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))
123 #define HOPc(pos,off) \
124 (char *)(reginfo->is_utf8_target \
125 ? reghop3((U8*)pos, off, \
126 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
129 #define HOPBACKc(pos, off) \
130 (char*)(reginfo->is_utf8_target \
131 ? reghopmaybe3((U8*)pos, (SSize_t)0-off, (U8*)(reginfo->strbeg)) \
132 : (pos - off >= reginfo->strbeg) \
136 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
137 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
139 /* lim must be +ve. Returns NULL on overshoot */
140 #define HOPMAYBE3(pos,off,lim) \
141 (reginfo->is_utf8_target \
142 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
143 : ((U8*)pos + off <= lim) \
147 /* like HOP3, but limits the result to <= lim even for the non-utf8 case.
148 * off must be >=0; args should be vars rather than expressions */
149 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
150 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
151 : (U8*)((pos + off) > lim ? lim : (pos + off)))
152 #define HOP3clim(pos,off,lim) ((char*)HOP3lim(pos,off,lim))
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. */
448 #ifndef PERL_IN_XSUB_RE
451 Perl_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
453 /* Returns a boolean as to whether or not 'character' is a member of the
454 * Posix character class given by 'classnum' that should be equivalent to a
455 * value in the typedef '_char_class_number'.
457 * Ideally this could be replaced by a just an array of function pointers
458 * to the C library functions that implement the macros this calls.
459 * However, to compile, the precise function signatures are required, and
460 * these may vary from platform to to platform. To avoid having to figure
461 * out what those all are on each platform, I (khw) am using this method,
462 * which adds an extra layer of function call overhead (unless the C
463 * optimizer strips it away). But we don't particularly care about
464 * performance with locales anyway. */
466 switch ((_char_class_number) classnum) {
467 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
468 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
469 case _CC_ENUM_ASCII: return isASCII_LC(character);
470 case _CC_ENUM_BLANK: return isBLANK_LC(character);
471 case _CC_ENUM_CASED: return isLOWER_LC(character)
472 || isUPPER_LC(character);
473 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
474 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
475 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
476 case _CC_ENUM_LOWER: return isLOWER_LC(character);
477 case _CC_ENUM_PRINT: return isPRINT_LC(character);
478 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
479 case _CC_ENUM_SPACE: return isSPACE_LC(character);
480 case _CC_ENUM_UPPER: return isUPPER_LC(character);
481 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
482 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
483 default: /* VERTSPACE should never occur in locales */
484 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
487 NOT_REACHED; /* NOTREACHED */
494 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
496 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
497 * 'character' is a member of the Posix character class given by 'classnum'
498 * that should be equivalent to a value in the typedef
499 * '_char_class_number'.
501 * This just calls isFOO_lc on the code point for the character if it is in
502 * the range 0-255. Outside that range, all characters use Unicode
503 * rules, ignoring any locale. So use the Unicode function if this class
504 * requires a swash, and use the Unicode macro otherwise. */
506 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
508 if (UTF8_IS_INVARIANT(*character)) {
509 return isFOO_lc(classnum, *character);
511 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
512 return isFOO_lc(classnum,
513 EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1)));
516 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, character + UTF8SKIP(character));
518 if (classnum < _FIRST_NON_SWASH_CC) {
520 /* Initialize the swash unless done already */
521 if (! PL_utf8_swash_ptrs[classnum]) {
522 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
523 PL_utf8_swash_ptrs[classnum] =
524 _core_swash_init("utf8",
527 PL_XPosix_ptrs[classnum], &flags);
530 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
532 TRUE /* is UTF */ ));
535 switch ((_char_class_number) classnum) {
536 case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character);
537 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
538 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
539 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
543 return FALSE; /* Things like CNTRL are always below 256 */
547 * pregexec and friends
550 #ifndef PERL_IN_XSUB_RE
552 - pregexec - match a regexp against a string
555 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
556 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
557 /* stringarg: the point in the string at which to begin matching */
558 /* strend: pointer to null at end of string */
559 /* strbeg: real beginning of string */
560 /* minend: end of match must be >= minend bytes after stringarg. */
561 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
562 * itself is accessed via the pointers above */
563 /* nosave: For optimizations. */
565 PERL_ARGS_ASSERT_PREGEXEC;
568 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
569 nosave ? 0 : REXEC_COPY_STR);
575 /* re_intuit_start():
577 * Based on some optimiser hints, try to find the earliest position in the
578 * string where the regex could match.
580 * rx: the regex to match against
581 * sv: the SV being matched: only used for utf8 flag; the string
582 * itself is accessed via the pointers below. Note that on
583 * something like an overloaded SV, SvPOK(sv) may be false
584 * and the string pointers may point to something unrelated to
586 * strbeg: real beginning of string
587 * strpos: the point in the string at which to begin matching
588 * strend: pointer to the byte following the last char of the string
589 * flags currently unused; set to 0
590 * data: currently unused; set to NULL
592 * The basic idea of re_intuit_start() is to use some known information
593 * about the pattern, namely:
595 * a) the longest known anchored substring (i.e. one that's at a
596 * constant offset from the beginning of the pattern; but not
597 * necessarily at a fixed offset from the beginning of the
599 * b) the longest floating substring (i.e. one that's not at a constant
600 * offset from the beginning of the pattern);
601 * c) Whether the pattern is anchored to the string; either
602 * an absolute anchor: /^../, or anchored to \n: /^.../m,
603 * or anchored to pos(): /\G/;
604 * d) A start class: a real or synthetic character class which
605 * represents which characters are legal at the start of the pattern;
607 * to either quickly reject the match, or to find the earliest position
608 * within the string at which the pattern might match, thus avoiding
609 * running the full NFA engine at those earlier locations, only to
610 * eventually fail and retry further along.
612 * Returns NULL if the pattern can't match, or returns the address within
613 * the string which is the earliest place the match could occur.
615 * The longest of the anchored and floating substrings is called 'check'
616 * and is checked first. The other is called 'other' and is checked
617 * second. The 'other' substring may not be present. For example,
619 * /(abc|xyz)ABC\d{0,3}DEFG/
623 * check substr (float) = "DEFG", offset 6..9 chars
624 * other substr (anchored) = "ABC", offset 3..3 chars
627 * Be aware that during the course of this function, sometimes 'anchored'
628 * refers to a substring being anchored relative to the start of the
629 * pattern, and sometimes to the pattern itself being anchored relative to
630 * the string. For example:
632 * /\dabc/: "abc" is anchored to the pattern;
633 * /^\dabc/: "abc" is anchored to the pattern and the string;
634 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
635 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
636 * but the pattern is anchored to the string.
640 Perl_re_intuit_start(pTHX_
643 const char * const strbeg,
647 re_scream_pos_data *data)
649 struct regexp *const prog = ReANY(rx);
650 SSize_t start_shift = prog->check_offset_min;
651 /* Should be nonnegative! */
652 SSize_t end_shift = 0;
653 /* current lowest pos in string where the regex can start matching */
654 char *rx_origin = strpos;
656 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
657 U8 other_ix = 1 - prog->substrs->check_ix;
659 char *other_last = strpos;/* latest pos 'other' substr already checked to */
660 char *check_at = NULL; /* check substr found at this pos */
661 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
662 RXi_GET_DECL(prog,progi);
663 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
664 regmatch_info *const reginfo = ®info_buf;
665 GET_RE_DEBUG_FLAGS_DECL;
667 PERL_ARGS_ASSERT_RE_INTUIT_START;
668 PERL_UNUSED_ARG(flags);
669 PERL_UNUSED_ARG(data);
671 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
672 "Intuit: trying to determine minimum start position...\n"));
674 /* for now, assume that all substr offsets are positive. If at some point
675 * in the future someone wants to do clever things with lookbehind and
676 * -ve offsets, they'll need to fix up any code in this function
677 * which uses these offsets. See the thread beginning
678 * <20140113145929.GF27210@iabyn.com>
680 assert(prog->substrs->data[0].min_offset >= 0);
681 assert(prog->substrs->data[0].max_offset >= 0);
682 assert(prog->substrs->data[1].min_offset >= 0);
683 assert(prog->substrs->data[1].max_offset >= 0);
684 assert(prog->substrs->data[2].min_offset >= 0);
685 assert(prog->substrs->data[2].max_offset >= 0);
687 /* for now, assume that if both present, that the floating substring
688 * doesn't start before the anchored substring.
689 * If you break this assumption (e.g. doing better optimisations
690 * with lookahead/behind), then you'll need to audit the code in this
691 * function carefully first
694 ! ( (prog->anchored_utf8 || prog->anchored_substr)
695 && (prog->float_utf8 || prog->float_substr))
696 || (prog->float_min_offset >= prog->anchored_offset));
698 /* byte rather than char calculation for efficiency. It fails
699 * to quickly reject some cases that can't match, but will reject
700 * them later after doing full char arithmetic */
701 if (prog->minlen > strend - strpos) {
702 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
703 " String too short...\n"));
707 RX_MATCH_UTF8_set(rx,utf8_target);
708 reginfo->is_utf8_target = cBOOL(utf8_target);
709 reginfo->info_aux = NULL;
710 reginfo->strbeg = strbeg;
711 reginfo->strend = strend;
712 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
714 /* not actually used within intuit, but zero for safety anyway */
715 reginfo->poscache_maxiter = 0;
718 if ((!prog->anchored_utf8 && prog->anchored_substr)
719 || (!prog->float_utf8 && prog->float_substr))
720 to_utf8_substr(prog);
721 check = prog->check_utf8;
723 if (!prog->check_substr && prog->check_utf8) {
724 if (! to_byte_substr(prog)) {
725 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
728 check = prog->check_substr;
731 /* dump the various substring data */
732 DEBUG_OPTIMISE_MORE_r({
734 for (i=0; i<=2; i++) {
735 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
736 : prog->substrs->data[i].substr);
740 Perl_re_printf( aTHX_
741 " substrs[%d]: min=%" IVdf " max=%" IVdf " end shift=%" IVdf
742 " useful=%" IVdf " utf8=%d [%s]\n",
744 (IV)prog->substrs->data[i].min_offset,
745 (IV)prog->substrs->data[i].max_offset,
746 (IV)prog->substrs->data[i].end_shift,
753 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
755 /* ml_anch: check after \n?
757 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
758 * with /.*.../, these flags will have been added by the
760 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
761 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
763 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
764 && !(prog->intflags & PREGf_IMPLICIT);
766 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
767 /* we are only allowed to match at BOS or \G */
769 /* trivially reject if there's a BOS anchor and we're not at BOS.
771 * Note that we don't try to do a similar quick reject for
772 * \G, since generally the caller will have calculated strpos
773 * based on pos() and gofs, so the string is already correctly
774 * anchored by definition; and handling the exceptions would
775 * be too fiddly (e.g. REXEC_IGNOREPOS).
777 if ( strpos != strbeg
778 && (prog->intflags & PREGf_ANCH_SBOL))
780 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
781 " Not at start...\n"));
785 /* in the presence of an anchor, the anchored (relative to the
786 * start of the regex) substr must also be anchored relative
787 * to strpos. So quickly reject if substr isn't found there.
788 * This works for \G too, because the caller will already have
789 * subtracted gofs from pos, and gofs is the offset from the
790 * \G to the start of the regex. For example, in /.abc\Gdef/,
791 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
792 * caller will have set strpos=pos()-4; we look for the substr
793 * at position pos()-4+1, which lines up with the "a" */
795 if (prog->check_offset_min == prog->check_offset_max) {
796 /* Substring at constant offset from beg-of-str... */
797 SSize_t slen = SvCUR(check);
798 char *s = HOP3c(strpos, prog->check_offset_min, strend);
800 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
801 " Looking for check substr at fixed offset %" IVdf "...\n",
802 (IV)prog->check_offset_min));
805 /* In this case, the regex is anchored at the end too.
806 * Unless it's a multiline match, the lengths must match
807 * exactly, give or take a \n. NB: slen >= 1 since
808 * the last char of check is \n */
810 && ( strend - s > slen
811 || strend - s < slen - 1
812 || (strend - s == slen && strend[-1] != '\n')))
814 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
815 " String too long...\n"));
818 /* Now should match s[0..slen-2] */
821 if (slen && (strend - s < slen
822 || *SvPVX_const(check) != *s
823 || (slen > 1 && (memNE(SvPVX_const(check), s, slen)))))
825 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
826 " String not equal...\n"));
831 goto success_at_start;
836 end_shift = prog->check_end_shift;
838 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
840 Perl_croak(aTHX_ "panic: end_shift: %" IVdf " pattern:\n%s\n ",
841 (IV)end_shift, RX_PRECOMP(prog));
846 /* This is the (re)entry point of the main loop in this function.
847 * The goal of this loop is to:
848 * 1) find the "check" substring in the region rx_origin..strend
849 * (adjusted by start_shift / end_shift). If not found, reject
851 * 2) If it exists, look for the "other" substr too if defined; for
852 * example, if the check substr maps to the anchored substr, then
853 * check the floating substr, and vice-versa. If not found, go
854 * back to (1) with rx_origin suitably incremented.
855 * 3) If we find an rx_origin position that doesn't contradict
856 * either of the substrings, then check the possible additional
857 * constraints on rx_origin of /^.../m or a known start class.
858 * If these fail, then depending on which constraints fail, jump
859 * back to here, or to various other re-entry points further along
860 * that skip some of the first steps.
861 * 4) If we pass all those tests, update the BmUSEFUL() count on the
862 * substring. If the start position was determined to be at the
863 * beginning of the string - so, not rejected, but not optimised,
864 * since we have to run regmatch from position 0 - decrement the
865 * BmUSEFUL() count. Otherwise increment it.
869 /* first, look for the 'check' substring */
875 DEBUG_OPTIMISE_MORE_r({
876 Perl_re_printf( aTHX_
877 " At restart: rx_origin=%" IVdf " Check offset min: %" IVdf
878 " Start shift: %" IVdf " End shift %" IVdf
879 " Real end Shift: %" IVdf "\n",
880 (IV)(rx_origin - strbeg),
881 (IV)prog->check_offset_min,
884 (IV)prog->check_end_shift);
887 end_point = HOP3(strend, -end_shift, strbeg);
888 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
893 /* If the regex is absolutely anchored to either the start of the
894 * string (SBOL) or to pos() (ANCH_GPOS), then
895 * check_offset_max represents an upper bound on the string where
896 * the substr could start. For the ANCH_GPOS case, we assume that
897 * the caller of intuit will have already set strpos to
898 * pos()-gofs, so in this case strpos + offset_max will still be
899 * an upper bound on the substr.
902 && prog->intflags & PREGf_ANCH
903 && prog->check_offset_max != SSize_t_MAX)
905 SSize_t len = SvCUR(check) - !!SvTAIL(check);
906 const char * const anchor =
907 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
909 /* do a bytes rather than chars comparison. It's conservative;
910 * so it skips doing the HOP if the result can't possibly end
911 * up earlier than the old value of end_point.
913 if ((char*)end_point - anchor > prog->check_offset_max) {
914 end_point = HOP3lim((U8*)anchor,
915 prog->check_offset_max,
921 check_at = fbm_instr( start_point, end_point,
922 check, multiline ? FBMrf_MULTILINE : 0);
924 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
925 " doing 'check' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
926 (IV)((char*)start_point - strbeg),
927 (IV)((char*)end_point - strbeg),
928 (IV)(check_at ? check_at - strbeg : -1)
931 /* Update the count-of-usability, remove useless subpatterns,
935 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
936 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
937 Perl_re_printf( aTHX_ " %s %s substr %s%s%s",
938 (check_at ? "Found" : "Did not find"),
939 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
940 ? "anchored" : "floating"),
943 (check_at ? " at offset " : "...\n") );
948 /* set rx_origin to the minimum position where the regex could start
949 * matching, given the constraint of the just-matched check substring.
950 * But don't set it lower than previously.
953 if (check_at - rx_origin > prog->check_offset_max)
954 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
955 /* Finish the diagnostic message */
956 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
957 "%ld (rx_origin now %" IVdf ")...\n",
958 (long)(check_at - strbeg),
959 (IV)(rx_origin - strbeg)
964 /* now look for the 'other' substring if defined */
966 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
967 : prog->substrs->data[other_ix].substr)
969 /* Take into account the "other" substring. */
973 struct reg_substr_datum *other;
976 other = &prog->substrs->data[other_ix];
978 /* if "other" is anchored:
979 * we've previously found a floating substr starting at check_at.
980 * This means that the regex origin must lie somewhere
981 * between min (rx_origin): HOP3(check_at, -check_offset_max)
982 * and max: HOP3(check_at, -check_offset_min)
983 * (except that min will be >= strpos)
984 * So the fixed substr must lie somewhere between
985 * HOP3(min, anchored_offset)
986 * HOP3(max, anchored_offset) + SvCUR(substr)
989 /* if "other" is floating
990 * Calculate last1, the absolute latest point where the
991 * floating substr could start in the string, ignoring any
992 * constraints from the earlier fixed match. It is calculated
995 * strend - prog->minlen (in chars) is the absolute latest
996 * position within the string where the origin of the regex
997 * could appear. The latest start point for the floating
998 * substr is float_min_offset(*) on from the start of the
999 * regex. last1 simply combines thee two offsets.
1001 * (*) You might think the latest start point should be
1002 * float_max_offset from the regex origin, and technically
1003 * you'd be correct. However, consider
1005 * Here, float min, max are 3,5 and minlen is 7.
1006 * This can match either
1010 * In the first case, the regex matches minlen chars; in the
1011 * second, minlen+1, in the third, minlen+2.
1012 * In the first case, the floating offset is 3 (which equals
1013 * float_min), in the second, 4, and in the third, 5 (which
1014 * equals float_max). In all cases, the floating string bcd
1015 * can never start more than 4 chars from the end of the
1016 * string, which equals minlen - float_min. As the substring
1017 * starts to match more than float_min from the start of the
1018 * regex, it makes the regex match more than minlen chars,
1019 * and the two cancel each other out. So we can always use
1020 * float_min - minlen, rather than float_max - minlen for the
1021 * latest position in the string.
1023 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1024 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1027 assert(prog->minlen >= other->min_offset);
1028 last1 = HOP3c(strend,
1029 other->min_offset - prog->minlen, strbeg);
1031 if (other_ix) {/* i.e. if (other-is-float) */
1032 /* last is the latest point where the floating substr could
1033 * start, *given* any constraints from the earlier fixed
1034 * match. This constraint is that the floating string starts
1035 * <= float_max_offset chars from the regex origin (rx_origin).
1036 * If this value is less than last1, use it instead.
1038 assert(rx_origin <= last1);
1040 /* this condition handles the offset==infinity case, and
1041 * is a short-cut otherwise. Although it's comparing a
1042 * byte offset to a char length, it does so in a safe way,
1043 * since 1 char always occupies 1 or more bytes,
1044 * so if a string range is (last1 - rx_origin) bytes,
1045 * it will be less than or equal to (last1 - rx_origin)
1046 * chars; meaning it errs towards doing the accurate HOP3
1047 * rather than just using last1 as a short-cut */
1048 (last1 - rx_origin) < other->max_offset
1050 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1053 assert(strpos + start_shift <= check_at);
1054 last = HOP4c(check_at, other->min_offset - start_shift,
1058 s = HOP3c(rx_origin, other->min_offset, strend);
1059 if (s < other_last) /* These positions already checked */
1062 must = utf8_target ? other->utf8_substr : other->substr;
1063 assert(SvPOK(must));
1066 char *to = last + SvCUR(must) - (SvTAIL(must)!=0);
1072 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1073 " skipping 'other' fbm scan: %" IVdf " > %" IVdf "\n",
1074 (IV)(from - strbeg),
1080 (unsigned char*)from,
1083 multiline ? FBMrf_MULTILINE : 0
1085 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1086 " doing 'other' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
1087 (IV)(from - strbeg),
1089 (IV)(s ? s - strbeg : -1)
1095 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1096 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1097 Perl_re_printf( aTHX_ " %s %s substr %s%s",
1098 s ? "Found" : "Contradicts",
1099 other_ix ? "floating" : "anchored",
1100 quoted, RE_SV_TAIL(must));
1105 /* last1 is latest possible substr location. If we didn't
1106 * find it before there, we never will */
1107 if (last >= last1) {
1108 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1109 "; giving up...\n"));
1113 /* try to find the check substr again at a later
1114 * position. Maybe next time we'll find the "other" substr
1116 other_last = HOP3c(last, 1, strend) /* highest failure */;
1118 other_ix /* i.e. if other-is-float */
1119 ? HOP3c(rx_origin, 1, strend)
1120 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1121 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1122 "; about to retry %s at offset %ld (rx_origin now %" IVdf ")...\n",
1123 (other_ix ? "floating" : "anchored"),
1124 (long)(HOP3c(check_at, 1, strend) - strbeg),
1125 (IV)(rx_origin - strbeg)
1130 if (other_ix) { /* if (other-is-float) */
1131 /* other_last is set to s, not s+1, since its possible for
1132 * a floating substr to fail first time, then succeed
1133 * second time at the same floating position; e.g.:
1134 * "-AB--AABZ" =~ /\wAB\d*Z/
1135 * The first time round, anchored and float match at
1136 * "-(AB)--AAB(Z)" then fail on the initial \w character
1137 * class. Second time round, they match at "-AB--A(AB)(Z)".
1142 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1143 other_last = HOP3c(s, 1, strend);
1145 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1146 " at offset %ld (rx_origin now %" IVdf ")...\n",
1148 (IV)(rx_origin - strbeg)
1154 DEBUG_OPTIMISE_MORE_r(
1155 Perl_re_printf( aTHX_
1156 " Check-only match: offset min:%" IVdf " max:%" IVdf
1157 " check_at:%" IVdf " rx_origin:%" IVdf " rx_origin-check_at:%" IVdf
1158 " strend:%" IVdf "\n",
1159 (IV)prog->check_offset_min,
1160 (IV)prog->check_offset_max,
1161 (IV)(check_at-strbeg),
1162 (IV)(rx_origin-strbeg),
1163 (IV)(rx_origin-check_at),
1169 postprocess_substr_matches:
1171 /* handle the extra constraint of /^.../m if present */
1173 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1176 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1177 " looking for /^/m anchor"));
1179 /* we have failed the constraint of a \n before rx_origin.
1180 * Find the next \n, if any, even if it's beyond the current
1181 * anchored and/or floating substrings. Whether we should be
1182 * scanning ahead for the next \n or the next substr is debatable.
1183 * On the one hand you'd expect rare substrings to appear less
1184 * often than \n's. On the other hand, searching for \n means
1185 * we're effectively flipping between check_substr and "\n" on each
1186 * iteration as the current "rarest" string candidate, which
1187 * means for example that we'll quickly reject the whole string if
1188 * hasn't got a \n, rather than trying every substr position
1192 s = HOP3c(strend, - prog->minlen, strpos);
1193 if (s <= rx_origin ||
1194 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1196 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1197 " Did not find /%s^%s/m...\n",
1198 PL_colors[0], PL_colors[1]));
1202 /* earliest possible origin is 1 char after the \n.
1203 * (since *rx_origin == '\n', it's safe to ++ here rather than
1204 * HOP(rx_origin, 1)) */
1207 if (prog->substrs->check_ix == 0 /* check is anchored */
1208 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1210 /* Position contradicts check-string; either because
1211 * check was anchored (and thus has no wiggle room),
1212 * or check was float and rx_origin is above the float range */
1213 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1214 " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
1215 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1219 /* if we get here, the check substr must have been float,
1220 * is in range, and we may or may not have had an anchored
1221 * "other" substr which still contradicts */
1222 assert(prog->substrs->check_ix); /* check is float */
1224 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1225 /* whoops, the anchored "other" substr exists, so we still
1226 * contradict. On the other hand, the float "check" substr
1227 * didn't contradict, so just retry the anchored "other"
1229 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1230 " Found /%s^%s/m, rescanning for anchored from offset %" IVdf " (rx_origin now %" IVdf ")...\n",
1231 PL_colors[0], PL_colors[1],
1232 (IV)(rx_origin - strbeg + prog->anchored_offset),
1233 (IV)(rx_origin - strbeg)
1235 goto do_other_substr;
1238 /* success: we don't contradict the found floating substring
1239 * (and there's no anchored substr). */
1240 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1241 " Found /%s^%s/m with rx_origin %ld...\n",
1242 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1245 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1246 " (multiline anchor test skipped)\n"));
1252 /* if we have a starting character class, then test that extra constraint.
1253 * (trie stclasses are too expensive to use here, we are better off to
1254 * leave it to regmatch itself) */
1256 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1257 const U8* const str = (U8*)STRING(progi->regstclass);
1259 /* XXX this value could be pre-computed */
1260 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1261 ? (reginfo->is_utf8_pat
1262 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1263 : STR_LEN(progi->regstclass))
1267 /* latest pos that a matching float substr constrains rx start to */
1268 char *rx_max_float = NULL;
1270 /* if the current rx_origin is anchored, either by satisfying an
1271 * anchored substring constraint, or a /^.../m constraint, then we
1272 * can reject the current origin if the start class isn't found
1273 * at the current position. If we have a float-only match, then
1274 * rx_origin is constrained to a range; so look for the start class
1275 * in that range. if neither, then look for the start class in the
1276 * whole rest of the string */
1278 /* XXX DAPM it's not clear what the minlen test is for, and why
1279 * it's not used in the floating case. Nothing in the test suite
1280 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1281 * Here are some old comments, which may or may not be correct:
1283 * minlen == 0 is possible if regstclass is \b or \B,
1284 * and the fixed substr is ''$.
1285 * Since minlen is already taken into account, rx_origin+1 is
1286 * before strend; accidentally, minlen >= 1 guaranties no false
1287 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1288 * 0) below assumes that regstclass does not come from lookahead...
1289 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1290 * This leaves EXACTF-ish only, which are dealt with in
1294 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1295 endpos = HOP3clim(rx_origin, (prog->minlen ? cl_l : 0), strend);
1296 else if (prog->float_substr || prog->float_utf8) {
1297 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1298 endpos = HOP3clim(rx_max_float, cl_l, strend);
1303 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1304 " looking for class: start_shift: %" IVdf " check_at: %" IVdf
1305 " rx_origin: %" IVdf " endpos: %" IVdf "\n",
1306 (IV)start_shift, (IV)(check_at - strbeg),
1307 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1309 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1312 if (endpos == strend) {
1313 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1314 " Could not match STCLASS...\n") );
1317 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1318 " This position contradicts STCLASS...\n") );
1319 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1320 && !(prog->intflags & PREGf_IMPLICIT))
1323 /* Contradict one of substrings */
1324 if (prog->anchored_substr || prog->anchored_utf8) {
1325 if (prog->substrs->check_ix == 1) { /* check is float */
1326 /* Have both, check_string is floating */
1327 assert(rx_origin + start_shift <= check_at);
1328 if (rx_origin + start_shift != check_at) {
1329 /* not at latest position float substr could match:
1330 * Recheck anchored substring, but not floating.
1331 * The condition above is in bytes rather than
1332 * chars for efficiency. It's conservative, in
1333 * that it errs on the side of doing 'goto
1334 * do_other_substr'. In this case, at worst,
1335 * an extra anchored search may get done, but in
1336 * practice the extra fbm_instr() is likely to
1337 * get skipped anyway. */
1338 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1339 " about to retry anchored at offset %ld (rx_origin now %" IVdf ")...\n",
1340 (long)(other_last - strbeg),
1341 (IV)(rx_origin - strbeg)
1343 goto do_other_substr;
1351 /* In the presence of ml_anch, we might be able to
1352 * find another \n without breaking the current float
1355 /* strictly speaking this should be HOP3c(..., 1, ...),
1356 * but since we goto a block of code that's going to
1357 * search for the next \n if any, its safe here */
1359 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1360 " about to look for /%s^%s/m starting at rx_origin %ld...\n",
1361 PL_colors[0], PL_colors[1],
1362 (long)(rx_origin - strbeg)) );
1363 goto postprocess_substr_matches;
1366 /* strictly speaking this can never be true; but might
1367 * be if we ever allow intuit without substrings */
1368 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1371 rx_origin = rx_max_float;
1374 /* at this point, any matching substrings have been
1375 * contradicted. Start again... */
1377 rx_origin = HOP3c(rx_origin, 1, strend);
1379 /* uses bytes rather than char calculations for efficiency.
1380 * It's conservative: it errs on the side of doing 'goto restart',
1381 * where there is code that does a proper char-based test */
1382 if (rx_origin + start_shift + end_shift > strend) {
1383 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1384 " Could not match STCLASS...\n") );
1387 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1388 " about to look for %s substr starting at offset %ld (rx_origin now %" IVdf ")...\n",
1389 (prog->substrs->check_ix ? "floating" : "anchored"),
1390 (long)(rx_origin + start_shift - strbeg),
1391 (IV)(rx_origin - strbeg)
1398 if (rx_origin != s) {
1399 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1400 " By STCLASS: moving %ld --> %ld\n",
1401 (long)(rx_origin - strbeg), (long)(s - strbeg))
1405 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1406 " Does not contradict STCLASS...\n");
1411 /* Decide whether using the substrings helped */
1413 if (rx_origin != strpos) {
1414 /* Fixed substring is found far enough so that the match
1415 cannot start at strpos. */
1417 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " try at offset...\n"));
1418 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1421 /* The found rx_origin position does not prohibit matching at
1422 * strpos, so calling intuit didn't gain us anything. Decrement
1423 * the BmUSEFUL() count on the check substring, and if we reach
1425 if (!(prog->intflags & PREGf_NAUGHTY)
1427 prog->check_utf8 /* Could be deleted already */
1428 && --BmUSEFUL(prog->check_utf8) < 0
1429 && (prog->check_utf8 == prog->float_utf8)
1431 prog->check_substr /* Could be deleted already */
1432 && --BmUSEFUL(prog->check_substr) < 0
1433 && (prog->check_substr == prog->float_substr)
1436 /* If flags & SOMETHING - do not do it many times on the same match */
1437 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " ... Disabling check substring...\n"));
1438 /* XXX Does the destruction order has to change with utf8_target? */
1439 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1440 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1441 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1442 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1443 check = NULL; /* abort */
1444 /* XXXX This is a remnant of the old implementation. It
1445 looks wasteful, since now INTUIT can use many
1446 other heuristics. */
1447 prog->extflags &= ~RXf_USE_INTUIT;
1451 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1452 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1453 PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );
1457 fail_finish: /* Substring not found */
1458 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1459 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1461 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch rejected by optimizer%s\n",
1462 PL_colors[4], PL_colors[5]));
1467 #define DECL_TRIE_TYPE(scan) \
1468 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1469 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1470 trie_utf8l, trie_flu8 } \
1471 trie_type = ((scan->flags == EXACT) \
1472 ? (utf8_target ? trie_utf8 : trie_plain) \
1473 : (scan->flags == EXACTL) \
1474 ? (utf8_target ? trie_utf8l : trie_plain) \
1475 : (scan->flags == EXACTFA) \
1477 ? trie_utf8_exactfa_fold \
1478 : trie_latin_utf8_exactfa_fold) \
1479 : (scan->flags == EXACTFLU8 \
1483 : trie_latin_utf8_fold)))
1485 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1488 U8 flags = FOLD_FLAGS_FULL; \
1489 switch (trie_type) { \
1491 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1492 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1493 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1495 goto do_trie_utf8_fold; \
1496 case trie_utf8_exactfa_fold: \
1497 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1499 case trie_utf8_fold: \
1500 do_trie_utf8_fold: \
1501 if ( foldlen>0 ) { \
1502 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1507 len = UTF8SKIP(uc); \
1508 uvc = _toFOLD_utf8_flags( (const U8*) uc, uc + len, foldbuf, &foldlen, \
1510 skiplen = UVCHR_SKIP( uvc ); \
1511 foldlen -= skiplen; \
1512 uscan = foldbuf + skiplen; \
1515 case trie_latin_utf8_exactfa_fold: \
1516 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1518 case trie_latin_utf8_fold: \
1519 if ( foldlen>0 ) { \
1520 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1526 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1527 skiplen = UVCHR_SKIP( uvc ); \
1528 foldlen -= skiplen; \
1529 uscan = foldbuf + skiplen; \
1533 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1534 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1535 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1539 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1546 charid = trie->charmap[ uvc ]; \
1550 if (widecharmap) { \
1551 SV** const svpp = hv_fetch(widecharmap, \
1552 (char*)&uvc, sizeof(UV), 0); \
1554 charid = (U16)SvIV(*svpp); \
1559 #define DUMP_EXEC_POS(li,s,doutf8,depth) \
1560 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1561 startpos, doutf8, depth)
1563 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1567 && (ln == 1 || folder(s, pat_string, ln)) \
1568 && (reginfo->intuit || regtry(reginfo, &s)) )\
1574 #define REXEC_FBC_UTF8_SCAN(CODE) \
1576 while (s < strend) { \
1582 #define REXEC_FBC_SCAN(CODE) \
1584 while (s < strend) { \
1590 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1591 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1593 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1602 #define REXEC_FBC_CLASS_SCAN(COND) \
1603 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1605 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1614 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1615 if (utf8_target) { \
1616 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1619 REXEC_FBC_CLASS_SCAN(COND); \
1622 /* The three macros below are slightly different versions of the same logic.
1624 * The first is for /a and /aa when the target string is UTF-8. This can only
1625 * match ascii, but it must advance based on UTF-8. The other two handle the
1626 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1627 * for the boundary (or non-boundary) between a word and non-word character.
1628 * The utf8 and non-utf8 cases have the same logic, but the details must be
1629 * different. Find the "wordness" of the character just prior to this one, and
1630 * compare it with the wordness of this one. If they differ, we have a
1631 * boundary. At the beginning of the string, pretend that the previous
1632 * character was a new-line.
1634 * All these macros uncleanly have side-effects with each other and outside
1635 * variables. So far it's been too much trouble to clean-up
1637 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1638 * a word character or not.
1639 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1641 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1643 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1644 * are looking for a boundary or for a non-boundary. If we are looking for a
1645 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1646 * see if this tentative match actually works, and if so, to quit the loop
1647 * here. And vice-versa if we are looking for a non-boundary.
1649 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1650 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1651 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1652 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1653 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1654 * complement. But in that branch we complement tmp, meaning that at the
1655 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1656 * which means at the top of the loop in the next iteration, it is
1657 * TEST_NON_UTF8(s-1) */
1658 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1659 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1660 tmp = TEST_NON_UTF8(tmp); \
1661 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1662 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1664 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1671 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1672 * TEST_UTF8 is a macro that for the same input code points returns identically
1673 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1674 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1675 if (s == reginfo->strbeg) { \
1678 else { /* Back-up to the start of the previous character */ \
1679 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1680 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1681 0, UTF8_ALLOW_DEFAULT); \
1683 tmp = TEST_UV(tmp); \
1684 LOAD_UTF8_CHARCLASS_ALNUM(); \
1685 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1686 if (tmp == ! (TEST_UTF8((U8 *) s, (U8 *) reginfo->strend))) { \
1695 /* Like the above two macros. UTF8_CODE is the complete code for handling
1696 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1698 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1699 if (utf8_target) { \
1702 else { /* Not utf8 */ \
1703 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1704 tmp = TEST_NON_UTF8(tmp); \
1705 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1706 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1715 /* Here, things have been set up by the previous code so that tmp is the \
1716 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1717 * utf8ness of the target). We also have to check if this matches against \
1718 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1719 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1721 if (tmp == ! TEST_NON_UTF8('\n')) { \
1728 /* This is the macro to use when we want to see if something that looks like it
1729 * could match, actually does, and if so exits the loop */
1730 #define REXEC_FBC_TRYIT \
1731 if ((reginfo->intuit || regtry(reginfo, &s))) \
1734 /* The only difference between the BOUND and NBOUND cases is that
1735 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1736 * NBOUND. This is accomplished by passing it as either the if or else clause,
1737 * with the other one being empty (PLACEHOLDER is defined as empty).
1739 * The TEST_FOO parameters are for operating on different forms of input, but
1740 * all should be ones that return identically for the same underlying code
1742 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1744 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1745 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1747 #define FBC_BOUND_A(TEST_NON_UTF8) \
1749 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1750 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1752 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1754 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1755 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1757 #define FBC_NBOUND_A(TEST_NON_UTF8) \
1759 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1760 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1764 S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) {
1765 IV cp_out = Perl__invlist_search(invlist, cp_in);
1766 assert(cp_out >= 0);
1769 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
1770 invmap[S_get_break_val_cp_checked(invlist, cp)]
1772 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
1773 invmap[_invlist_search(invlist, cp)]
1776 /* Takes a pointer to an inversion list, a pointer to its corresponding
1777 * inversion map, and a code point, and returns the code point's value
1778 * according to the two arrays. It assumes that all code points have a value.
1779 * This is used as the base macro for macros for particular properties */
1780 #define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \
1781 _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp)
1783 /* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
1784 * of a code point, returning the value for the first code point in the string.
1785 * And it takes the particular macro name that finds the desired value given a
1786 * code point. Merely convert the UTF-8 to code point and call the cp macro */
1787 #define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \
1788 (__ASSERT_(pos < strend) \
1789 /* Note assumes is valid UTF-8 */ \
1790 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))
1792 /* Returns the GCB value for the input code point */
1793 #define getGCB_VAL_CP(cp) \
1794 _generic_GET_BREAK_VAL_CP( \
1799 /* Returns the GCB value for the first code point in the UTF-8 encoded string
1800 * bounded by pos and strend */
1801 #define getGCB_VAL_UTF8(pos, strend) \
1802 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)
1804 /* Returns the LB value for the input code point */
1805 #define getLB_VAL_CP(cp) \
1806 _generic_GET_BREAK_VAL_CP( \
1811 /* Returns the LB value for the first code point in the UTF-8 encoded string
1812 * bounded by pos and strend */
1813 #define getLB_VAL_UTF8(pos, strend) \
1814 _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend)
1817 /* Returns the SB value for the input code point */
1818 #define getSB_VAL_CP(cp) \
1819 _generic_GET_BREAK_VAL_CP( \
1824 /* Returns the SB value for the first code point in the UTF-8 encoded string
1825 * bounded by pos and strend */
1826 #define getSB_VAL_UTF8(pos, strend) \
1827 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)
1829 /* Returns the WB value for the input code point */
1830 #define getWB_VAL_CP(cp) \
1831 _generic_GET_BREAK_VAL_CP( \
1836 /* Returns the WB value for the first code point in the UTF-8 encoded string
1837 * bounded by pos and strend */
1838 #define getWB_VAL_UTF8(pos, strend) \
1839 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)
1841 /* We know what class REx starts with. Try to find this position... */
1842 /* if reginfo->intuit, its a dryrun */
1843 /* annoyingly all the vars in this routine have different names from their counterparts
1844 in regmatch. /grrr */
1846 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1847 const char *strend, regmatch_info *reginfo)
1850 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1851 char *pat_string; /* The pattern's exactish string */
1852 char *pat_end; /* ptr to end char of pat_string */
1853 re_fold_t folder; /* Function for computing non-utf8 folds */
1854 const U8 *fold_array; /* array for folding ords < 256 */
1860 I32 tmp = 1; /* Scratch variable? */
1861 const bool utf8_target = reginfo->is_utf8_target;
1862 UV utf8_fold_flags = 0;
1863 const bool is_utf8_pat = reginfo->is_utf8_pat;
1864 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1865 with a result inverts that result, as 0^1 =
1867 _char_class_number classnum;
1869 RXi_GET_DECL(prog,progi);
1871 PERL_ARGS_ASSERT_FIND_BYCLASS;
1873 /* We know what class it must start with. */
1876 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1878 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) {
1879 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
1886 REXEC_FBC_UTF8_CLASS_SCAN(
1887 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1889 else if (ANYOF_FLAGS(c)) {
1890 REXEC_FBC_CLASS_SCAN(reginclass(prog,c, (U8*)s, (U8*)s+1, 0));
1893 REXEC_FBC_CLASS_SCAN(ANYOF_BITMAP_TEST(c, *((U8*)s)));
1897 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1898 assert(! is_utf8_pat);
1901 if (is_utf8_pat || utf8_target) {
1902 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1903 goto do_exactf_utf8;
1905 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1906 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1907 goto do_exactf_non_utf8; /* isn't dealt with by these */
1909 case EXACTF: /* This node only generated for non-utf8 patterns */
1910 assert(! is_utf8_pat);
1912 utf8_fold_flags = 0;
1913 goto do_exactf_utf8;
1915 fold_array = PL_fold;
1917 goto do_exactf_non_utf8;
1920 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1921 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1922 utf8_fold_flags = FOLDEQ_LOCALE;
1923 goto do_exactf_utf8;
1925 fold_array = PL_fold_locale;
1926 folder = foldEQ_locale;
1927 goto do_exactf_non_utf8;
1931 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1933 goto do_exactf_utf8;
1936 if (! utf8_target) { /* All code points in this node require
1937 UTF-8 to express. */
1940 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
1941 | FOLDEQ_S2_FOLDS_SANE;
1942 goto do_exactf_utf8;
1945 if (is_utf8_pat || utf8_target) {
1946 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1947 goto do_exactf_utf8;
1950 /* Any 'ss' in the pattern should have been replaced by regcomp,
1951 * so we don't have to worry here about this single special case
1952 * in the Latin1 range */
1953 fold_array = PL_fold_latin1;
1954 folder = foldEQ_latin1;
1958 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1959 are no glitches with fold-length differences
1960 between the target string and pattern */
1962 /* The idea in the non-utf8 EXACTF* cases is to first find the
1963 * first character of the EXACTF* node and then, if necessary,
1964 * case-insensitively compare the full text of the node. c1 is the
1965 * first character. c2 is its fold. This logic will not work for
1966 * Unicode semantics and the german sharp ss, which hence should
1967 * not be compiled into a node that gets here. */
1968 pat_string = STRING(c);
1969 ln = STR_LEN(c); /* length to match in octets/bytes */
1971 /* We know that we have to match at least 'ln' bytes (which is the
1972 * same as characters, since not utf8). If we have to match 3
1973 * characters, and there are only 2 availabe, we know without
1974 * trying that it will fail; so don't start a match past the
1975 * required minimum number from the far end */
1976 e = HOP3c(strend, -((SSize_t)ln), s);
1981 c2 = fold_array[c1];
1982 if (c1 == c2) { /* If char and fold are the same */
1983 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1986 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1994 /* If one of the operands is in utf8, we can't use the simpler folding
1995 * above, due to the fact that many different characters can have the
1996 * same fold, or portion of a fold, or different- length fold */
1997 pat_string = STRING(c);
1998 ln = STR_LEN(c); /* length to match in octets/bytes */
1999 pat_end = pat_string + ln;
2000 lnc = is_utf8_pat /* length to match in characters */
2001 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
2004 /* We have 'lnc' characters to match in the pattern, but because of
2005 * multi-character folding, each character in the target can match
2006 * up to 3 characters (Unicode guarantees it will never exceed
2007 * this) if it is utf8-encoded; and up to 2 if not (based on the
2008 * fact that the Latin 1 folds are already determined, and the
2009 * only multi-char fold in that range is the sharp-s folding to
2010 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
2011 * string character. Adjust lnc accordingly, rounding up, so that
2012 * if we need to match at least 4+1/3 chars, that really is 5. */
2013 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
2014 lnc = (lnc + expansion - 1) / expansion;
2016 /* As in the non-UTF8 case, if we have to match 3 characters, and
2017 * only 2 are left, it's guaranteed to fail, so don't start a
2018 * match that would require us to go beyond the end of the string
2020 e = HOP3c(strend, -((SSize_t)lnc), s);
2022 /* XXX Note that we could recalculate e to stop the loop earlier,
2023 * as the worst case expansion above will rarely be met, and as we
2024 * go along we would usually find that e moves further to the left.
2025 * This would happen only after we reached the point in the loop
2026 * where if there were no expansion we should fail. Unclear if
2027 * worth the expense */
2030 char *my_strend= (char *)strend;
2031 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
2032 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
2033 && (reginfo->intuit || regtry(reginfo, &s)) )
2037 s += (utf8_target) ? UTF8SKIP(s) : 1;
2043 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2044 if (FLAGS(c) != TRADITIONAL_BOUND) {
2045 if (! IN_UTF8_CTYPE_LOCALE) {
2046 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2047 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2052 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2056 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2057 if (FLAGS(c) != TRADITIONAL_BOUND) {
2058 if (! IN_UTF8_CTYPE_LOCALE) {
2059 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2060 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2065 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2068 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2070 assert(FLAGS(c) == TRADITIONAL_BOUND);
2072 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2075 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2077 assert(FLAGS(c) == TRADITIONAL_BOUND);
2079 FBC_BOUND_A(isWORDCHAR_A);
2082 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2084 assert(FLAGS(c) == TRADITIONAL_BOUND);
2086 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2089 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2091 assert(FLAGS(c) == TRADITIONAL_BOUND);
2093 FBC_NBOUND_A(isWORDCHAR_A);
2097 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
2098 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2109 switch((bound_type) FLAGS(c)) {
2110 case TRADITIONAL_BOUND:
2111 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2114 if (s == reginfo->strbeg) {
2115 if (reginfo->intuit || regtry(reginfo, &s))
2120 /* Didn't match. Try at the next position (if there is one) */
2121 s += (utf8_target) ? UTF8SKIP(s) : 1;
2122 if (UNLIKELY(s >= reginfo->strend)) {
2128 GCB_enum before = getGCB_VAL_UTF8(
2130 (U8*)(reginfo->strbeg)),
2131 (U8*) reginfo->strend);
2132 while (s < strend) {
2133 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
2134 (U8*) reginfo->strend);
2135 if ( (to_complement ^ isGCB(before,
2137 (U8*) reginfo->strbeg,
2140 && (reginfo->intuit || regtry(reginfo, &s)))
2148 else { /* Not utf8. Everything is a GCB except between CR and
2150 while (s < strend) {
2151 if ((to_complement ^ ( UCHARAT(s - 1) != '\r'
2152 || UCHARAT(s) != '\n'))
2153 && (reginfo->intuit || regtry(reginfo, &s)))
2161 /* And, since this is a bound, it can match after the final
2162 * character in the string */
2163 if ((reginfo->intuit || regtry(reginfo, &s))) {
2169 if (s == reginfo->strbeg) {
2170 if (reginfo->intuit || regtry(reginfo, &s)) {
2173 s += (utf8_target) ? UTF8SKIP(s) : 1;
2174 if (UNLIKELY(s >= reginfo->strend)) {
2180 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
2182 (U8*)(reginfo->strbeg)),
2183 (U8*) reginfo->strend);
2184 while (s < strend) {
2185 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
2186 if (to_complement ^ isLB(before,
2188 (U8*) reginfo->strbeg,
2190 (U8*) reginfo->strend,
2192 && (reginfo->intuit || regtry(reginfo, &s)))
2200 else { /* Not utf8. */
2201 LB_enum before = getLB_VAL_CP((U8) *(s -1));
2202 while (s < strend) {
2203 LB_enum after = getLB_VAL_CP((U8) *s);
2204 if (to_complement ^ isLB(before,
2206 (U8*) reginfo->strbeg,
2208 (U8*) reginfo->strend,
2210 && (reginfo->intuit || regtry(reginfo, &s)))
2219 if (reginfo->intuit || regtry(reginfo, &s)) {
2226 if (s == reginfo->strbeg) {
2227 if (reginfo->intuit || regtry(reginfo, &s)) {
2230 s += (utf8_target) ? UTF8SKIP(s) : 1;
2231 if (UNLIKELY(s >= reginfo->strend)) {
2237 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
2239 (U8*)(reginfo->strbeg)),
2240 (U8*) reginfo->strend);
2241 while (s < strend) {
2242 SB_enum after = getSB_VAL_UTF8((U8*) s,
2243 (U8*) reginfo->strend);
2244 if ((to_complement ^ isSB(before,
2246 (U8*) reginfo->strbeg,
2248 (U8*) reginfo->strend,
2250 && (reginfo->intuit || regtry(reginfo, &s)))
2258 else { /* Not utf8. */
2259 SB_enum before = getSB_VAL_CP((U8) *(s -1));
2260 while (s < strend) {
2261 SB_enum after = getSB_VAL_CP((U8) *s);
2262 if ((to_complement ^ isSB(before,
2264 (U8*) reginfo->strbeg,
2266 (U8*) reginfo->strend,
2268 && (reginfo->intuit || regtry(reginfo, &s)))
2277 /* Here are at the final position in the target string. The SB
2278 * value is always true here, so matches, depending on other
2280 if (reginfo->intuit || regtry(reginfo, &s)) {
2287 if (s == reginfo->strbeg) {
2288 if (reginfo->intuit || regtry(reginfo, &s)) {
2291 s += (utf8_target) ? UTF8SKIP(s) : 1;
2292 if (UNLIKELY(s >= reginfo->strend)) {
2298 /* We are at a boundary between char_sub_0 and char_sub_1.
2299 * We also keep track of the value for char_sub_-1 as we
2300 * loop through the line. Context may be needed to make a
2301 * determination, and if so, this can save having to
2303 WB_enum previous = WB_UNKNOWN;
2304 WB_enum before = getWB_VAL_UTF8(
2307 (U8*)(reginfo->strbeg)),
2308 (U8*) reginfo->strend);
2309 while (s < strend) {
2310 WB_enum after = getWB_VAL_UTF8((U8*) s,
2311 (U8*) reginfo->strend);
2312 if ((to_complement ^ isWB(previous,
2315 (U8*) reginfo->strbeg,
2317 (U8*) reginfo->strend,
2319 && (reginfo->intuit || regtry(reginfo, &s)))
2328 else { /* Not utf8. */
2329 WB_enum previous = WB_UNKNOWN;
2330 WB_enum before = getWB_VAL_CP((U8) *(s -1));
2331 while (s < strend) {
2332 WB_enum after = getWB_VAL_CP((U8) *s);
2333 if ((to_complement ^ isWB(previous,
2336 (U8*) reginfo->strbeg,
2338 (U8*) reginfo->strend,
2340 && (reginfo->intuit || regtry(reginfo, &s)))
2350 if (reginfo->intuit || regtry(reginfo, &s)) {
2357 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2358 is_LNBREAK_latin1_safe(s, strend)
2362 /* The argument to all the POSIX node types is the class number to pass to
2363 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2370 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2371 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
2372 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
2387 /* The complement of something that matches only ASCII matches all
2388 * non-ASCII, plus everything in ASCII that isn't in the class. */
2389 REXEC_FBC_UTF8_CLASS_SCAN( ! isASCII_utf8_safe(s, strend)
2390 || ! _generic_isCC_A(*s, FLAGS(c)));
2399 /* Don't need to worry about utf8, as it can match only a single
2400 * byte invariant character. */
2401 REXEC_FBC_CLASS_SCAN(
2402 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
2410 if (! utf8_target) {
2411 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
2417 classnum = (_char_class_number) FLAGS(c);
2418 if (classnum < _FIRST_NON_SWASH_CC) {
2419 while (s < strend) {
2421 /* We avoid loading in the swash as long as possible, but
2422 * should we have to, we jump to a separate loop. This
2423 * extra 'if' statement is what keeps this code from being
2424 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
2425 if (UTF8_IS_ABOVE_LATIN1(*s)) {
2426 goto found_above_latin1;
2428 if ((UTF8_IS_INVARIANT(*s)
2429 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2431 || ( UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, strend)
2432 && to_complement ^ cBOOL(
2433 _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*s,
2437 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2449 else switch (classnum) { /* These classes are implemented as
2451 case _CC_ENUM_SPACE:
2452 REXEC_FBC_UTF8_CLASS_SCAN(
2453 to_complement ^ cBOOL(isSPACE_utf8_safe(s, strend)));
2456 case _CC_ENUM_BLANK:
2457 REXEC_FBC_UTF8_CLASS_SCAN(
2458 to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
2461 case _CC_ENUM_XDIGIT:
2462 REXEC_FBC_UTF8_CLASS_SCAN(
2463 to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
2466 case _CC_ENUM_VERTSPACE:
2467 REXEC_FBC_UTF8_CLASS_SCAN(
2468 to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
2471 case _CC_ENUM_CNTRL:
2472 REXEC_FBC_UTF8_CLASS_SCAN(
2473 to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
2477 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2478 NOT_REACHED; /* NOTREACHED */
2483 found_above_latin1: /* Here we have to load a swash to get the result
2484 for the current code point */
2485 if (! PL_utf8_swash_ptrs[classnum]) {
2486 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2487 PL_utf8_swash_ptrs[classnum] =
2488 _core_swash_init("utf8",
2491 PL_XPosix_ptrs[classnum], &flags);
2494 /* This is a copy of the loop above for swash classes, though using the
2495 * FBC macro instead of being expanded out. Since we've loaded the
2496 * swash, we don't have to check for that each time through the loop */
2497 REXEC_FBC_UTF8_CLASS_SCAN(
2498 to_complement ^ cBOOL(_generic_utf8_safe(
2502 swash_fetch(PL_utf8_swash_ptrs[classnum],
2510 /* what trie are we using right now */
2511 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2512 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2513 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2515 const char *last_start = strend - trie->minlen;
2517 const char *real_start = s;
2519 STRLEN maxlen = trie->maxlen;
2521 U8 **points; /* map of where we were in the input string
2522 when reading a given char. For ASCII this
2523 is unnecessary overhead as the relationship
2524 is always 1:1, but for Unicode, especially
2525 case folded Unicode this is not true. */
2526 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2530 GET_RE_DEBUG_FLAGS_DECL;
2532 /* We can't just allocate points here. We need to wrap it in
2533 * an SV so it gets freed properly if there is a croak while
2534 * running the match */
2537 sv_points=newSV(maxlen * sizeof(U8 *));
2538 SvCUR_set(sv_points,
2539 maxlen * sizeof(U8 *));
2540 SvPOK_on(sv_points);
2541 sv_2mortal(sv_points);
2542 points=(U8**)SvPV_nolen(sv_points );
2543 if ( trie_type != trie_utf8_fold
2544 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2547 bitmap=(U8*)trie->bitmap;
2549 bitmap=(U8*)ANYOF_BITMAP(c);
2551 /* this is the Aho-Corasick algorithm modified a touch
2552 to include special handling for long "unknown char" sequences.
2553 The basic idea being that we use AC as long as we are dealing
2554 with a possible matching char, when we encounter an unknown char
2555 (and we have not encountered an accepting state) we scan forward
2556 until we find a legal starting char.
2557 AC matching is basically that of trie matching, except that when
2558 we encounter a failing transition, we fall back to the current
2559 states "fail state", and try the current char again, a process
2560 we repeat until we reach the root state, state 1, or a legal
2561 transition. If we fail on the root state then we can either
2562 terminate if we have reached an accepting state previously, or
2563 restart the entire process from the beginning if we have not.
2566 while (s <= last_start) {
2567 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2575 U8 *uscan = (U8*)NULL;
2576 U8 *leftmost = NULL;
2578 U32 accepted_word= 0;
2582 while ( state && uc <= (U8*)strend ) {
2584 U32 word = aho->states[ state ].wordnum;
2588 DEBUG_TRIE_EXECUTE_r(
2589 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2590 dump_exec_pos( (char *)uc, c, strend, real_start,
2591 (char *)uc, utf8_target, 0 );
2592 Perl_re_printf( aTHX_
2593 " Scanning for legal start char...\n");
2597 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2601 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2607 if (uc >(U8*)last_start) break;
2611 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2612 if (!leftmost || lpos < leftmost) {
2613 DEBUG_r(accepted_word=word);
2619 points[pointpos++ % maxlen]= uc;
2620 if (foldlen || uc < (U8*)strend) {
2621 REXEC_TRIE_READ_CHAR(trie_type, trie,
2623 uscan, len, uvc, charid, foldlen,
2625 DEBUG_TRIE_EXECUTE_r({
2626 dump_exec_pos( (char *)uc, c, strend,
2627 real_start, s, utf8_target, 0);
2628 Perl_re_printf( aTHX_
2629 " Charid:%3u CP:%4" UVxf " ",
2641 word = aho->states[ state ].wordnum;
2643 base = aho->states[ state ].trans.base;
2645 DEBUG_TRIE_EXECUTE_r({
2647 dump_exec_pos( (char *)uc, c, strend, real_start,
2648 s, utf8_target, 0 );
2649 Perl_re_printf( aTHX_
2650 "%sState: %4" UVxf ", word=%" UVxf,
2651 failed ? " Fail transition to " : "",
2652 (UV)state, (UV)word);
2658 ( ((offset = base + charid
2659 - 1 - trie->uniquecharcount)) >= 0)
2660 && ((U32)offset < trie->lasttrans)
2661 && trie->trans[offset].check == state
2662 && (tmp=trie->trans[offset].next))
2664 DEBUG_TRIE_EXECUTE_r(
2665 Perl_re_printf( aTHX_ " - legal\n"));
2670 DEBUG_TRIE_EXECUTE_r(
2671 Perl_re_printf( aTHX_ " - fail\n"));
2673 state = aho->fail[state];
2677 /* we must be accepting here */
2678 DEBUG_TRIE_EXECUTE_r(
2679 Perl_re_printf( aTHX_ " - accepting\n"));
2688 if (!state) state = 1;
2691 if ( aho->states[ state ].wordnum ) {
2692 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2693 if (!leftmost || lpos < leftmost) {
2694 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2699 s = (char*)leftmost;
2700 DEBUG_TRIE_EXECUTE_r({
2701 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
2702 (UV)accepted_word, (IV)(s - real_start)
2705 if (reginfo->intuit || regtry(reginfo, &s)) {
2711 DEBUG_TRIE_EXECUTE_r({
2712 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
2715 DEBUG_TRIE_EXECUTE_r(
2716 Perl_re_printf( aTHX_ "No match.\n"));
2725 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2732 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2733 * flags have same meanings as with regexec_flags() */
2736 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2743 struct regexp *const prog = ReANY(rx);
2745 if (flags & REXEC_COPY_STR) {
2748 DEBUG_C(Perl_re_printf( aTHX_
2749 "Copy on write: regexp capture, type %d\n",
2751 /* Create a new COW SV to share the match string and store
2752 * in saved_copy, unless the current COW SV in saved_copy
2753 * is valid and suitable for our purpose */
2754 if (( prog->saved_copy
2755 && SvIsCOW(prog->saved_copy)
2756 && SvPOKp(prog->saved_copy)
2759 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2761 /* just reuse saved_copy SV */
2762 if (RXp_MATCH_COPIED(prog)) {
2763 Safefree(prog->subbeg);
2764 RXp_MATCH_COPIED_off(prog);
2768 /* create new COW SV to share string */
2769 RX_MATCH_COPY_FREE(rx);
2770 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2772 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2773 assert (SvPOKp(prog->saved_copy));
2774 prog->sublen = strend - strbeg;
2775 prog->suboffset = 0;
2776 prog->subcoffset = 0;
2781 SSize_t max = strend - strbeg;
2784 if ( (flags & REXEC_COPY_SKIP_POST)
2785 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2786 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2787 ) { /* don't copy $' part of string */
2790 /* calculate the right-most part of the string covered
2791 * by a capture. Due to lookahead, this may be to
2792 * the right of $&, so we have to scan all captures */
2793 while (n <= prog->lastparen) {
2794 if (prog->offs[n].end > max)
2795 max = prog->offs[n].end;
2799 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2800 ? prog->offs[0].start
2802 assert(max >= 0 && max <= strend - strbeg);
2805 if ( (flags & REXEC_COPY_SKIP_PRE)
2806 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2807 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2808 ) { /* don't copy $` part of string */
2811 /* calculate the left-most part of the string covered
2812 * by a capture. Due to lookbehind, this may be to
2813 * the left of $&, so we have to scan all captures */
2814 while (min && n <= prog->lastparen) {
2815 if ( prog->offs[n].start != -1
2816 && prog->offs[n].start < min)
2818 min = prog->offs[n].start;
2822 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2823 && min > prog->offs[0].end
2825 min = prog->offs[0].end;
2829 assert(min >= 0 && min <= max && min <= strend - strbeg);
2832 if (RX_MATCH_COPIED(rx)) {
2833 if (sublen > prog->sublen)
2835 (char*)saferealloc(prog->subbeg, sublen+1);
2838 prog->subbeg = (char*)safemalloc(sublen+1);
2839 Copy(strbeg + min, prog->subbeg, sublen, char);
2840 prog->subbeg[sublen] = '\0';
2841 prog->suboffset = min;
2842 prog->sublen = sublen;
2843 RX_MATCH_COPIED_on(rx);
2845 prog->subcoffset = prog->suboffset;
2846 if (prog->suboffset && utf8_target) {
2847 /* Convert byte offset to chars.
2848 * XXX ideally should only compute this if @-/@+
2849 * has been seen, a la PL_sawampersand ??? */
2851 /* If there's a direct correspondence between the
2852 * string which we're matching and the original SV,
2853 * then we can use the utf8 len cache associated with
2854 * the SV. In particular, it means that under //g,
2855 * sv_pos_b2u() will use the previously cached
2856 * position to speed up working out the new length of
2857 * subcoffset, rather than counting from the start of
2858 * the string each time. This stops
2859 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2860 * from going quadratic */
2861 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2862 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2863 SV_GMAGIC|SV_CONST_RETURN);
2865 prog->subcoffset = utf8_length((U8*)strbeg,
2866 (U8*)(strbeg+prog->suboffset));
2870 RX_MATCH_COPY_FREE(rx);
2871 prog->subbeg = strbeg;
2872 prog->suboffset = 0;
2873 prog->subcoffset = 0;
2874 prog->sublen = strend - strbeg;
2882 - regexec_flags - match a regexp against a string
2885 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2886 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2887 /* stringarg: the point in the string at which to begin matching */
2888 /* strend: pointer to null at end of string */
2889 /* strbeg: real beginning of string */
2890 /* minend: end of match must be >= minend bytes after stringarg. */
2891 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2892 * itself is accessed via the pointers above */
2893 /* data: May be used for some additional optimizations.
2894 Currently unused. */
2895 /* flags: For optimizations. See REXEC_* in regexp.h */
2898 struct regexp *const prog = ReANY(rx);
2902 SSize_t minlen; /* must match at least this many chars */
2903 SSize_t dontbother = 0; /* how many characters not to try at end */
2904 const bool utf8_target = cBOOL(DO_UTF8(sv));
2906 RXi_GET_DECL(prog,progi);
2907 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2908 regmatch_info *const reginfo = ®info_buf;
2909 regexp_paren_pair *swap = NULL;
2911 GET_RE_DEBUG_FLAGS_DECL;
2913 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2914 PERL_UNUSED_ARG(data);
2916 /* Be paranoid... */
2918 Perl_croak(aTHX_ "NULL regexp parameter");
2922 debug_start_match(rx, utf8_target, stringarg, strend,
2926 startpos = stringarg;
2928 /* set these early as they may be used by the HOP macros below */
2929 reginfo->strbeg = strbeg;
2930 reginfo->strend = strend;
2931 reginfo->is_utf8_target = cBOOL(utf8_target);
2933 if (prog->intflags & PREGf_GPOS_SEEN) {
2936 /* set reginfo->ganch, the position where \G can match */
2939 (flags & REXEC_IGNOREPOS)
2940 ? stringarg /* use start pos rather than pos() */
2941 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2942 /* Defined pos(): */
2943 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2944 : strbeg; /* pos() not defined; use start of string */
2946 DEBUG_GPOS_r(Perl_re_printf( aTHX_
2947 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
2949 /* in the presence of \G, we may need to start looking earlier in
2950 * the string than the suggested start point of stringarg:
2951 * if prog->gofs is set, then that's a known, fixed minimum
2954 * /ab|c\G/: gofs = 1
2955 * or if the minimum offset isn't known, then we have to go back
2956 * to the start of the string, e.g. /w+\G/
2959 if (prog->intflags & PREGf_ANCH_GPOS) {
2961 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
2963 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
2965 DEBUG_r(Perl_re_printf( aTHX_
2966 "fail: ganch-gofs before earliest possible start\n"));
2971 startpos = reginfo->ganch;
2973 else if (prog->gofs) {
2974 startpos = HOPBACKc(startpos, prog->gofs);
2978 else if (prog->intflags & PREGf_GPOS_FLOAT)
2982 minlen = prog->minlen;
2983 if ((startpos + minlen) > strend || startpos < strbeg) {
2984 DEBUG_r(Perl_re_printf( aTHX_
2985 "Regex match can't succeed, so not even tried\n"));
2989 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2990 * which will call destuctors to reset PL_regmatch_state, free higher
2991 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2992 * regmatch_info_aux_eval */
2994 oldsave = PL_savestack_ix;
2998 if ((prog->extflags & RXf_USE_INTUIT)
2999 && !(flags & REXEC_CHECKED))
3001 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
3006 if (prog->extflags & RXf_CHECK_ALL) {
3007 /* we can match based purely on the result of INTUIT.
3008 * Set up captures etc just for $& and $-[0]
3009 * (an intuit-only match wont have $1,$2,..) */
3010 assert(!prog->nparens);
3012 /* s/// doesn't like it if $& is earlier than where we asked it to
3013 * start searching (which can happen on something like /.\G/) */
3014 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3017 /* this should only be possible under \G */
3018 assert(prog->intflags & PREGf_GPOS_SEEN);
3019 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3020 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3024 /* match via INTUIT shouldn't have any captures.
3025 * Let @-, @+, $^N know */
3026 prog->lastparen = prog->lastcloseparen = 0;
3027 RX_MATCH_UTF8_set(rx, utf8_target);
3028 prog->offs[0].start = s - strbeg;
3029 prog->offs[0].end = utf8_target
3030 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
3031 : s - strbeg + prog->minlenret;
3032 if ( !(flags & REXEC_NOT_FIRST) )
3033 S_reg_set_capture_string(aTHX_ rx,
3035 sv, flags, utf8_target);
3041 multiline = prog->extflags & RXf_PMf_MULTILINE;
3043 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
3044 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3045 "String too short [regexec_flags]...\n"));
3049 /* Check validity of program. */
3050 if (UCHARAT(progi->program) != REG_MAGIC) {
3051 Perl_croak(aTHX_ "corrupted regexp program");
3054 RX_MATCH_TAINTED_off(rx);
3055 RX_MATCH_UTF8_set(rx, utf8_target);
3057 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3058 reginfo->intuit = 0;
3059 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3060 reginfo->warned = FALSE;
3062 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
3063 /* see how far we have to get to not match where we matched before */
3064 reginfo->till = stringarg + minend;
3066 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
3067 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3068 S_cleanup_regmatch_info_aux has executed (registered by
3069 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3070 magic belonging to this SV.
3071 Not newSVsv, either, as it does not COW.
3073 reginfo->sv = newSV(0);
3074 SvSetSV_nosteal(reginfo->sv, sv);
3075 SAVEFREESV(reginfo->sv);
3078 /* reserve next 2 or 3 slots in PL_regmatch_state:
3079 * slot N+0: may currently be in use: skip it
3080 * slot N+1: use for regmatch_info_aux struct
3081 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3082 * slot N+3: ready for use by regmatch()
3086 regmatch_state *old_regmatch_state;
3087 regmatch_slab *old_regmatch_slab;
3088 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3090 /* on first ever match, allocate first slab */
3091 if (!PL_regmatch_slab) {
3092 Newx(PL_regmatch_slab, 1, regmatch_slab);
3093 PL_regmatch_slab->prev = NULL;
3094 PL_regmatch_slab->next = NULL;
3095 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3098 old_regmatch_state = PL_regmatch_state;
3099 old_regmatch_slab = PL_regmatch_slab;
3101 for (i=0; i <= max; i++) {
3103 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3105 reginfo->info_aux_eval =
3106 reginfo->info_aux->info_aux_eval =
3107 &(PL_regmatch_state->u.info_aux_eval);
3109 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3110 PL_regmatch_state = S_push_slab(aTHX);
3113 /* note initial PL_regmatch_state position; at end of match we'll
3114 * pop back to there and free any higher slabs */
3116 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3117 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
3118 reginfo->info_aux->poscache = NULL;
3120 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
3122 if ((prog->extflags & RXf_EVAL_SEEN))
3123 S_setup_eval_state(aTHX_ reginfo);
3125 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
3128 /* If there is a "must appear" string, look for it. */
3130 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
3131 /* We have to be careful. If the previous successful match
3132 was from this regex we don't want a subsequent partially
3133 successful match to clobber the old results.
3134 So when we detect this possibility we add a swap buffer
3135 to the re, and switch the buffer each match. If we fail,
3136 we switch it back; otherwise we leave it swapped.
3139 /* do we need a save destructor here for eval dies? */
3140 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
3141 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3142 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
3150 if (prog->recurse_locinput)
3151 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3153 /* Simplest case: anchored match need be tried only once, or with
3154 * MBOL, only at the beginning of each line.
3156 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3157 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3158 * match at the start of the string then it won't match anywhere else
3159 * either; while with /.*.../, if it doesn't match at the beginning,
3160 * the earliest it could match is at the start of the next line */
3162 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3165 if (regtry(reginfo, &s))
3168 if (!(prog->intflags & PREGf_ANCH_MBOL))
3171 /* didn't match at start, try at other newline positions */
3174 dontbother = minlen - 1;
3175 end = HOP3c(strend, -dontbother, strbeg) - 1;
3177 /* skip to next newline */
3179 while (s <= end) { /* note it could be possible to match at the end of the string */
3180 /* NB: newlines are the same in unicode as they are in latin */
3183 if (prog->check_substr || prog->check_utf8) {
3184 /* note that with PREGf_IMPLICIT, intuit can only fail
3185 * or return the start position, so it's of limited utility.
3186 * Nevertheless, I made the decision that the potential for
3187 * quick fail was still worth it - DAPM */
3188 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3192 if (regtry(reginfo, &s))
3196 } /* end anchored search */
3198 if (prog->intflags & PREGf_ANCH_GPOS)
3200 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3201 assert(prog->intflags & PREGf_GPOS_SEEN);
3202 /* For anchored \G, the only position it can match from is
3203 * (ganch-gofs); we already set startpos to this above; if intuit
3204 * moved us on from there, we can't possibly succeed */
3205 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
3206 if (s == startpos && regtry(reginfo, &s))
3211 /* Messy cases: unanchored match. */
3212 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
3213 /* we have /x+whatever/ */
3214 /* it must be a one character string (XXXX Except is_utf8_pat?) */
3220 if (! prog->anchored_utf8) {
3221 to_utf8_substr(prog);
3223 ch = SvPVX_const(prog->anchored_utf8)[0];
3226 DEBUG_EXECUTE_r( did_match = 1 );
3227 if (regtry(reginfo, &s)) goto got_it;
3229 while (s < strend && *s == ch)
3236 if (! prog->anchored_substr) {
3237 if (! to_byte_substr(prog)) {
3238 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3241 ch = SvPVX_const(prog->anchored_substr)[0];
3244 DEBUG_EXECUTE_r( did_match = 1 );
3245 if (regtry(reginfo, &s)) goto got_it;
3247 while (s < strend && *s == ch)
3252 DEBUG_EXECUTE_r(if (!did_match)
3253 Perl_re_printf( aTHX_
3254 "Did not find anchored character...\n")
3257 else if (prog->anchored_substr != NULL
3258 || prog->anchored_utf8 != NULL
3259 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
3260 && prog->float_max_offset < strend - s)) {
3265 char *last1; /* Last position checked before */
3269 if (prog->anchored_substr || prog->anchored_utf8) {
3271 if (! prog->anchored_utf8) {
3272 to_utf8_substr(prog);
3274 must = prog->anchored_utf8;
3277 if (! prog->anchored_substr) {
3278 if (! to_byte_substr(prog)) {
3279 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3282 must = prog->anchored_substr;
3284 back_max = back_min = prog->anchored_offset;
3287 if (! prog->float_utf8) {
3288 to_utf8_substr(prog);
3290 must = prog->float_utf8;
3293 if (! prog->float_substr) {
3294 if (! to_byte_substr(prog)) {
3295 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3298 must = prog->float_substr;
3300 back_max = prog->float_max_offset;
3301 back_min = prog->float_min_offset;
3307 last = HOP3c(strend, /* Cannot start after this */
3308 -(SSize_t)(CHR_SVLEN(must)
3309 - (SvTAIL(must) != 0) + back_min), strbeg);
3311 if (s > reginfo->strbeg)
3312 last1 = HOPc(s, -1);
3314 last1 = s - 1; /* bogus */
3316 /* XXXX check_substr already used to find "s", can optimize if
3317 check_substr==must. */
3319 strend = HOPc(strend, -dontbother);
3320 while ( (s <= last) &&
3321 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
3322 (unsigned char*)strend, must,
3323 multiline ? FBMrf_MULTILINE : 0)) ) {
3324 DEBUG_EXECUTE_r( did_match = 1 );
3325 if (HOPc(s, -back_max) > last1) {
3326 last1 = HOPc(s, -back_min);
3327 s = HOPc(s, -back_max);
3330 char * const t = (last1 >= reginfo->strbeg)
3331 ? HOPc(last1, 1) : last1 + 1;
3333 last1 = HOPc(s, -back_min);
3337 while (s <= last1) {
3338 if (regtry(reginfo, &s))
3341 s++; /* to break out of outer loop */
3348 while (s <= last1) {
3349 if (regtry(reginfo, &s))
3355 DEBUG_EXECUTE_r(if (!did_match) {
3356 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
3357 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
3358 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n",
3359 ((must == prog->anchored_substr || must == prog->anchored_utf8)
3360 ? "anchored" : "floating"),
3361 quoted, RE_SV_TAIL(must));
3365 else if ( (c = progi->regstclass) ) {
3367 const OPCODE op = OP(progi->regstclass);
3368 /* don't bother with what can't match */
3369 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
3370 strend = HOPc(strend, -(minlen - 1));
3373 SV * const prop = sv_newmortal();
3374 regprop(prog, prop, c, reginfo, NULL);
3376 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
3378 Perl_re_printf( aTHX_
3379 "Matching stclass %.*s against %s (%d bytes)\n",
3380 (int)SvCUR(prop), SvPVX_const(prop),
3381 quoted, (int)(strend - s));
3384 if (find_byclass(prog, c, s, strend, reginfo))
3386 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n"));
3390 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3398 if (! prog->float_utf8) {
3399 to_utf8_substr(prog);
3401 float_real = prog->float_utf8;
3404 if (! prog->float_substr) {
3405 if (! to_byte_substr(prog)) {
3406 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3409 float_real = prog->float_substr;
3412 little = SvPV_const(float_real, len);
3413 if (SvTAIL(float_real)) {
3414 /* This means that float_real contains an artificial \n on
3415 * the end due to the presence of something like this:
3416 * /foo$/ where we can match both "foo" and "foo\n" at the
3417 * end of the string. So we have to compare the end of the
3418 * string first against the float_real without the \n and
3419 * then against the full float_real with the string. We
3420 * have to watch out for cases where the string might be
3421 * smaller than the float_real or the float_real without
3423 char *checkpos= strend - len;
3425 Perl_re_printf( aTHX_
3426 "%sChecking for float_real.%s\n",
3427 PL_colors[4], PL_colors[5]));
3428 if (checkpos + 1 < strbeg) {
3429 /* can't match, even if we remove the trailing \n
3430 * string is too short to match */
3432 Perl_re_printf( aTHX_
3433 "%sString shorter than required trailing substring, cannot match.%s\n",
3434 PL_colors[4], PL_colors[5]));
3436 } else if (memEQ(checkpos + 1, little, len - 1)) {
3437 /* can match, the end of the string matches without the
3439 last = checkpos + 1;
3440 } else if (checkpos < strbeg) {
3441 /* cant match, string is too short when the "\n" is
3444 Perl_re_printf( aTHX_
3445 "%sString does not contain required trailing substring, cannot match.%s\n",
3446 PL_colors[4], PL_colors[5]));
3448 } else if (!multiline) {
3449 /* non multiline match, so compare with the "\n" at the
3450 * end of the string */
3451 if (memEQ(checkpos, little, len)) {
3455 Perl_re_printf( aTHX_
3456 "%sString does not contain required trailing substring, cannot match.%s\n",
3457 PL_colors[4], PL_colors[5]));
3461 /* multiline match, so we have to search for a place
3462 * where the full string is located */
3468 last = rninstr(s, strend, little, little + len);
3470 last = strend; /* matching "$" */
3473 /* at one point this block contained a comment which was
3474 * probably incorrect, which said that this was a "should not
3475 * happen" case. Even if it was true when it was written I am
3476 * pretty sure it is not anymore, so I have removed the comment
3477 * and replaced it with this one. Yves */
3479 Perl_re_printf( aTHX_
3480 "%sString does not contain required substring, cannot match.%s\n",
3481 PL_colors[4], PL_colors[5]
3485 dontbother = strend - last + prog->float_min_offset;
3487 if (minlen && (dontbother < minlen))
3488 dontbother = minlen - 1;
3489 strend -= dontbother; /* this one's always in bytes! */
3490 /* We don't know much -- general case. */
3493 if (regtry(reginfo, &s))
3502 if (regtry(reginfo, &s))
3504 } while (s++ < strend);
3512 /* s/// doesn't like it if $& is earlier than where we asked it to
3513 * start searching (which can happen on something like /.\G/) */
3514 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3515 && (prog->offs[0].start < stringarg - strbeg))
3517 /* this should only be possible under \G */
3518 assert(prog->intflags & PREGf_GPOS_SEEN);
3519 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3520 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3526 Perl_re_exec_indentf( aTHX_
3527 "rex=0x%" UVxf " freeing offs: 0x%" UVxf "\n",
3535 /* clean up; this will trigger destructors that will free all slabs
3536 * above the current one, and cleanup the regmatch_info_aux
3537 * and regmatch_info_aux_eval sructs */
3539 LEAVE_SCOPE(oldsave);
3541 if (RXp_PAREN_NAMES(prog))
3542 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3544 /* make sure $`, $&, $', and $digit will work later */
3545 if ( !(flags & REXEC_NOT_FIRST) )
3546 S_reg_set_capture_string(aTHX_ rx,
3547 strbeg, reginfo->strend,
3548 sv, flags, utf8_target);
3553 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n",
3554 PL_colors[4], PL_colors[5]));
3556 /* clean up; this will trigger destructors that will free all slabs
3557 * above the current one, and cleanup the regmatch_info_aux
3558 * and regmatch_info_aux_eval sructs */
3560 LEAVE_SCOPE(oldsave);
3563 /* we failed :-( roll it back */
3564 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3565 "rex=0x%" UVxf " rolling back offs: freeing=0x%" UVxf " restoring=0x%" UVxf "\n",
3571 Safefree(prog->offs);
3578 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3579 * Do inc before dec, in case old and new rex are the same */
3580 #define SET_reg_curpm(Re2) \
3581 if (reginfo->info_aux_eval) { \
3582 (void)ReREFCNT_inc(Re2); \
3583 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3584 PM_SETRE((PL_reg_curpm), (Re2)); \
3589 - regtry - try match at specific point
3591 STATIC bool /* 0 failure, 1 success */
3592 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3595 REGEXP *const rx = reginfo->prog;
3596 regexp *const prog = ReANY(rx);
3599 U32 depth = 0; /* used by REGCP_SET */
3601 RXi_GET_DECL(prog,progi);
3602 GET_RE_DEBUG_FLAGS_DECL;
3604 PERL_ARGS_ASSERT_REGTRY;
3606 reginfo->cutpoint=NULL;
3608 prog->offs[0].start = *startposp - reginfo->strbeg;
3609 prog->lastparen = 0;
3610 prog->lastcloseparen = 0;
3612 /* XXXX What this code is doing here?!!! There should be no need
3613 to do this again and again, prog->lastparen should take care of
3616 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3617 * Actually, the code in regcppop() (which Ilya may be meaning by
3618 * prog->lastparen), is not needed at all by the test suite
3619 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3620 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3621 * Meanwhile, this code *is* needed for the
3622 * above-mentioned test suite tests to succeed. The common theme
3623 * on those tests seems to be returning null fields from matches.
3624 * --jhi updated by dapm */
3626 /* After encountering a variant of the issue mentioned above I think
3627 * the point Ilya was making is that if we properly unwind whenever
3628 * we set lastparen to a smaller value then we should not need to do
3629 * this every time, only when needed. So if we have tests that fail if
3630 * we remove this, then it suggests somewhere else we are improperly
3631 * unwinding the lastparen/paren buffers. See UNWIND_PARENS() and
3632 * places it is called, and related regcp() routines. - Yves */
3634 if (prog->nparens) {
3635 regexp_paren_pair *pp = prog->offs;
3637 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3645 result = regmatch(reginfo, *startposp, progi->program + 1);
3647 prog->offs[0].end = result;
3650 if (reginfo->cutpoint)
3651 *startposp= reginfo->cutpoint;
3652 REGCP_UNWIND(lastcp);
3657 #define sayYES goto yes
3658 #define sayNO goto no
3659 #define sayNO_SILENT goto no_silent
3661 /* we dont use STMT_START/END here because it leads to
3662 "unreachable code" warnings, which are bogus, but distracting. */
3663 #define CACHEsayNO \
3664 if (ST.cache_mask) \
3665 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3668 /* this is used to determine how far from the left messages like
3669 'failed...' are printed in regexec.c. It should be set such that
3670 messages are inline with the regop output that created them.
3672 #define REPORT_CODE_OFF 29
3673 #define INDENT_CHARS(depth) ((int)(depth) % 20)
3676 Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...)
3680 PerlIO *f= Perl_debug_log;
3681 PERL_ARGS_ASSERT_RE_EXEC_INDENTF;
3682 va_start(ap, depth);
3683 PerlIO_printf(f, "%*s|%4" UVuf "| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" );
3684 result = PerlIO_vprintf(f, fmt, ap);
3688 #endif /* DEBUGGING */
3691 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3692 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3693 #define CHRTEST_NOT_A_CP_1 -999
3694 #define CHRTEST_NOT_A_CP_2 -998
3696 /* grab a new slab and return the first slot in it */
3698 STATIC regmatch_state *
3701 regmatch_slab *s = PL_regmatch_slab->next;
3703 Newx(s, 1, regmatch_slab);
3704 s->prev = PL_regmatch_slab;
3706 PL_regmatch_slab->next = s;
3708 PL_regmatch_slab = s;
3709 return SLAB_FIRST(s);
3713 /* push a new state then goto it */
3715 #define PUSH_STATE_GOTO(state, node, input) \
3716 pushinput = input; \
3718 st->resume_state = state; \
3721 /* push a new state with success backtracking, then goto it */
3723 #define PUSH_YES_STATE_GOTO(state, node, input) \
3724 pushinput = input; \
3726 st->resume_state = state; \
3727 goto push_yes_state;
3734 regmatch() - main matching routine
3736 This is basically one big switch statement in a loop. We execute an op,
3737 set 'next' to point the next op, and continue. If we come to a point which
3738 we may need to backtrack to on failure such as (A|B|C), we push a
3739 backtrack state onto the backtrack stack. On failure, we pop the top
3740 state, and re-enter the loop at the state indicated. If there are no more
3741 states to pop, we return failure.
3743 Sometimes we also need to backtrack on success; for example /A+/, where
3744 after successfully matching one A, we need to go back and try to
3745 match another one; similarly for lookahead assertions: if the assertion
3746 completes successfully, we backtrack to the state just before the assertion
3747 and then carry on. In these cases, the pushed state is marked as
3748 'backtrack on success too'. This marking is in fact done by a chain of
3749 pointers, each pointing to the previous 'yes' state. On success, we pop to
3750 the nearest yes state, discarding any intermediate failure-only states.
3751 Sometimes a yes state is pushed just to force some cleanup code to be
3752 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3753 it to free the inner regex.
3755 Note that failure backtracking rewinds the cursor position, while
3756 success backtracking leaves it alone.
3758 A pattern is complete when the END op is executed, while a subpattern
3759 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3760 ops trigger the "pop to last yes state if any, otherwise return true"
3763 A common convention in this function is to use A and B to refer to the two
3764 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3765 the subpattern to be matched possibly multiple times, while B is the entire
3766 rest of the pattern. Variable and state names reflect this convention.
3768 The states in the main switch are the union of ops and failure/success of
3769 substates associated with with that op. For example, IFMATCH is the op
3770 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3771 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3772 successfully matched A and IFMATCH_A_fail is a state saying that we have
3773 just failed to match A. Resume states always come in pairs. The backtrack
3774 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3775 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3776 on success or failure.
3778 The struct that holds a backtracking state is actually a big union, with
3779 one variant for each major type of op. The variable st points to the
3780 top-most backtrack struct. To make the code clearer, within each
3781 block of code we #define ST to alias the relevant union.
3783 Here's a concrete example of a (vastly oversimplified) IFMATCH
3789 #define ST st->u.ifmatch
3791 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3792 ST.foo = ...; // some state we wish to save
3794 // push a yes backtrack state with a resume value of
3795 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3797 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3800 case IFMATCH_A: // we have successfully executed A; now continue with B
3802 bar = ST.foo; // do something with the preserved value
3805 case IFMATCH_A_fail: // A failed, so the assertion failed
3806 ...; // do some housekeeping, then ...
3807 sayNO; // propagate the failure
3814 For any old-timers reading this who are familiar with the old recursive
3815 approach, the code above is equivalent to:
3817 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3826 ...; // do some housekeeping, then ...
3827 sayNO; // propagate the failure
3830 The topmost backtrack state, pointed to by st, is usually free. If you
3831 want to claim it, populate any ST.foo fields in it with values you wish to
3832 save, then do one of
3834 PUSH_STATE_GOTO(resume_state, node, newinput);
3835 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3837 which sets that backtrack state's resume value to 'resume_state', pushes a
3838 new free entry to the top of the backtrack stack, then goes to 'node'.
3839 On backtracking, the free slot is popped, and the saved state becomes the
3840 new free state. An ST.foo field in this new top state can be temporarily
3841 accessed to retrieve values, but once the main loop is re-entered, it
3842 becomes available for reuse.
3844 Note that the depth of the backtrack stack constantly increases during the
3845 left-to-right execution of the pattern, rather than going up and down with
3846 the pattern nesting. For example the stack is at its maximum at Z at the
3847 end of the pattern, rather than at X in the following:
3849 /(((X)+)+)+....(Y)+....Z/
3851 The only exceptions to this are lookahead/behind assertions and the cut,
3852 (?>A), which pop all the backtrack states associated with A before
3855 Backtrack state structs are allocated in slabs of about 4K in size.
3856 PL_regmatch_state and st always point to the currently active state,
3857 and PL_regmatch_slab points to the slab currently containing
3858 PL_regmatch_state. The first time regmatch() is called, the first slab is
3859 allocated, and is never freed until interpreter destruction. When the slab
3860 is full, a new one is allocated and chained to the end. At exit from
3861 regmatch(), slabs allocated since entry are freed.
3866 #define DEBUG_STATE_pp(pp) \
3868 DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \
3869 Perl_re_printf( aTHX_ \
3870 "%*s" pp " %s%s%s%s%s\n", \
3871 INDENT_CHARS(depth), "", \
3872 PL_reg_name[st->resume_state], \
3873 ((st==yes_state||st==mark_state) ? "[" : ""), \
3874 ((st==yes_state) ? "Y" : ""), \
3875 ((st==mark_state) ? "M" : ""), \
3876 ((st==yes_state||st==mark_state) ? "]" : "") \
3881 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3886 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3887 const char *start, const char *end, const char *blurb)
3889 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3891 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3896 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3897 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3899 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3900 start, end - start, 60);
3902 Perl_re_printf( aTHX_
3903 "%s%s REx%s %s against %s\n",
3904 PL_colors[4], blurb, PL_colors[5], s0, s1);
3906 if (utf8_target||utf8_pat)
3907 Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n",
3908 utf8_pat ? "pattern" : "",
3909 utf8_pat && utf8_target ? " and " : "",
3910 utf8_target ? "string" : ""
3916 S_dump_exec_pos(pTHX_ const char *locinput,
3917 const regnode *scan,
3918 const char *loc_regeol,
3919 const char *loc_bostr,
3920 const char *loc_reg_starttry,
3921 const bool utf8_target,
3925 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3926 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3927 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3928 /* The part of the string before starttry has one color
3929 (pref0_len chars), between starttry and current
3930 position another one (pref_len - pref0_len chars),
3931 after the current position the third one.
3932 We assume that pref0_len <= pref_len, otherwise we
3933 decrease pref0_len. */
3934 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3935 ? (5 + taill) - l : locinput - loc_bostr;
3938 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3940 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3942 pref0_len = pref_len - (locinput - loc_reg_starttry);
3943 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3944 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3945 ? (5 + taill) - pref_len : loc_regeol - locinput);
3946 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3950 if (pref0_len > pref_len)
3951 pref0_len = pref_len;
3953 const int is_uni = utf8_target ? 1 : 0;
3955 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3956 (locinput - pref_len),pref0_len, 60, 4, 5);
3958 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3959 (locinput - pref_len + pref0_len),
3960 pref_len - pref0_len, 60, 2, 3);
3962 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3963 locinput, loc_regeol - locinput, 10, 0, 1);
3965 const STRLEN tlen=len0+len1+len2;
3966 Perl_re_printf( aTHX_
3967 "%4" IVdf " <%.*s%.*s%s%.*s>%*s|%4u| ",
3968 (IV)(locinput - loc_bostr),
3971 (docolor ? "" : "> <"),
3973 (int)(tlen > 19 ? 0 : 19 - tlen),
3981 /* reg_check_named_buff_matched()
3982 * Checks to see if a named buffer has matched. The data array of
3983 * buffer numbers corresponding to the buffer is expected to reside
3984 * in the regexp->data->data array in the slot stored in the ARG() of
3985 * node involved. Note that this routine doesn't actually care about the
3986 * name, that information is not preserved from compilation to execution.
3987 * Returns the index of the leftmost defined buffer with the given name
3988 * or 0 if non of the buffers matched.
3991 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3994 RXi_GET_DECL(rex,rexi);
3995 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3996 I32 *nums=(I32*)SvPVX(sv_dat);
3998 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
4000 for ( n=0; n<SvIVX(sv_dat); n++ ) {
4001 if ((I32)rex->lastparen >= nums[n] &&
4002 rex->offs[nums[n]].end != -1)
4012 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
4013 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
4015 /* This function determines if there are one or two characters that match
4016 * the first character of the passed-in EXACTish node <text_node>, and if
4017 * so, returns them in the passed-in pointers.
4019 * If it determines that no possible character in the target string can
4020 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
4021 * the first character in <text_node> requires UTF-8 to represent, and the
4022 * target string isn't in UTF-8.)
4024 * If there are more than two characters that could match the beginning of
4025 * <text_node>, or if more context is required to determine a match or not,
4026 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
4028 * The motiviation behind this function is to allow the caller to set up
4029 * tight loops for matching. If <text_node> is of type EXACT, there is
4030 * only one possible character that can match its first character, and so
4031 * the situation is quite simple. But things get much more complicated if
4032 * folding is involved. It may be that the first character of an EXACTFish
4033 * node doesn't participate in any possible fold, e.g., punctuation, so it
4034 * can be matched only by itself. The vast majority of characters that are
4035 * in folds match just two things, their lower and upper-case equivalents.
4036 * But not all are like that; some have multiple possible matches, or match
4037 * sequences of more than one character. This function sorts all that out.
4039 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
4040 * loop of trying to match A*, we know we can't exit where the thing
4041 * following it isn't a B. And something can't be a B unless it is the
4042 * beginning of B. By putting a quick test for that beginning in a tight
4043 * loop, we can rule out things that can't possibly be B without having to
4044 * break out of the loop, thus avoiding work. Similarly, if A is a single
4045 * character, we can make a tight loop matching A*, using the outputs of
4048 * If the target string to match isn't in UTF-8, and there aren't
4049 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
4050 * the one or two possible octets (which are characters in this situation)
4051 * that can match. In all cases, if there is only one character that can
4052 * match, *<c1p> and *<c2p> will be identical.
4054 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
4055 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
4056 * can match the beginning of <text_node>. They should be declared with at
4057 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
4058 * undefined what these contain.) If one or both of the buffers are
4059 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
4060 * corresponding invariant. If variant, the corresponding *<c1p> and/or
4061 * *<c2p> will be set to a negative number(s) that shouldn't match any code
4062 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
4063 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
4065 const bool utf8_target = reginfo->is_utf8_target;
4067 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
4068 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
4069 bool use_chrtest_void = FALSE;
4070 const bool is_utf8_pat = reginfo->is_utf8_pat;
4072 /* Used when we have both utf8 input and utf8 output, to avoid converting
4073 * to/from code points */
4074 bool utf8_has_been_setup = FALSE;
4078 U8 *pat = (U8*)STRING(text_node);
4079 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
4081 if (OP(text_node) == EXACT || OP(text_node) == EXACTL) {
4083 /* In an exact node, only one thing can be matched, that first
4084 * character. If both the pat and the target are UTF-8, we can just
4085 * copy the input to the output, avoiding finding the code point of
4090 else if (utf8_target) {
4091 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
4092 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
4093 utf8_has_been_setup = TRUE;
4096 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
4099 else { /* an EXACTFish node */
4100 U8 *pat_end = pat + STR_LEN(text_node);
4102 /* An EXACTFL node has at least some characters unfolded, because what
4103 * they match is not known until now. So, now is the time to fold
4104 * the first few of them, as many as are needed to determine 'c1' and
4105 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
4106 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
4107 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
4108 * need to fold as many characters as a single character can fold to,
4109 * so that later we can check if the first ones are such a multi-char
4110 * fold. But, in such a pattern only locale-problematic characters
4111 * aren't folded, so we can skip this completely if the first character
4112 * in the node isn't one of the tricky ones */
4113 if (OP(text_node) == EXACTFL) {
4115 if (! is_utf8_pat) {
4116 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
4118 folded[0] = folded[1] = 's';
4120 pat_end = folded + 2;
4123 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
4128 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
4130 *(d++) = (U8) toFOLD_LC(*s);
4135 _toFOLD_utf8_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_skip_len_mg(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 SSize_t endref = 0; /* offset of end of backref when ln is start */
5347 char *locinput = startpos;
5348 char *pushinput; /* where to continue after a PUSH */
5349 I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */
5351 bool result = 0; /* return value of S_regmatch */
5352 U32 depth = 0; /* depth of backtrack stack */
5353 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
5354 const U32 max_nochange_depth =
5355 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
5356 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
5357 regmatch_state *yes_state = NULL; /* state to pop to on success of
5359 /* mark_state piggy backs on the yes_state logic so that when we unwind
5360 the stack on success we can update the mark_state as we go */
5361 regmatch_state *mark_state = NULL; /* last mark state we have seen */
5362 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
5363 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
5365 bool no_final = 0; /* prevent failure from backtracking? */
5366 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
5367 char *startpoint = locinput;
5368 SV *popmark = NULL; /* are we looking for a mark? */
5369 SV *sv_commit = NULL; /* last mark name seen in failure */
5370 SV *sv_yes_mark = NULL; /* last mark name we have seen
5371 during a successful match */
5372 U32 lastopen = 0; /* last open we saw */
5373 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
5374 SV* const oreplsv = GvSVn(PL_replgv);
5375 /* these three flags are set by various ops to signal information to
5376 * the very next op. They have a useful lifetime of exactly one loop
5377 * iteration, and are not preserved or restored by state pushes/pops
5379 bool sw = 0; /* the condition value in (?(cond)a|b) */
5380 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
5381 int logical = 0; /* the following EVAL is:
5385 or the following IFMATCH/UNLESSM is:
5386 false: plain (?=foo)
5387 true: used as a condition: (?(?=foo))
5389 PAD* last_pad = NULL;
5391 U8 gimme = G_SCALAR;
5392 CV *caller_cv = NULL; /* who called us */
5393 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
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;
5399 I32 orig_savestack_ix = PL_savestack_ix;
5401 /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */
5402 #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL))
5403 # define SOLARIS_BAD_OPTIMIZER
5404 const U32 *pl_charclass_dup = PL_charclass;
5405 # define PL_charclass pl_charclass_dup
5409 GET_RE_DEBUG_FLAGS_DECL;
5412 /* protect against undef(*^R) */
5413 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
5415 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
5416 multicall_oldcatch = 0;
5417 PERL_UNUSED_VAR(multicall_cop);
5419 PERL_ARGS_ASSERT_REGMATCH;
5421 st = PL_regmatch_state;
5423 /* Note that nextchr is a byte even in UTF */
5427 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
5428 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5429 Perl_re_printf( aTHX_ "regmatch start\n" );
5432 while (scan != NULL) {
5435 next = scan + NEXT_OFF(scan);
5438 state_num = OP(scan);
5442 if (state_num <= REGNODE_MAX) {
5443 SV * const prop = sv_newmortal();
5444 regnode *rnext = regnext(scan);
5446 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5447 regprop(rex, prop, scan, reginfo, NULL);
5448 Perl_re_printf( aTHX_
5449 "%*s%" IVdf ":%s(%" IVdf ")\n",
5450 INDENT_CHARS(depth), "",
5451 (IV)(scan - rexi->program),
5453 (PL_regkind[OP(scan)] == END || !rnext) ?
5454 0 : (IV)(rnext - rexi->program));
5461 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
5463 switch (state_num) {
5464 case SBOL: /* /^../ and /\A../ */
5465 if (locinput == reginfo->strbeg)
5469 case MBOL: /* /^../m */
5470 if (locinput == reginfo->strbeg ||
5471 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
5478 if (locinput == reginfo->ganch)
5482 case KEEPS: /* \K */
5483 /* update the startpoint */
5484 st->u.keeper.val = rex->offs[0].start;
5485 rex->offs[0].start = locinput - reginfo->strbeg;
5486 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
5487 NOT_REACHED; /* NOTREACHED */
5489 case KEEPS_next_fail:
5490 /* rollback the start point change */
5491 rex->offs[0].start = st->u.keeper.val;
5493 NOT_REACHED; /* NOTREACHED */
5495 case MEOL: /* /..$/m */
5496 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5500 case SEOL: /* /..$/ */
5501 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5503 if (reginfo->strend - locinput > 1)
5508 if (!NEXTCHR_IS_EOS)
5512 case SANY: /* /./s */
5515 goto increment_locinput;
5517 case REG_ANY: /* /./ */
5518 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
5520 goto increment_locinput;
5524 #define ST st->u.trie
5525 case TRIEC: /* (ab|cd) with known charclass */
5526 /* In this case the charclass data is available inline so
5527 we can fail fast without a lot of extra overhead.
5529 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
5531 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n",
5532 depth, PL_colors[4], PL_colors[5])
5535 NOT_REACHED; /* NOTREACHED */
5538 case TRIE: /* (ab|cd) */
5539 /* the basic plan of execution of the trie is:
5540 * At the beginning, run though all the states, and
5541 * find the longest-matching word. Also remember the position
5542 * of the shortest matching word. For example, this pattern:
5545 * when matched against the string "abcde", will generate
5546 * accept states for all words except 3, with the longest
5547 * matching word being 4, and the shortest being 2 (with
5548 * the position being after char 1 of the string).
5550 * Then for each matching word, in word order (i.e. 1,2,4,5),
5551 * we run the remainder of the pattern; on each try setting
5552 * the current position to the character following the word,
5553 * returning to try the next word on failure.
5555 * We avoid having to build a list of words at runtime by
5556 * using a compile-time structure, wordinfo[].prev, which
5557 * gives, for each word, the previous accepting word (if any).
5558 * In the case above it would contain the mappings 1->2, 2->0,
5559 * 3->0, 4->5, 5->1. We can use this table to generate, from
5560 * the longest word (4 above), a list of all words, by
5561 * following the list of prev pointers; this gives us the
5562 * unordered list 4,5,1,2. Then given the current word we have
5563 * just tried, we can go through the list and find the
5564 * next-biggest word to try (so if we just failed on word 2,
5565 * the next in the list is 4).
5567 * Since at runtime we don't record the matching position in
5568 * the string for each word, we have to work that out for
5569 * each word we're about to process. The wordinfo table holds
5570 * the character length of each word; given that we recorded
5571 * at the start: the position of the shortest word and its
5572 * length in chars, we just need to move the pointer the
5573 * difference between the two char lengths. Depending on
5574 * Unicode status and folding, that's cheap or expensive.
5576 * This algorithm is optimised for the case where are only a
5577 * small number of accept states, i.e. 0,1, or maybe 2.
5578 * With lots of accepts states, and having to try all of them,
5579 * it becomes quadratic on number of accept states to find all
5584 /* what type of TRIE am I? (utf8 makes this contextual) */
5585 DECL_TRIE_TYPE(scan);
5587 /* what trie are we using right now */
5588 reg_trie_data * const trie
5589 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
5590 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
5591 U32 state = trie->startstate;
5593 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
5594 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5596 && UTF8_IS_ABOVE_LATIN1(nextchr)
5597 && scan->flags == EXACTL)
5599 /* We only output for EXACTL, as we let the folder
5600 * output this message for EXACTFLU8 to avoid
5602 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
5607 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
5609 if (trie->states[ state ].wordnum) {
5611 Perl_re_exec_indentf( aTHX_ "%smatched empty string...%s\n",
5612 depth, PL_colors[4], PL_colors[5])
5618 Perl_re_exec_indentf( aTHX_ "%sfailed to match trie start class...%s\n",
5619 depth, PL_colors[4], PL_colors[5])
5626 U8 *uc = ( U8* )locinput;
5630 U8 *uscan = (U8*)NULL;
5631 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
5632 U32 charcount = 0; /* how many input chars we have matched */
5633 U32 accepted = 0; /* have we seen any accepting states? */
5635 ST.jump = trie->jump;
5638 ST.longfold = FALSE; /* char longer if folded => it's harder */
5641 /* fully traverse the TRIE; note the position of the
5642 shortest accept state and the wordnum of the longest
5645 while ( state && uc <= (U8*)(reginfo->strend) ) {
5646 U32 base = trie->states[ state ].trans.base;
5650 wordnum = trie->states[ state ].wordnum;
5652 if (wordnum) { /* it's an accept state */
5655 /* record first match position */
5657 ST.firstpos = (U8*)locinput;
5662 ST.firstchars = charcount;
5665 if (!ST.nextword || wordnum < ST.nextword)
5666 ST.nextword = wordnum;
5667 ST.topword = wordnum;
5670 DEBUG_TRIE_EXECUTE_r({
5671 DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth );
5673 PerlIO_printf( Perl_debug_log,
5674 "%*s%sState: %4" UVxf " Accepted: %c ",
5675 INDENT_CHARS(depth), "", PL_colors[4],
5676 (UV)state, (accepted ? 'Y' : 'N'));
5679 /* read a char and goto next state */
5680 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
5682 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
5683 uscan, len, uvc, charid, foldlen,
5690 base + charid - 1 - trie->uniquecharcount)) >= 0)
5692 && ((U32)offset < trie->lasttrans)
5693 && trie->trans[offset].check == state)
5695 state = trie->trans[offset].next;
5706 DEBUG_TRIE_EXECUTE_r(
5707 Perl_re_printf( aTHX_
5708 "Charid:%3x CP:%4" UVxf " After State: %4" UVxf "%s\n",
5709 charid, uvc, (UV)state, PL_colors[5] );
5715 /* calculate total number of accept states */
5720 w = trie->wordinfo[w].prev;
5723 ST.accepted = accepted;
5727 Perl_re_exec_indentf( aTHX_ "%sgot %" IVdf " possible matches%s\n",
5729 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
5731 goto trie_first_try; /* jump into the fail handler */
5733 NOT_REACHED; /* NOTREACHED */
5735 case TRIE_next_fail: /* we failed - try next alternative */
5739 /* undo any captures done in the tail part of a branch,
5741 * /(?:X(.)(.)|Y(.)).../
5742 * where the trie just matches X then calls out to do the
5743 * rest of the branch */
5744 REGCP_UNWIND(ST.cp);
5745 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5747 if (!--ST.accepted) {
5749 Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n",
5757 /* Find next-highest word to process. Note that this code
5758 * is O(N^2) per trie run (O(N) per branch), so keep tight */
5761 U16 const nextword = ST.nextword;
5762 reg_trie_wordinfo * const wordinfo
5763 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
5764 for (word=ST.topword; word; word=wordinfo[word].prev) {
5765 if (word > nextword && (!min || word < min))
5778 ST.lastparen = rex->lastparen;
5779 ST.lastcloseparen = rex->lastcloseparen;
5783 /* find start char of end of current word */
5785 U32 chars; /* how many chars to skip */
5786 reg_trie_data * const trie
5787 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
5789 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
5791 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
5796 /* the hard option - fold each char in turn and find
5797 * its folded length (which may be different */
5798 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
5806 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
5814 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
5819 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
5835 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
5836 ? ST.jump[ST.nextword]
5840 Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n",
5848 if ( ST.accepted > 1 || has_cutgroup || ST.jump ) {
5849 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
5850 NOT_REACHED; /* NOTREACHED */
5852 /* only one choice left - just continue */
5854 AV *const trie_words
5855 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
5856 SV ** const tmp = trie_words
5857 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL;
5858 SV *sv= tmp ? sv_newmortal() : NULL;
5860 Perl_re_exec_indentf( aTHX_ "%sonly one match left, short-circuiting: #%d <%s>%s\n",
5861 depth, PL_colors[4],
5863 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
5864 PL_colors[0], PL_colors[1],
5865 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
5867 : "not compiled under -Dr",
5871 locinput = (char*)uc;
5872 continue; /* execute rest of RE */
5877 case EXACTL: /* /abc/l */
5878 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5880 /* Complete checking would involve going through every character
5881 * matched by the string to see if any is above latin1. But the
5882 * comparision otherwise might very well be a fast assembly
5883 * language routine, and I (khw) don't think slowing things down
5884 * just to check for this warning is worth it. So this just checks
5885 * the first character */
5886 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
5887 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
5890 case EXACT: { /* /abc/ */
5891 char *s = STRING(scan);
5893 if (utf8_target != is_utf8_pat) {
5894 /* The target and the pattern have differing utf8ness. */
5896 const char * const e = s + ln;
5899 /* The target is utf8, the pattern is not utf8.
5900 * Above-Latin1 code points can't match the pattern;
5901 * invariants match exactly, and the other Latin1 ones need
5902 * to be downgraded to a single byte in order to do the
5903 * comparison. (If we could be confident that the target
5904 * is not malformed, this could be refactored to have fewer
5905 * tests by just assuming that if the first bytes match, it
5906 * is an invariant, but there are tests in the test suite
5907 * dealing with (??{...}) which violate this) */
5909 if (l >= reginfo->strend
5910 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
5914 if (UTF8_IS_INVARIANT(*(U8*)l)) {
5921 if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
5931 /* The target is not utf8, the pattern is utf8. */
5933 if (l >= reginfo->strend
5934 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
5938 if (UTF8_IS_INVARIANT(*(U8*)s)) {
5945 if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
5957 /* The target and the pattern have the same utf8ness. */
5958 /* Inline the first character, for speed. */
5959 if (reginfo->strend - locinput < ln
5960 || UCHARAT(s) != nextchr
5961 || (ln > 1 && memNE(s, locinput, ln)))
5970 case EXACTFL: { /* /abc/il */
5972 const U8 * fold_array;
5974 U32 fold_utf8_flags;
5976 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5977 folder = foldEQ_locale;
5978 fold_array = PL_fold_locale;
5979 fold_utf8_flags = FOLDEQ_LOCALE;
5982 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
5983 is effectively /u; hence to match, target
5985 if (! utf8_target) {
5988 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S1_ALREADY_FOLDED
5989 | FOLDEQ_S1_FOLDS_SANE;
5990 folder = foldEQ_latin1;
5991 fold_array = PL_fold_latin1;
5994 case EXACTFU_SS: /* /\x{df}/iu */
5995 case EXACTFU: /* /abc/iu */
5996 folder = foldEQ_latin1;
5997 fold_array = PL_fold_latin1;
5998 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
6001 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
6003 assert(! is_utf8_pat);
6005 case EXACTFA: /* /abc/iaa */
6006 folder = foldEQ_latin1;
6007 fold_array = PL_fold_latin1;
6008 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6011 case EXACTF: /* /abc/i This node only generated for
6012 non-utf8 patterns */
6013 assert(! is_utf8_pat);
6015 fold_array = PL_fold;
6016 fold_utf8_flags = 0;
6024 || state_num == EXACTFU_SS
6025 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
6027 /* Either target or the pattern are utf8, or has the issue where
6028 * the fold lengths may differ. */
6029 const char * const l = locinput;
6030 char *e = reginfo->strend;
6032 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
6033 l, &e, 0, utf8_target, fold_utf8_flags))
6041 /* Neither the target nor the pattern are utf8 */
6042 if (UCHARAT(s) != nextchr
6044 && UCHARAT(s) != fold_array[nextchr])
6048 if (reginfo->strend - locinput < ln)
6050 if (ln > 1 && ! folder(s, locinput, ln))
6056 case NBOUNDL: /* /\B/l */
6060 case BOUNDL: /* /\b/l */
6063 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6065 if (FLAGS(scan) != TRADITIONAL_BOUND) {
6066 if (! IN_UTF8_CTYPE_LOCALE) {
6067 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
6068 B_ON_NON_UTF8_LOCALE_IS_WRONG);
6074 if (locinput == reginfo->strbeg)
6075 b1 = isWORDCHAR_LC('\n');
6077 b1 = isWORDCHAR_LC_utf8_safe(reghop3((U8*)locinput, -1,
6078 (U8*)(reginfo->strbeg)),
6079 (U8*)(reginfo->strend));
6081 b2 = (NEXTCHR_IS_EOS)
6082 ? isWORDCHAR_LC('\n')
6083 : isWORDCHAR_LC_utf8_safe((U8*) locinput,
6084 (U8*) reginfo->strend);
6086 else { /* Here the string isn't utf8 */
6087 b1 = (locinput == reginfo->strbeg)
6088 ? isWORDCHAR_LC('\n')
6089 : isWORDCHAR_LC(UCHARAT(locinput - 1));
6090 b2 = (NEXTCHR_IS_EOS)
6091 ? isWORDCHAR_LC('\n')
6092 : isWORDCHAR_LC(nextchr);
6094 if (to_complement ^ (b1 == b2)) {
6100 case NBOUND: /* /\B/ */
6104 case BOUND: /* /\b/ */
6108 goto bound_ascii_match_only;
6110 case NBOUNDA: /* /\B/a */
6114 case BOUNDA: /* /\b/a */
6118 bound_ascii_match_only:
6119 /* Here the string isn't utf8, or is utf8 and only ascii characters
6120 * are to match \w. In the latter case looking at the byte just
6121 * prior to the current one may be just the final byte of a
6122 * multi-byte character. This is ok. There are two cases:
6123 * 1) it is a single byte character, and then the test is doing
6124 * just what it's supposed to.
6125 * 2) it is a multi-byte character, in which case the final byte is
6126 * never mistakable for ASCII, and so the test will say it is
6127 * not a word character, which is the correct answer. */
6128 b1 = (locinput == reginfo->strbeg)
6129 ? isWORDCHAR_A('\n')
6130 : isWORDCHAR_A(UCHARAT(locinput - 1));
6131 b2 = (NEXTCHR_IS_EOS)
6132 ? isWORDCHAR_A('\n')
6133 : isWORDCHAR_A(nextchr);
6134 if (to_complement ^ (b1 == b2)) {
6140 case NBOUNDU: /* /\B/u */
6144 case BOUNDU: /* /\b/u */
6147 if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) {
6150 else if (utf8_target) {
6152 switch((bound_type) FLAGS(scan)) {
6153 case TRADITIONAL_BOUND:
6156 b1 = (locinput == reginfo->strbeg)
6157 ? 0 /* isWORDCHAR_L1('\n') */
6158 : isWORDCHAR_utf8_safe(
6159 reghop3((U8*)locinput,
6161 (U8*)(reginfo->strbeg)),
6162 (U8*) reginfo->strend);
6163 b2 = (NEXTCHR_IS_EOS)
6164 ? 0 /* isWORDCHAR_L1('\n') */
6165 : isWORDCHAR_utf8_safe((U8*)locinput,
6166 (U8*) reginfo->strend);
6167 match = cBOOL(b1 != b2);
6171 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6172 match = TRUE; /* GCB always matches at begin and
6176 /* Find the gcb values of previous and current
6177 * chars, then see if is a break point */
6178 match = isGCB(getGCB_VAL_UTF8(
6179 reghop3((U8*)locinput,
6181 (U8*)(reginfo->strbeg)),
6182 (U8*) reginfo->strend),
6183 getGCB_VAL_UTF8((U8*) locinput,
6184 (U8*) reginfo->strend),
6185 (U8*) reginfo->strbeg,
6192 if (locinput == reginfo->strbeg) {
6195 else if (NEXTCHR_IS_EOS) {
6199 match = isLB(getLB_VAL_UTF8(
6200 reghop3((U8*)locinput,
6202 (U8*)(reginfo->strbeg)),
6203 (U8*) reginfo->strend),
6204 getLB_VAL_UTF8((U8*) locinput,
6205 (U8*) reginfo->strend),
6206 (U8*) reginfo->strbeg,
6208 (U8*) reginfo->strend,
6213 case SB_BOUND: /* Always matches at begin and end */
6214 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6218 match = isSB(getSB_VAL_UTF8(
6219 reghop3((U8*)locinput,
6221 (U8*)(reginfo->strbeg)),
6222 (U8*) reginfo->strend),
6223 getSB_VAL_UTF8((U8*) locinput,
6224 (U8*) reginfo->strend),
6225 (U8*) reginfo->strbeg,
6227 (U8*) reginfo->strend,
6233 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6237 match = isWB(WB_UNKNOWN,
6239 reghop3((U8*)locinput,
6241 (U8*)(reginfo->strbeg)),
6242 (U8*) reginfo->strend),
6243 getWB_VAL_UTF8((U8*) locinput,
6244 (U8*) reginfo->strend),
6245 (U8*) reginfo->strbeg,
6247 (U8*) reginfo->strend,
6253 else { /* Not utf8 target */
6254 switch((bound_type) FLAGS(scan)) {
6255 case TRADITIONAL_BOUND:
6258 b1 = (locinput == reginfo->strbeg)
6259 ? 0 /* isWORDCHAR_L1('\n') */
6260 : isWORDCHAR_L1(UCHARAT(locinput - 1));
6261 b2 = (NEXTCHR_IS_EOS)
6262 ? 0 /* isWORDCHAR_L1('\n') */
6263 : isWORDCHAR_L1(nextchr);
6264 match = cBOOL(b1 != b2);
6269 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6270 match = TRUE; /* GCB always matches at begin and
6273 else { /* Only CR-LF combo isn't a GCB in 0-255
6275 match = UCHARAT(locinput - 1) != '\r'
6276 || UCHARAT(locinput) != '\n';
6281 if (locinput == reginfo->strbeg) {
6284 else if (NEXTCHR_IS_EOS) {
6288 match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)),
6289 getLB_VAL_CP(UCHARAT(locinput)),
6290 (U8*) reginfo->strbeg,
6292 (U8*) reginfo->strend,
6297 case SB_BOUND: /* Always matches at begin and end */
6298 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6302 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)),
6303 getSB_VAL_CP(UCHARAT(locinput)),
6304 (U8*) reginfo->strbeg,
6306 (U8*) reginfo->strend,
6312 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6316 match = isWB(WB_UNKNOWN,
6317 getWB_VAL_CP(UCHARAT(locinput -1)),
6318 getWB_VAL_CP(UCHARAT(locinput)),
6319 (U8*) reginfo->strbeg,
6321 (U8*) reginfo->strend,
6328 if (to_complement ^ ! match) {
6333 case ANYOFL: /* /[abc]/l */
6334 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6336 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE)
6338 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
6341 case ANYOFD: /* /[abc]/d */
6342 case ANYOF: /* /[abc]/ */
6345 if (utf8_target && ! UTF8_IS_INVARIANT(*locinput)) {
6346 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
6349 locinput += UTF8SKIP(locinput);
6352 if (!REGINCLASS(rex, scan, (U8*)locinput, utf8_target))
6358 /* The argument (FLAGS) to all the POSIX node types is the class number
6361 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
6365 case POSIXL: /* \w or [:punct:] etc. under /l */
6366 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6370 /* Use isFOO_lc() for characters within Latin1. (Note that
6371 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6372 * wouldn't be invariant) */
6373 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6374 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
6382 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6383 /* An above Latin-1 code point, or malformed */
6384 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
6386 goto utf8_posix_above_latin1;
6389 /* Here is a UTF-8 variant code point below 256 and the target is
6391 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
6392 EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6393 *(locinput + 1))))))
6398 goto increment_locinput;
6400 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
6404 case POSIXD: /* \w or [:punct:] etc. under /d */
6410 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
6412 if (NEXTCHR_IS_EOS) {
6416 /* All UTF-8 variants match */
6417 if (! UTF8_IS_INVARIANT(nextchr)) {
6418 goto increment_locinput;
6424 case POSIXA: /* \w or [:punct:] etc. under /a */
6427 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
6428 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
6429 * character is a single byte */
6431 if (NEXTCHR_IS_EOS) {
6437 if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
6443 /* Here we are either not in utf8, or we matched a utf8-invariant,
6444 * so the next char is the next byte */
6448 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
6452 case POSIXU: /* \w or [:punct:] etc. under /u */
6454 if (NEXTCHR_IS_EOS) {
6458 /* Use _generic_isCC() for characters within Latin1. (Note that
6459 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6460 * wouldn't be invariant) */
6461 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6462 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
6469 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6470 if (! (to_complement
6471 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6479 else { /* Handle above Latin-1 code points */
6480 utf8_posix_above_latin1:
6481 classnum = (_char_class_number) FLAGS(scan);
6482 if (classnum < _FIRST_NON_SWASH_CC) {
6484 /* Here, uses a swash to find such code points. Load if if
6485 * not done already */
6486 if (! PL_utf8_swash_ptrs[classnum]) {
6487 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
6488 PL_utf8_swash_ptrs[classnum]
6489 = _core_swash_init("utf8",
6492 PL_XPosix_ptrs[classnum], &flags);
6494 if (! (to_complement
6495 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
6496 (U8 *) locinput, TRUE))))
6501 else { /* Here, uses macros to find above Latin-1 code points */
6503 case _CC_ENUM_SPACE:
6504 if (! (to_complement
6505 ^ cBOOL(is_XPERLSPACE_high(locinput))))
6510 case _CC_ENUM_BLANK:
6511 if (! (to_complement
6512 ^ cBOOL(is_HORIZWS_high(locinput))))
6517 case _CC_ENUM_XDIGIT:
6518 if (! (to_complement
6519 ^ cBOOL(is_XDIGIT_high(locinput))))
6524 case _CC_ENUM_VERTSPACE:
6525 if (! (to_complement
6526 ^ cBOOL(is_VERTWS_high(locinput))))
6531 default: /* The rest, e.g. [:cntrl:], can't match
6533 if (! to_complement) {
6539 locinput += UTF8SKIP(locinput);
6543 case CLUMP: /* Match \X: logical Unicode character. This is defined as
6544 a Unicode extended Grapheme Cluster */
6547 if (! utf8_target) {
6549 /* Match either CR LF or '.', as all the other possibilities
6551 locinput++; /* Match the . or CR */
6552 if (nextchr == '\r' /* And if it was CR, and the next is LF,
6554 && locinput < reginfo->strend
6555 && UCHARAT(locinput) == '\n')
6562 /* Get the gcb type for the current character */
6563 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput,
6564 (U8*) reginfo->strend);
6566 /* Then scan through the input until we get to the first
6567 * character whose type is supposed to be a gcb with the
6568 * current character. (There is always a break at the
6570 locinput += UTF8SKIP(locinput);
6571 while (locinput < reginfo->strend) {
6572 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput,
6573 (U8*) reginfo->strend);
6574 if (isGCB(prev_gcb, cur_gcb,
6575 (U8*) reginfo->strbeg, (U8*) locinput,
6582 locinput += UTF8SKIP(locinput);
6589 case NREFFL: /* /\g{name}/il */
6590 { /* The capture buffer cases. The ones beginning with N for the
6591 named buffers just convert to the equivalent numbered and
6592 pretend they were called as the corresponding numbered buffer
6594 /* don't initialize these in the declaration, it makes C++
6599 const U8 *fold_array;
6602 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6603 folder = foldEQ_locale;
6604 fold_array = PL_fold_locale;
6606 utf8_fold_flags = FOLDEQ_LOCALE;
6609 case NREFFA: /* /\g{name}/iaa */
6610 folder = foldEQ_latin1;
6611 fold_array = PL_fold_latin1;
6613 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6616 case NREFFU: /* /\g{name}/iu */
6617 folder = foldEQ_latin1;
6618 fold_array = PL_fold_latin1;
6620 utf8_fold_flags = 0;
6623 case NREFF: /* /\g{name}/i */
6625 fold_array = PL_fold;
6627 utf8_fold_flags = 0;
6630 case NREF: /* /\g{name}/ */
6634 utf8_fold_flags = 0;
6637 /* For the named back references, find the corresponding buffer
6639 n = reg_check_named_buff_matched(rex,scan);
6644 goto do_nref_ref_common;
6646 case REFFL: /* /\1/il */
6647 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6648 folder = foldEQ_locale;
6649 fold_array = PL_fold_locale;
6650 utf8_fold_flags = FOLDEQ_LOCALE;
6653 case REFFA: /* /\1/iaa */
6654 folder = foldEQ_latin1;
6655 fold_array = PL_fold_latin1;
6656 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6659 case REFFU: /* /\1/iu */
6660 folder = foldEQ_latin1;
6661 fold_array = PL_fold_latin1;
6662 utf8_fold_flags = 0;
6665 case REFF: /* /\1/i */
6667 fold_array = PL_fold;
6668 utf8_fold_flags = 0;
6671 case REF: /* /\1/ */
6674 utf8_fold_flags = 0;
6678 n = ARG(scan); /* which paren pair */
6681 ln = rex->offs[n].start;
6682 endref = rex->offs[n].end;
6683 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
6684 if (rex->lastparen < n || ln == -1 || endref == -1)
6685 sayNO; /* Do not match unless seen CLOSEn. */
6689 s = reginfo->strbeg + ln;
6690 if (type != REF /* REF can do byte comparison */
6691 && (utf8_target || type == REFFU || type == REFFL))
6693 char * limit = reginfo->strend;
6695 /* This call case insensitively compares the entire buffer
6696 * at s, with the current input starting at locinput, but
6697 * not going off the end given by reginfo->strend, and
6698 * returns in <limit> upon success, how much of the
6699 * current input was matched */
6700 if (! foldEQ_utf8_flags(s, NULL, endref - ln, utf8_target,
6701 locinput, &limit, 0, utf8_target, utf8_fold_flags))
6709 /* Not utf8: Inline the first character, for speed. */
6710 if (!NEXTCHR_IS_EOS &&
6711 UCHARAT(s) != nextchr &&
6713 UCHARAT(s) != fold_array[nextchr]))
6716 if (locinput + ln > reginfo->strend)
6718 if (ln > 1 && (type == REF
6719 ? memNE(s, locinput, ln)
6720 : ! folder(s, locinput, ln)))
6726 case NOTHING: /* null op; e.g. the 'nothing' following
6727 * the '*' in m{(a+|b)*}' */
6729 case TAIL: /* placeholder while compiling (A|B|C) */
6733 #define ST st->u.eval
6734 #define CUR_EVAL cur_eval->u.eval
6740 regexp_internal *rei;
6741 regnode *startpoint;
6744 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
6745 arg= (U32)ARG(scan);
6746 if (cur_eval && cur_eval->locinput == locinput) {
6747 if ( ++nochange_depth > max_nochange_depth )
6749 "Pattern subroutine nesting without pos change"
6750 " exceeded limit in regex");
6757 startpoint = scan + ARG2L(scan);
6758 EVAL_CLOSE_PAREN_SET( st, arg );
6759 /* Detect infinite recursion
6761 * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/
6762 * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever.
6763 * So we track the position in the string we are at each time
6764 * we recurse and if we try to enter the same routine twice from
6765 * the same position we throw an error.
6767 if ( rex->recurse_locinput[arg] == locinput ) {
6768 /* FIXME: we should show the regop that is failing as part
6769 * of the error message. */
6770 Perl_croak(aTHX_ "Infinite recursion in regex");
6772 ST.prev_recurse_locinput= rex->recurse_locinput[arg];
6773 rex->recurse_locinput[arg]= locinput;
6776 GET_RE_DEBUG_FLAGS_DECL;
6778 Perl_re_exec_indentf( aTHX_
6779 "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n",
6780 depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg]
6786 /* Save all the positions seen so far. */
6787 ST.cp = regcppush(rex, 0, maxopenparen);
6788 REGCP_SET(ST.lastcp);
6790 /* and then jump to the code we share with EVAL */
6791 goto eval_recurse_doit;
6794 case EVAL: /* /(?{...})B/ /(??{A})B/ and /(?(?{...})X|Y)B/ */
6795 if (cur_eval && cur_eval->locinput==locinput) {
6796 if ( ++nochange_depth > max_nochange_depth )
6797 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
6802 /* execute the code in the {...} */
6806 OP * const oop = PL_op;
6807 COP * const ocurcop = PL_curcop;
6811 /* save *all* paren positions */
6812 regcppush(rex, 0, maxopenparen);
6813 REGCP_SET(ST.lastcp);
6816 caller_cv = find_runcv(NULL);
6820 if (rexi->data->what[n] == 'r') { /* code from an external qr */
6822 (REGEXP*)(rexi->data->data[n])
6824 nop = (OP*)rexi->data->data[n+1];
6826 else if (rexi->data->what[n] == 'l') { /* literal code */
6828 nop = (OP*)rexi->data->data[n];
6829 assert(CvDEPTH(newcv));
6832 /* literal with own CV */
6833 assert(rexi->data->what[n] == 'L');
6834 newcv = rex->qr_anoncv;
6835 nop = (OP*)rexi->data->data[n];
6838 /* Some notes about MULTICALL and the context and save stacks.
6841 * /...(?{ my $x)}...(?{ my $y)}...(?{ my $z)}.../
6842 * since codeblocks don't introduce a new scope (so that
6843 * local() etc accumulate), at the end of a successful
6844 * match there will be a SAVEt_CLEARSV on the savestack
6845 * for each of $x, $y, $z. If the three code blocks above
6846 * happen to have come from different CVs (e.g. via
6847 * embedded qr//s), then we must ensure that during any
6848 * savestack unwinding, PL_comppad always points to the
6849 * right pad at each moment. We achieve this by
6850 * interleaving SAVEt_COMPPAD's on the savestack whenever
6851 * there is a change of pad.
6852 * In theory whenever we call a code block, we should
6853 * push a CXt_SUB context, then pop it on return from
6854 * that code block. This causes a bit of an issue in that
6855 * normally popping a context also clears the savestack
6856 * back to cx->blk_oldsaveix, but here we specifically
6857 * don't want to clear the save stack on exit from the
6859 * Also for efficiency we don't want to keep pushing and
6860 * popping the single SUB context as we backtrack etc.
6861 * So instead, we push a single context the first time
6862 * we need, it, then hang onto it until the end of this
6863 * function. Whenever we encounter a new code block, we
6864 * update the CV etc if that's changed. During the times
6865 * in this function where we're not executing a code
6866 * block, having the SUB context still there is a bit
6867 * naughty - but we hope that no-one notices.
6868 * When the SUB context is initially pushed, we fake up
6869 * cx->blk_oldsaveix to be as if we'd pushed this context
6870 * on first entry to S_regmatch rather than at some random
6871 * point during the regexe execution. That way if we
6872 * croak, popping the context stack will ensure that
6873 * *everything* SAVEd by this function is undone and then
6874 * the context popped, rather than e.g., popping the
6875 * context (and restoring the original PL_comppad) then
6876 * popping more of the savestack and restoring a bad
6880 /* If this is the first EVAL, push a MULTICALL. On
6881 * subsequent calls, if we're executing a different CV, or
6882 * if PL_comppad has got messed up from backtracking
6883 * through SAVECOMPPADs, then refresh the context.
6885 if (newcv != last_pushed_cv || PL_comppad != last_pad)
6887 U8 flags = (CXp_SUB_RE |
6888 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
6890 if (last_pushed_cv) {
6891 CHANGE_MULTICALL_FLAGS(newcv, flags);
6894 PUSH_MULTICALL_FLAGS(newcv, flags);
6896 /* see notes above */
6897 CX_CUR()->blk_oldsaveix = orig_savestack_ix;
6899 last_pushed_cv = newcv;
6902 /* these assignments are just to silence compiler
6904 multicall_cop = NULL;
6906 last_pad = PL_comppad;
6908 /* the initial nextstate you would normally execute
6909 * at the start of an eval (which would cause error
6910 * messages to come from the eval), may be optimised
6911 * away from the execution path in the regex code blocks;
6912 * so manually set PL_curcop to it initially */
6914 OP *o = cUNOPx(nop)->op_first;
6915 assert(o->op_type == OP_NULL);
6916 if (o->op_targ == OP_SCOPE) {
6917 o = cUNOPo->op_first;
6920 assert(o->op_targ == OP_LEAVE);
6921 o = cUNOPo->op_first;
6922 assert(o->op_type == OP_ENTER);
6926 if (o->op_type != OP_STUB) {
6927 assert( o->op_type == OP_NEXTSTATE
6928 || o->op_type == OP_DBSTATE
6929 || (o->op_type == OP_NULL
6930 && ( o->op_targ == OP_NEXTSTATE
6931 || o->op_targ == OP_DBSTATE
6935 PL_curcop = (COP*)o;
6940 DEBUG_STATE_r( Perl_re_printf( aTHX_
6941 " re EVAL PL_op=0x%" UVxf "\n", PTR2UV(nop)) );
6943 rex->offs[0].end = locinput - reginfo->strbeg;
6944 if (reginfo->info_aux_eval->pos_magic)
6945 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
6946 reginfo->sv, reginfo->strbeg,
6947 locinput - reginfo->strbeg);
6950 SV *sv_mrk = get_sv("REGMARK", 1);
6951 sv_setsv(sv_mrk, sv_yes_mark);
6954 /* we don't use MULTICALL here as we want to call the
6955 * first op of the block of interest, rather than the
6956 * first op of the sub. Also, we don't want to free
6957 * the savestack frame */
6958 before = (IV)(SP-PL_stack_base);
6960 CALLRUNOPS(aTHX); /* Scalar context. */
6962 if ((IV)(SP-PL_stack_base) == before)
6963 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
6969 /* before restoring everything, evaluate the returned
6970 * value, so that 'uninit' warnings don't use the wrong
6971 * PL_op or pad. Also need to process any magic vars
6972 * (e.g. $1) *before* parentheses are restored */
6977 if (logical == 0) /* (?{})/ */
6978 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
6979 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
6980 sw = cBOOL(SvTRUE(ret));
6983 else { /* /(??{}) */
6984 /* if its overloaded, let the regex compiler handle
6985 * it; otherwise extract regex, or stringify */
6986 if (SvGMAGICAL(ret))
6987 ret = sv_mortalcopy(ret);
6988 if (!SvAMAGIC(ret)) {
6992 if (SvTYPE(sv) == SVt_REGEXP)
6993 re_sv = (REGEXP*) sv;
6994 else if (SvSMAGICAL(ret)) {
6995 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
6997 re_sv = (REGEXP *) mg->mg_obj;
7000 /* force any undef warnings here */
7001 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
7002 ret = sv_mortalcopy(ret);
7003 (void) SvPV_force_nolen(ret);
7009 /* *** Note that at this point we don't restore
7010 * PL_comppad, (or pop the CxSUB) on the assumption it may
7011 * be used again soon. This is safe as long as nothing
7012 * in the regexp code uses the pad ! */
7014 PL_curcop = ocurcop;
7015 regcp_restore(rex, ST.lastcp, &maxopenparen);
7016 PL_curpm_under = PL_curpm;
7017 PL_curpm = PL_reg_curpm;
7020 PUSH_STATE_GOTO(EVAL_B, next, locinput);
7025 /* only /(??{})/ from now on */
7028 /* extract RE object from returned value; compiling if
7032 re_sv = reg_temp_copy(NULL, re_sv);
7037 if (SvUTF8(ret) && IN_BYTES) {
7038 /* In use 'bytes': make a copy of the octet
7039 * sequence, but without the flag on */
7041 const char *const p = SvPV(ret, len);
7042 ret = newSVpvn_flags(p, len, SVs_TEMP);
7044 if (rex->intflags & PREGf_USE_RE_EVAL)
7045 pm_flags |= PMf_USE_RE_EVAL;
7047 /* if we got here, it should be an engine which
7048 * supports compiling code blocks and stuff */
7049 assert(rex->engine && rex->engine->op_comp);
7050 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
7051 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
7052 rex->engine, NULL, NULL,
7053 /* copy /msixn etc to inner pattern */
7058 & (SVs_TEMP | SVs_GMG | SVf_ROK))
7059 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
7060 /* This isn't a first class regexp. Instead, it's
7061 caching a regexp onto an existing, Perl visible
7063 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
7069 RXp_MATCH_COPIED_off(re);
7070 re->subbeg = rex->subbeg;
7071 re->sublen = rex->sublen;
7072 re->suboffset = rex->suboffset;
7073 re->subcoffset = rex->subcoffset;
7075 re->lastcloseparen = 0;
7078 debug_start_match(re_sv, utf8_target, locinput,
7079 reginfo->strend, "Matching embedded");
7081 startpoint = rei->program + 1;
7082 EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0;
7083 * close_paren only for GOSUB */
7084 ST.prev_recurse_locinput= NULL; /* only used for GOSUB */
7085 /* Save all the seen positions so far. */
7086 ST.cp = regcppush(rex, 0, maxopenparen);
7087 REGCP_SET(ST.lastcp);
7088 /* and set maxopenparen to 0, since we are starting a "fresh" match */
7090 /* run the pattern returned from (??{...}) */
7092 eval_recurse_doit: /* Share code with GOSUB below this line
7093 * At this point we expect the stack context to be
7094 * set up correctly */
7096 /* invalidate the S-L poscache. We're now executing a
7097 * different set of WHILEM ops (and their associated
7098 * indexes) against the same string, so the bits in the
7099 * cache are meaningless. Setting maxiter to zero forces
7100 * the cache to be invalidated and zeroed before reuse.
7101 * XXX This is too dramatic a measure. Ideally we should
7102 * save the old cache and restore when running the outer
7104 reginfo->poscache_maxiter = 0;
7106 /* the new regexp might have a different is_utf8_pat than we do */
7107 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
7109 ST.prev_rex = rex_sv;
7110 ST.prev_curlyx = cur_curlyx;
7112 SET_reg_curpm(rex_sv);
7117 ST.prev_eval = cur_eval;
7119 /* now continue from first node in postoned RE */
7120 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, startpoint, locinput);
7121 NOT_REACHED; /* NOTREACHED */
7124 case EVAL_postponed_AB: /* cleanup after a successful (??{A})B */
7125 /* note: this is called twice; first after popping B, then A */
7127 Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n",
7128 depth, cur_eval, ST.prev_eval);
7131 #define SET_RECURSE_LOCINPUT(STR,VAL)\
7132 if ( cur_eval && CUR_EVAL.close_paren ) {\
7134 Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\
7136 CUR_EVAL.close_paren - 1,\
7140 rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\
7143 SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput);
7145 rex_sv = ST.prev_rex;
7146 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7147 SET_reg_curpm(rex_sv);
7148 rex = ReANY(rex_sv);
7149 rexi = RXi_GET(rex);
7151 /* preserve $^R across LEAVE's. See Bug 121070. */
7152 SV *save_sv= GvSV(PL_replgv);
7153 SvREFCNT_inc(save_sv);
7154 regcpblow(ST.cp); /* LEAVE in disguise */
7155 sv_setsv(GvSV(PL_replgv), save_sv);
7156 SvREFCNT_dec(save_sv);
7158 cur_eval = ST.prev_eval;
7159 cur_curlyx = ST.prev_curlyx;
7161 /* Invalidate cache. See "invalidate" comment above. */
7162 reginfo->poscache_maxiter = 0;
7163 if ( nochange_depth )
7166 SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput);
7170 case EVAL_B_fail: /* unsuccessful B in (?{...})B */
7171 REGCP_UNWIND(ST.lastcp);
7174 case EVAL_postponed_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
7175 /* note: this is called twice; first after popping B, then A */
7177 Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n",
7178 depth, cur_eval, ST.prev_eval);
7181 SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput);
7183 rex_sv = ST.prev_rex;
7184 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7185 SET_reg_curpm(rex_sv);
7186 rex = ReANY(rex_sv);
7187 rexi = RXi_GET(rex);
7189 REGCP_UNWIND(ST.lastcp);
7190 regcppop(rex, &maxopenparen);
7191 cur_eval = ST.prev_eval;
7192 cur_curlyx = ST.prev_curlyx;
7194 /* Invalidate cache. See "invalidate" comment above. */
7195 reginfo->poscache_maxiter = 0;
7196 if ( nochange_depth )
7199 SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput);
7204 n = ARG(scan); /* which paren pair */
7205 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
7206 if (n > maxopenparen)
7208 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
7209 "rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf " tmp; maxopenparen=%" UVuf "\n",
7214 (IV)rex->offs[n].start_tmp,
7220 /* XXX really need to log other places start/end are set too */
7221 #define CLOSE_CAPTURE \
7222 rex->offs[n].start = rex->offs[n].start_tmp; \
7223 rex->offs[n].end = locinput - reginfo->strbeg; \
7224 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
7225 "rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf "..%" IVdf "\n", \
7228 PTR2UV(rex->offs), \
7230 (IV)rex->offs[n].start, \
7231 (IV)rex->offs[n].end \
7235 n = ARG(scan); /* which paren pair */
7237 if (n > rex->lastparen)
7239 rex->lastcloseparen = n;
7240 if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) )
7245 case ACCEPT: /* (*ACCEPT) */
7247 sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7251 cursor && OP(cursor)!=END;
7252 cursor=regnext(cursor))
7254 if ( OP(cursor)==CLOSE ){
7256 if ( n <= lastopen ) {
7258 if (n > rex->lastparen)
7260 rex->lastcloseparen = n;
7261 if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) )
7270 case GROUPP: /* (?(1)) */
7271 n = ARG(scan); /* which paren pair */
7272 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
7275 case NGROUPP: /* (?(<name>)) */
7276 /* reg_check_named_buff_matched returns 0 for no match */
7277 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
7280 case INSUBP: /* (?(R)) */
7282 /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg
7283 * of SCAN is already set up as matches a eval.close_paren */
7284 sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n);
7287 case DEFINEP: /* (?(DEFINE)) */
7291 case IFTHEN: /* (?(cond)A|B) */
7292 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7294 next = NEXTOPER(NEXTOPER(scan));
7296 next = scan + ARG(scan);
7297 if (OP(next) == IFTHEN) /* Fake one. */
7298 next = NEXTOPER(NEXTOPER(next));
7302 case LOGICAL: /* modifier for EVAL and IFMATCH */
7303 logical = scan->flags;
7306 /*******************************************************************
7308 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
7309 pattern, where A and B are subpatterns. (For simple A, CURLYM or
7310 STAR/PLUS/CURLY/CURLYN are used instead.)
7312 A*B is compiled as <CURLYX><A><WHILEM><B>
7314 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
7315 state, which contains the current count, initialised to -1. It also sets
7316 cur_curlyx to point to this state, with any previous value saved in the
7319 CURLYX then jumps straight to the WHILEM op, rather than executing A,
7320 since the pattern may possibly match zero times (i.e. it's a while {} loop
7321 rather than a do {} while loop).
7323 Each entry to WHILEM represents a successful match of A. The count in the
7324 CURLYX block is incremented, another WHILEM state is pushed, and execution
7325 passes to A or B depending on greediness and the current count.
7327 For example, if matching against the string a1a2a3b (where the aN are
7328 substrings that match /A/), then the match progresses as follows: (the
7329 pushed states are interspersed with the bits of strings matched so far):
7332 <CURLYX cnt=0><WHILEM>
7333 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
7334 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
7335 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
7336 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
7338 (Contrast this with something like CURLYM, which maintains only a single
7342 a1 <CURLYM cnt=1> a2
7343 a1 a2 <CURLYM cnt=2> a3
7344 a1 a2 a3 <CURLYM cnt=3> b
7347 Each WHILEM state block marks a point to backtrack to upon partial failure
7348 of A or B, and also contains some minor state data related to that
7349 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
7350 overall state, such as the count, and pointers to the A and B ops.
7352 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
7353 must always point to the *current* CURLYX block, the rules are:
7355 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
7356 and set cur_curlyx to point the new block.
7358 When popping the CURLYX block after a successful or unsuccessful match,
7359 restore the previous cur_curlyx.
7361 When WHILEM is about to execute B, save the current cur_curlyx, and set it
7362 to the outer one saved in the CURLYX block.
7364 When popping the WHILEM block after a successful or unsuccessful B match,
7365 restore the previous cur_curlyx.
7367 Here's an example for the pattern (AI* BI)*BO
7368 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
7371 curlyx backtrack stack
7372 ------ ---------------
7374 CO <CO prev=NULL> <WO>
7375 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7376 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7377 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
7379 At this point the pattern succeeds, and we work back down the stack to
7380 clean up, restoring as we go:
7382 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7383 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7384 CO <CO prev=NULL> <WO>
7387 *******************************************************************/
7389 #define ST st->u.curlyx
7391 case CURLYX: /* start of /A*B/ (for complex A) */
7393 /* No need to save/restore up to this paren */
7394 I32 parenfloor = scan->flags;
7396 assert(next); /* keep Coverity happy */
7397 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
7400 /* XXXX Probably it is better to teach regpush to support
7401 parenfloor > maxopenparen ... */
7402 if (parenfloor > (I32)rex->lastparen)
7403 parenfloor = rex->lastparen; /* Pessimization... */
7405 ST.prev_curlyx= cur_curlyx;
7407 ST.cp = PL_savestack_ix;
7409 /* these fields contain the state of the current curly.
7410 * they are accessed by subsequent WHILEMs */
7411 ST.parenfloor = parenfloor;
7416 ST.count = -1; /* this will be updated by WHILEM */
7417 ST.lastloc = NULL; /* this will be updated by WHILEM */
7419 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
7420 NOT_REACHED; /* NOTREACHED */
7423 case CURLYX_end: /* just finished matching all of A*B */
7424 cur_curlyx = ST.prev_curlyx;
7426 NOT_REACHED; /* NOTREACHED */
7428 case CURLYX_end_fail: /* just failed to match all of A*B */
7430 cur_curlyx = ST.prev_curlyx;
7432 NOT_REACHED; /* NOTREACHED */
7436 #define ST st->u.whilem
7438 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
7440 /* see the discussion above about CURLYX/WHILEM */
7445 assert(cur_curlyx); /* keep Coverity happy */
7447 min = ARG1(cur_curlyx->u.curlyx.me);
7448 max = ARG2(cur_curlyx->u.curlyx.me);
7449 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
7450 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
7451 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
7452 ST.cache_offset = 0;
7456 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: matched %ld out of %d..%d\n",
7457 depth, (long)n, min, max)
7460 /* First just match a string of min A's. */
7463 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, maxopenparen);
7464 cur_curlyx->u.curlyx.lastloc = locinput;
7465 REGCP_SET(ST.lastcp);
7467 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
7468 NOT_REACHED; /* NOTREACHED */
7471 /* If degenerate A matches "", assume A done. */
7473 if (locinput == cur_curlyx->u.curlyx.lastloc) {
7474 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: empty match detected, trying continuation...\n",
7477 goto do_whilem_B_max;
7480 /* super-linear cache processing.
7482 * The idea here is that for certain types of CURLYX/WHILEM -
7483 * principally those whose upper bound is infinity (and
7484 * excluding regexes that have things like \1 and other very
7485 * non-regular expresssiony things), then if a pattern like
7486 * /....A*.../ fails and we backtrack to the WHILEM, then we
7487 * make a note that this particular WHILEM op was at string
7488 * position 47 (say) when the rest of pattern failed. Then, if
7489 * we ever find ourselves back at that WHILEM, and at string
7490 * position 47 again, we can just fail immediately rather than
7491 * running the rest of the pattern again.
7493 * This is very handy when patterns start to go
7494 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
7495 * with a combinatorial explosion of backtracking.
7497 * The cache is implemented as a bit array, with one bit per
7498 * string byte position per WHILEM op (up to 16) - so its
7499 * between 0.25 and 2x the string size.
7501 * To avoid allocating a poscache buffer every time, we do an
7502 * initially countdown; only after we have executed a WHILEM
7503 * op (string-length x #WHILEMs) times do we allocate the
7506 * The top 4 bits of scan->flags byte say how many different
7507 * relevant CURLLYX/WHILEM op pairs there are, while the
7508 * bottom 4-bits is the identifying index number of this
7514 if (!reginfo->poscache_maxiter) {
7515 /* start the countdown: Postpone detection until we
7516 * know the match is not *that* much linear. */
7517 reginfo->poscache_maxiter
7518 = (reginfo->strend - reginfo->strbeg + 1)
7520 /* possible overflow for long strings and many CURLYX's */
7521 if (reginfo->poscache_maxiter < 0)
7522 reginfo->poscache_maxiter = I32_MAX;
7523 reginfo->poscache_iter = reginfo->poscache_maxiter;
7526 if (reginfo->poscache_iter-- == 0) {
7527 /* initialise cache */
7528 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
7529 regmatch_info_aux *const aux = reginfo->info_aux;
7530 if (aux->poscache) {
7531 if ((SSize_t)reginfo->poscache_size < size) {
7532 Renew(aux->poscache, size, char);
7533 reginfo->poscache_size = size;
7535 Zero(aux->poscache, size, char);
7538 reginfo->poscache_size = size;
7539 Newxz(aux->poscache, size, char);
7541 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
7542 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
7543 PL_colors[4], PL_colors[5])
7547 if (reginfo->poscache_iter < 0) {
7548 /* have we already failed at this position? */
7549 SSize_t offset, mask;
7551 reginfo->poscache_iter = -1; /* stop eventual underflow */
7552 offset = (scan->flags & 0xf) - 1
7553 + (locinput - reginfo->strbeg)
7555 mask = 1 << (offset % 8);
7557 if (reginfo->info_aux->poscache[offset] & mask) {
7558 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "whilem: (cache) already tried at this position...\n",
7561 cur_curlyx->u.curlyx.count--;
7562 sayNO; /* cache records failure */
7564 ST.cache_offset = offset;
7565 ST.cache_mask = mask;
7569 /* Prefer B over A for minimal matching. */
7571 if (cur_curlyx->u.curlyx.minmod) {
7572 ST.save_curlyx = cur_curlyx;
7573 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
7574 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
7576 NOT_REACHED; /* NOTREACHED */
7579 /* Prefer A over B for maximal matching. */
7581 if (n < max) { /* More greed allowed? */
7582 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
7584 cur_curlyx->u.curlyx.lastloc = locinput;
7585 REGCP_SET(ST.lastcp);
7586 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
7587 NOT_REACHED; /* NOTREACHED */
7589 goto do_whilem_B_max;
7591 NOT_REACHED; /* NOTREACHED */
7593 case WHILEM_B_min: /* just matched B in a minimal match */
7594 case WHILEM_B_max: /* just matched B in a maximal match */
7595 cur_curlyx = ST.save_curlyx;
7597 NOT_REACHED; /* NOTREACHED */
7599 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
7600 cur_curlyx = ST.save_curlyx;
7601 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7602 cur_curlyx->u.curlyx.count--;
7604 NOT_REACHED; /* NOTREACHED */
7606 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
7607 REGCP_UNWIND(ST.lastcp);
7608 regcppop(rex, &maxopenparen);
7610 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
7611 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7612 cur_curlyx->u.curlyx.count--;
7614 NOT_REACHED; /* NOTREACHED */
7616 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
7617 REGCP_UNWIND(ST.lastcp);
7618 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
7619 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "whilem: failed, trying continuation...\n",
7623 if (cur_curlyx->u.curlyx.count >= REG_INFTY
7624 && ckWARN(WARN_REGEXP)
7625 && !reginfo->warned)
7627 reginfo->warned = TRUE;
7628 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
7629 "Complex regular subexpression recursion limit (%d) "
7635 ST.save_curlyx = cur_curlyx;
7636 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
7637 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
7639 NOT_REACHED; /* NOTREACHED */
7641 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
7642 cur_curlyx = ST.save_curlyx;
7644 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
7645 /* Maximum greed exceeded */
7646 if (cur_curlyx->u.curlyx.count >= REG_INFTY
7647 && ckWARN(WARN_REGEXP)
7648 && !reginfo->warned)
7650 reginfo->warned = TRUE;
7651 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
7652 "Complex regular subexpression recursion "
7653 "limit (%d) exceeded",
7656 cur_curlyx->u.curlyx.count--;
7660 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "trying longer...\n", depth)
7662 /* Try grabbing another A and see if it helps. */
7663 cur_curlyx->u.curlyx.lastloc = locinput;
7664 PUSH_STATE_GOTO(WHILEM_A_min,
7665 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
7667 NOT_REACHED; /* NOTREACHED */
7670 #define ST st->u.branch
7672 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
7673 next = scan + ARG(scan);
7676 scan = NEXTOPER(scan);
7679 case BRANCH: /* /(...|A|...)/ */
7680 scan = NEXTOPER(scan); /* scan now points to inner node */
7681 ST.lastparen = rex->lastparen;
7682 ST.lastcloseparen = rex->lastcloseparen;
7683 ST.next_branch = next;
7686 /* Now go into the branch */
7688 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
7690 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
7692 NOT_REACHED; /* NOTREACHED */
7694 case CUTGROUP: /* /(*THEN)/ */
7695 sv_yes_mark = st->u.mark.mark_name = scan->flags
7696 ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ])
7698 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
7699 NOT_REACHED; /* NOTREACHED */
7701 case CUTGROUP_next_fail:
7704 if (st->u.mark.mark_name)
7705 sv_commit = st->u.mark.mark_name;
7707 NOT_REACHED; /* NOTREACHED */
7711 NOT_REACHED; /* NOTREACHED */
7713 case BRANCH_next_fail: /* that branch failed; try the next, if any */
7718 REGCP_UNWIND(ST.cp);
7719 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7720 scan = ST.next_branch;
7721 /* no more branches? */
7722 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
7724 Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n",
7731 continue; /* execute next BRANCH[J] op */
7734 case MINMOD: /* next op will be non-greedy, e.g. A*? */
7739 #define ST st->u.curlym
7741 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
7743 /* This is an optimisation of CURLYX that enables us to push
7744 * only a single backtracking state, no matter how many matches
7745 * there are in {m,n}. It relies on the pattern being constant
7746 * length, with no parens to influence future backrefs
7750 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7752 ST.lastparen = rex->lastparen;
7753 ST.lastcloseparen = rex->lastcloseparen;
7755 /* if paren positive, emulate an OPEN/CLOSE around A */
7757 U32 paren = ST.me->flags;
7758 if (paren > maxopenparen)
7759 maxopenparen = paren;
7760 scan += NEXT_OFF(scan); /* Skip former OPEN. */
7768 ST.c1 = CHRTEST_UNINIT;
7771 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
7774 curlym_do_A: /* execute the A in /A{m,n}B/ */
7775 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
7776 NOT_REACHED; /* NOTREACHED */
7778 case CURLYM_A: /* we've just matched an A */
7780 /* after first match, determine A's length: u.curlym.alen */
7781 if (ST.count == 1) {
7782 if (reginfo->is_utf8_target) {
7783 char *s = st->locinput;
7784 while (s < locinput) {
7790 ST.alen = locinput - st->locinput;
7793 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
7796 Perl_re_exec_indentf( aTHX_ "CURLYM now matched %" IVdf " times, len=%" IVdf "...\n",
7797 depth, (IV) ST.count, (IV)ST.alen)
7800 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7804 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
7805 if ( max == REG_INFTY || ST.count < max )
7806 goto curlym_do_A; /* try to match another A */
7808 goto curlym_do_B; /* try to match B */
7810 case CURLYM_A_fail: /* just failed to match an A */
7811 REGCP_UNWIND(ST.cp);
7814 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
7815 || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7818 curlym_do_B: /* execute the B in /A{m,n}B/ */
7819 if (ST.c1 == CHRTEST_UNINIT) {
7820 /* calculate c1 and c2 for possible match of 1st char
7821 * following curly */
7822 ST.c1 = ST.c2 = CHRTEST_VOID;
7824 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
7825 regnode *text_node = ST.B;
7826 if (! HAS_TEXT(text_node))
7827 FIND_NEXT_IMPT(text_node);
7830 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
7832 But the former is redundant in light of the latter.
7834 if this changes back then the macro for
7835 IS_TEXT and friends need to change.
7837 if (PL_regkind[OP(text_node)] == EXACT) {
7838 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
7839 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
7849 Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%" IVdf "...\n",
7850 depth, (IV)ST.count)
7852 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
7853 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
7854 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
7855 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
7857 /* simulate B failing */
7859 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%" UVXf " c1=0x%" UVXf " c2=0x%" UVXf "\n",
7861 valid_utf8_to_uvchr((U8 *) locinput, NULL),
7862 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
7863 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
7865 state_num = CURLYM_B_fail;
7866 goto reenter_switch;
7869 else if (nextchr != ST.c1 && nextchr != ST.c2) {
7870 /* simulate B failing */
7872 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
7874 (int) nextchr, ST.c1, ST.c2)
7876 state_num = CURLYM_B_fail;
7877 goto reenter_switch;
7882 /* emulate CLOSE: mark current A as captured */
7883 I32 paren = ST.me->flags;
7885 rex->offs[paren].start
7886 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
7887 rex->offs[paren].end = locinput - reginfo->strbeg;
7888 if ((U32)paren > rex->lastparen)
7889 rex->lastparen = paren;
7890 rex->lastcloseparen = paren;
7893 rex->offs[paren].end = -1;
7895 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
7904 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
7905 NOT_REACHED; /* NOTREACHED */
7907 case CURLYM_B_fail: /* just failed to match a B */
7908 REGCP_UNWIND(ST.cp);
7909 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7911 I32 max = ARG2(ST.me);
7912 if (max != REG_INFTY && ST.count == max)
7914 goto curlym_do_A; /* try to match a further A */
7916 /* backtrack one A */
7917 if (ST.count == ARG1(ST.me) /* min */)
7920 SET_locinput(HOPc(locinput, -ST.alen));
7921 goto curlym_do_B; /* try to match B */
7924 #define ST st->u.curly
7926 #define CURLY_SETPAREN(paren, success) \
7929 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
7930 rex->offs[paren].end = locinput - reginfo->strbeg; \
7931 if (paren > rex->lastparen) \
7932 rex->lastparen = paren; \
7933 rex->lastcloseparen = paren; \
7936 rex->offs[paren].end = -1; \
7937 rex->lastparen = ST.lastparen; \
7938 rex->lastcloseparen = ST.lastcloseparen; \
7942 case STAR: /* /A*B/ where A is width 1 char */
7946 scan = NEXTOPER(scan);
7949 case PLUS: /* /A+B/ where A is width 1 char */
7953 scan = NEXTOPER(scan);
7956 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
7957 ST.paren = scan->flags; /* Which paren to set */
7958 ST.lastparen = rex->lastparen;
7959 ST.lastcloseparen = rex->lastcloseparen;
7960 if (ST.paren > maxopenparen)
7961 maxopenparen = ST.paren;
7962 ST.min = ARG1(scan); /* min to match */
7963 ST.max = ARG2(scan); /* max to match */
7964 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
7969 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
7972 case CURLY: /* /A{m,n}B/ where A is width 1 char */
7974 ST.min = ARG1(scan); /* min to match */
7975 ST.max = ARG2(scan); /* max to match */
7976 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7979 * Lookahead to avoid useless match attempts
7980 * when we know what character comes next.
7982 * Used to only do .*x and .*?x, but now it allows
7983 * for )'s, ('s and (?{ ... })'s to be in the way
7984 * of the quantifier and the EXACT-like node. -- japhy
7987 assert(ST.min <= ST.max);
7988 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
7989 ST.c1 = ST.c2 = CHRTEST_VOID;
7992 regnode *text_node = next;
7994 if (! HAS_TEXT(text_node))
7995 FIND_NEXT_IMPT(text_node);
7997 if (! HAS_TEXT(text_node))
7998 ST.c1 = ST.c2 = CHRTEST_VOID;
8000 if ( PL_regkind[OP(text_node)] != EXACT ) {
8001 ST.c1 = ST.c2 = CHRTEST_VOID;
8005 /* Currently we only get here when
8007 PL_rekind[OP(text_node)] == EXACT
8009 if this changes back then the macro for IS_TEXT and
8010 friends need to change. */
8011 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
8012 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
8024 char *li = locinput;
8027 regrepeat(rex, &li, ST.A, reginfo, ST.min)
8033 if (ST.c1 == CHRTEST_VOID)
8034 goto curly_try_B_min;
8036 ST.oldloc = locinput;
8038 /* set ST.maxpos to the furthest point along the
8039 * string that could possibly match */
8040 if (ST.max == REG_INFTY) {
8041 ST.maxpos = reginfo->strend - 1;
8043 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
8046 else if (utf8_target) {
8047 int m = ST.max - ST.min;
8048 for (ST.maxpos = locinput;
8049 m >0 && ST.maxpos < reginfo->strend; m--)
8050 ST.maxpos += UTF8SKIP(ST.maxpos);
8053 ST.maxpos = locinput + ST.max - ST.min;
8054 if (ST.maxpos >= reginfo->strend)
8055 ST.maxpos = reginfo->strend - 1;
8057 goto curly_try_B_min_known;
8061 /* avoid taking address of locinput, so it can remain
8063 char *li = locinput;
8064 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max);
8065 if (ST.count < ST.min)
8068 if ((ST.count > ST.min)
8069 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
8071 /* A{m,n} must come at the end of the string, there's
8072 * no point in backing off ... */
8074 /* ...except that $ and \Z can match before *and* after
8075 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
8076 We may back off by one in this case. */
8077 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
8081 goto curly_try_B_max;
8083 NOT_REACHED; /* NOTREACHED */
8085 case CURLY_B_min_known_fail:
8086 /* failed to find B in a non-greedy match where c1,c2 valid */
8088 REGCP_UNWIND(ST.cp);
8090 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8092 /* Couldn't or didn't -- move forward. */
8093 ST.oldloc = locinput;
8095 locinput += UTF8SKIP(locinput);
8099 curly_try_B_min_known:
8100 /* find the next place where 'B' could work, then call B */
8104 n = (ST.oldloc == locinput) ? 0 : 1;
8105 if (ST.c1 == ST.c2) {
8106 /* set n to utf8_distance(oldloc, locinput) */
8107 while (locinput <= ST.maxpos
8108 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
8110 locinput += UTF8SKIP(locinput);
8115 /* set n to utf8_distance(oldloc, locinput) */
8116 while (locinput <= ST.maxpos
8117 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
8118 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
8120 locinput += UTF8SKIP(locinput);
8125 else { /* Not utf8_target */
8126 if (ST.c1 == ST.c2) {
8127 while (locinput <= ST.maxpos &&
8128 UCHARAT(locinput) != ST.c1)
8132 while (locinput <= ST.maxpos
8133 && UCHARAT(locinput) != ST.c1
8134 && UCHARAT(locinput) != ST.c2)
8137 n = locinput - ST.oldloc;
8139 if (locinput > ST.maxpos)
8142 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
8143 * at b; check that everything between oldloc and
8144 * locinput matches */
8145 char *li = ST.oldloc;
8147 if (regrepeat(rex, &li, ST.A, reginfo, n) < n)
8149 assert(n == REG_INFTY || locinput == li);
8151 CURLY_SETPAREN(ST.paren, ST.count);
8152 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8154 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
8156 NOT_REACHED; /* NOTREACHED */
8158 case CURLY_B_min_fail:
8159 /* failed to find B in a non-greedy match where c1,c2 invalid */
8161 REGCP_UNWIND(ST.cp);
8163 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8165 /* failed -- move forward one */
8167 char *li = locinput;
8168 if (!regrepeat(rex, &li, ST.A, reginfo, 1)) {
8175 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
8176 ST.count > 0)) /* count overflow ? */
8179 CURLY_SETPAREN(ST.paren, ST.count);
8180 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8182 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
8186 NOT_REACHED; /* NOTREACHED */
8189 /* a successful greedy match: now try to match B */
8190 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8193 bool could_match = locinput < reginfo->strend;
8195 /* If it could work, try it. */
8196 if (ST.c1 != CHRTEST_VOID && could_match) {
8197 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
8199 could_match = memEQ(locinput,
8204 UTF8SKIP(locinput));
8207 could_match = UCHARAT(locinput) == ST.c1
8208 || UCHARAT(locinput) == ST.c2;
8211 if (ST.c1 == CHRTEST_VOID || could_match) {
8212 CURLY_SETPAREN(ST.paren, ST.count);
8213 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
8214 NOT_REACHED; /* NOTREACHED */
8219 case CURLY_B_max_fail:
8220 /* failed to find B in a greedy match */
8222 REGCP_UNWIND(ST.cp);
8224 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8227 if (--ST.count < ST.min)
8229 locinput = HOPc(locinput, -1);
8230 goto curly_try_B_max;
8234 case END: /* last op of main pattern */
8237 /* we've just finished A in /(??{A})B/; now continue with B */
8238 SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput);
8239 st->u.eval.prev_rex = rex_sv; /* inner */
8241 /* Save *all* the positions. */
8242 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
8243 rex_sv = CUR_EVAL.prev_rex;
8244 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
8245 SET_reg_curpm(rex_sv);
8246 rex = ReANY(rex_sv);
8247 rexi = RXi_GET(rex);
8249 st->u.eval.prev_curlyx = cur_curlyx;
8250 cur_curlyx = CUR_EVAL.prev_curlyx;
8252 REGCP_SET(st->u.eval.lastcp);
8254 /* Restore parens of the outer rex without popping the
8256 regcp_restore(rex, CUR_EVAL.lastcp, &maxopenparen);
8258 st->u.eval.prev_eval = cur_eval;
8259 cur_eval = CUR_EVAL.prev_eval;
8261 Perl_re_exec_indentf( aTHX_ "EVAL trying tail ... (cur_eval=%p)\n",
8263 if ( nochange_depth )
8266 SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput);
8268 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, st->u.eval.prev_eval->u.eval.B,
8269 locinput); /* match B */
8272 if (locinput < reginfo->till) {
8273 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
8274 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
8276 (long)(locinput - startpos),
8277 (long)(reginfo->till - startpos),
8280 sayNO_SILENT; /* Cannot match: too short. */
8282 sayYES; /* Success! */
8284 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
8286 Perl_re_exec_indentf( aTHX_ "%ssubpattern success...%s\n",
8287 depth, PL_colors[4], PL_colors[5]));
8288 sayYES; /* Success! */
8291 #define ST st->u.ifmatch
8296 case SUSPEND: /* (?>A) */
8298 newstart = locinput;
8301 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
8303 goto ifmatch_trivial_fail_test;
8305 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
8307 ifmatch_trivial_fail_test:
8309 char * const s = HOPBACKc(locinput, scan->flags);
8314 sw = 1 - cBOOL(ST.wanted);
8318 next = scan + ARG(scan);
8326 newstart = locinput;
8330 ST.logical = logical;
8331 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
8333 /* execute body of (?...A) */
8334 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
8335 NOT_REACHED; /* NOTREACHED */
8338 case IFMATCH_A_fail: /* body of (?...A) failed */
8339 ST.wanted = !ST.wanted;
8342 case IFMATCH_A: /* body of (?...A) succeeded */
8344 sw = cBOOL(ST.wanted);
8346 else if (!ST.wanted)
8349 if (OP(ST.me) != SUSPEND) {
8350 /* restore old position except for (?>...) */
8351 locinput = st->locinput;
8353 scan = ST.me + ARG(ST.me);
8356 continue; /* execute B */
8360 case LONGJMP: /* alternative with many branches compiles to
8361 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
8362 next = scan + ARG(scan);
8367 case COMMIT: /* (*COMMIT) */
8368 reginfo->cutpoint = reginfo->strend;
8371 case PRUNE: /* (*PRUNE) */
8373 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8374 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
8375 NOT_REACHED; /* NOTREACHED */
8377 case COMMIT_next_fail:
8381 NOT_REACHED; /* NOTREACHED */
8383 case OPFAIL: /* (*FAIL) */
8385 sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8387 /* deal with (?(?!)X|Y) properly,
8388 * make sure we trigger the no branch
8389 * of the trailing IFTHEN structure*/
8395 NOT_REACHED; /* NOTREACHED */
8397 #define ST st->u.mark
8398 case MARKPOINT: /* (*MARK:foo) */
8399 ST.prev_mark = mark_state;
8400 ST.mark_name = sv_commit = sv_yes_mark
8401 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8403 ST.mark_loc = locinput;
8404 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
8405 NOT_REACHED; /* NOTREACHED */
8407 case MARKPOINT_next:
8408 mark_state = ST.prev_mark;
8410 NOT_REACHED; /* NOTREACHED */
8412 case MARKPOINT_next_fail:
8413 if (popmark && sv_eq(ST.mark_name,popmark))
8415 if (ST.mark_loc > startpoint)
8416 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8417 popmark = NULL; /* we found our mark */
8418 sv_commit = ST.mark_name;
8421 Perl_re_exec_indentf( aTHX_ "%ssetting cutpoint to mark:%" SVf "...%s\n",
8423 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
8426 mark_state = ST.prev_mark;
8427 sv_yes_mark = mark_state ?
8428 mark_state->u.mark.mark_name : NULL;
8430 NOT_REACHED; /* NOTREACHED */
8432 case SKIP: /* (*SKIP) */
8434 /* (*SKIP) : if we fail we cut here*/
8435 ST.mark_name = NULL;
8436 ST.mark_loc = locinput;
8437 PUSH_STATE_GOTO(SKIP_next,next, locinput);
8439 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
8440 otherwise do nothing. Meaning we need to scan
8442 regmatch_state *cur = mark_state;
8443 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8446 if ( sv_eq( cur->u.mark.mark_name,
8449 ST.mark_name = find;
8450 PUSH_STATE_GOTO( SKIP_next, next, locinput);
8452 cur = cur->u.mark.prev_mark;
8455 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
8458 case SKIP_next_fail:
8460 /* (*CUT:NAME) - Set up to search for the name as we
8461 collapse the stack*/
8462 popmark = ST.mark_name;
8464 /* (*CUT) - No name, we cut here.*/
8465 if (ST.mark_loc > startpoint)
8466 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8467 /* but we set sv_commit to latest mark_name if there
8468 is one so they can test to see how things lead to this
8471 sv_commit=mark_state->u.mark.mark_name;
8475 NOT_REACHED; /* NOTREACHED */
8478 case LNBREAK: /* \R */
8479 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
8486 PerlIO_printf(Perl_error_log, "%" UVxf " %d\n",
8487 PTR2UV(scan), OP(scan));
8488 Perl_croak(aTHX_ "regexp memory corruption");
8490 /* this is a point to jump to in order to increment
8491 * locinput by one character */
8493 assert(!NEXTCHR_IS_EOS);
8495 locinput += PL_utf8skip[nextchr];
8496 /* locinput is allowed to go 1 char off the end (signifying
8497 * EOS), but not 2+ */
8498 if (locinput > reginfo->strend)
8507 /* switch break jumps here */
8508 scan = next; /* prepare to execute the next op and ... */
8509 continue; /* ... jump back to the top, reusing st */
8513 /* push a state that backtracks on success */
8514 st->u.yes.prev_yes_state = yes_state;
8518 /* push a new regex state, then continue at scan */
8520 regmatch_state *newst;
8523 regmatch_state *cur = st;
8524 regmatch_state *curyes = yes_state;
8526 regmatch_slab *slab = PL_regmatch_slab;
8527 for (i = 0; i < 3 && i <= depth; cur--,i++) {
8528 if (cur < SLAB_FIRST(slab)) {
8530 cur = SLAB_LAST(slab);
8532 Perl_re_exec_indentf( aTHX_ "%4s #%-3d %-10s %s\n",
8535 depth - i, PL_reg_name[cur->resume_state],
8536 (curyes == cur) ? "yes" : ""
8539 curyes = cur->u.yes.prev_yes_state;
8542 DEBUG_STATE_pp("push")
8545 st->locinput = locinput;
8547 if (newst > SLAB_LAST(PL_regmatch_slab))
8548 newst = S_push_slab(aTHX);
8549 PL_regmatch_state = newst;
8551 locinput = pushinput;
8557 #ifdef SOLARIS_BAD_OPTIMIZER
8558 # undef PL_charclass
8562 * We get here only if there's trouble -- normally "case END" is
8563 * the terminating point.
8565 Perl_croak(aTHX_ "corrupted regexp pointers");
8566 NOT_REACHED; /* NOTREACHED */
8570 /* we have successfully completed a subexpression, but we must now
8571 * pop to the state marked by yes_state and continue from there */
8572 assert(st != yes_state);
8574 while (st != yes_state) {
8576 if (st < SLAB_FIRST(PL_regmatch_slab)) {
8577 PL_regmatch_slab = PL_regmatch_slab->prev;
8578 st = SLAB_LAST(PL_regmatch_slab);
8582 DEBUG_STATE_pp("pop (no final)");
8584 DEBUG_STATE_pp("pop (yes)");
8590 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
8591 || yes_state > SLAB_LAST(PL_regmatch_slab))
8593 /* not in this slab, pop slab */
8594 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
8595 PL_regmatch_slab = PL_regmatch_slab->prev;
8596 st = SLAB_LAST(PL_regmatch_slab);
8598 depth -= (st - yes_state);
8601 yes_state = st->u.yes.prev_yes_state;
8602 PL_regmatch_state = st;
8605 locinput= st->locinput;
8606 state_num = st->resume_state + no_final;
8607 goto reenter_switch;
8610 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n",
8611 PL_colors[4], PL_colors[5]));
8613 if (reginfo->info_aux_eval) {
8614 /* each successfully executed (?{...}) block does the equivalent of
8615 * local $^R = do {...}
8616 * When popping the save stack, all these locals would be undone;
8617 * bypass this by setting the outermost saved $^R to the latest
8619 /* I dont know if this is needed or works properly now.
8620 * see code related to PL_replgv elsewhere in this file.
8623 if (oreplsv != GvSV(PL_replgv))
8624 sv_setsv(oreplsv, GvSV(PL_replgv));
8631 Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n",
8633 PL_colors[4], PL_colors[5])
8645 /* there's a previous state to backtrack to */
8647 if (st < SLAB_FIRST(PL_regmatch_slab)) {
8648 PL_regmatch_slab = PL_regmatch_slab->prev;
8649 st = SLAB_LAST(PL_regmatch_slab);
8651 PL_regmatch_state = st;
8652 locinput= st->locinput;
8654 DEBUG_STATE_pp("pop");
8656 if (yes_state == st)
8657 yes_state = st->u.yes.prev_yes_state;
8659 state_num = st->resume_state + 1; /* failure = success + 1 */
8661 goto reenter_switch;
8666 if (rex->intflags & PREGf_VERBARG_SEEN) {
8667 SV *sv_err = get_sv("REGERROR", 1);
8668 SV *sv_mrk = get_sv("REGMARK", 1);
8670 sv_commit = &PL_sv_no;
8672 sv_yes_mark = &PL_sv_yes;
8675 sv_commit = &PL_sv_yes;
8676 sv_yes_mark = &PL_sv_no;
8680 sv_setsv(sv_err, sv_commit);
8681 sv_setsv(sv_mrk, sv_yes_mark);
8685 if (last_pushed_cv) {
8687 /* see "Some notes about MULTICALL" above */
8689 PERL_UNUSED_VAR(SP);
8692 LEAVE_SCOPE(orig_savestack_ix);
8694 assert(!result || locinput - reginfo->strbeg >= 0);
8695 return result ? locinput - reginfo->strbeg : -1;
8699 - regrepeat - repeatedly match something simple, report how many
8701 * What 'simple' means is a node which can be the operand of a quantifier like
8704 * startposp - pointer a pointer to the start position. This is updated
8705 * to point to the byte following the highest successful
8707 * p - the regnode to be repeatedly matched against.
8708 * reginfo - struct holding match state, such as strend
8709 * max - maximum number of things to match.
8710 * depth - (for debugging) backtracking depth.
8713 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
8714 regmatch_info *const reginfo, I32 max _pDEPTH)
8716 char *scan; /* Pointer to current position in target string */
8718 char *loceol = reginfo->strend; /* local version */
8719 I32 hardcount = 0; /* How many matches so far */
8720 bool utf8_target = reginfo->is_utf8_target;
8721 unsigned int to_complement = 0; /* Invert the result? */
8723 _char_class_number classnum;
8725 PERL_ARGS_ASSERT_REGREPEAT;
8728 if (max == REG_INFTY)
8730 else if (! utf8_target && loceol - scan > max)
8731 loceol = scan + max;
8733 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
8734 * to the maximum of how far we should go in it (leaving it set to the real
8735 * end, if the maximum permissible would take us beyond that). This allows
8736 * us to make the loop exit condition that we haven't gone past <loceol> to
8737 * also mean that we haven't exceeded the max permissible count, saving a
8738 * test each time through the loop. But it assumes that the OP matches a
8739 * single byte, which is true for most of the OPs below when applied to a
8740 * non-UTF-8 target. Those relatively few OPs that don't have this
8741 * characteristic will have to compensate.
8743 * There is no adjustment for UTF-8 targets, as the number of bytes per
8744 * character varies. OPs will have to test both that the count is less
8745 * than the max permissible (using <hardcount> to keep track), and that we
8746 * are still within the bounds of the string (using <loceol>. A few OPs
8747 * match a single byte no matter what the encoding. They can omit the max
8748 * test if, for the UTF-8 case, they do the adjustment that was skipped
8751 * Thus, the code above sets things up for the common case; and exceptional
8752 * cases need extra work; the common case is to make sure <scan> doesn't
8753 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
8754 * count doesn't exceed the maximum permissible */
8759 while (scan < loceol && hardcount < max && *scan != '\n') {
8760 scan += UTF8SKIP(scan);
8764 while (scan < loceol && *scan != '\n')
8770 while (scan < loceol && hardcount < max) {
8771 scan += UTF8SKIP(scan);
8779 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8780 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
8781 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
8785 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8789 /* Can use a simple loop if the pattern char to match on is invariant
8790 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
8791 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
8792 * true iff it doesn't matter if the argument is in UTF-8 or not */
8793 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
8794 if (utf8_target && loceol - scan > max) {
8795 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
8796 * since here, to match at all, 1 char == 1 byte */
8797 loceol = scan + max;
8799 while (scan < loceol && UCHARAT(scan) == c) {
8803 else if (reginfo->is_utf8_pat) {
8805 STRLEN scan_char_len;
8807 /* When both target and pattern are UTF-8, we have to do
8809 while (hardcount < max
8811 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
8812 && memEQ(scan, STRING(p), scan_char_len))
8814 scan += scan_char_len;
8818 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
8820 /* Target isn't utf8; convert the character in the UTF-8
8821 * pattern to non-UTF8, and do a simple loop */
8822 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
8823 while (scan < loceol && UCHARAT(scan) == c) {
8826 } /* else pattern char is above Latin1, can't possibly match the
8831 /* Here, the string must be utf8; pattern isn't, and <c> is
8832 * different in utf8 than not, so can't compare them directly.
8833 * Outside the loop, find the two utf8 bytes that represent c, and
8834 * then look for those in sequence in the utf8 string */
8835 U8 high = UTF8_TWO_BYTE_HI(c);
8836 U8 low = UTF8_TWO_BYTE_LO(c);
8838 while (hardcount < max
8839 && scan + 1 < loceol
8840 && UCHARAT(scan) == high
8841 && UCHARAT(scan + 1) == low)
8849 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
8850 assert(! reginfo->is_utf8_pat);
8853 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
8857 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8858 utf8_flags = FOLDEQ_LOCALE;
8861 case EXACTF: /* This node only generated for non-utf8 patterns */
8862 assert(! reginfo->is_utf8_pat);
8867 if (! utf8_target) {
8870 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
8871 | FOLDEQ_S2_FOLDS_SANE;
8876 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
8880 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
8882 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8884 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
8887 if (c1 == CHRTEST_VOID) {
8888 /* Use full Unicode fold matching */
8889 char *tmpeol = reginfo->strend;
8890 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
8891 while (hardcount < max
8892 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
8893 STRING(p), NULL, pat_len,
8894 reginfo->is_utf8_pat, utf8_flags))
8897 tmpeol = reginfo->strend;
8901 else if (utf8_target) {
8903 while (scan < loceol
8905 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
8907 scan += UTF8SKIP(scan);
8912 while (scan < loceol
8914 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
8915 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
8917 scan += UTF8SKIP(scan);
8922 else if (c1 == c2) {
8923 while (scan < loceol && UCHARAT(scan) == c1) {
8928 while (scan < loceol &&
8929 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
8938 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8940 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) {
8941 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
8947 while (hardcount < max
8949 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
8951 scan += UTF8SKIP(scan);
8955 else if (ANYOF_FLAGS(p)) {
8956 while (scan < loceol
8957 && reginclass(prog, p, (U8*)scan, (U8*)scan+1, 0))
8961 while (scan < loceol && ANYOF_BITMAP_TEST(p, *((U8*)scan)))
8966 /* The argument (FLAGS) to all the POSIX node types is the class number */
8973 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8974 if (! utf8_target) {
8975 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
8981 while (hardcount < max && scan < loceol
8982 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
8985 scan += UTF8SKIP(scan);
8998 if (utf8_target && loceol - scan > max) {
9000 /* We didn't adjust <loceol> at the beginning of this routine
9001 * because is UTF-8, but it is actually ok to do so, since here, to
9002 * match, 1 char == 1 byte. */
9003 loceol = scan + max;
9005 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
9018 if (! utf8_target) {
9019 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
9025 /* The complement of something that matches only ASCII matches all
9026 * non-ASCII, plus everything in ASCII that isn't in the class. */
9027 while (hardcount < max && scan < loceol
9028 && ( ! isASCII_utf8_safe(scan, reginfo->strend)
9029 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
9031 scan += UTF8SKIP(scan);
9042 if (! utf8_target) {
9043 while (scan < loceol && to_complement
9044 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
9051 classnum = (_char_class_number) FLAGS(p);
9052 if (classnum < _FIRST_NON_SWASH_CC) {
9054 /* Here, a swash is needed for above-Latin1 code points.
9055 * Process as many Latin1 code points using the built-in rules.
9056 * Go to another loop to finish processing upon encountering
9057 * the first Latin1 code point. We could do that in this loop
9058 * as well, but the other way saves having to test if the swash
9059 * has been loaded every time through the loop: extra space to
9061 while (hardcount < max && scan < loceol) {
9062 if (UTF8_IS_INVARIANT(*scan)) {
9063 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
9070 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
9071 if (! (to_complement
9072 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*scan,
9081 goto found_above_latin1;
9088 /* For these character classes, the knowledge of how to handle
9089 * every code point is compiled in to Perl via a macro. This
9090 * code is written for making the loops as tight as possible.
9091 * It could be refactored to save space instead */
9093 case _CC_ENUM_SPACE:
9094 while (hardcount < max
9097 ^ cBOOL(isSPACE_utf8_safe(scan, loceol))))
9099 scan += UTF8SKIP(scan);
9103 case _CC_ENUM_BLANK:
9104 while (hardcount < max
9107 ^ cBOOL(isBLANK_utf8_safe(scan, loceol))))
9109 scan += UTF8SKIP(scan);
9113 case _CC_ENUM_XDIGIT:
9114 while (hardcount < max
9117 ^ cBOOL(isXDIGIT_utf8_safe(scan, loceol))))
9119 scan += UTF8SKIP(scan);
9123 case _CC_ENUM_VERTSPACE:
9124 while (hardcount < max
9127 ^ cBOOL(isVERTWS_utf8_safe(scan, loceol))))
9129 scan += UTF8SKIP(scan);
9133 case _CC_ENUM_CNTRL:
9134 while (hardcount < max
9137 ^ cBOOL(isCNTRL_utf8_safe(scan, loceol))))
9139 scan += UTF8SKIP(scan);
9144 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
9150 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
9152 /* Load the swash if not already present */
9153 if (! PL_utf8_swash_ptrs[classnum]) {
9154 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
9155 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
9159 PL_XPosix_ptrs[classnum], &flags);
9162 while (hardcount < max && scan < loceol
9163 && to_complement ^ cBOOL(_generic_utf8_safe(
9167 swash_fetch(PL_utf8_swash_ptrs[classnum],
9171 scan += UTF8SKIP(scan);
9178 while (hardcount < max && scan < loceol &&
9179 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
9184 /* LNBREAK can match one or two latin chars, which is ok, but we
9185 * have to use hardcount in this situation, and throw away the
9186 * adjustment to <loceol> done before the switch statement */
9187 loceol = reginfo->strend;
9188 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
9197 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9211 /* These are all 0 width, so match right here or not at all. */
9215 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
9216 NOT_REACHED; /* NOTREACHED */
9223 c = scan - *startposp;
9227 GET_RE_DEBUG_FLAGS_DECL;
9229 SV * const prop = sv_newmortal();
9230 regprop(prog, prop, p, reginfo, NULL);
9231 Perl_re_exec_indentf( aTHX_ "%s can match %" IVdf " times out of %" IVdf "...\n",
9232 depth, SvPVX_const(prop),(IV)c,(IV)max);
9240 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
9242 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
9243 create a copy so that changes the caller makes won't change the shared one.
9244 If <altsvp> is non-null, will return NULL in it, for back-compat.
9247 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
9249 PERL_ARGS_ASSERT_REGCLASS_SWASH;
9255 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL));
9258 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
9261 - reginclass - determine if a character falls into a character class
9263 n is the ANYOF-type regnode
9264 p is the target string
9265 p_end points to one byte beyond the end of the target string
9266 utf8_target tells whether p is in UTF-8.
9268 Returns true if matched; false otherwise.
9270 Note that this can be a synthetic start class, a combination of various
9271 nodes, so things you think might be mutually exclusive, such as locale,
9272 aren't. It can match both locale and non-locale
9277 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
9280 const char flags = ANYOF_FLAGS(n);
9284 PERL_ARGS_ASSERT_REGINCLASS;
9286 /* If c is not already the code point, get it. Note that
9287 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
9288 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
9290 const U32 utf8n_flags = UTF8_ALLOW_DEFAULT;
9291 c = utf8n_to_uvchr(p, p_end - p, &c_len, utf8n_flags | UTF8_CHECK_ONLY);
9292 if (c_len == (STRLEN)-1) {
9293 _force_out_malformed_utf8_message(p, p_end,
9295 1 /* 1 means die */ );
9296 NOT_REACHED; /* NOTREACHED */
9298 if (c > 255 && OP(n) == ANYOFL && ! ANYOFL_UTF8_LOCALE_REQD(flags)) {
9299 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
9303 /* If this character is potentially in the bitmap, check it */
9304 if (c < NUM_ANYOF_CODE_POINTS) {
9305 if (ANYOF_BITMAP_TEST(n, c))
9308 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9315 else if (flags & ANYOF_LOCALE_FLAGS) {
9316 if ((flags & ANYOFL_FOLD)
9318 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
9322 else if (ANYOF_POSIXL_TEST_ANY_SET(n)
9326 /* The data structure is arranged so bits 0, 2, 4, ... are set
9327 * if the class includes the Posix character class given by
9328 * bit/2; and 1, 3, 5, ... are set if the class includes the
9329 * complemented Posix class given by int(bit/2). So we loop
9330 * through the bits, each time changing whether we complement
9331 * the result or not. Suppose for the sake of illustration
9332 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
9333 * is set, it means there is a match for this ANYOF node if the
9334 * character is in the class given by the expression (0 / 2 = 0
9335 * = \w). If it is in that class, isFOO_lc() will return 1,
9336 * and since 'to_complement' is 0, the result will stay TRUE,
9337 * and we exit the loop. Suppose instead that bit 0 is 0, but
9338 * bit 1 is 1. That means there is a match if the character
9339 * matches \W. We won't bother to call isFOO_lc() on bit 0,
9340 * but will on bit 1. On the second iteration 'to_complement'
9341 * will be 1, so the exclusive or will reverse things, so we
9342 * are testing for \W. On the third iteration, 'to_complement'
9343 * will be 0, and we would be testing for \s; the fourth
9344 * iteration would test for \S, etc.
9346 * Note that this code assumes that all the classes are closed
9347 * under folding. For example, if a character matches \w, then
9348 * its fold does too; and vice versa. This should be true for
9349 * any well-behaved locale for all the currently defined Posix
9350 * classes, except for :lower: and :upper:, which are handled
9351 * by the pseudo-class :cased: which matches if either of the
9352 * other two does. To get rid of this assumption, an outer
9353 * loop could be used below to iterate over both the source
9354 * character, and its fold (if different) */
9357 int to_complement = 0;
9359 while (count < ANYOF_MAX) {
9360 if (ANYOF_POSIXL_TEST(n, count)
9361 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
9374 /* If the bitmap didn't (or couldn't) match, and something outside the
9375 * bitmap could match, try that. */
9377 if (c >= NUM_ANYOF_CODE_POINTS
9378 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
9380 match = TRUE; /* Everything above the bitmap matches */
9382 /* Here doesn't match everything above the bitmap. If there is
9383 * some information available beyond the bitmap, we may find a
9384 * match in it. If so, this is most likely because the code point
9385 * is outside the bitmap range. But rarely, it could be because of
9386 * some other reason. If so, various flags are set to indicate
9387 * this possibility. On ANYOFD nodes, there may be matches that
9388 * happen only when the target string is UTF-8; or for other node
9389 * types, because runtime lookup is needed, regardless of the
9390 * UTF-8ness of the target string. Finally, under /il, there may
9391 * be some matches only possible if the locale is a UTF-8 one. */
9392 else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP
9393 && ( c >= NUM_ANYOF_CODE_POINTS
9394 || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)
9395 && ( UNLIKELY(OP(n) != ANYOFD)
9396 || (utf8_target && ! isASCII_uni(c)
9397 # if NUM_ANYOF_CODE_POINTS > 256
9401 || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags)
9402 && IN_UTF8_CTYPE_LOCALE)))
9404 SV* only_utf8_locale = NULL;
9405 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
9406 &only_utf8_locale, NULL);
9412 } else { /* Convert to utf8 */
9413 utf8_p = utf8_buffer;
9414 append_utf8_from_native_byte(*p, &utf8_p);
9415 utf8_p = utf8_buffer;
9418 if (swash_fetch(sw, utf8_p, TRUE)) {
9422 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
9423 match = _invlist_contains_cp(only_utf8_locale, c);
9427 if (UNICODE_IS_SUPER(c)
9429 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9431 && ckWARN_d(WARN_NON_UNICODE))
9433 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
9434 "Matched non-Unicode code point 0x%04" UVXf " against Unicode property; may not be portable", c);
9438 #if ANYOF_INVERT != 1
9439 /* Depending on compiler optimization cBOOL takes time, so if don't have to
9441 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
9444 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
9445 return (flags & ANYOF_INVERT) ^ match;
9449 S_reghop3(U8 *s, SSize_t off, const U8* lim)
9451 /* return the position 'off' UTF-8 characters away from 's', forward if
9452 * 'off' >= 0, backwards if negative. But don't go outside of position
9453 * 'lim', which better be < s if off < 0 */
9455 PERL_ARGS_ASSERT_REGHOP3;
9458 while (off-- && s < lim) {
9459 /* XXX could check well-formedness here */
9464 while (off++ && s > lim) {
9466 if (UTF8_IS_CONTINUED(*s)) {
9467 while (s > lim && UTF8_IS_CONTINUATION(*s))
9469 if (! UTF8_IS_START(*s)) {
9470 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9473 /* XXX could check well-formedness here */
9480 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
9482 PERL_ARGS_ASSERT_REGHOP4;
9485 while (off-- && s < rlim) {
9486 /* XXX could check well-formedness here */
9491 while (off++ && s > llim) {
9493 if (UTF8_IS_CONTINUED(*s)) {
9494 while (s > llim && UTF8_IS_CONTINUATION(*s))
9496 if (! UTF8_IS_START(*s)) {
9497 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9500 /* XXX could check well-formedness here */
9506 /* like reghop3, but returns NULL on overrun, rather than returning last
9510 S_reghopmaybe3(U8* s, SSize_t off, const U8* const lim)
9512 PERL_ARGS_ASSERT_REGHOPMAYBE3;
9515 while (off-- && s < lim) {
9516 /* XXX could check well-formedness here */
9523 while (off++ && s > lim) {
9525 if (UTF8_IS_CONTINUED(*s)) {
9526 while (s > lim && UTF8_IS_CONTINUATION(*s))
9528 if (! UTF8_IS_START(*s)) {
9529 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9532 /* XXX could check well-formedness here */
9541 /* when executing a regex that may have (?{}), extra stuff needs setting
9542 up that will be visible to the called code, even before the current
9543 match has finished. In particular:
9545 * $_ is localised to the SV currently being matched;
9546 * pos($_) is created if necessary, ready to be updated on each call-out
9548 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
9549 isn't set until the current pattern is successfully finished), so that
9550 $1 etc of the match-so-far can be seen;
9551 * save the old values of subbeg etc of the current regex, and set then
9552 to the current string (again, this is normally only done at the end
9557 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
9560 regexp *const rex = ReANY(reginfo->prog);
9561 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
9563 eval_state->rex = rex;
9566 /* Make $_ available to executed code. */
9567 if (reginfo->sv != DEFSV) {
9569 DEFSV_set(reginfo->sv);
9572 if (!(mg = mg_find_mglob(reginfo->sv))) {
9573 /* prepare for quick setting of pos */
9574 mg = sv_magicext_mglob(reginfo->sv);
9577 eval_state->pos_magic = mg;
9578 eval_state->pos = mg->mg_len;
9579 eval_state->pos_flags = mg->mg_flags;
9582 eval_state->pos_magic = NULL;
9584 if (!PL_reg_curpm) {
9585 /* PL_reg_curpm is a fake PMOP that we can attach the current
9586 * regex to and point PL_curpm at, so that $1 et al are visible
9587 * within a /(?{})/. It's just allocated once per interpreter the
9588 * first time its needed */
9589 Newxz(PL_reg_curpm, 1, PMOP);
9592 SV* const repointer = &PL_sv_undef;
9593 /* this regexp is also owned by the new PL_reg_curpm, which
9594 will try to free it. */
9595 av_push(PL_regex_padav, repointer);
9596 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
9597 PL_regex_pad = AvARRAY(PL_regex_padav);
9601 SET_reg_curpm(reginfo->prog);
9602 eval_state->curpm = PL_curpm;
9603 PL_curpm_under = PL_curpm;
9604 PL_curpm = PL_reg_curpm;
9605 if (RXp_MATCH_COPIED(rex)) {
9606 /* Here is a serious problem: we cannot rewrite subbeg,
9607 since it may be needed if this match fails. Thus
9608 $` inside (?{}) could fail... */
9609 eval_state->subbeg = rex->subbeg;
9610 eval_state->sublen = rex->sublen;
9611 eval_state->suboffset = rex->suboffset;
9612 eval_state->subcoffset = rex->subcoffset;
9614 eval_state->saved_copy = rex->saved_copy;
9616 RXp_MATCH_COPIED_off(rex);
9619 eval_state->subbeg = NULL;
9620 rex->subbeg = (char *)reginfo->strbeg;
9622 rex->subcoffset = 0;
9623 rex->sublen = reginfo->strend - reginfo->strbeg;
9627 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
9630 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
9632 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
9633 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
9636 Safefree(aux->poscache);
9640 /* undo the effects of S_setup_eval_state() */
9642 if (eval_state->subbeg) {
9643 regexp * const rex = eval_state->rex;
9644 rex->subbeg = eval_state->subbeg;
9645 rex->sublen = eval_state->sublen;
9646 rex->suboffset = eval_state->suboffset;
9647 rex->subcoffset = eval_state->subcoffset;
9649 rex->saved_copy = eval_state->saved_copy;
9651 RXp_MATCH_COPIED_on(rex);
9653 if (eval_state->pos_magic)
9655 eval_state->pos_magic->mg_len = eval_state->pos;
9656 eval_state->pos_magic->mg_flags =
9657 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
9658 | (eval_state->pos_flags & MGf_BYTES);
9661 PL_curpm = eval_state->curpm;
9664 PL_regmatch_state = aux->old_regmatch_state;
9665 PL_regmatch_slab = aux->old_regmatch_slab;
9667 /* free all slabs above current one - this must be the last action
9668 * of this function, as aux and eval_state are allocated within
9669 * slabs and may be freed here */
9671 s = PL_regmatch_slab->next;
9673 PL_regmatch_slab->next = NULL;
9675 regmatch_slab * const osl = s;
9684 S_to_utf8_substr(pTHX_ regexp *prog)
9686 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
9687 * on the converted value */
9691 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
9694 if (prog->substrs->data[i].substr
9695 && !prog->substrs->data[i].utf8_substr) {
9696 SV* const sv = newSVsv(prog->substrs->data[i].substr);
9697 prog->substrs->data[i].utf8_substr = sv;
9698 sv_utf8_upgrade(sv);
9699 if (SvVALID(prog->substrs->data[i].substr)) {
9700 if (SvTAIL(prog->substrs->data[i].substr)) {
9701 /* Trim the trailing \n that fbm_compile added last
9703 SvCUR_set(sv, SvCUR(sv) - 1);
9704 /* Whilst this makes the SV technically "invalid" (as its
9705 buffer is no longer followed by "\0") when fbm_compile()
9706 adds the "\n" back, a "\0" is restored. */
9707 fbm_compile(sv, FBMcf_TAIL);
9711 if (prog->substrs->data[i].substr == prog->check_substr)
9712 prog->check_utf8 = sv;
9718 S_to_byte_substr(pTHX_ regexp *prog)
9720 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
9721 * on the converted value; returns FALSE if can't be converted. */
9725 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
9728 if (prog->substrs->data[i].utf8_substr
9729 && !prog->substrs->data[i].substr) {
9730 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
9731 if (! sv_utf8_downgrade(sv, TRUE)) {
9734 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
9735 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
9736 /* Trim the trailing \n that fbm_compile added last
9738 SvCUR_set(sv, SvCUR(sv) - 1);
9739 fbm_compile(sv, FBMcf_TAIL);
9743 prog->substrs->data[i].substr = sv;
9744 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
9745 prog->check_substr = sv;
9753 Perl__is_grapheme(pTHX_ const U8 * strbeg, const U8 * s, const U8 * strend, const UV cp)
9755 /* Temporary helper function for toke.c. Verify that the code point 'cp'
9756 * is a stand-alone grapheme. The UTF-8 for 'cp' begins at position 's' in
9757 * the larger string bounded by 'strbeg' and 'strend'.
9759 * 'cp' needs to be assigned (if not a future version of the Unicode
9760 * Standard could make it something that combines with adjacent characters,
9761 * so code using it would then break), and there has to be a GCB break
9762 * before and after the character. */
9764 GCB_enum cp_gcb_val, prev_cp_gcb_val, next_cp_gcb_val;
9765 const U8 * prev_cp_start;
9767 PERL_ARGS_ASSERT__IS_GRAPHEME;
9769 /* Unassigned code points are forbidden */
9770 if (UNLIKELY(! ELEMENT_RANGE_MATCHES_INVLIST(
9771 _invlist_search(PL_Assigned_invlist, cp))))
9776 cp_gcb_val = getGCB_VAL_CP(cp);
9778 /* Find the GCB value of the previous code point in the input */
9779 prev_cp_start = utf8_hop_back(s, -1, strbeg);
9780 if (UNLIKELY(prev_cp_start == s)) {
9781 prev_cp_gcb_val = GCB_EDGE;
9784 prev_cp_gcb_val = getGCB_VAL_UTF8(prev_cp_start, strend);
9787 /* And check that is a grapheme boundary */
9788 if (! isGCB(prev_cp_gcb_val, cp_gcb_val, strbeg, s,
9789 TRUE /* is UTF-8 encoded */ ))
9794 /* Similarly verify there is a break between the current character and the
9798 next_cp_gcb_val = GCB_EDGE;
9801 next_cp_gcb_val = getGCB_VAL_UTF8(s, strend);
9804 return isGCB(cp_gcb_val, next_cp_gcb_val, strbeg, s, TRUE);
9811 * ex: set ts=8 sts=4 sw=4 et: