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, (U8 *) reginfo->strend))) { \
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_safe);
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_safe);
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_safe);
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_safe);
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_safe);
2110 switch((bound_type) FLAGS(c)) {
2111 case TRADITIONAL_BOUND:
2112 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
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_safe(s, strend)
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_safe(s, strend)));
2457 case _CC_ENUM_BLANK:
2458 REXEC_FBC_UTF8_CLASS_SCAN(
2459 to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
2462 case _CC_ENUM_XDIGIT:
2463 REXEC_FBC_UTF8_CLASS_SCAN(
2464 to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
2467 case _CC_ENUM_VERTSPACE:
2468 REXEC_FBC_UTF8_CLASS_SCAN(
2469 to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
2472 case _CC_ENUM_CNTRL:
2473 REXEC_FBC_UTF8_CLASS_SCAN(
2474 to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
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_safe(
2503 swash_fetch(PL_utf8_swash_ptrs[classnum],
2511 /* what trie are we using right now */
2512 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2513 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2514 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2516 const char *last_start = strend - trie->minlen;
2518 const char *real_start = s;
2520 STRLEN maxlen = trie->maxlen;
2522 U8 **points; /* map of where we were in the input string
2523 when reading a given char. For ASCII this
2524 is unnecessary overhead as the relationship
2525 is always 1:1, but for Unicode, especially
2526 case folded Unicode this is not true. */
2527 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2531 GET_RE_DEBUG_FLAGS_DECL;
2533 /* We can't just allocate points here. We need to wrap it in
2534 * an SV so it gets freed properly if there is a croak while
2535 * running the match */
2538 sv_points=newSV(maxlen * sizeof(U8 *));
2539 SvCUR_set(sv_points,
2540 maxlen * sizeof(U8 *));
2541 SvPOK_on(sv_points);
2542 sv_2mortal(sv_points);
2543 points=(U8**)SvPV_nolen(sv_points );
2544 if ( trie_type != trie_utf8_fold
2545 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2548 bitmap=(U8*)trie->bitmap;
2550 bitmap=(U8*)ANYOF_BITMAP(c);
2552 /* this is the Aho-Corasick algorithm modified a touch
2553 to include special handling for long "unknown char" sequences.
2554 The basic idea being that we use AC as long as we are dealing
2555 with a possible matching char, when we encounter an unknown char
2556 (and we have not encountered an accepting state) we scan forward
2557 until we find a legal starting char.
2558 AC matching is basically that of trie matching, except that when
2559 we encounter a failing transition, we fall back to the current
2560 states "fail state", and try the current char again, a process
2561 we repeat until we reach the root state, state 1, or a legal
2562 transition. If we fail on the root state then we can either
2563 terminate if we have reached an accepting state previously, or
2564 restart the entire process from the beginning if we have not.
2567 while (s <= last_start) {
2568 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2576 U8 *uscan = (U8*)NULL;
2577 U8 *leftmost = NULL;
2579 U32 accepted_word= 0;
2583 while ( state && uc <= (U8*)strend ) {
2585 U32 word = aho->states[ state ].wordnum;
2589 DEBUG_TRIE_EXECUTE_r(
2590 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2591 dump_exec_pos( (char *)uc, c, strend, real_start,
2592 (char *)uc, utf8_target, 0 );
2593 Perl_re_printf( aTHX_
2594 " Scanning for legal start char...\n");
2598 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2602 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2608 if (uc >(U8*)last_start) break;
2612 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2613 if (!leftmost || lpos < leftmost) {
2614 DEBUG_r(accepted_word=word);
2620 points[pointpos++ % maxlen]= uc;
2621 if (foldlen || uc < (U8*)strend) {
2622 REXEC_TRIE_READ_CHAR(trie_type, trie,
2624 uscan, len, uvc, charid, foldlen,
2626 DEBUG_TRIE_EXECUTE_r({
2627 dump_exec_pos( (char *)uc, c, strend,
2628 real_start, s, utf8_target, 0);
2629 Perl_re_printf( aTHX_
2630 " Charid:%3u CP:%4" UVxf " ",
2642 word = aho->states[ state ].wordnum;
2644 base = aho->states[ state ].trans.base;
2646 DEBUG_TRIE_EXECUTE_r({
2648 dump_exec_pos( (char *)uc, c, strend, real_start,
2649 s, utf8_target, 0 );
2650 Perl_re_printf( aTHX_
2651 "%sState: %4" UVxf ", word=%" UVxf,
2652 failed ? " Fail transition to " : "",
2653 (UV)state, (UV)word);
2659 ( ((offset = base + charid
2660 - 1 - trie->uniquecharcount)) >= 0)
2661 && ((U32)offset < trie->lasttrans)
2662 && trie->trans[offset].check == state
2663 && (tmp=trie->trans[offset].next))
2665 DEBUG_TRIE_EXECUTE_r(
2666 Perl_re_printf( aTHX_ " - legal\n"));
2671 DEBUG_TRIE_EXECUTE_r(
2672 Perl_re_printf( aTHX_ " - fail\n"));
2674 state = aho->fail[state];
2678 /* we must be accepting here */
2679 DEBUG_TRIE_EXECUTE_r(
2680 Perl_re_printf( aTHX_ " - accepting\n"));
2689 if (!state) state = 1;
2692 if ( aho->states[ state ].wordnum ) {
2693 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2694 if (!leftmost || lpos < leftmost) {
2695 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2700 s = (char*)leftmost;
2701 DEBUG_TRIE_EXECUTE_r({
2702 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
2703 (UV)accepted_word, (IV)(s - real_start)
2706 if (reginfo->intuit || regtry(reginfo, &s)) {
2712 DEBUG_TRIE_EXECUTE_r({
2713 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
2716 DEBUG_TRIE_EXECUTE_r(
2717 Perl_re_printf( aTHX_ "No match.\n"));
2726 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2733 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2734 * flags have same meanings as with regexec_flags() */
2737 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2744 struct regexp *const prog = ReANY(rx);
2746 if (flags & REXEC_COPY_STR) {
2749 DEBUG_C(Perl_re_printf( aTHX_
2750 "Copy on write: regexp capture, type %d\n",
2752 /* Create a new COW SV to share the match string and store
2753 * in saved_copy, unless the current COW SV in saved_copy
2754 * is valid and suitable for our purpose */
2755 if (( prog->saved_copy
2756 && SvIsCOW(prog->saved_copy)
2757 && SvPOKp(prog->saved_copy)
2760 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2762 /* just reuse saved_copy SV */
2763 if (RXp_MATCH_COPIED(prog)) {
2764 Safefree(prog->subbeg);
2765 RXp_MATCH_COPIED_off(prog);
2769 /* create new COW SV to share string */
2770 RX_MATCH_COPY_FREE(rx);
2771 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2773 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2774 assert (SvPOKp(prog->saved_copy));
2775 prog->sublen = strend - strbeg;
2776 prog->suboffset = 0;
2777 prog->subcoffset = 0;
2782 SSize_t max = strend - strbeg;
2785 if ( (flags & REXEC_COPY_SKIP_POST)
2786 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2787 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2788 ) { /* don't copy $' part of string */
2791 /* calculate the right-most part of the string covered
2792 * by a capture. Due to lookahead, this may be to
2793 * the right of $&, so we have to scan all captures */
2794 while (n <= prog->lastparen) {
2795 if (prog->offs[n].end > max)
2796 max = prog->offs[n].end;
2800 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2801 ? prog->offs[0].start
2803 assert(max >= 0 && max <= strend - strbeg);
2806 if ( (flags & REXEC_COPY_SKIP_PRE)
2807 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2808 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2809 ) { /* don't copy $` part of string */
2812 /* calculate the left-most part of the string covered
2813 * by a capture. Due to lookbehind, this may be to
2814 * the left of $&, so we have to scan all captures */
2815 while (min && n <= prog->lastparen) {
2816 if ( prog->offs[n].start != -1
2817 && prog->offs[n].start < min)
2819 min = prog->offs[n].start;
2823 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2824 && min > prog->offs[0].end
2826 min = prog->offs[0].end;
2830 assert(min >= 0 && min <= max && min <= strend - strbeg);
2833 if (RX_MATCH_COPIED(rx)) {
2834 if (sublen > prog->sublen)
2836 (char*)saferealloc(prog->subbeg, sublen+1);
2839 prog->subbeg = (char*)safemalloc(sublen+1);
2840 Copy(strbeg + min, prog->subbeg, sublen, char);
2841 prog->subbeg[sublen] = '\0';
2842 prog->suboffset = min;
2843 prog->sublen = sublen;
2844 RX_MATCH_COPIED_on(rx);
2846 prog->subcoffset = prog->suboffset;
2847 if (prog->suboffset && utf8_target) {
2848 /* Convert byte offset to chars.
2849 * XXX ideally should only compute this if @-/@+
2850 * has been seen, a la PL_sawampersand ??? */
2852 /* If there's a direct correspondence between the
2853 * string which we're matching and the original SV,
2854 * then we can use the utf8 len cache associated with
2855 * the SV. In particular, it means that under //g,
2856 * sv_pos_b2u() will use the previously cached
2857 * position to speed up working out the new length of
2858 * subcoffset, rather than counting from the start of
2859 * the string each time. This stops
2860 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2861 * from going quadratic */
2862 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2863 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2864 SV_GMAGIC|SV_CONST_RETURN);
2866 prog->subcoffset = utf8_length((U8*)strbeg,
2867 (U8*)(strbeg+prog->suboffset));
2871 RX_MATCH_COPY_FREE(rx);
2872 prog->subbeg = strbeg;
2873 prog->suboffset = 0;
2874 prog->subcoffset = 0;
2875 prog->sublen = strend - strbeg;
2883 - regexec_flags - match a regexp against a string
2886 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2887 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2888 /* stringarg: the point in the string at which to begin matching */
2889 /* strend: pointer to null at end of string */
2890 /* strbeg: real beginning of string */
2891 /* minend: end of match must be >= minend bytes after stringarg. */
2892 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2893 * itself is accessed via the pointers above */
2894 /* data: May be used for some additional optimizations.
2895 Currently unused. */
2896 /* flags: For optimizations. See REXEC_* in regexp.h */
2899 struct regexp *const prog = ReANY(rx);
2903 SSize_t minlen; /* must match at least this many chars */
2904 SSize_t dontbother = 0; /* how many characters not to try at end */
2905 const bool utf8_target = cBOOL(DO_UTF8(sv));
2907 RXi_GET_DECL(prog,progi);
2908 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2909 regmatch_info *const reginfo = ®info_buf;
2910 regexp_paren_pair *swap = NULL;
2912 GET_RE_DEBUG_FLAGS_DECL;
2914 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2915 PERL_UNUSED_ARG(data);
2917 /* Be paranoid... */
2919 Perl_croak(aTHX_ "NULL regexp parameter");
2923 debug_start_match(rx, utf8_target, stringarg, strend,
2927 startpos = stringarg;
2929 /* set these early as they may be used by the HOP macros below */
2930 reginfo->strbeg = strbeg;
2931 reginfo->strend = strend;
2932 reginfo->is_utf8_target = cBOOL(utf8_target);
2934 if (prog->intflags & PREGf_GPOS_SEEN) {
2937 /* set reginfo->ganch, the position where \G can match */
2940 (flags & REXEC_IGNOREPOS)
2941 ? stringarg /* use start pos rather than pos() */
2942 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2943 /* Defined pos(): */
2944 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2945 : strbeg; /* pos() not defined; use start of string */
2947 DEBUG_GPOS_r(Perl_re_printf( aTHX_
2948 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
2950 /* in the presence of \G, we may need to start looking earlier in
2951 * the string than the suggested start point of stringarg:
2952 * if prog->gofs is set, then that's a known, fixed minimum
2955 * /ab|c\G/: gofs = 1
2956 * or if the minimum offset isn't known, then we have to go back
2957 * to the start of the string, e.g. /w+\G/
2960 if (prog->intflags & PREGf_ANCH_GPOS) {
2962 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
2964 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
2966 DEBUG_r(Perl_re_printf( aTHX_
2967 "fail: ganch-gofs before earliest possible start\n"));
2972 startpos = reginfo->ganch;
2974 else if (prog->gofs) {
2975 startpos = HOPBACKc(startpos, prog->gofs);
2979 else if (prog->intflags & PREGf_GPOS_FLOAT)
2983 minlen = prog->minlen;
2984 if ((startpos + minlen) > strend || startpos < strbeg) {
2985 DEBUG_r(Perl_re_printf( aTHX_
2986 "Regex match can't succeed, so not even tried\n"));
2990 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2991 * which will call destuctors to reset PL_regmatch_state, free higher
2992 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2993 * regmatch_info_aux_eval */
2995 oldsave = PL_savestack_ix;
2999 if ((prog->extflags & RXf_USE_INTUIT)
3000 && !(flags & REXEC_CHECKED))
3002 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
3007 if (prog->extflags & RXf_CHECK_ALL) {
3008 /* we can match based purely on the result of INTUIT.
3009 * Set up captures etc just for $& and $-[0]
3010 * (an intuit-only match wont have $1,$2,..) */
3011 assert(!prog->nparens);
3013 /* s/// doesn't like it if $& is earlier than where we asked it to
3014 * start searching (which can happen on something like /.\G/) */
3015 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3018 /* this should only be possible under \G */
3019 assert(prog->intflags & PREGf_GPOS_SEEN);
3020 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3021 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3025 /* match via INTUIT shouldn't have any captures.
3026 * Let @-, @+, $^N know */
3027 prog->lastparen = prog->lastcloseparen = 0;
3028 RX_MATCH_UTF8_set(rx, utf8_target);
3029 prog->offs[0].start = s - strbeg;
3030 prog->offs[0].end = utf8_target
3031 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
3032 : s - strbeg + prog->minlenret;
3033 if ( !(flags & REXEC_NOT_FIRST) )
3034 S_reg_set_capture_string(aTHX_ rx,
3036 sv, flags, utf8_target);
3042 multiline = prog->extflags & RXf_PMf_MULTILINE;
3044 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
3045 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3046 "String too short [regexec_flags]...\n"));
3050 /* Check validity of program. */
3051 if (UCHARAT(progi->program) != REG_MAGIC) {
3052 Perl_croak(aTHX_ "corrupted regexp program");
3055 RX_MATCH_TAINTED_off(rx);
3056 RX_MATCH_UTF8_set(rx, utf8_target);
3058 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3059 reginfo->intuit = 0;
3060 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3061 reginfo->warned = FALSE;
3063 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
3064 /* see how far we have to get to not match where we matched before */
3065 reginfo->till = stringarg + minend;
3067 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
3068 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3069 S_cleanup_regmatch_info_aux has executed (registered by
3070 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3071 magic belonging to this SV.
3072 Not newSVsv, either, as it does not COW.
3074 reginfo->sv = newSV(0);
3075 SvSetSV_nosteal(reginfo->sv, sv);
3076 SAVEFREESV(reginfo->sv);
3079 /* reserve next 2 or 3 slots in PL_regmatch_state:
3080 * slot N+0: may currently be in use: skip it
3081 * slot N+1: use for regmatch_info_aux struct
3082 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3083 * slot N+3: ready for use by regmatch()
3087 regmatch_state *old_regmatch_state;
3088 regmatch_slab *old_regmatch_slab;
3089 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3091 /* on first ever match, allocate first slab */
3092 if (!PL_regmatch_slab) {
3093 Newx(PL_regmatch_slab, 1, regmatch_slab);
3094 PL_regmatch_slab->prev = NULL;
3095 PL_regmatch_slab->next = NULL;
3096 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3099 old_regmatch_state = PL_regmatch_state;
3100 old_regmatch_slab = PL_regmatch_slab;
3102 for (i=0; i <= max; i++) {
3104 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3106 reginfo->info_aux_eval =
3107 reginfo->info_aux->info_aux_eval =
3108 &(PL_regmatch_state->u.info_aux_eval);
3110 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3111 PL_regmatch_state = S_push_slab(aTHX);
3114 /* note initial PL_regmatch_state position; at end of match we'll
3115 * pop back to there and free any higher slabs */
3117 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3118 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
3119 reginfo->info_aux->poscache = NULL;
3121 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
3123 if ((prog->extflags & RXf_EVAL_SEEN))
3124 S_setup_eval_state(aTHX_ reginfo);
3126 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
3129 /* If there is a "must appear" string, look for it. */
3131 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
3132 /* We have to be careful. If the previous successful match
3133 was from this regex we don't want a subsequent partially
3134 successful match to clobber the old results.
3135 So when we detect this possibility we add a swap buffer
3136 to the re, and switch the buffer each match. If we fail,
3137 we switch it back; otherwise we leave it swapped.
3140 /* do we need a save destructor here for eval dies? */
3141 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
3142 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3143 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
3151 if (prog->recurse_locinput)
3152 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3154 /* Simplest case: anchored match need be tried only once, or with
3155 * MBOL, only at the beginning of each line.
3157 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3158 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3159 * match at the start of the string then it won't match anywhere else
3160 * either; while with /.*.../, if it doesn't match at the beginning,
3161 * the earliest it could match is at the start of the next line */
3163 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3166 if (regtry(reginfo, &s))
3169 if (!(prog->intflags & PREGf_ANCH_MBOL))
3172 /* didn't match at start, try at other newline positions */
3175 dontbother = minlen - 1;
3176 end = HOP3c(strend, -dontbother, strbeg) - 1;
3178 /* skip to next newline */
3180 while (s <= end) { /* note it could be possible to match at the end of the string */
3181 /* NB: newlines are the same in unicode as they are in latin */
3184 if (prog->check_substr || prog->check_utf8) {
3185 /* note that with PREGf_IMPLICIT, intuit can only fail
3186 * or return the start position, so it's of limited utility.
3187 * Nevertheless, I made the decision that the potential for
3188 * quick fail was still worth it - DAPM */
3189 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3193 if (regtry(reginfo, &s))
3197 } /* end anchored search */
3199 if (prog->intflags & PREGf_ANCH_GPOS)
3201 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3202 assert(prog->intflags & PREGf_GPOS_SEEN);
3203 /* For anchored \G, the only position it can match from is
3204 * (ganch-gofs); we already set startpos to this above; if intuit
3205 * moved us on from there, we can't possibly succeed */
3206 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
3207 if (s == startpos && regtry(reginfo, &s))
3212 /* Messy cases: unanchored match. */
3213 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
3214 /* we have /x+whatever/ */
3215 /* it must be a one character string (XXXX Except is_utf8_pat?) */
3221 if (! prog->anchored_utf8) {
3222 to_utf8_substr(prog);
3224 ch = SvPVX_const(prog->anchored_utf8)[0];
3227 DEBUG_EXECUTE_r( did_match = 1 );
3228 if (regtry(reginfo, &s)) goto got_it;
3230 while (s < strend && *s == ch)
3237 if (! prog->anchored_substr) {
3238 if (! to_byte_substr(prog)) {
3239 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3242 ch = SvPVX_const(prog->anchored_substr)[0];
3245 DEBUG_EXECUTE_r( did_match = 1 );
3246 if (regtry(reginfo, &s)) goto got_it;
3248 while (s < strend && *s == ch)
3253 DEBUG_EXECUTE_r(if (!did_match)
3254 Perl_re_printf( aTHX_
3255 "Did not find anchored character...\n")
3258 else if (prog->anchored_substr != NULL
3259 || prog->anchored_utf8 != NULL
3260 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
3261 && prog->float_max_offset < strend - s)) {
3266 char *last1; /* Last position checked before */
3270 if (prog->anchored_substr || prog->anchored_utf8) {
3272 if (! prog->anchored_utf8) {
3273 to_utf8_substr(prog);
3275 must = prog->anchored_utf8;
3278 if (! prog->anchored_substr) {
3279 if (! to_byte_substr(prog)) {
3280 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3283 must = prog->anchored_substr;
3285 back_max = back_min = prog->anchored_offset;
3288 if (! prog->float_utf8) {
3289 to_utf8_substr(prog);
3291 must = prog->float_utf8;
3294 if (! prog->float_substr) {
3295 if (! to_byte_substr(prog)) {
3296 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3299 must = prog->float_substr;
3301 back_max = prog->float_max_offset;
3302 back_min = prog->float_min_offset;
3308 last = HOP3c(strend, /* Cannot start after this */
3309 -(SSize_t)(CHR_SVLEN(must)
3310 - (SvTAIL(must) != 0) + back_min), strbeg);
3312 if (s > reginfo->strbeg)
3313 last1 = HOPc(s, -1);
3315 last1 = s - 1; /* bogus */
3317 /* XXXX check_substr already used to find "s", can optimize if
3318 check_substr==must. */
3320 strend = HOPc(strend, -dontbother);
3321 while ( (s <= last) &&
3322 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
3323 (unsigned char*)strend, must,
3324 multiline ? FBMrf_MULTILINE : 0)) ) {
3325 DEBUG_EXECUTE_r( did_match = 1 );
3326 if (HOPc(s, -back_max) > last1) {
3327 last1 = HOPc(s, -back_min);
3328 s = HOPc(s, -back_max);
3331 char * const t = (last1 >= reginfo->strbeg)
3332 ? HOPc(last1, 1) : last1 + 1;
3334 last1 = HOPc(s, -back_min);
3338 while (s <= last1) {
3339 if (regtry(reginfo, &s))
3342 s++; /* to break out of outer loop */
3349 while (s <= last1) {
3350 if (regtry(reginfo, &s))
3356 DEBUG_EXECUTE_r(if (!did_match) {
3357 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
3358 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
3359 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n",
3360 ((must == prog->anchored_substr || must == prog->anchored_utf8)
3361 ? "anchored" : "floating"),
3362 quoted, RE_SV_TAIL(must));
3366 else if ( (c = progi->regstclass) ) {
3368 const OPCODE op = OP(progi->regstclass);
3369 /* don't bother with what can't match */
3370 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
3371 strend = HOPc(strend, -(minlen - 1));
3374 SV * const prop = sv_newmortal();
3375 regprop(prog, prop, c, reginfo, NULL);
3377 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
3379 Perl_re_printf( aTHX_
3380 "Matching stclass %.*s against %s (%d bytes)\n",
3381 (int)SvCUR(prop), SvPVX_const(prop),
3382 quoted, (int)(strend - s));
3385 if (find_byclass(prog, c, s, strend, reginfo))
3387 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n"));
3391 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3399 if (! prog->float_utf8) {
3400 to_utf8_substr(prog);
3402 float_real = prog->float_utf8;
3405 if (! prog->float_substr) {
3406 if (! to_byte_substr(prog)) {
3407 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3410 float_real = prog->float_substr;
3413 little = SvPV_const(float_real, len);
3414 if (SvTAIL(float_real)) {
3415 /* This means that float_real contains an artificial \n on
3416 * the end due to the presence of something like this:
3417 * /foo$/ where we can match both "foo" and "foo\n" at the
3418 * end of the string. So we have to compare the end of the
3419 * string first against the float_real without the \n and
3420 * then against the full float_real with the string. We
3421 * have to watch out for cases where the string might be
3422 * smaller than the float_real or the float_real without
3424 char *checkpos= strend - len;
3426 Perl_re_printf( aTHX_
3427 "%sChecking for float_real.%s\n",
3428 PL_colors[4], PL_colors[5]));
3429 if (checkpos + 1 < strbeg) {
3430 /* can't match, even if we remove the trailing \n
3431 * string is too short to match */
3433 Perl_re_printf( aTHX_
3434 "%sString shorter than required trailing substring, cannot match.%s\n",
3435 PL_colors[4], PL_colors[5]));
3437 } else if (memEQ(checkpos + 1, little, len - 1)) {
3438 /* can match, the end of the string matches without the
3440 last = checkpos + 1;
3441 } else if (checkpos < strbeg) {
3442 /* cant match, string is too short when the "\n" is
3445 Perl_re_printf( aTHX_
3446 "%sString does not contain required trailing substring, cannot match.%s\n",
3447 PL_colors[4], PL_colors[5]));
3449 } else if (!multiline) {
3450 /* non multiline match, so compare with the "\n" at the
3451 * end of the string */
3452 if (memEQ(checkpos, little, len)) {
3456 Perl_re_printf( aTHX_
3457 "%sString does not contain required trailing substring, cannot match.%s\n",
3458 PL_colors[4], PL_colors[5]));
3462 /* multiline match, so we have to search for a place
3463 * where the full string is located */
3469 last = rninstr(s, strend, little, little + len);
3471 last = strend; /* matching "$" */
3474 /* at one point this block contained a comment which was
3475 * probably incorrect, which said that this was a "should not
3476 * happen" case. Even if it was true when it was written I am
3477 * pretty sure it is not anymore, so I have removed the comment
3478 * and replaced it with this one. Yves */
3480 Perl_re_printf( aTHX_
3481 "%sString does not contain required substring, cannot match.%s\n",
3482 PL_colors[4], PL_colors[5]
3486 dontbother = strend - last + prog->float_min_offset;
3488 if (minlen && (dontbother < minlen))
3489 dontbother = minlen - 1;
3490 strend -= dontbother; /* this one's always in bytes! */
3491 /* We don't know much -- general case. */
3494 if (regtry(reginfo, &s))
3503 if (regtry(reginfo, &s))
3505 } while (s++ < strend);
3513 /* s/// doesn't like it if $& is earlier than where we asked it to
3514 * start searching (which can happen on something like /.\G/) */
3515 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3516 && (prog->offs[0].start < stringarg - strbeg))
3518 /* this should only be possible under \G */
3519 assert(prog->intflags & PREGf_GPOS_SEEN);
3520 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3521 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3527 Perl_re_exec_indentf( aTHX_
3528 "rex=0x%" UVxf " freeing offs: 0x%" UVxf "\n",
3536 /* clean up; this will trigger destructors that will free all slabs
3537 * above the current one, and cleanup the regmatch_info_aux
3538 * and regmatch_info_aux_eval sructs */
3540 LEAVE_SCOPE(oldsave);
3542 if (RXp_PAREN_NAMES(prog))
3543 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3545 /* make sure $`, $&, $', and $digit will work later */
3546 if ( !(flags & REXEC_NOT_FIRST) )
3547 S_reg_set_capture_string(aTHX_ rx,
3548 strbeg, reginfo->strend,
3549 sv, flags, utf8_target);
3554 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n",
3555 PL_colors[4], PL_colors[5]));
3557 /* clean up; this will trigger destructors that will free all slabs
3558 * above the current one, and cleanup the regmatch_info_aux
3559 * and regmatch_info_aux_eval sructs */
3561 LEAVE_SCOPE(oldsave);
3564 /* we failed :-( roll it back */
3565 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3566 "rex=0x%" UVxf " rolling back offs: freeing=0x%" UVxf " restoring=0x%" UVxf "\n",
3572 Safefree(prog->offs);
3579 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3580 * Do inc before dec, in case old and new rex are the same */
3581 #define SET_reg_curpm(Re2) \
3582 if (reginfo->info_aux_eval) { \
3583 (void)ReREFCNT_inc(Re2); \
3584 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3585 PM_SETRE((PL_reg_curpm), (Re2)); \
3590 - regtry - try match at specific point
3592 STATIC bool /* 0 failure, 1 success */
3593 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3596 REGEXP *const rx = reginfo->prog;
3597 regexp *const prog = ReANY(rx);
3600 U32 depth = 0; /* used by REGCP_SET */
3602 RXi_GET_DECL(prog,progi);
3603 GET_RE_DEBUG_FLAGS_DECL;
3605 PERL_ARGS_ASSERT_REGTRY;
3607 reginfo->cutpoint=NULL;
3609 prog->offs[0].start = *startposp - reginfo->strbeg;
3610 prog->lastparen = 0;
3611 prog->lastcloseparen = 0;
3613 /* XXXX What this code is doing here?!!! There should be no need
3614 to do this again and again, prog->lastparen should take care of
3617 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3618 * Actually, the code in regcppop() (which Ilya may be meaning by
3619 * prog->lastparen), is not needed at all by the test suite
3620 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3621 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3622 * Meanwhile, this code *is* needed for the
3623 * above-mentioned test suite tests to succeed. The common theme
3624 * on those tests seems to be returning null fields from matches.
3625 * --jhi updated by dapm */
3627 /* After encountering a variant of the issue mentioned above I think
3628 * the point Ilya was making is that if we properly unwind whenever
3629 * we set lastparen to a smaller value then we should not need to do
3630 * this every time, only when needed. So if we have tests that fail if
3631 * we remove this, then it suggests somewhere else we are improperly
3632 * unwinding the lastparen/paren buffers. See UNWIND_PARENS() and
3633 * places it is called, and related regcp() routines. - Yves */
3635 if (prog->nparens) {
3636 regexp_paren_pair *pp = prog->offs;
3638 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3646 result = regmatch(reginfo, *startposp, progi->program + 1);
3648 prog->offs[0].end = result;
3651 if (reginfo->cutpoint)
3652 *startposp= reginfo->cutpoint;
3653 REGCP_UNWIND(lastcp);
3658 #define sayYES goto yes
3659 #define sayNO goto no
3660 #define sayNO_SILENT goto no_silent
3662 /* we dont use STMT_START/END here because it leads to
3663 "unreachable code" warnings, which are bogus, but distracting. */
3664 #define CACHEsayNO \
3665 if (ST.cache_mask) \
3666 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3669 /* this is used to determine how far from the left messages like
3670 'failed...' are printed in regexec.c. It should be set such that
3671 messages are inline with the regop output that created them.
3673 #define REPORT_CODE_OFF 29
3674 #define INDENT_CHARS(depth) ((int)(depth) % 20)
3677 Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...)
3681 PerlIO *f= Perl_debug_log;
3682 PERL_ARGS_ASSERT_RE_EXEC_INDENTF;
3683 va_start(ap, depth);
3684 PerlIO_printf(f, "%*s|%4" UVuf "| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" );
3685 result = PerlIO_vprintf(f, fmt, ap);
3689 #endif /* DEBUGGING */
3692 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3693 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3694 #define CHRTEST_NOT_A_CP_1 -999
3695 #define CHRTEST_NOT_A_CP_2 -998
3697 /* grab a new slab and return the first slot in it */
3699 STATIC regmatch_state *
3702 regmatch_slab *s = PL_regmatch_slab->next;
3704 Newx(s, 1, regmatch_slab);
3705 s->prev = PL_regmatch_slab;
3707 PL_regmatch_slab->next = s;
3709 PL_regmatch_slab = s;
3710 return SLAB_FIRST(s);
3714 /* push a new state then goto it */
3716 #define PUSH_STATE_GOTO(state, node, input) \
3717 pushinput = input; \
3719 st->resume_state = state; \
3722 /* push a new state with success backtracking, then goto it */
3724 #define PUSH_YES_STATE_GOTO(state, node, input) \
3725 pushinput = input; \
3727 st->resume_state = state; \
3728 goto push_yes_state;
3735 regmatch() - main matching routine
3737 This is basically one big switch statement in a loop. We execute an op,
3738 set 'next' to point the next op, and continue. If we come to a point which
3739 we may need to backtrack to on failure such as (A|B|C), we push a
3740 backtrack state onto the backtrack stack. On failure, we pop the top
3741 state, and re-enter the loop at the state indicated. If there are no more
3742 states to pop, we return failure.
3744 Sometimes we also need to backtrack on success; for example /A+/, where
3745 after successfully matching one A, we need to go back and try to
3746 match another one; similarly for lookahead assertions: if the assertion
3747 completes successfully, we backtrack to the state just before the assertion
3748 and then carry on. In these cases, the pushed state is marked as
3749 'backtrack on success too'. This marking is in fact done by a chain of
3750 pointers, each pointing to the previous 'yes' state. On success, we pop to
3751 the nearest yes state, discarding any intermediate failure-only states.
3752 Sometimes a yes state is pushed just to force some cleanup code to be
3753 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3754 it to free the inner regex.
3756 Note that failure backtracking rewinds the cursor position, while
3757 success backtracking leaves it alone.
3759 A pattern is complete when the END op is executed, while a subpattern
3760 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3761 ops trigger the "pop to last yes state if any, otherwise return true"
3764 A common convention in this function is to use A and B to refer to the two
3765 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3766 the subpattern to be matched possibly multiple times, while B is the entire
3767 rest of the pattern. Variable and state names reflect this convention.
3769 The states in the main switch are the union of ops and failure/success of
3770 substates associated with with that op. For example, IFMATCH is the op
3771 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3772 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3773 successfully matched A and IFMATCH_A_fail is a state saying that we have
3774 just failed to match A. Resume states always come in pairs. The backtrack
3775 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3776 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3777 on success or failure.
3779 The struct that holds a backtracking state is actually a big union, with
3780 one variant for each major type of op. The variable st points to the
3781 top-most backtrack struct. To make the code clearer, within each
3782 block of code we #define ST to alias the relevant union.
3784 Here's a concrete example of a (vastly oversimplified) IFMATCH
3790 #define ST st->u.ifmatch
3792 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3793 ST.foo = ...; // some state we wish to save
3795 // push a yes backtrack state with a resume value of
3796 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3798 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3801 case IFMATCH_A: // we have successfully executed A; now continue with B
3803 bar = ST.foo; // do something with the preserved value
3806 case IFMATCH_A_fail: // A failed, so the assertion failed
3807 ...; // do some housekeeping, then ...
3808 sayNO; // propagate the failure
3815 For any old-timers reading this who are familiar with the old recursive
3816 approach, the code above is equivalent to:
3818 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3827 ...; // do some housekeeping, then ...
3828 sayNO; // propagate the failure
3831 The topmost backtrack state, pointed to by st, is usually free. If you
3832 want to claim it, populate any ST.foo fields in it with values you wish to
3833 save, then do one of
3835 PUSH_STATE_GOTO(resume_state, node, newinput);
3836 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3838 which sets that backtrack state's resume value to 'resume_state', pushes a
3839 new free entry to the top of the backtrack stack, then goes to 'node'.
3840 On backtracking, the free slot is popped, and the saved state becomes the
3841 new free state. An ST.foo field in this new top state can be temporarily
3842 accessed to retrieve values, but once the main loop is re-entered, it
3843 becomes available for reuse.
3845 Note that the depth of the backtrack stack constantly increases during the
3846 left-to-right execution of the pattern, rather than going up and down with
3847 the pattern nesting. For example the stack is at its maximum at Z at the
3848 end of the pattern, rather than at X in the following:
3850 /(((X)+)+)+....(Y)+....Z/
3852 The only exceptions to this are lookahead/behind assertions and the cut,
3853 (?>A), which pop all the backtrack states associated with A before
3856 Backtrack state structs are allocated in slabs of about 4K in size.
3857 PL_regmatch_state and st always point to the currently active state,
3858 and PL_regmatch_slab points to the slab currently containing
3859 PL_regmatch_state. The first time regmatch() is called, the first slab is
3860 allocated, and is never freed until interpreter destruction. When the slab
3861 is full, a new one is allocated and chained to the end. At exit from
3862 regmatch(), slabs allocated since entry are freed.
3867 #define DEBUG_STATE_pp(pp) \
3869 DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \
3870 Perl_re_printf( aTHX_ \
3871 "%*s" pp " %s%s%s%s%s\n", \
3872 INDENT_CHARS(depth), "", \
3873 PL_reg_name[st->resume_state], \
3874 ((st==yes_state||st==mark_state) ? "[" : ""), \
3875 ((st==yes_state) ? "Y" : ""), \
3876 ((st==mark_state) ? "M" : ""), \
3877 ((st==yes_state||st==mark_state) ? "]" : "") \
3882 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3887 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3888 const char *start, const char *end, const char *blurb)
3890 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3892 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3897 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3898 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3900 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3901 start, end - start, 60);
3903 Perl_re_printf( aTHX_
3904 "%s%s REx%s %s against %s\n",
3905 PL_colors[4], blurb, PL_colors[5], s0, s1);
3907 if (utf8_target||utf8_pat)
3908 Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n",
3909 utf8_pat ? "pattern" : "",
3910 utf8_pat && utf8_target ? " and " : "",
3911 utf8_target ? "string" : ""
3917 S_dump_exec_pos(pTHX_ const char *locinput,
3918 const regnode *scan,
3919 const char *loc_regeol,
3920 const char *loc_bostr,
3921 const char *loc_reg_starttry,
3922 const bool utf8_target,
3926 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3927 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3928 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3929 /* The part of the string before starttry has one color
3930 (pref0_len chars), between starttry and current
3931 position another one (pref_len - pref0_len chars),
3932 after the current position the third one.
3933 We assume that pref0_len <= pref_len, otherwise we
3934 decrease pref0_len. */
3935 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3936 ? (5 + taill) - l : locinput - loc_bostr;
3939 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3941 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3943 pref0_len = pref_len - (locinput - loc_reg_starttry);
3944 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3945 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3946 ? (5 + taill) - pref_len : loc_regeol - locinput);
3947 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3951 if (pref0_len > pref_len)
3952 pref0_len = pref_len;
3954 const int is_uni = utf8_target ? 1 : 0;
3956 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3957 (locinput - pref_len),pref0_len, 60, 4, 5);
3959 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3960 (locinput - pref_len + pref0_len),
3961 pref_len - pref0_len, 60, 2, 3);
3963 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3964 locinput, loc_regeol - locinput, 10, 0, 1);
3966 const STRLEN tlen=len0+len1+len2;
3967 Perl_re_printf( aTHX_
3968 "%4" IVdf " <%.*s%.*s%s%.*s>%*s|%4u| ",
3969 (IV)(locinput - loc_bostr),
3972 (docolor ? "" : "> <"),
3974 (int)(tlen > 19 ? 0 : 19 - tlen),
3982 /* reg_check_named_buff_matched()
3983 * Checks to see if a named buffer has matched. The data array of
3984 * buffer numbers corresponding to the buffer is expected to reside
3985 * in the regexp->data->data array in the slot stored in the ARG() of
3986 * node involved. Note that this routine doesn't actually care about the
3987 * name, that information is not preserved from compilation to execution.
3988 * Returns the index of the leftmost defined buffer with the given name
3989 * or 0 if non of the buffers matched.
3992 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3995 RXi_GET_DECL(rex,rexi);
3996 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3997 I32 *nums=(I32*)SvPVX(sv_dat);
3999 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
4001 for ( n=0; n<SvIVX(sv_dat); n++ ) {
4002 if ((I32)rex->lastparen >= nums[n] &&
4003 rex->offs[nums[n]].end != -1)
4013 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
4014 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
4016 /* This function determines if there are one or two characters that match
4017 * the first character of the passed-in EXACTish node <text_node>, and if
4018 * so, returns them in the passed-in pointers.
4020 * If it determines that no possible character in the target string can
4021 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
4022 * the first character in <text_node> requires UTF-8 to represent, and the
4023 * target string isn't in UTF-8.)
4025 * If there are more than two characters that could match the beginning of
4026 * <text_node>, or if more context is required to determine a match or not,
4027 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
4029 * The motiviation behind this function is to allow the caller to set up
4030 * tight loops for matching. If <text_node> is of type EXACT, there is
4031 * only one possible character that can match its first character, and so
4032 * the situation is quite simple. But things get much more complicated if
4033 * folding is involved. It may be that the first character of an EXACTFish
4034 * node doesn't participate in any possible fold, e.g., punctuation, so it
4035 * can be matched only by itself. The vast majority of characters that are
4036 * in folds match just two things, their lower and upper-case equivalents.
4037 * But not all are like that; some have multiple possible matches, or match
4038 * sequences of more than one character. This function sorts all that out.
4040 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
4041 * loop of trying to match A*, we know we can't exit where the thing
4042 * following it isn't a B. And something can't be a B unless it is the
4043 * beginning of B. By putting a quick test for that beginning in a tight
4044 * loop, we can rule out things that can't possibly be B without having to
4045 * break out of the loop, thus avoiding work. Similarly, if A is a single
4046 * character, we can make a tight loop matching A*, using the outputs of
4049 * If the target string to match isn't in UTF-8, and there aren't
4050 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
4051 * the one or two possible octets (which are characters in this situation)
4052 * that can match. In all cases, if there is only one character that can
4053 * match, *<c1p> and *<c2p> will be identical.
4055 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
4056 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
4057 * can match the beginning of <text_node>. They should be declared with at
4058 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
4059 * undefined what these contain.) If one or both of the buffers are
4060 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
4061 * corresponding invariant. If variant, the corresponding *<c1p> and/or
4062 * *<c2p> will be set to a negative number(s) that shouldn't match any code
4063 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
4064 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
4066 const bool utf8_target = reginfo->is_utf8_target;
4068 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
4069 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
4070 bool use_chrtest_void = FALSE;
4071 const bool is_utf8_pat = reginfo->is_utf8_pat;
4073 /* Used when we have both utf8 input and utf8 output, to avoid converting
4074 * to/from code points */
4075 bool utf8_has_been_setup = FALSE;
4079 U8 *pat = (U8*)STRING(text_node);
4080 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
4082 if (OP(text_node) == EXACT || OP(text_node) == EXACTL) {
4084 /* In an exact node, only one thing can be matched, that first
4085 * character. If both the pat and the target are UTF-8, we can just
4086 * copy the input to the output, avoiding finding the code point of
4091 else if (utf8_target) {
4092 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
4093 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
4094 utf8_has_been_setup = TRUE;
4097 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
4100 else { /* an EXACTFish node */
4101 U8 *pat_end = pat + STR_LEN(text_node);
4103 /* An EXACTFL node has at least some characters unfolded, because what
4104 * they match is not known until now. So, now is the time to fold
4105 * the first few of them, as many as are needed to determine 'c1' and
4106 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
4107 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
4108 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
4109 * need to fold as many characters as a single character can fold to,
4110 * so that later we can check if the first ones are such a multi-char
4111 * fold. But, in such a pattern only locale-problematic characters
4112 * aren't folded, so we can skip this completely if the first character
4113 * in the node isn't one of the tricky ones */
4114 if (OP(text_node) == EXACTFL) {
4116 if (! is_utf8_pat) {
4117 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
4119 folded[0] = folded[1] = 's';
4121 pat_end = folded + 2;
4124 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
4129 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
4131 *(d++) = (U8) toFOLD_LC(*s);
4136 _to_utf8_fold_flags(s,
4139 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
4150 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
4151 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
4153 /* Multi-character folds require more context to sort out. Also
4154 * PL_utf8_foldclosures used below doesn't handle them, so have to
4155 * be handled outside this routine */
4156 use_chrtest_void = TRUE;
4158 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
4159 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
4161 /* Load the folds hash, if not already done */
4163 if (! PL_utf8_foldclosures) {
4164 _load_PL_utf8_foldclosures();
4167 /* The fold closures data structure is a hash with the keys
4168 * being the UTF-8 of every character that is folded to, like
4169 * 'k', and the values each an array of all code points that
4170 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
4171 * Multi-character folds are not included */
4172 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
4177 /* Not found in the hash, therefore there are no folds
4178 * containing it, so there is only a single character that
4182 else { /* Does participate in folds */
4183 AV* list = (AV*) *listp;
4184 if (av_tindex_nomg(list) != 1) {
4186 /* If there aren't exactly two folds to this, it is
4187 * outside the scope of this function */
4188 use_chrtest_void = TRUE;
4190 else { /* There are two. Get them */
4191 SV** c_p = av_fetch(list, 0, FALSE);
4193 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
4197 c_p = av_fetch(list, 1, FALSE);
4199 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
4203 /* Folds that cross the 255/256 boundary are forbidden
4204 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
4205 * one is ASCIII. Since the pattern character is above
4206 * 255, and its only other match is below 256, the only
4207 * legal match will be to itself. We have thrown away
4208 * the original, so have to compute which is the one
4210 if ((c1 < 256) != (c2 < 256)) {
4211 if ((OP(text_node) == EXACTFL
4212 && ! IN_UTF8_CTYPE_LOCALE)
4213 || ((OP(text_node) == EXACTFA
4214 || OP(text_node) == EXACTFA_NO_TRIE)
4215 && (isASCII(c1) || isASCII(c2))))
4228 else /* Here, c1 is <= 255 */
4230 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
4231 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
4232 && ((OP(text_node) != EXACTFA
4233 && OP(text_node) != EXACTFA_NO_TRIE)
4236 /* Here, there could be something above Latin1 in the target
4237 * which folds to this character in the pattern. All such
4238 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
4239 * than two characters involved in their folds, so are outside
4240 * the scope of this function */
4241 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4242 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
4245 use_chrtest_void = TRUE;
4248 else { /* Here nothing above Latin1 can fold to the pattern
4250 switch (OP(text_node)) {
4252 case EXACTFL: /* /l rules */
4253 c2 = PL_fold_locale[c1];
4256 case EXACTF: /* This node only generated for non-utf8
4258 assert(! is_utf8_pat);
4259 if (! utf8_target) { /* /d rules */
4264 /* /u rules for all these. This happens to work for
4265 * EXACTFA as nothing in Latin1 folds to ASCII */
4266 case EXACTFA_NO_TRIE: /* This node only generated for
4267 non-utf8 patterns */
4268 assert(! is_utf8_pat);
4273 c2 = PL_fold_latin1[c1];
4277 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
4278 NOT_REACHED; /* NOTREACHED */
4284 /* Here have figured things out. Set up the returns */
4285 if (use_chrtest_void) {
4286 *c2p = *c1p = CHRTEST_VOID;
4288 else if (utf8_target) {
4289 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
4290 uvchr_to_utf8(c1_utf8, c1);
4291 uvchr_to_utf8(c2_utf8, c2);
4294 /* Invariants are stored in both the utf8 and byte outputs; Use
4295 * negative numbers otherwise for the byte ones. Make sure that the
4296 * byte ones are the same iff the utf8 ones are the same */
4297 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
4298 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
4301 ? CHRTEST_NOT_A_CP_1
4302 : CHRTEST_NOT_A_CP_2;
4304 else if (c1 > 255) {
4305 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
4310 *c1p = *c2p = c2; /* c2 is the only representable value */
4312 else { /* c1 is representable; see about c2 */
4314 *c2p = (c2 < 256) ? c2 : c1;
4321 S_isGCB(pTHX_ const GCB_enum before, const GCB_enum after, const U8 * const strbeg, const U8 * const curpos, const bool utf8_target)
4323 /* returns a boolean indicating if there is a Grapheme Cluster Boundary
4324 * between the inputs. See http://www.unicode.org/reports/tr29/. */
4326 PERL_ARGS_ASSERT_ISGCB;
4328 switch (GCB_table[before][after]) {
4335 case GCB_RI_then_RI:
4338 U8 * temp_pos = (U8 *) curpos;
4340 /* Do not break within emoji flag sequences. That is, do not
4341 * break between regional indicator (RI) symbols if there is an
4342 * odd number of RI characters before the break point.
4343 * GB12 ^ (RI RI)* RI × RI
4344 * GB13 [^RI] (RI RI)* RI × RI */
4346 while (backup_one_GCB(strbeg,
4348 utf8_target) == GCB_Regional_Indicator)
4353 return RI_count % 2 != 1;
4356 case GCB_EX_then_EM:
4358 /* GB10 ( E_Base | E_Base_GAZ ) Extend* × E_Modifier */
4360 U8 * temp_pos = (U8 *) curpos;
4364 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4366 while (prev == GCB_Extend);
4368 return prev != GCB_E_Base && prev != GCB_E_Base_GAZ;
4376 Perl_re_printf( aTHX_ "Unhandled GCB pair: GCB_table[%d, %d] = %d\n",
4377 before, after, GCB_table[before][after]);
4384 S_backup_one_GCB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4388 PERL_ARGS_ASSERT_BACKUP_ONE_GCB;
4390 if (*curpos < strbeg) {
4395 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4396 U8 * prev_prev_char_pos;
4398 if (! prev_char_pos) {
4402 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4403 gcb = getGCB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4404 *curpos = prev_char_pos;
4405 prev_char_pos = prev_prev_char_pos;
4408 *curpos = (U8 *) strbeg;
4413 if (*curpos - 2 < strbeg) {
4414 *curpos = (U8 *) strbeg;
4418 gcb = getGCB_VAL_CP(*(*curpos - 1));
4424 /* Combining marks attach to most classes that precede them, but this defines
4425 * the exceptions (from TR14) */
4426 #define LB_CM_ATTACHES_TO(prev) ( ! ( prev == LB_EDGE \
4427 || prev == LB_Mandatory_Break \
4428 || prev == LB_Carriage_Return \
4429 || prev == LB_Line_Feed \
4430 || prev == LB_Next_Line \
4431 || prev == LB_Space \
4432 || prev == LB_ZWSpace))
4435 S_isLB(pTHX_ LB_enum before,
4437 const U8 * const strbeg,
4438 const U8 * const curpos,
4439 const U8 * const strend,
4440 const bool utf8_target)
4442 U8 * temp_pos = (U8 *) curpos;
4443 LB_enum prev = before;
4445 /* Is the boundary between 'before' and 'after' line-breakable?
4446 * Most of this is just a table lookup of a generated table from Unicode
4447 * rules. But some rules require context to decide, and so have to be
4448 * implemented in code */
4450 PERL_ARGS_ASSERT_ISLB;
4452 /* Rule numbers in the comments below are as of Unicode 9.0 */
4456 switch (LB_table[before][after]) {
4461 case LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4464 case LB_SP_foo + LB_BREAKABLE:
4465 case LB_SP_foo + LB_NOBREAK:
4466 case LB_SP_foo + LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4468 /* When we have something following a SP, we have to look at the
4469 * context in order to know what to do.
4471 * SP SP should not reach here because LB7: Do not break before
4472 * spaces. (For two spaces in a row there is nothing that
4473 * overrides that) */
4474 assert(after != LB_Space);
4476 /* Here we have a space followed by a non-space. Mostly this is a
4477 * case of LB18: "Break after spaces". But there are complications
4478 * as the handling of spaces is somewhat tricky. They are in a
4479 * number of rules, which have to be applied in priority order, but
4480 * something earlier in the string can cause a rule to be skipped
4481 * and a lower priority rule invoked. A prime example is LB7 which
4482 * says don't break before a space. But rule LB8 (lower priority)
4483 * says that the first break opportunity after a ZW is after any
4484 * span of spaces immediately after it. If a ZW comes before a SP
4485 * in the input, rule LB8 applies, and not LB7. Other such rules
4486 * involve combining marks which are rules 9 and 10, but they may
4487 * override higher priority rules if they come earlier in the
4488 * string. Since we're doing random access into the middle of the
4489 * string, we have to look for rules that should get applied based
4490 * on both string position and priority. Combining marks do not
4491 * attach to either ZW nor SP, so we don't have to consider them
4494 * To check for LB8, we have to find the first non-space character
4495 * before this span of spaces */
4497 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4499 while (prev == LB_Space);
4501 /* LB8 Break before any character following a zero-width space,
4502 * even if one or more spaces intervene.
4504 * So if we have a ZW just before this span, and to get here this
4505 * is the final space in the span. */
4506 if (prev == LB_ZWSpace) {
4510 /* Here, not ZW SP+. There are several rules that have higher
4511 * priority than LB18 and can be resolved now, as they don't depend
4512 * on anything earlier in the string (except ZW, which we have
4513 * already handled). One of these rules is LB11 Do not break
4514 * before Word joiner, but we have specially encoded that in the
4515 * lookup table so it is caught by the single test below which
4516 * catches the other ones. */
4517 if (LB_table[LB_Space][after] - LB_SP_foo
4518 == LB_NOBREAK_EVEN_WITH_SP_BETWEEN)
4523 /* If we get here, we have to XXX consider combining marks. */
4524 if (prev == LB_Combining_Mark) {
4526 /* What happens with these depends on the character they
4529 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4531 while (prev == LB_Combining_Mark);
4533 /* Most times these attach to and inherit the characteristics
4534 * of that character, but not always, and when not, they are to
4535 * be treated as AL by rule LB10. */
4536 if (! LB_CM_ATTACHES_TO(prev)) {
4537 prev = LB_Alphabetic;
4541 /* Here, we have the character preceding the span of spaces all set
4542 * up. We follow LB18: "Break after spaces" unless the table shows
4543 * that is overriden */
4544 return LB_table[prev][after] != LB_NOBREAK_EVEN_WITH_SP_BETWEEN;
4548 /* We don't know how to treat the CM except by looking at the first
4549 * non-CM character preceding it. ZWJ is treated as CM */
4551 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4553 while (prev == LB_Combining_Mark || prev == LB_ZWJ);
4555 /* Here, 'prev' is that first earlier non-CM character. If the CM
4556 * attatches to it, then it inherits the behavior of 'prev'. If it
4557 * doesn't attach, it is to be treated as an AL */
4558 if (! LB_CM_ATTACHES_TO(prev)) {
4559 prev = LB_Alphabetic;
4564 case LB_HY_or_BA_then_foo + LB_BREAKABLE:
4565 case LB_HY_or_BA_then_foo + LB_NOBREAK:
4567 /* LB21a Don't break after Hebrew + Hyphen.
4568 * HL (HY | BA) × */
4570 if (backup_one_LB(strbeg, &temp_pos, utf8_target)
4571 == LB_Hebrew_Letter)
4576 return LB_table[prev][after] - LB_HY_or_BA_then_foo == LB_BREAKABLE;
4578 case LB_PR_or_PO_then_OP_or_HY + LB_BREAKABLE:
4579 case LB_PR_or_PO_then_OP_or_HY + LB_NOBREAK:
4581 /* LB25a (PR | PO) × ( OP | HY )? NU */
4582 if (advance_one_LB(&temp_pos, strend, utf8_target) == LB_Numeric) {
4586 return LB_table[prev][after] - LB_PR_or_PO_then_OP_or_HY
4589 case LB_SY_or_IS_then_various + LB_BREAKABLE:
4590 case LB_SY_or_IS_then_various + LB_NOBREAK:
4592 /* LB25d NU (SY | IS)* × (NU | SY | IS | CL | CP ) */
4594 LB_enum temp = prev;
4596 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4598 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric);
4599 if (temp == LB_Numeric) {
4603 return LB_table[prev][after] - LB_SY_or_IS_then_various
4607 case LB_various_then_PO_or_PR + LB_BREAKABLE:
4608 case LB_various_then_PO_or_PR + LB_NOBREAK:
4610 /* LB25e NU (SY | IS)* (CL | CP)? × (PO | PR) */
4612 LB_enum temp = prev;
4613 if (temp == LB_Close_Punctuation || temp == LB_Close_Parenthesis)
4615 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4617 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric) {
4618 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4620 if (temp == LB_Numeric) {
4623 return LB_various_then_PO_or_PR;
4626 case LB_RI_then_RI + LB_NOBREAK:
4627 case LB_RI_then_RI + LB_BREAKABLE:
4631 /* LB30a Break between two regional indicator symbols if and
4632 * only if there are an even number of regional indicators
4633 * preceding the position of the break.
4635 * sot (RI RI)* RI × RI
4636 * [^RI] (RI RI)* RI × RI */
4638 while (backup_one_LB(strbeg,
4640 utf8_target) == LB_Regional_Indicator)
4645 return RI_count % 2 == 0;
4653 Perl_re_printf( aTHX_ "Unhandled LB pair: LB_table[%d, %d] = %d\n",
4654 before, after, LB_table[before][after]);
4661 S_advance_one_LB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4665 PERL_ARGS_ASSERT_ADVANCE_ONE_LB;
4667 if (*curpos >= strend) {
4672 *curpos += UTF8SKIP(*curpos);
4673 if (*curpos >= strend) {
4676 lb = getLB_VAL_UTF8(*curpos, strend);
4680 if (*curpos >= strend) {
4683 lb = getLB_VAL_CP(**curpos);
4690 S_backup_one_LB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4694 PERL_ARGS_ASSERT_BACKUP_ONE_LB;
4696 if (*curpos < strbeg) {
4701 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4702 U8 * prev_prev_char_pos;
4704 if (! prev_char_pos) {
4708 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4709 lb = getLB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4710 *curpos = prev_char_pos;
4711 prev_char_pos = prev_prev_char_pos;
4714 *curpos = (U8 *) strbeg;
4719 if (*curpos - 2 < strbeg) {
4720 *curpos = (U8 *) strbeg;
4724 lb = getLB_VAL_CP(*(*curpos - 1));
4731 S_isSB(pTHX_ SB_enum before,
4733 const U8 * const strbeg,
4734 const U8 * const curpos,
4735 const U8 * const strend,
4736 const bool utf8_target)
4738 /* returns a boolean indicating if there is a Sentence Boundary Break
4739 * between the inputs. See http://www.unicode.org/reports/tr29/ */
4741 U8 * lpos = (U8 *) curpos;
4742 bool has_para_sep = FALSE;
4743 bool has_sp = FALSE;
4745 PERL_ARGS_ASSERT_ISSB;
4747 /* Break at the start and end of text.
4750 But unstated in Unicode is don't break if the text is empty */
4751 if (before == SB_EDGE || after == SB_EDGE) {
4752 return before != after;
4755 /* SB 3: Do not break within CRLF. */
4756 if (before == SB_CR && after == SB_LF) {
4760 /* Break after paragraph separators. CR and LF are considered
4761 * so because Unicode views text as like word processing text where there
4762 * are no newlines except between paragraphs, and the word processor takes
4763 * care of wrapping without there being hard line-breaks in the text *./
4764 SB4. Sep | CR | LF ÷ */
4765 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4769 /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF.
4770 * (See Section 6.2, Replacing Ignore Rules.)
4771 SB5. X (Extend | Format)* → X */
4772 if (after == SB_Extend || after == SB_Format) {
4774 /* Implied is that the these characters attach to everything
4775 * immediately prior to them except for those separator-type
4776 * characters. And the rules earlier have already handled the case
4777 * when one of those immediately precedes the extend char */
4781 if (before == SB_Extend || before == SB_Format) {
4782 U8 * temp_pos = lpos;
4783 const SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4784 if ( backup != SB_EDGE
4793 /* Here, both 'before' and 'backup' are these types; implied is that we
4794 * don't break between them */
4795 if (backup == SB_Extend || backup == SB_Format) {
4800 /* Do not break after ambiguous terminators like period, if they are
4801 * immediately followed by a number or lowercase letter, if they are
4802 * between uppercase letters, if the first following letter (optionally
4803 * after certain punctuation) is lowercase, or if they are followed by
4804 * "continuation" punctuation such as comma, colon, or semicolon. For
4805 * example, a period may be an abbreviation or numeric period, and thus may
4806 * not mark the end of a sentence.
4808 * SB6. ATerm × Numeric */
4809 if (before == SB_ATerm && after == SB_Numeric) {
4813 /* SB7. (Upper | Lower) ATerm × Upper */
4814 if (before == SB_ATerm && after == SB_Upper) {
4815 U8 * temp_pos = lpos;
4816 SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4817 if (backup == SB_Upper || backup == SB_Lower) {
4822 /* The remaining rules that aren't the final one, all require an STerm or
4823 * an ATerm after having backed up over some Close* Sp*, and in one case an
4824 * optional Paragraph separator, although one rule doesn't have any Sp's in it.
4825 * So do that backup now, setting flags if either Sp or a paragraph
4826 * separator are found */
4828 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4829 has_para_sep = TRUE;
4830 before = backup_one_SB(strbeg, &lpos, utf8_target);
4833 if (before == SB_Sp) {
4836 before = backup_one_SB(strbeg, &lpos, utf8_target);
4838 while (before == SB_Sp);
4841 while (before == SB_Close) {
4842 before = backup_one_SB(strbeg, &lpos, utf8_target);
4845 /* The next few rules apply only when the backed-up-to is an ATerm, and in
4846 * most cases an STerm */
4847 if (before == SB_STerm || before == SB_ATerm) {
4849 /* So, here the lhs matches
4850 * (STerm | ATerm) Close* Sp* (Sep | CR | LF)?
4851 * and we have set flags if we found an Sp, or the optional Sep,CR,LF.
4852 * The rules that apply here are:
4854 * SB8 ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR
4855 | LF | STerm | ATerm) )* Lower
4856 SB8a (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm)
4857 SB9 (STerm | ATerm) Close* × (Close | Sp | Sep | CR | LF)
4858 SB10 (STerm | ATerm) Close* Sp* × (Sp | Sep | CR | LF)
4859 SB11 (STerm | ATerm) Close* Sp* (Sep | CR | LF)? ÷
4862 /* And all but SB11 forbid having seen a paragraph separator */
4863 if (! has_para_sep) {
4864 if (before == SB_ATerm) { /* SB8 */
4865 U8 * rpos = (U8 *) curpos;
4866 SB_enum later = after;
4868 while ( later != SB_OLetter
4869 && later != SB_Upper
4870 && later != SB_Lower
4874 && later != SB_STerm
4875 && later != SB_ATerm
4876 && later != SB_EDGE)
4878 later = advance_one_SB(&rpos, strend, utf8_target);
4880 if (later == SB_Lower) {
4885 if ( after == SB_SContinue /* SB8a */
4886 || after == SB_STerm
4887 || after == SB_ATerm)
4892 if (! has_sp) { /* SB9 applies only if there was no Sp* */
4893 if ( after == SB_Close
4903 /* SB10. This and SB9 could probably be combined some way, but khw
4904 * has decided to follow the Unicode rule book precisely for
4905 * simplified maintenance */
4919 /* Otherwise, do not break.
4926 S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4930 PERL_ARGS_ASSERT_ADVANCE_ONE_SB;
4932 if (*curpos >= strend) {
4938 *curpos += UTF8SKIP(*curpos);
4939 if (*curpos >= strend) {
4942 sb = getSB_VAL_UTF8(*curpos, strend);
4943 } while (sb == SB_Extend || sb == SB_Format);
4948 if (*curpos >= strend) {
4951 sb = getSB_VAL_CP(**curpos);
4952 } while (sb == SB_Extend || sb == SB_Format);
4959 S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4963 PERL_ARGS_ASSERT_BACKUP_ONE_SB;
4965 if (*curpos < strbeg) {
4970 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4971 if (! prev_char_pos) {
4975 /* Back up over Extend and Format. curpos is always just to the right
4976 * of the characater whose value we are getting */
4978 U8 * prev_prev_char_pos;
4979 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1,
4982 sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4983 *curpos = prev_char_pos;
4984 prev_char_pos = prev_prev_char_pos;
4987 *curpos = (U8 *) strbeg;
4990 } while (sb == SB_Extend || sb == SB_Format);
4994 if (*curpos - 2 < strbeg) {
4995 *curpos = (U8 *) strbeg;
4999 sb = getSB_VAL_CP(*(*curpos - 1));
5000 } while (sb == SB_Extend || sb == SB_Format);
5007 S_isWB(pTHX_ WB_enum previous,
5010 const U8 * const strbeg,
5011 const U8 * const curpos,
5012 const U8 * const strend,
5013 const bool utf8_target)
5015 /* Return a boolean as to if the boundary between 'before' and 'after' is
5016 * a Unicode word break, using their published algorithm, but tailored for
5017 * Perl by treating spans of white space as one unit. Context may be
5018 * needed to make this determination. If the value for the character
5019 * before 'before' is known, it is passed as 'previous'; otherwise that
5020 * should be set to WB_UNKNOWN. The other input parameters give the
5021 * boundaries and current position in the matching of the string. That
5022 * is, 'curpos' marks the position where the character whose wb value is
5023 * 'after' begins. See http://www.unicode.org/reports/tr29/ */
5025 U8 * before_pos = (U8 *) curpos;
5026 U8 * after_pos = (U8 *) curpos;
5027 WB_enum prev = before;
5030 PERL_ARGS_ASSERT_ISWB;
5032 /* Rule numbers in the comments below are as of Unicode 9.0 */
5036 switch (WB_table[before][after]) {
5043 case WB_hs_then_hs: /* 2 horizontal spaces in a row */
5044 next = advance_one_WB(&after_pos, strend, utf8_target,
5045 FALSE /* Don't skip Extend nor Format */ );
5046 /* A space immediately preceeding an Extend or Format is attached
5047 * to by them, and hence gets separated from previous spaces.
5048 * Otherwise don't break between horizontal white space */
5049 return next == WB_Extend || next == WB_Format;
5051 /* WB4 Ignore Format and Extend characters, except when they appear at
5052 * the beginning of a region of text. This code currently isn't
5053 * general purpose, but it works as the rules are currently and likely
5054 * to be laid out. The reason it works is that when 'they appear at
5055 * the beginning of a region of text', the rule is to break before
5056 * them, just like any other character. Therefore, the default rule
5057 * applies and we don't have to look in more depth. Should this ever
5058 * change, we would have to have 2 'case' statements, like in the rules
5059 * below, and backup a single character (not spacing over the extend
5060 * ones) and then see if that is one of the region-end characters and
5062 case WB_Ex_or_FO_or_ZWJ_then_foo:
5063 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5066 case WB_DQ_then_HL + WB_BREAKABLE:
5067 case WB_DQ_then_HL + WB_NOBREAK:
5069 /* WB7c Hebrew_Letter Double_Quote × Hebrew_Letter */
5071 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5072 == WB_Hebrew_Letter)
5077 return WB_table[before][after] - WB_DQ_then_HL == WB_BREAKABLE;
5079 case WB_HL_then_DQ + WB_BREAKABLE:
5080 case WB_HL_then_DQ + WB_NOBREAK:
5082 /* WB7b Hebrew_Letter × Double_Quote Hebrew_Letter */
5084 if (advance_one_WB(&after_pos, strend, utf8_target,
5085 TRUE /* Do skip Extend and Format */ )
5086 == WB_Hebrew_Letter)
5091 return WB_table[before][after] - WB_HL_then_DQ == WB_BREAKABLE;
5093 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_NOBREAK:
5094 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_BREAKABLE:
5096 /* WB6 (ALetter | Hebrew_Letter) × (MidLetter | MidNumLet
5097 * | Single_Quote) (ALetter | Hebrew_Letter) */
5099 next = advance_one_WB(&after_pos, strend, utf8_target,
5100 TRUE /* Do skip Extend and Format */ );
5102 if (next == WB_ALetter || next == WB_Hebrew_Letter)
5107 return WB_table[before][after]
5108 - WB_LE_or_HL_then_MB_or_ML_or_SQ == WB_BREAKABLE;
5110 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_NOBREAK:
5111 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_BREAKABLE:
5113 /* WB7 (ALetter | Hebrew_Letter) (MidLetter | MidNumLet
5114 * | Single_Quote) × (ALetter | Hebrew_Letter) */
5116 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5117 if (prev == WB_ALetter || prev == WB_Hebrew_Letter)
5122 return WB_table[before][after]
5123 - WB_MB_or_ML_or_SQ_then_LE_or_HL == WB_BREAKABLE;
5125 case WB_MB_or_MN_or_SQ_then_NU + WB_NOBREAK:
5126 case WB_MB_or_MN_or_SQ_then_NU + WB_BREAKABLE:
5128 /* WB11 Numeric (MidNum | (MidNumLet | Single_Quote)) × Numeric
5131 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5137 return WB_table[before][after]
5138 - WB_MB_or_MN_or_SQ_then_NU == WB_BREAKABLE;
5140 case WB_NU_then_MB_or_MN_or_SQ + WB_NOBREAK:
5141 case WB_NU_then_MB_or_MN_or_SQ + WB_BREAKABLE:
5143 /* WB12 Numeric × (MidNum | MidNumLet | Single_Quote) Numeric */
5145 if (advance_one_WB(&after_pos, strend, utf8_target,
5146 TRUE /* Do skip Extend and Format */ )
5152 return WB_table[before][after]
5153 - WB_NU_then_MB_or_MN_or_SQ == WB_BREAKABLE;
5155 case WB_RI_then_RI + WB_NOBREAK:
5156 case WB_RI_then_RI + WB_BREAKABLE:
5160 /* Do not break within emoji flag sequences. That is, do not
5161 * break between regional indicator (RI) symbols if there is an
5162 * odd number of RI characters before the potential break
5165 * WB15 ^ (RI RI)* RI × RI
5166 * WB16 [^RI] (RI RI)* RI × RI */
5168 while (backup_one_WB(&previous,
5171 utf8_target) == WB_Regional_Indicator)
5176 return RI_count % 2 != 1;
5184 Perl_re_printf( aTHX_ "Unhandled WB pair: WB_table[%d, %d] = %d\n",
5185 before, after, WB_table[before][after]);
5192 S_advance_one_WB(pTHX_ U8 ** curpos,
5193 const U8 * const strend,
5194 const bool utf8_target,
5195 const bool skip_Extend_Format)
5199 PERL_ARGS_ASSERT_ADVANCE_ONE_WB;
5201 if (*curpos >= strend) {
5207 /* Advance over Extend and Format */
5209 *curpos += UTF8SKIP(*curpos);
5210 if (*curpos >= strend) {
5213 wb = getWB_VAL_UTF8(*curpos, strend);
5214 } while ( skip_Extend_Format
5215 && (wb == WB_Extend || wb == WB_Format));
5220 if (*curpos >= strend) {
5223 wb = getWB_VAL_CP(**curpos);
5224 } while ( skip_Extend_Format
5225 && (wb == WB_Extend || wb == WB_Format));
5232 S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5236 PERL_ARGS_ASSERT_BACKUP_ONE_WB;
5238 /* If we know what the previous character's break value is, don't have
5240 if (*previous != WB_UNKNOWN) {
5243 /* But we need to move backwards by one */
5245 *curpos = reghopmaybe3(*curpos, -1, strbeg);
5247 *previous = WB_EDGE;
5248 *curpos = (U8 *) strbeg;
5251 *previous = WB_UNKNOWN;
5256 *previous = (*curpos <= strbeg) ? WB_EDGE : WB_UNKNOWN;
5259 /* And we always back up over these three types */
5260 if (wb != WB_Extend && wb != WB_Format && wb != WB_ZWJ) {
5265 if (*curpos < strbeg) {
5270 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5271 if (! prev_char_pos) {
5275 /* Back up over Extend and Format. curpos is always just to the right
5276 * of the characater whose value we are getting */
5278 U8 * prev_prev_char_pos;
5279 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos,
5283 wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5284 *curpos = prev_char_pos;
5285 prev_char_pos = prev_prev_char_pos;
5288 *curpos = (U8 *) strbeg;
5291 } while (wb == WB_Extend || wb == WB_Format || wb == WB_ZWJ);
5295 if (*curpos - 2 < strbeg) {
5296 *curpos = (U8 *) strbeg;
5300 wb = getWB_VAL_CP(*(*curpos - 1));
5301 } while (wb == WB_Extend || wb == WB_Format);
5307 #define EVAL_CLOSE_PAREN_IS(st,expr) \
5310 ( ( st )->u.eval.close_paren ) && \
5311 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5314 #define EVAL_CLOSE_PAREN_IS_TRUE(st,expr) \
5317 ( ( st )->u.eval.close_paren ) && \
5319 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5323 #define EVAL_CLOSE_PAREN_SET(st,expr) \
5324 (st)->u.eval.close_paren = ( (expr) + 1 )
5326 #define EVAL_CLOSE_PAREN_CLEAR(st) \
5327 (st)->u.eval.close_paren = 0
5329 /* returns -1 on failure, $+[0] on success */
5331 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
5334 const bool utf8_target = reginfo->is_utf8_target;
5335 const U32 uniflags = UTF8_ALLOW_DEFAULT;
5336 REGEXP *rex_sv = reginfo->prog;
5337 regexp *rex = ReANY(rex_sv);
5338 RXi_GET_DECL(rex,rexi);
5339 /* the current state. This is a cached copy of PL_regmatch_state */
5341 /* cache heavy used fields of st in registers */
5344 U32 n = 0; /* general value; init to avoid compiler warning */
5345 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
5346 char *locinput = startpos;
5347 char *pushinput; /* where to continue after a PUSH */
5348 I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */
5350 bool result = 0; /* return value of S_regmatch */
5351 U32 depth = 0; /* depth of backtrack stack */
5352 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
5353 const U32 max_nochange_depth =
5354 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
5355 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
5356 regmatch_state *yes_state = NULL; /* state to pop to on success of
5358 /* mark_state piggy backs on the yes_state logic so that when we unwind
5359 the stack on success we can update the mark_state as we go */
5360 regmatch_state *mark_state = NULL; /* last mark state we have seen */
5361 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
5362 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
5364 bool no_final = 0; /* prevent failure from backtracking? */
5365 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
5366 char *startpoint = locinput;
5367 SV *popmark = NULL; /* are we looking for a mark? */
5368 SV *sv_commit = NULL; /* last mark name seen in failure */
5369 SV *sv_yes_mark = NULL; /* last mark name we have seen
5370 during a successful match */
5371 U32 lastopen = 0; /* last open we saw */
5372 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
5373 SV* const oreplsv = GvSVn(PL_replgv);
5374 /* these three flags are set by various ops to signal information to
5375 * the very next op. They have a useful lifetime of exactly one loop
5376 * iteration, and are not preserved or restored by state pushes/pops
5378 bool sw = 0; /* the condition value in (?(cond)a|b) */
5379 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
5380 int logical = 0; /* the following EVAL is:
5384 or the following IFMATCH/UNLESSM is:
5385 false: plain (?=foo)
5386 true: used as a condition: (?(?=foo))
5388 PAD* last_pad = NULL;
5390 U8 gimme = G_SCALAR;
5391 CV *caller_cv = NULL; /* who called us */
5392 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
5393 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
5394 U32 maxopenparen = 0; /* max '(' index seen so far */
5395 int to_complement; /* Invert the result? */
5396 _char_class_number classnum;
5397 bool is_utf8_pat = reginfo->is_utf8_pat;
5400 /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */
5401 #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL))
5402 # define SOLARIS_BAD_OPTIMIZER
5403 const U32 *pl_charclass_dup = PL_charclass;
5404 # define PL_charclass pl_charclass_dup
5408 GET_RE_DEBUG_FLAGS_DECL;
5411 /* protect against undef(*^R) */
5412 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
5414 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
5415 multicall_oldcatch = 0;
5416 PERL_UNUSED_VAR(multicall_cop);
5418 PERL_ARGS_ASSERT_REGMATCH;
5420 st = PL_regmatch_state;
5422 /* Note that nextchr is a byte even in UTF */
5426 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
5427 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5428 Perl_re_printf( aTHX_ "regmatch start\n" );
5431 while (scan != NULL) {
5434 next = scan + NEXT_OFF(scan);
5437 state_num = OP(scan);
5441 if (state_num <= REGNODE_MAX) {
5442 SV * const prop = sv_newmortal();
5443 regnode *rnext = regnext(scan);
5445 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5446 regprop(rex, prop, scan, reginfo, NULL);
5447 Perl_re_printf( aTHX_
5448 "%*s%" IVdf ":%s(%" IVdf ")\n",
5449 INDENT_CHARS(depth), "",
5450 (IV)(scan - rexi->program),
5452 (PL_regkind[OP(scan)] == END || !rnext) ?
5453 0 : (IV)(rnext - rexi->program));
5460 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
5462 switch (state_num) {
5463 case SBOL: /* /^../ and /\A../ */
5464 if (locinput == reginfo->strbeg)
5468 case MBOL: /* /^../m */
5469 if (locinput == reginfo->strbeg ||
5470 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
5477 if (locinput == reginfo->ganch)
5481 case KEEPS: /* \K */
5482 /* update the startpoint */
5483 st->u.keeper.val = rex->offs[0].start;
5484 rex->offs[0].start = locinput - reginfo->strbeg;
5485 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
5486 NOT_REACHED; /* NOTREACHED */
5488 case KEEPS_next_fail:
5489 /* rollback the start point change */
5490 rex->offs[0].start = st->u.keeper.val;
5492 NOT_REACHED; /* NOTREACHED */
5494 case MEOL: /* /..$/m */
5495 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5499 case SEOL: /* /..$/ */
5500 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5502 if (reginfo->strend - locinput > 1)
5507 if (!NEXTCHR_IS_EOS)
5511 case SANY: /* /./s */
5514 goto increment_locinput;
5516 case REG_ANY: /* /./ */
5517 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
5519 goto increment_locinput;
5523 #define ST st->u.trie
5524 case TRIEC: /* (ab|cd) with known charclass */
5525 /* In this case the charclass data is available inline so
5526 we can fail fast without a lot of extra overhead.
5528 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
5530 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n",
5531 depth, PL_colors[4], PL_colors[5])
5534 NOT_REACHED; /* NOTREACHED */
5537 case TRIE: /* (ab|cd) */
5538 /* the basic plan of execution of the trie is:
5539 * At the beginning, run though all the states, and
5540 * find the longest-matching word. Also remember the position
5541 * of the shortest matching word. For example, this pattern:
5544 * when matched against the string "abcde", will generate
5545 * accept states for all words except 3, with the longest
5546 * matching word being 4, and the shortest being 2 (with
5547 * the position being after char 1 of the string).
5549 * Then for each matching word, in word order (i.e. 1,2,4,5),
5550 * we run the remainder of the pattern; on each try setting
5551 * the current position to the character following the word,
5552 * returning to try the next word on failure.
5554 * We avoid having to build a list of words at runtime by
5555 * using a compile-time structure, wordinfo[].prev, which
5556 * gives, for each word, the previous accepting word (if any).
5557 * In the case above it would contain the mappings 1->2, 2->0,
5558 * 3->0, 4->5, 5->1. We can use this table to generate, from
5559 * the longest word (4 above), a list of all words, by
5560 * following the list of prev pointers; this gives us the
5561 * unordered list 4,5,1,2. Then given the current word we have
5562 * just tried, we can go through the list and find the
5563 * next-biggest word to try (so if we just failed on word 2,
5564 * the next in the list is 4).
5566 * Since at runtime we don't record the matching position in
5567 * the string for each word, we have to work that out for
5568 * each word we're about to process. The wordinfo table holds
5569 * the character length of each word; given that we recorded
5570 * at the start: the position of the shortest word and its
5571 * length in chars, we just need to move the pointer the
5572 * difference between the two char lengths. Depending on
5573 * Unicode status and folding, that's cheap or expensive.
5575 * This algorithm is optimised for the case where are only a
5576 * small number of accept states, i.e. 0,1, or maybe 2.
5577 * With lots of accepts states, and having to try all of them,
5578 * it becomes quadratic on number of accept states to find all
5583 /* what type of TRIE am I? (utf8 makes this contextual) */
5584 DECL_TRIE_TYPE(scan);
5586 /* what trie are we using right now */
5587 reg_trie_data * const trie
5588 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
5589 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
5590 U32 state = trie->startstate;
5592 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
5593 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5595 && UTF8_IS_ABOVE_LATIN1(nextchr)
5596 && scan->flags == EXACTL)
5598 /* We only output for EXACTL, as we let the folder
5599 * output this message for EXACTFLU8 to avoid
5601 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
5606 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
5608 if (trie->states[ state ].wordnum) {
5610 Perl_re_exec_indentf( aTHX_ "%smatched empty string...%s\n",
5611 depth, PL_colors[4], PL_colors[5])
5617 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n",
5618 depth, PL_colors[4], PL_colors[5])
5625 U8 *uc = ( U8* )locinput;
5629 U8 *uscan = (U8*)NULL;
5630 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
5631 U32 charcount = 0; /* how many input chars we have matched */
5632 U32 accepted = 0; /* have we seen any accepting states? */
5634 ST.jump = trie->jump;
5637 ST.longfold = FALSE; /* char longer if folded => it's harder */
5640 /* fully traverse the TRIE; note the position of the
5641 shortest accept state and the wordnum of the longest
5644 while ( state && uc <= (U8*)(reginfo->strend) ) {
5645 U32 base = trie->states[ state ].trans.base;
5649 wordnum = trie->states[ state ].wordnum;
5651 if (wordnum) { /* it's an accept state */
5654 /* record first match position */
5656 ST.firstpos = (U8*)locinput;
5661 ST.firstchars = charcount;
5664 if (!ST.nextword || wordnum < ST.nextword)
5665 ST.nextword = wordnum;
5666 ST.topword = wordnum;
5669 DEBUG_TRIE_EXECUTE_r({
5670 DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth );
5672 PerlIO_printf( Perl_debug_log,
5673 "%*s%sState: %4" UVxf " Accepted: %c ",
5674 INDENT_CHARS(depth), "", PL_colors[4],
5675 (UV)state, (accepted ? 'Y' : 'N'));
5678 /* read a char and goto next state */
5679 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
5681 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
5682 uscan, len, uvc, charid, foldlen,
5689 base + charid - 1 - trie->uniquecharcount)) >= 0)
5691 && ((U32)offset < trie->lasttrans)
5692 && trie->trans[offset].check == state)
5694 state = trie->trans[offset].next;
5705 DEBUG_TRIE_EXECUTE_r(
5706 Perl_re_printf( aTHX_
5707 "Charid:%3x CP:%4" UVxf " After State: %4" UVxf "%s\n",
5708 charid, uvc, (UV)state, PL_colors[5] );
5714 /* calculate total number of accept states */
5719 w = trie->wordinfo[w].prev;
5722 ST.accepted = accepted;
5726 Perl_re_exec_indentf( aTHX_ "%sgot %" IVdf " possible matches%s\n",
5728 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
5730 goto trie_first_try; /* jump into the fail handler */
5732 NOT_REACHED; /* NOTREACHED */
5734 case TRIE_next_fail: /* we failed - try next alternative */
5738 REGCP_UNWIND(ST.cp);
5739 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5741 if (!--ST.accepted) {
5743 Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n",
5751 /* Find next-highest word to process. Note that this code
5752 * is O(N^2) per trie run (O(N) per branch), so keep tight */
5755 U16 const nextword = ST.nextword;
5756 reg_trie_wordinfo * const wordinfo
5757 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
5758 for (word=ST.topword; word; word=wordinfo[word].prev) {
5759 if (word > nextword && (!min || word < min))
5772 ST.lastparen = rex->lastparen;
5773 ST.lastcloseparen = rex->lastcloseparen;
5777 /* find start char of end of current word */
5779 U32 chars; /* how many chars to skip */
5780 reg_trie_data * const trie
5781 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
5783 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
5785 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
5790 /* the hard option - fold each char in turn and find
5791 * its folded length (which may be different */
5792 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
5800 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
5808 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
5813 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
5829 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
5830 ? ST.jump[ST.nextword]
5834 Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n",
5842 if ( ST.accepted > 1 || has_cutgroup || ST.jump ) {
5843 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
5844 NOT_REACHED; /* NOTREACHED */
5846 /* only one choice left - just continue */
5848 AV *const trie_words
5849 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
5850 SV ** const tmp = trie_words
5851 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL;
5852 SV *sv= tmp ? sv_newmortal() : NULL;
5854 Perl_re_exec_indentf( aTHX_ "%sonly one match left, short-circuiting: #%d <%s>%s\n",
5855 depth, PL_colors[4],
5857 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
5858 PL_colors[0], PL_colors[1],
5859 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
5861 : "not compiled under -Dr",
5865 locinput = (char*)uc;
5866 continue; /* execute rest of RE */
5871 case EXACTL: /* /abc/l */
5872 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5874 /* Complete checking would involve going through every character
5875 * matched by the string to see if any is above latin1. But the
5876 * comparision otherwise might very well be a fast assembly
5877 * language routine, and I (khw) don't think slowing things down
5878 * just to check for this warning is worth it. So this just checks
5879 * the first character */
5880 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
5881 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
5884 case EXACT: { /* /abc/ */
5885 char *s = STRING(scan);
5887 if (utf8_target != is_utf8_pat) {
5888 /* The target and the pattern have differing utf8ness. */
5890 const char * const e = s + ln;
5893 /* The target is utf8, the pattern is not utf8.
5894 * Above-Latin1 code points can't match the pattern;
5895 * invariants match exactly, and the other Latin1 ones need
5896 * to be downgraded to a single byte in order to do the
5897 * comparison. (If we could be confident that the target
5898 * is not malformed, this could be refactored to have fewer
5899 * tests by just assuming that if the first bytes match, it
5900 * is an invariant, but there are tests in the test suite
5901 * dealing with (??{...}) which violate this) */
5903 if (l >= reginfo->strend
5904 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
5908 if (UTF8_IS_INVARIANT(*(U8*)l)) {
5915 if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
5925 /* The target is not utf8, the pattern is utf8. */
5927 if (l >= reginfo->strend
5928 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
5932 if (UTF8_IS_INVARIANT(*(U8*)s)) {
5939 if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
5951 /* The target and the pattern have the same utf8ness. */
5952 /* Inline the first character, for speed. */
5953 if (reginfo->strend - locinput < ln
5954 || UCHARAT(s) != nextchr
5955 || (ln > 1 && memNE(s, locinput, ln)))
5964 case EXACTFL: { /* /abc/il */
5966 const U8 * fold_array;
5968 U32 fold_utf8_flags;
5970 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5971 folder = foldEQ_locale;
5972 fold_array = PL_fold_locale;
5973 fold_utf8_flags = FOLDEQ_LOCALE;
5976 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
5977 is effectively /u; hence to match, target
5979 if (! utf8_target) {
5982 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S1_ALREADY_FOLDED
5983 | FOLDEQ_S1_FOLDS_SANE;
5984 folder = foldEQ_latin1;
5985 fold_array = PL_fold_latin1;
5988 case EXACTFU_SS: /* /\x{df}/iu */
5989 case EXACTFU: /* /abc/iu */
5990 folder = foldEQ_latin1;
5991 fold_array = PL_fold_latin1;
5992 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
5995 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
5997 assert(! is_utf8_pat);
5999 case EXACTFA: /* /abc/iaa */
6000 folder = foldEQ_latin1;
6001 fold_array = PL_fold_latin1;
6002 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6005 case EXACTF: /* /abc/i This node only generated for
6006 non-utf8 patterns */
6007 assert(! is_utf8_pat);
6009 fold_array = PL_fold;
6010 fold_utf8_flags = 0;
6018 || state_num == EXACTFU_SS
6019 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
6021 /* Either target or the pattern are utf8, or has the issue where
6022 * the fold lengths may differ. */
6023 const char * const l = locinput;
6024 char *e = reginfo->strend;
6026 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
6027 l, &e, 0, utf8_target, fold_utf8_flags))
6035 /* Neither the target nor the pattern are utf8 */
6036 if (UCHARAT(s) != nextchr
6038 && UCHARAT(s) != fold_array[nextchr])
6042 if (reginfo->strend - locinput < ln)
6044 if (ln > 1 && ! folder(s, locinput, ln))
6050 case NBOUNDL: /* /\B/l */
6054 case BOUNDL: /* /\b/l */
6057 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6059 if (FLAGS(scan) != TRADITIONAL_BOUND) {
6060 if (! IN_UTF8_CTYPE_LOCALE) {
6061 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
6062 B_ON_NON_UTF8_LOCALE_IS_WRONG);
6068 if (locinput == reginfo->strbeg)
6069 b1 = isWORDCHAR_LC('\n');
6071 b1 = isWORDCHAR_LC_utf8_safe(reghop3((U8*)locinput, -1,
6072 (U8*)(reginfo->strbeg)),
6073 (U8*)(reginfo->strend));
6075 b2 = (NEXTCHR_IS_EOS)
6076 ? isWORDCHAR_LC('\n')
6077 : isWORDCHAR_LC_utf8_safe((U8*) locinput,
6078 (U8*) reginfo->strend);
6080 else { /* Here the string isn't utf8 */
6081 b1 = (locinput == reginfo->strbeg)
6082 ? isWORDCHAR_LC('\n')
6083 : isWORDCHAR_LC(UCHARAT(locinput - 1));
6084 b2 = (NEXTCHR_IS_EOS)
6085 ? isWORDCHAR_LC('\n')
6086 : isWORDCHAR_LC(nextchr);
6088 if (to_complement ^ (b1 == b2)) {
6094 case NBOUND: /* /\B/ */
6098 case BOUND: /* /\b/ */
6102 goto bound_ascii_match_only;
6104 case NBOUNDA: /* /\B/a */
6108 case BOUNDA: /* /\b/a */
6112 bound_ascii_match_only:
6113 /* Here the string isn't utf8, or is utf8 and only ascii characters
6114 * are to match \w. In the latter case looking at the byte just
6115 * prior to the current one may be just the final byte of a
6116 * multi-byte character. This is ok. There are two cases:
6117 * 1) it is a single byte character, and then the test is doing
6118 * just what it's supposed to.
6119 * 2) it is a multi-byte character, in which case the final byte is
6120 * never mistakable for ASCII, and so the test will say it is
6121 * not a word character, which is the correct answer. */
6122 b1 = (locinput == reginfo->strbeg)
6123 ? isWORDCHAR_A('\n')
6124 : isWORDCHAR_A(UCHARAT(locinput - 1));
6125 b2 = (NEXTCHR_IS_EOS)
6126 ? isWORDCHAR_A('\n')
6127 : isWORDCHAR_A(nextchr);
6128 if (to_complement ^ (b1 == b2)) {
6134 case NBOUNDU: /* /\B/u */
6138 case BOUNDU: /* /\b/u */
6141 if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) {
6144 else if (utf8_target) {
6146 switch((bound_type) FLAGS(scan)) {
6147 case TRADITIONAL_BOUND:
6150 b1 = (locinput == reginfo->strbeg)
6151 ? 0 /* isWORDCHAR_L1('\n') */
6152 : isWORDCHAR_utf8_safe(
6153 reghop3((U8*)locinput,
6155 (U8*)(reginfo->strbeg)),
6156 (U8*) reginfo->strend);
6157 b2 = (NEXTCHR_IS_EOS)
6158 ? 0 /* isWORDCHAR_L1('\n') */
6159 : isWORDCHAR_utf8_safe((U8*)locinput,
6160 (U8*) reginfo->strend);
6161 match = cBOOL(b1 != b2);
6165 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6166 match = TRUE; /* GCB always matches at begin and
6170 /* Find the gcb values of previous and current
6171 * chars, then see if is a break point */
6172 match = isGCB(getGCB_VAL_UTF8(
6173 reghop3((U8*)locinput,
6175 (U8*)(reginfo->strbeg)),
6176 (U8*) reginfo->strend),
6177 getGCB_VAL_UTF8((U8*) locinput,
6178 (U8*) reginfo->strend),
6179 (U8*) reginfo->strbeg,
6186 if (locinput == reginfo->strbeg) {
6189 else if (NEXTCHR_IS_EOS) {
6193 match = isLB(getLB_VAL_UTF8(
6194 reghop3((U8*)locinput,
6196 (U8*)(reginfo->strbeg)),
6197 (U8*) reginfo->strend),
6198 getLB_VAL_UTF8((U8*) locinput,
6199 (U8*) reginfo->strend),
6200 (U8*) reginfo->strbeg,
6202 (U8*) reginfo->strend,
6207 case SB_BOUND: /* Always matches at begin and end */
6208 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6212 match = isSB(getSB_VAL_UTF8(
6213 reghop3((U8*)locinput,
6215 (U8*)(reginfo->strbeg)),
6216 (U8*) reginfo->strend),
6217 getSB_VAL_UTF8((U8*) locinput,
6218 (U8*) reginfo->strend),
6219 (U8*) reginfo->strbeg,
6221 (U8*) reginfo->strend,
6227 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6231 match = isWB(WB_UNKNOWN,
6233 reghop3((U8*)locinput,
6235 (U8*)(reginfo->strbeg)),
6236 (U8*) reginfo->strend),
6237 getWB_VAL_UTF8((U8*) locinput,
6238 (U8*) reginfo->strend),
6239 (U8*) reginfo->strbeg,
6241 (U8*) reginfo->strend,
6247 else { /* Not utf8 target */
6248 switch((bound_type) FLAGS(scan)) {
6249 case TRADITIONAL_BOUND:
6252 b1 = (locinput == reginfo->strbeg)
6253 ? 0 /* isWORDCHAR_L1('\n') */
6254 : isWORDCHAR_L1(UCHARAT(locinput - 1));
6255 b2 = (NEXTCHR_IS_EOS)
6256 ? 0 /* isWORDCHAR_L1('\n') */
6257 : isWORDCHAR_L1(nextchr);
6258 match = cBOOL(b1 != b2);
6263 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6264 match = TRUE; /* GCB always matches at begin and
6267 else { /* Only CR-LF combo isn't a GCB in 0-255
6269 match = UCHARAT(locinput - 1) != '\r'
6270 || UCHARAT(locinput) != '\n';
6275 if (locinput == reginfo->strbeg) {
6278 else if (NEXTCHR_IS_EOS) {
6282 match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)),
6283 getLB_VAL_CP(UCHARAT(locinput)),
6284 (U8*) reginfo->strbeg,
6286 (U8*) reginfo->strend,
6291 case SB_BOUND: /* Always matches at begin and end */
6292 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6296 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)),
6297 getSB_VAL_CP(UCHARAT(locinput)),
6298 (U8*) reginfo->strbeg,
6300 (U8*) reginfo->strend,
6306 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6310 match = isWB(WB_UNKNOWN,
6311 getWB_VAL_CP(UCHARAT(locinput -1)),
6312 getWB_VAL_CP(UCHARAT(locinput)),
6313 (U8*) reginfo->strbeg,
6315 (U8*) reginfo->strend,
6322 if (to_complement ^ ! match) {
6327 case ANYOFL: /* /[abc]/l */
6328 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6330 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE)
6332 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
6335 case ANYOFD: /* /[abc]/d */
6336 case ANYOF: /* /[abc]/ */
6339 if (utf8_target && ! UTF8_IS_INVARIANT(*locinput)) {
6340 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
6343 locinput += UTF8SKIP(locinput);
6346 if (!REGINCLASS(rex, scan, (U8*)locinput, utf8_target))
6352 /* The argument (FLAGS) to all the POSIX node types is the class number
6355 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
6359 case POSIXL: /* \w or [:punct:] etc. under /l */
6360 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6364 /* Use isFOO_lc() for characters within Latin1. (Note that
6365 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6366 * wouldn't be invariant) */
6367 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6368 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
6376 if (! UTF8_IS_DOWNGRADEABLE_START(nextchr)) { /* An above Latin-1 code point */
6377 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
6378 goto utf8_posix_above_latin1;
6381 /* Here is a UTF-8 variant code point below 256 and the target is
6383 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
6384 EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6385 *(locinput + 1))))))
6390 goto increment_locinput;
6392 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
6396 case POSIXD: /* \w or [:punct:] etc. under /d */
6402 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
6404 if (NEXTCHR_IS_EOS) {
6408 /* All UTF-8 variants match */
6409 if (! UTF8_IS_INVARIANT(nextchr)) {
6410 goto increment_locinput;
6416 case POSIXA: /* \w or [:punct:] etc. under /a */
6419 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
6420 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
6421 * character is a single byte */
6423 if (NEXTCHR_IS_EOS) {
6429 if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
6435 /* Here we are either not in utf8, or we matched a utf8-invariant,
6436 * so the next char is the next byte */
6440 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
6444 case POSIXU: /* \w or [:punct:] etc. under /u */
6446 if (NEXTCHR_IS_EOS) {
6450 /* Use _generic_isCC() for characters within Latin1. (Note that
6451 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6452 * wouldn't be invariant) */
6453 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6454 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
6461 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
6462 if (! (to_complement
6463 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6471 else { /* Handle above Latin-1 code points */
6472 utf8_posix_above_latin1:
6473 classnum = (_char_class_number) FLAGS(scan);
6474 if (classnum < _FIRST_NON_SWASH_CC) {
6476 /* Here, uses a swash to find such code points. Load if if
6477 * not done already */
6478 if (! PL_utf8_swash_ptrs[classnum]) {
6479 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
6480 PL_utf8_swash_ptrs[classnum]
6481 = _core_swash_init("utf8",
6484 PL_XPosix_ptrs[classnum], &flags);
6486 if (! (to_complement
6487 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
6488 (U8 *) locinput, TRUE))))
6493 else { /* Here, uses macros to find above Latin-1 code points */
6495 case _CC_ENUM_SPACE:
6496 if (! (to_complement
6497 ^ cBOOL(is_XPERLSPACE_high(locinput))))
6502 case _CC_ENUM_BLANK:
6503 if (! (to_complement
6504 ^ cBOOL(is_HORIZWS_high(locinput))))
6509 case _CC_ENUM_XDIGIT:
6510 if (! (to_complement
6511 ^ cBOOL(is_XDIGIT_high(locinput))))
6516 case _CC_ENUM_VERTSPACE:
6517 if (! (to_complement
6518 ^ cBOOL(is_VERTWS_high(locinput))))
6523 default: /* The rest, e.g. [:cntrl:], can't match
6525 if (! to_complement) {
6531 locinput += UTF8SKIP(locinput);
6535 case CLUMP: /* Match \X: logical Unicode character. This is defined as
6536 a Unicode extended Grapheme Cluster */
6539 if (! utf8_target) {
6541 /* Match either CR LF or '.', as all the other possibilities
6543 locinput++; /* Match the . or CR */
6544 if (nextchr == '\r' /* And if it was CR, and the next is LF,
6546 && locinput < reginfo->strend
6547 && UCHARAT(locinput) == '\n')
6554 /* Get the gcb type for the current character */
6555 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput,
6556 (U8*) reginfo->strend);
6558 /* Then scan through the input until we get to the first
6559 * character whose type is supposed to be a gcb with the
6560 * current character. (There is always a break at the
6562 locinput += UTF8SKIP(locinput);
6563 while (locinput < reginfo->strend) {
6564 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput,
6565 (U8*) reginfo->strend);
6566 if (isGCB(prev_gcb, cur_gcb,
6567 (U8*) reginfo->strbeg, (U8*) locinput,
6574 locinput += UTF8SKIP(locinput);
6581 case NREFFL: /* /\g{name}/il */
6582 { /* The capture buffer cases. The ones beginning with N for the
6583 named buffers just convert to the equivalent numbered and
6584 pretend they were called as the corresponding numbered buffer
6586 /* don't initialize these in the declaration, it makes C++
6591 const U8 *fold_array;
6594 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6595 folder = foldEQ_locale;
6596 fold_array = PL_fold_locale;
6598 utf8_fold_flags = FOLDEQ_LOCALE;
6601 case NREFFA: /* /\g{name}/iaa */
6602 folder = foldEQ_latin1;
6603 fold_array = PL_fold_latin1;
6605 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6608 case NREFFU: /* /\g{name}/iu */
6609 folder = foldEQ_latin1;
6610 fold_array = PL_fold_latin1;
6612 utf8_fold_flags = 0;
6615 case NREFF: /* /\g{name}/i */
6617 fold_array = PL_fold;
6619 utf8_fold_flags = 0;
6622 case NREF: /* /\g{name}/ */
6626 utf8_fold_flags = 0;
6629 /* For the named back references, find the corresponding buffer
6631 n = reg_check_named_buff_matched(rex,scan);
6636 goto do_nref_ref_common;
6638 case REFFL: /* /\1/il */
6639 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6640 folder = foldEQ_locale;
6641 fold_array = PL_fold_locale;
6642 utf8_fold_flags = FOLDEQ_LOCALE;
6645 case REFFA: /* /\1/iaa */
6646 folder = foldEQ_latin1;
6647 fold_array = PL_fold_latin1;
6648 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6651 case REFFU: /* /\1/iu */
6652 folder = foldEQ_latin1;
6653 fold_array = PL_fold_latin1;
6654 utf8_fold_flags = 0;
6657 case REFF: /* /\1/i */
6659 fold_array = PL_fold;
6660 utf8_fold_flags = 0;
6663 case REF: /* /\1/ */
6666 utf8_fold_flags = 0;
6670 n = ARG(scan); /* which paren pair */
6673 ln = rex->offs[n].start;
6674 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
6675 if (rex->lastparen < n || ln == -1)
6676 sayNO; /* Do not match unless seen CLOSEn. */
6677 if (ln == rex->offs[n].end)
6680 s = reginfo->strbeg + ln;
6681 if (type != REF /* REF can do byte comparison */
6682 && (utf8_target || type == REFFU || type == REFFL))
6684 char * limit = reginfo->strend;
6686 /* This call case insensitively compares the entire buffer
6687 * at s, with the current input starting at locinput, but
6688 * not going off the end given by reginfo->strend, and
6689 * returns in <limit> upon success, how much of the
6690 * current input was matched */
6691 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
6692 locinput, &limit, 0, utf8_target, utf8_fold_flags))
6700 /* Not utf8: Inline the first character, for speed. */
6701 if (!NEXTCHR_IS_EOS &&
6702 UCHARAT(s) != nextchr &&
6704 UCHARAT(s) != fold_array[nextchr]))
6706 ln = rex->offs[n].end - ln;
6707 if (locinput + ln > reginfo->strend)
6709 if (ln > 1 && (type == REF
6710 ? memNE(s, locinput, ln)
6711 : ! folder(s, locinput, ln)))
6717 case NOTHING: /* null op; e.g. the 'nothing' following
6718 * the '*' in m{(a+|b)*}' */
6720 case TAIL: /* placeholder while compiling (A|B|C) */
6724 #define ST st->u.eval
6725 #define CUR_EVAL cur_eval->u.eval
6731 regexp_internal *rei;
6732 regnode *startpoint;
6735 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
6736 arg= (U32)ARG(scan);
6737 if (cur_eval && cur_eval->locinput == locinput) {
6738 if ( ++nochange_depth > max_nochange_depth )
6740 "Pattern subroutine nesting without pos change"
6741 " exceeded limit in regex");
6748 startpoint = scan + ARG2L(scan);
6749 EVAL_CLOSE_PAREN_SET( st, arg );
6750 /* Detect infinite recursion
6752 * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/
6753 * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever.
6754 * So we track the position in the string we are at each time
6755 * we recurse and if we try to enter the same routine twice from
6756 * the same position we throw an error.
6758 if ( rex->recurse_locinput[arg] == locinput ) {
6759 /* FIXME: we should show the regop that is failing as part
6760 * of the error message. */
6761 Perl_croak(aTHX_ "Infinite recursion in regex");
6763 ST.prev_recurse_locinput= rex->recurse_locinput[arg];
6764 rex->recurse_locinput[arg]= locinput;
6767 GET_RE_DEBUG_FLAGS_DECL;
6769 Perl_re_exec_indentf( aTHX_
6770 "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n",
6771 depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg]
6777 /* Save all the positions seen so far. */
6778 ST.cp = regcppush(rex, 0, maxopenparen);
6779 REGCP_SET(ST.lastcp);
6781 /* and then jump to the code we share with EVAL */
6782 goto eval_recurse_doit;
6785 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
6786 if (cur_eval && cur_eval->locinput==locinput) {
6787 if ( ++nochange_depth > max_nochange_depth )
6788 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
6793 /* execute the code in the {...} */
6797 OP * const oop = PL_op;
6798 COP * const ocurcop = PL_curcop;
6802 /* save *all* paren positions */
6803 regcppush(rex, 0, maxopenparen);
6804 REGCP_SET(runops_cp);
6807 caller_cv = find_runcv(NULL);
6811 if (rexi->data->what[n] == 'r') { /* code from an external qr */
6813 (REGEXP*)(rexi->data->data[n])
6815 nop = (OP*)rexi->data->data[n+1];
6817 else if (rexi->data->what[n] == 'l') { /* literal code */
6819 nop = (OP*)rexi->data->data[n];
6820 assert(CvDEPTH(newcv));
6823 /* literal with own CV */
6824 assert(rexi->data->what[n] == 'L');
6825 newcv = rex->qr_anoncv;
6826 nop = (OP*)rexi->data->data[n];
6829 /* normally if we're about to execute code from the same
6830 * CV that we used previously, we just use the existing
6831 * CX stack entry. However, its possible that in the
6832 * meantime we may have backtracked, popped from the save
6833 * stack, and undone the SAVECOMPPAD(s) associated with
6834 * PUSH_MULTICALL; in which case PL_comppad no longer
6835 * points to newcv's pad. */
6836 if (newcv != last_pushed_cv || PL_comppad != last_pad)
6838 U8 flags = (CXp_SUB_RE |
6839 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
6840 if (last_pushed_cv) {
6841 /* PUSH/POP_MULTICALL save and restore the
6842 * caller's PL_comppad; if we call multiple subs
6843 * using the same CX block, we have to save and
6844 * unwind the varying PL_comppad's ourselves,
6845 * especially restoring the right PL_comppad on
6846 * backtrack - so save it on the save stack */
6848 CHANGE_MULTICALL_FLAGS(newcv, flags);
6851 PUSH_MULTICALL_FLAGS(newcv, flags);
6853 last_pushed_cv = newcv;
6856 /* these assignments are just to silence compiler
6858 multicall_cop = NULL;
6860 last_pad = PL_comppad;
6862 /* the initial nextstate you would normally execute
6863 * at the start of an eval (which would cause error
6864 * messages to come from the eval), may be optimised
6865 * away from the execution path in the regex code blocks;
6866 * so manually set PL_curcop to it initially */
6868 OP *o = cUNOPx(nop)->op_first;
6869 assert(o->op_type == OP_NULL);
6870 if (o->op_targ == OP_SCOPE) {
6871 o = cUNOPo->op_first;
6874 assert(o->op_targ == OP_LEAVE);
6875 o = cUNOPo->op_first;
6876 assert(o->op_type == OP_ENTER);
6880 if (o->op_type != OP_STUB) {
6881 assert( o->op_type == OP_NEXTSTATE
6882 || o->op_type == OP_DBSTATE
6883 || (o->op_type == OP_NULL
6884 && ( o->op_targ == OP_NEXTSTATE
6885 || o->op_targ == OP_DBSTATE
6889 PL_curcop = (COP*)o;
6894 DEBUG_STATE_r( Perl_re_printf( aTHX_
6895 " re EVAL PL_op=0x%" UVxf "\n", PTR2UV(nop)) );
6897 rex->offs[0].end = locinput - reginfo->strbeg;
6898 if (reginfo->info_aux_eval->pos_magic)
6899 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
6900 reginfo->sv, reginfo->strbeg,
6901 locinput - reginfo->strbeg);
6904 SV *sv_mrk = get_sv("REGMARK", 1);
6905 sv_setsv(sv_mrk, sv_yes_mark);
6908 /* we don't use MULTICALL here as we want to call the
6909 * first op of the block of interest, rather than the
6910 * first op of the sub. Also, we don't want to free
6911 * the savestack frame */
6912 before = (IV)(SP-PL_stack_base);
6914 CALLRUNOPS(aTHX); /* Scalar context. */
6916 if ((IV)(SP-PL_stack_base) == before)
6917 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
6923 /* before restoring everything, evaluate the returned
6924 * value, so that 'uninit' warnings don't use the wrong
6925 * PL_op or pad. Also need to process any magic vars
6926 * (e.g. $1) *before* parentheses are restored */
6931 if (logical == 0) /* (?{})/ */
6932 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
6933 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
6934 sw = cBOOL(SvTRUE(ret));
6937 else { /* /(??{}) */
6938 /* if its overloaded, let the regex compiler handle
6939 * it; otherwise extract regex, or stringify */
6940 if (SvGMAGICAL(ret))
6941 ret = sv_mortalcopy(ret);
6942 if (!SvAMAGIC(ret)) {
6946 if (SvTYPE(sv) == SVt_REGEXP)
6947 re_sv = (REGEXP*) sv;
6948 else if (SvSMAGICAL(ret)) {
6949 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
6951 re_sv = (REGEXP *) mg->mg_obj;
6954 /* force any undef warnings here */
6955 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
6956 ret = sv_mortalcopy(ret);
6957 (void) SvPV_force_nolen(ret);
6963 /* *** Note that at this point we don't restore
6964 * PL_comppad, (or pop the CxSUB) on the assumption it may
6965 * be used again soon. This is safe as long as nothing
6966 * in the regexp code uses the pad ! */
6968 PL_curcop = ocurcop;
6969 regcp_restore(rex, runops_cp, &maxopenparen);
6970 PL_curpm_under = PL_curpm;
6971 PL_curpm = PL_reg_curpm;
6977 /* only /(??{})/ from now on */
6980 /* extract RE object from returned value; compiling if
6984 re_sv = reg_temp_copy(NULL, re_sv);
6989 if (SvUTF8(ret) && IN_BYTES) {
6990 /* In use 'bytes': make a copy of the octet
6991 * sequence, but without the flag on */
6993 const char *const p = SvPV(ret, len);
6994 ret = newSVpvn_flags(p, len, SVs_TEMP);
6996 if (rex->intflags & PREGf_USE_RE_EVAL)
6997 pm_flags |= PMf_USE_RE_EVAL;
6999 /* if we got here, it should be an engine which
7000 * supports compiling code blocks and stuff */
7001 assert(rex->engine && rex->engine->op_comp);
7002 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
7003 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
7004 rex->engine, NULL, NULL,
7005 /* copy /msixn etc to inner pattern */
7010 & (SVs_TEMP | SVs_GMG | SVf_ROK))
7011 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
7012 /* This isn't a first class regexp. Instead, it's
7013 caching a regexp onto an existing, Perl visible
7015 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
7021 RXp_MATCH_COPIED_off(re);
7022 re->subbeg = rex->subbeg;
7023 re->sublen = rex->sublen;
7024 re->suboffset = rex->suboffset;
7025 re->subcoffset = rex->subcoffset;
7027 re->lastcloseparen = 0;
7030 debug_start_match(re_sv, utf8_target, locinput,
7031 reginfo->strend, "Matching embedded");
7033 startpoint = rei->program + 1;
7034 EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0;
7035 * close_paren only for GOSUB */
7036 ST.prev_recurse_locinput= NULL; /* only used for GOSUB */
7037 /* Save all the seen positions so far. */
7038 ST.cp = regcppush(rex, 0, maxopenparen);
7039 REGCP_SET(ST.lastcp);
7040 /* and set maxopenparen to 0, since we are starting a "fresh" match */
7042 /* run the pattern returned from (??{...}) */
7044 eval_recurse_doit: /* Share code with GOSUB below this line
7045 * At this point we expect the stack context to be
7046 * set up correctly */
7048 /* invalidate the S-L poscache. We're now executing a
7049 * different set of WHILEM ops (and their associated
7050 * indexes) against the same string, so the bits in the
7051 * cache are meaningless. Setting maxiter to zero forces
7052 * the cache to be invalidated and zeroed before reuse.
7053 * XXX This is too dramatic a measure. Ideally we should
7054 * save the old cache and restore when running the outer
7056 reginfo->poscache_maxiter = 0;
7058 /* the new regexp might have a different is_utf8_pat than we do */
7059 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
7061 ST.prev_rex = rex_sv;
7062 ST.prev_curlyx = cur_curlyx;
7064 SET_reg_curpm(rex_sv);
7069 ST.prev_eval = cur_eval;
7071 /* now continue from first node in postoned RE */
7072 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
7073 NOT_REACHED; /* NOTREACHED */
7076 case EVAL_AB: /* cleanup after a successful (??{A})B */
7077 /* note: this is called twice; first after popping B, then A */
7079 Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n",
7080 depth, cur_eval, ST.prev_eval);
7083 #define SET_RECURSE_LOCINPUT(STR,VAL)\
7084 if ( cur_eval && CUR_EVAL.close_paren ) {\
7086 Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\
7088 CUR_EVAL.close_paren - 1,\
7092 rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\
7095 SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput);
7097 rex_sv = ST.prev_rex;
7098 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7099 SET_reg_curpm(rex_sv);
7100 rex = ReANY(rex_sv);
7101 rexi = RXi_GET(rex);
7103 /* preserve $^R across LEAVE's. See Bug 121070. */
7104 SV *save_sv= GvSV(PL_replgv);
7105 SvREFCNT_inc(save_sv);
7106 regcpblow(ST.cp); /* LEAVE in disguise */
7107 sv_setsv(GvSV(PL_replgv), save_sv);
7108 SvREFCNT_dec(save_sv);
7110 cur_eval = ST.prev_eval;
7111 cur_curlyx = ST.prev_curlyx;
7113 /* Invalidate cache. See "invalidate" comment above. */
7114 reginfo->poscache_maxiter = 0;
7115 if ( nochange_depth )
7118 SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput);
7122 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
7123 /* note: this is called twice; first after popping B, then A */
7125 Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n",
7126 depth, cur_eval, ST.prev_eval);
7129 SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput);
7131 rex_sv = ST.prev_rex;
7132 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7133 SET_reg_curpm(rex_sv);
7134 rex = ReANY(rex_sv);
7135 rexi = RXi_GET(rex);
7137 REGCP_UNWIND(ST.lastcp);
7138 regcppop(rex, &maxopenparen);
7139 cur_eval = ST.prev_eval;
7140 cur_curlyx = ST.prev_curlyx;
7142 /* Invalidate cache. See "invalidate" comment above. */
7143 reginfo->poscache_maxiter = 0;
7144 if ( nochange_depth )
7147 SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput);
7152 n = ARG(scan); /* which paren pair */
7153 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
7154 if (n > maxopenparen)
7156 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
7157 "rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf " tmp; maxopenparen=%" UVuf "\n",
7162 (IV)rex->offs[n].start_tmp,
7168 /* XXX really need to log other places start/end are set too */
7169 #define CLOSE_CAPTURE \
7170 rex->offs[n].start = rex->offs[n].start_tmp; \
7171 rex->offs[n].end = locinput - reginfo->strbeg; \
7172 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
7173 "rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf "..%" IVdf "\n", \
7176 PTR2UV(rex->offs), \
7178 (IV)rex->offs[n].start, \
7179 (IV)rex->offs[n].end \
7183 n = ARG(scan); /* which paren pair */
7185 if (n > rex->lastparen)
7187 rex->lastcloseparen = n;
7188 if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) )
7193 case ACCEPT: /* (*ACCEPT) */
7195 sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7199 cursor && OP(cursor)!=END;
7200 cursor=regnext(cursor))
7202 if ( OP(cursor)==CLOSE ){
7204 if ( n <= lastopen ) {
7206 if (n > rex->lastparen)
7208 rex->lastcloseparen = n;
7209 if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) )
7218 case GROUPP: /* (?(1)) */
7219 n = ARG(scan); /* which paren pair */
7220 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
7223 case NGROUPP: /* (?(<name>)) */
7224 /* reg_check_named_buff_matched returns 0 for no match */
7225 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
7228 case INSUBP: /* (?(R)) */
7230 /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg
7231 * of SCAN is already set up as matches a eval.close_paren */
7232 sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n);
7235 case DEFINEP: /* (?(DEFINE)) */
7239 case IFTHEN: /* (?(cond)A|B) */
7240 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7242 next = NEXTOPER(NEXTOPER(scan));
7244 next = scan + ARG(scan);
7245 if (OP(next) == IFTHEN) /* Fake one. */
7246 next = NEXTOPER(NEXTOPER(next));
7250 case LOGICAL: /* modifier for EVAL and IFMATCH */
7251 logical = scan->flags;
7254 /*******************************************************************
7256 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
7257 pattern, where A and B are subpatterns. (For simple A, CURLYM or
7258 STAR/PLUS/CURLY/CURLYN are used instead.)
7260 A*B is compiled as <CURLYX><A><WHILEM><B>
7262 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
7263 state, which contains the current count, initialised to -1. It also sets
7264 cur_curlyx to point to this state, with any previous value saved in the
7267 CURLYX then jumps straight to the WHILEM op, rather than executing A,
7268 since the pattern may possibly match zero times (i.e. it's a while {} loop
7269 rather than a do {} while loop).
7271 Each entry to WHILEM represents a successful match of A. The count in the
7272 CURLYX block is incremented, another WHILEM state is pushed, and execution
7273 passes to A or B depending on greediness and the current count.
7275 For example, if matching against the string a1a2a3b (where the aN are
7276 substrings that match /A/), then the match progresses as follows: (the
7277 pushed states are interspersed with the bits of strings matched so far):
7280 <CURLYX cnt=0><WHILEM>
7281 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
7282 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
7283 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
7284 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
7286 (Contrast this with something like CURLYM, which maintains only a single
7290 a1 <CURLYM cnt=1> a2
7291 a1 a2 <CURLYM cnt=2> a3
7292 a1 a2 a3 <CURLYM cnt=3> b
7295 Each WHILEM state block marks a point to backtrack to upon partial failure
7296 of A or B, and also contains some minor state data related to that
7297 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
7298 overall state, such as the count, and pointers to the A and B ops.
7300 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
7301 must always point to the *current* CURLYX block, the rules are:
7303 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
7304 and set cur_curlyx to point the new block.
7306 When popping the CURLYX block after a successful or unsuccessful match,
7307 restore the previous cur_curlyx.
7309 When WHILEM is about to execute B, save the current cur_curlyx, and set it
7310 to the outer one saved in the CURLYX block.
7312 When popping the WHILEM block after a successful or unsuccessful B match,
7313 restore the previous cur_curlyx.
7315 Here's an example for the pattern (AI* BI)*BO
7316 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
7319 curlyx backtrack stack
7320 ------ ---------------
7322 CO <CO prev=NULL> <WO>
7323 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7324 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7325 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
7327 At this point the pattern succeeds, and we work back down the stack to
7328 clean up, restoring as we go:
7330 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7331 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7332 CO <CO prev=NULL> <WO>
7335 *******************************************************************/
7337 #define ST st->u.curlyx
7339 case CURLYX: /* start of /A*B/ (for complex A) */
7341 /* No need to save/restore up to this paren */
7342 I32 parenfloor = scan->flags;
7344 assert(next); /* keep Coverity happy */
7345 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
7348 /* XXXX Probably it is better to teach regpush to support
7349 parenfloor > maxopenparen ... */
7350 if (parenfloor > (I32)rex->lastparen)
7351 parenfloor = rex->lastparen; /* Pessimization... */
7353 ST.prev_curlyx= cur_curlyx;
7355 ST.cp = PL_savestack_ix;
7357 /* these fields contain the state of the current curly.
7358 * they are accessed by subsequent WHILEMs */
7359 ST.parenfloor = parenfloor;
7364 ST.count = -1; /* this will be updated by WHILEM */
7365 ST.lastloc = NULL; /* this will be updated by WHILEM */
7367 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
7368 NOT_REACHED; /* NOTREACHED */
7371 case CURLYX_end: /* just finished matching all of A*B */
7372 cur_curlyx = ST.prev_curlyx;
7374 NOT_REACHED; /* NOTREACHED */
7376 case CURLYX_end_fail: /* just failed to match all of A*B */
7378 cur_curlyx = ST.prev_curlyx;
7380 NOT_REACHED; /* NOTREACHED */
7384 #define ST st->u.whilem
7386 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
7388 /* see the discussion above about CURLYX/WHILEM */
7393 assert(cur_curlyx); /* keep Coverity happy */
7395 min = ARG1(cur_curlyx->u.curlyx.me);
7396 max = ARG2(cur_curlyx->u.curlyx.me);
7397 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
7398 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
7399 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
7400 ST.cache_offset = 0;
7404 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: matched %ld out of %d..%d\n",
7405 depth, (long)n, min, max)
7408 /* First just match a string of min A's. */
7411 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, maxopenparen);
7412 cur_curlyx->u.curlyx.lastloc = locinput;
7413 REGCP_SET(ST.lastcp);
7415 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
7416 NOT_REACHED; /* NOTREACHED */
7419 /* If degenerate A matches "", assume A done. */
7421 if (locinput == cur_curlyx->u.curlyx.lastloc) {
7422 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: empty match detected, trying continuation...\n",
7425 goto do_whilem_B_max;
7428 /* super-linear cache processing.
7430 * The idea here is that for certain types of CURLYX/WHILEM -
7431 * principally those whose upper bound is infinity (and
7432 * excluding regexes that have things like \1 and other very
7433 * non-regular expresssiony things), then if a pattern like
7434 * /....A*.../ fails and we backtrack to the WHILEM, then we
7435 * make a note that this particular WHILEM op was at string
7436 * position 47 (say) when the rest of pattern failed. Then, if
7437 * we ever find ourselves back at that WHILEM, and at string
7438 * position 47 again, we can just fail immediately rather than
7439 * running the rest of the pattern again.
7441 * This is very handy when patterns start to go
7442 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
7443 * with a combinatorial explosion of backtracking.
7445 * The cache is implemented as a bit array, with one bit per
7446 * string byte position per WHILEM op (up to 16) - so its
7447 * between 0.25 and 2x the string size.
7449 * To avoid allocating a poscache buffer every time, we do an
7450 * initially countdown; only after we have executed a WHILEM
7451 * op (string-length x #WHILEMs) times do we allocate the
7454 * The top 4 bits of scan->flags byte say how many different
7455 * relevant CURLLYX/WHILEM op pairs there are, while the
7456 * bottom 4-bits is the identifying index number of this
7462 if (!reginfo->poscache_maxiter) {
7463 /* start the countdown: Postpone detection until we
7464 * know the match is not *that* much linear. */
7465 reginfo->poscache_maxiter
7466 = (reginfo->strend - reginfo->strbeg + 1)
7468 /* possible overflow for long strings and many CURLYX's */
7469 if (reginfo->poscache_maxiter < 0)
7470 reginfo->poscache_maxiter = I32_MAX;
7471 reginfo->poscache_iter = reginfo->poscache_maxiter;
7474 if (reginfo->poscache_iter-- == 0) {
7475 /* initialise cache */
7476 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
7477 regmatch_info_aux *const aux = reginfo->info_aux;
7478 if (aux->poscache) {
7479 if ((SSize_t)reginfo->poscache_size < size) {
7480 Renew(aux->poscache, size, char);
7481 reginfo->poscache_size = size;
7483 Zero(aux->poscache, size, char);
7486 reginfo->poscache_size = size;
7487 Newxz(aux->poscache, size, char);
7489 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
7490 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
7491 PL_colors[4], PL_colors[5])
7495 if (reginfo->poscache_iter < 0) {
7496 /* have we already failed at this position? */
7497 SSize_t offset, mask;
7499 reginfo->poscache_iter = -1; /* stop eventual underflow */
7500 offset = (scan->flags & 0xf) - 1
7501 + (locinput - reginfo->strbeg)
7503 mask = 1 << (offset % 8);
7505 if (reginfo->info_aux->poscache[offset] & mask) {
7506 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: (cache) already tried at this position...\n",
7509 cur_curlyx->u.curlyx.count--;
7510 sayNO; /* cache records failure */
7512 ST.cache_offset = offset;
7513 ST.cache_mask = mask;
7517 /* Prefer B over A for minimal matching. */
7519 if (cur_curlyx->u.curlyx.minmod) {
7520 ST.save_curlyx = cur_curlyx;
7521 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
7522 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
7524 REGCP_SET(ST.lastcp);
7525 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
7527 NOT_REACHED; /* NOTREACHED */
7530 /* Prefer A over B for maximal matching. */
7532 if (n < max) { /* More greed allowed? */
7533 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
7535 cur_curlyx->u.curlyx.lastloc = locinput;
7536 REGCP_SET(ST.lastcp);
7537 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
7538 NOT_REACHED; /* NOTREACHED */
7540 goto do_whilem_B_max;
7542 NOT_REACHED; /* NOTREACHED */
7544 case WHILEM_B_min: /* just matched B in a minimal match */
7545 case WHILEM_B_max: /* just matched B in a maximal match */
7546 cur_curlyx = ST.save_curlyx;
7548 NOT_REACHED; /* NOTREACHED */
7550 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
7551 cur_curlyx = ST.save_curlyx;
7552 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7553 cur_curlyx->u.curlyx.count--;
7555 NOT_REACHED; /* NOTREACHED */
7557 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
7559 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
7560 REGCP_UNWIND(ST.lastcp);
7561 regcppop(rex, &maxopenparen);
7562 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7563 cur_curlyx->u.curlyx.count--;
7565 NOT_REACHED; /* NOTREACHED */
7567 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
7568 REGCP_UNWIND(ST.lastcp);
7569 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
7570 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "whilem: failed, trying continuation...\n",
7574 if (cur_curlyx->u.curlyx.count >= REG_INFTY
7575 && ckWARN(WARN_REGEXP)
7576 && !reginfo->warned)
7578 reginfo->warned = TRUE;
7579 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
7580 "Complex regular subexpression recursion limit (%d) "
7586 ST.save_curlyx = cur_curlyx;
7587 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
7588 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
7590 NOT_REACHED; /* NOTREACHED */
7592 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
7593 cur_curlyx = ST.save_curlyx;
7594 REGCP_UNWIND(ST.lastcp);
7595 regcppop(rex, &maxopenparen);
7597 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
7598 /* Maximum greed exceeded */
7599 if (cur_curlyx->u.curlyx.count >= REG_INFTY
7600 && ckWARN(WARN_REGEXP)
7601 && !reginfo->warned)
7603 reginfo->warned = TRUE;
7604 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
7605 "Complex regular subexpression recursion "
7606 "limit (%d) exceeded",
7609 cur_curlyx->u.curlyx.count--;
7613 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "trying longer...\n", depth)
7615 /* Try grabbing another A and see if it helps. */
7616 cur_curlyx->u.curlyx.lastloc = locinput;
7617 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
7619 REGCP_SET(ST.lastcp);
7620 PUSH_STATE_GOTO(WHILEM_A_min,
7621 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
7623 NOT_REACHED; /* NOTREACHED */
7626 #define ST st->u.branch
7628 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
7629 next = scan + ARG(scan);
7632 scan = NEXTOPER(scan);
7635 case BRANCH: /* /(...|A|...)/ */
7636 scan = NEXTOPER(scan); /* scan now points to inner node */
7637 ST.lastparen = rex->lastparen;
7638 ST.lastcloseparen = rex->lastcloseparen;
7639 ST.next_branch = next;
7642 /* Now go into the branch */
7644 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
7646 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
7648 NOT_REACHED; /* NOTREACHED */
7650 case CUTGROUP: /* /(*THEN)/ */
7651 sv_yes_mark = st->u.mark.mark_name = scan->flags
7652 ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ])
7654 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
7655 NOT_REACHED; /* NOTREACHED */
7657 case CUTGROUP_next_fail:
7660 if (st->u.mark.mark_name)
7661 sv_commit = st->u.mark.mark_name;
7663 NOT_REACHED; /* NOTREACHED */
7667 NOT_REACHED; /* NOTREACHED */
7669 case BRANCH_next_fail: /* that branch failed; try the next, if any */
7674 REGCP_UNWIND(ST.cp);
7675 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7676 scan = ST.next_branch;
7677 /* no more branches? */
7678 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
7680 Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n",
7687 continue; /* execute next BRANCH[J] op */
7690 case MINMOD: /* next op will be non-greedy, e.g. A*? */
7695 #define ST st->u.curlym
7697 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
7699 /* This is an optimisation of CURLYX that enables us to push
7700 * only a single backtracking state, no matter how many matches
7701 * there are in {m,n}. It relies on the pattern being constant
7702 * length, with no parens to influence future backrefs
7706 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7708 ST.lastparen = rex->lastparen;
7709 ST.lastcloseparen = rex->lastcloseparen;
7711 /* if paren positive, emulate an OPEN/CLOSE around A */
7713 U32 paren = ST.me->flags;
7714 if (paren > maxopenparen)
7715 maxopenparen = paren;
7716 scan += NEXT_OFF(scan); /* Skip former OPEN. */
7724 ST.c1 = CHRTEST_UNINIT;
7727 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
7730 curlym_do_A: /* execute the A in /A{m,n}B/ */
7731 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
7732 NOT_REACHED; /* NOTREACHED */
7734 case CURLYM_A: /* we've just matched an A */
7736 /* after first match, determine A's length: u.curlym.alen */
7737 if (ST.count == 1) {
7738 if (reginfo->is_utf8_target) {
7739 char *s = st->locinput;
7740 while (s < locinput) {
7746 ST.alen = locinput - st->locinput;
7749 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
7752 Perl_re_exec_indentf( aTHX_ "CURLYM now matched %" IVdf " times, len=%" IVdf "...\n",
7753 depth, (IV) ST.count, (IV)ST.alen)
7756 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7760 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
7761 if ( max == REG_INFTY || ST.count < max )
7762 goto curlym_do_A; /* try to match another A */
7764 goto curlym_do_B; /* try to match B */
7766 case CURLYM_A_fail: /* just failed to match an A */
7767 REGCP_UNWIND(ST.cp);
7770 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
7771 || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7774 curlym_do_B: /* execute the B in /A{m,n}B/ */
7775 if (ST.c1 == CHRTEST_UNINIT) {
7776 /* calculate c1 and c2 for possible match of 1st char
7777 * following curly */
7778 ST.c1 = ST.c2 = CHRTEST_VOID;
7780 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
7781 regnode *text_node = ST.B;
7782 if (! HAS_TEXT(text_node))
7783 FIND_NEXT_IMPT(text_node);
7786 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
7788 But the former is redundant in light of the latter.
7790 if this changes back then the macro for
7791 IS_TEXT and friends need to change.
7793 if (PL_regkind[OP(text_node)] == EXACT) {
7794 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
7795 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
7805 Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%" IVdf "...\n",
7806 depth, (IV)ST.count)
7808 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
7809 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
7810 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
7811 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
7813 /* simulate B failing */
7815 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%" UVXf " c1=0x%" UVXf " c2=0x%" UVXf "\n",
7817 valid_utf8_to_uvchr((U8 *) locinput, NULL),
7818 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
7819 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
7821 state_num = CURLYM_B_fail;
7822 goto reenter_switch;
7825 else if (nextchr != ST.c1 && nextchr != ST.c2) {
7826 /* simulate B failing */
7828 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
7830 (int) nextchr, ST.c1, ST.c2)
7832 state_num = CURLYM_B_fail;
7833 goto reenter_switch;
7838 /* emulate CLOSE: mark current A as captured */
7839 I32 paren = ST.me->flags;
7841 rex->offs[paren].start
7842 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
7843 rex->offs[paren].end = locinput - reginfo->strbeg;
7844 if ((U32)paren > rex->lastparen)
7845 rex->lastparen = paren;
7846 rex->lastcloseparen = paren;
7849 rex->offs[paren].end = -1;
7851 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7860 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
7861 NOT_REACHED; /* NOTREACHED */
7863 case CURLYM_B_fail: /* just failed to match a B */
7864 REGCP_UNWIND(ST.cp);
7865 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7867 I32 max = ARG2(ST.me);
7868 if (max != REG_INFTY && ST.count == max)
7870 goto curlym_do_A; /* try to match a further A */
7872 /* backtrack one A */
7873 if (ST.count == ARG1(ST.me) /* min */)
7876 SET_locinput(HOPc(locinput, -ST.alen));
7877 goto curlym_do_B; /* try to match B */
7880 #define ST st->u.curly
7882 #define CURLY_SETPAREN(paren, success) \
7885 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
7886 rex->offs[paren].end = locinput - reginfo->strbeg; \
7887 if (paren > rex->lastparen) \
7888 rex->lastparen = paren; \
7889 rex->lastcloseparen = paren; \
7892 rex->offs[paren].end = -1; \
7893 rex->lastparen = ST.lastparen; \
7894 rex->lastcloseparen = ST.lastcloseparen; \
7898 case STAR: /* /A*B/ where A is width 1 char */
7902 scan = NEXTOPER(scan);
7905 case PLUS: /* /A+B/ where A is width 1 char */
7909 scan = NEXTOPER(scan);
7912 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
7913 ST.paren = scan->flags; /* Which paren to set */
7914 ST.lastparen = rex->lastparen;
7915 ST.lastcloseparen = rex->lastcloseparen;
7916 if (ST.paren > maxopenparen)
7917 maxopenparen = ST.paren;
7918 ST.min = ARG1(scan); /* min to match */
7919 ST.max = ARG2(scan); /* max to match */
7920 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
7925 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
7928 case CURLY: /* /A{m,n}B/ where A is width 1 char */
7930 ST.min = ARG1(scan); /* min to match */
7931 ST.max = ARG2(scan); /* max to match */
7932 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7935 * Lookahead to avoid useless match attempts
7936 * when we know what character comes next.
7938 * Used to only do .*x and .*?x, but now it allows
7939 * for )'s, ('s and (?{ ... })'s to be in the way
7940 * of the quantifier and the EXACT-like node. -- japhy
7943 assert(ST.min <= ST.max);
7944 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
7945 ST.c1 = ST.c2 = CHRTEST_VOID;
7948 regnode *text_node = next;
7950 if (! HAS_TEXT(text_node))
7951 FIND_NEXT_IMPT(text_node);
7953 if (! HAS_TEXT(text_node))
7954 ST.c1 = ST.c2 = CHRTEST_VOID;
7956 if ( PL_regkind[OP(text_node)] != EXACT ) {
7957 ST.c1 = ST.c2 = CHRTEST_VOID;
7961 /* Currently we only get here when
7963 PL_rekind[OP(text_node)] == EXACT
7965 if this changes back then the macro for IS_TEXT and
7966 friends need to change. */
7967 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
7968 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
7980 char *li = locinput;
7983 regrepeat(rex, &li, ST.A, reginfo, ST.min)
7989 if (ST.c1 == CHRTEST_VOID)
7990 goto curly_try_B_min;
7992 ST.oldloc = locinput;
7994 /* set ST.maxpos to the furthest point along the
7995 * string that could possibly match */
7996 if (ST.max == REG_INFTY) {
7997 ST.maxpos = reginfo->strend - 1;
7999 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
8002 else if (utf8_target) {
8003 int m = ST.max - ST.min;
8004 for (ST.maxpos = locinput;
8005 m >0 && ST.maxpos < reginfo->strend; m--)
8006 ST.maxpos += UTF8SKIP(ST.maxpos);
8009 ST.maxpos = locinput + ST.max - ST.min;
8010 if (ST.maxpos >= reginfo->strend)
8011 ST.maxpos = reginfo->strend - 1;
8013 goto curly_try_B_min_known;
8017 /* avoid taking address of locinput, so it can remain
8019 char *li = locinput;
8020 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max);
8021 if (ST.count < ST.min)
8024 if ((ST.count > ST.min)
8025 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
8027 /* A{m,n} must come at the end of the string, there's
8028 * no point in backing off ... */
8030 /* ...except that $ and \Z can match before *and* after
8031 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
8032 We may back off by one in this case. */
8033 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
8037 goto curly_try_B_max;
8039 NOT_REACHED; /* NOTREACHED */
8041 case CURLY_B_min_known_fail:
8042 /* failed to find B in a non-greedy match where c1,c2 valid */
8044 REGCP_UNWIND(ST.cp);
8046 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8048 /* Couldn't or didn't -- move forward. */
8049 ST.oldloc = locinput;
8051 locinput += UTF8SKIP(locinput);
8055 curly_try_B_min_known:
8056 /* find the next place where 'B' could work, then call B */
8060 n = (ST.oldloc == locinput) ? 0 : 1;
8061 if (ST.c1 == ST.c2) {
8062 /* set n to utf8_distance(oldloc, locinput) */
8063 while (locinput <= ST.maxpos
8064 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
8066 locinput += UTF8SKIP(locinput);
8071 /* set n to utf8_distance(oldloc, locinput) */
8072 while (locinput <= ST.maxpos
8073 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
8074 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
8076 locinput += UTF8SKIP(locinput);
8081 else { /* Not utf8_target */
8082 if (ST.c1 == ST.c2) {
8083 while (locinput <= ST.maxpos &&
8084 UCHARAT(locinput) != ST.c1)
8088 while (locinput <= ST.maxpos
8089 && UCHARAT(locinput) != ST.c1
8090 && UCHARAT(locinput) != ST.c2)
8093 n = locinput - ST.oldloc;
8095 if (locinput > ST.maxpos)
8098 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
8099 * at b; check that everything between oldloc and
8100 * locinput matches */
8101 char *li = ST.oldloc;
8103 if (regrepeat(rex, &li, ST.A, reginfo, n) < n)
8105 assert(n == REG_INFTY || locinput == li);
8107 CURLY_SETPAREN(ST.paren, ST.count);
8108 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8110 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
8112 NOT_REACHED; /* NOTREACHED */
8114 case CURLY_B_min_fail:
8115 /* failed to find B in a non-greedy match where c1,c2 invalid */
8117 REGCP_UNWIND(ST.cp);
8119 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8121 /* failed -- move forward one */
8123 char *li = locinput;
8124 if (!regrepeat(rex, &li, ST.A, reginfo, 1)) {
8131 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
8132 ST.count > 0)) /* count overflow ? */
8135 CURLY_SETPAREN(ST.paren, ST.count);
8136 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8138 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
8142 NOT_REACHED; /* NOTREACHED */
8145 /* a successful greedy match: now try to match B */
8146 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8149 bool could_match = locinput < reginfo->strend;
8151 /* If it could work, try it. */
8152 if (ST.c1 != CHRTEST_VOID && could_match) {
8153 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
8155 could_match = memEQ(locinput,
8160 UTF8SKIP(locinput));
8163 could_match = UCHARAT(locinput) == ST.c1
8164 || UCHARAT(locinput) == ST.c2;
8167 if (ST.c1 == CHRTEST_VOID || could_match) {
8168 CURLY_SETPAREN(ST.paren, ST.count);
8169 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
8170 NOT_REACHED; /* NOTREACHED */
8175 case CURLY_B_max_fail:
8176 /* failed to find B in a greedy match */
8178 REGCP_UNWIND(ST.cp);
8180 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8183 if (--ST.count < ST.min)
8185 locinput = HOPc(locinput, -1);
8186 goto curly_try_B_max;
8190 case END: /* last op of main pattern */
8193 /* we've just finished A in /(??{A})B/; now continue with B */
8194 SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput);
8195 st->u.eval.prev_rex = rex_sv; /* inner */
8197 /* Save *all* the positions. */
8198 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
8199 rex_sv = CUR_EVAL.prev_rex;
8200 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
8201 SET_reg_curpm(rex_sv);
8202 rex = ReANY(rex_sv);
8203 rexi = RXi_GET(rex);
8205 st->u.eval.prev_curlyx = cur_curlyx;
8206 cur_curlyx = CUR_EVAL.prev_curlyx;
8208 REGCP_SET(st->u.eval.lastcp);
8210 /* Restore parens of the outer rex without popping the
8212 regcp_restore(rex, CUR_EVAL.lastcp, &maxopenparen);
8214 st->u.eval.prev_eval = cur_eval;
8215 cur_eval = CUR_EVAL.prev_eval;
8217 Perl_re_exec_indentf( aTHX_ "EVAL trying tail ... (cur_eval=%p)\n",
8219 if ( nochange_depth )
8222 SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput);
8224 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
8225 locinput); /* match B */
8228 if (locinput < reginfo->till) {
8229 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
8230 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
8232 (long)(locinput - startpos),
8233 (long)(reginfo->till - startpos),
8236 sayNO_SILENT; /* Cannot match: too short. */
8238 sayYES; /* Success! */
8240 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
8242 Perl_re_exec_indentf( aTHX_ "%ssubpattern success...%s\n",
8243 depth, PL_colors[4], PL_colors[5]));
8244 sayYES; /* Success! */
8247 #define ST st->u.ifmatch
8252 case SUSPEND: /* (?>A) */
8254 newstart = locinput;
8257 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
8259 goto ifmatch_trivial_fail_test;
8261 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
8263 ifmatch_trivial_fail_test:
8265 char * const s = HOPBACKc(locinput, scan->flags);
8270 sw = 1 - cBOOL(ST.wanted);
8274 next = scan + ARG(scan);
8282 newstart = locinput;
8286 ST.logical = logical;
8287 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
8289 /* execute body of (?...A) */
8290 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
8291 NOT_REACHED; /* NOTREACHED */
8294 case IFMATCH_A_fail: /* body of (?...A) failed */
8295 ST.wanted = !ST.wanted;
8298 case IFMATCH_A: /* body of (?...A) succeeded */
8300 sw = cBOOL(ST.wanted);
8302 else if (!ST.wanted)
8305 if (OP(ST.me) != SUSPEND) {
8306 /* restore old position except for (?>...) */
8307 locinput = st->locinput;
8309 scan = ST.me + ARG(ST.me);
8312 continue; /* execute B */
8316 case LONGJMP: /* alternative with many branches compiles to
8317 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
8318 next = scan + ARG(scan);
8323 case COMMIT: /* (*COMMIT) */
8324 reginfo->cutpoint = reginfo->strend;
8327 case PRUNE: /* (*PRUNE) */
8329 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8330 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
8331 NOT_REACHED; /* NOTREACHED */
8333 case COMMIT_next_fail:
8337 NOT_REACHED; /* NOTREACHED */
8339 case OPFAIL: /* (*FAIL) */
8341 sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8343 /* deal with (?(?!)X|Y) properly,
8344 * make sure we trigger the no branch
8345 * of the trailing IFTHEN structure*/
8351 NOT_REACHED; /* NOTREACHED */
8353 #define ST st->u.mark
8354 case MARKPOINT: /* (*MARK:foo) */
8355 ST.prev_mark = mark_state;
8356 ST.mark_name = sv_commit = sv_yes_mark
8357 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8359 ST.mark_loc = locinput;
8360 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
8361 NOT_REACHED; /* NOTREACHED */
8363 case MARKPOINT_next:
8364 mark_state = ST.prev_mark;
8366 NOT_REACHED; /* NOTREACHED */
8368 case MARKPOINT_next_fail:
8369 if (popmark && sv_eq(ST.mark_name,popmark))
8371 if (ST.mark_loc > startpoint)
8372 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8373 popmark = NULL; /* we found our mark */
8374 sv_commit = ST.mark_name;
8377 Perl_re_exec_indentf( aTHX_ "%ssetting cutpoint to mark:%" SVf "...%s\n",
8379 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
8382 mark_state = ST.prev_mark;
8383 sv_yes_mark = mark_state ?
8384 mark_state->u.mark.mark_name : NULL;
8386 NOT_REACHED; /* NOTREACHED */
8388 case SKIP: /* (*SKIP) */
8390 /* (*SKIP) : if we fail we cut here*/
8391 ST.mark_name = NULL;
8392 ST.mark_loc = locinput;
8393 PUSH_STATE_GOTO(SKIP_next,next, locinput);
8395 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
8396 otherwise do nothing. Meaning we need to scan
8398 regmatch_state *cur = mark_state;
8399 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8402 if ( sv_eq( cur->u.mark.mark_name,
8405 ST.mark_name = find;
8406 PUSH_STATE_GOTO( SKIP_next, next, locinput);
8408 cur = cur->u.mark.prev_mark;
8411 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
8414 case SKIP_next_fail:
8416 /* (*CUT:NAME) - Set up to search for the name as we
8417 collapse the stack*/
8418 popmark = ST.mark_name;
8420 /* (*CUT) - No name, we cut here.*/
8421 if (ST.mark_loc > startpoint)
8422 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8423 /* but we set sv_commit to latest mark_name if there
8424 is one so they can test to see how things lead to this
8427 sv_commit=mark_state->u.mark.mark_name;
8431 NOT_REACHED; /* NOTREACHED */
8434 case LNBREAK: /* \R */
8435 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
8442 PerlIO_printf(Perl_error_log, "%" UVxf " %d\n",
8443 PTR2UV(scan), OP(scan));
8444 Perl_croak(aTHX_ "regexp memory corruption");
8446 /* this is a point to jump to in order to increment
8447 * locinput by one character */
8449 assert(!NEXTCHR_IS_EOS);
8451 locinput += PL_utf8skip[nextchr];
8452 /* locinput is allowed to go 1 char off the end, but not 2+ */
8453 if (locinput > reginfo->strend)
8462 /* switch break jumps here */
8463 scan = next; /* prepare to execute the next op and ... */
8464 continue; /* ... jump back to the top, reusing st */
8468 /* push a state that backtracks on success */
8469 st->u.yes.prev_yes_state = yes_state;
8473 /* push a new regex state, then continue at scan */
8475 regmatch_state *newst;
8478 regmatch_state *cur = st;
8479 regmatch_state *curyes = yes_state;
8481 regmatch_slab *slab = PL_regmatch_slab;
8482 for (;curd > -1 && (depth-curd < 3);cur--,curd--) {
8483 if (cur < SLAB_FIRST(slab)) {
8485 cur = SLAB_LAST(slab);
8487 Perl_re_exec_indentf( aTHX_ "#%-3d %-10s %s\n",
8489 curd, PL_reg_name[cur->resume_state],
8490 (curyes == cur) ? "yes" : ""
8493 curyes = cur->u.yes.prev_yes_state;
8496 DEBUG_STATE_pp("push")
8499 st->locinput = locinput;
8501 if (newst > SLAB_LAST(PL_regmatch_slab))
8502 newst = S_push_slab(aTHX);
8503 PL_regmatch_state = newst;
8505 locinput = pushinput;
8511 #ifdef SOLARIS_BAD_OPTIMIZER
8512 # undef PL_charclass
8516 * We get here only if there's trouble -- normally "case END" is
8517 * the terminating point.
8519 Perl_croak(aTHX_ "corrupted regexp pointers");
8520 NOT_REACHED; /* NOTREACHED */
8524 /* we have successfully completed a subexpression, but we must now
8525 * pop to the state marked by yes_state and continue from there */
8526 assert(st != yes_state);
8528 while (st != yes_state) {
8530 if (st < SLAB_FIRST(PL_regmatch_slab)) {
8531 PL_regmatch_slab = PL_regmatch_slab->prev;
8532 st = SLAB_LAST(PL_regmatch_slab);
8536 DEBUG_STATE_pp("pop (no final)");
8538 DEBUG_STATE_pp("pop (yes)");
8544 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
8545 || yes_state > SLAB_LAST(PL_regmatch_slab))
8547 /* not in this slab, pop slab */
8548 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
8549 PL_regmatch_slab = PL_regmatch_slab->prev;
8550 st = SLAB_LAST(PL_regmatch_slab);
8552 depth -= (st - yes_state);
8555 yes_state = st->u.yes.prev_yes_state;
8556 PL_regmatch_state = st;
8559 locinput= st->locinput;
8560 state_num = st->resume_state + no_final;
8561 goto reenter_switch;
8564 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n",
8565 PL_colors[4], PL_colors[5]));
8567 if (reginfo->info_aux_eval) {
8568 /* each successfully executed (?{...}) block does the equivalent of
8569 * local $^R = do {...}
8570 * When popping the save stack, all these locals would be undone;
8571 * bypass this by setting the outermost saved $^R to the latest
8573 /* I dont know if this is needed or works properly now.
8574 * see code related to PL_replgv elsewhere in this file.
8577 if (oreplsv != GvSV(PL_replgv))
8578 sv_setsv(oreplsv, GvSV(PL_replgv));
8585 Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n",
8587 PL_colors[4], PL_colors[5])
8599 /* there's a previous state to backtrack to */
8601 if (st < SLAB_FIRST(PL_regmatch_slab)) {
8602 PL_regmatch_slab = PL_regmatch_slab->prev;
8603 st = SLAB_LAST(PL_regmatch_slab);
8605 PL_regmatch_state = st;
8606 locinput= st->locinput;
8608 DEBUG_STATE_pp("pop");
8610 if (yes_state == st)
8611 yes_state = st->u.yes.prev_yes_state;
8613 state_num = st->resume_state + 1; /* failure = success + 1 */
8615 goto reenter_switch;
8620 if (rex->intflags & PREGf_VERBARG_SEEN) {
8621 SV *sv_err = get_sv("REGERROR", 1);
8622 SV *sv_mrk = get_sv("REGMARK", 1);
8624 sv_commit = &PL_sv_no;
8626 sv_yes_mark = &PL_sv_yes;
8629 sv_commit = &PL_sv_yes;
8630 sv_yes_mark = &PL_sv_no;
8634 sv_setsv(sv_err, sv_commit);
8635 sv_setsv(sv_mrk, sv_yes_mark);
8639 if (last_pushed_cv) {
8642 PERL_UNUSED_VAR(SP);
8645 assert(!result || locinput - reginfo->strbeg >= 0);
8646 return result ? locinput - reginfo->strbeg : -1;
8650 - regrepeat - repeatedly match something simple, report how many
8652 * What 'simple' means is a node which can be the operand of a quantifier like
8655 * startposp - pointer a pointer to the start position. This is updated
8656 * to point to the byte following the highest successful
8658 * p - the regnode to be repeatedly matched against.
8659 * reginfo - struct holding match state, such as strend
8660 * max - maximum number of things to match.
8661 * depth - (for debugging) backtracking depth.
8664 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
8665 regmatch_info *const reginfo, I32 max _pDEPTH)
8667 char *scan; /* Pointer to current position in target string */
8669 char *loceol = reginfo->strend; /* local version */
8670 I32 hardcount = 0; /* How many matches so far */
8671 bool utf8_target = reginfo->is_utf8_target;
8672 unsigned int to_complement = 0; /* Invert the result? */
8674 _char_class_number classnum;
8676 PERL_ARGS_ASSERT_REGREPEAT;
8679 if (max == REG_INFTY)
8681 else if (! utf8_target && loceol - scan > max)
8682 loceol = scan + max;
8684 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
8685 * to the maximum of how far we should go in it (leaving it set to the real
8686 * end, if the maximum permissible would take us beyond that). This allows
8687 * us to make the loop exit condition that we haven't gone past <loceol> to
8688 * also mean that we haven't exceeded the max permissible count, saving a
8689 * test each time through the loop. But it assumes that the OP matches a
8690 * single byte, which is true for most of the OPs below when applied to a
8691 * non-UTF-8 target. Those relatively few OPs that don't have this
8692 * characteristic will have to compensate.
8694 * There is no adjustment for UTF-8 targets, as the number of bytes per
8695 * character varies. OPs will have to test both that the count is less
8696 * than the max permissible (using <hardcount> to keep track), and that we
8697 * are still within the bounds of the string (using <loceol>. A few OPs
8698 * match a single byte no matter what the encoding. They can omit the max
8699 * test if, for the UTF-8 case, they do the adjustment that was skipped
8702 * Thus, the code above sets things up for the common case; and exceptional
8703 * cases need extra work; the common case is to make sure <scan> doesn't
8704 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
8705 * count doesn't exceed the maximum permissible */
8710 while (scan < loceol && hardcount < max && *scan != '\n') {
8711 scan += UTF8SKIP(scan);
8715 while (scan < loceol && *scan != '\n')
8721 while (scan < loceol && hardcount < max) {
8722 scan += UTF8SKIP(scan);
8730 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8731 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
8732 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
8736 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8740 /* Can use a simple loop if the pattern char to match on is invariant
8741 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
8742 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
8743 * true iff it doesn't matter if the argument is in UTF-8 or not */
8744 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
8745 if (utf8_target && loceol - scan > max) {
8746 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
8747 * since here, to match at all, 1 char == 1 byte */
8748 loceol = scan + max;
8750 while (scan < loceol && UCHARAT(scan) == c) {
8754 else if (reginfo->is_utf8_pat) {
8756 STRLEN scan_char_len;
8758 /* When both target and pattern are UTF-8, we have to do
8760 while (hardcount < max
8762 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
8763 && memEQ(scan, STRING(p), scan_char_len))
8765 scan += scan_char_len;
8769 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
8771 /* Target isn't utf8; convert the character in the UTF-8
8772 * pattern to non-UTF8, and do a simple loop */
8773 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
8774 while (scan < loceol && UCHARAT(scan) == c) {
8777 } /* else pattern char is above Latin1, can't possibly match the
8782 /* Here, the string must be utf8; pattern isn't, and <c> is
8783 * different in utf8 than not, so can't compare them directly.
8784 * Outside the loop, find the two utf8 bytes that represent c, and
8785 * then look for those in sequence in the utf8 string */
8786 U8 high = UTF8_TWO_BYTE_HI(c);
8787 U8 low = UTF8_TWO_BYTE_LO(c);
8789 while (hardcount < max
8790 && scan + 1 < loceol
8791 && UCHARAT(scan) == high
8792 && UCHARAT(scan + 1) == low)
8800 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
8801 assert(! reginfo->is_utf8_pat);
8804 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
8808 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8809 utf8_flags = FOLDEQ_LOCALE;
8812 case EXACTF: /* This node only generated for non-utf8 patterns */
8813 assert(! reginfo->is_utf8_pat);
8818 if (! utf8_target) {
8821 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
8822 | FOLDEQ_S2_FOLDS_SANE;
8827 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
8831 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
8833 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8835 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
8838 if (c1 == CHRTEST_VOID) {
8839 /* Use full Unicode fold matching */
8840 char *tmpeol = reginfo->strend;
8841 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
8842 while (hardcount < max
8843 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
8844 STRING(p), NULL, pat_len,
8845 reginfo->is_utf8_pat, utf8_flags))
8848 tmpeol = reginfo->strend;
8852 else if (utf8_target) {
8854 while (scan < loceol
8856 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
8858 scan += UTF8SKIP(scan);
8863 while (scan < loceol
8865 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
8866 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
8868 scan += UTF8SKIP(scan);
8873 else if (c1 == c2) {
8874 while (scan < loceol && UCHARAT(scan) == c1) {
8879 while (scan < loceol &&
8880 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
8889 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8891 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) {
8892 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
8898 while (hardcount < max
8900 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
8902 scan += UTF8SKIP(scan);
8906 else if (ANYOF_FLAGS(p)) {
8907 while (scan < loceol
8908 && reginclass(prog, p, (U8*)scan, (U8*)scan+1, 0))
8912 while (scan < loceol && ANYOF_BITMAP_TEST(p, *((U8*)scan)))
8917 /* The argument (FLAGS) to all the POSIX node types is the class number */
8924 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8925 if (! utf8_target) {
8926 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
8932 while (hardcount < max && scan < loceol
8933 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
8936 scan += UTF8SKIP(scan);
8949 if (utf8_target && loceol - scan > max) {
8951 /* We didn't adjust <loceol> at the beginning of this routine
8952 * because is UTF-8, but it is actually ok to do so, since here, to
8953 * match, 1 char == 1 byte. */
8954 loceol = scan + max;
8956 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
8969 if (! utf8_target) {
8970 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
8976 /* The complement of something that matches only ASCII matches all
8977 * non-ASCII, plus everything in ASCII that isn't in the class. */
8978 while (hardcount < max && scan < loceol
8979 && ( ! isASCII_utf8_safe(scan, reginfo->strend)
8980 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
8982 scan += UTF8SKIP(scan);
8993 if (! utf8_target) {
8994 while (scan < loceol && to_complement
8995 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
9002 classnum = (_char_class_number) FLAGS(p);
9003 if (classnum < _FIRST_NON_SWASH_CC) {
9005 /* Here, a swash is needed for above-Latin1 code points.
9006 * Process as many Latin1 code points using the built-in rules.
9007 * Go to another loop to finish processing upon encountering
9008 * the first Latin1 code point. We could do that in this loop
9009 * as well, but the other way saves having to test if the swash
9010 * has been loaded every time through the loop: extra space to
9012 while (hardcount < max && scan < loceol) {
9013 if (UTF8_IS_INVARIANT(*scan)) {
9014 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
9021 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
9022 if (! (to_complement
9023 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*scan,
9032 goto found_above_latin1;
9039 /* For these character classes, the knowledge of how to handle
9040 * every code point is compiled in to Perl via a macro. This
9041 * code is written for making the loops as tight as possible.
9042 * It could be refactored to save space instead */
9044 case _CC_ENUM_SPACE:
9045 while (hardcount < max
9048 ^ cBOOL(isSPACE_utf8_safe(scan, loceol))))
9050 scan += UTF8SKIP(scan);
9054 case _CC_ENUM_BLANK:
9055 while (hardcount < max
9058 ^ cBOOL(isBLANK_utf8_safe(scan, loceol))))
9060 scan += UTF8SKIP(scan);
9064 case _CC_ENUM_XDIGIT:
9065 while (hardcount < max
9068 ^ cBOOL(isXDIGIT_utf8_safe(scan, loceol))))
9070 scan += UTF8SKIP(scan);
9074 case _CC_ENUM_VERTSPACE:
9075 while (hardcount < max
9078 ^ cBOOL(isVERTWS_utf8_safe(scan, loceol))))
9080 scan += UTF8SKIP(scan);
9084 case _CC_ENUM_CNTRL:
9085 while (hardcount < max
9088 ^ cBOOL(isCNTRL_utf8_safe(scan, loceol))))
9090 scan += UTF8SKIP(scan);
9095 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
9101 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
9103 /* Load the swash if not already present */
9104 if (! PL_utf8_swash_ptrs[classnum]) {
9105 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
9106 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
9110 PL_XPosix_ptrs[classnum], &flags);
9113 while (hardcount < max && scan < loceol
9114 && to_complement ^ cBOOL(_generic_utf8_safe(
9118 swash_fetch(PL_utf8_swash_ptrs[classnum],
9122 scan += UTF8SKIP(scan);
9129 while (hardcount < max && scan < loceol &&
9130 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
9135 /* LNBREAK can match one or two latin chars, which is ok, but we
9136 * have to use hardcount in this situation, and throw away the
9137 * adjustment to <loceol> done before the switch statement */
9138 loceol = reginfo->strend;
9139 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
9148 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9162 /* These are all 0 width, so match right here or not at all. */
9166 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
9167 NOT_REACHED; /* NOTREACHED */
9174 c = scan - *startposp;
9178 GET_RE_DEBUG_FLAGS_DECL;
9180 SV * const prop = sv_newmortal();
9181 regprop(prog, prop, p, reginfo, NULL);
9182 Perl_re_exec_indentf( aTHX_ "%s can match %" IVdf " times out of %" IVdf "...\n",
9183 depth, SvPVX_const(prop),(IV)c,(IV)max);
9191 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
9193 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
9194 create a copy so that changes the caller makes won't change the shared one.
9195 If <altsvp> is non-null, will return NULL in it, for back-compat.
9198 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
9200 PERL_ARGS_ASSERT_REGCLASS_SWASH;
9206 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL));
9209 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
9212 - reginclass - determine if a character falls into a character class
9214 n is the ANYOF-type regnode
9215 p is the target string
9216 p_end points to one byte beyond the end of the target string
9217 utf8_target tells whether p is in UTF-8.
9219 Returns true if matched; false otherwise.
9221 Note that this can be a synthetic start class, a combination of various
9222 nodes, so things you think might be mutually exclusive, such as locale,
9223 aren't. It can match both locale and non-locale
9228 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
9231 const char flags = ANYOF_FLAGS(n);
9235 PERL_ARGS_ASSERT_REGINCLASS;
9237 /* If c is not already the code point, get it. Note that
9238 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
9239 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
9241 const U32 utf8n_flags = UTF8_ALLOW_DEFAULT;
9242 c = utf8n_to_uvchr(p, p_end - p, &c_len, utf8n_flags | UTF8_CHECK_ONLY);
9243 if (c_len == (STRLEN)-1) {
9244 _force_out_malformed_utf8_message(p, p_end,
9246 1 /* 1 means die */ );
9247 NOT_REACHED; /* NOTREACHED */
9249 if (c > 255 && OP(n) == ANYOFL && ! ANYOFL_UTF8_LOCALE_REQD(flags)) {
9250 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
9254 /* If this character is potentially in the bitmap, check it */
9255 if (c < NUM_ANYOF_CODE_POINTS) {
9256 if (ANYOF_BITMAP_TEST(n, c))
9259 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9266 else if (flags & ANYOF_LOCALE_FLAGS) {
9267 if ((flags & ANYOFL_FOLD)
9269 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
9273 else if (ANYOF_POSIXL_TEST_ANY_SET(n)
9277 /* The data structure is arranged so bits 0, 2, 4, ... are set
9278 * if the class includes the Posix character class given by
9279 * bit/2; and 1, 3, 5, ... are set if the class includes the
9280 * complemented Posix class given by int(bit/2). So we loop
9281 * through the bits, each time changing whether we complement
9282 * the result or not. Suppose for the sake of illustration
9283 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
9284 * is set, it means there is a match for this ANYOF node if the
9285 * character is in the class given by the expression (0 / 2 = 0
9286 * = \w). If it is in that class, isFOO_lc() will return 1,
9287 * and since 'to_complement' is 0, the result will stay TRUE,
9288 * and we exit the loop. Suppose instead that bit 0 is 0, but
9289 * bit 1 is 1. That means there is a match if the character
9290 * matches \W. We won't bother to call isFOO_lc() on bit 0,
9291 * but will on bit 1. On the second iteration 'to_complement'
9292 * will be 1, so the exclusive or will reverse things, so we
9293 * are testing for \W. On the third iteration, 'to_complement'
9294 * will be 0, and we would be testing for \s; the fourth
9295 * iteration would test for \S, etc.
9297 * Note that this code assumes that all the classes are closed
9298 * under folding. For example, if a character matches \w, then
9299 * its fold does too; and vice versa. This should be true for
9300 * any well-behaved locale for all the currently defined Posix
9301 * classes, except for :lower: and :upper:, which are handled
9302 * by the pseudo-class :cased: which matches if either of the
9303 * other two does. To get rid of this assumption, an outer
9304 * loop could be used below to iterate over both the source
9305 * character, and its fold (if different) */
9308 int to_complement = 0;
9310 while (count < ANYOF_MAX) {
9311 if (ANYOF_POSIXL_TEST(n, count)
9312 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
9325 /* If the bitmap didn't (or couldn't) match, and something outside the
9326 * bitmap could match, try that. */
9328 if (c >= NUM_ANYOF_CODE_POINTS
9329 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
9331 match = TRUE; /* Everything above the bitmap matches */
9333 /* Here doesn't match everything above the bitmap. If there is
9334 * some information available beyond the bitmap, we may find a
9335 * match in it. If so, this is most likely because the code point
9336 * is outside the bitmap range. But rarely, it could be because of
9337 * some other reason. If so, various flags are set to indicate
9338 * this possibility. On ANYOFD nodes, there may be matches that
9339 * happen only when the target string is UTF-8; or for other node
9340 * types, because runtime lookup is needed, regardless of the
9341 * UTF-8ness of the target string. Finally, under /il, there may
9342 * be some matches only possible if the locale is a UTF-8 one. */
9343 else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP
9344 && ( c >= NUM_ANYOF_CODE_POINTS
9345 || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)
9346 && ( UNLIKELY(OP(n) != ANYOFD)
9347 || (utf8_target && ! isASCII_uni(c)
9348 # if NUM_ANYOF_CODE_POINTS > 256
9352 || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags)
9353 && IN_UTF8_CTYPE_LOCALE)))
9355 SV* only_utf8_locale = NULL;
9356 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
9357 &only_utf8_locale, NULL);
9363 } else { /* Convert to utf8 */
9364 utf8_p = utf8_buffer;
9365 append_utf8_from_native_byte(*p, &utf8_p);
9366 utf8_p = utf8_buffer;
9369 if (swash_fetch(sw, utf8_p, TRUE)) {
9373 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
9374 match = _invlist_contains_cp(only_utf8_locale, c);
9378 if (UNICODE_IS_SUPER(c)
9380 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9382 && ckWARN_d(WARN_NON_UNICODE))
9384 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
9385 "Matched non-Unicode code point 0x%04" UVXf " against Unicode property; may not be portable", c);
9389 #if ANYOF_INVERT != 1
9390 /* Depending on compiler optimization cBOOL takes time, so if don't have to
9392 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
9395 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
9396 return (flags & ANYOF_INVERT) ^ match;
9400 S_reghop3(U8 *s, SSize_t off, const U8* lim)
9402 /* return the position 'off' UTF-8 characters away from 's', forward if
9403 * 'off' >= 0, backwards if negative. But don't go outside of position
9404 * 'lim', which better be < s if off < 0 */
9406 PERL_ARGS_ASSERT_REGHOP3;
9409 while (off-- && s < lim) {
9410 /* XXX could check well-formedness here */
9415 while (off++ && s > lim) {
9417 if (UTF8_IS_CONTINUED(*s)) {
9418 while (s > lim && UTF8_IS_CONTINUATION(*s))
9420 if (! UTF8_IS_START(*s)) {
9421 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9424 /* XXX could check well-formedness here */
9431 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
9433 PERL_ARGS_ASSERT_REGHOP4;
9436 while (off-- && s < rlim) {
9437 /* XXX could check well-formedness here */
9442 while (off++ && s > llim) {
9444 if (UTF8_IS_CONTINUED(*s)) {
9445 while (s > llim && UTF8_IS_CONTINUATION(*s))
9447 if (! UTF8_IS_START(*s)) {
9448 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9451 /* XXX could check well-formedness here */
9457 /* like reghop3, but returns NULL on overrun, rather than returning last
9461 S_reghopmaybe3(U8* s, SSize_t off, const U8* const lim)
9463 PERL_ARGS_ASSERT_REGHOPMAYBE3;
9466 while (off-- && s < lim) {
9467 /* XXX could check well-formedness here */
9474 while (off++ && s > lim) {
9476 if (UTF8_IS_CONTINUED(*s)) {
9477 while (s > lim && UTF8_IS_CONTINUATION(*s))
9479 if (! UTF8_IS_START(*s)) {
9480 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9483 /* XXX could check well-formedness here */
9492 /* when executing a regex that may have (?{}), extra stuff needs setting
9493 up that will be visible to the called code, even before the current
9494 match has finished. In particular:
9496 * $_ is localised to the SV currently being matched;
9497 * pos($_) is created if necessary, ready to be updated on each call-out
9499 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
9500 isn't set until the current pattern is successfully finished), so that
9501 $1 etc of the match-so-far can be seen;
9502 * save the old values of subbeg etc of the current regex, and set then
9503 to the current string (again, this is normally only done at the end
9508 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
9511 regexp *const rex = ReANY(reginfo->prog);
9512 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
9514 eval_state->rex = rex;
9517 /* Make $_ available to executed code. */
9518 if (reginfo->sv != DEFSV) {
9520 DEFSV_set(reginfo->sv);
9523 if (!(mg = mg_find_mglob(reginfo->sv))) {
9524 /* prepare for quick setting of pos */
9525 mg = sv_magicext_mglob(reginfo->sv);
9528 eval_state->pos_magic = mg;
9529 eval_state->pos = mg->mg_len;
9530 eval_state->pos_flags = mg->mg_flags;
9533 eval_state->pos_magic = NULL;
9535 if (!PL_reg_curpm) {
9536 /* PL_reg_curpm is a fake PMOP that we can attach the current
9537 * regex to and point PL_curpm at, so that $1 et al are visible
9538 * within a /(?{})/. It's just allocated once per interpreter the
9539 * first time its needed */
9540 Newxz(PL_reg_curpm, 1, PMOP);
9543 SV* const repointer = &PL_sv_undef;
9544 /* this regexp is also owned by the new PL_reg_curpm, which
9545 will try to free it. */
9546 av_push(PL_regex_padav, repointer);
9547 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
9548 PL_regex_pad = AvARRAY(PL_regex_padav);
9552 SET_reg_curpm(reginfo->prog);
9553 eval_state->curpm = PL_curpm;
9554 PL_curpm_under = PL_curpm;
9555 PL_curpm = PL_reg_curpm;
9556 if (RXp_MATCH_COPIED(rex)) {
9557 /* Here is a serious problem: we cannot rewrite subbeg,
9558 since it may be needed if this match fails. Thus
9559 $` inside (?{}) could fail... */
9560 eval_state->subbeg = rex->subbeg;
9561 eval_state->sublen = rex->sublen;
9562 eval_state->suboffset = rex->suboffset;
9563 eval_state->subcoffset = rex->subcoffset;
9565 eval_state->saved_copy = rex->saved_copy;
9567 RXp_MATCH_COPIED_off(rex);
9570 eval_state->subbeg = NULL;
9571 rex->subbeg = (char *)reginfo->strbeg;
9573 rex->subcoffset = 0;
9574 rex->sublen = reginfo->strend - reginfo->strbeg;
9578 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
9581 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
9583 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
9584 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
9587 Safefree(aux->poscache);
9591 /* undo the effects of S_setup_eval_state() */
9593 if (eval_state->subbeg) {
9594 regexp * const rex = eval_state->rex;
9595 rex->subbeg = eval_state->subbeg;
9596 rex->sublen = eval_state->sublen;
9597 rex->suboffset = eval_state->suboffset;
9598 rex->subcoffset = eval_state->subcoffset;
9600 rex->saved_copy = eval_state->saved_copy;
9602 RXp_MATCH_COPIED_on(rex);
9604 if (eval_state->pos_magic)
9606 eval_state->pos_magic->mg_len = eval_state->pos;
9607 eval_state->pos_magic->mg_flags =
9608 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
9609 | (eval_state->pos_flags & MGf_BYTES);
9612 PL_curpm = eval_state->curpm;
9615 PL_regmatch_state = aux->old_regmatch_state;
9616 PL_regmatch_slab = aux->old_regmatch_slab;
9618 /* free all slabs above current one - this must be the last action
9619 * of this function, as aux and eval_state are allocated within
9620 * slabs and may be freed here */
9622 s = PL_regmatch_slab->next;
9624 PL_regmatch_slab->next = NULL;
9626 regmatch_slab * const osl = s;
9635 S_to_utf8_substr(pTHX_ regexp *prog)
9637 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
9638 * on the converted value */
9642 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
9645 if (prog->substrs->data[i].substr
9646 && !prog->substrs->data[i].utf8_substr) {
9647 SV* const sv = newSVsv(prog->substrs->data[i].substr);
9648 prog->substrs->data[i].utf8_substr = sv;
9649 sv_utf8_upgrade(sv);
9650 if (SvVALID(prog->substrs->data[i].substr)) {
9651 if (SvTAIL(prog->substrs->data[i].substr)) {
9652 /* Trim the trailing \n that fbm_compile added last
9654 SvCUR_set(sv, SvCUR(sv) - 1);
9655 /* Whilst this makes the SV technically "invalid" (as its
9656 buffer is no longer followed by "\0") when fbm_compile()
9657 adds the "\n" back, a "\0" is restored. */
9658 fbm_compile(sv, FBMcf_TAIL);
9662 if (prog->substrs->data[i].substr == prog->check_substr)
9663 prog->check_utf8 = sv;
9669 S_to_byte_substr(pTHX_ regexp *prog)
9671 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
9672 * on the converted value; returns FALSE if can't be converted. */
9676 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
9679 if (prog->substrs->data[i].utf8_substr
9680 && !prog->substrs->data[i].substr) {
9681 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
9682 if (! sv_utf8_downgrade(sv, TRUE)) {
9685 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
9686 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
9687 /* Trim the trailing \n that fbm_compile added last
9689 SvCUR_set(sv, SvCUR(sv) - 1);
9690 fbm_compile(sv, FBMcf_TAIL);
9694 prog->substrs->data[i].substr = sv;
9695 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
9696 prog->check_substr = sv;
9704 * ex: set ts=8 sts=4 sw=4 et: