 || o->op_type == OP_DBSTATE
6876 || (o->op_type == OP_NULL
6877 && ( o->op_targ == OP_NEXTSTATE
6878 || o->op_targ == OP_DBSTATE
6882 PL_curcop = (COP*)o;
6887 DEBUG_STATE_r( Perl_re_printf( aTHX_
6888 " re EVAL PL_op=0x%" UVxf "\n", PTR2UV(nop)) );
6890 rex->offs[0].end = locinput - reginfo->strbeg;
6891 if (reginfo->info_aux_eval->pos_magic)
6892 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
6893 reginfo->sv, reginfo->strbeg,
6894 locinput - reginfo->strbeg);
6897 SV *sv_mrk = get_sv("REGMARK", 1);
6898 sv_setsv(sv_mrk, sv_yes_mark);
6901 /* we don't use MULTICALL here as we want to call the
6902 * first op of the block of interest, rather than the
6903 * first op of the sub. Also, we don't want to free
6904 * the savestack frame */
6905 before = (IV)(SP-PL_stack_base);
6907 CALLRUNOPS(aTHX); /* Scalar context. */
6909 if ((IV)(SP-PL_stack_base) == before)
6910 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
6916 /* before restoring everything, evaluate the returned
6917 * value, so that 'uninit' warnings don't use the wrong
6918 * PL_op or pad. Also need to process any magic vars
6919 * (e.g. $1) *before* parentheses are restored */
6924 if (logical == 0) /* (?{})/ */
6925 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
6926 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
6927 sw = cBOOL(SvTRUE(ret));
6930 else { /* /(??{}) */
6931 /* if its overloaded, let the regex compiler handle
6932 * it; otherwise extract regex, or stringify */
6933 if (SvGMAGICAL(ret))
6934 ret = sv_mortalcopy(ret);
6935 if (!SvAMAGIC(ret)) {
6939 if (SvTYPE(sv) == SVt_REGEXP)
6940 re_sv = (REGEXP*) sv;
6941 else if (SvSMAGICAL(ret)) {
6942 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
6944 re_sv = (REGEXP *) mg->mg_obj;
6947 /* force any undef warnings here */
6948 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
6949 ret = sv_mortalcopy(ret);
6950 (void) SvPV_force_nolen(ret);
6956 /* *** Note that at this point we don't restore
6957 * PL_comppad, (or pop the CxSUB) on the assumption it may
6958 * be used again soon. This is safe as long as nothing
6959 * in the regexp code uses the pad ! */
6961 PL_curcop = ocurcop;
6962 regcp_restore(rex, runops_cp, &maxopenparen);
6963 PL_curpm_under = PL_curpm;
6964 PL_curpm = PL_reg_curpm;
6970 /* only /(??{})/ from now on */
6973 /* extract RE object from returned value; compiling if
6977 re_sv = reg_temp_copy(NULL, re_sv);
6982 if (SvUTF8(ret) && IN_BYTES) {
6983 /* In use 'bytes': make a copy of the octet
6984 * sequence, but without the flag on */
6986 const char *const p = SvPV(ret, len);
6987 ret = newSVpvn_flags(p, len, SVs_TEMP);
6989 if (rex->intflags & PREGf_USE_RE_EVAL)
6990 pm_flags |= PMf_USE_RE_EVAL;
6992 /* if we got here, it should be an engine which
6993 * supports compiling code blocks and stuff */
6994 assert(rex->engine && rex->engine->op_comp);
6995 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
6996 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
6997 rex->engine, NULL, NULL,
6998 /* copy /msixn etc to inner pattern */
7003 & (SVs_TEMP | SVs_GMG | SVf_ROK))
7004 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
7005 /* This isn't a first class regexp. Instead, it's
7006 caching a regexp onto an existing, Perl visible
7008 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
7014 RXp_MATCH_COPIED_off(re);
7015 re->subbeg = rex->subbeg;
7016 re->sublen = rex->sublen;
7017 re->suboffset = rex->suboffset;
7018 re->subcoffset = rex->subcoffset;
7020 re->lastcloseparen = 0;
7023 debug_start_match(re_sv, utf8_target, locinput,
7024 reginfo->strend, "Matching embedded");
7026 startpoint = rei->program + 1;
7027 EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0;
7028 * close_paren only for GOSUB */
7029 ST.prev_recurse_locinput= NULL; /* only used for GOSUB */
7030 /* Save all the seen positions so far. */
7031 ST.cp = regcppush(rex, 0, maxopenparen);
7032 REGCP_SET(ST.lastcp);
7033 /* and set maxopenparen to 0, since we are starting a "fresh" match */
7035 /* run the pattern returned from (??{...}) */
7037 eval_recurse_doit: /* Share code with GOSUB below this line
7038 * At this point we expect the stack context to be
7039 * set up correctly */
7041 /* invalidate the S-L poscache. We're now executing a
7042 * different set of WHILEM ops (and their associated
7043 * indexes) against the same string, so the bits in the
7044 * cache are meaningless. Setting maxiter to zero forces
7045 * the cache to be invalidated and zeroed before reuse.
7046 * XXX This is too dramatic a measure. Ideally we should
7047 * save the old cache and restore when running the outer
7049 reginfo->poscache_maxiter = 0;
7051 /* the new regexp might have a different is_utf8_pat than we do */
7052 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
7054 ST.prev_rex = rex_sv;
7055 ST.prev_curlyx = cur_curlyx;
7057 SET_reg_curpm(rex_sv);
7062 ST.prev_eval = cur_eval;
7064 /* now continue from first node in postoned RE */
7065 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
7066 NOT_REACHED; /* NOTREACHED */
7069 case EVAL_AB: /* cleanup after a successful (??{A})B */
7070 /* note: this is called twice; first after popping B, then A */
7072 Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n",
7073 depth, cur_eval, ST.prev_eval);
7076 #define SET_RECURSE_LOCINPUT(STR,VAL)\
7077 if ( cur_eval && CUR_EVAL.close_paren ) {\
7079 Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\
7081 CUR_EVAL.close_paren - 1,\
7085 rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\
7088 SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput);
7090 rex_sv = ST.prev_rex;
7091 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7092 SET_reg_curpm(rex_sv);
7093 rex = ReANY(rex_sv);
7094 rexi = RXi_GET(rex);
7096 /* preserve $^R across LEAVE's. See Bug 121070. */
7097 SV *save_sv= GvSV(PL_replgv);
7098 SvREFCNT_inc(save_sv);
7099 regcpblow(ST.cp); /* LEAVE in disguise */
7100 sv_setsv(GvSV(PL_replgv), save_sv);
7101 SvREFCNT_dec(save_sv);
7103 cur_eval = ST.prev_eval;
7104 cur_curlyx = ST.prev_curlyx;
7106 /* Invalidate cache. See "invalidate" comment above. */
7107 reginfo->poscache_maxiter = 0;
7108 if ( nochange_depth )
7111 SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput);
7115 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
7116 /* note: this is called twice; first after popping B, then A */
7118 Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n",
7119 depth, cur_eval, ST.prev_eval);
7122 SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput);
7124 rex_sv = ST.prev_rex;
7125 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7126 SET_reg_curpm(rex_sv);
7127 rex = ReANY(rex_sv);
7128 rexi = RXi_GET(rex);
7130 REGCP_UNWIND(ST.lastcp);
7131 regcppop(rex, &maxopenparen);
7132 cur_eval = ST.prev_eval;
7133 cur_curlyx = ST.prev_curlyx;
7135 /* Invalidate cache. See "invalidate" comment above. */
7136 reginfo->poscache_maxiter = 0;
7137 if ( nochange_depth )
7140 SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput);
7145 n = ARG(scan); /* which paren pair */
7146 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
7147 if (n > maxopenparen)
7149 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
7150 "rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf " tmp; maxopenparen=%" UVuf "\n",
7155 (IV)rex->offs[n].start_tmp,
7161 /* XXX really need to log other places start/end are set too */
7162 #define CLOSE_CAPTURE \
7163 rex->offs[n].start = rex->offs[n].start_tmp; \
7164 rex->offs[n].end = locinput - reginfo->strbeg; \
7165 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
7166 "rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf "..%" IVdf "\n", \
7169 PTR2UV(rex->offs), \
7171 (IV)rex->offs[n].start, \
7172 (IV)rex->offs[n].end \
7176 n = ARG(scan); /* which paren pair */
7178 if (n > rex->lastparen)
7180 rex->lastcloseparen = n;
7181 if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) )
7186 case ACCEPT: /* (*ACCEPT) */
7188 sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7192 cursor && OP(cursor)!=END;
7193 cursor=regnext(cursor))
7195 if ( OP(cursor)==CLOSE ){
7197 if ( n <= lastopen ) {
7199 if (n > rex->lastparen)
7201 rex->lastcloseparen = n;
7202 if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) )
7211 case GROUPP: /* (?(1)) */
7212 n = ARG(scan); /* which paren pair */
7213 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
7216 case NGROUPP: /* (?(<name>)) */
7217 /* reg_check_named_buff_matched returns 0 for no match */
7218 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
7221 case INSUBP: /* (?(R)) */
7223 /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg
7224 * of SCAN is already set up as matches a eval.close_paren */
7225 sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n);
7228 case DEFINEP: /* (?(DEFINE)) */
7232 case IFTHEN: /* (?(cond)A|B) */
7233 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7235 next = NEXTOPER(NEXTOPER(scan));
7237 next = scan + ARG(scan);
7238 if (OP(next) == IFTHEN) /* Fake one. */
7239 next = NEXTOPER(NEXTOPER(next));
7243 case LOGICAL: /* modifier for EVAL and IFMATCH */
7244 logical = scan->flags;
7247 /*******************************************************************
7249 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
7250 pattern, where A and B are subpatterns. (For simple A, CURLYM or
7251 STAR/PLUS/CURLY/CURLYN are used instead.)
7253 A*B is compiled as <CURLYX><A><WHILEM><B>
7255 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
7256 state, which contains the current count, initialised to -1. It also sets
7257 cur_curlyx to point to this state, with any previous value saved in the
7260 CURLYX then jumps straight to the WHILEM op, rather than executing A,
7261 since the pattern may possibly match zero times (i.e. it's a while {} loop
7262 rather than a do {} while loop).
7264 Each entry to WHILEM represents a successful match of A. The count in the
7265 CURLYX block is incremented, another WHILEM state is pushed, and execution
7266 passes to A or B depending on greediness and the current count.
7268 For example, if matching against the string a1a2a3b (where the aN are
7269 substrings that match /A/), then the match progresses as follows: (the
7270 pushed states are interspersed with the bits of strings matched so far):
7273 <CURLYX cnt=0><WHILEM>
7274 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
7275 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
7276 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
7277 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
7279 (Contrast this with something like CURLYM, which maintains only a single
7283 a1 <CURLYM cnt=1> a2
7284 a1 a2 <CURLYM cnt=2> a3
7285 a1 a2 a3 <CURLYM cnt=3> b
7288 Each WHILEM state block marks a point to backtrack to upon partial failure
7289 of A or B, and also contains some minor state data related to that
7290 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
7291 overall state, such as the count, and pointers to the A and B ops.
7293 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
7294 must always point to the *current* CURLYX block, the rules are:
7296 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
7297 and set cur_curlyx to point the new block.
7299 When popping the CURLYX block after a successful or unsuccessful match,
7300 restore the previous cur_curlyx.
7302 When WHILEM is about to execute B, save the current cur_curlyx, and set it
7303 to the outer one saved in the CURLYX block.
7305 When popping the WHILEM block after a successful or unsuccessful B match,
7306 restore the previous cur_curlyx.
7308 Here's an example for the pattern (AI* BI)*BO
7309 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
7312 curlyx backtrack stack
7313 ------ ---------------
7315 CO <CO prev=NULL> <WO>
7316 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7317 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7318 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
7320 At this point the pattern succeeds, and we work back down the stack to
7321 clean up, restoring as we go:
7323 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7324 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7325 CO <CO prev=NULL> <WO>
7328 *******************************************************************/
7330 #define ST st->u.curlyx
7332 case CURLYX: /* start of /A*B/ (for complex A) */
7334 /* No need to save/restore up to this paren */
7335 I32 parenfloor = scan->flags;
7337 assert(next); /* keep Coverity happy */
7338 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
7341 /* XXXX Probably it is better to teach regpush to support
7342 parenfloor > maxopenparen ... */
7343 if (parenfloor > (I32)rex->lastparen)
7344 parenfloor = rex->lastparen; /* Pessimization... */
7346 ST.prev_curlyx= cur_curlyx;
7348 ST.cp = PL_savestack_ix;
7350 /* these fields contain the state of the current curly.
7351 * they are accessed by subsequent WHILEMs */
7352 ST.parenfloor = parenfloor;
7357 ST.count = -1; /* this will be updated by WHILEM */
7358 ST.lastloc = NULL; /* this will be updated by WHILEM */
7360 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
7361 NOT_REACHED; /* NOTREACHED */
7364 case CURLYX_end: /* just finished matching all of A*B */
7365 cur_curlyx = ST.prev_curlyx;
7367 NOT_REACHED; /* NOTREACHED */
7369 case CURLYX_end_fail: /* just failed to match all of A*B */
7371 cur_curlyx = ST.prev_curlyx;
7373 NOT_REACHED; /* NOTREACHED */
7377 #define ST st->u.whilem
7379 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
7381 /* see the discussion above about CURLYX/WHILEM */
7386 assert(cur_curlyx); /* keep Coverity happy */
7388 min = ARG1(cur_curlyx->u.curlyx.me);
7389 max = ARG2(cur_curlyx->u.curlyx.me);
7390 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
7391 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
7392 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
7393 ST.cache_offset = 0;
7397 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: matched %ld out of %d..%d\n",
7398 depth, (long)n, min, max)
7401 /* First just match a string of min A's. */
7404 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, maxopenparen);
7405 cur_curlyx->u.curlyx.lastloc = locinput;
7406 REGCP_SET(ST.lastcp);
7408 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
7409 NOT_REACHED; /* NOTREACHED */
7412 /* If degenerate A matches "", assume A done. */
7414 if (locinput == cur_curlyx->u.curlyx.lastloc) {
7415 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: empty match detected, trying continuation...\n",
7418 goto do_whilem_B_max;
7421 /* super-linear cache processing.
7423 * The idea here is that for certain types of CURLYX/WHILEM -
7424 * principally those whose upper bound is infinity (and
7425 * excluding regexes that have things like \1 and other very
7426 * non-regular expresssiony things), then if a pattern like
7427 * /....A*.../ fails and we backtrack to the WHILEM, then we
7428 * make a note that this particular WHILEM op was at string
7429 * position 47 (say) when the rest of pattern failed. Then, if
7430 * we ever find ourselves back at that WHILEM, and at string
7431 * position 47 again, we can just fail immediately rather than
7432 * running the rest of the pattern again.
7434 * This is very handy when patterns start to go
7435 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
7436 * with a combinatorial explosion of backtracking.
7438 * The cache is implemented as a bit array, with one bit per
7439 * string byte position per WHILEM op (up to 16) - so its
7440 * between 0.25 and 2x the string size.
7442 * To avoid allocating a poscache buffer every time, we do an
7443 * initially countdown; only after we have executed a WHILEM
7444 * op (string-length x #WHILEMs) times do we allocate the
7447 * The top 4 bits of scan->flags byte say how many different
7448 * relevant CURLLYX/WHILEM op pairs there are, while the
7449 * bottom 4-bits is the identifying index number of this
7455 if (!reginfo->poscache_maxiter) {
7456 /* start the countdown: Postpone detection until we
7457 * know the match is not *that* much linear. */
7458 reginfo->poscache_maxiter
7459 = (reginfo->strend - reginfo->strbeg + 1)
7461 /* possible overflow for long strings and many CURLYX's */
7462 if (reginfo->poscache_maxiter < 0)
7463 reginfo->poscache_maxiter = I32_MAX;
7464 reginfo->poscache_iter = reginfo->poscache_maxiter;
7467 if (reginfo->poscache_iter-- == 0) {
7468 /* initialise cache */
7469 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
7470 regmatch_info_aux *const aux = reginfo->info_aux;
7471 if (aux->poscache) {
7472 if ((SSize_t)reginfo->poscache_size < size) {
7473 Renew(aux->poscache, size, char);
7474 reginfo->poscache_size = size;
7476 Zero(aux->poscache, size, char);
7479 reginfo->poscache_size = size;
7480 Newxz(aux->poscache, size, char);
7482 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
7483 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
7484 PL_colors[4], PL_colors[5])
7488 if (reginfo->poscache_iter < 0) {
7489 /* have we already failed at this position? */
7490 SSize_t offset, mask;
7492 reginfo->poscache_iter = -1; /* stop eventual underflow */
7493 offset = (scan->flags & 0xf) - 1
7494 + (locinput - reginfo->strbeg)
7496 mask = 1 << (offset % 8);
7498 if (reginfo->info_aux->poscache[offset] & mask) {
7499 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: (cache) already tried at this position...\n",
7502 cur_curlyx->u.curlyx.count--;
7503 sayNO; /* cache records failure */
7505 ST.cache_offset = offset;
7506 ST.cache_mask = mask;
7510 /* Prefer B over A for minimal matching. */
7512 if (cur_curlyx->u.curlyx.minmod) {
7513 ST.save_curlyx = cur_curlyx;
7514 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
7515 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
7517 REGCP_SET(ST.lastcp);
7518 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
7520 NOT_REACHED; /* NOTREACHED */
7523 /* Prefer A over B for maximal matching. */
7525 if (n < max) { /* More greed allowed? */
7526 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
7528 cur_curlyx->u.curlyx.lastloc = locinput;
7529 REGCP_SET(ST.lastcp);
7530 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
7531 NOT_REACHED; /* NOTREACHED */
7533 goto do_whilem_B_max;
7535 NOT_REACHED; /* NOTREACHED */
7537 case WHILEM_B_min: /* just matched B in a minimal match */
7538 case WHILEM_B_max: /* just matched B in a maximal match */
7539 cur_curlyx = ST.save_curlyx;
7541 NOT_REACHED; /* NOTREACHED */
7543 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
7544 cur_curlyx = ST.save_curlyx;
7545 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7546 cur_curlyx->u.curlyx.count--;
7548 NOT_REACHED; /* NOTREACHED */
7550 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
7552 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
7553 REGCP_UNWIND(ST.lastcp);
7554 regcppop(rex, &maxopenparen);
7555 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7556 cur_curlyx->u.curlyx.count--;
7558 NOT_REACHED; /* NOTREACHED */
7560 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
7561 REGCP_UNWIND(ST.lastcp);
7562 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
7563 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "whilem: failed, trying continuation...\n",
7567 if (cur_curlyx->u.curlyx.count >= REG_INFTY
7568 && ckWARN(WARN_REGEXP)
7569 && !reginfo->warned)
7571 reginfo->warned = TRUE;
7572 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
7573 "Complex regular subexpression recursion limit (%d) "
7579 ST.save_curlyx = cur_curlyx;
7580 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
7581 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
7583 NOT_REACHED; /* NOTREACHED */
7585 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
7586 cur_curlyx = ST.save_curlyx;
7587 REGCP_UNWIND(ST.lastcp);
7588 regcppop(rex, &maxopenparen);
7590 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
7591 /* Maximum greed exceeded */
7592 if (cur_curlyx->u.curlyx.count >= REG_INFTY
7593 && ckWARN(WARN_REGEXP)
7594 && !reginfo->warned)
7596 reginfo->warned = TRUE;
7597 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
7598 "Complex regular subexpression recursion "
7599 "limit (%d) exceeded",
7602 cur_curlyx->u.curlyx.count--;
7606 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "trying longer...\n", depth)
7608 /* Try grabbing another A and see if it helps. */
7609 cur_curlyx->u.curlyx.lastloc = locinput;
7610 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
7612 REGCP_SET(ST.lastcp);
7613 PUSH_STATE_GOTO(WHILEM_A_min,
7614 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
7616 NOT_REACHED; /* NOTREACHED */
7619 #define ST st->u.branch
7621 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
7622 next = scan + ARG(scan);
7625 scan = NEXTOPER(scan);
7628 case BRANCH: /* /(...|A|...)/ */
7629 scan = NEXTOPER(scan); /* scan now points to inner node */
7630 ST.lastparen = rex->lastparen;
7631 ST.lastcloseparen = rex->lastcloseparen;
7632 ST.next_branch = next;
7635 /* Now go into the branch */
7637 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
7639 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
7641 NOT_REACHED; /* NOTREACHED */
7643 case CUTGROUP: /* /(*THEN)/ */
7644 sv_yes_mark = st->u.mark.mark_name = scan->flags
7645 ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ])
7647 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
7648 NOT_REACHED; /* NOTREACHED */
7650 case CUTGROUP_next_fail:
7653 if (st->u.mark.mark_name)
7654 sv_commit = st->u.mark.mark_name;
7656 NOT_REACHED; /* NOTREACHED */
7660 NOT_REACHED; /* NOTREACHED */
7662 case BRANCH_next_fail: /* that branch failed; try the next, if any */
7667 REGCP_UNWIND(ST.cp);
7668 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7669 scan = ST.next_branch;
7670 /* no more branches? */
7671 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
7673 Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n",
7680 continue; /* execute next BRANCH[J] op */
7683 case MINMOD: /* next op will be non-greedy, e.g. A*? */
7688 #define ST st->u.curlym
7690 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
7692 /* This is an optimisation of CURLYX that enables us to push
7693 * only a single backtracking state, no matter how many matches
7694 * there are in {m,n}. It relies on the pattern being constant
7695 * length, with no parens to influence future backrefs
7699 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7701 ST.lastparen = rex->lastparen;
7702 ST.lastcloseparen = rex->lastcloseparen;
7704 /* if paren positive, emulate an OPEN/CLOSE around A */
7706 U32 paren = ST.me->flags;
7707 if (paren > maxopenparen)
7708 maxopenparen = paren;
7709 scan += NEXT_OFF(scan); /* Skip former OPEN. */
7717 ST.c1 = CHRTEST_UNINIT;
7720 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
7723 curlym_do_A: /* execute the A in /A{m,n}B/ */
7724 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
7725 NOT_REACHED; /* NOTREACHED */
7727 case CURLYM_A: /* we've just matched an A */
7729 /* after first match, determine A's length: u.curlym.alen */
7730 if (ST.count == 1) {
7731 if (reginfo->is_utf8_target) {
7732 char *s = st->locinput;
7733 while (s < locinput) {
7739 ST.alen = locinput - st->locinput;
7742 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
7745 Perl_re_exec_indentf( aTHX_ "CURLYM now matched %" IVdf " times, len=%" IVdf "...\n",
7746 depth, (IV) ST.count, (IV)ST.alen)
7749 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7753 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
7754 if ( max == REG_INFTY || ST.count < max )
7755 goto curlym_do_A; /* try to match another A */
7757 goto curlym_do_B; /* try to match B */
7759 case CURLYM_A_fail: /* just failed to match an A */
7760 REGCP_UNWIND(ST.cp);
7763 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
7764 || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7767 curlym_do_B: /* execute the B in /A{m,n}B/ */
7768 if (ST.c1 == CHRTEST_UNINIT) {
7769 /* calculate c1 and c2 for possible match of 1st char
7770 * following curly */
7771 ST.c1 = ST.c2 = CHRTEST_VOID;
7773 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
7774 regnode *text_node = ST.B;
7775 if (! HAS_TEXT(text_node))
7776 FIND_NEXT_IMPT(text_node);
7779 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
7781 But the former is redundant in light of the latter.
7783 if this changes back then the macro for
7784 IS_TEXT and friends need to change.
7786 if (PL_regkind[OP(text_node)] == EXACT) {
7787 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
7788 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
7798 Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%" IVdf "...\n",
7799 depth, (IV)ST.count)
7801 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
7802 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
7803 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
7804 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
7806 /* simulate B failing */
7808 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%" UVXf " c1=0x%" UVXf " c2=0x%" UVXf "\n",
7810 valid_utf8_to_uvchr((U8 *) locinput, NULL),
7811 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
7812 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
7814 state_num = CURLYM_B_fail;
7815 goto reenter_switch;
7818 else if (nextchr != ST.c1 && nextchr != ST.c2) {
7819 /* simulate B failing */
7821 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
7823 (int) nextchr, ST.c1, ST.c2)
7825 state_num = CURLYM_B_fail;
7826 goto reenter_switch;
7831 /* emulate CLOSE: mark current A as captured */
7832 I32 paren = ST.me->flags;
7834 rex->offs[paren].start
7835 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
7836 rex->offs[paren].end = locinput - reginfo->strbeg;
7837 if ((U32)paren > rex->lastparen)
7838 rex->lastparen = paren;
7839 rex->lastcloseparen = paren;
7842 rex->offs[paren].end = -1;
7844 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7853 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
7854 NOT_REACHED; /* NOTREACHED */
7856 case CURLYM_B_fail: /* just failed to match a B */
7857 REGCP_UNWIND(ST.cp);
7858 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7860 I32 max = ARG2(ST.me);
7861 if (max != REG_INFTY && ST.count == max)
7863 goto curlym_do_A; /* try to match a further A */
7865 /* backtrack one A */
7866 if (ST.count == ARG1(ST.me) /* min */)
7869 SET_locinput(HOPc(locinput, -ST.alen));
7870 goto curlym_do_B; /* try to match B */
7873 #define ST st->u.curly
7875 #define CURLY_SETPAREN(paren, success) \
7878 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
7879 rex->offs[paren].end = locinput - reginfo->strbeg; \
7880 if (paren > rex->lastparen) \
7881 rex->lastparen = paren; \
7882 rex->lastcloseparen = paren; \
7885 rex->offs[paren].end = -1; \
7886 rex->lastparen = ST.lastparen; \
7887 rex->lastcloseparen = ST.lastcloseparen; \
7891 case STAR: /* /A*B/ where A is width 1 char */
7895 scan = NEXTOPER(scan);
7898 case PLUS: /* /A+B/ where A is width 1 char */
7902 scan = NEXTOPER(scan);
7905 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
7906 ST.paren = scan->flags; /* Which paren to set */
7907 ST.lastparen = rex->lastparen;
7908 ST.lastcloseparen = rex->lastcloseparen;
7909 if (ST.paren > maxopenparen)
7910 maxopenparen = ST.paren;
7911 ST.min = ARG1(scan); /* min to match */
7912 ST.max = ARG2(scan); /* max to match */
7913 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
7918 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
7921 case CURLY: /* /A{m,n}B/ where A is width 1 char */
7923 ST.min = ARG1(scan); /* min to match */
7924 ST.max = ARG2(scan); /* max to match */
7925 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7928 * Lookahead to avoid useless match attempts
7929 * when we know what character comes next.
7931 * Used to only do .*x and .*?x, but now it allows
7932 * for )'s, ('s and (?{ ... })'s to be in the way
7933 * of the quantifier and the EXACT-like node. -- japhy
7936 assert(ST.min <= ST.max);
7937 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
7938 ST.c1 = ST.c2 = CHRTEST_VOID;
7941 regnode *text_node = next;
7943 if (! HAS_TEXT(text_node))
7944 FIND_NEXT_IMPT(text_node);
7946 if (! HAS_TEXT(text_node))
7947 ST.c1 = ST.c2 = CHRTEST_VOID;
7949 if ( PL_regkind[OP(text_node)] != EXACT ) {
7950 ST.c1 = ST.c2 = CHRTEST_VOID;
7954 /* Currently we only get here when
7956 PL_rekind[OP(text_node)] == EXACT
7958 if this changes back then the macro for IS_TEXT and
7959 friends need to change. */
7960 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
7961 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
7973 char *li = locinput;
7976 regrepeat(rex, &li, ST.A, reginfo, ST.min)
7982 if (ST.c1 == CHRTEST_VOID)
7983 goto curly_try_B_min;
7985 ST.oldloc = locinput;
7987 /* set ST.maxpos to the furthest point along the
7988 * string that could possibly match */
7989 if (ST.max == REG_INFTY) {
7990 ST.maxpos = reginfo->strend - 1;
7992 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
7995 else if (utf8_target) {
7996 int m = ST.max - ST.min;
7997 for (ST.maxpos = locinput;
7998 m >0 && ST.maxpos < reginfo->strend; m--)
7999 ST.maxpos += UTF8SKIP(ST.maxpos);
8002 ST.maxpos = locinput + ST.max - ST.min;
8003 if (ST.maxpos >= reginfo->strend)
8004 ST.maxpos = reginfo->strend - 1;
8006 goto curly_try_B_min_known;
8010 /* avoid taking address of locinput, so it can remain
8012 char *li = locinput;
8013 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max);
8014 if (ST.count < ST.min)
8017 if ((ST.count > ST.min)
8018 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
8020 /* A{m,n} must come at the end of the string, there's
8021 * no point in backing off ... */
8023 /* ...except that $ and \Z can match before *and* after
8024 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
8025 We may back off by one in this case. */
8026 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
8030 goto curly_try_B_max;
8032 NOT_REACHED; /* NOTREACHED */
8034 case CURLY_B_min_known_fail:
8035 /* failed to find B in a non-greedy match where c1,c2 valid */
8037 REGCP_UNWIND(ST.cp);
8039 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8041 /* Couldn't or didn't -- move forward. */
8042 ST.oldloc = locinput;
8044 locinput += UTF8SKIP(locinput);
8048 curly_try_B_min_known:
8049 /* find the next place where 'B' could work, then call B */
8053 n = (ST.oldloc == locinput) ? 0 : 1;
8054 if (ST.c1 == ST.c2) {
8055 /* set n to utf8_distance(oldloc, locinput) */
8056 while (locinput <= ST.maxpos
8057 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
8059 locinput += UTF8SKIP(locinput);
8064 /* set n to utf8_distance(oldloc, locinput) */
8065 while (locinput <= ST.maxpos
8066 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
8067 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
8069 locinput += UTF8SKIP(locinput);
8074 else { /* Not utf8_target */
8075 if (ST.c1 == ST.c2) {
8076 while (locinput <= ST.maxpos &&
8077 UCHARAT(locinput) != ST.c1)
8081 while (locinput <= ST.maxpos
8082 && UCHARAT(locinput) != ST.c1
8083 && UCHARAT(locinput) != ST.c2)
8086 n = locinput - ST.oldloc;
8088 if (locinput > ST.maxpos)
8091 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
8092 * at b; check that everything between oldloc and
8093 * locinput matches */
8094 char *li = ST.oldloc;
8096 if (regrepeat(rex, &li, ST.A, reginfo, n) < n)
8098 assert(n == REG_INFTY || locinput == li);
8100 CURLY_SETPAREN(ST.paren, ST.count);
8101 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8103 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
8105 NOT_REACHED; /* NOTREACHED */
8107 case CURLY_B_min_fail:
8108 /* failed to find B in a non-greedy match where c1,c2 invalid */
8110 REGCP_UNWIND(ST.cp);
8112 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8114 /* failed -- move forward one */
8116 char *li = locinput;
8117 if (!regrepeat(rex, &li, ST.A, reginfo, 1)) {
8124 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
8125 ST.count > 0)) /* count overflow ? */
8128 CURLY_SETPAREN(ST.paren, ST.count);
8129 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8131 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
8135 NOT_REACHED; /* NOTREACHED */
8138 /* a successful greedy match: now try to match B */
8139 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8142 bool could_match = locinput < reginfo->strend;
8144 /* If it could work, try it. */
8145 if (ST.c1 != CHRTEST_VOID && could_match) {
8146 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
8148 could_match = memEQ(locinput,
8153 UTF8SKIP(locinput));
8156 could_match = UCHARAT(locinput) == ST.c1
8157 || UCHARAT(locinput) == ST.c2;
8160 if (ST.c1 == CHRTEST_VOID || could_match) {
8161 CURLY_SETPAREN(ST.paren, ST.count);
8162 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
8163 NOT_REACHED; /* NOTREACHED */
8168 case CURLY_B_max_fail:
8169 /* failed to find B in a greedy match */
8171 REGCP_UNWIND(ST.cp);
8173 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8176 if (--ST.count < ST.min)
8178 locinput = HOPc(locinput, -1);
8179 goto curly_try_B_max;
8183 case END: /* last op of main pattern */
8186 /* we've just finished A in /(??{A})B/; now continue with B */
8187 SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput);
8188 st->u.eval.prev_rex = rex_sv; /* inner */
8190 /* Save *all* the positions. */
8191 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
8192 rex_sv = CUR_EVAL.prev_rex;
8193 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
8194 SET_reg_curpm(rex_sv);
8195 rex = ReANY(rex_sv);
8196 rexi = RXi_GET(rex);
8198 st->u.eval.prev_curlyx = cur_curlyx;
8199 cur_curlyx = CUR_EVAL.prev_curlyx;
8201 REGCP_SET(st->u.eval.lastcp);
8203 /* Restore parens of the outer rex without popping the
8205 regcp_restore(rex, CUR_EVAL.lastcp, &maxopenparen);
8207 st->u.eval.prev_eval = cur_eval;
8208 cur_eval = CUR_EVAL.prev_eval;
8210 Perl_re_exec_indentf( aTHX_ "EVAL trying tail ... (cur_eval=%p)\n",
8212 if ( nochange_depth )
8215 SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput);
8217 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
8218 locinput); /* match B */
8221 if (locinput < reginfo->till) {
8222 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
8223 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
8225 (long)(locinput - startpos),
8226 (long)(reginfo->till - startpos),
8229 sayNO_SILENT; /* Cannot match: too short. */
8231 sayYES; /* Success! */
8233 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
8235 Perl_re_exec_indentf( aTHX_ "%ssubpattern success...%s\n",
8236 depth, PL_colors[4], PL_colors[5]));
8237 sayYES; /* Success! */
8240 #define ST st->u.ifmatch
8245 case SUSPEND: /* (?>A) */
8247 newstart = locinput;
8250 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
8252 goto ifmatch_trivial_fail_test;
8254 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
8256 ifmatch_trivial_fail_test:
8258 char * const s = HOPBACKc(locinput, scan->flags);
8263 sw = 1 - cBOOL(ST.wanted);
8267 next = scan + ARG(scan);
8275 newstart = locinput;
8279 ST.logical = logical;
8280 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
8282 /* execute body of (?...A) */
8283 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
8284 NOT_REACHED; /* NOTREACHED */
8287 case IFMATCH_A_fail: /* body of (?...A) failed */
8288 ST.wanted = !ST.wanted;
8291 case IFMATCH_A: /* body of (?...A) succeeded */
8293 sw = cBOOL(ST.wanted);
8295 else if (!ST.wanted)
8298 if (OP(ST.me) != SUSPEND) {
8299 /* restore old position except for (?>...) */
8300 locinput = st->locinput;
8302 scan = ST.me + ARG(ST.me);
8305 continue; /* execute B */
8309 case LONGJMP: /* alternative with many branches compiles to
8310 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
8311 next = scan + ARG(scan);
8316 case COMMIT: /* (*COMMIT) */
8317 reginfo->cutpoint = reginfo->strend;
8320 case PRUNE: /* (*PRUNE) */
8322 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8323 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
8324 NOT_REACHED; /* NOTREACHED */
8326 case COMMIT_next_fail:
8330 NOT_REACHED; /* NOTREACHED */
8332 case OPFAIL: /* (*FAIL) */
8334 sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8336 /* deal with (?(?!)X|Y) properly,
8337 * make sure we trigger the no branch
8338 * of the trailing IFTHEN structure*/
8344 NOT_REACHED; /* NOTREACHED */
8346 #define ST st->u.mark
8347 case MARKPOINT: /* (*MARK:foo) */
8348 ST.prev_mark = mark_state;
8349 ST.mark_name = sv_commit = sv_yes_mark
8350 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8352 ST.mark_loc = locinput;
8353 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
8354 NOT_REACHED; /* NOTREACHED */
8356 case MARKPOINT_next:
8357 mark_state = ST.prev_mark;
8359 NOT_REACHED; /* NOTREACHED */
8361 case MARKPOINT_next_fail:
8362 if (popmark && sv_eq(ST.mark_name,popmark))
8364 if (ST.mark_loc > startpoint)
8365 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8366 popmark = NULL; /* we found our mark */
8367 sv_commit = ST.mark_name;
8370 Perl_re_exec_indentf( aTHX_ "%ssetting cutpoint to mark:%" SVf "...%s\n",
8372 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
8375 mark_state = ST.prev_mark;
8376 sv_yes_mark = mark_state ?
8377 mark_state->u.mark.mark_name : NULL;
8379 NOT_REACHED; /* NOTREACHED */
8381 case SKIP: /* (*SKIP) */
8383 /* (*SKIP) : if we fail we cut here*/
8384 ST.mark_name = NULL;
8385 ST.mark_loc = locinput;
8386 PUSH_STATE_GOTO(SKIP_next,next, locinput);
8388 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
8389 otherwise do nothing. Meaning we need to scan
8391 regmatch_state *cur = mark_state;
8392 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8395 if ( sv_eq( cur->u.mark.mark_name,
8398 ST.mark_name = find;
8399 PUSH_STATE_GOTO( SKIP_next, next, locinput);
8401 cur = cur->u.mark.prev_mark;
8404 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
8407 case SKIP_next_fail:
8409 /* (*CUT:NAME) - Set up to search for the name as we
8410 collapse the stack*/
8411 popmark = ST.mark_name;
8413 /* (*CUT) - No name, we cut here.*/
8414 if (ST.mark_loc > startpoint)
8415 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8416 /* but we set sv_commit to latest mark_name if there
8417 is one so they can test to see how things lead to this
8420 sv_commit=mark_state->u.mark.mark_name;
8424 NOT_REACHED; /* NOTREACHED */
8427 case LNBREAK: /* \R */
8428 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
8435 PerlIO_printf(Perl_error_log, "%" UVxf " %d\n",
8436 PTR2UV(scan), OP(scan));
8437 Perl_croak(aTHX_ "regexp memory corruption");
8439 /* this is a point to jump to in order to increment
8440 * locinput by one character */
8442 assert(!NEXTCHR_IS_EOS);
8444 locinput += PL_utf8skip[nextchr];
8445 /* locinput is allowed to go 1 char off the end, but not 2+ */
8446 if (locinput > reginfo->strend)
8455 /* switch break jumps here */
8456 scan = next; /* prepare to execute the next op and ... */
8457 continue; /* ... jump back to the top, reusing st */
8461 /* push a state that backtracks on success */
8462 st->u.yes.prev_yes_state = yes_state;
8466 /* push a new regex state, then continue at scan */
8468 regmatch_state *newst;
8471 regmatch_state *cur = st;
8472 regmatch_state *curyes = yes_state;
8474 regmatch_slab *slab = PL_regmatch_slab;
8475 for (;curd > -1 && (depth-curd < 3);cur--,curd--) {
8476 if (cur < SLAB_FIRST(slab)) {
8478 cur = SLAB_LAST(slab);
8480 Perl_re_exec_indentf( aTHX_ "#%-3d %-10s %s\n",
8482 curd, PL_reg_name[cur->resume_state],
8483 (curyes == cur) ? "yes" : ""
8486 curyes = cur->u.yes.prev_yes_state;
8489 DEBUG_STATE_pp("push")
8492 st->locinput = locinput;
8494 if (newst > SLAB_LAST(PL_regmatch_slab))
8495 newst = S_push_slab(aTHX);
8496 PL_regmatch_state = newst;
8498 locinput = pushinput;
8504 #ifdef SOLARIS_BAD_OPTIMIZER
8505 # undef PL_charclass
8509 * We get here only if there's trouble -- normally "case END" is
8510 * the terminating point.
8512 Perl_croak(aTHX_ "corrupted regexp pointers");
8513 NOT_REACHED; /* NOTREACHED */
8517 /* we have successfully completed a subexpression, but we must now
8518 * pop to the state marked by yes_state and continue from there */
8519 assert(st != yes_state);
8521 while (st != yes_state) {
8523 if (st < SLAB_FIRST(PL_regmatch_slab)) {
8524 PL_regmatch_slab = PL_regmatch_slab->prev;
8525 st = SLAB_LAST(PL_regmatch_slab);
8529 DEBUG_STATE_pp("pop (no final)");
8531 DEBUG_STATE_pp("pop (yes)");
8537 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
8538 || yes_state > SLAB_LAST(PL_regmatch_slab))
8540 /* not in this slab, pop slab */
8541 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
8542 PL_regmatch_slab = PL_regmatch_slab->prev;
8543 st = SLAB_LAST(PL_regmatch_slab);
8545 depth -= (st - yes_state);
8548 yes_state = st->u.yes.prev_yes_state;
8549 PL_regmatch_state = st;
8552 locinput= st->locinput;
8553 state_num = st->resume_state + no_final;
8554 goto reenter_switch;
8557 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n",
8558 PL_colors[4], PL_colors[5]));
8560 if (reginfo->info_aux_eval) {
8561 /* each successfully executed (?{...}) block does the equivalent of
8562 * local $^R = do {...}
8563 * When popping the save stack, all these locals would be undone;
8564 * bypass this by setting the outermost saved $^R to the latest
8566 /* I dont know if this is needed or works properly now.
8567 * see code related to PL_replgv elsewhere in this file.
8570 if (oreplsv != GvSV(PL_replgv))
8571 sv_setsv(oreplsv, GvSV(PL_replgv));
8578 Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n",
8580 PL_colors[4], PL_colors[5])
8592 /* there's a previous state to backtrack to */
8594 if (st < SLAB_FIRST(PL_regmatch_slab)) {
8595 PL_regmatch_slab = PL_regmatch_slab->prev;
8596 st = SLAB_LAST(PL_regmatch_slab);
8598 PL_regmatch_state = st;
8599 locinput= st->locinput;
8601 DEBUG_STATE_pp("pop");
8603 if (yes_state == st)
8604 yes_state = st->u.yes.prev_yes_state;
8606 state_num = st->resume_state + 1; /* failure = success + 1 */
8608 goto reenter_switch;
8613 if (rex->intflags & PREGf_VERBARG_SEEN) {
8614 SV *sv_err = get_sv("REGERROR", 1);
8615 SV *sv_mrk = get_sv("REGMARK", 1);
8617 sv_commit = &PL_sv_no;
8619 sv_yes_mark = &PL_sv_yes;
8622 sv_commit = &PL_sv_yes;
8623 sv_yes_mark = &PL_sv_no;
8627 sv_setsv(sv_err, sv_commit);
8628 sv_setsv(sv_mrk, sv_yes_mark);
8632 if (last_pushed_cv) {
8635 PERL_UNUSED_VAR(SP);
8638 assert(!result || locinput - reginfo->strbeg >= 0);
8639 return result ? locinput - reginfo->strbeg : -1;
8643 - regrepeat - repeatedly match something simple, report how many
8645 * What 'simple' means is a node which can be the operand of a quantifier like
8648 * startposp - pointer a pointer to the start position. This is updated
8649 * to point to the byte following the highest successful
8651 * p - the regnode to be repeatedly matched against.
8652 * reginfo - struct holding match state, such as strend
8653 * max - maximum number of things to match.
8654 * depth - (for debugging) backtracking depth.
8657 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
8658 regmatch_info *const reginfo, I32 max _pDEPTH)
8660 char *scan; /* Pointer to current position in target string */
8662 char *loceol = reginfo->strend; /* local version */
8663 I32 hardcount = 0; /* How many matches so far */
8664 bool utf8_target = reginfo->is_utf8_target;
8665 unsigned int to_complement = 0; /* Invert the result? */
8667 _char_class_number classnum;
8669 PERL_ARGS_ASSERT_REGREPEAT;
8672 if (max == REG_INFTY)
8674 else if (! utf8_target && loceol - scan > max)
8675 loceol = scan + max;
8677 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
8678 * to the maximum of how far we should go in it (leaving it set to the real
8679 * end, if the maximum permissible would take us beyond that). This allows
8680 * us to make the loop exit condition that we haven't gone past <loceol> to
8681 * also mean that we haven't exceeded the max permissible count, saving a
8682 * test each time through the loop. But it assumes that the OP matches a
8683 * single byte, which is true for most of the OPs below when applied to a
8684 * non-UTF-8 target. Those relatively few OPs that don't have this
8685 * characteristic will have to compensate.
8687 * There is no adjustment for UTF-8 targets, as the number of bytes per
8688 * character varies. OPs will have to test both that the count is less
8689 * than the max permissible (using <hardcount> to keep track), and that we
8690 * are still within the bounds of the string (using <loceol>. A few OPs
8691 * match a single byte no matter what the encoding. They can omit the max
8692 * test if, for the UTF-8 case, they do the adjustment that was skipped
8695 * Thus, the code above sets things up for the common case; and exceptional
8696 * cases need extra work; the common case is to make sure <scan> doesn't
8697 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
8698 * count doesn't exceed the maximum permissible */
8703 while (scan < loceol && hardcount < max && *scan != '\n') {
8704 scan += UTF8SKIP(scan);
8708 while (scan < loceol && *scan != '\n')
8714 while (scan < loceol && hardcount < max) {
8715 scan += UTF8SKIP(scan);
8723 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8724 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
8725 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
8729 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8733 /* Can use a simple loop if the pattern char to match on is invariant
8734 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
8735 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
8736 * true iff it doesn't matter if the argument is in UTF-8 or not */
8737 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
8738 if (utf8_target && loceol - scan > max) {
8739 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
8740 * since here, to match at all, 1 char == 1 byte */
8741 loceol = scan + max;
8743 while (scan < loceol && UCHARAT(scan) == c) {
8747 else if (reginfo->is_utf8_pat) {
8749 STRLEN scan_char_len;
8751 /* When both target and pattern are UTF-8, we have to do
8753 while (hardcount < max
8755 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
8756 && memEQ(scan, STRING(p), scan_char_len))
8758 scan += scan_char_len;
8762 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
8764 /* Target isn't utf8; convert the character in the UTF-8
8765 * pattern to non-UTF8, and do a simple loop */
8766 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
8767 while (scan < loceol && UCHARAT(scan) == c) {
8770 } /* else pattern char is above Latin1, can't possibly match the
8775 /* Here, the string must be utf8; pattern isn't, and <c> is
8776 * different in utf8 than not, so can't compare them directly.
8777 * Outside the loop, find the two utf8 bytes that represent c, and
8778 * then look for those in sequence in the utf8 string */
8779 U8 high = UTF8_TWO_BYTE_HI(c);
8780 U8 low = UTF8_TWO_BYTE_LO(c);
8782 while (hardcount < max
8783 && scan + 1 < loceol
8784 && UCHARAT(scan) == high
8785 && UCHARAT(scan + 1) == low)
8793 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
8794 assert(! reginfo->is_utf8_pat);
8797 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
8801 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8802 utf8_flags = FOLDEQ_LOCALE;
8805 case EXACTF: /* This node only generated for non-utf8 patterns */
8806 assert(! reginfo->is_utf8_pat);
8811 if (! utf8_target) {
8814 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
8815 | FOLDEQ_S2_FOLDS_SANE;
8820 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
8824 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
8826 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8828 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
8831 if (c1 == CHRTEST_VOID) {
8832 /* Use full Unicode fold matching */
8833 char *tmpeol = reginfo->strend;
8834 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
8835 while (hardcount < max
8836 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
8837 STRING(p), NULL, pat_len,
8838 reginfo->is_utf8_pat, utf8_flags))
8841 tmpeol = reginfo->strend;
8845 else if (utf8_target) {
8847 while (scan < loceol
8849 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
8851 scan += UTF8SKIP(scan);
8856 while (scan < loceol
8858 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
8859 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
8861 scan += UTF8SKIP(scan);
8866 else if (c1 == c2) {
8867 while (scan < loceol && UCHARAT(scan) == c1) {
8872 while (scan < loceol &&
8873 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
8882 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8884 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) {
8885 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
8891 while (hardcount < max
8893 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
8895 scan += UTF8SKIP(scan);
8899 else if (ANYOF_FLAGS(p)) {
8900 while (scan < loceol
8901 && reginclass(prog, p, (U8*)scan, (U8*)scan+1, 0))
8905 while (scan < loceol && ANYOF_BITMAP_TEST(p, *((U8*)scan)))
8910 /* The argument (FLAGS) to all the POSIX node types is the class number */
8917 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8918 if (! utf8_target) {
8919 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
8925 while (hardcount < max && scan < loceol
8926 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
8929 scan += UTF8SKIP(scan);
8942 if (utf8_target && loceol - scan > max) {
8944 /* We didn't adjust <loceol> at the beginning of this routine
8945 * because is UTF-8, but it is actually ok to do so, since here, to
8946 * match, 1 char == 1 byte. */
8947 loceol = scan + max;
8949 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
8962 if (! utf8_target) {
8963 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
8969 /* The complement of something that matches only ASCII matches all
8970 * non-ASCII, plus everything in ASCII that isn't in the class. */
8971 while (hardcount < max && scan < loceol
8972 && (! isASCII_utf8(scan)
8973 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
8975 scan += UTF8SKIP(scan);
8986 if (! utf8_target) {
8987 while (scan < loceol && to_complement
8988 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
8995 classnum = (_char_class_number) FLAGS(p);
8996 if (classnum < _FIRST_NON_SWASH_CC) {
8998 /* Here, a swash is needed for above-Latin1 code points.
8999 * Process as many Latin1 code points using the built-in rules.
9000 * Go to another loop to finish processing upon encountering
9001 * the first Latin1 code point. We could do that in this loop
9002 * as well, but the other way saves having to test if the swash
9003 * has been loaded every time through the loop: extra space to
9005 while (hardcount < max && scan < loceol) {
9006 if (UTF8_IS_INVARIANT(*scan)) {
9007 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
9014 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
9015 if (! (to_complement
9016 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*scan,
9025 goto found_above_latin1;
9032 /* For these character classes, the knowledge of how to handle
9033 * every code point is compiled in to Perl via a macro. This
9034 * code is written for making the loops as tight as possible.
9035 * It could be refactored to save space instead */
9037 case _CC_ENUM_SPACE:
9038 while (hardcount < max
9040 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
9042 scan += UTF8SKIP(scan);
9046 case _CC_ENUM_BLANK:
9047 while (hardcount < max
9049 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
9051 scan += UTF8SKIP(scan);
9055 case _CC_ENUM_XDIGIT:
9056 while (hardcount < max
9058 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
9060 scan += UTF8SKIP(scan);
9064 case _CC_ENUM_VERTSPACE:
9065 while (hardcount < max
9067 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
9069 scan += UTF8SKIP(scan);
9073 case _CC_ENUM_CNTRL:
9074 while (hardcount < max
9076 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
9078 scan += UTF8SKIP(scan);
9083 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
9089 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
9091 /* Load the swash if not already present */
9092 if (! PL_utf8_swash_ptrs[classnum]) {
9093 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
9094 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
9098 PL_XPosix_ptrs[classnum], &flags);
9101 while (hardcount < max && scan < loceol
9102 && to_complement ^ cBOOL(_generic_utf8(
9105 swash_fetch(PL_utf8_swash_ptrs[classnum],
9109 scan += UTF8SKIP(scan);
9116 while (hardcount < max && scan < loceol &&
9117 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
9122 /* LNBREAK can match one or two latin chars, which is ok, but we
9123 * have to use hardcount in this situation, and throw away the
9124 * adjustment to <loceol> done before the switch statement */
9125 loceol = reginfo->strend;
9126 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
9135 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9149 /* These are all 0 width, so match right here or not at all. */
9153 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
9154 NOT_REACHED; /* NOTREACHED */
9161 c = scan - *startposp;
9165 GET_RE_DEBUG_FLAGS_DECL;
9167 SV * const prop = sv_newmortal();
9168 regprop(prog, prop, p, reginfo, NULL);
9169 Perl_re_exec_indentf( aTHX_ "%s can match %" IVdf " times out of %" IVdf "...\n",
9170 depth, SvPVX_const(prop),(IV)c,(IV)max);
9178 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
9180 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
9181 create a copy so that changes the caller makes won't change the shared one.
9182 If <altsvp> is non-null, will return NULL in it, for back-compat.
9185 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
9187 PERL_ARGS_ASSERT_REGCLASS_SWASH;
9193 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL));
9196 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
9199 - reginclass - determine if a character falls into a character class
9201 n is the ANYOF-type regnode
9202 p is the target string
9203 p_end points to one byte beyond the end of the target string
9204 utf8_target tells whether p is in UTF-8.
9206 Returns true if matched; false otherwise.
9208 Note that this can be a synthetic start class, a combination of various
9209 nodes, so things you think might be mutually exclusive, such as locale,
9210 aren't. It can match both locale and non-locale
9215 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
9218 const char flags = ANYOF_FLAGS(n);
9222 PERL_ARGS_ASSERT_REGINCLASS;
9224 /* If c is not already the code point, get it. Note that
9225 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
9226 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
9228 c = utf8n_to_uvchr(p, p_end - p, &c_len, ( UTF8_ALLOW_DEFAULT
9229 | UTF8_CHECK_ONLY));
9230 if (c_len == (STRLEN)-1)
9231 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
9232 if (c > 255 && OP(n) == ANYOFL && ! ANYOFL_UTF8_LOCALE_REQD(flags)) {
9233 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
9237 /* If this character is potentially in the bitmap, check it */
9238 if (c < NUM_ANYOF_CODE_POINTS) {
9239 if (ANYOF_BITMAP_TEST(n, c))
9242 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9249 else if (flags & ANYOF_LOCALE_FLAGS) {
9250 if ((flags & ANYOFL_FOLD)
9252 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
9256 else if (ANYOF_POSIXL_TEST_ANY_SET(n)
9260 /* The data structure is arranged so bits 0, 2, 4, ... are set
9261 * if the class includes the Posix character class given by
9262 * bit/2; and 1, 3, 5, ... are set if the class includes the
9263 * complemented Posix class given by int(bit/2). So we loop
9264 * through the bits, each time changing whether we complement
9265 * the result or not. Suppose for the sake of illustration
9266 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
9267 * is set, it means there is a match for this ANYOF node if the
9268 * character is in the class given by the expression (0 / 2 = 0
9269 * = \w). If it is in that class, isFOO_lc() will return 1,
9270 * and since 'to_complement' is 0, the result will stay TRUE,
9271 * and we exit the loop. Suppose instead that bit 0 is 0, but
9272 * bit 1 is 1. That means there is a match if the character
9273 * matches \W. We won't bother to call isFOO_lc() on bit 0,
9274 * but will on bit 1. On the second iteration 'to_complement'
9275 * will be 1, so the exclusive or will reverse things, so we
9276 * are testing for \W. On the third iteration, 'to_complement'
9277 * will be 0, and we would be testing for \s; the fourth
9278 * iteration would test for \S, etc.
9280 * Note that this code assumes that all the classes are closed
9281 * under folding. For example, if a character matches \w, then
9282 * its fold does too; and vice versa. This should be true for
9283 * any well-behaved locale for all the currently defined Posix
9284 * classes, except for :lower: and :upper:, which are handled
9285 * by the pseudo-class :cased: which matches if either of the
9286 * other two does. To get rid of this assumption, an outer
9287 * loop could be used below to iterate over both the source
9288 * character, and its fold (if different) */
9291 int to_complement = 0;
9293 while (count < ANYOF_MAX) {
9294 if (ANYOF_POSIXL_TEST(n, count)
9295 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
9308 /* If the bitmap didn't (or couldn't) match, and something outside the
9309 * bitmap could match, try that. */
9311 if (c >= NUM_ANYOF_CODE_POINTS
9312 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
9314 match = TRUE; /* Everything above the bitmap matches */
9316 /* Here doesn't match everything above the bitmap. If there is
9317 * some information available beyond the bitmap, we may find a
9318 * match in it. If so, this is most likely because the code point
9319 * is outside the bitmap range. But rarely, it could be because of
9320 * some other reason. If so, various flags are set to indicate
9321 * this possibility. On ANYOFD nodes, there may be matches that
9322 * happen only when the target string is UTF-8; or for other node
9323 * types, because runtime lookup is needed, regardless of the
9324 * UTF-8ness of the target string. Finally, under /il, there may
9325 * be some matches only possible if the locale is a UTF-8 one. */
9326 else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP
9327 && ( c >= NUM_ANYOF_CODE_POINTS
9328 || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)
9329 && ( UNLIKELY(OP(n) != ANYOFD)
9330 || (utf8_target && ! isASCII_uni(c)
9331 # if NUM_ANYOF_CODE_POINTS > 256
9335 || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags)
9336 && IN_UTF8_CTYPE_LOCALE)))
9338 SV* only_utf8_locale = NULL;
9339 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
9340 &only_utf8_locale, NULL);
9346 } else { /* Convert to utf8 */
9347 utf8_p = utf8_buffer;
9348 append_utf8_from_native_byte(*p, &utf8_p);
9349 utf8_p = utf8_buffer;
9352 if (swash_fetch(sw, utf8_p, TRUE)) {
9356 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
9357 match = _invlist_contains_cp(only_utf8_locale, c);
9361 if (UNICODE_IS_SUPER(c)
9363 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9365 && ckWARN_d(WARN_NON_UNICODE))
9367 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
9368 "Matched non-Unicode code point 0x%04" UVXf " against Unicode property; may not be portable", c);
9372 #if ANYOF_INVERT != 1
9373 /* Depending on compiler optimization cBOOL takes time, so if don't have to
9375 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
9378 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
9379 return (flags & ANYOF_INVERT) ^ match;
9383 S_reghop3(U8 *s, SSize_t off, const U8* lim)
9385 /* return the position 'off' UTF-8 characters away from 's', forward if
9386 * 'off' >= 0, backwards if negative. But don't go outside of position
9387 * 'lim', which better be < s if off < 0 */
9389 PERL_ARGS_ASSERT_REGHOP3;
9392 while (off-- && s < lim) {
9393 /* XXX could check well-formedness here */
9398 while (off++ && s > lim) {
9400 if (UTF8_IS_CONTINUED(*s)) {
9401 while (s > lim && UTF8_IS_CONTINUATION(*s))
9403 if (! UTF8_IS_START(*s)) {
9404 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9407 /* XXX could check well-formedness here */
9414 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
9416 PERL_ARGS_ASSERT_REGHOP4;
9419 while (off-- && s < rlim) {
9420 /* XXX could check well-formedness here */
9425 while (off++ && s > llim) {
9427 if (UTF8_IS_CONTINUED(*s)) {
9428 while (s > llim && UTF8_IS_CONTINUATION(*s))
9430 if (! UTF8_IS_START(*s)) {
9431 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9434 /* XXX could check well-formedness here */
9440 /* like reghop3, but returns NULL on overrun, rather than returning last
9444 S_reghopmaybe3(U8* s, SSize_t off, const U8* const lim)
9446 PERL_ARGS_ASSERT_REGHOPMAYBE3;
9449 while (off-- && s < lim) {
9450 /* XXX could check well-formedness here */
9457 while (off++ && s > lim) {
9459 if (UTF8_IS_CONTINUED(*s)) {
9460 while (s > lim && UTF8_IS_CONTINUATION(*s))
9462 if (! UTF8_IS_START(*s)) {
9463 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9466 /* XXX could check well-formedness here */
9475 /* when executing a regex that may have (?{}), extra stuff needs setting
9476 up that will be visible to the called code, even before the current
9477 match has finished. In particular:
9479 * $_ is localised to the SV currently being matched;
9480 * pos($_) is created if necessary, ready to be updated on each call-out
9482 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
9483 isn't set until the current pattern is successfully finished), so that
9484 $1 etc of the match-so-far can be seen;
9485 * save the old values of subbeg etc of the current regex, and set then
9486 to the current string (again, this is normally only done at the end
9491 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
9494 regexp *const rex = ReANY(reginfo->prog);
9495 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
9497 eval_state->rex = rex;
9500 /* Make $_ available to executed code. */
9501 if (reginfo->sv != DEFSV) {
9503 DEFSV_set(reginfo->sv);
9506 if (!(mg = mg_find_mglob(reginfo->sv))) {
9507 /* prepare for quick setting of pos */
9508 mg = sv_magicext_mglob(reginfo->sv);
9511 eval_state->pos_magic = mg;
9512 eval_state->pos = mg->mg_len;
9513 eval_state->pos_flags = mg->mg_flags;
9516 eval_state->pos_magic = NULL;
9518 if (!PL_reg_curpm) {
9519 /* PL_reg_curpm is a fake PMOP that we can attach the current
9520 * regex to and point PL_curpm at, so that $1 et al are visible
9521 * within a /(?{})/. It's just allocated once per interpreter the
9522 * first time its needed */
9523 Newxz(PL_reg_curpm, 1, PMOP);
9526 SV* const repointer = &PL_sv_undef;
9527 /* this regexp is also owned by the new PL_reg_curpm, which
9528 will try to free it. */
9529 av_push(PL_regex_padav, repointer);
9530 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
9531 PL_regex_pad = AvARRAY(PL_regex_padav);
9535 SET_reg_curpm(reginfo->prog);
9536 eval_state->curpm = PL_curpm;
9537 PL_curpm_under = PL_curpm;
9538 PL_curpm = PL_reg_curpm;
9539 if (RXp_MATCH_COPIED(rex)) {
9540 /* Here is a serious problem: we cannot rewrite subbeg,
9541 since it may be needed if this match fails. Thus
9542 $` inside (?{}) could fail... */
9543 eval_state->subbeg = rex->subbeg;
9544 eval_state->sublen = rex->sublen;
9545 eval_state->suboffset = rex->suboffset;
9546 eval_state->subcoffset = rex->subcoffset;
9548 eval_state->saved_copy = rex->saved_copy;
9550 RXp_MATCH_COPIED_off(rex);
9553 eval_state->subbeg = NULL;
9554 rex->subbeg = (char *)reginfo->strbeg;
9556 rex->subcoffset = 0;
9557 rex->sublen = reginfo->strend - reginfo->strbeg;
9561 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
9564 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
9566 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
9567 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
9570 Safefree(aux->poscache);
9574 /* undo the effects of S_setup_eval_state() */
9576 if (eval_state->subbeg) {
9577 regexp * const rex = eval_state->rex;
9578 rex->subbeg = eval_state->subbeg;
9579 rex->sublen = eval_state->sublen;
9580 rex->suboffset = eval_state->suboffset;
9581 rex->subcoffset = eval_state->subcoffset;
9583 rex->saved_copy = eval_state->saved_copy;
9585 RXp_MATCH_COPIED_on(rex);
9587 if (eval_state->pos_magic)
9589 eval_state->pos_magic->mg_len = eval_state->pos;
9590 eval_state->pos_magic->mg_flags =
9591 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
9592 | (eval_state->pos_flags & MGf_BYTES);
9595 PL_curpm = eval_state->curpm;
9598 PL_regmatch_state = aux->old_regmatch_state;
9599 PL_regmatch_slab = aux->old_regmatch_slab;
9601 /* free all slabs above current one - this must be the last action
9602 * of this function, as aux and eval_state are allocated within
9603 * slabs and may be freed here */
9605 s = PL_regmatch_slab->next;
9607 PL_regmatch_slab->next = NULL;
9609 regmatch_slab * const osl = s;
9618 S_to_utf8_substr(pTHX_ regexp *prog)
9620 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
9621 * on the converted value */
9625 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
9628 if (prog->substrs->data[i].substr
9629 && !prog->substrs->data[i].utf8_substr) {
9630 SV* const sv = newSVsv(prog->substrs->data[i].substr);
9631 prog->substrs->data[i].utf8_substr = sv;
9632 sv_utf8_upgrade(sv);
9633 if (SvVALID(prog->substrs->data[i].substr)) {
9634 if (SvTAIL(prog->substrs->data[i].substr)) {
9635 /* Trim the trailing \n that fbm_compile added last
9637 SvCUR_set(sv, SvCUR(sv) - 1);
9638 /* Whilst this makes the SV technically "invalid" (as its
9639 buffer is no longer followed by "\0") when fbm_compile()
9640 adds the "\n" back, a "\0" is restored. */
9641 fbm_compile(sv, FBMcf_TAIL);
9645 if (prog->substrs->data[i].substr == prog->check_substr)
9646 prog->check_utf8 = sv;
9652 S_to_byte_substr(pTHX_ regexp *prog)
9654 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
9655 * on the converted value; returns FALSE if can't be converted. */
9659 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
9662 if (prog->substrs->data[i].utf8_substr
9663 && !prog->substrs->data[i].substr) {
9664 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
9665 if (! sv_utf8_downgrade(sv, TRUE)) {
9668 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
9669 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
9670 /* Trim the trailing \n that fbm_compile added last
9672 SvCUR_set(sv, SvCUR(sv) - 1);
9673 fbm_compile(sv, FBMcf_TAIL);
9677 prog->substrs->data[i].substr = sv;
9678 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
9679 prog->check_substr = sv;
9687 * ex: set ts=8 sts=4 sw=4 et: